Revert "Fix timestamps for slow-paced event rates"
[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>
2629549e 25#include <sys/prctl.h>
7fc90dca
MD
26#include <sys/mman.h>
27#include <sys/stat.h>
58d4b2a2
MD
28#include <sys/types.h>
29#include <sys/wait.h>
7fc90dca 30#include <fcntl.h>
2691221a
MD
31#include <unistd.h>
32#include <errno.h>
d9e99d10 33#include <pthread.h>
11ff9c7d
MD
34#include <semaphore.h>
35#include <time.h>
1ea11eab 36#include <assert.h>
e822f505 37#include <signal.h>
95259bd0 38#include <urcu/uatomic.h>
80e2814b 39#include <urcu/futex.h>
c117fb1b 40#include <urcu/compiler.h>
1ea11eab 41
4318ae1b 42#include <lttng/ust-events.h>
4318ae1b 43#include <lttng/ust-abi.h>
4318ae1b 44#include <lttng/ust.h>
44c72f10
MD
45#include <ust-comm.h>
46#include <usterr-signal-safe.h>
47#include "tracepoint-internal.h"
b751f722 48#include "ltt-tracer-core.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;
2629549e 162 int prctl_ret;
2691221a 163 struct {
e44418f3
MD
164 uint32_t major;
165 uint32_t minor;
2691221a 166 pid_t pid;
5c33bde8 167 pid_t ppid;
2691221a 168 uid_t uid;
83610856 169 gid_t gid;
c117fb1b 170 uint32_t bits_per_long;
2629549e 171 char name[16]; /* process name */
2691221a
MD
172 } reg_msg;
173
e44418f3
MD
174 reg_msg.major = LTTNG_UST_COMM_VERSION_MAJOR;
175 reg_msg.minor = LTTNG_UST_COMM_VERSION_MINOR;
2691221a 176 reg_msg.pid = getpid();
5c33bde8 177 reg_msg.ppid = getppid();
2691221a 178 reg_msg.uid = getuid();
83610856 179 reg_msg.gid = getgid();
c117fb1b 180 reg_msg.bits_per_long = CAA_BITS_PER_LONG;
2629549e
MD
181 prctl_ret = prctl(PR_GET_NAME, (unsigned long) reg_msg.name, 0, 0, 0);
182 if (prctl_ret) {
183 ERR("Error executing prctl");
184 return -errno;
185 }
2691221a 186
57773204 187 ret = ustcomm_send_unix_sock(socket, &reg_msg, sizeof(reg_msg));
2691221a
MD
188 if (ret >= 0 && ret != sizeof(reg_msg))
189 return -EIO;
190 return ret;
191}
192
d9e99d10 193static
57773204 194int send_reply(int sock, struct ustcomm_ust_reply *lur)
d9e99d10 195{
9eb62b9c 196 ssize_t len;
d3a492d1 197
57773204 198 len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
d3a492d1 199 switch (len) {
a4be8962 200 case sizeof(*lur):
d3a492d1
MD
201 DBG("message successfully sent");
202 return 0;
203 case -1:
204 if (errno == ECONNRESET) {
205 printf("remote end closed connection\n");
206 return 0;
207 }
208 return -1;
209 default:
210 printf("incorrect message size: %zd\n", len);
211 return -1;
212 }
213}
214
215static
edaa1431 216int handle_register_done(struct sock_info *sock_info)
11ff9c7d
MD
217{
218 int ret;
219
edaa1431
MD
220 if (sock_info->constructor_sem_posted)
221 return 0;
222 sock_info->constructor_sem_posted = 1;
56cd7e2f
MD
223 if (uatomic_read(&sem_count) <= 0) {
224 return 0;
225 }
95259bd0
MD
226 ret = uatomic_add_return(&sem_count, -1);
227 if (ret == 0) {
228 ret = sem_post(&constructor_wait);
229 assert(!ret);
230 }
11ff9c7d
MD
231 return 0;
232}
233
234static
235int handle_message(struct sock_info *sock_info,
57773204 236 int sock, struct ustcomm_ust_msg *lum)
d3a492d1 237{
1ea11eab 238 int ret = 0;
b61ce3b2 239 const struct lttng_ust_objd_ops *ops;
57773204 240 struct ustcomm_ust_reply lur;
