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 |
5e1b7b8b | 23 | #define _GNU_SOURCE |
2691221a MD |
24 | #include <sys/types.h> |
25 | #include <sys/socket.h> | |
7fc90dca MD |
26 | #include <sys/mman.h> |
27 | #include <sys/stat.h> | |
58d4b2a2 MD |
28 | #include <sys/types.h> |
29 | #include <sys/wait.h> | |
7fc90dca | 30 | #include <fcntl.h> |
2691221a MD |
31 | #include <unistd.h> |
32 | #include <errno.h> | |
d9e99d10 | 33 | #include <pthread.h> |
11ff9c7d MD |
34 | #include <semaphore.h> |
35 | #include <time.h> | |
1ea11eab | 36 | #include <assert.h> |
e822f505 | 37 | #include <signal.h> |
6f97f9c2 | 38 | #include <limits.h> |
95259bd0 | 39 | #include <urcu/uatomic.h> |
80e2814b | 40 | #include <urcu/futex.h> |
c117fb1b | 41 | #include <urcu/compiler.h> |
1ea11eab | 42 | |
4318ae1b | 43 | #include <lttng/ust-events.h> |
4318ae1b | 44 | #include <lttng/ust-abi.h> |
4318ae1b | 45 | #include <lttng/ust.h> |
7bc53e94 | 46 | #include <lttng/ust-error.h> |
74d81a6c | 47 | #include <lttng/ust-ctl.h> |
8c90a710 | 48 | #include <urcu/tls-compat.h> |
44c72f10 | 49 | #include <ust-comm.h> |
6548fca4 | 50 | #include <ust-fd.h> |
44c72f10 | 51 | #include <usterr-signal-safe.h> |
cd54f6d9 | 52 | #include <helper.h> |
44c72f10 | 53 | #include "tracepoint-internal.h" |
7dd08bec | 54 | #include "lttng-tracer-core.h" |
08114193 | 55 | #include "compat.h" |
6f97f9c2 | 56 | #include "../libringbuffer/rb-init.h" |
cf73e0fe | 57 | #include "lttng-ust-statedump.h" |
f9364363 | 58 | #include "clock.h" |
5e1b7b8b | 59 | #include "../libringbuffer/getcpu.h" |
13efba44 | 60 | #include "getenv.h" |
edaa1431 MD |
61 | |
62 | /* | |
63 | * Has lttng ust comm constructor been called ? | |
64 | */ | |
65 | static int initialized; | |
66 | ||
1ea11eab | 67 | /* |
17dfb34b MD |
68 | * The ust_lock/ust_unlock lock is used as a communication thread mutex. |
69 | * Held when handling a command, also held by fork() to deal with | |
70 | * removal of threads, and by exit path. | |
3327ac33 MD |
71 | * |
72 | * The UST lock is the centralized mutex across UST tracing control and | |
73 | * probe registration. | |
74 | * | |
75 | * ust_exit_mutex must never nest in ust_mutex. | |
d58d1454 | 76 | * |
4770bd47 MD |
77 | * ust_fork_mutex must never nest in ust_mutex. |
78 | * | |
d58d1454 MD |
79 | * ust_mutex_nest is a per-thread nesting counter, allowing the perf |
80 | * counter lazy initialization called by events within the statedump, | |
81 | * which traces while the ust_mutex is held. | |
4770bd47 MD |
82 | * |
83 | * ust_lock nests within the dynamic loader lock (within glibc) because | |
84 | * it is taken within the library constructor. | |
3327ac33 MD |
85 | */ |
86 | static pthread_mutex_t ust_mutex = PTHREAD_MUTEX_INITIALIZER; | |
87 | ||
d58d1454 MD |
88 | /* Allow nesting the ust_mutex within the same thread. */ |
89 | static DEFINE_URCU_TLS(int, ust_mutex_nest); | |
90 | ||
3327ac33 MD |
91 | /* |
92 | * ust_exit_mutex protects thread_active variable wrt thread exit. It | |
93 | * cannot be done by ust_mutex because pthread_cancel(), which takes an | |
94 | * internal libc lock, cannot nest within ust_mutex. | |
95 | * | |
96 | * It never nests within a ust_mutex. | |
1ea11eab | 97 | */ |
3327ac33 | 98 | static pthread_mutex_t ust_exit_mutex = PTHREAD_MUTEX_INITIALIZER; |
1ea11eab | 99 | |
458d678c PW |
100 | /* |
101 | * ust_fork_mutex protects base address statedump tracing against forks. It | |
102 | * prevents the dynamic loader lock to be taken (by base address statedump | |
103 | * tracing) while a fork is happening, thus preventing deadlock issues with | |
104 | * the dynamic loader lock. | |
105 | */ | |
106 | static pthread_mutex_t ust_fork_mutex = PTHREAD_MUTEX_INITIALIZER; | |
107 | ||
1ea11eab MD |
108 | /* Should the ust comm thread quit ? */ |
109 | static int lttng_ust_comm_should_quit; | |
110 | ||
07b57e5e MD |
111 | /* |
112 | * This variable can be tested by applications to check whether | |
113 | * lttng-ust is loaded. They simply have to define their own | |
114 | * "lttng_ust_loaded" weak symbol, and test it. It is set to 1 by the | |
115 | * library constructor. | |
116 | */ | |
117 | int lttng_ust_loaded __attribute__((weak)); | |
118 | ||
3327ac33 | 119 | /* |
d58d1454 | 120 | * Return 0 on success, -1 if should quit. |
3327ac33 | 121 | * The lock is taken in both cases. |
d58d1454 | 122 | * Signal-safe. |
3327ac33 MD |
123 | */ |
124 | int ust_lock(void) | |
125 | { | |
d58d1454 | 126 | sigset_t sig_all_blocked, orig_mask; |
e446ad80 | 127 | int ret, oldstate; |
d58d1454 | 128 | |
e446ad80 MD |
129 | ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate); |
130 | if (ret) { | |
131 | ERR("pthread_setcancelstate: %s", strerror(ret)); | |
132 | } | |
133 | if (oldstate != PTHREAD_CANCEL_ENABLE) { | |
134 | ERR("pthread_setcancelstate: unexpected oldstate"); | |
135 | } | |
d58d1454 MD |
136 | sigfillset(&sig_all_blocked); |
137 | ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask); | |
138 | if (ret) { | |
139 | ERR("pthread_sigmask: %s", strerror(ret)); | |
140 | } | |
141 | if (!URCU_TLS(ust_mutex_nest)++) | |
142 | pthread_mutex_lock(&ust_mutex); | |
143 | ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL); | |
144 | if (ret) { | |
145 | ERR("pthread_sigmask: %s", strerror(ret)); | |
146 | } | |
3327ac33 MD |
147 | if (lttng_ust_comm_should_quit) { |
148 | return -1; | |
149 | } else { | |
150 | return 0; | |
151 | } | |
152 | } | |
153 | ||
154 | /* | |
155 | * ust_lock_nocheck() can be used in constructors/destructors, because | |
156 | * they are already nested within the dynamic loader lock, and therefore | |
157 | * have exclusive access against execution of liblttng-ust destructor. | |
d58d1454 | 158 | * Signal-safe. |
3327ac33 MD |
159 | */ |
160 | void ust_lock_nocheck(void) | |
161 | { | |
d58d1454 | 162 | sigset_t sig_all_blocked, orig_mask; |
e446ad80 | 163 | int ret, oldstate; |
d58d1454 | 164 | |
e446ad80 MD |
165 | ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate); |
166 | if (ret) { | |
167 | ERR("pthread_setcancelstate: %s", strerror(ret)); | |
168 | } | |
169 | if (oldstate != PTHREAD_CANCEL_ENABLE) { | |
170 | ERR("pthread_setcancelstate: unexpected oldstate"); | |
171 | } | |
d58d1454 MD |
172 | sigfillset(&sig_all_blocked); |
173 | ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask); | |
174 | if (ret) { | |
175 | ERR("pthread_sigmask: %s", strerror(ret)); | |
176 | } | |
177 | if (!URCU_TLS(ust_mutex_nest)++) | |
178 | pthread_mutex_lock(&ust_mutex); | |
179 | ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL); | |
180 | if (ret) { | |
181 | ERR("pthread_sigmask: %s", strerror(ret)); | |
182 | } | |
3327ac33 MD |
183 | } |
184 | ||
d58d1454 MD |
185 | /* |
186 | * Signal-safe. | |
187 | */ | |
3327ac33 MD |
188 | void ust_unlock(void) |
189 | { | |
d58d1454 | 190 | sigset_t sig_all_blocked, orig_mask; |
e446ad80 | 191 | int ret, oldstate; |
d58d1454 MD |
192 | |
193 | sigfillset(&sig_all_blocked); | |
194 | ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask); | |
195 | if (ret) { | |
196 | ERR("pthread_sigmask: %s", strerror(ret)); | |
197 | } | |
198 | if (!--URCU_TLS(ust_mutex_nest)) | |
199 | pthread_mutex_unlock(&ust_mutex); | |
200 | ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL); | |
201 | if (ret) { | |
202 | ERR("pthread_sigmask: %s", strerror(ret)); | |
203 | } | |
e446ad80 MD |
204 | ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &oldstate); |
205 | if (ret) { | |
206 | ERR("pthread_setcancelstate: %s", strerror(ret)); | |
207 | } | |
208 | if (oldstate != PTHREAD_CANCEL_DISABLE) { | |
209 | ERR("pthread_setcancelstate: unexpected oldstate"); | |
210 | } | |
3327ac33 MD |
211 | } |
212 | ||
11ff9c7d MD |
213 | /* |
214 | * Wait for either of these before continuing to the main | |
215 | * program: | |
216 | * - the register_done message from sessiond daemon | |
217 | * (will let the sessiond daemon enable sessions before main | |
218 | * starts.) | |
219 | * - sessiond daemon is not reachable. | |
220 | * - timeout (ensuring applications are resilient to session | |
221 | * daemon problems). | |
222 | */ | |
223 | static sem_t constructor_wait; | |
950aab0c MD |
224 | /* |
225 | * Doing this for both the global and local sessiond. | |
226 | */ | |
95259bd0 | 227 | static int sem_count = { 2 }; |
11ff9c7d | 228 | |
e8508a49 MD |
229 | /* |
230 | * Counting nesting within lttng-ust. Used to ensure that calling fork() | |
231 | * from liblttng-ust does not execute the pre/post fork handlers. | |
232 | */ | |
8c90a710 | 233 | static DEFINE_URCU_TLS(int, lttng_ust_nest_count); |
e8508a49 | 234 | |
1ea11eab MD |
235 | /* |
236 | * Info about socket and associated listener thread. | |
237 | */ | |
238 | struct sock_info { | |
11ff9c7d | 239 | const char *name; |
1ea11eab | 240 | pthread_t ust_listener; /* listener thread */ |
46050b1a | 241 | int root_handle; |
8d20bf54 MD |
242 | int constructor_sem_posted; |
243 | int allowed; | |
44e073f5 | 244 | int global; |
