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> | |
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> |
80e2814b | 38 | #include <urcu/futex.h> |
c117fb1b | 39 | #include <urcu/compiler.h> |
1ea11eab | 40 | |
4318ae1b | 41 | #include <lttng/ust-events.h> |
4318ae1b | 42 | #include <lttng/ust-abi.h> |
4318ae1b | 43 | #include <lttng/ust.h> |
7bc53e94 | 44 | #include <lttng/ust-error.h> |
44c72f10 MD |
45 | #include <ust-comm.h> |
46 | #include <usterr-signal-safe.h> | |
cd54f6d9 | 47 | #include <helper.h> |
44c72f10 | 48 | #include "tracepoint-internal.h" |
b751f722 | 49 | #include "ltt-tracer-core.h" |
08114193 | 50 | #include "compat.h" |
f645cfa7 | 51 | #include "../libringbuffer/tlsfixup.h" |
edaa1431 MD |
52 | |
53 | /* | |
54 | * Has lttng ust comm constructor been called ? | |
55 | */ | |
56 | static int initialized; | |
57 | ||
1ea11eab | 58 | /* |
17dfb34b MD |
59 | * The ust_lock/ust_unlock lock is used as a communication thread mutex. |
60 | * Held when handling a command, also held by fork() to deal with | |
61 | * removal of threads, and by exit path. | |
1ea11eab | 62 | */ |
1ea11eab MD |
63 | |
64 | /* Should the ust comm thread quit ? */ | |
65 | static int lttng_ust_comm_should_quit; | |
66 | ||
11ff9c7d MD |
67 | /* |
68 | * Wait for either of these before continuing to the main | |
69 | * program: | |
70 | * - the register_done message from sessiond daemon | |
71 | * (will let the sessiond daemon enable sessions before main | |
72 | * starts.) | |
73 | * - sessiond daemon is not reachable. | |
74 | * - timeout (ensuring applications are resilient to session | |
75 | * daemon problems). | |
76 | */ | |
77 | static sem_t constructor_wait; | |
950aab0c MD |
78 | /* |
79 | * Doing this for both the global and local sessiond. | |
80 | */ | |
95259bd0 | 81 | static int sem_count = { 2 }; |
11ff9c7d | 82 | |
e8508a49 MD |
83 | /* |
84 | * Counting nesting within lttng-ust. Used to ensure that calling fork() | |
85 | * from liblttng-ust does not execute the pre/post fork handlers. | |
86 | */ | |
87 | static int __thread lttng_ust_nest_count; | |
88 | ||
1ea11eab MD |
89 | /* |
90 | * Info about socket and associated listener thread. | |
91 | */ | |
92 | struct sock_info { | |
11ff9c7d | 93 | const char *name; |
1ea11eab | 94 | pthread_t ust_listener; /* listener thread */ |
46050b1a | 95 | int root_handle; |
8d20bf54 MD |
96 | int constructor_sem_posted; |
97 | int allowed; | |
44e073f5 | 98 | int global; |
7fc90dca MD |
99 | |
100 | char sock_path[PATH_MAX]; | |
101 | int socket; | |
102 | ||
103 | char wait_shm_path[PATH_MAX]; | |
104 | char *wait_shm_mmap; | |
1ea11eab | 105 | }; |
2691221a MD |
106 | |
107 | /* Socket from app (connect) to session daemon (listen) for communication */ | |
1ea11eab | 108 | struct sock_info global_apps = { |
11ff9c7d | 109 | .name = "global", |
44e073f5 | 110 | .global = 1, |
7fc90dca | 111 | |
46050b1a | 112 | .root_handle = -1, |
8d20bf54 | 113 | .allowed = 1, |
7fc90dca MD |
114 | |
115 | .sock_path = DEFAULT_GLOBAL_APPS_UNIX_SOCK, | |
116 | .socket = -1, | |
117 | ||
118 | .wait_shm_path = DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH, | |
1ea11eab | 119 | }; |
2691221a MD |
120 | |
121 | /* TODO: allow global_apps_sock_path override */ | |
122 | ||
1ea11eab | 123 | struct sock_info local_apps = { |
11ff9c7d | 124 | .name = "local", |
44e073f5 | 125 | .global = 0, |
46050b1a | 126 | .root_handle = -1, |
8d20bf54 | 127 | .allowed = 0, /* Check setuid bit first */ |
7fc90dca MD |
128 | |
129 | .socket = -1, | |
1ea11eab | 130 | }; |
2691221a | 131 | |
37ed587a MD |
132 | static int wait_poll_fallback; |
133 | ||
edaa1431 MD |
134 | extern void ltt_ring_buffer_client_overwrite_init(void); |
135 | extern void ltt_ring_buffer_client_discard_init(void); | |
136 | extern void ltt_ring_buffer_metadata_client_init(void); | |
137 | extern void ltt_ring_buffer_client_overwrite_exit(void); | |
138 | extern void ltt_ring_buffer_client_discard_exit(void); | |
139 | extern void ltt_ring_buffer_metadata_client_exit(void); | |
140 | ||
a903623f MD |
141 | /* |
142 | * Force a read (imply TLS fixup for dlopen) of TLS variables. | |
143 | */ | |
144 | static | |
145 | void lttng_fixup_nest_count_tls(void) | |
146 | { | |
147 | asm volatile ("" : : "m" (lttng_ust_nest_count)); | |
148 | } | |
149 | ||
2691221a | 150 | static |
8d20bf54 | 151 | int setup_local_apps(void) |
2691221a MD |
152 | { |
153 | const char *home_dir; | |
7fc90dca | 154 | uid_t uid; |
2691221a | 155 | |
7fc90dca | 156 | uid = getuid(); |
8d20bf54 MD |
157 | /* |
158 | * Disallow per-user tracing for setuid binaries. | |
159 | */ | |
7fc90dca | 160 | if (uid != geteuid()) { |
9ec6895c | 161 | assert(local_apps.allowed == 0); |
d0a1ae63 | 162 | return 0; |
8d20bf54 | 163 | } |
2691221a | 164 | home_dir = (const char *) getenv("HOME"); |
9ec6895c MD |
165 | if (!home_dir) { |
166 | WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing."); | |
167 | assert(local_apps.allowed == 0); | |
2691221a | 168 | return -ENOENT; |
9ec6895c MD |
169 | } |
170 | local_apps.allowed = 1; | |
1ea11eab | 171 | snprintf(local_apps.sock_path, PATH_MAX, |
2691221a | 172 | DEFAULT_HOME_APPS_UNIX_SOCK, home_dir); |
7fc90dca MD |
173 | snprintf(local_apps.wait_shm_path, PATH_MAX, |
174 | DEFAULT_HOME_APPS_WAIT_SHM_PATH, uid); | |
2691221a MD |
175 | return 0; |
176 | } | |
177 | ||
178 | static | |
179 | int register_app_to_sessiond(int socket) | |
180 | { | |
181 | ssize_t ret; | |
182 | struct { | |
e44418f3 MD |
