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> |
95c25348 | 37 | #include <dlfcn.h> |
95259bd0 | 38 | #include <urcu/uatomic.h> |
80e2814b | 39 | #include <urcu/futex.h> |
c117fb1b | 40 | #include <urcu/compiler.h> |
1ea11eab | 41 | |
4318ae1b | 42 | #include <lttng/ust-events.h> |
4318ae1b | 43 | #include <lttng/ust-abi.h> |
4318ae1b | 44 | #include <lttng/ust.h> |
7bc53e94 | 45 | #include <lttng/ust-error.h> |
74d81a6c | 46 | #include <lttng/ust-ctl.h> |
8c90a710 | 47 | #include <urcu/tls-compat.h> |
44c72f10 MD |
48 | #include <ust-comm.h> |
49 | #include <usterr-signal-safe.h> | |
cd54f6d9 | 50 | #include <helper.h> |
44c72f10 | 51 | #include "tracepoint-internal.h" |
7dd08bec | 52 | #include "lttng-tracer-core.h" |
08114193 | 53 | #include "compat.h" |
f645cfa7 | 54 | #include "../libringbuffer/tlsfixup.h" |
13436238 | 55 | #include "../liblttng-ust-baddr/lttng-ust-baddr.h" |
edaa1431 MD |
56 | |
57 | /* | |
58 | * Has lttng ust comm constructor been called ? | |
59 | */ | |
60 | static int initialized; | |
61 | ||
1ea11eab | 62 | /* |
17dfb34b MD |
63 | * The ust_lock/ust_unlock lock is used as a communication thread mutex. |
64 | * Held when handling a command, also held by fork() to deal with | |
65 | * removal of threads, and by exit path. | |
1ea11eab | 66 | */ |
1ea11eab MD |
67 | |
68 | /* Should the ust comm thread quit ? */ | |
69 | static int lttng_ust_comm_should_quit; | |
70 | ||
11ff9c7d MD |
71 | /* |
72 | * Wait for either of these before continuing to the main | |
73 | * program: | |
74 | * - the register_done message from sessiond daemon | |
75 | * (will let the sessiond daemon enable sessions before main | |
76 | * starts.) | |
77 | * - sessiond daemon is not reachable. | |
78 | * - timeout (ensuring applications are resilient to session | |
79 | * daemon problems). | |
80 | */ | |
81 | static sem_t constructor_wait; | |
950aab0c MD |
82 | /* |
83 | * Doing this for both the global and local sessiond. | |
84 | */ | |
95259bd0 | 85 | static int sem_count = { 2 }; |
11ff9c7d | 86 | |
e8508a49 MD |
87 | /* |
88 | * Counting nesting within lttng-ust. Used to ensure that calling fork() | |
89 | * from liblttng-ust does not execute the pre/post fork handlers. | |
90 | */ | |
8c90a710 | 91 | static DEFINE_URCU_TLS(int, lttng_ust_nest_count); |
e8508a49 | 92 | |
1ea11eab MD |
93 | /* |
94 | * Info about socket and associated listener thread. | |
95 | */ | |
96 | struct sock_info { | |
11ff9c7d | 97 | const char *name; |
1ea11eab | 98 | pthread_t ust_listener; /* listener thread */ |
46050b1a | 99 | int root_handle; |
8d20bf54 MD |
100 | int constructor_sem_posted; |
101 | int allowed; | |
44e073f5 | 102 | int global; |
e33f3265 | 103 | int thread_active; |
7fc90dca MD |
104 | |
105 | char sock_path[PATH_MAX]; | |
106 | int socket; | |
32ce8569 | 107 | int notify_socket; |
7fc90dca MD |
108 | |
109 | char wait_shm_path[PATH_MAX]; | |
110 | char *wait_shm_mmap; | |
95c25348 | 111 | struct lttng_session *session_enabled; |
1ea11eab | 112 | }; |
2691221a MD |
113 | |
114 | /* Socket from app (connect) to session daemon (listen) for communication */ | |
1ea11eab | 115 | struct sock_info global_apps = { |
11ff9c7d | 116 | .name = "global", |
44e073f5 | 117 | .global = 1, |
7fc90dca | 118 | |
46050b1a | 119 | .root_handle = -1, |
8d20bf54 | 120 | .allowed = 1, |
e33f3265 | 121 | .thread_active = 0, |
7fc90dca | 122 | |
32ce8569 | 123 | .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME, |
7fc90dca | 124 | .socket = -1, |
32ce8569 | 125 | .notify_socket = -1, |
7fc90dca | 126 | |
32ce8569 | 127 | .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME, |
95c25348 PW |
128 | |
129 | .session_enabled = NULL, | |
1ea11eab | 130 | }; |
2691221a MD |
131 | |
132 | /* TODO: allow global_apps_sock_path override */ | |
133 | ||
1ea11eab | 134 | struct sock_info local_apps = { |
11ff9c7d | 135 | .name = "local", |
44e073f5 | 136 | .global = 0, |
46050b1a | 137 | .root_handle = -1, |
8d20bf54 | 138 | .allowed = 0, /* Check setuid bit first */ |
e33f3265 | 139 | .thread_active = 0, |
7fc90dca MD |
140 | |
141 | .socket = -1, | |
32ce8569 | 142 | .notify_socket = -1, |
95c25348 PW |
143 | |
144 | .session_enabled = NULL, | |
1ea11eab | 145 | }; |
2691221a | 146 | |
37ed587a MD |
147 | static int wait_poll_fallback; |
148 | ||
74d81a6c MD |
149 | static const char *cmd_name_mapping[] = { |
150 | [ LTTNG_UST_RELEASE ] = "Release", | |
151 | [ LTTNG_UST_SESSION ] = "Create Session", | |
152 | [ LTTNG_UST_TRACER_VERSION ] = "Get Tracer Version", | |
153 | ||
154 | [ LTTNG_UST_TRACEPOINT_LIST ] = "Create Tracepoint List", | |
155 | [ LTTNG_UST_WAIT_QUIESCENT ] = "Wait for Quiescent State", | |
156 | [ LTTNG_UST_REGISTER_DONE ] = "Registration Done", | |
157 | [ LTTNG_UST_TRACEPOINT_FIELD_LIST ] = "Create Tracepoint Field List", | |
158 | ||
159 | /* Session FD commands */ | |
160 | [ LTTNG_UST_CHANNEL ] = "Create Channel", | |
161 | [ LTTNG_UST_SESSION_START ] = "Start Session", | |
162 | [ LTTNG_UST_SESSION_STOP ] = "Stop Session", | |
163 | ||
164 | /* Channel FD commands */ | |
165 | [ LTTNG_UST_STREAM ] = "Create Stream", | |
166 | [ LTTNG_UST_EVENT ] = "Create Event", | |
167 | ||
168 | /* Event and Channel FD commands */ | |
169 | [ LTTNG_UST_CONTEXT ] = "Create Context", | |
170 | [ LTTNG_UST_FLUSH_BUFFER ] = "Flush Buffer", | |
171 | ||
172 | /* Event, Channel and Session commands */ | |
173 | [ LTTNG_UST_ENABLE ] = "Enable", | |
174 | [ LTTNG_UST_DISABLE ] = "Disable", | |
175 | ||
176 | /* Tracepoint list commands */ | |
177 | [ LTTNG_UST_TRACEPOINT_LIST_GET ] = "List Next Tracepoint", | |
178 | [ LTTNG_UST_TRACEPOINT_FIELD_LIST_GET ] = "List Next Tracepoint Field", | |
179 | ||
180 | /* Event FD commands */ | |
181 | [ LTTNG_UST_FILTER ] = "Create Filter", | |
75582b3a | 182 | [ LTTNG_UST_EXCLUSION ] = "Add exclusions to event", |
74d81a6c MD |
183 | }; |
184 | ||
ff517991 MD |
185 | static const char *str_timeout; |
186 | static int got_timeout_env; | |
187 | ||
7dd08bec | 188 | extern void lttng_ring_buffer_client_overwrite_init(void); |
34a91bdb | 189 | extern void lttng_ring_buffer_client_overwrite_rt_init(void); |
7dd08bec | 190 | extern void lttng_ring_buffer_client_discard_init(void); |
34a91bdb | 191 | extern void lttng_ring_buffer_client_discard_rt_init(void); |
7dd08bec MD |
192 | extern void lttng_ring_buffer_metadata_client_init(void); |
193 | extern void lttng_ring_buffer_client_overwrite_exit(void); | |
34a91bdb | 194 | extern void lttng_ring_buffer_client_overwrite_rt_exit(void); |
7dd08bec | 195 | extern void lttng_ring_buffer_client_discard_exit(void); |
34a91bdb | 196 | extern void lttng_ring_buffer_client_discard_rt_exit(void); |
7dd08bec | 197 | extern void lttng_ring_buffer_metadata_client_exit(void); |
edaa1431 | 198 | |
3c6f6263 AM |
199 | /* |
200 | * Returns the HOME directory path. Caller MUST NOT free(3) the returned | |
