Move private ABI counter client symbols to dedicated header
[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 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 static char *get_map_shm(struct sock_info *sock_info);
366
367 ssize_t lttng_ust_read(int fd, void *buf, size_t len)
368 {
369 ssize_t ret;
370 size_t copied = 0, to_copy = len;
371
372 do {
373 ret = read(fd, buf + copied, to_copy);
374 if (ret > 0) {
375 copied += ret;
376 to_copy -= ret;
377 }
378 } while ((ret > 0 && to_copy > 0)
379 || (ret < 0 && errno == EINTR));
380 if (ret > 0) {
381 ret = copied;
382 }
383 return ret;
384 }
385 /*
386 * Returns the HOME directory path. Caller MUST NOT free(3) the returned
387 * pointer.
388 */
389 static
390 const char *get_lttng_home_dir(void)
391 {
392 const char *val;
393
394 val = (const char *) lttng_ust_getenv("LTTNG_HOME");
395 if (val != NULL) {
396 return val;
397 }
398 return (const char *) lttng_ust_getenv("HOME");
399 }
400
401 /*
402 * Force a read (imply TLS fixup for dlopen) of TLS variables.
403 */
404 static
405 void lttng_fixup_nest_count_tls(void)
406 {
407 asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count)));
408 }
409
410 static
411 void lttng_fixup_ust_mutex_nest_tls(void)
412 {
413 asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest)));
414 }
415
416 /*
417 * Fixup lttng-ust urcu TLS.
418 */
419 static
420 void lttng_fixup_lttng_ust_urcu_tls(void)
421 {
422 (void) lttng_ust_urcu_read_ongoing();
423 }
424
425 void lttng_ust_fixup_tls(void)
426 {
427 lttng_fixup_lttng_ust_urcu_tls();
428 lttng_fixup_ringbuffer_tls();
429 lttng_fixup_vtid_tls();
430 lttng_fixup_nest_count_tls();
431 lttng_fixup_procname_tls();
432 lttng_fixup_ust_mutex_nest_tls();
433 lttng_ust_fixup_perf_counter_tls();
434 lttng_ust_fixup_fd_tracker_tls();
435 lttng_fixup_cgroup_ns_tls();
436 lttng_fixup_ipc_ns_tls();
437 lttng_fixup_net_ns_tls();
438 lttng_fixup_time_ns_tls();
439 lttng_fixup_uts_ns_tls();
440 }
441
442 int lttng_get_notify_socket(void *owner)
443 {
444 struct sock_info *info = owner;
445
446 return info->notify_socket;
447 }
448
449
450 char* lttng_ust_sockinfo_get_procname(void *owner)
451 {
452 struct sock_info *info = owner;
453
454 return info->procname;
455 }
456
457 static
458 void print_cmd(int cmd, int handle)
459 {
460 const char *cmd_name = "Unknown";
461
462 if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping)
463 && cmd_name_mapping[cmd]) {
464 cmd_name = cmd_name_mapping[cmd];
465 }
466 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
467 cmd_name, cmd,
468 lttng_ust_obj_get_name(handle), handle);
469 }
470
471 static
472 int setup_global_apps(void)
473 {
474 int ret = 0;
475 assert(!global_apps.wait_shm_mmap);
476
477 global_apps.wait_shm_mmap = get_map_shm(&global_apps);
478 if (!global_apps.wait_shm_mmap) {
479 WARN("Unable to get map shm for global apps. Disabling LTTng-UST global tracing.");
480 global_apps.allowed = 0;
481 ret = -EIO;
482 goto error;
483 }
484
485 global_apps.allowed = 1;
486 lttng_pthread_getname_np(global_apps.procname, LTTNG_UST_ABI_PROCNAME_LEN);
487 error:
488 return ret;
489 }
490 static
491 int setup_local_apps(void)
492 {
493 int ret = 0;
494 const char *home_dir;
495 uid_t uid;
496
497 assert(!local_apps.wait_shm_mmap);
498
499 uid = getuid();
500 /*
501 * Disallow per-user tracing for setuid binaries.
502 */
503 if (uid != geteuid()) {
504 assert(local_apps.allowed == 0);
505 ret = 0;
506 goto end;
507 }
508 home_dir = get_lttng_home_dir();
509 if (!home_dir) {
510 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
511 assert(local_apps.allowed == 0);
512 ret = -ENOENT;
513 goto end;
514 }
515 local_apps.allowed = 1;
516 snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
517 home_dir,
518 LTTNG_DEFAULT_HOME_RUNDIR,
519 LTTNG_UST_SOCK_FILENAME);
520 snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
521 LTTNG_UST_WAIT_FILENAME,
522 uid);
523
524 local_apps.wait_shm_mmap = get_map_shm(&local_apps);
525 if (!local_apps.wait_shm_mmap) {
526 WARN("Unable to get map shm for local apps. Disabling LTTng-UST per-user tracing.");
527 local_apps.allowed = 0;
528 ret = -EIO;
529 goto end;
530 }
531
532 lttng_pthread_getname_np(local_apps.procname, LTTNG_UST_ABI_PROCNAME_LEN);
533 end:
534 return ret;
535 }
536
537 /*
538 * Get socket timeout, in ms.
539 * -1: wait forever. 0: don't wait. >0: timeout, in ms.
540 */
541 static
542 long get_timeout(void)
543 {
544 long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
545
546 if (!got_timeout_env) {
547 str_timeout = lttng_ust_getenv("LTTNG_UST_REGISTER_TIMEOUT");
548 got_timeout_env = 1;
549 }
550 if (str_timeout)
551 constructor_delay_ms = strtol(str_timeout, NULL, 10);
552 /* All negative values are considered as "-1". */
553 if (constructor_delay_ms < -1)
554 constructor_delay_ms = -1;
555 return constructor_delay_ms;
556 }
557
558 /* Timeout for notify socket send and recv. */
559 static
560 long get_notify_sock_timeout(void)
561 {
562 return get_timeout();
563 }
564
565 /* Timeout for connecting to cmd and notify sockets. */
566 static
567 long get_connect_sock_timeout(void)
568 {
569 return get_timeout();
570 }
571
572 /*
573 * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
574 */
575 static
576 int get_constructor_timeout(struct timespec *constructor_timeout)
577 {
578 long constructor_delay_ms;
579 int ret;
580
581 constructor_delay_ms = get_timeout();
582
583 switch (constructor_delay_ms) {
584 case -1:/* fall-through */
585 case 0:
586 return constructor_delay_ms;
587 default:
588 break;
589 }
590
591 /*
592 * If we are unable to find the current time, don't wait.
593 */
594 ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
595 if (ret) {
596 /* Don't wait. */
597 return 0;
598 }
599 constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
600 constructor_timeout->tv_nsec +=
601 (constructor_delay_ms % 1000UL) * 1000000UL;
602 if (constructor_timeout->tv_nsec >= 1000000000UL) {
603 constructor_timeout->tv_sec++;
604 constructor_timeout->tv_nsec -= 1000000000UL;
605 }
606 /* Timeout wait (constructor_delay_ms). */
607 return 1;
608 }
609
610 static
611 void get_allow_blocking(void)
612 {
613 const char *str_allow_blocking =
614 lttng_ust_getenv("LTTNG_UST_ALLOW_BLOCKING");
615
616 if (str_allow_blocking) {
617 DBG("%s environment variable is set",
618 "LTTNG_UST_ALLOW_BLOCKING");
619 lttng_ust_ringbuffer_set_allow_blocking();
620 }
621 }
622
623 static
624 int register_to_sessiond(int socket, enum ustctl_socket_type type)
625 {
626 return ustcomm_send_reg_msg(socket,
627 type,
628 CAA_BITS_PER_LONG,
629 lttng_ust_rb_alignof(uint8_t) * CHAR_BIT,
630 lttng_ust_rb_alignof(uint16_t) * CHAR_BIT,
631 lttng_ust_rb_alignof(uint32_t) * CHAR_BIT,
632 lttng_ust_rb_alignof(uint64_t) * CHAR_BIT,
633 lttng_ust_rb_alignof(unsigned long) * CHAR_BIT);
634 }
635
636 static
637 int send_reply(int sock, struct ustcomm_ust_reply *lur)
638 {
639 ssize_t len;
640
641 len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
642 switch (len) {
643 case sizeof(*lur):
644 DBG("message successfully sent");
645 return 0;
646 default:
647 if (len == -ECONNRESET) {
648 DBG("remote end closed connection");
649 return 0;
650 }
651 if (len < 0)
652 return len;
653 DBG("incorrect message size: %zd", len);
654 return -EINVAL;
655 }
656 }
657
658 static
659 void decrement_sem_count(unsigned int count)
660 {
661 int ret;
662
663 assert(uatomic_read(&sem_count) >= count);
664
665 if (uatomic_read(&sem_count) <= 0) {
666 return;
667 }
668
669 ret = uatomic_add_return(&sem_count, -count);
670 if (ret == 0) {
671 ret = sem_post(&constructor_wait);
672 assert(!ret);
673 }
674 }
675
676 static
677 int handle_register_done(struct sock_info *sock_info)
678 {
679 if (sock_info->registration_done)
680 return 0;
681 sock_info->registration_done = 1;
682
683 decrement_sem_count(1);
684 if (!sock_info->statedump_pending) {
685 sock_info->initial_statedump_done = 1;
686 decrement_sem_count(1);
687 }
688
689 return 0;
690 }
691
692 static
693 int handle_register_failed(struct sock_info *sock_info)
694 {
695 if (sock_info->registration_done)
696 return 0;
697 sock_info->registration_done = 1;
698 sock_info->initial_statedump_done = 1;
699
700 decrement_sem_count(2);
701
702 return 0;
703 }
704
705 /*
706 * Only execute pending statedump after the constructor semaphore has
707 * been posted by the current listener thread. This means statedump will
708 * only be performed after the "registration done" command is received
709 * from this thread's session daemon.
