0a039feffdb7f8d64085da19118b0fea3a886e42
[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 EfficiOS Inc.
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 <lttng/ust-cancelstate.h>
42 #include <urcu/tls-compat.h>
43 #include "lib/lttng-ust/futex.h"
44 #include "common/ustcomm.h"
45 #include "common/ust-fd.h"
46 #include "common/logging.h"
47 #include "common/macros.h"
48 #include "common/tracepoint.h"
49 #include "lttng-tracer-core.h"
50 #include "common/compat/pthread.h"
51 #include "common/procname.h"
52 #include "common/ringbuffer/rb-init.h"
53 #include "lttng-ust-statedump.h"
54 #include "common/clock.h"
55 #include "common/getenv.h"
56 #include "lib/lttng-ust/events.h"
57 #include "context-internal.h"
58 #include "common/align.h"
59 #include "common/counter-clients/clients.h"
60 #include "common/ringbuffer-clients/clients.h"
61
62 /*
63 * Has lttng ust comm constructor been called ?
64 */
65 static int initialized;
66
67 /*
68 * The ust_lock/ust_unlock lock is used as a communication thread mutex.
69 * Held when handling a command, also held by fork() to deal with
70 * removal of threads, and by exit path.
71 *
72 * The UST lock is the centralized mutex across UST tracing control and
73 * probe registration.
74 *
75 * ust_exit_mutex must never nest in ust_mutex.
76 *
77 * ust_fork_mutex must never nest in ust_mutex.
78 *
79 * ust_mutex_nest is a per-thread nesting counter, allowing the perf
80 * counter lazy initialization called by events within the statedump,
81 * which traces while the ust_mutex is held.
82 *
83 * ust_lock nests within the dynamic loader lock (within glibc) because
84 * it is taken within the library constructor.
85 *
86 * The ust fd tracker lock nests within the ust_mutex.
87 */
88 static pthread_mutex_t ust_mutex = PTHREAD_MUTEX_INITIALIZER;
89
90 /* Allow nesting the ust_mutex within the same thread. */
91 static DEFINE_URCU_TLS(int, ust_mutex_nest);
92
93 /*
94 * ust_exit_mutex protects thread_active variable wrt thread exit. It
95 * cannot be done by ust_mutex because pthread_cancel(), which takes an
96 * internal libc lock, cannot nest within ust_mutex.
97 *
98 * It never nests within a ust_mutex.
99 */
100 static pthread_mutex_t ust_exit_mutex = PTHREAD_MUTEX_INITIALIZER;
101
102 /*
103 * ust_fork_mutex protects base address statedump tracing against forks. It
104 * prevents the dynamic loader lock to be taken (by base address statedump
105 * tracing) while a fork is happening, thus preventing deadlock issues with
106 * the dynamic loader lock.
107 */
108 static pthread_mutex_t ust_fork_mutex = PTHREAD_MUTEX_INITIALIZER;
109
110 /* Should the ust comm thread quit ? */
111 static int lttng_ust_comm_should_quit;
112
113 /*
114 * This variable can be tested by applications to check whether
115 * lttng-ust is loaded. They simply have to define their own
116 * "lttng_ust_loaded" weak symbol, and test it. It is set to 1 by the
117 * library constructor.
118 */
119 int lttng_ust_loaded __attribute__((weak));
120
121 /*
122 * Notes on async-signal-safety of ust lock: a few libc functions are used
123 * which are not strictly async-signal-safe:
124 *
125 * - pthread_setcancelstate
126 * - pthread_mutex_lock
127 * - pthread_mutex_unlock
128 *
129 * As of glibc 2.35, the implementation of pthread_setcancelstate only
130 * touches TLS data, and it appears to be safe to use from signal
131 * handlers. If the libc implementation changes, this will need to be
132 * revisited, and we may ask glibc to provide an async-signal-safe
133 * pthread_setcancelstate.
134 *
135 * As of glibc 2.35, the implementation of pthread_mutex_lock/unlock
136 * for fast mutexes only relies on the pthread_mutex_t structure.
137 * Disabling signals around all uses of this mutex ensures
138 * signal-safety. If the libc implementation changes and eventually uses
139 * other global resources, this will need to be revisited and we may
140 * need to implement our own mutex.
141 */
142
143 /*
144 * Return 0 on success, -1 if should quit.
145 * The lock is taken in both cases.
146 * Signal-safe.
147 */
148 int ust_lock(void)
149 {
150 sigset_t sig_all_blocked, orig_mask;
151 int ret;
152
153 if (lttng_ust_cancelstate_disable_push()) {
154 ERR("lttng_ust_cancelstate_disable_push");
155 }
156 sigfillset(&sig_all_blocked);
157 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
158 if (ret) {
159 ERR("pthread_sigmask: ret=%d", ret);
160 }
161 if (!URCU_TLS(ust_mutex_nest)++)
162 pthread_mutex_lock(&ust_mutex);
163 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
164 if (ret) {
165 ERR("pthread_sigmask: ret=%d", ret);
166 }
167 if (lttng_ust_comm_should_quit) {
168 return -1;
169 } else {
170 return 0;
171 }
172 }
173
174 /*
175 * ust_lock_nocheck() can be used in constructors/destructors, because
176 * they are already nested within the dynamic loader lock, and therefore
177 * have exclusive access against execution of liblttng-ust destructor.
178 * Signal-safe.
179 */
180 void ust_lock_nocheck(void)
181 {
182 sigset_t sig_all_blocked, orig_mask;
183 int ret;
184
185 if (lttng_ust_cancelstate_disable_push()) {
186 ERR("lttng_ust_cancelstate_disable_push");
187 }
188 sigfillset(&sig_all_blocked);
189 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
190 if (ret) {
191 ERR("pthread_sigmask: ret=%d", ret);
192 }
193 if (!URCU_TLS(ust_mutex_nest)++)
194 pthread_mutex_lock(&ust_mutex);
195 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
196 if (ret) {
197 ERR("pthread_sigmask: ret=%d", ret);
198 }
199 }
200
201 /*
202 * Signal-safe.
203 */
204 void ust_unlock(void)
205 {
206 sigset_t sig_all_blocked, orig_mask;
207 int ret;
208
209 sigfillset(&sig_all_blocked);
210 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
211 if (ret) {
212 ERR("pthread_sigmask: ret=%d", ret);
213 }
214 if (!--URCU_TLS(ust_mutex_nest))
215 pthread_mutex_unlock(&ust_mutex);
216 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
217 if (ret) {
218 ERR("pthread_sigmask: ret=%d", ret);
219 }
220 if (lttng_ust_cancelstate_disable_pop()) {
221 ERR("lttng_ust_cancelstate_disable_pop");
222 }
223 }
224
225 /*
226 * Wait for either of these before continuing to the main
227 * program:
228 * - the register_done message from sessiond daemon
229 * (will let the sessiond daemon enable sessions before main
230 * starts.)
231 * - sessiond daemon is not reachable.
232 * - timeout (ensuring applications are resilient to session
233 * daemon problems).
234 */
235 static sem_t constructor_wait;
236 /*
237 * Doing this for both the global and local sessiond.
238 */
239 enum {
240 sem_count_initial_value = 4,
241 };
242
243 static int sem_count = sem_count_initial_value;
244
245 /*
246 * Counting nesting within lttng-ust. Used to ensure that calling fork()
247 * from liblttng-ust does not execute the pre/post fork handlers.
248 */
249 static DEFINE_URCU_TLS(int, lttng_ust_nest_count);
250
251 /*
252 * Info about socket and associated listener thread.
253 */
254 struct sock_info {
255 const char *name;
256 pthread_t ust_listener; /* listener thread */
257 int root_handle;
258 int registration_done;
259 int allowed;
260 int global;
261 int thread_active;
262
263 char sock_path[PATH_MAX];
264 int socket;
265 int notify_socket;
266
267 char wait_shm_path[PATH_MAX];
268 char *wait_shm_mmap;
269 /* Keep track of lazy state dump not performed yet. */
270 int statedump_pending;
271 int initial_statedump_done;
272 /* Keep procname for statedump */
273 char procname[LTTNG_UST_CONTEXT_PROCNAME_LEN];
274 };
275
276 /* Socket from app (connect) to session daemon (listen) for communication */
277 static struct sock_info global_apps = {
278 .name = "global",
279 .global = 1,
280
281 .root_handle = -1,
282 .registration_done = 0,
283 .allowed = 0,
284 .thread_active = 0,
285
286 .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME,
287 .socket = -1,
288 .notify_socket = -1,
289
290 .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME,
291
292 .statedump_pending = 0,
293 .initial_statedump_done = 0,
294 .procname[0] = '\0'
295 };
296
297 /* TODO: allow global_apps_sock_path override */
298
299 static struct sock_info local_apps = {
300 .name = "local",
301 .global = 0,
302 .root_handle = -1,
303 .registration_done = 0,
304 .allowed = 0, /* Check setuid bit first */
305 .thread_active = 0,
306
307 .socket = -1,
308 .notify_socket = -1,
309
310 .statedump_pending = 0,
311 .initial_statedump_done = 0,
312 .procname[0] = '\0'
313 };
314
315 static int wait_poll_fallback;
316
317 static const char *cmd_name_mapping[] = {
318 [ LTTNG_UST_ABI_RELEASE ] = "Release",
319 [ LTTNG_UST_ABI_SESSION ] = "Create Session",
320 [ LTTNG_UST_ABI_TRACER_VERSION ] = "Get Tracer Version",
321
322 [ LTTNG_UST_ABI_TRACEPOINT_LIST ] = "Create Tracepoint List",
323 [ LTTNG_UST_ABI_WAIT_QUIESCENT ] = "Wait for Quiescent State",
324 [ LTTNG_UST_ABI_REGISTER_DONE ] = "Registration Done",
325 [ LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST ] = "Create Tracepoint Field List",
326
327 [ LTTNG_UST_ABI_EVENT_NOTIFIER_GROUP_CREATE ] = "Create event notifier group",
328
329 /* Session FD commands */
330 [ LTTNG_UST_ABI_CHANNEL ] = "Create Channel",
331 [ LTTNG_UST_ABI_SESSION_START ] = "Start Session",
332 [ LTTNG_UST_ABI_SESSION_STOP ] = "Stop Session",
333
334 /* Channel FD commands */
335 [ LTTNG_UST_ABI_STREAM ] = "Create Stream",
336 [ LTTNG_UST_ABI_EVENT ] = "Create Event",
337
338 /* Event and Channel FD commands */
339 [ LTTNG_UST_ABI_CONTEXT ] = "Create Context",
340 [ LTTNG_UST_ABI_FLUSH_BUFFER ] = "Flush Buffer",
341
342 /* Event, Channel and Session commands */
343 [ LTTNG_UST_ABI_ENABLE ] = "Enable",
344 [ LTTNG_UST_ABI_DISABLE ] = "Disable",
345
346 /* Tracepoint list commands */
347 [ LTTNG_UST_ABI_TRACEPOINT_LIST_GET ] = "List Next Tracepoint",
348 [ LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST_GET ] = "List Next Tracepoint Field",
349
350 /* Event FD commands */
351 [ LTTNG_UST_ABI_FILTER ] = "Create Filter",
352 [ LTTNG_UST_ABI_EXCLUSION ] = "Add exclusions to event",
353
354 /* Event notifier group commands */
355 [ LTTNG_UST_ABI_EVENT_NOTIFIER_CREATE ] = "Create event notifier",
356
357 /* Session and event notifier group commands */
358 [ LTTNG_UST_ABI_COUNTER ] = "Create Counter",
359
360 /* Counter commands */
361 [ LTTNG_UST_ABI_COUNTER_GLOBAL ] = "Create Counter Global",
362 [ LTTNG_UST_ABI_COUNTER_CPU ] = "Create Counter CPU",
363 };
364
365 static const char *str_timeout;
366 static int got_timeout_env;
367
368 static char *get_map_shm(struct sock_info *sock_info);
369
370 /*
371 * Returns the HOME directory path. Caller MUST NOT free(3) the returned
372 * pointer.