201 | * pointer. | |
202 | */ | |
203 | static | |
204 | const char *get_lttng_home_dir(void) | |
205 | { | |
206 | const char *val; | |
207 | ||
208 | val = (const char *) getenv("LTTNG_HOME"); | |
209 | if (val != NULL) { | |
210 | return val; | |
211 | } | |
212 | return (const char *) getenv("HOME"); | |
213 | } | |
214 | ||
a903623f MD |
215 | /* |
216 | * Force a read (imply TLS fixup for dlopen) of TLS variables. | |
217 | */ | |
218 | static | |
219 | void lttng_fixup_nest_count_tls(void) | |
220 | { | |
8c90a710 | 221 | asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count))); |
a903623f MD |
222 | } |
223 | ||
32ce8569 MD |
224 | int lttng_get_notify_socket(void *owner) |
225 | { | |
226 | struct sock_info *info = owner; | |
227 | ||
228 | return info->notify_socket; | |
229 | } | |
230 | ||
74d81a6c MD |
231 | static |
232 | void print_cmd(int cmd, int handle) | |
233 | { | |
234 | const char *cmd_name = "Unknown"; | |
235 | ||
fd67a004 MD |
236 | if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping) |
237 | && cmd_name_mapping[cmd]) { | |
74d81a6c MD |
238 | cmd_name = cmd_name_mapping[cmd]; |
239 | } | |
fd67a004 MD |
240 | DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)", |
241 | cmd_name, cmd, | |
74d81a6c MD |
242 | lttng_ust_obj_get_name(handle), handle); |
243 | } | |
244 | ||
2691221a | 245 | static |
8d20bf54 | 246 | int setup_local_apps(void) |
2691221a MD |
247 | { |
248 | const char *home_dir; | |
7fc90dca | 249 | uid_t uid; |
2691221a | 250 | |
7fc90dca | 251 | uid = getuid(); |
8d20bf54 MD |
252 | /* |
253 | * Disallow per-user tracing for setuid binaries. | |
254 | */ | |
7fc90dca | 255 | if (uid != geteuid()) { |
9ec6895c | 256 | assert(local_apps.allowed == 0); |
d0a1ae63 | 257 | return 0; |
8d20bf54 | 258 | } |
3c6f6263 | 259 | home_dir = get_lttng_home_dir(); |
9ec6895c MD |
260 | if (!home_dir) { |
261 | WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing."); | |
262 | assert(local_apps.allowed == 0); | |
2691221a | 263 | return -ENOENT; |
9ec6895c MD |
264 | } |
265 | local_apps.allowed = 1; | |
32ce8569 MD |
266 | snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s", |
267 | home_dir, | |
268 | LTTNG_DEFAULT_HOME_RUNDIR, | |
269 | LTTNG_UST_SOCK_FILENAME); | |
270 | snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u", | |
271 | LTTNG_UST_WAIT_FILENAME, | |
272 | uid); | |
2691221a MD |
273 | return 0; |
274 | } | |
275 | ||
ff517991 MD |
276 | /* |
277 | * Get notify_sock timeout, in ms. | |
278 | * -1: don't wait. 0: wait forever. >0: timeout, in ms. | |
279 | */ | |
280 | static | |
281 | long get_timeout(void) | |
282 | { | |
283 | long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS; | |
284 | ||
285 | if (!got_timeout_env) { | |
286 | str_timeout = getenv("LTTNG_UST_REGISTER_TIMEOUT"); | |
287 | got_timeout_env = 1; | |
288 | } | |
289 | if (str_timeout) | |
290 | constructor_delay_ms = strtol(str_timeout, NULL, 10); | |
291 | return constructor_delay_ms; | |
292 | } | |
293 | ||
294 | static | |
295 | long get_notify_sock_timeout(void) | |
296 | { | |
297 | return get_timeout(); | |
298 | } | |
299 | ||
300 | /* | |
301 | * Return values: -1: don't wait. 0: wait forever. 1: timeout wait. | |
302 | */ | |
303 | static | |
304 | int get_constructor_timeout(struct timespec *constructor_timeout) | |
305 | { | |
306 | long constructor_delay_ms; | |
307 | int ret; | |
308 | ||
309 | constructor_delay_ms = get_timeout(); | |
310 | ||
311 | switch (constructor_delay_ms) { | |
312 | case -1:/* fall-through */ | |
313 | case 0: | |
314 | return constructor_delay_ms; | |
315 | default: | |
316 | break; | |
317 | } | |
318 | ||
319 | /* | |
320 | * If we are unable to find the current time, don't wait. | |
321 | */ | |
322 | ret = clock_gettime(CLOCK_REALTIME, constructor_timeout); | |
323 | if (ret) { | |
324 | return -1; | |
325 | } | |
326 | constructor_timeout->tv_sec += constructor_delay_ms / 1000UL; | |
327 | constructor_timeout->tv_nsec += | |
328 | (constructor_delay_ms % 1000UL) * 1000000UL; | |
329 | if (constructor_timeout->tv_nsec >= 1000000000UL) { | |
330 | constructor_timeout->tv_sec++; | |
331 | constructor_timeout->tv_nsec -= 1000000000UL; | |
332 | } | |
333 | return 1; | |
334 | } | |
335 | ||
2691221a | 336 | static |
32ce8569 | 337 | int register_to_sessiond(int socket, enum ustctl_socket_type type) |
2691221a | 338 | { |
32ce8569 MD |
339 | return ustcomm_send_reg_msg(socket, |
340 | type, | |
341 | CAA_BITS_PER_LONG, | |
342 | lttng_alignof(uint8_t) * CHAR_BIT, | |
343 | lttng_alignof(uint16_t) * CHAR_BIT, | |
344 | lttng_alignof(uint32_t) * CHAR_BIT, | |
345 | lttng_alignof(uint64_t) * CHAR_BIT, | |
346 | lttng_alignof(unsigned long) * CHAR_BIT); | |
2691221a MD |
347 | } |
348 | ||
d9e99d10 | 349 | static |
57773204 | 350 | int send_reply(int sock, struct ustcomm_ust_reply *lur) |
d9e99d10 | 351 | { |
9eb62b9c | 352 | ssize_t len; |
d3a492d1 | 353 | |
57773204 | 354 | len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur)); |
d3a492d1 | 355 | switch (len) { |
a4be8962 | 356 | case sizeof(*lur): |
d3a492d1 MD |
357 | DBG("message successfully sent"); |
358 | return 0; | |
7bc53e94 MD |
359 | default: |
360 | if (len == -ECONNRESET) { | |
361 | DBG("remote end closed connection"); | |
d3a492d1 MD |
362 | return 0; |
363 | } | |
7bc53e94 MD |
364 | if (len < 0) |
365 | return len; | |
366 | DBG("incorrect message size: %zd", len); | |
367 | return -EINVAL; | |
d3a492d1 MD |
368 | } |
369 | } | |
370 | ||
371 | static | |
edaa1431 | 372 | int handle_register_done(struct sock_info *sock_info) |
11ff9c7d MD |
373 | { |
374 | int ret; | |
375 | ||
edaa1431 MD |
376 | if (sock_info->constructor_sem_posted) |
377 | return 0; | |
378 | sock_info->constructor_sem_posted = 1; | |
56cd7e2f MD |
379 | if (uatomic_read(&sem_count) <= 0) { |
380 | return 0; | |
381 | } | |
95259bd0 MD |
382 | ret = uatomic_add_return(&sem_count, -1); |
383 | if (ret == 0) { | |
384 | ret = sem_post(&constructor_wait); | |
385 | assert(!ret); | |
386 | } | |
11ff9c7d MD |
387 | return 0; |
388 | } | |
389 | ||
390 | static | |
391 | int handle_message(struct sock_info *sock_info, | |
57773204 | 392 | int sock, struct ustcomm_ust_msg *lum) |
d3a492d1 | 393 | { |
1ea11eab | 394 | int ret = 0; |
b61ce3b2 | 395 | const struct lttng_ust_objd_ops *ops; |
57773204 | 396 | struct ustcomm_ust_reply lur; |
ef9ff354 | 397 | union ust_args args; |
40003310 | 398 | ssize_t len; |
1ea11eab | 399 | |
17dfb34b | 400 | ust_lock(); |
1ea11eab | 401 | |
46050b1a MD |
402 | memset(&lur, 0, sizeof(lur)); |
403 | ||
1ea11eab | 404 | if (lttng_ust_comm_should_quit) { |
74d81a6c | 405 | ret = -LTTNG_UST_ERR_EXITING; |
1ea11eab MD |
406 | goto end; |
407 | } | |
9eb62b9c | 408 | |
46050b1a MD |
409 | ops = objd_ops(lum->handle); |
410 | if (!ops) { | |
411 | ret = -ENOENT; | |
