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> |
74d81a6c | 45 | #include <lttng/ust-ctl.h> |
8c90a710 | 46 | #include <urcu/tls-compat.h> |
44c72f10 MD |
47 | #include <ust-comm.h> |
48 | #include <usterr-signal-safe.h> | |
cd54f6d9 | 49 | #include <helper.h> |
44c72f10 | 50 | #include "tracepoint-internal.h" |
7dd08bec | 51 | #include "lttng-tracer-core.h" |
08114193 | 52 | #include "compat.h" |
f645cfa7 | 53 | #include "../libringbuffer/tlsfixup.h" |
edaa1431 MD |
54 | |
55 | /* | |
56 | * Has lttng ust comm constructor been called ? | |
57 | */ | |
58 | static int initialized; | |
59 | ||
1ea11eab | 60 | /* |
17dfb34b MD |
61 | * The ust_lock/ust_unlock lock is used as a communication thread mutex. |
62 | * Held when handling a command, also held by fork() to deal with | |
63 | * removal of threads, and by exit path. | |
1ea11eab | 64 | */ |
1ea11eab MD |
65 | |
66 | /* Should the ust comm thread quit ? */ | |
67 | static int lttng_ust_comm_should_quit; | |
68 | ||
11ff9c7d MD |
69 | /* |
70 | * Wait for either of these before continuing to the main | |
71 | * program: | |
72 | * - the register_done message from sessiond daemon | |
73 | * (will let the sessiond daemon enable sessions before main | |
74 | * starts.) | |
75 | * - sessiond daemon is not reachable. | |
76 | * - timeout (ensuring applications are resilient to session | |
77 | * daemon problems). | |
78 | */ | |
79 | static sem_t constructor_wait; | |
950aab0c MD |
80 | /* |
81 | * Doing this for both the global and local sessiond. | |
82 | */ | |
95259bd0 | 83 | static int sem_count = { 2 }; |
11ff9c7d | 84 | |
e8508a49 MD |
85 | /* |
86 | * Counting nesting within lttng-ust. Used to ensure that calling fork() | |
87 | * from liblttng-ust does not execute the pre/post fork handlers. | |
88 | */ | |
8c90a710 | 89 | static DEFINE_URCU_TLS(int, lttng_ust_nest_count); |
e8508a49 | 90 | |
1ea11eab MD |
91 | /* |
92 | * Info about socket and associated listener thread. | |
93 | */ | |
94 | struct sock_info { | |
11ff9c7d | 95 | const char *name; |
1ea11eab | 96 | pthread_t ust_listener; /* listener thread */ |
46050b1a | 97 | int root_handle; |
8d20bf54 MD |
98 | int constructor_sem_posted; |
99 | int allowed; | |
44e073f5 | 100 | int global; |
e33f3265 | 101 | int thread_active; |
7fc90dca MD |
102 | |
103 | char sock_path[PATH_MAX]; | |
104 | int socket; | |
32ce8569 | 105 | int notify_socket; |
7fc90dca MD |
106 | |
107 | char wait_shm_path[PATH_MAX]; | |
108 | char *wait_shm_mmap; | |
1ea11eab | 109 | }; |
2691221a MD |
110 | |
111 | /* Socket from app (connect) to session daemon (listen) for communication */ | |
1ea11eab | 112 | struct sock_info global_apps = { |
11ff9c7d | 113 | .name = "global", |
44e073f5 | 114 | .global = 1, |
7fc90dca | 115 | |
46050b1a | 116 | .root_handle = -1, |
8d20bf54 | 117 | .allowed = 1, |
e33f3265 | 118 | .thread_active = 0, |
7fc90dca | 119 | |
32ce8569 | 120 | .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME, |
7fc90dca | 121 | .socket = -1, |
32ce8569 | 122 | .notify_socket = -1, |
7fc90dca | 123 | |
32ce8569 | 124 | .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME, |
1ea11eab | 125 | }; |
2691221a MD |
126 | |
127 | /* TODO: allow global_apps_sock_path override */ | |
128 | ||
1ea11eab | 129 | struct sock_info local_apps = { |
11ff9c7d | 130 | .name = "local", |
44e073f5 | 131 | .global = 0, |
46050b1a | 132 | .root_handle = -1, |
8d20bf54 | 133 | .allowed = 0, /* Check setuid bit first */ |
e33f3265 | 134 | .thread_active = 0, |
7fc90dca MD |
135 | |
136 | .socket = -1, | |
32ce8569 | 137 | .notify_socket = -1, |
1ea11eab | 138 | }; |
2691221a | 139 | |
37ed587a MD |
140 | static int wait_poll_fallback; |
141 | ||
74d81a6c MD |
142 | static const char *cmd_name_mapping[] = { |
143 | [ LTTNG_UST_RELEASE ] = "Release", | |
144 | [ LTTNG_UST_SESSION ] = "Create Session", | |
145 | [ LTTNG_UST_TRACER_VERSION ] = "Get Tracer Version", | |
146 | ||
147 | [ LTTNG_UST_TRACEPOINT_LIST ] = "Create Tracepoint List", | |
148 | [ LTTNG_UST_WAIT_QUIESCENT ] = "Wait for Quiescent State", | |
149 | [ LTTNG_UST_REGISTER_DONE ] = "Registration Done", | |
150 | [ LTTNG_UST_TRACEPOINT_FIELD_LIST ] = "Create Tracepoint Field List", | |
151 | ||
152 | /* Session FD commands */ | |
153 | [ LTTNG_UST_CHANNEL ] = "Create Channel", | |
154 | [ LTTNG_UST_SESSION_START ] = "Start Session", | |
155 | [ LTTNG_UST_SESSION_STOP ] = "Stop Session", | |
156 | ||
157 | /* Channel FD commands */ | |
158 | [ LTTNG_UST_STREAM ] = "Create Stream", | |
159 | [ LTTNG_UST_EVENT ] = "Create Event", | |
160 | ||
161 | /* Event and Channel FD commands */ | |
162 | [ LTTNG_UST_CONTEXT ] = "Create Context", | |
163 | [ LTTNG_UST_FLUSH_BUFFER ] = "Flush Buffer", | |
164 | ||
165 | /* Event, Channel and Session commands */ | |
166 | [ LTTNG_UST_ENABLE ] = "Enable", | |
167 | [ LTTNG_UST_DISABLE ] = "Disable", | |
168 | ||
169 | /* Tracepoint list commands */ | |
170 | [ LTTNG_UST_TRACEPOINT_LIST_GET ] = "List Next Tracepoint", | |
171 | [ LTTNG_UST_TRACEPOINT_FIELD_LIST_GET ] = "List Next Tracepoint Field", | |
172 | ||
173 | /* Event FD commands */ | |
174 | [ LTTNG_UST_FILTER ] = "Create Filter", | |
175 | }; | |
176 | ||
ff517991 MD |
177 | static const char *str_timeout; |
178 | static int got_timeout_env; | |
179 | ||
7dd08bec | 180 | extern void lttng_ring_buffer_client_overwrite_init(void); |
34a91bdb | 181 | extern void lttng_ring_buffer_client_overwrite_rt_init(void); |
7dd08bec | 182 | extern void lttng_ring_buffer_client_discard_init(void); |
34a91bdb | 183 | extern void lttng_ring_buffer_client_discard_rt_init(void); |
7dd08bec MD |
184 | extern void lttng_ring_buffer_metadata_client_init(void); |
185 | extern void lttng_ring_buffer_client_overwrite_exit(void); | |
34a91bdb | 186 | extern void lttng_ring_buffer_client_overwrite_rt_exit(void); |
7dd08bec | 187 | extern void lttng_ring_buffer_client_discard_exit(void); |
34a91bdb | 188 | extern void lttng_ring_buffer_client_discard_rt_exit(void); |
7dd08bec | 189 | extern void lttng_ring_buffer_metadata_client_exit(void); |
edaa1431 | 190 | |
3c6f6263 AM |
191 | /* |
192 | * Returns the HOME directory path. Caller MUST NOT free(3) the returned | |
193 | * pointer. | |
194 | */ | |
195 | static | |
