Commit | Line | Data |
---|---|---|
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 | ||
22 | #include <sys/types.h> | |
23 | #include <sys/socket.h> | |
2629549e | 24 | #include <sys/prctl.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> |
1ea11eab | 38 | |
edaa1431 MD |
39 | #include <lttng-ust-comm.h> |
40 | #include <ust/usterr-signal-safe.h> | |
41 | #include <ust/lttng-ust-abi.h> | |
42 | #include <ust/tracepoint.h> | |
d0a1ae63 | 43 | #include <ust/tracepoint-internal.h> |
e822f505 | 44 | #include <ust/ust.h> |
b751f722 | 45 | #include "ltt-tracer-core.h" |
edaa1431 MD |
46 | |
47 | /* | |
48 | * Has lttng ust comm constructor been called ? | |
49 | */ | |
50 | static int initialized; | |
51 | ||
1ea11eab | 52 | /* |
17dfb34b MD |
53 | * The ust_lock/ust_unlock lock is used as a communication thread mutex. |
54 | * Held when handling a command, also held by fork() to deal with | |
55 | * removal of threads, and by exit path. | |
1ea11eab | 56 | */ |
1ea11eab MD |
57 | |
58 | /* Should the ust comm thread quit ? */ | |
59 | static int lttng_ust_comm_should_quit; | |
60 | ||
11ff9c7d MD |
61 | /* |
62 | * Wait for either of these before continuing to the main | |
63 | * program: | |
64 | * - the register_done message from sessiond daemon | |
65 | * (will let the sessiond daemon enable sessions before main | |
66 | * starts.) | |
67 | * - sessiond daemon is not reachable. | |
68 | * - timeout (ensuring applications are resilient to session | |
69 | * daemon problems). | |
70 | */ | |
71 | static sem_t constructor_wait; | |
950aab0c MD |
72 | /* |
73 | * Doing this for both the global and local sessiond. | |
74 | */ | |
95259bd0 | 75 | static int sem_count = { 2 }; |
11ff9c7d | 76 | |
1ea11eab MD |
77 | /* |
78 | * Info about socket and associated listener thread. | |
79 | */ | |
80 | struct sock_info { | |
11ff9c7d | 81 | const char *name; |
1ea11eab | 82 | pthread_t ust_listener; /* listener thread */ |
46050b1a | 83 | int root_handle; |
8d20bf54 MD |
84 | int constructor_sem_posted; |
85 | int allowed; | |
44e073f5 | 86 | int global; |
7fc90dca MD |
87 | |
88 | char sock_path[PATH_MAX]; | |
89 | int socket; | |
90 | ||
91 | char wait_shm_path[PATH_MAX]; | |
92 | char *wait_shm_mmap; | |
1ea11eab | 93 | }; |
2691221a MD |
94 | |
95 | /* Socket from app (connect) to session daemon (listen) for communication */ | |
1ea11eab | 96 | struct sock_info global_apps = { |
11ff9c7d | 97 | .name = "global", |
44e073f5 | 98 | .global = 1, |
7fc90dca | 99 | |
46050b1a | 100 | .root_handle = -1, |
8d20bf54 | 101 | .allowed = 1, |
7fc90dca MD |
102 | |
103 | .sock_path = DEFAULT_GLOBAL_APPS_UNIX_SOCK, | |
104 | .socket = -1, | |
105 | ||
106 | .wait_shm_path = DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH, | |
1ea11eab | 107 | }; |
2691221a MD |
108 | |
109 | /* TODO: allow global_apps_sock_path override */ | |
110 | ||
1ea11eab | 111 | struct sock_info local_apps = { |
11ff9c7d | 112 | .name = "local", |
44e073f5 | 113 | .global = 0, |
46050b1a | 114 | .root_handle = -1, |
8d20bf54 | 115 | .allowed = 0, /* Check setuid bit first */ |
7fc90dca MD |
116 | |
117 | .socket = -1, | |
1ea11eab | 118 | }; |
2691221a | 119 | |
edaa1431 MD |
120 | extern void ltt_ring_buffer_client_overwrite_init(void); |
121 | extern void ltt_ring_buffer_client_discard_init(void); | |
122 | extern void ltt_ring_buffer_metadata_client_init(void); | |
123 | extern void ltt_ring_buffer_client_overwrite_exit(void); | |
124 | extern void ltt_ring_buffer_client_discard_exit(void); | |
125 | extern void ltt_ring_buffer_metadata_client_exit(void); | |
126 | ||
