f180c3bc8277874ddd673954b1e3bf8467b7e95a
[lttng-ust.git] / src / lib / lttng-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 <urcu/compiler.h>
29 #include <lttng/urcu/urcu-ust.h>
30
31 #include <lttng/ust-utils.h>
32 #include <lttng/ust-events.h>
33 #include <lttng/ust-abi.h>
34 #include <lttng/ust-fork.h>
35 #include <lttng/ust-error.h>
36 #include <lttng/ust-ctl.h>
37 #include <lttng/ust-libc-wrapper.h>
38 #include <lttng/ust-thread.h>
39 #include <lttng/ust-tracer.h>
40 #include <lttng/ust-common.h>
41 #include <urcu/tls-compat.h>
42 #include "lib/lttng-ust/futex.h"
43 #include "common/ustcomm.h"
44 #include "common/ust-fd.h"
45 #include "common/logging.h"
46 #include "common/macros.h"
47 #include "common/tracepoint.h"
48 #include "lttng-tracer-core.h"
49 #include "common/compat/pthread.h"
50 #include "common/procname.h"
51 #include "common/ringbuffer/rb-init.h"
52 #include "lttng-ust-statedump.h"
53 #include "common/clock.h"
54 #include "common/getenv.h"
55 #include "lib/lttng-ust/events.h"
56 #include "context-internal.h"
57 #include "common/align.h"
58 #include "common/counter-clients/clients.h"
59 #include "common/ringbuffer-clients/clients.h"
60
61 /*
62 * Has lttng ust comm constructor been called ?
63 */
64 static int initialized;
65
66 /*
67 * The ust_lock/ust_unlock lock is used as a communication thread mutex.
68 * Held when handling a command, also held by fork() to deal with
69 * removal of threads, and by exit path.
70 *
71 * The UST lock is the centralized mutex across UST tracing control and
72 * probe registration.
73 *
74 * ust_exit_mutex must never nest in ust_mutex.
75 *
76 * ust_fork_mutex must never nest in ust_mutex.
77 *
78 * ust_mutex_nest is a per-thread nesting counter, allowing the perf
79 * counter lazy initialization called by events within the statedump,
80 * which traces while the ust_mutex is held.
81 *
82 * ust_lock nests within the dynamic loader lock (within glibc) because
83 * it is taken within the library constructor.
84 *
85 * The ust fd tracker lock nests within the ust_mutex.
86 */
87 static pthread_mutex_t ust_mutex = PTHREAD_MUTEX_INITIALIZER;
88
89 /* Allow nesting the ust_mutex within the same thread. */
90 static DEFINE_URCU_TLS(int, ust_mutex_nest);
91
92 /*
93 * ust_exit_mutex protects thread_active variable wrt thread exit. It
94 * cannot be done by ust_mutex because pthread_cancel(), which takes an
95 * internal libc lock, cannot nest within ust_mutex.
96 *
97 * It never nests within a ust_mutex.
98 */
99 static pthread_mutex_t ust_exit_mutex = PTHREAD_MUTEX_INITIALIZER;
100
101 /*
102 * ust_fork_mutex protects base address statedump tracing against forks. It
103 * prevents the dynamic loader lock to be taken (by base address statedump
104 * tracing) while a fork is happening, thus preventing deadlock issues with
105 * the dynamic loader lock.
106 */
107 static pthread_mutex_t ust_fork_mutex = PTHREAD_MUTEX_INITIALIZER;
108
109 /* Should the ust comm thread quit ? */
110 static int lttng_ust_comm_should_quit;
111
112 /*
113 * This variable can be tested by applications to check whether
114 * lttng-ust is loaded. They simply have to define their own
115 * "lttng_ust_loaded" weak symbol, and test it. It is set to 1 by the
116 * library constructor.
117 */
118 int lttng_ust_loaded __attribute__((weak));
119
120 /*
121 * Return 0 on success, -1 if should quit.
122 * The lock is taken in both cases.
123 * Signal-safe.
124 */
125 int ust_lock(void)
126 {
127 sigset_t sig_all_blocked, orig_mask;
128 int ret, oldstate;
129
130 ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
131 if (ret) {
132 ERR("pthread_setcancelstate: %s", strerror(ret));
133 }
134 if (oldstate != PTHREAD_CANCEL_ENABLE) {
135 ERR("pthread_setcancelstate: unexpected oldstate");
136 }
137 sigfillset(&sig_all_blocked);
138 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
139 if (ret) {
140 ERR("pthread_sigmask: %s", strerror(ret));
141 }
142 if (!URCU_TLS(ust_mutex_nest)++)
143 pthread_mutex_lock(&ust_mutex);
144 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
145 if (ret) {
146 ERR("pthread_sigmask: %s", strerror(ret));
147 }
148 if (lttng_ust_comm_should_quit) {
149 return -1;
150 } else {
151 return 0;
152 }
153 }
154
155 /*
156 * ust_lock_nocheck() can be used in constructors/destructors, because
157 * they are already nested within the dynamic loader lock, and therefore
158 * have exclusive access against execution of liblttng-ust destructor.
159 * Signal-safe.
160 */
161 void ust_lock_nocheck(void)
162 {
163 sigset_t sig_all_blocked, orig_mask;
164 int ret, oldstate;
165
166 ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
167 if (ret) {
168 ERR("pthread_setcancelstate: %s", strerror(ret));
169 }
170 if (oldstate != PTHREAD_CANCEL_ENABLE) {
171 ERR("pthread_setcancelstate: unexpected oldstate");
172 }
173 sigfillset(&sig_all_blocked);
174 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
175 if (ret) {
176 ERR("pthread_sigmask: %s", strerror(ret));
177 }
178 if (!URCU_TLS(ust_mutex_nest)++)
179 pthread_mutex_lock(&ust_mutex);
180 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
181 if (ret) {
182 ERR("pthread_sigmask: %s", strerror(ret));
183 }
184 }
185
186 /*
187 * Signal-safe.
188 */
189 void ust_unlock(void)
190 {
191 sigset_t sig_all_blocked, orig_mask;
192 int ret, oldstate;
193
194 sigfillset(&sig_all_blocked);
195 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
196 if (ret) {
197 ERR("pthread_sigmask: %s", strerror(ret));
198 }
199 if (!--URCU_TLS(ust_mutex_nest))
200 pthread_mutex_unlock(&ust_mutex);
201 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
202 if (ret) {
203 ERR("pthread_sigmask: %s", strerror(ret));
204 }
205 ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &oldstate);
206 if (ret) {
207 ERR("pthread_setcancelstate: %s", strerror(ret));
208 }
209 if (oldstate != PTHREAD_CANCEL_DISABLE) {
210 ERR("pthread_setcancelstate: unexpected oldstate");
211 }
212 }
213
214 /*
215 * Wait for either of these before continuing to the main
216 * program:
217 * - the register_done message from sessiond daemon
218 * (will let the sessiond daemon enable sessions before main
219 * starts.)
220 * - sessiond daemon is not reachable.
221 * - timeout (ensuring applications are resilient to session
222 * daemon problems).
223 */
224 static sem_t constructor_wait;
225 /*
226 * Doing this for both the global and local sessiond.
227 */
228 enum {
229 sem_count_initial_value = 4,
230 };
231
232 static int sem_count = sem_count_initial_value;
233
234 /*
235 * Counting nesting within lttng-ust. Used to ensure that calling fork()
236 * from liblttng-ust does not execute the pre/post fork handlers.
237 */
238 static DEFINE_URCU_TLS(int, lttng_ust_nest_count);
239
240 /*
241 * Info about socket and associated listener thread.
242 */
243 struct sock_info {
244 const char *name;
245 pthread_t ust_listener; /* listener thread */
246 int root_handle;
247 int registration_done;
248 int allowed;
249 int global;
250 int thread_active;
251
252 char sock_path[PATH_MAX];
253 int socket;
254 int notify_socket;
255
256 char wait_shm_path[PATH_MAX];
257 char *wait_shm_mmap;
258 /* Keep track of lazy state dump not performed yet. */
259 int statedump_pending;
260 int initial_statedump_done;
261 /* Keep procname for statedump */
262 char procname[LTTNG_UST_ABI_PROCNAME_LEN];
263 };
264
265 /* Socket from app (connect) to session daemon (listen) for communication */
266 static struct sock_info global_apps = {
267 .name = "global",
268 .global = 1,
269
270 .root_handle = -1,
271 .registration_done = 0,
272 .allowed = 0,
273 .thread_active = 0,
274
275 .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME,
276 .socket = -1,
277 .notify_socket = -1,
278
279 .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME,
280
281 .statedump_pending = 0,
282 .initial_statedump_done = 0,
283 .procname[0] = '\0'
284 };
285
286 /* TODO: allow global_apps_sock_path override */
287
288 static struct sock_info local_apps = {
289 .name = "local",
290 .global = 0,
291 .root_handle = -1,
292 .registration_done = 0,
293 .allowed = 0, /* Check setuid bit first */
294 .thread_active = 0,
295
296 .socket = -1,
297 .notify_socket = -1,
298
299 .statedump_pending = 0,
300 .initial_statedump_done = 0,
301 .procname[0] = '\0'
302 };
303
304 static int wait_poll_fallback;
305
306 static const char *cmd_name_mapping[] = {
307 [ LTTNG_UST_ABI_RELEASE ] = "Release",
308 [ LTTNG_UST_ABI_SESSION ] = "Create Session",
309 [ LTTNG_UST_ABI_TRACER_VERSION ] = "Get Tracer Version",
310
311 [ LTTNG_UST_ABI_TRACEPOINT_LIST ] = "Create Tracepoint List",
312 [ LTTNG_UST_ABI_WAIT_QUIESCENT ] = "Wait for Quiescent State",
313 [ LTTNG_UST_ABI_REGISTER_DONE ] = "Registration Done",
314 [ LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST ] = "Create Tracepoint Field List",
315
316 [ LTTNG_UST_ABI_EVENT_NOTIFIER_GROUP_CREATE ] = "Create event notifier group",
317
318 /* Session FD commands */
319 [ LTTNG_UST_ABI_CHANNEL ] = "Create Channel",
320 [ LTTNG_UST_ABI_SESSION_START ] = "Start Session",
321 [ LTTNG_UST_ABI_SESSION_STOP ] = "Stop Session",
322
323 /* Channel FD commands */
324 [ LTTNG_UST_ABI_STREAM ] = "Create Stream",
325 [ LTTNG_UST_ABI_EVENT ] = "Create Event",
326
327 /* Event and Channel FD commands */
328 [ LTTNG_UST_ABI_CONTEXT ] = "Create Context",
329 [ LTTNG_UST_ABI_FLUSH_BUFFER ] = "Flush Buffer",
330
331 /* Event, Channel and Session commands */
332 [ LTTNG_UST_ABI_ENABLE ] = "Enable",
333 [ LTTNG_UST_ABI_DISABLE ] = "Disable",
334
335 /* Tracepoint list commands */
336 [ LTTNG_UST_ABI_TRACEPOINT_LIST_GET ] = "List Next Tracepoint",
337 [ LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST_GET ] = "List Next Tracepoint Field",
338
339 /* Event FD commands */
340 [ LTTNG_UST_ABI_FILTER ] = "Create Filter",
341 [ LTTNG_UST_ABI_EXCLUSION ] = "Add exclusions to event",
342
343 /* Event notifier group commands */
344 [ LTTNG_UST_ABI_EVENT_NOTIFIER_CREATE ] = "Create event notifier",
345
346 /* Session and event notifier group commands */
347 [ LTTNG_UST_ABI_COUNTER ] = "Create Counter",
348
349 /* Counter commands */
350 [ LTTNG_UST_ABI_COUNTER_GLOBAL ] = "Create Counter Global",
351 [ LTTNG_UST_ABI_COUNTER_CPU ] = "Create Counter CPU",
352 };
353
354 static const char *str_timeout;
355 static int got_timeout_env;
356
357 static char *get_map_shm(struct sock_info *sock_info);
358
359 /*
360 * Returns the HOME directory path. Caller MUST NOT free(3) the returned
361 * pointer.
