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> | |
4c1ee94e | 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" |
f645cfa7 | 49 | #include "../libringbuffer/tlsfixup.h" |
edaa1431 MD |
50 | |
51 | /* | |
52 | * Has lttng ust comm constructor been called ? | |
53 | */ | |
54 | static int initialized; | |
55 | ||
1ea11eab | 56 | /* |
17dfb34b MD |
57 | * The ust_lock/ust_unlock lock is used as a communication thread mutex. |
58 | * Held when handling a command, also held by fork() to deal with | |
59 | * removal of threads, and by exit path. | |
1ea11eab | 60 | */ |
1ea11eab MD |
61 | |
62 | /* Should the ust comm thread quit ? */ | |
63 | static int lttng_ust_comm_should_quit; | |
64 | ||
11ff9c7d MD |
65 | /* |
66 | * Wait for either of these before continuing to the main | |
67 | * program: | |
68 | * - the register_done message from sessiond daemon | |
69 | * (will let the sessiond daemon enable sessions before main | |
70 | * starts.) | |
71 | * - sessiond daemon is not reachable. | |
72 | * - timeout (ensuring applications are resilient to session | |
73 | * daemon problems). | |
74 | */ | |
75 | static sem_t constructor_wait; | |
950aab0c MD |
76 | /* |
77 | * Doing this for both the global and local sessiond. | |
78 | */ | |
95259bd0 | 79 | static int sem_count = { 2 }; |
11ff9c7d | 80 | |
1ea11eab MD |
81 | /* |
82 | * Info about socket and associated listener thread. | |
83 | */ | |
84 | struct sock_info { | |
11ff9c7d | 85 | const char *name; |
1ea11eab | 86 | pthread_t ust_listener; /* listener thread */ |
46050b1a | 87 | int root_handle; |
8d20bf54 MD |
88 | int constructor_sem_posted; |
89 | int allowed; | |
44e073f5 | 90 | int global; |
7fc90dca MD |
91 | |
92 | char sock_path[PATH_MAX]; | |
93 | int socket; | |
94 | ||
95 | char wait_shm_path[PATH_MAX]; | |
96 | char *wait_shm_mmap; | |
1ea11eab | 97 | }; |
2691221a MD |
98 | |
99 | /* Socket from app (connect) to session daemon (listen) for communication */ | |
1ea11eab | 100 | struct sock_info global_apps = { |
11ff9c7d | 101 | .name = "global", |
44e073f5 | 102 | .global = 1, |
7fc90dca | 103 | |
46050b1a | 104 | .root_handle = -1, |
8d20bf54 | 105 | .allowed = 1, |
7fc90dca MD |
106 | |
107 | .sock_path = DEFAULT_GLOBAL_APPS_UNIX_SOCK, | |
108 | .socket = -1, | |
109 | ||
110 | .wait_shm_path = DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH, | |
1ea11eab | 111 | }; |
2691221a MD |
112 | |
113 | /* TODO: allow global_apps_sock_path override */ | |
114 | ||
1ea11eab | 115 | struct sock_info local_apps = { |
11ff9c7d | 116 | .name = "local", |
44e073f5 | 117 | .global = 0, |
46050b1a | 118 | .root_handle = -1, |
8d20bf54 | 119 | .allowed = 0, /* Check setuid bit first */ |
7fc90dca MD |
120 | |
121 | .socket = -1, | |
1ea11eab | 122 | }; |
2691221a | 123 | |
37ed587a MD |
124 | static int wait_poll_fallback; |
125 | ||
edaa1431 MD |
126 | extern void ltt_ring_buffer_client_overwrite_init(void); |
127 | extern void ltt_ring_buffer_client_discard_init(void); | |
128 | extern void ltt_ring_buffer_metadata_client_init(void); | |
129 | extern void ltt_ring_buffer_client_overwrite_exit(void); | |
130 | extern void ltt_ring_buffer_client_discard_exit(void); | |
131 | extern void ltt_ring_buffer_metadata_client_exit(void); | |
132 | ||
2691221a | 133 | static |
8d20bf54 | 134 | int setup_local_apps(void) |
2691221a MD |
135 | { |
136 | const char *home_dir; | |
7fc90dca | 137 | uid_t uid; |
2691221a | 138 | |
7fc90dca | 139 | uid = getuid(); |
8d20bf54 MD |
140 | /* |
141 | * Disallow per-user tracing for setuid binaries. | |
142 | */ | |
7fc90dca | 143 | if (uid != geteuid()) { |
8d20bf54 | 144 | local_apps.allowed = 0; |
d0a1ae63 | 145 | return 0; |
8d20bf54 MD |
146 | } else { |
147 | local_apps.allowed = 1; | |
148 | } | |
2691221a MD |
149 | home_dir = (const char *) getenv("HOME"); |
150 | if (!home_dir) | |
151 | return -ENOENT; | |
1ea11eab | 152 | snprintf(local_apps.sock_path, PATH_MAX, |
2691221a | 153 | DEFAULT_HOME_APPS_UNIX_SOCK, home_dir); |
7fc90dca MD |
154 | snprintf(local_apps.wait_shm_path, PATH_MAX, |
155 | DEFAULT_HOME_APPS_WAIT_SHM_PATH, uid); | |
2691221a MD |
156 | return 0; |
157 | } | |
158 | ||
159 | static | |
160 | int register_app_to_sessiond(int socket) | |
161 | { | |
162 | ssize_t ret; | |
4c1ee94e | 163 | int prctl_ret; |
2691221a | 164 | struct { |
e44418f3 MD |
165 | uint32_t major; |
166 | uint32_t minor; | |
2691221a | 167 | pid_t pid; |
5c33bde8 | 168 | pid_t ppid; |
2691221a | 169 | uid_t uid; |
83610856 | 170 | gid_t gid; |
c117fb1b | 171 | uint32_t bits_per_long; |
