Add DBG message when registering a probe
[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;
1ea11eab 242
17dfb34b 243 ust_lock();
1ea11eab 244
46050b1a
MD
245 memset(&lur, 0, sizeof(lur));
246
1ea11eab 247 if (lttng_ust_comm_should_quit) {
46050b1a 248 ret = -EPERM;
1ea11eab
MD
249 goto end;
250 }
9eb62b9c 251
46050b1a
MD
252 ops = objd_ops(lum->handle);
253 if (!ops) {
254 ret = -ENOENT;
255 goto end;
1ea11eab 256 }
46050b1a
MD
257
258 switch (lum->cmd) {
11ff9c7d
MD
259 case LTTNG_UST_REGISTER_DONE:
260 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
edaa1431 261 ret = handle_register_done(sock_info);
11ff9c7d
MD
262 else
263 ret = -EINVAL;
264 break;
46050b1a
MD
265 case LTTNG_UST_RELEASE:
266 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
267 ret = -EPERM;
268 else
d4419b81 269 ret = lttng_ust_objd_unref(lum->handle);
d9e99d10
MD
270 break;
271 default:
46050b1a
MD
272 if (ops->cmd)
273 ret = ops->cmd(lum->handle, lum->cmd,
274 (unsigned long) &lum->u);
275 else
276 ret = -ENOSYS;
277 break;
d9e99d10 278 }
46050b1a 279
1ea11eab 280end:
46050b1a
MD
281 lur.handle = lum->handle;
282 lur.cmd = lum->cmd;
283 lur.ret_val = ret;
284 if (ret >= 0) {
57773204 285 lur.ret_code = USTCOMM_OK;
46050b1a 286 } else {
57773204 287 //lur.ret_code = USTCOMM_SESSION_FAIL;
193183fb 288 lur.ret_code = ret;
46050b1a 289 }
193183fb
MD
290 switch (lum->cmd) {
291 case LTTNG_UST_STREAM:
381c0f1e
MD
292 /*
293 * Special-case reply to send stream info.
294 * Use lum.u output.
295 */
296 lur.u.stream.memory_map_size = lum->u.stream.memory_map_size;
193183fb
MD
297 shm_fd = lum->u.stream.shm_fd;
298 wait_fd = lum->u.stream.wait_fd;
299 break;
824f40b8 300 case LTTNG_UST_METADATA:
193183fb
MD
301 case LTTNG_UST_CHANNEL:
302 lur.u.channel.memory_map_size = lum->u.channel.memory_map_size;
303 shm_fd = lum->u.channel.shm_fd;
304 wait_fd = lum->u.channel.wait_fd;
305 break;
23c8854a 306 case LTTNG_UST_TRACER_VERSION:
57773204
MD
307 lur.u.version = lum->u.version;
308 break;
b115631f 309 case LTTNG_UST_TRACEPOINT_LIST_GET:
cbef6901 310 memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
b115631f 311 break;
381c0f1e 312 }
46050b1a 313 ret = send_reply(sock, &lur);
193183fb
MD
314 if (ret < 0) {
315 perror("error sending reply");
316 goto error;
317 }
46050b1a 318
824f40b8
MD
319 if ((lum->cmd == LTTNG_UST_STREAM
320 || lum->cmd == LTTNG_UST_CHANNEL
321 || lum->cmd == LTTNG_UST_METADATA)
57773204 322 && lur.ret_code == USTCOMM_OK) {
381c0f1e 323 /* we also need to send the file descriptors. */
57773204 324 ret = ustcomm_send_fds_unix_sock(sock,
193183fb 325 &shm_fd, &shm_fd,
381c0f1e
MD
326 1, sizeof(int));
327 if (ret < 0) {
328 perror("send shm_fd");
329 goto error;
330 }
57773204 331 ret = ustcomm_send_fds_unix_sock(sock,
193183fb 332 &wait_fd, &wait_fd,
381c0f1e
MD
333 1, sizeof(int));
334 if (ret < 0) {
335 perror("send wait_fd");
336 goto error;
337 }
338 }
339error:
17dfb34b 340 ust_unlock();
1ea11eab 341 return ret;
d9e99d10
MD
342}
343
46050b1a
MD
344static
345void cleanup_sock_info(struct sock_info *sock_info)
346{
347 int ret;
348
349 if (sock_info->socket != -1) {
350 ret = close(sock_info->socket);
351 if (ret) {
7fc90dca 352 ERR("Error closing apps socket");
46050b1a
MD
353 }
354 sock_info->socket = -1;
355 }
356 if (sock_info->root_handle != -1) {
d4419b81 357 ret = lttng_ust_objd_unref(sock_info->root_handle);
46050b1a
MD
358 if (ret) {
359 ERR("Error unref root handle");
360 }
361 sock_info->root_handle = -1;
362 }
318dfea9 363 sock_info->constructor_sem_posted = 0;
7fc90dca
MD
364 if (sock_info->wait_shm_mmap) {
365 ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE));
366 if (ret) {
367 ERR("Error unmapping wait shm");
368 }
369 sock_info->wait_shm_mmap = NULL;
370 }
371}
372
58d4b2a2 373/*
33bbeb90
MD
374 * Using fork to set umask in the child process (not multi-thread safe).
