Fix: document ust_fork_mutex nesting
[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
32ce8569
MD
338int lttng_get_notify_socket(void *owner)
339{
340 struct sock_info *info = owner;
341
342 return info->notify_socket;
343}
344
74d81a6c
MD
345static
346void print_cmd(int cmd, int handle)
347{
348 const char *cmd_name = "Unknown";
349
fd67a004
MD
350 if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping)
351 && cmd_name_mapping[cmd]) {
74d81a6c
MD
352 cmd_name = cmd_name_mapping[cmd];
353 }
fd67a004
MD
354 DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
355 cmd_name, cmd,
74d81a6c
MD
356 lttng_ust_obj_get_name(handle), handle);
357}
358
2691221a 359static
8d20bf54 360int setup_local_apps(void)
2691221a
MD
361{
362 const char *home_dir;
7fc90dca 363 uid_t uid;
2691221a 364
7fc90dca 365 uid = getuid();
8d20bf54
MD
366 /*
367 * Disallow per-user tracing for setuid binaries.
368 */
7fc90dca 369 if (uid != geteuid()) {
9ec6895c 370 assert(local_apps.allowed == 0);
d0a1ae63 371 return 0;
8d20bf54 372 }
3c6f6263 373 home_dir = get_lttng_home_dir();
9ec6895c
MD
374 if (!home_dir) {
375 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
376 assert(local_apps.allowed == 0);
2691221a 377 return -ENOENT;
9ec6895c
MD
378 }
379 local_apps.allowed = 1;
32ce8569
MD
380 snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
381 home_dir,
382 LTTNG_DEFAULT_HOME_RUNDIR,
383 LTTNG_UST_SOCK_FILENAME);
384 snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
385 LTTNG_UST_WAIT_FILENAME,
386 uid);
2691221a
MD
387 return 0;
388}
389
ff517991
MD
390/*
391 * Get notify_sock timeout, in ms.
392 * -1: don't wait. 0: wait forever. >0: timeout, in ms.
393 */
394static
395long get_timeout(void)
396{
397 long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
398
399 if (!got_timeout_env) {
400 str_timeout = getenv("LTTNG_UST_REGISTER_TIMEOUT");
401 got_timeout_env = 1;
402 }
403 if (str_timeout)
404 constructor_delay_ms = strtol(str_timeout, NULL, 10);
405 return constructor_delay_ms;
406}
407
408static
409long get_notify_sock_timeout(void)
410{
411 return get_timeout();
412}
413
414/*
415 * Return values: -1: don't wait. 0: wait forever. 1: timeout wait.
416 */
417static
418int get_constructor_timeout(struct timespec *constructor_timeout)
419{
420 long constructor_delay_ms;
421 int ret;
422
423 constructor_delay_ms = get_timeout();
424
425 switch (constructor_delay_ms) {
426 case -1:/* fall-through */
427 case 0:
428 return constructor_delay_ms;
429 default:
430 break;
431 }
432
433 /*
434 * If we are unable to find the current time, don't wait.
435 */
436 ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
437 if (ret) {
438 return -1;
439 }
440 constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
441 constructor_timeout->tv_nsec +=
442 (constructor_delay_ms % 1000UL) * 1000000UL;
443 if (constructor_timeout->tv_nsec >= 1000000000UL) {
444 constructor_timeout->tv_sec++;
445 constructor_timeout->tv_nsec -= 1000000000UL;
446 }
447 return 1;
448}
449
2691221a 450static
32ce8569 451int register_to_sessiond(int socket, enum ustctl_socket_type type)
2691221a 452{
32ce8569
MD
453 return ustcomm_send_reg_msg(socket,
454 type,
455 CAA_BITS_PER_LONG,
456 lttng_alignof(uint8_t) * CHAR_BIT,
457 lttng_alignof(uint16_t) * CHAR_BIT,
458 lttng_alignof(uint32_t) * CHAR_BIT,
459 lttng_alignof(uint64_t) * CHAR_BIT,
460 lttng_alignof(unsigned long) * CHAR_BIT);
2691221a
MD
461}
462
d9e99d10 463static
57773204 464int send_reply(int sock, struct ustcomm_ust_reply *lur)
d9e99d10 465{
9eb62b9c 466 ssize_t len;
d3a492d1 467
57773204 468 len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
d3a492d1 469 switch (len) {
a4be8962 470 case sizeof(*lur):
d3a492d1
MD
471 DBG("message successfully sent");
472 return 0;
7bc53e94
MD
473 default:
474 if (len == -ECONNRESET) {
475 DBG("remote end closed connection");
d3a492d1
MD
476 return 0;
477 }
7bc53e94
MD
478 if (len < 0)
479 return len;
480 DBG("incorrect message size: %zd", len);
481 return -EINVAL;
d3a492d1
MD
482 }
483}
484
485static
edaa1431 486int handle_register_done(struct sock_info *sock_info)
11ff9c7d
MD
487{
488 int ret;
489
edaa1431
MD
490 if (sock_info->constructor_sem_posted)
491 return 0;
492 sock_info->constructor_sem_posted = 1;
56cd7e2f
MD
493 if (uatomic_read(&sem_count) <= 0) {
494 return 0;
495 }
95259bd0
MD
496 ret = uatomic_add_return(&sem_count, -1);
497 if (ret == 0) {
498 ret = sem_post(&constructor_wait);
499 assert(!ret);
500 }
11ff9c7d
MD
501 return 0;
502}
503
37dddb65
MD
504/*
505 * Only execute pending statedump after the constructor semaphore has
506 * been posted by each listener thread. This means statedump will only
507 * be performed after the "registration done" command is received from
508 * each session daemon the application is connected to.
509 *
510 * This ensures we don't run into deadlock issues with the dynamic
511 * loader mutex, which is held while the constructor is called and
512 * waiting on the constructor semaphore. All operations requiring this
513 * dynamic loader lock need to be postponed using this mechanism.