183 | uint32_t major; |
184 | uint32_t minor; | |
2691221a | 185 | pid_t pid; |
5c33bde8 | 186 | pid_t ppid; |
2691221a | 187 | uid_t uid; |
83610856 | 188 | gid_t gid; |
c117fb1b | 189 | uint32_t bits_per_long; |
2629549e | 190 | char name[16]; /* process name */ |
2691221a MD |
191 | } reg_msg; |
192 | ||
e44418f3 MD |
193 | reg_msg.major = LTTNG_UST_COMM_VERSION_MAJOR; |
194 | reg_msg.minor = LTTNG_UST_COMM_VERSION_MINOR; | |
2691221a | 195 | reg_msg.pid = getpid(); |
5c33bde8 | 196 | reg_msg.ppid = getppid(); |
2691221a | 197 | reg_msg.uid = getuid(); |
83610856 | 198 | reg_msg.gid = getgid(); |
c117fb1b | 199 | reg_msg.bits_per_long = CAA_BITS_PER_LONG; |
08114193 | 200 | lttng_ust_getprocname(reg_msg.name); |
2691221a | 201 | |
57773204 | 202 | ret = ustcomm_send_unix_sock(socket, ®_msg, sizeof(reg_msg)); |
2691221a MD |
203 | if (ret >= 0 && ret != sizeof(reg_msg)) |
204 | return -EIO; | |
205 | return ret; | |
206 | } | |
207 | ||
d9e99d10 | 208 | static |
57773204 | 209 | int send_reply(int sock, struct ustcomm_ust_reply *lur) |
d9e99d10 | 210 | { |
9eb62b9c | 211 | ssize_t len; |
d3a492d1 | 212 | |
57773204 | 213 | len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur)); |
d3a492d1 | 214 | switch (len) { |
a4be8962 | 215 | case sizeof(*lur): |
d3a492d1 MD |
216 | DBG("message successfully sent"); |
217 | return 0; | |
7bc53e94 MD |
218 | default: |
219 | if (len == -ECONNRESET) { | |
220 | DBG("remote end closed connection"); | |
d3a492d1 MD |
221 | return 0; |
222 | } | |
7bc53e94 MD |
223 | if (len < 0) |
224 | return len; | |
225 | DBG("incorrect message size: %zd", len); | |
226 | return -EINVAL; | |
d3a492d1 MD |
227 | } |
228 | } | |
229 | ||
230 | static | |
edaa1431 | 231 | int handle_register_done(struct sock_info *sock_info) |
11ff9c7d MD |
232 | { |
233 | int ret; | |
234 | ||
edaa1431 MD |
235 | if (sock_info->constructor_sem_posted) |
236 | return 0; | |
237 | sock_info->constructor_sem_posted = 1; | |
56cd7e2f MD |
238 | if (uatomic_read(&sem_count) <= 0) { |
239 | return 0; | |
240 | } | |
95259bd0 MD |
241 | ret = uatomic_add_return(&sem_count, -1); |
242 | if (ret == 0) { | |
243 | ret = sem_post(&constructor_wait); | |
244 | assert(!ret); | |
245 | } | |
11ff9c7d MD |
246 | return 0; |
247 | } | |
248 | ||
249 | static | |
250 | int handle_message(struct sock_info *sock_info, | |
57773204 | 251 | int sock, struct ustcomm_ust_msg *lum) |
d3a492d1 | 252 | { |
1ea11eab | 253 | int ret = 0; |
b61ce3b2 | 254 | const struct lttng_ust_objd_ops *ops; |
57773204 | 255 | struct ustcomm_ust_reply lur; |
193183fb | 256 | int shm_fd, wait_fd; |
ef9ff354 | 257 | union ust_args args; |
40003310 | 258 | ssize_t len; |
1ea11eab | 259 | |
17dfb34b | 260 | ust_lock(); |
1ea11eab | 261 | |
46050b1a MD |
262 | memset(&lur, 0, sizeof(lur)); |
263 | ||
1ea11eab | 264 | if (lttng_ust_comm_should_quit) { |
46050b1a | 265 | ret = -EPERM; |
1ea11eab MD |
266 | goto end; |
267 | } | |
9eb62b9c | 268 | |
46050b1a MD |
269 | ops = objd_ops(lum->handle); |
270 | if (!ops) { | |
271 | ret = -ENOENT; | |
272 | goto end; | |
1ea11eab | 273 | } |
46050b1a MD |
274 | |
275 | switch (lum->cmd) { | |
11ff9c7d MD |
276 | case LTTNG_UST_REGISTER_DONE: |
277 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
edaa1431 | 278 | ret = handle_register_done(sock_info); |
11ff9c7d MD |
279 | else |
280 | ret = -EINVAL; | |
281 | break; | |
46050b1a MD |
282 | case LTTNG_UST_RELEASE: |
283 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
284 | ret = -EPERM; | |
285 | else | |
d4419b81 | 286 | ret = lttng_ust_objd_unref(lum->handle); |
d9e99d10 | 287 | break; |
2d78951a MD |
288 | case LTTNG_UST_FILTER: |
289 | { | |
290 | /* Receive filter data */ | |
cd54f6d9 | 291 | struct lttng_ust_filter_bytecode *bytecode; |
2d78951a | 292 | |
cd54f6d9 | 293 | if (lum->u.filter.data_size > FILTER_BYTECODE_MAX_LEN) { |
7bc53e94 | 294 | ERR("Filter data size is too large: %u bytes", |
2d78951a MD |
295 | lum->u.filter.data_size); |
296 | ret = -EINVAL; | |
297 | goto error; | |
298 | } | |
2734ca65 | 299 | |
885b1dfd | 300 | if (lum->u.filter.reloc_offset > lum->u.filter.data_size) { |
7bc53e94 | 301 | ERR("Filter reloc offset %u is not within data", |
2734ca65 CB |
302 | lum->u.filter.reloc_offset); |
303 | ret = -EINVAL; | |
304 | goto error; | |
305 | } | |
306 | ||
cd54f6d9 MD |
307 | bytecode = zmalloc(sizeof(*bytecode) + lum->u.filter.data_size); |
308 | if (!bytecode) { | |
309 | ret = -ENOMEM; | |
310 | goto error; | |
311 | } | |
312 | len = ustcomm_recv_unix_sock(sock, bytecode->data, | |
2d78951a MD |
313 | lum->u.filter.data_size); |
314 | switch (len) { | |
315 | case 0: /* orderly shutdown */ | |
316 | ret = 0; | |
cd54f6d9 | 317 | free(bytecode); |
2d78951a | 318 | goto error; |
2d78951a MD |
319 | default: |
320 | if (len == lum->u.filter.data_size) { | |
7bc53e94 | 321 | DBG("filter data received"); |
2d78951a | 322 | break; |
7bc53e94 MD |
323 | } else if (len < 0) { |
324 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
325 | if (len == -ECONNRESET) { | |
326 | ERR("%s remote end closed connection", sock_info->name); | |
327 | ret = len; | |
328 | free(bytecode); | |
329 | goto error; | |
330 | } | |
331 | ret = len; | |
332 | goto end; | |
2d78951a | 333 | } else { |
7bc53e94 | 334 | DBG("incorrect filter data message size: %zd", len); |
2d78951a | 335 | ret = -EINVAL; |
cd54f6d9 | 336 | free(bytecode); |
2d78951a MD |
337 | goto end; |
338 | } | |
339 | } | |
cd54f6d9 MD |
340 | bytecode->len = lum->u.filter.data_size; |
