Fix: add urcu-bp TLS fixup
[lttng-ust.git] / liblttng-ust / lttng-ust-comm.c
CommitLineData
2691221a
MD
1/*
2 * lttng-ust-comm.c
3 *
4 * Copyright (C) 2011 David Goulet <david.goulet@polymtl.ca>
5 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
6 *
7 * This library is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; only
10 * version 2.1 of the License.
11 *
12 * This library is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
16 *
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with this library; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20 */
21
80e2814b 22#define _LGPL_SOURCE
2691221a
MD
23#include <sys/types.h>
24#include <sys/socket.h>
7fc90dca
MD
25#include <sys/mman.h>
26#include <sys/stat.h>
58d4b2a2
MD
27#include <sys/types.h>
28#include <sys/wait.h>
7fc90dca 29#include <fcntl.h>
2691221a
MD
30#include <unistd.h>
31#include <errno.h>
d9e99d10 32#include <pthread.h>
11ff9c7d
MD
33#include <semaphore.h>
34#include <time.h>
1ea11eab 35#include <assert.h>
e822f505 36#include <signal.h>
95259bd0 37#include <urcu/uatomic.h>
80e2814b 38#include <urcu/futex.h>
c117fb1b 39#include <urcu/compiler.h>
1ea11eab 40
4318ae1b 41#include <lttng/ust-events.h>
4318ae1b 42#include <lttng/ust-abi.h>
4318ae1b 43#include <lttng/ust.h>
7bc53e94 44#include <lttng/ust-error.h>
74d81a6c 45#include <lttng/ust-ctl.h>
8c90a710 46#include <urcu/tls-compat.h>
44c72f10
MD
47#include <ust-comm.h>
48#include <usterr-signal-safe.h>
cd54f6d9 49#include <helper.h>
44c72f10 50#include "tracepoint-internal.h"
7dd08bec 51#include "lttng-tracer-core.h"
08114193 52#include "compat.h"
f645cfa7 53#include "../libringbuffer/tlsfixup.h"
394598c1 54#include "lttng-ust-baddr.h"
edaa1431
MD
55
56/*
57 * Has lttng ust comm constructor been called ?
58 */
59static int initialized;
60
1ea11eab 61/*
17dfb34b
MD
62 * The ust_lock/ust_unlock lock is used as a communication thread mutex.
63 * Held when handling a command, also held by fork() to deal with
64 * removal of threads, and by exit path.
3327ac33
MD
65 *
66 * The UST lock is the centralized mutex across UST tracing control and
67 * probe registration.
68 *
69 * ust_exit_mutex must never nest in ust_mutex.
d58d1454 70 *
4770bd47
MD
71 * ust_fork_mutex must never nest in ust_mutex.
72 *
d58d1454
MD
73 * ust_mutex_nest is a per-thread nesting counter, allowing the perf
74 * counter lazy initialization called by events within the statedump,
75 * which traces while the ust_mutex is held.
4770bd47
MD
76 *
77 * ust_lock nests within the dynamic loader lock (within glibc) because
78 * it is taken within the library constructor.
3327ac33
MD
79 */
80static pthread_mutex_t ust_mutex = PTHREAD_MUTEX_INITIALIZER;
81
d58d1454
MD
82/* Allow nesting the ust_mutex within the same thread. */
83static DEFINE_URCU_TLS(int, ust_mutex_nest);
84
3327ac33
MD
85/*
86 * ust_exit_mutex protects thread_active variable wrt thread exit. It
87 * cannot be done by ust_mutex because pthread_cancel(), which takes an
88 * internal libc lock, cannot nest within ust_mutex.
89 *
90 * It never nests within a ust_mutex.
1ea11eab 91 */
3327ac33 92static pthread_mutex_t ust_exit_mutex = PTHREAD_MUTEX_INITIALIZER;
1ea11eab 93
458d678c
PW
94/*
95 * ust_fork_mutex protects base address statedump tracing against forks. It
96 * prevents the dynamic loader lock to be taken (by base address statedump
97 * tracing) while a fork is happening, thus preventing deadlock issues with
98 * the dynamic loader lock.
99 */
100static pthread_mutex_t ust_fork_mutex = PTHREAD_MUTEX_INITIALIZER;
101
1ea11eab
MD
102/* Should the ust comm thread quit ? */
103static int lttng_ust_comm_should_quit;
104
3327ac33 105/*
d58d1454 106 * Return 0 on success, -1 if should quit.
3327ac33 107 * The lock is taken in both cases.
d58d1454 108 * Signal-safe.
3327ac33
MD
109 */
110int ust_lock(void)
111{
d58d1454
MD
112 sigset_t sig_all_blocked, orig_mask;
113 int ret;
114
115 sigfillset(&sig_all_blocked);
116 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
117 if (ret) {
118 ERR("pthread_sigmask: %s", strerror(ret));
119 }
120 if (!URCU_TLS(ust_mutex_nest)++)
121 pthread_mutex_lock(&ust_mutex);
122 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
123 if (ret) {
124 ERR("pthread_sigmask: %s", strerror(ret));
125 }
3327ac33
MD
126 if (lttng_ust_comm_should_quit) {
127 return -1;
128 } else {
129 return 0;
130 }
131}
132
133/*
134 * ust_lock_nocheck() can be used in constructors/destructors, because
135 * they are already nested within the dynamic loader lock, and therefore
136 * have exclusive access against execution of liblttng-ust destructor.
d58d1454 137 * Signal-safe.
3327ac33
MD
138 */
139void ust_lock_nocheck(void)
140{
d58d1454
MD
141 sigset_t sig_all_blocked, orig_mask;
142 int ret;
143
144 sigfillset(&sig_all_blocked);
145 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
146 if (ret) {
147 ERR("pthread_sigmask: %s", strerror(ret));
148 }
149 if (!URCU_TLS(ust_mutex_nest)++)
150 pthread_mutex_lock(&ust_mutex);
151 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
152 if (ret) {
153 ERR("pthread_sigmask: %s", strerror(ret));
154 }
3327ac33
MD
155}
156
d58d1454
MD
157/*
158 * Signal-safe.
159 */
3327ac33
MD
160void ust_unlock(void)
161{
d58d1454
MD
162 sigset_t sig_all_blocked, orig_mask;
163 int ret;
164
165 sigfillset(&sig_all_blocked);
166 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
167 if (ret) {
168 ERR("pthread_sigmask: %s", strerror(ret));
169 }
170 if (!--URCU_TLS(ust_mutex_nest))
171 pthread_mutex_unlock(&ust_mutex);
172 ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
173 if (ret) {
174 ERR("pthread_sigmask: %s", strerror(ret));
175 }
3327ac33
MD
176}
177
11ff9c7d
MD
178/*
179 * Wait for either of these before continuing to the main
180 * program:
181 * - the register_done message from sessiond daemon
182 * (will let the sessiond daemon enable sessions before main
183 * starts.)
184 * - sessiond daemon is not reachable.
185 * - timeout (ensuring applications are resilient to session
186 * daemon problems).
187 */
188static sem_t constructor_wait;
950aab0c
MD
189/*
190 * Doing this for both the global and local sessiond.
191 */
95259bd0 192static int sem_count = { 2 };
11ff9c7d 193
e8508a49
MD
194/*
195 * Counting nesting within lttng-ust. Used to ensure that calling fork()
196 * from liblttng-ust does not execute the pre/post fork handlers.
197 */
8c90a710 198static DEFINE_URCU_TLS(int, lttng_ust_nest_count);
e8508a49 199
1ea11eab
MD
200/*
201 * Info about socket and associated listener thread.