196 | const char *get_lttng_home_dir(void) | |
197 | { | |
198 | const char *val; | |
199 | ||
200 | val = (const char *) getenv("LTTNG_HOME"); | |
201 | if (val != NULL) { | |
202 | return val; | |
203 | } | |
204 | return (const char *) getenv("HOME"); | |
205 | } | |
206 | ||
a903623f MD |
207 | /* |
208 | * Force a read (imply TLS fixup for dlopen) of TLS variables. | |
209 | */ | |
210 | static | |
211 | void lttng_fixup_nest_count_tls(void) | |
212 | { | |
8c90a710 | 213 | asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count))); |
a903623f MD |
214 | } |
215 | ||
32ce8569 MD |
216 | int lttng_get_notify_socket(void *owner) |
217 | { | |
218 | struct sock_info *info = owner; | |
219 | ||
220 | return info->notify_socket; | |
221 | } | |
222 | ||
74d81a6c MD |
223 | static |
224 | void print_cmd(int cmd, int handle) | |
225 | { | |
226 | const char *cmd_name = "Unknown"; | |
227 | ||
fd67a004 MD |
228 | if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping) |
229 | && cmd_name_mapping[cmd]) { | |
74d81a6c MD |
230 | cmd_name = cmd_name_mapping[cmd]; |
231 | } | |
fd67a004 MD |
232 | DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)", |
233 | cmd_name, cmd, | |
74d81a6c MD |
234 | lttng_ust_obj_get_name(handle), handle); |
235 | } | |
236 | ||
2691221a | 237 | static |
8d20bf54 | 238 | int setup_local_apps(void) |
2691221a MD |
239 | { |
240 | const char *home_dir; | |
7fc90dca | 241 | uid_t uid; |
2691221a | 242 | |
7fc90dca | 243 | uid = getuid(); |
8d20bf54 MD |
244 | /* |
245 | * Disallow per-user tracing for setuid binaries. | |
246 | */ | |
7fc90dca | 247 | if (uid != geteuid()) { |
9ec6895c | 248 | assert(local_apps.allowed == 0); |
d0a1ae63 | 249 | return 0; |
8d20bf54 | 250 | } |
3c6f6263 | 251 | home_dir = get_lttng_home_dir(); |
9ec6895c MD |
252 | if (!home_dir) { |
253 | WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing."); | |
254 | assert(local_apps.allowed == 0); | |
2691221a | 255 | return -ENOENT; |
9ec6895c MD |
256 | } |
257 | local_apps.allowed = 1; | |
32ce8569 MD |
258 | snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s", |
259 | home_dir, | |
260 | LTTNG_DEFAULT_HOME_RUNDIR, | |
261 | LTTNG_UST_SOCK_FILENAME); | |
262 | snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u", | |
263 | LTTNG_UST_WAIT_FILENAME, | |
264 | uid); | |
2691221a MD |
265 | return 0; |
266 | } | |
267 | ||
ff517991 MD |
268 | /* |
269 | * Get notify_sock timeout, in ms. | |
270 | * -1: don't wait. 0: wait forever. >0: timeout, in ms. | |
271 | */ | |
272 | static | |
273 | long get_timeout(void) | |
274 | { | |
275 | long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS; | |
276 | ||
277 | if (!got_timeout_env) { | |
278 | str_timeout = getenv("LTTNG_UST_REGISTER_TIMEOUT"); | |
279 | got_timeout_env = 1; | |
280 | } | |
281 | if (str_timeout) | |
282 | constructor_delay_ms = strtol(str_timeout, NULL, 10); | |
283 | return constructor_delay_ms; | |
284 | } | |
285 | ||
286 | static | |
287 | long get_notify_sock_timeout(void) | |
288 | { | |
289 | return get_timeout(); | |
290 | } | |
291 | ||
292 | /* | |
293 | * Return values: -1: don't wait. 0: wait forever. 1: timeout wait. | |
294 | */ | |
295 | static | |
296 | int get_constructor_timeout(struct timespec *constructor_timeout) | |
297 | { | |
298 | long constructor_delay_ms; | |
299 | int ret; | |
300 | ||
301 | constructor_delay_ms = get_timeout(); | |
302 | ||
303 | switch (constructor_delay_ms) { | |
304 | case -1:/* fall-through */ | |
305 | case 0: | |
306 | return constructor_delay_ms; | |
307 | default: | |
308 | break; | |
309 | } | |
310 | ||
311 | /* | |
312 | * If we are unable to find the current time, don't wait. | |
313 | */ | |
314 | ret = clock_gettime(CLOCK_REALTIME, constructor_timeout); | |
315 | if (ret) { | |
316 | return -1; | |
317 | } | |
318 | constructor_timeout->tv_sec += constructor_delay_ms / 1000UL; | |
319 | constructor_timeout->tv_nsec += | |
320 | (constructor_delay_ms % 1000UL) * 1000000UL; | |
321 | if (constructor_timeout->tv_nsec >= 1000000000UL) { | |
322 | constructor_timeout->tv_sec++; | |
323 | constructor_timeout->tv_nsec -= 1000000000UL; | |
324 | } | |
325 | return 1; | |
326 | } | |
327 | ||
2691221a | 328 | static |
32ce8569 | 329 | int register_to_sessiond(int socket, enum ustctl_socket_type type) |
2691221a | 330 | { |
32ce8569 MD |
331 | return ustcomm_send_reg_msg(socket, |
332 | type, | |
333 | CAA_BITS_PER_LONG, | |
334 | lttng_alignof(uint8_t) * CHAR_BIT, | |
335 | lttng_alignof(uint16_t) * CHAR_BIT, | |
336 | lttng_alignof(uint32_t) * CHAR_BIT, | |
337 | lttng_alignof(uint64_t) * CHAR_BIT, | |
338 | lttng_alignof(unsigned long) * CHAR_BIT); | |
2691221a MD |
339 | } |
340 | ||
d9e99d10 | 341 | static |
57773204 | 342 | int send_reply(int sock, struct ustcomm_ust_reply *lur) |
d9e99d10 | 343 | { |
9eb62b9c | 344 | ssize_t len; |
d3a492d1 | 345 | |
57773204 | 346 | len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur)); |
d3a492d1 | 347 | switch (len) { |
a4be8962 | 348 | case sizeof(*lur): |
d3a492d1 MD |
349 | DBG("message successfully sent"); |
350 | return 0; | |
7bc53e94 MD |
351 | default: |
352 | if (len == -ECONNRESET) { | |
353 | DBG("remote end closed connection"); | |
d3a492d1 MD |
354 | return 0; |
355 | } | |
7bc53e94 MD |
356 | if (len < 0) |
357 | return len; | |
358 | DBG("incorrect message size: %zd", len); | |
359 | return -EINVAL; | |
d3a492d1 MD |
360 | } |
361 | } | |
362 | ||
363 | static | |
edaa1431 | 364 | int handle_register_done(struct sock_info *sock_info) |
11ff9c7d MD |
365 | { |
366 | int ret; | |
367 | ||
edaa1431 MD |
368 | if (sock_info->constructor_sem_posted) |
369 | return 0; | |
370 | sock_info->constructor_sem_posted = 1; | |
56cd7e2f MD |
371 | if (uatomic_read(&sem_count) <= 0) { |
372 | return 0; | |
373 | } | |
95259bd0 MD |
374 | ret = uatomic_add_return(&sem_count, -1); |
375 | if (ret == 0) { | |
376 | ret = sem_post(&constructor_wait); | |
377 | assert(!ret); | |
378 | } | |
11ff9c7d MD |
379 | return 0; |
380 | } | |
381 | ||
382 | static | |
383 | int handle_message(struct sock_info *sock_info, | |
57773204 | 384 | int sock, struct ustcomm_ust_msg *lum) |
d3a492d1 | 385 | { |
1ea11eab | 386 | int ret = 0; |
