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 | ||
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> |
1ea11eab | 40 | |
4318ae1b MD |
41 | #include <lttng/ust-comm.h> |
42 | #include <lttng/ust-events.h> | |
43 | #include <lttng/usterr-signal-safe.h> | |
44 | #include <lttng/ust-abi.h> | |
45 | #include <lttng/tracepoint.h> | |
46 | #include <lttng/tracepoint-internal.h> | |
47 | #include <lttng/ust.h> | |
b751f722 | 48 | #include "ltt-tracer-core.h" |
edaa1431 MD |
49 | |
50 | /* | |
51 | * Has lttng ust comm constructor been called ? | |
52 | */ | |
53 | static 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 ? */ | |
62 | static 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 | */ | |
74 | static sem_t constructor_wait; | |
950aab0c MD |
75 | /* |
76 | * Doing this for both the global and local sessiond. | |
77 | */ | |
95259bd0 | 78 | static int sem_count = { 2 }; |
11ff9c7d | 79 | |
1ea11eab MD |
80 | /* |
81 | * Info about socket and associated listener thread. | |
82 | */ | |
83 | struct 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 | 99 | struct 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 | 114 | struct 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 |
123 | static int wait_poll_fallback; |
124 | ||
edaa1431 MD |
125 | extern void ltt_ring_buffer_client_overwrite_init(void); |
126 | extern void ltt_ring_buffer_client_discard_init(void); | |
127 | extern void ltt_ring_buffer_metadata_client_init(void); | |
128 | extern void ltt_ring_buffer_client_overwrite_exit(void); | |
129 | extern void ltt_ring_buffer_client_discard_exit(void); | |
130 | extern void ltt_ring_buffer_metadata_client_exit(void); | |
131 | ||
2691221a | 132 | static |
8d20bf54 | 133 | int 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 | ||
158 | static | |
159 | int 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; |
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(); |
2629549e MD |
179 | prctl_ret = prctl(PR_GET_NAME, (unsigned long) reg_msg.name, 0, 0, 0); |
180 | if (prctl_ret) { | |
181 | ERR("Error executing prctl"); | |
182 | return -errno; | |
183 | } | |
2691221a | 184 | |
57773204 | 185 | ret = ustcomm_send_unix_sock(socket, ®_msg, sizeof(reg_msg)); |
2691221a MD |
186 | if (ret >= 0 && ret != sizeof(reg_msg)) |
187 | return -EIO; | |
188 | return ret; | |
189 | } | |
190 | ||
d9e99d10 | 191 | static |
57773204 | 192 | int send_reply(int sock, struct ustcomm_ust_reply *lur) |
d9e99d10 | 193 | { |
9eb62b9c | 194 | ssize_t len; |
d3a492d1 | 195 | |
57773204 | 196 | len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur)); |
d3a492d1 | 197 | switch (len) { |
a4be8962 | 198 | case sizeof(*lur): |
d3a492d1 MD |
199 | DBG("message successfully sent"); |
200 | return 0; | |
201 | case -1: | |
202 | if (errno == ECONNRESET) { | |
203 | printf("remote end closed connection\n"); | |
204 | return 0; | |
205 | } | |
206 | return -1; | |
207 | default: | |
208 | printf("incorrect message size: %zd\n", len); | |
209 | return -1; | |
210 | } | |
211 | } | |
212 | ||
213 | static | |
edaa1431 | 214 | int handle_register_done(struct sock_info *sock_info) |
11ff9c7d MD |
215 | { |
216 | int ret; | |
217 | ||
edaa1431 MD |
218 | if (sock_info->constructor_sem_posted) |
219 | return 0; | |
220 | sock_info->constructor_sem_posted = 1; | |
56cd7e2f MD |
221 | if (uatomic_read(&sem_count) <= 0) { |
222 | return 0; | |
223 | } | |
95259bd0 MD |
224 | ret = uatomic_add_return(&sem_count, -1); |
225 | if (ret == 0) { | |
226 | ret = sem_post(&constructor_wait); | |
227 | assert(!ret); | |
228 | } | |
11ff9c7d MD |
229 | return 0; |