202 */
203struct sock_info {
11ff9c7d 204 const char *name;
1ea11eab 205 pthread_t ust_listener; /* listener thread */
46050b1a 206 int root_handle;
8d20bf54
MD
207 int constructor_sem_posted;
208 int allowed;
44e073f5 209 int global;
e33f3265 210 int thread_active;
7fc90dca
MD
211
212 char sock_path[PATH_MAX];
213 int socket;
32ce8569 214 int notify_socket;
7fc90dca
MD
215
216 char wait_shm_path[PATH_MAX];
217 char *wait_shm_mmap;
37dddb65
MD
218 /* Keep track of lazy state dump not performed yet. */
219 int statedump_pending;
1ea11eab 220};
2691221a
MD
221
222/* Socket from app (connect) to session daemon (listen) for communication */
1ea11eab 223struct sock_info global_apps = {
11ff9c7d 224 .name = "global",
44e073f5 225 .global = 1,
7fc90dca 226
46050b1a 227 .root_handle = -1,
8d20bf54 228 .allowed = 1,
e33f3265 229 .thread_active = 0,
7fc90dca 230
32ce8569 231 .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME,
7fc90dca 232 .socket = -1,
32ce8569 233 .notify_socket = -1,
7fc90dca 234
32ce8569 235 .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME,
95c25348 236
37dddb65 237 .statedump_pending = 0,
1ea11eab 238};
2691221a
MD
239
240/* TODO: allow global_apps_sock_path override */
241
1ea11eab 242struct sock_info local_apps = {
11ff9c7d 243 .name = "local",
44e073f5 244 .global = 0,
46050b1a 245 .root_handle = -1,
8d20bf54 246 .allowed = 0, /* Check setuid bit first */
e33f3265 247 .thread_active = 0,
7fc90dca
MD
248
249 .socket = -1,
32ce8569 250 .notify_socket = -1,
95c25348 251
37dddb65 252 .statedump_pending = 0,
1ea11eab 253};
2691221a 254
37ed587a
MD
255static int wait_poll_fallback;
256
74d81a6c
MD
257static const char *cmd_name_mapping[] = {
258 [ LTTNG_UST_RELEASE ] = "Release",
259 [ LTTNG_UST_SESSION ] = "Create Session",
260 [ LTTNG_UST_TRACER_VERSION ] = "Get Tracer Version",
261
262 [ LTTNG_UST_TRACEPOINT_LIST ] = "Create Tracepoint List",
263 [ LTTNG_UST_WAIT_QUIESCENT ] = "Wait for Quiescent State",
264 [ LTTNG_UST_REGISTER_DONE ] = "Registration Done",
265 [ LTTNG_UST_TRACEPOINT_FIELD_LIST ] = "Create Tracepoint Field List",
266
267 /* Session FD commands */
268 [ LTTNG_UST_CHANNEL ] = "Create Channel",
269 [ LTTNG_UST_SESSION_START ] = "Start Session",
270 [ LTTNG_UST_SESSION_STOP ] = "Stop Session",
271
272 /* Channel FD commands */
273 [ LTTNG_UST_STREAM ] = "Create Stream",
274 [ LTTNG_UST_EVENT ] = "Create Event",
275
276 /* Event and Channel FD commands */
277 [ LTTNG_UST_CONTEXT ] = "Create Context",
278 [ LTTNG_UST_FLUSH_BUFFER ] = "Flush Buffer",
279
280 /* Event, Channel and Session commands */
281 [ LTTNG_UST_ENABLE ] = "Enable",
282 [ LTTNG_UST_DISABLE ] = "Disable",
283
284 /* Tracepoint list commands */
285 [ LTTNG_UST_TRACEPOINT_LIST_GET ] = "List Next Tracepoint",
286 [ LTTNG_UST_TRACEPOINT_FIELD_LIST_GET ] = "List Next Tracepoint Field",
287
288 /* Event FD commands */
289 [ LTTNG_UST_FILTER ] = "Create Filter",
75582b3a 290 [ LTTNG_UST_EXCLUSION ] = "Add exclusions to event",
74d81a6c
MD
291};
292
ff517991
MD
293static const char *str_timeout;
294static int got_timeout_env;
295
7dd08bec 296extern void lttng_ring_buffer_client_overwrite_init(void);
34a91bdb 297extern void lttng_ring_buffer_client_overwrite_rt_init(void);
7dd08bec 298extern void lttng_ring_buffer_client_discard_init(void);
34a91bdb 299extern void lttng_ring_buffer_client_discard_rt_init(void);
7dd08bec
MD
300extern void lttng_ring_buffer_metadata_client_init(void);
301extern void lttng_ring_buffer_client_overwrite_exit(void);
34a91bdb 302extern void lttng_ring_buffer_client_overwrite_rt_exit(void);
7dd08bec 303extern void lttng_ring_buffer_client_discard_exit(void);
34a91bdb 304extern void lttng_ring_buffer_client_discard_rt_exit(void);
7dd08bec 305extern void lttng_ring_buffer_metadata_client_exit(void);
edaa1431 306
3c6f6263
AM
307/*
308 * Returns the HOME directory path. Caller MUST NOT free(3) the returned
309 * pointer.
310 */
311static
312const char *get_lttng_home_dir(void)
313{
314 const char *val;
315
316 val = (const char *) getenv("LTTNG_HOME");
317 if (val != NULL) {
318 return val;
319 }
320 return (const char *) getenv("HOME");
321}
322
a903623f
MD
323/*
324 * Force a read (imply TLS fixup for dlopen) of TLS variables.
325 */
326static
327void lttng_fixup_nest_count_tls(void)
328{
8c90a710 329 asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count)));
a903623f
MD
330}
331
d58d1454
MD
332static
333void lttng_fixup_ust_mutex_nest_tls(void)
334{
335 asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest)));
336}
337
3df53fae
MD
338/*
339 * Fixup urcu bp TLS.
340 */
341static
342void lttng_fixup_urcu_bp_tls(void)
343{
344 rcu_read_lock();
345 rcu_read_unlock();
346}
347
32ce8569
MD
348int lttng_get_notify_socket(void *owner)
349{
350 struct sock_info *info = owner;
351
352 return info->notify_socket;
353}
354
74d81a6c
MD
355static
356void print_cmd(int cmd, int handle)
357{
358 const char *cmd_name = "Unknown";
359
fd67a004
MD
360 if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping)
361 && cmd_name_mapping[cmd]) {
74d81a6c
MD
362 cmd_name = cmd_name_mapping[cmd];
363 }
fd67a004
MD
364 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
365 cmd_name, cmd,
74d81a6c
MD
366 lttng_ust_obj_get_name(handle), handle);
367}
368
2691221a 369static
8d20bf54 370int setup_local_apps(void)
2691221a
MD
371{
372 const char *home_dir;
7fc90dca 373 uid_t uid;
2691221a 374
7fc90dca 375 uid = getuid();
8d20bf54
MD
376 /*
377 * Disallow per-user tracing for setuid binaries.
378 */
7fc90dca 379 if (uid != geteuid()) {
9ec6895c 380 assert(local_apps.allowed == 0);
d0a1ae63 381 return 0;
8d20bf54 382 }
3c6f6263 383 home_dir = get_lttng_home_dir();
9ec6895c
MD
384 if (!home_dir) {
385 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
386 assert(local_apps.allowed == 0);
2691221a 387 return -ENOENT;
9ec6895c
MD
388 }
389 local_apps.allowed = 1;
32ce8569
MD
390 snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
391 home_dir,
392 LTTNG_DEFAULT_HOME_RUNDIR,
393 LTTNG_UST_SOCK_FILENAME);
394 snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
395 LTTNG_UST_WAIT_FILENAME,
396 uid);
2691221a
MD
397 return 0;
398}
399
ff517991
MD
400/*
401 * Get notify_sock timeout, in ms.
402 * -1: don't wait. 0: wait forever. >0: timeout, in ms.
403 */
404static
405long get_timeout(void)
406{
407 long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
408
409 if (!got_timeout_env) {
410 str_timeout = getenv("LTTNG_UST_REGISTER_TIMEOUT");
411 got_timeout_env = 1;
412 }
413 if (str_timeout)
414 constructor_delay_ms = strtol(str_timeout, NULL, 10);
415 return constructor_delay_ms;
416}
417
418static
419long get_notify_sock_timeout(void)
420{
421 return get_timeout();
422}
423
424/*
425 * Return values: -1: don't wait. 0: wait forever. 1: timeout wait.
426 */
427static
428int get_constructor_timeout(struct timespec *constructor_timeout)
429{
430 long constructor_delay_ms;
431 int ret;
432
433 constructor_delay_ms = get_timeout();
434
435 switch (constructor_delay_ms) {
436 case -1:/* fall-through */
437 case 0:
438 return constructor_delay_ms;
439 default:
440 break;
441 }
442
443 /*
444 * If we are unable to find the current time, don't wait.