362 */
363 static
364 const char *get_lttng_home_dir(void)
365 {
366 const char *val;
367
368 val = (const char *) lttng_ust_getenv("LTTNG_HOME");
369 if (val != NULL) {
370 return val;
371 }
372 return (const char *) lttng_ust_getenv("HOME");
373 }
374
375 /*
376 * Force a read (imply TLS allocation for dlopen) of TLS variables.
377 */
378 static
379 void lttng_nest_count_alloc_tls(void)
380 {
381 asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count)));
382 }
383
384 static
385 void lttng_ust_mutex_nest_alloc_tls(void)
386 {
387 asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest)));
388 }
389
390 /*
391 * Allocate lttng-ust urcu TLS.
392 */
393 static
394 void lttng_lttng_ust_urcu_alloc_tls(void)
395 {
396 (void) lttng_ust_urcu_read_ongoing();
397 }
398
399 void lttng_ust_alloc_tls(void)
400 {
401 lttng_lttng_ust_urcu_alloc_tls();
402 lttng_ringbuffer_alloc_tls();
403 lttng_vtid_alloc_tls();
404 lttng_nest_count_alloc_tls();
405 lttng_procname_alloc_tls();
406 lttng_ust_mutex_nest_alloc_tls();
407 lttng_ust_perf_counter_alloc_tls();
408 lttng_ust_common_alloc_tls();
409 lttng_cgroup_ns_alloc_tls();
410 lttng_ipc_ns_alloc_tls();
411 lttng_net_ns_alloc_tls();
412 lttng_time_ns_alloc_tls();
413 lttng_uts_ns_alloc_tls();
414 lttng_ust_ring_buffer_client_discard_alloc_tls();
415 lttng_ust_ring_buffer_client_discard_rt_alloc_tls();
416 lttng_ust_ring_buffer_client_overwrite_alloc_tls();
417 lttng_ust_ring_buffer_client_overwrite_rt_alloc_tls();
418 }
419
420 /*
421 * LTTng-UST uses Global Dynamic model TLS variables rather than IE
422 * model because many versions of glibc don't preallocate a pool large
423 * enough for TLS variables IE model defined in other shared libraries,
424 * and causes issues when using LTTng-UST for Java tracing.
425 *
426 * Because of this use of Global Dynamic TLS variables, users wishing to
427 * trace from signal handlers need to explicitly trigger the lazy
428 * allocation of those variables for each thread before using them.
429 * This can be triggered by calling lttng_ust_init_thread().
430 */
431 void lttng_ust_init_thread(void)
432 {
433 /*
434 * Because those TLS variables are global dynamic, we need to
435 * ensure those are initialized before a signal handler nesting over
436 * this thread attempts to use them.
437 */
438 lttng_ust_alloc_tls();
439 }
440
441 int lttng_get_notify_socket(void *owner)
442 {
443 struct sock_info *info = owner;
444
445 return info->notify_socket;
446 }
447
448
449 char* lttng_ust_sockinfo_get_procname(void *owner)
450 {
451 struct sock_info *info = owner;
452
453 return info->procname;
454 }
455
456 static
457 void print_cmd(int cmd, int handle)
458 {
459 const char *cmd_name = "Unknown";
460
461 if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping)
462 && cmd_name_mapping[cmd]) {
463 cmd_name = cmd_name_mapping[cmd];
464 }
465 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
466 cmd_name, cmd,
467 lttng_ust_obj_get_name(handle), handle);
468 }
469
470 static
471 int setup_global_apps(void)
472 {
473 int ret = 0;
474 assert(!global_apps.wait_shm_mmap);
475
476 global_apps.wait_shm_mmap = get_map_shm(&global_apps);
477 if (!global_apps.wait_shm_mmap) {
478 WARN("Unable to get map shm for global apps. Disabling LTTng-UST global tracing.");
479 global_apps.allowed = 0;
480 ret = -EIO;
481 goto error;
482 }
483
484 global_apps.allowed = 1;
485 lttng_pthread_getname_np(global_apps.procname, LTTNG_UST_ABI_PROCNAME_LEN);
486 error:
487 return ret;
488 }
489 static
490 int setup_local_apps(void)
491 {
492 int ret = 0;
493 const char *home_dir;
494 uid_t uid;
495
496 assert(!local_apps.wait_shm_mmap);
497
498 uid = getuid();
499 /*
500 * Disallow per-user tracing for setuid binaries.
501 */
502 if (uid != geteuid()) {
503 assert(local_apps.allowed == 0);
504 ret = 0;
505 goto end;
506 }
507 home_dir = get_lttng_home_dir();
508 if (!home_dir) {
509 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
510 assert(local_apps.allowed == 0);
511 ret = -ENOENT;
512 goto end;
513 }
514 local_apps.allowed = 1;
515 snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
516 home_dir,
517 LTTNG_DEFAULT_HOME_RUNDIR,
518 LTTNG_UST_SOCK_FILENAME);
519 snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
520 LTTNG_UST_WAIT_FILENAME,
521 uid);
522
523 local_apps.wait_shm_mmap = get_map_shm(&local_apps);
524 if (!local_apps.wait_shm_mmap) {
525 WARN("Unable to get map shm for local apps. Disabling LTTng-UST per-user tracing.");
526 local_apps.allowed = 0;
527 ret = -EIO;
528 goto end;
529 }
530
531 lttng_pthread_getname_np(local_apps.procname, LTTNG_UST_ABI_PROCNAME_LEN);
532 end:
533 return ret;
534 }
535
536 /*
537 * Get socket timeout, in ms.
538 * -1: wait forever. 0: don't wait. >0: timeout, in ms.
539 */
540 static
541 long get_timeout(void)
542 {
543 long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
544
545 if (!got_timeout_env) {
546 str_timeout = lttng_ust_getenv("LTTNG_UST_REGISTER_TIMEOUT");
547 got_timeout_env = 1;
548 }
549 if (str_timeout)
550 constructor_delay_ms = strtol(str_timeout, NULL, 10);
551 /* All negative values are considered as "-1". */
552 if (constructor_delay_ms < -1)
553 constructor_delay_ms = -1;
554 return constructor_delay_ms;
555 }
556
557 /* Timeout for notify socket send and recv. */
558 static
559 long get_notify_sock_timeout(void)
560 {
561 return get_timeout();
562 }
563
564 /* Timeout for connecting to cmd and notify sockets. */
565 static
566 long get_connect_sock_timeout(void)
567 {
568 return get_timeout();
569 }
570
571 /*
572 * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
573 */
574 static
575 int get_constructor_timeout(struct timespec *constructor_timeout)
576 {
577 long constructor_delay_ms;
578 int ret;
579
580 constructor_delay_ms = get_timeout();
581
582 switch (constructor_delay_ms) {
583 case -1:/* fall-through */
584 case 0:
585 return constructor_delay_ms;
586 default:
587 break;
588 }
589
590 /*
591 * If we are unable to find the current time, don't wait.
592 */
593 ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
594 if (ret) {
595 /* Don't wait. */
596 return 0;
597 }
598 constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
599 constructor_timeout->tv_nsec +=
600 (constructor_delay_ms % 1000UL) * 1000000UL;
601 if (constructor_timeout->tv_nsec >= 1000000000UL) {
602 constructor_timeout->tv_sec++;
603 constructor_timeout->tv_nsec -= 1000000000UL;
604 }
605 /* Timeout wait (constructor_delay_ms). */
606 return 1;
607 }
608
609 static
610 void get_allow_blocking(void)
611 {
612 const char *str_allow_blocking =
613 lttng_ust_getenv("LTTNG_UST_ALLOW_BLOCKING");
614
615 if (str_allow_blocking) {
616 DBG("%s environment variable is set",
617 "LTTNG_UST_ALLOW_BLOCKING");
618 lttng_ust_ringbuffer_set_allow_blocking();
619 }
620 }
621
622 static
623 int register_to_sessiond(int socket, enum lttng_ust_ctl_socket_type type)
624 {
625 return ustcomm_send_reg_msg(socket,
626 type,
627 CAA_BITS_PER_LONG,
628 lttng_ust_rb_alignof(uint8_t) * CHAR_BIT,
629 lttng_ust_rb_alignof(uint16_t) * CHAR_BIT,
630 lttng_ust_rb_alignof(uint32_t) * CHAR_BIT,
631 lttng_ust_rb_alignof(uint64_t) * CHAR_BIT,
632 lttng_ust_rb_alignof(unsigned long) * CHAR_BIT);
633 }
634
635 static
636 int send_reply(int sock, struct ustcomm_ust_reply *lur)
637 {
638 ssize_t len;
639
640 len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
641 switch (len) {
642 case sizeof(*lur):
643 DBG("message successfully sent");
644 return 0;
645 default:
646 if (len == -ECONNRESET) {
647 DBG("remote end closed connection");
648 return 0;
649 }
650 if (len < 0)
651 return len;
652 DBG("incorrect message size: %zd", len);
653 return -EINVAL;
654 }
655 }
656
657 static
658 void decrement_sem_count(unsigned int count)
659 {
660 int ret;
661
662 assert(uatomic_read(&sem_count) >= count);
663
664 if (uatomic_read(&sem_count) <= 0) {
665 return;
666 }
667
668 ret = uatomic_add_return(&sem_count, -count);
669 if (ret == 0) {
670 ret = sem_post(&constructor_wait);
671 assert(!ret);
672 }
673 }
674
675 static
676 int handle_register_done(struct sock_info *sock_info)
677 {
678 if (sock_info->registration_done)
679 return 0;
680 sock_info->registration_done = 1;
681
682 decrement_sem_count(1);
683 if (!sock_info->statedump_pending) {
684 sock_info->initial_statedump_done = 1;
685 decrement_sem_count(1);
686 }
687
688 return 0;
689 }
690
691 static
692 int handle_register_failed(struct sock_info *sock_info)
693 {
694 if (sock_info->registration_done)
695 return 0;
696 sock_info->registration_done = 1;
697 sock_info->initial_statedump_done = 1;
698
699 decrement_sem_count(2);
700
701 return 0;
702 }
703
704 /*
705 * Only execute pending statedump after the constructor semaphore has
706 * been posted by the current listener thread. This means statedump will
707 * only be performed after the "registration done" command is received
708 * from this thread's session daemon.