375 * We deal with the shm_open vs ftruncate race (happening when the
376 * sessiond owns the shm and does not let everybody modify it, to ensure
377 * safety against shm_unlink) by simply letting the mmap fail and
378 * retrying after a few seconds.
379 * For global shm, everybody has rw access to it until the sessiond
380 * starts.
58d4b2a2 381 */
7fc90dca 382static
58d4b2a2 383int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
7fc90dca 384{
7fc90dca 385 int wait_shm_fd, ret;
58d4b2a2 386 pid_t pid;
44e073f5 387
58d4b2a2 388 /*
33bbeb90 389 * Try to open read-only.
58d4b2a2 390 */
33bbeb90 391 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
58d4b2a2
MD
392 if (wait_shm_fd >= 0) {
393 goto end;
394 } else if (wait_shm_fd < 0 && errno != ENOENT) {
395 /*
33bbeb90
MD
396 * Real-only open did not work, and it's not because the
397 * entry was not present. It's a failure that prohibits
398 * using shm.
58d4b2a2 399 */
7fc90dca 400 ERR("Error opening shm %s", sock_info->wait_shm_path);
58d4b2a2 401 goto end;
7fc90dca
MD
402 }
403 /*
58d4b2a2
MD
404 * If the open failed because the file did not exist, try
405 * creating it ourself.
7fc90dca 406 */
58d4b2a2
MD
407 pid = fork();
408 if (pid > 0) {
409 int status;
410
411 /*
412 * Parent: wait for child to return, in which case the
413 * shared memory map will have been created.
414 */
415 pid = wait(&status);
416 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
417 wait_shm_fd = -1;
418 goto end;
7fc90dca 419 }
58d4b2a2
MD
420 /*
421 * Try to open read-only again after creation.
422 */
33bbeb90 423 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
58d4b2a2
MD
424 if (wait_shm_fd < 0) {
425 /*
426 * Real-only open did not work. It's a failure
427 * that prohibits using shm.
428 */
429 ERR("Error opening shm %s", sock_info->wait_shm_path);
430 goto end;
431 }
432 goto end;
433 } else if (pid == 0) {
434 int create_mode;
435
436 /* Child */
33bbeb90 437 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
58d4b2a2 438 if (sock_info->global)
33bbeb90 439 create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
58d4b2a2
MD
440 /*
441 * We're alone in a child process, so we can modify the
442 * process-wide umask.
443 */
33bbeb90 444 umask(~create_mode);
58d4b2a2 445 /*
33bbeb90
MD
446 * Try creating shm (or get rw access).
447 * We don't do an exclusive open, because we allow other
448 * processes to create+ftruncate it concurrently.
58d4b2a2
MD
449 */
450 wait_shm_fd = shm_open(sock_info->wait_shm_path,
451 O_RDWR | O_CREAT, create_mode);
452 if (wait_shm_fd >= 0) {
453 ret = ftruncate(wait_shm_fd, mmap_size);
454 if (ret) {
455 PERROR("ftruncate");
456 exit(EXIT_FAILURE);
457 }
58d4b2a2
MD
458 exit(EXIT_SUCCESS);
459 }
33bbeb90
MD
460 /*
461 * For local shm, we need to have rw access to accept
462 * opening it: this means the local sessiond will be
463 * able to wake us up. For global shm, we open it even
464 * if rw access is not granted, because the root.root
465 * sessiond will be able to override all rights and wake
466 * us up.
467 */
468 if (!sock_info->global && errno != EACCES) {
58d4b2a2
MD
469 ERR("Error opening shm %s", sock_info->wait_shm_path);
470 exit(EXIT_FAILURE);
471 }
472 /*
33bbeb90
MD
473 * The shm exists, but we cannot open it RW. Report
474 * success.