2629549e | 172 | char name[16]; /* process name */ |
2691221a MD |
173 | } reg_msg; |
174 | ||
e44418f3 MD |
175 | reg_msg.major = LTTNG_UST_COMM_VERSION_MAJOR; |
176 | reg_msg.minor = LTTNG_UST_COMM_VERSION_MINOR; | |
2691221a | 177 | reg_msg.pid = getpid(); |
5c33bde8 | 178 | reg_msg.ppid = getppid(); |
2691221a | 179 | reg_msg.uid = getuid(); |
83610856 | 180 | reg_msg.gid = getgid(); |
c117fb1b | 181 | reg_msg.bits_per_long = CAA_BITS_PER_LONG; |
4c1ee94e MD |
182 | prctl_ret = prctl(PR_GET_NAME, (unsigned long) reg_msg.name, 0, 0, 0); |
183 | if (prctl_ret) { | |
184 | ERR("Error executing prctl"); | |
185 | return -errno; | |
186 | } | |
2691221a | 187 | |
57773204 | 188 | ret = ustcomm_send_unix_sock(socket, ®_msg, sizeof(reg_msg)); |
2691221a MD |
189 | if (ret >= 0 && ret != sizeof(reg_msg)) |
190 | return -EIO; | |
191 | return ret; | |
192 | } | |
193 | ||
d9e99d10 | 194 | static |
57773204 | 195 | int send_reply(int sock, struct ustcomm_ust_reply *lur) |
d9e99d10 | 196 | { |
9eb62b9c | 197 | ssize_t len; |
d3a492d1 | 198 | |
57773204 | 199 | len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur)); |
d3a492d1 | 200 | switch (len) { |
a4be8962 | 201 | case sizeof(*lur): |
d3a492d1 MD |
202 | DBG("message successfully sent"); |
203 | return 0; | |
204 | case -1: | |
205 | if (errno == ECONNRESET) { | |
206 | printf("remote end closed connection\n"); | |
207 | return 0; | |
208 | } | |
209 | return -1; | |
210 | default: | |
211 | printf("incorrect message size: %zd\n", len); | |
212 | return -1; | |
213 | } | |
214 | } | |
215 | ||
216 | static | |
edaa1431 | 217 | int handle_register_done(struct sock_info *sock_info) |
11ff9c7d MD |
218 | { |
219 | int ret; | |
220 | ||
edaa1431 MD |
221 | if (sock_info->constructor_sem_posted) |
222 | return 0; | |
223 | sock_info->constructor_sem_posted = 1; | |
56cd7e2f MD |
224 | if (uatomic_read(&sem_count) <= 0) { |
225 | return 0; | |
226 | } | |
95259bd0 MD |
227 | ret = uatomic_add_return(&sem_count, -1); |
228 | if (ret == 0) { | |
229 | ret = sem_post(&constructor_wait); | |
230 | assert(!ret); | |
231 | } | |
11ff9c7d MD |
232 | return 0; |
233 | } | |
234 | ||
235 | static | |
236 | int handle_message(struct sock_info *sock_info, | |
57773204 | 237 | int sock, struct ustcomm_ust_msg *lum) |
d3a492d1 | 238 | { |
1ea11eab | 239 | int ret = 0; |
b61ce3b2 | 240 | const struct lttng_ust_objd_ops *ops; |
57773204 | 241 | struct ustcomm_ust_reply lur; |
193183fb | 242 | int shm_fd, wait_fd; |
ef9ff354 | 243 | union ust_args args; |
1ea11eab | 244 | |
17dfb34b | 245 | ust_lock(); |
1ea11eab | 246 | |
46050b1a MD |
247 | memset(&lur, 0, sizeof(lur)); |
248 | ||
1ea11eab | 249 | if (lttng_ust_comm_should_quit) { |
46050b1a | 250 | ret = -EPERM; |
1ea11eab MD |
251 | goto end; |
252 | } | |
9eb62b9c | 253 | |
46050b1a MD |
254 | ops = objd_ops(lum->handle); |
255 | if (!ops) { | |
256 | ret = -ENOENT; | |
257 | goto end; | |
1ea11eab | 258 | } |
46050b1a MD |
259 | |
260 | switch (lum->cmd) { | |
11ff9c7d MD |
261 | case LTTNG_UST_REGISTER_DONE: |
262 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
edaa1431 | 263 | ret = handle_register_done(sock_info); |
11ff9c7d MD |
264 | else |
265 | ret = -EINVAL; | |
266 | break; | |
46050b1a MD |
267 | case LTTNG_UST_RELEASE: |
268 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
269 | ret = -EPERM; | |
270 | else | |
d4419b81 | 271 | ret = lttng_ust_objd_unref(lum->handle); |
d9e99d10 MD |
272 | break; |
273 | default: | |
46050b1a MD |
274 | if (ops->cmd) |
275 | ret = ops->cmd(lum->handle, lum->cmd, | |
ef9ff354 MD |
276 | (unsigned long) &lum->u, |
277 | &args); | |
46050b1a MD |
278 | else |
279 | ret = -ENOSYS; | |
280 | break; | |
d9e99d10 | 281 | } |
46050b1a | 282 | |
1ea11eab | 283 | end: |
46050b1a MD |
284 | lur.handle = lum->handle; |
285 | lur.cmd = lum->cmd; | |
286 | lur.ret_val = ret; | |
287 | if (ret >= 0) { | |
57773204 | 288 | lur.ret_code = USTCOMM_OK; |
46050b1a | 289 | } else { |
57773204 | 290 | //lur.ret_code = USTCOMM_SESSION_FAIL; |
193183fb | 291 | lur.ret_code = ret; |
46050b1a | 292 | } |
e6ea14c5 MD |
293 | if (ret >= 0) { |
294 | switch (lum->cmd) { | |
295 | case LTTNG_UST_STREAM: | |
296 | /* | |
297 | * Special-case reply to send stream info. | |
298 | * Use lum.u output. | |
299 | */ | |
300 | lur.u.stream.memory_map_size = *args.stream.memory_map_size; | |
301 | shm_fd = *args.stream.shm_fd; | |
302 | wait_fd = *args.stream.wait_fd; | |
303 | break; | |
304 | case LTTNG_UST_METADATA: | |
305 | case LTTNG_UST_CHANNEL: | |
