fix: Handle sys_futex with async cancel, add missing pthread_join
[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>
44c72f10
MD
44#include <ust-comm.h>
45#include <usterr-signal-safe.h>
46#include "tracepoint-internal.h"
b751f722 47#include "ltt-tracer-core.h"
48621a42 48#include "compat.h"
edaa1431
MD
49
50/*
51 * Has lttng ust comm constructor been called ?
52 */
53static int initialized;
54
1ea11eab 55/*
17dfb34b
MD
56 * The ust_lock/ust_unlock lock is used as a communication thread mutex.
57 * Held when handling a command, also held by fork() to deal with
58 * removal of threads, and by exit path.
1ea11eab 59 */
1ea11eab
MD
60
61/* Should the ust comm thread quit ? */
62static int lttng_ust_comm_should_quit;
63
11ff9c7d
MD
64/*
65 * Wait for either of these before continuing to the main
66 * program:
67 * - the register_done message from sessiond daemon
68 * (will let the sessiond daemon enable sessions before main
69 * starts.)
70 * - sessiond daemon is not reachable.
71 * - timeout (ensuring applications are resilient to session
72 * daemon problems).
73 */
74static sem_t constructor_wait;
950aab0c
MD
75/*
76 * Doing this for both the global and local sessiond.
77 */
95259bd0 78static int sem_count = { 2 };
11ff9c7d 79
1ea11eab
MD
80/*
81 * Info about socket and associated listener thread.
82 */
83struct sock_info {
11ff9c7d 84 const char *name;
1ea11eab 85 pthread_t ust_listener; /* listener thread */
46050b1a 86 int root_handle;
8d20bf54
MD
87 int constructor_sem_posted;
88 int allowed;
44e073f5 89 int global;
7fc90dca
MD
90
91 char sock_path[PATH_MAX];
92 int socket;
93
94 char wait_shm_path[PATH_MAX];
95 char *wait_shm_mmap;
1ea11eab 96};
2691221a
MD
97
98/* Socket from app (connect) to session daemon (listen) for communication */
1ea11eab 99struct sock_info global_apps = {
11ff9c7d 100 .name = "global",
44e073f5 101 .global = 1,
7fc90dca 102
46050b1a 103 .root_handle = -1,
8d20bf54 104 .allowed = 1,
7fc90dca
MD
105
106 .sock_path = DEFAULT_GLOBAL_APPS_UNIX_SOCK,
107 .socket = -1,
108
109 .wait_shm_path = DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH,
1ea11eab 110};
2691221a
MD
111
112/* TODO: allow global_apps_sock_path override */
113
1ea11eab 114struct sock_info local_apps = {
11ff9c7d 115 .name = "local",
44e073f5 116 .global = 0,
46050b1a 117 .root_handle = -1,
8d20bf54 118 .allowed = 0, /* Check setuid bit first */
7fc90dca
MD
119
120 .socket = -1,
1ea11eab 121};
2691221a 122
37ed587a
MD
123static int wait_poll_fallback;
124
edaa1431
MD
125extern void ltt_ring_buffer_client_overwrite_init(void);
126extern void ltt_ring_buffer_client_discard_init(void);
127extern void ltt_ring_buffer_metadata_client_init(void);
128extern void ltt_ring_buffer_client_overwrite_exit(void);
129extern void ltt_ring_buffer_client_discard_exit(void);
130extern void ltt_ring_buffer_metadata_client_exit(void);
131
2691221a 132static
8d20bf54 133int setup_local_apps(void)
2691221a
MD
134{
135 const char *home_dir;
7fc90dca 136 uid_t uid;
2691221a 137
7fc90dca 138 uid = getuid();
8d20bf54
MD
139 /*
140 * Disallow per-user tracing for setuid binaries.
