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" |
7dd08bec | 49 | #include "lttng-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; |
f70c1cd1 | 99 | int thread_active; |
7fc90dca MD |
100 | |
101 | char sock_path[PATH_MAX]; | |
102 | int socket; | |
103 | ||
104 | char wait_shm_path[PATH_MAX]; | |
105 | char *wait_shm_mmap; | |
1ea11eab | 106 | }; |
2691221a MD |
107 | |
108 | /* Socket from app (connect) to session daemon (listen) for communication */ | |
1ea11eab | 109 | struct sock_info global_apps = { |
11ff9c7d | 110 | .name = "global", |
44e073f5 | 111 | .global = 1, |
7fc90dca | 112 | |
46050b1a | 113 | .root_handle = -1, |
8d20bf54 | 114 | .allowed = 1, |
f70c1cd1 | 115 | .thread_active = 0, |
7fc90dca MD |
116 | |
117 | .sock_path = DEFAULT_GLOBAL_APPS_UNIX_SOCK, | |
118 | .socket = -1, | |
119 | ||
120 | .wait_shm_path = DEFAULT_GLOBAL_APPS_WAIT_SHM_PATH, | |
1ea11eab | 121 | }; |
2691221a MD |
122 | |
123 | /* TODO: allow global_apps_sock_path override */ | |
124 | ||
1ea11eab | 125 | struct sock_info local_apps = { |
11ff9c7d | 126 | .name = "local", |
44e073f5 | 127 | .global = 0, |
46050b1a | 128 | .root_handle = -1, |
8d20bf54 | 129 | .allowed = 0, /* Check setuid bit first */ |
f70c1cd1 | 130 | .thread_active = 0, |
7fc90dca MD |
131 | |
132 | .socket = -1, | |
1ea11eab | 133 | }; |
2691221a | 134 | |
37ed587a MD |
135 | static int wait_poll_fallback; |
136 | ||
7dd08bec MD |
137 | extern void lttng_ring_buffer_client_overwrite_init(void); |
138 | extern void lttng_ring_buffer_client_discard_init(void); | |
139 | extern void lttng_ring_buffer_metadata_client_init(void); | |
140 | extern void lttng_ring_buffer_client_overwrite_exit(void); | |
141 | extern void lttng_ring_buffer_client_discard_exit(void); | |
142 | extern void lttng_ring_buffer_metadata_client_exit(void); | |
edaa1431 | 143 | |
a903623f MD |
144 | /* |
145 | * Force a read (imply TLS fixup for dlopen) of TLS variables. | |
146 | */ | |
147 | static | |
148 | void lttng_fixup_nest_count_tls(void) | |
149 | { | |
150 | asm volatile ("" : : "m" (lttng_ust_nest_count)); | |
151 | } | |
152 | ||
2691221a | 153 | static |
8d20bf54 | 154 | int setup_local_apps(void) |
2691221a MD |
155 | { |
156 | const char *home_dir; | |
7fc90dca | 157 | uid_t uid; |
2691221a | 158 | |
7fc90dca | 159 | uid = getuid(); |
8d20bf54 MD |
160 | /* |
161 | * Disallow per-user tracing for setuid binaries. | |
162 | */ | |
7fc90dca | 163 | if (uid != geteuid()) { |
9ec6895c | 164 | assert(local_apps.allowed == 0); |
d0a1ae63 | 165 | return 0; |
8d20bf54 | 166 | } |
2691221a | 167 | home_dir = (const char *) getenv("HOME"); |
9ec6895c MD |
168 | if (!home_dir) { |
169 | WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing."); | |
170 | assert(local_apps.allowed == 0); | |
2691221a | 171 | return -ENOENT; |
9ec6895c MD |
172 | } |
173 | local_apps.allowed = 1; | |
1ea11eab | 174 | snprintf(local_apps.sock_path, PATH_MAX, |
2691221a | 175 | DEFAULT_HOME_APPS_UNIX_SOCK, home_dir); |
7fc90dca MD |
176 | snprintf(local_apps.wait_shm_path, PATH_MAX, |
177 | DEFAULT_HOME_APPS_WAIT_SHM_PATH, uid); | |
2691221a MD |
178 | return 0; |
179 | } | |
180 | ||
181 | static | |
182 | int register_app_to_sessiond(int socket) | |
183 | { | |
184 | ssize_t ret; | |
185 | struct { | |
e44418f3 MD |
186 | uint32_t major; |
187 | uint32_t minor; | |
2691221a | 188 | pid_t pid; |
5c33bde8 | 189 | pid_t ppid; |
2691221a | 190 | uid_t uid; |
83610856 | 191 | gid_t gid; |
c117fb1b | 192 | uint32_t bits_per_long; |
2629549e | 193 | char name[16]; /* process name */ |
2691221a MD |
194 | } reg_msg; |
195 | ||
e44418f3 MD |
196 | reg_msg.major = LTTNG_UST_COMM_VERSION_MAJOR; |
197 | reg_msg.minor = LTTNG_UST_COMM_VERSION_MINOR; | |
2691221a | 198 | reg_msg.pid = getpid(); |
5c33bde8 | 199 | reg_msg.ppid = getppid(); |
2691221a | 200 | reg_msg.uid = getuid(); |
83610856 | 201 | reg_msg.gid = getgid(); |
c117fb1b | 202 | reg_msg.bits_per_long = CAA_BITS_PER_LONG; |
08114193 | 203 | lttng_ust_getprocname(reg_msg.name); |
2691221a | 204 | |
57773204 | 205 | ret = ustcomm_send_unix_sock(socket, ®_msg, sizeof(reg_msg)); |
2691221a MD |
206 | if (ret >= 0 && ret != sizeof(reg_msg)) |
207 | return -EIO; | |
208 | return ret; | |
209 | } | |
210 | ||
d9e99d10 | 211 | static |
57773204 | 212 | int send_reply(int sock, struct ustcomm_ust_reply *lur) |
d9e99d10 | 213 | { |
9eb62b9c | 214 | ssize_t len; |
d3a492d1 | 215 | |
57773204 | 216 | len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur)); |
d3a492d1 | 217 | switch (len) { |
a4be8962 | 218 | case sizeof(*lur): |
d3a492d1 MD |
219 | DBG("message successfully sent"); |
220 | return 0; | |
7bc53e94 MD |
221 | default: |
222 | if (len == -ECONNRESET) { | |
223 | DBG("remote end closed connection"); | |
d3a492d1 MD |
224 | return 0; |
225 | } | |
7bc53e94 MD |
226 | if (len < 0) |
227 | return len; | |
228 | DBG("incorrect message size: %zd", len); | |
229 | return -EINVAL; | |
d3a492d1 MD |
230 | } |
231 | } | |
232 | ||
233 | static | |
edaa1431 | 234 | int handle_register_done(struct sock_info *sock_info) |
11ff9c7d MD |
235 | { |
236 | int ret; | |
237 | ||
edaa1431 MD |
238 | if (sock_info->constructor_sem_posted) |
239 | return 0; | |
240 | sock_info->constructor_sem_posted = 1; | |
56cd7e2f MD |
241 | if (uatomic_read(&sem_count) <= 0) { |
242 | return 0; | |