412 | goto end; | |
1ea11eab | 413 | } |
46050b1a MD |
414 | |
415 | switch (lum->cmd) { | |
11ff9c7d MD |
416 | case LTTNG_UST_REGISTER_DONE: |
417 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
edaa1431 | 418 | ret = handle_register_done(sock_info); |
11ff9c7d MD |
419 | else |
420 | ret = -EINVAL; | |
421 | break; | |
46050b1a MD |
422 | case LTTNG_UST_RELEASE: |
423 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
424 | ret = -EPERM; | |
425 | else | |
1849ef7c | 426 | ret = lttng_ust_objd_unref(lum->handle, 1); |
d9e99d10 | 427 | break; |
2d78951a MD |
428 | case LTTNG_UST_FILTER: |
429 | { | |
430 | /* Receive filter data */ | |
f488575f | 431 | struct lttng_ust_filter_bytecode_node *bytecode; |
2d78951a | 432 | |
cd54f6d9 | 433 | if (lum->u.filter.data_size > FILTER_BYTECODE_MAX_LEN) { |
7bc53e94 | 434 | ERR("Filter data size is too large: %u bytes", |
2d78951a MD |
435 | lum->u.filter.data_size); |
436 | ret = -EINVAL; | |
437 | goto error; | |
438 | } | |
2734ca65 | 439 | |
885b1dfd | 440 | if (lum->u.filter.reloc_offset > lum->u.filter.data_size) { |
7bc53e94 | 441 | ERR("Filter reloc offset %u is not within data", |
2734ca65 CB |
442 | lum->u.filter.reloc_offset); |
443 | ret = -EINVAL; | |
444 | goto error; | |
445 | } | |
446 | ||
cd54f6d9 MD |
447 | bytecode = zmalloc(sizeof(*bytecode) + lum->u.filter.data_size); |
448 | if (!bytecode) { | |
449 | ret = -ENOMEM; | |
450 | goto error; | |
451 | } | |
f488575f | 452 | len = ustcomm_recv_unix_sock(sock, bytecode->bc.data, |
2d78951a MD |
453 | lum->u.filter.data_size); |
454 | switch (len) { | |
455 | case 0: /* orderly shutdown */ | |
456 | ret = 0; | |
cd54f6d9 | 457 | free(bytecode); |
2d78951a | 458 | goto error; |
2d78951a MD |
459 | default: |
460 | if (len == lum->u.filter.data_size) { | |
7bc53e94 | 461 | DBG("filter data received"); |
2d78951a | 462 | break; |
7bc53e94 MD |
463 | } else if (len < 0) { |
464 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
465 | if (len == -ECONNRESET) { | |
466 | ERR("%s remote end closed connection", sock_info->name); | |
467 | ret = len; | |
468 | free(bytecode); | |
469 | goto error; | |
470 | } | |
471 | ret = len; | |
eb8bf361 | 472 | free(bytecode); |
7bc53e94 | 473 | goto end; |
2d78951a | 474 | } else { |
7bc53e94 | 475 | DBG("incorrect filter data message size: %zd", len); |
2d78951a | 476 | ret = -EINVAL; |
cd54f6d9 | 477 | free(bytecode); |
2d78951a MD |
478 | goto end; |
479 | } | |
480 | } | |
f488575f MD |
481 | bytecode->bc.len = lum->u.filter.data_size; |
482 | bytecode->bc.reloc_offset = lum->u.filter.reloc_offset; | |
3f6fd224 | 483 | bytecode->bc.seqnum = lum->u.filter.seqnum; |
cd54f6d9 | 484 | if (ops->cmd) { |
2d78951a | 485 | ret = ops->cmd(lum->handle, lum->cmd, |
cd54f6d9 | 486 | (unsigned long) bytecode, |
f59ed768 | 487 | &args, sock_info); |
cd54f6d9 MD |
488 | if (ret) { |
489 | free(bytecode); | |
490 | } | |
491 | /* don't free bytecode if everything went fine. */ | |
492 | } else { | |
2d78951a | 493 | ret = -ENOSYS; |
cd54f6d9 MD |
494 | free(bytecode); |
495 | } | |
2d78951a MD |
496 | break; |
497 | } | |
86e36163 JI |
498 | case LTTNG_UST_EXCLUSION: |
499 | { | |
500 | /* Receive exclusion names */ | |
501 | struct lttng_ust_excluder_node *node; | |
502 | unsigned int count; | |
503 | ||
504 | count = lum->u.exclusion.count; | |
505 | if (count == 0) { | |
506 | /* There are no names to read */ | |
507 | ret = 0; | |
508 | goto error; | |
509 | } | |
510 | node = zmalloc(sizeof(*node) + | |
511 | count * LTTNG_UST_SYM_NAME_LEN); | |
512 | if (!node) { | |
513 | ret = -ENOMEM; | |
514 | goto error; | |
515 | } | |
516 | node->excluder.count = count; | |
517 | len = ustcomm_recv_unix_sock(sock, node->excluder.names, | |
518 | count * LTTNG_UST_SYM_NAME_LEN); | |
519 | switch (len) { | |
520 | case 0: /* orderly shutdown */ | |
521 | ret = 0; | |
522 | free(node); | |
523 | goto error; | |
524 | default: | |
525 | if (len == count * LTTNG_UST_SYM_NAME_LEN) { | |
526 | DBG("Exclusion data received"); | |
527 | break; | |
528 | } else if (len < 0) { | |
529 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
530 | if (len == -ECONNRESET) { | |
531 | ERR("%s remote end closed connection", sock_info->name); | |
532 | ret = len; | |
533 | free(node); | |
534 | goto error; | |
535 | } | |
536 | ret = len; | |
537 | free(node); | |
538 | goto end; | |
539 | } else { | |
540 | DBG("Incorrect exclusion data message size: %zd", len); | |
541 | ret = -EINVAL; | |
542 | free(node); | |
543 | goto end; | |
544 | } | |
545 | } | |
546 | if (ops->cmd) { | |
547 | ret = ops->cmd(lum->handle, lum->cmd, | |
548 | (unsigned long) node, | |
549 | &args, sock_info); | |
550 | if (ret) { | |
551 | free(node); | |
552 | } | |
553 | /* Don't free exclusion data if everything went fine. */ | |
554 | } else { | |
555 | ret = -ENOSYS; | |
556 | free(node); | |
557 | } | |
558 | break; | |
559 | } | |
74d81a6c MD |
560 | case LTTNG_UST_CHANNEL: |
561 | { | |
562 | void *chan_data; | |
ff0f5728 | 563 | int wakeup_fd; |
74d81a6c MD |
564 | |
565 | len = ustcomm_recv_channel_from_sessiond(sock, | |
ff0f5728 MD |
566 | &chan_data, lum->u.channel.len, |
567 | &wakeup_fd); | |
74d81a6c MD |
568 | switch (len) { |
569 | case 0: /* orderly shutdown */ | |
570 | ret = 0; | |
571 | goto error; | |
572 | default: | |
573 | if (len == lum->u.channel.len) { | |
574 | DBG("channel data received"); | |
575 | break; | |
576 | } else if (len < 0) { | |
577 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
578 | if (len == -ECONNRESET) { | |
579 | ERR("%s remote end closed connection", sock_info->name); | |
580 | ret = len; | |
581 | goto error; | |
582 | } | |
583 | ret = len; | |
584 | goto end; | |
585 | } else { | |
586 | DBG("incorrect channel data message size: %zd", len); | |
587 | ret = -EINVAL; | |
588 | goto end; | |
589 | } | |
590 | } | |
591 | args.channel.chan_data = chan_data; | |
ff0f5728 | 592 | args.channel.wakeup_fd = wakeup_fd; |
74d81a6c MD |
593 | if (ops->cmd) |
594 | ret = ops->cmd(lum->handle, lum->cmd, | |
595 | (unsigned long) &lum->u, | |
596 | &args, sock_info); | |
597 | else | |
598 | ret = -ENOSYS; | |
599 | break; | |
600 | } | |
601 | case LTTNG_UST_STREAM: | |
602 | { | |
603 | /* Receive shm_fd, wakeup_fd */ | |
604 | ret = ustcomm_recv_stream_from_sessiond(sock, | |
605 | &lum->u.stream.len, | |
606 | &args.stream.shm_fd, | |
607 | &args.stream.wakeup_fd); | |
608 | if (ret) { | |
609 | goto end; | |
610 | } | |
611 | if (ops->cmd) | |
612 | ret = ops->cmd(lum->handle, lum->cmd, | |
613 | (unsigned long) &lum->u, | |
614 | &args, sock_info); | |
615 | else | |
616 | ret = -ENOSYS; | |
617 | break; | |
618 | } | |
d9e99d10 | 619 | default: |
46050b1a MD |
620 | if (ops->cmd) |
621 | ret = ops->cmd(lum->handle, lum->cmd, | |
ef9ff354 | 622 | (unsigned long) &lum->u, |
f59ed768 | 623 | &args, sock_info); |