193183fb 241 int shm_fd, wait_fd;
ef9ff354 242 union ust_args args;
1ea11eab 243
17dfb34b 244 ust_lock();
1ea11eab 245
46050b1a
MD
246 memset(&lur, 0, sizeof(lur));
247
1ea11eab 248 if (lttng_ust_comm_should_quit) {
46050b1a 249 ret = -EPERM;
1ea11eab
MD
250 goto end;
251 }
9eb62b9c 252
46050b1a
MD
253 ops = objd_ops(lum->handle);
254 if (!ops) {
255 ret = -ENOENT;
256 goto end;
1ea11eab 257 }
46050b1a
MD
258
259 switch (lum->cmd) {
11ff9c7d
MD
260 case LTTNG_UST_REGISTER_DONE:
261 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
edaa1431 262 ret = handle_register_done(sock_info);
11ff9c7d
MD
263 else
264 ret = -EINVAL;
265 break;
46050b1a
MD
266 case LTTNG_UST_RELEASE:
267 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
268 ret = -EPERM;
269 else
d4419b81 270 ret = lttng_ust_objd_unref(lum->handle);
d9e99d10
MD
271 break;
272 default:
46050b1a
MD
273 if (ops->cmd)
274 ret = ops->cmd(lum->handle, lum->cmd,
ef9ff354
MD
275 (unsigned long) &lum->u,
276 &args);
46050b1a
MD
277 else
278 ret = -ENOSYS;
279 break;
d9e99d10 280 }
46050b1a 281
1ea11eab 282end:
46050b1a
MD
283 lur.handle = lum->handle;
284 lur.cmd = lum->cmd;
285 lur.ret_val = ret;
286 if (ret >= 0) {
57773204 287 lur.ret_code = USTCOMM_OK;
46050b1a 288 } else {
57773204 289 //lur.ret_code = USTCOMM_SESSION_FAIL;
193183fb 290 lur.ret_code = ret;
46050b1a 291 }
193183fb
MD
292 switch (lum->cmd) {
293 case LTTNG_UST_STREAM:
381c0f1e
MD
294 /*
295 * Special-case reply to send stream info.
296 * Use lum.u output.
297 */
ef9ff354
MD
298 lur.u.stream.memory_map_size = *args.stream.memory_map_size;
299 shm_fd = *args.stream.shm_fd;
300 wait_fd = *args.stream.wait_fd;
193183fb 301 break;
824f40b8 302 case LTTNG_UST_METADATA:
193183fb 303 case LTTNG_UST_CHANNEL:
ef9ff354
MD
304 lur.u.channel.memory_map_size = *args.channel.memory_map_size;
305 shm_fd = *args.channel.shm_fd;
306 wait_fd = *args.channel.wait_fd;
193183fb 307 break;
23c8854a 308 case LTTNG_UST_TRACER_VERSION:
57773204
MD
309 lur.u.version = lum->u.version;
310 break;
b115631f 311 case LTTNG_UST_TRACEPOINT_LIST_GET:
cbef6901 312 memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
b115631f 313 break;
381c0f1e 314 }
46050b1a 315 ret = send_reply(sock, &lur);
193183fb
MD
316 if (ret < 0) {
317 perror("error sending reply");
318 goto error;
319 }
46050b1a 320
824f40b8
MD
321 if ((lum->cmd == LTTNG_UST_STREAM
322 || lum->cmd == LTTNG_UST_CHANNEL
323 || lum->cmd == LTTNG_UST_METADATA)
57773204 324 && lur.ret_code == USTCOMM_OK) {
381c0f1e 325 /* we also need to send the file descriptors. */
57773204 326 ret = ustcomm_send_fds_unix_sock(sock,
193183fb 327 &shm_fd, &shm_fd,
381c0f1e
MD
328 1, sizeof(int));
329 if (ret < 0) {
330 perror("send shm_fd");
331 goto error;
332 }
57773204 333 ret = ustcomm_send_fds_unix_sock(sock,
193183fb 334 &wait_fd, &wait_fd,
381c0f1e
MD
335 1, sizeof(int));
336 if (ret < 0) {
337 perror("send wait_fd");
338 goto error;
339 }
340 }
ef9ff354
MD
341 /*
342 * We still have the memory map reference, and the fds have been
3ad5f707
MD
343 * sent to the sessiond. We can therefore close those fds. Note
344 * that we keep the write side of the wait_fd open, but close
345 * the read side.