243 | } | |
95259bd0 MD |
244 | ret = uatomic_add_return(&sem_count, -1); |
245 | if (ret == 0) { | |
246 | ret = sem_post(&constructor_wait); | |
247 | assert(!ret); | |
248 | } | |
11ff9c7d MD |
249 | return 0; |
250 | } | |
251 | ||
252 | static | |
253 | int handle_message(struct sock_info *sock_info, | |
57773204 | 254 | int sock, struct ustcomm_ust_msg *lum) |
d3a492d1 | 255 | { |
1ea11eab | 256 | int ret = 0; |
b61ce3b2 | 257 | const struct lttng_ust_objd_ops *ops; |
57773204 | 258 | struct ustcomm_ust_reply lur; |
193183fb | 259 | int shm_fd, wait_fd; |
ef9ff354 | 260 | union ust_args args; |
40003310 | 261 | ssize_t len; |
1ea11eab | 262 | |
17dfb34b | 263 | ust_lock(); |
1ea11eab | 264 | |
46050b1a MD |
265 | memset(&lur, 0, sizeof(lur)); |
266 | ||
1ea11eab | 267 | if (lttng_ust_comm_should_quit) { |
46050b1a | 268 | ret = -EPERM; |
1ea11eab MD |
269 | goto end; |
270 | } | |
9eb62b9c | 271 | |
46050b1a MD |
272 | ops = objd_ops(lum->handle); |
273 | if (!ops) { | |
274 | ret = -ENOENT; | |
275 | goto end; | |
1ea11eab | 276 | } |
46050b1a MD |
277 | |
278 | switch (lum->cmd) { | |
11ff9c7d MD |
279 | case LTTNG_UST_REGISTER_DONE: |
280 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
edaa1431 | 281 | ret = handle_register_done(sock_info); |
11ff9c7d MD |
282 | else |
283 | ret = -EINVAL; | |
284 | break; | |
46050b1a MD |
285 | case LTTNG_UST_RELEASE: |
286 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
287 | ret = -EPERM; | |
288 | else | |
d4419b81 | 289 | ret = lttng_ust_objd_unref(lum->handle); |
d9e99d10 | 290 | break; |
2d78951a MD |
291 | case LTTNG_UST_FILTER: |
292 | { | |
293 | /* Receive filter data */ | |
f488575f | 294 | struct lttng_ust_filter_bytecode_node *bytecode; |
2d78951a | 295 | |
cd54f6d9 | 296 | if (lum->u.filter.data_size > FILTER_BYTECODE_MAX_LEN) { |
7bc53e94 | 297 | ERR("Filter data size is too large: %u bytes", |
2d78951a MD |
298 | lum->u.filter.data_size); |
299 | ret = -EINVAL; | |
300 | goto error; | |
301 | } | |
2734ca65 | 302 | |
885b1dfd | 303 | if (lum->u.filter.reloc_offset > lum->u.filter.data_size) { |
7bc53e94 | 304 | ERR("Filter reloc offset %u is not within data", |
2734ca65 CB |
305 | lum->u.filter.reloc_offset); |
306 | ret = -EINVAL; | |
307 | goto error; | |
308 | } | |
309 | ||
cd54f6d9 MD |
310 | bytecode = zmalloc(sizeof(*bytecode) + lum->u.filter.data_size); |
311 | if (!bytecode) { | |
312 | ret = -ENOMEM; | |
313 | goto error; | |
314 | } | |
f488575f | 315 | len = ustcomm_recv_unix_sock(sock, bytecode->bc.data, |
2d78951a MD |
316 | lum->u.filter.data_size); |
317 | switch (len) { | |
318 | case 0: /* orderly shutdown */ | |
319 | ret = 0; | |
cd54f6d9 | 320 | free(bytecode); |
2d78951a | 321 | goto error; |
2d78951a MD |
322 | default: |
323 | if (len == lum->u.filter.data_size) { | |
7bc53e94 | 324 | DBG("filter data received"); |
2d78951a | 325 | break; |
7bc53e94 MD |
326 | } else if (len < 0) { |
327 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
328 | if (len == -ECONNRESET) { | |
329 | ERR("%s remote end closed connection", sock_info->name); | |
330 | ret = len; | |
331 | free(bytecode); | |
332 | goto error; | |
333 | } | |
334 | ret = len; | |
335 | goto end; | |
2d78951a | 336 | } else { |
7bc53e94 | 337 | DBG("incorrect filter data message size: %zd", len); |
2d78951a | 338 | ret = -EINVAL; |
cd54f6d9 | 339 | free(bytecode); |
2d78951a MD |
340 | goto end; |
341 | } | |
342 | } | |
f488575f MD |
343 | bytecode->bc.len = lum->u.filter.data_size; |
344 | bytecode->bc.reloc_offset = lum->u.filter.reloc_offset; | |
3f6fd224 | 345 | bytecode->bc.seqnum = lum->u.filter.seqnum; |
cd54f6d9 | 346 | if (ops->cmd) { |
2d78951a | 347 | ret = ops->cmd(lum->handle, lum->cmd, |
cd54f6d9 | 348 | (unsigned long) bytecode, |
f59ed768 | 349 | &args, sock_info); |
cd54f6d9 MD |
350 | if (ret) { |
351 | free(bytecode); | |
352 | } | |
353 | /* don't free bytecode if everything went fine. */ | |
354 | } else { | |
2d78951a | 355 | ret = -ENOSYS; |
cd54f6d9 MD |
356 | free(bytecode); |
357 | } | |
2d78951a MD |
358 | break; |
359 | } | |
d9e99d10 | 360 | default: |
46050b1a MD |
361 | if (ops->cmd) |
362 | ret = ops->cmd(lum->handle, lum->cmd, | |
ef9ff354 | 363 | (unsigned long) &lum->u, |
f59ed768 | 364 | &args, sock_info); |
46050b1a MD |
365 | else |
366 | ret = -ENOSYS; | |
367 | break; | |
d9e99d10 | 368 | } |
46050b1a | 369 | |
1ea11eab | 370 | end: |
46050b1a MD |
371 | lur.handle = lum->handle; |
372 | lur.cmd = lum->cmd; | |
373 | lur.ret_val = ret; | |
374 | if (ret >= 0) { | |
7bc53e94 | 375 | lur.ret_code = LTTNG_UST_OK; |
46050b1a | 376 | } else { |
7bc53e94 MD |
377 | /* |
378 | * Use -LTTNG_UST_ERR as wildcard for UST internal | |
379 | * error that are not caused by the transport, except if | |
380 | * we already have a more precise error message to | |
381 | * report. | |
382 | */ | |
64b2564e DG |
383 | if (ret > -LTTNG_UST_ERR) { |
384 | /* Translate code to UST error. */ | |
385 | switch (ret) { | |
386 | case -EEXIST: | |
387 | lur.ret_code = -LTTNG_UST_ERR_EXIST; | |
388 | break; | |
389 | case -EINVAL: | |
390 | lur.ret_code = -LTTNG_UST_ERR_INVAL; | |
391 | break; | |
392 | case -ENOENT: | |
393 | lur.ret_code = -LTTNG_UST_ERR_NOENT; | |
394 | break; | |
395 | case -EPERM: | |
396 | lur.ret_code = -LTTNG_UST_ERR_PERM; | |
397 | break; | |