46050b1a MD |
624 | else |
625 | ret = -ENOSYS; | |
626 | break; | |
d9e99d10 | 627 | } |
46050b1a | 628 | |
1ea11eab | 629 | end: |
46050b1a MD |
630 | lur.handle = lum->handle; |
631 | lur.cmd = lum->cmd; | |
632 | lur.ret_val = ret; | |
633 | if (ret >= 0) { | |
7bc53e94 | 634 | lur.ret_code = LTTNG_UST_OK; |
46050b1a | 635 | } else { |
7bc53e94 MD |
636 | /* |
637 | * Use -LTTNG_UST_ERR as wildcard for UST internal | |
638 | * error that are not caused by the transport, except if | |
639 | * we already have a more precise error message to | |
640 | * report. | |
641 | */ | |
64b2564e DG |
642 | if (ret > -LTTNG_UST_ERR) { |
643 | /* Translate code to UST error. */ | |
644 | switch (ret) { | |
645 | case -EEXIST: | |
646 | lur.ret_code = -LTTNG_UST_ERR_EXIST; | |
647 | break; | |
648 | case -EINVAL: | |
649 | lur.ret_code = -LTTNG_UST_ERR_INVAL; | |
650 | break; | |
651 | case -ENOENT: | |
652 | lur.ret_code = -LTTNG_UST_ERR_NOENT; | |
653 | break; | |
654 | case -EPERM: | |
655 | lur.ret_code = -LTTNG_UST_ERR_PERM; | |
656 | break; | |
657 | case -ENOSYS: | |
658 | lur.ret_code = -LTTNG_UST_ERR_NOSYS; | |
659 | break; | |
660 | default: | |
661 | lur.ret_code = -LTTNG_UST_ERR; | |
662 | break; | |
663 | } | |
664 | } else { | |
7bc53e94 | 665 | lur.ret_code = ret; |
64b2564e | 666 | } |
46050b1a | 667 | } |
e6ea14c5 MD |
668 | if (ret >= 0) { |
669 | switch (lum->cmd) { | |
e6ea14c5 MD |
670 | case LTTNG_UST_TRACER_VERSION: |
671 | lur.u.version = lum->u.version; | |
672 | break; | |
673 | case LTTNG_UST_TRACEPOINT_LIST_GET: | |
674 | memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint)); | |
675 | break; | |
676 | } | |
381c0f1e | 677 | } |
74d81a6c | 678 | DBG("Return value: %d", lur.ret_val); |
46050b1a | 679 | ret = send_reply(sock, &lur); |
193183fb | 680 | if (ret < 0) { |
7bc53e94 | 681 | DBG("error sending reply"); |
193183fb MD |
682 | goto error; |
683 | } | |
46050b1a | 684 | |
40003310 MD |
685 | /* |
686 | * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field | |
687 | * after the reply. | |
688 | */ | |
7bc53e94 | 689 | if (lur.ret_code == LTTNG_UST_OK) { |
40003310 MD |
690 | switch (lum->cmd) { |
691 | case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET: | |
692 | len = ustcomm_send_unix_sock(sock, | |
693 | &args.field_list.entry, | |
694 | sizeof(args.field_list.entry)); | |
7bc53e94 MD |
695 | if (len < 0) { |
696 | ret = len; | |
697 | goto error; | |
698 | } | |
40003310 | 699 | if (len != sizeof(args.field_list.entry)) { |
7bc53e94 | 700 | ret = -EINVAL; |
40003310 MD |
701 | goto error; |
702 | } | |
703 | } | |
704 | } | |
ef9ff354 | 705 | |
381c0f1e | 706 | error: |
17dfb34b | 707 | ust_unlock(); |
1ea11eab | 708 | return ret; |
d9e99d10 MD |
709 | } |
710 | ||
46050b1a | 711 | static |
efe0de09 | 712 | void cleanup_sock_info(struct sock_info *sock_info, int exiting) |
46050b1a MD |
713 | { |
714 | int ret; | |
715 | ||
5b14aab3 MD |
716 | if (sock_info->root_handle != -1) { |
717 | ret = lttng_ust_objd_unref(sock_info->root_handle, 1); | |
718 | if (ret) { | |
719 | ERR("Error unref root handle"); | |
720 | } | |
721 | sock_info->root_handle = -1; | |
722 | } | |
723 | sock_info->constructor_sem_posted = 0; | |
724 | ||
725 | /* | |
726 | * wait_shm_mmap, socket and notify socket are used by listener | |
727 | * threads outside of the ust lock, so we cannot tear them down | |
728 | * ourselves, because we cannot join on these threads. Leave | |
729 | * responsibility of cleaning up these resources to the OS | |
730 | * process exit. | |
731 | */ | |
732 | if (exiting) | |
733 | return; | |
734 | ||
46050b1a | 735 | if (sock_info->socket != -1) { |
e6973a89 | 736 | ret = ustcomm_close_unix_sock(sock_info->socket); |
46050b1a | 737 | if (ret) { |
32ce8569 | 738 | ERR("Error closing ust cmd socket"); |
46050b1a MD |
739 | } |
740 | sock_info->socket = -1; | |
741 | } | |
32ce8569 MD |
742 | if (sock_info->notify_socket != -1) { |
743 | ret = ustcomm_close_unix_sock(sock_info->notify_socket); | |
744 | if (ret) { | |
745 | ERR("Error closing ust notify socket"); | |
746 | } | |
747 | sock_info->notify_socket = -1; | |
748 | } | |
5b14aab3 | 749 | if (sock_info->wait_shm_mmap) { |
7fc90dca MD |
750 | ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE)); |
751 | if (ret) { | |
752 | ERR("Error unmapping wait shm"); | |
753 | } | |
754 | sock_info->wait_shm_mmap = NULL; | |
755 | } | |
756 | } | |
757 | ||
58d4b2a2 | 758 | /* |
33bbeb90 MD |
759 | * Using fork to set umask in the child process (not multi-thread safe). |
760 | * We deal with the shm_open vs ftruncate race (happening when the | |
761 | * sessiond owns the shm and does not let everybody modify it, to ensure | |
762 | * safety against shm_unlink) by simply letting the mmap fail and | |
763 | * retrying after a few seconds. | |
764 | * For global shm, everybody has rw access to it until the sessiond | |
765 | * starts. | |
58d4b2a2 | 766 | */ |
7fc90dca | 767 | static |
58d4b2a2 | 768 | int get_wait_shm(struct sock_info *sock_info, size_t mmap_size) |
7fc90dca | 769 | { |
7fc90dca | 770 | int wait_shm_fd, ret; |
58d4b2a2 | 771 | pid_t pid; |
44e073f5 | 772 | |
58d4b2a2 | 773 | /* |
33bbeb90 | 774 | * Try to open read-only. |
58d4b2a2 | 775 | */ |
33bbeb90 | 776 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 | 777 | if (wait_shm_fd >= 0) { |
7aa76730 MD |
778 | int32_t tmp_read; |
779 | ssize_t len; | |
780 | size_t bytes_read = 0; | |
781 | ||
782 | /* | |
783 | * Try to read the fd. If unable to do so, try opening | |
784 | * it in write mode. | |
785 | */ | |
786 | do { | |
787 | len = read(wait_shm_fd, | |
788 | &((char *) &tmp_read)[bytes_read], | |
789 | sizeof(tmp_read) - bytes_read); | |
790 | if (len > 0) { | |
791 | bytes_read += len; | |
792 | } | |
793 | } while ((len < 0 && errno == EINTR) | |
794 | || (len > 0 && bytes_read < sizeof(tmp_read))); | |
795 | if (bytes_read != sizeof(tmp_read)) { | |
796 | ret = close(wait_shm_fd); | |
797 | if (ret) { | |
798 | ERR("close wait_shm_fd"); | |
799 | } | |
800 | goto open_write; | |
801 | } | |
58d4b2a2 MD |
802 | goto end; |
803 | } else if (wait_shm_fd < 0 && errno != ENOENT) { | |
804 | /* | |
33bbeb90 MD |
805 | * Real-only open did not work, and it's not because the |
806 | * entry was not present. It's a failure that prohibits | |
807 | * using shm. | |
58d4b2a2 | 808 | */ |
7fc90dca | 809 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
58d4b2a2 | 810 | goto end; |
7fc90dca | 811 | } |
7aa76730 MD |
812 | |
813 | open_write: | |
7fc90dca | 814 | /* |
7aa76730 MD |
815 | * If the open failed because the file did not exist, or because |
816 | * the file was not truncated yet, try creating it ourself. | |
7fc90dca | 817 | */ |
8c90a710 | 818 | URCU_TLS(lttng_ust_nest_count)++; |
58d4b2a2 | 819 | pid = fork(); |
8c90a710 | 820 | URCU_TLS(lttng_ust_nest_count)--; |