2691221a | 127 | static |
8d20bf54 | 128 | int setup_local_apps(void) |
2691221a MD |
129 | { |
130 | const char *home_dir; | |
7fc90dca | 131 | uid_t uid; |
2691221a | 132 | |
7fc90dca | 133 | uid = getuid(); |
8d20bf54 MD |
134 | /* |
135 | * Disallow per-user tracing for setuid binaries. | |
136 | */ | |
7fc90dca | 137 | if (uid != geteuid()) { |
8d20bf54 | 138 | local_apps.allowed = 0; |
d0a1ae63 | 139 | return 0; |
8d20bf54 MD |
140 | } else { |
141 | local_apps.allowed = 1; | |
142 | } | |
2691221a MD |
143 | home_dir = (const char *) getenv("HOME"); |
144 | if (!home_dir) | |
145 | return -ENOENT; | |
1ea11eab | 146 | snprintf(local_apps.sock_path, PATH_MAX, |
2691221a | 147 | DEFAULT_HOME_APPS_UNIX_SOCK, home_dir); |
7fc90dca MD |
148 | snprintf(local_apps.wait_shm_path, PATH_MAX, |
149 | DEFAULT_HOME_APPS_WAIT_SHM_PATH, uid); | |
2691221a MD |
150 | return 0; |
151 | } | |
152 | ||
153 | static | |
154 | int register_app_to_sessiond(int socket) | |
155 | { | |
156 | ssize_t ret; | |
2629549e | 157 | int prctl_ret; |
2691221a | 158 | struct { |
e44418f3 MD |
159 | uint32_t major; |
160 | uint32_t minor; | |
2691221a | 161 | pid_t pid; |
5c33bde8 | 162 | pid_t ppid; |
2691221a | 163 | uid_t uid; |
83610856 | 164 | gid_t gid; |
2629549e | 165 | char name[16]; /* process name */ |
2691221a MD |
166 | } reg_msg; |
167 | ||
e44418f3 MD |
168 | reg_msg.major = LTTNG_UST_COMM_VERSION_MAJOR; |
169 | reg_msg.minor = LTTNG_UST_COMM_VERSION_MINOR; | |
2691221a | 170 | reg_msg.pid = getpid(); |
5c33bde8 | 171 | reg_msg.ppid = getppid(); |
2691221a | 172 | reg_msg.uid = getuid(); |
83610856 | 173 | reg_msg.gid = getgid(); |
2629549e MD |
174 | prctl_ret = prctl(PR_GET_NAME, (unsigned long) reg_msg.name, 0, 0, 0); |
175 | if (prctl_ret) { | |
176 | ERR("Error executing prctl"); | |
177 | return -errno; | |
178 | } | |
2691221a MD |
179 | |
180 | ret = lttcomm_send_unix_sock(socket, ®_msg, sizeof(reg_msg)); | |
181 | if (ret >= 0 && ret != sizeof(reg_msg)) | |
182 | return -EIO; | |
183 | return ret; | |
184 | } | |
185 | ||
d9e99d10 | 186 | static |
d3a492d1 | 187 | int send_reply(int sock, struct lttcomm_ust_reply *lur) |
d9e99d10 | 188 | { |
9eb62b9c | 189 | ssize_t len; |
d3a492d1 | 190 | |
a4be8962 | 191 | len = lttcomm_send_unix_sock(sock, lur, sizeof(*lur)); |
d3a492d1 | 192 | switch (len) { |
a4be8962 | 193 | case sizeof(*lur): |
d3a492d1 MD |
194 | DBG("message successfully sent"); |
195 | return 0; | |
196 | case -1: | |
197 | if (errno == ECONNRESET) { | |
198 | printf("remote end closed connection\n"); | |
199 | return 0; | |
200 | } | |
201 | return -1; | |
202 | default: | |
203 | printf("incorrect message size: %zd\n", len); | |
204 | return -1; | |
205 | } | |
206 | } | |
207 | ||
208 | static | |
edaa1431 | 209 | int handle_register_done(struct sock_info *sock_info) |
11ff9c7d MD |
210 | { |
211 | int ret; | |
212 | ||
edaa1431 MD |
213 | if (sock_info->constructor_sem_posted) |
214 | return 0; | |
215 | sock_info->constructor_sem_posted = 1; | |
95259bd0 MD |
216 | ret = uatomic_add_return(&sem_count, -1); |
217 | if (ret == 0) { | |
218 | ret = sem_post(&constructor_wait); | |
219 | assert(!ret); | |
220 | } | |
11ff9c7d MD |
221 | return 0; |
222 | } | |
223 | ||
224 | static | |
225 | int handle_message(struct sock_info *sock_info, | |
226 | int sock, struct lttcomm_ust_msg *lum) | |
d3a492d1 | 227 | { |
1ea11eab | 228 | int ret = 0; |
46050b1a MD |
229 | const struct objd_ops *ops; |
230 | struct lttcomm_ust_reply lur; | |
1ea11eab | 231 | |
17dfb34b | 232 | ust_lock(); |
1ea11eab | 233 | |
46050b1a MD |
234 | memset(&lur, 0, sizeof(lur)); |
235 | ||