58d4b2a2
MD
475 */
476 exit(EXIT_SUCCESS);
477 } else {
478 return -1;
7fc90dca 479 }
58d4b2a2 480end:
33bbeb90
MD
481 if (wait_shm_fd >= 0 && !sock_info->global) {
482 struct stat statbuf;
483
484 /*
485 * Ensure that our user is the owner of the shm file for
486 * local shm. If we do not own the file, it means our
487 * sessiond will not have access to wake us up (there is
488 * probably a rogue process trying to fake our
489 * sessiond). Fallback to polling method in this case.
490 */
491 ret = fstat(wait_shm_fd, &statbuf);
492 if (ret) {
493 PERROR("fstat");
494 goto error_close;
495 }
496 if (statbuf.st_uid != getuid())
497 goto error_close;
498 }
58d4b2a2 499 return wait_shm_fd;
33bbeb90
MD
500
501error_close:
502 ret = close(wait_shm_fd);
503 if (ret) {
504 PERROR("Error closing fd");
505 }
506 return -1;
58d4b2a2
MD
507}
508
509static
510char *get_map_shm(struct sock_info *sock_info)
511{
512 size_t mmap_size = sysconf(_SC_PAGE_SIZE);
513 int wait_shm_fd, ret;
514 char *wait_shm_mmap;
515
516 wait_shm_fd = get_wait_shm(sock_info, mmap_size);
517 if (wait_shm_fd < 0) {
518 goto error;
44e073f5 519 }
7fc90dca
MD
520 wait_shm_mmap = mmap(NULL, mmap_size, PROT_READ,
521 MAP_SHARED, wait_shm_fd, 0);
7fc90dca
MD
522 /* close shm fd immediately after taking the mmap reference */
523 ret = close(wait_shm_fd);
524 if (ret) {
33bbeb90
MD
525 PERROR("Error closing fd");
526 }
527 if (wait_shm_mmap == MAP_FAILED) {
528 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
529 goto error;
7fc90dca
MD
530 }
531 return wait_shm_mmap;
532
533error:
534 return NULL;
535}
536
537static
538void wait_for_sessiond(struct sock_info *sock_info)
539{
80e2814b
MD
540 int ret;
541
7fc90dca
MD
542 ust_lock();
543 if (lttng_ust_comm_should_quit) {
544 goto quit;
545 }
37ed587a
MD
546 if (wait_poll_fallback) {
547 goto error;
548 }
7fc90dca
MD
549 if (!sock_info->wait_shm_mmap) {
550 sock_info->wait_shm_mmap = get_map_shm(sock_info);
551 if (!sock_info->wait_shm_mmap)
552 goto error;
553 }
554 ust_unlock();
555
556 DBG("Waiting for %s apps sessiond", sock_info->name);
80e2814b
MD
557 /* Wait for futex wakeup */
558 if (uatomic_read((int32_t *) sock_info->wait_shm_mmap) == 0) {
559 ret = futex_async((int32_t *) sock_info->wait_shm_mmap,
560 FUTEX_WAIT, 0, NULL, NULL, 0);
80e2814b 561 if (ret < 0) {
37ed587a
MD
562 if (errno == EFAULT) {
563 wait_poll_fallback = 1;
a8b870ad 564 DBG(
37ed587a
MD
565"Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
566"do not support FUTEX_WAKE on read-only memory mappings correctly. "
567"Please upgrade your kernel "
568"(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
569"mainline). LTTng-UST will use polling mode fallback.");
1cb5dd98 570 PERROR("futex");
37ed587a 571 }
80e2814b
MD
572 }
573 }
7fc90dca
MD
574 return;
575
576quit:
577 ust_unlock();
578 return;
579
580error:
581 ust_unlock();
7fc90dca 582 return;
46050b1a
MD
583}
584
1ea11eab
MD
585/*
586 * This thread does not allocate any resource, except within
587 * handle_message, within mutex protection. This mutex protects against
588 * fork and exit.
589 * The other moment it allocates resources is at socket connexion, which
590 * is also protected by the mutex.
591 */
d9e99d10
MD
592static
593void *ust_listener_thread(void *arg)
594{
1ea11eab 595 struct sock_info *sock_info = arg;
c0eedf81 596 int sock, ret, prev_connect_failed = 0, has_waited = 0;
d9e99d10 597
9eb62b9c
MD
598 /* Restart trying to connect to the session daemon */
599restart:
c0eedf81
MD
600 if (prev_connect_failed) {
601 /* Wait for sessiond availability with pipe */
602 wait_for_sessiond(sock_info);
603 if (has_waited) {
604 has_waited = 0;
605 /*
606 * Sleep for 5 seconds before retrying after a
607 * sequence of failure / wait / failure. This
608 * deals with a killed or broken session daemon.