341 | bytecode->reloc_offset = lum->u.filter.reloc_offset; | |
cd54f6d9 | 342 | if (ops->cmd) { |
2d78951a | 343 | ret = ops->cmd(lum->handle, lum->cmd, |
cd54f6d9 | 344 | (unsigned long) bytecode, |
f59ed768 | 345 | &args, sock_info); |
cd54f6d9 MD |
346 | if (ret) { |
347 | free(bytecode); | |
348 | } | |
349 | /* don't free bytecode if everything went fine. */ | |
350 | } else { | |
2d78951a | 351 | ret = -ENOSYS; |
cd54f6d9 MD |
352 | free(bytecode); |
353 | } | |
2d78951a MD |
354 | break; |
355 | } | |
d9e99d10 | 356 | default: |
46050b1a MD |
357 | if (ops->cmd) |
358 | ret = ops->cmd(lum->handle, lum->cmd, | |
ef9ff354 | 359 | (unsigned long) &lum->u, |
f59ed768 | 360 | &args, sock_info); |
46050b1a MD |
361 | else |
362 | ret = -ENOSYS; | |
363 | break; | |
d9e99d10 | 364 | } |
46050b1a | 365 | |
1ea11eab | 366 | end: |
46050b1a MD |
367 | lur.handle = lum->handle; |
368 | lur.cmd = lum->cmd; | |
369 | lur.ret_val = ret; | |
370 | if (ret >= 0) { | |
7bc53e94 | 371 | lur.ret_code = LTTNG_UST_OK; |
46050b1a | 372 | } else { |
7bc53e94 MD |
373 | /* |
374 | * Use -LTTNG_UST_ERR as wildcard for UST internal | |
375 | * error that are not caused by the transport, except if | |
376 | * we already have a more precise error message to | |
377 | * report. | |
378 | */ | |
64b2564e DG |
379 | if (ret > -LTTNG_UST_ERR) { |
380 | /* Translate code to UST error. */ | |
381 | switch (ret) { | |
382 | case -EEXIST: | |
383 | lur.ret_code = -LTTNG_UST_ERR_EXIST; | |
384 | break; | |
385 | case -EINVAL: | |
386 | lur.ret_code = -LTTNG_UST_ERR_INVAL; | |
387 | break; | |
388 | case -ENOENT: | |
389 | lur.ret_code = -LTTNG_UST_ERR_NOENT; | |
390 | break; | |
391 | case -EPERM: | |
392 | lur.ret_code = -LTTNG_UST_ERR_PERM; | |
393 | break; | |
394 | case -ENOSYS: | |
395 | lur.ret_code = -LTTNG_UST_ERR_NOSYS; | |
396 | break; | |
397 | default: | |
398 | lur.ret_code = -LTTNG_UST_ERR; | |
399 | break; | |
400 | } | |
401 | } else { | |
7bc53e94 | 402 | lur.ret_code = ret; |
64b2564e | 403 | } |
46050b1a | 404 | } |
e6ea14c5 MD |
405 | if (ret >= 0) { |
406 | switch (lum->cmd) { | |
407 | case LTTNG_UST_STREAM: | |
408 | /* | |
409 | * Special-case reply to send stream info. | |
410 | * Use lum.u output. | |
411 | */ | |
412 | lur.u.stream.memory_map_size = *args.stream.memory_map_size; | |
413 | shm_fd = *args.stream.shm_fd; | |
414 | wait_fd = *args.stream.wait_fd; | |
415 | break; | |
416 | case LTTNG_UST_METADATA: | |
417 | case LTTNG_UST_CHANNEL: | |
418 | lur.u.channel.memory_map_size = *args.channel.memory_map_size; | |
419 | shm_fd = *args.channel.shm_fd; | |
420 | wait_fd = *args.channel.wait_fd; | |
421 | break; | |
422 | case LTTNG_UST_TRACER_VERSION: | |
423 | lur.u.version = lum->u.version; | |
424 | break; | |
425 | case LTTNG_UST_TRACEPOINT_LIST_GET: | |
426 | memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint)); | |
427 | break; | |
428 | } | |
381c0f1e | 429 | } |
46050b1a | 430 | ret = send_reply(sock, &lur); |
193183fb | 431 | if (ret < 0) { |
7bc53e94 | 432 | DBG("error sending reply"); |
193183fb MD |
433 | goto error; |
434 | } | |
46050b1a | 435 | |
824f40b8 MD |
436 | if ((lum->cmd == LTTNG_UST_STREAM |
437 | || lum->cmd == LTTNG_UST_CHANNEL | |
438 | || lum->cmd == LTTNG_UST_METADATA) | |
7bc53e94 | 439 | && lur.ret_code == LTTNG_UST_OK) { |
50be2fc4 MD |
440 | int sendret = 0; |
441 | ||
381c0f1e | 442 | /* we also need to send the file descriptors. */ |
57773204 | 443 | ret = ustcomm_send_fds_unix_sock(sock, |
193183fb | 444 | &shm_fd, &shm_fd, |
381c0f1e MD |
445 | 1, sizeof(int)); |
446 | if (ret < 0) { | |
7bc53e94 | 447 | ERR("send shm_fd"); |
50be2fc4 | 448 | sendret = ret; |
381c0f1e | 449 | } |
50be2fc4 MD |
450 | /* |
451 | * The sessiond expects 2 file descriptors, even upon | |
452 | * error. | |
453 | */ | |
57773204 | 454 | ret = ustcomm_send_fds_unix_sock(sock, |
193183fb | 455 | &wait_fd, &wait_fd, |
381c0f1e MD |
456 | 1, sizeof(int)); |
457 | if (ret < 0) { | |
458 | perror("send wait_fd"); | |
459 | goto error; | |
460 | } | |
50be2fc4 MD |
461 | if (sendret) { |
462 | ret = sendret; | |
463 | goto error; | |
464 | } | |
381c0f1e | 465 | } |
40003310 MD |
466 | /* |
467 | * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field | |
468 | * after the reply. | |
469 | */ | |
7bc53e94 | 470 | if (lur.ret_code == LTTNG_UST_OK) { |
40003310 MD |
471 | switch (lum->cmd) { |
472 | case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET: | |
473 | len = ustcomm_send_unix_sock(sock, | |
474 | &args.field_list.entry, | |
475 | sizeof(args.field_list.entry)); | |
7bc53e94 MD |
476 | if (len < 0) { |
477 | ret = len; | |
478 | goto error; | |
479 | } | |
40003310 | 480 | if (len != sizeof(args.field_list.entry)) { |
7bc53e94 | 481 | ret = -EINVAL; |
40003310 MD |
482 | goto error; |
483 | } | |
484 | } | |
485 | } | |
ef9ff354 MD |
486 | /* |
487 | * We still have the memory map reference, and the fds have been | |
3ad5f707 MD |
488 | * sent to the sessiond. We can therefore close those fds. Note |
489 | * that we keep the write side of the wait_fd open, but close | |
490 | * the read side. | |
ef9ff354 | 491 | */ |
7bc53e94 | 492 | if (lur.ret_code == LTTNG_UST_OK) { |
ef9ff354 MD |
493 | switch (lum->cmd) { |
494 | case LTTNG_UST_STREAM: | |
495 | if (shm_fd >= 0) { | |
496 | ret = close(shm_fd); | |
497 | if (ret) { | |
498 | PERROR("Error closing stream shm_fd"); | |
499 | } | |
500 | *args.stream.shm_fd = -1; | |
501 | } | |
502 | if (wait_fd >= 0) { | |