e33f3265 | 245 | int thread_active; |
7fc90dca MD |
246 | |
247 | char sock_path[PATH_MAX]; | |
248 | int socket; | |
32ce8569 | 249 | int notify_socket; |
7fc90dca MD |
250 | |
251 | char wait_shm_path[PATH_MAX]; | |
252 | char *wait_shm_mmap; | |
37dddb65 MD |
253 | /* Keep track of lazy state dump not performed yet. */ |
254 | int statedump_pending; | |
1ea11eab | 255 | }; |
2691221a MD |
256 | |
257 | /* Socket from app (connect) to session daemon (listen) for communication */ | |
1ea11eab | 258 | struct sock_info global_apps = { |
11ff9c7d | 259 | .name = "global", |
44e073f5 | 260 | .global = 1, |
7fc90dca | 261 | |
46050b1a | 262 | .root_handle = -1, |
8d20bf54 | 263 | .allowed = 1, |
e33f3265 | 264 | .thread_active = 0, |
7fc90dca | 265 | |
32ce8569 | 266 | .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME, |
7fc90dca | 267 | .socket = -1, |
32ce8569 | 268 | .notify_socket = -1, |
7fc90dca | 269 | |
32ce8569 | 270 | .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME, |
95c25348 | 271 | |
37dddb65 | 272 | .statedump_pending = 0, |
1ea11eab | 273 | }; |
2691221a MD |
274 | |
275 | /* TODO: allow global_apps_sock_path override */ | |
276 | ||
1ea11eab | 277 | struct sock_info local_apps = { |
11ff9c7d | 278 | .name = "local", |
44e073f5 | 279 | .global = 0, |
46050b1a | 280 | .root_handle = -1, |
8d20bf54 | 281 | .allowed = 0, /* Check setuid bit first */ |
e33f3265 | 282 | .thread_active = 0, |
7fc90dca MD |
283 | |
284 | .socket = -1, | |
32ce8569 | 285 | .notify_socket = -1, |
95c25348 | 286 | |
37dddb65 | 287 | .statedump_pending = 0, |
1ea11eab | 288 | }; |
2691221a | 289 | |
37ed587a MD |
290 | static int wait_poll_fallback; |
291 | ||
74d81a6c MD |
292 | static const char *cmd_name_mapping[] = { |
293 | [ LTTNG_UST_RELEASE ] = "Release", | |
294 | [ LTTNG_UST_SESSION ] = "Create Session", | |
295 | [ LTTNG_UST_TRACER_VERSION ] = "Get Tracer Version", | |
296 | ||
297 | [ LTTNG_UST_TRACEPOINT_LIST ] = "Create Tracepoint List", | |
298 | [ LTTNG_UST_WAIT_QUIESCENT ] = "Wait for Quiescent State", | |
299 | [ LTTNG_UST_REGISTER_DONE ] = "Registration Done", | |
300 | [ LTTNG_UST_TRACEPOINT_FIELD_LIST ] = "Create Tracepoint Field List", | |
301 | ||
302 | /* Session FD commands */ | |
303 | [ LTTNG_UST_CHANNEL ] = "Create Channel", | |
304 | [ LTTNG_UST_SESSION_START ] = "Start Session", | |
305 | [ LTTNG_UST_SESSION_STOP ] = "Stop Session", | |
306 | ||
307 | /* Channel FD commands */ | |
308 | [ LTTNG_UST_STREAM ] = "Create Stream", | |
309 | [ LTTNG_UST_EVENT ] = "Create Event", | |
310 | ||
311 | /* Event and Channel FD commands */ | |
312 | [ LTTNG_UST_CONTEXT ] = "Create Context", | |
313 | [ LTTNG_UST_FLUSH_BUFFER ] = "Flush Buffer", | |
314 | ||
315 | /* Event, Channel and Session commands */ | |
316 | [ LTTNG_UST_ENABLE ] = "Enable", | |
317 | [ LTTNG_UST_DISABLE ] = "Disable", | |
318 | ||
319 | /* Tracepoint list commands */ | |
320 | [ LTTNG_UST_TRACEPOINT_LIST_GET ] = "List Next Tracepoint", | |
321 | [ LTTNG_UST_TRACEPOINT_FIELD_LIST_GET ] = "List Next Tracepoint Field", | |
322 | ||
323 | /* Event FD commands */ | |
324 | [ LTTNG_UST_FILTER ] = "Create Filter", | |
75582b3a | 325 | [ LTTNG_UST_EXCLUSION ] = "Add exclusions to event", |
74d81a6c MD |
326 | }; |
327 | ||
ff517991 MD |
328 | static const char *str_timeout; |
329 | static int got_timeout_env; | |
330 | ||
7dd08bec | 331 | extern void lttng_ring_buffer_client_overwrite_init(void); |
34a91bdb | 332 | extern void lttng_ring_buffer_client_overwrite_rt_init(void); |
7dd08bec | 333 | extern void lttng_ring_buffer_client_discard_init(void); |
34a91bdb | 334 | extern void lttng_ring_buffer_client_discard_rt_init(void); |
7dd08bec MD |
335 | extern void lttng_ring_buffer_metadata_client_init(void); |
336 | extern void lttng_ring_buffer_client_overwrite_exit(void); | |
34a91bdb | 337 | extern void lttng_ring_buffer_client_overwrite_rt_exit(void); |
7dd08bec | 338 | extern void lttng_ring_buffer_client_discard_exit(void); |
34a91bdb | 339 | extern void lttng_ring_buffer_client_discard_rt_exit(void); |
7dd08bec | 340 | extern void lttng_ring_buffer_metadata_client_exit(void); |
edaa1431 | 341 | |
405be658 MD |
342 | ssize_t lttng_ust_read(int fd, void *buf, size_t len) |
343 | { | |
344 | ssize_t ret; | |
345 | size_t copied = 0, to_copy = len; | |
346 | ||
347 | do { | |
348 | ret = read(fd, buf + copied, to_copy); | |
349 | if (ret > 0) { | |
350 | copied += ret; | |
351 | to_copy -= ret; | |
352 | } | |
353 | } while ((ret > 0 && to_copy > 0) | |
354 | || (ret < 0 && errno == EINTR)); | |
355 | if (ret > 0) { | |
356 | ret = copied; | |
357 | } | |
358 | return ret; | |
359 | } | |
3c6f6263 AM |
360 | /* |
361 | * Returns the HOME directory path. Caller MUST NOT free(3) the returned | |
362 | * pointer. | |
363 | */ | |
364 | static | |
365 | const char *get_lttng_home_dir(void) | |
366 | { | |
367 | const char *val; | |
368 | ||
6f626d28 | 369 | val = (const char *) lttng_getenv("LTTNG_HOME"); |
3c6f6263 AM |
370 | if (val != NULL) { |
371 | return val; | |
372 | } | |
6f626d28 | 373 | return (const char *) lttng_getenv("HOME"); |
3c6f6263 AM |
374 | } |
375 | ||
a903623f MD |
376 | /* |
377 | * Force a read (imply TLS fixup for dlopen) of TLS variables. | |
378 | */ | |
379 | static | |
380 | void lttng_fixup_nest_count_tls(void) | |
381 | { | |
8c90a710 | 382 | asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count))); |
a903623f MD |
383 | } |
384 | ||
d58d1454 MD |
385 | static |
386 | void lttng_fixup_ust_mutex_nest_tls(void) | |
387 | { | |
388 | asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest))); | |
389 | } | |
390 | ||
1556a549 MD |
391 | /* |
392 | * Fixup urcu bp TLS. | |
393 | */ | |
394 | static | |
395 | void lttng_fixup_urcu_bp_tls(void) | |
396 | { | |
397 | rcu_read_lock(); | |
398 | rcu_read_unlock(); | |
399 | } | |
400 | ||
c362addf MD |
401 | void lttng_ust_fixup_tls(void) |
402 | { | |
403 | lttng_fixup_urcu_bp_tls(); | |
404 | lttng_fixup_ringbuffer_tls(); | |
405 | lttng_fixup_vtid_tls(); | |
406 | lttng_fixup_nest_count_tls(); | |
407 | lttng_fixup_procname_tls(); | |
408 | lttng_fixup_ust_mutex_nest_tls(); | |
6548fca4 | 409 | lttng_ust_fixup_fd_tracker_tls(); |
c362addf MD |
410 | } |
411 | ||
32ce8569 MD |
412 | int lttng_get_notify_socket(void *owner) |
413 | { | |
414 | struct sock_info *info = owner; | |
415 | ||
416 | return info->notify_socket; | |
417 | } | |
418 | ||
74d81a6c MD |
419 | static |
420 | void print_cmd(int cmd, int handle) | |
421 | { | |
422 | const char *cmd_name = "Unknown"; | |
423 | ||
fd67a004 MD |
424 | if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping) |
425 | && cmd_name_mapping[cmd]) { | |
74d81a6c MD |
426 | cmd_name = cmd_name_mapping[cmd]; |
427 | } | |
fd67a004 MD |
428 | DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)", |
429 | cmd_name, cmd, | |
74d81a6c MD |
430 | lttng_ust_obj_get_name(handle), handle); |
431 | } | |
432 | ||
2691221a | 433 | static |
8d20bf54 | 434 | int setup_local_apps(void) |
2691221a MD |
435 | { |
436 | const char *home_dir; | |
7fc90dca | 437 | uid_t uid; |
2691221a | 438 | |
7fc90dca | 439 | uid = getuid(); |
8d20bf54 MD |
440 | /* |
441 | * Disallow per-user tracing for setuid binaries. | |
442 | */ | |
7fc90dca | 443 | if (uid != geteuid()) { |
9ec6895c | 444 | assert(local_apps.allowed == 0); |
d0a1ae63 | 445 | return 0; |
8d20bf54 | 446 | } |
3c6f6263 | 447 | home_dir = get_lttng_home_dir(); |
9ec6895c MD |
448 | if (!home_dir) { |
449 | WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing."); | |
450 | assert(local_apps.allowed == 0); | |
2691221a | 451 | return -ENOENT; |
9ec6895c MD |
452 | } |
453 | local_apps.allowed = 1; | |
32ce8569 MD |
454 | snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s", |
455 | home_dir, | |
456 | LTTNG_DEFAULT_HOME_RUNDIR, | |
457 | LTTNG_UST_SOCK_FILENAME); | |
458 | snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u", | |
459 | LTTNG_UST_WAIT_FILENAME, | |
460 | uid); | |
2691221a MD |
461 | return 0; |
462 | } | |
463 | ||
ff517991 | 464 | /* |
451d66b2 | 465 | * Get socket timeout, in ms. |
28515902 | 466 | * -1: wait forever. 0: don't wait. >0: timeout, in ms. |
ff517991 MD |
467 | */ |
468 | static | |
469 | long get_timeout(void) | |
470 | { | |
471 | long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS; | |
472 | ||
473 | if (!got_timeout_env) { | |
6f626d28 | 474 | str_timeout = lttng_getenv("LTTNG_UST_REGISTER_TIMEOUT"); |
ff517991 MD |
475 | got_timeout_env = 1; |
476 | } | |
477 | if (str_timeout) | |
478 | constructor_delay_ms = strtol(str_timeout, NULL, 10); | |
5cf81d53 MD |
479 | /* All negative values are considered as "-1". */ |
480 | if (constructor_delay_ms < -1) | |
481 | constructor_delay_ms = -1; | |
ff517991 MD |
482 | return constructor_delay_ms; |
483 | } | |
484 | ||
451d66b2 | 485 | /* Timeout for notify socket send and recv. */ |
ff517991 MD |
486 | static |
487 | long get_notify_sock_timeout(void) | |
488 | { | |
489 | return get_timeout(); | |