58d4b2a2 MD |
821 | if (pid > 0) { |
822 | int status; | |
823 | ||
824 | /* | |
825 | * Parent: wait for child to return, in which case the | |
826 | * shared memory map will have been created. | |
827 | */ | |
828 | pid = wait(&status); | |
b7d3cb32 | 829 | if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) { |
58d4b2a2 MD |
830 | wait_shm_fd = -1; |
831 | goto end; | |
7fc90dca | 832 | } |
58d4b2a2 MD |
833 | /* |
834 | * Try to open read-only again after creation. | |
835 | */ | |
33bbeb90 | 836 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 MD |
837 | if (wait_shm_fd < 0) { |
838 | /* | |
839 | * Real-only open did not work. It's a failure | |
840 | * that prohibits using shm. | |
841 | */ | |
842 | ERR("Error opening shm %s", sock_info->wait_shm_path); | |
843 | goto end; | |
844 | } | |
845 | goto end; | |
846 | } else if (pid == 0) { | |
847 | int create_mode; | |
848 | ||
849 | /* Child */ | |
33bbeb90 | 850 | create_mode = S_IRUSR | S_IWUSR | S_IRGRP; |
58d4b2a2 | 851 | if (sock_info->global) |
33bbeb90 | 852 | create_mode |= S_IROTH | S_IWGRP | S_IWOTH; |
58d4b2a2 MD |
853 | /* |
854 | * We're alone in a child process, so we can modify the | |
855 | * process-wide umask. | |
856 | */ | |
33bbeb90 | 857 | umask(~create_mode); |
58d4b2a2 | 858 | /* |
33bbeb90 MD |
859 | * Try creating shm (or get rw access). |
860 | * We don't do an exclusive open, because we allow other | |
861 | * processes to create+ftruncate it concurrently. | |
58d4b2a2 MD |
862 | */ |
863 | wait_shm_fd = shm_open(sock_info->wait_shm_path, | |
864 | O_RDWR | O_CREAT, create_mode); | |
865 | if (wait_shm_fd >= 0) { | |
866 | ret = ftruncate(wait_shm_fd, mmap_size); | |
867 | if (ret) { | |
868 | PERROR("ftruncate"); | |
b0c1425d | 869 | _exit(EXIT_FAILURE); |
58d4b2a2 | 870 | } |
b0c1425d | 871 | _exit(EXIT_SUCCESS); |
58d4b2a2 | 872 | } |
33bbeb90 MD |
873 | /* |
874 | * For local shm, we need to have rw access to accept | |
875 | * opening it: this means the local sessiond will be | |
876 | * able to wake us up. For global shm, we open it even | |
877 | * if rw access is not granted, because the root.root | |
878 | * sessiond will be able to override all rights and wake | |
879 | * us up. | |
880 | */ | |
881 | if (!sock_info->global && errno != EACCES) { | |
58d4b2a2 | 882 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
5d3bc5ed | 883 | _exit(EXIT_FAILURE); |
58d4b2a2 MD |
884 | } |
885 | /* | |
33bbeb90 MD |
886 | * The shm exists, but we cannot open it RW. Report |
887 | * success. | |
58d4b2a2 | 888 | */ |
5d3bc5ed | 889 | _exit(EXIT_SUCCESS); |
58d4b2a2 MD |
890 | } else { |
891 | return -1; | |
7fc90dca | 892 | } |
58d4b2a2 | 893 | end: |
33bbeb90 MD |
894 | if (wait_shm_fd >= 0 && !sock_info->global) { |
895 | struct stat statbuf; | |
896 | ||
897 | /* | |
898 | * Ensure that our user is the owner of the shm file for | |
899 | * local shm. If we do not own the file, it means our | |
900 | * sessiond will not have access to wake us up (there is | |
901 | * probably a rogue process trying to fake our | |
902 | * sessiond). Fallback to polling method in this case. | |
903 | */ | |
904 | ret = fstat(wait_shm_fd, &statbuf); | |
905 | if (ret) { | |
906 | PERROR("fstat"); | |
907 | goto error_close; | |
908 | } | |
909 | if (statbuf.st_uid != getuid()) | |
910 | goto error_close; | |
911 | } | |
58d4b2a2 | 912 | return wait_shm_fd; |
33bbeb90 MD |
913 | |
914 | error_close: | |
915 | ret = close(wait_shm_fd); | |
916 | if (ret) { | |
917 | PERROR("Error closing fd"); | |
918 | } | |
919 | return -1; | |
58d4b2a2 MD |
920 | } |
921 | ||
922 | static | |
923 | char *get_map_shm(struct sock_info *sock_info) | |
924 | { | |
925 | size_t mmap_size = sysconf(_SC_PAGE_SIZE); | |
926 | int wait_shm_fd, ret; | |
927 | char *wait_shm_mmap; | |
928 | ||
929 | wait_shm_fd = get_wait_shm(sock_info, mmap_size); | |
930 | if (wait_shm_fd < 0) { | |
931 | goto error; | |
44e073f5 | 932 | } |
7fc90dca MD |
933 | wait_shm_mmap = mmap(NULL, mmap_size, PROT_READ, |
934 | MAP_SHARED, wait_shm_fd, 0); | |
7fc90dca MD |
935 | /* close shm fd immediately after taking the mmap reference */ |
936 | ret = close(wait_shm_fd); | |
937 | if (ret) { | |
33bbeb90 MD |
938 | PERROR("Error closing fd"); |
939 | } | |
940 | if (wait_shm_mmap == MAP_FAILED) { | |
941 | DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode."); | |
942 | goto error; | |
7fc90dca MD |
943 | } |
944 | return wait_shm_mmap; | |
945 | ||
946 | error: | |
947 | return NULL; | |
948 | } | |
949 | ||
950 | static | |
951 | void wait_for_sessiond(struct sock_info *sock_info) | |
952 | { | |
efe0de09 | 953 | int ret; |
80e2814b | 954 | |
7fc90dca MD |
955 | ust_lock(); |
956 | if (lttng_ust_comm_should_quit) { | |
957 | goto quit; | |
958 | } | |
37ed587a MD |
959 | if (wait_poll_fallback) { |
960 | goto error; | |
961 | } | |
7fc90dca MD |
962 | if (!sock_info->wait_shm_mmap) { |
963 | sock_info->wait_shm_mmap = get_map_shm(sock_info); | |
964 | if (!sock_info->wait_shm_mmap) | |
965 | goto error; | |
966 | } | |
967 | ust_unlock(); | |
968 | ||
969 | DBG("Waiting for %s apps sessiond", sock_info->name); | |
80e2814b MD |
970 | /* Wait for futex wakeup */ |
971 | if (uatomic_read((int32_t *) sock_info->wait_shm_mmap) == 0) { | |
972 | ret = futex_async((int32_t *) sock_info->wait_shm_mmap, | |
973 | FUTEX_WAIT, 0, NULL, NULL, 0); | |
80e2814b | 974 | if (ret < 0) { |
37ed587a MD |
975 | if (errno == EFAULT) { |
976 | wait_poll_fallback = 1; | |
a8b870ad | 977 | DBG( |
37ed587a MD |
978 | "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) " |
979 | "do not support FUTEX_WAKE on read-only memory mappings correctly. " | |
980 | "Please upgrade your kernel " | |
981 | "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel " | |
982 | "mainline). LTTng-UST will use polling mode fallback."); | |
cd27263b MD |
983 | if (ust_debug()) |
984 | PERROR("futex"); | |
37ed587a | 985 | } |
80e2814b MD |
986 | } |
987 | } | |
7fc90dca MD |
988 | return; |
989 | ||
990 | quit: | |
991 | ust_unlock(); | |
992 | return; | |
993 | ||
994 | error: | |
995 | ust_unlock(); | |
7fc90dca | 996 | return; |
46050b1a MD |
997 | } |
998 | ||
1ea11eab MD |
999 | /* |
1000 | * This thread does not allocate any resource, except within | |
1001 | * handle_message, within mutex protection. This mutex protects against | |
1002 | * fork and exit. | |
98bf993f | 1003 | * The other moment it allocates resources is at socket connection, which |
1ea11eab MD |
1004 | * is also protected by the mutex. |
1005 | */ | |
d9e99d10 MD |
1006 | static |
1007 | void *ust_listener_thread(void *arg) | |
1008 | { | |
1ea11eab | 1009 | struct sock_info *sock_info = arg; |
c0eedf81 | 1010 | int sock, ret, prev_connect_failed = 0, has_waited = 0; |
ff517991 | 1011 | long timeout; |
d9e99d10 | 1012 | |
9eb62b9c MD |