b61ce3b2 | 387 | const struct lttng_ust_objd_ops *ops; |
57773204 | 388 | struct ustcomm_ust_reply lur; |
ef9ff354 | 389 | union ust_args args; |
40003310 | 390 | ssize_t len; |
1ea11eab | 391 | |
17dfb34b | 392 | ust_lock(); |
1ea11eab | 393 | |
46050b1a MD |
394 | memset(&lur, 0, sizeof(lur)); |
395 | ||
1ea11eab | 396 | if (lttng_ust_comm_should_quit) { |
74d81a6c | 397 | ret = -LTTNG_UST_ERR_EXITING; |
1ea11eab MD |
398 | goto end; |
399 | } | |
9eb62b9c | 400 | |
46050b1a MD |
401 | ops = objd_ops(lum->handle); |
402 | if (!ops) { | |
403 | ret = -ENOENT; | |
404 | goto end; | |
1ea11eab | 405 | } |
46050b1a MD |
406 | |
407 | switch (lum->cmd) { | |
11ff9c7d MD |
408 | case LTTNG_UST_REGISTER_DONE: |
409 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
edaa1431 | 410 | ret = handle_register_done(sock_info); |
11ff9c7d MD |
411 | else |
412 | ret = -EINVAL; | |
413 | break; | |
46050b1a MD |
414 | case LTTNG_UST_RELEASE: |
415 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
416 | ret = -EPERM; | |
417 | else | |
1849ef7c | 418 | ret = lttng_ust_objd_unref(lum->handle, 1); |
d9e99d10 | 419 | break; |
2d78951a MD |
420 | case LTTNG_UST_FILTER: |
421 | { | |
422 | /* Receive filter data */ | |
f488575f | 423 | struct lttng_ust_filter_bytecode_node *bytecode; |
2d78951a | 424 | |
cd54f6d9 | 425 | if (lum->u.filter.data_size > FILTER_BYTECODE_MAX_LEN) { |
7bc53e94 | 426 | ERR("Filter data size is too large: %u bytes", |
2d78951a MD |
427 | lum->u.filter.data_size); |
428 | ret = -EINVAL; | |
429 | goto error; | |
430 | } | |
2734ca65 | 431 | |
885b1dfd | 432 | if (lum->u.filter.reloc_offset > lum->u.filter.data_size) { |
7bc53e94 | 433 | ERR("Filter reloc offset %u is not within data", |
2734ca65 CB |
434 | lum->u.filter.reloc_offset); |
435 | ret = -EINVAL; | |
436 | goto error; | |
437 | } | |
438 | ||
cd54f6d9 MD |
439 | bytecode = zmalloc(sizeof(*bytecode) + lum->u.filter.data_size); |
440 | if (!bytecode) { | |
441 | ret = -ENOMEM; | |
442 | goto error; | |
443 | } | |
f488575f | 444 | len = ustcomm_recv_unix_sock(sock, bytecode->bc.data, |
2d78951a MD |
445 | lum->u.filter.data_size); |
446 | switch (len) { | |
447 | case 0: /* orderly shutdown */ | |
448 | ret = 0; | |
cd54f6d9 | 449 | free(bytecode); |
2d78951a | 450 | goto error; |
2d78951a MD |
451 | default: |
452 | if (len == lum->u.filter.data_size) { | |
7bc53e94 | 453 | DBG("filter data received"); |
2d78951a | 454 | break; |
7bc53e94 MD |
455 | } else if (len < 0) { |
456 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
457 | if (len == -ECONNRESET) { | |
458 | ERR("%s remote end closed connection", sock_info->name); | |
459 | ret = len; | |
460 | free(bytecode); | |
461 | goto error; | |
462 | } | |
463 | ret = len; | |
eb8bf361 | 464 | free(bytecode); |
7bc53e94 | 465 | goto end; |
2d78951a | 466 | } else { |
7bc53e94 | 467 | DBG("incorrect filter data message size: %zd", len); |
2d78951a | 468 | ret = -EINVAL; |
cd54f6d9 | 469 | free(bytecode); |
2d78951a MD |
470 | goto end; |
471 | } | |
472 | } | |
f488575f MD |
473 | bytecode->bc.len = lum->u.filter.data_size; |
474 | bytecode->bc.reloc_offset = lum->u.filter.reloc_offset; | |
3f6fd224 | 475 | bytecode->bc.seqnum = lum->u.filter.seqnum; |
cd54f6d9 | 476 | if (ops->cmd) { |
2d78951a | 477 | ret = ops->cmd(lum->handle, lum->cmd, |
cd54f6d9 | 478 | (unsigned long) bytecode, |
f59ed768 | 479 | &args, sock_info); |
cd54f6d9 MD |
480 | if (ret) { |
481 | free(bytecode); | |
482 | } | |
483 | /* don't free bytecode if everything went fine. */ | |
484 | } else { | |
2d78951a | 485 | ret = -ENOSYS; |
cd54f6d9 MD |
486 | free(bytecode); |
487 | } | |
2d78951a MD |
488 | break; |
489 | } | |
74d81a6c MD |
490 | case LTTNG_UST_CHANNEL: |
491 | { | |
492 | void *chan_data; | |
ff0f5728 | 493 | int wakeup_fd; |
74d81a6c MD |
494 | |
495 | len = ustcomm_recv_channel_from_sessiond(sock, | |
ff0f5728 MD |
496 | &chan_data, lum->u.channel.len, |
497 | &wakeup_fd); | |
74d81a6c MD |
498 | switch (len) { |
499 | case 0: /* orderly shutdown */ | |
500 | ret = 0; | |
501 | goto error; | |
502 | default: | |
503 | if (len == lum->u.channel.len) { | |
504 | DBG("channel data received"); | |
505 | break; | |
506 | } else if (len < 0) { | |
507 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
508 | if (len == -ECONNRESET) { | |
509 | ERR("%s remote end closed connection", sock_info->name); | |
510 | ret = len; | |
511 | goto error; | |
512 | } | |
513 | ret = len; | |
514 | goto end; | |
515 | } else { | |
516 | DBG("incorrect channel data message size: %zd", len); | |
517 | ret = -EINVAL; | |
518 | goto end; | |
519 | } | |
520 | } | |
521 | args.channel.chan_data = chan_data; | |
ff0f5728 | 522 | args.channel.wakeup_fd = wakeup_fd; |
74d81a6c MD |
523 | if (ops->cmd) |
524 | ret = ops->cmd(lum->handle, lum->cmd, | |
525 | (unsigned long) &lum->u, | |
526 | &args, sock_info); | |
527 | else | |
528 | ret = -ENOSYS; | |
529 | break; | |
530 | } | |
531 | case LTTNG_UST_STREAM: | |
532 | { | |
533 | /* Receive shm_fd, wakeup_fd */ | |
534 | ret = ustcomm_recv_stream_from_sessiond(sock, | |
535 | &lum->u.stream.len, | |
536 | &args.stream.shm_fd, | |
537 | &args.stream.wakeup_fd); | |
538 | if (ret) { | |
539 | goto end; | |
540 | } | |
541 | if (ops->cmd) | |
542 | ret = ops->cmd(lum->handle, lum->cmd, | |
543 | (unsigned long) &lum->u, | |
544 | &args, sock_info); | |
545 | else | |
546 | ret = -ENOSYS; | |
547 | break; | |
548 | } | |
d9e99d10 | 549 | default: |
46050b1a MD |
550 | if (ops->cmd) |
551 | ret = ops->cmd(lum->handle, lum->cmd, | |
ef9ff354 | 552 | (unsigned long) &lum->u, |
f59ed768 | 553 | &args, sock_info); |
46050b1a MD |
554 | else |
555 | ret = -ENOSYS; | |
556 | break; | |
d9e99d10 | 557 | } |
46050b1a | 558 | |
1ea11eab | 559 | end: |
46050b1a MD |
560 | lur.handle = lum->handle; |
561 | lur.cmd = lum->cmd; | |
562 | lur.ret_val = ret; | |
563 | if (ret >= 0) { | |
7bc53e94 | 564 | lur.ret_code = LTTNG_UST_OK; |
46050b1a | 565 | } else { |
7bc53e94 MD |
566 | /* |
567 | * Use -LTTNG_UST_ERR as wildcard for UST internal | |
568 | * error that are not caused by the transport, except if | |
569 | * we already have a more precise error message to | |
570 | * report. | |
571 | */ | |
64b2564e DG |