398 | case -ENOSYS: | |
399 | lur.ret_code = -LTTNG_UST_ERR_NOSYS; | |
400 | break; | |
401 | default: | |
402 | lur.ret_code = -LTTNG_UST_ERR; | |
403 | break; | |
404 | } | |
405 | } else { | |
7bc53e94 | 406 | lur.ret_code = ret; |
64b2564e | 407 | } |
46050b1a | 408 | } |
e6ea14c5 MD |
409 | if (ret >= 0) { |
410 | switch (lum->cmd) { | |
411 | case LTTNG_UST_STREAM: | |
412 | /* | |
413 | * Special-case reply to send stream info. | |
414 | * Use lum.u output. | |
415 | */ | |
416 | lur.u.stream.memory_map_size = *args.stream.memory_map_size; | |
417 | shm_fd = *args.stream.shm_fd; | |
418 | wait_fd = *args.stream.wait_fd; | |
419 | break; | |
420 | case LTTNG_UST_METADATA: | |
421 | case LTTNG_UST_CHANNEL: | |
422 | lur.u.channel.memory_map_size = *args.channel.memory_map_size; | |
423 | shm_fd = *args.channel.shm_fd; | |
424 | wait_fd = *args.channel.wait_fd; | |
425 | break; | |
426 | case LTTNG_UST_TRACER_VERSION: | |
427 | lur.u.version = lum->u.version; | |
428 | break; | |
429 | case LTTNG_UST_TRACEPOINT_LIST_GET: | |
430 | memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint)); | |
431 | break; | |
432 | } | |
381c0f1e | 433 | } |
46050b1a | 434 | ret = send_reply(sock, &lur); |
193183fb | 435 | if (ret < 0) { |
7bc53e94 | 436 | DBG("error sending reply"); |
193183fb MD |
437 | goto error; |
438 | } | |
46050b1a | 439 | |
824f40b8 MD |
440 | if ((lum->cmd == LTTNG_UST_STREAM |
441 | || lum->cmd == LTTNG_UST_CHANNEL | |
442 | || lum->cmd == LTTNG_UST_METADATA) | |
7bc53e94 | 443 | && lur.ret_code == LTTNG_UST_OK) { |
50be2fc4 MD |
444 | int sendret = 0; |
445 | ||
381c0f1e | 446 | /* we also need to send the file descriptors. */ |
57773204 | 447 | ret = ustcomm_send_fds_unix_sock(sock, |
193183fb | 448 | &shm_fd, &shm_fd, |
381c0f1e MD |
449 | 1, sizeof(int)); |
450 | if (ret < 0) { | |
7bc53e94 | 451 | ERR("send shm_fd"); |
50be2fc4 | 452 | sendret = ret; |
381c0f1e | 453 | } |
50be2fc4 MD |
454 | /* |
455 | * The sessiond expects 2 file descriptors, even upon | |
456 | * error. | |
457 | */ | |
57773204 | 458 | ret = ustcomm_send_fds_unix_sock(sock, |
193183fb | 459 | &wait_fd, &wait_fd, |
381c0f1e MD |
460 | 1, sizeof(int)); |
461 | if (ret < 0) { | |
462 | perror("send wait_fd"); | |
463 | goto error; | |
464 | } | |
50be2fc4 MD |
465 | if (sendret) { |
466 | ret = sendret; | |
467 | goto error; | |
468 | } | |
381c0f1e | 469 | } |
40003310 MD |
470 | /* |
471 | * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field | |
472 | * after the reply. | |
473 | */ | |
7bc53e94 | 474 | if (lur.ret_code == LTTNG_UST_OK) { |
40003310 MD |
475 | switch (lum->cmd) { |
476 | case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET: | |
477 | len = ustcomm_send_unix_sock(sock, | |
478 | &args.field_list.entry, | |
479 | sizeof(args.field_list.entry)); | |
7bc53e94 MD |
480 | if (len < 0) { |
481 | ret = len; | |
482 | goto error; | |
483 | } | |
40003310 | 484 | if (len != sizeof(args.field_list.entry)) { |
7bc53e94 | 485 | ret = -EINVAL; |
40003310 MD |
486 | goto error; |
487 | } | |
488 | } | |
489 | } | |
ef9ff354 MD |
490 | /* |
491 | * We still have the memory map reference, and the fds have been | |
3ad5f707 MD |
492 | * sent to the sessiond. We can therefore close those fds. Note |
493 | * that we keep the write side of the wait_fd open, but close | |
494 | * the read side. | |
ef9ff354 | 495 | */ |
7bc53e94 | 496 | if (lur.ret_code == LTTNG_UST_OK) { |
ef9ff354 MD |
497 | switch (lum->cmd) { |
498 | case LTTNG_UST_STREAM: | |
499 | if (shm_fd >= 0) { | |
500 | ret = close(shm_fd); | |
501 | if (ret) { | |
502 | PERROR("Error closing stream shm_fd"); | |
503 | } | |
504 | *args.stream.shm_fd = -1; | |
505 | } | |
506 | if (wait_fd >= 0) { | |
507 | ret = close(wait_fd); | |
508 | if (ret) { | |
509 | PERROR("Error closing stream wait_fd"); | |
510 | } | |
511 | *args.stream.wait_fd = -1; | |
512 | } | |
513 | break; | |
514 | case LTTNG_UST_METADATA: | |
515 | case LTTNG_UST_CHANNEL: | |
516 | if (shm_fd >= 0) { | |
517 | ret = close(shm_fd); | |
518 | if (ret) { | |
519 | PERROR("Error closing channel shm_fd"); | |
520 | } | |
521 | *args.channel.shm_fd = -1; | |
522 | } | |
523 | if (wait_fd >= 0) { | |
524 | ret = close(wait_fd); | |
525 | if (ret) { | |
526 | PERROR("Error closing channel wait_fd"); | |
527 | } | |
528 | *args.channel.wait_fd = -1; | |
529 | } | |
530 | break; | |
531 | } | |
532 | } | |
533 | ||
381c0f1e | 534 | error: |
17dfb34b | 535 | ust_unlock(); |
1ea11eab | 536 | return ret; |
d9e99d10 MD |
537 | } |
538 | ||
46050b1a | 539 | static |
efe0de09 | 540 | void cleanup_sock_info(struct sock_info *sock_info, int exiting) |
46050b1a MD |
541 | { |
542 | int ret; | |
543 | ||
544 | if (sock_info->socket != -1) { | |
e6973a89 | 545 | ret = ustcomm_close_unix_sock(sock_info->socket); |
46050b1a | 546 | if (ret) { |
7fc90dca | 547 | ERR("Error closing apps socket"); |
46050b1a MD |
548 | } |
549 | sock_info->socket = -1; | |
550 | } | |
551 | if (sock_info->root_handle != -1) { | |
d4419b81 | 552 | ret = lttng_ust_objd_unref(sock_info->root_handle); |
46050b1a MD |
553 | if (ret) { |
554 | ERR("Error unref root handle"); | |
555 | } | |
556 | sock_info->root_handle = -1; | |
557 | } | |
318dfea9 | 558 | sock_info->constructor_sem_posted = 0; |
efe0de09 MD |
559 | /* |
560 | * wait_shm_mmap is used by listener threads outside of the | |