445 */
446 ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
447 if (ret) {
448 return -1;
449 }
450 constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
451 constructor_timeout->tv_nsec +=
452 (constructor_delay_ms % 1000UL) * 1000000UL;
453 if (constructor_timeout->tv_nsec >= 1000000000UL) {
454 constructor_timeout->tv_sec++;
455 constructor_timeout->tv_nsec -= 1000000000UL;
456 }
457 return 1;
458}
459
2691221a 460static
32ce8569 461int register_to_sessiond(int socket, enum ustctl_socket_type type)
2691221a 462{
32ce8569
MD
463 return ustcomm_send_reg_msg(socket,
464 type,
465 CAA_BITS_PER_LONG,
466 lttng_alignof(uint8_t) * CHAR_BIT,
467 lttng_alignof(uint16_t) * CHAR_BIT,
468 lttng_alignof(uint32_t) * CHAR_BIT,
469 lttng_alignof(uint64_t) * CHAR_BIT,
470 lttng_alignof(unsigned long) * CHAR_BIT);
2691221a
MD
471}
472
d9e99d10 473static
57773204 474int send_reply(int sock, struct ustcomm_ust_reply *lur)
d9e99d10 475{
9eb62b9c 476 ssize_t len;
d3a492d1 477
57773204 478 len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
d3a492d1 479 switch (len) {
a4be8962 480 case sizeof(*lur):
d3a492d1
MD
481 DBG("message successfully sent");
482 return 0;
7bc53e94
MD
483 default:
484 if (len == -ECONNRESET) {
485 DBG("remote end closed connection");
d3a492d1
MD
486 return 0;
487 }
7bc53e94
MD
488 if (len < 0)
489 return len;
490 DBG("incorrect message size: %zd", len);
491 return -EINVAL;
d3a492d1
MD
492 }
493}
494
495static
edaa1431 496int handle_register_done(struct sock_info *sock_info)
11ff9c7d
MD
497{
498 int ret;
499
edaa1431
MD
500 if (sock_info->constructor_sem_posted)
501 return 0;
502 sock_info->constructor_sem_posted = 1;
56cd7e2f
MD
503 if (uatomic_read(&sem_count) <= 0) {
504 return 0;
505 }
95259bd0
MD
506 ret = uatomic_add_return(&sem_count, -1);
507 if (ret == 0) {
508 ret = sem_post(&constructor_wait);
509 assert(!ret);
510 }
11ff9c7d
MD
511 return 0;
512}
513
37dddb65
MD
514/*
515 * Only execute pending statedump after the constructor semaphore has
516 * been posted by each listener thread. This means statedump will only
517 * be performed after the "registration done" command is received from
518 * each session daemon the application is connected to.
519 *
520 * This ensures we don't run into deadlock issues with the dynamic
521 * loader mutex, which is held while the constructor is called and
522 * waiting on the constructor semaphore. All operations requiring this
523 * dynamic loader lock need to be postponed using this mechanism.
524 */
525static
526void handle_pending_statedump(struct sock_info *sock_info)
527{
528 int ctor_passed = sock_info->constructor_sem_posted;
529
530 if (ctor_passed && sock_info->statedump_pending) {
531 sock_info->statedump_pending = 0;
2932a87f 532 pthread_mutex_lock(&ust_fork_mutex);
37dddb65 533 lttng_handle_pending_statedump(sock_info);
458d678c 534 pthread_mutex_unlock(&ust_fork_mutex);
37dddb65
MD
535 }
536}
537
11ff9c7d
MD
538static
539int handle_message(struct sock_info *sock_info,
57773204 540 int sock, struct ustcomm_ust_msg *lum)
d3a492d1 541{
1ea11eab 542 int ret = 0;
b61ce3b2 543 const struct lttng_ust_objd_ops *ops;
57773204 544 struct ustcomm_ust_reply lur;
ef9ff354 545 union ust_args args;
40003310 546 ssize_t len;
1ea11eab 547
46050b1a
MD
548 memset(&lur, 0, sizeof(lur));
549
3327ac33 550 if (ust_lock()) {
74d81a6c 551 ret = -LTTNG_UST_ERR_EXITING;
1ea11eab
MD
552 goto end;
553 }
9eb62b9c 554
46050b1a
MD
555 ops = objd_ops(lum->handle);
556 if (!ops) {
557 ret = -ENOENT;
558 goto end;
1ea11eab 559 }
46050b1a
MD
560
561 switch (lum->cmd) {
11ff9c7d
MD
562 case LTTNG_UST_REGISTER_DONE:
563 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
edaa1431 564 ret = handle_register_done(sock_info);
11ff9c7d
MD
565 else
566 ret = -EINVAL;
567 break;
46050b1a
MD
568 case LTTNG_UST_RELEASE:
569 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
570 ret = -EPERM;
571 else
1849ef7c 572 ret = lttng_ust_objd_unref(lum->handle, 1);
d9e99d10 573 break;
2d78951a
MD
574 case LTTNG_UST_FILTER:
575 {
576 /* Receive filter data */
f488575f 577 struct lttng_ust_filter_bytecode_node *bytecode;
2d78951a 578
cd54f6d9 579 if (lum->u.filter.data_size > FILTER_BYTECODE_MAX_LEN) {
7bc53e94 580 ERR("Filter data size is too large: %u bytes",
2d78951a
MD
581 lum->u.filter.data_size);
582 ret = -EINVAL;
583 goto error;
584 }
2734ca65 585
885b1dfd 586 if (lum->u.filter.reloc_offset > lum->u.filter.data_size) {
7bc53e94 587 ERR("Filter reloc offset %u is not within data",
2734ca65
CB
588 lum->u.filter.reloc_offset);
589 ret = -EINVAL;
590 goto error;
591 }
592
cd54f6d9
MD
593 bytecode = zmalloc(sizeof(*bytecode) + lum->u.filter.data_size);
594 if (!bytecode) {
595 ret = -ENOMEM;
596 goto error;
597 }
f488575f 598 len = ustcomm_recv_unix_sock(sock, bytecode->bc.data,
2d78951a
MD
599 lum->u.filter.data_size);
600 switch (len) {
601 case 0: /* orderly shutdown */
602 ret = 0;
cd54f6d9 603 free(bytecode);
2d78951a 604 goto error;
2d78951a
MD
605 default:
606 if (len == lum->u.filter.data_size) {
7bc53e94 607 DBG("filter data received");
2d78951a 608 break;
7bc53e94
MD
609 } else if (len < 0) {
610 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
611 if (len == -ECONNRESET) {
612 ERR("%s remote end closed connection", sock_info->name);
613 ret = len;
614 free(bytecode);
615 goto error;
616 }
617 ret = len;
eb8bf361 618 free(bytecode);
7bc53e94 619 goto end;
2d78951a 620 } else {
7bc53e94 621 DBG("incorrect filter data message size: %zd", len);
2d78951a 622 ret = -EINVAL;
cd54f6d9 623 free(bytecode);
2d78951a
MD
624 goto end;
625 }
626 }
f488575f
MD
627 bytecode->bc.len = lum->u.filter.data_size;
628 bytecode->bc.reloc_offset = lum->u.filter.reloc_offset;
3f6fd224 629 bytecode->bc.seqnum = lum->u.filter.seqnum;
cd54f6d9 630 if (ops->cmd) {
2d78951a 631 ret = ops->cmd(lum->handle, lum->cmd,
cd54f6d9 632 (unsigned long) bytecode,
f59ed768 633 &args, sock_info);
cd54f6d9
MD
634 if (ret) {
635 free(bytecode);
636 }
637 /* don't free bytecode if everything went fine. */
638 } else {
2d78951a 639 ret = -ENOSYS;
cd54f6d9
MD
640 free(bytecode);
641 }
2d78951a
MD
642 break;
643 }
86e36163
JI
644 case LTTNG_UST_EXCLUSION:
645 {
646 /* Receive exclusion names */
647 struct lttng_ust_excluder_node *node;
648 unsigned int count;
649
650 count = lum->u.exclusion.count;
651 if (count == 0) {
652 /* There are no names to read */
653 ret = 0;
654 goto error;
655 }
656 node = zmalloc(sizeof(*node) +
657 count * LTTNG_UST_SYM_NAME_LEN);
658 if (!node) {
659 ret = -ENOMEM;
660 goto error;
661 }
662 node->excluder.count = count;
663 len = ustcomm_recv_unix_sock(sock, node->excluder.names,
664 count * LTTNG_UST_SYM_NAME_LEN);
665 switch (len) {
666 case 0: /* orderly shutdown */
667 ret = 0;
668 free(node);
669 goto error;
670 default:
671 if (len == count * LTTNG_UST_SYM_NAME_LEN) {
672 DBG("Exclusion data received");
673 break;
674 } else if (len < 0) {
675 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
676 if (len == -ECONNRESET) {
677 ERR("%s remote end closed connection", sock_info->name);
678 ret = len;
679 free(node);
680 goto error;
681 }
682 ret = len;
683 free(node);
684 goto end;
685 } else {
686 DBG("Incorrect exclusion data message size: %zd", len);
687 ret = -EINVAL;
688 free(node);
689 goto end;
690 }
691 }
692 if (ops->cmd) {
693 ret = ops->cmd(lum->handle, lum->cmd,
694 (unsigned long) node,
695 &args, sock_info);
696 if (ret) {
697 free(node);
698 }
699 /* Don't free exclusion data if everything went fine. */
700 } else {
701 ret = -ENOSYS;
702 free(node);
703 }
704 break;
705 }
74d81a6c
MD
706 case LTTNG_UST_CHANNEL:
707 {
708 void *chan_data;
ff0f5728 709 int wakeup_fd;
74d81a6c
MD
710
711 len = ustcomm_recv_channel_from_sessiond(sock,
ff0f5728
MD
712 &chan_data, lum->u.channel.len,
713 &wakeup_fd);
74d81a6c
MD
714 switch (len) {
715 case 0: /* orderly shutdown */
716 ret = 0;
717 goto error;
718 default:
719 if (len == lum->u.channel.len) {
720 DBG("channel data received");
721 break;
722 } else if (len < 0) {
723 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
724 if (len == -ECONNRESET) {
725 ERR("%s remote end closed connection", sock_info->name);
726 ret = len;
727 goto error;
728 }
729 ret = len;
730 goto end;
731 } else {
732 DBG("incorrect channel data message size: %zd", len);
733 ret = -EINVAL;
734 goto end;
735 }
736 }
737 args.channel.chan_data = chan_data;
ff0f5728 738 args.channel.wakeup_fd = wakeup_fd;
74d81a6c
MD
739 if (ops->cmd)
740 ret = ops->cmd(lum->handle, lum->cmd,
741 (unsigned long) &lum->u,
742 &args, sock_info);
743 else
744 ret = -ENOSYS;
745 break;
746 }
747 case LTTNG_UST_STREAM:
748 {
749 /* Receive shm_fd, wakeup_fd */
750 ret = ustcomm_recv_stream_from_sessiond(sock,
751 &lum->u.stream.len,
752 &args.stream.shm_fd,
753 &args.stream.wakeup_fd);
754 if (ret) {
755 goto end;
756 }
757 if (ops->cmd)
758 ret = ops->cmd(lum->handle, lum->cmd,
759 (unsigned long) &lum->u,
760 &args, sock_info);
761 else
762 ret = -ENOSYS;
763 break;
764 }
d9e99d10 765 default:
46050b1a
MD
766 if (ops->cmd)
767 ret = ops->cmd(lum->handle, lum->cmd,
ef9ff354 768 (unsigned long) &lum->u,
f59ed768 769 &args, sock_info);
46050b1a
MD
770 else
771 ret = -ENOSYS;
772 break;
d9e99d10 773 }
46050b1a 774
1ea11eab 775end:
46050b1a
MD
776 lur.handle = lum->handle;
777 lur.cmd = lum->cmd;
778 lur.ret_val = ret;
779 if (ret >= 0) {
7bc53e94 780 lur.ret_code = LTTNG_UST_OK;
46050b1a 781 } else {
7bc53e94
MD
782 /*
783 * Use -LTTNG_UST_ERR as wildcard for UST internal
784 * error that are not caused by the transport, except if
785 * we already have a more precise error message to
786 * report.
787 */
64b2564e
DG
788 if (ret > -LTTNG_UST_ERR) {
789 /* Translate code to UST error. */
790 switch (ret) {
791 case -EEXIST:
792 lur.ret_code = -LTTNG_UST_ERR_EXIST;
793 break;
794 case -EINVAL:
795 lur.ret_code = -LTTNG_UST_ERR_INVAL;
796 break;
797 case -ENOENT:
798 lur.ret_code = -LTTNG_UST_ERR_NOENT;
799 break;
800 case -EPERM:
801 lur.ret_code = -LTTNG_UST_ERR_PERM;
802 break;
803 case -ENOSYS:
804 lur.ret_code = -LTTNG_UST_ERR_NOSYS;
805 break;
806 default:
807 lur.ret_code = -LTTNG_UST_ERR;
808 break;
809 }
810 } else {
7bc53e94 811 lur.ret_code = ret;
64b2564e 812 }
46050b1a 813 }
e6ea14c5
MD
814 if (ret >= 0) {
815 switch (lum->cmd) {
e6ea14c5
MD
816 case LTTNG_UST_TRACER_VERSION:
817 lur.u.version = lum->u.version;
818 break;
819 case LTTNG_UST_TRACEPOINT_LIST_GET:
820 memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
821 break;
822 }
381c0f1e 823 }
74d81a6c 824 DBG("Return value: %d", lur.ret_val);
46050b1a 825 ret = send_reply(sock, &lur);
193183fb 826 if (ret < 0) {
7bc53e94 827 DBG("error sending reply");
193183fb
MD
828 goto error;
829 }
46050b1a 830
40003310
MD
831 /*
832 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
833 * after the reply.
834 */
7bc53e94 835 if (lur.ret_code == LTTNG_UST_OK) {
40003310
MD
836 switch (lum->cmd) {
837 case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET:
838 len = ustcomm_send_unix_sock(sock,
839 &args.field_list.entry,
840 sizeof(args.field_list.entry));
7bc53e94
MD
841 if (len < 0) {
842 ret = len;
843 goto error;
844 }
40003310 845 if (len != sizeof(args.field_list.entry)) {
7bc53e94 846 ret = -EINVAL;
40003310
MD
847 goto error;
848 }
849 }
850 }
ef9ff354 851
381c0f1e 852error:
17dfb34b 853 ust_unlock();
d9e99d10 854
37dddb65
MD
855 /*
856 * Performed delayed statedump operations outside of the UST
857 * lock. We need to take the dynamic loader lock before we take
858 * the UST lock internally within handle_pending_statedump().
859 */
860 handle_pending_statedump(sock_info);
246be17e 861
37dddb65 862 return ret;
246be17e
PW
863}
864
46050b1a 865static
efe0de09 866void cleanup_sock_info(struct sock_info *sock_info, int exiting)
46050b1a
MD
867{
868 int ret;
869
5b14aab3
MD
870 if (sock_info->root_handle != -1) {
871 ret = lttng_ust_objd_unref(sock_info->root_handle, 1);
872 if (ret) {
873 ERR("Error unref root handle");
874 }
875 sock_info->root_handle = -1;
876 }
877 sock_info->constructor_sem_posted = 0;
878
879 /*
880 * wait_shm_mmap, socket and notify socket are used by listener
881 * threads outside of the ust lock, so we cannot tear them down
882 * ourselves, because we cannot join on these threads. Leave
883 * responsibility of cleaning up these resources to the OS
884 * process exit.