141 */
7fc90dca 142 if (uid != geteuid()) {
8d20bf54 143 local_apps.allowed = 0;
d0a1ae63 144 return 0;
8d20bf54
MD
145 } else {
146 local_apps.allowed = 1;
147 }
2691221a
MD
148 home_dir = (const char *) getenv("HOME");
149 if (!home_dir)
150 return -ENOENT;
1ea11eab 151 snprintf(local_apps.sock_path, PATH_MAX,
2691221a 152 DEFAULT_HOME_APPS_UNIX_SOCK, home_dir);
7fc90dca
MD
153 snprintf(local_apps.wait_shm_path, PATH_MAX,
154 DEFAULT_HOME_APPS_WAIT_SHM_PATH, uid);
2691221a
MD
155 return 0;
156}
157
158static
159int register_app_to_sessiond(int socket)
160{
161 ssize_t ret;
162 struct {
e44418f3
MD
163 uint32_t major;
164 uint32_t minor;
2691221a 165 pid_t pid;
5c33bde8 166 pid_t ppid;
2691221a 167 uid_t uid;
83610856 168 gid_t gid;
c117fb1b 169 uint32_t bits_per_long;
2629549e 170 char name[16]; /* process name */
2691221a
MD
171 } reg_msg;
172
e44418f3
MD
173 reg_msg.major = LTTNG_UST_COMM_VERSION_MAJOR;
174 reg_msg.minor = LTTNG_UST_COMM_VERSION_MINOR;
2691221a 175 reg_msg.pid = getpid();
5c33bde8 176 reg_msg.ppid = getppid();
2691221a 177 reg_msg.uid = getuid();
83610856 178 reg_msg.gid = getgid();
c117fb1b 179 reg_msg.bits_per_long = CAA_BITS_PER_LONG;
48621a42 180 lttng_ust_getprocname(reg_msg.name);
2691221a 181
57773204 182 ret = ustcomm_send_unix_sock(socket, &reg_msg, sizeof(reg_msg));
2691221a
MD
183 if (ret >= 0 && ret != sizeof(reg_msg))
184 return -EIO;
185 return ret;
186}
187
d9e99d10 188static
57773204 189int send_reply(int sock, struct ustcomm_ust_reply *lur)
d9e99d10 190{
9eb62b9c 191 ssize_t len;
d3a492d1 192
57773204 193 len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
d3a492d1 194 switch (len) {
a4be8962 195 case sizeof(*lur):
d3a492d1
MD
196 DBG("message successfully sent");
197 return 0;
198 case -1:
199 if (errno == ECONNRESET) {
200 printf("remote end closed connection\n");
201 return 0;
202 }
203 return -1;
204 default:
205 printf("incorrect message size: %zd\n", len);
206 return -1;
207 }
208}
209
210static
edaa1431 211int handle_register_done(struct sock_info *sock_info)
11ff9c7d
MD
212{
213 int ret;
214
edaa1431
MD
215 if (sock_info->constructor_sem_posted)
216 return 0;
217 sock_info->constructor_sem_posted = 1;
56cd7e2f
MD
218 if (uatomic_read(&sem_count) <= 0) {
219 return 0;
220 }
95259bd0
MD
221 ret = uatomic_add_return(&sem_count, -1);
222 if (ret == 0) {
223 ret = sem_post(&constructor_wait);
224 assert(!ret);
225 }
11ff9c7d
MD
226 return 0;
227}
228
229static
230int handle_message(struct sock_info *sock_info,
57773204 231 int sock, struct ustcomm_ust_msg *lum)
d3a492d1 232{
1ea11eab 233 int ret = 0;
b61ce3b2 234 const struct lttng_ust_objd_ops *ops;
57773204 235 struct ustcomm_ust_reply lur;
193183fb 236 int shm_fd, wait_fd;
ef9ff354 237 union ust_args args;
1ea11eab 238
17dfb34b 239 ust_lock();
1ea11eab 240
46050b1a
MD
241 memset(&lur, 0, sizeof(lur));
242
1ea11eab 243 if (lttng_ust_comm_should_quit) {
46050b1a 244 ret = -EPERM;
1ea11eab
MD
245 goto end;
246 }
9eb62b9c 247
46050b1a
MD
248 ops = objd_ops(lum->handle);
249 if (!ops) {
250 ret = -ENOENT;
251 goto end;
1ea11eab 252 }
46050b1a
MD
253
254 switch (lum->cmd) {
11ff9c7d
MD
255 case LTTNG_UST_REGISTER_DONE:
256 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
edaa1431 257 ret = handle_register_done(sock_info);
11ff9c7d
MD
258 else
259 ret = -EINVAL;
260 break;
46050b1a
MD
261 case LTTNG_UST_RELEASE:
262 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
263 ret = -EPERM;
264 else
d4419b81 265 ret = lttng_ust_objd_unref(lum->handle);
d9e99d10
MD
266 break;
267 default:
46050b1a
MD
268 if (ops->cmd)
269 ret = ops->cmd(lum->handle, lum->cmd,
ef9ff354
MD
270 (unsigned long) &lum->u,
271 &args);
46050b1a
MD
272 else
273 ret = -ENOSYS;
274 break;
d9e99d10 275 }
46050b1a 276
1ea11eab 277end:
46050b1a
MD
278 lur.handle = lum->handle;
279 lur.cmd = lum->cmd;
280 lur.ret_val = ret;
281 if (ret >= 0) {
57773204 282 lur.ret_code = USTCOMM_OK;
46050b1a 283 } else {
57773204 284 //lur.ret_code = USTCOMM_SESSION_FAIL;
193183fb 285 lur.ret_code = ret;
46050b1a 286 }
193183fb
MD
287 switch (lum->cmd) {
288 case LTTNG_UST_STREAM:
381c0f1e
MD
289 /*
290 * Special-case reply to send stream info.