709 *
710 * This ensures we don't run into deadlock issues with the dynamic
711 * loader mutex, which is held while the constructor is called and
712 * waiting on the constructor semaphore. All operations requiring this
713 * dynamic loader lock need to be postponed using this mechanism.
714 *
715 * In a scenario with two session daemons connected to the application,
716 * it is possible that the first listener thread which receives the
717 * registration done command issues its statedump while the dynamic
718 * loader lock is still held by the application constructor waiting on
719 * the semaphore. It will however be allowed to proceed when the
720 * second session daemon sends the registration done command to the
721 * second listener thread. This situation therefore does not produce
722 * a deadlock.
723 */
724 static
725 void handle_pending_statedump(struct sock_info *sock_info)
726 {
727 if (sock_info->registration_done && sock_info->statedump_pending) {
728 sock_info->statedump_pending = 0;
729 pthread_mutex_lock(&ust_fork_mutex);
730 lttng_handle_pending_statedump(sock_info);
731 pthread_mutex_unlock(&ust_fork_mutex);
732
733 if (!sock_info->initial_statedump_done) {
734 sock_info->initial_statedump_done = 1;
735 decrement_sem_count(1);
736 }
737 }
738 }
739
740 static inline
741 const char *bytecode_type_str(uint32_t cmd)
742 {
743 switch (cmd) {
744 case LTTNG_UST_ABI_CAPTURE:
745 return "capture";
746 case LTTNG_UST_ABI_FILTER:
747 return "filter";
748 default:
749 abort();
750 }
751 }
752
753 static
754 int handle_bytecode_recv(struct sock_info *sock_info,
755 int sock, struct ustcomm_ust_msg *lum)
756 {
757 struct lttng_ust_bytecode_node *bytecode = NULL;
758 enum lttng_ust_bytecode_type type;
759 const struct lttng_ust_abi_objd_ops *ops;
760 uint32_t data_size, data_size_max, reloc_offset;
761 uint64_t seqnum;
762 ssize_t len;
763 int ret = 0;
764
765 switch (lum->cmd) {
766 case LTTNG_UST_ABI_FILTER:
767 type = LTTNG_UST_BYTECODE_TYPE_FILTER;
768 data_size = lum->u.filter.data_size;
769 data_size_max = LTTNG_UST_ABI_FILTER_BYTECODE_MAX_LEN;
770 reloc_offset = lum->u.filter.reloc_offset;
771 seqnum = lum->u.filter.seqnum;
772 break;
773 case LTTNG_UST_ABI_CAPTURE:
774 type = LTTNG_UST_BYTECODE_TYPE_CAPTURE;
775 data_size = lum->u.capture.data_size;
776 data_size_max = LTTNG_UST_ABI_CAPTURE_BYTECODE_MAX_LEN;
777 reloc_offset = lum->u.capture.reloc_offset;
778 seqnum = lum->u.capture.seqnum;
779 break;
780 default:
781 abort();
782 }
783
784 if (data_size > data_size_max) {
785 ERR("Bytecode %s data size is too large: %u bytes",
786 bytecode_type_str(lum->cmd), data_size);
787 ret = -EINVAL;
788 goto end;
789 }
790
791 if (reloc_offset > data_size) {
792 ERR("Bytecode %s reloc offset %u is not within data",
793 bytecode_type_str(lum->cmd), reloc_offset);
794 ret = -EINVAL;
795 goto end;
796 }
797
798 /* Allocate the structure AND the `data[]` field. */
799 bytecode = zmalloc(sizeof(*bytecode) + data_size);
800 if (!bytecode) {
801 ret = -ENOMEM;
802 goto end;
803 }
804
805 bytecode->bc.len = data_size;
806 bytecode->bc.reloc_offset = reloc_offset;
807 bytecode->bc.seqnum = seqnum;
808 bytecode->type = type;
809
810 len = ustcomm_recv_unix_sock(sock, bytecode->bc.data, bytecode->bc.len);
811 switch (len) {
812 case 0: /* orderly shutdown */
813 ret = 0;
814 goto end;
815 default:
816 if (len == bytecode->bc.len) {
817 DBG("Bytecode %s data received",
818 bytecode_type_str(lum->cmd));
819 break;
820 } else if (len < 0) {
821 DBG("Receive failed from lttng-sessiond with errno %d",
822 (int) -len);
823 if (len == -ECONNRESET) {
824 ERR("%s remote end closed connection",
825 sock_info->name);
826 ret = len;
827 goto end;
828 }
829 ret = len;
830 goto end;
831 } else {
832 DBG("Incorrect %s bytecode data message size: %zd",
833 bytecode_type_str(lum->cmd), len);
834 ret = -EINVAL;
835 goto end;
836 }
837 }
838
839 ops = lttng_ust_abi_objd_ops(lum->handle);
840 if (!ops) {
841 ret = -ENOENT;
842 goto end;
843 }
844
845 if (ops->cmd)
846 ret = ops->cmd(lum->handle, lum->cmd,
847 (unsigned long) &bytecode,
848 NULL, sock_info);
849 else
850 ret = -ENOSYS;
851
852 end:
853 free(bytecode);
854 return ret;
855 }
856
857 static
858 void prepare_cmd_reply(struct ustcomm_ust_reply *lur, uint32_t handle, uint32_t cmd, int ret)
859 {
860 lur->handle = handle;
861 lur->cmd = cmd;
862 lur->ret_val = ret;
863 if (ret >= 0) {
864 lur->ret_code = LTTNG_UST_OK;
865 } else {
866 /*
867 * Use -LTTNG_UST_ERR as wildcard for UST internal
868 * error that are not caused by the transport, except if
869 * we already have a more precise error message to
870 * report.
871 */
872 if (ret > -LTTNG_UST_ERR) {
873 /* Translate code to UST error. */
874 switch (ret) {
875 case -EEXIST:
876 lur->ret_code = -LTTNG_UST_ERR_EXIST;
877 break;
878 case -EINVAL:
879 lur->ret_code = -LTTNG_UST_ERR_INVAL;
880 break;
881 case -ENOENT:
882 lur->ret_code = -LTTNG_UST_ERR_NOENT;
883 break;
884 case -EPERM:
885 lur->ret_code = -LTTNG_UST_ERR_PERM;
886 break;
887 case -ENOSYS:
888 lur->ret_code = -LTTNG_UST_ERR_NOSYS;
889 break;
890 default:
891 lur->ret_code = -LTTNG_UST_ERR;
892 break;
893 }
894 } else {
895 lur->ret_code = ret;
896 }
897 }
898 }
899
900 static
901 int handle_message(struct sock_info *sock_info,
902 int sock, struct ustcomm_ust_msg *lum)
903 {
904 int ret = 0;
905 const struct lttng_ust_abi_objd_ops *ops;
906 struct ustcomm_ust_reply lur;
907 union lttng_ust_abi_args args;
908 char ctxstr[LTTNG_UST_ABI_SYM_NAME_LEN]; /* App context string. */
909 ssize_t len;
910
911 memset(&lur, 0, sizeof(lur));
912
913 if (ust_lock()) {
914 ret = -LTTNG_UST_ERR_EXITING;
915 goto error;
916 }
917
918 ops = lttng_ust_abi_objd_ops(lum->handle);
919 if (!ops) {
920 ret = -ENOENT;
921 goto error;
922 }
923
924 switch (lum->cmd) {
925 case LTTNG_UST_ABI_FILTER:
926 case LTTNG_UST_ABI_EXCLUSION:
927 case LTTNG_UST_ABI_CHANNEL:
928 case LTTNG_UST_ABI_STREAM:
929 case LTTNG_UST_ABI_CONTEXT:
930 /*
931 * Those commands send additional payload after struct
932 * ustcomm_ust_msg, which makes it pretty much impossible to
933 * deal with "unknown command" errors without leaving the
934 * communication pipe in a out-of-sync state. This is part of
935 * the ABI between liblttng-ust-ctl and liblttng-ust, and
936 * should be fixed on the next breaking
937 * LTTNG_UST_ABI_MAJOR_VERSION protocol bump by indicating the
938 * total command message length as part of a message header so
939 * that the protocol can recover from invalid command errors.
940 */
941 break;
942
943 case LTTNG_UST_ABI_CAPTURE:
944 case LTTNG_UST_ABI_COUNTER:
945 case LTTNG_UST_ABI_COUNTER_GLOBAL:
946 case LTTNG_UST_ABI_COUNTER_CPU:
947 case LTTNG_UST_ABI_EVENT_NOTIFIER_CREATE:
948 case LTTNG_UST_ABI_EVENT_NOTIFIER_GROUP_CREATE:
949 /*
950 * Those commands expect a reply to the struct ustcomm_ust_msg
951 * before sending additional payload.
952 */
953 prepare_cmd_reply(&lur, lum->handle, lum->cmd, 0);
954
955 ret = send_reply(sock, &lur);
956 if (ret < 0) {
957 DBG("error sending reply");
958 goto error;
959 }
960 break;
961
962 default:
963 /*
964 * Other commands either don't send additional payload, or are
965 * unknown.