306 | lur.u.channel.memory_map_size = *args.channel.memory_map_size; | |
307 | shm_fd = *args.channel.shm_fd; | |
308 | wait_fd = *args.channel.wait_fd; | |
309 | break; | |
310 | case LTTNG_UST_TRACER_VERSION: | |
311 | lur.u.version = lum->u.version; | |
312 | break; | |
313 | case LTTNG_UST_TRACEPOINT_LIST_GET: | |
314 | memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint)); | |
315 | break; | |
316 | } | |
381c0f1e | 317 | } |
46050b1a | 318 | ret = send_reply(sock, &lur); |
193183fb MD |
319 | if (ret < 0) { |
320 | perror("error sending reply"); | |
321 | goto error; | |
322 | } | |
46050b1a | 323 | |
824f40b8 MD |
324 | if ((lum->cmd == LTTNG_UST_STREAM |
325 | || lum->cmd == LTTNG_UST_CHANNEL | |
326 | || lum->cmd == LTTNG_UST_METADATA) | |
57773204 | 327 | && lur.ret_code == USTCOMM_OK) { |
5f003977 MD |
328 | int sendret = 0; |
329 | ||
381c0f1e | 330 | /* we also need to send the file descriptors. */ |
57773204 | 331 | ret = ustcomm_send_fds_unix_sock(sock, |
193183fb | 332 | &shm_fd, &shm_fd, |
381c0f1e MD |
333 | 1, sizeof(int)); |
334 | if (ret < 0) { | |
335 | perror("send shm_fd"); | |
5f003977 | 336 | sendret = ret; |
381c0f1e | 337 | } |
5f003977 MD |
338 | /* |
339 | * The sessiond expects 2 file descriptors, even upon | |
340 | * error. | |
341 | */ | |
57773204 | 342 | ret = ustcomm_send_fds_unix_sock(sock, |
193183fb | 343 | &wait_fd, &wait_fd, |
381c0f1e MD |
344 | 1, sizeof(int)); |
345 | if (ret < 0) { | |
346 | perror("send wait_fd"); | |
347 | goto error; | |
348 | } | |
5f003977 MD |
349 | if (sendret) { |
350 | ret = sendret; | |
351 | goto error; | |
352 | } | |
381c0f1e | 353 | } |
ef9ff354 MD |
354 | /* |
355 | * We still have the memory map reference, and the fds have been | |
3ad5f707 MD |
356 | * sent to the sessiond. We can therefore close those fds. Note |
357 | * that we keep the write side of the wait_fd open, but close | |
358 | * the read side. | |
ef9ff354 MD |
359 | */ |
360 | if (lur.ret_code == USTCOMM_OK) { | |
361 | switch (lum->cmd) { | |
362 | case LTTNG_UST_STREAM: | |
363 | if (shm_fd >= 0) { | |
364 | ret = close(shm_fd); | |
365 | if (ret) { | |
366 | PERROR("Error closing stream shm_fd"); | |
367 | } | |
368 | *args.stream.shm_fd = -1; | |
369 | } | |
370 | if (wait_fd >= 0) { | |
371 | ret = close(wait_fd); | |
372 | if (ret) { | |
373 | PERROR("Error closing stream wait_fd"); | |
374 | } | |
375 | *args.stream.wait_fd = -1; | |
376 | } | |
377 | break; | |
378 | case LTTNG_UST_METADATA: | |
379 | case LTTNG_UST_CHANNEL: | |
380 | if (shm_fd >= 0) { | |
381 | ret = close(shm_fd); | |
382 | if (ret) { | |
383 | PERROR("Error closing channel shm_fd"); | |
384 | } | |
385 | *args.channel.shm_fd = -1; | |
386 | } | |
387 | if (wait_fd >= 0) { | |
388 | ret = close(wait_fd); | |
389 | if (ret) { | |
390 | PERROR("Error closing channel wait_fd"); | |
391 | } | |
392 | *args.channel.wait_fd = -1; | |
393 | } | |
394 | break; | |
395 | } | |
396 | } | |
397 | ||
381c0f1e | 398 | error: |
17dfb34b | 399 | ust_unlock(); |
1ea11eab | 400 | return ret; |
d9e99d10 MD |
401 | } |
402 | ||
46050b1a | 403 | static |
efe0de09 | 404 | void cleanup_sock_info(struct sock_info *sock_info, int exiting) |
46050b1a MD |
405 | { |
406 | int ret; | |
407 | ||
408 | if (sock_info->socket != -1) { | |
e6973a89 | 409 | ret = ustcomm_close_unix_sock(sock_info->socket); |
46050b1a | 410 | if (ret) { |
7fc90dca | 411 | ERR("Error closing apps socket"); |
46050b1a MD |
412 | } |
413 | sock_info->socket = -1; | |
414 | } | |
415 | if (sock_info->root_handle != -1) { | |
d4419b81 | 416 | ret = lttng_ust_objd_unref(sock_info->root_handle); |
46050b1a MD |
417 | if (ret) { |
418 | ERR("Error unref root handle"); | |
419 | } | |
420 | sock_info->root_handle = -1; | |
421 | } | |
318dfea9 | 422 | sock_info->constructor_sem_posted = 0; |
efe0de09 MD |
423 | /* |
424 | * wait_shm_mmap is used by listener threads outside of the | |
425 | * ust lock, so we cannot tear it down ourselves, because we | |
426 | * cannot join on these threads. Leave this task to the OS | |
427 | * process exit. | |
428 | */ | |
429 | if (!exiting && sock_info->wait_shm_mmap) { | |
7fc90dca MD |
430 | ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE)); |
431 | if (ret) { | |
432 | ERR("Error unmapping wait shm"); | |
433 | } | |
434 | sock_info->wait_shm_mmap = NULL; | |
435 | } | |
436 | } | |
437 | ||
58d4b2a2 | 438 | /* |
33bbeb90 MD |
439 | * Using fork to set umask in the child process (not multi-thread safe). |
440 | * We deal with the shm_open vs ftruncate race (happening when the | |