514 */
515static
516void handle_pending_statedump(struct sock_info *sock_info)
517{
518 int ctor_passed = sock_info->constructor_sem_posted;
519
520 if (ctor_passed && sock_info->statedump_pending) {
458d678c 521 pthread_mutex_lock(&ust_fork_mutex);
37dddb65
MD
522 sock_info->statedump_pending = 0;
523 lttng_handle_pending_statedump(sock_info);
458d678c 524 pthread_mutex_unlock(&ust_fork_mutex);
37dddb65
MD
525 }
526}
527
11ff9c7d
MD
528static
529int handle_message(struct sock_info *sock_info,
57773204 530 int sock, struct ustcomm_ust_msg *lum)
d3a492d1 531{
1ea11eab 532 int ret = 0;
b61ce3b2 533 const struct lttng_ust_objd_ops *ops;
57773204 534 struct ustcomm_ust_reply lur;
ef9ff354 535 union ust_args args;
40003310 536 ssize_t len;
1ea11eab 537
46050b1a
MD
538 memset(&lur, 0, sizeof(lur));
539
3327ac33 540 if (ust_lock()) {
74d81a6c 541 ret = -LTTNG_UST_ERR_EXITING;
1ea11eab
MD
542 goto end;
543 }
9eb62b9c 544
46050b1a
MD
545 ops = objd_ops(lum->handle);
546 if (!ops) {
547 ret = -ENOENT;
548 goto end;
1ea11eab 549 }
46050b1a
MD
550
551 switch (lum->cmd) {
11ff9c7d
MD
552 case LTTNG_UST_REGISTER_DONE:
553 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
edaa1431 554 ret = handle_register_done(sock_info);
11ff9c7d
MD
555 else
556 ret = -EINVAL;
557 break;
46050b1a
MD
558 case LTTNG_UST_RELEASE:
559 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
560 ret = -EPERM;
561 else
1849ef7c 562 ret = lttng_ust_objd_unref(lum->handle, 1);
d9e99d10 563 break;
2d78951a
MD
564 case LTTNG_UST_FILTER:
565 {
566 /* Receive filter data */
f488575f 567 struct lttng_ust_filter_bytecode_node *bytecode;
2d78951a 568
cd54f6d9 569 if (lum->u.filter.data_size > FILTER_BYTECODE_MAX_LEN) {
7bc53e94 570 ERR("Filter data size is too large: %u bytes",
2d78951a
MD
571 lum->u.filter.data_size);
572 ret = -EINVAL;
573 goto error;
574 }
2734ca65 575
885b1dfd 576 if (lum->u.filter.reloc_offset > lum->u.filter.data_size) {
7bc53e94 577 ERR("Filter reloc offset %u is not within data",
2734ca65
CB
578 lum->u.filter.reloc_offset);
579 ret = -EINVAL;
580 goto error;
581 }
582
cd54f6d9
MD
583 bytecode = zmalloc(sizeof(*bytecode) + lum->u.filter.data_size);
584 if (!bytecode) {
585 ret = -ENOMEM;
586 goto error;
587 }
f488575f 588 len = ustcomm_recv_unix_sock(sock, bytecode->bc.data,
2d78951a
MD
589 lum->u.filter.data_size);
590 switch (len) {
591 case 0: /* orderly shutdown */
592 ret = 0;
cd54f6d9 593 free(bytecode);
2d78951a 594 goto error;
2d78951a
MD
595 default:
596 if (len == lum->u.filter.data_size) {
7bc53e94 597 DBG("filter data received");
2d78951a 598 break;
7bc53e94
MD
599 } else if (len < 0) {
600 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
601 if (len == -ECONNRESET) {
602 ERR("%s remote end closed connection", sock_info->name);
603 ret = len;
604 free(bytecode);
605 goto error;
606 }
607 ret = len;
eb8bf361 608 free(bytecode);
7bc53e94 609 goto end;
2d78951a 610 } else {
7bc53e94 611 DBG("incorrect filter data message size: %zd", len);
2d78951a 612 ret = -EINVAL;
cd54f6d9 613 free(bytecode);
2d78951a
MD
614 goto end;
615 }
616 }
f488575f
MD
617 bytecode->bc.len = lum->u.filter.data_size;
618 bytecode->bc.reloc_offset = lum->u.filter.reloc_offset;
3f6fd224 619 bytecode->bc.seqnum = lum->u.filter.seqnum;
cd54f6d9 620 if (ops->cmd) {
2d78951a 621 ret = ops->cmd(lum->handle, lum->cmd,
cd54f6d9 622 (unsigned long) bytecode,
f59ed768 623 &args, sock_info);
cd54f6d9
MD
624 if (ret) {
625 free(bytecode);
626 }
627 /* don't free bytecode if everything went fine. */
628 } else {
2d78951a 629 ret = -ENOSYS;
cd54f6d9
MD
630 free(bytecode);
631 }
2d78951a
MD
632 break;
633 }
86e36163
JI
634 case LTTNG_UST_EXCLUSION:
635 {
636 /* Receive exclusion names */
637 struct lttng_ust_excluder_node *node;
638 unsigned int count;
639
640 count = lum->u.exclusion.count;
641 if (count == 0) {
642 /* There are no names to read */
643 ret = 0;
644 goto error;
645 }
646 node = zmalloc(sizeof(*node) +
647 count * LTTNG_UST_SYM_NAME_LEN);
648 if (!node) {
649 ret = -ENOMEM;
650 goto error;
651 }
652 node->excluder.count = count;
653 len = ustcomm_recv_unix_sock(sock, node->excluder.names,
654 count * LTTNG_UST_SYM_NAME_LEN);
655 switch (len) {
656 case 0: /* orderly shutdown */
657 ret = 0;
658 free(node);
659 goto error;
660 default:
661 if (len == count * LTTNG_UST_SYM_NAME_LEN) {
662 DBG("Exclusion data received");
663 break;
664 } else if (len < 0) {
665 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
666 if (len == -ECONNRESET) {
667 ERR("%s remote end closed connection", sock_info->name);
668 ret = len;
669 free(node);
670 goto error;
671 }
672 ret = len;
673 free(node);
674 goto end;
675 } else {
676 DBG("Incorrect exclusion data message size: %zd", len);
677 ret = -EINVAL;
678 free(node);
679 goto end;
680 }
681 }
682 if (ops->cmd) {
683 ret = ops->cmd(lum->handle, lum->cmd,
684 (unsigned long) node,
685 &args, sock_info);
686 if (ret) {
687 free(node);
688 }
689 /* Don't free exclusion data if everything went fine. */
690 } else {
691 ret = -ENOSYS;
692 free(node);
693 }
694 break;
695 }
74d81a6c
MD
696 case LTTNG_UST_CHANNEL:
697 {
698 void *chan_data;
ff0f5728 699 int wakeup_fd;
74d81a6c
MD
700
701 len = ustcomm_recv_channel_from_sessiond(sock,
ff0f5728
MD
702 &chan_data, lum->u.channel.len,
703 &wakeup_fd);
74d81a6c
MD
704 switch (len) {
705 case 0: /* orderly shutdown */
706 ret = 0;
707 goto error;