503 | ret = close(wait_fd); | |
504 | if (ret) { | |
505 | PERROR("Error closing stream wait_fd"); | |
506 | } | |
507 | *args.stream.wait_fd = -1; | |
508 | } | |
509 | break; | |
510 | case LTTNG_UST_METADATA: | |
511 | case LTTNG_UST_CHANNEL: | |
512 | if (shm_fd >= 0) { | |
513 | ret = close(shm_fd); | |
514 | if (ret) { | |
515 | PERROR("Error closing channel shm_fd"); | |
516 | } | |
517 | *args.channel.shm_fd = -1; | |
518 | } | |
519 | if (wait_fd >= 0) { | |
520 | ret = close(wait_fd); | |
521 | if (ret) { | |
522 | PERROR("Error closing channel wait_fd"); | |
523 | } | |
524 | *args.channel.wait_fd = -1; | |
525 | } | |
526 | break; | |
527 | } | |
528 | } | |
529 | ||
381c0f1e | 530 | error: |
17dfb34b | 531 | ust_unlock(); |
1ea11eab | 532 | return ret; |
d9e99d10 MD |
533 | } |
534 | ||
46050b1a | 535 | static |
efe0de09 | 536 | void cleanup_sock_info(struct sock_info *sock_info, int exiting) |
46050b1a MD |
537 | { |
538 | int ret; | |
539 | ||
540 | if (sock_info->socket != -1) { | |
e6973a89 | 541 | ret = ustcomm_close_unix_sock(sock_info->socket); |
46050b1a | 542 | if (ret) { |
7fc90dca | 543 | ERR("Error closing apps socket"); |
46050b1a MD |
544 | } |
545 | sock_info->socket = -1; | |
546 | } | |
547 | if (sock_info->root_handle != -1) { | |
d4419b81 | 548 | ret = lttng_ust_objd_unref(sock_info->root_handle); |
46050b1a MD |
549 | if (ret) { |
550 | ERR("Error unref root handle"); | |
551 | } | |
552 | sock_info->root_handle = -1; | |
553 | } | |
318dfea9 | 554 | sock_info->constructor_sem_posted = 0; |
efe0de09 MD |
555 | /* |
556 | * wait_shm_mmap is used by listener threads outside of the | |
557 | * ust lock, so we cannot tear it down ourselves, because we | |
558 | * cannot join on these threads. Leave this task to the OS | |
559 | * process exit. | |
560 | */ | |
561 | if (!exiting && sock_info->wait_shm_mmap) { | |
7fc90dca MD |
562 | ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE)); |
563 | if (ret) { | |
564 | ERR("Error unmapping wait shm"); | |
565 | } | |
566 | sock_info->wait_shm_mmap = NULL; | |
567 | } | |
568 | } | |
569 | ||
58d4b2a2 | 570 | /* |
33bbeb90 MD |
571 | * Using fork to set umask in the child process (not multi-thread safe). |
572 | * We deal with the shm_open vs ftruncate race (happening when the | |
573 | * sessiond owns the shm and does not let everybody modify it, to ensure | |
574 | * safety against shm_unlink) by simply letting the mmap fail and | |
575 | * retrying after a few seconds. | |
576 | * For global shm, everybody has rw access to it until the sessiond | |
577 | * starts. | |
58d4b2a2 | 578 | */ |
7fc90dca | 579 | static |
58d4b2a2 | 580 | int get_wait_shm(struct sock_info *sock_info, size_t mmap_size) |
7fc90dca | 581 | { |
7fc90dca | 582 | int wait_shm_fd, ret; |
58d4b2a2 | 583 | pid_t pid; |
44e073f5 | 584 | |
58d4b2a2 | 585 | /* |
33bbeb90 | 586 | * Try to open read-only. |
58d4b2a2 | 587 | */ |
33bbeb90 | 588 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 MD |
589 | if (wait_shm_fd >= 0) { |
590 | goto end; | |
591 | } else if (wait_shm_fd < 0 && errno != ENOENT) { | |
592 | /* | |
33bbeb90 MD |
593 | * Real-only open did not work, and it's not because the |
594 | * entry was not present. It's a failure that prohibits | |
595 | * using shm. | |
58d4b2a2 | 596 | */ |
7fc90dca | 597 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
58d4b2a2 | 598 | goto end; |
7fc90dca MD |
599 | } |
600 | /* | |
58d4b2a2 MD |
601 | * If the open failed because the file did not exist, try |
602 | * creating it ourself. | |
7fc90dca | 603 | */ |
e8508a49 | 604 | lttng_ust_nest_count++; |
58d4b2a2 | 605 | pid = fork(); |
e8508a49 | 606 | lttng_ust_nest_count--; |
58d4b2a2 MD |
607 | if (pid > 0) { |
608 | int status; | |
609 | ||
610 | /* | |
611 | * Parent: wait for child to return, in which case the | |
612 | * shared memory map will have been created. | |
613 | */ | |
614 | pid = wait(&status); | |
b7d3cb32 | 615 | if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) { |
58d4b2a2 MD |
616 | wait_shm_fd = -1; |
617 | goto end; | |
7fc90dca | 618 | } |
58d4b2a2 MD |
619 | /* |
620 | * Try to open read-only again after creation. | |
621 | */ | |
33bbeb90 | 622 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 MD |
623 | if (wait_shm_fd < 0) { |
624 | /* | |
625 | * Real-only open did not work. It's a failure | |
626 | * that prohibits using shm. | |
627 | */ | |
628 | ERR("Error opening shm %s", sock_info->wait_shm_path); | |
629 | goto end; | |
630 | } | |
631 | goto end; | |
632 | } else if (pid == 0) { | |
633 | int create_mode; | |
634 | ||
635 | /* Child */ | |
33bbeb90 | 636 | create_mode = S_IRUSR | S_IWUSR | S_IRGRP; |
58d4b2a2 | 637 | if (sock_info->global) |
33bbeb90 | 638 | create_mode |= S_IROTH | S_IWGRP | S_IWOTH; |
58d4b2a2 MD |
639 | /* |
640 | * We're alone in a child process, so we can modify the | |
641 | * process-wide umask. | |
642 | */ | |
33bbeb90 | 643 | umask(~create_mode); |
58d4b2a2 | 644 | /* |
33bbeb90 MD |
645 | * Try creating shm (or get rw access). |
646 | * We don't do an exclusive open, because we allow other | |
647 | * processes to create+ftruncate it concurrently. | |
58d4b2a2 MD |
648 | */ |
649 | wait_shm_fd = shm_open(sock_info->wait_shm_path, | |
650 | O_RDWR | O_CREAT, create_mode); | |
651 | if (wait_shm_fd >= 0) { | |
652 | ret = ftruncate(wait_shm_fd, mmap_size); | |
653 | if (ret) { | |
654 | PERROR("ftruncate"); | |
b0c1425d | 655 | _exit(EXIT_FAILURE); |
58d4b2a2 | 656 | } |
b0c1425d | 657 | _exit(EXIT_SUCCESS); |