966 */
967 break;
968 }
969
970 switch (lum->cmd) {
971 case LTTNG_UST_ABI_REGISTER_DONE:
972 if (lum->handle == LTTNG_UST_ABI_ROOT_HANDLE)
973 ret = handle_register_done(sock_info);
974 else
975 ret = -EINVAL;
976 break;
977 case LTTNG_UST_ABI_RELEASE:
978 if (lum->handle == LTTNG_UST_ABI_ROOT_HANDLE)
979 ret = -EPERM;
980 else
981 ret = lttng_ust_abi_objd_unref(lum->handle, 1);
982 break;
983 case LTTNG_UST_ABI_CAPTURE:
984 case LTTNG_UST_ABI_FILTER:
985 ret = handle_bytecode_recv(sock_info, sock, lum);
986 if (ret)
987 goto error;
988 break;
989 case LTTNG_UST_ABI_EXCLUSION:
990 {
991 /* Receive exclusion names */
992 struct lttng_ust_excluder_node *node;
993 unsigned int count;
994
995 count = lum->u.exclusion.count;
996 if (count == 0) {
997 /* There are no names to read */
998 ret = 0;
999 goto error;
1000 }
1001 node = zmalloc(sizeof(*node) +
1002 count * LTTNG_UST_ABI_SYM_NAME_LEN);
1003 if (!node) {
1004 ret = -ENOMEM;
1005 goto error;
1006 }
1007 node->excluder.count = count;
1008 len = ustcomm_recv_unix_sock(sock, node->excluder.names,
1009 count * LTTNG_UST_ABI_SYM_NAME_LEN);
1010 switch (len) {
1011 case 0: /* orderly shutdown */
1012 ret = 0;
1013 free(node);
1014 goto error;
1015 default:
1016 if (len == count * LTTNG_UST_ABI_SYM_NAME_LEN) {
1017 DBG("Exclusion data received");
1018 break;
1019 } else if (len < 0) {
1020 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1021 if (len == -ECONNRESET) {
1022 ERR("%s remote end closed connection", sock_info->name);
1023 ret = len;
1024 free(node);
1025 goto error;
1026 }
1027 ret = len;
1028 free(node);
1029 goto error;
1030 } else {
1031 DBG("Incorrect exclusion data message size: %zd", len);
1032 ret = -EINVAL;
1033 free(node);
1034 goto error;
1035 }
1036 }
1037 if (ops->cmd)
1038 ret = ops->cmd(lum->handle, lum->cmd,
1039 (unsigned long) &node,
1040 &args, sock_info);
1041 else
1042 ret = -ENOSYS;
1043 free(node);
1044 break;
1045 }
1046 case LTTNG_UST_ABI_EVENT_NOTIFIER_GROUP_CREATE:
1047 {
1048 int event_notifier_notif_fd, close_ret;
1049
1050 len = ustcomm_recv_event_notifier_notif_fd_from_sessiond(sock,
1051 &event_notifier_notif_fd);
1052 switch (len) {
1053 case 0: /* orderly shutdown */
1054 ret = 0;
1055 goto error;
1056 case 1:
1057 break;
1058 default:
1059 if (len < 0) {
1060 DBG("Receive failed from lttng-sessiond with errno %d",
1061 (int) -len);
1062 if (len == -ECONNRESET) {
1063 ERR("%s remote end closed connection",
1064 sock_info->name);
1065 ret = len;
1066 goto error;
1067 }
1068 ret = len;
1069 goto error;
1070 } else {
1071 DBG("Incorrect event notifier fd message size: %zd",
1072 len);
1073 ret = -EINVAL;
1074 goto error;
1075 }
1076 }
1077 args.event_notifier_handle.event_notifier_notif_fd =
1078 event_notifier_notif_fd;
1079 if (ops->cmd)
1080 ret = ops->cmd(lum->handle, lum->cmd,
1081 (unsigned long) &lum->u,
1082 &args, sock_info);
1083 else
1084 ret = -ENOSYS;
1085 if (args.event_notifier_handle.event_notifier_notif_fd >= 0) {
1086 lttng_ust_lock_fd_tracker();
1087 close_ret = close(args.event_notifier_handle.event_notifier_notif_fd);
1088 lttng_ust_unlock_fd_tracker();
1089 if (close_ret)
1090 PERROR("close");
1091 }
1092 break;
1093 }
1094 case LTTNG_UST_ABI_CHANNEL:
1095 {
1096 void *chan_data;
1097 int wakeup_fd;
1098
1099 len = ustcomm_recv_channel_from_sessiond(sock,
1100 &chan_data, lum->u.channel.len,
1101 &wakeup_fd);
1102 switch (len) {
1103 case 0: /* orderly shutdown */
1104 ret = 0;
1105 goto error;
1106 default:
1107 if (len == lum->u.channel.len) {
1108 DBG("channel data received");
1109 break;
1110 } else if (len < 0) {
1111 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1112 if (len == -ECONNRESET) {
1113 ERR("%s remote end closed connection", sock_info->name);
1114 ret = len;
1115 goto error;
1116 }
1117 ret = len;
1118 goto error;
1119 } else {
1120 DBG("incorrect channel data message size: %zd", len);
1121 ret = -EINVAL;
1122 goto error;
1123 }
1124 }
1125 args.channel.chan_data = chan_data;
1126 args.channel.wakeup_fd = wakeup_fd;
1127 if (ops->cmd)
1128 ret = ops->cmd(lum->handle, lum->cmd,
1129 (unsigned long) &lum->u,
1130 &args, sock_info);
1131 else
1132 ret = -ENOSYS;
1133 if (args.channel.wakeup_fd >= 0) {
1134 int close_ret;
1135
1136 lttng_ust_lock_fd_tracker();
1137 close_ret = close(args.channel.wakeup_fd);
1138 lttng_ust_unlock_fd_tracker();
1139 args.channel.wakeup_fd = -1;
1140 if (close_ret)
1141 PERROR("close");
1142 }
1143 free(args.channel.chan_data);
1144 break;
1145 }
1146 case LTTNG_UST_ABI_STREAM:
1147 {
1148 int close_ret;
1149
1150 /* Receive shm_fd, wakeup_fd */
1151 ret = ustcomm_recv_stream_from_sessiond(sock,
1152 NULL,
1153 &args.stream.shm_fd,
1154 &args.stream.wakeup_fd);
1155 if (ret) {
1156 goto error;
1157 }
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 if (args.stream.shm_fd >= 0) {
1166 lttng_ust_lock_fd_tracker();
1167 close_ret = close(args.stream.shm_fd);
1168 lttng_ust_unlock_fd_tracker();
1169 args.stream.shm_fd = -1;
1170 if (close_ret)
1171 PERROR("close");
1172 }
1173 if (args.stream.wakeup_fd >= 0) {
1174 lttng_ust_lock_fd_tracker();
1175 close_ret = close(args.stream.wakeup_fd);
1176 lttng_ust_unlock_fd_tracker();
1177 args.stream.wakeup_fd = -1;
1178 if (close_ret)
1179 PERROR("close");
1180 }
1181 break;
1182 }
1183 case LTTNG_UST_ABI_CONTEXT:
1184 switch (lum->u.context.ctx) {
1185 case LTTNG_UST_ABI_CONTEXT_APP_CONTEXT:
1186 {
1187 char *p;
1188 size_t ctxlen, recvlen;
1189
1190 ctxlen = strlen("$app.") + lum->u.context.u.app_ctx.provider_name_len - 1
1191 + strlen(":") + lum->u.context.u.app_ctx.ctx_name_len;
1192 if (ctxlen >= LTTNG_UST_ABI_SYM_NAME_LEN) {
1193 ERR("Application context string length size is too large: %zu bytes",
1194 ctxlen);
1195 ret = -EINVAL;
1196 goto error;
1197 }
1198 strcpy(ctxstr, "$app.");
1199 p = &ctxstr[strlen("$app.")];
1200 recvlen = ctxlen - strlen("$app.");
1201 len = ustcomm_recv_unix_sock(sock, p, recvlen);
1202 switch (len) {
1203 case 0: /* orderly shutdown */
1204 ret = 0;
1205 goto error;
1206 default:
1207 if (len == recvlen) {
1208 DBG("app context data received");
1209 break;
1210 } else if (len < 0) {
1211 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1212 if (len == -ECONNRESET) {
1213 ERR("%s remote end closed connection", sock_info->name);
1214 ret = len;
1215 goto error;
1216 }
1217 ret = len;
1218 goto error;
1219 } else {
1220 DBG("incorrect app context data message size: %zd", len);
1221 ret = -EINVAL;
1222 goto error;
1223 }
1224 }
1225 /* Put : between provider and ctxname. */
1226 p[lum->u.context.u.app_ctx.provider_name_len - 1] = ':';
1227 args.app_context.ctxname = ctxstr;
1228 break;
1229 }
1230 default:
1231 break;
1232 }
1233 if (ops->cmd) {
1234 ret = ops->cmd(lum->handle, lum->cmd,
1235 (unsigned long) &lum->u,
1236 &args, sock_info);
1237 } else {
1238 ret = -ENOSYS;
1239 }
1240 break;
1241 case LTTNG_UST_ABI_COUNTER:
1242 {
1243 void *counter_data;
1244
1245 len = ustcomm_recv_counter_from_sessiond(sock,
1246 &counter_data, lum->u.counter.len);
1247 switch (len) {
1248 case 0: /* orderly shutdown */
1249 ret = 0;
1250 goto error;
1251 default:
1252 if (len == lum->u.counter.len) {
1253 DBG("counter data received");
1254 break;
1255 } else if (len < 0) {
1256 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1257 if (len == -ECONNRESET) {
1258 ERR("%s remote end closed connection", sock_info->name);
1259 ret = len;
1260 goto error;
1261 }
1262 ret = len;
1263 goto error;
1264 } else {
1265 DBG("incorrect counter data message size: %zd", len);
1266 ret = -EINVAL;
1267 goto error;
1268 }
1269 }
1270 args.counter.counter_data = counter_data;
1271 if (ops->cmd)
1272 ret = ops->cmd(lum->handle, lum->cmd,
1273 (unsigned long) &lum->u,
1274 &args, sock_info);
1275 else
1276 ret = -ENOSYS;
1277 free(args.counter.counter_data);
1278 break;
1279 }
1280 case LTTNG_UST_ABI_COUNTER_GLOBAL:
1281 {
1282 /* Receive shm_fd */
1283 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1284 &args.counter_shm.shm_fd);
1285 if (ret) {
1286 goto error;
1287 }
1288
1289 if (ops->cmd)
1290 ret = ops->cmd(lum->handle, lum->cmd,
1291 (unsigned long) &lum->u,
1292 &args, sock_info);
1293 else
1294 ret = -ENOSYS;
1295 if (args.counter_shm.shm_fd >= 0) {
1296 int close_ret;
1297
1298 lttng_ust_lock_fd_tracker();
1299 close_ret = close(args.counter_shm.shm_fd);
1300 lttng_ust_unlock_fd_tracker();
1301 args.counter_shm.shm_fd = -1;
1302 if (close_ret)
1303 PERROR("close");
1304 }
1305 break;
1306 }
1307 case LTTNG_UST_ABI_COUNTER_CPU:
1308 {
1309 /* Receive shm_fd */
1310 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1311 &args.counter_shm.shm_fd);
1312 if (ret) {
1313 goto error;
1314 }
1315