708 default:
709 if (len == lum->u.channel.len) {
710 DBG("channel data received");
711 break;
712 } else if (len < 0) {
713 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
714 if (len == -ECONNRESET) {
715 ERR("%s remote end closed connection", sock_info->name);
716 ret = len;
717 goto error;
718 }
719 ret = len;
720 goto end;
721 } else {
722 DBG("incorrect channel data message size: %zd", len);
723 ret = -EINVAL;
724 goto end;
725 }
726 }
727 args.channel.chan_data = chan_data;
ff0f5728 728 args.channel.wakeup_fd = wakeup_fd;
74d81a6c
MD
729 if (ops->cmd)
730 ret = ops->cmd(lum->handle, lum->cmd,
731 (unsigned long) &lum->u,
732 &args, sock_info);
733 else
734 ret = -ENOSYS;
735 break;
736 }
737 case LTTNG_UST_STREAM:
738 {
739 /* Receive shm_fd, wakeup_fd */
740 ret = ustcomm_recv_stream_from_sessiond(sock,
741 &lum->u.stream.len,
742 &args.stream.shm_fd,
743 &args.stream.wakeup_fd);
744 if (ret) {
745 goto end;
746 }
747 if (ops->cmd)
748 ret = ops->cmd(lum->handle, lum->cmd,
749 (unsigned long) &lum->u,
750 &args, sock_info);
751 else
752 ret = -ENOSYS;
753 break;
754 }
d9e99d10 755 default:
46050b1a
MD
756 if (ops->cmd)
757 ret = ops->cmd(lum->handle, lum->cmd,
ef9ff354 758 (unsigned long) &lum->u,
f59ed768 759 &args, sock_info);
46050b1a
MD
760 else
761 ret = -ENOSYS;
762 break;
d9e99d10 763 }
46050b1a 764
1ea11eab 765end:
46050b1a
MD
766 lur.handle = lum->handle;
767 lur.cmd = lum->cmd;
768 lur.ret_val = ret;
769 if (ret >= 0) {
7bc53e94 770 lur.ret_code = LTTNG_UST_OK;
46050b1a 771 } else {
7bc53e94
MD
772 /*
773 * Use -LTTNG_UST_ERR as wildcard for UST internal
774 * error that are not caused by the transport, except if
775 * we already have a more precise error message to
776 * report.
777 */
64b2564e
DG
778 if (ret > -LTTNG_UST_ERR) {
779 /* Translate code to UST error. */
780 switch (ret) {
781 case -EEXIST:
782 lur.ret_code = -LTTNG_UST_ERR_EXIST;
783 break;
784 case -EINVAL:
785 lur.ret_code = -LTTNG_UST_ERR_INVAL;
786 break;
787 case -ENOENT:
788 lur.ret_code = -LTTNG_UST_ERR_NOENT;
789 break;
790 case -EPERM:
791 lur.ret_code = -LTTNG_UST_ERR_PERM;
792 break;
793 case -ENOSYS:
794 lur.ret_code = -LTTNG_UST_ERR_NOSYS;
795 break;
796 default:
797 lur.ret_code = -LTTNG_UST_ERR;
798 break;
799 }
800 } else {
7bc53e94 801 lur.ret_code = ret;
64b2564e 802 }
46050b1a 803 }
e6ea14c5
MD
804 if (ret >= 0) {
805 switch (lum->cmd) {
e6ea14c5
MD
806 case LTTNG_UST_TRACER_VERSION:
807 lur.u.version = lum->u.version;
808 break;
809 case LTTNG_UST_TRACEPOINT_LIST_GET:
810 memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
811 break;
812 }
381c0f1e 813 }
74d81a6c 814 DBG("Return value: %d", lur.ret_val);
46050b1a 815 ret = send_reply(sock, &lur);
193183fb 816 if (ret < 0) {
7bc53e94 817 DBG("error sending reply");
193183fb
MD
818 goto error;
819 }
46050b1a 820
40003310
MD
821 /*
822 * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
823 * after the reply.
824 */
7bc53e94 825 if (lur.ret_code == LTTNG_UST_OK) {
40003310
MD
826 switch (lum->cmd) {
827 case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET:
828 len = ustcomm_send_unix_sock(sock,
829 &args.field_list.entry,
830 sizeof(args.field_list.entry));
7bc53e94
MD
831 if (len < 0) {
832 ret = len;
833 goto error;
834 }
40003310 835 if (len != sizeof(args.field_list.entry)) {
7bc53e94 836 ret = -EINVAL;
40003310
MD
837 goto error;
838 }
839 }
840 }
ef9ff354 841
381c0f1e 842error:
17dfb34b 843 ust_unlock();
d9e99d10 844
37dddb65
MD
845 /*
846 * Performed delayed statedump operations outside of the UST
847 * lock. We need to take the dynamic loader lock before we take
848 * the UST lock internally within handle_pending_statedump().
849 */
850 handle_pending_statedump(sock_info);
246be17e 851
37dddb65 852 return ret;
246be17e
PW
853}
854
46050b1a 855static
efe0de09 856void cleanup_sock_info(struct sock_info *sock_info, int exiting)
46050b1a
MD
857{
858 int ret;
859
5b14aab3
MD
860 if (sock_info->root_handle != -1) {
861 ret = lttng_ust_objd_unref(sock_info->root_handle, 1);
862 if (ret) {
863 ERR("Error unref root handle");
864 }
865 sock_info->root_handle = -1;
866 }
867 sock_info->constructor_sem_posted = 0;
868
869 /*
870 * wait_shm_mmap, socket and notify socket are used by listener
871 * threads outside of the ust lock, so we cannot tear them down
872 * ourselves, because we cannot join on these threads. Leave
873 * responsibility of cleaning up these resources to the OS
874 * process exit.
875 */
876 if (exiting)
877 return;
878
46050b1a 879 if (sock_info->socket != -1) {
e6973a89 880 ret = ustcomm_close_unix_sock(sock_info->socket);
46050b1a 881 if (ret) {
32ce8569 882 ERR("Error closing ust cmd socket");
46050b1a
MD
883 }
884 sock_info->socket = -1;
885 }
32ce8569
MD
886 if (sock_info->notify_socket != -1) {
887 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
888 if (ret) {
889 ERR("Error closing ust notify socket");
890 }
891 sock_info->notify_socket = -1;
892 }
5b14aab3 893 if (sock_info->wait_shm_mmap) {
172d6b68
MD
894 long page_size;
895
896 page_size = sysconf(_SC_PAGE_SIZE);
897 if (page_size > 0) {
898 ret = munmap(sock_info->wait_shm_mmap, page_size);
899 if (ret) {
900 ERR("Error unmapping wait shm");
901 }
7fc90dca
MD
902 }
903 sock_info->wait_shm_mmap = NULL;
904 }
905}
906
58d4b2a2 907/*
33bbeb90
MD
908 * Using fork to set umask in the child process (not multi-thread safe).
