]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
fd38203a LP |
2 | |
3 | #include <sys/epoll.h> | |
4 | #include <sys/timerfd.h> | |
5 | #include <sys/wait.h> | |
6 | ||
cde93897 | 7 | #include "sd-daemon.h" |
07630cea LP |
8 | #include "sd-event.h" |
9 | #include "sd-id128.h" | |
10 | ||
b5efdb8a | 11 | #include "alloc-util.h" |
f8f3f926 | 12 | #include "env-util.h" |
a137a1c3 | 13 | #include "event-source.h" |
3ffd4af2 | 14 | #include "fd-util.h" |
97ef5391 | 15 | #include "fs-util.h" |
fd38203a | 16 | #include "hashmap.h" |
07630cea LP |
17 | #include "list.h" |
18 | #include "macro.h" | |
0a970718 | 19 | #include "memory-util.h" |
f5947a5e | 20 | #include "missing_syscall.h" |
07630cea | 21 | #include "prioq.h" |
4a0b58c4 | 22 | #include "process-util.h" |
6e9feda3 | 23 | #include "set.h" |
24882e06 | 24 | #include "signal-util.h" |
55cbfaa5 | 25 | #include "string-table.h" |
07630cea | 26 | #include "string-util.h" |
442ac269 | 27 | #include "strxcpyx.h" |
07630cea | 28 | #include "time-util.h" |
fd38203a | 29 | |
c2ba3ad6 | 30 | #define DEFAULT_ACCURACY_USEC (250 * USEC_PER_MSEC) |
fd38203a | 31 | |
f8f3f926 LP |
32 | static bool EVENT_SOURCE_WATCH_PIDFD(sd_event_source *s) { |
33 | /* Returns true if this is a PID event source and can be implemented by watching EPOLLIN */ | |
34 | return s && | |
35 | s->type == SOURCE_CHILD && | |
36 | s->child.pidfd >= 0 && | |
37 | s->child.options == WEXITED; | |
38 | } | |
39 | ||
55cbfaa5 DM |
40 | static const char* const event_source_type_table[_SOURCE_EVENT_SOURCE_TYPE_MAX] = { |
41 | [SOURCE_IO] = "io", | |
42 | [SOURCE_TIME_REALTIME] = "realtime", | |
43 | [SOURCE_TIME_BOOTTIME] = "bootime", | |
44 | [SOURCE_TIME_MONOTONIC] = "monotonic", | |
45 | [SOURCE_TIME_REALTIME_ALARM] = "realtime-alarm", | |
46 | [SOURCE_TIME_BOOTTIME_ALARM] = "boottime-alarm", | |
47 | [SOURCE_SIGNAL] = "signal", | |
48 | [SOURCE_CHILD] = "child", | |
49 | [SOURCE_DEFER] = "defer", | |
50 | [SOURCE_POST] = "post", | |
51 | [SOURCE_EXIT] = "exit", | |
52 | [SOURCE_WATCHDOG] = "watchdog", | |
97ef5391 | 53 | [SOURCE_INOTIFY] = "inotify", |
55cbfaa5 DM |
54 | }; |
55 | ||
56 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(event_source_type, int); | |
57 | ||
a8548816 | 58 | #define EVENT_SOURCE_IS_TIME(t) IN_SET((t), SOURCE_TIME_REALTIME, SOURCE_TIME_BOOTTIME, SOURCE_TIME_MONOTONIC, SOURCE_TIME_REALTIME_ALARM, SOURCE_TIME_BOOTTIME_ALARM) |
6a0f1f6d | 59 | |
fd38203a | 60 | struct sd_event { |
da7e457c | 61 | unsigned n_ref; |
fd38203a LP |
62 | |
63 | int epoll_fd; | |
cde93897 | 64 | int watchdog_fd; |
fd38203a LP |
65 | |
66 | Prioq *pending; | |
67 | Prioq *prepare; | |
c2ba3ad6 | 68 | |
a8548816 | 69 | /* timerfd_create() only supports these five clocks so far. We |
6a0f1f6d LP |
70 | * can add support for more clocks when the kernel learns to |
71 | * deal with them, too. */ | |
72 | struct clock_data realtime; | |
a8548816 | 73 | struct clock_data boottime; |
6a0f1f6d LP |
74 | struct clock_data monotonic; |
75 | struct clock_data realtime_alarm; | |
76 | struct clock_data boottime_alarm; | |
fd38203a | 77 | |
da7e457c LP |
78 | usec_t perturb; |
79 | ||
9da4cb2b LP |
80 | sd_event_source **signal_sources; /* indexed by signal number */ |
81 | Hashmap *signal_data; /* indexed by priority */ | |
fd38203a LP |
82 | |
83 | Hashmap *child_sources; | |
baf76283 | 84 | unsigned n_enabled_child_sources; |
fd38203a | 85 | |
6e9feda3 LP |
86 | Set *post_sources; |
87 | ||
6203e07a | 88 | Prioq *exit; |
fd38203a | 89 | |
97ef5391 LP |
90 | Hashmap *inotify_data; /* indexed by priority */ |
91 | ||
92 | /* A list of inode structures that still have an fd open, that we need to close before the next loop iteration */ | |
93 | LIST_HEAD(struct inode_data, inode_data_to_close); | |
94 | ||
95 | /* A list of inotify objects that already have events buffered which aren't processed yet */ | |
96 | LIST_HEAD(struct inotify_data, inotify_data_buffered); | |
97 | ||
da7e457c | 98 | pid_t original_pid; |
c2ba3ad6 | 99 | |