373 */
374 static
375 const char *get_lttng_home_dir(void)
376 {
377 const char *val;
378
379 val = (const char *) lttng_ust_getenv("LTTNG_HOME");
380 if (val != NULL) {
381 return val;
382 }
383 return (const char *) lttng_ust_getenv("HOME");
384 }
385
386 /*
387 * Force a read (imply TLS allocation for dlopen) of TLS variables.
388 */
389 static
390 void lttng_ust_nest_count_alloc_tls(void)
391 {
392 asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count)));
393 }
394
395 static
396 void lttng_ust_mutex_nest_alloc_tls(void)
397 {
398 asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest)));
399 }
400
401 /*
402 * Allocate lttng-ust urcu TLS.
403 */
404 static
405 void lttng_ust_urcu_alloc_tls(void)
406 {
407 (void) lttng_ust_urcu_read_ongoing();
408 }
409
410 void lttng_ust_common_init_thread(int flags)
411 {
412 lttng_ust_urcu_alloc_tls();
413 lttng_ringbuffer_alloc_tls();
414 lttng_ust_vtid_init_thread(flags);
415 lttng_ust_nest_count_alloc_tls();
416 lttng_ust_procname_init_thread(flags);
417 lttng_ust_mutex_nest_alloc_tls();
418 lttng_ust_perf_counter_init_thread(flags);
419 lttng_ust_common_alloc_tls();
420 lttng_ust_cgroup_ns_init_thread(flags);
421 lttng_ust_ipc_ns_init_thread(flags);
422 lttng_ust_net_ns_init_thread(flags);
423 lttng_ust_time_ns_init_thread(flags);
424 lttng_ust_uts_ns_init_thread(flags);
425 lttng_ust_ring_buffer_client_discard_alloc_tls();
426 lttng_ust_ring_buffer_client_discard_rt_alloc_tls();
427 lttng_ust_ring_buffer_client_overwrite_alloc_tls();
428 lttng_ust_ring_buffer_client_overwrite_rt_alloc_tls();
429 }
430
431 /*
432 * LTTng-UST uses Global Dynamic model TLS variables rather than IE
433 * model because many versions of glibc don't preallocate a pool large
434 * enough for TLS variables IE model defined in other shared libraries,
435 * and causes issues when using LTTng-UST for Java tracing.
436 *
437 * Because of this use of Global Dynamic TLS variables, users wishing to
438 * trace from signal handlers need to explicitly trigger the lazy
439 * allocation of those variables for each thread before using them.
440 * This can be triggered by calling lttng_ust_init_thread().
441 */
442 void lttng_ust_init_thread(void)
443 {
444 /*
445 * Because those TLS variables are global dynamic, we need to
446 * ensure those are initialized before a signal handler nesting over
447 * this thread attempts to use them.
448 */
449 lttng_ust_common_init_thread(LTTNG_UST_INIT_THREAD_MASK);
450 }
451
452 int lttng_get_notify_socket(void *owner)
453 {
454 struct sock_info *info = owner;
455
456 return info->notify_socket;
457 }
458
459
460 char* lttng_ust_sockinfo_get_procname(void *owner)
461 {
462 struct sock_info *info = owner;
463
464 return info->procname;
465 }
466
467 static
468 void print_cmd(int cmd, int handle)
469 {
470 const char *cmd_name = "Unknown";
471
472 if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping)
473 && cmd_name_mapping[cmd]) {
474 cmd_name = cmd_name_mapping[cmd];
475 }
476 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
477 cmd_name, cmd,
478 lttng_ust_obj_get_name(handle), handle);
479 }
480
481 static
482 int setup_global_apps(void)
483 {
484 int ret = 0;
485 assert(!global_apps.wait_shm_mmap);
486
487 global_apps.wait_shm_mmap = get_map_shm(&global_apps);
488 if (!global_apps.wait_shm_mmap) {
489 WARN("Unable to get map shm for global apps. Disabling LTTng-UST global tracing.");
490 global_apps.allowed = 0;
491 ret = -EIO;
492 goto error;
493 }
494
495 global_apps.allowed = 1;
496 lttng_pthread_getname_np(global_apps.procname, LTTNG_UST_CONTEXT_PROCNAME_LEN);
497 error:
498 return ret;
499 }
500 static
501 int setup_local_apps(void)
502 {
503 int ret = 0;
504 const char *home_dir;
505 uid_t uid;
506
507 assert(!local_apps.wait_shm_mmap);
508
509 uid = getuid();
510 /*
511 * Disallow per-user tracing for setuid binaries.
512 */
513 if (uid != geteuid()) {
514 assert(local_apps.allowed == 0);
515 ret = 0;
516 goto end;
517 }
518 home_dir = get_lttng_home_dir();
519 if (!home_dir) {
520 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
521 assert(local_apps.allowed == 0);
522 ret = -ENOENT;
523 goto end;
524 }
525 local_apps.allowed = 1;
526 snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
527 home_dir,
528 LTTNG_DEFAULT_HOME_RUNDIR,
529 LTTNG_UST_SOCK_FILENAME);
530 snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
531 LTTNG_UST_WAIT_FILENAME,
532 uid);
533
534 local_apps.wait_shm_mmap = get_map_shm(&local_apps);
535 if (!local_apps.wait_shm_mmap) {
536 WARN("Unable to get map shm for local apps. Disabling LTTng-UST per-user tracing.");
537 local_apps.allowed = 0;
538 ret = -EIO;
539 goto end;
540 }
541
542 lttng_pthread_getname_np(local_apps.procname, LTTNG_UST_CONTEXT_PROCNAME_LEN);
543 end:
544 return ret;
545 }
546
547 /*
548 * Get socket timeout, in ms.
549 * -1: wait forever. 0: don't wait. >0: timeout, in ms.
550 */
551 static
552 long get_timeout(void)
553 {
554 long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
555
556 if (!got_timeout_env) {
557 str_timeout = lttng_ust_getenv("LTTNG_UST_REGISTER_TIMEOUT");
558 got_timeout_env = 1;
559 }
560 if (str_timeout)
561 constructor_delay_ms = strtol(str_timeout, NULL, 10);
562 /* All negative values are considered as "-1". */
563 if (constructor_delay_ms < -1)
564 constructor_delay_ms = -1;
565 return constructor_delay_ms;
566 }
567
568 /* Timeout for notify socket send and recv. */
569 static
570 long get_notify_sock_timeout(void)
571 {
572 return get_timeout();
573 }
574
575 /* Timeout for connecting to cmd and notify sockets. */
576 static
577 long get_connect_sock_timeout(void)
578 {
579 return get_timeout();
580 }
581
582 /*
583 * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
584 */
585 static
586 int get_constructor_timeout(struct timespec *constructor_timeout)
587 {
588 long constructor_delay_ms;
589 int ret;
590
591 constructor_delay_ms = get_timeout();
592
593 switch (constructor_delay_ms) {
594 case -1:/* fall-through */
595 case 0:
596 return constructor_delay_ms;
597 default:
598 break;
599 }
600
601 /*
602 * If we are unable to find the current time, don't wait.
603 */
604 ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
605 if (ret) {
606 /* Don't wait. */
607 return 0;
608 }
609 constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
610 constructor_timeout->tv_nsec +=
611 (constructor_delay_ms % 1000UL) * 1000000UL;
612 if (constructor_timeout->tv_nsec >= 1000000000UL) {
613 constructor_timeout->tv_sec++;
614 constructor_timeout->tv_nsec -= 1000000000UL;
615 }
616 /* Timeout wait (constructor_delay_ms). */
617 return 1;
618 }
619
620 static
621 void get_allow_blocking(void)
622 {
623 const char *str_allow_blocking =
624 lttng_ust_getenv("LTTNG_UST_ALLOW_BLOCKING");
625
626 if (str_allow_blocking) {
627 DBG("%s environment variable is set",
628 "LTTNG_UST_ALLOW_BLOCKING");
629 lttng_ust_ringbuffer_set_allow_blocking();
630 }
631 }
632
633 static
634 int register_to_sessiond(int socket, enum lttng_ust_ctl_socket_type type,
635 const char *procname)
636 {
637 return ustcomm_send_reg_msg(socket,
638 type,
639 CAA_BITS_PER_LONG,
640 lttng_ust_rb_alignof(uint8_t) * CHAR_BIT,
641 lttng_ust_rb_alignof(uint16_t) * CHAR_BIT,
642 lttng_ust_rb_alignof(uint32_t) * CHAR_BIT,
643 lttng_ust_rb_alignof(uint64_t) * CHAR_BIT,
644 lttng_ust_rb_alignof(unsigned long) * CHAR_BIT,
645 procname);
646 }
647
648 static
649 int send_reply(int sock, struct ustcomm_ust_reply *lur)
650 {
651 ssize_t len;
652
653 len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
654 switch (len) {
655 case sizeof(*lur):
656 DBG("message successfully sent");
657 return 0;
658 default:
659 if (len == -ECONNRESET) {
660 DBG("remote end closed connection");
661 return 0;
662 }
663 if (len < 0)
664 return len;
665 DBG("incorrect message size: %zd", len);
666 return -EINVAL;
667 }
668 }
669
670 static
671 void decrement_sem_count(unsigned int count)
672 {
673 int ret;
674
675 assert(uatomic_read(&sem_count) >= count);
676
677 if (uatomic_read(&sem_count) <= 0) {
678 return;
679 }
680
681 ret = uatomic_add_return(&sem_count, -count);
682 if (ret == 0) {
683 ret = sem_post(&constructor_wait);
684 assert(!ret);
685 }
686 }
687
688 static
689 int handle_register_done(struct sock_info *sock_info)
690 {
691 if (sock_info->registration_done)
692 return 0;
693 sock_info->registration_done = 1;
694
695 decrement_sem_count(1);
696 if (!sock_info->statedump_pending) {
697 sock_info->initial_statedump_done = 1;
698 decrement_sem_count(1);
699 }
700
701 return 0;
702 }
703
704 static
705 int handle_register_failed(struct sock_info *sock_info)
706 {
707 if (sock_info->registration_done)
708 return 0;
709 sock_info->registration_done = 1;
710 sock_info->initial_statedump_done = 1;
711
712 decrement_sem_count(2);
713
714 return 0;
715 }
716
717 /*
718 * Only execute pending statedump after the constructor semaphore has
719 * been posted by the current listener thread. This means statedump will
720 * only be performed after the "registration done" command is received
721 * from this thread's session daemon.