230 | } | |
231 | ||
232 | static | |
233 | int handle_message(struct sock_info *sock_info, | |
57773204 | 234 | int sock, struct ustcomm_ust_msg *lum) |
d3a492d1 | 235 | { |
1ea11eab | 236 | int ret = 0; |
b61ce3b2 | 237 | const struct lttng_ust_objd_ops *ops; |
57773204 | 238 | struct ustcomm_ust_reply lur; |
193183fb | 239 | int shm_fd, wait_fd; |
1ea11eab | 240 | |
17dfb34b | 241 | ust_lock(); |
1ea11eab | 242 | |
46050b1a MD |
243 | memset(&lur, 0, sizeof(lur)); |
244 | ||
1ea11eab | 245 | if (lttng_ust_comm_should_quit) { |
46050b1a | 246 | ret = -EPERM; |
1ea11eab MD |
247 | goto end; |
248 | } | |
9eb62b9c | 249 | |
46050b1a MD |
250 | ops = objd_ops(lum->handle); |
251 | if (!ops) { | |
252 | ret = -ENOENT; | |
253 | goto end; | |
1ea11eab | 254 | } |
46050b1a MD |
255 | |
256 | switch (lum->cmd) { | |
11ff9c7d MD |
257 | case LTTNG_UST_REGISTER_DONE: |
258 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
edaa1431 | 259 | ret = handle_register_done(sock_info); |
11ff9c7d MD |
260 | else |
261 | ret = -EINVAL; | |
262 | break; | |
46050b1a MD |
263 | case LTTNG_UST_RELEASE: |
264 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
265 | ret = -EPERM; | |
266 | else | |
d4419b81 | 267 | ret = lttng_ust_objd_unref(lum->handle); |
d9e99d10 MD |
268 | break; |
269 | default: | |
46050b1a MD |
270 | if (ops->cmd) |
271 | ret = ops->cmd(lum->handle, lum->cmd, | |
272 | (unsigned long) &lum->u); | |
273 | else | |
274 | ret = -ENOSYS; | |
275 | break; | |
d9e99d10 | 276 | } |
46050b1a | 277 | |
1ea11eab | 278 | end: |
46050b1a MD |
279 | lur.handle = lum->handle; |
280 | lur.cmd = lum->cmd; | |
281 | lur.ret_val = ret; | |
282 | if (ret >= 0) { | |
57773204 | 283 | lur.ret_code = USTCOMM_OK; |
46050b1a | 284 | } else { |
57773204 | 285 | //lur.ret_code = USTCOMM_SESSION_FAIL; |
193183fb | 286 | lur.ret_code = ret; |
46050b1a | 287 | } |
193183fb MD |
288 | switch (lum->cmd) { |
289 | case LTTNG_UST_STREAM: | |
381c0f1e MD |
290 | /* |
291 | * Special-case reply to send stream info. | |
292 | * Use lum.u output. | |
293 | */ | |
294 | lur.u.stream.memory_map_size = lum->u.stream.memory_map_size; | |
193183fb MD |
295 | shm_fd = lum->u.stream.shm_fd; |
296 | wait_fd = lum->u.stream.wait_fd; | |
297 | break; | |
824f40b8 | 298 | case LTTNG_UST_METADATA: |
193183fb MD |
299 | case LTTNG_UST_CHANNEL: |
300 | lur.u.channel.memory_map_size = lum->u.channel.memory_map_size; | |
301 | shm_fd = lum->u.channel.shm_fd; | |
302 | wait_fd = lum->u.channel.wait_fd; | |
303 | break; | |
57773204 MD |
304 | case LTTNG_UST_VERSION: |
305 | lur.u.version = lum->u.version; | |
306 | break; | |
381c0f1e | 307 | } |
46050b1a | 308 | ret = send_reply(sock, &lur); |
193183fb MD |
309 | if (ret < 0) { |
310 | perror("error sending reply"); | |
311 | goto error; | |
312 | } | |
46050b1a | 313 | |
824f40b8 MD |
314 | if ((lum->cmd == LTTNG_UST_STREAM |
315 | || lum->cmd == LTTNG_UST_CHANNEL | |
316 | || lum->cmd == LTTNG_UST_METADATA) | |
57773204 | 317 | && lur.ret_code == USTCOMM_OK) { |
381c0f1e | 318 | /* we also need to send the file descriptors. */ |
57773204 | 319 | ret = ustcomm_send_fds_unix_sock(sock, |
193183fb | 320 | &shm_fd, &shm_fd, |
381c0f1e MD |
321 | 1, sizeof(int)); |
322 | if (ret < 0) { | |
323 | perror("send shm_fd"); | |
324 | goto error; | |
325 | } | |
57773204 | 326 | ret = ustcomm_send_fds_unix_sock(sock, |
193183fb | 327 | &wait_fd, &wait_fd, |
381c0f1e MD |
328 | 1, sizeof(int)); |
329 | if (ret < 0) { | |
330 | perror("send wait_fd"); | |
331 | goto error; | |
332 | } | |
333 | } | |
334 | error: | |
17dfb34b | 335 | ust_unlock(); |