60a3b1e1 | 100 | uint64_t iteration; |
e475d10c | 101 | triple_timestamp timestamp; |
da7e457c | 102 | int state; |
eaa3cbef | 103 | |
6203e07a | 104 | bool exit_requested:1; |
da7e457c | 105 | bool need_process_child:1; |
cde93897 | 106 | bool watchdog:1; |
34b87517 | 107 | bool profile_delays:1; |
afc6adb5 | 108 | |
6203e07a LP |
109 | int exit_code; |
110 | ||
afc6adb5 LP |
111 | pid_t tid; |
112 | sd_event **default_event_ptr; | |
cde93897 LP |
113 | |
114 | usec_t watchdog_last, watchdog_period; | |
15b38f93 LP |
115 | |
116 | unsigned n_sources; | |
a71fe8b8 | 117 | |
5cddd924 LP |
118 | struct epoll_event *event_queue; |
119 | size_t event_queue_allocated; | |
120 | ||
a71fe8b8 | 121 | LIST_HEAD(sd_event_source, sources); |
34b87517 VC |
122 | |
123 | usec_t last_run, last_log; | |
124 | unsigned delays[sizeof(usec_t) * 8]; | |
fd38203a LP |
125 | }; |
126 | ||
b937d761 NM |
127 | static thread_local sd_event *default_event = NULL; |
128 | ||
a71fe8b8 | 129 | static void source_disconnect(sd_event_source *s); |
97ef5391 | 130 | static void event_gc_inode_data(sd_event *e, struct inode_data *d); |
a71fe8b8 | 131 | |
b937d761 NM |
132 | static sd_event *event_resolve(sd_event *e) { |
133 | return e == SD_EVENT_DEFAULT ? default_event : e; | |
134 | } | |
135 | ||
fd38203a LP |
136 | static int pending_prioq_compare(const void *a, const void *b) { |
137 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 138 | int r; |
fd38203a LP |
139 | |
140 | assert(x->pending); | |
141 | assert(y->pending); | |
142 | ||
baf76283 LP |
143 | /* Enabled ones first */ |
144 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
fd38203a | 145 | return -1; |
baf76283 | 146 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) |
fd38203a LP |
147 | return 1; |
148 | ||
149 | /* Lower priority values first */ | |
9c57a73b YW |
150 | r = CMP(x->priority, y->priority); |
151 | if (r != 0) | |
152 | return r; | |
fd38203a LP |
153 | |
154 | /* Older entries first */ | |
9c57a73b | 155 | return CMP(x->pending_iteration, y->pending_iteration); |
fd38203a LP |
156 | } |
157 | ||
158 | static int prepare_prioq_compare(const void *a, const void *b) { | |
159 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 160 | int r; |
fd38203a LP |
161 | |
162 | assert(x->prepare); | |
163 | assert(y->prepare); | |
164 | ||
8046c457 KK |
165 | /* Enabled ones first */ |
166 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
167 | return -1; | |
168 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) | |
169 | return 1; | |
170 | ||
fd38203a LP |
171 | /* Move most recently prepared ones last, so that we can stop |
172 | * preparing as soon as we hit one that has already been | |
173 | * prepared in the current iteration */ | |
9c57a73b YW |
174 | r = CMP(x->prepare_iteration, y->prepare_iteration); |
175 | if (r != 0) | |
176 | return r; | |
fd38203a | 177 | |
fd38203a | 178 | /* Lower priority values first */ |
9c57a73b | 179 | return CMP(x->priority, y->priority); |
fd38203a LP |
180 | } |
181 | ||
c2ba3ad6 | 182 | static int earliest_time_prioq_compare(const void *a, const void *b) { |
fd38203a LP |
183 | const sd_event_source *x = a, *y = b; |
184 | ||
6a0f1f6d LP |
185 | assert(EVENT_SOURCE_IS_TIME(x->type)); |
186 | assert(x->type == y->type); | |
fd38203a | 187 | |
baf76283 LP |
188 | /* Enabled ones first */ |
189 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
fd38203a | 190 | return -1; |
baf76283 | 191 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) |
fd38203a LP |
192 | return 1; |
193 | ||
194 | /* Move the pending ones to the end */ | |
195 | if (!x->pending && y->pending) | |
196 | return -1; | |
197 | if (x->pending && !y->pending) | |
198 | return 1; | |
199 | ||
200 | /* Order by time */ | |
9c57a73b | 201 | return CMP(x->time.next, y->time.next); |
fd38203a LP |
202 | } |
203 | ||
1bce0ffa LP |
204 | static usec_t time_event_source_latest(const sd_event_source *s) { |
205 | return usec_add(s->time.next, s->time.accuracy); | |
206 | } | |
207 | ||