609 */
610 sleep(5);
611 }
612 has_waited = 1;
613 prev_connect_failed = 0;
614 }
17dfb34b 615 ust_lock();
1ea11eab
MD
616
617 if (lttng_ust_comm_should_quit) {
17dfb34b 618 ust_unlock();
1ea11eab
MD
619 goto quit;
620 }
9eb62b9c 621
1ea11eab
MD
622 if (sock_info->socket != -1) {
623 ret = close(sock_info->socket);
624 if (ret) {
11ff9c7d 625 ERR("Error closing %s apps socket", sock_info->name);
1ea11eab
MD
626 }
627 sock_info->socket = -1;
628 }
46050b1a 629
9eb62b9c 630 /* Register */
57773204 631 ret = ustcomm_connect_unix_sock(sock_info->sock_path);
9eb62b9c 632 if (ret < 0) {
4d3c9523 633 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
c0eedf81 634 prev_connect_failed = 1;
11ff9c7d
MD
635 /*
636 * If we cannot find the sessiond daemon, don't delay
637 * constructor execution.
638 */
edaa1431 639 ret = handle_register_done(sock_info);
11ff9c7d 640 assert(!ret);
17dfb34b 641 ust_unlock();
1ea11eab 642 goto restart;
46050b1a
MD
643 }
644
645 sock_info->socket = sock = ret;
646
647 /*
648 * Create only one root handle per listener thread for the whole
649 * process lifetime.
650 */
651 if (sock_info->root_handle == -1) {
652 ret = lttng_abi_create_root_handle();
a51070bb 653 if (ret < 0) {
46050b1a 654 ERR("Error creating root handle");
17dfb34b 655 ust_unlock();
46050b1a
MD
656 goto quit;
657 }
658 sock_info->root_handle = ret;
9eb62b9c 659 }
1ea11eab 660
9eb62b9c
MD
661 ret = register_app_to_sessiond(sock);
662 if (ret < 0) {
11ff9c7d 663 ERR("Error registering to %s apps socket", sock_info->name);
c0eedf81 664 prev_connect_failed = 1;
11ff9c7d
MD
665 /*
666 * If we cannot register to the sessiond daemon, don't
667 * delay constructor execution.
668 */
edaa1431 669 ret = handle_register_done(sock_info);
11ff9c7d 670 assert(!ret);
17dfb34b 671 ust_unlock();
9eb62b9c
MD
672 goto restart;
673 }
17dfb34b 674 ust_unlock();
46050b1a 675
d9e99d10
MD
676 for (;;) {
677 ssize_t len;
57773204 678 struct ustcomm_ust_msg lum;
d9e99d10 679
57773204 680 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
d9e99d10
MD
681 switch (len) {
682 case 0: /* orderly shutdown */
11ff9c7d 683 DBG("%s ltt-sessiond has performed an orderly shutdown\n", sock_info->name);
8236ba10
MD
684 ust_lock();
685 /*
686 * Either sessiond has shutdown or refused us by closing the socket.
687 * In either case, we don't want to delay construction execution,
688 * and we need to wait before retry.
689 */
690 prev_connect_failed = 1;
691 /*
692 * If we cannot register to the sessiond daemon, don't
693 * delay constructor execution.
694 */
695 ret = handle_register_done(sock_info);
696 assert(!ret);
697 ust_unlock();
d9e99d10 698 goto end;
e7723462 699 case sizeof(lum):
d9e99d10 700 DBG("message received\n");
11ff9c7d 701 ret = handle_message(sock_info, sock, &lum);
2a80c9d8 702 if (ret < 0) {
11ff9c7d 703 ERR("Error handling message for %s socket", sock_info->name);
d9e99d10
MD
704 }
705 continue;
706 case -1:
8236ba10 707 DBG("Receive failed from lttng-sessiond with errno %d", errno);
d9e99d10 708 if (errno == ECONNRESET) {
11ff9c7d 709 ERR("%s remote end closed connection\n", sock_info->name);
d9e99d10
MD
710 goto end;
711 }
712 goto end;
713 default:
11ff9c7d 714 ERR("incorrect message size (%s socket): %zd\n", sock_info->name, len);
d9e99d10
MD
715 continue;
716 }
717
718 }
719end:
9eb62b9c 720 goto restart; /* try to reconnect */
1ea11eab 721quit:
d9e99d10
MD
722 return NULL;
723}
724
cf12a773
MD
725/*
726 * Return values: -1: don't wait. 0: wait forever. 1: timeout wait.