1ea11eab | 236 | if (lttng_ust_comm_should_quit) { |
46050b1a | 237 | ret = -EPERM; |
1ea11eab MD |
238 | goto end; |
239 | } | |
9eb62b9c | 240 | |
46050b1a MD |
241 | ops = objd_ops(lum->handle); |
242 | if (!ops) { | |
243 | ret = -ENOENT; | |
244 | goto end; | |
1ea11eab | 245 | } |
46050b1a MD |
246 | |
247 | switch (lum->cmd) { | |
11ff9c7d MD |
248 | case LTTNG_UST_REGISTER_DONE: |
249 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
edaa1431 | 250 | ret = handle_register_done(sock_info); |
11ff9c7d MD |
251 | else |
252 | ret = -EINVAL; | |
253 | break; | |
46050b1a MD |
254 | case LTTNG_UST_RELEASE: |
255 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
256 | ret = -EPERM; | |
257 | else | |
258 | ret = objd_unref(lum->handle); | |
d9e99d10 MD |
259 | break; |
260 | default: | |
46050b1a MD |
261 | if (ops->cmd) |
262 | ret = ops->cmd(lum->handle, lum->cmd, | |
263 | (unsigned long) &lum->u); | |
264 | else | |
265 | ret = -ENOSYS; | |
266 | break; | |
d9e99d10 | 267 | } |
46050b1a | 268 | |
1ea11eab | 269 | end: |
46050b1a MD |
270 | lur.handle = lum->handle; |
271 | lur.cmd = lum->cmd; | |
272 | lur.ret_val = ret; | |
273 | if (ret >= 0) { | |
274 | lur.ret_code = LTTCOMM_OK; | |
275 | } else { | |
276 | lur.ret_code = LTTCOMM_SESSION_FAIL; | |
277 | } | |
278 | ret = send_reply(sock, &lur); | |
279 | ||
17dfb34b | 280 | ust_unlock(); |
1ea11eab | 281 | return ret; |
d9e99d10 MD |
282 | } |
283 | ||
46050b1a MD |
284 | static |
285 | void cleanup_sock_info(struct sock_info *sock_info) | |
286 | { | |
287 | int ret; | |
288 | ||
289 | if (sock_info->socket != -1) { | |
290 | ret = close(sock_info->socket); | |
291 | if (ret) { | |
7fc90dca | 292 | ERR("Error closing apps socket"); |
46050b1a MD |
293 | } |
294 | sock_info->socket = -1; | |
295 | } | |
296 | if (sock_info->root_handle != -1) { | |
297 | ret = objd_unref(sock_info->root_handle); | |
298 | if (ret) { | |
299 | ERR("Error unref root handle"); | |
300 | } | |
301 | sock_info->root_handle = -1; | |
302 | } | |
318dfea9 | 303 | sock_info->constructor_sem_posted = 0; |
7fc90dca MD |
304 | if (sock_info->wait_shm_mmap) { |
305 | ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE)); | |
306 | if (ret) { | |
307 | ERR("Error unmapping wait shm"); | |
308 | } | |
309 | sock_info->wait_shm_mmap = NULL; | |
310 | } | |
311 | } | |
312 | ||
58d4b2a2 | 313 | /* |
33bbeb90 MD |
314 | * Using fork to set umask in the child process (not multi-thread safe). |
315 | * We deal with the shm_open vs ftruncate race (happening when the | |
316 | * sessiond owns the shm and does not let everybody modify it, to ensure | |
317 | * safety against shm_unlink) by simply letting the mmap fail and | |
318 | * retrying after a few seconds. | |
319 | * For global shm, everybody has rw access to it until the sessiond | |
320 | * starts. | |
58d4b2a2 | 321 | */ |
7fc90dca | 322 | static |
58d4b2a2 | 323 | int get_wait_shm(struct sock_info *sock_info, size_t mmap_size) |
7fc90dca | 324 | { |
7fc90dca | 325 | int wait_shm_fd, ret; |
58d4b2a2 | 326 | pid_t pid; |
44e073f5 | 327 | |
58d4b2a2 | 328 | /* |
33bbeb90 | 329 | * Try to open read-only. |
58d4b2a2 | 330 | */ |
33bbeb90 | 331 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 MD |
332 | if (wait_shm_fd >= 0) { |
333 | goto end; | |
334 | } else if (wait_shm_fd < 0 && errno != ENOENT) { | |
335 | /* | |
33bbeb90 MD |
336 | * Real-only open did not work, and it's not because the |
337 | * entry was not present. It's a failure that prohibits | |
338 | * using shm. | |
58d4b2a2 | 339 | */ |
7fc90dca | 340 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