291 * Use lum.u output.
292 */
ef9ff354
MD
293 lur.u.stream.memory_map_size = *args.stream.memory_map_size;
294 shm_fd = *args.stream.shm_fd;
295 wait_fd = *args.stream.wait_fd;
193183fb 296 break;
824f40b8 297 case LTTNG_UST_METADATA:
193183fb 298 case LTTNG_UST_CHANNEL:
ef9ff354
MD
299 lur.u.channel.memory_map_size = *args.channel.memory_map_size;
300 shm_fd = *args.channel.shm_fd;
301 wait_fd = *args.channel.wait_fd;
193183fb 302 break;
23c8854a 303 case LTTNG_UST_TRACER_VERSION:
57773204
MD
304 lur.u.version = lum->u.version;
305 break;
b115631f 306 case LTTNG_UST_TRACEPOINT_LIST_GET:
cbef6901 307 memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
b115631f 308 break;
381c0f1e 309 }
46050b1a 310 ret = send_reply(sock, &lur);
193183fb
MD
311 if (ret < 0) {
312 perror("error sending reply");
313 goto error;
314 }
46050b1a 315
824f40b8
MD
316 if ((lum->cmd == LTTNG_UST_STREAM
317 || lum->cmd == LTTNG_UST_CHANNEL
318 || lum->cmd == LTTNG_UST_METADATA)
57773204 319 && lur.ret_code == USTCOMM_OK) {
381c0f1e 320 /* we also need to send the file descriptors. */
57773204 321 ret = ustcomm_send_fds_unix_sock(sock,
193183fb 322 &shm_fd, &shm_fd,
381c0f1e
MD
323 1, sizeof(int));
324 if (ret < 0) {
325 perror("send shm_fd");
326 goto error;
327 }
57773204 328 ret = ustcomm_send_fds_unix_sock(sock,
193183fb 329 &wait_fd, &wait_fd,
381c0f1e
MD
330 1, sizeof(int));
331 if (ret < 0) {
332 perror("send wait_fd");
333 goto error;
334 }
335 }
ef9ff354
MD
336 /*
337 * We still have the memory map reference, and the fds have been
3ad5f707
MD
338 * sent to the sessiond. We can therefore close those fds. Note
339 * that we keep the write side of the wait_fd open, but close
340 * the read side.
ef9ff354
MD
341 */
342 if (lur.ret_code == USTCOMM_OK) {
343 switch (lum->cmd) {
344 case LTTNG_UST_STREAM:
345 if (shm_fd >= 0) {
346 ret = close(shm_fd);
347 if (ret) {
348 PERROR("Error closing stream shm_fd");
349 }
350 *args.stream.shm_fd = -1;
351 }
352 if (wait_fd >= 0) {
353 ret = close(wait_fd);
354 if (ret) {
355 PERROR("Error closing stream wait_fd");
356 }
357 *args.stream.wait_fd = -1;
358 }
359 break;
360 case LTTNG_UST_METADATA:
361 case LTTNG_UST_CHANNEL:
362 if (shm_fd >= 0) {
363 ret = close(shm_fd);
364 if (ret) {
365 PERROR("Error closing channel shm_fd");
366 }
367 *args.channel.shm_fd = -1;
368 }
369 if (wait_fd >= 0) {
370 ret = close(wait_fd);
371 if (ret) {
372 PERROR("Error closing channel wait_fd");
373 }
374 *args.channel.wait_fd = -1;
375 }
376 break;
377 }
378 }
379
381c0f1e 380error:
17dfb34b 381 ust_unlock();
1ea11eab 382 return ret;
d9e99d10
MD
383}
384
46050b1a 385static
f5f94532 386void cleanup_sock_info(struct sock_info *sock_info)
46050b1a
MD
387{
388 int ret;
389
390 if (sock_info->socket != -1) {
391 ret = close(sock_info->socket);
392 if (ret) {
7fc90dca 393 ERR("Error closing apps socket");
46050b1a
MD
394 }
395 sock_info->socket = -1;
396 }
397 if (sock_info->root_handle != -1) {
d4419b81 398 ret = lttng_ust_objd_unref(sock_info->root_handle);
46050b1a
MD
399 if (ret) {
400 ERR("Error unref root handle");
401 }
402 sock_info->root_handle = -1;
403 }
318dfea9 404 sock_info->constructor_sem_posted = 0;
f5f94532 405 if (sock_info->wait_shm_mmap) {
7fc90dca
MD
406 ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE));
407 if (ret) {
408 ERR("Error unmapping wait shm");
409 }
410 sock_info->wait_shm_mmap = NULL;
411 }
412}
413
58d4b2a2 414/*
33bbeb90
MD
415 * Using fork to set umask in the child process (not multi-thread safe).