ef9ff354
MD
346 */
347 if (lur.ret_code == USTCOMM_OK) {
348 switch (lum->cmd) {
349 case LTTNG_UST_STREAM:
350 if (shm_fd >= 0) {
351 ret = close(shm_fd);
352 if (ret) {
353 PERROR("Error closing stream shm_fd");
354 }
355 *args.stream.shm_fd = -1;
356 }
357 if (wait_fd >= 0) {
358 ret = close(wait_fd);
359 if (ret) {
360 PERROR("Error closing stream wait_fd");
361 }
362 *args.stream.wait_fd = -1;
363 }
364 break;
365 case LTTNG_UST_METADATA:
366 case LTTNG_UST_CHANNEL:
367 if (shm_fd >= 0) {
368 ret = close(shm_fd);
369 if (ret) {
370 PERROR("Error closing channel shm_fd");
371 }
372 *args.channel.shm_fd = -1;
373 }
374 if (wait_fd >= 0) {
375 ret = close(wait_fd);
376 if (ret) {
377 PERROR("Error closing channel wait_fd");
378 }
379 *args.channel.wait_fd = -1;
380 }
381 break;
382 }
383 }
384
381c0f1e 385error:
17dfb34b 386 ust_unlock();
1ea11eab 387 return ret;
d9e99d10
MD
388}
389
46050b1a
MD
390static
391void cleanup_sock_info(struct sock_info *sock_info)
392{
393 int ret;
394
395 if (sock_info->socket != -1) {
396 ret = close(sock_info->socket);
397 if (ret) {
7fc90dca 398 ERR("Error closing apps socket");
46050b1a
MD
399 }
400 sock_info->socket = -1;
401 }
402 if (sock_info->root_handle != -1) {
d4419b81 403 ret = lttng_ust_objd_unref(sock_info->root_handle);
46050b1a
MD
404 if (ret) {
405 ERR("Error unref root handle");
406 }
407 sock_info->root_handle = -1;
408 }
318dfea9 409 sock_info->constructor_sem_posted = 0;
7fc90dca
MD
410 if (sock_info->wait_shm_mmap) {
411 ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE));
412 if (ret) {
413 ERR("Error unmapping wait shm");
414 }
415 sock_info->wait_shm_mmap = NULL;
416 }
417}
418
58d4b2a2 419/*
33bbeb90
MD
420 * Using fork to set umask in the child process (not multi-thread safe).
421 * We deal with the shm_open vs ftruncate race (happening when the
422 * sessiond owns the shm and does not let everybody modify it, to ensure
423 * safety against shm_unlink) by simply letting the mmap fail and
424 * retrying after a few seconds.
425 * For global shm, everybody has rw access to it until the sessiond
426 * starts.
58d4b2a2 427 */
7fc90dca 428static
58d4b2a2 429int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
7fc90dca 430{
7fc90dca 431 int wait_shm_fd, ret;
58d4b2a2 432 pid_t pid;
44e073f5 433
58d4b2a2 434 /*
33bbeb90 435 * Try to open read-only.
58d4b2a2 436 */
33bbeb90 437 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
58d4b2a2
MD
438 if (wait_shm_fd >= 0) {
439 goto end;
440 } else if (wait_shm_fd < 0 && errno != ENOENT) {
441 /*
33bbeb90
MD
442 * Real-only open did not work, and it's not because the
443 * entry was not present. It's a failure that prohibits
444 * using shm.
58d4b2a2 445 */
7fc90dca 446 ERR("Error opening shm %s", sock_info->wait_shm_path);
58d4b2a2 447 goto end;
7fc90dca
MD
448 }
449 /*
58d4b2a2
MD
450 * If the open failed because the file did not exist, try
451 * creating it ourself.