710 *
711 * This ensures we don't run into deadlock issues with the dynamic
712 * loader mutex, which is held while the constructor is called and
713 * waiting on the constructor semaphore. All operations requiring this
714 * dynamic loader lock need to be postponed using this mechanism.
715 *
716 * In a scenario with two session daemons connected to the application,
717 * it is possible that the first listener thread which receives the
718 * registration done command issues its statedump while the dynamic
719 * loader lock is still held by the application constructor waiting on
720 * the semaphore. It will however be allowed to proceed when the
721 * second session daemon sends the registration done command to the
722 * second listener thread. This situation therefore does not produce
723 * a deadlock.
724 */
725 static
726 void handle_pending_statedump(struct sock_info *sock_info)
727 {
728 if (sock_info->registration_done && sock_info->statedump_pending) {
729 sock_info->statedump_pending = 0;
730 pthread_mutex_lock(&ust_fork_mutex);
731 lttng_handle_pending_statedump(sock_info);
732 pthread_mutex_unlock(&ust_fork_mutex);
733
734 if (!sock_info->initial_statedump_done) {
735 sock_info->initial_statedump_done = 1;
736 decrement_sem_count(1);
737 }
738 }
739 }
740
741 static inline
742 const char *bytecode_type_str(uint32_t cmd)
743 {
744 switch (cmd) {
745 case LTTNG_UST_ABI_CAPTURE:
746 return "capture";
747 case LTTNG_UST_ABI_FILTER:
748 return "filter";
749 default:
750 abort();
751 }
752 }
753
754 static
755 int handle_bytecode_recv(struct sock_info *sock_info,
756 int sock, struct ustcomm_ust_msg *lum)
757 {
758 struct lttng_ust_bytecode_node *bytecode = NULL;
759 enum lttng_ust_bytecode_type type;
760 const struct lttng_ust_abi_objd_ops *ops;
761 uint32_t data_size, data_size_max, reloc_offset;
762 uint64_t seqnum;
763 ssize_t len;
764 int ret = 0;
765
766 switch (lum->cmd) {
767 case LTTNG_UST_ABI_FILTER:
768 type = LTTNG_UST_BYTECODE_TYPE_FILTER;
769 data_size = lum->u.filter.data_size;
770 data_size_max = LTTNG_UST_ABI_FILTER_BYTECODE_MAX_LEN;
771 reloc_offset = lum->u.filter.reloc_offset;
772 seqnum = lum->u.filter.seqnum;
773 break;
774 case LTTNG_UST_ABI_CAPTURE:
775 type = LTTNG_UST_BYTECODE_TYPE_CAPTURE;
776 data_size = lum->u.capture.data_size;
777 data_size_max = LTTNG_UST_ABI_CAPTURE_BYTECODE_MAX_LEN;
778 reloc_offset = lum->u.capture.reloc_offset;
779 seqnum = lum->u.capture.seqnum;
780 break;
781 default:
782 abort();
783 }
784
785 if (data_size > data_size_max) {
786 ERR("Bytecode %s data size is too large: %u bytes",
787 bytecode_type_str(lum->cmd), data_size);
788 ret = -EINVAL;
789 goto end;
790 }
791
792 if (reloc_offset > data_size) {
793 ERR("Bytecode %s reloc offset %u is not within data",
794 bytecode_type_str(lum->cmd), reloc_offset);
795 ret = -EINVAL;
796 goto end;
797 }
798
799 /* Allocate the structure AND the `data[]` field. */
800 bytecode = zmalloc(sizeof(*bytecode) + data_size);
801 if (!bytecode) {
802 ret = -ENOMEM;
803 goto end;
804 }
805
806 bytecode->bc.len = data_size;
807 bytecode->bc.reloc_offset = reloc_offset;
808 bytecode->bc.seqnum = seqnum;
809 bytecode->type = type;
810
811 len = ustcomm_recv_unix_sock(sock, bytecode->bc.data, bytecode->bc.len);
812 switch (len) {
813 case 0: /* orderly shutdown */
814 ret = 0;
815 goto end;
816 default:
817 if (len == bytecode->bc.len) {
818 DBG("Bytecode %s data received",
819 bytecode_type_str(lum->cmd));
820 break;
821 } else if (len < 0) {
822 DBG("Receive failed from lttng-sessiond with errno %d",
823 (int) -len);
824 if (len == -ECONNRESET) {
825 ERR("%s remote end closed connection",
826 sock_info->name);
827 ret = len;
828 goto end;
829 }
830 ret = len;
831 goto end;
832 } else {
833 DBG("Incorrect %s bytecode data message size: %zd",
834 bytecode_type_str(lum->cmd), len);
835 ret = -EINVAL;
836 goto end;
837 }
838 }
839
840 ops = lttng_ust_abi_objd_ops(lum->handle);
841 if (!ops) {
842 ret = -ENOENT;
843 goto end;
844 }
845
846 if (ops->cmd)
847 ret = ops->cmd(lum->handle, lum->cmd,
848 (unsigned long) &bytecode,
849 NULL, sock_info);
850 else
851 ret = -ENOSYS;
852
853 end:
854 free(bytecode);
855 return ret;
856 }
857
858 static
859 int handle_message(struct sock_info *sock_info,
860 int sock, struct ustcomm_ust_msg *lum)
861 {
862 int ret = 0;
863 const struct lttng_ust_abi_objd_ops *ops;
864 struct ustcomm_ust_reply lur;
865 union lttng_ust_abi_args args;
866 char ctxstr[LTTNG_UST_ABI_SYM_NAME_LEN]; /* App context string. */
867 ssize_t len;
868
869 memset(&lur, 0, sizeof(lur));
870
871 if (ust_lock()) {
872 ret = -LTTNG_UST_ERR_EXITING;
873 goto error;
874 }
875
876 ops = lttng_ust_abi_objd_ops(lum->handle);
877 if (!ops) {
878 ret = -ENOENT;
879 goto error;
880 }
881
882 switch (lum->cmd) {
883 case LTTNG_UST_ABI_REGISTER_DONE:
884 if (lum->handle == LTTNG_UST_ABI_ROOT_HANDLE)
885 ret = handle_register_done(sock_info);
886 else
887 ret = -EINVAL;
888 break;
889 case LTTNG_UST_ABI_RELEASE:
890 if (lum->handle == LTTNG_UST_ABI_ROOT_HANDLE)
891 ret = -EPERM;
892 else
893 ret = lttng_ust_abi_objd_unref(lum->handle, 1);
894 break;
895 case LTTNG_UST_ABI_CAPTURE:
896 case LTTNG_UST_ABI_FILTER:
897 ret = handle_bytecode_recv(sock_info, sock, lum);
898 if (ret)
899 goto error;
900 break;
901 case LTTNG_UST_ABI_EXCLUSION:
902 {
903 /* Receive exclusion names */
904 struct lttng_ust_excluder_node *node;
905 unsigned int count;
906
907 count = lum->u.exclusion.count;
908 if (count == 0) {
909 /* There are no names to read */
910 ret = 0;
911 goto error;
912 }
913 node = zmalloc(sizeof(*node) +
914 count * LTTNG_UST_ABI_SYM_NAME_LEN);
915 if (!node) {
916 ret = -ENOMEM;
917 goto error;
918 }
919 node->excluder.count = count;
920 len = ustcomm_recv_unix_sock(sock, node->excluder.names,
921 count * LTTNG_UST_ABI_SYM_NAME_LEN);
922 switch (len) {
923 case 0: /* orderly shutdown */
924 ret = 0;
925 free(node);
926 goto error;
927 default:
928 if (len == count * LTTNG_UST_ABI_SYM_NAME_LEN) {
929 DBG("Exclusion data received");
930 break;
931 } else if (len < 0) {
932 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
933 if (len == -ECONNRESET) {
934 ERR("%s remote end closed connection", sock_info->name);
935 ret = len;
936 free(node);
937 goto error;
938 }
939 ret = len;
940 free(node);
941 goto error;
942 } else {
943 DBG("Incorrect exclusion data message size: %zd", len);
944 ret = -EINVAL;