1ea11eab | 336 | return ret; |
d9e99d10 MD |
337 | } |
338 | ||
46050b1a MD |
339 | static |
340 | void cleanup_sock_info(struct sock_info *sock_info) | |
341 | { | |
342 | int ret; | |
343 | ||
344 | if (sock_info->socket != -1) { | |
345 | ret = close(sock_info->socket); | |
346 | if (ret) { | |
7fc90dca | 347 | ERR("Error closing apps socket"); |
46050b1a MD |
348 | } |
349 | sock_info->socket = -1; | |
350 | } | |
351 | if (sock_info->root_handle != -1) { | |
d4419b81 | 352 | ret = lttng_ust_objd_unref(sock_info->root_handle); |
46050b1a MD |
353 | if (ret) { |
354 | ERR("Error unref root handle"); | |
355 | } | |
356 | sock_info->root_handle = -1; | |
357 | } | |
318dfea9 | 358 | sock_info->constructor_sem_posted = 0; |
7fc90dca MD |
359 | if (sock_info->wait_shm_mmap) { |
360 | ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE)); | |
361 | if (ret) { | |
362 | ERR("Error unmapping wait shm"); | |
363 | } | |
364 | sock_info->wait_shm_mmap = NULL; | |
365 | } | |
366 | } | |
367 | ||
58d4b2a2 | 368 | /* |
33bbeb90 MD |
369 | * Using fork to set umask in the child process (not multi-thread safe). |
370 | * We deal with the shm_open vs ftruncate race (happening when the | |
371 | * sessiond owns the shm and does not let everybody modify it, to ensure | |
372 | * safety against shm_unlink) by simply letting the mmap fail and | |
373 | * retrying after a few seconds. | |
374 | * For global shm, everybody has rw access to it until the sessiond | |
375 | * starts. | |
58d4b2a2 | 376 | */ |
7fc90dca | 377 | static |
58d4b2a2 | 378 | int get_wait_shm(struct sock_info *sock_info, size_t mmap_size) |
7fc90dca | 379 | { |
7fc90dca | 380 | int wait_shm_fd, ret; |
58d4b2a2 | 381 | pid_t pid; |
44e073f5 | 382 | |
58d4b2a2 | 383 | /* |
33bbeb90 | 384 | * Try to open read-only. |
58d4b2a2 | 385 | */ |
33bbeb90 | 386 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 MD |
387 | if (wait_shm_fd >= 0) { |
388 | goto end; | |
389 | } else if (wait_shm_fd < 0 && errno != ENOENT) { | |
390 | /* | |
33bbeb90 MD |
391 | * Real-only open did not work, and it's not because the |
392 | * entry was not present. It's a failure that prohibits | |
393 | * using shm. | |
58d4b2a2 | 394 | */ |
7fc90dca | 395 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
58d4b2a2 | 396 | goto end; |
7fc90dca MD |
397 | } |
398 | /* | |
58d4b2a2 MD |
399 | * If the open failed because the file did not exist, try |
400 | * creating it ourself. | |
7fc90dca | 401 | */ |
58d4b2a2 MD |
402 | pid = fork(); |
403 | if (pid > 0) { | |
404 | int status; | |
405 | ||
406 | /* | |
407 | * Parent: wait for child to return, in which case the | |
408 | * shared memory map will have been created. | |
409 | */ | |
410 | pid = wait(&status); | |
411 | if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) { | |
412 | wait_shm_fd = -1; | |
413 | goto end; | |
7fc90dca | 414 | } |
58d4b2a2 MD |
415 | /* |
416 | * Try to open read-only again after creation. | |
417 | */ | |
33bbeb90 | 418 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 MD |
419 | if (wait_shm_fd < 0) { |
420 | /* | |
421 | * Real-only open did not work. It's a failure | |
422 | * that prohibits using shm. | |
423 | */ | |
424 | ERR("Error opening shm %s", sock_info->wait_shm_path); | |
425 | goto end; | |
426 | } | |
427 | goto end; | |
428 | } else if (pid == 0) { | |
429 | int create_mode; | |
430 | ||
431 | /* Child */ | |
33bbeb90 | 432 | create_mode = S_IRUSR | S_IWUSR | S_IRGRP; |
58d4b2a2 | 433 | if (sock_info->global) |
33bbeb90 | 434 | create_mode |= S_IROTH | S_IWGRP | S_IWOTH; |
58d4b2a2 MD |
435 | /* |