909 * We deal with the shm_open vs ftruncate race (happening when the
910 * sessiond owns the shm and does not let everybody modify it, to ensure
911 * safety against shm_unlink) by simply letting the mmap fail and
912 * retrying after a few seconds.
913 * For global shm, everybody has rw access to it until the sessiond
914 * starts.
58d4b2a2 915 */
7fc90dca 916static
58d4b2a2 917int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
7fc90dca 918{
7fc90dca 919 int wait_shm_fd, ret;
58d4b2a2 920 pid_t pid;
44e073f5 921
58d4b2a2 922 /*
33bbeb90 923 * Try to open read-only.
58d4b2a2 924 */
33bbeb90 925 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
58d4b2a2 926 if (wait_shm_fd >= 0) {
7aa76730
MD
927 int32_t tmp_read;
928 ssize_t len;
929 size_t bytes_read = 0;
930
931 /*
932 * Try to read the fd. If unable to do so, try opening
933 * it in write mode.
934 */
935 do {
936 len = read(wait_shm_fd,
937 &((char *) &tmp_read)[bytes_read],
938 sizeof(tmp_read) - bytes_read);
939 if (len > 0) {
940 bytes_read += len;
941 }
942 } while ((len < 0 && errno == EINTR)
943 || (len > 0 && bytes_read < sizeof(tmp_read)));
944 if (bytes_read != sizeof(tmp_read)) {
945 ret = close(wait_shm_fd);
946 if (ret) {
947 ERR("close wait_shm_fd");
948 }
949 goto open_write;
950 }
58d4b2a2
MD
951 goto end;
952 } else if (wait_shm_fd < 0 && errno != ENOENT) {
953 /*
33bbeb90
MD
954 * Real-only open did not work, and it's not because the
955 * entry was not present. It's a failure that prohibits
956 * using shm.
58d4b2a2 957 */
7fc90dca 958 ERR("Error opening shm %s", sock_info->wait_shm_path);
58d4b2a2 959 goto end;
7fc90dca 960 }
7aa76730
MD
961
962open_write:
7fc90dca 963 /*
7aa76730
MD
964 * If the open failed because the file did not exist, or because
965 * the file was not truncated yet, try creating it ourself.
7fc90dca 966 */
8c90a710 967 URCU_TLS(lttng_ust_nest_count)++;
58d4b2a2 968 pid = fork();
8c90a710 969 URCU_TLS(lttng_ust_nest_count)--;
58d4b2a2
MD
970 if (pid > 0) {
971 int status;
972
973 /*
974 * Parent: wait for child to return, in which case the
975 * shared memory map will have been created.
976 */
977 pid = wait(&status);
b7d3cb32 978 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
58d4b2a2
MD
979 wait_shm_fd = -1;
980 goto end;
7fc90dca 981 }
58d4b2a2
MD
982 /*
983 * Try to open read-only again after creation.
984 */
33bbeb90 985 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
58d4b2a2
MD
986 if (wait_shm_fd < 0) {
987 /*
988 * Real-only open did not work. It's a failure
989 * that prohibits using shm.
990 */
991 ERR("Error opening shm %s", sock_info->wait_shm_path);
992 goto end;
993 }
994 goto end;
995 } else if (pid == 0) {
996 int create_mode;
997
998 /* Child */
33bbeb90 999 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
58d4b2a2 1000 if (sock_info->global)
33bbeb90 1001 create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
58d4b2a2
MD
1002 /*
1003 * We're alone in a child process, so we can modify the
1004 * process-wide umask.
1005 */
33bbeb90 1006 umask(~create_mode);
58d4b2a2 1007 /*
33bbeb90
MD
1008 * Try creating shm (or get rw access).
1009 * We don't do an exclusive open, because we allow other
1010 * processes to create+ftruncate it concurrently.
58d4b2a2
MD
1011 */
1012 wait_shm_fd = shm_open(sock_info->wait_shm_path,
1013 O_RDWR | O_CREAT, create_mode);
1014 if (wait_shm_fd >= 0) {
1015 ret = ftruncate(wait_shm_fd, mmap_size);
1016 if (ret) {
1017 PERROR("ftruncate");
b0c1425d 1018 _exit(EXIT_FAILURE);
58d4b2a2 1019 }
b0c1425d 1020 _exit(EXIT_SUCCESS);
58d4b2a2 1021 }
33bbeb90
MD
1022 /*
1023 * For local shm, we need to have rw access to accept
1024 * opening it: this means the local sessiond will be
1025 * able to wake us up. For global shm, we open it even
1026 * if rw access is not granted, because the root.root
1027 * sessiond will be able to override all rights and wake
1028 * us up.
1029 */
1030 if (!sock_info->global && errno != EACCES) {
58d4b2a2 1031 ERR("Error opening shm %s", sock_info->wait_shm_path);
5d3bc5ed 1032 _exit(EXIT_FAILURE);
58d4b2a2
MD
1033 }
1034 /*
33bbeb90
MD
1035 * The shm exists, but we cannot open it RW. Report
1036 * success.
58d4b2a2 1037 */
5d3bc5ed 1038 _exit(EXIT_SUCCESS);
58d4b2a2
MD
1039 } else {
1040 return -1;
7fc90dca 1041 }
58d4b2a2 1042end:
33bbeb90
MD
1043 if (wait_shm_fd >= 0 && !sock_info->global) {
1044 struct stat statbuf;
1045
1046 /*
1047 * Ensure that our user is the owner of the shm file for
1048 * local shm. If we do not own the file, it means our
1049 * sessiond will not have access to wake us up (there is
1050 * probably a rogue process trying to fake our
1051 * sessiond). Fallback to polling method in this case.