416 * We deal with the shm_open vs ftruncate race (happening when the
417 * sessiond owns the shm and does not let everybody modify it, to ensure
418 * safety against shm_unlink) by simply letting the mmap fail and
419 * retrying after a few seconds.
420 * For global shm, everybody has rw access to it until the sessiond
421 * starts.
58d4b2a2 422 */
7fc90dca 423static
58d4b2a2 424int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
7fc90dca 425{
7fc90dca 426 int wait_shm_fd, ret;
58d4b2a2 427 pid_t pid;
44e073f5 428
58d4b2a2 429 /*
33bbeb90 430 * Try to open read-only.
58d4b2a2 431 */
33bbeb90 432 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
58d4b2a2
MD
433 if (wait_shm_fd >= 0) {
434 goto end;
435 } else if (wait_shm_fd < 0 && errno != ENOENT) {
436 /*
33bbeb90
MD
437 * Real-only open did not work, and it's not because the
438 * entry was not present. It's a failure that prohibits
439 * using shm.
58d4b2a2 440 */
7fc90dca 441 ERR("Error opening shm %s", sock_info->wait_shm_path);
58d4b2a2 442 goto end;
7fc90dca
MD
443 }
444 /*
58d4b2a2
MD
445 * If the open failed because the file did not exist, try
446 * creating it ourself.
7fc90dca 447 */
58d4b2a2
MD
448 pid = fork();
449 if (pid > 0) {
450 int status;
451
452 /*
453 * Parent: wait for child to return, in which case the
454 * shared memory map will have been created.
455 */
456 pid = wait(&status);
b7d3cb32 457 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
58d4b2a2
MD
458 wait_shm_fd = -1;
459 goto end;
7fc90dca 460 }
58d4b2a2
MD
461 /*
462 * Try to open read-only again after creation.
463 */
33bbeb90 464 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
58d4b2a2
MD
465 if (wait_shm_fd < 0) {
466 /*
467 * Real-only open did not work. It's a failure
468 * that prohibits using shm.
469 */
470 ERR("Error opening shm %s", sock_info->wait_shm_path);
471 goto end;
472 }
473 goto end;
474 } else if (pid == 0) {
475 int create_mode;
476
477 /* Child */
33bbeb90 478 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
58d4b2a2 479 if (sock_info->global)
33bbeb90 480 create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
58d4b2a2
MD
481 /*
482 * We're alone in a child process, so we can modify the
483 * process-wide umask.
484 */
33bbeb90 485 umask(~create_mode);
58d4b2a2 486 /*
33bbeb90
MD
487 * Try creating shm (or get rw access).
488 * We don't do an exclusive open, because we allow other
489 * processes to create+ftruncate it concurrently.
58d4b2a2
MD
490 */
491 wait_shm_fd = shm_open(sock_info->wait_shm_path,
492 O_RDWR | O_CREAT, create_mode);
493 if (wait_shm_fd >= 0) {
494 ret = ftruncate(wait_shm_fd, mmap_size);
495 if (ret) {
496 PERROR("ftruncate");
497 exit(EXIT_FAILURE);
498 }
58d4b2a2
MD
499 exit(EXIT_SUCCESS);
500 }
33bbeb90
MD
501 /*
502 * For local shm, we need to have rw access to accept
503 * opening it: this means the local sessiond will be
504 * able to wake us up. For global shm, we open it even
505 * if rw access is not granted, because the root.root
506 * sessiond will be able to override all rights and wake
507 * us up.
508 */
509 if (!sock_info->global && errno != EACCES) {
58d4b2a2
MD
510 ERR("Error opening shm %s", sock_info->wait_shm_path);
511 exit(EXIT_FAILURE);
512 }
513 /*
33bbeb90
MD
514 * The shm exists, but we cannot open it RW. Report
515 * success.
58d4b2a2
MD
516 */
517 exit(EXIT_SUCCESS);
518 } else {
519 return -1;
7fc90dca 520 }
58d4b2a2 521end:
33bbeb90
MD
522 if (wait_shm_fd >= 0 && !sock_info->global) {
523 struct stat statbuf;
524
525 /*
526 * Ensure that our user is the owner of the shm file for
527 * local shm. If we do not own the file, it means our
528 * sessiond will not have access to wake us up (there is
529 * probably a rogue process trying to fake our
530 * sessiond). Fallback to polling method in this case.