572 | if (ret > -LTTNG_UST_ERR) { |
573 | /* Translate code to UST error. */ | |
574 | switch (ret) { | |
575 | case -EEXIST: | |
576 | lur.ret_code = -LTTNG_UST_ERR_EXIST; | |
577 | break; | |
578 | case -EINVAL: | |
579 | lur.ret_code = -LTTNG_UST_ERR_INVAL; | |
580 | break; | |
581 | case -ENOENT: | |
582 | lur.ret_code = -LTTNG_UST_ERR_NOENT; | |
583 | break; | |
584 | case -EPERM: | |
585 | lur.ret_code = -LTTNG_UST_ERR_PERM; | |
586 | break; | |
587 | case -ENOSYS: | |
588 | lur.ret_code = -LTTNG_UST_ERR_NOSYS; | |
589 | break; | |
590 | default: | |
591 | lur.ret_code = -LTTNG_UST_ERR; | |
592 | break; | |
593 | } | |
594 | } else { | |
7bc53e94 | 595 | lur.ret_code = ret; |
64b2564e | 596 | } |
46050b1a | 597 | } |
e6ea14c5 MD |
598 | if (ret >= 0) { |
599 | switch (lum->cmd) { | |
e6ea14c5 MD |
600 | case LTTNG_UST_TRACER_VERSION: |
601 | lur.u.version = lum->u.version; | |
602 | break; | |
603 | case LTTNG_UST_TRACEPOINT_LIST_GET: | |
604 | memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint)); | |
605 | break; | |
606 | } | |
381c0f1e | 607 | } |
74d81a6c | 608 | DBG("Return value: %d", lur.ret_val); |
46050b1a | 609 | ret = send_reply(sock, &lur); |
193183fb | 610 | if (ret < 0) { |
7bc53e94 | 611 | DBG("error sending reply"); |
193183fb MD |
612 | goto error; |
613 | } | |
46050b1a | 614 | |
40003310 MD |
615 | /* |
616 | * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field | |
617 | * after the reply. | |
618 | */ | |
7bc53e94 | 619 | if (lur.ret_code == LTTNG_UST_OK) { |
40003310 MD |
620 | switch (lum->cmd) { |
621 | case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET: | |
622 | len = ustcomm_send_unix_sock(sock, | |
623 | &args.field_list.entry, | |
624 | sizeof(args.field_list.entry)); | |
7bc53e94 MD |
625 | if (len < 0) { |
626 | ret = len; | |
627 | goto error; | |
628 | } | |
40003310 | 629 | if (len != sizeof(args.field_list.entry)) { |
7bc53e94 | 630 | ret = -EINVAL; |
40003310 MD |
631 | goto error; |
632 | } | |
633 | } | |
634 | } | |
ef9ff354 | 635 | |
381c0f1e | 636 | error: |
17dfb34b | 637 | ust_unlock(); |
1ea11eab | 638 | return ret; |
d9e99d10 MD |
639 | } |
640 | ||
46050b1a | 641 | static |
efe0de09 | 642 | void cleanup_sock_info(struct sock_info *sock_info, int exiting) |
46050b1a MD |
643 | { |
644 | int ret; | |
645 | ||
5b14aab3 MD |
646 | if (sock_info->root_handle != -1) { |
647 | ret = lttng_ust_objd_unref(sock_info->root_handle, 1); | |
648 | if (ret) { | |
649 | ERR("Error unref root handle"); | |
650 | } | |
651 | sock_info->root_handle = -1; | |
652 | } | |
653 | sock_info->constructor_sem_posted = 0; | |
654 | ||
655 | /* | |
656 | * wait_shm_mmap, socket and notify socket are used by listener | |
657 | * threads outside of the ust lock, so we cannot tear them down | |
658 | * ourselves, because we cannot join on these threads. Leave | |
659 | * responsibility of cleaning up these resources to the OS | |
660 | * process exit. | |
661 | */ | |
662 | if (exiting) | |
663 | return; | |
664 | ||
46050b1a | 665 | if (sock_info->socket != -1) { |
e6973a89 | 666 | ret = ustcomm_close_unix_sock(sock_info->socket); |
46050b1a | 667 | if (ret) { |
32ce8569 | 668 | ERR("Error closing ust cmd socket"); |
46050b1a MD |
669 | } |
670 | sock_info->socket = -1; | |
671 | } | |
32ce8569 MD |
672 | if (sock_info->notify_socket != -1) { |
673 | ret = ustcomm_close_unix_sock(sock_info->notify_socket); | |
674 | if (ret) { | |
675 | ERR("Error closing ust notify socket"); | |
676 | } | |
677 | sock_info->notify_socket = -1; | |
678 | } | |
5b14aab3 | 679 | if (sock_info->wait_shm_mmap) { |
7fc90dca MD |
680 | ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE)); |
681 | if (ret) { | |
682 | ERR("Error unmapping wait shm"); | |
683 | } | |
684 | sock_info->wait_shm_mmap = NULL; | |
685 | } | |
686 | } | |
687 | ||
58d4b2a2 | 688 | /* |
33bbeb90 MD |
689 | * Using fork to set umask in the child process (not multi-thread safe). |
690 | * We deal with the shm_open vs ftruncate race (happening when the | |
691 | * sessiond owns the shm and does not let everybody modify it, to ensure | |
692 | * safety against shm_unlink) by simply letting the mmap fail and | |
693 | * retrying after a few seconds. | |
694 | * For global shm, everybody has rw access to it until the sessiond | |
695 | * starts. | |
58d4b2a2 | 696 | */ |
7fc90dca | 697 | static |
58d4b2a2 | 698 | int get_wait_shm(struct sock_info *sock_info, size_t mmap_size) |
7fc90dca | 699 | { |
7fc90dca | 700 | int wait_shm_fd, ret; |
58d4b2a2 | 701 | pid_t pid; |
44e073f5 | 702 | |
58d4b2a2 | 703 | /* |
33bbeb90 | 704 | * Try to open read-only. |
58d4b2a2 | 705 | */ |
33bbeb90 | 706 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 MD |
707 | if (wait_shm_fd >= 0) { |
708 | goto end; | |
709 | } else if (wait_shm_fd < 0 && errno != ENOENT) { | |
710 | /* | |
33bbeb90 MD |
711 | * Real-only open did not work, and it's not because the |
712 | * entry was not present. It's a failure that prohibits | |
713 | * using shm. | |
58d4b2a2 | 714 | */ |
7fc90dca | 715 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
58d4b2a2 | 716 | goto end; |
7fc90dca MD |
717 | } |
718 | /* | |
58d4b2a2 MD |
719 | * If the open failed because the file did not exist, try |
720 | * creating it ourself. | |
7fc90dca | 721 | */ |
8c90a710 | 722 | URCU_TLS(lttng_ust_nest_count)++; |
58d4b2a2 | 723 | pid = fork(); |
8c90a710 | 724 | URCU_TLS(lttng_ust_nest_count)--; |
58d4b2a2 MD |
725 | if (pid > 0) { |
726 | int status; | |
727 | ||
728 | /* | |
729 | * Parent: wait for child to return, in which case the | |
730 | * shared memory map will have been created. | |
731 | */ | |
732 | pid = wait(&status); | |
b7d3cb32 | 733 | if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) { |
58d4b2a2 MD |
734 | wait_shm_fd = -1; |
735 | goto end; | |
7fc90dca | 736 | } |
58d4b2a2 MD |
737 | /* |
738 | * Try to open read-only again after creation. | |
739 | */ | |
33bbeb90 | 740 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 MD |
741 | if (wait_shm_fd < 0) { |
742 | /* | |
743 | * Real-only open did not work. It's a failure | |
744 | * that prohibits using shm. | |
745 | */ | |
746 | ERR("Error opening shm %s", sock_info->wait_shm_path); | |
747 | goto end; | |