1316 if (ops->cmd)
1317 ret = ops->cmd(lum->handle, lum->cmd,
1318 (unsigned long) &lum->u,
1319 &args, sock_info);
1320 else
1321 ret = -ENOSYS;
1322 if (args.counter_shm.shm_fd >= 0) {
1323 int close_ret;
1324
1325 lttng_ust_lock_fd_tracker();
1326 close_ret = close(args.counter_shm.shm_fd);
1327 lttng_ust_unlock_fd_tracker();
1328 args.counter_shm.shm_fd = -1;
1329 if (close_ret)
1330 PERROR("close");
1331 }
1332 break;
1333 }
1334 case LTTNG_UST_ABI_EVENT_NOTIFIER_CREATE:
1335 {
1336 /* Receive struct lttng_ust_event_notifier */
1337 struct lttng_ust_abi_event_notifier event_notifier;
1338
1339 if (sizeof(event_notifier) != lum->u.event_notifier.len) {
1340 DBG("incorrect event notifier data message size: %u", lum->u.event_notifier.len);
1341 ret = -EINVAL;
1342 goto error;
1343 }
1344 len = ustcomm_recv_unix_sock(sock, &event_notifier, sizeof(event_notifier));
1345 switch (len) {
1346 case 0: /* orderly shutdown */
1347 ret = 0;
1348 goto error;
1349 default:
1350 if (len == sizeof(event_notifier)) {
1351 DBG("event notifier data received");
1352 break;
1353 } else if (len < 0) {
1354 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1355 if (len == -ECONNRESET) {
1356 ERR("%s remote end closed connection", sock_info->name);
1357 ret = len;
1358 goto error;
1359 }
1360 ret = len;
1361 goto error;
1362 } else {
1363 DBG("incorrect event notifier data message size: %zd", len);
1364 ret = -EINVAL;
1365 goto error;
1366 }
1367 }
1368 if (ops->cmd)
1369 ret = ops->cmd(lum->handle, lum->cmd,
1370 (unsigned long) &event_notifier,
1371 &args, sock_info);
1372 else
1373 ret = -ENOSYS;
1374 break;
1375 }
1376
1377 default:
1378 if (ops->cmd)
1379 ret = ops->cmd(lum->handle, lum->cmd,
1380 (unsigned long) &lum->u,
1381 &args, sock_info);
1382 else
1383 ret = -ENOSYS;
1384 break;
1385 }
1386
1387 prepare_cmd_reply(&lur, lum->handle, lum->cmd, ret);
1388
1389 if (ret >= 0) {
1390 switch (lum->cmd) {
1391 case LTTNG_UST_ABI_TRACER_VERSION:
1392 lur.u.version = lum->u.version;
1393 break;
1394 case LTTNG_UST_ABI_TRACEPOINT_LIST_GET:
1395 memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
1396 break;
1397 }
1398 }
1399 DBG("Return value: %d", lur.ret_val);
1400
1401 ust_unlock();
1402
1403 /*
1404 * Performed delayed statedump operations outside of the UST
1405 * lock. We need to take the dynamic loader lock before we take
1406 * the UST lock internally within handle_pending_statedump().
1407 */
1408 handle_pending_statedump(sock_info);
1409
1410 if (ust_lock()) {
1411 ret = -LTTNG_UST_ERR_EXITING;
1412 goto error;
1413 }
1414
1415 ret = send_reply(sock, &lur);
1416 if (ret < 0) {
1417 DBG("error sending reply");
1418 goto error;
1419 }
1420
1421 /*
1422 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1423 * after the reply.
1424 */
1425 if (lur.ret_code == LTTNG_UST_OK) {
1426 switch (lum->cmd) {
1427 case LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST_GET:
1428 len = ustcomm_send_unix_sock(sock,
1429 &args.field_list.entry,
1430 sizeof(args.field_list.entry));
1431 if (len < 0) {
1432 ret = len;
1433 goto error;
1434 }
1435 if (len != sizeof(args.field_list.entry)) {
1436 ret = -EINVAL;
1437 goto error;
1438 }
1439 }
1440 }
1441
1442 error:
1443 ust_unlock();
1444
1445 return ret;
1446 }
1447
1448 static
1449 void cleanup_sock_info(struct sock_info *sock_info, int exiting)
1450 {
1451 int ret;
1452
1453 if (sock_info->root_handle != -1) {
1454 ret = lttng_ust_abi_objd_unref(sock_info->root_handle, 1);
1455 if (ret) {
1456 ERR("Error unref root handle");
1457 }
1458 sock_info->root_handle = -1;
1459 }
1460 sock_info->registration_done = 0;
1461 sock_info->initial_statedump_done = 0;
1462
1463 /*
1464 * wait_shm_mmap, socket and notify socket are used by listener
1465 * threads outside of the ust lock, so we cannot tear them down
1466 * ourselves, because we cannot join on these threads. Leave
1467 * responsibility of cleaning up these resources to the OS
1468 * process exit.
1469 */
1470 if (exiting)
1471 return;
1472
1473 if (sock_info->socket != -1) {
1474 ret = ustcomm_close_unix_sock(sock_info->socket);
1475 if (ret) {
1476 ERR("Error closing ust cmd socket");
1477 }
1478 sock_info->socket = -1;
1479 }
1480 if (sock_info->notify_socket != -1) {
1481 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1482 if (ret) {
1483 ERR("Error closing ust notify socket");
1484 }
1485 sock_info->notify_socket = -1;
1486 }
1487 if (sock_info->wait_shm_mmap) {
1488 long page_size;
1489
1490 page_size = LTTNG_UST_PAGE_SIZE;
1491 if (page_size <= 0) {
1492 if (!page_size) {
1493 errno = EINVAL;
1494 }
1495 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1496 } else {
1497 ret = munmap(sock_info->wait_shm_mmap, page_size);
1498 if (ret) {
1499 ERR("Error unmapping wait shm");
1500 }
1501 }
1502 sock_info->wait_shm_mmap = NULL;
1503 }
1504 }
1505
1506 /*
1507 * Using fork to set umask in the child process (not multi-thread safe).
1508 * We deal with the shm_open vs ftruncate race (happening when the
1509 * sessiond owns the shm and does not let everybody modify it, to ensure
1510 * safety against shm_unlink) by simply letting the mmap fail and
1511 * retrying after a few seconds.
1512 * For global shm, everybody has rw access to it until the sessiond
1513 * starts.
1514 */
1515 static
1516 int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
1517 {
1518 int wait_shm_fd, ret;
1519 pid_t pid;
1520
1521 /*
1522 * Try to open read-only.
1523 */
1524 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1525 if (wait_shm_fd >= 0) {
1526 int32_t tmp_read;
1527 ssize_t len;
1528 size_t bytes_read = 0;
1529
1530 /*
1531 * Try to read the fd. If unable to do so, try opening
1532 * it in write mode.
1533 */
1534 do {
1535 len = read(wait_shm_fd,
1536 &((char *) &tmp_read)[bytes_read],
1537 sizeof(tmp_read) - bytes_read);
1538 if (len > 0) {
1539 bytes_read += len;
1540 }
1541 } while ((len < 0 && errno == EINTR)
1542 || (len > 0 && bytes_read < sizeof(tmp_read)));
1543 if (bytes_read != sizeof(tmp_read)) {
1544 ret = close(wait_shm_fd);
1545 if (ret) {
1546 ERR("close wait_shm_fd");
1547 }
1548 goto open_write;
1549 }
1550 goto end;
1551 } else if (wait_shm_fd < 0 && errno != ENOENT) {
1552 /*
1553 * Real-only open did not work, and it's not because the
1554 * entry was not present. It's a failure that prohibits
1555 * using shm.
1556 */
1557 ERR("Error opening shm %s", sock_info->wait_shm_path);
1558 goto end;
1559 }
1560
1561 open_write:
1562 /*
1563 * If the open failed because the file did not exist, or because
1564 * the file was not truncated yet, try creating it ourself.
1565 */
1566 URCU_TLS(lttng_ust_nest_count)++;
1567 pid = fork();
1568 URCU_TLS(lttng_ust_nest_count)--;
1569 if (pid > 0) {
1570 int status;
1571
1572 /*
1573 * Parent: wait for child to return, in which case the
1574 * shared memory map will have been created.
1575 */
1576 pid = wait(&status);
1577 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
1578 wait_shm_fd = -1;
1579 goto end;
1580 }
1581 /*
1582 * Try to open read-only again after creation.
1583 */
1584 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1585 if (wait_shm_fd < 0) {
1586 /*
1587 * Real-only open did not work. It's a failure
1588 * that prohibits using shm.
1589 */
1590 ERR("Error opening shm %s", sock_info->wait_shm_path);
1591 goto end;
1592 }
1593 goto end;
1594 } else if (pid == 0) {
1595 int create_mode;
1596
1597 /* Child */
1598 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
1599 if (sock_info->global)
1600 create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
1601 /*
1602 * We're alone in a child process, so we can modify the
1603 * process-wide umask.
1604 */
1605 umask(~create_mode);
1606 /*
1607 * Try creating shm (or get rw access).
1608 * We don't do an exclusive open, because we allow other
1609 * processes to create+ftruncate it concurrently.
1610 */
1611 wait_shm_fd = shm_open(sock_info->wait_shm_path,
1612 O_RDWR | O_CREAT, create_mode);
1613 if (wait_shm_fd >= 0) {
1614 ret = ftruncate(wait_shm_fd, mmap_size);
1615 if (ret) {
1616 PERROR("ftruncate");
1617 _exit(EXIT_FAILURE);
1618 }
1619 _exit(EXIT_SUCCESS);
1620 }
1621 /*
1622 * For local shm, we need to have rw access to accept
1623 * opening it: this means the local sessiond will be
1624 * able to wake us up. For global shm, we open it even
1625 * if rw access is not granted, because the root.root
1626 * sessiond will be able to override all rights and wake
1627 * us up.
1628 */
1629 if (!sock_info->global && errno != EACCES) {
1630 ERR("Error opening shm %s", sock_info->wait_shm_path);
1631 _exit(EXIT_FAILURE);
1632 }
1633 /*
1634 * The shm exists, but we cannot open it RW. Report
1635 * success.