561 | * ust lock, so we cannot tear it down ourselves, because we | |
562 | * cannot join on these threads. Leave this task to the OS | |
563 | * process exit. | |
564 | */ | |
565 | if (!exiting && sock_info->wait_shm_mmap) { | |
7fc90dca MD |
566 | ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE)); |
567 | if (ret) { | |
568 | ERR("Error unmapping wait shm"); | |
569 | } | |
570 | sock_info->wait_shm_mmap = NULL; | |
571 | } | |
572 | } | |
573 | ||
58d4b2a2 | 574 | /* |
33bbeb90 MD |
575 | * Using fork to set umask in the child process (not multi-thread safe). |
576 | * We deal with the shm_open vs ftruncate race (happening when the | |
577 | * sessiond owns the shm and does not let everybody modify it, to ensure | |
578 | * safety against shm_unlink) by simply letting the mmap fail and | |
579 | * retrying after a few seconds. | |
580 | * For global shm, everybody has rw access to it until the sessiond | |
581 | * starts. | |
58d4b2a2 | 582 | */ |
7fc90dca | 583 | static |
58d4b2a2 | 584 | int get_wait_shm(struct sock_info *sock_info, size_t mmap_size) |
7fc90dca | 585 | { |
7fc90dca | 586 | int wait_shm_fd, ret; |
58d4b2a2 | 587 | pid_t pid; |
44e073f5 | 588 | |
58d4b2a2 | 589 | /* |
33bbeb90 | 590 | * Try to open read-only. |
58d4b2a2 | 591 | */ |
33bbeb90 | 592 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 MD |
593 | if (wait_shm_fd >= 0) { |
594 | goto end; | |
595 | } else if (wait_shm_fd < 0 && errno != ENOENT) { | |
596 | /* | |
33bbeb90 MD |
597 | * Real-only open did not work, and it's not because the |
598 | * entry was not present. It's a failure that prohibits | |
599 | * using shm. | |
58d4b2a2 | 600 | */ |
7fc90dca | 601 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
58d4b2a2 | 602 | goto end; |
7fc90dca MD |
603 | } |
604 | /* | |
58d4b2a2 MD |
605 | * If the open failed because the file did not exist, try |
606 | * creating it ourself. | |
7fc90dca | 607 | */ |
e8508a49 | 608 | lttng_ust_nest_count++; |
58d4b2a2 | 609 | pid = fork(); |
e8508a49 | 610 | lttng_ust_nest_count--; |
58d4b2a2 MD |
611 | if (pid > 0) { |
612 | int status; | |
613 | ||
614 | /* | |
615 | * Parent: wait for child to return, in which case the | |
616 | * shared memory map will have been created. | |
617 | */ | |
618 | pid = wait(&status); | |
b7d3cb32 | 619 | if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) { |
58d4b2a2 MD |
620 | wait_shm_fd = -1; |
621 | goto end; | |
7fc90dca | 622 | } |
58d4b2a2 MD |
623 | /* |
624 | * Try to open read-only again after creation. | |
625 | */ | |
33bbeb90 | 626 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 MD |
627 | if (wait_shm_fd < 0) { |
628 | /* | |
629 | * Real-only open did not work. It's a failure | |
630 | * that prohibits using shm. | |
631 | */ | |
632 | ERR("Error opening shm %s", sock_info->wait_shm_path); | |
633 | goto end; | |
634 | } | |
635 | goto end; | |
636 | } else if (pid == 0) { | |
637 | int create_mode; | |
638 | ||
639 | /* Child */ | |
33bbeb90 | 640 | create_mode = S_IRUSR | S_IWUSR | S_IRGRP; |
58d4b2a2 | 641 | if (sock_info->global) |
33bbeb90 | 642 | create_mode |= S_IROTH | S_IWGRP | S_IWOTH; |
58d4b2a2 MD |
643 | /* |
644 | * We're alone in a child process, so we can modify the | |
645 | * process-wide umask. | |
646 | */ | |
33bbeb90 | 647 | umask(~create_mode); |
58d4b2a2 | 648 | /* |
33bbeb90 MD |
649 | * Try creating shm (or get rw access). |
650 | * We don't do an exclusive open, because we allow other | |
651 | * processes to create+ftruncate it concurrently. | |
58d4b2a2 MD |
652 | */ |
653 | wait_shm_fd = shm_open(sock_info->wait_shm_path, | |
654 | O_RDWR | O_CREAT, create_mode); | |
655 | if (wait_shm_fd >= 0) { | |
656 | ret = ftruncate(wait_shm_fd, mmap_size); | |
657 | if (ret) { | |
658 | PERROR("ftruncate"); | |
b0c1425d | 659 | _exit(EXIT_FAILURE); |
58d4b2a2 | 660 | } |
b0c1425d | 661 | _exit(EXIT_SUCCESS); |
58d4b2a2 | 662 | } |
33bbeb90 MD |
663 | /* |
664 | * For local shm, we need to have rw access to accept | |
665 | * opening it: this means the local sessiond will be | |
666 | * able to wake us up. For global shm, we open it even | |
667 | * if rw access is not granted, because the root.root | |
668 | * sessiond will be able to override all rights and wake | |
669 | * us up. | |
670 | */ | |
671 | if (!sock_info->global && errno != EACCES) { | |
58d4b2a2 | 672 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
5d3bc5ed | 673 | _exit(EXIT_FAILURE); |
58d4b2a2 MD |
674 | } |
675 | /* | |
33bbeb90 MD |
676 | * The shm exists, but we cannot open it RW. Report |
677 | * success. | |
58d4b2a2 | 678 | */ |
5d3bc5ed | 679 | _exit(EXIT_SUCCESS); |
58d4b2a2 MD |
680 | } else { |
681 | return -1; | |
7fc90dca | 682 | } |
58d4b2a2 | 683 | end: |
33bbeb90 MD |
684 | if (wait_shm_fd >= 0 && !sock_info->global) { |
685 | struct stat statbuf; | |
686 | ||
687 | /* | |
688 | * Ensure that our user is the owner of the shm file for | |
689 | * local shm. If we do not own the file, it means our | |
690 | * sessiond will not have access to wake us up (there is | |
691 | * probably a rogue process trying to fake our | |
692 | * sessiond). Fallback to polling method in this case. | |
693 | */ | |
694 | ret = fstat(wait_shm_fd, &statbuf); | |
695 | if (ret) { | |
696 | PERROR("fstat"); | |
697 | goto error_close; | |