945 free(node);
946 goto error;
947 }
948 }
949 if (ops->cmd)
950 ret = ops->cmd(lum->handle, lum->cmd,
951 (unsigned long) &node,
952 &args, sock_info);
953 else
954 ret = -ENOSYS;
955 free(node);
956 break;
957 }
958 case LTTNG_UST_ABI_EVENT_NOTIFIER_GROUP_CREATE:
959 {
960 int event_notifier_notif_fd, close_ret;
961
962 len = ustcomm_recv_event_notifier_notif_fd_from_sessiond(sock,
963 &event_notifier_notif_fd);
964 switch (len) {
965 case 0: /* orderly shutdown */
966 ret = 0;
967 goto error;
968 case 1:
969 break;
970 default:
971 if (len < 0) {
972 DBG("Receive failed from lttng-sessiond with errno %d",
973 (int) -len);
974 if (len == -ECONNRESET) {
975 ERR("%s remote end closed connection",
976 sock_info->name);
977 ret = len;
978 goto error;
979 }
980 ret = len;
981 goto error;
982 } else {
983 DBG("Incorrect event notifier fd message size: %zd",
984 len);
985 ret = -EINVAL;
986 goto error;
987 }
988 }
989 args.event_notifier_handle.event_notifier_notif_fd =
990 event_notifier_notif_fd;
991 if (ops->cmd)
992 ret = ops->cmd(lum->handle, lum->cmd,
993 (unsigned long) &lum->u,
994 &args, sock_info);
995 else
996 ret = -ENOSYS;
997 if (args.event_notifier_handle.event_notifier_notif_fd >= 0) {
998 lttng_ust_lock_fd_tracker();
999 close_ret = close(args.event_notifier_handle.event_notifier_notif_fd);
1000 lttng_ust_unlock_fd_tracker();
1001 if (close_ret)
1002 PERROR("close");
1003 }
1004 break;
1005 }
1006 case LTTNG_UST_ABI_CHANNEL:
1007 {
1008 void *chan_data;
1009 int wakeup_fd;
1010
1011 len = ustcomm_recv_channel_from_sessiond(sock,
1012 &chan_data, lum->u.channel.len,
1013 &wakeup_fd);
1014 switch (len) {
1015 case 0: /* orderly shutdown */
1016 ret = 0;
1017 goto error;
1018 default:
1019 if (len == lum->u.channel.len) {
1020 DBG("channel data received");
1021 break;
1022 } else if (len < 0) {
1023 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1024 if (len == -ECONNRESET) {
1025 ERR("%s remote end closed connection", sock_info->name);
1026 ret = len;
1027 goto error;
1028 }
1029 ret = len;
1030 goto error;
1031 } else {
1032 DBG("incorrect channel data message size: %zd", len);
1033 ret = -EINVAL;
1034 goto error;
1035 }
1036 }
1037 args.channel.chan_data = chan_data;
1038 args.channel.wakeup_fd = wakeup_fd;
1039 if (ops->cmd)
1040 ret = ops->cmd(lum->handle, lum->cmd,
1041 (unsigned long) &lum->u,
1042 &args, sock_info);
1043 else
1044 ret = -ENOSYS;
1045 if (args.channel.wakeup_fd >= 0) {
1046 int close_ret;
1047
1048 lttng_ust_lock_fd_tracker();
1049 close_ret = close(args.channel.wakeup_fd);
1050 lttng_ust_unlock_fd_tracker();
1051 args.channel.wakeup_fd = -1;
1052 if (close_ret)
1053 PERROR("close");
1054 }
1055 free(args.channel.chan_data);
1056 break;
1057 }
1058 case LTTNG_UST_ABI_STREAM:
1059 {
1060 int close_ret;
1061
1062 /* Receive shm_fd, wakeup_fd */
1063 ret = ustcomm_recv_stream_from_sessiond(sock,
1064 NULL,
1065 &args.stream.shm_fd,
1066 &args.stream.wakeup_fd);
1067 if (ret) {
1068 goto error;
1069 }
1070
1071 if (ops->cmd)
1072 ret = ops->cmd(lum->handle, lum->cmd,
1073 (unsigned long) &lum->u,
1074 &args, sock_info);
1075 else
1076 ret = -ENOSYS;
1077 if (args.stream.shm_fd >= 0) {
1078 lttng_ust_lock_fd_tracker();
1079 close_ret = close(args.stream.shm_fd);
1080 lttng_ust_unlock_fd_tracker();
1081 args.stream.shm_fd = -1;
1082 if (close_ret)
1083 PERROR("close");
1084 }
1085 if (args.stream.wakeup_fd >= 0) {
1086 lttng_ust_lock_fd_tracker();
1087 close_ret = close(args.stream.wakeup_fd);
1088 lttng_ust_unlock_fd_tracker();
1089 args.stream.wakeup_fd = -1;
1090 if (close_ret)
1091 PERROR("close");
1092 }
1093 break;
1094 }
1095 case LTTNG_UST_ABI_CONTEXT:
1096 switch (lum->u.context.ctx) {
1097 case LTTNG_UST_ABI_CONTEXT_APP_CONTEXT:
1098 {
1099 char *p;
1100 size_t ctxlen, recvlen;
1101
1102 ctxlen = strlen("$app.") + lum->u.context.u.app_ctx.provider_name_len - 1
1103 + strlen(":") + lum->u.context.u.app_ctx.ctx_name_len;
1104 if (ctxlen >= LTTNG_UST_ABI_SYM_NAME_LEN) {
1105 ERR("Application context string length size is too large: %zu bytes",
1106 ctxlen);
1107 ret = -EINVAL;
1108 goto error;
1109 }
1110 strcpy(ctxstr, "$app.");
1111 p = &ctxstr[strlen("$app.")];
1112 recvlen = ctxlen - strlen("$app.");
1113 len = ustcomm_recv_unix_sock(sock, p, recvlen);
1114 switch (len) {
1115 case 0: /* orderly shutdown */
1116 ret = 0;
1117 goto error;
1118 default:
1119 if (len == recvlen) {
1120 DBG("app context data received");
1121 break;
1122 } else if (len < 0) {
1123 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1124 if (len == -ECONNRESET) {
1125 ERR("%s remote end closed connection", sock_info->name);
1126 ret = len;
1127 goto error;
1128 }
1129 ret = len;
1130 goto error;
1131 } else {
1132 DBG("incorrect app context data message size: %zd", len);
1133 ret = -EINVAL;
1134 goto error;
1135 }
1136 }
1137 /* Put : between provider and ctxname. */
1138 p[lum->u.context.u.app_ctx.provider_name_len - 1] = ':';
1139 args.app_context.ctxname = ctxstr;
1140 break;
1141 }
1142 default:
1143 break;
1144 }
1145 if (ops->cmd) {
1146 ret = ops->cmd(lum->handle, lum->cmd,
1147 (unsigned long) &lum->u,
1148 &args, sock_info);
1149 } else {
1150 ret = -ENOSYS;
1151 }
1152 break;
1153 case LTTNG_UST_ABI_COUNTER:
1154 {
1155 void *counter_data;
1156
1157 len = ustcomm_recv_counter_from_sessiond(sock,
1158 &counter_data, lum->u.counter.len);
1159 switch (len) {
1160 case 0: /* orderly shutdown */
1161 ret = 0;
1162 goto error;
1163 default:
1164 if (len == lum->u.counter.len) {
1165 DBG("counter data received");
1166 break;
1167 } else if (len < 0) {
1168 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1169 if (len == -ECONNRESET) {
1170 ERR("%s remote end closed connection", sock_info->name);
1171 ret = len;
1172 goto error;
1173 }
1174 ret = len;
1175 goto error;
1176 } else {
1177 DBG("incorrect counter data message size: %zd", len);
1178 ret = -EINVAL;
1179 goto error;
1180 }
1181 }
1182 args.counter.counter_data = counter_data;
1183 if (ops->cmd)
1184 ret = ops->cmd(lum->handle, lum->cmd,
1185 (unsigned long) &lum->u,
1186 &args, sock_info);
1187 else
1188 ret = -ENOSYS;
1189 free(args.counter.counter_data);
1190 break;
1191 }
1192 case LTTNG_UST_ABI_COUNTER_GLOBAL:
1193 {
1194 /* Receive shm_fd */
1195 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1196 &args.counter_shm.shm_fd);
1197 if (ret) {
1198 goto error;
1199 }
1200
1201 if (ops->cmd)