1052 */
1053 ret = fstat(wait_shm_fd, &statbuf);
1054 if (ret) {
1055 PERROR("fstat");
1056 goto error_close;
1057 }
1058 if (statbuf.st_uid != getuid())
1059 goto error_close;
1060 }
58d4b2a2 1061 return wait_shm_fd;
33bbeb90
MD
1062
1063error_close:
1064 ret = close(wait_shm_fd);
1065 if (ret) {
1066 PERROR("Error closing fd");
1067 }
1068 return -1;
58d4b2a2
MD
1069}
1070
1071static
1072char *get_map_shm(struct sock_info *sock_info)
1073{
172d6b68 1074 long page_size;
58d4b2a2
MD
1075 int wait_shm_fd, ret;
1076 char *wait_shm_mmap;
1077
172d6b68
MD
1078 page_size = sysconf(_SC_PAGE_SIZE);
1079 if (page_size < 0) {
1080 goto error;
1081 }
1082
1083 wait_shm_fd = get_wait_shm(sock_info, page_size);
58d4b2a2
MD
1084 if (wait_shm_fd < 0) {
1085 goto error;
44e073f5 1086 }
172d6b68 1087 wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
7fc90dca 1088 MAP_SHARED, wait_shm_fd, 0);
7fc90dca
MD
1089 /* close shm fd immediately after taking the mmap reference */
1090 ret = close(wait_shm_fd);
1091 if (ret) {
33bbeb90
MD
1092 PERROR("Error closing fd");
1093 }
1094 if (wait_shm_mmap == MAP_FAILED) {
1095 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1096 goto error;
7fc90dca
MD
1097 }
1098 return wait_shm_mmap;
1099
1100error:
1101 return NULL;
1102}
1103
1104static
1105void wait_for_sessiond(struct sock_info *sock_info)
1106{
efe0de09 1107 int ret;
80e2814b 1108
3327ac33 1109 if (ust_lock()) {
7fc90dca
MD
1110 goto quit;
1111 }
37ed587a
MD
1112 if (wait_poll_fallback) {
1113 goto error;
1114 }
7fc90dca
MD
1115 if (!sock_info->wait_shm_mmap) {
1116 sock_info->wait_shm_mmap = get_map_shm(sock_info);
1117 if (!sock_info->wait_shm_mmap)
1118 goto error;
1119 }
1120 ust_unlock();
1121
1122 DBG("Waiting for %s apps sessiond", sock_info->name);
80e2814b
MD
1123 /* Wait for futex wakeup */
1124 if (uatomic_read((int32_t *) sock_info->wait_shm_mmap) == 0) {
1125 ret = futex_async((int32_t *) sock_info->wait_shm_mmap,
1126 FUTEX_WAIT, 0, NULL, NULL, 0);
80e2814b 1127 if (ret < 0) {
37ed587a
MD
1128 if (errno == EFAULT) {
1129 wait_poll_fallback = 1;
a8b870ad 1130 DBG(
37ed587a
MD
1131"Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1132"do not support FUTEX_WAKE on read-only memory mappings correctly. "
1133"Please upgrade your kernel "
1134"(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1135"mainline). LTTng-UST will use polling mode fallback.");
cd27263b
MD
1136 if (ust_debug())
1137 PERROR("futex");
37ed587a 1138 }
80e2814b
MD
1139 }
1140 }
7fc90dca
MD
1141 return;
1142
1143quit:
1144 ust_unlock();
1145 return;
1146
1147error:
1148 ust_unlock();
7fc90dca 1149 return;
46050b1a
MD
1150}
1151
1ea11eab
MD
1152/*
1153 * This thread does not allocate any resource, except within
1154 * handle_message, within mutex protection. This mutex protects against
1155 * fork and exit.
98bf993f 1156 * The other moment it allocates resources is at socket connection, which
1ea11eab
MD
1157 * is also protected by the mutex.
1158 */
d9e99d10
MD
1159static
1160void *ust_listener_thread(void *arg)
1161{
1ea11eab 1162 struct sock_info *sock_info = arg;
c0eedf81 1163 int sock, ret, prev_connect_failed = 0, has_waited = 0;
ff517991 1164 long timeout;
d9e99d10 1165
9eb62b9c
MD
1166 /* Restart trying to connect to the session daemon */
1167restart:
c0eedf81
MD
1168 if (prev_connect_failed) {
1169 /* Wait for sessiond availability with pipe */
1170 wait_for_sessiond(sock_info);
1171 if (has_waited) {
1172 has_waited = 0;
1173 /*
1174 * Sleep for 5 seconds before retrying after a
1175 * sequence of failure / wait / failure. This
1176 * deals with a killed or broken session daemon.
1177 */
1178 sleep(5);
1179 }
1180 has_waited = 1;
1181 prev_connect_failed = 0;
1182 }
9eb62b9c 1183
1ea11eab 1184 if (sock_info->socket != -1) {
e6973a89 1185 ret = ustcomm_close_unix_sock(sock_info->socket);
1ea11eab 1186 if (ret) {
32ce8569
MD
1187 ERR("Error closing %s ust cmd socket",
1188 sock_info->name);
1ea11eab
MD
1189 }
1190 sock_info->socket = -1;
1191 }
32ce8569
MD
1192 if (sock_info->notify_socket != -1) {
1193 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1194 if (ret) {
1195 ERR("Error closing %s ust notify socket",
1196 sock_info->name);
1197 }
1198 sock_info->notify_socket = -1;
1199 }
46050b1a 1200
321f2351
MD
1201 /*
1202 * Register. We need to perform both connect and sending
1203 * registration message before doing the next connect otherwise
1204 * we may reach unix socket connect queue max limits and block
1205 * on the 2nd connect while the session daemon is awaiting the
1206 * first connect registration message.
1207 */
1208 /* Connect cmd socket */
1209 ret = ustcomm_connect_unix_sock(sock_info->sock_path);
1210 if (ret < 0) {
1211 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1212 prev_connect_failed = 1;
5b14aab3 1213
3327ac33 1214 if (ust_lock()) {
321f2351 1215 goto quit;
32ce8569 1216 }
46050b1a 1217
e3426ddc 1218 /*
321f2351
MD
1219 * If we cannot find the sessiond daemon, don't delay
1220 * constructor execution.
e3426ddc 1221 */
321f2351
MD
1222 ret = handle_register_done(sock_info);
1223 assert(!ret);
1224 ust_unlock();
1225 goto restart;
27fe9f21 1226 }
321f2351 1227 sock_info->socket = ret;
27fe9f21 1228
3327ac33 1229 if (ust_lock()) {
5b14aab3
MD
1230 goto quit;
1231 }
1232
46050b1a
MD
1233 /*
1234 * Create only one root handle per listener thread for the whole
f59ed768
MD
1235 * process lifetime, so we ensure we get ID which is statically
1236 * assigned to the root handle.