531 */
532 ret = fstat(wait_shm_fd, &statbuf);
533 if (ret) {
534 PERROR("fstat");
535 goto error_close;
536 }
537 if (statbuf.st_uid != getuid())
538 goto error_close;
539 }
58d4b2a2 540 return wait_shm_fd;
33bbeb90
MD
541
542error_close:
543 ret = close(wait_shm_fd);
544 if (ret) {
545 PERROR("Error closing fd");
546 }
547 return -1;
58d4b2a2
MD
548}
549
550static
551char *get_map_shm(struct sock_info *sock_info)
552{
553 size_t mmap_size = sysconf(_SC_PAGE_SIZE);
554 int wait_shm_fd, ret;
555 char *wait_shm_mmap;
556
557 wait_shm_fd = get_wait_shm(sock_info, mmap_size);
558 if (wait_shm_fd < 0) {
559 goto error;
44e073f5 560 }
7fc90dca
MD
561 wait_shm_mmap = mmap(NULL, mmap_size, PROT_READ,
562 MAP_SHARED, wait_shm_fd, 0);
7fc90dca
MD
563 /* close shm fd immediately after taking the mmap reference */
564 ret = close(wait_shm_fd);
565 if (ret) {
33bbeb90
MD
566 PERROR("Error closing fd");
567 }
568 if (wait_shm_mmap == MAP_FAILED) {
569 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
570 goto error;
7fc90dca
MD
571 }
572 return wait_shm_mmap;
573
574error:
575 return NULL;
576}
577
578static
579void wait_for_sessiond(struct sock_info *sock_info)
580{
f5f94532 581 int ret, oldtype;
80e2814b 582
7fc90dca
MD
583 ust_lock();
584 if (lttng_ust_comm_should_quit) {
585 goto quit;
586 }
37ed587a
MD
587 if (wait_poll_fallback) {
588 goto error;
589 }
7fc90dca
MD
590 if (!sock_info->wait_shm_mmap) {
591 sock_info->wait_shm_mmap = get_map_shm(sock_info);
592 if (!sock_info->wait_shm_mmap)
593 goto error;
594 }
595 ust_unlock();
596
597 DBG("Waiting for %s apps sessiond", sock_info->name);
f5f94532
MD
598 /*
599 * sys_futex does not honor pthread cancel requests. Set to
600 * async.
601 */
602 ret = pthread_setcanceltype(PTHREAD_CANCEL_ASYNCHRONOUS, &oldtype);
603 if (ret) {
604 ERR("Error setting thread cancel type");
605 }
80e2814b
MD
606 /* Wait for futex wakeup */
607 if (uatomic_read((int32_t *) sock_info->wait_shm_mmap) == 0) {
608 ret = futex_async((int32_t *) sock_info->wait_shm_mmap,
609 FUTEX_WAIT, 0, NULL, NULL, 0);
80e2814b 610 if (ret < 0) {
37ed587a
MD
611 if (errno == EFAULT) {
612 wait_poll_fallback = 1;
a8b870ad 613 DBG(
37ed587a
MD
614"Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
615"do not support FUTEX_WAKE on read-only memory mappings correctly. "
616"Please upgrade your kernel "
617"(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
618"mainline). LTTng-UST will use polling mode fallback.");
cd27263b
MD
619 if (ust_debug())
620 PERROR("futex");
37ed587a 621 }
80e2814b
MD
622 }
623 }
f5f94532
MD
624 ret = pthread_setcanceltype(oldtype, &oldtype);
625 if (ret) {
626 ERR("Error setting thread cancel type");
627 }
7fc90dca
MD
628 return;
629
630quit:
631 ust_unlock();
632 return;
633
634error:
635 ust_unlock();
7fc90dca 636 return;
46050b1a
MD
637}
638
1ea11eab
MD
639/*
640 * This thread does not allocate any resource, except within
641 * handle_message, within mutex protection. This mutex protects against
642 * fork and exit.
643 * The other moment it allocates resources is at socket connexion, which
644 * is also protected by the mutex.
645 */
d9e99d10
MD
646static
647void *ust_listener_thread(void *arg)
648{
1ea11eab 649 struct sock_info *sock_info = arg;
c0eedf81 650 int sock, ret, prev_connect_failed = 0, has_waited = 0;
d9e99d10 651
9eb62b9c
MD
652 /* Restart trying to connect to the session daemon */
653restart:
c0eedf81
MD
654 if (prev_connect_failed) {
655 /* Wait for sessiond availability with pipe */
656 wait_for_sessiond(sock_info);
657 if (has_waited) {
658 has_waited = 0;
659 /*
660 * Sleep for 5 seconds before retrying after a
661 * sequence of failure / wait / failure. This
662 * deals with a killed or broken session daemon.