885 */
886 if (exiting)
887 return;
888
46050b1a 889 if (sock_info->socket != -1) {
e6973a89 890 ret = ustcomm_close_unix_sock(sock_info->socket);
46050b1a 891 if (ret) {
32ce8569 892 ERR("Error closing ust cmd socket");
46050b1a
MD
893 }
894 sock_info->socket = -1;
895 }
32ce8569
MD
896 if (sock_info->notify_socket != -1) {
897 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
898 if (ret) {
899 ERR("Error closing ust notify socket");
900 }
901 sock_info->notify_socket = -1;
902 }
5b14aab3 903 if (sock_info->wait_shm_mmap) {
172d6b68
MD
904 long page_size;
905
906 page_size = sysconf(_SC_PAGE_SIZE);
907 if (page_size > 0) {
908 ret = munmap(sock_info->wait_shm_mmap, page_size);
909 if (ret) {
910 ERR("Error unmapping wait shm");
911 }
7fc90dca
MD
912 }
913 sock_info->wait_shm_mmap = NULL;
914 }
915}
916
58d4b2a2 917/*
33bbeb90
MD
918 * Using fork to set umask in the child process (not multi-thread safe).
919 * We deal with the shm_open vs ftruncate race (happening when the
920 * sessiond owns the shm and does not let everybody modify it, to ensure
921 * safety against shm_unlink) by simply letting the mmap fail and
922 * retrying after a few seconds.
923 * For global shm, everybody has rw access to it until the sessiond
924 * starts.
58d4b2a2 925 */
7fc90dca 926static
58d4b2a2 927int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
7fc90dca 928{
7fc90dca 929 int wait_shm_fd, ret;
58d4b2a2 930 pid_t pid;
44e073f5 931
58d4b2a2 932 /*
33bbeb90 933 * Try to open read-only.
58d4b2a2 934 */
33bbeb90 935 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
58d4b2a2 936 if (wait_shm_fd >= 0) {
7aa76730
MD
937 int32_t tmp_read;
938 ssize_t len;
939 size_t bytes_read = 0;
940
941 /*
942 * Try to read the fd. If unable to do so, try opening
943 * it in write mode.
944 */
945 do {
946 len = read(wait_shm_fd,
947 &((char *) &tmp_read)[bytes_read],
948 sizeof(tmp_read) - bytes_read);
949 if (len > 0) {
950 bytes_read += len;
951 }
952 } while ((len < 0 && errno == EINTR)
953 || (len > 0 && bytes_read < sizeof(tmp_read)));
954 if (bytes_read != sizeof(tmp_read)) {
955 ret = close(wait_shm_fd);
956 if (ret) {
957 ERR("close wait_shm_fd");
958 }
959 goto open_write;
960 }
58d4b2a2
MD
961 goto end;
962 } else if (wait_shm_fd < 0 && errno != ENOENT) {
963 /*
33bbeb90
MD
964 * Real-only open did not work, and it's not because the
965 * entry was not present. It's a failure that prohibits
966 * using shm.
58d4b2a2 967 */
7fc90dca 968 ERR("Error opening shm %s", sock_info->wait_shm_path);
58d4b2a2 969 goto end;
7fc90dca 970 }
7aa76730
MD
971
972open_write:
7fc90dca 973 /*
7aa76730
MD
974 * If the open failed because the file did not exist, or because
975 * the file was not truncated yet, try creating it ourself.
7fc90dca 976 */
8c90a710 977 URCU_TLS(lttng_ust_nest_count)++;
58d4b2a2 978 pid = fork();
8c90a710 979 URCU_TLS(lttng_ust_nest_count)--;
58d4b2a2
MD
980 if (pid > 0) {
981 int status;
982
983 /*
984 * Parent: wait for child to return, in which case the
985 * shared memory map will have been created.
986 */
987 pid = wait(&status);
b7d3cb32 988 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
58d4b2a2
MD
989 wait_shm_fd = -1;
990 goto end;
7fc90dca 991 }
58d4b2a2
MD
992 /*
993 * Try to open read-only again after creation.
994 */
33bbeb90 995 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
58d4b2a2
MD
996 if (wait_shm_fd < 0) {
997 /*
998 * Real-only open did not work. It's a failure
999 * that prohibits using shm.
1000 */
1001 ERR("Error opening shm %s", sock_info->wait_shm_path);
1002 goto end;
1003 }
1004 goto end;
1005 } else if (pid == 0) {
1006 int create_mode;
1007
1008 /* Child */
33bbeb90 1009 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
58d4b2a2 1010 if (sock_info->global)
33bbeb90 1011 create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
58d4b2a2
MD
1012 /*
1013 * We're alone in a child process, so we can modify the
1014 * process-wide umask.
1015 */
33bbeb90 1016 umask(~create_mode);
58d4b2a2 1017 /*
33bbeb90
MD
1018 * Try creating shm (or get rw access).
1019 * We don't do an exclusive open, because we allow other
1020 * processes to create+ftruncate it concurrently.
58d4b2a2
MD
1021 */
1022 wait_shm_fd = shm_open(sock_info->wait_shm_path,
1023 O_RDWR | O_CREAT, create_mode);
1024 if (wait_shm_fd >= 0) {
1025 ret = ftruncate(wait_shm_fd, mmap_size);
1026 if (ret) {
1027 PERROR("ftruncate");
b0c1425d 1028 _exit(EXIT_FAILURE);
58d4b2a2 1029 }
b0c1425d 1030 _exit(EXIT_SUCCESS);
58d4b2a2 1031 }
33bbeb90
MD
1032 /*
1033 * For local shm, we need to have rw access to accept
1034 * opening it: this means the local sessiond will be
1035 * able to wake us up. For global shm, we open it even
1036 * if rw access is not granted, because the root.root
1037 * sessiond will be able to override all rights and wake
1038 * us up.
1039 */
1040 if (!sock_info->global && errno != EACCES) {
58d4b2a2 1041 ERR("Error opening shm %s", sock_info->wait_shm_path);
5d3bc5ed 1042 _exit(EXIT_FAILURE);
58d4b2a2
MD
1043 }
1044 /*
33bbeb90
MD
1045 * The shm exists, but we cannot open it RW. Report
1046 * success.
58d4b2a2 1047 */
5d3bc5ed 1048 _exit(EXIT_SUCCESS);
58d4b2a2
MD
1049 } else {
1050 return -1;
7fc90dca 1051 }
58d4b2a2 1052end:
33bbeb90
MD
1053 if (wait_shm_fd >= 0 && !sock_info->global) {
1054 struct stat statbuf;
1055
1056 /*
1057 * Ensure that our user is the owner of the shm file for
1058 * local shm. If we do not own the file, it means our
1059 * sessiond will not have access to wake us up (there is
1060 * probably a rogue process trying to fake our
1061 * sessiond). Fallback to polling method in this case.