698 | } | |
699 | if (statbuf.st_uid != getuid()) | |
700 | goto error_close; | |
701 | } | |
58d4b2a2 | 702 | return wait_shm_fd; |
33bbeb90 MD |
703 | |
704 | error_close: | |
705 | ret = close(wait_shm_fd); | |
706 | if (ret) { | |
707 | PERROR("Error closing fd"); | |
708 | } | |
709 | return -1; | |
58d4b2a2 MD |
710 | } |
711 | ||
712 | static | |
713 | char *get_map_shm(struct sock_info *sock_info) | |
714 | { | |
715 | size_t mmap_size = sysconf(_SC_PAGE_SIZE); | |
716 | int wait_shm_fd, ret; | |
717 | char *wait_shm_mmap; | |
718 | ||
719 | wait_shm_fd = get_wait_shm(sock_info, mmap_size); | |
720 | if (wait_shm_fd < 0) { | |
721 | goto error; | |
44e073f5 | 722 | } |
7fc90dca MD |
723 | wait_shm_mmap = mmap(NULL, mmap_size, PROT_READ, |
724 | MAP_SHARED, wait_shm_fd, 0); | |
7fc90dca MD |
725 | /* close shm fd immediately after taking the mmap reference */ |
726 | ret = close(wait_shm_fd); | |
727 | if (ret) { | |
33bbeb90 MD |
728 | PERROR("Error closing fd"); |
729 | } | |
730 | if (wait_shm_mmap == MAP_FAILED) { | |
731 | DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode."); | |
732 | goto error; | |
7fc90dca MD |
733 | } |
734 | return wait_shm_mmap; | |
735 | ||
736 | error: | |
737 | return NULL; | |
738 | } | |
739 | ||
740 | static | |
741 | void wait_for_sessiond(struct sock_info *sock_info) | |
742 | { | |
efe0de09 | 743 | int ret; |
80e2814b | 744 | |
7fc90dca MD |
745 | ust_lock(); |
746 | if (lttng_ust_comm_should_quit) { | |
747 | goto quit; | |
748 | } | |
37ed587a MD |
749 | if (wait_poll_fallback) { |
750 | goto error; | |
751 | } | |
7fc90dca MD |
752 | if (!sock_info->wait_shm_mmap) { |
753 | sock_info->wait_shm_mmap = get_map_shm(sock_info); | |
754 | if (!sock_info->wait_shm_mmap) | |
755 | goto error; | |
756 | } | |
757 | ust_unlock(); | |
758 | ||
759 | DBG("Waiting for %s apps sessiond", sock_info->name); | |
80e2814b MD |
760 | /* Wait for futex wakeup */ |
761 | if (uatomic_read((int32_t *) sock_info->wait_shm_mmap) == 0) { | |
762 | ret = futex_async((int32_t *) sock_info->wait_shm_mmap, | |
763 | FUTEX_WAIT, 0, NULL, NULL, 0); | |
80e2814b | 764 | if (ret < 0) { |
37ed587a MD |
765 | if (errno == EFAULT) { |
766 | wait_poll_fallback = 1; | |
a8b870ad | 767 | DBG( |
37ed587a MD |
768 | "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) " |
769 | "do not support FUTEX_WAKE on read-only memory mappings correctly. " | |
770 | "Please upgrade your kernel " | |
771 | "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel " | |
772 | "mainline). LTTng-UST will use polling mode fallback."); | |
cd27263b MD |
773 | if (ust_debug()) |
774 | PERROR("futex"); | |
37ed587a | 775 | } |
80e2814b MD |
776 | } |
777 | } | |
7fc90dca MD |
778 | return; |
779 | ||
780 | quit: | |
781 | ust_unlock(); | |
782 | return; | |
783 | ||
784 | error: | |
785 | ust_unlock(); | |
7fc90dca | 786 | return; |
46050b1a MD |
787 | } |
788 | ||
1ea11eab MD |
789 | /* |
790 | * This thread does not allocate any resource, except within | |
791 | * handle_message, within mutex protection. This mutex protects against | |
792 | * fork and exit. | |
98bf993f | 793 | * The other moment it allocates resources is at socket connection, which |
1ea11eab MD |
794 | * is also protected by the mutex. |
795 | */ | |
d9e99d10 MD |
796 | static |
797 | void *ust_listener_thread(void *arg) | |
798 | { | |
1ea11eab | 799 | struct sock_info *sock_info = arg; |
c0eedf81 | 800 | int sock, ret, prev_connect_failed = 0, has_waited = 0; |
d9e99d10 | 801 | |
9eb62b9c MD |
802 | /* Restart trying to connect to the session daemon */ |
803 | restart: | |
c0eedf81 MD |
804 | if (prev_connect_failed) { |
805 | /* Wait for sessiond availability with pipe */ | |
806 | wait_for_sessiond(sock_info); | |
807 | if (has_waited) { | |
808 | has_waited = 0; | |
809 | /* | |
810 | * Sleep for 5 seconds before retrying after a | |
811 | * sequence of failure / wait / failure. This | |
812 | * deals with a killed or broken session daemon. | |
813 | */ | |
814 | sleep(5); | |
815 | } | |
816 | has_waited = 1; | |
817 | prev_connect_failed = 0; | |
818 | } | |
17dfb34b | 819 | ust_lock(); |
1ea11eab MD |
820 | |
821 | if (lttng_ust_comm_should_quit) { | |
1ea11eab MD |
822 | goto quit; |
823 | } | |
9eb62b9c | 824 | |
1ea11eab | 825 | if (sock_info->socket != -1) { |
e6973a89 | 826 | ret = ustcomm_close_unix_sock(sock_info->socket); |
1ea11eab | 827 | if (ret) { |
11ff9c7d | 828 | ERR("Error closing %s apps socket", sock_info->name); |
1ea11eab MD |
829 | } |
830 | sock_info->socket = -1; | |
831 | } | |
46050b1a | 832 | |
9eb62b9c | 833 | /* Register */ |
57773204 | 834 | ret = ustcomm_connect_unix_sock(sock_info->sock_path); |
9eb62b9c | 835 | if (ret < 0) { |
4d3c9523 | 836 | DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name); |
c0eedf81 | 837 | prev_connect_failed = 1; |
11ff9c7d MD |
838 | /* |
839 | * If we cannot find the sessiond daemon, don't delay | |
840 | * constructor execution. | |
841 | */ | |
edaa1431 | 842 | ret = handle_register_done(sock_info); |
11ff9c7d | 843 | assert(!ret); |
17dfb34b | 844 | ust_unlock(); |
1ea11eab | 845 | goto restart; |
46050b1a MD |
846 | } |
847 | ||
848 | sock_info->socket = sock = ret; | |
849 | ||
850 | /* | |
851 | * Create only one root handle per listener thread for the whole | |
f59ed768 MD |