722 *
723 * This ensures we don't run into deadlock issues with the dynamic
724 * loader mutex, which is held while the constructor is called and
725 * waiting on the constructor semaphore. All operations requiring this
726 * dynamic loader lock need to be postponed using this mechanism.
727 *
728 * In a scenario with two session daemons connected to the application,
729 * it is possible that the first listener thread which receives the
730 * registration done command issues its statedump while the dynamic
731 * loader lock is still held by the application constructor waiting on
732 * the semaphore. It will however be allowed to proceed when the
733 * second session daemon sends the registration done command to the
734 * second listener thread. This situation therefore does not produce
735 * a deadlock.
736 */
737 static
738 void handle_pending_statedump(struct sock_info *sock_info)
739 {
740 if (sock_info->registration_done && sock_info->statedump_pending) {
741 sock_info->statedump_pending = 0;
742 pthread_mutex_lock(&ust_fork_mutex);
743 lttng_handle_pending_statedump(sock_info);
744 pthread_mutex_unlock(&ust_fork_mutex);
745
746 if (!sock_info->initial_statedump_done) {
747 sock_info->initial_statedump_done = 1;
748 decrement_sem_count(1);
749 }
750 }
751 }
752
753 static inline
754 const char *bytecode_type_str(uint32_t cmd)
755 {
756 switch (cmd) {
757 case LTTNG_UST_ABI_CAPTURE:
758 return "capture";
759 case LTTNG_UST_ABI_FILTER:
760 return "filter";
761 default:
762 abort();
763 }
764 }
765
766 static
767 int handle_bytecode_recv(struct sock_info *sock_info,
768 int sock, struct ustcomm_ust_msg *lum)
769 {
770 struct lttng_ust_bytecode_node *bytecode = NULL;
771 enum lttng_ust_bytecode_type type;
772 const struct lttng_ust_abi_objd_ops *ops;
773 uint32_t data_size, data_size_max, reloc_offset;
774 uint64_t seqnum;
775 ssize_t len;
776 int ret = 0;
777
778 switch (lum->cmd) {
779 case LTTNG_UST_ABI_FILTER:
780 type = LTTNG_UST_BYTECODE_TYPE_FILTER;
781 data_size = lum->u.filter.data_size;
782 data_size_max = LTTNG_UST_ABI_FILTER_BYTECODE_MAX_LEN;
783 reloc_offset = lum->u.filter.reloc_offset;
784 seqnum = lum->u.filter.seqnum;
785 break;
786 case LTTNG_UST_ABI_CAPTURE:
787 type = LTTNG_UST_BYTECODE_TYPE_CAPTURE;
788 data_size = lum->u.capture.data_size;
789 data_size_max = LTTNG_UST_ABI_CAPTURE_BYTECODE_MAX_LEN;
790 reloc_offset = lum->u.capture.reloc_offset;
791 seqnum = lum->u.capture.seqnum;
792 break;
793 default:
794 abort();
795 }
796
797 if (data_size > data_size_max) {
798 ERR("Bytecode %s data size is too large: %u bytes",
799 bytecode_type_str(lum->cmd), data_size);
800 ret = -EINVAL;
801 goto end;
802 }
803
804 if (reloc_offset > data_size) {
805 ERR("Bytecode %s reloc offset %u is not within data",
806 bytecode_type_str(lum->cmd), reloc_offset);
807 ret = -EINVAL;
808 goto end;
809 }
810
811 /* Allocate the structure AND the `data[]` field. */
812 bytecode = zmalloc(sizeof(*bytecode) + data_size);
813 if (!bytecode) {
814 ret = -ENOMEM;
815 goto end;
816 }
817
818 bytecode->bc.len = data_size;
819 bytecode->bc.reloc_offset = reloc_offset;
820 bytecode->bc.seqnum = seqnum;
821 bytecode->type = type;
822
823 len = ustcomm_recv_unix_sock(sock, bytecode->bc.data, bytecode->bc.len);
824 switch (len) {
825 case 0: /* orderly shutdown */
826 ret = 0;
827 goto end;
828 default:
829 if (len == bytecode->bc.len) {
830 DBG("Bytecode %s data received",
831 bytecode_type_str(lum->cmd));
832 break;
833 } else if (len < 0) {
834 DBG("Receive failed from lttng-sessiond with errno %d",
835 (int) -len);
836 if (len == -ECONNRESET) {
837 ERR("%s remote end closed connection",
838 sock_info->name);
839 ret = len;
840 goto end;
841 }
842 ret = len;
843 goto end;
844 } else {
845 DBG("Incorrect %s bytecode data message size: %zd",
846 bytecode_type_str(lum->cmd), len);
847 ret = -EINVAL;
848 goto end;
849 }
850 }
851
852 ops = lttng_ust_abi_objd_ops(lum->handle);
853 if (!ops) {
854 ret = -ENOENT;
855 goto end;
856 }
857
858 if (ops->cmd)
859 ret = ops->cmd(lum->handle, lum->cmd,
860 (unsigned long) &bytecode,
861 NULL, sock_info);
862 else
863 ret = -ENOSYS;
864
865 end:
866 free(bytecode);
867 return ret;
868 }
869
870 static
871 void prepare_cmd_reply(struct ustcomm_ust_reply *lur, uint32_t handle, uint32_t cmd, int ret)
872 {
873 lur->handle = handle;
874 lur->cmd = cmd;
875 lur->ret_val = ret;
876 if (ret >= 0) {
877 lur->ret_code = LTTNG_UST_OK;
878 } else {
879 /*
880 * Use -LTTNG_UST_ERR as wildcard for UST internal
881 * error that are not caused by the transport, except if
882 * we already have a more precise error message to
883 * report.
884 */
885 if (ret > -LTTNG_UST_ERR) {
886 /* Translate code to UST error. */
887 switch (ret) {
888 case -EEXIST:
889 lur->ret_code = -LTTNG_UST_ERR_EXIST;
890 break;
891 case -EINVAL:
892 lur->ret_code = -LTTNG_UST_ERR_INVAL;
893 break;
894 case -ENOENT:
895 lur->ret_code = -LTTNG_UST_ERR_NOENT;
896 break;
897 case -EPERM:
898 lur->ret_code = -LTTNG_UST_ERR_PERM;
899 break;
900 case -ENOSYS:
901 lur->ret_code = -LTTNG_UST_ERR_NOSYS;
902 break;
903 default:
904 lur->ret_code = -LTTNG_UST_ERR;
905 break;
906 }
907 } else {
908 lur->ret_code = ret;
909 }
910 }
911 }
912
913 static
914 int handle_message(struct sock_info *sock_info,
915 int sock, struct ustcomm_ust_msg *lum)
916 {
917 int ret = 0;
918 const struct lttng_ust_abi_objd_ops *ops;
919 struct ustcomm_ust_reply lur;
920 union lttng_ust_abi_args args;
921 char ctxstr[LTTNG_UST_ABI_SYM_NAME_LEN]; /* App context string. */
922 ssize_t len;
923
924 memset(&lur, 0, sizeof(lur));
925
926 if (ust_lock()) {
927 ret = -LTTNG_UST_ERR_EXITING;
928 goto error;
929 }
930
931 ops = lttng_ust_abi_objd_ops(lum->handle);
932 if (!ops) {
933 ret = -ENOENT;
934 goto error;
935 }
936
937 switch (lum->cmd) {
938 case LTTNG_UST_ABI_FILTER:
939 case LTTNG_UST_ABI_EXCLUSION:
940 case LTTNG_UST_ABI_CHANNEL:
941 case LTTNG_UST_ABI_STREAM:
942 case LTTNG_UST_ABI_CONTEXT:
943 /*
944 * Those commands send additional payload after struct
945 * ustcomm_ust_msg, which makes it pretty much impossible to
946 * deal with "unknown command" errors without leaving the
947 * communication pipe in a out-of-sync state. This is part of
948 * the ABI between liblttng-ust-ctl and liblttng-ust, and
949 * should be fixed on the next breaking
950 * LTTNG_UST_ABI_MAJOR_VERSION protocol bump by indicating the
951 * total command message length as part of a message header so
952 * that the protocol can recover from invalid command errors.
953 */
954 break;
955
956 case LTTNG_UST_ABI_CAPTURE:
957 case LTTNG_UST_ABI_COUNTER:
958 case LTTNG_UST_ABI_COUNTER_GLOBAL:
959 case LTTNG_UST_ABI_COUNTER_CPU:
960 case LTTNG_UST_ABI_EVENT_NOTIFIER_CREATE:
961 case LTTNG_UST_ABI_EVENT_NOTIFIER_GROUP_CREATE:
962 /*
963 * Those commands expect a reply to the struct ustcomm_ust_msg
964 * before sending additional payload.
965 */
966 prepare_cmd_reply(&lur, lum->handle, lum->cmd, 0);
967
968 ret = send_reply(sock, &lur);
969 if (ret < 0) {
970 DBG("error sending reply");
971 goto error;
972 }
973 break;
974
975 default:
976 /*
977 * Other commands either don't send additional payload, or are
978 * unknown.