1062 */
1063 ret = fstat(wait_shm_fd, &statbuf);
1064 if (ret) {
1065 PERROR("fstat");
1066 goto error_close;
1067 }
1068 if (statbuf.st_uid != getuid())
1069 goto error_close;
1070 }
58d4b2a2 1071 return wait_shm_fd;
33bbeb90
MD
1072
1073error_close:
1074 ret = close(wait_shm_fd);
1075 if (ret) {
1076 PERROR("Error closing fd");
1077 }
1078 return -1;
58d4b2a2
MD
1079}
1080
1081static
1082char *get_map_shm(struct sock_info *sock_info)
1083{
172d6b68 1084 long page_size;
58d4b2a2
MD
1085 int wait_shm_fd, ret;
1086 char *wait_shm_mmap;
1087
172d6b68
MD
1088 page_size = sysconf(_SC_PAGE_SIZE);
1089 if (page_size < 0) {
1090 goto error;
1091 }
1092
1093 wait_shm_fd = get_wait_shm(sock_info, page_size);
58d4b2a2
MD
1094 if (wait_shm_fd < 0) {
1095 goto error;
44e073f5 1096 }
172d6b68 1097 wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
7fc90dca 1098 MAP_SHARED, wait_shm_fd, 0);
7fc90dca
MD
1099 /* close shm fd immediately after taking the mmap reference */
1100 ret = close(wait_shm_fd);
1101 if (ret) {
33bbeb90
MD
1102 PERROR("Error closing fd");
1103 }
1104 if (wait_shm_mmap == MAP_FAILED) {
1105 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1106 goto error;
7fc90dca
MD
1107 }
1108 return wait_shm_mmap;
1109
1110error:
1111 return NULL;
1112}
1113
1114static
1115void wait_for_sessiond(struct sock_info *sock_info)
1116{
efe0de09 1117 int ret;
80e2814b 1118
3327ac33 1119 if (ust_lock()) {
7fc90dca
MD
1120 goto quit;
1121 }
37ed587a
MD
1122 if (wait_poll_fallback) {
1123 goto error;
1124 }
7fc90dca
MD
1125 if (!sock_info->wait_shm_mmap) {
1126 sock_info->wait_shm_mmap = get_map_shm(sock_info);
1127 if (!sock_info->wait_shm_mmap)
1128 goto error;
1129 }
1130 ust_unlock();
1131
1132 DBG("Waiting for %s apps sessiond", sock_info->name);
80e2814b
MD
1133 /* Wait for futex wakeup */
1134 if (uatomic_read((int32_t *) sock_info->wait_shm_mmap) == 0) {
1135 ret = futex_async((int32_t *) sock_info->wait_shm_mmap,
1136 FUTEX_WAIT, 0, NULL, NULL, 0);
80e2814b 1137 if (ret < 0) {
37ed587a
MD
1138 if (errno == EFAULT) {
1139 wait_poll_fallback = 1;
a8b870ad 1140 DBG(
37ed587a
MD
1141"Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1142"do not support FUTEX_WAKE on read-only memory mappings correctly. "
1143"Please upgrade your kernel "
1144"(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1145"mainline). LTTng-UST will use polling mode fallback.");
cd27263b
MD
1146 if (ust_debug())
1147 PERROR("futex");
37ed587a 1148 }
80e2814b
MD
1149 }
1150 }
7fc90dca
MD
1151 return;
1152
1153quit:
1154 ust_unlock();
1155 return;
1156
1157error:
1158 ust_unlock();
7fc90dca 1159 return;
46050b1a
MD
1160}
1161
1ea11eab
MD
1162/*
1163 * This thread does not allocate any resource, except within
1164 * handle_message, within mutex protection. This mutex protects against
1165 * fork and exit.
98bf993f 1166 * The other moment it allocates resources is at socket connection, which
1ea11eab
MD
1167 * is also protected by the mutex.
1168 */
d9e99d10
MD
1169static
1170void *ust_listener_thread(void *arg)
1171{
1ea11eab 1172 struct sock_info *sock_info = arg;
c0eedf81 1173 int sock, ret, prev_connect_failed = 0, has_waited = 0;
ff517991 1174 long timeout;
d9e99d10 1175
9eb62b9c
MD
1176 /* Restart trying to connect to the session daemon */
1177restart:
c0eedf81
MD
1178 if (prev_connect_failed) {
1179 /* Wait for sessiond availability with pipe */
1180 wait_for_sessiond(sock_info);
1181 if (has_waited) {
1182 has_waited = 0;
1183 /*
1184 * Sleep for 5 seconds before retrying after a
1185 * sequence of failure / wait / failure. This
1186 * deals with a killed or broken session daemon.
1187 */
1188 sleep(5);
1189 }
1190 has_waited = 1;
1191 prev_connect_failed = 0;
1192 }
9eb62b9c 1193
1ea11eab 1194 if (sock_info->socket != -1) {
e6973a89 1195 ret = ustcomm_close_unix_sock(sock_info->socket);
1ea11eab 1196 if (ret) {
32ce8569
MD
1197 ERR("Error closing %s ust cmd socket",
1198 sock_info->name);
1ea11eab
MD
1199 }
1200 sock_info->socket = -1;
1201 }
32ce8569
MD
1202 if (sock_info->notify_socket != -1) {
1203 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1204 if (ret) {
1205 ERR("Error closing %s ust notify socket",
1206 sock_info->name);
1207 }
1208 sock_info->notify_socket = -1;
1209 }
46050b1a 1210
321f2351
MD
1211 /*
1212 * Register. We need to perform both connect and sending
1213 * registration message before doing the next connect otherwise
1214 * we may reach unix socket connect queue max limits and block
1215 * on the 2nd connect while the session daemon is awaiting the
1216 * first connect registration message.
1217 */
1218 /* Connect cmd socket */
1219 ret = ustcomm_connect_unix_sock(sock_info->sock_path);
1220 if (ret < 0) {
1221 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1222 prev_connect_failed = 1;
5b14aab3 1223
3327ac33 1224 if (ust_lock()) {
321f2351 1225 goto quit;
32ce8569 1226 }
46050b1a 1227
e3426ddc 1228 /*
321f2351
MD
1229 * If we cannot find the sessiond daemon, don't delay
1230 * constructor execution.
e3426ddc 1231 */
321f2351
MD
1232 ret = handle_register_done(sock_info);
1233 assert(!ret);
1234 ust_unlock();
1235 goto restart;
27fe9f21 1236 }
321f2351 1237 sock_info->socket = ret;
27fe9f21 1238
3327ac33 1239 if (ust_lock()) {
5b14aab3
MD
1240 goto quit;
1241 }
1242
46050b1a
MD
1243 /*
1244 * Create only one root handle per listener thread for the whole
f59ed768
MD
1245 * process lifetime, so we ensure we get ID which is statically
1246 * assigned to the root handle.
46050b1a
MD
1247 */
1248 if (sock_info->root_handle == -1) {
1249 ret = lttng_abi_create_root_handle();
a51070bb 1250 if (ret < 0) {
46050b1a 1251 ERR("Error creating root handle");
46050b1a
MD
1252 goto quit;
1253 }
1254 sock_info->root_handle = ret;
9eb62b9c 1255 }
1ea11eab 1256
32ce8569 1257 ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD);
9eb62b9c 1258 if (ret < 0) {
32ce8569
MD
1259 ERR("Error registering to %s ust cmd socket",
1260 sock_info->name);
c0eedf81 1261 prev_connect_failed = 1;
11ff9c7d
MD
1262 /*
1263 * If we cannot register to the sessiond daemon, don't
1264 * delay constructor execution.
1265 */
edaa1431 1266 ret = handle_register_done(sock_info);
11ff9c7d 1267 assert(!ret);
17dfb34b 1268 ust_unlock();
9eb62b9c
MD
1269 goto restart;
1270 }
321f2351
MD
1271
1272 ust_unlock();
1273
1274 /* Connect notify socket */
1275 ret = ustcomm_connect_unix_sock(sock_info->sock_path);
1276 if (ret < 0) {
1277 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1278 prev_connect_failed = 1;
1279
3327ac33 1280 if (ust_lock()) {
321f2351
MD
1281 goto quit;
1282 }
1283
1284 /*
1285 * If we cannot find the sessiond daemon, don't delay
1286 * constructor execution.
1287 */
1288 ret = handle_register_done(sock_info);
1289 assert(!ret);
1290 ust_unlock();
1291 goto restart;
1292 }
1293 sock_info->notify_socket = ret;
1294
1295 timeout = get_notify_sock_timeout();
1296 if (timeout >= 0) {
1297 /*
1298 * Give at least 10ms to sessiond to reply to
1299 * notifications.
1300 */
1301 if (timeout < 10)
1302 timeout = 10;
1303 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
1304 timeout);
1305 if (ret < 0) {
1306 WARN("Error setting socket receive timeout");
1307 }
1308 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
1309 timeout);
1310 if (ret < 0) {
1311 WARN("Error setting socket send timeout");
1312 }
1313 } else if (timeout < -1) {
1314 WARN("Unsupported timeout value %ld", timeout);
1315 }
1316
3327ac33 1317 if (ust_lock()) {
321f2351
MD
1318 goto quit;
1319 }
1320
32ce8569
MD
1321 ret = register_to_sessiond(sock_info->notify_socket,
1322 USTCTL_SOCKET_NOTIFY);
1323 if (ret < 0) {
1324 ERR("Error registering to %s ust notify socket",
1325 sock_info->name);
1326 prev_connect_failed = 1;
1327 /*
1328 * If we cannot register to the sessiond daemon, don't
1329 * delay constructor execution.