852 | * process lifetime, so we ensure we get ID which is statically |
853 | * assigned to the root handle. | |
46050b1a MD |
854 | */ |
855 | if (sock_info->root_handle == -1) { | |
856 | ret = lttng_abi_create_root_handle(); | |
a51070bb | 857 | if (ret < 0) { |
46050b1a | 858 | ERR("Error creating root handle"); |
46050b1a MD |
859 | goto quit; |
860 | } | |
861 | sock_info->root_handle = ret; | |
9eb62b9c | 862 | } |
1ea11eab | 863 | |
9eb62b9c MD |
864 | ret = register_app_to_sessiond(sock); |
865 | if (ret < 0) { | |
11ff9c7d | 866 | ERR("Error registering to %s apps socket", sock_info->name); |
c0eedf81 | 867 | prev_connect_failed = 1; |
11ff9c7d MD |
868 | /* |
869 | * If we cannot register to the sessiond daemon, don't | |
870 | * delay constructor execution. | |
871 | */ | |
edaa1431 | 872 | ret = handle_register_done(sock_info); |
11ff9c7d | 873 | assert(!ret); |
17dfb34b | 874 | ust_unlock(); |
9eb62b9c MD |
875 | goto restart; |
876 | } | |
17dfb34b | 877 | ust_unlock(); |
46050b1a | 878 | |
d9e99d10 MD |
879 | for (;;) { |
880 | ssize_t len; | |
57773204 | 881 | struct ustcomm_ust_msg lum; |
d9e99d10 | 882 | |
57773204 | 883 | len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum)); |
d9e99d10 MD |
884 | switch (len) { |
885 | case 0: /* orderly shutdown */ | |
7dd08bec | 886 | DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name); |
8236ba10 | 887 | ust_lock(); |
235ebca2 | 888 | if (lttng_ust_comm_should_quit) { |
235ebca2 MD |
889 | goto quit; |
890 | } | |
8236ba10 MD |
891 | /* |
892 | * Either sessiond has shutdown or refused us by closing the socket. | |
893 | * In either case, we don't want to delay construction execution, | |
894 | * and we need to wait before retry. | |
895 | */ | |
896 | prev_connect_failed = 1; | |
897 | /* | |
898 | * If we cannot register to the sessiond daemon, don't | |
899 | * delay constructor execution. | |
900 | */ | |
901 | ret = handle_register_done(sock_info); | |
902 | assert(!ret); | |
903 | ust_unlock(); | |
d9e99d10 | 904 | goto end; |
e7723462 | 905 | case sizeof(lum): |
7bc53e94 | 906 | DBG("message received"); |
11ff9c7d | 907 | ret = handle_message(sock_info, sock, &lum); |
7bc53e94 | 908 | if (ret) { |
11ff9c7d | 909 | ERR("Error handling message for %s socket", sock_info->name); |
d9e99d10 MD |
910 | } |
911 | continue; | |
7bc53e94 MD |
912 | default: |
913 | if (len < 0) { | |
914 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
915 | } else { | |
916 | DBG("incorrect message size (%s socket): %zd", sock_info->name, len); | |
917 | } | |
918 | if (len == -ECONNRESET) { | |
919 | DBG("%s remote end closed connection", sock_info->name); | |
d9e99d10 MD |
920 | goto end; |
921 | } | |
922 | goto end; | |
d9e99d10 MD |
923 | } |
924 | ||
925 | } | |
926 | end: | |
f59ed768 | 927 | ust_lock(); |
235ebca2 | 928 | if (lttng_ust_comm_should_quit) { |
235ebca2 MD |
929 | goto quit; |
930 | } | |
f59ed768 MD |
931 | /* Cleanup socket handles before trying to reconnect */ |
932 | lttng_ust_objd_table_owner_cleanup(sock_info); | |
933 | ust_unlock(); | |
9eb62b9c | 934 | goto restart; /* try to reconnect */ |
f70c1cd1 | 935 | |
1ea11eab | 936 | quit: |
f70c1cd1 MD |
937 | sock_info->thread_active = 0; |
938 | ust_unlock(); | |
d9e99d10 MD |
939 | return NULL; |
940 | } | |
941 | ||
cf12a773 MD |
942 | /* |
943 | * Return values: -1: don't wait. 0: wait forever. 1: timeout wait. | |
944 | */ | |
11ff9c7d MD |
945 | static |
946 | int get_timeout(struct timespec *constructor_timeout) | |
947 | { | |
cf12a773 MD |
948 | long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS; |
949 | char *str_delay; | |
11ff9c7d MD |
950 | int ret; |
951 | ||
69400ac4 | 952 | str_delay = getenv("LTTNG_UST_REGISTER_TIMEOUT"); |
cf12a773 MD |
953 | if (str_delay) { |
954 | constructor_delay_ms = strtol(str_delay, NULL, 10); | |
955 | } | |
956 | ||
957 | switch (constructor_delay_ms) { | |
958 | case -1:/* fall-through */ | |
959 | case 0: | |
960 | return constructor_delay_ms; | |
961 | default: | |
962 | break; | |
963 | } | |
964 | ||
965 | /* | |
966 | * If we are unable to find the current time, don't wait. | |
967 | */ | |
968 | ret = clock_gettime(CLOCK_REALTIME, constructor_timeout); | |
969 | if (ret) { | |
970 | return -1; | |
971 | } | |
95259bd0 MD |
972 | constructor_timeout->tv_sec += constructor_delay_ms / 1000UL; |
973 | constructor_timeout->tv_nsec += | |
974 | (constructor_delay_ms % 1000UL) * 1000000UL; | |
11ff9c7d MD |
975 | if (constructor_timeout->tv_nsec >= 1000000000UL) { |
976 | constructor_timeout->tv_sec++; | |
977 | constructor_timeout->tv_nsec -= 1000000000UL; | |
978 | } | |
cf12a773 | 979 | return 1; |
11ff9c7d | 980 | } |
d9e99d10 | 981 | |
2691221a MD |
982 | /* |
983 | * sessiond monitoring thread: monitor presence of global and per-user | |
984 | * sessiond by polling the application common named pipe. | |
985 | */ | |
edaa1431 | 986 | void __attribute__((constructor)) lttng_ust_init(void) |
2691221a | 987 | { |
11ff9c7d | 988 | struct timespec constructor_timeout; |
ae6a58bf | 989 | sigset_t sig_all_blocked, orig_parent_mask; |
1879f67f | 990 | pthread_attr_t thread_attr; |
cf12a773 | 991 | int timeout_mode; |
2691221a MD |
992 | int ret; |
993 | ||
edaa1431 MD |
994 | if (uatomic_xchg(&initialized, 1) == 1) |
995 | return; | |
996 | ||
eddd8d5d MD |
997 | /* |