c2ba3ad6 LP |
208 | static int latest_time_prioq_compare(const void *a, const void *b) { |
209 | const sd_event_source *x = a, *y = b; | |
210 | ||
6a0f1f6d LP |
211 | assert(EVENT_SOURCE_IS_TIME(x->type)); |
212 | assert(x->type == y->type); | |
c2ba3ad6 | 213 | |
baf76283 LP |
214 | /* Enabled ones first */ |
215 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
c2ba3ad6 | 216 | return -1; |
baf76283 | 217 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) |
c2ba3ad6 LP |
218 | return 1; |
219 | ||
220 | /* Move the pending ones to the end */ | |
221 | if (!x->pending && y->pending) | |
222 | return -1; | |
223 | if (x->pending && !y->pending) | |
224 | return 1; | |
225 | ||
226 | /* Order by time */ | |
9c57a73b | 227 | return CMP(time_event_source_latest(x), time_event_source_latest(y)); |
c2ba3ad6 LP |
228 | } |
229 | ||
6203e07a | 230 | static int exit_prioq_compare(const void *a, const void *b) { |
da7e457c LP |
231 | const sd_event_source *x = a, *y = b; |
232 | ||
6203e07a LP |
233 | assert(x->type == SOURCE_EXIT); |
234 | assert(y->type == SOURCE_EXIT); | |
da7e457c | 235 | |
baf76283 LP |
236 | /* Enabled ones first */ |
237 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
da7e457c | 238 | return -1; |
baf76283 | 239 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) |
da7e457c LP |
240 | return 1; |
241 | ||
242 | /* Lower priority values first */ | |
6dd91b36 | 243 | return CMP(x->priority, y->priority); |
da7e457c LP |
244 | } |
245 | ||
6a0f1f6d LP |
246 | static void free_clock_data(struct clock_data *d) { |
247 | assert(d); | |
9da4cb2b | 248 | assert(d->wakeup == WAKEUP_CLOCK_DATA); |
6a0f1f6d LP |
249 | |
250 | safe_close(d->fd); | |
251 | prioq_free(d->earliest); | |
252 | prioq_free(d->latest); | |
253 | } | |
254 | ||
8301aa0b | 255 | static sd_event *event_free(sd_event *e) { |
a71fe8b8 LP |
256 | sd_event_source *s; |
257 | ||
fd38203a | 258 | assert(e); |
a71fe8b8 LP |
259 | |
260 | while ((s = e->sources)) { | |
261 | assert(s->floating); | |
262 | source_disconnect(s); | |
263 | sd_event_source_unref(s); | |
264 | } | |
265 | ||
15b38f93 | 266 | assert(e->n_sources == 0); |
fd38203a | 267 | |
afc6adb5 LP |
268 | if (e->default_event_ptr) |
269 | *(e->default_event_ptr) = NULL; | |
270 | ||
03e334a1 | 271 | safe_close(e->epoll_fd); |
03e334a1 | 272 | safe_close(e->watchdog_fd); |
cde93897 | 273 | |
6a0f1f6d | 274 | free_clock_data(&e->realtime); |
a8548816 | 275 | free_clock_data(&e->boottime); |
6a0f1f6d LP |
276 | free_clock_data(&e->monotonic); |
277 | free_clock_data(&e->realtime_alarm); | |
278 | free_clock_data(&e->boottime_alarm); | |
279 | ||
fd38203a LP |
280 | prioq_free(e->pending); |
281 | prioq_free(e->prepare); | |
6203e07a | 282 | prioq_free(e->exit); |
fd38203a LP |
283 | |
284 | free(e->signal_sources); | |
9da4cb2b | 285 | hashmap_free(e->signal_data); |
fd38203a | 286 | |
97ef5391 LP |
287 | hashmap_free(e->inotify_data); |
288 | ||
fd38203a | 289 | hashmap_free(e->child_sources); |
6e9feda3 | 290 | set_free(e->post_sources); |
8301aa0b | 291 | |
5cddd924 LP |
292 | free(e->event_queue); |
293 | ||
8301aa0b | 294 | return mfree(e); |
fd38203a LP |
295 | } |
296 | ||
f7262a9f | 297 | _public_ int sd_event_new(sd_event** ret) { |
fd38203a LP |
298 | sd_event *e; |
299 | int r; | |
300 | ||
305f78bf | 301 | assert_return(ret, -EINVAL); |
fd38203a | 302 | |
d08eb1fa | 303 | e = new(sd_event, 1); |
fd38203a LP |
304 | if (!e) |
305 | return -ENOMEM; | |
306 | ||
d08eb1fa LP |
307 | *e = (sd_event) { |
308 | .n_ref = 1, | |
309 | .epoll_fd = -1, | |
310 | .watchdog_fd = -1, | |
311 | .realtime.wakeup = WAKEUP_CLOCK_DATA, | |
312 | .realtime.fd = -1, | |
313 | .realtime.next = USEC_INFINITY, | |
314 | .boottime.wakeup = WAKEUP_CLOCK_DATA, | |
315 | .boottime.fd = -1, | |
316 | .boottime.next = USEC_INFINITY, | |
317 | .monotonic.wakeup = WAKEUP_CLOCK_DATA, | |
318 | .monotonic.fd = -1, | |
319 | .monotonic.next = USEC_INFINITY, | |
320 | .realtime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