1636 */
1637 _exit(EXIT_SUCCESS);
1638 } else {
1639 return -1;
1640 }
1641 end:
1642 if (wait_shm_fd >= 0 && !sock_info->global) {
1643 struct stat statbuf;
1644
1645 /*
1646 * Ensure that our user is the owner of the shm file for
1647 * local shm. If we do not own the file, it means our
1648 * sessiond will not have access to wake us up (there is
1649 * probably a rogue process trying to fake our
1650 * sessiond). Fallback to polling method in this case.
1651 */
1652 ret = fstat(wait_shm_fd, &statbuf);
1653 if (ret) {
1654 PERROR("fstat");
1655 goto error_close;
1656 }
1657 if (statbuf.st_uid != getuid())
1658 goto error_close;
1659 }
1660 return wait_shm_fd;
1661
1662 error_close:
1663 ret = close(wait_shm_fd);
1664 if (ret) {
1665 PERROR("Error closing fd");
1666 }
1667 return -1;
1668 }
1669
1670 static
1671 char *get_map_shm(struct sock_info *sock_info)
1672 {
1673 long page_size;
1674 int wait_shm_fd, ret;
1675 char *wait_shm_mmap;
1676
1677 page_size = sysconf(_SC_PAGE_SIZE);
1678 if (page_size <= 0) {
1679 if (!page_size) {
1680 errno = EINVAL;
1681 }
1682 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1683 goto error;
1684 }
1685
1686 lttng_ust_lock_fd_tracker();
1687 wait_shm_fd = get_wait_shm(sock_info, page_size);
1688 if (wait_shm_fd < 0) {
1689 lttng_ust_unlock_fd_tracker();
1690 goto error;
1691 }
1692
1693 ret = lttng_ust_add_fd_to_tracker(wait_shm_fd);
1694 if (ret < 0) {
1695 ret = close(wait_shm_fd);
1696 if (!ret) {
1697 PERROR("Error closing fd");
1698 }
1699 lttng_ust_unlock_fd_tracker();
1700 goto error;
1701 }
1702
1703 wait_shm_fd = ret;
1704 lttng_ust_unlock_fd_tracker();
1705
1706 wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
1707 MAP_SHARED, wait_shm_fd, 0);
1708
1709 /* close shm fd immediately after taking the mmap reference */
1710 lttng_ust_lock_fd_tracker();
1711 ret = close(wait_shm_fd);
1712 if (!ret) {
1713 lttng_ust_delete_fd_from_tracker(wait_shm_fd);
1714 } else {
1715 PERROR("Error closing fd");
1716 }
1717 lttng_ust_unlock_fd_tracker();
1718
1719 if (wait_shm_mmap == MAP_FAILED) {
1720 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1721 goto error;
1722 }
1723 return wait_shm_mmap;
1724
1725 error:
1726 return NULL;
1727 }
1728
1729 static
1730 void wait_for_sessiond(struct sock_info *sock_info)
1731 {
1732 /* Use ust_lock to check if we should quit. */
1733 if (ust_lock()) {
1734 goto quit;
1735 }
1736 if (wait_poll_fallback) {
1737 goto error;
1738 }
1739 ust_unlock();
1740
1741 assert(sock_info->wait_shm_mmap);
1742
1743 DBG("Waiting for %s apps sessiond", sock_info->name);
1744 /* Wait for futex wakeup */
1745 if (uatomic_read((int32_t *) sock_info->wait_shm_mmap))
1746 goto end_wait;
1747
1748 while (lttng_ust_futex_async((int32_t *) sock_info->wait_shm_mmap,
1749 FUTEX_WAIT, 0, NULL, NULL, 0)) {
1750 switch (errno) {
1751 case EWOULDBLOCK:
1752 /* Value already changed. */
1753 goto end_wait;
1754 case EINTR:
1755 /* Retry if interrupted by signal. */
1756 break; /* Get out of switch. */
1757 case EFAULT:
1758 wait_poll_fallback = 1;
1759 DBG(
1760 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1761 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1762 "Please upgrade your kernel "
1763 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1764 "mainline). LTTng-UST will use polling mode fallback.");
1765 if (lttng_ust_logging_debug_enabled())
1766 PERROR("futex");
1767 goto end_wait;
1768 }
1769 }
1770 end_wait:
1771 return;
1772
1773 quit:
1774 ust_unlock();
1775 return;
1776
1777 error:
1778 ust_unlock();
1779 return;
1780 }
1781
1782 /*
1783 * This thread does not allocate any resource, except within
1784 * handle_message, within mutex protection. This mutex protects against
1785 * fork and exit.
1786 * The other moment it allocates resources is at socket connection, which
1787 * is also protected by the mutex.
1788 */
1789 static
1790 void *ust_listener_thread(void *arg)
1791 {
1792 struct sock_info *sock_info = arg;
1793 int sock, ret, prev_connect_failed = 0, has_waited = 0, fd;
1794 long timeout;
1795
1796 lttng_ust_alloc_tls();
1797 /*
1798 * If available, add '-ust' to the end of this thread's
1799 * process name
1800 */
1801 ret = lttng_ust_setustprocname();
1802 if (ret) {
1803 ERR("Unable to set UST process name");
1804 }
1805
1806 /* Restart trying to connect to the session daemon */
1807 restart:
1808 if (prev_connect_failed) {
1809 /* Wait for sessiond availability with pipe */
1810 wait_for_sessiond(sock_info);
1811 if (has_waited) {
1812 has_waited = 0;
1813 /*
1814 * Sleep for 5 seconds before retrying after a
1815 * sequence of failure / wait / failure. This
1816 * deals with a killed or broken session daemon.
1817 */
1818 sleep(5);
1819 } else {
1820 has_waited = 1;
1821 }
1822 prev_connect_failed = 0;
1823 }
1824
1825 if (ust_lock()) {
1826 goto quit;
1827 }
1828
1829 if (sock_info->socket != -1) {
1830 /* FD tracker is updated by ustcomm_close_unix_sock() */
1831 ret = ustcomm_close_unix_sock(sock_info->socket);
1832 if (ret) {
1833 ERR("Error closing %s ust cmd socket",
1834 sock_info->name);
1835 }
1836 sock_info->socket = -1;
1837 }
1838 if (sock_info->notify_socket != -1) {
1839 /* FD tracker is updated by ustcomm_close_unix_sock() */
1840 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1841 if (ret) {
1842 ERR("Error closing %s ust notify socket",
1843 sock_info->name);
1844 }
1845 sock_info->notify_socket = -1;
1846 }
1847
1848
1849 /*
1850 * Register. We need to perform both connect and sending
1851 * registration message before doing the next connect otherwise
1852 * we may reach unix socket connect queue max limits and block
1853 * on the 2nd connect while the session daemon is awaiting the
1854 * first connect registration message.
1855 */
1856 /* Connect cmd socket */
1857 lttng_ust_lock_fd_tracker();
1858 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1859 get_connect_sock_timeout());
1860 if (ret < 0) {
1861 lttng_ust_unlock_fd_tracker();
1862 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1863 prev_connect_failed = 1;
1864
1865 /*
1866 * If we cannot find the sessiond daemon, don't delay
1867 * constructor execution.
1868 */
1869 ret = handle_register_failed(sock_info);
1870 assert(!ret);
1871 ust_unlock();
1872 goto restart;
1873 }
1874 fd = ret;
1875 ret = lttng_ust_add_fd_to_tracker(fd);
1876 if (ret < 0) {
1877 ret = close(fd);
1878 if (ret) {
1879 PERROR("close on sock_info->socket");
1880 }
1881 ret = -1;
1882 lttng_ust_unlock_fd_tracker();
1883 ust_unlock();
1884 goto quit;
1885 }
1886
1887 sock_info->socket = ret;
1888 lttng_ust_unlock_fd_tracker();
1889
1890 ust_unlock();
1891 /*
1892 * Unlock/relock ust lock because connect is blocking (with
1893 * timeout). Don't delay constructors on the ust lock for too
1894 * long.
1895 */
1896 if (ust_lock()) {
1897 goto quit;
1898 }
1899
1900 /*
1901 * Create only one root handle per listener thread for the whole
1902 * process lifetime, so we ensure we get ID which is statically
1903 * assigned to the root handle.
1904 */
1905 if (sock_info->root_handle == -1) {
1906 ret = lttng_abi_create_root_handle();
1907 if (ret < 0) {
1908 ERR("Error creating root handle");
1909 goto quit;
1910 }
1911 sock_info->root_handle = ret;
1912 }
1913
1914 ret = register_to_sessiond(sock_info->socket, LTTNG_UST_CTL_SOCKET_CMD);
1915 if (ret < 0) {
1916 ERR("Error registering to %s ust cmd socket",
1917 sock_info->name);
1918 prev_connect_failed = 1;
1919 /*
1920 * If we cannot register to the sessiond daemon, don't
1921 * delay constructor execution.
1922 */
1923 ret = handle_register_failed(sock_info);
1924 assert(!ret);
1925 ust_unlock();
1926 goto restart;
1927 }
1928
1929 ust_unlock();
1930 /*
1931 * Unlock/relock ust lock because connect is blocking (with
1932 * timeout). Don't delay constructors on the ust lock for too
1933 * long.
1934 */
1935 if (ust_lock()) {
1936 goto quit;
1937 }
1938
1939 /* Connect notify socket */
1940 lttng_ust_lock_fd_tracker();
1941 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1942 get_connect_sock_timeout());
1943 if (ret < 0) {
1944 lttng_ust_unlock_fd_tracker();
1945 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1946 prev_connect_failed = 1;
1947
1948 /*
1949 * If we cannot find the sessiond daemon, don't delay
1950 * constructor execution.
1951 */
1952 ret = handle_register_failed(sock_info);
1953 assert(!ret);
1954 ust_unlock();
1955 goto restart;
1956 }
1957
1958 fd = ret;
1959 ret = lttng_ust_add_fd_to_tracker(fd);
1960 if (ret < 0) {
1961 ret = close(fd);
1962 if (ret) {
1963 PERROR("close on sock_info->notify_socket");
1964 }
1965 ret = -1;
1966 lttng_ust_unlock_fd_tracker();
1967 ust_unlock();
1968 goto quit;
1969 }
1970
1971 sock_info->notify_socket = ret;
1972 lttng_ust_unlock_fd_tracker();
1973
1974 ust_unlock();
1975 /*
1976 * Unlock/relock ust lock because connect is blocking (with
1977 * timeout). Don't delay constructors on the ust lock for too
1978 * long.