1202 ret = ops->cmd(lum->handle, lum->cmd,
1203 (unsigned long) &lum->u,
1204 &args, sock_info);
1205 else
1206 ret = -ENOSYS;
1207 if (args.counter_shm.shm_fd >= 0) {
1208 int close_ret;
1209
1210 lttng_ust_lock_fd_tracker();
1211 close_ret = close(args.counter_shm.shm_fd);
1212 lttng_ust_unlock_fd_tracker();
1213 args.counter_shm.shm_fd = -1;
1214 if (close_ret)
1215 PERROR("close");
1216 }
1217 break;
1218 }
1219 case LTTNG_UST_ABI_COUNTER_CPU:
1220 {
1221 /* Receive shm_fd */
1222 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1223 &args.counter_shm.shm_fd);
1224 if (ret) {
1225 goto error;
1226 }
1227
1228 if (ops->cmd)
1229 ret = ops->cmd(lum->handle, lum->cmd,
1230 (unsigned long) &lum->u,
1231 &args, sock_info);
1232 else
1233 ret = -ENOSYS;
1234 if (args.counter_shm.shm_fd >= 0) {
1235 int close_ret;
1236
1237 lttng_ust_lock_fd_tracker();
1238 close_ret = close(args.counter_shm.shm_fd);
1239 lttng_ust_unlock_fd_tracker();
1240 args.counter_shm.shm_fd = -1;
1241 if (close_ret)
1242 PERROR("close");
1243 }
1244 break;
1245 }
1246 case LTTNG_UST_ABI_EVENT_NOTIFIER_CREATE:
1247 {
1248 /* Receive struct lttng_ust_event_notifier */
1249 struct lttng_ust_abi_event_notifier event_notifier;
1250
1251 if (sizeof(event_notifier) != lum->u.event_notifier.len) {
1252 DBG("incorrect event notifier data message size: %u", lum->u.event_notifier.len);
1253 ret = -EINVAL;
1254 goto error;
1255 }
1256 len = ustcomm_recv_unix_sock(sock, &event_notifier, sizeof(event_notifier));
1257 switch (len) {
1258 case 0: /* orderly shutdown */
1259 ret = 0;
1260 goto error;
1261 default:
1262 if (len == sizeof(event_notifier)) {
1263 DBG("event notifier data received");
1264 break;
1265 } else if (len < 0) {
1266 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1267 if (len == -ECONNRESET) {
1268 ERR("%s remote end closed connection", sock_info->name);
1269 ret = len;
1270 goto error;
1271 }
1272 ret = len;
1273 goto error;
1274 } else {
1275 DBG("incorrect event notifier data message size: %zd", len);
1276 ret = -EINVAL;
1277 goto error;
1278 }
1279 }
1280 if (ops->cmd)
1281 ret = ops->cmd(lum->handle, lum->cmd,
1282 (unsigned long) &event_notifier,
1283 &args, sock_info);
1284 else
1285 ret = -ENOSYS;
1286 break;
1287 }
1288
1289 default:
1290 if (ops->cmd)
1291 ret = ops->cmd(lum->handle, lum->cmd,
1292 (unsigned long) &lum->u,
1293 &args, sock_info);
1294 else
1295 ret = -ENOSYS;
1296 break;
1297 }
1298
1299 lur.handle = lum->handle;
1300 lur.cmd = lum->cmd;
1301 lur.ret_val = ret;
1302 if (ret >= 0) {
1303 lur.ret_code = LTTNG_UST_OK;
1304 } else {
1305 /*
1306 * Use -LTTNG_UST_ERR as wildcard for UST internal
1307 * error that are not caused by the transport, except if
1308 * we already have a more precise error message to
1309 * report.
1310 */
1311 if (ret > -LTTNG_UST_ERR) {
1312 /* Translate code to UST error. */
1313 switch (ret) {
1314 case -EEXIST:
1315 lur.ret_code = -LTTNG_UST_ERR_EXIST;
1316 break;
1317 case -EINVAL:
1318 lur.ret_code = -LTTNG_UST_ERR_INVAL;
1319 break;
1320 case -ENOENT:
1321 lur.ret_code = -LTTNG_UST_ERR_NOENT;
1322 break;
1323 case -EPERM:
1324 lur.ret_code = -LTTNG_UST_ERR_PERM;
1325 break;
1326 case -ENOSYS:
1327 lur.ret_code = -LTTNG_UST_ERR_NOSYS;
1328 break;
1329 default:
1330 lur.ret_code = -LTTNG_UST_ERR;
1331 break;
1332 }
1333 } else {
1334 lur.ret_code = ret;
1335 }
1336 }
1337 if (ret >= 0) {
1338 switch (lum->cmd) {
1339 case LTTNG_UST_ABI_TRACER_VERSION:
1340 lur.u.version = lum->u.version;
1341 break;
1342 case LTTNG_UST_ABI_TRACEPOINT_LIST_GET:
1343 memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
1344 break;
1345 }
1346 }
1347 DBG("Return value: %d", lur.ret_val);
1348
1349 ust_unlock();
1350
1351 /*
1352 * Performed delayed statedump operations outside of the UST
1353 * lock. We need to take the dynamic loader lock before we take
1354 * the UST lock internally within handle_pending_statedump().
1355 */
1356 handle_pending_statedump(sock_info);
1357
1358 if (ust_lock()) {
1359 ret = -LTTNG_UST_ERR_EXITING;
1360 goto error;
1361 }
1362
1363 ret = send_reply(sock, &lur);
1364 if (ret < 0) {
1365 DBG("error sending reply");
1366 goto error;
1367 }
1368
1369 /*
1370 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1371 * after the reply.
1372 */
1373 if (lur.ret_code == LTTNG_UST_OK) {
1374 switch (lum->cmd) {
1375 case LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST_GET:
1376 len = ustcomm_send_unix_sock(sock,
1377 &args.field_list.entry,
1378 sizeof(args.field_list.entry));
1379 if (len < 0) {
1380 ret = len;
1381 goto error;
1382 }
1383 if (len != sizeof(args.field_list.entry)) {
1384 ret = -EINVAL;
1385 goto error;
1386 }
1387 }
1388 }
1389
1390 error:
1391 ust_unlock();
1392
1393 return ret;
1394 }
1395
1396 static
1397 void cleanup_sock_info(struct sock_info *sock_info, int exiting)
1398 {
1399 int ret;
1400
1401 if (sock_info->root_handle != -1) {
1402 ret = lttng_ust_abi_objd_unref(sock_info->root_handle, 1);
1403 if (ret) {
1404 ERR("Error unref root handle");
1405 }
1406 sock_info->root_handle = -1;
1407 }
1408 sock_info->registration_done = 0;
1409 sock_info->initial_statedump_done = 0;
1410
1411 /*
1412 * wait_shm_mmap, socket and notify socket are used by listener
1413 * threads outside of the ust lock, so we cannot tear them down
1414 * ourselves, because we cannot join on these threads. Leave
1415 * responsibility of cleaning up these resources to the OS
1416 * process exit.
1417 */
1418 if (exiting)
1419 return;
1420
1421 if (sock_info->socket != -1) {
1422 ret = ustcomm_close_unix_sock(sock_info->socket);
1423 if (ret) {
1424 ERR("Error closing ust cmd socket");
1425 }
1426 sock_info->socket = -1;
1427 }
1428 if (sock_info->notify_socket != -1) {
1429 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1430 if (ret) {
1431 ERR("Error closing ust notify socket");
1432 }
1433 sock_info->notify_socket = -1;
1434 }
1435 if (sock_info->wait_shm_mmap) {
1436 long page_size;
1437
1438 page_size = LTTNG_UST_PAGE_SIZE;
1439 if (page_size <= 0) {
1440 if (!page_size) {
1441 errno = EINVAL;
1442 }
1443 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1444 } else {
1445 ret = munmap(sock_info->wait_shm_mmap, page_size);
1446 if (ret) {
1447 ERR("Error unmapping wait shm");
1448 }
1449 }
1450 sock_info->wait_shm_mmap = NULL;
1451 }
1452 }
1453
1454 /*
1455 * Using fork to set umask in the child process (not multi-thread safe).