321 | .realtime_alarm.fd = -1, | |
322 | .realtime_alarm.next = USEC_INFINITY, | |
323 | .boottime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
324 | .boottime_alarm.fd = -1, | |
325 | .boottime_alarm.next = USEC_INFINITY, | |
326 | .perturb = USEC_INFINITY, | |
327 | .original_pid = getpid_cached(), | |
328 | }; | |
fd38203a | 329 | |
c983e776 EV |
330 | r = prioq_ensure_allocated(&e->pending, pending_prioq_compare); |
331 | if (r < 0) | |
fd38203a | 332 | goto fail; |
fd38203a LP |
333 | |
334 | e->epoll_fd = epoll_create1(EPOLL_CLOEXEC); | |
335 | if (e->epoll_fd < 0) { | |
336 | r = -errno; | |
337 | goto fail; | |
338 | } | |
339 | ||
7fe2903c LP |
340 | e->epoll_fd = fd_move_above_stdio(e->epoll_fd); |
341 | ||
34b87517 | 342 | if (secure_getenv("SD_EVENT_PROFILE_DELAYS")) { |
34a6843d | 343 | log_debug("Event loop profiling enabled. Logarithmic histogram of event loop iterations in the range 2^0 ... 2^63 us will be logged every 5s."); |
34b87517 VC |
344 | e->profile_delays = true; |
345 | } | |
346 | ||
fd38203a LP |
347 | *ret = e; |
348 | return 0; | |
349 | ||
350 | fail: | |
351 | event_free(e); | |
352 | return r; | |
353 | } | |
354 | ||
8301aa0b | 355 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event, sd_event, event_free); |
fd38203a | 356 | |
afd15bbb ZJS |
357 | _public_ sd_event_source* sd_event_source_disable_unref(sd_event_source *s) { |
358 | if (s) | |
359 | (void) sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
360 | return sd_event_source_unref(s); | |
361 | } | |
362 | ||
eaa3cbef LP |
363 | static bool event_pid_changed(sd_event *e) { |
364 | assert(e); | |
365 | ||
a2360a46 | 366 | /* We don't support people creating an event loop and keeping |
eaa3cbef LP |
367 | * it around over a fork(). Let's complain. */ |
368 | ||
df0ff127 | 369 | return e->original_pid != getpid_cached(); |
eaa3cbef LP |
370 | } |
371 | ||
366e6411 | 372 | static void source_io_unregister(sd_event_source *s) { |
fd38203a LP |
373 | assert(s); |
374 | assert(s->type == SOURCE_IO); | |
375 | ||
f6806734 | 376 | if (event_pid_changed(s->event)) |
366e6411 | 377 | return; |
f6806734 | 378 | |
fd38203a | 379 | if (!s->io.registered) |
366e6411 | 380 | return; |
fd38203a | 381 | |
d1cf2023 | 382 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->io.fd, NULL) < 0) |
55cbfaa5 DM |
383 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll: %m", |
384 | strna(s->description), event_source_type_to_string(s->type)); | |
fd38203a LP |
385 | |
386 | s->io.registered = false; | |
fd38203a LP |
387 | } |
388 | ||
305f78bf LP |
389 | static int source_io_register( |
390 | sd_event_source *s, | |
391 | int enabled, | |
392 | uint32_t events) { | |
393 | ||
fd38203a LP |
394 | assert(s); |
395 | assert(s->type == SOURCE_IO); | |
baf76283 | 396 | assert(enabled != SD_EVENT_OFF); |
fd38203a | 397 | |
1eac7948 | 398 | struct epoll_event ev = { |
a82f89aa LP |
399 | .events = events | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0), |
400 | .data.ptr = s, | |
401 | }; | |
fd38203a | 402 | |
15c689d7 | 403 | if (epoll_ctl(s->event->epoll_fd, |
1eac7948 ZJS |
404 | s->io.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, |
405 | s->io.fd, | |
15c689d7 | 406 | &ev) < 0) |
fd38203a LP |
407 | return -errno; |
408 | ||
409 | s->io.registered = true; | |
410 | ||
411 | return 0; | |
412 | } | |
413 | ||
f8f3f926 LP |
414 | static void source_child_pidfd_unregister(sd_event_source *s) { |
415 | assert(s); | |
416 | assert(s->type == SOURCE_CHILD); | |
417 | ||
418 | if (event_pid_changed(s->event)) | |
419 | return; | |
420 | ||
421 | if (!s->child.registered) | |
422 | return; | |
423 | ||
424 | if (EVENT_SOURCE_WATCH_PIDFD(s)) | |
425 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->child.pidfd, NULL) < 0) | |
426 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll: %m", | |
427 | strna(s->description), event_source_type_to_string(s->type)); | |
428 | ||
429 | s->child.registered = false; | |
430 | } | |
431 | ||