7fc90dca 452 */
58d4b2a2
MD
453 pid = fork();
454 if (pid > 0) {
455 int status;
456
457 /*
458 * Parent: wait for child to return, in which case the
459 * shared memory map will have been created.
460 */
461 pid = wait(&status);
b7d3cb32 462 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
58d4b2a2
MD
463 wait_shm_fd = -1;
464 goto end;
7fc90dca 465 }
58d4b2a2
MD
466 /*
467 * Try to open read-only again after creation.
468 */
33bbeb90 469 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
58d4b2a2
MD
470 if (wait_shm_fd < 0) {
471 /*
472 * Real-only open did not work. It's a failure
473 * that prohibits using shm.
474 */
475 ERR("Error opening shm %s", sock_info->wait_shm_path);
476 goto end;
477 }
478 goto end;
479 } else if (pid == 0) {
480 int create_mode;
481
482 /* Child */
33bbeb90 483 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
58d4b2a2 484 if (sock_info->global)
33bbeb90 485 create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
58d4b2a2
MD
486 /*
487 * We're alone in a child process, so we can modify the
488 * process-wide umask.
489 */
33bbeb90 490 umask(~create_mode);
58d4b2a2 491 /*
33bbeb90
MD
492 * Try creating shm (or get rw access).
493 * We don't do an exclusive open, because we allow other
494 * processes to create+ftruncate it concurrently.
58d4b2a2
MD
495 */
496 wait_shm_fd = shm_open(sock_info->wait_shm_path,
497 O_RDWR | O_CREAT, create_mode);
498 if (wait_shm_fd >= 0) {
499 ret = ftruncate(wait_shm_fd, mmap_size);
500 if (ret) {
501 PERROR("ftruncate");
502 exit(EXIT_FAILURE);
503 }
58d4b2a2
MD
504 exit(EXIT_SUCCESS);
505 }
33bbeb90
MD
506 /*
507 * For local shm, we need to have rw access to accept
508 * opening it: this means the local sessiond will be
509 * able to wake us up. For global shm, we open it even
510 * if rw access is not granted, because the root.root
511 * sessiond will be able to override all rights and wake
512 * us up.
513 */
514 if (!sock_info->global && errno != EACCES) {
58d4b2a2
MD
515 ERR("Error opening shm %s", sock_info->wait_shm_path);
516 exit(EXIT_FAILURE);
517 }
518 /*
33bbeb90
MD
519 * The shm exists, but we cannot open it RW. Report
520 * success.
58d4b2a2
MD
521 */
522 exit(EXIT_SUCCESS);
523 } else {
524 return -1;
7fc90dca 525 }
58d4b2a2 526end:
33bbeb90
MD
527 if (wait_shm_fd >= 0 && !sock_info->global) {
528 struct stat statbuf;
529
530 /*
531 * Ensure that our user is the owner of the shm file for
532 * local shm. If we do not own the file, it means our
533 * sessiond will not have access to wake us up (there is
534 * probably a rogue process trying to fake our
535 * sessiond). Fallback to polling method in this case.