58d4b2a2 | 658 | } |
33bbeb90 MD |
659 | /* |
660 | * For local shm, we need to have rw access to accept | |
661 | * opening it: this means the local sessiond will be | |
662 | * able to wake us up. For global shm, we open it even | |
663 | * if rw access is not granted, because the root.root | |
664 | * sessiond will be able to override all rights and wake | |
665 | * us up. | |
666 | */ | |
667 | if (!sock_info->global && errno != EACCES) { | |
58d4b2a2 | 668 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
5d3bc5ed | 669 | _exit(EXIT_FAILURE); |
58d4b2a2 MD |
670 | } |
671 | /* | |
33bbeb90 MD |
672 | * The shm exists, but we cannot open it RW. Report |
673 | * success. | |
58d4b2a2 | 674 | */ |
5d3bc5ed | 675 | _exit(EXIT_SUCCESS); |
58d4b2a2 MD |
676 | } else { |
677 | return -1; | |
7fc90dca | 678 | } |
58d4b2a2 | 679 | end: |
33bbeb90 MD |
680 | if (wait_shm_fd >= 0 && !sock_info->global) { |
681 | struct stat statbuf; | |
682 | ||
683 | /* | |
684 | * Ensure that our user is the owner of the shm file for | |
685 | * local shm. If we do not own the file, it means our | |
686 | * sessiond will not have access to wake us up (there is | |
687 | * probably a rogue process trying to fake our | |
688 | * sessiond). Fallback to polling method in this case. | |
689 | */ | |
690 | ret = fstat(wait_shm_fd, &statbuf); | |
691 | if (ret) { | |
692 | PERROR("fstat"); | |
693 | goto error_close; | |
694 | } | |
695 | if (statbuf.st_uid != getuid()) | |
696 | goto error_close; | |
697 | } | |
58d4b2a2 | 698 | return wait_shm_fd; |
33bbeb90 MD |
699 | |
700 | error_close: | |
701 | ret = close(wait_shm_fd); | |
702 | if (ret) { | |
703 | PERROR("Error closing fd"); | |
704 | } | |
705 | return -1; | |
58d4b2a2 MD |
706 | } |
707 | ||
708 | static | |
709 | char *get_map_shm(struct sock_info *sock_info) | |
710 | { | |
711 | size_t mmap_size = sysconf(_SC_PAGE_SIZE); | |
712 | int wait_shm_fd, ret; | |
713 | char *wait_shm_mmap; | |
714 | ||
715 | wait_shm_fd = get_wait_shm(sock_info, mmap_size); | |
716 | if (wait_shm_fd < 0) { | |
717 | goto error; | |
44e073f5 | 718 | } |
7fc90dca MD |
719 | wait_shm_mmap = mmap(NULL, mmap_size, PROT_READ, |
720 | MAP_SHARED, wait_shm_fd, 0); | |
7fc90dca MD |
721 | /* close shm fd immediately after taking the mmap reference */ |
722 | ret = close(wait_shm_fd); | |
723 | if (ret) { | |
33bbeb90 MD |
724 | PERROR("Error closing fd"); |
725 | } | |
726 | if (wait_shm_mmap == MAP_FAILED) { | |
727 | DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode."); | |
728 | goto error; | |
7fc90dca MD |
729 | } |
730 | return wait_shm_mmap; | |
731 | ||
732 | error: | |
733 | return NULL; | |
734 | } | |
735 | ||
736 | static | |
737 | void wait_for_sessiond(struct sock_info *sock_info) | |
738 | { | |
efe0de09 | 739 | int ret; |
80e2814b | 740 | |
7fc90dca MD |
741 | ust_lock(); |
742 | if (lttng_ust_comm_should_quit) { | |
743 | goto quit; | |
744 | } | |
37ed587a MD |
745 | if (wait_poll_fallback) { |
746 | goto error; | |
747 | } | |
7fc90dca MD |
748 | if (!sock_info->wait_shm_mmap) { |
749 | sock_info->wait_shm_mmap = get_map_shm(sock_info); | |
750 | if (!sock_info->wait_shm_mmap) | |
751 | goto error; | |
752 | } | |
753 | ust_unlock(); | |
754 | ||
755 | DBG("Waiting for %s apps sessiond", sock_info->name); | |
80e2814b MD |
756 | /* Wait for futex wakeup */ |
757 | if (uatomic_read((int32_t *) sock_info->wait_shm_mmap) == 0) { | |
758 | ret = futex_async((int32_t *) sock_info->wait_shm_mmap, | |
759 | FUTEX_WAIT, 0, NULL, NULL, 0); | |
80e2814b | 760 | if (ret < 0) { |
37ed587a MD |
761 | if (errno == EFAULT) { |
762 | wait_poll_fallback = 1; | |
a8b870ad | 763 | DBG( |
37ed587a MD |
764 | "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) " |
765 | "do not support FUTEX_WAKE on read-only memory mappings correctly. " | |
766 | "Please upgrade your kernel " | |
767 | "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel " | |
768 | "mainline). LTTng-UST will use polling mode fallback."); | |
cd27263b MD |
769 | if (ust_debug()) |
770 | PERROR("futex"); | |
37ed587a | 771 | } |
80e2814b MD |
772 | } |
773 | } | |
7fc90dca MD |
774 | return; |
775 | ||
776 | quit: | |
777 | ust_unlock(); | |
778 | return; | |
779 | ||
780 | error: | |
781 | ust_unlock(); | |
7fc90dca | 782 | return; |
46050b1a MD |
783 | } |
784 | ||
1ea11eab MD |
785 | /* |
786 | * This thread does not allocate any resource, except within | |
787 | * handle_message, within mutex protection. This mutex protects against | |
788 | * fork and exit. | |
98bf993f | 789 | * The other moment it allocates resources is at socket connection, which |
1ea11eab MD |
790 | * is also protected by the mutex. |
791 | */ | |
d9e99d10 MD |
792 | static |
793 | void *ust_listener_thread(void *arg) | |
794 | { | |
1ea11eab | 795 | struct sock_info *sock_info = arg; |
c0eedf81 | 796 | int sock, ret, prev_connect_failed = 0, has_waited = 0; |
d9e99d10 | 797 | |
9eb62b9c MD |
798 | /* Restart trying to connect to the session daemon */ |
799 | restart: | |
c0eedf81 MD |
800 | if (prev_connect_failed) { |
801 | /* Wait for sessiond availability with pipe */ | |
802 | wait_for_sessiond(sock_info); | |
803 | if (has_waited) { | |
804 | has_waited = 0; | |
805 | /* | |
806 | * Sleep for 5 seconds before retrying after a | |
807 | * sequence of failure / wait / failure. This | |
808 | * deals with a killed or broken session daemon. | |
809 | */ | |
810 | sleep(5); | |
811 | } | |
812 | has_waited = 1; | |
813 | prev_connect_failed = 0; | |