1456 * We deal with the shm_open vs ftruncate race (happening when the
1457 * sessiond owns the shm and does not let everybody modify it, to ensure
1458 * safety against shm_unlink) by simply letting the mmap fail and
1459 * retrying after a few seconds.
1460 * For global shm, everybody has rw access to it until the sessiond
1461 * starts.
1462 */
1463 static
1464 int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
1465 {
1466 int wait_shm_fd, ret;
1467 pid_t pid;
1468
1469 /*
1470 * Try to open read-only.
1471 */
1472 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1473 if (wait_shm_fd >= 0) {
1474 int32_t tmp_read;
1475 ssize_t len;
1476 size_t bytes_read = 0;
1477
1478 /*
1479 * Try to read the fd. If unable to do so, try opening
1480 * it in write mode.
1481 */
1482 do {
1483 len = read(wait_shm_fd,
1484 &((char *) &tmp_read)[bytes_read],
1485 sizeof(tmp_read) - bytes_read);
1486 if (len > 0) {
1487 bytes_read += len;
1488 }
1489 } while ((len < 0 && errno == EINTR)
1490 || (len > 0 && bytes_read < sizeof(tmp_read)));
1491 if (bytes_read != sizeof(tmp_read)) {
1492 ret = close(wait_shm_fd);
1493 if (ret) {
1494 ERR("close wait_shm_fd");
1495 }
1496 goto open_write;
1497 }
1498 goto end;
1499 } else if (wait_shm_fd < 0 && errno != ENOENT) {
1500 /*
1501 * Real-only open did not work, and it's not because the
1502 * entry was not present. It's a failure that prohibits
1503 * using shm.
1504 */
1505 ERR("Error opening shm %s", sock_info->wait_shm_path);
1506 goto end;
1507 }
1508
1509 open_write:
1510 /*
1511 * If the open failed because the file did not exist, or because
1512 * the file was not truncated yet, try creating it ourself.
1513 */
1514 URCU_TLS(lttng_ust_nest_count)++;
1515 pid = fork();
1516 URCU_TLS(lttng_ust_nest_count)--;
1517 if (pid > 0) {
1518 int status;
1519
1520 /*
1521 * Parent: wait for child to return, in which case the
1522 * shared memory map will have been created.
1523 */
1524 pid = wait(&status);
1525 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
1526 wait_shm_fd = -1;
1527 goto end;
1528 }
1529 /*
1530 * Try to open read-only again after creation.
1531 */
1532 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1533 if (wait_shm_fd < 0) {
1534 /*
1535 * Real-only open did not work. It's a failure
1536 * that prohibits using shm.
1537 */
1538 ERR("Error opening shm %s", sock_info->wait_shm_path);
1539 goto end;
1540 }
1541 goto end;
1542 } else if (pid == 0) {
1543 int create_mode;
1544
1545 /* Child */
1546 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
1547 if (sock_info->global)
1548 create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
1549 /*
1550 * We're alone in a child process, so we can modify the
1551 * process-wide umask.
1552 */
1553 umask(~create_mode);
1554 /*
1555 * Try creating shm (or get rw access).
1556 * We don't do an exclusive open, because we allow other
1557 * processes to create+ftruncate it concurrently.
1558 */
1559 wait_shm_fd = shm_open(sock_info->wait_shm_path,
1560 O_RDWR | O_CREAT, create_mode);
1561 if (wait_shm_fd >= 0) {
1562 ret = ftruncate(wait_shm_fd, mmap_size);
1563 if (ret) {
1564 PERROR("ftruncate");
1565 _exit(EXIT_FAILURE);
1566 }
1567 _exit(EXIT_SUCCESS);
1568 }
1569 /*
1570 * For local shm, we need to have rw access to accept
1571 * opening it: this means the local sessiond will be
1572 * able to wake us up. For global shm, we open it even
1573 * if rw access is not granted, because the root.root
1574 * sessiond will be able to override all rights and wake
1575 * us up.
1576 */
1577 if (!sock_info->global && errno != EACCES) {
1578 ERR("Error opening shm %s", sock_info->wait_shm_path);
1579 _exit(EXIT_FAILURE);
1580 }
1581 /*
1582 * The shm exists, but we cannot open it RW. Report
1583 * success.
1584 */
1585 _exit(EXIT_SUCCESS);
1586 } else {
1587 return -1;
1588 }
1589 end:
1590 if (wait_shm_fd >= 0 && !sock_info->global) {
1591 struct stat statbuf;
1592
1593 /*
1594 * Ensure that our user is the owner of the shm file for
1595 * local shm. If we do not own the file, it means our
1596 * sessiond will not have access to wake us up (there is
1597 * probably a rogue process trying to fake our
1598 * sessiond). Fallback to polling method in this case.
1599 */
1600 ret = fstat(wait_shm_fd, &statbuf);
1601 if (ret) {
1602 PERROR("fstat");
1603 goto error_close;
1604 }
1605 if (statbuf.st_uid != getuid())
1606 goto error_close;
1607 }
1608 return wait_shm_fd;
1609
1610 error_close:
1611 ret = close(wait_shm_fd);
1612 if (ret) {
1613 PERROR("Error closing fd");
1614 }
1615 return -1;
1616 }
1617
1618 static
1619 char *get_map_shm(struct sock_info *sock_info)
1620 {
1621 long page_size;
1622 int wait_shm_fd, ret;
1623 char *wait_shm_mmap;
1624
1625 page_size = sysconf(_SC_PAGE_SIZE);
1626 if (page_size <= 0) {
1627 if (!page_size) {
1628 errno = EINVAL;
1629 }
1630 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1631 goto error;
1632 }
1633
1634 lttng_ust_lock_fd_tracker();
1635 wait_shm_fd = get_wait_shm(sock_info, page_size);
1636 if (wait_shm_fd < 0) {
1637 lttng_ust_unlock_fd_tracker();
1638 goto error;
1639 }
1640
1641 ret = lttng_ust_add_fd_to_tracker(wait_shm_fd);
1642 if (ret < 0) {
1643 ret = close(wait_shm_fd);
1644 if (!ret) {
1645 PERROR("Error closing fd");
1646 }
1647 lttng_ust_unlock_fd_tracker();
1648 goto error;
1649 }
1650
1651 wait_shm_fd = ret;
1652 lttng_ust_unlock_fd_tracker();
1653
1654 wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
1655 MAP_SHARED, wait_shm_fd, 0);
1656
1657 /* close shm fd immediately after taking the mmap reference */
1658 lttng_ust_lock_fd_tracker();
1659 ret = close(wait_shm_fd);
1660 if (!ret) {
1661 lttng_ust_delete_fd_from_tracker(wait_shm_fd);
1662 } else {
1663 PERROR("Error closing fd");
1664 }
1665 lttng_ust_unlock_fd_tracker();
1666
1667 if (wait_shm_mmap == MAP_FAILED) {
1668 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1669 goto error;
1670 }
1671 return wait_shm_mmap;
1672
1673 error:
1674 return NULL;
1675 }
1676
1677 static
1678 void wait_for_sessiond(struct sock_info *sock_info)
1679 {
1680 /* Use ust_lock to check if we should quit. */
1681 if (ust_lock()) {
1682 goto quit;
1683 }
1684 if (wait_poll_fallback) {
1685 goto error;
1686 }
1687 ust_unlock();
1688
1689 assert(sock_info->wait_shm_mmap);
1690
1691 DBG("Waiting for %s apps sessiond", sock_info->name);
1692 /* Wait for futex wakeup */
1693 if (uatomic_read((int32_t *) sock_info->wait_shm_mmap))
1694 goto end_wait;
1695
1696 while (lttng_ust_futex_async((int32_t *) sock_info->wait_shm_mmap,
1697 FUTEX_WAIT, 0, NULL, NULL, 0)) {
1698 switch (errno) {
1699 case EWOULDBLOCK:
1700 /* Value already changed. */
1701 goto end_wait;
1702 case EINTR:
1703 /* Retry if interrupted by signal. */
1704 break; /* Get out of switch. */
1705 case EFAULT:
1706 wait_poll_fallback = 1;
1707 DBG(
1708 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1709 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1710 "Please upgrade your kernel "
1711 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1712 "mainline). LTTng-UST will use polling mode fallback.");
1713 if (ust_err_debug_enabled())
1714 PERROR("futex");
1715 goto end_wait;
1716 }
1717 }
1718 end_wait:
1719 return;
1720
1721 quit:
1722 ust_unlock();
1723 return;
1724
1725 error:
1726 ust_unlock();
1727 return;
1728 }
1729
1730 /*
1731 * This thread does not allocate any resource, except within
1732 * handle_message, within mutex protection. This mutex protects against
1733 * fork and exit.