58d4b2a2 | 341 | goto end; |
7fc90dca MD |
342 | } |
343 | /* | |
58d4b2a2 MD |
344 | * If the open failed because the file did not exist, try |
345 | * creating it ourself. | |
7fc90dca | 346 | */ |
58d4b2a2 MD |
347 | pid = fork(); |
348 | if (pid > 0) { | |
349 | int status; | |
350 | ||
351 | /* | |
352 | * Parent: wait for child to return, in which case the | |
353 | * shared memory map will have been created. | |
354 | */ | |
355 | pid = wait(&status); | |
356 | if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) { | |
357 | wait_shm_fd = -1; | |
358 | goto end; | |
7fc90dca | 359 | } |
58d4b2a2 MD |
360 | /* |
361 | * Try to open read-only again after creation. | |
362 | */ | |
33bbeb90 | 363 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 MD |
364 | if (wait_shm_fd < 0) { |
365 | /* | |
366 | * Real-only open did not work. It's a failure | |
367 | * that prohibits using shm. | |
368 | */ | |
369 | ERR("Error opening shm %s", sock_info->wait_shm_path); | |
370 | goto end; | |
371 | } | |
372 | goto end; | |
373 | } else if (pid == 0) { | |
374 | int create_mode; | |
375 | ||
376 | /* Child */ | |
33bbeb90 | 377 | create_mode = S_IRUSR | S_IWUSR | S_IRGRP; |
58d4b2a2 | 378 | if (sock_info->global) |
33bbeb90 | 379 | create_mode |= S_IROTH | S_IWGRP | S_IWOTH; |
58d4b2a2 MD |
380 | /* |
381 | * We're alone in a child process, so we can modify the | |
382 | * process-wide umask. | |
383 | */ | |
33bbeb90 | 384 | umask(~create_mode); |
58d4b2a2 | 385 | /* |
33bbeb90 MD |
386 | * Try creating shm (or get rw access). |
387 | * We don't do an exclusive open, because we allow other | |
388 | * processes to create+ftruncate it concurrently. | |
58d4b2a2 MD |
389 | */ |
390 | wait_shm_fd = shm_open(sock_info->wait_shm_path, | |
391 | O_RDWR | O_CREAT, create_mode); | |
392 | if (wait_shm_fd >= 0) { | |
393 | ret = ftruncate(wait_shm_fd, mmap_size); | |
394 | if (ret) { | |
395 | PERROR("ftruncate"); | |
396 | exit(EXIT_FAILURE); | |
397 | } | |
58d4b2a2 MD |
398 | exit(EXIT_SUCCESS); |
399 | } | |
33bbeb90 MD |
400 | /* |
401 | * For local shm, we need to have rw access to accept | |
402 | * opening it: this means the local sessiond will be | |
403 | * able to wake us up. For global shm, we open it even | |
404 | * if rw access is not granted, because the root.root | |
405 | * sessiond will be able to override all rights and wake | |
406 | * us up. | |
407 | */ | |
408 | if (!sock_info->global && errno != EACCES) { | |
58d4b2a2 MD |
409 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
410 | exit(EXIT_FAILURE); | |
411 | } | |
412 | /* | |
33bbeb90 MD |
413 | * The shm exists, but we cannot open it RW. Report |
414 | * success. | |
58d4b2a2 MD |
415 | */ |
416 | exit(EXIT_SUCCESS); | |
417 | } else { | |
418 | return -1; | |
7fc90dca | 419 | } |
58d4b2a2 | 420 | end: |
33bbeb90 MD |
421 | if (wait_shm_fd >= 0 && !sock_info->global) { |
422 | struct stat statbuf; | |
423 | ||
424 | /* | |
425 | * Ensure that our user is the owner of the shm file for | |
426 | * local shm. If we do not own the file, it means our | |
427 | * sessiond will not have access to wake us up (there is | |
428 | * probably a rogue process trying to fake our | |
429 | * sessiond). Fallback to polling method in this case. | |
430 | */ | |
431 | ret = fstat(wait_shm_fd, &statbuf); | |
432 | if (ret) { | |
433 | PERROR("fstat"); | |
434 | goto error_close; | |
435 | } | |
436 | if (statbuf.st_uid != getuid()) | |
437 | goto error_close; | |
438 | } | |
58d4b2a2 | 439 | return wait_shm_fd; |
33bbeb90 MD |
440 | |
441 | error_close: | |
442 | ret = close(wait_shm_fd); | |