536 */
537 ret = fstat(wait_shm_fd, &statbuf);
538 if (ret) {
539 PERROR("fstat");
540 goto error_close;
541 }
542 if (statbuf.st_uid != getuid())
543 goto error_close;
544 }
58d4b2a2 545 return wait_shm_fd;
33bbeb90
MD
546
547error_close:
548 ret = close(wait_shm_fd);
549 if (ret) {
550 PERROR("Error closing fd");
551 }
552 return -1;
58d4b2a2
MD
553}
554
555static
556char *get_map_shm(struct sock_info *sock_info)
557{
558 size_t mmap_size = sysconf(_SC_PAGE_SIZE);
559 int wait_shm_fd, ret;
560 char *wait_shm_mmap;
561
562 wait_shm_fd = get_wait_shm(sock_info, mmap_size);
563 if (wait_shm_fd < 0) {
564 goto error;
44e073f5 565 }
7fc90dca
MD
566 wait_shm_mmap = mmap(NULL, mmap_size, PROT_READ,
567 MAP_SHARED, wait_shm_fd, 0);
7fc90dca
MD
568 /* close shm fd immediately after taking the mmap reference */
569 ret = close(wait_shm_fd);
570 if (ret) {
33bbeb90
MD
571 PERROR("Error closing fd");
572 }
573 if (wait_shm_mmap == MAP_FAILED) {
574 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
575 goto error;
7fc90dca
MD
576 }
577 return wait_shm_mmap;
578
579error:
580 return NULL;
581}
582
583static
584void wait_for_sessiond(struct sock_info *sock_info)
585{
80e2814b
MD
586 int ret;
587
7fc90dca
MD
588 ust_lock();
589 if (lttng_ust_comm_should_quit) {
590 goto quit;
591 }
37ed587a
MD
592 if (wait_poll_fallback) {
593 goto error;
594 }
7fc90dca
MD
595 if (!sock_info->wait_shm_mmap) {
596 sock_info->wait_shm_mmap = get_map_shm(sock_info);
597 if (!sock_info->wait_shm_mmap)
598 goto error;
599 }
600 ust_unlock();
601
602 DBG("Waiting for %s apps sessiond", sock_info->name);
80e2814b
MD
603 /* Wait for futex wakeup */
604 if (uatomic_read((int32_t *) sock_info->wait_shm_mmap) == 0) {
605 ret = futex_async((int32_t *) sock_info->wait_shm_mmap,
606 FUTEX_WAIT, 0, NULL, NULL, 0);
80e2814b 607 if (ret < 0) {
37ed587a
MD
608 if (errno == EFAULT) {
609 wait_poll_fallback = 1;
a8b870ad 610 DBG(
37ed587a
MD
611"Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
612"do not support FUTEX_WAKE on read-only memory mappings correctly. "
613"Please upgrade your kernel "
614"(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
615"mainline). LTTng-UST will use polling mode fallback.");
cd27263b
MD
616 if (ust_debug())
617 PERROR("futex");
37ed587a 618 }
80e2814b
MD
619 }
620 }
7fc90dca
MD
621 return;
622
623quit:
624 ust_unlock();
625 return;
626
627error:
628 ust_unlock();
7fc90dca 629 return;
46050b1a
MD
630}
631
1ea11eab
MD
632/*
633 * This thread does not allocate any resource, except within
634 * handle_message, within mutex protection. This mutex protects against
635 * fork and exit.
636 * The other moment it allocates resources is at socket connexion, which
637 * is also protected by the mutex.
638 */
d9e99d10
MD
639static
640void *ust_listener_thread(void *arg)
641{
1ea11eab 642 struct sock_info *sock_info = arg;
c0eedf81 643 int sock, ret, prev_connect_failed = 0, has_waited = 0;
d9e99d10 644
9eb62b9c
MD
645 /* Restart trying to connect to the session daemon */
646restart:
c0eedf81
MD
647 if (prev_connect_failed) {
648 /* Wait for sessiond availability with pipe */
649 wait_for_sessiond(sock_info);
650 if (has_waited) {
651 has_waited = 0;
652 /*
653 * Sleep for 5 seconds before retrying after a
654 * sequence of failure / wait / failure. This
655 * deals with a killed or broken session daemon.
656 */
657 sleep(5);
658 }
659 has_waited = 1;
660 prev_connect_failed = 0;
661 }
17dfb34b 662 ust_lock();
1ea11eab
MD
663
664 if (lttng_ust_comm_should_quit) {
17dfb34b 665 ust_unlock();
1ea11eab
MD
666 goto quit;
667 }
9eb62b9c 668
1ea11eab
MD
669 if (sock_info->socket != -1) {
670 ret = close(sock_info->socket);
671 if (ret) {
11ff9c7d 672 ERR("Error closing %s apps socket", sock_info->name);
1ea11eab
MD
673 }
674 sock_info->socket = -1;
675 }
46050b1a 676
9eb62b9c 677 /* Register */
57773204 678 ret = ustcomm_connect_unix_sock(sock_info->sock_path);
9eb62b9c 679 if (ret < 0) {
4d3c9523 680 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
c0eedf81 681 prev_connect_failed = 1;
11ff9c7d
MD
682 /*
683 * If we cannot find the sessiond daemon, don't delay
684 * constructor execution.