490 | } | |
491 | ||
451d66b2 MD |
492 | /* Timeout for connecting to cmd and notify sockets. */ |
493 | static | |
494 | long get_connect_sock_timeout(void) | |
495 | { | |
496 | return get_timeout(); | |
497 | } | |
498 | ||
ff517991 | 499 | /* |
28515902 | 500 | * Return values: -1: wait forever. 0: don't wait. 1: timeout wait. |
ff517991 MD |
501 | */ |
502 | static | |
503 | int get_constructor_timeout(struct timespec *constructor_timeout) | |
504 | { | |
505 | long constructor_delay_ms; | |
506 | int ret; | |
507 | ||
508 | constructor_delay_ms = get_timeout(); | |
509 | ||
510 | switch (constructor_delay_ms) { | |
511 | case -1:/* fall-through */ | |
512 | case 0: | |
513 | return constructor_delay_ms; | |
514 | default: | |
515 | break; | |
516 | } | |
517 | ||
518 | /* | |
519 | * If we are unable to find the current time, don't wait. | |
520 | */ | |
521 | ret = clock_gettime(CLOCK_REALTIME, constructor_timeout); | |
522 | if (ret) { | |
28515902 JG |
523 | /* Don't wait. */ |
524 | return 0; | |
ff517991 MD |
525 | } |
526 | constructor_timeout->tv_sec += constructor_delay_ms / 1000UL; | |
527 | constructor_timeout->tv_nsec += | |
528 | (constructor_delay_ms % 1000UL) * 1000000UL; | |
529 | if (constructor_timeout->tv_nsec >= 1000000000UL) { | |
530 | constructor_timeout->tv_sec++; | |
531 | constructor_timeout->tv_nsec -= 1000000000UL; | |
532 | } | |
28515902 | 533 | /* Timeout wait (constructor_delay_ms). */ |
ff517991 MD |
534 | return 1; |
535 | } | |
536 | ||
6f97f9c2 | 537 | static |
9050321d | 538 | void get_allow_blocking(void) |
6f97f9c2 | 539 | { |
9050321d MD |
540 | const char *str_allow_blocking = |
541 | lttng_getenv("LTTNG_UST_ALLOW_BLOCKING"); | |
542 | ||
543 | if (str_allow_blocking) { | |
544 | DBG("%s environment variable is set", | |
545 | "LTTNG_UST_ALLOW_BLOCKING"); | |
546 | lttng_ust_ringbuffer_set_allow_blocking(); | |
6f97f9c2 MD |
547 | } |
548 | } | |
549 | ||
2691221a | 550 | static |
32ce8569 | 551 | int register_to_sessiond(int socket, enum ustctl_socket_type type) |
2691221a | 552 | { |
32ce8569 MD |
553 | return ustcomm_send_reg_msg(socket, |
554 | type, | |
555 | CAA_BITS_PER_LONG, | |
556 | lttng_alignof(uint8_t) * CHAR_BIT, | |
557 | lttng_alignof(uint16_t) * CHAR_BIT, | |
558 | lttng_alignof(uint32_t) * CHAR_BIT, | |
559 | lttng_alignof(uint64_t) * CHAR_BIT, | |
560 | lttng_alignof(unsigned long) * CHAR_BIT); | |
2691221a MD |
561 | } |
562 | ||
d9e99d10 | 563 | static |
57773204 | 564 | int send_reply(int sock, struct ustcomm_ust_reply *lur) |
d9e99d10 | 565 | { |
9eb62b9c | 566 | ssize_t len; |
d3a492d1 | 567 | |
57773204 | 568 | len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur)); |
d3a492d1 | 569 | switch (len) { |
a4be8962 | 570 | case sizeof(*lur): |
d3a492d1 MD |
571 | DBG("message successfully sent"); |
572 | return 0; | |
7bc53e94 MD |
573 | default: |
574 | if (len == -ECONNRESET) { | |
575 | DBG("remote end closed connection"); | |
d3a492d1 MD |
576 | return 0; |
577 | } | |
7bc53e94 MD |
578 | if (len < 0) |
579 | return len; | |
580 | DBG("incorrect message size: %zd", len); | |
581 | return -EINVAL; | |
d3a492d1 MD |
582 | } |
583 | } | |
584 | ||
585 | static | |
edaa1431 | 586 | int handle_register_done(struct sock_info *sock_info) |
11ff9c7d MD |
587 | { |
588 | int ret; | |
589 | ||
edaa1431 MD |
590 | if (sock_info->constructor_sem_posted) |
591 | return 0; | |
592 | sock_info->constructor_sem_posted = 1; | |
56cd7e2f MD |
593 | if (uatomic_read(&sem_count) <= 0) { |
594 | return 0; | |
595 | } | |
95259bd0 MD |
596 | ret = uatomic_add_return(&sem_count, -1); |
597 | if (ret == 0) { | |
598 | ret = sem_post(&constructor_wait); | |
599 | assert(!ret); | |
600 | } | |
11ff9c7d MD |
601 | return 0; |
602 | } | |
603 | ||
37dddb65 MD |
604 | /* |
605 | * Only execute pending statedump after the constructor semaphore has | |
606 | * been posted by each listener thread. This means statedump will only | |
607 | * be performed after the "registration done" command is received from | |
608 | * each session daemon the application is connected to. | |
609 | * | |
610 | * This ensures we don't run into deadlock issues with the dynamic | |
611 | * loader mutex, which is held while the constructor is called and | |
612 | * waiting on the constructor semaphore. All operations requiring this | |
613 | * dynamic loader lock need to be postponed using this mechanism. | |
614 | */ | |
615 | static | |
616 | void handle_pending_statedump(struct sock_info *sock_info) | |
617 | { | |
618 | int ctor_passed = sock_info->constructor_sem_posted; | |
619 | ||
620 | if (ctor_passed && sock_info->statedump_pending) { | |
621 | sock_info->statedump_pending = 0; | |
2932a87f | 622 | pthread_mutex_lock(&ust_fork_mutex); |
37dddb65 | 623 | lttng_handle_pending_statedump(sock_info); |
458d678c | 624 | pthread_mutex_unlock(&ust_fork_mutex); |
37dddb65 MD |
625 | } |
626 | } | |
627 | ||
11ff9c7d MD |
628 | static |
629 | int handle_message(struct sock_info *sock_info, | |
57773204 | 630 | int sock, struct ustcomm_ust_msg *lum) |
d3a492d1 | 631 | { |
1ea11eab | 632 | int ret = 0; |
b61ce3b2 | 633 | const struct lttng_ust_objd_ops *ops; |
57773204 | 634 | struct ustcomm_ust_reply lur; |
ef9ff354 | 635 | union ust_args args; |
8e696cfa | 636 | char ctxstr[LTTNG_UST_SYM_NAME_LEN]; /* App context string. */ |
40003310 | 637 | ssize_t len; |
1ea11eab | 638 | |
46050b1a MD |
639 | memset(&lur, 0, sizeof(lur)); |
640 | ||
3327ac33 | 641 | if (ust_lock()) { |
74d81a6c | 642 | ret = -LTTNG_UST_ERR_EXITING; |
0dafcd63 | 643 | goto error; |
1ea11eab | 644 | } |
9eb62b9c | 645 | |
46050b1a MD |
646 | ops = objd_ops(lum->handle); |
647 | if (!ops) { | |
648 | ret = -ENOENT; | |
0dafcd63 | 649 | goto error; |
1ea11eab | 650 | } |
46050b1a MD |
651 | |
652 | switch (lum->cmd) { | |
11ff9c7d MD |
653 | case LTTNG_UST_REGISTER_DONE: |
654 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
edaa1431 | 655 | ret = handle_register_done(sock_info); |
11ff9c7d MD |
656 | else |
657 | ret = -EINVAL; | |
658 | break; | |
46050b1a MD |
659 | case LTTNG_UST_RELEASE: |
660 | if (lum->handle == LTTNG_UST_ROOT_HANDLE) | |
661 | ret = -EPERM; | |
662 | else | |
1849ef7c | 663 | ret = lttng_ust_objd_unref(lum->handle, 1); |
d9e99d10 | 664 | break; |
2d78951a MD |
665 | case LTTNG_UST_FILTER: |
666 | { | |
667 | /* Receive filter data */ | |
f488575f | 668 | struct lttng_ust_filter_bytecode_node *bytecode; |
2d78951a | 669 | |
cd54f6d9 | 670 | if (lum->u.filter.data_size > FILTER_BYTECODE_MAX_LEN) { |
7bc53e94 | 671 | ERR("Filter data size is too large: %u bytes", |
2d78951a MD |
672 | lum->u.filter.data_size); |
673 | ret = -EINVAL; | |
674 | goto error; | |
675 | } | |
2734ca65 | 676 | |
885b1dfd | 677 | if (lum->u.filter.reloc_offset > lum->u.filter.data_size) { |
7bc53e94 | 678 | ERR("Filter reloc offset %u is not within data", |
2734ca65 CB |
679 | lum->u.filter.reloc_offset); |
680 | ret = -EINVAL; | |
681 | goto error; | |
682 | } | |
683 | ||
cd54f6d9 MD |
684 | bytecode = zmalloc(sizeof(*bytecode) + lum->u.filter.data_size); |
685 | if (!bytecode) { | |
686 | ret = -ENOMEM; | |
687 | goto error; | |
688 | } | |
f488575f | 689 | len = ustcomm_recv_unix_sock(sock, bytecode->bc.data, |
2d78951a MD |
690 | lum->u.filter.data_size); |
691 | switch (len) { | |
692 | case 0: /* orderly shutdown */ | |
693 | ret = 0; | |
cd54f6d9 | 694 | free(bytecode); |
2d78951a | 695 | goto error; |
2d78951a MD |
696 | default: |
697 | if (len == lum->u.filter.data_size) { | |
7bc53e94 | 698 | DBG("filter data received"); |
2d78951a | 699 | break; |
7bc53e94 MD |
700 | } else if (len < 0) { |
701 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
702 | if (len == -ECONNRESET) { | |
703 | ERR("%s remote end closed connection", sock_info->name); | |
704 | ret = len; | |
705 | free(bytecode); | |
706 | goto error; | |
707 | } | |
708 | ret = len; | |
eb8bf361 | 709 | free(bytecode); |
0dafcd63 | 710 | goto error; |
2d78951a | 711 | } else { |
7bc53e94 | 712 | DBG("incorrect filter data message size: %zd", len); |
2d78951a | 713 | ret = -EINVAL; |
cd54f6d9 | 714 | free(bytecode); |
0dafcd63 | 715 | goto error; |
2d78951a MD |
716 | } |
717 | } | |
f488575f MD |
718 | bytecode->bc.len = lum->u.filter.data_size; |
719 | bytecode->bc.reloc_offset = lum->u.filter.reloc_offset; | |
3f6fd224 | 720 | bytecode->bc.seqnum = lum->u.filter.seqnum; |
cd54f6d9 | 721 | if (ops->cmd) { |
2d78951a | 722 | ret = ops->cmd(lum->handle, lum->cmd, |
cd54f6d9 | 723 | (unsigned long) bytecode, |
f59ed768 | 724 | &args, sock_info); |
cd54f6d9 MD |
725 | if (ret) { |
726 | free(bytecode); | |
727 | } | |
728 | /* don't free bytecode if everything went fine. */ | |
729 | } else { | |
2d78951a | 730 | ret = -ENOSYS; |
cd54f6d9 MD |
731 | free(bytecode); |
732 | } | |
2d78951a MD |
733 | break; |
734 | } | |
86e36163 JI |
735 | case LTTNG_UST_EXCLUSION: |
736 | { | |
737 | /* Receive exclusion names */ | |
738 | struct lttng_ust_excluder_node *node; | |