443 | if (ret) { | |
444 | PERROR("Error closing fd"); | |
445 | } | |
446 | return -1; | |
58d4b2a2 MD |
447 | } |
448 | ||
449 | static | |
450 | char *get_map_shm(struct sock_info *sock_info) | |
451 | { | |
452 | size_t mmap_size = sysconf(_SC_PAGE_SIZE); | |
453 | int wait_shm_fd, ret; | |
454 | char *wait_shm_mmap; | |
455 | ||
456 | wait_shm_fd = get_wait_shm(sock_info, mmap_size); | |
457 | if (wait_shm_fd < 0) { | |
458 | goto error; | |
44e073f5 | 459 | } |
7fc90dca MD |
460 | wait_shm_mmap = mmap(NULL, mmap_size, PROT_READ, |
461 | MAP_SHARED, wait_shm_fd, 0); | |
7fc90dca MD |
462 | /* close shm fd immediately after taking the mmap reference */ |
463 | ret = close(wait_shm_fd); | |
464 | if (ret) { | |
33bbeb90 MD |
465 | PERROR("Error closing fd"); |
466 | } | |
467 | if (wait_shm_mmap == MAP_FAILED) { | |
468 | DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode."); | |
469 | goto error; | |
7fc90dca MD |
470 | } |
471 | return wait_shm_mmap; | |
472 | ||
473 | error: | |
474 | return NULL; | |
475 | } | |
476 | ||
477 | static | |
478 | void wait_for_sessiond(struct sock_info *sock_info) | |
479 | { | |
480 | ust_lock(); | |
481 | if (lttng_ust_comm_should_quit) { | |
482 | goto quit; | |
483 | } | |
484 | if (!sock_info->wait_shm_mmap) { | |
485 | sock_info->wait_shm_mmap = get_map_shm(sock_info); | |
486 | if (!sock_info->wait_shm_mmap) | |
487 | goto error; | |
488 | } | |
489 | ust_unlock(); | |
490 | ||
491 | DBG("Waiting for %s apps sessiond", sock_info->name); | |
492 | /* Wait for futex wakeup TODO */ | |
493 | sleep(5); | |
494 | ||
495 | return; | |
496 | ||
497 | quit: | |
498 | ust_unlock(); | |
499 | return; | |
500 | ||
501 | error: | |
502 | ust_unlock(); | |
503 | /* Error handling: fallback on a 5 seconds sleep. */ | |
504 | sleep(5); | |
505 | return; | |
46050b1a MD |
506 | } |
507 | ||
1ea11eab MD |
508 | /* |
509 | * This thread does not allocate any resource, except within | |
510 | * handle_message, within mutex protection. This mutex protects against | |
511 | * fork and exit. | |
512 | * The other moment it allocates resources is at socket connexion, which | |
513 | * is also protected by the mutex. | |
514 | */ | |
d9e99d10 MD |
515 | static |
516 | void *ust_listener_thread(void *arg) | |
517 | { | |
1ea11eab MD |
518 | struct sock_info *sock_info = arg; |
519 | int sock, ret; | |
d9e99d10 | 520 | |
9eb62b9c MD |
521 | /* Restart trying to connect to the session daemon */ |
522 | restart: | |
17dfb34b | 523 | ust_lock(); |
1ea11eab MD |
524 | |
525 | if (lttng_ust_comm_should_quit) { | |
17dfb34b | 526 | ust_unlock(); |
1ea11eab MD |
527 | goto quit; |
528 | } | |
9eb62b9c | 529 | |
1ea11eab MD |
530 | if (sock_info->socket != -1) { |
531 | ret = close(sock_info->socket); | |
532 | if (ret) { | |
11ff9c7d | 533 | ERR("Error closing %s apps socket", sock_info->name); |
1ea11eab MD |
534 | } |
535 | sock_info->socket = -1; | |
536 | } | |
46050b1a | 537 | |
9eb62b9c | 538 | /* Register */ |
1ea11eab | 539 | ret = lttcomm_connect_unix_sock(sock_info->sock_path); |
9eb62b9c | 540 | if (ret < 0) { |
11ff9c7d MD |
541 | ERR("Error connecting to %s apps socket", sock_info->name); |
542 | /* | |
543 | * If we cannot find the sessiond daemon, don't delay | |
544 | * constructor execution. | |
545 | */ | |
edaa1431 | 546 | ret = handle_register_done(sock_info); |
11ff9c7d | 547 | assert(!ret); |
17dfb34b | 548 | ust_unlock(); |
7fc90dca MD |
549 | |
550 | /* Wait for sessiond availability with pipe */ | |
551 | wait_for_sessiond(sock_info); | |
1ea11eab | 552 | goto restart; |
46050b1a MD |
553 | } |
554 | ||