1013 | /* Restart trying to connect to the session daemon */ |
1014 | restart: | |
c0eedf81 MD |
1015 | if (prev_connect_failed) { |
1016 | /* Wait for sessiond availability with pipe */ | |
1017 | wait_for_sessiond(sock_info); | |
1018 | if (has_waited) { | |
1019 | has_waited = 0; | |
1020 | /* | |
1021 | * Sleep for 5 seconds before retrying after a | |
1022 | * sequence of failure / wait / failure. This | |
1023 | * deals with a killed or broken session daemon. | |
1024 | */ | |
1025 | sleep(5); | |
1026 | } | |
1027 | has_waited = 1; | |
1028 | prev_connect_failed = 0; | |
1029 | } | |
9eb62b9c | 1030 | |
1ea11eab | 1031 | if (sock_info->socket != -1) { |
e6973a89 | 1032 | ret = ustcomm_close_unix_sock(sock_info->socket); |
1ea11eab | 1033 | if (ret) { |
32ce8569 MD |
1034 | ERR("Error closing %s ust cmd socket", |
1035 | sock_info->name); | |
1ea11eab MD |
1036 | } |
1037 | sock_info->socket = -1; | |
1038 | } | |
32ce8569 MD |
1039 | if (sock_info->notify_socket != -1) { |
1040 | ret = ustcomm_close_unix_sock(sock_info->notify_socket); | |
1041 | if (ret) { | |
1042 | ERR("Error closing %s ust notify socket", | |
1043 | sock_info->name); | |
1044 | } | |
1045 | sock_info->notify_socket = -1; | |
1046 | } | |
46050b1a | 1047 | |
321f2351 MD |
1048 | /* |
1049 | * Register. We need to perform both connect and sending | |
1050 | * registration message before doing the next connect otherwise | |
1051 | * we may reach unix socket connect queue max limits and block | |
1052 | * on the 2nd connect while the session daemon is awaiting the | |
1053 | * first connect registration message. | |
1054 | */ | |
1055 | /* Connect cmd socket */ | |
1056 | ret = ustcomm_connect_unix_sock(sock_info->sock_path); | |
1057 | if (ret < 0) { | |
1058 | DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name); | |
1059 | prev_connect_failed = 1; | |
5b14aab3 | 1060 | |
321f2351 | 1061 | ust_lock(); |
5b14aab3 | 1062 | |
321f2351 MD |
1063 | if (lttng_ust_comm_should_quit) { |
1064 | goto quit; | |
32ce8569 | 1065 | } |
46050b1a | 1066 | |
e3426ddc | 1067 | /* |
321f2351 MD |
1068 | * If we cannot find the sessiond daemon, don't delay |
1069 | * constructor execution. | |
e3426ddc | 1070 | */ |
321f2351 MD |
1071 | ret = handle_register_done(sock_info); |
1072 | assert(!ret); | |
1073 | ust_unlock(); | |
1074 | goto restart; | |
27fe9f21 | 1075 | } |
321f2351 | 1076 | sock_info->socket = ret; |
27fe9f21 | 1077 | |
5b14aab3 MD |
1078 | ust_lock(); |
1079 | ||
1080 | if (lttng_ust_comm_should_quit) { | |
1081 | goto quit; | |
1082 | } | |
1083 | ||
46050b1a MD |
1084 | /* |
1085 | * Create only one root handle per listener thread for the whole | |
f59ed768 MD |
1086 | * process lifetime, so we ensure we get ID which is statically |
1087 | * assigned to the root handle. | |
46050b1a MD |
1088 | */ |
1089 | if (sock_info->root_handle == -1) { | |
1090 | ret = lttng_abi_create_root_handle(); | |
a51070bb | 1091 | if (ret < 0) { |
46050b1a | 1092 | ERR("Error creating root handle"); |
46050b1a MD |
1093 | goto quit; |
1094 | } | |
1095 | sock_info->root_handle = ret; | |
9eb62b9c | 1096 | } |
1ea11eab | 1097 | |
32ce8569 | 1098 | ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD); |
9eb62b9c | 1099 | if (ret < 0) { |
32ce8569 MD |
1100 | ERR("Error registering to %s ust cmd socket", |
1101 | sock_info->name); | |
c0eedf81 | 1102 | prev_connect_failed = 1; |
11ff9c7d MD |
1103 | /* |
1104 | * If we cannot register to the sessiond daemon, don't | |
1105 | * delay constructor execution. | |
1106 | */ | |
edaa1431 | 1107 | ret = handle_register_done(sock_info); |
11ff9c7d | 1108 | assert(!ret); |
17dfb34b | 1109 | ust_unlock(); |
9eb62b9c MD |
1110 | goto restart; |
1111 | } | |
321f2351 MD |
1112 | |
1113 | ust_unlock(); | |
1114 | ||
1115 | /* Connect notify socket */ | |
1116 | ret = ustcomm_connect_unix_sock(sock_info->sock_path); | |
1117 | if (ret < 0) { | |
1118 | DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name); | |
1119 | prev_connect_failed = 1; | |
1120 | ||
1121 | ust_lock(); | |
1122 | ||
1123 | if (lttng_ust_comm_should_quit) { | |
1124 | goto quit; | |
1125 | } | |
1126 | ||
1127 | /* | |
1128 | * If we cannot find the sessiond daemon, don't delay | |
1129 | * constructor execution. | |
1130 | */ | |
1131 | ret = handle_register_done(sock_info); | |
1132 | assert(!ret); | |
1133 | ust_unlock(); | |
1134 | goto restart; | |
1135 | } | |
1136 | sock_info->notify_socket = ret; | |
1137 | ||
1138 | timeout = get_notify_sock_timeout(); | |
1139 | if (timeout >= 0) { | |
1140 | /* | |
1141 | * Give at least 10ms to sessiond to reply to | |
1142 | * notifications. | |
1143 | */ | |
1144 | if (timeout < 10) | |
1145 | timeout = 10; | |
1146 | ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket, | |
1147 | timeout); | |
1148 | if (ret < 0) { | |
1149 | WARN("Error setting socket receive timeout"); | |
1150 | } | |
1151 | ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket, | |
1152 | timeout); | |
1153 | if (ret < 0) { | |
1154 | WARN("Error setting socket send timeout"); | |
1155 | } | |
1156 | } else if (timeout < -1) { | |
1157 | WARN("Unsupported timeout value %ld", timeout); | |
1158 | } | |
1159 | ||
1160 | ust_lock(); | |
1161 | ||
1162 | if (lttng_ust_comm_should_quit) { | |
1163 | goto quit; | |
1164 | } | |
1165 | ||
32ce8569 MD |
1166 | ret = register_to_sessiond(sock_info->notify_socket, |
1167 | USTCTL_SOCKET_NOTIFY); | |
1168 | if (ret < 0) { | |
1169 | ERR("Error registering to %s ust notify socket", | |
1170 | sock_info->name); | |
1171 | prev_connect_failed = 1; | |
1172 | /* | |
1173 | * If we cannot register to the sessiond daemon, don't | |
1174 | * delay constructor execution. | |
1175 | */ | |
1176 | ret = handle_register_done(sock_info); | |
1177 | assert(!ret); | |
1178 | ust_unlock(); | |
1179 | goto restart; | |
1180 | } | |
1181 | sock = sock_info->socket; | |
1182 | ||
17dfb34b | 1183 | ust_unlock(); |
46050b1a | 1184 | |
d9e99d10 MD |
1185 | for (;;) { |
1186 | ssize_t len; | |
57773204 | 1187 | struct ustcomm_ust_msg lum; |
d9e99d10 | 1188 | |
57773204 | 1189 | len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum)); |
d9e99d10 MD |
1190 | switch (len) { |
1191 | case 0: /* orderly shutdown */ | |
7dd08bec | 1192 | DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name); |
8236ba10 | 1193 | ust_lock(); |
d5e1fea6 | 1194 | if (lttng_ust_comm_should_quit) { |
d5e1fea6 MD |
1195 | goto quit; |
1196 | } | |
8236ba10 MD |
1197 | /* |
1198 | * Either sessiond has shutdown or refused us by closing the socket. | |
1199 | * In either case, we don't want to delay construction execution, | |
1200 | * and we need to wait before retry. | |
1201 | */ | |
1202 | prev_connect_failed = 1; | |
1203 | /* | |
1204 | * If we cannot register to the sessiond daemon, don't | |
1205 | * delay constructor execution. | |
1206 | */ | |