727 */
11ff9c7d
MD
728static
729int get_timeout(struct timespec *constructor_timeout)
730{
cf12a773
MD
731 long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
732 char *str_delay;
11ff9c7d
MD
733 int ret;
734
69400ac4 735 str_delay = getenv("LTTNG_UST_REGISTER_TIMEOUT");
cf12a773
MD
736 if (str_delay) {
737 constructor_delay_ms = strtol(str_delay, NULL, 10);
738 }
739
740 switch (constructor_delay_ms) {
741 case -1:/* fall-through */
742 case 0:
743 return constructor_delay_ms;
744 default:
745 break;
746 }
747
748 /*
749 * If we are unable to find the current time, don't wait.
750 */
751 ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
752 if (ret) {
753 return -1;
754 }
95259bd0
MD
755 constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
756 constructor_timeout->tv_nsec +=
757 (constructor_delay_ms % 1000UL) * 1000000UL;
11ff9c7d
MD
758 if (constructor_timeout->tv_nsec >= 1000000000UL) {
759 constructor_timeout->tv_sec++;
760 constructor_timeout->tv_nsec -= 1000000000UL;
761 }
cf12a773 762 return 1;
11ff9c7d 763}
d9e99d10 764
2691221a
MD
765/*
766 * sessiond monitoring thread: monitor presence of global and per-user
767 * sessiond by polling the application common named pipe.
768 */
769/* TODO */
770
edaa1431 771void __attribute__((constructor)) lttng_ust_init(void)
2691221a 772{
11ff9c7d 773 struct timespec constructor_timeout;
cf12a773 774 int timeout_mode;
2691221a
MD
775 int ret;
776
edaa1431
MD
777 if (uatomic_xchg(&initialized, 1) == 1)
778 return;
779
780 /*
781 * We want precise control over the order in which we construct
782 * our sub-libraries vs starting to receive commands from
783 * sessiond (otherwise leading to errors when trying to create
784 * sessiond before the init functions are completed).
785 */
2691221a 786 init_usterr();
edaa1431
MD
787 init_tracepoint();
788 ltt_ring_buffer_metadata_client_init();
789 ltt_ring_buffer_client_overwrite_init();
790 ltt_ring_buffer_client_discard_init();
2691221a 791
cf12a773 792 timeout_mode = get_timeout(&constructor_timeout);
11ff9c7d 793
95259bd0 794 ret = sem_init(&constructor_wait, 0, 0);
11ff9c7d
MD
795 assert(!ret);
796
8d20bf54 797 ret = setup_local_apps();
2691221a 798 if (ret) {
8d20bf54 799 ERR("Error setting up to local apps");
2691221a 800 }
1ea11eab
MD
801 ret = pthread_create(&local_apps.ust_listener, NULL,
802 ust_listener_thread, &local_apps);
11ff9c7d 803
8d20bf54
MD
804 if (local_apps.allowed) {
805 ret = pthread_create(&global_apps.ust_listener, NULL,
806 ust_listener_thread, &global_apps);
807 } else {
808 handle_register_done(&local_apps);
809 }
810
cf12a773
MD
811 switch (timeout_mode) {
812 case 1: /* timeout wait */
95259bd0
MD
813 do {
814 ret = sem_timedwait(&constructor_wait,
815 &constructor_timeout);
816 } while (ret < 0 && errno == EINTR);
cf12a773
MD
817 if (ret < 0 && errno == ETIMEDOUT) {
818 ERR("Timed out waiting for ltt-sessiond");
819 } else {
820 assert(!ret);
821 }
822 break;
7b766b16 823 case -1:/* wait forever */
95259bd0
MD
824 do {
825 ret = sem_wait(&constructor_wait);
826 } while (ret < 0 && errno == EINTR);
11ff9c7d 827 assert(!ret);
cf12a773 828 break;
7b766b16 829 case 0: /* no timeout */
cf12a773 830 break;
11ff9c7d 831 }
2691221a
MD
832}
833
17dfb34b
MD
834static
835void lttng_ust_cleanup(int exiting)
836{
837 cleanup_sock_info(&global_apps);
838 if (local_apps.allowed) {
839 cleanup_sock_info(&local_apps);
840 }
841 lttng_ust_abi_exit();
003fedf4 842 lttng_ust_events_exit();
17dfb34b
MD
843 ltt_ring_buffer_client_discard_exit();
844 ltt_ring_buffer_client_overwrite_exit();
845 ltt_ring_buffer_metadata_client_exit();
846 exit_tracepoint();
847 if (!exiting) {
848 /* Reinitialize values for fork */
849 sem_count = 2;
850 lttng_ust_comm_should_quit = 0;
851 initialized = 0;
852 }
853}
854
edaa1431 855void __attribute__((destructor)) lttng_ust_exit(void)
2691221a
MD
856{
857 int ret;
858
9eb62b9c
MD
859 /*
860 * Using pthread_cancel here because:
861 * A) we don't want to hang application teardown.