1330 */
1331 ret = handle_register_done(sock_info);
1332 assert(!ret);
1333 ust_unlock();
1334 goto restart;
1335 }
1336 sock = sock_info->socket;
1337
17dfb34b 1338 ust_unlock();
46050b1a 1339
d9e99d10
MD
1340 for (;;) {
1341 ssize_t len;
57773204 1342 struct ustcomm_ust_msg lum;
d9e99d10 1343
57773204 1344 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
d9e99d10
MD
1345 switch (len) {
1346 case 0: /* orderly shutdown */
7dd08bec 1347 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
3327ac33 1348 if (ust_lock()) {
d5e1fea6
MD
1349 goto quit;
1350 }
8236ba10
MD
1351 /*
1352 * Either sessiond has shutdown or refused us by closing the socket.
1353 * In either case, we don't want to delay construction execution,
1354 * and we need to wait before retry.
1355 */
1356 prev_connect_failed = 1;
1357 /*
1358 * If we cannot register to the sessiond daemon, don't
1359 * delay constructor execution.
1360 */
1361 ret = handle_register_done(sock_info);
1362 assert(!ret);
1363 ust_unlock();
d9e99d10 1364 goto end;
e7723462 1365 case sizeof(lum):
74d81a6c 1366 print_cmd(lum.cmd, lum.handle);
11ff9c7d 1367 ret = handle_message(sock_info, sock, &lum);
7bc53e94 1368 if (ret) {
11ff9c7d 1369 ERR("Error handling message for %s socket", sock_info->name);
d9e99d10
MD
1370 }
1371 continue;
7bc53e94
MD
1372 default:
1373 if (len < 0) {
1374 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1375 } else {
1376 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
1377 }
1378 if (len == -ECONNRESET) {
1379 DBG("%s remote end closed connection", sock_info->name);
d9e99d10
MD
1380 goto end;
1381 }
1382 goto end;
d9e99d10
MD
1383 }
1384
1385 }
1386end:
3327ac33 1387 if (ust_lock()) {
d5e1fea6
MD
1388 goto quit;
1389 }
f59ed768
MD
1390 /* Cleanup socket handles before trying to reconnect */
1391 lttng_ust_objd_table_owner_cleanup(sock_info);
1392 ust_unlock();
9eb62b9c 1393 goto restart; /* try to reconnect */
e33f3265 1394
1ea11eab 1395quit:
e33f3265 1396 ust_unlock();
3327ac33
MD
1397
1398 pthread_mutex_lock(&ust_exit_mutex);
1399 sock_info->thread_active = 0;
1400 pthread_mutex_unlock(&ust_exit_mutex);
d9e99d10
MD
1401 return NULL;
1402}
1403
2594a5b4
MD
1404/*
1405 * Weak symbol to call when the ust malloc wrapper is not loaded.
1406 */
1407__attribute__((weak))
1408void lttng_ust_malloc_wrapper_init(void)
1409{
1410}
1411
2691221a
MD
1412/*
1413 * sessiond monitoring thread: monitor presence of global and per-user
1414 * sessiond by polling the application common named pipe.
1415 */
edaa1431 1416void __attribute__((constructor)) lttng_ust_init(void)
2691221a 1417{
11ff9c7d 1418 struct timespec constructor_timeout;
ae6a58bf 1419 sigset_t sig_all_blocked, orig_parent_mask;
1879f67f 1420 pthread_attr_t thread_attr;
cf12a773 1421 int timeout_mode;
2691221a
MD
1422 int ret;
1423
edaa1431
MD
1424 if (uatomic_xchg(&initialized, 1) == 1)
1425 return;
1426
eddd8d5d
MD
1427 /*
1428 * Fixup interdependency between TLS fixup mutex (which happens
1429 * to be the dynamic linker mutex) and ust_lock, taken within
1430 * the ust lock.
1431 */
3df53fae 1432 lttng_fixup_urcu_bp_tls();
f645cfa7 1433 lttng_fixup_ringbuffer_tls();
4158a15a 1434 lttng_fixup_vtid_tls();
a903623f 1435 lttng_fixup_nest_count_tls();
009745db 1436 lttng_fixup_procname_tls();
d58d1454 1437 lttng_fixup_ust_mutex_nest_tls();
eddd8d5d 1438
edaa1431
MD
1439 /*
1440 * We want precise control over the order in which we construct
1441 * our sub-libraries vs starting to receive commands from
1442 * sessiond (otherwise leading to errors when trying to create
1443 * sessiond before the init functions are completed).
1444 */
2691221a 1445 init_usterr();
edaa1431 1446 init_tracepoint();
bd703713 1447 lttng_ust_baddr_statedump_init();
7dd08bec
MD
1448 lttng_ring_buffer_metadata_client_init();
1449 lttng_ring_buffer_client_overwrite_init();
34a91bdb 1450 lttng_ring_buffer_client_overwrite_rt_init();
7dd08bec 1451 lttng_ring_buffer_client_discard_init();
34a91bdb 1452 lttng_ring_buffer_client_discard_rt_init();
d58d1454 1453 lttng_perf_counter_init();
a0a3bef9 1454 lttng_context_init();
2594a5b4
MD
1455 /*
1456 * Invoke ust malloc wrapper init before starting other threads.
1457 */
1458 lttng_ust_malloc_wrapper_init();
2691221a 1459
ff517991 1460 timeout_mode = get_constructor_timeout(&constructor_timeout);
11ff9c7d 1461
95259bd0 1462 ret = sem_init(&constructor_wait, 0, 0);
11ff9c7d
MD
1463 assert(!ret);
1464
8d20bf54 1465 ret = setup_local_apps();
2691221a 1466 if (ret) {
9ec6895c 1467 DBG("local apps setup returned %d", ret);
2691221a 1468 }
ae6a58bf
WP
1469
1470 /* A new thread created by pthread_create inherits the signal mask
1471 * from the parent. To avoid any signal being received by the
1472 * listener thread, we block all signals temporarily in the parent,
1473 * while we create the listener thread.
1474 */
1475 sigfillset(&sig_all_blocked);
1476 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
1477 if (ret) {
d94d802c 1478 ERR("pthread_sigmask: %s", strerror(ret));
ae6a58bf
WP
1479 }
1480
1879f67f
MG
1481 ret = pthread_attr_init(&thread_attr);
1482 if (ret) {
1483 ERR("pthread_attr_init: %s", strerror(ret));
1484 }
1485 ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
1486 if (ret) {
1487 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
1488 }
1489
c0bbbd5a 1490 pthread_mutex_lock(&ust_exit_mutex);
1879f67f 1491 ret = pthread_create(&global_apps.ust_listener, &thread_attr,
dde70ea0 1492 ust_listener_thread, &global_apps);
d94d802c
MD
1493 if (ret) {
1494 ERR("pthread_create global: %s", strerror(ret));
1495 }
e33f3265 1496 global_apps.thread_active = 1;
c0bbbd5a 1497 pthread_mutex_unlock(&ust_exit_mutex);
e33f3265 1498
8d20bf54 1499 if (local_apps.allowed) {
c0bbbd5a 1500 pthread_mutex_lock(&ust_exit_mutex);
1879f67f 1501 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
dde70ea0 1502 ust_listener_thread, &local_apps);
d94d802c
MD
1503 if (ret) {
1504 ERR("pthread_create local: %s", strerror(ret));
1505 }
e33f3265 1506 local_apps.thread_active = 1;
c0bbbd5a 1507 pthread_mutex_unlock(&ust_exit_mutex);
8d20bf54
MD
1508 } else {
1509 handle_register_done(&local_apps);
1510 }
1879f67f
MG
1511 ret = pthread_attr_destroy(&thread_attr);
1512 if (ret) {
1513 ERR("pthread_attr_destroy: %s", strerror(ret));
1514 }
8d20bf54 1515
ae6a58bf
WP
1516 /* Restore original signal mask in parent */
1517 ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
1518 if (ret) {
d94d802c 1519 ERR("pthread_sigmask: %s", strerror(ret));
ae6a58bf
WP
1520 }
1521
cf12a773
MD
1522 switch (timeout_mode) {
1523 case 1: /* timeout wait */
95259bd0
MD
1524 do {
1525 ret = sem_timedwait(&constructor_wait,
1526 &constructor_timeout);
1527 } while (ret < 0 && errno == EINTR);
cf12a773 1528 if (ret < 0 && errno == ETIMEDOUT) {
7dd08bec 1529 ERR("Timed out waiting for lttng-sessiond");
cf12a773
MD
1530 } else {
1531 assert(!ret);
1532 }
1533 break;
7b766b16 1534 case -1:/* wait forever */
95259bd0
MD
1535 do {
1536 ret = sem_wait(&constructor_wait);
1537 } while (ret < 0 && errno == EINTR);
11ff9c7d 1538 assert(!ret);
cf12a773 1539 break;
7b766b16 1540 case 0: /* no timeout */
cf12a773 1541 break;
11ff9c7d 1542 }
2691221a
MD
1543}
1544
17dfb34b
MD
1545static
1546void lttng_ust_cleanup(int exiting)
1547{
efe0de09 1548 cleanup_sock_info(&global_apps, exiting);
17dfb34b 1549 if (local_apps.allowed) {
efe0de09 1550 cleanup_sock_info(&local_apps, exiting);
17dfb34b 1551 }
efe0de09
MD
1552 /*
1553 * The teardown in this function all affect data structures
1554 * accessed under the UST lock by the listener thread. This
1555 * lock, along with the lttng_ust_comm_should_quit flag, ensure
1556 * that none of these threads are accessing this data at this
1557 * point.