998 | * Fixup interdependency between TLS fixup mutex (which happens | |
999 | * to be the dynamic linker mutex) and ust_lock, taken within | |
1000 | * the ust lock. | |
1001 | */ | |
1002 | lttng_fixup_event_tls(); | |
f645cfa7 | 1003 | lttng_fixup_ringbuffer_tls(); |
4158a15a | 1004 | lttng_fixup_vtid_tls(); |
a903623f | 1005 | lttng_fixup_nest_count_tls(); |
009745db | 1006 | lttng_fixup_procname_tls(); |
eddd8d5d | 1007 | |
edaa1431 MD |
1008 | /* |
1009 | * We want precise control over the order in which we construct | |
1010 | * our sub-libraries vs starting to receive commands from | |
1011 | * sessiond (otherwise leading to errors when trying to create | |
1012 | * sessiond before the init functions are completed). | |
1013 | */ | |
2691221a | 1014 | init_usterr(); |
edaa1431 | 1015 | init_tracepoint(); |
7dd08bec MD |
1016 | lttng_ring_buffer_metadata_client_init(); |
1017 | lttng_ring_buffer_client_overwrite_init(); | |
1018 | lttng_ring_buffer_client_discard_init(); | |
2691221a | 1019 | |
cf12a773 | 1020 | timeout_mode = get_timeout(&constructor_timeout); |
11ff9c7d | 1021 | |
95259bd0 | 1022 | ret = sem_init(&constructor_wait, 0, 0); |
11ff9c7d MD |
1023 | assert(!ret); |
1024 | ||
8d20bf54 | 1025 | ret = setup_local_apps(); |
2691221a | 1026 | if (ret) { |
9ec6895c | 1027 | DBG("local apps setup returned %d", ret); |
2691221a | 1028 | } |
ae6a58bf WP |
1029 | |
1030 | /* A new thread created by pthread_create inherits the signal mask | |
1031 | * from the parent. To avoid any signal being received by the | |
1032 | * listener thread, we block all signals temporarily in the parent, | |
1033 | * while we create the listener thread. | |
1034 | */ | |
1035 | sigfillset(&sig_all_blocked); | |
1036 | ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask); | |
1037 | if (ret) { | |
d94d802c | 1038 | ERR("pthread_sigmask: %s", strerror(ret)); |
ae6a58bf WP |
1039 | } |
1040 | ||
1879f67f MG |
1041 | ret = pthread_attr_init(&thread_attr); |
1042 | if (ret) { | |
1043 | ERR("pthread_attr_init: %s", strerror(ret)); | |
1044 | } | |
1045 | ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED); | |
1046 | if (ret) { | |
1047 | ERR("pthread_attr_setdetachstate: %s", strerror(ret)); | |
1048 | } | |
1049 | ||
f70c1cd1 | 1050 | ust_lock(); |
1879f67f | 1051 | ret = pthread_create(&global_apps.ust_listener, &thread_attr, |
dde70ea0 | 1052 | ust_listener_thread, &global_apps); |
d94d802c MD |
1053 | if (ret) { |
1054 | ERR("pthread_create global: %s", strerror(ret)); | |
1055 | } | |
f70c1cd1 MD |
1056 | global_apps.thread_active = 1; |
1057 | ust_unlock(); | |
1058 | ||
8d20bf54 | 1059 | if (local_apps.allowed) { |
f70c1cd1 | 1060 | ust_lock(); |
1879f67f | 1061 | ret = pthread_create(&local_apps.ust_listener, &thread_attr, |
dde70ea0 | 1062 | ust_listener_thread, &local_apps); |
d94d802c MD |
1063 | if (ret) { |
1064 | ERR("pthread_create local: %s", strerror(ret)); | |
1065 | } | |
f70c1cd1 MD |
1066 | local_apps.thread_active = 1; |
1067 | ust_unlock(); | |
8d20bf54 MD |
1068 | } else { |
1069 | handle_register_done(&local_apps); | |
1070 | } | |
1879f67f MG |
1071 | ret = pthread_attr_destroy(&thread_attr); |
1072 | if (ret) { | |
1073 | ERR("pthread_attr_destroy: %s", strerror(ret)); | |
1074 | } | |
8d20bf54 | 1075 | |
ae6a58bf WP |
1076 | /* Restore original signal mask in parent */ |
1077 | ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL); | |
1078 | if (ret) { | |
d94d802c | 1079 | ERR("pthread_sigmask: %s", strerror(ret)); |
ae6a58bf WP |
1080 | } |
1081 | ||
cf12a773 MD |
1082 | switch (timeout_mode) { |
1083 | case 1: /* timeout wait */ | |
95259bd0 MD |
1084 | do { |
1085 | ret = sem_timedwait(&constructor_wait, | |
1086 | &constructor_timeout); | |
1087 | } while (ret < 0 && errno == EINTR); | |
cf12a773 | 1088 | if (ret < 0 && errno == ETIMEDOUT) { |
7dd08bec | 1089 | ERR("Timed out waiting for lttng-sessiond"); |
cf12a773 MD |
1090 | } else { |
1091 | assert(!ret); | |
1092 | } | |
1093 | break; | |
7b766b16 | 1094 | case -1:/* wait forever */ |
95259bd0 MD |
1095 | do { |
1096 | ret = sem_wait(&constructor_wait); | |
1097 | } while (ret < 0 && errno == EINTR); | |
11ff9c7d | 1098 | assert(!ret); |
cf12a773 | 1099 | break; |
7b766b16 | 1100 | case 0: /* no timeout */ |
cf12a773 | 1101 | break; |
11ff9c7d | 1102 | } |
2691221a MD |
1103 | } |
1104 | ||
17dfb34b MD |
1105 | static |
1106 | void lttng_ust_cleanup(int exiting) | |
1107 | { | |
efe0de09 | 1108 | cleanup_sock_info(&global_apps, exiting); |
17dfb34b | 1109 | if (local_apps.allowed) { |
efe0de09 | 1110 | cleanup_sock_info(&local_apps, exiting); |
17dfb34b | 1111 | } |
efe0de09 MD |
1112 | /* |
1113 | * The teardown in this function all affect data structures | |
1114 | * accessed under the UST lock by the listener thread. This | |
1115 | * lock, along with the lttng_ust_comm_should_quit flag, ensure | |
1116 | * that none of these threads are accessing this data at this | |
1117 | * point. | |
1118 | */ | |
17dfb34b | 1119 | lttng_ust_abi_exit(); |
003fedf4 | 1120 | lttng_ust_events_exit(); |
7dd08bec MD |
1121 | lttng_ring_buffer_client_discard_exit(); |
1122 | lttng_ring_buffer_client_overwrite_exit(); | |
1123 | lttng_ring_buffer_metadata_client_exit(); | |
17dfb34b MD |
1124 | exit_tracepoint(); |
1125 | if (!exiting) { | |
1126 | /* Reinitialize values for fork */ | |