436 | * We're alone in a child process, so we can modify the | |
437 | * process-wide umask. | |
438 | */ | |
33bbeb90 | 439 | umask(~create_mode); |
58d4b2a2 | 440 | /* |
33bbeb90 MD |
441 | * Try creating shm (or get rw access). |
442 | * We don't do an exclusive open, because we allow other | |
443 | * processes to create+ftruncate it concurrently. | |
58d4b2a2 MD |
444 | */ |
445 | wait_shm_fd = shm_open(sock_info->wait_shm_path, | |
446 | O_RDWR | O_CREAT, create_mode); | |
447 | if (wait_shm_fd >= 0) { | |
448 | ret = ftruncate(wait_shm_fd, mmap_size); | |
449 | if (ret) { | |
450 | PERROR("ftruncate"); | |
451 | exit(EXIT_FAILURE); | |
452 | } | |
58d4b2a2 MD |
453 | exit(EXIT_SUCCESS); |
454 | } | |
33bbeb90 MD |
455 | /* |
456 | * For local shm, we need to have rw access to accept | |
457 | * opening it: this means the local sessiond will be | |
458 | * able to wake us up. For global shm, we open it even | |
459 | * if rw access is not granted, because the root.root | |
460 | * sessiond will be able to override all rights and wake | |
461 | * us up. | |
462 | */ | |
463 | if (!sock_info->global && errno != EACCES) { | |
58d4b2a2 MD |
464 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
465 | exit(EXIT_FAILURE); | |
466 | } | |
467 | /* | |
33bbeb90 MD |
468 | * The shm exists, but we cannot open it RW. Report |
469 | * success. | |
58d4b2a2 MD |
470 | */ |
471 | exit(EXIT_SUCCESS); | |
472 | } else { | |
473 | return -1; | |
7fc90dca | 474 | } |
58d4b2a2 | 475 | end: |
33bbeb90 MD |
476 | if (wait_shm_fd >= 0 && !sock_info->global) { |
477 | struct stat statbuf; | |
478 | ||
479 | /* | |
480 | * Ensure that our user is the owner of the shm file for | |
481 | * local shm. If we do not own the file, it means our | |
482 | * sessiond will not have access to wake us up (there is | |
483 | * probably a rogue process trying to fake our | |
484 | * sessiond). Fallback to polling method in this case. | |
485 | */ | |
486 | ret = fstat(wait_shm_fd, &statbuf); | |
487 | if (ret) { | |
488 | PERROR("fstat"); | |
489 | goto error_close; | |
490 | } | |
491 | if (statbuf.st_uid != getuid()) | |
492 | goto error_close; | |
493 | } | |
58d4b2a2 | 494 | return wait_shm_fd; |
33bbeb90 MD |
495 | |
496 | error_close: | |
497 | ret = close(wait_shm_fd); | |
498 | if (ret) { | |
499 | PERROR("Error closing fd"); | |
500 | } | |
501 | return -1; | |
58d4b2a2 MD |
502 | } |
503 | ||
504 | static | |
505 | char *get_map_shm(struct sock_info *sock_info) | |
506 | { | |
507 | size_t mmap_size = sysconf(_SC_PAGE_SIZE); | |
508 | int wait_shm_fd, ret; | |
509 | char *wait_shm_mmap; | |
510 | ||
511 | wait_shm_fd = get_wait_shm(sock_info, mmap_size); | |
512 | if (wait_shm_fd < 0) { | |
513 | goto error; | |
44e073f5 | 514 | } |
7fc90dca MD |
515 | wait_shm_mmap = mmap(NULL, mmap_size, PROT_READ, |
516 | MAP_SHARED, wait_shm_fd, 0); | |
7fc90dca MD |
517 | /* close shm fd immediately after taking the mmap reference */ |
518 | ret = close(wait_shm_fd); | |
519 | if (ret) { | |
33bbeb90 MD |
520 | PERROR("Error closing fd"); |
521 | } | |
522 | if (wait_shm_mmap == MAP_FAILED) { | |
523 | DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode."); | |
524 | goto error; | |
7fc90dca MD |
525 | } |
526 | return wait_shm_mmap; | |
527 | ||
528 | error: | |
529 | return NULL; | |
530 | } | |
531 | ||
532 | static | |
533 | void wait_for_sessiond(struct sock_info *sock_info) | |
534 | { | |
80e2814b MD |
535 | int ret; |
536 | ||
7fc90dca MD |
537 | ust_lock(); |
538 | if (lttng_ust_comm_should_quit) { | |
539 | goto quit; | |
540 | } | |
37ed587a MD |
541 | if (wait_poll_fallback) { |
542 | goto error; | |