555 | sock_info->socket = sock = ret; | |
556 | ||
557 | /* | |
558 | * Create only one root handle per listener thread for the whole | |
559 | * process lifetime. | |
560 | */ | |
561 | if (sock_info->root_handle == -1) { | |
562 | ret = lttng_abi_create_root_handle(); | |
563 | if (ret) { | |
564 | ERR("Error creating root handle"); | |
17dfb34b | 565 | ust_unlock(); |
46050b1a MD |
566 | goto quit; |
567 | } | |
568 | sock_info->root_handle = ret; | |
9eb62b9c | 569 | } |
1ea11eab | 570 | |
9eb62b9c MD |
571 | ret = register_app_to_sessiond(sock); |
572 | if (ret < 0) { | |
11ff9c7d MD |
573 | ERR("Error registering to %s apps socket", sock_info->name); |
574 | /* | |
575 | * If we cannot register to the sessiond daemon, don't | |
576 | * delay constructor execution. | |
577 | */ | |
edaa1431 | 578 | ret = handle_register_done(sock_info); |
11ff9c7d | 579 | assert(!ret); |
17dfb34b | 580 | ust_unlock(); |
7fc90dca | 581 | wait_for_sessiond(sock_info); |
9eb62b9c MD |
582 | goto restart; |
583 | } | |
17dfb34b | 584 | ust_unlock(); |
46050b1a | 585 | |
d9e99d10 MD |
586 | for (;;) { |
587 | ssize_t len; | |
e7723462 | 588 | struct lttcomm_ust_msg lum; |
d9e99d10 | 589 | |
e7723462 | 590 | len = lttcomm_recv_unix_sock(sock, &lum, sizeof(lum)); |
d9e99d10 MD |
591 | switch (len) { |
592 | case 0: /* orderly shutdown */ | |
11ff9c7d | 593 | DBG("%s ltt-sessiond has performed an orderly shutdown\n", sock_info->name); |
d9e99d10 | 594 | goto end; |
e7723462 | 595 | case sizeof(lum): |
d9e99d10 | 596 | DBG("message received\n"); |
11ff9c7d | 597 | ret = handle_message(sock_info, sock, &lum); |
2a80c9d8 | 598 | if (ret < 0) { |
11ff9c7d | 599 | ERR("Error handling message for %s socket", sock_info->name); |
d9e99d10 MD |
600 | } |
601 | continue; | |
602 | case -1: | |
603 | if (errno == ECONNRESET) { | |
11ff9c7d | 604 | ERR("%s remote end closed connection\n", sock_info->name); |
d9e99d10 MD |
605 | goto end; |
606 | } | |
607 | goto end; | |
608 | default: | |
11ff9c7d | 609 | ERR("incorrect message size (%s socket): %zd\n", sock_info->name, len); |
d9e99d10 MD |
610 | continue; |
611 | } | |
612 | ||
613 | } | |
614 | end: | |
9eb62b9c | 615 | goto restart; /* try to reconnect */ |
1ea11eab | 616 | quit: |
d9e99d10 MD |
617 | return NULL; |
618 | } | |
619 | ||
cf12a773 MD |
620 | /* |
621 | * Return values: -1: don't wait. 0: wait forever. 1: timeout wait. | |
622 | */ | |
11ff9c7d MD |
623 | static |
624 | int get_timeout(struct timespec *constructor_timeout) | |
625 | { | |
cf12a773 MD |
626 | long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS; |
627 | char *str_delay; | |
11ff9c7d MD |
628 | int ret; |
629 | ||
cf12a773 MD |
630 | str_delay = getenv("UST_REGISTER_TIMEOUT"); |
631 | if (str_delay) { | |
632 | constructor_delay_ms = strtol(str_delay, NULL, 10); | |
633 | } | |
634 | ||
635 | switch (constructor_delay_ms) { | |
636 | case -1:/* fall-through */ | |
637 | case 0: | |
638 | return constructor_delay_ms; | |
639 | default: | |
640 | break; | |
641 | } | |
642 | ||
643 | /* | |
644 | * If we are unable to find the current time, don't wait. | |
645 | */ | |
646 | ret = clock_gettime(CLOCK_REALTIME, constructor_timeout); | |
647 | if (ret) { | |
648 | return -1; | |
649 | } | |
95259bd0 MD |
650 | constructor_timeout->tv_sec += constructor_delay_ms / 1000UL; |
651 | constructor_timeout->tv_nsec += | |
652 | (constructor_delay_ms % 1000UL) * 1000000UL; | |
11ff9c7d MD |
653 | if (constructor_timeout->tv_nsec >= 1000000000UL) { |
654 | constructor_timeout->tv_sec++; | |
655 | constructor_timeout->tv_nsec -= 1000000000UL; | |
656 | } | |