1734 * The other moment it allocates resources is at socket connection, which
1735 * is also protected by the mutex.
1736 */
1737 static
1738 void *ust_listener_thread(void *arg)
1739 {
1740 struct sock_info *sock_info = arg;
1741 int sock, ret, prev_connect_failed = 0, has_waited = 0, fd;
1742 long timeout;
1743
1744 lttng_ust_fixup_tls();
1745 /*
1746 * If available, add '-ust' to the end of this thread's
1747 * process name
1748 */
1749 ret = lttng_ust_setustprocname();
1750 if (ret) {
1751 ERR("Unable to set UST process name");
1752 }
1753
1754 /* Restart trying to connect to the session daemon */
1755 restart:
1756 if (prev_connect_failed) {
1757 /* Wait for sessiond availability with pipe */
1758 wait_for_sessiond(sock_info);
1759 if (has_waited) {
1760 has_waited = 0;
1761 /*
1762 * Sleep for 5 seconds before retrying after a
1763 * sequence of failure / wait / failure. This
1764 * deals with a killed or broken session daemon.
1765 */
1766 sleep(5);
1767 } else {
1768 has_waited = 1;
1769 }
1770 prev_connect_failed = 0;
1771 }
1772
1773 if (ust_lock()) {
1774 goto quit;
1775 }
1776
1777 if (sock_info->socket != -1) {
1778 /* FD tracker is updated by ustcomm_close_unix_sock() */
1779 ret = ustcomm_close_unix_sock(sock_info->socket);
1780 if (ret) {
1781 ERR("Error closing %s ust cmd socket",
1782 sock_info->name);
1783 }
1784 sock_info->socket = -1;
1785 }
1786 if (sock_info->notify_socket != -1) {
1787 /* FD tracker is updated by ustcomm_close_unix_sock() */
1788 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1789 if (ret) {
1790 ERR("Error closing %s ust notify socket",
1791 sock_info->name);
1792 }
1793 sock_info->notify_socket = -1;
1794 }
1795
1796
1797 /*
1798 * Register. We need to perform both connect and sending
1799 * registration message before doing the next connect otherwise
1800 * we may reach unix socket connect queue max limits and block
1801 * on the 2nd connect while the session daemon is awaiting the
1802 * first connect registration message.
1803 */
1804 /* Connect cmd socket */
1805 lttng_ust_lock_fd_tracker();
1806 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1807 get_connect_sock_timeout());
1808 if (ret < 0) {
1809 lttng_ust_unlock_fd_tracker();
1810 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1811 prev_connect_failed = 1;
1812
1813 /*
1814 * If we cannot find the sessiond daemon, don't delay
1815 * constructor execution.
1816 */
1817 ret = handle_register_failed(sock_info);
1818 assert(!ret);
1819 ust_unlock();
1820 goto restart;
1821 }
1822 fd = ret;
1823 ret = lttng_ust_add_fd_to_tracker(fd);
1824 if (ret < 0) {
1825 ret = close(fd);
1826 if (ret) {
1827 PERROR("close on sock_info->socket");
1828 }
1829 ret = -1;
1830 lttng_ust_unlock_fd_tracker();
1831 ust_unlock();
1832 goto quit;
1833 }
1834
1835 sock_info->socket = ret;
1836 lttng_ust_unlock_fd_tracker();
1837
1838 ust_unlock();
1839 /*
1840 * Unlock/relock ust lock because connect is blocking (with
1841 * timeout). Don't delay constructors on the ust lock for too
1842 * long.
1843 */
1844 if (ust_lock()) {
1845 goto quit;
1846 }
1847
1848 /*
1849 * Create only one root handle per listener thread for the whole
1850 * process lifetime, so we ensure we get ID which is statically
1851 * assigned to the root handle.
1852 */
1853 if (sock_info->root_handle == -1) {
1854 ret = lttng_abi_create_root_handle();
1855 if (ret < 0) {
1856 ERR("Error creating root handle");
1857 goto quit;
1858 }
1859 sock_info->root_handle = ret;
1860 }
1861
1862 ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD);
1863 if (ret < 0) {
1864 ERR("Error registering to %s ust cmd socket",
1865 sock_info->name);
1866 prev_connect_failed = 1;
1867 /*
1868 * If we cannot register to the sessiond daemon, don't
1869 * delay constructor execution.
1870 */
1871 ret = handle_register_failed(sock_info);
1872 assert(!ret);
1873 ust_unlock();
1874 goto restart;
1875 }
1876
1877 ust_unlock();
1878 /*
1879 * Unlock/relock ust lock because connect is blocking (with
1880 * timeout). Don't delay constructors on the ust lock for too
1881 * long.
1882 */
1883 if (ust_lock()) {
1884 goto quit;
1885 }
1886
1887 /* Connect notify socket */
1888 lttng_ust_lock_fd_tracker();
1889 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1890 get_connect_sock_timeout());
1891 if (ret < 0) {
1892 lttng_ust_unlock_fd_tracker();
1893 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1894 prev_connect_failed = 1;
1895
1896 /*
1897 * If we cannot find the sessiond daemon, don't delay
1898 * constructor execution.
1899 */
1900 ret = handle_register_failed(sock_info);
1901 assert(!ret);
1902 ust_unlock();
1903 goto restart;
1904 }
1905
1906 fd = ret;
1907 ret = lttng_ust_add_fd_to_tracker(fd);
1908 if (ret < 0) {
1909 ret = close(fd);
1910 if (ret) {
1911 PERROR("close on sock_info->notify_socket");
1912 }
1913 ret = -1;
1914 lttng_ust_unlock_fd_tracker();
1915 ust_unlock();
1916 goto quit;
1917 }
1918
1919 sock_info->notify_socket = ret;
1920 lttng_ust_unlock_fd_tracker();
1921
1922 ust_unlock();
1923 /*
1924 * Unlock/relock ust lock because connect is blocking (with
1925 * timeout). Don't delay constructors on the ust lock for too
1926 * long.
1927 */
1928 if (ust_lock()) {
1929 goto quit;
1930 }
1931
1932 timeout = get_notify_sock_timeout();
1933 if (timeout >= 0) {
1934 /*
1935 * Give at least 10ms to sessiond to reply to
1936 * notifications.
1937 */
1938 if (timeout < 10)
1939 timeout = 10;
1940 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
1941 timeout);
1942 if (ret < 0) {
1943 WARN("Error setting socket receive timeout");
1944 }
1945 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
1946 timeout);
1947 if (ret < 0) {
1948 WARN("Error setting socket send timeout");
1949 }
1950 } else if (timeout < -1) {
1951 WARN("Unsupported timeout value %ld", timeout);
1952 }
1953
1954 ret = register_to_sessiond(sock_info->notify_socket,
1955 USTCTL_SOCKET_NOTIFY);
1956 if (ret < 0) {
1957 ERR("Error registering to %s ust notify socket",
1958 sock_info->name);
1959 prev_connect_failed = 1;
1960 /*
1961 * If we cannot register to the sessiond daemon, don't
1962 * delay constructor execution.
1963 */
1964 ret = handle_register_failed(sock_info);
1965 assert(!ret);
1966 ust_unlock();
1967 goto restart;
1968 }
1969 sock = sock_info->socket;
1970
1971 ust_unlock();
1972
1973 for (;;) {
1974 ssize_t len;
1975 struct ustcomm_ust_msg lum;
1976
1977 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
1978 switch (len) {
1979 case 0: /* orderly shutdown */
1980 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
1981 if (ust_lock()) {
1982 goto quit;
1983 }
1984 /*
1985 * Either sessiond has shutdown or refused us by closing the socket.
1986 * In either case, we don't want to delay construction execution,
1987 * and we need to wait before retry.
1988 */
1989 prev_connect_failed = 1;
1990 /*
1991 * If we cannot register to the sessiond daemon, don't
1992 * delay constructor execution.
1993 */
1994 ret = handle_register_failed(sock_info);
1995 assert(!ret);
1996 ust_unlock();
1997 goto end;
1998 case sizeof(lum):
1999 print_cmd(lum.cmd, lum.handle);
2000 ret = handle_message(sock_info, sock, &lum);
2001 if (ret) {
2002 ERR("Error handling message for %s socket",
2003 sock_info->name);
2004 /*
2005 * Close socket if protocol error is
2006 * detected.