739 | unsigned int count; | |
740 | ||
741 | count = lum->u.exclusion.count; | |
742 | if (count == 0) { | |
743 | /* There are no names to read */ | |
744 | ret = 0; | |
745 | goto error; | |
746 | } | |
747 | node = zmalloc(sizeof(*node) + | |
748 | count * LTTNG_UST_SYM_NAME_LEN); | |
749 | if (!node) { | |
750 | ret = -ENOMEM; | |
751 | goto error; | |
752 | } | |
753 | node->excluder.count = count; | |
754 | len = ustcomm_recv_unix_sock(sock, node->excluder.names, | |
755 | count * LTTNG_UST_SYM_NAME_LEN); | |
756 | switch (len) { | |
757 | case 0: /* orderly shutdown */ | |
758 | ret = 0; | |
759 | free(node); | |
760 | goto error; | |
761 | default: | |
762 | if (len == count * LTTNG_UST_SYM_NAME_LEN) { | |
763 | DBG("Exclusion data received"); | |
764 | break; | |
765 | } else if (len < 0) { | |
766 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
767 | if (len == -ECONNRESET) { | |
768 | ERR("%s remote end closed connection", sock_info->name); | |
769 | ret = len; | |
770 | free(node); | |
771 | goto error; | |
772 | } | |
773 | ret = len; | |
774 | free(node); | |
0dafcd63 | 775 | goto error; |
86e36163 JI |
776 | } else { |
777 | DBG("Incorrect exclusion data message size: %zd", len); | |
778 | ret = -EINVAL; | |
779 | free(node); | |
0dafcd63 | 780 | goto error; |
86e36163 JI |
781 | } |
782 | } | |
783 | if (ops->cmd) { | |
784 | ret = ops->cmd(lum->handle, lum->cmd, | |
785 | (unsigned long) node, | |
786 | &args, sock_info); | |
787 | if (ret) { | |
788 | free(node); | |
789 | } | |
790 | /* Don't free exclusion data if everything went fine. */ | |
791 | } else { | |
792 | ret = -ENOSYS; | |
793 | free(node); | |
794 | } | |
795 | break; | |
796 | } | |
74d81a6c MD |
797 | case LTTNG_UST_CHANNEL: |
798 | { | |
799 | void *chan_data; | |
ff0f5728 | 800 | int wakeup_fd; |
74d81a6c MD |
801 | |
802 | len = ustcomm_recv_channel_from_sessiond(sock, | |
ff0f5728 MD |
803 | &chan_data, lum->u.channel.len, |
804 | &wakeup_fd); | |
74d81a6c MD |
805 | switch (len) { |
806 | case 0: /* orderly shutdown */ | |
807 | ret = 0; | |
808 | goto error; | |
809 | default: | |
810 | if (len == lum->u.channel.len) { | |
811 | DBG("channel data received"); | |
812 | break; | |
813 | } else if (len < 0) { | |
814 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
815 | if (len == -ECONNRESET) { | |
816 | ERR("%s remote end closed connection", sock_info->name); | |
817 | ret = len; | |
818 | goto error; | |
819 | } | |
820 | ret = len; | |
0dafcd63 | 821 | goto error; |
74d81a6c MD |
822 | } else { |
823 | DBG("incorrect channel data message size: %zd", len); | |
824 | ret = -EINVAL; | |
0dafcd63 | 825 | goto error; |
74d81a6c MD |
826 | } |
827 | } | |
828 | args.channel.chan_data = chan_data; | |
ff0f5728 | 829 | args.channel.wakeup_fd = wakeup_fd; |
74d81a6c MD |
830 | if (ops->cmd) |
831 | ret = ops->cmd(lum->handle, lum->cmd, | |
832 | (unsigned long) &lum->u, | |
833 | &args, sock_info); | |
834 | else | |
835 | ret = -ENOSYS; | |
836 | break; | |
837 | } | |
838 | case LTTNG_UST_STREAM: | |
839 | { | |
840 | /* Receive shm_fd, wakeup_fd */ | |
841 | ret = ustcomm_recv_stream_from_sessiond(sock, | |
842 | &lum->u.stream.len, | |
843 | &args.stream.shm_fd, | |
844 | &args.stream.wakeup_fd); | |
845 | if (ret) { | |
0dafcd63 | 846 | goto error; |
74d81a6c MD |
847 | } |
848 | if (ops->cmd) | |
849 | ret = ops->cmd(lum->handle, lum->cmd, | |
850 | (unsigned long) &lum->u, | |
851 | &args, sock_info); | |
852 | else | |
853 | ret = -ENOSYS; | |
854 | break; | |
855 | } | |
8e696cfa MD |
856 | case LTTNG_UST_CONTEXT: |
857 | switch (lum->u.context.ctx) { | |
858 | case LTTNG_UST_CONTEXT_APP_CONTEXT: | |
859 | { | |
860 | char *p; | |
861 | size_t ctxlen, recvlen; | |
862 | ||
863 | ctxlen = strlen("$app.") + lum->u.context.u.app_ctx.provider_name_len - 1 | |
864 | + strlen(":") + lum->u.context.u.app_ctx.ctx_name_len; | |
865 | if (ctxlen >= LTTNG_UST_SYM_NAME_LEN) { | |
866 | ERR("Application context string length size is too large: %zu bytes", | |
867 | ctxlen); | |
868 | ret = -EINVAL; | |
869 | goto error; | |
870 | } | |
871 | strcpy(ctxstr, "$app."); | |
872 | p = &ctxstr[strlen("$app.")]; | |
873 | recvlen = ctxlen - strlen("$app."); | |
874 | len = ustcomm_recv_unix_sock(sock, p, recvlen); | |
875 | switch (len) { | |
876 | case 0: /* orderly shutdown */ | |
877 | ret = 0; | |
878 | goto error; | |
879 | default: | |
880 | if (len == recvlen) { | |
881 | DBG("app context data received"); | |
882 | break; | |
883 | } else if (len < 0) { | |
884 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
885 | if (len == -ECONNRESET) { | |
886 | ERR("%s remote end closed connection", sock_info->name); | |
887 | ret = len; | |
888 | goto error; | |
889 | } | |
890 | ret = len; | |
891 | goto error; | |
892 | } else { | |
893 | DBG("incorrect app context data message size: %zd", len); | |
894 | ret = -EINVAL; | |
895 | goto error; | |
896 | } | |
897 | } | |
898 | /* Put : between provider and ctxname. */ | |
899 | p[lum->u.context.u.app_ctx.provider_name_len - 1] = ':'; | |
900 | args.app_context.ctxname = ctxstr; | |
901 | break; | |
902 | } | |
903 | default: | |
904 | break; | |
905 | } | |
906 | if (ops->cmd) { | |
907 | ret = ops->cmd(lum->handle, lum->cmd, | |
908 | (unsigned long) &lum->u, | |
909 | &args, sock_info); | |
910 | } else { | |
911 | ret = -ENOSYS; | |
912 | } | |
913 | break; | |
d9e99d10 | 914 | default: |
46050b1a MD |
915 | if (ops->cmd) |
916 | ret = ops->cmd(lum->handle, lum->cmd, | |
ef9ff354 | 917 | (unsigned long) &lum->u, |
f59ed768 | 918 | &args, sock_info); |
46050b1a MD |
919 | else |
920 | ret = -ENOSYS; | |
921 | break; | |
d9e99d10 | 922 | } |
46050b1a | 923 | |
46050b1a MD |
924 | lur.handle = lum->handle; |
925 | lur.cmd = lum->cmd; | |
926 | lur.ret_val = ret; | |
927 | if (ret >= 0) { | |
7bc53e94 | 928 | lur.ret_code = LTTNG_UST_OK; |
46050b1a | 929 | } else { |
7bc53e94 MD |
930 | /* |
931 | * Use -LTTNG_UST_ERR as wildcard for UST internal | |
932 | * error that are not caused by the transport, except if | |
933 | * we already have a more precise error message to | |
934 | * report. | |
935 | */ | |
64b2564e DG |
936 | if (ret > -LTTNG_UST_ERR) { |
937 | /* Translate code to UST error. */ | |
938 | switch (ret) { | |
939 | case -EEXIST: | |
940 | lur.ret_code = -LTTNG_UST_ERR_EXIST; | |
941 | break; | |
942 | case -EINVAL: | |
943 | lur.ret_code = -LTTNG_UST_ERR_INVAL; | |
944 | break; | |
945 | case -ENOENT: | |
946 | lur.ret_code = -LTTNG_UST_ERR_NOENT; | |
947 | break; | |
948 | case -EPERM: | |
949 | lur.ret_code = -LTTNG_UST_ERR_PERM; | |
950 | break; | |
951 | case -ENOSYS: | |
952 | lur.ret_code = -LTTNG_UST_ERR_NOSYS; | |
953 | break; | |
954 | default: | |
955 | lur.ret_code = -LTTNG_UST_ERR; | |
956 | break; | |
957 | } | |
958 | } else { | |
7bc53e94 | 959 | lur.ret_code = ret; |
64b2564e | 960 | } |
46050b1a | 961 | } |
e6ea14c5 MD |
962 | if (ret >= 0) { |
963 | switch (lum->cmd) { | |
e6ea14c5 MD |
964 | case LTTNG_UST_TRACER_VERSION: |
965 | lur.u.version = lum->u.version; | |
966 | break; | |
967 | case LTTNG_UST_TRACEPOINT_LIST_GET: | |
968 | memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint)); | |
969 | break; | |
970 | } | |
381c0f1e | 971 | } |
74d81a6c | 972 | DBG("Return value: %d", lur.ret_val); |
4c62d8d1 MD |
973 | |
974 | ust_unlock(); | |
975 | ||
976 | /* | |
977 | * Performed delayed statedump operations outside of the UST | |
978 | * lock. We need to take the dynamic loader lock before we take | |
979 | * the UST lock internally within handle_pending_statedump(). | |
980 | */ | |
981 | handle_pending_statedump(sock_info); | |
982 | ||
983 | if (ust_lock()) { | |
984 | ret = -LTTNG_UST_ERR_EXITING; | |
985 | goto error; | |
986 | } | |
987 | ||
46050b1a | 988 | ret = send_reply(sock, &lur); |
193183fb | 989 | if (ret < 0) { |
7bc53e94 | 990 | DBG("error sending reply"); |
193183fb MD |
991 | goto error; |
992 | } | |
46050b1a | 993 | |
40003310 MD |
994 | /* |
995 | * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field | |
996 | * after the reply. | |
997 | */ | |
7bc53e94 | 998 | if (lur.ret_code == LTTNG_UST_OK) { |
40003310 MD |
999 | switch (lum->cmd) { |
1000 | case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET: | |
1001 | len = ustcomm_send_unix_sock(sock, | |
1002 | &args.field_list.entry, | |
1003 | sizeof(args.field_list.entry)); | |
7bc53e94 MD |
1004 | if (len < 0) { |
1005 | ret = len; | |
1006 | goto error; | |
1007 | } | |
40003310 | 1008 | if (len != sizeof(args.field_list.entry)) { |
7bc53e94 | 1009 | ret = -EINVAL; |
40003310 MD |
1010 | goto error; |
1011 | } | |
1012 | } | |
1013 | } | |
ef9ff354 | 1014 | |
381c0f1e | 1015 | error: |
17dfb34b | 1016 | ust_unlock(); |
d9e99d10 | 1017 | |
37dddb65 | 1018 | return ret; |
246be17e PW |
1019 | } |
1020 | ||
46050b1a | 1021 | static |
efe0de09 | 1022 | void cleanup_sock_info(struct sock_info *sock_info, int exiting) |
46050b1a MD |
1023 | { |
1024 | int ret; | |
1025 | ||
5b14aab3 MD |