432 | static int source_child_pidfd_register(sd_event_source *s, int enabled) { | |
433 | int r; | |
434 | ||
435 | assert(s); | |
436 | assert(s->type == SOURCE_CHILD); | |
437 | assert(enabled != SD_EVENT_OFF); | |
438 | ||
439 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { | |
1eac7948 | 440 | struct epoll_event ev = { |
f8f3f926 LP |
441 | .events = EPOLLIN | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0), |
442 | .data.ptr = s, | |
443 | }; | |
444 | ||
445 | if (s->child.registered) | |
446 | r = epoll_ctl(s->event->epoll_fd, EPOLL_CTL_MOD, s->child.pidfd, &ev); | |
447 | else | |
448 | r = epoll_ctl(s->event->epoll_fd, EPOLL_CTL_ADD, s->child.pidfd, &ev); | |
449 | if (r < 0) | |
450 | return -errno; | |
451 | } | |
452 | ||
453 | s->child.registered = true; | |
454 | return 0; | |
455 | } | |
456 | ||
6a0f1f6d LP |
457 | static clockid_t event_source_type_to_clock(EventSourceType t) { |
458 | ||
459 | switch (t) { | |
460 | ||
461 | case SOURCE_TIME_REALTIME: | |
462 | return CLOCK_REALTIME; | |
463 | ||
a8548816 TG |
464 | case SOURCE_TIME_BOOTTIME: |
465 | return CLOCK_BOOTTIME; | |
466 | ||
6a0f1f6d LP |
467 | case SOURCE_TIME_MONOTONIC: |
468 | return CLOCK_MONOTONIC; | |
469 | ||
470 | case SOURCE_TIME_REALTIME_ALARM: | |
471 | return CLOCK_REALTIME_ALARM; | |
472 | ||
473 | case SOURCE_TIME_BOOTTIME_ALARM: | |
474 | return CLOCK_BOOTTIME_ALARM; | |
475 | ||
476 | default: | |
477 | return (clockid_t) -1; | |
478 | } | |
479 | } | |
480 | ||
481 | static EventSourceType clock_to_event_source_type(clockid_t clock) { | |
482 | ||
483 | switch (clock) { | |
484 | ||
485 | case CLOCK_REALTIME: | |
486 | return SOURCE_TIME_REALTIME; | |
487 | ||
a8548816 TG |
488 | case CLOCK_BOOTTIME: |
489 | return SOURCE_TIME_BOOTTIME; | |
490 | ||
6a0f1f6d LP |
491 | case CLOCK_MONOTONIC: |
492 | return SOURCE_TIME_MONOTONIC; | |
493 | ||
494 | case CLOCK_REALTIME_ALARM: | |
495 | return SOURCE_TIME_REALTIME_ALARM; | |
496 | ||
497 | case CLOCK_BOOTTIME_ALARM: | |
498 | return SOURCE_TIME_BOOTTIME_ALARM; | |
499 | ||
500 | default: | |
501 | return _SOURCE_EVENT_SOURCE_TYPE_INVALID; | |
502 | } | |
503 | } | |
504 | ||
505 | static struct clock_data* event_get_clock_data(sd_event *e, EventSourceType t) { | |
506 | assert(e); | |
507 | ||
508 | switch (t) { | |
509 | ||
510 | case SOURCE_TIME_REALTIME: | |
511 | return &e->realtime; | |
512 | ||
a8548816 TG |
513 | case SOURCE_TIME_BOOTTIME: |
514 | return &e->boottime; | |
515 | ||
6a0f1f6d LP |
516 | case SOURCE_TIME_MONOTONIC: |
517 | return &e->monotonic; | |
518 | ||
519 | case SOURCE_TIME_REALTIME_ALARM: | |
520 | return &e->realtime_alarm; | |
521 | ||
522 | case SOURCE_TIME_BOOTTIME_ALARM: | |
523 | return &e->boottime_alarm; | |
524 | ||
525 | default: | |
526 | return NULL; | |
527 | } | |
528 | } | |
529 | ||
3e4eb8e7 YW |
530 | static void event_free_signal_data(sd_event *e, struct signal_data *d) { |
531 | assert(e); | |
532 | ||
533 | if (!d) | |
534 | return; | |
535 | ||
536 | hashmap_remove(e->signal_data, &d->priority); | |
537 | safe_close(d->fd); | |
538 | free(d); | |
539 | } | |
540 | ||
9da4cb2b LP |
541 | static int event_make_signal_data( |
542 | sd_event *e, | |
543 | int sig, | |
544 | struct signal_data **ret) { | |
4807d2d0 | 545 | |
9da4cb2b LP |
546 | struct signal_data *d; |
547 | bool added = false; | |
548 | sigset_t ss_copy; | |
549 | int64_t priority; | |
f95387cd ZJS |
550 | int r; |
551 | ||
552 | assert(e); | |
553 | ||
f6806734 | 554 | if (event_pid_changed(e)) |
9da4cb2b | 555 | return -ECHILD; |
f6806734 | 556 | |
9da4cb2b LP |
557 | if (e->signal_sources && e->signal_sources[sig]) |
558 | priority = e->signal_sources[sig]->priority; | |
559 | else | |
de05913d | 560 | priority = SD_EVENT_PRIORITY_NORMAL; |
f95387cd | 561 | |
9da4cb2b LP |
562 | d = hashmap_get(e->signal_data, &priority); |
563 | if (d) { | |
564 | if (sigismember(&d->sigset, sig) > 0) { | |
565 | if (ret) | |
566 | *ret = d; | |
567 | return 0; | |
568 | } | |
569 | } else { | |