685 */
edaa1431 686 ret = handle_register_done(sock_info);
11ff9c7d 687 assert(!ret);
17dfb34b 688 ust_unlock();
1ea11eab 689 goto restart;
46050b1a
MD
690 }
691
692 sock_info->socket = sock = ret;
693
694 /*
695 * Create only one root handle per listener thread for the whole
696 * process lifetime.
697 */
698 if (sock_info->root_handle == -1) {
699 ret = lttng_abi_create_root_handle();
a51070bb 700 if (ret < 0) {
46050b1a 701 ERR("Error creating root handle");
17dfb34b 702 ust_unlock();
46050b1a
MD
703 goto quit;
704 }
705 sock_info->root_handle = ret;
9eb62b9c 706 }
1ea11eab 707
9eb62b9c
MD
708 ret = register_app_to_sessiond(sock);
709 if (ret < 0) {
11ff9c7d 710 ERR("Error registering to %s apps socket", sock_info->name);
c0eedf81 711 prev_connect_failed = 1;
11ff9c7d
MD
712 /*
713 * If we cannot register to the sessiond daemon, don't
714 * delay constructor execution.
715 */
edaa1431 716 ret = handle_register_done(sock_info);
11ff9c7d 717 assert(!ret);
17dfb34b 718 ust_unlock();
9eb62b9c
MD
719 goto restart;
720 }
17dfb34b 721 ust_unlock();
46050b1a 722
d9e99d10
MD
723 for (;;) {
724 ssize_t len;
57773204 725 struct ustcomm_ust_msg lum;
d9e99d10 726
57773204 727 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
d9e99d10
MD
728 switch (len) {
729 case 0: /* orderly shutdown */
11ff9c7d 730 DBG("%s ltt-sessiond has performed an orderly shutdown\n", sock_info->name);
8236ba10
MD
731 ust_lock();
732 /*
733 * Either sessiond has shutdown or refused us by closing the socket.
734 * In either case, we don't want to delay construction execution,
735 * and we need to wait before retry.
736 */
737 prev_connect_failed = 1;
738 /*
739 * If we cannot register to the sessiond daemon, don't
740 * delay constructor execution.
741 */
742 ret = handle_register_done(sock_info);
743 assert(!ret);
744 ust_unlock();
d9e99d10 745 goto end;
e7723462 746 case sizeof(lum):
d9e99d10 747 DBG("message received\n");
11ff9c7d 748 ret = handle_message(sock_info, sock, &lum);
2a80c9d8 749 if (ret < 0) {
11ff9c7d 750 ERR("Error handling message for %s socket", sock_info->name);
d9e99d10
MD
751 }
752 continue;
753 case -1:
8236ba10 754 DBG("Receive failed from lttng-sessiond with errno %d", errno);
d9e99d10 755 if (errno == ECONNRESET) {
11ff9c7d 756 ERR("%s remote end closed connection\n", sock_info->name);
d9e99d10
MD
757 goto end;
758 }
759 goto end;
760 default:
11ff9c7d 761 ERR("incorrect message size (%s socket): %zd\n", sock_info->name, len);
d9e99d10
MD
762 continue;
763 }
764
765 }
766end:
9eb62b9c 767 goto restart; /* try to reconnect */
1ea11eab 768quit:
d9e99d10
MD
769 return NULL;
770}
771
cf12a773
MD
772/*
773 * Return values: -1: don't wait. 0: wait forever. 1: timeout wait.
774 */
11ff9c7d
MD
775static
776int get_timeout(struct timespec *constructor_timeout)
777{
cf12a773
MD
778 long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
779 char *str_delay;
11ff9c7d
MD
780 int ret;
781
69400ac4 782 str_delay = getenv("LTTNG_UST_REGISTER_TIMEOUT");
cf12a773
MD
783 if (str_delay) {
784 constructor_delay_ms = strtol(str_delay, NULL, 10);
785 }
786
787 switch (constructor_delay_ms) {
788 case -1:/* fall-through */
789 case 0:
790 return constructor_delay_ms;
791 default:
792 break;
793 }
794
795 /*
796 * If we are unable to find the current time, don't wait.