748 | } | |
749 | goto end; | |
750 | } else if (pid == 0) { | |
751 | int create_mode; | |
752 | ||
753 | /* Child */ | |
33bbeb90 | 754 | create_mode = S_IRUSR | S_IWUSR | S_IRGRP; |
58d4b2a2 | 755 | if (sock_info->global) |
33bbeb90 | 756 | create_mode |= S_IROTH | S_IWGRP | S_IWOTH; |
58d4b2a2 MD |
757 | /* |
758 | * We're alone in a child process, so we can modify the | |
759 | * process-wide umask. | |
760 | */ | |
33bbeb90 | 761 | umask(~create_mode); |
58d4b2a2 | 762 | /* |
33bbeb90 MD |
763 | * Try creating shm (or get rw access). |
764 | * We don't do an exclusive open, because we allow other | |
765 | * processes to create+ftruncate it concurrently. | |
58d4b2a2 MD |
766 | */ |
767 | wait_shm_fd = shm_open(sock_info->wait_shm_path, | |
768 | O_RDWR | O_CREAT, create_mode); | |
769 | if (wait_shm_fd >= 0) { | |
770 | ret = ftruncate(wait_shm_fd, mmap_size); | |
771 | if (ret) { | |
772 | PERROR("ftruncate"); | |
b0c1425d | 773 | _exit(EXIT_FAILURE); |
58d4b2a2 | 774 | } |
b0c1425d | 775 | _exit(EXIT_SUCCESS); |
58d4b2a2 | 776 | } |
33bbeb90 MD |
777 | /* |
778 | * For local shm, we need to have rw access to accept | |
779 | * opening it: this means the local sessiond will be | |
780 | * able to wake us up. For global shm, we open it even | |
781 | * if rw access is not granted, because the root.root | |
782 | * sessiond will be able to override all rights and wake | |
783 | * us up. | |
784 | */ | |
785 | if (!sock_info->global && errno != EACCES) { | |
58d4b2a2 | 786 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
5d3bc5ed | 787 | _exit(EXIT_FAILURE); |
58d4b2a2 MD |
788 | } |
789 | /* | |
33bbeb90 MD |
790 | * The shm exists, but we cannot open it RW. Report |
791 | * success. | |
58d4b2a2 | 792 | */ |
5d3bc5ed | 793 | _exit(EXIT_SUCCESS); |
58d4b2a2 MD |
794 | } else { |
795 | return -1; | |
7fc90dca | 796 | } |
58d4b2a2 | 797 | end: |
33bbeb90 MD |
798 | if (wait_shm_fd >= 0 && !sock_info->global) { |
799 | struct stat statbuf; | |
800 | ||
801 | /* | |
802 | * Ensure that our user is the owner of the shm file for | |
803 | * local shm. If we do not own the file, it means our | |
804 | * sessiond will not have access to wake us up (there is | |
805 | * probably a rogue process trying to fake our | |
806 | * sessiond). Fallback to polling method in this case. | |
807 | */ | |
808 | ret = fstat(wait_shm_fd, &statbuf); | |
809 | if (ret) { | |
810 | PERROR("fstat"); | |
811 | goto error_close; | |
812 | } | |
813 | if (statbuf.st_uid != getuid()) | |
814 | goto error_close; | |
815 | } | |
58d4b2a2 | 816 | return wait_shm_fd; |
33bbeb90 MD |
817 | |
818 | error_close: | |
819 | ret = close(wait_shm_fd); | |
820 | if (ret) { | |
821 | PERROR("Error closing fd"); | |
822 | } | |
823 | return -1; | |
58d4b2a2 MD |
824 | } |
825 | ||
826 | static | |
827 | char *get_map_shm(struct sock_info *sock_info) | |
828 | { | |
829 | size_t mmap_size = sysconf(_SC_PAGE_SIZE); | |
830 | int wait_shm_fd, ret; | |
831 | char *wait_shm_mmap; | |
832 | ||
833 | wait_shm_fd = get_wait_shm(sock_info, mmap_size); | |
834 | if (wait_shm_fd < 0) { | |
835 | goto error; | |
44e073f5 | 836 | } |
7fc90dca MD |
837 | wait_shm_mmap = mmap(NULL, mmap_size, PROT_READ, |
838 | MAP_SHARED, wait_shm_fd, 0); | |
7fc90dca MD |
839 | /* close shm fd immediately after taking the mmap reference */ |
840 | ret = close(wait_shm_fd); | |
841 | if (ret) { | |
33bbeb90 MD |
842 | PERROR("Error closing fd"); |
843 | } | |
844 | if (wait_shm_mmap == MAP_FAILED) { | |
845 | DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode."); | |
846 | goto error; | |
7fc90dca MD |
847 | } |
848 | return wait_shm_mmap; | |
849 | ||
850 | error: | |
851 | return NULL; | |
852 | } | |
853 | ||
854 | static | |
855 | void wait_for_sessiond(struct sock_info *sock_info) | |
856 | { | |
efe0de09 | 857 | int ret; |
80e2814b | 858 | |
7fc90dca MD |
859 | ust_lock(); |
860 | if (lttng_ust_comm_should_quit) { | |
861 | goto quit; | |
862 | } | |
37ed587a MD |
863 | if (wait_poll_fallback) { |
864 | goto error; | |
865 | } | |
7fc90dca MD |
866 | if (!sock_info->wait_shm_mmap) { |
867 | sock_info->wait_shm_mmap = get_map_shm(sock_info); | |
868 | if (!sock_info->wait_shm_mmap) | |
869 | goto error; | |
870 | } | |
871 | ust_unlock(); | |
872 | ||
873 | DBG("Waiting for %s apps sessiond", sock_info->name); | |
80e2814b MD |
874 | /* Wait for futex wakeup */ |
875 | if (uatomic_read((int32_t *) sock_info->wait_shm_mmap) == 0) { | |
876 | ret = futex_async((int32_t *) sock_info->wait_shm_mmap, | |
877 | FUTEX_WAIT, 0, NULL, NULL, 0); | |
80e2814b | 878 | if (ret < 0) { |
37ed587a MD |
879 | if (errno == EFAULT) { |
880 | wait_poll_fallback = 1; | |
a8b870ad | 881 | DBG( |
37ed587a MD |
882 | "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) " |
883 | "do not support FUTEX_WAKE on read-only memory mappings correctly. " | |
884 | "Please upgrade your kernel " | |
885 | "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel " | |
886 | "mainline). LTTng-UST will use polling mode fallback."); | |
cd27263b MD |
887 | if (ust_debug()) |
888 | PERROR("futex"); | |
37ed587a | 889 | } |
80e2814b MD |
890 | } |
891 | } | |
7fc90dca MD |
892 | return; |
893 | ||
894 | quit: | |
895 | ust_unlock(); | |
896 | return; | |
897 | ||
898 | error: | |
899 | ust_unlock(); | |
7fc90dca | 900 | return; |
46050b1a MD |
901 | } |
902 | ||
1ea11eab MD |
903 | /* |
904 | * This thread does not allocate any resource, except within | |
905 | * handle_message, within mutex protection. This mutex protects against | |
906 | * fork and exit. | |
98bf993f | 907 | * The other moment it allocates resources is at socket connection, which |
1ea11eab MD |
908 | * is also protected by the mutex. |
909 | */ | |
d9e99d10 MD |
910 | static |
911 | void *ust_listener_thread(void *arg) | |
912 | { | |
1ea11eab | 913 | struct sock_info *sock_info = arg; |
c0eedf81 | 914 | int sock, ret, prev_connect_failed = 0, has_waited = 0; |
32ce8569 MD |
915 | int open_sock[2]; |
916 | int i; | |
ff517991 | 917 | long timeout; |
d9e99d10 | 918 | |
9eb62b9c MD |
919 | /* Restart trying to connect to the session daemon */ |
920 | restart: | |
c0eedf81 MD |
921 | if (prev_connect_failed) { |
922 | /* Wait for sessiond availability with pipe */ | |