570 | r = hashmap_ensure_allocated(&e->signal_data, &uint64_hash_ops); | |
571 | if (r < 0) | |
572 | return r; | |
573 | ||
d08eb1fa | 574 | d = new(struct signal_data, 1); |
9da4cb2b LP |
575 | if (!d) |
576 | return -ENOMEM; | |
577 | ||
d08eb1fa LP |
578 | *d = (struct signal_data) { |
579 | .wakeup = WAKEUP_SIGNAL_DATA, | |
580 | .fd = -1, | |
581 | .priority = priority, | |
582 | }; | |
9da4cb2b LP |
583 | |
584 | r = hashmap_put(e->signal_data, &d->priority, d); | |
90f604d1 ZJS |
585 | if (r < 0) { |
586 | free(d); | |
9da4cb2b | 587 | return r; |
90f604d1 | 588 | } |
f95387cd | 589 | |
9da4cb2b LP |
590 | added = true; |
591 | } | |
592 | ||
593 | ss_copy = d->sigset; | |
594 | assert_se(sigaddset(&ss_copy, sig) >= 0); | |
595 | ||
596 | r = signalfd(d->fd, &ss_copy, SFD_NONBLOCK|SFD_CLOEXEC); | |
597 | if (r < 0) { | |
598 | r = -errno; | |
599 | goto fail; | |
600 | } | |
601 | ||
602 | d->sigset = ss_copy; | |
f95387cd | 603 | |
9da4cb2b LP |
604 | if (d->fd >= 0) { |
605 | if (ret) | |
606 | *ret = d; | |
f95387cd | 607 | return 0; |
9da4cb2b LP |
608 | } |
609 | ||
7fe2903c | 610 | d->fd = fd_move_above_stdio(r); |
f95387cd | 611 | |
1eac7948 | 612 | struct epoll_event ev = { |
a82f89aa LP |
613 | .events = EPOLLIN, |
614 | .data.ptr = d, | |
615 | }; | |
f95387cd | 616 | |
15c689d7 | 617 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { |
9da4cb2b LP |
618 | r = -errno; |
619 | goto fail; | |
f95387cd ZJS |
620 | } |
621 | ||
9da4cb2b LP |
622 | if (ret) |
623 | *ret = d; | |
624 | ||
f95387cd | 625 | return 0; |
9da4cb2b LP |
626 | |
627 | fail: | |
3e4eb8e7 YW |
628 | if (added) |
629 | event_free_signal_data(e, d); | |
9da4cb2b LP |
630 | |
631 | return r; | |
632 | } | |
633 | ||
634 | static void event_unmask_signal_data(sd_event *e, struct signal_data *d, int sig) { | |
635 | assert(e); | |
636 | assert(d); | |
637 | ||
638 | /* Turns off the specified signal in the signal data | |
639 | * object. If the signal mask of the object becomes empty that | |
640 | * way removes it. */ | |
641 | ||
642 | if (sigismember(&d->sigset, sig) == 0) | |
643 | return; | |
644 | ||
645 | assert_se(sigdelset(&d->sigset, sig) >= 0); | |
646 | ||
647 | if (sigisemptyset(&d->sigset)) { | |
9da4cb2b | 648 | /* If all the mask is all-zero we can get rid of the structure */ |
3e4eb8e7 | 649 | event_free_signal_data(e, d); |
9da4cb2b LP |
650 | return; |
651 | } | |
652 | ||
653 | assert(d->fd >= 0); | |
654 | ||
655 | if (signalfd(d->fd, &d->sigset, SFD_NONBLOCK|SFD_CLOEXEC) < 0) | |
656 | log_debug_errno(errno, "Failed to unset signal bit, ignoring: %m"); | |
657 | } | |
658 | ||
659 | static void event_gc_signal_data(sd_event *e, const int64_t *priority, int sig) { | |
660 | struct signal_data *d; | |
661 | static const int64_t zero_priority = 0; | |
662 | ||
663 | assert(e); | |
664 | ||
f8f3f926 LP |
665 | /* Rechecks if the specified signal is still something we are interested in. If not, we'll unmask it, |
666 | * and possibly drop the signalfd for it. */ | |
9da4cb2b LP |
667 | |
668 | if (sig == SIGCHLD && | |
669 | e->n_enabled_child_sources > 0) | |
670 | return; | |
671 | ||
672 | if (e->signal_sources && | |
673 | e->signal_sources[sig] && | |
674 | e->signal_sources[sig]->enabled != SD_EVENT_OFF) | |
675 | return; | |
676 | ||
677 | /* | |
678 | * The specified signal might be enabled in three different queues: | |
679 | * | |
680 | * 1) the one that belongs to the priority passed (if it is non-NULL) | |
681 | * 2) the one that belongs to the priority of the event source of the signal (if there is one) | |
682 | * 3) the 0 priority (to cover the SIGCHLD case) | |
683 | * | |
684 | * Hence, let's remove it from all three here. | |
685 | */ | |
686 | ||
687 | if (priority) { | |
688 | d = hashmap_get(e->signal_data, priority); | |
689 | if (d) | |
690 | event_unmask_signal_data(e, d, sig); | |
691 | } | |
692 | ||
693 | if (e->signal_sources && e->signal_sources[sig]) { | |
694 | d = hashmap_get(e->signal_data, &e->signal_sources[sig]->priority); | |