543 | } | |
7fc90dca MD |
544 | if (!sock_info->wait_shm_mmap) { |
545 | sock_info->wait_shm_mmap = get_map_shm(sock_info); | |
546 | if (!sock_info->wait_shm_mmap) | |
547 | goto error; | |
548 | } | |
549 | ust_unlock(); | |
550 | ||
551 | DBG("Waiting for %s apps sessiond", sock_info->name); | |
80e2814b MD |
552 | /* Wait for futex wakeup */ |
553 | if (uatomic_read((int32_t *) sock_info->wait_shm_mmap) == 0) { | |
554 | ret = futex_async((int32_t *) sock_info->wait_shm_mmap, | |
555 | FUTEX_WAIT, 0, NULL, NULL, 0); | |
80e2814b | 556 | if (ret < 0) { |
37ed587a MD |
557 | if (errno == EFAULT) { |
558 | wait_poll_fallback = 1; | |
f0d03086 | 559 | WARN( |
37ed587a MD |
560 | "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) " |
561 | "do not support FUTEX_WAKE on read-only memory mappings correctly. " | |
562 | "Please upgrade your kernel " | |
563 | "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel " | |
564 | "mainline). LTTng-UST will use polling mode fallback."); | |
565 | } | |
80e2814b | 566 | PERROR("futex"); |
80e2814b MD |
567 | } |
568 | } | |
7fc90dca MD |
569 | return; |
570 | ||
571 | quit: | |
572 | ust_unlock(); | |
573 | return; | |
574 | ||
575 | error: | |
576 | ust_unlock(); | |
7fc90dca | 577 | return; |
46050b1a MD |
578 | } |
579 | ||
1ea11eab MD |
580 | /* |
581 | * This thread does not allocate any resource, except within | |
582 | * handle_message, within mutex protection. This mutex protects against | |
583 | * fork and exit. | |
584 | * The other moment it allocates resources is at socket connexion, which | |
585 | * is also protected by the mutex. | |
586 | */ | |
d9e99d10 MD |
587 | static |
588 | void *ust_listener_thread(void *arg) | |
589 | { | |
1ea11eab | 590 | struct sock_info *sock_info = arg; |
c0eedf81 | 591 | int sock, ret, prev_connect_failed = 0, has_waited = 0; |
d9e99d10 | 592 | |
9eb62b9c MD |
593 | /* Restart trying to connect to the session daemon */ |
594 | restart: | |
c0eedf81 MD |
595 | if (prev_connect_failed) { |
596 | /* Wait for sessiond availability with pipe */ | |
597 | wait_for_sessiond(sock_info); | |
598 | if (has_waited) { | |
599 | has_waited = 0; | |
600 | /* | |
601 | * Sleep for 5 seconds before retrying after a | |
602 | * sequence of failure / wait / failure. This | |
603 | * deals with a killed or broken session daemon. | |
604 | */ | |
605 | sleep(5); | |
606 | } | |
607 | has_waited = 1; | |
608 | prev_connect_failed = 0; | |
609 | } | |
17dfb34b | 610 | ust_lock(); |
1ea11eab MD |
611 | |
612 | if (lttng_ust_comm_should_quit) { | |
17dfb34b | 613 | ust_unlock(); |
1ea11eab MD |
614 | goto quit; |
615 | } | |
9eb62b9c | 616 | |
1ea11eab MD |
617 | if (sock_info->socket != -1) { |
618 | ret = close(sock_info->socket); | |
619 | if (ret) { | |
11ff9c7d | 620 | ERR("Error closing %s apps socket", sock_info->name); |
1ea11eab MD |
621 | } |
622 | sock_info->socket = -1; | |
623 | } | |
46050b1a | 624 | |
9eb62b9c | 625 | /* Register */ |
57773204 | 626 | ret = ustcomm_connect_unix_sock(sock_info->sock_path); |
9eb62b9c | 627 | if (ret < 0) { |
11ff9c7d | 628 | ERR("Error connecting to %s apps socket", sock_info->name); |
c0eedf81 | 629 | prev_connect_failed = 1; |
11ff9c7d MD |
630 | /* |
631 | * If we cannot find the sessiond daemon, don't delay | |
632 | * constructor execution. | |
633 | */ | |
edaa1431 | 634 | ret = handle_register_done(sock_info); |
11ff9c7d | 635 | assert(!ret); |
17dfb34b | 636 | ust_unlock(); |
1ea11eab | 637 | goto restart; |
46050b1a MD |
638 | } |
639 | ||
640 | sock_info->socket = sock = ret; | |
641 | ||
642 | /* | |