46050b1a
MD
1237 */
1238 if (sock_info->root_handle == -1) {
1239 ret = lttng_abi_create_root_handle();
a51070bb 1240 if (ret < 0) {
46050b1a 1241 ERR("Error creating root handle");
46050b1a
MD
1242 goto quit;
1243 }
1244 sock_info->root_handle = ret;
9eb62b9c 1245 }
1ea11eab 1246
32ce8569 1247 ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD);
9eb62b9c 1248 if (ret < 0) {
32ce8569
MD
1249 ERR("Error registering to %s ust cmd socket",
1250 sock_info->name);
c0eedf81 1251 prev_connect_failed = 1;
11ff9c7d
MD
1252 /*
1253 * If we cannot register to the sessiond daemon, don't
1254 * delay constructor execution.
1255 */
edaa1431 1256 ret = handle_register_done(sock_info);
11ff9c7d 1257 assert(!ret);
17dfb34b 1258 ust_unlock();
9eb62b9c
MD
1259 goto restart;
1260 }
321f2351
MD
1261
1262 ust_unlock();
1263
1264 /* Connect notify socket */
1265 ret = ustcomm_connect_unix_sock(sock_info->sock_path);
1266 if (ret < 0) {
1267 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1268 prev_connect_failed = 1;
1269
3327ac33 1270 if (ust_lock()) {
321f2351
MD
1271 goto quit;
1272 }
1273
1274 /*
1275 * If we cannot find the sessiond daemon, don't delay
1276 * constructor execution.
1277 */
1278 ret = handle_register_done(sock_info);
1279 assert(!ret);
1280 ust_unlock();
1281 goto restart;
1282 }
1283 sock_info->notify_socket = ret;
1284
1285 timeout = get_notify_sock_timeout();
1286 if (timeout >= 0) {
1287 /*
1288 * Give at least 10ms to sessiond to reply to
1289 * notifications.
1290 */
1291 if (timeout < 10)
1292 timeout = 10;
1293 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
1294 timeout);
1295 if (ret < 0) {
1296 WARN("Error setting socket receive timeout");
1297 }
1298 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
1299 timeout);
1300 if (ret < 0) {
1301 WARN("Error setting socket send timeout");
1302 }
1303 } else if (timeout < -1) {
1304 WARN("Unsupported timeout value %ld", timeout);
1305 }
1306
3327ac33 1307 if (ust_lock()) {
321f2351
MD
1308 goto quit;
1309 }
1310
32ce8569
MD
1311 ret = register_to_sessiond(sock_info->notify_socket,
1312 USTCTL_SOCKET_NOTIFY);
1313 if (ret < 0) {
1314 ERR("Error registering to %s ust notify socket",
1315 sock_info->name);
1316 prev_connect_failed = 1;
1317 /*
1318 * If we cannot register to the sessiond daemon, don't
1319 * delay constructor execution.
1320 */
1321 ret = handle_register_done(sock_info);
1322 assert(!ret);
1323 ust_unlock();
1324 goto restart;
1325 }
1326 sock = sock_info->socket;
1327
17dfb34b 1328 ust_unlock();
46050b1a 1329
d9e99d10
MD
1330 for (;;) {
1331 ssize_t len;
57773204 1332 struct ustcomm_ust_msg lum;
d9e99d10 1333
57773204 1334 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
d9e99d10
MD
1335 switch (len) {
1336 case 0: /* orderly shutdown */
7dd08bec 1337 DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
3327ac33 1338 if (ust_lock()) {
d5e1fea6
MD
1339 goto quit;
1340 }
8236ba10
MD
1341 /*
1342 * Either sessiond has shutdown or refused us by closing the socket.
1343 * In either case, we don't want to delay construction execution,
1344 * and we need to wait before retry.
1345 */
1346 prev_connect_failed = 1;
1347 /*
1348 * If we cannot register to the sessiond daemon, don't
1349 * delay constructor execution.
1350 */
1351 ret = handle_register_done(sock_info);
1352 assert(!ret);
1353 ust_unlock();
d9e99d10 1354 goto end;
e7723462 1355 case sizeof(lum):
74d81a6c 1356 print_cmd(lum.cmd, lum.handle);
11ff9c7d 1357 ret = handle_message(sock_info, sock, &lum);
7bc53e94 1358 if (ret) {
11ff9c7d 1359 ERR("Error handling message for %s socket", sock_info->name);
d9e99d10
MD
1360 }
1361 continue;
7bc53e94
MD
1362 default:
1363 if (len < 0) {
1364 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1365 } else {
1366 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
1367 }
1368 if (len == -ECONNRESET) {
1369 DBG("%s remote end closed connection", sock_info->name);
d9e99d10
MD
1370 goto end;
1371 }
1372 goto end;
d9e99d10
MD
1373 }
1374
1375 }
1376end:
3327ac33 1377 if (ust_lock()) {
d5e1fea6
MD
1378 goto quit;
1379 }
f59ed768
MD
1380 /* Cleanup socket handles before trying to reconnect */
1381 lttng_ust_objd_table_owner_cleanup(sock_info);
1382 ust_unlock();
9eb62b9c 1383 goto restart; /* try to reconnect */
e33f3265 1384
1ea11eab 1385quit:
e33f3265 1386 ust_unlock();
3327ac33
MD
1387
1388 pthread_mutex_lock(&ust_exit_mutex);
1389 sock_info->thread_active = 0;
1390 pthread_mutex_unlock(&ust_exit_mutex);
d9e99d10
MD
1391 return NULL;
1392}
1393
2594a5b4
MD
1394/*
1395 * Weak symbol to call when the ust malloc wrapper is not loaded.
1396 */
1397__attribute__((weak))
1398void lttng_ust_malloc_wrapper_init(void)
1399{
1400}
1401
2691221a
MD
1402/*
1403 * sessiond monitoring thread: monitor presence of global and per-user
1404 * sessiond by polling the application common named pipe.
1405 */
edaa1431 1406void __attribute__((constructor)) lttng_ust_init(void)
2691221a 1407{
11ff9c7d 1408 struct timespec constructor_timeout;
ae6a58bf 1409 sigset_t sig_all_blocked, orig_parent_mask;
1879f67f 1410 pthread_attr_t thread_attr;
cf12a773 1411 int timeout_mode;
2691221a
MD
1412 int ret;
1413
edaa1431
MD
1414 if (uatomic_xchg(&initialized, 1) == 1)
1415 return;
1416
eddd8d5d
MD
1417 /*
1418 * Fixup interdependency between TLS fixup mutex (which happens
1419 * to be the dynamic linker mutex) and ust_lock, taken within
1420 * the ust lock.