695 | if (d) | |
696 | event_unmask_signal_data(e, d, sig); | |
697 | } | |
698 | ||
699 | d = hashmap_get(e->signal_data, &zero_priority); | |
700 | if (d) | |
701 | event_unmask_signal_data(e, d, sig); | |
f95387cd ZJS |
702 | } |
703 | ||
e1951c16 MS |
704 | static void event_source_pp_prioq_reshuffle(sd_event_source *s) { |
705 | assert(s); | |
706 | ||
707 | /* Reshuffles the pending + prepare prioqs. Called whenever the dispatch order changes, i.e. when | |
708 | * they are enabled/disabled or marked pending and such. */ | |
709 | ||
710 | if (s->pending) | |
711 | prioq_reshuffle(s->event->pending, s, &s->pending_index); | |
712 | ||
713 | if (s->prepare) | |
714 | prioq_reshuffle(s->event->prepare, s, &s->prepare_index); | |
715 | } | |
716 | ||
717 | static void event_source_time_prioq_reshuffle(sd_event_source *s) { | |
718 | struct clock_data *d; | |
719 | ||
720 | assert(s); | |
721 | assert(EVENT_SOURCE_IS_TIME(s->type)); | |
722 | ||
723 | /* Called whenever the event source's timer ordering properties changed, i.e. time, accuracy, | |
724 | * pending, enable state. Makes sure the two prioq's are ordered properly again. */ | |
725 | assert_se(d = event_get_clock_data(s->event, s->type)); | |
726 | prioq_reshuffle(d->earliest, s, &s->time.earliest_index); | |
727 | prioq_reshuffle(d->latest, s, &s->time.latest_index); | |
728 | d->needs_rearm = true; | |
729 | } | |
730 | ||
a71fe8b8 LP |
731 | static void source_disconnect(sd_event_source *s) { |
732 | sd_event *event; | |
733 | ||
fd38203a LP |
734 | assert(s); |
735 | ||
a71fe8b8 LP |
736 | if (!s->event) |
737 | return; | |
15b38f93 | 738 | |
a71fe8b8 | 739 | assert(s->event->n_sources > 0); |
fd38203a | 740 | |
a71fe8b8 | 741 | switch (s->type) { |
fd38203a | 742 | |
a71fe8b8 LP |
743 | case SOURCE_IO: |
744 | if (s->io.fd >= 0) | |
745 | source_io_unregister(s); | |
fd38203a | 746 | |
a71fe8b8 | 747 | break; |
6a0f1f6d | 748 | |
a71fe8b8 | 749 | case SOURCE_TIME_REALTIME: |
a8548816 | 750 | case SOURCE_TIME_BOOTTIME: |
a71fe8b8 LP |
751 | case SOURCE_TIME_MONOTONIC: |
752 | case SOURCE_TIME_REALTIME_ALARM: | |
753 | case SOURCE_TIME_BOOTTIME_ALARM: { | |
754 | struct clock_data *d; | |
fd38203a | 755 | |
a71fe8b8 LP |
756 | d = event_get_clock_data(s->event, s->type); |
757 | assert(d); | |
758 | ||
759 | prioq_remove(d->earliest, s, &s->time.earliest_index); | |
760 | prioq_remove(d->latest, s, &s->time.latest_index); | |
212bbb17 | 761 | d->needs_rearm = true; |
a71fe8b8 LP |
762 | break; |
763 | } | |
764 | ||
765 | case SOURCE_SIGNAL: | |
766 | if (s->signal.sig > 0) { | |
9da4cb2b | 767 | |
a71fe8b8 LP |
768 | if (s->event->signal_sources) |
769 | s->event->signal_sources[s->signal.sig] = NULL; | |
4807d2d0 | 770 | |
9da4cb2b | 771 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
6a0f1f6d | 772 | } |
fd38203a | 773 | |
a71fe8b8 | 774 | break; |
fd38203a | 775 | |
a71fe8b8 LP |
776 | case SOURCE_CHILD: |
777 | if (s->child.pid > 0) { | |
778 | if (s->enabled != SD_EVENT_OFF) { | |
779 | assert(s->event->n_enabled_child_sources > 0); | |
780 | s->event->n_enabled_child_sources--; | |
4807d2d0 | 781 | } |
fd38203a | 782 | |
4a0b58c4 | 783 | (void) hashmap_remove(s->event->child_sources, PID_TO_PTR(s->child.pid)); |
a71fe8b8 | 784 | } |
fd38203a | 785 | |
f8f3f926 LP |
786 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
787 | source_child_pidfd_unregister(s); | |
788 | else | |
789 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
790 | ||
a71fe8b8 | 791 | break; |
fd38203a | 792 | |
a71fe8b8 LP |
793 | case SOURCE_DEFER: |
794 | /* nothing */ | |
795 | break; | |
fd38203a | 796 | |
a71fe8b8 LP |
797 | case SOURCE_POST: |
798 | set_remove(s->event->post_sources, s); | |
799 | break; | |
da7e457c | 800 | |
a71fe8b8 LP |
801 | case SOURCE_EXIT: |
802 | prioq_remove(s->event->exit, s, &s->exit.prioq_index); | |
803 | break; | |
0eb2e0e3 | 804 | |
97ef5391 LP |
805 | case SOURCE_INOTIFY: { |
806 | struct inode_data *inode_data; | |
807 | ||
808 | inode_data = s->inotify.inode_data; | |