1026 | if (sock_info->root_handle != -1) { |
1027 | ret = lttng_ust_objd_unref(sock_info->root_handle, 1); | |
1028 | if (ret) { | |
1029 | ERR("Error unref root handle"); | |
1030 | } | |
1031 | sock_info->root_handle = -1; | |
1032 | } | |
1033 | sock_info->constructor_sem_posted = 0; | |
1034 | ||
1035 | /* | |
1036 | * wait_shm_mmap, socket and notify socket are used by listener | |
1037 | * threads outside of the ust lock, so we cannot tear them down | |
1038 | * ourselves, because we cannot join on these threads. Leave | |
1039 | * responsibility of cleaning up these resources to the OS | |
1040 | * process exit. | |
1041 | */ | |
1042 | if (exiting) | |
1043 | return; | |
1044 | ||
46050b1a | 1045 | if (sock_info->socket != -1) { |
e6973a89 | 1046 | ret = ustcomm_close_unix_sock(sock_info->socket); |
46050b1a | 1047 | if (ret) { |
32ce8569 | 1048 | ERR("Error closing ust cmd socket"); |
46050b1a MD |
1049 | } |
1050 | sock_info->socket = -1; | |
1051 | } | |
32ce8569 MD |
1052 | if (sock_info->notify_socket != -1) { |
1053 | ret = ustcomm_close_unix_sock(sock_info->notify_socket); | |
1054 | if (ret) { | |
1055 | ERR("Error closing ust notify socket"); | |
1056 | } | |
1057 | sock_info->notify_socket = -1; | |
1058 | } | |
5b14aab3 | 1059 | if (sock_info->wait_shm_mmap) { |
172d6b68 MD |
1060 | long page_size; |
1061 | ||
1062 | page_size = sysconf(_SC_PAGE_SIZE); | |
2657d1ba MD |
1063 | if (page_size <= 0) { |
1064 | if (!page_size) { | |
1065 | errno = EINVAL; | |
1066 | } | |
1067 | PERROR("Error in sysconf(_SC_PAGE_SIZE)"); | |
1068 | } else { | |
172d6b68 MD |
1069 | ret = munmap(sock_info->wait_shm_mmap, page_size); |
1070 | if (ret) { | |
1071 | ERR("Error unmapping wait shm"); | |
1072 | } | |
7fc90dca MD |
1073 | } |
1074 | sock_info->wait_shm_mmap = NULL; | |
1075 | } | |
1076 | } | |
1077 | ||
58d4b2a2 | 1078 | /* |
33bbeb90 MD |
1079 | * Using fork to set umask in the child process (not multi-thread safe). |
1080 | * We deal with the shm_open vs ftruncate race (happening when the | |
1081 | * sessiond owns the shm and does not let everybody modify it, to ensure | |
1082 | * safety against shm_unlink) by simply letting the mmap fail and | |
1083 | * retrying after a few seconds. | |
1084 | * For global shm, everybody has rw access to it until the sessiond | |
1085 | * starts. | |
58d4b2a2 | 1086 | */ |
7fc90dca | 1087 | static |
58d4b2a2 | 1088 | int get_wait_shm(struct sock_info *sock_info, size_t mmap_size) |
7fc90dca | 1089 | { |
7fc90dca | 1090 | int wait_shm_fd, ret; |
58d4b2a2 | 1091 | pid_t pid; |
44e073f5 | 1092 | |
58d4b2a2 | 1093 | /* |
33bbeb90 | 1094 | * Try to open read-only. |
58d4b2a2 | 1095 | */ |
33bbeb90 | 1096 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 | 1097 | if (wait_shm_fd >= 0) { |
7aa76730 MD |
1098 | int32_t tmp_read; |
1099 | ssize_t len; | |
1100 | size_t bytes_read = 0; | |
1101 | ||
1102 | /* | |
1103 | * Try to read the fd. If unable to do so, try opening | |
1104 | * it in write mode. | |
1105 | */ | |
1106 | do { | |
1107 | len = read(wait_shm_fd, | |
1108 | &((char *) &tmp_read)[bytes_read], | |
1109 | sizeof(tmp_read) - bytes_read); | |
1110 | if (len > 0) { | |
1111 | bytes_read += len; | |
1112 | } | |
1113 | } while ((len < 0 && errno == EINTR) | |
1114 | || (len > 0 && bytes_read < sizeof(tmp_read))); | |
1115 | if (bytes_read != sizeof(tmp_read)) { | |
1116 | ret = close(wait_shm_fd); | |
1117 | if (ret) { | |
1118 | ERR("close wait_shm_fd"); | |
1119 | } | |
1120 | goto open_write; | |
1121 | } | |
58d4b2a2 MD |
1122 | goto end; |
1123 | } else if (wait_shm_fd < 0 && errno != ENOENT) { | |
1124 | /* | |
33bbeb90 MD |
1125 | * Real-only open did not work, and it's not because the |
1126 | * entry was not present. It's a failure that prohibits | |
1127 | * using shm. | |
58d4b2a2 | 1128 | */ |
7fc90dca | 1129 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
58d4b2a2 | 1130 | goto end; |
7fc90dca | 1131 | } |
7aa76730 MD |
1132 | |
1133 | open_write: | |
7fc90dca | 1134 | /* |
7aa76730 MD |
1135 | * If the open failed because the file did not exist, or because |
1136 | * the file was not truncated yet, try creating it ourself. | |
7fc90dca | 1137 | */ |
8c90a710 | 1138 | URCU_TLS(lttng_ust_nest_count)++; |
58d4b2a2 | 1139 | pid = fork(); |
8c90a710 | 1140 | URCU_TLS(lttng_ust_nest_count)--; |
58d4b2a2 MD |
1141 | if (pid > 0) { |
1142 | int status; | |
1143 | ||
1144 | /* | |
1145 | * Parent: wait for child to return, in which case the | |
1146 | * shared memory map will have been created. | |
1147 | */ | |
1148 | pid = wait(&status); | |
b7d3cb32 | 1149 | if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) { |
58d4b2a2 MD |
1150 | wait_shm_fd = -1; |
1151 | goto end; | |
7fc90dca | 1152 | } |
58d4b2a2 MD |
1153 | /* |
1154 | * Try to open read-only again after creation. | |
1155 | */ | |
33bbeb90 | 1156 | wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0); |
58d4b2a2 MD |
1157 | if (wait_shm_fd < 0) { |
1158 | /* | |
1159 | * Real-only open did not work. It's a failure | |
1160 | * that prohibits using shm. | |
1161 | */ | |
1162 | ERR("Error opening shm %s", sock_info->wait_shm_path); | |
1163 | goto end; | |
1164 | } | |
1165 | goto end; | |
1166 | } else if (pid == 0) { | |
1167 | int create_mode; | |
1168 | ||
1169 | /* Child */ | |
33bbeb90 | 1170 | create_mode = S_IRUSR | S_IWUSR | S_IRGRP; |
58d4b2a2 | 1171 | if (sock_info->global) |
33bbeb90 | 1172 | create_mode |= S_IROTH | S_IWGRP | S_IWOTH; |
58d4b2a2 MD |
1173 | /* |
1174 | * We're alone in a child process, so we can modify the | |
1175 | * process-wide umask. | |
1176 | */ | |
33bbeb90 | 1177 | umask(~create_mode); |
58d4b2a2 | 1178 | /* |
33bbeb90 MD |
1179 | * Try creating shm (or get rw access). |
1180 | * We don't do an exclusive open, because we allow other | |
1181 | * processes to create+ftruncate it concurrently. | |
58d4b2a2 MD |
1182 | */ |
1183 | wait_shm_fd = shm_open(sock_info->wait_shm_path, | |
1184 | O_RDWR | O_CREAT, create_mode); | |
1185 | if (wait_shm_fd >= 0) { | |
1186 | ret = ftruncate(wait_shm_fd, mmap_size); | |
1187 | if (ret) { | |
1188 | PERROR("ftruncate"); | |
b0c1425d | 1189 | _exit(EXIT_FAILURE); |
58d4b2a2 | 1190 | } |
b0c1425d | 1191 | _exit(EXIT_SUCCESS); |
58d4b2a2 | 1192 | } |
33bbeb90 MD |
1193 | /* |
1194 | * For local shm, we need to have rw access to accept | |
1195 | * opening it: this means the local sessiond will be | |
1196 | * able to wake us up. For global shm, we open it even | |
1197 | * if rw access is not granted, because the root.root | |
1198 | * sessiond will be able to override all rights and wake | |
1199 | * us up. | |
1200 | */ | |
1201 | if (!sock_info->global && errno != EACCES) { | |
58d4b2a2 | 1202 | ERR("Error opening shm %s", sock_info->wait_shm_path); |
5d3bc5ed | 1203 | _exit(EXIT_FAILURE); |
58d4b2a2 MD |
1204 | } |
1205 | /* | |
33bbeb90 MD |
1206 | * The shm exists, but we cannot open it RW. Report |
1207 | * success. | |
58d4b2a2 | 1208 | */ |
5d3bc5ed | 1209 | _exit(EXIT_SUCCESS); |
58d4b2a2 MD |
1210 | } else { |
1211 | return -1; | |
7fc90dca | 1212 | } |
58d4b2a2 | 1213 | end: |
33bbeb90 MD |
1214 | if (wait_shm_fd >= 0 && !sock_info->global) { |
1215 | struct stat statbuf; | |
1216 | ||
1217 | /* | |
1218 | * Ensure that our user is the owner of the shm file for | |
1219 | * local shm. If we do not own the file, it means our | |
1220 | * sessiond will not have access to wake us up (there is | |
1221 | * probably a rogue process trying to fake our | |
1222 | * sessiond). Fallback to polling method in this case. | |
1223 | */ | |
1224 | ret = fstat(wait_shm_fd, &statbuf); | |
1225 | if (ret) { | |
1226 | PERROR("fstat"); | |
1227 | goto error_close; | |
1228 | } | |
1229 | if (statbuf.st_uid != getuid()) | |
1230 | goto error_close; | |
1231 | } | |
58d4b2a2 | 1232 | return wait_shm_fd; |
33bbeb90 MD |
1233 | |
1234 | error_close: | |
1235 | ret = close(wait_shm_fd); | |
1236 | if (ret) { | |
1237 | PERROR("Error closing fd"); | |
1238 | } | |
1239 | return -1; | |
58d4b2a2 MD |
1240 | } |
1241 | ||
1242 | static | |
1243 | char *get_map_shm(struct sock_info *sock_info) | |
1244 | { | |
172d6b68 | 1245 | long page_size; |
58d4b2a2 MD |
1246 | int wait_shm_fd, ret; |
1247 | char *wait_shm_mmap; | |
1248 | ||
172d6b68 | 1249 | page_size = sysconf(_SC_PAGE_SIZE); |
2657d1ba MD |
1250 | if (page_size <= 0) { |
1251 | if (!page_size) { | |
1252 | errno = EINVAL; | |
1253 | } | |
1254 | PERROR("Error in sysconf(_SC_PAGE_SIZE)"); | |
172d6b68 MD |
1255 | goto error; |
1256 | } | |
1257 | ||
6548fca4 | 1258 | lttng_ust_lock_fd_tracker(); |
172d6b68 | 1259 | wait_shm_fd = get_wait_shm(sock_info, page_size); |
58d4b2a2 | 1260 | if (wait_shm_fd < 0) { |
6548fca4 | 1261 | lttng_ust_unlock_fd_tracker(); |
58d4b2a2 | 1262 | goto error; |
44e073f5 | 1263 | } |
6548fca4 MD |
1264 | lttng_ust_add_fd_to_tracker(wait_shm_fd); |
1265 | lttng_ust_unlock_fd_tracker(); | |
1266 | ||
172d6b68 | 1267 | wait_shm_mmap = mmap(NULL, page_size, PROT_READ, |