1979 */
1980 if (ust_lock()) {
1981 goto quit;
1982 }
1983
1984 timeout = get_notify_sock_timeout();
1985 if (timeout >= 0) {
1986 /*
1987 * Give at least 10ms to sessiond to reply to
1988 * notifications.
1989 */
1990 if (timeout < 10)
1991 timeout = 10;
1992 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
1993 timeout);
1994 if (ret < 0) {
1995 WARN("Error setting socket receive timeout");
1996 }
1997 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
1998 timeout);
1999 if (ret < 0) {
2000 WARN("Error setting socket send timeout");
2001 }
2002 } else if (timeout < -1) {
2003 WARN("Unsupported timeout value %ld", timeout);
2004 }
2005
2006 ret = register_to_sessiond(sock_info->notify_socket,
2007 LTTNG_UST_CTL_SOCKET_NOTIFY);
2008 if (ret < 0) {
2009 ERR("Error registering to %s ust notify socket",
2010 sock_info->name);
2011 prev_connect_failed = 1;
2012 /*
2013 * If we cannot register to the sessiond daemon, don't
2014 * delay constructor execution.
2015 */
2016 ret = handle_register_failed(sock_info);
2017 assert(!ret);
2018 ust_unlock();
2019 goto restart;
2020 }
2021 sock = sock_info->socket;
2022
2023 ust_unlock();
2024
2025 for (;;) {
2026 ssize_t len;
2027 struct ustcomm_ust_msg lum;
2028
2029 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
2030 switch (len) {
2031 case 0: /* orderly shutdown */
2032 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
2033 if (ust_lock()) {
2034 goto quit;
2035 }
2036 /*
2037 * Either sessiond has shutdown or refused us by closing the socket.
2038 * In either case, we don't want to delay construction execution,
2039 * and we need to wait before retry.
2040 */
2041 prev_connect_failed = 1;
2042 /*
2043 * If we cannot register to the sessiond daemon, don't
2044 * delay constructor execution.
2045 */
2046 ret = handle_register_failed(sock_info);
2047 assert(!ret);
2048 ust_unlock();
2049 goto end;
2050 case sizeof(lum):
2051 print_cmd(lum.cmd, lum.handle);
2052 ret = handle_message(sock_info, sock, &lum);
2053 if (ret) {
2054 ERR("Error handling message for %s socket",
2055 sock_info->name);
2056 /*
2057 * Close socket if protocol error is
2058 * detected.
2059 */
2060 goto end;
2061 }
2062 continue;
2063 default:
2064 if (len < 0) {
2065 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
2066 } else {
2067 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
2068 }
2069 if (len == -ECONNRESET) {
2070 DBG("%s remote end closed connection", sock_info->name);
2071 goto end;
2072 }
2073 goto end;
2074 }
2075
2076 }
2077 end:
2078 if (ust_lock()) {
2079 goto quit;
2080 }
2081 /* Cleanup socket handles before trying to reconnect */
2082 lttng_ust_abi_objd_table_owner_cleanup(sock_info);
2083 ust_unlock();
2084 goto restart; /* try to reconnect */
2085
2086 quit:
2087 ust_unlock();
2088
2089 pthread_mutex_lock(&ust_exit_mutex);
2090 sock_info->thread_active = 0;
2091 pthread_mutex_unlock(&ust_exit_mutex);
2092 return NULL;
2093 }
2094
2095 /*
2096 * Weak symbol to call when the ust malloc wrapper is not loaded.
2097 */
2098 __attribute__((weak))
2099 void lttng_ust_libc_wrapper_malloc_ctor(void)
2100 {
2101 }
2102
2103 /*
2104 * Use a symbol of the previous ABI to detect if liblttng-ust.so.0 is loaded in
2105 * the current process.
2106 */
2107 #define LTTNG_UST_SONAME_0_SYM "ltt_probe_register"
2108
2109 static
2110 void lttng_ust_check_soname_0(void)
2111 {
2112 if (!dlsym(RTLD_DEFAULT, LTTNG_UST_SONAME_0_SYM))
2113 return;
2114
2115 CRIT("Incompatible library ABIs detected within the same process. "
2116 "The process is likely linked against different major soname of LTTng-UST which is unsupported. "
2117 "The detection was triggered by lookup of ABI 0 symbol \"%s\" in the Global Symbol Table\n",
2118 LTTNG_UST_SONAME_0_SYM);
2119 }
2120
2121 /*
2122 * Expose a canary symbol of the previous ABI to ensure we catch uses of a
2123 * liblttng-ust.so.0 dlopen'd after .so.1 has been loaded. Use a different
2124 * symbol than the detection code to ensure we don't detect ourself.
2125 */
2126 void init_usterr(void);
2127 void init_usterr(void)
2128 {
2129 CRIT("Incompatible library ABIs detected within the same process. "
2130 "The process is likely linked against different major soname of LTTng-UST which is unsupported. "
2131 "The detection was triggered by canary symbol \"%s\"\n", __func__);
2132 }
2133
2134 /*
2135 * sessiond monitoring thread: monitor presence of global and per-user
2136 * sessiond by polling the application common named pipe.
2137 */
2138 static
2139 void lttng_ust_ctor(void)
2140 __attribute__((constructor));
2141 static
2142 void lttng_ust_ctor(void)
2143 {
2144 struct timespec constructor_timeout;
2145 sigset_t sig_all_blocked, orig_parent_mask;
2146 pthread_attr_t thread_attr;
2147 int timeout_mode;
2148 int ret;
2149 void *handle;
2150
2151 if (uatomic_xchg(&initialized, 1) == 1)
2152 return;
2153
2154 /*
2155 * Fixup interdependency between TLS allocation mutex (which happens
2156 * to be the dynamic linker mutex) and ust_lock, taken within
2157 * the ust lock.
2158 */
2159 lttng_ust_alloc_tls();
2160
2161 lttng_ust_loaded = 1;
2162
2163 /*
2164 * Check if we find a symbol of the previous ABI in the current process
2165 * as different ABIs of liblttng-ust can't co-exist in a process. If we
2166 * do so, emit a critical log message which will also abort if the
2167 * LTTNG_UST_ABORT_ON_CRITICAL environment variable is set.
2168 */
2169 lttng_ust_check_soname_0();
2170
2171 /*
2172 * We need to ensure that the liblttng-ust library is not unloaded to avoid
2173 * the unloading of code used by the ust_listener_threads as we can not
2174 * reliably know when they exited. To do that, manually load
2175 * liblttng-ust.so to increment the dynamic loader's internal refcount for
2176 * this library so it never becomes zero, thus never gets unloaded from the
2177 * address space of the process. Since we are already running in the
2178 * constructor of the LTTNG_UST_LIB_SONAME library, calling dlopen will
2179 * simply increment the refcount and no additionnal work is needed by the
2180 * dynamic loader as the shared library is already loaded in the address
2181 * space. As a safe guard, we use the RTLD_NODELETE flag to prevent
2182 * unloading of the UST library if its refcount becomes zero (which should
2183 * never happen). Do the return value check but discard the handle at the
2184 * end of the function as it's not needed.
2185 */
2186 handle = dlopen(LTTNG_UST_LIB_SONAME, RTLD_LAZY | RTLD_NODELETE);
2187 if (!handle) {
2188 ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SONAME);
2189 } else {
2190 DBG("dlopened liblttng-ust shared library (%s).", LTTNG_UST_LIB_SONAME);
2191 }
2192
2193 /*
2194 * We want precise control over the order in which we construct
2195 * our sub-libraries vs starting to receive commands from
2196 * sessiond (otherwise leading to errors when trying to create
2197 * sessiond before the init functions are completed).
2198 */
2199
2200 /*
2201 * Both the logging and getenv lazy-initialization uses getenv()
2202 * internally and thus needs to be explicitly initialized in
2203 * liblttng-ust before we start any threads as an unsuspecting normally
2204 * single threaded application using liblttng-ust could be using
2205 * setenv() which is not thread-safe.
2206 */
2207 lttng_ust_logging_init();
2208 lttng_ust_getenv_init();
2209
2210 /* Call the liblttng-ust-common constructor. */
2211 lttng_ust_common_ctor();
2212
2213 lttng_ust_tp_init();
2214 lttng_ust_statedump_init();
2215 lttng_ust_ring_buffer_clients_init();
2216 lttng_ust_counter_clients_init();
2217 lttng_perf_counter_init();
2218 /*
2219 * Invoke ust malloc wrapper init before starting other threads.
2220 */
2221 lttng_ust_libc_wrapper_malloc_ctor();
2222
2223 timeout_mode = get_constructor_timeout(&constructor_timeout);
2224
2225 get_allow_blocking();
2226
2227 ret = sem_init(&constructor_wait, 0, 0);
2228 if (ret) {
2229 PERROR("sem_init");
2230 }
2231
2232 ret = setup_global_apps();
2233 if (ret) {
2234 assert(global_apps.allowed == 0);
2235 DBG("global apps setup returned %d", ret);
2236 }
2237
2238 ret = setup_local_apps();
2239 if (ret) {
2240 assert(local_apps.allowed == 0);
2241 DBG("local apps setup returned %d", ret);
2242 }
2243
2244 /* A new thread created by pthread_create inherits the signal mask
2245 * from the parent. To avoid any signal being received by the
2246 * listener thread, we block all signals temporarily in the parent,
2247 * while we create the listener thread.