814 | } | |
17dfb34b | 815 | ust_lock(); |
1ea11eab MD |
816 | |
817 | if (lttng_ust_comm_should_quit) { | |
17dfb34b | 818 | ust_unlock(); |
1ea11eab MD |
819 | goto quit; |
820 | } | |
9eb62b9c | 821 | |
1ea11eab | 822 | if (sock_info->socket != -1) { |
e6973a89 | 823 | ret = ustcomm_close_unix_sock(sock_info->socket); |
1ea11eab | 824 | if (ret) { |
11ff9c7d | 825 | ERR("Error closing %s apps socket", sock_info->name); |
1ea11eab MD |
826 | } |
827 | sock_info->socket = -1; | |
828 | } | |
46050b1a | 829 | |
9eb62b9c | 830 | /* Register */ |
57773204 | 831 | ret = ustcomm_connect_unix_sock(sock_info->sock_path); |
9eb62b9c | 832 | if (ret < 0) { |
4d3c9523 | 833 | DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name); |
c0eedf81 | 834 | prev_connect_failed = 1; |
11ff9c7d MD |
835 | /* |
836 | * If we cannot find the sessiond daemon, don't delay | |
837 | * constructor execution. | |
838 | */ | |
edaa1431 | 839 | ret = handle_register_done(sock_info); |
11ff9c7d | 840 | assert(!ret); |
17dfb34b | 841 | ust_unlock(); |
1ea11eab | 842 | goto restart; |
46050b1a MD |
843 | } |
844 | ||
845 | sock_info->socket = sock = ret; | |
846 | ||
847 | /* | |
848 | * Create only one root handle per listener thread for the whole | |
f59ed768 MD |
849 | * process lifetime, so we ensure we get ID which is statically |
850 | * assigned to the root handle. | |
46050b1a MD |
851 | */ |
852 | if (sock_info->root_handle == -1) { | |
853 | ret = lttng_abi_create_root_handle(); | |
a51070bb | 854 | if (ret < 0) { |
46050b1a | 855 | ERR("Error creating root handle"); |
17dfb34b | 856 | ust_unlock(); |
46050b1a MD |
857 | goto quit; |
858 | } | |
859 | sock_info->root_handle = ret; | |
9eb62b9c | 860 | } |
1ea11eab | 861 | |
9eb62b9c MD |
862 | ret = register_app_to_sessiond(sock); |
863 | if (ret < 0) { | |
11ff9c7d | 864 | ERR("Error registering to %s apps socket", sock_info->name); |
c0eedf81 | 865 | prev_connect_failed = 1; |
11ff9c7d MD |
866 | /* |
867 | * If we cannot register to the sessiond daemon, don't | |
868 | * delay constructor execution. | |
869 | */ | |
edaa1431 | 870 | ret = handle_register_done(sock_info); |
11ff9c7d | 871 | assert(!ret); |
17dfb34b | 872 | ust_unlock(); |
9eb62b9c MD |
873 | goto restart; |
874 | } | |
17dfb34b | 875 | ust_unlock(); |
46050b1a | 876 | |
d9e99d10 MD |
877 | for (;;) { |
878 | ssize_t len; | |
57773204 | 879 | struct ustcomm_ust_msg lum; |
d9e99d10 | 880 | |
57773204 | 881 | len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum)); |
d9e99d10 MD |
882 | switch (len) { |
883 | case 0: /* orderly shutdown */ | |
7bc53e94 | 884 | DBG("%s ltt-sessiond has performed an orderly shutdown", sock_info->name); |
8236ba10 MD |
885 | ust_lock(); |
886 | /* | |
887 | * Either sessiond has shutdown or refused us by closing the socket. | |
888 | * In either case, we don't want to delay construction execution, | |
889 | * and we need to wait before retry. | |
890 | */ | |
891 | prev_connect_failed = 1; | |
892 | /* | |
893 | * If we cannot register to the sessiond daemon, don't | |
894 | * delay constructor execution. | |
895 | */ | |
896 | ret = handle_register_done(sock_info); | |
897 | assert(!ret); | |
898 | ust_unlock(); | |
d9e99d10 | 899 | goto end; |
e7723462 | 900 | case sizeof(lum): |
7bc53e94 | 901 | DBG("message received"); |
11ff9c7d | 902 | ret = handle_message(sock_info, sock, &lum); |
7bc53e94 | 903 | if (ret) { |
11ff9c7d | 904 | ERR("Error handling message for %s socket", sock_info->name); |
d9e99d10 MD |
905 | } |
906 | continue; | |
7bc53e94 MD |
907 | default: |
908 | if (len < 0) { | |
909 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
910 | } else { | |
911 | DBG("incorrect message size (%s socket): %zd", sock_info->name, len); | |
912 | } | |
913 | if (len == -ECONNRESET) { | |
914 | DBG("%s remote end closed connection", sock_info->name); | |
d9e99d10 MD |
915 | goto end; |
916 | } | |
917 | goto end; | |
d9e99d10 MD |
918 | } |
919 | ||
920 | } | |
921 | end: | |
f59ed768 MD |
922 | ust_lock(); |
923 | /* Cleanup socket handles before trying to reconnect */ | |
924 | lttng_ust_objd_table_owner_cleanup(sock_info); | |
925 | ust_unlock(); | |
9eb62b9c | 926 | goto restart; /* try to reconnect */ |
1ea11eab | 927 | quit: |
d9e99d10 MD |
928 | return NULL; |
929 | } | |
930 | ||
cf12a773 MD |
931 | /* |
932 | * Return values: -1: don't wait. 0: wait forever. 1: timeout wait. | |
933 | */ | |
11ff9c7d MD |
934 | static |
935 | int get_timeout(struct timespec *constructor_timeout) | |
936 | { | |
cf12a773 MD |
937 | long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS; |
938 | char *str_delay; | |
11ff9c7d MD |
939 | int ret; |
940 | ||
69400ac4 | 941 | str_delay = getenv("LTTNG_UST_REGISTER_TIMEOUT"); |
cf12a773 MD |
942 | if (str_delay) { |
943 | constructor_delay_ms = strtol(str_delay, NULL, 10); | |
944 | } | |
945 | ||
946 | switch (constructor_delay_ms) { | |
947 | case -1:/* fall-through */ | |
948 | case 0: | |
949 | return constructor_delay_ms; | |
950 | default: | |
951 | break; | |
952 | } | |
953 | ||
954 | /* | |
955 | * If we are unable to find the current time, don't wait. | |
956 | */ | |
957 | ret = clock_gettime(CLOCK_REALTIME, constructor_timeout); | |
958 | if (ret) { | |
959 | return -1; | |
960 | } | |
95259bd0 MD |
961 | constructor_timeout->tv_sec += constructor_delay_ms / 1000UL; |
962 | constructor_timeout->tv_nsec += | |
963 | (constructor_delay_ms % 1000UL) * 1000000UL; | |
11ff9c7d MD |
964 | if (constructor_timeout->tv_nsec >= 1000000000UL) { |