1207 | ret = handle_register_done(sock_info); | |
1208 | assert(!ret); | |
1209 | ust_unlock(); | |
d9e99d10 | 1210 | goto end; |
e7723462 | 1211 | case sizeof(lum): |
74d81a6c | 1212 | print_cmd(lum.cmd, lum.handle); |
11ff9c7d | 1213 | ret = handle_message(sock_info, sock, &lum); |
7bc53e94 | 1214 | if (ret) { |
11ff9c7d | 1215 | ERR("Error handling message for %s socket", sock_info->name); |
95c25348 | 1216 | } else { |
13436238 PW |
1217 | struct lttng_session *session; |
1218 | ||
1219 | session = sock_info->session_enabled; | |
95c25348 PW |
1220 | if (session) { |
1221 | sock_info->session_enabled = NULL; | |
1222 | lttng_ust_baddr_statedump(session); | |
1223 | } | |
d9e99d10 MD |
1224 | } |
1225 | continue; | |
7bc53e94 MD |
1226 | default: |
1227 | if (len < 0) { | |
1228 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
1229 | } else { | |
1230 | DBG("incorrect message size (%s socket): %zd", sock_info->name, len); | |
1231 | } | |
1232 | if (len == -ECONNRESET) { | |
1233 | DBG("%s remote end closed connection", sock_info->name); | |
d9e99d10 MD |
1234 | goto end; |
1235 | } | |
1236 | goto end; | |
d9e99d10 MD |
1237 | } |
1238 | ||
1239 | } | |
1240 | end: | |
f59ed768 | 1241 | ust_lock(); |
d5e1fea6 | 1242 | if (lttng_ust_comm_should_quit) { |
d5e1fea6 MD |
1243 | goto quit; |
1244 | } | |
f59ed768 MD |
1245 | /* Cleanup socket handles before trying to reconnect */ |
1246 | lttng_ust_objd_table_owner_cleanup(sock_info); | |
1247 | ust_unlock(); | |
9eb62b9c | 1248 | goto restart; /* try to reconnect */ |
e33f3265 | 1249 | |
1ea11eab | 1250 | quit: |
e33f3265 MD |
1251 | sock_info->thread_active = 0; |
1252 | ust_unlock(); | |
d9e99d10 MD |
1253 | return NULL; |
1254 | } | |
1255 | ||
2691221a MD |
1256 | /* |
1257 | * sessiond monitoring thread: monitor presence of global and per-user | |
1258 | * sessiond by polling the application common named pipe. | |
1259 | */ | |
edaa1431 | 1260 | void __attribute__((constructor)) lttng_ust_init(void) |
2691221a | 1261 | { |
11ff9c7d | 1262 | struct timespec constructor_timeout; |
ae6a58bf | 1263 | sigset_t sig_all_blocked, orig_parent_mask; |
1879f67f | 1264 | pthread_attr_t thread_attr; |
cf12a773 | 1265 | int timeout_mode; |
2691221a MD |
1266 | int ret; |
1267 | ||
edaa1431 MD |
1268 | if (uatomic_xchg(&initialized, 1) == 1) |
1269 | return; | |
1270 | ||
eddd8d5d MD |
1271 | /* |
1272 | * Fixup interdependency between TLS fixup mutex (which happens | |
1273 | * to be the dynamic linker mutex) and ust_lock, taken within | |
1274 | * the ust lock. | |
1275 | */ | |
f645cfa7 | 1276 | lttng_fixup_ringbuffer_tls(); |
4158a15a | 1277 | lttng_fixup_vtid_tls(); |
a903623f | 1278 | lttng_fixup_nest_count_tls(); |
009745db | 1279 | lttng_fixup_procname_tls(); |
eddd8d5d | 1280 | |
edaa1431 MD |
1281 | /* |
1282 | * We want precise control over the order in which we construct | |
1283 | * our sub-libraries vs starting to receive commands from | |
1284 | * sessiond (otherwise leading to errors when trying to create | |
1285 | * sessiond before the init functions are completed). | |
1286 | */ | |
2691221a | 1287 | init_usterr(); |
edaa1431 | 1288 | init_tracepoint(); |
7dd08bec MD |
1289 | lttng_ring_buffer_metadata_client_init(); |
1290 | lttng_ring_buffer_client_overwrite_init(); | |
34a91bdb | 1291 | lttng_ring_buffer_client_overwrite_rt_init(); |
7dd08bec | 1292 | lttng_ring_buffer_client_discard_init(); |
34a91bdb | 1293 | lttng_ring_buffer_client_discard_rt_init(); |
a0a3bef9 | 1294 | lttng_context_init(); |
2691221a | 1295 | |
ff517991 | 1296 | timeout_mode = get_constructor_timeout(&constructor_timeout); |
11ff9c7d | 1297 | |
95259bd0 | 1298 | ret = sem_init(&constructor_wait, 0, 0); |
11ff9c7d MD |
1299 | assert(!ret); |
1300 | ||
8d20bf54 | 1301 | ret = setup_local_apps(); |
2691221a | 1302 | if (ret) { |
9ec6895c | 1303 | DBG("local apps setup returned %d", ret); |
2691221a | 1304 | } |
ae6a58bf WP |
1305 | |
1306 | /* A new thread created by pthread_create inherits the signal mask | |
1307 | * from the parent. To avoid any signal being received by the | |
1308 | * listener thread, we block all signals temporarily in the parent, | |
1309 | * while we create the listener thread. | |
1310 | */ | |
1311 | sigfillset(&sig_all_blocked); | |
1312 | ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask); | |
1313 | if (ret) { | |
d94d802c | 1314 | ERR("pthread_sigmask: %s", strerror(ret)); |
ae6a58bf WP |
1315 | } |
1316 | ||
1879f67f MG |
1317 | ret = pthread_attr_init(&thread_attr); |
1318 | if (ret) { | |
1319 | ERR("pthread_attr_init: %s", strerror(ret)); | |
1320 | } | |
1321 | ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED); | |
1322 | if (ret) { | |
1323 | ERR("pthread_attr_setdetachstate: %s", strerror(ret)); | |
1324 | } | |
1325 | ||
e33f3265 | 1326 | ust_lock(); |
1879f67f | 1327 | ret = pthread_create(&global_apps.ust_listener, &thread_attr, |
dde70ea0 | 1328 | ust_listener_thread, &global_apps); |
d94d802c MD |
1329 | if (ret) { |
1330 | ERR("pthread_create global: %s", strerror(ret)); | |
1331 | } | |
e33f3265 MD |
1332 | global_apps.thread_active = 1; |
1333 | ust_unlock(); | |
1334 | ||
8d20bf54 | 1335 | if (local_apps.allowed) { |
e33f3265 | 1336 | ust_lock(); |
1879f67f | 1337 | ret = pthread_create(&local_apps.ust_listener, &thread_attr, |
dde70ea0 | 1338 | ust_listener_thread, &local_apps); |
d94d802c MD |
1339 | if (ret) { |
1340 | ERR("pthread_create local: %s", strerror(ret)); | |
1341 | } | |
e33f3265 MD |
1342 | local_apps.thread_active = 1; |
1343 | ust_unlock(); | |
8d20bf54 MD |
1344 | } else { |
1345 | handle_register_done(&local_apps); | |
1346 | } | |
1879f67f MG |
1347 | ret = pthread_attr_destroy(&thread_attr); |
1348 | if (ret) { | |
1349 | ERR("pthread_attr_destroy: %s", strerror(ret)); | |
1350 | } | |
8d20bf54 | 1351 | |
ae6a58bf WP |
1352 | /* Restore original signal mask in parent */ |
1353 | ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL); | |
1354 | if (ret) { | |
d94d802c | 1355 | ERR("pthread_sigmask: %s", strerror(ret)); |
ae6a58bf WP |
1356 | } |
1357 | ||
cf12a773 MD |
1358 | switch (timeout_mode) { |
1359 | case 1: /* timeout wait */ | |
95259bd0 MD |
1360 | do { |
1361 | ret = sem_timedwait(&constructor_wait, | |
1362 | &constructor_timeout); | |
1363 | } while (ret < 0 && errno == EINTR); | |
cf12a773 | 1364 | if (ret < 0 && errno == ETIMEDOUT) { |
7dd08bec | 1365 | ERR("Timed out waiting for lttng-sessiond"); |
cf12a773 MD |
1366 | } else { |
1367 | assert(!ret); | |
1368 | } | |
1369 | break; | |
7b766b16 | 1370 | case -1:/* wait forever */ |
95259bd0 MD |
1371 | do { |
1372 | ret = sem_wait(&constructor_wait); | |
1373 | } while (ret < 0 && errno == EINTR); | |
11ff9c7d | 1374 | assert(!ret); |
cf12a773 | 1375 | break; |
7b766b16 | 1376 | case 0: /* no timeout */ |
cf12a773 | 1377 | break; |
11ff9c7d | 1378 | } |
2691221a MD |