441 | * sessiond owns the shm and does not let everybody modify it, to ensure | |
442 | * safety against shm_unlink) by simply letting the mmap fail and | |
443 | * retrying after a few seconds. | |
444 | * For global shm, everybody has rw access to it until the sessiond | |
445 | * starts. | |
58d4b2a2 | 446 | */ |
7fc90dca | 447 | static |
58d4b2a2 | 448 | int get_wait_shm(struct sock_info *sock_info, size_t mmap_size) |
7fc90dca | 449 | { |
7fc90dca | 450 | int wait_shm_fd, ret; |
58d4b2a2 | 451 | pid_t pid; |
44e073f5 | 452 | |
58d4b2a2 | 453 | /* |
33bbeb90 | 454 | * Try to open read-only. |
58d4b2a2 | 455 | */ |
33bbeb90 | 456 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 MD |
457 | if (wait_shm_fd >= 0) { |
458 | goto end; | |
459 | } else if (wait_shm_fd < 0 && errno != ENOENT) { | |
460 | /* | |
33bbeb90 MD |
461 | * Real-only open did not work, and it's not because the |
462 | * entry was not present. It's a failure that prohibits | |
463 | * using shm. | |
58d4b2a2 | 464 | */ |
7fc90dca | 465 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
58d4b2a2 | 466 | goto end; |
7fc90dca MD |
467 | } |
468 | /* | |
58d4b2a2 MD |
469 | * If the open failed because the file did not exist, try |
470 | * creating it ourself. | |
7fc90dca | 471 | */ |
58d4b2a2 MD |
472 | pid = fork(); |
473 | if (pid > 0) { | |
474 | int status; | |
475 | ||
476 | /* | |
477 | * Parent: wait for child to return, in which case the | |
478 | * shared memory map will have been created. | |
479 | */ | |
480 | pid = wait(&status); | |
b7d3cb32 | 481 | if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) { |
58d4b2a2 MD |
482 | wait_shm_fd = -1; |
483 | goto end; | |
7fc90dca | 484 | } |
58d4b2a2 MD |
485 | /* |
486 | * Try to open read-only again after creation. | |
487 | */ | |
33bbeb90 | 488 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 MD |
489 | if (wait_shm_fd < 0) { |
490 | /* | |
491 | * Real-only open did not work. It's a failure | |
492 | * that prohibits using shm. | |
493 | */ | |
494 | ERR("Error opening shm %s", sock_info->wait_shm_path); | |
495 | goto end; | |
496 | } | |
497 | goto end; | |
498 | } else if (pid == 0) { | |
499 | int create_mode; | |
500 | ||
501 | /* Child */ | |
33bbeb90 | 502 | create_mode = S_IRUSR | S_IWUSR | S_IRGRP; |
58d4b2a2 | 503 | if (sock_info->global) |
33bbeb90 | 504 | create_mode |= S_IROTH | S_IWGRP | S_IWOTH; |
58d4b2a2 MD |
505 | /* |
506 | * We're alone in a child process, so we can modify the | |
507 | * process-wide umask. | |
508 | */ | |
33bbeb90 | 509 | umask(~create_mode); |
58d4b2a2 | 510 | /* |
33bbeb90 MD |
511 | * Try creating shm (or get rw access). |
512 | * We don't do an exclusive open, because we allow other | |
513 | * processes to create+ftruncate it concurrently. | |
58d4b2a2 MD |
514 | */ |
515 | wait_shm_fd = shm_open(sock_info->wait_shm_path, | |
516 | O_RDWR | O_CREAT, create_mode); | |
517 | if (wait_shm_fd >= 0) { | |
518 | ret = ftruncate(wait_shm_fd, mmap_size); | |
519 | if (ret) { | |
520 | PERROR("ftruncate"); | |
521 | exit(EXIT_FAILURE); | |
522 | } | |
58d4b2a2 MD |
523 | exit(EXIT_SUCCESS); |
524 | } | |
33bbeb90 MD |
525 | /* |
526 | * For local shm, we need to have rw access to accept | |
527 | * opening it: this means the local sessiond will be | |
528 | * able to wake us up. For global shm, we open it even | |
529 | * if rw access is not granted, because the root.root | |
530 | * sessiond will be able to override all rights and wake | |
531 | * us up. | |
532 | */ | |
533 | if (!sock_info->global && errno != EACCES) { | |
58d4b2a2 MD |
534 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
535 | exit(EXIT_FAILURE); | |
536 | } | |
537 | /* | |
33bbeb90 MD |
538 | * The shm exists, but we cannot open it RW. Report |
539 | * success. | |
58d4b2a2 MD |
540 | */ |
541 | exit(EXIT_SUCCESS); | |
542 | } else { | |
543 | return -1; | |
7fc90dca | 544 | } |
58d4b2a2 | 545 | end: |
33bbeb90 MD |
546 | if (wait_shm_fd >= 0 && !sock_info->global) { |
547 | struct stat statbuf; | |
548 | ||
549 | /* | |
550 | * Ensure that our user is the owner of the shm file for | |
551 | * local shm. If we do not own the file, it means our | |
552 | * sessiond will not have access to wake us up (there is | |
553 | * probably a rogue process trying to fake our | |
554 | * sessiond). Fallback to polling method in this case. | |
555 | */ | |
556 | ret = fstat(wait_shm_fd, &statbuf); | |
557 | if (ret) { | |
558 | PERROR("fstat"); | |
559 | goto error_close; | |
560 | } | |