cf12a773 | 657 | return 1; |
11ff9c7d | 658 | } |
d9e99d10 | 659 | |
2691221a MD |
660 | /* |
661 | * sessiond monitoring thread: monitor presence of global and per-user | |
662 | * sessiond by polling the application common named pipe. | |
663 | */ | |
664 | /* TODO */ | |
665 | ||
edaa1431 | 666 | void __attribute__((constructor)) lttng_ust_init(void) |
2691221a | 667 | { |
11ff9c7d | 668 | struct timespec constructor_timeout; |
cf12a773 | 669 | int timeout_mode; |
2691221a MD |
670 | int ret; |
671 | ||
edaa1431 MD |
672 | if (uatomic_xchg(&initialized, 1) == 1) |
673 | return; | |
674 | ||
675 | /* | |
676 | * We want precise control over the order in which we construct | |
677 | * our sub-libraries vs starting to receive commands from | |
678 | * sessiond (otherwise leading to errors when trying to create | |
679 | * sessiond before the init functions are completed). | |
680 | */ | |
2691221a | 681 | init_usterr(); |
edaa1431 MD |
682 | init_tracepoint(); |
683 | ltt_ring_buffer_metadata_client_init(); | |
684 | ltt_ring_buffer_client_overwrite_init(); | |
685 | ltt_ring_buffer_client_discard_init(); | |
2691221a | 686 | |
cf12a773 | 687 | timeout_mode = get_timeout(&constructor_timeout); |
11ff9c7d | 688 | |
95259bd0 | 689 | ret = sem_init(&constructor_wait, 0, 0); |
11ff9c7d MD |
690 | assert(!ret); |
691 | ||
8d20bf54 | 692 | ret = setup_local_apps(); |
2691221a | 693 | if (ret) { |
8d20bf54 | 694 | ERR("Error setting up to local apps"); |
2691221a | 695 | } |
1ea11eab MD |
696 | ret = pthread_create(&local_apps.ust_listener, NULL, |
697 | ust_listener_thread, &local_apps); | |
11ff9c7d | 698 | |
8d20bf54 MD |
699 | if (local_apps.allowed) { |
700 | ret = pthread_create(&global_apps.ust_listener, NULL, | |
701 | ust_listener_thread, &global_apps); | |
702 | } else { | |
703 | handle_register_done(&local_apps); | |
704 | } | |
705 | ||
cf12a773 MD |
706 | switch (timeout_mode) { |
707 | case 1: /* timeout wait */ | |
95259bd0 MD |
708 | do { |
709 | ret = sem_timedwait(&constructor_wait, | |
710 | &constructor_timeout); | |
711 | } while (ret < 0 && errno == EINTR); | |
cf12a773 MD |
712 | if (ret < 0 && errno == ETIMEDOUT) { |
713 | ERR("Timed out waiting for ltt-sessiond"); | |
714 | } else { | |
715 | assert(!ret); | |
716 | } | |
717 | break; | |
7b766b16 | 718 | case -1:/* wait forever */ |
95259bd0 MD |
719 | do { |
720 | ret = sem_wait(&constructor_wait); | |
721 | } while (ret < 0 && errno == EINTR); | |
11ff9c7d | 722 | assert(!ret); |
cf12a773 | 723 | break; |
7b766b16 | 724 | case 0: /* no timeout */ |
cf12a773 | 725 | break; |
11ff9c7d | 726 | } |
2691221a MD |
727 | } |
728 | ||
17dfb34b MD |
729 | static |
730 | void lttng_ust_cleanup(int exiting) | |
731 | { | |
732 | cleanup_sock_info(&global_apps); | |
733 | if (local_apps.allowed) { | |
734 | cleanup_sock_info(&local_apps); | |
735 | } | |
736 | lttng_ust_abi_exit(); | |
737 | ltt_events_exit(); | |
738 | ltt_ring_buffer_client_discard_exit(); | |
739 | ltt_ring_buffer_client_overwrite_exit(); | |
740 | ltt_ring_buffer_metadata_client_exit(); | |
741 | exit_tracepoint(); | |
742 | if (!exiting) { | |
743 | /* Reinitialize values for fork */ | |
744 | sem_count = 2; | |
745 | lttng_ust_comm_should_quit = 0; | |
746 | initialized = 0; | |
747 | } | |
748 | } | |
749 | ||
edaa1431 | 750 | void __attribute__((destructor)) lttng_ust_exit(void) |
2691221a MD |
751 | { |
752 | int ret; | |
753 | ||
9eb62b9c MD |
754 | /* |
755 | * Using pthread_cancel here because: | |
756 | * A) we don't want to hang application teardown. | |
757 | * B) the thread is not allocating any resource. | |