643 | * Create only one root handle per listener thread for the whole | |
644 | * process lifetime. | |
645 | */ | |
646 | if (sock_info->root_handle == -1) { | |
647 | ret = lttng_abi_create_root_handle(); | |
a51070bb | 648 | if (ret < 0) { |
46050b1a | 649 | ERR("Error creating root handle"); |
17dfb34b | 650 | ust_unlock(); |
46050b1a MD |
651 | goto quit; |
652 | } | |
653 | sock_info->root_handle = ret; | |
9eb62b9c | 654 | } |
1ea11eab | 655 | |
9eb62b9c MD |
656 | ret = register_app_to_sessiond(sock); |
657 | if (ret < 0) { | |
11ff9c7d | 658 | ERR("Error registering to %s apps socket", sock_info->name); |
c0eedf81 | 659 | prev_connect_failed = 1; |
11ff9c7d MD |
660 | /* |
661 | * If we cannot register to the sessiond daemon, don't | |
662 | * delay constructor execution. | |
663 | */ | |
edaa1431 | 664 | ret = handle_register_done(sock_info); |
11ff9c7d | 665 | assert(!ret); |
17dfb34b | 666 | ust_unlock(); |
9eb62b9c MD |
667 | goto restart; |
668 | } | |
17dfb34b | 669 | ust_unlock(); |
46050b1a | 670 | |
d9e99d10 MD |
671 | for (;;) { |
672 | ssize_t len; | |
57773204 | 673 | struct ustcomm_ust_msg lum; |
d9e99d10 | 674 | |
57773204 | 675 | len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum)); |
d9e99d10 MD |
676 | switch (len) { |
677 | case 0: /* orderly shutdown */ | |
11ff9c7d | 678 | DBG("%s ltt-sessiond has performed an orderly shutdown\n", sock_info->name); |
d9e99d10 | 679 | goto end; |
e7723462 | 680 | case sizeof(lum): |
d9e99d10 | 681 | DBG("message received\n"); |
11ff9c7d | 682 | ret = handle_message(sock_info, sock, &lum); |
2a80c9d8 | 683 | if (ret < 0) { |
11ff9c7d | 684 | ERR("Error handling message for %s socket", sock_info->name); |
d9e99d10 MD |
685 | } |
686 | continue; | |
687 | case -1: | |
688 | if (errno == ECONNRESET) { | |
11ff9c7d | 689 | ERR("%s remote end closed connection\n", sock_info->name); |
d9e99d10 MD |
690 | goto end; |
691 | } | |
692 | goto end; | |
693 | default: | |
11ff9c7d | 694 | ERR("incorrect message size (%s socket): %zd\n", sock_info->name, len); |
d9e99d10 MD |
695 | continue; |
696 | } | |
697 | ||
698 | } | |
699 | end: | |
9eb62b9c | 700 | goto restart; /* try to reconnect */ |
1ea11eab | 701 | quit: |
d9e99d10 MD |
702 | return NULL; |
703 | } | |
704 | ||
cf12a773 MD |
705 | /* |
706 | * Return values: -1: don't wait. 0: wait forever. 1: timeout wait. | |
707 | */ | |
11ff9c7d MD |
708 | static |
709 | int get_timeout(struct timespec *constructor_timeout) | |
710 | { | |
cf12a773 MD |
711 | long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS; |
712 | char *str_delay; | |
11ff9c7d MD |
713 | int ret; |
714 | ||
cf12a773 MD |
715 | str_delay = getenv("UST_REGISTER_TIMEOUT"); |
716 | if (str_delay) { | |
717 | constructor_delay_ms = strtol(str_delay, NULL, 10); | |
718 | } | |
719 | ||
720 | switch (constructor_delay_ms) { | |
721 | case -1:/* fall-through */ | |
722 | case 0: | |
723 | return constructor_delay_ms; | |
724 | default: | |
725 | break; | |
726 | } | |
727 | ||
728 | /* | |
729 | * If we are unable to find the current time, don't wait. | |
730 | */ | |
731 | ret = clock_gettime(CLOCK_REALTIME, constructor_timeout); | |
732 | if (ret) { | |
733 | return -1; | |
734 | } | |
95259bd0 MD |
735 | constructor_timeout->tv_sec += constructor_delay_ms / 1000UL; |
736 | constructor_timeout->tv_nsec += | |
737 | (constructor_delay_ms % 1000UL) * 1000000UL; | |
11ff9c7d MD |
738 | if (constructor_timeout->tv_nsec >= 1000000000UL) { |
739 | constructor_timeout->tv_sec++; | |
740 | constructor_timeout->tv_nsec -= 1000000000UL; | |
741 | } | |
cf12a773 | 742 | return 1; |
11ff9c7d | 743 | } |
d9e99d10 | 744 | |