561 | if (statbuf.st_uid != getuid()) | |
562 | goto error_close; | |
563 | } | |
58d4b2a2 | 564 | return wait_shm_fd; |
33bbeb90 MD |
565 | |
566 | error_close: | |
567 | ret = close(wait_shm_fd); | |
568 | if (ret) { | |
569 | PERROR("Error closing fd"); | |
570 | } | |
571 | return -1; | |
58d4b2a2 MD |
572 | } |
573 | ||
574 | static | |
575 | char *get_map_shm(struct sock_info *sock_info) | |
576 | { | |
577 | size_t mmap_size = sysconf(_SC_PAGE_SIZE); | |
578 | int wait_shm_fd, ret; | |
579 | char *wait_shm_mmap; | |
580 | ||
581 | wait_shm_fd = get_wait_shm(sock_info, mmap_size); | |
582 | if (wait_shm_fd < 0) { | |
583 | goto error; | |
44e073f5 | 584 | } |
7fc90dca MD |
585 | wait_shm_mmap = mmap(NULL, mmap_size, PROT_READ, |
586 | MAP_SHARED, wait_shm_fd, 0); | |
7fc90dca MD |
587 | /* close shm fd immediately after taking the mmap reference */ |
588 | ret = close(wait_shm_fd); | |
589 | if (ret) { | |
33bbeb90 MD |
590 | PERROR("Error closing fd"); |
591 | } | |
592 | if (wait_shm_mmap == MAP_FAILED) { | |
593 | DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode."); | |
594 | goto error; | |
7fc90dca MD |
595 | } |
596 | return wait_shm_mmap; | |
597 | ||
598 | error: | |
599 | return NULL; | |
600 | } | |
601 | ||
602 | static | |
603 | void wait_for_sessiond(struct sock_info *sock_info) | |
604 | { | |
efe0de09 | 605 | int ret; |
80e2814b | 606 | |
7fc90dca MD |
607 | ust_lock(); |
608 | if (lttng_ust_comm_should_quit) { | |
609 | goto quit; | |
610 | } | |
37ed587a MD |
611 | if (wait_poll_fallback) { |
612 | goto error; | |
613 | } | |
7fc90dca MD |
614 | if (!sock_info->wait_shm_mmap) { |
615 | sock_info->wait_shm_mmap = get_map_shm(sock_info); | |
616 | if (!sock_info->wait_shm_mmap) | |
617 | goto error; | |
618 | } | |
619 | ust_unlock(); | |
620 | ||
621 | DBG("Waiting for %s apps sessiond", sock_info->name); | |
80e2814b MD |
622 | /* Wait for futex wakeup */ |
623 | if (uatomic_read((int32_t *) sock_info->wait_shm_mmap) == 0) { | |
624 | ret = futex_async((int32_t *) sock_info->wait_shm_mmap, | |
625 | FUTEX_WAIT, 0, NULL, NULL, 0); | |
80e2814b | 626 | if (ret < 0) { |
37ed587a MD |
627 | if (errno == EFAULT) { |
628 | wait_poll_fallback = 1; | |
a8b870ad | 629 | DBG( |
37ed587a MD |
630 | "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) " |
631 | "do not support FUTEX_WAKE on read-only memory mappings correctly. " | |
632 | "Please upgrade your kernel " | |
633 | "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel " | |
634 | "mainline). LTTng-UST will use polling mode fallback."); | |
cd27263b MD |
635 | if (ust_debug()) |
636 | PERROR("futex"); | |
37ed587a | 637 | } |
80e2814b MD |
638 | } |
639 | } | |
7fc90dca MD |
640 | return; |
641 | ||
642 | quit: | |
643 | ust_unlock(); | |
644 | return; | |
645 | ||
646 | error: | |
647 | ust_unlock(); | |
7fc90dca | 648 | return; |
46050b1a MD |
649 | } |
650 | ||
1ea11eab MD |
651 | /* |
652 | * This thread does not allocate any resource, except within | |
653 | * handle_message, within mutex protection. This mutex protects against | |
654 | * fork and exit. | |
655 | * The other moment it allocates resources is at socket connexion, which | |
656 | * is also protected by the mutex. | |
657 | */ | |
d9e99d10 MD |
658 | static |
659 | void *ust_listener_thread(void *arg) | |
660 | { | |
1ea11eab | 661 | struct sock_info *sock_info = arg; |
c0eedf81 | 662 | int sock, ret, prev_connect_failed = 0, has_waited = 0; |
d9e99d10 | 663 | |
9eb62b9c MD |
664 | /* Restart trying to connect to the session daemon */ |
665 | restart: | |
c0eedf81 MD |
666 | if (prev_connect_failed) { |
667 | /* Wait for sessiond availability with pipe */ | |
668 | wait_for_sessiond(sock_info); | |
669 | if (has_waited) { | |
670 | has_waited = 0; | |
671 | /* | |
672 | * Sleep for 5 seconds before retrying after a | |
673 | * sequence of failure / wait / failure. This | |
674 | * deals with a killed or broken session daemon. | |
675 | */ | |
676 | sleep(5); | |
677 | } | |
678 | has_waited = 1; | |
679 | prev_connect_failed = 0; | |
680 | } | |
17dfb34b | 681 | ust_lock(); |
1ea11eab MD |
682 | |
683 | if (lttng_ust_comm_should_quit) { | |
17dfb34b | 684 | ust_unlock(); |
1ea11eab MD |
685 | goto quit; |
686 | } | |
9eb62b9c | 687 | |
1ea11eab | 688 | if (sock_info->socket != -1) { |
e6973a89 | 689 | ret = ustcomm_close_unix_sock(sock_info->socket); |
1ea11eab | 690 | if (ret) { |
11ff9c7d | 691 | ERR("Error closing %s apps socket", sock_info->name); |
1ea11eab MD |
692 | } |
693 | sock_info->socket = -1; | |