979 */
980 break;
981 }
982
983 switch (lum->cmd) {
984 case LTTNG_UST_ABI_REGISTER_DONE:
985 if (lum->handle == LTTNG_UST_ABI_ROOT_HANDLE)
986 ret = handle_register_done(sock_info);
987 else
988 ret = -EINVAL;
989 break;
990 case LTTNG_UST_ABI_RELEASE:
991 if (lum->handle == LTTNG_UST_ABI_ROOT_HANDLE)
992 ret = -EPERM;
993 else
994 ret = lttng_ust_abi_objd_unref(lum->handle, 1);
995 break;
996 case LTTNG_UST_ABI_CAPTURE:
997 case LTTNG_UST_ABI_FILTER:
998 ret = handle_bytecode_recv(sock_info, sock, lum);
999 if (ret)
1000 goto error;
1001 break;
1002 case LTTNG_UST_ABI_EXCLUSION:
1003 {
1004 /* Receive exclusion names */
1005 struct lttng_ust_excluder_node *node;
1006 unsigned int count;
1007
1008 count = lum->u.exclusion.count;
1009 if (count == 0) {
1010 /* There are no names to read */
1011 ret = 0;
1012 goto error;
1013 }
1014 node = zmalloc(sizeof(*node) +
1015 count * LTTNG_UST_ABI_SYM_NAME_LEN);
1016 if (!node) {
1017 ret = -ENOMEM;
1018 goto error;
1019 }
1020 node->excluder.count = count;
1021 len = ustcomm_recv_unix_sock(sock, node->excluder.names,
1022 count * LTTNG_UST_ABI_SYM_NAME_LEN);
1023 switch (len) {
1024 case 0: /* orderly shutdown */
1025 ret = 0;
1026 free(node);
1027 goto error;
1028 default:
1029 if (len == count * LTTNG_UST_ABI_SYM_NAME_LEN) {
1030 DBG("Exclusion data received");
1031 break;
1032 } else if (len < 0) {
1033 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1034 if (len == -ECONNRESET) {
1035 ERR("%s remote end closed connection", sock_info->name);
1036 ret = len;
1037 free(node);
1038 goto error;
1039 }
1040 ret = len;
1041 free(node);
1042 goto error;
1043 } else {
1044 DBG("Incorrect exclusion data message size: %zd", len);
1045 ret = -EINVAL;
1046 free(node);
1047 goto error;
1048 }
1049 }
1050 if (ops->cmd)
1051 ret = ops->cmd(lum->handle, lum->cmd,
1052 (unsigned long) &node,
1053 &args, sock_info);
1054 else
1055 ret = -ENOSYS;
1056 free(node);
1057 break;
1058 }
1059 case LTTNG_UST_ABI_EVENT_NOTIFIER_GROUP_CREATE:
1060 {
1061 int event_notifier_notif_fd, close_ret;
1062
1063 len = ustcomm_recv_event_notifier_notif_fd_from_sessiond(sock,
1064 &event_notifier_notif_fd);
1065 switch (len) {
1066 case 0: /* orderly shutdown */
1067 ret = 0;
1068 goto error;
1069 case 1:
1070 break;
1071 default:
1072 if (len < 0) {
1073 DBG("Receive failed from lttng-sessiond with errno %d",
1074 (int) -len);
1075 if (len == -ECONNRESET) {
1076 ERR("%s remote end closed connection",
1077 sock_info->name);
1078 ret = len;
1079 goto error;
1080 }
1081 ret = len;
1082 goto error;
1083 } else {
1084 DBG("Incorrect event notifier fd message size: %zd",
1085 len);
1086 ret = -EINVAL;
1087 goto error;
1088 }
1089 }
1090 args.event_notifier_handle.event_notifier_notif_fd =
1091 event_notifier_notif_fd;
1092 if (ops->cmd)
1093 ret = ops->cmd(lum->handle, lum->cmd,
1094 (unsigned long) &lum->u,
1095 &args, sock_info);
1096 else
1097 ret = -ENOSYS;
1098 if (args.event_notifier_handle.event_notifier_notif_fd >= 0) {
1099 lttng_ust_lock_fd_tracker();
1100 close_ret = close(args.event_notifier_handle.event_notifier_notif_fd);
1101 lttng_ust_unlock_fd_tracker();
1102 if (close_ret)
1103 PERROR("close");
1104 }
1105 break;
1106 }
1107 case LTTNG_UST_ABI_CHANNEL:
1108 {
1109 void *chan_data;
1110 int wakeup_fd;
1111
1112 len = ustcomm_recv_channel_from_sessiond(sock,
1113 &chan_data, lum->u.channel.len,
1114 &wakeup_fd);
1115 switch (len) {
1116 case 0: /* orderly shutdown */
1117 ret = 0;
1118 goto error;
1119 default:
1120 if (len == lum->u.channel.len) {
1121 DBG("channel data received");
1122 break;
1123 } else if (len < 0) {
1124 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1125 if (len == -ECONNRESET) {
1126 ERR("%s remote end closed connection", sock_info->name);
1127 ret = len;
1128 goto error;
1129 }
1130 ret = len;
1131 goto error;
1132 } else {
1133 DBG("incorrect channel data message size: %zd", len);
1134 ret = -EINVAL;
1135 goto error;
1136 }
1137 }
1138 args.channel.chan_data = chan_data;
1139 args.channel.wakeup_fd = wakeup_fd;
1140 if (ops->cmd)
1141 ret = ops->cmd(lum->handle, lum->cmd,
1142 (unsigned long) &lum->u,
1143 &args, sock_info);
1144 else
1145 ret = -ENOSYS;
1146 if (args.channel.wakeup_fd >= 0) {
1147 int close_ret;
1148
1149 lttng_ust_lock_fd_tracker();
1150 close_ret = close(args.channel.wakeup_fd);
1151 lttng_ust_unlock_fd_tracker();
1152 args.channel.wakeup_fd = -1;
1153 if (close_ret)
1154 PERROR("close");
1155 }
1156 free(args.channel.chan_data);
1157 break;
1158 }
1159 case LTTNG_UST_ABI_STREAM:
1160 {
1161 int close_ret;
1162
1163 /* Receive shm_fd, wakeup_fd */
1164 ret = ustcomm_recv_stream_from_sessiond(sock,
1165 NULL,
1166 &args.stream.shm_fd,
1167 &args.stream.wakeup_fd);
1168 if (ret) {
1169 goto error;
1170 }
1171
1172 if (ops->cmd)
1173 ret = ops->cmd(lum->handle, lum->cmd,
1174 (unsigned long) &lum->u,
1175 &args, sock_info);
1176 else
1177 ret = -ENOSYS;
1178 if (args.stream.shm_fd >= 0) {
1179 lttng_ust_lock_fd_tracker();
1180 close_ret = close(args.stream.shm_fd);
1181 lttng_ust_unlock_fd_tracker();
1182 args.stream.shm_fd = -1;
1183 if (close_ret)
1184 PERROR("close");
1185 }
1186 if (args.stream.wakeup_fd >= 0) {
1187 lttng_ust_lock_fd_tracker();
1188 close_ret = close(args.stream.wakeup_fd);
1189 lttng_ust_unlock_fd_tracker();
1190 args.stream.wakeup_fd = -1;
1191 if (close_ret)
1192 PERROR("close");
1193 }
1194 break;
1195 }
1196 case LTTNG_UST_ABI_CONTEXT:
1197 switch (lum->u.context.ctx) {
1198 case LTTNG_UST_ABI_CONTEXT_APP_CONTEXT:
1199 {
1200 char *p;
1201 size_t ctxlen, recvlen;
1202
1203 ctxlen = strlen("$app.") + lum->u.context.u.app_ctx.provider_name_len - 1
1204 + strlen(":") + lum->u.context.u.app_ctx.ctx_name_len;
1205 if (ctxlen >= LTTNG_UST_ABI_SYM_NAME_LEN) {
1206 ERR("Application context string length size is too large: %zu bytes",
1207 ctxlen);
1208 ret = -EINVAL;
1209 goto error;
1210 }
1211 strcpy(ctxstr, "$app.");
1212 p = &ctxstr[strlen("$app.")];
1213 recvlen = ctxlen - strlen("$app.");
1214 len = ustcomm_recv_unix_sock(sock, p, recvlen);
1215 switch (len) {
1216 case 0: /* orderly shutdown */
1217 ret = 0;
1218 goto error;
1219 default:
1220 if (len == recvlen) {
1221 DBG("app context data received");
1222 break;
1223 } else if (len < 0) {
1224 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1225 if (len == -ECONNRESET) {
1226 ERR("%s remote end closed connection", sock_info->name);
1227 ret = len;
1228 goto error;
1229 }
1230 ret = len;
1231 goto error;
1232 } else {
1233 DBG("incorrect app context data message size: %zd", len);
1234 ret = -EINVAL;
1235 goto error;
1236 }
1237 }
1238 /* Put : between provider and ctxname. */
1239 p[lum->u.context.u.app_ctx.provider_name_len - 1] = ':';
1240 args.app_context.ctxname = ctxstr;
1241 break;
1242 }
1243 default:
1244 break;
1245 }
1246 if (ops->cmd) {
1247 ret = ops->cmd(lum->handle, lum->cmd,
1248 (unsigned long) &lum->u,
1249 &args, sock_info);
1250 } else {
1251 ret = -ENOSYS;
1252 }
1253 break;
1254 case LTTNG_UST_ABI_COUNTER:
1255 {
1256 void *counter_data;
1257
1258 len = ustcomm_recv_counter_from_sessiond(sock,
1259 &counter_data, lum->u.counter.len);
1260 switch (len) {
1261 case 0: /* orderly shutdown */
1262 ret = 0;
1263 goto error;
1264 default:
1265 if (len == lum->u.counter.len) {
1266 DBG("counter data received");
1267 break;
1268 } else if (len < 0) {
1269 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1270 if (len == -ECONNRESET) {
1271 ERR("%s remote end closed connection", sock_info->name);
1272 ret = len;
1273 goto error;
1274 }
1275 ret = len;
1276 goto error;
1277 } else {
1278 DBG("incorrect counter data message size: %zd", len);
1279 ret = -EINVAL;
1280 goto error;
1281 }
1282 }
1283 args.counter.counter_data = counter_data;
1284 if (ops->cmd)
1285 ret = ops->cmd(lum->handle, lum->cmd,
1286 (unsigned long) &lum->u,
1287 &args, sock_info);
1288 else
1289 ret = -ENOSYS;
1290 free(args.counter.counter_data);
1291 break;
1292 }
1293 case LTTNG_UST_ABI_COUNTER_GLOBAL:
1294 {
1295 /* Receive shm_fd */
1296 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1297 &args.counter_shm.shm_fd);
1298 if (ret) {
1299 goto error;
1300 }
1301
1302 if (ops->cmd)
1303 ret = ops->cmd(lum->handle, lum->cmd,
1304 (unsigned long) &lum->u,
1305 &args, sock_info);
1306 else
1307 ret = -ENOSYS;
1308 if (args.counter_shm.shm_fd >= 0) {
1309 int close_ret;
1310
1311 lttng_ust_lock_fd_tracker();
1312 close_ret = close(args.counter_shm.shm_fd);
1313 lttng_ust_unlock_fd_tracker();
1314 args.counter_shm.shm_fd = -1;
1315 if (close_ret)
1316 PERROR("close");
1317 }
1318 break;
1319 }
1320 case LTTNG_UST_ABI_COUNTER_CPU:
1321 {
1322 /* Receive shm_fd */
1323 ret = ustcomm_recv_counter_shm_from_sessiond(sock,
1324 &args.counter_shm.shm_fd);