2007 */
2008 goto end;
2009 }
2010 continue;
2011 default:
2012 if (len < 0) {
2013 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
2014 } else {
2015 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
2016 }
2017 if (len == -ECONNRESET) {
2018 DBG("%s remote end closed connection", sock_info->name);
2019 goto end;
2020 }
2021 goto end;
2022 }
2023
2024 }
2025 end:
2026 if (ust_lock()) {
2027 goto quit;
2028 }
2029 /* Cleanup socket handles before trying to reconnect */
2030 lttng_ust_abi_objd_table_owner_cleanup(sock_info);
2031 ust_unlock();
2032 goto restart; /* try to reconnect */
2033
2034 quit:
2035 ust_unlock();
2036
2037 pthread_mutex_lock(&ust_exit_mutex);
2038 sock_info->thread_active = 0;
2039 pthread_mutex_unlock(&ust_exit_mutex);
2040 return NULL;
2041 }
2042
2043 /*
2044 * Weak symbol to call when the ust malloc wrapper is not loaded.
2045 */
2046 __attribute__((weak))
2047 void lttng_ust_libc_wrapper_malloc_init(void)
2048 {
2049 }
2050
2051 /*
2052 * sessiond monitoring thread: monitor presence of global and per-user
2053 * sessiond by polling the application common named pipe.
2054 */
2055 void __attribute__((constructor)) lttng_ust_init(void)
2056 {
2057 struct timespec constructor_timeout;
2058 sigset_t sig_all_blocked, orig_parent_mask;
2059 pthread_attr_t thread_attr;
2060 int timeout_mode;
2061 int ret;
2062 void *handle;
2063
2064 if (uatomic_xchg(&initialized, 1) == 1)
2065 return;
2066
2067 /*
2068 * Fixup interdependency between TLS fixup mutex (which happens
2069 * to be the dynamic linker mutex) and ust_lock, taken within
2070 * the ust lock.
2071 */
2072 lttng_ust_fixup_tls();
2073
2074 lttng_ust_loaded = 1;
2075
2076 /*
2077 * We need to ensure that the liblttng-ust library is not unloaded to avoid
2078 * the unloading of code used by the ust_listener_threads as we can not
2079 * reliably know when they exited. To do that, manually load
2080 * liblttng-ust.so to increment the dynamic loader's internal refcount for
2081 * this library so it never becomes zero, thus never gets unloaded from the
2082 * address space of the process. Since we are already running in the
2083 * constructor of the LTTNG_UST_LIB_SONAME library, calling dlopen will
2084 * simply increment the refcount and no additionnal work is needed by the
2085 * dynamic loader as the shared library is already loaded in the address
2086 * space. As a safe guard, we use the RTLD_NODELETE flag to prevent
2087 * unloading of the UST library if its refcount becomes zero (which should
2088 * never happen). Do the return value check but discard the handle at the
2089 * end of the function as it's not needed.
2090 */
2091 handle = dlopen(LTTNG_UST_LIB_SONAME, RTLD_LAZY | RTLD_NODELETE);
2092 if (!handle) {
2093 ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SONAME);
2094 }
2095
2096 /*
2097 * We want precise control over the order in which we construct
2098 * our sub-libraries vs starting to receive commands from
2099 * sessiond (otherwise leading to errors when trying to create
2100 * sessiond before the init functions are completed).
2101 */
2102 ust_err_init();
2103 lttng_ust_getenv_init(); /* Needs ust_err_init() to be completed. */
2104 lttng_ust_tp_init();
2105 lttng_ust_init_fd_tracker();
2106 lttng_ust_clock_init();
2107 lttng_ust_getcpu_init();
2108 lttng_ust_statedump_init();
2109 lttng_ring_buffer_metadata_client_init();
2110 lttng_ring_buffer_client_overwrite_init();
2111 lttng_ring_buffer_client_overwrite_rt_init();
2112 lttng_ring_buffer_client_discard_init();
2113 lttng_ring_buffer_client_discard_rt_init();
2114 lttng_counter_client_percpu_32_modular_init();
2115 lttng_counter_client_percpu_64_modular_init();
2116 lttng_perf_counter_init();
2117 /*
2118 * Invoke ust malloc wrapper init before starting other threads.
2119 */
2120 lttng_ust_libc_wrapper_malloc_init();
2121
2122 timeout_mode = get_constructor_timeout(&constructor_timeout);
2123
2124 get_allow_blocking();
2125
2126 ret = sem_init(&constructor_wait, 0, 0);
2127 if (ret) {
2128 PERROR("sem_init");
2129 }
2130
2131 ret = setup_global_apps();
2132 if (ret) {
2133 assert(global_apps.allowed == 0);
2134 DBG("global apps setup returned %d", ret);
2135 }
2136
2137 ret = setup_local_apps();
2138 if (ret) {
2139 assert(local_apps.allowed == 0);
2140 DBG("local apps setup returned %d", ret);
2141 }
2142
2143 /* A new thread created by pthread_create inherits the signal mask
2144 * from the parent. To avoid any signal being received by the
2145 * listener thread, we block all signals temporarily in the parent,
2146 * while we create the listener thread.
2147 */
2148 sigfillset(&sig_all_blocked);
2149 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
2150 if (ret) {
2151 ERR("pthread_sigmask: %s", strerror(ret));
2152 }
2153
2154 ret = pthread_attr_init(&thread_attr);
2155 if (ret) {
2156 ERR("pthread_attr_init: %s", strerror(ret));
2157 }
2158 ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
2159 if (ret) {
2160 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
2161 }
2162
2163 if (global_apps.allowed) {
2164 pthread_mutex_lock(&ust_exit_mutex);
2165 ret = pthread_create(&global_apps.ust_listener, &thread_attr,
2166 ust_listener_thread, &global_apps);
2167 if (ret) {
2168 ERR("pthread_create global: %s", strerror(ret));
2169 }
2170 global_apps.thread_active = 1;
2171 pthread_mutex_unlock(&ust_exit_mutex);
2172 } else {
2173 handle_register_done(&global_apps);
2174 }
2175
2176 if (local_apps.allowed) {
2177 pthread_mutex_lock(&ust_exit_mutex);
2178 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
2179 ust_listener_thread, &local_apps);
2180 if (ret) {
2181 ERR("pthread_create local: %s", strerror(ret));
2182 }
2183 local_apps.thread_active = 1;
2184 pthread_mutex_unlock(&ust_exit_mutex);
2185 } else {
2186 handle_register_done(&local_apps);
2187 }
2188 ret = pthread_attr_destroy(&thread_attr);
2189 if (ret) {
2190 ERR("pthread_attr_destroy: %s", strerror(ret));
2191 }
2192
2193 /* Restore original signal mask in parent */
2194 ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
2195 if (ret) {
2196 ERR("pthread_sigmask: %s", strerror(ret));
2197 }
2198
2199 switch (timeout_mode) {
2200 case 1: /* timeout wait */
2201 do {
2202 ret = sem_timedwait(&constructor_wait,
2203 &constructor_timeout);
2204 } while (ret < 0 && errno == EINTR);
2205 if (ret < 0) {
2206 switch (errno) {
2207 case ETIMEDOUT:
2208 ERR("Timed out waiting for lttng-sessiond");
2209 break;
2210 case EINVAL:
2211 PERROR("sem_timedwait");
2212 break;
2213 default:
2214 ERR("Unexpected error \"%s\" returned by sem_timedwait",
2215 strerror(errno));
2216 }
2217 }
2218 break;
2219 case -1:/* wait forever */
2220 do {
2221 ret = sem_wait(&constructor_wait);
2222 } while (ret < 0 && errno == EINTR);
2223 if (ret < 0) {
2224 switch (errno) {
2225 case EINVAL:
2226 PERROR("sem_wait");
2227 break;
2228 default:
2229 ERR("Unexpected error \"%s\" returned by sem_wait",
2230 strerror(errno));
2231 }
2232 }
2233 break;
2234 case 0: /* no timeout */
2235 break;
2236 }
2237 }
2238
2239 static
2240 void lttng_ust_cleanup(int exiting)
2241 {
2242 cleanup_sock_info(&global_apps, exiting);
2243 cleanup_sock_info(&local_apps, exiting);
2244 local_apps.allowed = 0;
2245 global_apps.allowed = 0;
2246 /*
2247 * The teardown in this function all affect data structures
2248 * accessed under the UST lock by the listener thread. This
2249 * lock, along with the lttng_ust_comm_should_quit flag, ensure
2250 * that none of these threads are accessing this data at this
2251 * point.