694 | } | |
46050b1a | 695 | |
9eb62b9c | 696 | /* Register */ |
57773204 | 697 | ret = ustcomm_connect_unix_sock(sock_info->sock_path); |
9eb62b9c | 698 | if (ret < 0) { |
4d3c9523 | 699 | DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name); |
c0eedf81 | 700 | prev_connect_failed = 1; |
11ff9c7d MD |
701 | /* |
702 | * If we cannot find the sessiond daemon, don't delay | |
703 | * constructor execution. | |
704 | */ | |
edaa1431 | 705 | ret = handle_register_done(sock_info); |
11ff9c7d | 706 | assert(!ret); |
17dfb34b | 707 | ust_unlock(); |
1ea11eab | 708 | goto restart; |
46050b1a MD |
709 | } |
710 | ||
711 | sock_info->socket = sock = ret; | |
712 | ||
713 | /* | |
714 | * Create only one root handle per listener thread for the whole | |
715 | * process lifetime. | |
716 | */ | |
717 | if (sock_info->root_handle == -1) { | |
718 | ret = lttng_abi_create_root_handle(); | |
a51070bb | 719 | if (ret < 0) { |
46050b1a | 720 | ERR("Error creating root handle"); |
17dfb34b | 721 | ust_unlock(); |
46050b1a MD |
722 | goto quit; |
723 | } | |
724 | sock_info->root_handle = ret; | |
9eb62b9c | 725 | } |
1ea11eab | 726 | |
9eb62b9c MD |
727 | ret = register_app_to_sessiond(sock); |
728 | if (ret < 0) { | |
11ff9c7d | 729 | ERR("Error registering to %s apps socket", sock_info->name); |
c0eedf81 | 730 | prev_connect_failed = 1; |
11ff9c7d MD |
731 | /* |
732 | * If we cannot register to the sessiond daemon, don't | |
733 | * delay constructor execution. | |
734 | */ | |
edaa1431 | 735 | ret = handle_register_done(sock_info); |
11ff9c7d | 736 | assert(!ret); |
17dfb34b | 737 | ust_unlock(); |
9eb62b9c MD |
738 | goto restart; |
739 | } | |
17dfb34b | 740 | ust_unlock(); |
46050b1a | 741 | |
d9e99d10 MD |
742 | for (;;) { |
743 | ssize_t len; | |
57773204 | 744 | struct ustcomm_ust_msg lum; |
d9e99d10 | 745 | |
57773204 | 746 | len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum)); |
d9e99d10 MD |
747 | switch (len) { |
748 | case 0: /* orderly shutdown */ | |
11ff9c7d | 749 | DBG("%s ltt-sessiond has performed an orderly shutdown\n", sock_info->name); |
8236ba10 MD |
750 | ust_lock(); |
751 | /* | |
752 | * Either sessiond has shutdown or refused us by closing the socket. | |
753 | * In either case, we don't want to delay construction execution, | |
754 | * and we need to wait before retry. | |
755 | */ | |
756 | prev_connect_failed = 1; | |
757 | /* | |
758 | * If we cannot register to the sessiond daemon, don't | |
759 | * delay constructor execution. | |
760 | */ | |
761 | ret = handle_register_done(sock_info); | |
762 | assert(!ret); | |
763 | ust_unlock(); | |
d9e99d10 | 764 | goto end; |
e7723462 | 765 | case sizeof(lum): |
d9e99d10 | 766 | DBG("message received\n"); |
11ff9c7d | 767 | ret = handle_message(sock_info, sock, &lum); |
2a80c9d8 | 768 | if (ret < 0) { |
11ff9c7d | 769 | ERR("Error handling message for %s socket", sock_info->name); |
d9e99d10 MD |
770 | } |
771 | continue; | |
772 | case -1: | |
8236ba10 | 773 | DBG("Receive failed from lttng-sessiond with errno %d", errno); |
d9e99d10 | 774 | if (errno == ECONNRESET) { |
11ff9c7d | 775 | ERR("%s remote end closed connection\n", sock_info->name); |
d9e99d10 MD |
776 | goto end; |
777 | } | |
778 | goto end; | |
779 | default: | |
11ff9c7d | 780 | ERR("incorrect message size (%s socket): %zd\n", sock_info->name, len); |
d9e99d10 MD |
781 | continue; |
782 | } | |
783 | ||
784 | } | |
785 | end: | |
9eb62b9c | 786 | goto restart; /* try to reconnect */ |
1ea11eab | 787 | quit: |
d9e99d10 MD |
788 | return NULL; |
789 | } | |
790 | ||
cf12a773 MD |
791 | /* |
792 | * Return values: -1: don't wait. 0: wait forever. 1: timeout wait. | |
793 | */ | |
11ff9c7d MD |
794 | static |
795 | int get_timeout(struct timespec *constructor_timeout) | |
796 | { | |
cf12a773 MD |
797 | long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS; |
798 | char *str_delay; | |
11ff9c7d MD |
799 | int ret; |
800 | ||
69400ac4 | 801 | str_delay = getenv("LTTNG_UST_REGISTER_TIMEOUT"); |
cf12a773 MD |
802 | if (str_delay) { |
803 | constructor_delay_ms = strtol(str_delay, NULL, 10); | |
804 | } | |
805 | ||
806 | switch (constructor_delay_ms) { | |
807 | case -1:/* fall-through */ | |
808 | case 0: | |
809 | return constructor_delay_ms; | |
810 | default: | |
811 | break; | |
812 | } | |
813 | ||
814 | /* | |
815 | * If we are unable to find the current time, don't wait. | |
816 | */ | |
817 | ret = clock_gettime(CLOCK_REALTIME, constructor_timeout); | |
818 | if (ret) { | |
819 | return -1; | |
820 | } | |
95259bd0 MD |