1127 | sem_count = 2; | |
1128 | lttng_ust_comm_should_quit = 0; | |
1129 | initialized = 0; | |
1130 | } | |
1131 | } | |
1132 | ||
edaa1431 | 1133 | void __attribute__((destructor)) lttng_ust_exit(void) |
2691221a MD |
1134 | { |
1135 | int ret; | |
1136 | ||
9eb62b9c MD |
1137 | /* |
1138 | * Using pthread_cancel here because: | |
1139 | * A) we don't want to hang application teardown. | |
1140 | * B) the thread is not allocating any resource. | |
1141 | */ | |
1ea11eab MD |
1142 | |
1143 | /* | |
1144 | * Require the communication thread to quit. Synchronize with | |
1145 | * mutexes to ensure it is not in a mutex critical section when | |
1146 | * pthread_cancel is later called. | |
1147 | */ | |
17dfb34b | 1148 | ust_lock(); |
1ea11eab | 1149 | lttng_ust_comm_should_quit = 1; |
1ea11eab | 1150 | |
f5f94532 | 1151 | /* cancel threads */ |
f70c1cd1 MD |
1152 | if (global_apps.thread_active) { |
1153 | ret = pthread_cancel(global_apps.ust_listener); | |
1154 | if (ret) { | |
1155 | ERR("Error cancelling global ust listener thread: %s", | |
1156 | strerror(ret)); | |
1157 | } else { | |
1158 | global_apps.thread_active = 0; | |
1159 | } | |
2691221a | 1160 | } |
f70c1cd1 | 1161 | if (local_apps.thread_active) { |
8d20bf54 MD |
1162 | ret = pthread_cancel(local_apps.ust_listener); |
1163 | if (ret) { | |
d94d802c MD |
1164 | ERR("Error cancelling local ust listener thread: %s", |
1165 | strerror(ret)); | |
f70c1cd1 MD |
1166 | } else { |
1167 | local_apps.thread_active = 0; | |
8d20bf54 | 1168 | } |
8d20bf54 | 1169 | } |
f70c1cd1 MD |
1170 | ust_unlock(); |
1171 | ||
efe0de09 MD |
1172 | /* |
1173 | * Do NOT join threads: use of sys_futex makes it impossible to | |
1174 | * join the threads without using async-cancel, but async-cancel | |
1175 | * is delivered by a signal, which could hit the target thread | |
1176 | * anywhere in its code path, including while the ust_lock() is | |
1177 | * held, causing a deadlock for the other thread. Let the OS | |
1178 | * cleanup the threads if there are stalled in a syscall. | |
1179 | */ | |
17dfb34b | 1180 | lttng_ust_cleanup(1); |
2691221a | 1181 | } |
e822f505 MD |
1182 | |
1183 | /* | |
1184 | * We exclude the worker threads across fork and clone (except | |
1185 | * CLONE_VM), because these system calls only keep the forking thread | |
1186 | * running in the child. Therefore, we don't want to call fork or clone | |
1187 | * in the middle of an tracepoint or ust tracing state modification. | |
1188 | * Holding this mutex protects these structures across fork and clone. | |
1189 | */ | |
b728d87e | 1190 | void ust_before_fork(sigset_t *save_sigset) |
e822f505 MD |
1191 | { |
1192 | /* | |
1193 | * Disable signals. This is to avoid that the child intervenes | |
1194 | * before it is properly setup for tracing. It is safer to | |
1195 | * disable all signals, because then we know we are not breaking | |
1196 | * anything by restoring the original mask. | |
1197 | */ | |
1198 | sigset_t all_sigs; | |
1199 | int ret; | |
1200 | ||
e8508a49 MD |
1201 | if (lttng_ust_nest_count) |
1202 | return; | |
e822f505 MD |
1203 | /* Disable signals */ |
1204 | sigfillset(&all_sigs); | |
b728d87e | 1205 | ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset); |
e822f505 MD |
1206 | if (ret == -1) { |
1207 | PERROR("sigprocmask"); | |
1208 | } | |
17dfb34b | 1209 | ust_lock(); |
e822f505 MD |
1210 | rcu_bp_before_fork(); |
1211 | } | |
1212 | ||
b728d87e | 1213 | static void ust_after_fork_common(sigset_t *restore_sigset) |
e822f505 MD |
1214 | { |
1215 | int ret; | |
1216 | ||
17dfb34b MD |
1217 | DBG("process %d", getpid()); |
1218 | ust_unlock(); | |
e822f505 | 1219 | /* Restore signals */ |
23c8854a | 1220 | ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL); |
e822f505 MD |
1221 | if (ret == -1) { |
1222 | PERROR("sigprocmask"); | |
1223 | } | |
1224 | } | |
1225 | ||
b728d87e | 1226 | void ust_after_fork_parent(sigset_t *restore_sigset) |
e822f505 | 1227 | { |
e8508a49 MD |
1228 | if (lttng_ust_nest_count) |
1229 | return; | |
17dfb34b | 1230 | DBG("process %d", getpid()); |
e822f505 MD |
1231 | rcu_bp_after_fork_parent(); |
1232 | /* Release mutexes and reenable signals */ | |
b728d87e | 1233 | ust_after_fork_common(restore_sigset); |
e822f505 MD |
1234 | } |
1235 | ||
17dfb34b MD |
1236 | /* |
1237 | * After fork, in the child, we need to cleanup all the leftover state, | |
1238 | * except the worker thread which already magically disappeared thanks | |
1239 | * to the weird Linux fork semantics. After tyding up, we call | |
1240 | * lttng_ust_init() again to start over as a new PID. | |
1241 | * | |
1242 | * This is meant for forks() that have tracing in the child between the | |
1243 | * fork and following exec call (if there is any). | |
1244 | */ | |
b728d87e | 1245 | void ust_after_fork_child(sigset_t *restore_sigset) |
e822f505 | 1246 | { |
e8508a49 MD |
1247 | if (lttng_ust_nest_count) |
1248 | return; | |
17dfb34b | 1249 | DBG("process %d", getpid()); |
e822f505 MD |
1250 | /* Release urcu mutexes */ |
1251 | rcu_bp_after_fork_child(); | |
17dfb34b | 1252 | lttng_ust_cleanup(0); |
a93bfc45 | 1253 | lttng_context_vtid_reset(); |
e822f505 | 1254 | /* Release mutexes and reenable signals */ |
b728d87e | 1255 | ust_after_fork_common(restore_sigset); |
318dfea9 | 1256 | lttng_ust_init(); |
e822f505 | 1257 | } |