2252 */
2253 lttng_ust_abi_exit();
2254 lttng_ust_abi_events_exit();
2255 lttng_perf_counter_exit();
2256 lttng_ring_buffer_client_discard_rt_exit();
2257 lttng_ring_buffer_client_discard_exit();
2258 lttng_ring_buffer_client_overwrite_rt_exit();
2259 lttng_ring_buffer_client_overwrite_exit();
2260 lttng_ring_buffer_metadata_client_exit();
2261 lttng_counter_client_percpu_32_modular_exit();
2262 lttng_counter_client_percpu_64_modular_exit();
2263 lttng_ust_statedump_destroy();
2264 lttng_ust_tp_exit();
2265 if (!exiting) {
2266 /* Reinitialize values for fork */
2267 sem_count = sem_count_initial_value;
2268 lttng_ust_comm_should_quit = 0;
2269 initialized = 0;
2270 }
2271 }
2272
2273 void __attribute__((destructor)) lttng_ust_exit(void)
2274 {
2275 int ret;
2276
2277 /*
2278 * Using pthread_cancel here because:
2279 * A) we don't want to hang application teardown.
2280 * B) the thread is not allocating any resource.
2281 */
2282
2283 /*
2284 * Require the communication thread to quit. Synchronize with
2285 * mutexes to ensure it is not in a mutex critical section when
2286 * pthread_cancel is later called.
2287 */
2288 ust_lock_nocheck();
2289 lttng_ust_comm_should_quit = 1;
2290 ust_unlock();
2291
2292 pthread_mutex_lock(&ust_exit_mutex);
2293 /* cancel threads */
2294 if (global_apps.thread_active) {
2295 ret = pthread_cancel(global_apps.ust_listener);
2296 if (ret) {
2297 ERR("Error cancelling global ust listener thread: %s",
2298 strerror(ret));
2299 } else {
2300 global_apps.thread_active = 0;
2301 }
2302 }
2303 if (local_apps.thread_active) {
2304 ret = pthread_cancel(local_apps.ust_listener);
2305 if (ret) {
2306 ERR("Error cancelling local ust listener thread: %s",
2307 strerror(ret));
2308 } else {
2309 local_apps.thread_active = 0;
2310 }
2311 }
2312 pthread_mutex_unlock(&ust_exit_mutex);
2313
2314 /*
2315 * Do NOT join threads: use of sys_futex makes it impossible to
2316 * join the threads without using async-cancel, but async-cancel
2317 * is delivered by a signal, which could hit the target thread
2318 * anywhere in its code path, including while the ust_lock() is
2319 * held, causing a deadlock for the other thread. Let the OS
2320 * cleanup the threads if there are stalled in a syscall.
2321 */
2322 lttng_ust_cleanup(1);
2323 }
2324
2325 static
2326 void ust_context_ns_reset(void)
2327 {
2328 lttng_context_pid_ns_reset();
2329 lttng_context_cgroup_ns_reset();
2330 lttng_context_ipc_ns_reset();
2331 lttng_context_mnt_ns_reset();
2332 lttng_context_net_ns_reset();
2333 lttng_context_user_ns_reset();
2334 lttng_context_time_ns_reset();
2335 lttng_context_uts_ns_reset();
2336 }
2337
2338 static
2339 void ust_context_vuids_reset(void)
2340 {
2341 lttng_context_vuid_reset();
2342 lttng_context_veuid_reset();
2343 lttng_context_vsuid_reset();
2344 }
2345
2346 static
2347 void ust_context_vgids_reset(void)
2348 {
2349 lttng_context_vgid_reset();
2350 lttng_context_vegid_reset();
2351 lttng_context_vsgid_reset();
2352 }
2353
2354 /*
2355 * We exclude the worker threads across fork and clone (except
2356 * CLONE_VM), because these system calls only keep the forking thread
2357 * running in the child. Therefore, we don't want to call fork or clone
2358 * in the middle of an tracepoint or ust tracing state modification.
2359 * Holding this mutex protects these structures across fork and clone.
2360 */
2361 void lttng_ust_before_fork(sigset_t *save_sigset)
2362 {
2363 /*
2364 * Disable signals. This is to avoid that the child intervenes
2365 * before it is properly setup for tracing. It is safer to
2366 * disable all signals, because then we know we are not breaking
2367 * anything by restoring the original mask.
2368 */
2369 sigset_t all_sigs;
2370 int ret;
2371
2372 /* Fixup lttng-ust TLS. */
2373 lttng_ust_fixup_tls();
2374
2375 if (URCU_TLS(lttng_ust_nest_count))
2376 return;
2377 /* Disable signals */
2378 sigfillset(&all_sigs);
2379 ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
2380 if (ret == -1) {
2381 PERROR("sigprocmask");
2382 }
2383
2384 pthread_mutex_lock(&ust_fork_mutex);
2385
2386 ust_lock_nocheck();
2387 lttng_ust_urcu_before_fork();
2388 lttng_ust_lock_fd_tracker();
2389 lttng_perf_lock();
2390 }
2391
2392 static void ust_after_fork_common(sigset_t *restore_sigset)
2393 {
2394 int ret;
2395
2396 DBG("process %d", getpid());
2397 lttng_perf_unlock();
2398 lttng_ust_unlock_fd_tracker();
2399 ust_unlock();
2400
2401 pthread_mutex_unlock(&ust_fork_mutex);
2402
2403 /* Restore signals */
2404 ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
2405 if (ret == -1) {
2406 PERROR("sigprocmask");
2407 }
2408 }
2409
2410 void lttng_ust_after_fork_parent(sigset_t *restore_sigset)
2411 {
2412 if (URCU_TLS(lttng_ust_nest_count))
2413 return;
2414 DBG("process %d", getpid());
2415 lttng_ust_urcu_after_fork_parent();
2416 /* Release mutexes and reenable signals */
2417 ust_after_fork_common(restore_sigset);
2418 }
2419
2420 /*
2421 * After fork, in the child, we need to cleanup all the leftover state,
2422 * except the worker thread which already magically disappeared thanks
2423 * to the weird Linux fork semantics. After tyding up, we call
2424 * lttng_ust_init() again to start over as a new PID.
2425 *
2426 * This is meant for forks() that have tracing in the child between the
2427 * fork and following exec call (if there is any).
2428 */
2429 void lttng_ust_after_fork_child(sigset_t *restore_sigset)
2430 {
2431 if (URCU_TLS(lttng_ust_nest_count))
2432 return;
2433 lttng_context_vpid_reset();
2434 lttng_context_vtid_reset();
2435 lttng_ust_context_procname_reset();
2436 ust_context_ns_reset();
2437 ust_context_vuids_reset();
2438 ust_context_vgids_reset();
2439 DBG("process %d", getpid());
2440 /* Release urcu mutexes */
2441 lttng_ust_urcu_after_fork_child();
2442 lttng_ust_cleanup(0);
2443 /* Release mutexes and reenable signals */
2444 ust_after_fork_common(restore_sigset);
2445 lttng_ust_init();
2446 }
2447
2448 void lttng_ust_after_setns(void)
2449 {
2450 ust_context_ns_reset();
2451 ust_context_vuids_reset();
2452 ust_context_vgids_reset();
2453 }
2454
2455 void lttng_ust_after_unshare(void)
2456 {
2457 ust_context_ns_reset();
2458 ust_context_vuids_reset();
2459 ust_context_vgids_reset();
2460 }
2461
2462 void lttng_ust_after_setuid(void)
2463 {
2464 ust_context_vuids_reset();
2465 }
2466
2467 void lttng_ust_after_seteuid(void)
2468 {
2469 ust_context_vuids_reset();
2470 }
2471
2472 void lttng_ust_after_setreuid(void)
2473 {
2474 ust_context_vuids_reset();
2475 }
2476
2477 void lttng_ust_after_setresuid(void)
2478 {
2479 ust_context_vuids_reset();
2480 }
2481
2482 void lttng_ust_after_setgid(void)
2483 {
2484 ust_context_vgids_reset();
2485 }
2486
2487 void lttng_ust_after_setegid(void)
2488 {
2489 ust_context_vgids_reset();
2490 }
2491
2492 void lttng_ust_after_setregid(void)
2493 {
2494 ust_context_vgids_reset();
2495 }
2496
2497 void lttng_ust_after_setresgid(void)
2498 {
2499 ust_context_vgids_reset();
2500 }
2501
2502 void lttng_ust_sockinfo_session_enabled(void *owner)
2503 {
2504 struct sock_info *sock_info = owner;
2505 sock_info->statedump_pending = 1;
2506 }
This page took 0.076892 seconds and 5 git commands to generate.