923 | wait_for_sessiond(sock_info); | |
924 | if (has_waited) { | |
925 | has_waited = 0; | |
926 | /* | |
927 | * Sleep for 5 seconds before retrying after a | |
928 | * sequence of failure / wait / failure. This | |
929 | * deals with a killed or broken session daemon. | |
930 | */ | |
931 | sleep(5); | |
932 | } | |
933 | has_waited = 1; | |
934 | prev_connect_failed = 0; | |
935 | } | |
9eb62b9c | 936 | |
1ea11eab | 937 | if (sock_info->socket != -1) { |
e6973a89 | 938 | ret = ustcomm_close_unix_sock(sock_info->socket); |
1ea11eab | 939 | if (ret) { |
32ce8569 MD |
940 | ERR("Error closing %s ust cmd socket", |
941 | sock_info->name); | |
1ea11eab MD |
942 | } |
943 | sock_info->socket = -1; | |
944 | } | |
32ce8569 MD |
945 | if (sock_info->notify_socket != -1) { |
946 | ret = ustcomm_close_unix_sock(sock_info->notify_socket); | |
947 | if (ret) { | |
948 | ERR("Error closing %s ust notify socket", | |
949 | sock_info->name); | |
950 | } | |
951 | sock_info->notify_socket = -1; | |
952 | } | |
46050b1a | 953 | |
9eb62b9c | 954 | /* Register */ |
32ce8569 MD |
955 | for (i = 0; i < 2; i++) { |
956 | ret = ustcomm_connect_unix_sock(sock_info->sock_path); | |
957 | if (ret < 0) { | |
958 | DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name); | |
959 | prev_connect_failed = 1; | |
5b14aab3 MD |
960 | |
961 | ust_lock(); | |
962 | ||
963 | if (lttng_ust_comm_should_quit) { | |
964 | goto quit; | |
965 | } | |
966 | ||
32ce8569 MD |
967 | /* |
968 | * If we cannot find the sessiond daemon, don't delay | |
969 | * constructor execution. | |
970 | */ | |
971 | ret = handle_register_done(sock_info); | |
972 | assert(!ret); | |
973 | ust_unlock(); | |
974 | goto restart; | |
975 | } | |
976 | open_sock[i] = ret; | |
46050b1a MD |
977 | } |
978 | ||
32ce8569 MD |
979 | sock_info->socket = open_sock[0]; |
980 | sock_info->notify_socket = open_sock[1]; | |
46050b1a | 981 | |
27fe9f21 | 982 | timeout = get_notify_sock_timeout(); |
e3426ddc MD |
983 | if (timeout >= 0) { |
984 | /* | |
985 | * Give at least 10ms to sessiond to reply to | |
986 | * notifications. | |
987 | */ | |
988 | if (timeout < 10) | |
989 | timeout = 10; | |
27fe9f21 MD |
990 | ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket, |
991 | timeout); | |
992 | if (ret < 0) { | |
993 | WARN("Error setting socket receive timeout"); | |
994 | } | |
995 | ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket, | |
996 | timeout); | |
997 | if (ret < 0) { | |
998 | WARN("Error setting socket send timeout"); | |
999 | } | |
e3426ddc | 1000 | } else if (timeout < -1) { |
5d62bf6b | 1001 | WARN("Unsupported timeout value %ld", timeout); |
27fe9f21 MD |
1002 | } |
1003 | ||
5b14aab3 MD |
1004 | ust_lock(); |
1005 | ||
1006 | if (lttng_ust_comm_should_quit) { | |
1007 | goto quit; | |
1008 | } | |
1009 | ||
46050b1a MD |
1010 | /* |
1011 | * Create only one root handle per listener thread for the whole | |
f59ed768 MD |
1012 | * process lifetime, so we ensure we get ID which is statically |
1013 | * assigned to the root handle. | |
46050b1a MD |
1014 | */ |
1015 | if (sock_info->root_handle == -1) { | |
1016 | ret = lttng_abi_create_root_handle(); | |
a51070bb | 1017 | if (ret < 0) { |
46050b1a | 1018 | ERR("Error creating root handle"); |
46050b1a MD |
1019 | goto quit; |
1020 | } | |
1021 | sock_info->root_handle = ret; | |
9eb62b9c | 1022 | } |
1ea11eab | 1023 | |
32ce8569 | 1024 | ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD); |
9eb62b9c | 1025 | if (ret < 0) { |
32ce8569 MD |
1026 | ERR("Error registering to %s ust cmd socket", |
1027 | sock_info->name); | |
c0eedf81 | 1028 | prev_connect_failed = 1; |
11ff9c7d MD |
1029 | /* |
1030 | * If we cannot register to the sessiond daemon, don't | |
1031 | * delay constructor execution. | |
1032 | */ | |
edaa1431 | 1033 | ret = handle_register_done(sock_info); |
11ff9c7d | 1034 | assert(!ret); |
17dfb34b | 1035 | ust_unlock(); |
9eb62b9c MD |
1036 | goto restart; |
1037 | } | |
32ce8569 MD |
1038 | ret = register_to_sessiond(sock_info->notify_socket, |
1039 | USTCTL_SOCKET_NOTIFY); | |
1040 | if (ret < 0) { | |
1041 | ERR("Error registering to %s ust notify socket", | |
1042 | sock_info->name); | |
1043 | prev_connect_failed = 1; | |
1044 | /* | |
1045 | * If we cannot register to the sessiond daemon, don't | |
1046 | * delay constructor execution. | |
1047 | */ | |
1048 | ret = handle_register_done(sock_info); | |
1049 | assert(!ret); | |
1050 | ust_unlock(); | |
1051 | goto restart; | |
1052 | } | |
1053 | sock = sock_info->socket; | |
1054 | ||
17dfb34b | 1055 | ust_unlock(); |
46050b1a | 1056 | |
d9e99d10 MD |
1057 | for (;;) { |
1058 | ssize_t len; | |
57773204 | 1059 | struct ustcomm_ust_msg lum; |
d9e99d10 | 1060 | |
57773204 | 1061 | len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum)); |
d9e99d10 MD |
1062 | switch (len) { |
1063 | case 0: /* orderly shutdown */ | |
7dd08bec | 1064 | DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name); |
8236ba10 | 1065 | ust_lock(); |
d5e1fea6 | 1066 | if (lttng_ust_comm_should_quit) { |
d5e1fea6 MD |
1067 | goto quit; |
1068 | } | |
8236ba10 MD |
1069 | /* |
1070 | * Either sessiond has shutdown or refused us by closing the socket. | |
1071 | * In either case, we don't want to delay construction execution, | |
1072 | * and we need to wait before retry. | |
1073 | */ | |
1074 | prev_connect_failed = 1; | |
1075 | /* | |
1076 | * If we cannot register to the sessiond daemon, don't | |
1077 | * delay constructor execution. | |
1078 | */ | |
1079 | ret = handle_register_done(sock_info); | |
1080 | assert(!ret); | |
1081 | ust_unlock(); | |
d9e99d10 | 1082 | goto end; |
e7723462 | 1083 | case sizeof(lum): |
74d81a6c | 1084 | print_cmd(lum.cmd, lum.handle); |
11ff9c7d | 1085 | ret = handle_message(sock_info, sock, &lum); |
7bc53e94 | 1086 | if (ret) { |
11ff9c7d | 1087 | ERR("Error handling message for %s socket", sock_info->name); |
d9e99d10 MD |
1088 | } |
1089 | continue; | |
7bc53e94 MD |
1090 | default: |
1091 | if (len < 0) { | |
1092 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
1093 | } else { | |
1094 | DBG("incorrect message size (%s socket): %zd", sock_info->name, len); | |
1095 | } | |
1096 | if (len == -ECONNRESET) { | |