862 * B) the thread is not allocating any resource.
863 */
1ea11eab
MD
864
865 /*
866 * Require the communication thread to quit. Synchronize with
867 * mutexes to ensure it is not in a mutex critical section when
868 * pthread_cancel is later called.
869 */
17dfb34b 870 ust_lock();
1ea11eab 871 lttng_ust_comm_should_quit = 1;
17dfb34b 872 ust_unlock();
1ea11eab 873
1ea11eab 874 ret = pthread_cancel(global_apps.ust_listener);
9eb62b9c
MD
875 if (ret) {
876 ERR("Error cancelling global ust listener thread");
2691221a 877 }
8d20bf54
MD
878 if (local_apps.allowed) {
879 ret = pthread_cancel(local_apps.ust_listener);
880 if (ret) {
881 ERR("Error cancelling local ust listener thread");
882 }
8d20bf54 883 }
17dfb34b 884 lttng_ust_cleanup(1);
2691221a 885}
e822f505
MD
886
887/*
888 * We exclude the worker threads across fork and clone (except
889 * CLONE_VM), because these system calls only keep the forking thread
890 * running in the child. Therefore, we don't want to call fork or clone
891 * in the middle of an tracepoint or ust tracing state modification.
892 * Holding this mutex protects these structures across fork and clone.
893 */
b728d87e 894void ust_before_fork(sigset_t *save_sigset)
e822f505
MD
895{
896 /*
897 * Disable signals. This is to avoid that the child intervenes
898 * before it is properly setup for tracing. It is safer to
899 * disable all signals, because then we know we are not breaking
900 * anything by restoring the original mask.
901 */
902 sigset_t all_sigs;
903 int ret;
904
905 /* Disable signals */
906 sigfillset(&all_sigs);
b728d87e 907 ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
e822f505
MD
908 if (ret == -1) {
909 PERROR("sigprocmask");
910 }
17dfb34b 911 ust_lock();
e822f505
MD
912 rcu_bp_before_fork();
913}
914
b728d87e 915static void ust_after_fork_common(sigset_t *restore_sigset)
e822f505
MD
916{
917 int ret;
918
17dfb34b
MD
919 DBG("process %d", getpid());
920 ust_unlock();
e822f505 921 /* Restore signals */
23c8854a 922 ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
e822f505
MD
923 if (ret == -1) {
924 PERROR("sigprocmask");
925 }
926}
927
b728d87e 928void ust_after_fork_parent(sigset_t *restore_sigset)
e822f505 929{
17dfb34b 930 DBG("process %d", getpid());
e822f505
MD
931 rcu_bp_after_fork_parent();
932 /* Release mutexes and reenable signals */
b728d87e 933 ust_after_fork_common(restore_sigset);
e822f505
MD
934}
935
17dfb34b
MD
936/*
937 * After fork, in the child, we need to cleanup all the leftover state,
938 * except the worker thread which already magically disappeared thanks
939 * to the weird Linux fork semantics. After tyding up, we call
940 * lttng_ust_init() again to start over as a new PID.
941 *
942 * This is meant for forks() that have tracing in the child between the
943 * fork and following exec call (if there is any).
944 */
b728d87e 945void ust_after_fork_child(sigset_t *restore_sigset)
e822f505 946{
17dfb34b 947 DBG("process %d", getpid());
e822f505
MD
948 /* Release urcu mutexes */
949 rcu_bp_after_fork_child();
17dfb34b 950 lttng_ust_cleanup(0);
a93bfc45 951 lttng_context_vtid_reset();
e822f505 952 /* Release mutexes and reenable signals */
b728d87e 953 ust_after_fork_common(restore_sigset);
318dfea9 954 lttng_ust_init();
e822f505 955}
This page took 0.076309 seconds and 4 git commands to generate.