663 */
664 sleep(5);
665 }
666 has_waited = 1;
667 prev_connect_failed = 0;
668 }
17dfb34b 669 ust_lock();
1ea11eab
MD
670
671 if (lttng_ust_comm_should_quit) {
17dfb34b 672 ust_unlock();
1ea11eab
MD
673 goto quit;
674 }
9eb62b9c 675
1ea11eab
MD
676 if (sock_info->socket != -1) {
677 ret = close(sock_info->socket);
678 if (ret) {
11ff9c7d 679 ERR("Error closing %s apps socket", sock_info->name);
1ea11eab
MD
680 }
681 sock_info->socket = -1;
682 }
46050b1a 683
9eb62b9c 684 /* Register */
57773204 685 ret = ustcomm_connect_unix_sock(sock_info->sock_path);
9eb62b9c 686 if (ret < 0) {
4d3c9523 687 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
c0eedf81 688 prev_connect_failed = 1;
11ff9c7d
MD
689 /*
690 * If we cannot find the sessiond daemon, don't delay
691 * constructor execution.
692 */
edaa1431 693 ret = handle_register_done(sock_info);
11ff9c7d 694 assert(!ret);
17dfb34b 695 ust_unlock();
1ea11eab 696 goto restart;
46050b1a
MD
697 }
698
699 sock_info->socket = sock = ret;
700
701 /*
702 * Create only one root handle per listener thread for the whole
703 * process lifetime.
704 */
705 if (sock_info->root_handle == -1) {
706 ret = lttng_abi_create_root_handle();
a51070bb 707 if (ret < 0) {
46050b1a 708 ERR("Error creating root handle");
17dfb34b 709 ust_unlock();
46050b1a
MD
710 goto quit;
711 }
712 sock_info->root_handle = ret;
9eb62b9c 713 }
1ea11eab 714
9eb62b9c
MD
715 ret = register_app_to_sessiond(sock);
716 if (ret < 0) {
11ff9c7d 717 ERR("Error registering to %s apps socket", sock_info->name);
c0eedf81 718 prev_connect_failed = 1;
11ff9c7d
MD
719 /*
720 * If we cannot register to the sessiond daemon, don't
721 * delay constructor execution.
722 */
edaa1431 723 ret = handle_register_done(sock_info);
11ff9c7d 724 assert(!ret);
17dfb34b 725 ust_unlock();
9eb62b9c
MD
726 goto restart;
727 }
17dfb34b 728 ust_unlock();
46050b1a 729
d9e99d10
MD
730 for (;;) {
731 ssize_t len;
57773204 732 struct ustcomm_ust_msg lum;
d9e99d10 733
57773204 734 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
d9e99d10
MD
735 switch (len) {
736 case 0: /* orderly shutdown */
11ff9c7d 737 DBG("%s ltt-sessiond has performed an orderly shutdown\n", sock_info->name);
8236ba10
MD
738 ust_lock();
739 /*
740 * Either sessiond has shutdown or refused us by closing the socket.
741 * In either case, we don't want to delay construction execution,
742 * and we need to wait before retry.
743 */
744 prev_connect_failed = 1;
745 /*
746 * If we cannot register to the sessiond daemon, don't
747 * delay constructor execution.
748 */
749 ret = handle_register_done(sock_info);
750 assert(!ret);
751 ust_unlock();
d9e99d10 752 goto end;
e7723462 753 case sizeof(lum):
d9e99d10 754 DBG("message received\n");
11ff9c7d 755 ret = handle_message(sock_info, sock, &lum);
2a80c9d8 756 if (ret < 0) {
11ff9c7d 757 ERR("Error handling message for %s socket", sock_info->name);
d9e99d10
MD
758 }
759 continue;
760 case -1:
8236ba10 761 DBG("Receive failed from lttng-sessiond with errno %d", errno);
d9e99d10 762 if (errno == ECONNRESET) {
11ff9c7d 763 ERR("%s remote end closed connection\n", sock_info->name);
d9e99d10
MD
764 goto end;
765 }
766 goto end;
767 default:
11ff9c7d 768 ERR("incorrect message size (%s socket): %zd\n", sock_info->name, len);
d9e99d10
MD
769 continue;
770 }
771
772 }
773end:
9eb62b9c 774 goto restart; /* try to reconnect */
1ea11eab 775quit:
d9e99d10
MD
776 return NULL;
777}
778
cf12a773
MD
779/*
780 * Return values: -1: don't wait. 0: wait forever. 1: timeout wait.