2248 */
2249 sigfillset(&sig_all_blocked);
2250 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
2251 if (ret) {
2252 ERR("pthread_sigmask: %s", strerror(ret));
2253 }
2254
2255 ret = pthread_attr_init(&thread_attr);
2256 if (ret) {
2257 ERR("pthread_attr_init: %s", strerror(ret));
2258 }
2259 ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
2260 if (ret) {
2261 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
2262 }
2263
2264 if (global_apps.allowed) {
2265 pthread_mutex_lock(&ust_exit_mutex);
2266 ret = pthread_create(&global_apps.ust_listener, &thread_attr,
2267 ust_listener_thread, &global_apps);
2268 if (ret) {
2269 ERR("pthread_create global: %s", strerror(ret));
2270 }
2271 global_apps.thread_active = 1;
2272 pthread_mutex_unlock(&ust_exit_mutex);
2273 } else {
2274 handle_register_done(&global_apps);
2275 }
2276
2277 if (local_apps.allowed) {
2278 pthread_mutex_lock(&ust_exit_mutex);
2279 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
2280 ust_listener_thread, &local_apps);
2281 if (ret) {
2282 ERR("pthread_create local: %s", strerror(ret));
2283 }
2284 local_apps.thread_active = 1;
2285 pthread_mutex_unlock(&ust_exit_mutex);
2286 } else {
2287 handle_register_done(&local_apps);
2288 }
2289 ret = pthread_attr_destroy(&thread_attr);
2290 if (ret) {
2291 ERR("pthread_attr_destroy: %s", strerror(ret));
2292 }
2293
2294 /* Restore original signal mask in parent */
2295 ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
2296 if (ret) {
2297 ERR("pthread_sigmask: %s", strerror(ret));
2298 }
2299
2300 switch (timeout_mode) {
2301 case 1: /* timeout wait */
2302 do {
2303 ret = sem_timedwait(&constructor_wait,
2304 &constructor_timeout);
2305 } while (ret < 0 && errno == EINTR);
2306 if (ret < 0) {
2307 switch (errno) {
2308 case ETIMEDOUT:
2309 ERR("Timed out waiting for lttng-sessiond");
2310 break;
2311 case EINVAL:
2312 PERROR("sem_timedwait");
2313 break;
2314 default:
2315 ERR("Unexpected error \"%s\" returned by sem_timedwait",
2316 strerror(errno));
2317 }
2318 }
2319 break;
2320 case -1:/* wait forever */
2321 do {
2322 ret = sem_wait(&constructor_wait);
2323 } while (ret < 0 && errno == EINTR);
2324 if (ret < 0) {
2325 switch (errno) {
2326 case EINVAL:
2327 PERROR("sem_wait");
2328 break;
2329 default:
2330 ERR("Unexpected error \"%s\" returned by sem_wait",
2331 strerror(errno));
2332 }
2333 }
2334 break;
2335 case 0: /* no timeout */
2336 break;
2337 }
2338 }
2339
2340 static
2341 void lttng_ust_cleanup(int exiting)
2342 {
2343 cleanup_sock_info(&global_apps, exiting);
2344 cleanup_sock_info(&local_apps, exiting);
2345 local_apps.allowed = 0;
2346 global_apps.allowed = 0;
2347 /*
2348 * The teardown in this function all affect data structures
2349 * accessed under the UST lock by the listener thread. This
2350 * lock, along with the lttng_ust_comm_should_quit flag, ensure
2351 * that none of these threads are accessing this data at this
2352 * point.
2353 */
2354 lttng_ust_abi_exit();
2355 lttng_ust_abi_events_exit();
2356 lttng_perf_counter_exit();
2357 lttng_ust_ring_buffer_clients_exit();
2358 lttng_ust_counter_clients_exit();
2359 lttng_ust_statedump_destroy();
2360 lttng_ust_tp_exit();
2361 if (!exiting) {
2362 /* Reinitialize values for fork */
2363 sem_count = sem_count_initial_value;
2364 lttng_ust_comm_should_quit = 0;
2365 initialized = 0;
2366 }
2367 }
2368
2369 static
2370 void lttng_ust_exit(void)
2371 __attribute__((destructor));
2372 static
2373 void lttng_ust_exit(void)
2374 {
2375 int ret;
2376
2377 /*
2378 * Using pthread_cancel here because:
2379 * A) we don't want to hang application teardown.
2380 * B) the thread is not allocating any resource.
2381 */
2382
2383 /*
2384 * Require the communication thread to quit. Synchronize with
2385 * mutexes to ensure it is not in a mutex critical section when
2386 * pthread_cancel is later called.
2387 */
2388 ust_lock_nocheck();
2389 lttng_ust_comm_should_quit = 1;
2390 ust_unlock();
2391
2392 pthread_mutex_lock(&ust_exit_mutex);
2393 /* cancel threads */
2394 if (global_apps.thread_active) {
2395 ret = pthread_cancel(global_apps.ust_listener);
2396 if (ret) {
2397 ERR("Error cancelling global ust listener thread: %s",
2398 strerror(ret));
2399 } else {
2400 global_apps.thread_active = 0;
2401 }
2402 }
2403 if (local_apps.thread_active) {
2404 ret = pthread_cancel(local_apps.ust_listener);
2405 if (ret) {
2406 ERR("Error cancelling local ust listener thread: %s",
2407 strerror(ret));
2408 } else {
2409 local_apps.thread_active = 0;
2410 }
2411 }
2412 pthread_mutex_unlock(&ust_exit_mutex);
2413
2414 /*
2415 * Do NOT join threads: use of sys_futex makes it impossible to
2416 * join the threads without using async-cancel, but async-cancel
2417 * is delivered by a signal, which could hit the target thread
2418 * anywhere in its code path, including while the ust_lock() is
2419 * held, causing a deadlock for the other thread. Let the OS
2420 * cleanup the threads if there are stalled in a syscall.
2421 */
2422 lttng_ust_cleanup(1);
2423 }
2424
2425 static
2426 void ust_context_ns_reset(void)
2427 {
2428 lttng_context_pid_ns_reset();
2429 lttng_context_cgroup_ns_reset();
2430 lttng_context_ipc_ns_reset();
2431 lttng_context_mnt_ns_reset();
2432 lttng_context_net_ns_reset();
2433 lttng_context_user_ns_reset();
2434 lttng_context_time_ns_reset();
2435 lttng_context_uts_ns_reset();
2436 }
2437
2438 static
2439 void ust_context_vuids_reset(void)
2440 {
2441 lttng_context_vuid_reset();
2442 lttng_context_veuid_reset();
2443 lttng_context_vsuid_reset();
2444 }
2445
2446 static
2447 void ust_context_vgids_reset(void)
2448 {
2449 lttng_context_vgid_reset();
2450 lttng_context_vegid_reset();
2451 lttng_context_vsgid_reset();
2452 }
2453
2454 /*
2455 * We exclude the worker threads across fork and clone (except
2456 * CLONE_VM), because these system calls only keep the forking thread
2457 * running in the child. Therefore, we don't want to call fork or clone
2458 * in the middle of an tracepoint or ust tracing state modification.
2459 * Holding this mutex protects these structures across fork and clone.
2460 */
2461 void lttng_ust_before_fork(sigset_t *save_sigset)
2462 {
2463 /*
2464 * Disable signals. This is to avoid that the child intervenes
2465 * before it is properly setup for tracing. It is safer to
2466 * disable all signals, because then we know we are not breaking
2467 * anything by restoring the original mask.
2468 */
2469 sigset_t all_sigs;
2470 int ret;
2471
2472 /* Allocate lttng-ust TLS. */
2473 lttng_ust_alloc_tls();
2474
2475 if (URCU_TLS(lttng_ust_nest_count))
2476 return;
2477 /* Disable signals */
2478 sigfillset(&all_sigs);
2479 ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
2480 if (ret == -1) {
2481 PERROR("sigprocmask");
2482 }
2483
2484 pthread_mutex_lock(&ust_fork_mutex);
2485
2486 ust_lock_nocheck();
2487 lttng_ust_urcu_before_fork();
2488 lttng_ust_lock_fd_tracker();
2489 lttng_perf_lock();
2490 }
2491
2492 static void ust_after_fork_common(sigset_t *restore_sigset)
2493 {
2494 int ret;
2495
2496 DBG("process %d", getpid());
2497 lttng_perf_unlock();
2498 lttng_ust_unlock_fd_tracker();
2499 ust_unlock();
2500
2501 pthread_mutex_unlock(&ust_fork_mutex);
2502
2503 /* Restore signals */
2504 ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
2505 if (ret == -1) {
2506 PERROR("sigprocmask");
2507 }
2508 }
2509
2510 void lttng_ust_after_fork_parent(sigset_t *restore_sigset)
2511 {
2512 if (URCU_TLS(lttng_ust_nest_count))
2513 return;
2514 DBG("process %d", getpid());
2515 lttng_ust_urcu_after_fork_parent();
2516 /* Release mutexes and reenable signals */
2517 ust_after_fork_common(restore_sigset);
2518 }
2519
2520 /*
2521 * After fork, in the child, we need to cleanup all the leftover state,
2522 * except the worker thread which already magically disappeared thanks
2523 * to the weird Linux fork semantics. After tyding up, we call
2524 * lttng_ust_ctor() again to start over as a new PID.
2525 *
2526 * This is meant for forks() that have tracing in the child between the
2527 * fork and following exec call (if there is any).
2528 */
2529 void lttng_ust_after_fork_child(sigset_t *restore_sigset)
2530 {
2531 if (URCU_TLS(lttng_ust_nest_count))
2532 return;
2533 lttng_context_vpid_reset();
2534 lttng_context_vtid_reset();
2535 lttng_ust_context_procname_reset();
2536 ust_context_ns_reset();
2537 ust_context_vuids_reset();
2538 ust_context_vgids_reset();
2539 DBG("process %d", getpid());
2540 /* Release urcu mutexes */
2541 lttng_ust_urcu_after_fork_child();
2542 lttng_ust_cleanup(0);
2543 /* Release mutexes and reenable signals */
2544 ust_after_fork_common(restore_sigset);
2545 lttng_ust_ctor();
2546 }
2547
2548 void lttng_ust_after_setns(void)
2549 {
2550 ust_context_ns_reset();
2551 ust_context_vuids_reset();
2552 ust_context_vgids_reset();
2553 }
2554
2555 void lttng_ust_after_unshare(void)
2556 {
2557 ust_context_ns_reset();
2558 ust_context_vuids_reset();
2559 ust_context_vgids_reset();
2560 }
2561
2562 void lttng_ust_after_setuid(void)
2563 {
2564 ust_context_vuids_reset();
2565 }
2566
2567 void lttng_ust_after_seteuid(void)
2568 {
2569 ust_context_vuids_reset();
2570 }
2571
2572 void lttng_ust_after_setreuid(void)
2573 {
2574 ust_context_vuids_reset();
2575 }
2576
2577 void lttng_ust_after_setresuid(void)
2578 {
2579 ust_context_vuids_reset();
2580 }
2581
2582 void lttng_ust_after_setgid(void)
2583 {
2584 ust_context_vgids_reset();
2585 }
2586
2587 void lttng_ust_after_setegid(void)
2588 {
2589 ust_context_vgids_reset();
2590 }
2591
2592 void lttng_ust_after_setregid(void)
2593 {
2594 ust_context_vgids_reset();
2595 }
2596
2597 void lttng_ust_after_setresgid(void)
2598 {
2599 ust_context_vgids_reset();
2600 }
2601
2602 void lttng_ust_sockinfo_session_enabled(void *owner)
2603 {
2604 struct sock_info *sock_info = owner;
2605 sock_info->statedump_pending = 1;
2606 }
This page took 0.087828 seconds and 3 git commands to generate.