1558 */
17dfb34b 1559 lttng_ust_abi_exit();
003fedf4 1560 lttng_ust_events_exit();
a0a3bef9 1561 lttng_context_exit();
d58d1454 1562 lttng_perf_counter_exit();
34a91bdb 1563 lttng_ring_buffer_client_discard_rt_exit();
7dd08bec 1564 lttng_ring_buffer_client_discard_exit();
34a91bdb 1565 lttng_ring_buffer_client_overwrite_rt_exit();
7dd08bec
MD
1566 lttng_ring_buffer_client_overwrite_exit();
1567 lttng_ring_buffer_metadata_client_exit();
bd703713 1568 lttng_ust_baddr_statedump_destroy();
17dfb34b
MD
1569 exit_tracepoint();
1570 if (!exiting) {
1571 /* Reinitialize values for fork */
1572 sem_count = 2;
1573 lttng_ust_comm_should_quit = 0;
1574 initialized = 0;
1575 }
1576}
1577
edaa1431 1578void __attribute__((destructor)) lttng_ust_exit(void)
2691221a
MD
1579{
1580 int ret;
1581
9eb62b9c
MD
1582 /*
1583 * Using pthread_cancel here because:
1584 * A) we don't want to hang application teardown.
1585 * B) the thread is not allocating any resource.
1586 */
1ea11eab
MD
1587
1588 /*
1589 * Require the communication thread to quit. Synchronize with
1590 * mutexes to ensure it is not in a mutex critical section when
1591 * pthread_cancel is later called.
1592 */
3327ac33 1593 ust_lock_nocheck();
1ea11eab 1594 lttng_ust_comm_should_quit = 1;
3327ac33 1595 ust_unlock();
1ea11eab 1596
3327ac33 1597 pthread_mutex_lock(&ust_exit_mutex);
f5f94532 1598 /* cancel threads */
e33f3265
MD
1599 if (global_apps.thread_active) {
1600 ret = pthread_cancel(global_apps.ust_listener);
1601 if (ret) {
1602 ERR("Error cancelling global ust listener thread: %s",
1603 strerror(ret));
1604 } else {
1605 global_apps.thread_active = 0;
1606 }
2691221a 1607 }
e33f3265 1608 if (local_apps.thread_active) {
8d20bf54
MD
1609 ret = pthread_cancel(local_apps.ust_listener);
1610 if (ret) {
d94d802c
MD
1611 ERR("Error cancelling local ust listener thread: %s",
1612 strerror(ret));
e33f3265
MD
1613 } else {
1614 local_apps.thread_active = 0;
8d20bf54 1615 }
8d20bf54 1616 }
3327ac33 1617 pthread_mutex_unlock(&ust_exit_mutex);
e33f3265 1618
efe0de09
MD
1619 /*
1620 * Do NOT join threads: use of sys_futex makes it impossible to
1621 * join the threads without using async-cancel, but async-cancel
1622 * is delivered by a signal, which could hit the target thread
1623 * anywhere in its code path, including while the ust_lock() is
1624 * held, causing a deadlock for the other thread. Let the OS
1625 * cleanup the threads if there are stalled in a syscall.
1626 */
17dfb34b 1627 lttng_ust_cleanup(1);
2691221a 1628}
e822f505
MD
1629
1630/*
1631 * We exclude the worker threads across fork and clone (except
1632 * CLONE_VM), because these system calls only keep the forking thread
1633 * running in the child. Therefore, we don't want to call fork or clone
1634 * in the middle of an tracepoint or ust tracing state modification.
1635 * Holding this mutex protects these structures across fork and clone.
1636 */
b728d87e 1637void ust_before_fork(sigset_t *save_sigset)
e822f505
MD
1638{
1639 /*
1640 * Disable signals. This is to avoid that the child intervenes
1641 * before it is properly setup for tracing. It is safer to
1642 * disable all signals, because then we know we are not breaking
1643 * anything by restoring the original mask.
1644 */
1645 sigset_t all_sigs;
1646 int ret;
1647
8c90a710 1648 if (URCU_TLS(lttng_ust_nest_count))
e8508a49 1649 return;
e822f505
MD
1650 /* Disable signals */
1651 sigfillset(&all_sigs);
b728d87e 1652 ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
e822f505
MD
1653 if (ret == -1) {
1654 PERROR("sigprocmask");
1655 }
458d678c
PW
1656
1657 pthread_mutex_lock(&ust_fork_mutex);
1658
3327ac33 1659 ust_lock_nocheck();
e822f505
MD
1660 rcu_bp_before_fork();
1661}
1662
b728d87e 1663static void ust_after_fork_common(sigset_t *restore_sigset)
e822f505
MD
1664{
1665 int ret;
1666
17dfb34b
MD
1667 DBG("process %d", getpid());
1668 ust_unlock();
458d678c
PW
1669
1670 pthread_mutex_unlock(&ust_fork_mutex);
1671
e822f505 1672 /* Restore signals */
23c8854a 1673 ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
e822f505
MD
1674 if (ret == -1) {
1675 PERROR("sigprocmask");
1676 }
1677}
1678
b728d87e 1679void ust_after_fork_parent(sigset_t *restore_sigset)
e822f505 1680{
8c90a710 1681 if (URCU_TLS(lttng_ust_nest_count))
e8508a49 1682 return;
17dfb34b 1683 DBG("process %d", getpid());
e822f505
MD
1684 rcu_bp_after_fork_parent();
1685 /* Release mutexes and reenable signals */
b728d87e 1686 ust_after_fork_common(restore_sigset);
e822f505
MD
1687}
1688
17dfb34b
MD
1689/*
1690 * After fork, in the child, we need to cleanup all the leftover state,
1691 * except the worker thread which already magically disappeared thanks
1692 * to the weird Linux fork semantics. After tyding up, we call
1693 * lttng_ust_init() again to start over as a new PID.
1694 *
1695 * This is meant for forks() that have tracing in the child between the
1696 * fork and following exec call (if there is any).
1697 */
b728d87e 1698void ust_after_fork_child(sigset_t *restore_sigset)
e822f505 1699{
8c90a710 1700 if (URCU_TLS(lttng_ust_nest_count))
e8508a49 1701 return;
17dfb34b 1702 DBG("process %d", getpid());
e822f505
MD
1703 /* Release urcu mutexes */
1704 rcu_bp_after_fork_child();
17dfb34b 1705 lttng_ust_cleanup(0);
a93bfc45 1706 lttng_context_vtid_reset();
e822f505 1707 /* Release mutexes and reenable signals */
b728d87e 1708 ust_after_fork_common(restore_sigset);
318dfea9 1709 lttng_ust_init();
e822f505 1710}
95c25348 1711
246be17e 1712void lttng_ust_sockinfo_session_enabled(void *owner)
95c25348
PW
1713{
1714 struct sock_info *sock_info = owner;
37dddb65 1715 sock_info->statedump_pending = 1;
95c25348 1716}
This page took 0.117185 seconds and 4 git commands to generate.