1325 if (ret) {
1326 goto error;
1327 }
1328
1329 if (ops->cmd)
1330 ret = ops->cmd(lum->handle, lum->cmd,
1331 (unsigned long) &lum->u,
1332 &args, sock_info);
1333 else
1334 ret = -ENOSYS;
1335 if (args.counter_shm.shm_fd >= 0) {
1336 int close_ret;
1337
1338 lttng_ust_lock_fd_tracker();
1339 close_ret = close(args.counter_shm.shm_fd);
1340 lttng_ust_unlock_fd_tracker();
1341 args.counter_shm.shm_fd = -1;
1342 if (close_ret)
1343 PERROR("close");
1344 }
1345 break;
1346 }
1347 case LTTNG_UST_ABI_EVENT_NOTIFIER_CREATE:
1348 {
1349 /* Receive struct lttng_ust_event_notifier */
1350 struct lttng_ust_abi_event_notifier event_notifier;
1351
1352 if (sizeof(event_notifier) != lum->u.event_notifier.len) {
1353 DBG("incorrect event notifier data message size: %u", lum->u.event_notifier.len);
1354 ret = -EINVAL;
1355 goto error;
1356 }
1357 len = ustcomm_recv_unix_sock(sock, &event_notifier, sizeof(event_notifier));
1358 switch (len) {
1359 case 0: /* orderly shutdown */
1360 ret = 0;
1361 goto error;
1362 default:
1363 if (len == sizeof(event_notifier)) {
1364 DBG("event notifier data received");
1365 break;
1366 } else if (len < 0) {
1367 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1368 if (len == -ECONNRESET) {
1369 ERR("%s remote end closed connection", sock_info->name);
1370 ret = len;
1371 goto error;
1372 }
1373 ret = len;
1374 goto error;
1375 } else {
1376 DBG("incorrect event notifier data message size: %zd", len);
1377 ret = -EINVAL;
1378 goto error;
1379 }
1380 }
1381 if (ops->cmd)
1382 ret = ops->cmd(lum->handle, lum->cmd,
1383 (unsigned long) &event_notifier,
1384 &args, sock_info);
1385 else
1386 ret = -ENOSYS;
1387 break;
1388 }
1389
1390 default:
1391 if (ops->cmd)
1392 ret = ops->cmd(lum->handle, lum->cmd,
1393 (unsigned long) &lum->u,
1394 &args, sock_info);
1395 else
1396 ret = -ENOSYS;
1397 break;
1398 }
1399
1400 prepare_cmd_reply(&lur, lum->handle, lum->cmd, ret);
1401
1402 if (ret >= 0) {
1403 switch (lum->cmd) {
1404 case LTTNG_UST_ABI_TRACER_VERSION:
1405 lur.u.version = lum->u.version;
1406 break;
1407 case LTTNG_UST_ABI_TRACEPOINT_LIST_GET:
1408 memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
1409 break;
1410 }
1411 }
1412 DBG("Return value: %d", lur.ret_val);
1413
1414 ust_unlock();
1415
1416 /*
1417 * Performed delayed statedump operations outside of the UST
1418 * lock. We need to take the dynamic loader lock before we take
1419 * the UST lock internally within handle_pending_statedump().
1420 */
1421 handle_pending_statedump(sock_info);
1422
1423 if (ust_lock()) {
1424 ret = -LTTNG_UST_ERR_EXITING;
1425 goto error;
1426 }
1427
1428 ret = send_reply(sock, &lur);
1429 if (ret < 0) {
1430 DBG("error sending reply");
1431 goto error;
1432 }
1433
1434 /*
1435 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1436 * after the reply.
1437 */
1438 if (lur.ret_code == LTTNG_UST_OK) {
1439 switch (lum->cmd) {
1440 case LTTNG_UST_ABI_TRACEPOINT_FIELD_LIST_GET:
1441 len = ustcomm_send_unix_sock(sock,
1442 &args.field_list.entry,
1443 sizeof(args.field_list.entry));
1444 if (len < 0) {
1445 ret = len;
1446 goto error;
1447 }
1448 if (len != sizeof(args.field_list.entry)) {
1449 ret = -EINVAL;
1450 goto error;
1451 }
1452 }
1453 }
1454
1455 error:
1456 ust_unlock();
1457
1458 return ret;
1459 }
1460
1461 static
1462 void cleanup_sock_info(struct sock_info *sock_info, int exiting)
1463 {
1464 int ret;
1465
1466 if (sock_info->root_handle != -1) {
1467 ret = lttng_ust_abi_objd_unref(sock_info->root_handle, 1);
1468 if (ret) {
1469 ERR("Error unref root handle");
1470 }
1471 sock_info->root_handle = -1;
1472 }
1473
1474
1475 /*
1476 * wait_shm_mmap, socket and notify socket are used by listener
1477 * threads outside of the ust lock, so we cannot tear them down
1478 * ourselves, because we cannot join on these threads. Leave
1479 * responsibility of cleaning up these resources to the OS
1480 * process exit.
1481 */
1482 if (exiting)
1483 return;
1484
1485 sock_info->registration_done = 0;
1486 sock_info->initial_statedump_done = 0;
1487
1488 if (sock_info->socket != -1) {
1489 ret = ustcomm_close_unix_sock(sock_info->socket);
1490 if (ret) {
1491 ERR("Error closing ust cmd socket");
1492 }
1493 sock_info->socket = -1;
1494 }
1495 if (sock_info->notify_socket != -1) {
1496 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1497 if (ret) {
1498 ERR("Error closing ust notify socket");
1499 }
1500 sock_info->notify_socket = -1;
1501 }
1502 if (sock_info->wait_shm_mmap) {
1503 long page_size;
1504
1505 page_size = LTTNG_UST_PAGE_SIZE;
1506 if (page_size <= 0) {
1507 if (!page_size) {
1508 errno = EINVAL;
1509 }
1510 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1511 } else {
1512 ret = munmap(sock_info->wait_shm_mmap, page_size);
1513 if (ret) {
1514 ERR("Error unmapping wait shm");
1515 }
1516 }
1517 sock_info->wait_shm_mmap = NULL;
1518 }
1519 }
1520
1521 /*
1522 * Using fork to set umask in the child process (not multi-thread safe).
1523 * We deal with the shm_open vs ftruncate race (happening when the
1524 * sessiond owns the shm and does not let everybody modify it, to ensure
1525 * safety against shm_unlink) by simply letting the mmap fail and
1526 * retrying after a few seconds.
1527 * For global shm, everybody has rw access to it until the sessiond
1528 * starts.
1529 */
1530 static
1531 int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
1532 {
1533 int wait_shm_fd, ret;
1534 pid_t pid;
1535
1536 /*
1537 * Try to open read-only.
1538 */
1539 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1540 if (wait_shm_fd >= 0) {
1541 int32_t tmp_read;
1542 ssize_t len;
1543 size_t bytes_read = 0;
1544
1545 /*
1546 * Try to read the fd. If unable to do so, try opening
1547 * it in write mode.
1548 */
1549 do {
1550 len = read(wait_shm_fd,
1551 &((char *) &tmp_read)[bytes_read],
1552 sizeof(tmp_read) - bytes_read);
1553 if (len > 0) {
1554 bytes_read += len;
1555 }
1556 } while ((len < 0 && errno == EINTR)
1557 || (len > 0 && bytes_read < sizeof(tmp_read)));
1558 if (bytes_read != sizeof(tmp_read)) {
1559 ret = close(wait_shm_fd);
1560 if (ret) {
1561 ERR("close wait_shm_fd");
1562 }
1563 goto open_write;
1564 }
1565 goto end;
1566 } else if (wait_shm_fd < 0 && errno != ENOENT) {
1567 /*
1568 * Real-only open did not work, and it's not because the
1569 * entry was not present. It's a failure that prohibits
1570 * using shm.
1571 */
1572 ERR("Error opening shm %s", sock_info->wait_shm_path);
1573 goto end;
1574 }
1575
1576 open_write:
1577 /*
1578 * If the open failed because the file did not exist, or because
1579 * the file was not truncated yet, try creating it ourself.
1580 */
1581 URCU_TLS(lttng_ust_nest_count)++;
1582 pid = fork();
1583 URCU_TLS(lttng_ust_nest_count)--;
1584 if (pid > 0) {
1585 int status;
1586
1587 /*
1588 * Parent: wait for child to return, in which case the
1589 * shared memory map will have been created.
1590 */
1591 pid = wait(&status);
1592 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
1593 wait_shm_fd = -1;
1594 goto end;
1595 }
1596 /*
1597 * Try to open read-only again after creation.
1598 */
1599 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1600 if (wait_shm_fd < 0) {
1601 /*
1602 * Real-only open did not work. It's a failure
1603 * that prohibits using shm.
1604 */
1605 ERR("Error opening shm %s", sock_info->wait_shm_path);
1606 goto end;
1607 }
1608 goto end;
1609 } else if (pid == 0) {
1610 int create_mode;
1611
1612 /* Child */
1613 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
1614 if (sock_info->global)
1615 create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
1616 /*
1617 * We're alone in a child process, so we can modify the
1618 * process-wide umask.
1619 */
1620 umask(~create_mode);
1621 /*
1622 * Try creating shm (or get rw access).
1623 * We don't do an exclusive open, because we allow other
1624 * processes to create+ftruncate it concurrently.
1625 */
1626 wait_shm_fd = shm_open(sock_info->wait_shm_path,
1627 O_RDWR | O_CREAT, create_mode);
1628 if (wait_shm_fd >= 0) {
1629 ret = ftruncate(wait_shm_fd, mmap_size);
1630 if (ret) {
1631 PERROR("ftruncate");
1632 _exit(EXIT_FAILURE);
1633 }
1634 _exit(EXIT_SUCCESS);
1635 }
1636 /*
1637 * For local shm, we need to have rw access to accept
1638 * opening it: this means the local sessiond will be
1639 * able to wake us up. For global shm, we open it even
1640 * if rw access is not granted, because the root.root
1641 * sessiond will be able to override all rights and wake
1642 * us up.
1643 */
1644 if (!sock_info->global && errno != EACCES) {
1645 ERR("Error opening shm %s", sock_info->wait_shm_path);
1646 _exit(EXIT_FAILURE);
1647 }
1648 /*
1649 * The shm exists, but we cannot open it RW. Report
1650 * success.