965 | constructor_timeout->tv_sec++; | |
966 | constructor_timeout->tv_nsec -= 1000000000UL; | |
967 | } | |
cf12a773 | 968 | return 1; |
11ff9c7d | 969 | } |
d9e99d10 | 970 | |
2691221a MD |
971 | /* |
972 | * sessiond monitoring thread: monitor presence of global and per-user | |
973 | * sessiond by polling the application common named pipe. | |
974 | */ | |
edaa1431 | 975 | void __attribute__((constructor)) lttng_ust_init(void) |
2691221a | 976 | { |
11ff9c7d | 977 | struct timespec constructor_timeout; |
ae6a58bf | 978 | sigset_t sig_all_blocked, orig_parent_mask; |
1879f67f | 979 | pthread_attr_t thread_attr; |
cf12a773 | 980 | int timeout_mode; |
2691221a MD |
981 | int ret; |
982 | ||
edaa1431 MD |
983 | if (uatomic_xchg(&initialized, 1) == 1) |
984 | return; | |
985 | ||
eddd8d5d MD |
986 | /* |
987 | * Fixup interdependency between TLS fixup mutex (which happens | |
988 | * to be the dynamic linker mutex) and ust_lock, taken within | |
989 | * the ust lock. | |
990 | */ | |
991 | lttng_fixup_event_tls(); | |
f645cfa7 | 992 | lttng_fixup_ringbuffer_tls(); |
4158a15a | 993 | lttng_fixup_vtid_tls(); |
a903623f | 994 | lttng_fixup_nest_count_tls(); |
009745db | 995 | lttng_fixup_procname_tls(); |
eddd8d5d | 996 | |
edaa1431 MD |
997 | /* |
998 | * We want precise control over the order in which we construct | |
999 | * our sub-libraries vs starting to receive commands from | |
1000 | * sessiond (otherwise leading to errors when trying to create | |
1001 | * sessiond before the init functions are completed). | |
1002 | */ | |
2691221a | 1003 | init_usterr(); |
edaa1431 MD |
1004 | init_tracepoint(); |
1005 | ltt_ring_buffer_metadata_client_init(); | |
1006 | ltt_ring_buffer_client_overwrite_init(); | |
1007 | ltt_ring_buffer_client_discard_init(); | |
2691221a | 1008 | |
cf12a773 | 1009 | timeout_mode = get_timeout(&constructor_timeout); |
11ff9c7d | 1010 | |
95259bd0 | 1011 | ret = sem_init(&constructor_wait, 0, 0); |
11ff9c7d MD |
1012 | assert(!ret); |
1013 | ||
8d20bf54 | 1014 | ret = setup_local_apps(); |
2691221a | 1015 | if (ret) { |
9ec6895c | 1016 | DBG("local apps setup returned %d", ret); |
2691221a | 1017 | } |
ae6a58bf WP |
1018 | |
1019 | /* A new thread created by pthread_create inherits the signal mask | |
1020 | * from the parent. To avoid any signal being received by the | |
1021 | * listener thread, we block all signals temporarily in the parent, | |
1022 | * while we create the listener thread. | |
1023 | */ | |
1024 | sigfillset(&sig_all_blocked); | |
1025 | ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask); | |
1026 | if (ret) { | |
d94d802c | 1027 | ERR("pthread_sigmask: %s", strerror(ret)); |
ae6a58bf WP |
1028 | } |
1029 | ||
1879f67f MG |
1030 | ret = pthread_attr_init(&thread_attr); |
1031 | if (ret) { | |
1032 | ERR("pthread_attr_init: %s", strerror(ret)); | |
1033 | } | |
1034 | ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED); | |
1035 | if (ret) { | |
1036 | ERR("pthread_attr_setdetachstate: %s", strerror(ret)); | |
1037 | } | |
1038 | ||
1039 | ret = pthread_create(&global_apps.ust_listener, &thread_attr, | |
dde70ea0 | 1040 | ust_listener_thread, &global_apps); |
d94d802c MD |
1041 | if (ret) { |
1042 | ERR("pthread_create global: %s", strerror(ret)); | |
1043 | } | |
8d20bf54 | 1044 | if (local_apps.allowed) { |
1879f67f | 1045 | ret = pthread_create(&local_apps.ust_listener, &thread_attr, |
dde70ea0 | 1046 | ust_listener_thread, &local_apps); |
d94d802c MD |
1047 | if (ret) { |
1048 | ERR("pthread_create local: %s", strerror(ret)); | |
1049 | } | |
8d20bf54 MD |
1050 | } else { |
1051 | handle_register_done(&local_apps); | |
1052 | } | |
1879f67f MG |
1053 | ret = pthread_attr_destroy(&thread_attr); |
1054 | if (ret) { | |
1055 | ERR("pthread_attr_destroy: %s", strerror(ret)); | |
1056 | } | |
8d20bf54 | 1057 | |
ae6a58bf WP |
1058 | /* Restore original signal mask in parent */ |
1059 | ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL); | |
1060 | if (ret) { | |
d94d802c | 1061 | ERR("pthread_sigmask: %s", strerror(ret)); |
ae6a58bf WP |
1062 | } |
1063 | ||
cf12a773 MD |
1064 | switch (timeout_mode) { |
1065 | case 1: /* timeout wait */ | |
95259bd0 MD |
1066 | do { |
1067 | ret = sem_timedwait(&constructor_wait, | |
1068 | &constructor_timeout); | |
1069 | } while (ret < 0 && errno == EINTR); | |
cf12a773 MD |
1070 | if (ret < 0 && errno == ETIMEDOUT) { |
1071 | ERR("Timed out waiting for ltt-sessiond"); | |
1072 | } else { | |
1073 | assert(!ret); | |
1074 | } | |
1075 | break; | |
7b766b16 | 1076 | case -1:/* wait forever */ |
95259bd0 MD |
1077 | do { |
1078 | ret = sem_wait(&constructor_wait); | |
1079 | } while (ret < 0 && errno == EINTR); | |
11ff9c7d | 1080 | assert(!ret); |
cf12a773 | 1081 | break; |
7b766b16 | 1082 | case 0: /* no timeout */ |
cf12a773 | 1083 | break; |
11ff9c7d | 1084 | } |
2691221a MD |
1085 | } |
1086 | ||
17dfb34b MD |
1087 | static |
1088 | void lttng_ust_cleanup(int exiting) | |
1089 | { | |
efe0de09 | 1090 | cleanup_sock_info(&global_apps, exiting); |
17dfb34b | 1091 | if (local_apps.allowed) { |
efe0de09 | 1092 | cleanup_sock_info(&local_apps, exiting); |
17dfb34b | 1093 | } |
efe0de09 MD |
1094 | /* |
1095 | * The teardown in this function all affect data structures | |
1096 | * accessed under the UST lock by the listener thread. This | |
1097 | * lock, along with the lttng_ust_comm_should_quit flag, ensure | |
1098 | * that none of these threads are accessing this data at this | |
1099 | * point. | |