1421 */
f645cfa7 1422 lttng_fixup_ringbuffer_tls();
4158a15a 1423 lttng_fixup_vtid_tls();
a903623f 1424 lttng_fixup_nest_count_tls();
009745db 1425 lttng_fixup_procname_tls();
d58d1454 1426 lttng_fixup_ust_mutex_nest_tls();
eddd8d5d 1427
edaa1431
MD
1428 /*
1429 * We want precise control over the order in which we construct
1430 * our sub-libraries vs starting to receive commands from
1431 * sessiond (otherwise leading to errors when trying to create
1432 * sessiond before the init functions are completed).
1433 */
2691221a 1434 init_usterr();
edaa1431 1435 init_tracepoint();
bd703713 1436 lttng_ust_baddr_statedump_init();
7dd08bec
MD
1437 lttng_ring_buffer_metadata_client_init();
1438 lttng_ring_buffer_client_overwrite_init();
34a91bdb 1439 lttng_ring_buffer_client_overwrite_rt_init();
7dd08bec 1440 lttng_ring_buffer_client_discard_init();
34a91bdb 1441 lttng_ring_buffer_client_discard_rt_init();
d58d1454 1442 lttng_perf_counter_init();
a0a3bef9 1443 lttng_context_init();
2594a5b4
MD
1444 /*
1445 * Invoke ust malloc wrapper init before starting other threads.
1446 */
1447 lttng_ust_malloc_wrapper_init();
2691221a 1448
ff517991 1449 timeout_mode = get_constructor_timeout(&constructor_timeout);
11ff9c7d 1450
95259bd0 1451 ret = sem_init(&constructor_wait, 0, 0);
11ff9c7d
MD
1452 assert(!ret);
1453
8d20bf54 1454 ret = setup_local_apps();
2691221a 1455 if (ret) {
9ec6895c 1456 DBG("local apps setup returned %d", ret);
2691221a 1457 }
ae6a58bf
WP
1458
1459 /* A new thread created by pthread_create inherits the signal mask
1460 * from the parent. To avoid any signal being received by the
1461 * listener thread, we block all signals temporarily in the parent,
1462 * while we create the listener thread.
1463 */
1464 sigfillset(&sig_all_blocked);
1465 ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
1466 if (ret) {
d94d802c 1467 ERR("pthread_sigmask: %s", strerror(ret));
ae6a58bf
WP
1468 }
1469
1879f67f
MG
1470 ret = pthread_attr_init(&thread_attr);
1471 if (ret) {
1472 ERR("pthread_attr_init: %s", strerror(ret));
1473 }
1474 ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
1475 if (ret) {
1476 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
1477 }
1478
c0bbbd5a 1479 pthread_mutex_lock(&ust_exit_mutex);
1879f67f 1480 ret = pthread_create(&global_apps.ust_listener, &thread_attr,
dde70ea0 1481 ust_listener_thread, &global_apps);
d94d802c
MD
1482 if (ret) {
1483 ERR("pthread_create global: %s", strerror(ret));
1484 }
e33f3265 1485 global_apps.thread_active = 1;
c0bbbd5a 1486 pthread_mutex_unlock(&ust_exit_mutex);
e33f3265 1487
8d20bf54 1488 if (local_apps.allowed) {
c0bbbd5a 1489 pthread_mutex_lock(&ust_exit_mutex);
1879f67f 1490 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
dde70ea0 1491 ust_listener_thread, &local_apps);
d94d802c
MD
1492 if (ret) {
1493 ERR("pthread_create local: %s", strerror(ret));
1494 }
e33f3265 1495 local_apps.thread_active = 1;
c0bbbd5a 1496 pthread_mutex_unlock(&ust_exit_mutex);
8d20bf54
MD
1497 } else {
1498 handle_register_done(&local_apps);
1499 }
1879f67f
MG
1500 ret = pthread_attr_destroy(&thread_attr);
1501 if (ret) {
1502 ERR("pthread_attr_destroy: %s", strerror(ret));
1503 }
8d20bf54 1504
ae6a58bf
WP
1505 /* Restore original signal mask in parent */
1506 ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
1507 if (ret) {
d94d802c 1508 ERR("pthread_sigmask: %s", strerror(ret));
ae6a58bf
WP
1509 }
1510
cf12a773
MD
1511 switch (timeout_mode) {
1512 case 1: /* timeout wait */
95259bd0
MD
1513 do {
1514 ret = sem_timedwait(&constructor_wait,
1515 &constructor_timeout);
1516 } while (ret < 0 && errno == EINTR);
cf12a773 1517 if (ret < 0 && errno == ETIMEDOUT) {
7dd08bec 1518 ERR("Timed out waiting for lttng-sessiond");
cf12a773
MD
1519 } else {
1520 assert(!ret);
1521 }
1522 break;
7b766b16 1523 case -1:/* wait forever */
95259bd0
MD
1524 do {
1525 ret = sem_wait(&constructor_wait);
1526 } while (ret < 0 && errno == EINTR);
11ff9c7d 1527 assert(!ret);
cf12a773 1528 break;
7b766b16 1529 case 0: /* no timeout */
cf12a773 1530 break;
11ff9c7d 1531 }
2691221a
MD
1532}
1533
17dfb34b
MD
1534static
1535void lttng_ust_cleanup(int exiting)
1536{
efe0de09 1537 cleanup_sock_info(&global_apps, exiting);
17dfb34b 1538 if (local_apps.allowed) {
efe0de09 1539 cleanup_sock_info(&local_apps, exiting);
17dfb34b 1540 }
efe0de09
MD
1541 /*
1542 * The teardown in this function all affect data structures
1543 * accessed under the UST lock by the listener thread. This
1544 * lock, along with the lttng_ust_comm_should_quit flag, ensure
1545 * that none of these threads are accessing this data at this
1546 * point.