797 */
798 ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
799 if (ret) {
800 return -1;
801 }
95259bd0
MD
802 constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
803 constructor_timeout->tv_nsec +=
804 (constructor_delay_ms % 1000UL) * 1000000UL;
11ff9c7d
MD
805 if (constructor_timeout->tv_nsec >= 1000000000UL) {
806 constructor_timeout->tv_sec++;
807 constructor_timeout->tv_nsec -= 1000000000UL;
808 }
cf12a773 809 return 1;
11ff9c7d 810}
d9e99d10 811
2691221a
MD
812/*
813 * sessiond monitoring thread: monitor presence of global and per-user
814 * sessiond by polling the application common named pipe.
815 */
816/* TODO */
817
edaa1431 818void __attribute__((constructor)) lttng_ust_init(void)
2691221a 819{
11ff9c7d 820 struct timespec constructor_timeout;
cf12a773 821 int timeout_mode;
2691221a
MD
822 int ret;
823
edaa1431
MD
824 if (uatomic_xchg(&initialized, 1) == 1)
825 return;
826
827 /*
828 * We want precise control over the order in which we construct
829 * our sub-libraries vs starting to receive commands from
830 * sessiond (otherwise leading to errors when trying to create
831 * sessiond before the init functions are completed).
832 */
2691221a 833 init_usterr();
edaa1431
MD
834 init_tracepoint();
835 ltt_ring_buffer_metadata_client_init();
836 ltt_ring_buffer_client_overwrite_init();
837 ltt_ring_buffer_client_discard_init();
2691221a 838
cf12a773 839 timeout_mode = get_timeout(&constructor_timeout);
11ff9c7d 840
95259bd0 841 ret = sem_init(&constructor_wait, 0, 0);
11ff9c7d
MD
842 assert(!ret);
843
8d20bf54 844 ret = setup_local_apps();
2691221a 845 if (ret) {
8d20bf54 846 ERR("Error setting up to local apps");
2691221a 847 }
1ea11eab
MD
848 ret = pthread_create(&local_apps.ust_listener, NULL,
849 ust_listener_thread, &local_apps);
11ff9c7d 850
8d20bf54
MD
851 if (local_apps.allowed) {
852 ret = pthread_create(&global_apps.ust_listener, NULL,
853 ust_listener_thread, &global_apps);
854 } else {
855 handle_register_done(&local_apps);
856 }
857
cf12a773
MD
858 switch (timeout_mode) {
859 case 1: /* timeout wait */
95259bd0
MD
860 do {
861 ret = sem_timedwait(&constructor_wait,
862 &constructor_timeout);
863 } while (ret < 0 && errno == EINTR);
cf12a773
MD
864 if (ret < 0 && errno == ETIMEDOUT) {
865 ERR("Timed out waiting for ltt-sessiond");
866 } else {
867 assert(!ret);
868 }
869 break;
7b766b16 870 case -1:/* wait forever */
95259bd0
MD
871 do {
872 ret = sem_wait(&constructor_wait);
873 } while (ret < 0 && errno == EINTR);
11ff9c7d 874 assert(!ret);
cf12a773 875 break;
7b766b16 876 case 0: /* no timeout */
cf12a773 877 break;
11ff9c7d 878 }
2691221a
MD
879}
880
17dfb34b
MD
881static
882void lttng_ust_cleanup(int exiting)
883{
884 cleanup_sock_info(&global_apps);
885 if (local_apps.allowed) {
886 cleanup_sock_info(&local_apps);
887 }
888 lttng_ust_abi_exit();
003fedf4 889 lttng_ust_events_exit();
17dfb34b
MD
890 ltt_ring_buffer_client_discard_exit();
891 ltt_ring_buffer_client_overwrite_exit();
892 ltt_ring_buffer_metadata_client_exit();
893 exit_tracepoint();
894 if (!exiting) {
895 /* Reinitialize values for fork */
896 sem_count = 2;
897 lttng_ust_comm_should_quit = 0;
898 initialized = 0;
899 }
900}
901
edaa1431 902void __attribute__((destructor)) lttng_ust_exit(void)
2691221a
MD
903{
904 int ret;
905
9eb62b9c
MD
906 /*
907 * Using pthread_cancel here because:
908 * A) we don't want to hang application teardown.