1100 | */ | |
17dfb34b | 1101 | lttng_ust_abi_exit(); |
003fedf4 | 1102 | lttng_ust_events_exit(); |
17dfb34b MD |
1103 | ltt_ring_buffer_client_discard_exit(); |
1104 | ltt_ring_buffer_client_overwrite_exit(); | |
1105 | ltt_ring_buffer_metadata_client_exit(); | |
1106 | exit_tracepoint(); | |
1107 | if (!exiting) { | |
1108 | /* Reinitialize values for fork */ | |
1109 | sem_count = 2; | |
1110 | lttng_ust_comm_should_quit = 0; | |
1111 | initialized = 0; | |
1112 | } | |
1113 | } | |
1114 | ||
edaa1431 | 1115 | void __attribute__((destructor)) lttng_ust_exit(void) |
2691221a MD |
1116 | { |
1117 | int ret; | |
1118 | ||
9eb62b9c MD |
1119 | /* |
1120 | * Using pthread_cancel here because: | |
1121 | * A) we don't want to hang application teardown. | |
1122 | * B) the thread is not allocating any resource. | |
1123 | */ | |
1ea11eab MD |
1124 | |
1125 | /* | |
1126 | * Require the communication thread to quit. Synchronize with | |
1127 | * mutexes to ensure it is not in a mutex critical section when | |
1128 | * pthread_cancel is later called. | |
1129 | */ | |
17dfb34b | 1130 | ust_lock(); |
1ea11eab | 1131 | lttng_ust_comm_should_quit = 1; |
17dfb34b | 1132 | ust_unlock(); |
1ea11eab | 1133 | |
f5f94532 | 1134 | /* cancel threads */ |
1ea11eab | 1135 | ret = pthread_cancel(global_apps.ust_listener); |
9eb62b9c | 1136 | if (ret) { |
d94d802c MD |
1137 | ERR("Error cancelling global ust listener thread: %s", |
1138 | strerror(ret)); | |
2691221a | 1139 | } |
8d20bf54 MD |
1140 | if (local_apps.allowed) { |
1141 | ret = pthread_cancel(local_apps.ust_listener); | |
1142 | if (ret) { | |
d94d802c MD |
1143 | ERR("Error cancelling local ust listener thread: %s", |
1144 | strerror(ret)); | |
8d20bf54 | 1145 | } |
8d20bf54 | 1146 | } |
efe0de09 MD |
1147 | /* |
1148 | * Do NOT join threads: use of sys_futex makes it impossible to | |
1149 | * join the threads without using async-cancel, but async-cancel | |
1150 | * is delivered by a signal, which could hit the target thread | |
1151 | * anywhere in its code path, including while the ust_lock() is | |
1152 | * held, causing a deadlock for the other thread. Let the OS | |
1153 | * cleanup the threads if there are stalled in a syscall. | |
1154 | */ | |
17dfb34b | 1155 | lttng_ust_cleanup(1); |
2691221a | 1156 | } |
e822f505 MD |
1157 | |
1158 | /* | |
1159 | * We exclude the worker threads across fork and clone (except | |
1160 | * CLONE_VM), because these system calls only keep the forking thread | |
1161 | * running in the child. Therefore, we don't want to call fork or clone | |
1162 | * in the middle of an tracepoint or ust tracing state modification. | |
1163 | * Holding this mutex protects these structures across fork and clone. | |
1164 | */ | |
b728d87e | 1165 | void ust_before_fork(sigset_t *save_sigset) |
e822f505 MD |
1166 | { |
1167 | /* | |
1168 | * Disable signals. This is to avoid that the child intervenes | |
1169 | * before it is properly setup for tracing. It is safer to | |
1170 | * disable all signals, because then we know we are not breaking | |
1171 | * anything by restoring the original mask. | |
1172 | */ | |
1173 | sigset_t all_sigs; | |
1174 | int ret; | |
1175 | ||
e8508a49 MD |
1176 | if (lttng_ust_nest_count) |
1177 | return; | |
e822f505 MD |
1178 | /* Disable signals */ |
1179 | sigfillset(&all_sigs); | |
b728d87e | 1180 | ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset); |
e822f505 MD |
1181 | if (ret == -1) { |
1182 | PERROR("sigprocmask"); | |
1183 | } | |
17dfb34b | 1184 | ust_lock(); |
e822f505 MD |
1185 | rcu_bp_before_fork(); |
1186 | } | |
1187 | ||
b728d87e | 1188 | static void ust_after_fork_common(sigset_t *restore_sigset) |
e822f505 MD |
1189 | { |
1190 | int ret; | |
1191 | ||
17dfb34b MD |
1192 | DBG("process %d", getpid()); |
1193 | ust_unlock(); | |
e822f505 | 1194 | /* Restore signals */ |
23c8854a | 1195 | ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL); |
e822f505 MD |
1196 | if (ret == -1) { |
1197 | PERROR("sigprocmask"); | |
1198 | } | |
1199 | } | |
1200 | ||
b728d87e | 1201 | void ust_after_fork_parent(sigset_t *restore_sigset) |
e822f505 | 1202 | { |
e8508a49 MD |
1203 | if (lttng_ust_nest_count) |
1204 | return; | |
17dfb34b | 1205 | DBG("process %d", getpid()); |
e822f505 MD |
1206 | rcu_bp_after_fork_parent(); |
1207 | /* Release mutexes and reenable signals */ | |
b728d87e | 1208 | ust_after_fork_common(restore_sigset); |
e822f505 MD |
1209 | } |
1210 | ||
17dfb34b MD |
1211 | /* |
1212 | * After fork, in the child, we need to cleanup all the leftover state, | |
1213 | * except the worker thread which already magically disappeared thanks | |
1214 | * to the weird Linux fork semantics. After tyding up, we call | |
1215 | * lttng_ust_init() again to start over as a new PID. | |
1216 | * | |
1217 | * This is meant for forks() that have tracing in the child between the | |
1218 | * fork and following exec call (if there is any). | |
1219 | */ | |
b728d87e | 1220 | void ust_after_fork_child(sigset_t *restore_sigset) |
e822f505 | 1221 | { |
e8508a49 MD |
1222 | if (lttng_ust_nest_count) |
1223 | return; | |
17dfb34b | 1224 | DBG("process %d", getpid()); |
e822f505 MD |
1225 | /* Release urcu mutexes */ |
1226 | rcu_bp_after_fork_child(); | |
17dfb34b | 1227 | lttng_ust_cleanup(0); |
a93bfc45 | 1228 | lttng_context_vtid_reset(); |
e822f505 | 1229 | /* Release mutexes and reenable signals */ |
b728d87e | 1230 | ust_after_fork_common(restore_sigset); |
318dfea9 | 1231 | lttng_ust_init(); |
e822f505 | 1232 | } |