2691221a MD |
745 | /* |
746 | * sessiond monitoring thread: monitor presence of global and per-user | |
747 | * sessiond by polling the application common named pipe. | |
748 | */ | |
749 | /* TODO */ | |
750 | ||
edaa1431 | 751 | void __attribute__((constructor)) lttng_ust_init(void) |
2691221a | 752 | { |
11ff9c7d | 753 | struct timespec constructor_timeout; |
cf12a773 | 754 | int timeout_mode; |
2691221a MD |
755 | int ret; |
756 | ||
edaa1431 MD |
757 | if (uatomic_xchg(&initialized, 1) == 1) |
758 | return; | |
759 | ||
760 | /* | |
761 | * We want precise control over the order in which we construct | |
762 | * our sub-libraries vs starting to receive commands from | |
763 | * sessiond (otherwise leading to errors when trying to create | |
764 | * sessiond before the init functions are completed). | |
765 | */ | |
2691221a | 766 | init_usterr(); |
edaa1431 MD |
767 | init_tracepoint(); |
768 | ltt_ring_buffer_metadata_client_init(); | |
769 | ltt_ring_buffer_client_overwrite_init(); | |
770 | ltt_ring_buffer_client_discard_init(); | |
2691221a | 771 | |
cf12a773 | 772 | timeout_mode = get_timeout(&constructor_timeout); |
11ff9c7d | 773 | |
95259bd0 | 774 | ret = sem_init(&constructor_wait, 0, 0); |
11ff9c7d MD |
775 | assert(!ret); |
776 | ||
8d20bf54 | 777 | ret = setup_local_apps(); |
2691221a | 778 | if (ret) { |
8d20bf54 | 779 | ERR("Error setting up to local apps"); |
2691221a | 780 | } |
1ea11eab MD |
781 | ret = pthread_create(&local_apps.ust_listener, NULL, |
782 | ust_listener_thread, &local_apps); | |
11ff9c7d | 783 | |
8d20bf54 MD |
784 | if (local_apps.allowed) { |
785 | ret = pthread_create(&global_apps.ust_listener, NULL, | |
786 | ust_listener_thread, &global_apps); | |
787 | } else { | |
788 | handle_register_done(&local_apps); | |
789 | } | |
790 | ||
cf12a773 MD |
791 | switch (timeout_mode) { |
792 | case 1: /* timeout wait */ | |
95259bd0 MD |
793 | do { |
794 | ret = sem_timedwait(&constructor_wait, | |
795 | &constructor_timeout); | |
796 | } while (ret < 0 && errno == EINTR); | |
cf12a773 MD |
797 | if (ret < 0 && errno == ETIMEDOUT) { |
798 | ERR("Timed out waiting for ltt-sessiond"); | |
799 | } else { | |
800 | assert(!ret); | |
801 | } | |
802 | break; | |
7b766b16 | 803 | case -1:/* wait forever */ |
95259bd0 MD |
804 | do { |
805 | ret = sem_wait(&constructor_wait); | |
806 | } while (ret < 0 && errno == EINTR); | |
11ff9c7d | 807 | assert(!ret); |
cf12a773 | 808 | break; |
7b766b16 | 809 | case 0: /* no timeout */ |
cf12a773 | 810 | break; |
11ff9c7d | 811 | } |
2691221a MD |
812 | } |
813 | ||
17dfb34b MD |
814 | static |
815 | void lttng_ust_cleanup(int exiting) | |
816 | { | |
817 | cleanup_sock_info(&global_apps); | |
818 | if (local_apps.allowed) { | |
819 | cleanup_sock_info(&local_apps); | |
820 | } | |
821 | lttng_ust_abi_exit(); | |
003fedf4 | 822 | lttng_ust_events_exit(); |
17dfb34b MD |
823 | ltt_ring_buffer_client_discard_exit(); |
824 | ltt_ring_buffer_client_overwrite_exit(); | |
825 | ltt_ring_buffer_metadata_client_exit(); | |
826 | exit_tracepoint(); | |
827 | if (!exiting) { | |
828 | /* Reinitialize values for fork */ | |
829 | sem_count = 2; | |
830 | lttng_ust_comm_should_quit = 0; | |
831 | initialized = 0; | |
832 | } | |
833 | } | |
834 | ||
edaa1431 | 835 | void __attribute__((destructor)) lttng_ust_exit(void) |
2691221a MD |
836 | { |
837 | int ret; | |
838 | ||
9eb62b9c MD |
839 | /* |
840 | * Using pthread_cancel here because: | |
841 | * A) we don't want to hang application teardown. | |
842 | * B) the thread is not allocating any resource. | |
843 | */ | |
1ea11eab MD |
844 | |