1097 | DBG("%s remote end closed connection", sock_info->name); | |
d9e99d10 MD |
1098 | goto end; |
1099 | } | |
1100 | goto end; | |
d9e99d10 MD |
1101 | } |
1102 | ||
1103 | } | |
1104 | end: | |
f59ed768 | 1105 | ust_lock(); |
d5e1fea6 | 1106 | if (lttng_ust_comm_should_quit) { |
d5e1fea6 MD |
1107 | goto quit; |
1108 | } | |
f59ed768 MD |
1109 | /* Cleanup socket handles before trying to reconnect */ |
1110 | lttng_ust_objd_table_owner_cleanup(sock_info); | |
1111 | ust_unlock(); | |
9eb62b9c | 1112 | goto restart; /* try to reconnect */ |
e33f3265 | 1113 | |
1ea11eab | 1114 | quit: |
e33f3265 MD |
1115 | sock_info->thread_active = 0; |
1116 | ust_unlock(); | |
d9e99d10 MD |
1117 | return NULL; |
1118 | } | |
1119 | ||
2691221a MD |
1120 | /* |
1121 | * sessiond monitoring thread: monitor presence of global and per-user | |
1122 | * sessiond by polling the application common named pipe. | |
1123 | */ | |
edaa1431 | 1124 | void __attribute__((constructor)) lttng_ust_init(void) |
2691221a | 1125 | { |
11ff9c7d | 1126 | struct timespec constructor_timeout; |
ae6a58bf | 1127 | sigset_t sig_all_blocked, orig_parent_mask; |
1879f67f | 1128 | pthread_attr_t thread_attr; |
cf12a773 | 1129 | int timeout_mode; |
2691221a MD |
1130 | int ret; |
1131 | ||
edaa1431 MD |
1132 | if (uatomic_xchg(&initialized, 1) == 1) |
1133 | return; | |
1134 | ||
eddd8d5d MD |
1135 | /* |
1136 | * Fixup interdependency between TLS fixup mutex (which happens | |
1137 | * to be the dynamic linker mutex) and ust_lock, taken within | |
1138 | * the ust lock. | |
1139 | */ | |
f645cfa7 | 1140 | lttng_fixup_ringbuffer_tls(); |
4158a15a | 1141 | lttng_fixup_vtid_tls(); |
a903623f | 1142 | lttng_fixup_nest_count_tls(); |
009745db | 1143 | lttng_fixup_procname_tls(); |
eddd8d5d | 1144 | |
edaa1431 MD |
1145 | /* |
1146 | * We want precise control over the order in which we construct | |
1147 | * our sub-libraries vs starting to receive commands from | |
1148 | * sessiond (otherwise leading to errors when trying to create | |
1149 | * sessiond before the init functions are completed). | |
1150 | */ | |
2691221a | 1151 | init_usterr(); |
edaa1431 | 1152 | init_tracepoint(); |
7dd08bec MD |
1153 | lttng_ring_buffer_metadata_client_init(); |
1154 | lttng_ring_buffer_client_overwrite_init(); | |
34a91bdb | 1155 | lttng_ring_buffer_client_overwrite_rt_init(); |
7dd08bec | 1156 | lttng_ring_buffer_client_discard_init(); |
34a91bdb | 1157 | lttng_ring_buffer_client_discard_rt_init(); |
a0a3bef9 | 1158 | lttng_context_init(); |
2691221a | 1159 | |
ff517991 | 1160 | timeout_mode = get_constructor_timeout(&constructor_timeout); |
11ff9c7d | 1161 | |
95259bd0 | 1162 | ret = sem_init(&constructor_wait, 0, 0); |
11ff9c7d MD |
1163 | assert(!ret); |
1164 | ||
8d20bf54 | 1165 | ret = setup_local_apps(); |
2691221a | 1166 | if (ret) { |
9ec6895c | 1167 | DBG("local apps setup returned %d", ret); |
2691221a | 1168 | } |
ae6a58bf WP |
1169 | |
1170 | /* A new thread created by pthread_create inherits the signal mask | |
1171 | * from the parent. To avoid any signal being received by the | |
1172 | * listener thread, we block all signals temporarily in the parent, | |
1173 | * while we create the listener thread. | |
1174 | */ | |
1175 | sigfillset(&sig_all_blocked); | |
1176 | ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask); | |
1177 | if (ret) { | |
d94d802c | 1178 | ERR("pthread_sigmask: %s", strerror(ret)); |
ae6a58bf WP |
1179 | } |
1180 | ||
1879f67f MG |
1181 | ret = pthread_attr_init(&thread_attr); |
1182 | if (ret) { | |
1183 | ERR("pthread_attr_init: %s", strerror(ret)); | |
1184 | } | |
1185 | ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED); | |
1186 | if (ret) { | |
1187 | ERR("pthread_attr_setdetachstate: %s", strerror(ret)); | |
1188 | } | |
1189 | ||
e33f3265 | 1190 | ust_lock(); |
1879f67f | 1191 | ret = pthread_create(&global_apps.ust_listener, &thread_attr, |
dde70ea0 | 1192 | ust_listener_thread, &global_apps); |
d94d802c MD |
1193 | if (ret) { |
1194 | ERR("pthread_create global: %s", strerror(ret)); | |
1195 | } | |
e33f3265 MD |
1196 | global_apps.thread_active = 1; |
1197 | ust_unlock(); | |
1198 | ||
8d20bf54 | 1199 | if (local_apps.allowed) { |
e33f3265 | 1200 | ust_lock(); |
1879f67f | 1201 | ret = pthread_create(&local_apps.ust_listener, &thread_attr, |
dde70ea0 | 1202 | ust_listener_thread, &local_apps); |
d94d802c MD |
1203 | if (ret) { |
1204 | ERR("pthread_create local: %s", strerror(ret)); | |
1205 | } | |
e33f3265 MD |
1206 | local_apps.thread_active = 1; |
1207 | ust_unlock(); | |
8d20bf54 MD |
1208 | } else { |
1209 | handle_register_done(&local_apps); | |
1210 | } | |
1879f67f MG |
1211 | ret = pthread_attr_destroy(&thread_attr); |
1212 | if (ret) { | |
1213 | ERR("pthread_attr_destroy: %s", strerror(ret)); | |
1214 | } | |
8d20bf54 | 1215 | |
ae6a58bf WP |
1216 | /* Restore original signal mask in parent */ |
1217 | ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL); | |
1218 | if (ret) { | |
d94d802c | 1219 | ERR("pthread_sigmask: %s", strerror(ret)); |
ae6a58bf WP |
1220 | } |
1221 | ||
cf12a773 MD |
1222 | switch (timeout_mode) { |
1223 | case 1: /* timeout wait */ | |
95259bd0 MD |
1224 | do { |
1225 | ret = sem_timedwait(&constructor_wait, | |
1226 | &constructor_timeout); | |
1227 | } while (ret < 0 && errno == EINTR); | |
cf12a773 | 1228 | if (ret < 0 && errno == ETIMEDOUT) { |
7dd08bec | 1229 | ERR("Timed out waiting for lttng-sessiond"); |
cf12a773 MD |
1230 | } else { |
1231 | assert(!ret); | |
1232 | } | |
1233 | break; | |
7b766b16 | 1234 | case -1:/* wait forever */ |
95259bd0 MD |
1235 | do { |
1236 | ret = sem_wait(&constructor_wait); | |
1237 | } while (ret < 0 && errno == EINTR); | |
11ff9c7d | 1238 | assert(!ret); |
cf12a773 | 1239 | break; |
7b766b16 | 1240 | case 0: /* no timeout */ |
cf12a773 | 1241 | break; |
11ff9c7d | 1242 | } |
2691221a MD |
1243 | } |
1244 | ||
17dfb34b MD |
1245 | static |
1246 | void lttng_ust_cleanup(int exiting) | |
1247 | { | |
efe0de09 | 1248 | cleanup_sock_info(&global_apps, exiting); |
17dfb34b | 1249 | if (local_apps.allowed) { |
efe0de09 | 1250 | cleanup_sock_info(&local_apps, exiting); |
17dfb34b | 1251 | } |
efe0de09 MD |
1252 | /* |