1651 */
1652 _exit(EXIT_SUCCESS);
1653 } else {
1654 return -1;
1655 }
1656 end:
1657 if (wait_shm_fd >= 0 && !sock_info->global) {
1658 struct stat statbuf;
1659
1660 /*
1661 * Ensure that our user is the owner of the shm file for
1662 * local shm. If we do not own the file, it means our
1663 * sessiond will not have access to wake us up (there is
1664 * probably a rogue process trying to fake our
1665 * sessiond). Fallback to polling method in this case.
1666 */
1667 ret = fstat(wait_shm_fd, &statbuf);
1668 if (ret) {
1669 PERROR("fstat");
1670 goto error_close;
1671 }
1672 if (statbuf.st_uid != getuid())
1673 goto error_close;
1674 }
1675 return wait_shm_fd;
1676
1677 error_close:
1678 ret = close(wait_shm_fd);
1679 if (ret) {
1680 PERROR("Error closing fd");
1681 }
1682 return -1;
1683 }
1684
1685 static
1686 char *get_map_shm(struct sock_info *sock_info)
1687 {
1688 long page_size;
1689 int wait_shm_fd, ret;
1690 char *wait_shm_mmap;
1691
1692 page_size = sysconf(_SC_PAGE_SIZE);
1693 if (page_size <= 0) {
1694 if (!page_size) {
1695 errno = EINVAL;
1696 }
1697 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1698 goto error;
1699 }
1700
1701 lttng_ust_lock_fd_tracker();
1702 wait_shm_fd = get_wait_shm(sock_info, page_size);
1703 if (wait_shm_fd < 0) {
1704 lttng_ust_unlock_fd_tracker();
1705 goto error;
1706 }
1707
1708 ret = lttng_ust_add_fd_to_tracker(wait_shm_fd);
1709 if (ret < 0) {
1710 ret = close(wait_shm_fd);
1711 if (!ret) {
1712 PERROR("Error closing fd");
1713 }
1714 lttng_ust_unlock_fd_tracker();
1715 goto error;
1716 }
1717
1718 wait_shm_fd = ret;
1719 lttng_ust_unlock_fd_tracker();
1720
1721 wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
1722 MAP_SHARED, wait_shm_fd, 0);
1723
1724 /* close shm fd immediately after taking the mmap reference */
1725 lttng_ust_lock_fd_tracker();
1726 ret = close(wait_shm_fd);
1727 if (!ret) {
1728 lttng_ust_delete_fd_from_tracker(wait_shm_fd);
1729 } else {
1730 PERROR("Error closing fd");
1731 }
1732 lttng_ust_unlock_fd_tracker();
1733
1734 if (wait_shm_mmap == MAP_FAILED) {
1735 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1736 goto error;
1737 }
1738 return wait_shm_mmap;
1739
1740 error:
1741 return NULL;
1742 }
1743
1744 static
1745 void wait_for_sessiond(struct sock_info *sock_info)
1746 {
1747 /* Use ust_lock to check if we should quit. */
1748 if (ust_lock()) {
1749 goto quit;
1750 }
1751 if (wait_poll_fallback) {
1752 goto error;
1753 }
1754 ust_unlock();
1755
1756 assert(sock_info->wait_shm_mmap);
1757
1758 DBG("Waiting for %s apps sessiond", sock_info->name);
1759 /* Wait for futex wakeup */
1760 while (!uatomic_read((int32_t *) sock_info->wait_shm_mmap)) {
1761 if (!lttng_ust_futex_async((int32_t *) sock_info->wait_shm_mmap, FUTEX_WAIT, 0, NULL, NULL, 0)) {
1762 /*
1763 * Prior queued wakeups queued by unrelated code
1764 * using the same address can cause futex wait to
1765 * return 0 even through the futex value is still
1766 * 0 (spurious wakeups). Check the value again
1767 * in user-space to validate whether it really
1768 * differs from 0.
1769 */
1770 continue;
1771 }
1772 switch (errno) {
1773 case EAGAIN:
1774 /* Value already changed. */
1775 goto end_wait;
1776 case EINTR:
1777 /* Retry if interrupted by signal. */
1778 break; /* Get out of switch. Check again. */
1779 case EFAULT:
1780 wait_poll_fallback = 1;
1781 DBG(
1782 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1783 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1784 "Please upgrade your kernel "
1785 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1786 "mainline). LTTng-UST will use polling mode fallback.");
1787 if (lttng_ust_logging_debug_enabled())
1788 PERROR("futex");
1789 goto end_wait;
1790 }
1791 }
1792 end_wait:
1793 return;
1794
1795 quit:
1796 ust_unlock();
1797 return;
1798
1799 error:
1800 ust_unlock();
1801 return;
1802 }
1803
1804 /*
1805 * This thread does not allocate any resource, except within
1806 * handle_message, within mutex protection. This mutex protects against
1807 * fork and exit.
1808 * The other moment it allocates resources is at socket connection, which
1809 * is also protected by the mutex.
1810 */
1811 static
1812 void *ust_listener_thread(void *arg)
1813 {
1814 struct sock_info *sock_info = arg;
1815 int sock, ret, prev_connect_failed = 0, has_waited = 0, fd;
1816 long timeout;
1817
1818 lttng_ust_common_init_thread(0);
1819 /*
1820 * If available, add '-ust' to the end of this thread's
1821 * process name
1822 */
1823 ret = lttng_ust_setustprocname();
1824 if (ret) {
1825 ERR("Unable to set UST process name");
1826 }
1827
1828 /* Restart trying to connect to the session daemon */
1829 restart:
1830 if (prev_connect_failed) {
1831 /* Wait for sessiond availability with pipe */
1832 wait_for_sessiond(sock_info);
1833 if (has_waited) {
1834 has_waited = 0;
1835 /*
1836 * Sleep for 5 seconds before retrying after a
1837 * sequence of failure / wait / failure. This
1838 * deals with a killed or broken session daemon.
1839 */
1840 sleep(5);
1841 } else {
1842 has_waited = 1;
1843 }
1844 prev_connect_failed = 0;
1845 }
1846
1847 if (ust_lock()) {
1848 goto quit;
1849 }
1850
1851 if (sock_info->socket != -1) {
1852 /* FD tracker is updated by ustcomm_close_unix_sock() */
1853 ret = ustcomm_close_unix_sock(sock_info->socket);
1854 if (ret) {
1855 ERR("Error closing %s ust cmd socket",
1856 sock_info->name);
1857 }
1858 sock_info->socket = -1;
1859 }
1860 if (sock_info->notify_socket != -1) {
1861 /* FD tracker is updated by ustcomm_close_unix_sock() */
1862 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1863 if (ret) {
1864 ERR("Error closing %s ust notify socket",
1865 sock_info->name);
1866 }
1867 sock_info->notify_socket = -1;
1868 }
1869
1870
1871 /*
1872 * Register. We need to perform both connect and sending
1873 * registration message before doing the next connect otherwise
1874 * we may reach unix socket connect queue max limits and block
1875 * on the 2nd connect while the session daemon is awaiting the
1876 * first connect registration message.
1877 */
1878 /* Connect cmd socket */
1879 lttng_ust_lock_fd_tracker();
1880 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1881 get_connect_sock_timeout());
1882 if (ret < 0) {
1883 lttng_ust_unlock_fd_tracker();
1884 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1885 prev_connect_failed = 1;
1886
1887 /*
1888 * If we cannot find the sessiond daemon, don't delay
1889 * constructor execution.
1890 */
1891 ret = handle_register_failed(sock_info);
1892 assert(!ret);
1893 ust_unlock();
1894 goto restart;
1895 }
1896 fd = ret;
1897 ret = lttng_ust_add_fd_to_tracker(fd);
1898 if (ret < 0) {
1899 ret = close(fd);
1900 if (ret) {
1901 PERROR("close on sock_info->socket");
1902 }
1903 ret = -1;
1904 lttng_ust_unlock_fd_tracker();
1905 ust_unlock();
1906 goto quit;
1907 }
1908
1909 sock_info->socket = ret;
1910 lttng_ust_unlock_fd_tracker();
1911
1912 ust_unlock();
1913 /*
1914 * Unlock/relock ust lock because connect is blocking (with
1915 * timeout). Don't delay constructors on the ust lock for too
1916 * long.
1917 */
1918 if (ust_lock()) {
1919 goto quit;
1920 }
1921
1922 /*
1923 * Create only one root handle per listener thread for the whole
1924 * process lifetime, so we ensure we get ID which is statically
1925 * assigned to the root handle.
1926 */
1927 if (sock_info->root_handle == -1) {
1928 ret = lttng_abi_create_root_handle();
1929 if (ret < 0) {
1930 ERR("Error creating root handle");
1931 goto quit;
1932 }
1933 sock_info->root_handle = ret;
1934 }
1935
1936 ret = register_to_sessiond(sock_info->socket, LTTNG_UST_CTL_SOCKET_CMD,
1937 sock_info->procname);
1938 if (ret < 0) {
1939 ERR("Error registering to %s ust cmd socket",
1940 sock_info->name);
1941 prev_connect_failed = 1;
1942 /*
1943 * If we cannot register to the sessiond daemon, don't
1944 * delay constructor execution.
1945 */
1946 ret = handle_register_failed(sock_info);
1947 assert(!ret);
1948 ust_unlock();
1949 goto restart;
1950 }
1951
1952 ust_unlock();
1953 /*
1954 * Unlock/relock ust lock because connect is blocking (with
1955 * timeout). Don't delay constructors on the ust lock for too
1956 * long.
1957 */
1958 if (ust_lock()) {
1959 goto quit;
1960 }
1961
1962 /* Connect notify socket */
1963 lttng_ust_lock_fd_tracker();
1964 ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1965 get_connect_sock_timeout());
1966 if (ret < 0) {
1967 lttng_ust_unlock_fd_tracker();
1968 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1969 prev_connect_failed = 1;
1970
1971 /*
1972 * If we cannot find the sessiond daemon, don't delay
1973 * constructor execution.
1974 */
1975 ret = handle_register_failed(sock_info);
1976 assert(!ret);
1977 ust_unlock();
1978 goto restart;
1979 }
1980
1981 fd = ret;
1982 ret = lttng_ust_add_fd_to_tracker(fd);
1983 if (ret < 0) {
1984 ret = close(fd);
1985 if (ret) {
1986 PERROR("close on sock_info->notify_socket");
1987 }
1988 ret = -1;
1989 lttng_ust_unlock_fd_tracker();
1990 ust_unlock();
1991 goto quit;
1992 }
1993
1994 sock_info->notify_socket = ret;
1995 lttng_ust_unlock_fd_tracker();
1996
1997 ust_unlock();
1998 /*
1999 * Unlock/relock ust lock because connect is blocking (with
2000 * timeout). Don't delay constructors on the ust lock for too
2001 * long.