821 | constructor_timeout->tv_sec += constructor_delay_ms / 1000UL; |
822 | constructor_timeout->tv_nsec += | |
823 | (constructor_delay_ms % 1000UL) * 1000000UL; | |
11ff9c7d MD |
824 | if (constructor_timeout->tv_nsec >= 1000000000UL) { |
825 | constructor_timeout->tv_sec++; | |
826 | constructor_timeout->tv_nsec -= 1000000000UL; | |
827 | } | |
cf12a773 | 828 | return 1; |
11ff9c7d | 829 | } |
d9e99d10 | 830 | |
2691221a MD |
831 | /* |
832 | * sessiond monitoring thread: monitor presence of global and per-user | |
833 | * sessiond by polling the application common named pipe. | |
834 | */ | |
835 | /* TODO */ | |
836 | ||
edaa1431 | 837 | void __attribute__((constructor)) lttng_ust_init(void) |
2691221a | 838 | { |
11ff9c7d | 839 | struct timespec constructor_timeout; |
cf12a773 | 840 | int timeout_mode; |
2691221a MD |
841 | int ret; |
842 | ||
edaa1431 MD |
843 | if (uatomic_xchg(&initialized, 1) == 1) |
844 | return; | |
845 | ||
eddd8d5d MD |
846 | /* |
847 | * Fixup interdependency between TLS fixup mutex (which happens | |
848 | * to be the dynamic linker mutex) and ust_lock, taken within | |
849 | * the ust lock. | |
850 | */ | |
851 | lttng_fixup_event_tls(); | |
f645cfa7 | 852 | lttng_fixup_ringbuffer_tls(); |
4158a15a | 853 | lttng_fixup_vtid_tls(); |
eddd8d5d | 854 | |
edaa1431 MD |
855 | /* |
856 | * We want precise control over the order in which we construct | |
857 | * our sub-libraries vs starting to receive commands from | |
858 | * sessiond (otherwise leading to errors when trying to create | |
859 | * sessiond before the init functions are completed). | |
860 | */ | |
2691221a | 861 | init_usterr(); |
edaa1431 MD |
862 | init_tracepoint(); |
863 | ltt_ring_buffer_metadata_client_init(); | |
864 | ltt_ring_buffer_client_overwrite_init(); | |
865 | ltt_ring_buffer_client_discard_init(); | |
2691221a | 866 | |
cf12a773 | 867 | timeout_mode = get_timeout(&constructor_timeout); |
11ff9c7d | 868 | |
95259bd0 | 869 | ret = sem_init(&constructor_wait, 0, 0); |
11ff9c7d MD |
870 | assert(!ret); |
871 | ||
8d20bf54 | 872 | ret = setup_local_apps(); |
2691221a | 873 | if (ret) { |
8d20bf54 | 874 | ERR("Error setting up to local apps"); |
2691221a | 875 | } |
1ea11eab MD |
876 | ret = pthread_create(&local_apps.ust_listener, NULL, |
877 | ust_listener_thread, &local_apps); | |
11ff9c7d | 878 | |
8d20bf54 MD |
879 | if (local_apps.allowed) { |
880 | ret = pthread_create(&global_apps.ust_listener, NULL, | |
881 | ust_listener_thread, &global_apps); | |
882 | } else { | |
883 | handle_register_done(&local_apps); | |
884 | } | |
885 | ||
cf12a773 MD |
886 | switch (timeout_mode) { |
887 | case 1: /* timeout wait */ | |
95259bd0 MD |
888 | do { |
889 | ret = sem_timedwait(&constructor_wait, | |
890 | &constructor_timeout); | |
891 | } while (ret < 0 && errno == EINTR); | |
cf12a773 MD |
892 | if (ret < 0 && errno == ETIMEDOUT) { |
893 | ERR("Timed out waiting for ltt-sessiond"); | |
894 | } else { | |
895 | assert(!ret); | |
896 | } | |
897 | break; | |
7b766b16 | 898 | case -1:/* wait forever */ |
95259bd0 MD |
899 | do { |
900 | ret = sem_wait(&constructor_wait); | |
901 | } while (ret < 0 && errno == EINTR); | |
11ff9c7d | 902 | assert(!ret); |
cf12a773 | 903 | break; |
7b766b16 | 904 | case 0: /* no timeout */ |
cf12a773 | 905 | break; |
11ff9c7d | 906 | } |
2691221a MD |
907 | } |
908 | ||
17dfb34b MD |
909 | static |
910 | void lttng_ust_cleanup(int exiting) | |
911 | { | |
efe0de09 | 912 | cleanup_sock_info(&global_apps, exiting); |
17dfb34b | 913 | if (local_apps.allowed) { |
efe0de09 | 914 | cleanup_sock_info(&local_apps, exiting); |
17dfb34b | 915 | } |
efe0de09 MD |
916 | /* |
917 | * The teardown in this function all affect data structures | |
918 | * accessed under the UST lock by the listener thread. This | |
919 | * lock, along with the lttng_ust_comm_should_quit flag, ensure | |
920 | * that none of these threads are accessing this data at this | |
921 | * point. | |
922 | */ | |
17dfb34b | 923 | lttng_ust_abi_exit(); |
003fedf4 | 924 | lttng_ust_events_exit(); |
17dfb34b MD |
925 | ltt_ring_buffer_client_discard_exit(); |
926 | ltt_ring_buffer_client_overwrite_exit(); | |
927 | ltt_ring_buffer_metadata_client_exit(); | |
928 | exit_tracepoint(); | |
929 | if (!exiting) { | |
930 | /* Reinitialize values for fork */ | |
931 | sem_count = 2; | |
932 | lttng_ust_comm_should_quit = 0; | |
933 | initialized = 0; | |
934 | } | |
935 | } | |
936 | ||
edaa1431 | 937 | void __attribute__((destructor)) lttng_ust_exit(void) |
2691221a MD |
938 | { |