758 | */ | |
1ea11eab MD |
759 | |
760 | /* | |
761 | * Require the communication thread to quit. Synchronize with | |
762 | * mutexes to ensure it is not in a mutex critical section when | |
763 | * pthread_cancel is later called. | |
764 | */ | |
17dfb34b | 765 | ust_lock(); |
1ea11eab | 766 | lttng_ust_comm_should_quit = 1; |
17dfb34b | 767 | ust_unlock(); |
1ea11eab | 768 | |
1ea11eab | 769 | ret = pthread_cancel(global_apps.ust_listener); |
9eb62b9c MD |
770 | if (ret) { |
771 | ERR("Error cancelling global ust listener thread"); | |
2691221a | 772 | } |
8d20bf54 MD |
773 | if (local_apps.allowed) { |
774 | ret = pthread_cancel(local_apps.ust_listener); | |
775 | if (ret) { | |
776 | ERR("Error cancelling local ust listener thread"); | |
777 | } | |
8d20bf54 | 778 | } |
17dfb34b | 779 | lttng_ust_cleanup(1); |
2691221a | 780 | } |
e822f505 MD |
781 | |
782 | /* | |
783 | * We exclude the worker threads across fork and clone (except | |
784 | * CLONE_VM), because these system calls only keep the forking thread | |
785 | * running in the child. Therefore, we don't want to call fork or clone | |
786 | * in the middle of an tracepoint or ust tracing state modification. | |
787 | * Holding this mutex protects these structures across fork and clone. | |
788 | */ | |
789 | void ust_before_fork(ust_fork_info_t *fork_info) | |
790 | { | |
791 | /* | |
792 | * Disable signals. This is to avoid that the child intervenes | |
793 | * before it is properly setup for tracing. It is safer to | |
794 | * disable all signals, because then we know we are not breaking | |
795 | * anything by restoring the original mask. | |
796 | */ | |
797 | sigset_t all_sigs; | |
798 | int ret; | |
799 | ||
800 | /* Disable signals */ | |
801 | sigfillset(&all_sigs); | |
802 | ret = sigprocmask(SIG_BLOCK, &all_sigs, &fork_info->orig_sigs); | |
803 | if (ret == -1) { | |
804 | PERROR("sigprocmask"); | |
805 | } | |
17dfb34b | 806 | ust_lock(); |
e822f505 MD |
807 | rcu_bp_before_fork(); |
808 | } | |
809 | ||
810 | static void ust_after_fork_common(ust_fork_info_t *fork_info) | |
811 | { | |
812 | int ret; | |
813 | ||
17dfb34b MD |
814 | DBG("process %d", getpid()); |
815 | ust_unlock(); | |
e822f505 MD |
816 | /* Restore signals */ |
817 | ret = sigprocmask(SIG_SETMASK, &fork_info->orig_sigs, NULL); | |
818 | if (ret == -1) { | |
819 | PERROR("sigprocmask"); | |
820 | } | |
821 | } | |
822 | ||
823 | void ust_after_fork_parent(ust_fork_info_t *fork_info) | |
824 | { | |
17dfb34b | 825 | DBG("process %d", getpid()); |
e822f505 MD |
826 | rcu_bp_after_fork_parent(); |
827 | /* Release mutexes and reenable signals */ | |
828 | ust_after_fork_common(fork_info); | |
829 | } | |
830 | ||
17dfb34b MD |
831 | /* |
832 | * After fork, in the child, we need to cleanup all the leftover state, | |
833 | * except the worker thread which already magically disappeared thanks | |
834 | * to the weird Linux fork semantics. After tyding up, we call | |
835 | * lttng_ust_init() again to start over as a new PID. | |
836 | * | |
837 | * This is meant for forks() that have tracing in the child between the | |
838 | * fork and following exec call (if there is any). | |
839 | */ | |
e822f505 MD |
840 | void ust_after_fork_child(ust_fork_info_t *fork_info) |
841 | { | |
17dfb34b | 842 | DBG("process %d", getpid()); |
e822f505 MD |
843 | /* Release urcu mutexes */ |
844 | rcu_bp_after_fork_child(); | |
17dfb34b | 845 | lttng_ust_cleanup(0); |
e822f505 MD |
846 | /* Release mutexes and reenable signals */ |
847 | ust_after_fork_common(fork_info); | |
318dfea9 | 848 | lttng_ust_init(); |
e822f505 | 849 | } |