2002 */
2003 if (ust_lock()) {
2004 goto quit;
2005 }
2006
2007 timeout = get_notify_sock_timeout();
2008 if (timeout >= 0) {
2009 /*
2010 * Give at least 10ms to sessiond to reply to
2011 * notifications.
2012 */
2013 if (timeout < 10)
2014 timeout = 10;
2015 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
2016 timeout);
2017 if (ret < 0) {
2018 WARN("Error setting socket receive timeout");
2019 }
2020 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
2021 timeout);
2022 if (ret < 0) {
2023 WARN("Error setting socket send timeout");
2024 }
2025 } else if (timeout < -1) {
2026 WARN("Unsupported timeout value %ld", timeout);
2027 }
2028
2029 ret = register_to_sessiond(sock_info->notify_socket,
2030 LTTNG_UST_CTL_SOCKET_NOTIFY, sock_info->procname);
2031 if (ret < 0) {
2032 ERR("Error registering to %s ust notify socket",
2033 sock_info->name);
2034 prev_connect_failed = 1;
2035 /*
2036 * If we cannot register to the sessiond daemon, don't
2037 * delay constructor execution.
2038 */
2039 ret = handle_register_failed(sock_info);
2040 assert(!ret);
2041 ust_unlock();
2042 goto restart;
2043 }
2044 sock = sock_info->socket;
2045
2046 ust_unlock();
2047
2048 for (;;) {
2049 ssize_t len;
2050 struct ustcomm_ust_msg lum;
2051
2052 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
2053 switch (len) {
2054 case 0: /* orderly shutdown */
2055 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
2056 if (ust_lock()) {
2057 goto quit;
2058 }
2059 /*
2060 * Either sessiond has shutdown or refused us by closing the socket.
2061 * In either case, we don't want to delay construction execution,
2062 * and we need to wait before retry.
2063 */
2064 prev_connect_failed = 1;
2065 /*
2066 * If we cannot register to the sessiond daemon, don't
2067 * delay constructor execution.
2068 */
2069 ret = handle_register_failed(sock_info);
2070 assert(!ret);
2071 ust_unlock();
2072 goto end;
2073 case sizeof(lum):
2074 print_cmd(lum.cmd, lum.handle);
2075 ret = handle_message(sock_info, sock, &lum);
2076 if (ret) {
2077 ERR("Error handling message for %s socket",
2078 sock_info->name);
2079 /*
2080 * Close socket if protocol error is
2081 * detected.
2082 */
2083 goto end;
2084 }
2085 continue;
2086 default:
2087 if (len < 0) {
2088 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
2089 } else {
2090 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
2091 }
2092 if (len == -ECONNRESET) {
2093 DBG("%s remote end closed connection", sock_info->name);
2094 goto end;
2095 }
2096 goto end;
2097 }
2098
2099 }
2100 end:
2101 if (ust_lock()) {
2102 goto quit;
2103 }
2104 /* Cleanup socket handles before trying to reconnect */
2105 lttng_ust_abi_objd_table_owner_cleanup(sock_info);
2106 ust_unlock();
2107 goto restart; /* try to reconnect */
2108
2109 quit:
2110 ust_unlock();
2111
2112 pthread_mutex_lock(&ust_exit_mutex);
2113 sock_info->thread_active = 0;
2114 pthread_mutex_unlock(&ust_exit_mutex);
2115 return NULL;
2116 }
2117
2118 /*
2119 * Weak symbol to call when the ust malloc wrapper is not loaded.
2120 */
2121 __attribute__((weak))
2122 void lttng_ust_libc_wrapper_malloc_ctor(void)
2123 {
2124 }
2125
2126 /*
2127 * Use a symbol of the previous ABI to detect if liblttng-ust.so.0 is loaded in
2128 * the current process.
2129 */
2130 #define LTTNG_UST_SONAME_0_SYM "ltt_probe_register"
2131
2132 static
2133 void lttng_ust_check_soname_0(void)
2134 {
2135 if (!dlsym(RTLD_DEFAULT, LTTNG_UST_SONAME_0_SYM))
2136 return;
2137
2138 CRIT("Incompatible library ABIs detected within the same process. "
2139 "The process is likely linked against different major soname of LTTng-UST which is unsupported. "
2140 "The detection was triggered by lookup of ABI 0 symbol \"%s\" in the Global Symbol Table\n",
2141 LTTNG_UST_SONAME_0_SYM);
2142 }
2143
2144 /*
2145 * Expose a canary symbol of the previous ABI to ensure we catch uses of a
2146 * liblttng-ust.so.0 dlopen'd after .so.1 has been loaded. Use a different
2147 * symbol than the detection code to ensure we don't detect ourself.
2148 *
2149 * This scheme will only work on systems where the global symbol table has
2150 * priority when resolving the symbols of a dlopened shared object, which is
2151 * the case on Linux but not on FreeBSD.
2152 */
2153 void init_usterr(void);
2154 void init_usterr(void)
2155 {
2156 CRIT("Incompatible library ABIs detected within the same process. "
2157 "The process is likely linked against different major soname of LTTng-UST which is unsupported. "
2158 "The detection was triggered by canary symbol \"%s\"\n", __func__);
2159 }
2160
2161 /*
2162 * sessiond monitoring thread: monitor presence of global and per-user
2163 * sessiond by polling the application common named pipe.
2164 */
2165 static
2166 void lttng_ust_ctor(void)
2167 __attribute__((constructor));
2168 static
2169 void lttng_ust_ctor(void)
2170 {
2171 struct timespec constructor_timeout;
2172 sigset_t sig_all_blocked, orig_parent_mask;
2173 pthread_attr_t thread_attr;
2174 int timeout_mode;
2175 int ret;
2176 void *handle;
2177
2178 if (uatomic_xchg(&initialized, 1) == 1)
2179 return;
2180
2181 /*
2182 * Fixup interdependency between TLS allocation mutex (which happens
2183 * to be the dynamic linker mutex) and ust_lock, taken within
2184 * the ust lock.
2185 */
2186 lttng_ust_common_init_thread(0);
2187
2188 lttng_ust_loaded = 1;
2189
2190 /*
2191 * Check if we find a symbol of the previous ABI in the current process
2192 * as different ABIs of liblttng-ust can't co-exist in a process. If we
2193 * do so, emit a critical log message which will also abort if the
2194 * LTTNG_UST_ABORT_ON_CRITICAL environment variable is set.
2195 */
2196 lttng_ust_check_soname_0();
2197
2198 /*
2199 * We need to ensure that the liblttng-ust library is not unloaded to avoid
2200 * the unloading of code used by the ust_listener_threads as we can not
2201 * reliably know when they exited. To do that, manually load
2202 * liblttng-ust.so to increment the dynamic loader's internal refcount for
2203 * this library so it never becomes zero, thus never gets unloaded from the
2204 * address space of the process. Since we are already running in the
2205 * constructor of the LTTNG_UST_LIB_SONAME library, calling dlopen will
2206 * simply increment the refcount and no additional work is needed by the
2207 * dynamic loader as the shared library is already loaded in the address
2208 * space. As a safe guard, we use the RTLD_NODELETE flag to prevent
2209 * unloading of the UST library if its refcount becomes zero (which should
2210 * never happen). Do the return value check but discard the handle at the
2211 * end of the function as it's not needed.
2212 */
2213 handle = dlopen(LTTNG_UST_LIB_SONAME, RTLD_LAZY | RTLD_NODELETE);
2214 if (!handle) {
2215 ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SONAME);
2216 } else {
2217 DBG("dlopened liblttng-ust shared library (%s).", LTTNG_UST_LIB_SONAME);
2218 }
2219
2220 /*
2221 * We want precise control over the order in which we construct
2222 * our sub-libraries vs starting to receive commands from
2223 * sessiond (otherwise leading to errors when trying to create
2224 * sessiond before the init functions are completed).
2225 */
2226
2227 /*
2228 * Both the logging and getenv lazy-initialization uses getenv()
2229 * internally and thus needs to be explicitly initialized in
2230 * liblttng-ust before we start any threads as an unsuspecting normally
2231 * single threaded application using liblttng-ust could be using
2232 * setenv() which is not thread-safe.
2233 */
2234 lttng_ust_logging_init();
2235 lttng_ust_getenv_init();
2236
2237 /* Call the liblttng-ust-common constructor. */
2238 lttng_ust_common_ctor();
2239
2240 lttng_ust_tp_init();
2241 lttng_ust_statedump_init();
2242 lttng_ust_ring_buffer_clients_init();
2243 lttng_ust_counter_clients_init();
2244 lttng_perf_counter_init();
2245 /*
2246 * Invoke ust malloc wrapper init before starting other threads.
2247 */
2248 lttng_ust_libc_wrapper_malloc_ctor();
2249
2250 timeout_mode = get_constructor_timeout(&constructor_timeout);
2251
2252 get_allow_blocking();
2253
2254 ret = sem_init(&constructor_wait, 0, 0);
2255 if (ret) {
2256 PERROR("sem_init");
2257 }
2258
2259 ret = setup_global_apps();
2260 if (ret) {
2261 assert(global_apps.allowed == 0);
2262 DBG("global apps setup returned %d", ret);
2263 }
2264
2265 ret = setup_local_apps();
2266 if (ret) {
2267 assert(local_apps.allowed == 0);
2268 DBG("local apps setup returned %d", ret);
2269 }
2270
2271 /* A new thread created by pthread_create inherits the signal mask
2272 * from the parent. To avoid any signal being received by the
2273 * listener thread, we block all signals temporarily in the parent,
2274 * while we create the listener thread.