781 */
11ff9c7d
MD
782static
783int get_timeout(struct timespec *constructor_timeout)
784{
cf12a773
MD
785 long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
786 char *str_delay;
11ff9c7d
MD
787 int ret;
788
69400ac4 789 str_delay = getenv("LTTNG_UST_REGISTER_TIMEOUT");
cf12a773
MD
790 if (str_delay) {
791 constructor_delay_ms = strtol(str_delay, NULL, 10);
792 }
793
794 switch (constructor_delay_ms) {
795 case -1:/* fall-through */
796 case 0:
797 return constructor_delay_ms;
798 default:
799 break;
800 }
801
802 /*
803 * If we are unable to find the current time, don't wait.
804 */
805 ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
806 if (ret) {
807 return -1;
808 }
95259bd0
MD
809 constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
810 constructor_timeout->tv_nsec +=
811 (constructor_delay_ms % 1000UL) * 1000000UL;
11ff9c7d
MD
812 if (constructor_timeout->tv_nsec >= 1000000000UL) {
813 constructor_timeout->tv_sec++;
814 constructor_timeout->tv_nsec -= 1000000000UL;
815 }
cf12a773 816 return 1;
11ff9c7d 817}
d9e99d10 818
2691221a
MD
819/*
820 * sessiond monitoring thread: monitor presence of global and per-user
821 * sessiond by polling the application common named pipe.
822 */
823/* TODO */
824
edaa1431 825void __attribute__((constructor)) lttng_ust_init(void)
2691221a 826{
11ff9c7d 827 struct timespec constructor_timeout;
cf12a773 828 int timeout_mode;
2691221a
MD
829 int ret;
830
edaa1431
MD
831 if (uatomic_xchg(&initialized, 1) == 1)
832 return;
833
834 /*
835 * We want precise control over the order in which we construct
836 * our sub-libraries vs starting to receive commands from
837 * sessiond (otherwise leading to errors when trying to create
838 * sessiond before the init functions are completed).
839 */
2691221a 840 init_usterr();
edaa1431
MD
841 init_tracepoint();
842 ltt_ring_buffer_metadata_client_init();
843 ltt_ring_buffer_client_overwrite_init();
844 ltt_ring_buffer_client_discard_init();
2691221a 845
cf12a773 846 timeout_mode = get_timeout(&constructor_timeout);
11ff9c7d 847
95259bd0 848 ret = sem_init(&constructor_wait, 0, 0);
11ff9c7d
MD
849 assert(!ret);
850
8d20bf54 851 ret = setup_local_apps();
2691221a 852 if (ret) {
8d20bf54 853 ERR("Error setting up to local apps");
2691221a 854 }
1ea11eab
MD
855 ret = pthread_create(&local_apps.ust_listener, NULL,
856 ust_listener_thread, &local_apps);
11ff9c7d 857
8d20bf54
MD
858 if (local_apps.allowed) {
859 ret = pthread_create(&global_apps.ust_listener, NULL,
860 ust_listener_thread, &global_apps);
861 } else {
862 handle_register_done(&local_apps);
863 }
864
cf12a773
MD
865 switch (timeout_mode) {
866 case 1: /* timeout wait */
95259bd0
MD
867 do {
868 ret = sem_timedwait(&constructor_wait,
869 &constructor_timeout);
870 } while (ret < 0 && errno == EINTR);
cf12a773
MD
871 if (ret < 0 && errno == ETIMEDOUT) {
872 ERR("Timed out waiting for ltt-sessiond");
873 } else {
874 assert(!ret);
875 }
876 break;
7b766b16 877 case -1:/* wait forever */
95259bd0
MD
878 do {
879 ret = sem_wait(&constructor_wait);
880 } while (ret < 0 && errno == EINTR);
11ff9c7d 881 assert(!ret);
cf12a773 882 break;
7b766b16 883 case 0: /* no timeout */
cf12a773 884 break;
11ff9c7d 885 }
2691221a
MD
886}
887
17dfb34b
MD
888static
889void lttng_ust_cleanup(int exiting)
890{
f5f94532 891 cleanup_sock_info(&global_apps);
17dfb34b 892 if (local_apps.allowed) {
f5f94532 893 cleanup_sock_info(&local_apps);
17dfb34b
MD
894 }
895 lttng_ust_abi_exit();
003fedf4 896 lttng_ust_events_exit();
17dfb34b
MD
897 ltt_ring_buffer_client_discard_exit();
898 ltt_ring_buffer_client_overwrite_exit();
899 ltt_ring_buffer_metadata_client_exit();
900 exit_tracepoint();
901 if (!exiting) {
902 /* Reinitialize values for fork */
903 sem_count = 2;
904 lttng_ust_comm_should_quit = 0;
905 initialized = 0;
906 }
907}
908
edaa1431 909void __attribute__((destructor)) lttng_ust_exit(void)
2691221a
MD
910{
911 int ret;
912
9eb62b9c
MD
913 /*
914 * Using pthread_cancel here because:
915 * A) we don't want to hang application teardown.