7fc90dca | 1268 | MAP_SHARED, wait_shm_fd, 0); |
6548fca4 | 1269 | |
7fc90dca | 1270 | /* close shm fd immediately after taking the mmap reference */ |
6548fca4 | 1271 | lttng_ust_lock_fd_tracker(); |
7fc90dca | 1272 | ret = close(wait_shm_fd); |
6548fca4 MD |
1273 | if (!ret) { |
1274 | lttng_ust_delete_fd_from_tracker(wait_shm_fd); | |
1275 | } else { | |
33bbeb90 MD |
1276 | PERROR("Error closing fd"); |
1277 | } | |
6548fca4 MD |
1278 | lttng_ust_unlock_fd_tracker(); |
1279 | ||
33bbeb90 MD |
1280 | if (wait_shm_mmap == MAP_FAILED) { |
1281 | DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode."); | |
1282 | goto error; | |
7fc90dca MD |
1283 | } |
1284 | return wait_shm_mmap; | |
1285 | ||
1286 | error: | |
1287 | return NULL; | |
1288 | } | |
1289 | ||
1290 | static | |
1291 | void wait_for_sessiond(struct sock_info *sock_info) | |
1292 | { | |
3327ac33 | 1293 | if (ust_lock()) { |
7fc90dca MD |
1294 | goto quit; |
1295 | } | |
37ed587a MD |
1296 | if (wait_poll_fallback) { |
1297 | goto error; | |
1298 | } | |
7fc90dca MD |
1299 | if (!sock_info->wait_shm_mmap) { |
1300 | sock_info->wait_shm_mmap = get_map_shm(sock_info); | |
1301 | if (!sock_info->wait_shm_mmap) | |
1302 | goto error; | |
1303 | } | |
1304 | ust_unlock(); | |
1305 | ||
1306 | DBG("Waiting for %s apps sessiond", sock_info->name); | |
80e2814b | 1307 | /* Wait for futex wakeup */ |
ee7fcec8 MD |
1308 | if (uatomic_read((int32_t *) sock_info->wait_shm_mmap)) |
1309 | goto end_wait; | |
1310 | ||
1311 | while (futex_async((int32_t *) sock_info->wait_shm_mmap, | |
1312 | FUTEX_WAIT, 0, NULL, NULL, 0)) { | |
1313 | switch (errno) { | |
1314 | case EWOULDBLOCK: | |
1315 | /* Value already changed. */ | |
1316 | goto end_wait; | |
1317 | case EINTR: | |
1318 | /* Retry if interrupted by signal. */ | |
1319 | break; /* Get out of switch. */ | |
1320 | case EFAULT: | |
1321 | wait_poll_fallback = 1; | |
1322 | DBG( | |
37ed587a MD |
1323 | "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) " |
1324 | "do not support FUTEX_WAKE on read-only memory mappings correctly. " | |
1325 | "Please upgrade your kernel " | |
1326 | "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel " | |
1327 | "mainline). LTTng-UST will use polling mode fallback."); | |
ee7fcec8 MD |
1328 | if (ust_debug()) |
1329 | PERROR("futex"); | |
1330 | goto end_wait; | |
80e2814b MD |
1331 | } |
1332 | } | |
ee7fcec8 | 1333 | end_wait: |
7fc90dca MD |
1334 | return; |
1335 | ||
1336 | quit: | |
1337 | ust_unlock(); | |
1338 | return; | |
1339 | ||
1340 | error: | |
1341 | ust_unlock(); | |
7fc90dca | 1342 | return; |
46050b1a MD |
1343 | } |
1344 | ||
1ea11eab MD |
1345 | /* |
1346 | * This thread does not allocate any resource, except within | |
1347 | * handle_message, within mutex protection. This mutex protects against | |
1348 | * fork and exit. | |
98bf993f | 1349 | * The other moment it allocates resources is at socket connection, which |
1ea11eab MD |
1350 | * is also protected by the mutex. |
1351 | */ | |
d9e99d10 MD |
1352 | static |
1353 | void *ust_listener_thread(void *arg) | |
1354 | { | |
1ea11eab | 1355 | struct sock_info *sock_info = arg; |
c0eedf81 | 1356 | int sock, ret, prev_connect_failed = 0, has_waited = 0; |
ff517991 | 1357 | long timeout; |
d9e99d10 | 1358 | |
c362addf | 1359 | lttng_ust_fixup_tls(); |
01f0e40c RB |
1360 | /* |
1361 | * If available, add '-ust' to the end of this thread's | |
1362 | * process name | |
1363 | */ | |
1364 | ret = lttng_ust_setustprocname(); | |
1365 | if (ret) { | |
1366 | ERR("Unable to set UST process name"); | |
1367 | } | |
1368 | ||
9eb62b9c MD |
1369 | /* Restart trying to connect to the session daemon */ |
1370 | restart: | |
c0eedf81 MD |
1371 | if (prev_connect_failed) { |
1372 | /* Wait for sessiond availability with pipe */ | |
1373 | wait_for_sessiond(sock_info); | |
1374 | if (has_waited) { | |
1375 | has_waited = 0; | |
1376 | /* | |
1377 | * Sleep for 5 seconds before retrying after a | |
1378 | * sequence of failure / wait / failure. This | |
1379 | * deals with a killed or broken session daemon. | |
1380 | */ | |
1381 | sleep(5); | |
eacc4aa4 MD |
1382 | } else { |
1383 | has_waited = 1; | |
c0eedf81 | 1384 | } |
c0eedf81 MD |
1385 | prev_connect_failed = 0; |
1386 | } | |
9eb62b9c | 1387 | |
1ea11eab | 1388 | if (sock_info->socket != -1) { |
6548fca4 | 1389 | /* FD tracker is updated by ustcomm_close_unix_sock() */ |
e6973a89 | 1390 | ret = ustcomm_close_unix_sock(sock_info->socket); |
1ea11eab | 1391 | if (ret) { |
32ce8569 MD |
1392 | ERR("Error closing %s ust cmd socket", |
1393 | sock_info->name); | |
1ea11eab MD |
1394 | } |
1395 | sock_info->socket = -1; | |
1396 | } | |
32ce8569 | 1397 | if (sock_info->notify_socket != -1) { |
6548fca4 | 1398 | /* FD tracker is updated by ustcomm_close_unix_sock() */ |
32ce8569 MD |
1399 | ret = ustcomm_close_unix_sock(sock_info->notify_socket); |
1400 | if (ret) { | |
1401 | ERR("Error closing %s ust notify socket", | |
1402 | sock_info->name); | |
1403 | } | |
1404 | sock_info->notify_socket = -1; | |
1405 | } | |
46050b1a | 1406 | |
6548fca4 MD |
1407 | if (ust_lock()) { |
1408 | goto quit; | |
1409 | } | |
1410 | ||
321f2351 MD |
1411 | /* |
1412 | * Register. We need to perform both connect and sending | |
1413 | * registration message before doing the next connect otherwise | |
1414 | * we may reach unix socket connect queue max limits and block | |
1415 | * on the 2nd connect while the session daemon is awaiting the | |
1416 | * first connect registration message. | |
1417 | */ | |
1418 | /* Connect cmd socket */ | |
6548fca4 | 1419 | lttng_ust_lock_fd_tracker(); |
451d66b2 MD |
1420 | ret = ustcomm_connect_unix_sock(sock_info->sock_path, |
1421 | get_connect_sock_timeout()); | |
321f2351 | 1422 | if (ret < 0) { |
6548fca4 | 1423 | lttng_ust_unlock_fd_tracker(); |
321f2351 MD |
1424 | DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name); |
1425 | prev_connect_failed = 1; | |
5b14aab3 | 1426 | |
e3426ddc | 1427 | /* |
321f2351 MD |
1428 | * If we cannot find the sessiond daemon, don't delay |
1429 | * constructor execution. | |
e3426ddc | 1430 | */ |
321f2351 MD |
1431 | ret = handle_register_done(sock_info); |
1432 | assert(!ret); | |
1433 | ust_unlock(); | |
1434 | goto restart; | |
27fe9f21 | 1435 | } |
6548fca4 MD |
1436 | lttng_ust_add_fd_to_tracker(ret); |
1437 | lttng_ust_unlock_fd_tracker(); | |
321f2351 | 1438 | sock_info->socket = ret; |
27fe9f21 | 1439 | |
6548fca4 MD |
1440 | ust_unlock(); |
1441 | /* | |
1442 | * Unlock/relock ust lock because connect is blocking (with | |
1443 | * timeout). Don't delay constructors on the ust lock for too | |
1444 | * long. | |
1445 | */ | |
3327ac33 | 1446 | if (ust_lock()) { |
5b14aab3 MD |
1447 | goto quit; |
1448 | } | |
1449 | ||
46050b1a MD |
1450 | /* |
1451 | * Create only one root handle per listener thread for the whole | |
f59ed768 MD |
1452 | * process lifetime, so we ensure we get ID which is statically |
1453 | * assigned to the root handle. | |
46050b1a MD |
1454 | */ |
1455 | if (sock_info->root_handle == -1) { | |
1456 | ret = lttng_abi_create_root_handle(); | |
a51070bb | 1457 | if (ret < 0) { |
46050b1a | 1458 | ERR("Error creating root handle"); |
46050b1a MD |
1459 | goto quit; |
1460 | } | |
1461 | sock_info->root_handle = ret; | |
9eb62b9c | 1462 | } |
1ea11eab | 1463 | |
32ce8569 | 1464 | ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD); |
9eb62b9c | 1465 | if (ret < 0) { |
32ce8569 MD |
1466 | ERR("Error registering to %s ust cmd socket", |
1467 | sock_info->name); | |
c0eedf81 | 1468 | prev_connect_failed = 1; |
11ff9c7d MD |
1469 | /* |
1470 | * If we cannot register to the sessiond daemon, don't | |
1471 | * delay constructor execution. | |
1472 | */ | |
edaa1431 | 1473 | ret = handle_register_done(sock_info); |
11ff9c7d | 1474 | assert(!ret); |
17dfb34b | 1475 | ust_unlock(); |
9eb62b9c MD |
1476 | goto restart; |
1477 | } | |
321f2351 MD |
1478 | |
1479 | ust_unlock(); | |
6548fca4 MD |
1480 | /* |
1481 | * Unlock/relock ust lock because connect is blocking (with | |
1482 | * timeout). Don't delay constructors on the ust lock for too | |
1483 | * long. | |
1484 | */ | |
1485 | if (ust_lock()) { | |
1486 | goto quit; | |
1487 | } | |
321f2351 MD |
1488 | |
1489 | /* Connect notify socket */ | |
6548fca4 | 1490 | lttng_ust_lock_fd_tracker(); |
451d66b2 MD |
1491 | ret = ustcomm_connect_unix_sock(sock_info->sock_path, |
1492 | get_connect_sock_timeout()); | |
321f2351 | 1493 | if (ret < 0) { |
6548fca4 | 1494 | lttng_ust_unlock_fd_tracker(); |
321f2351 MD |
1495 | DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name); |
1496 | prev_connect_failed = 1; | |
1497 | ||
321f2351 MD |
1498 | /* |
1499 | * If we cannot find the sessiond daemon, don't delay | |
1500 | * constructor execution. | |
1501 | */ | |
1502 | ret = handle_register_done(sock_info); | |
1503 | assert(!ret); | |
1504 | ust_unlock(); | |
1505 | goto restart; | |
1506 | } | |
6548fca4 MD |
1507 | lttng_ust_add_fd_to_tracker(ret); |