1547 */
17dfb34b 1548 lttng_ust_abi_exit();
003fedf4 1549 lttng_ust_events_exit();
a0a3bef9 1550 lttng_context_exit();
d58d1454 1551 lttng_perf_counter_exit();
34a91bdb 1552 lttng_ring_buffer_client_discard_rt_exit();
7dd08bec 1553 lttng_ring_buffer_client_discard_exit();
34a91bdb 1554 lttng_ring_buffer_client_overwrite_rt_exit();
7dd08bec
MD
1555 lttng_ring_buffer_client_overwrite_exit();
1556 lttng_ring_buffer_metadata_client_exit();
bd703713 1557 lttng_ust_baddr_statedump_destroy();
17dfb34b
MD
1558 exit_tracepoint();
1559 if (!exiting) {
1560 /* Reinitialize values for fork */
1561 sem_count = 2;
1562 lttng_ust_comm_should_quit = 0;
1563 initialized = 0;
1564 }
1565}
1566
edaa1431 1567void __attribute__((destructor)) lttng_ust_exit(void)
2691221a
MD
1568{
1569 int ret;
1570
9eb62b9c
MD
1571 /*
1572 * Using pthread_cancel here because:
1573 * A) we don't want to hang application teardown.
1574 * B) the thread is not allocating any resource.
1575 */
1ea11eab
MD
1576
1577 /*
1578 * Require the communication thread to quit. Synchronize with
1579 * mutexes to ensure it is not in a mutex critical section when
1580 * pthread_cancel is later called.
1581 */
3327ac33 1582 ust_lock_nocheck();
1ea11eab 1583 lttng_ust_comm_should_quit = 1;
3327ac33 1584 ust_unlock();
1ea11eab 1585
3327ac33 1586 pthread_mutex_lock(&ust_exit_mutex);
f5f94532 1587 /* cancel threads */
e33f3265
MD
1588 if (global_apps.thread_active) {
1589 ret = pthread_cancel(global_apps.ust_listener);
1590 if (ret) {
1591 ERR("Error cancelling global ust listener thread: %s",
1592 strerror(ret));
1593 } else {
1594 global_apps.thread_active = 0;
1595 }
2691221a 1596 }
e33f3265 1597 if (local_apps.thread_active) {
8d20bf54
MD
1598 ret = pthread_cancel(local_apps.ust_listener);
1599 if (ret) {
d94d802c
MD
1600 ERR("Error cancelling local ust listener thread: %s",
1601 strerror(ret));
e33f3265
MD
1602 } else {
1603 local_apps.thread_active = 0;
8d20bf54 1604 }
8d20bf54 1605 }
3327ac33 1606 pthread_mutex_unlock(&ust_exit_mutex);
e33f3265 1607
efe0de09
MD
1608 /*
1609 * Do NOT join threads: use of sys_futex makes it impossible to
1610 * join the threads without using async-cancel, but async-cancel
1611 * is delivered by a signal, which could hit the target thread
1612 * anywhere in its code path, including while the ust_lock() is
1613 * held, causing a deadlock for the other thread. Let the OS
1614 * cleanup the threads if there are stalled in a syscall.
1615 */
17dfb34b 1616 lttng_ust_cleanup(1);
2691221a 1617}
e822f505
MD
1618
1619/*
1620 * We exclude the worker threads across fork and clone (except
1621 * CLONE_VM), because these system calls only keep the forking thread
1622 * running in the child. Therefore, we don't want to call fork or clone
1623 * in the middle of an tracepoint or ust tracing state modification.
1624 * Holding this mutex protects these structures across fork and clone.
1625 */
b728d87e 1626void ust_before_fork(sigset_t *save_sigset)
e822f505
MD
1627{
1628 /*
1629 * Disable signals. This is to avoid that the child intervenes
1630 * before it is properly setup for tracing. It is safer to
1631 * disable all signals, because then we know we are not breaking
1632 * anything by restoring the original mask.
1633 */
1634 sigset_t all_sigs;
1635 int ret;
1636
8c90a710 1637 if (URCU_TLS(lttng_ust_nest_count))
e8508a49 1638 return;
e822f505
MD
1639 /* Disable signals */
1640 sigfillset(&all_sigs);
b728d87e 1641 ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
e822f505
MD
1642 if (ret == -1) {
1643 PERROR("sigprocmask");
1644 }
458d678c
PW
1645
1646 pthread_mutex_lock(&ust_fork_mutex);
1647
3327ac33 1648 ust_lock_nocheck();
e822f505
MD
1649 rcu_bp_before_fork();
1650}
1651
b728d87e 1652static void ust_after_fork_common(sigset_t *restore_sigset)
e822f505
MD
1653{
1654 int ret;
1655
17dfb34b
MD
1656 DBG("process %d", getpid());
1657 ust_unlock();
458d678c
PW
1658
1659 pthread_mutex_unlock(&ust_fork_mutex);
1660
e822f505 1661 /* Restore signals */
23c8854a 1662 ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
e822f505
MD
1663 if (ret == -1) {
1664 PERROR("sigprocmask");
1665 }
1666}
1667
b728d87e 1668void ust_after_fork_parent(sigset_t *restore_sigset)
e822f505 1669{
8c90a710 1670 if (URCU_TLS(lttng_ust_nest_count))
e8508a49 1671 return;
17dfb34b 1672 DBG("process %d", getpid());
e822f505
MD
1673 rcu_bp_after_fork_parent();
1674 /* Release mutexes and reenable signals */
b728d87e 1675 ust_after_fork_common(restore_sigset);
e822f505
MD
1676}
1677
17dfb34b
MD
1678/*
1679 * After fork, in the child, we need to cleanup all the leftover state,
1680 * except the worker thread which already magically disappeared thanks
1681 * to the weird Linux fork semantics. After tyding up, we call
1682 * lttng_ust_init() again to start over as a new PID.
1683 *
1684 * This is meant for forks() that have tracing in the child between the
1685 * fork and following exec call (if there is any).
1686 */
b728d87e 1687void ust_after_fork_child(sigset_t *restore_sigset)
e822f505 1688{
8c90a710 1689 if (URCU_TLS(lttng_ust_nest_count))
e8508a49 1690 return;
17dfb34b 1691 DBG("process %d", getpid());
e822f505
MD
1692 /* Release urcu mutexes */
1693 rcu_bp_after_fork_child();
17dfb34b 1694 lttng_ust_cleanup(0);
a93bfc45 1695 lttng_context_vtid_reset();
e822f505 1696 /* Release mutexes and reenable signals */
b728d87e 1697 ust_after_fork_common(restore_sigset);
318dfea9 1698 lttng_ust_init();
e822f505 1699}
95c25348 1700
246be17e 1701void lttng_ust_sockinfo_session_enabled(void *owner)
95c25348
PW
1702{
1703 struct sock_info *sock_info = owner;
37dddb65 1704 sock_info->statedump_pending = 1;
95c25348 1705}
This page took 0.120141 seconds and 4 git commands to generate.