909 * B) the thread is not allocating any resource.
910 */
1ea11eab
MD
911
912 /*
913 * Require the communication thread to quit. Synchronize with
914 * mutexes to ensure it is not in a mutex critical section when
915 * pthread_cancel is later called.
916 */
17dfb34b 917 ust_lock();
1ea11eab 918 lttng_ust_comm_should_quit = 1;
17dfb34b 919 ust_unlock();
1ea11eab 920
1ea11eab 921 ret = pthread_cancel(global_apps.ust_listener);
9eb62b9c
MD
922 if (ret) {
923 ERR("Error cancelling global ust listener thread");
2691221a 924 }
8d20bf54
MD
925 if (local_apps.allowed) {
926 ret = pthread_cancel(local_apps.ust_listener);
927 if (ret) {
928 ERR("Error cancelling local ust listener thread");
929 }
8d20bf54 930 }
17dfb34b 931 lttng_ust_cleanup(1);
2691221a 932}
e822f505
MD
933
934/*
935 * We exclude the worker threads across fork and clone (except
936 * CLONE_VM), because these system calls only keep the forking thread
937 * running in the child. Therefore, we don't want to call fork or clone
938 * in the middle of an tracepoint or ust tracing state modification.
939 * Holding this mutex protects these structures across fork and clone.
940 */
b728d87e 941void ust_before_fork(sigset_t *save_sigset)
e822f505
MD
942{
943 /*
944 * Disable signals. This is to avoid that the child intervenes
945 * before it is properly setup for tracing. It is safer to
946 * disable all signals, because then we know we are not breaking
947 * anything by restoring the original mask.
948 */
949 sigset_t all_sigs;
950 int ret;
951
952 /* Disable signals */
953 sigfillset(&all_sigs);
b728d87e 954 ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
e822f505
MD
955 if (ret == -1) {
956 PERROR("sigprocmask");
957 }
17dfb34b 958 ust_lock();
e822f505
MD
959 rcu_bp_before_fork();
960}
961
b728d87e 962static void ust_after_fork_common(sigset_t *restore_sigset)
e822f505
MD
963{
964 int ret;
965
17dfb34b
MD
966 DBG("process %d", getpid());
967 ust_unlock();
e822f505 968 /* Restore signals */
23c8854a 969 ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
e822f505
MD
970 if (ret == -1) {
971 PERROR("sigprocmask");
972 }
973}
974
b728d87e 975void ust_after_fork_parent(sigset_t *restore_sigset)
e822f505 976{
17dfb34b 977 DBG("process %d", getpid());
e822f505
MD
978 rcu_bp_after_fork_parent();
979 /* Release mutexes and reenable signals */
b728d87e 980 ust_after_fork_common(restore_sigset);
e822f505
MD
981}
982
17dfb34b
MD
983/*
984 * After fork, in the child, we need to cleanup all the leftover state,
985 * except the worker thread which already magically disappeared thanks
986 * to the weird Linux fork semantics. After tyding up, we call
987 * lttng_ust_init() again to start over as a new PID.
988 *
989 * This is meant for forks() that have tracing in the child between the
990 * fork and following exec call (if there is any).
991 */
b728d87e 992void ust_after_fork_child(sigset_t *restore_sigset)
e822f505 993{
17dfb34b 994 DBG("process %d", getpid());
e822f505
MD
995 /* Release urcu mutexes */
996 rcu_bp_after_fork_child();
17dfb34b 997 lttng_ust_cleanup(0);
a93bfc45 998 lttng_context_vtid_reset();
e822f505 999 /* Release mutexes and reenable signals */
b728d87e 1000 ust_after_fork_common(restore_sigset);
318dfea9 1001 lttng_ust_init();
e822f505 1002}
This page took 0.072867 seconds and 4 git commands to generate.