1508 | lttng_ust_unlock_fd_tracker(); | |
321f2351 MD |
1509 | sock_info->notify_socket = ret; |
1510 | ||
6548fca4 MD |
1511 | ust_unlock(); |
1512 | /* | |
1513 | * Unlock/relock ust lock because connect is blocking (with | |
1514 | * timeout). Don't delay constructors on the ust lock for too | |
1515 | * long. | |
1516 | */ | |
1517 | if (ust_lock()) { | |
1518 | goto quit; | |
1519 | } | |
1520 | ||
321f2351 MD |
1521 | timeout = get_notify_sock_timeout(); |
1522 | if (timeout >= 0) { | |
1523 | /* | |
1524 | * Give at least 10ms to sessiond to reply to | |
1525 | * notifications. | |
1526 | */ | |
1527 | if (timeout < 10) | |
1528 | timeout = 10; | |
1529 | ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket, | |
1530 | timeout); | |
1531 | if (ret < 0) { | |
1532 | WARN("Error setting socket receive timeout"); | |
1533 | } | |
1534 | ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket, | |
1535 | timeout); | |
1536 | if (ret < 0) { | |
1537 | WARN("Error setting socket send timeout"); | |
1538 | } | |
1539 | } else if (timeout < -1) { | |
1540 | WARN("Unsupported timeout value %ld", timeout); | |
1541 | } | |
1542 | ||
32ce8569 MD |
1543 | ret = register_to_sessiond(sock_info->notify_socket, |
1544 | USTCTL_SOCKET_NOTIFY); | |
1545 | if (ret < 0) { | |
1546 | ERR("Error registering to %s ust notify socket", | |
1547 | sock_info->name); | |
1548 | prev_connect_failed = 1; | |
1549 | /* | |
1550 | * If we cannot register to the sessiond daemon, don't | |
1551 | * delay constructor execution. | |
1552 | */ | |
1553 | ret = handle_register_done(sock_info); | |
1554 | assert(!ret); | |
1555 | ust_unlock(); | |
1556 | goto restart; | |
1557 | } | |
1558 | sock = sock_info->socket; | |
1559 | ||
17dfb34b | 1560 | ust_unlock(); |
46050b1a | 1561 | |
d9e99d10 MD |
1562 | for (;;) { |
1563 | ssize_t len; | |
57773204 | 1564 | struct ustcomm_ust_msg lum; |
d9e99d10 | 1565 | |
57773204 | 1566 | len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum)); |
d9e99d10 MD |
1567 | switch (len) { |
1568 | case 0: /* orderly shutdown */ | |
7dd08bec | 1569 | DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name); |
3327ac33 | 1570 | if (ust_lock()) { |
d5e1fea6 MD |
1571 | goto quit; |
1572 | } | |
8236ba10 MD |
1573 | /* |
1574 | * Either sessiond has shutdown or refused us by closing the socket. | |
1575 | * In either case, we don't want to delay construction execution, | |
1576 | * and we need to wait before retry. | |
1577 | */ | |
1578 | prev_connect_failed = 1; | |
1579 | /* | |
1580 | * If we cannot register to the sessiond daemon, don't | |
1581 | * delay constructor execution. | |
1582 | */ | |
1583 | ret = handle_register_done(sock_info); | |
1584 | assert(!ret); | |
1585 | ust_unlock(); | |
d9e99d10 | 1586 | goto end; |
e7723462 | 1587 | case sizeof(lum): |
74d81a6c | 1588 | print_cmd(lum.cmd, lum.handle); |
11ff9c7d | 1589 | ret = handle_message(sock_info, sock, &lum); |
7bc53e94 | 1590 | if (ret) { |
0dafcd63 MD |
1591 | ERR("Error handling message for %s socket", |
1592 | sock_info->name); | |
1593 | /* | |
1594 | * Close socket if protocol error is | |
1595 | * detected. | |
1596 | */ | |
1597 | goto end; | |
d9e99d10 MD |
1598 | } |
1599 | continue; | |
7bc53e94 MD |
1600 | default: |
1601 | if (len < 0) { | |
1602 | DBG("Receive failed from lttng-sessiond with errno %d", (int) -len); | |
1603 | } else { | |
1604 | DBG("incorrect message size (%s socket): %zd", sock_info->name, len); | |
1605 | } | |
1606 | if (len == -ECONNRESET) { | |
1607 | DBG("%s remote end closed connection", sock_info->name); | |
d9e99d10 MD |
1608 | goto end; |
1609 | } | |
1610 | goto end; | |
d9e99d10 MD |
1611 | } |
1612 | ||
1613 | } | |
1614 | end: | |
3327ac33 | 1615 | if (ust_lock()) { |
d5e1fea6 MD |
1616 | goto quit; |
1617 | } | |
f59ed768 MD |
1618 | /* Cleanup socket handles before trying to reconnect */ |
1619 | lttng_ust_objd_table_owner_cleanup(sock_info); | |
1620 | ust_unlock(); | |
9eb62b9c | 1621 | goto restart; /* try to reconnect */ |
e33f3265 | 1622 | |
1ea11eab | 1623 | quit: |
e33f3265 | 1624 | ust_unlock(); |
3327ac33 MD |
1625 | |
1626 | pthread_mutex_lock(&ust_exit_mutex); | |
1627 | sock_info->thread_active = 0; | |
1628 | pthread_mutex_unlock(&ust_exit_mutex); | |
d9e99d10 MD |
1629 | return NULL; |
1630 | } | |
1631 | ||
2594a5b4 MD |
1632 | /* |
1633 | * Weak symbol to call when the ust malloc wrapper is not loaded. | |
1634 | */ | |
1635 | __attribute__((weak)) | |
1636 | void lttng_ust_malloc_wrapper_init(void) | |
1637 | { | |
1638 | } | |
1639 | ||
2691221a MD |
1640 | /* |
1641 | * sessiond monitoring thread: monitor presence of global and per-user | |
1642 | * sessiond by polling the application common named pipe. | |
1643 | */ | |
edaa1431 | 1644 | void __attribute__((constructor)) lttng_ust_init(void) |
2691221a | 1645 | { |
11ff9c7d | 1646 | struct timespec constructor_timeout; |
ae6a58bf | 1647 | sigset_t sig_all_blocked, orig_parent_mask; |
1879f67f | 1648 | pthread_attr_t thread_attr; |
cf12a773 | 1649 | int timeout_mode; |
2691221a MD |
1650 | int ret; |
1651 | ||
edaa1431 MD |
1652 | if (uatomic_xchg(&initialized, 1) == 1) |
1653 | return; | |
1654 | ||
eddd8d5d MD |
1655 | /* |
1656 | * Fixup interdependency between TLS fixup mutex (which happens | |
1657 | * to be the dynamic linker mutex) and ust_lock, taken within | |
1658 | * the ust lock. | |
1659 | */ | |
c362addf | 1660 | lttng_ust_fixup_tls(); |
eddd8d5d | 1661 | |
07b57e5e MD |
1662 | lttng_ust_loaded = 1; |
1663 | ||
edaa1431 MD |
1664 | /* |
1665 | * We want precise control over the order in which we construct | |
1666 | * our sub-libraries vs starting to receive commands from | |
1667 | * sessiond (otherwise leading to errors when trying to create | |
1668 | * sessiond before the init functions are completed). | |
1669 | */ | |
2691221a | 1670 | init_usterr(); |
6f626d28 | 1671 | lttng_ust_getenv_init(); /* Needs init_usterr() to be completed. */ |
edaa1431 | 1672 | init_tracepoint(); |
6548fca4 | 1673 | lttng_ust_init_fd_tracker(); |
f9364363 | 1674 | lttng_ust_clock_init(); |
5e1b7b8b | 1675 | lttng_ust_getcpu_init(); |
cf73e0fe | 1676 | lttng_ust_statedump_init(); |
7dd08bec MD |
1677 | lttng_ring_buffer_metadata_client_init(); |
1678 | lttng_ring_buffer_client_overwrite_init(); | |
34a91bdb | 1679 | lttng_ring_buffer_client_overwrite_rt_init(); |
7dd08bec | 1680 | lttng_ring_buffer_client_discard_init(); |
34a91bdb | 1681 | lttng_ring_buffer_client_discard_rt_init(); |
d58d1454 | 1682 | lttng_perf_counter_init(); |
2594a5b4 MD |
1683 | /* |
1684 | * Invoke ust malloc wrapper init before starting other threads. | |
1685 | */ | |
1686 | lttng_ust_malloc_wrapper_init(); | |
2691221a | 1687 | |
ff517991 | 1688 | timeout_mode = get_constructor_timeout(&constructor_timeout); |
11ff9c7d | 1689 | |
9050321d | 1690 | get_allow_blocking(); |
6f97f9c2 | 1691 | |
95259bd0 | 1692 | ret = sem_init(&constructor_wait, 0, 0); |
8aadb54a MD |
1693 | if (ret) { |
1694 | PERROR("sem_init"); | |
1695 | } | |
11ff9c7d | 1696 | |
8d20bf54 | 1697 | ret = setup_local_apps(); |
2691221a | 1698 | if (ret) { |
9ec6895c | 1699 | DBG("local apps setup returned %d", ret); |
2691221a | 1700 | } |
ae6a58bf WP |
1701 | |
1702 | /* A new thread created by pthread_create inherits the signal mask | |
1703 | * from the parent. To avoid any signal being received by the | |
1704 | * listener thread, we block all signals temporarily in the parent, | |
1705 | * while we create the listener thread. | |
1706 | */ | |
1707 | sigfillset(&sig_all_blocked); | |
1708 | ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask); | |
1709 | if (ret) { | |
d94d802c | 1710 | ERR("pthread_sigmask: %s", strerror(ret)); |
ae6a58bf WP |
1711 | } |
1712 | ||
1879f67f MG |
1713 | ret = pthread_attr_init(&thread_attr); |
1714 | if (ret) { | |
1715 | ERR("pthread_attr_init: %s", strerror(ret)); | |
1716 | } | |
1717 | ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED); | |
1718 | if (ret) { | |
1719 | ERR("pthread_attr_setdetachstate: %s", strerror(ret)); | |
1720 | } | |
1721 | ||
c0bbbd5a | 1722 | pthread_mutex_lock(&ust_exit_mutex); |
1879f67f | 1723 | ret = pthread_create(&global_apps.ust_listener, &thread_attr, |
dde70ea0 | 1724 | ust_listener_thread, &global_apps); |
d94d802c MD |
1725 | if (ret) { |
1726 | ERR("pthread_create global: %s", strerror(ret)); | |
1727 | } | |
e33f3265 | 1728 | global_apps.thread_active = 1; |
c0bbbd5a | 1729 | pthread_mutex_unlock(&ust_exit_mutex); |
e33f3265 | 1730 | |
8d20bf54 | 1731 | if (local_apps.allowed) { |
c0bbbd5a | 1732 | pthread_mutex_lock(&ust_exit_mutex); |
1879f67f | 1733 | ret = pthread_create(&local_apps.ust_listener, &thread_attr, |
dde70ea0 | 1734 | ust_listener_thread, &local_apps); |
d94d802c MD |
1735 | if (ret) { |
1736 | ERR("pthread_create local: %s", strerror(ret)); | |
1737 | } | |
e33f3265 | 1738 | local_apps.thread_active = 1; |
c0bbbd5a | 1739 | pthread_mutex_unlock(&ust_exit_mutex); |
8d20bf54 MD |
1740 | } else { |