845 | /* | |
846 | * Require the communication thread to quit. Synchronize with | |
847 | * mutexes to ensure it is not in a mutex critical section when | |
848 | * pthread_cancel is later called. | |
849 | */ | |
17dfb34b | 850 | ust_lock(); |
1ea11eab | 851 | lttng_ust_comm_should_quit = 1; |
17dfb34b | 852 | ust_unlock(); |
1ea11eab | 853 | |
1ea11eab | 854 | ret = pthread_cancel(global_apps.ust_listener); |
9eb62b9c MD |
855 | if (ret) { |
856 | ERR("Error cancelling global ust listener thread"); | |
2691221a | 857 | } |
8d20bf54 MD |
858 | if (local_apps.allowed) { |
859 | ret = pthread_cancel(local_apps.ust_listener); | |
860 | if (ret) { | |
861 | ERR("Error cancelling local ust listener thread"); | |
862 | } | |
8d20bf54 | 863 | } |
17dfb34b | 864 | lttng_ust_cleanup(1); |
2691221a | 865 | } |
e822f505 MD |
866 | |
867 | /* | |
868 | * We exclude the worker threads across fork and clone (except | |
869 | * CLONE_VM), because these system calls only keep the forking thread | |
870 | * running in the child. Therefore, we don't want to call fork or clone | |
871 | * in the middle of an tracepoint or ust tracing state modification. | |
872 | * Holding this mutex protects these structures across fork and clone. | |
873 | */ | |
874 | void ust_before_fork(ust_fork_info_t *fork_info) | |
875 | { | |
876 | /* | |
877 | * Disable signals. This is to avoid that the child intervenes | |
878 | * before it is properly setup for tracing. It is safer to | |
879 | * disable all signals, because then we know we are not breaking | |
880 | * anything by restoring the original mask. | |
881 | */ | |
882 | sigset_t all_sigs; | |
883 | int ret; | |
884 | ||
885 | /* Disable signals */ | |
886 | sigfillset(&all_sigs); | |
887 | ret = sigprocmask(SIG_BLOCK, &all_sigs, &fork_info->orig_sigs); | |
888 | if (ret == -1) { | |
889 | PERROR("sigprocmask"); | |
890 | } | |
17dfb34b | 891 | ust_lock(); |
e822f505 MD |
892 | rcu_bp_before_fork(); |
893 | } | |
894 | ||
895 | static void ust_after_fork_common(ust_fork_info_t *fork_info) | |
896 | { | |
897 | int ret; | |
898 | ||
17dfb34b MD |
899 | DBG("process %d", getpid()); |
900 | ust_unlock(); | |
e822f505 MD |
901 | /* Restore signals */ |
902 | ret = sigprocmask(SIG_SETMASK, &fork_info->orig_sigs, NULL); | |
903 | if (ret == -1) { | |
904 | PERROR("sigprocmask"); | |
905 | } | |
906 | } | |
907 | ||
908 | void ust_after_fork_parent(ust_fork_info_t *fork_info) | |
909 | { | |
17dfb34b | 910 | DBG("process %d", getpid()); |
e822f505 MD |
911 | rcu_bp_after_fork_parent(); |
912 | /* Release mutexes and reenable signals */ | |
913 | ust_after_fork_common(fork_info); | |
914 | } | |
915 | ||
17dfb34b MD |
916 | /* |
917 | * After fork, in the child, we need to cleanup all the leftover state, | |
918 | * except the worker thread which already magically disappeared thanks | |
919 | * to the weird Linux fork semantics. After tyding up, we call | |
920 | * lttng_ust_init() again to start over as a new PID. | |
921 | * | |
922 | * This is meant for forks() that have tracing in the child between the | |
923 | * fork and following exec call (if there is any). | |
924 | */ | |
e822f505 MD |
925 | void ust_after_fork_child(ust_fork_info_t *fork_info) |
926 | { | |
17dfb34b | 927 | DBG("process %d", getpid()); |
e822f505 MD |
928 | /* Release urcu mutexes */ |
929 | rcu_bp_after_fork_child(); | |
17dfb34b | 930 | lttng_ust_cleanup(0); |
a93bfc45 | 931 | lttng_context_vtid_reset(); |
e822f505 MD |
932 | /* Release mutexes and reenable signals */ |
933 | ust_after_fork_common(fork_info); | |
318dfea9 | 934 | lttng_ust_init(); |
e822f505 | 935 | } |