916 * B) the thread is not allocating any resource.
917 */
1ea11eab
MD
918
919 /*
920 * Require the communication thread to quit. Synchronize with
921 * mutexes to ensure it is not in a mutex critical section when
922 * pthread_cancel is later called.
923 */
17dfb34b 924 ust_lock();
1ea11eab 925 lttng_ust_comm_should_quit = 1;
17dfb34b 926 ust_unlock();
1ea11eab 927
f5f94532 928 /* cancel threads */
1ea11eab 929 ret = pthread_cancel(global_apps.ust_listener);
9eb62b9c
MD
930 if (ret) {
931 ERR("Error cancelling global ust listener thread");
2691221a 932 }
8d20bf54
MD
933 if (local_apps.allowed) {
934 ret = pthread_cancel(local_apps.ust_listener);
935 if (ret) {
936 ERR("Error cancelling local ust listener thread");
937 }
8d20bf54 938 }
f5f94532
MD
939 /* join threads */
940 ret = pthread_join(global_apps.ust_listener, NULL);
941 if (ret) {
942 ERR("Error joining global ust listener thread");
943 }
944 if (local_apps.allowed) {
945 ret = pthread_join(local_apps.ust_listener, NULL);
946 if (ret) {
947 ERR("Error joining local ust listener thread");
948 }
949 }
17dfb34b 950 lttng_ust_cleanup(1);
2691221a 951}
e822f505
MD
952
953/*
954 * We exclude the worker threads across fork and clone (except
955 * CLONE_VM), because these system calls only keep the forking thread
956 * running in the child. Therefore, we don't want to call fork or clone
957 * in the middle of an tracepoint or ust tracing state modification.
958 * Holding this mutex protects these structures across fork and clone.
959 */
b728d87e 960void ust_before_fork(sigset_t *save_sigset)
e822f505
MD
961{
962 /*
963 * Disable signals. This is to avoid that the child intervenes
964 * before it is properly setup for tracing. It is safer to
965 * disable all signals, because then we know we are not breaking
966 * anything by restoring the original mask.
967 */
968 sigset_t all_sigs;
969 int ret;
970
971 /* Disable signals */
972 sigfillset(&all_sigs);
b728d87e 973 ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
e822f505
MD
974 if (ret == -1) {
975 PERROR("sigprocmask");
976 }
17dfb34b 977 ust_lock();
e822f505
MD
978 rcu_bp_before_fork();
979}
980
b728d87e 981static void ust_after_fork_common(sigset_t *restore_sigset)
e822f505
MD
982{
983 int ret;
984
17dfb34b
MD
985 DBG("process %d", getpid());
986 ust_unlock();
e822f505 987 /* Restore signals */
23c8854a 988 ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
e822f505
MD
989 if (ret == -1) {
990 PERROR("sigprocmask");
991 }
992}
993
b728d87e 994void ust_after_fork_parent(sigset_t *restore_sigset)
e822f505 995{
17dfb34b 996 DBG("process %d", getpid());
e822f505
MD
997 rcu_bp_after_fork_parent();
998 /* Release mutexes and reenable signals */
b728d87e 999 ust_after_fork_common(restore_sigset);
e822f505
MD
1000}
1001
17dfb34b
MD
1002/*
1003 * After fork, in the child, we need to cleanup all the leftover state,
1004 * except the worker thread which already magically disappeared thanks
1005 * to the weird Linux fork semantics. After tyding up, we call
1006 * lttng_ust_init() again to start over as a new PID.
1007 *
1008 * This is meant for forks() that have tracing in the child between the
1009 * fork and following exec call (if there is any).
1010 */
b728d87e 1011void ust_after_fork_child(sigset_t *restore_sigset)
e822f505 1012{
17dfb34b 1013 DBG("process %d", getpid());
e822f505
MD
1014 /* Release urcu mutexes */
1015 rcu_bp_after_fork_child();
17dfb34b 1016 lttng_ust_cleanup(0);
a93bfc45 1017 lttng_context_vtid_reset();
e822f505 1018 /* Release mutexes and reenable signals */
b728d87e 1019 ust_after_fork_common(restore_sigset);
318dfea9 1020 lttng_ust_init();
e822f505 1021}
This page took 0.076435 seconds and 4 git commands to generate.