1741 | handle_register_done(&local_apps); | |
1742 | } | |
1879f67f MG |
1743 | ret = pthread_attr_destroy(&thread_attr); |
1744 | if (ret) { | |
1745 | ERR("pthread_attr_destroy: %s", strerror(ret)); | |
1746 | } | |
8d20bf54 | 1747 | |
ae6a58bf WP |
1748 | /* Restore original signal mask in parent */ |
1749 | ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL); | |
1750 | if (ret) { | |
d94d802c | 1751 | ERR("pthread_sigmask: %s", strerror(ret)); |
ae6a58bf WP |
1752 | } |
1753 | ||
cf12a773 MD |
1754 | switch (timeout_mode) { |
1755 | case 1: /* timeout wait */ | |
95259bd0 MD |
1756 | do { |
1757 | ret = sem_timedwait(&constructor_wait, | |
1758 | &constructor_timeout); | |
1759 | } while (ret < 0 && errno == EINTR); | |
8aadb54a MD |
1760 | if (ret < 0) { |
1761 | switch (errno) { | |
1762 | case ETIMEDOUT: | |
1763 | ERR("Timed out waiting for lttng-sessiond"); | |
1764 | break; | |
1765 | case EINVAL: | |
1766 | PERROR("sem_timedwait"); | |
1767 | break; | |
1768 | default: | |
1769 | ERR("Unexpected error \"%s\" returned by sem_timedwait", | |
1770 | strerror(errno)); | |
1771 | } | |
cf12a773 MD |
1772 | } |
1773 | break; | |
7b766b16 | 1774 | case -1:/* wait forever */ |
95259bd0 MD |
1775 | do { |
1776 | ret = sem_wait(&constructor_wait); | |
1777 | } while (ret < 0 && errno == EINTR); | |
8aadb54a MD |
1778 | if (ret < 0) { |
1779 | switch (errno) { | |
1780 | case EINVAL: | |
1781 | PERROR("sem_wait"); | |
1782 | break; | |
1783 | default: | |
1784 | ERR("Unexpected error \"%s\" returned by sem_wait", | |
1785 | strerror(errno)); | |
1786 | } | |
1787 | } | |
cf12a773 | 1788 | break; |
7b766b16 | 1789 | case 0: /* no timeout */ |
cf12a773 | 1790 | break; |
11ff9c7d | 1791 | } |
2691221a MD |
1792 | } |
1793 | ||
17dfb34b MD |
1794 | static |
1795 | void lttng_ust_cleanup(int exiting) | |
1796 | { | |
efe0de09 | 1797 | cleanup_sock_info(&global_apps, exiting); |
932cfadb | 1798 | cleanup_sock_info(&local_apps, exiting); |
74f98bc9 | 1799 | local_apps.allowed = 0; |
efe0de09 MD |
1800 | /* |
1801 | * The teardown in this function all affect data structures | |
1802 | * accessed under the UST lock by the listener thread. This | |
1803 | * lock, along with the lttng_ust_comm_should_quit flag, ensure | |
1804 | * that none of these threads are accessing this data at this | |
1805 | * point. | |
1806 | */ | |
17dfb34b | 1807 | lttng_ust_abi_exit(); |
003fedf4 | 1808 | lttng_ust_events_exit(); |
d58d1454 | 1809 | lttng_perf_counter_exit(); |
34a91bdb | 1810 | lttng_ring_buffer_client_discard_rt_exit(); |
7dd08bec | 1811 | lttng_ring_buffer_client_discard_exit(); |
34a91bdb | 1812 | lttng_ring_buffer_client_overwrite_rt_exit(); |
7dd08bec MD |
1813 | lttng_ring_buffer_client_overwrite_exit(); |
1814 | lttng_ring_buffer_metadata_client_exit(); | |
cf73e0fe | 1815 | lttng_ust_statedump_destroy(); |
17dfb34b MD |
1816 | exit_tracepoint(); |
1817 | if (!exiting) { | |
1818 | /* Reinitialize values for fork */ | |
1819 | sem_count = 2; | |
1820 | lttng_ust_comm_should_quit = 0; | |
1821 | initialized = 0; | |
1822 | } | |
1823 | } | |
1824 | ||
edaa1431 | 1825 | void __attribute__((destructor)) lttng_ust_exit(void) |
2691221a MD |
1826 | { |
1827 | int ret; | |
1828 | ||
9eb62b9c MD |
1829 | /* |
1830 | * Using pthread_cancel here because: | |
1831 | * A) we don't want to hang application teardown. | |
1832 | * B) the thread is not allocating any resource. | |
1833 | */ | |
1ea11eab MD |
1834 | |
1835 | /* | |
1836 | * Require the communication thread to quit. Synchronize with | |
1837 | * mutexes to ensure it is not in a mutex critical section when | |
1838 | * pthread_cancel is later called. | |
1839 | */ | |
3327ac33 | 1840 | ust_lock_nocheck(); |
1ea11eab | 1841 | lttng_ust_comm_should_quit = 1; |
3327ac33 | 1842 | ust_unlock(); |
1ea11eab | 1843 | |
3327ac33 | 1844 | pthread_mutex_lock(&ust_exit_mutex); |
f5f94532 | 1845 | /* cancel threads */ |
e33f3265 MD |
1846 | if (global_apps.thread_active) { |
1847 | ret = pthread_cancel(global_apps.ust_listener); | |
1848 | if (ret) { | |
1849 | ERR("Error cancelling global ust listener thread: %s", | |
1850 | strerror(ret)); | |
1851 | } else { | |
1852 | global_apps.thread_active = 0; | |
1853 | } | |
2691221a | 1854 | } |
e33f3265 | 1855 | if (local_apps.thread_active) { |
8d20bf54 MD |
1856 | ret = pthread_cancel(local_apps.ust_listener); |
1857 | if (ret) { | |
d94d802c MD |
1858 | ERR("Error cancelling local ust listener thread: %s", |
1859 | strerror(ret)); | |
e33f3265 MD |
1860 | } else { |
1861 | local_apps.thread_active = 0; | |
8d20bf54 | 1862 | } |
8d20bf54 | 1863 | } |
3327ac33 | 1864 | pthread_mutex_unlock(&ust_exit_mutex); |
e33f3265 | 1865 | |
efe0de09 MD |
1866 | /* |
1867 | * Do NOT join threads: use of sys_futex makes it impossible to | |
1868 | * join the threads without using async-cancel, but async-cancel | |
1869 | * is delivered by a signal, which could hit the target thread | |
1870 | * anywhere in its code path, including while the ust_lock() is | |
1871 | * held, causing a deadlock for the other thread. Let the OS | |
1872 | * cleanup the threads if there are stalled in a syscall. | |
1873 | */ | |
17dfb34b | 1874 | lttng_ust_cleanup(1); |
2691221a | 1875 | } |
e822f505 MD |
1876 | |
1877 | /* | |
1878 | * We exclude the worker threads across fork and clone (except | |
1879 | * CLONE_VM), because these system calls only keep the forking thread | |
1880 | * running in the child. Therefore, we don't want to call fork or clone | |
1881 | * in the middle of an tracepoint or ust tracing state modification. | |
1882 | * Holding this mutex protects these structures across fork and clone. | |
1883 | */ | |
b728d87e | 1884 | void ust_before_fork(sigset_t *save_sigset) |
e822f505 MD |
1885 | { |
1886 | /* | |
1887 | * Disable signals. This is to avoid that the child intervenes | |
1888 | * before it is properly setup for tracing. It is safer to | |
1889 | * disable all signals, because then we know we are not breaking | |
1890 | * anything by restoring the original mask. | |
1891 | */ | |
1892 | sigset_t all_sigs; | |
1893 | int ret; | |
1894 | ||
c362addf MD |
1895 | /* Fixup lttng-ust TLS. */ |
1896 | lttng_ust_fixup_tls(); | |
1897 | ||
8c90a710 | 1898 | if (URCU_TLS(lttng_ust_nest_count)) |
e8508a49 | 1899 | return; |
e822f505 MD |
1900 | /* Disable signals */ |
1901 | sigfillset(&all_sigs); | |
b728d87e | 1902 | ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset); |
e822f505 MD |
1903 | if (ret == -1) { |
1904 | PERROR("sigprocmask"); | |
1905 | } | |
458d678c PW |
1906 | |
1907 | pthread_mutex_lock(&ust_fork_mutex); | |
1908 | ||
3327ac33 | 1909 | ust_lock_nocheck(); |
e822f505 MD |
1910 | rcu_bp_before_fork(); |
1911 | } | |
1912 | ||
b728d87e | 1913 | static void ust_after_fork_common(sigset_t *restore_sigset) |
e822f505 MD |
1914 | { |
1915 | int ret; | |
1916 | ||
17dfb34b MD |
1917 | DBG("process %d", getpid()); |
1918 | ust_unlock(); | |
458d678c PW |
1919 | |
1920 | pthread_mutex_unlock(&ust_fork_mutex); | |
1921 | ||
e822f505 | 1922 | /* Restore signals */ |
23c8854a | 1923 | ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL); |
e822f505 MD |
1924 | if (ret == -1) { |
1925 | PERROR("sigprocmask"); | |
1926 | } | |
1927 | } | |
1928 | ||
b728d87e | 1929 | void ust_after_fork_parent(sigset_t *restore_sigset) |
e822f505 | 1930 | { |
8c90a710 | 1931 | if (URCU_TLS(lttng_ust_nest_count)) |
e8508a49 | 1932 | return; |
17dfb34b | 1933 | DBG("process %d", getpid()); |
e822f505 MD |
1934 | rcu_bp_after_fork_parent(); |
1935 | /* Release mutexes and reenable signals */ | |
b728d87e | 1936 | ust_after_fork_common(restore_sigset); |
e822f505 MD |
1937 | } |
1938 | ||
17dfb34b MD |
1939 | /* |
1940 | * After fork, in the child, we need to cleanup all the leftover state, | |
1941 | * except the worker thread which already magically disappeared thanks | |
1942 | * to the weird Linux fork semantics. After tyding up, we call | |
1943 | * lttng_ust_init() again to start over as a new PID. | |
1944 | * | |
1945 | * This is meant for forks() that have tracing in the child between the | |
1946 | * fork and following exec call (if there is any). | |
1947 | */ | |
b728d87e | 1948 | void ust_after_fork_child(sigset_t *restore_sigset) |
e822f505 | 1949 | { |
8c90a710 | 1950 | if (URCU_TLS(lttng_ust_nest_count)) |
e8508a49 | 1951 | return; |
8478887d | 1952 | lttng_context_vtid_reset(); |
17dfb34b | 1953 | DBG("process %d", getpid()); |
e822f505 MD |
1954 | /* Release urcu mutexes */ |
1955 | rcu_bp_after_fork_child(); | |
17dfb34b | 1956 | lttng_ust_cleanup(0); |
e822f505 | 1957 | /* Release mutexes and reenable signals */ |
b728d87e | 1958 | ust_after_fork_common(restore_sigset); |
318dfea9 | 1959 | lttng_ust_init(); |
e822f505 | 1960 | } |
95c25348 | 1961 | |
246be17e | 1962 | void lttng_ust_sockinfo_session_enabled(void *owner) |
95c25348 PW |
1963 | { |
1964 | struct sock_info *sock_info = owner; | |
37dddb65 | 1965 | sock_info->statedump_pending = 1; |
95c25348 | 1966 | } |