1379 | } |
1380 | ||
17dfb34b MD |
1381 | static |
1382 | void lttng_ust_cleanup(int exiting) | |
1383 | { | |
efe0de09 | 1384 | cleanup_sock_info(&global_apps, exiting); |
17dfb34b | 1385 | if (local_apps.allowed) { |
efe0de09 | 1386 | cleanup_sock_info(&local_apps, exiting); |
17dfb34b | 1387 | } |
efe0de09 MD |
1388 | /* |
1389 | * The teardown in this function all affect data structures | |
1390 | * accessed under the UST lock by the listener thread. This | |
1391 | * lock, along with the lttng_ust_comm_should_quit flag, ensure | |
1392 | * that none of these threads are accessing this data at this | |
1393 | * point. | |
1394 | */ | |
17dfb34b | 1395 | lttng_ust_abi_exit(); |
003fedf4 | 1396 | lttng_ust_events_exit(); |
a0a3bef9 | 1397 | lttng_context_exit(); |
34a91bdb | 1398 | lttng_ring_buffer_client_discard_rt_exit(); |
7dd08bec | 1399 | lttng_ring_buffer_client_discard_exit(); |
34a91bdb | 1400 | lttng_ring_buffer_client_overwrite_rt_exit(); |
7dd08bec MD |
1401 | lttng_ring_buffer_client_overwrite_exit(); |
1402 | lttng_ring_buffer_metadata_client_exit(); | |
17dfb34b MD |
1403 | exit_tracepoint(); |
1404 | if (!exiting) { | |
1405 | /* Reinitialize values for fork */ | |
1406 | sem_count = 2; | |
1407 | lttng_ust_comm_should_quit = 0; | |
1408 | initialized = 0; | |
1409 | } | |
1410 | } | |
1411 | ||
edaa1431 | 1412 | void __attribute__((destructor)) lttng_ust_exit(void) |
2691221a MD |
1413 | { |
1414 | int ret; | |
1415 | ||
9eb62b9c MD |
1416 | /* |
1417 | * Using pthread_cancel here because: | |
1418 | * A) we don't want to hang application teardown. | |
1419 | * B) the thread is not allocating any resource. | |
1420 | */ | |
1ea11eab MD |
1421 | |
1422 | /* | |
1423 | * Require the communication thread to quit. Synchronize with | |
1424 | * mutexes to ensure it is not in a mutex critical section when | |
1425 | * pthread_cancel is later called. | |
1426 | */ | |
17dfb34b | 1427 | ust_lock(); |
1ea11eab | 1428 | lttng_ust_comm_should_quit = 1; |
1ea11eab | 1429 | |
f5f94532 | 1430 | /* cancel threads */ |
e33f3265 MD |
1431 | if (global_apps.thread_active) { |
1432 | ret = pthread_cancel(global_apps.ust_listener); | |
1433 | if (ret) { | |
1434 | ERR("Error cancelling global ust listener thread: %s", | |
1435 | strerror(ret)); | |
1436 | } else { | |
1437 | global_apps.thread_active = 0; | |
1438 | } | |
2691221a | 1439 | } |
e33f3265 | 1440 | if (local_apps.thread_active) { |
8d20bf54 MD |
1441 | ret = pthread_cancel(local_apps.ust_listener); |
1442 | if (ret) { | |
d94d802c MD |
1443 | ERR("Error cancelling local ust listener thread: %s", |
1444 | strerror(ret)); | |
e33f3265 MD |
1445 | } else { |
1446 | local_apps.thread_active = 0; | |
8d20bf54 | 1447 | } |
8d20bf54 | 1448 | } |
e33f3265 MD |
1449 | ust_unlock(); |
1450 | ||
efe0de09 MD |
1451 | /* |
1452 | * Do NOT join threads: use of sys_futex makes it impossible to | |
1453 | * join the threads without using async-cancel, but async-cancel | |
1454 | * is delivered by a signal, which could hit the target thread | |
1455 | * anywhere in its code path, including while the ust_lock() is | |
1456 | * held, causing a deadlock for the other thread. Let the OS | |
1457 | * cleanup the threads if there are stalled in a syscall. | |
1458 | */ | |
17dfb34b | 1459 | lttng_ust_cleanup(1); |
2691221a | 1460 | } |
e822f505 MD |
1461 | |
1462 | /* | |
1463 | * We exclude the worker threads across fork and clone (except | |
1464 | * CLONE_VM), because these system calls only keep the forking thread | |
1465 | * running in the child. Therefore, we don't want to call fork or clone | |
1466 | * in the middle of an tracepoint or ust tracing state modification. | |
1467 | * Holding this mutex protects these structures across fork and clone. | |
1468 | */ | |
b728d87e | 1469 | void ust_before_fork(sigset_t *save_sigset) |
e822f505 MD |
1470 | { |
1471 | /* | |
1472 | * Disable signals. This is to avoid that the child intervenes | |
1473 | * before it is properly setup for tracing. It is safer to | |
1474 | * disable all signals, because then we know we are not breaking | |
1475 | * anything by restoring the original mask. | |
1476 | */ | |
1477 | sigset_t all_sigs; | |
1478 | int ret; | |
1479 | ||
8c90a710 | 1480 | if (URCU_TLS(lttng_ust_nest_count)) |
e8508a49 | 1481 | return; |
e822f505 MD |
1482 | /* Disable signals */ |
1483 | sigfillset(&all_sigs); | |
b728d87e | 1484 | ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset); |
e822f505 MD |
1485 | if (ret == -1) { |
1486 | PERROR("sigprocmask"); | |
1487 | } | |
17dfb34b | 1488 | ust_lock(); |
e822f505 MD |
1489 | rcu_bp_before_fork(); |
1490 | } | |
1491 | ||
b728d87e | 1492 | static void ust_after_fork_common(sigset_t *restore_sigset) |
e822f505 MD |
1493 | { |
1494 | int ret; | |
1495 | ||
17dfb34b MD |
1496 | DBG("process %d", getpid()); |
1497 | ust_unlock(); | |
e822f505 | 1498 | /* Restore signals */ |
23c8854a | 1499 | ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL); |
e822f505 MD |
1500 | if (ret == -1) { |
1501 | PERROR("sigprocmask"); | |
1502 | } | |
1503 | } | |
1504 | ||
b728d87e | 1505 | void ust_after_fork_parent(sigset_t *restore_sigset) |
e822f505 | 1506 | { |
8c90a710 | 1507 | if (URCU_TLS(lttng_ust_nest_count)) |
e8508a49 | 1508 | return; |
17dfb34b | 1509 | DBG("process %d", getpid()); |
e822f505 MD |
1510 | rcu_bp_after_fork_parent(); |
1511 | /* Release mutexes and reenable signals */ | |
b728d87e | 1512 | ust_after_fork_common(restore_sigset); |
e822f505 MD |
1513 | } |
1514 | ||
17dfb34b MD |
1515 | /* |
1516 | * After fork, in the child, we need to cleanup all the leftover state, | |
1517 | * except the worker thread which already magically disappeared thanks | |
1518 | * to the weird Linux fork semantics. After tyding up, we call | |
1519 | * lttng_ust_init() again to start over as a new PID. | |
1520 | * | |
1521 | * This is meant for forks() that have tracing in the child between the | |
1522 | * fork and following exec call (if there is any). | |
1523 | */ | |
b728d87e | 1524 | void ust_after_fork_child(sigset_t *restore_sigset) |
e822f505 | 1525 | { |
8c90a710 | 1526 | if (URCU_TLS(lttng_ust_nest_count)) |
e8508a49 | 1527 | return; |
17dfb34b | 1528 | DBG("process %d", getpid()); |
e822f505 MD |
1529 | /* Release urcu mutexes */ |
1530 | rcu_bp_after_fork_child(); | |
17dfb34b | 1531 | lttng_ust_cleanup(0); |
a93bfc45 | 1532 | lttng_context_vtid_reset(); |
e822f505 | 1533 | /* Release mutexes and reenable signals */ |
b728d87e | 1534 | ust_after_fork_common(restore_sigset); |
318dfea9 | 1535 | lttng_ust_init(); |
e822f505 | 1536 | } |
95c25348 PW |
1537 | |
1538 | void lttng_ust_sockinfo_session_enabled(void *owner, | |
1539 | struct lttng_session *session_enabled) | |
1540 | { | |
1541 | struct sock_info *sock_info = owner; | |
1542 | sock_info->session_enabled = session_enabled; | |
1543 | } |