1253 | * The teardown in this function all affect data structures | |
1254 | * accessed under the UST lock by the listener thread. This | |
1255 | * lock, along with the lttng_ust_comm_should_quit flag, ensure | |
1256 | * that none of these threads are accessing this data at this | |
1257 | * point. | |
1258 | */ | |
17dfb34b | 1259 | lttng_ust_abi_exit(); |
003fedf4 | 1260 | lttng_ust_events_exit(); |
a0a3bef9 | 1261 | lttng_context_exit(); |
34a91bdb | 1262 | lttng_ring_buffer_client_discard_rt_exit(); |
7dd08bec | 1263 | lttng_ring_buffer_client_discard_exit(); |
34a91bdb | 1264 | lttng_ring_buffer_client_overwrite_rt_exit(); |
7dd08bec MD |
1265 | lttng_ring_buffer_client_overwrite_exit(); |
1266 | lttng_ring_buffer_metadata_client_exit(); | |
17dfb34b MD |
1267 | exit_tracepoint(); |
1268 | if (!exiting) { | |
1269 | /* Reinitialize values for fork */ | |
1270 | sem_count = 2; | |
1271 | lttng_ust_comm_should_quit = 0; | |
1272 | initialized = 0; | |
1273 | } | |
1274 | } | |
1275 | ||
edaa1431 | 1276 | void __attribute__((destructor)) lttng_ust_exit(void) |
2691221a MD |
1277 | { |
1278 | int ret; | |
1279 | ||
9eb62b9c MD |
1280 | /* |
1281 | * Using pthread_cancel here because: | |
1282 | * A) we don't want to hang application teardown. | |
1283 | * B) the thread is not allocating any resource. | |
1284 | */ | |
1ea11eab MD |
1285 | |
1286 | /* | |
1287 | * Require the communication thread to quit. Synchronize with | |
1288 | * mutexes to ensure it is not in a mutex critical section when | |
1289 | * pthread_cancel is later called. | |
1290 | */ | |
17dfb34b | 1291 | ust_lock(); |
1ea11eab | 1292 | lttng_ust_comm_should_quit = 1; |
1ea11eab | 1293 | |
f5f94532 | 1294 | /* cancel threads */ |
e33f3265 MD |
1295 | if (global_apps.thread_active) { |
1296 | ret = pthread_cancel(global_apps.ust_listener); | |
1297 | if (ret) { | |
1298 | ERR("Error cancelling global ust listener thread: %s", | |
1299 | strerror(ret)); | |
1300 | } else { | |
1301 | global_apps.thread_active = 0; | |
1302 | } | |
2691221a | 1303 | } |
e33f3265 | 1304 | if (local_apps.thread_active) { |
8d20bf54 MD |
1305 | ret = pthread_cancel(local_apps.ust_listener); |
1306 | if (ret) { | |
d94d802c MD |
1307 | ERR("Error cancelling local ust listener thread: %s", |
1308 | strerror(ret)); | |
e33f3265 MD |
1309 | } else { |
1310 | local_apps.thread_active = 0; | |
8d20bf54 | 1311 | } |
8d20bf54 | 1312 | } |
e33f3265 MD |
1313 | ust_unlock(); |
1314 | ||
efe0de09 MD |
1315 | /* |
1316 | * Do NOT join threads: use of sys_futex makes it impossible to | |
1317 | * join the threads without using async-cancel, but async-cancel | |
1318 | * is delivered by a signal, which could hit the target thread | |
1319 | * anywhere in its code path, including while the ust_lock() is | |
1320 | * held, causing a deadlock for the other thread. Let the OS | |
1321 | * cleanup the threads if there are stalled in a syscall. | |
1322 | */ | |
17dfb34b | 1323 | lttng_ust_cleanup(1); |
2691221a | 1324 | } |
e822f505 MD |
1325 | |
1326 | /* | |
1327 | * We exclude the worker threads across fork and clone (except | |
1328 | * CLONE_VM), because these system calls only keep the forking thread | |
1329 | * running in the child. Therefore, we don't want to call fork or clone | |
1330 | * in the middle of an tracepoint or ust tracing state modification. | |
1331 | * Holding this mutex protects these structures across fork and clone. | |
1332 | */ | |
b728d87e | 1333 | void ust_before_fork(sigset_t *save_sigset) |
e822f505 MD |
1334 | { |
1335 | /* | |
1336 | * Disable signals. This is to avoid that the child intervenes | |
1337 | * before it is properly setup for tracing. It is safer to | |
1338 | * disable all signals, because then we know we are not breaking | |
1339 | * anything by restoring the original mask. | |
1340 | */ | |
1341 | sigset_t all_sigs; | |
1342 | int ret; | |
1343 | ||
8c90a710 | 1344 | if (URCU_TLS(lttng_ust_nest_count)) |
e8508a49 | 1345 | return; |
e822f505 MD |
1346 | /* Disable signals */ |
1347 | sigfillset(&all_sigs); | |
b728d87e | 1348 | ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset); |
e822f505 MD |
1349 | if (ret == -1) { |
1350 | PERROR("sigprocmask"); | |
1351 | } | |
17dfb34b | 1352 | ust_lock(); |
e822f505 MD |
1353 | rcu_bp_before_fork(); |
1354 | } | |
1355 | ||
b728d87e | 1356 | static void ust_after_fork_common(sigset_t *restore_sigset) |
e822f505 MD |
1357 | { |
1358 | int ret; | |
1359 | ||
17dfb34b MD |
1360 | DBG("process %d", getpid()); |
1361 | ust_unlock(); | |
e822f505 | 1362 | /* Restore signals */ |
23c8854a | 1363 | ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL); |
e822f505 MD |
1364 | if (ret == -1) { |
1365 | PERROR("sigprocmask"); | |
1366 | } | |
1367 | } | |
1368 | ||
b728d87e | 1369 | void ust_after_fork_parent(sigset_t *restore_sigset) |
e822f505 | 1370 | { |
8c90a710 | 1371 | if (URCU_TLS(lttng_ust_nest_count)) |
e8508a49 | 1372 | return; |
17dfb34b | 1373 | DBG("process %d", getpid()); |
e822f505 MD |
1374 | rcu_bp_after_fork_parent(); |
1375 | /* Release mutexes and reenable signals */ | |
b728d87e | 1376 | ust_after_fork_common(restore_sigset); |
e822f505 MD |
1377 | } |
1378 | ||
17dfb34b MD |
1379 | /* |
1380 | * After fork, in the child, we need to cleanup all the leftover state, | |
1381 | * except the worker thread which already magically disappeared thanks | |
1382 | * to the weird Linux fork semantics. After tyding up, we call | |
1383 | * lttng_ust_init() again to start over as a new PID. | |
1384 | * | |
1385 | * This is meant for forks() that have tracing in the child between the | |
1386 | * fork and following exec call (if there is any). | |
1387 | */ | |
b728d87e | 1388 | void ust_after_fork_child(sigset_t *restore_sigset) |
e822f505 | 1389 | { |
8c90a710 | 1390 | if (URCU_TLS(lttng_ust_nest_count)) |
e8508a49 | 1391 | return; |
17dfb34b | 1392 | DBG("process %d", getpid()); |
e822f505 MD |
1393 | /* Release urcu mutexes */ |
1394 | rcu_bp_after_fork_child(); | |
17dfb34b | 1395 | lttng_ust_cleanup(0); |
a93bfc45 | 1396 | lttng_context_vtid_reset(); |
e822f505 | 1397 | /* Release mutexes and reenable signals */ |
b728d87e | 1398 | ust_after_fork_common(restore_sigset); |
318dfea9 | 1399 | lttng_ust_init(); |
e822f505 | 1400 | } |