939 | int ret; | |
940 | ||
9eb62b9c MD |
941 | /* |
942 | * Using pthread_cancel here because: | |
943 | * A) we don't want to hang application teardown. | |
944 | * B) the thread is not allocating any resource. | |
945 | */ | |
1ea11eab MD |
946 | |
947 | /* | |
948 | * Require the communication thread to quit. Synchronize with | |
949 | * mutexes to ensure it is not in a mutex critical section when | |
950 | * pthread_cancel is later called. | |
951 | */ | |
17dfb34b | 952 | ust_lock(); |
1ea11eab | 953 | lttng_ust_comm_should_quit = 1; |
17dfb34b | 954 | ust_unlock(); |
1ea11eab | 955 | |
f5f94532 | 956 | /* cancel threads */ |
1ea11eab | 957 | ret = pthread_cancel(global_apps.ust_listener); |
9eb62b9c MD |
958 | if (ret) { |
959 | ERR("Error cancelling global ust listener thread"); | |
2691221a | 960 | } |
8d20bf54 MD |
961 | if (local_apps.allowed) { |
962 | ret = pthread_cancel(local_apps.ust_listener); | |
963 | if (ret) { | |
964 | ERR("Error cancelling local ust listener thread"); | |
965 | } | |
8d20bf54 | 966 | } |
efe0de09 MD |
967 | /* |
968 | * Do NOT join threads: use of sys_futex makes it impossible to | |
969 | * join the threads without using async-cancel, but async-cancel | |
970 | * is delivered by a signal, which could hit the target thread | |
971 | * anywhere in its code path, including while the ust_lock() is | |
972 | * held, causing a deadlock for the other thread. Let the OS | |
973 | * cleanup the threads if there are stalled in a syscall. | |
974 | */ | |
17dfb34b | 975 | lttng_ust_cleanup(1); |
2691221a | 976 | } |
e822f505 MD |
977 | |
978 | /* | |
979 | * We exclude the worker threads across fork and clone (except | |
980 | * CLONE_VM), because these system calls only keep the forking thread | |
981 | * running in the child. Therefore, we don't want to call fork or clone | |
982 | * in the middle of an tracepoint or ust tracing state modification. | |
983 | * Holding this mutex protects these structures across fork and clone. | |
984 | */ | |
b728d87e | 985 | void ust_before_fork(sigset_t *save_sigset) |
e822f505 MD |
986 | { |
987 | /* | |
988 | * Disable signals. This is to avoid that the child intervenes | |
989 | * before it is properly setup for tracing. It is safer to | |
990 | * disable all signals, because then we know we are not breaking | |
991 | * anything by restoring the original mask. | |
992 | */ | |
993 | sigset_t all_sigs; | |
994 | int ret; | |
995 | ||
996 | /* Disable signals */ | |
997 | sigfillset(&all_sigs); | |
b728d87e | 998 | ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset); |
e822f505 MD |
999 | if (ret == -1) { |
1000 | PERROR("sigprocmask"); | |
1001 | } | |
17dfb34b | 1002 | ust_lock(); |
e822f505 MD |
1003 | rcu_bp_before_fork(); |
1004 | } | |
1005 | ||
b728d87e | 1006 | static void ust_after_fork_common(sigset_t *restore_sigset) |
e822f505 MD |
1007 | { |
1008 | int ret; | |
1009 | ||
17dfb34b MD |
1010 | DBG("process %d", getpid()); |
1011 | ust_unlock(); | |
e822f505 | 1012 | /* Restore signals */ |
23c8854a | 1013 | ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL); |
e822f505 MD |
1014 | if (ret == -1) { |
1015 | PERROR("sigprocmask"); | |
1016 | } | |
1017 | } | |
1018 | ||
b728d87e | 1019 | void ust_after_fork_parent(sigset_t *restore_sigset) |
e822f505 | 1020 | { |
17dfb34b | 1021 | DBG("process %d", getpid()); |
e822f505 MD |
1022 | rcu_bp_after_fork_parent(); |
1023 | /* Release mutexes and reenable signals */ | |
b728d87e | 1024 | ust_after_fork_common(restore_sigset); |
e822f505 MD |
1025 | } |
1026 | ||
17dfb34b MD |
1027 | /* |
1028 | * After fork, in the child, we need to cleanup all the leftover state, | |
1029 | * except the worker thread which already magically disappeared thanks | |
1030 | * to the weird Linux fork semantics. After tyding up, we call | |
1031 | * lttng_ust_init() again to start over as a new PID. | |
1032 | * | |
1033 | * This is meant for forks() that have tracing in the child between the | |
1034 | * fork and following exec call (if there is any). | |
1035 | */ | |
b728d87e | 1036 | void ust_after_fork_child(sigset_t *restore_sigset) |
e822f505 | 1037 | { |
17dfb34b | 1038 | DBG("process %d", getpid()); |
e822f505 MD |
1039 | /* Release urcu mutexes */ |
1040 | rcu_bp_after_fork_child(); | |
17dfb34b | 1041 | lttng_ust_cleanup(0); |
a93bfc45 | 1042 | lttng_context_vtid_reset(); |
e822f505 | 1043 | /* Release mutexes and reenable signals */ |
b728d87e | 1044 | ust_after_fork_common(restore_sigset); |
318dfea9 | 1045 | lttng_ust_init(); |
e822f505 | 1046 | } |