2275 */
2276 sigfillset(&sig_all_blocked);
2277 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
2278 if (ret) {
2279 ERR("pthread_sigmask: %s", strerror(ret));
2280 }
2281
2282 ret = pthread_attr_init(&thread_attr);
2283 if (ret) {
2284 ERR("pthread_attr_init: %s", strerror(ret));
2285 }
2286 ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
2287 if (ret) {
2288 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
2289 }
2290
2291 if (global_apps.allowed) {
2292 pthread_mutex_lock(&ust_exit_mutex);
2293 ret = pthread_create(&global_apps.ust_listener, &thread_attr,
2294 ust_listener_thread, &global_apps);
2295 if (ret) {
2296 ERR("pthread_create global: %s", strerror(ret));
2297 }
2298 global_apps.thread_active = 1;
2299 pthread_mutex_unlock(&ust_exit_mutex);
2300 } else {
2301 handle_register_done(&global_apps);
2302 }
2303
2304 if (local_apps.allowed) {
2305 pthread_mutex_lock(&ust_exit_mutex);
2306 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
2307 ust_listener_thread, &local_apps);
2308 if (ret) {
2309 ERR("pthread_create local: %s", strerror(ret));
2310 }
2311 local_apps.thread_active = 1;
2312 pthread_mutex_unlock(&ust_exit_mutex);
2313 } else {
2314 handle_register_done(&local_apps);
2315 }
2316 ret = pthread_attr_destroy(&thread_attr);
2317 if (ret) {
2318 ERR("pthread_attr_destroy: %s", strerror(ret));
2319 }
2320
2321 /* Restore original signal mask in parent */
2322 ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
2323 if (ret) {
2324 ERR("pthread_sigmask: %s", strerror(ret));
2325 }
2326
2327 switch (timeout_mode) {
2328 case 1: /* timeout wait */
2329 do {
2330 ret = sem_timedwait(&constructor_wait,
2331 &constructor_timeout);
2332 } while (ret < 0 && errno == EINTR);
2333 if (ret < 0) {
2334 switch (errno) {
2335 case ETIMEDOUT:
2336 ERR("Timed out waiting for lttng-sessiond");
2337 break;
2338 case EINVAL:
2339 PERROR("sem_timedwait");
2340 break;
2341 default:
2342 ERR("Unexpected error \"%s\" returned by sem_timedwait",
2343 strerror(errno));
2344 }
2345 }
2346 break;
2347 case -1:/* wait forever */
2348 do {
2349 ret = sem_wait(&constructor_wait);
2350 } while (ret < 0 && errno == EINTR);
2351 if (ret < 0) {
2352 switch (errno) {
2353 case EINVAL:
2354 PERROR("sem_wait");
2355 break;
2356 default:
2357 ERR("Unexpected error \"%s\" returned by sem_wait",
2358 strerror(errno));
2359 }
2360 }
2361 break;
2362 case 0: /* no timeout */
2363 break;
2364 }
2365 }
2366
2367 static
2368 void lttng_ust_cleanup(int exiting)
2369 {
2370 cleanup_sock_info(&global_apps, exiting);
2371 cleanup_sock_info(&local_apps, exiting);
2372 local_apps.allowed = 0;
2373 global_apps.allowed = 0;
2374 /*
2375 * The teardown in this function all affect data structures
2376 * accessed under the UST lock by the listener thread. This
2377 * lock, along with the lttng_ust_comm_should_quit flag, ensure
2378 * that none of these threads are accessing this data at this
2379 * point.
2380 */
2381 lttng_ust_abi_exit();
2382 lttng_ust_abi_events_exit();
2383 lttng_perf_counter_exit();
2384 lttng_ust_ring_buffer_clients_exit();
2385 lttng_ust_counter_clients_exit();
2386 lttng_ust_statedump_destroy();
2387 lttng_ust_tp_exit();
2388 if (!exiting) {
2389 /* Reinitialize values for fork */
2390 sem_count = sem_count_initial_value;
2391 lttng_ust_comm_should_quit = 0;
2392 initialized = 0;
2393 }
2394 }
2395
2396 static
2397 void lttng_ust_exit(void)
2398 __attribute__((destructor));
2399 static
2400 void lttng_ust_exit(void)
2401 {
2402 int ret;
2403
2404 /*
2405 * Using pthread_cancel here because:
2406 * A) we don't want to hang application teardown.
2407 * B) the thread is not allocating any resource.
2408 */
2409
2410 /*
2411 * Require the communication thread to quit. Synchronize with
2412 * mutexes to ensure it is not in a mutex critical section when
2413 * pthread_cancel is later called.
2414 */
2415 ust_lock_nocheck();
2416 lttng_ust_comm_should_quit = 1;
2417 ust_unlock();
2418
2419 pthread_mutex_lock(&ust_exit_mutex);
2420 /* cancel threads */
2421 if (global_apps.thread_active) {
2422 ret = pthread_cancel(global_apps.ust_listener);
2423 if (ret) {
2424 ERR("Error cancelling global ust listener thread: %s",
2425 strerror(ret));
2426 } else {
2427 global_apps.thread_active = 0;
2428 }
2429 }
2430 if (local_apps.thread_active) {
2431 ret = pthread_cancel(local_apps.ust_listener);
2432 if (ret) {
2433 ERR("Error cancelling local ust listener thread: %s",
2434 strerror(ret));
2435 } else {
2436 local_apps.thread_active = 0;
2437 }
2438 }
2439 pthread_mutex_unlock(&ust_exit_mutex);
2440
2441 /*
2442 * Do NOT join threads: use of sys_futex makes it impossible to
2443 * join the threads without using async-cancel, but async-cancel
2444 * is delivered by a signal, which could hit the target thread
2445 * anywhere in its code path, including while the ust_lock() is
2446 * held, causing a deadlock for the other thread. Let the OS
2447 * cleanup the threads if there are stalled in a syscall.
2448 */
2449 lttng_ust_cleanup(1);
2450 }
2451
2452 static
2453 void ust_context_ns_reset(void)
2454 {
2455 lttng_context_pid_ns_reset();
2456 lttng_context_cgroup_ns_reset();
2457 lttng_context_ipc_ns_reset();
2458 lttng_context_mnt_ns_reset();
2459 lttng_context_net_ns_reset();
2460 lttng_context_user_ns_reset();
2461 lttng_context_time_ns_reset();
2462 lttng_context_uts_ns_reset();
2463 }
2464
2465 static
2466 void ust_context_vuids_reset(void)
2467 {
2468 lttng_context_vuid_reset();
2469 lttng_context_veuid_reset();
2470 lttng_context_vsuid_reset();
2471 }
2472
2473 static
2474 void ust_context_vgids_reset(void)
2475 {
2476 lttng_context_vgid_reset();
2477 lttng_context_vegid_reset();
2478 lttng_context_vsgid_reset();
2479 }
2480
2481 /*
2482 * We exclude the worker threads across fork and clone (except
2483 * CLONE_VM), because these system calls only keep the forking thread
2484 * running in the child. Therefore, we don't want to call fork or clone
2485 * in the middle of an tracepoint or ust tracing state modification.
2486 * Holding this mutex protects these structures across fork and clone.
2487 */
2488 void lttng_ust_before_fork(sigset_t *save_sigset)
2489 {
2490 /*
2491 * Disable signals. This is to avoid that the child intervenes
2492 * before it is properly setup for tracing. It is safer to
2493 * disable all signals, because then we know we are not breaking
2494 * anything by restoring the original mask.
2495 */
2496 sigset_t all_sigs;
2497 int ret;
2498
2499 /* Allocate lttng-ust TLS. */
2500 lttng_ust_common_init_thread(0);
2501
2502 if (URCU_TLS(lttng_ust_nest_count))
2503 return;
2504 /* Disable signals */
2505 sigfillset(&all_sigs);
2506 ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
2507 if (ret == -1) {
2508 PERROR("sigprocmask");
2509 }
2510
2511 pthread_mutex_lock(&ust_fork_mutex);
2512
2513 ust_lock_nocheck();
2514 lttng_ust_urcu_before_fork();
2515 lttng_ust_lock_fd_tracker();
2516 lttng_perf_lock();
2517 }
2518
2519 static void ust_after_fork_common(sigset_t *restore_sigset)
2520 {
2521 int ret;
2522
2523 DBG("process %d", getpid());
2524 lttng_perf_unlock();
2525 lttng_ust_unlock_fd_tracker();
2526 ust_unlock();
2527
2528 pthread_mutex_unlock(&ust_fork_mutex);
2529
2530 /* Restore signals */
2531 ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
2532 if (ret == -1) {
2533 PERROR("sigprocmask");
2534 }
2535 }
2536
2537 void lttng_ust_after_fork_parent(sigset_t *restore_sigset)
2538 {
2539 if (URCU_TLS(lttng_ust_nest_count))
2540 return;
2541 DBG("process %d", getpid());
2542 lttng_ust_urcu_after_fork_parent();
2543 /* Release mutexes and re-enable signals */
2544 ust_after_fork_common(restore_sigset);
2545 }
2546
2547 /*
2548 * After fork, in the child, we need to cleanup all the leftover state,
2549 * except the worker thread which already magically disappeared thanks
2550 * to the weird Linux fork semantics. After tyding up, we call
2551 * lttng_ust_ctor() again to start over as a new PID.
2552 *
2553 * This is meant for forks() that have tracing in the child between the
2554 * fork and following exec call (if there is any).
2555 */
2556 void lttng_ust_after_fork_child(sigset_t *restore_sigset)
2557 {
2558 if (URCU_TLS(lttng_ust_nest_count))
2559 return;
2560 lttng_context_vpid_reset();
2561 lttng_context_vtid_reset();
2562 lttng_ust_context_procname_reset();
2563 ust_context_ns_reset();
2564 ust_context_vuids_reset();
2565 ust_context_vgids_reset();
2566 DBG("process %d", getpid());
2567 /* Release urcu mutexes */
2568 lttng_ust_urcu_after_fork_child();
2569 lttng_ust_cleanup(0);
2570 /* Release mutexes and re-enable signals */
2571 ust_after_fork_common(restore_sigset);
2572 lttng_ust_ctor();
2573 }
2574
2575 void lttng_ust_after_setns(void)
2576 {
2577 ust_context_ns_reset();
2578 ust_context_vuids_reset();
2579 ust_context_vgids_reset();
2580 }
2581
2582 void lttng_ust_after_unshare(void)
2583 {
2584 ust_context_ns_reset();
2585 ust_context_vuids_reset();
2586 ust_context_vgids_reset();
2587 }
2588
2589 void lttng_ust_after_setuid(void)
2590 {
2591 ust_context_vuids_reset();
2592 }
2593
2594 void lttng_ust_after_seteuid(void)
2595 {
2596 ust_context_vuids_reset();
2597 }
2598
2599 void lttng_ust_after_setreuid(void)
2600 {
2601 ust_context_vuids_reset();
2602 }
2603
2604 void lttng_ust_after_setresuid(void)
2605 {
2606 ust_context_vuids_reset();
2607 }
2608
2609 void lttng_ust_after_setgid(void)
2610 {
2611 ust_context_vgids_reset();
2612 }
2613
2614 void lttng_ust_after_setegid(void)
2615 {
2616 ust_context_vgids_reset();
2617 }
2618
2619 void lttng_ust_after_setregid(void)
2620 {
2621 ust_context_vgids_reset();
2622 }
2623
2624 void lttng_ust_after_setresgid(void)
2625 {
2626 ust_context_vgids_reset();
2627 }
2628
2629 void lttng_ust_sockinfo_session_enabled(void *owner)
2630 {
2631 struct sock_info *sock_info = owner;
2632 sock_info->statedump_pending = 1;
2633 }
This page took 0.109518 seconds and 3 git commands to generate.