809 | if (inode_data) { | |
810 | struct inotify_data *inotify_data; | |
811 | assert_se(inotify_data = inode_data->inotify_data); | |
812 | ||
813 | /* Detach this event source from the inode object */ | |
814 | LIST_REMOVE(inotify.by_inode_data, inode_data->event_sources, s); | |
815 | s->inotify.inode_data = NULL; | |
816 | ||
817 | if (s->pending) { | |
818 | assert(inotify_data->n_pending > 0); | |
819 | inotify_data->n_pending--; | |
820 | } | |
821 | ||
822 | /* Note that we don't reduce the inotify mask for the watch descriptor here if the inode is | |
823 | * continued to being watched. That's because inotify doesn't really have an API for that: we | |
824 | * can only change watch masks with access to the original inode either by fd or by path. But | |
825 | * paths aren't stable, and keeping an O_PATH fd open all the time would mean wasting an fd | |
f21f31b2 | 826 | * continuously and keeping the mount busy which we can't really do. We could reconstruct the |
97ef5391 LP |
827 | * original inode from /proc/self/fdinfo/$INOTIFY_FD (as all watch descriptors are listed |
828 | * there), but given the need for open_by_handle_at() which is privileged and not universally | |
829 | * available this would be quite an incomplete solution. Hence we go the other way, leave the | |
830 | * mask set, even if it is not minimized now, and ignore all events we aren't interested in | |
831 | * anymore after reception. Yes, this sucks, but … Linux … */ | |
832 | ||
833 | /* Maybe release the inode data (and its inotify) */ | |
834 | event_gc_inode_data(s->event, inode_data); | |
835 | } | |
836 | ||
837 | break; | |
838 | } | |
839 | ||
a71fe8b8 LP |
840 | default: |
841 | assert_not_reached("Wut? I shouldn't exist."); | |
842 | } | |
6e9feda3 | 843 | |
a71fe8b8 LP |
844 | if (s->pending) |
845 | prioq_remove(s->event->pending, s, &s->pending_index); | |
9d3e3aa5 | 846 | |
a71fe8b8 LP |
847 | if (s->prepare) |
848 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
fd38203a | 849 | |
e514aa1e | 850 | event = TAKE_PTR(s->event); |
a71fe8b8 LP |
851 | LIST_REMOVE(sources, event->sources, s); |
852 | event->n_sources--; | |
fd38203a | 853 | |
f5982559 LP |
854 | /* Note that we don't invalidate the type here, since we still need it in order to close the fd or |
855 | * pidfd associated with this event source, which we'll do only on source_free(). */ | |
856 | ||
a71fe8b8 LP |
857 | if (!s->floating) |
858 | sd_event_unref(event); | |
859 | } | |
860 | ||
861 | static void source_free(sd_event_source *s) { | |
862 | assert(s); | |
fd38203a | 863 | |
a71fe8b8 | 864 | source_disconnect(s); |
ab93297c NM |
865 | |
866 | if (s->type == SOURCE_IO && s->io.owned) | |
15723a1d LP |
867 | s->io.fd = safe_close(s->io.fd); |
868 | ||
f8f3f926 LP |
869 | if (s->type == SOURCE_CHILD) { |
870 | /* Eventually the kernel will do this automatically for us, but for now let's emulate this (unreliably) in userspace. */ | |
871 | ||
872 | if (s->child.process_owned) { | |
873 | ||
874 | if (!s->child.exited) { | |
875 | bool sent = false; | |
876 | ||
877 | if (s->child.pidfd >= 0) { | |
878 | if (pidfd_send_signal(s->child.pidfd, SIGKILL, NULL, 0) < 0) { | |
879 | if (errno == ESRCH) /* Already dead */ | |
880 | sent = true; | |
881 | else if (!ERRNO_IS_NOT_SUPPORTED(errno)) | |
882 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via pidfd_send_signal(), re-trying via kill(): %m", | |
883 | s->child.pid); | |
884 | } else | |
885 | sent = true; | |
886 | } | |
887 | ||
888 | if (!sent) | |
889 | if (kill(s->child.pid, SIGKILL) < 0) | |
890 | if (errno != ESRCH) /* Already dead */ | |
891 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via kill(), ignoring: %m", | |
892 | s->child.pid); | |
893 | } | |
894 | ||
895 | if (!s->child.waited) { | |
896 | siginfo_t si = {}; | |
897 | ||
898 | /* Reap the child if we can */ | |
899 | (void) waitid(P_PID, s->child.pid, &si, WEXITED); | |
900 | } | |
901 | } | |
902 | ||
903 | if (s->child.pidfd_owned) | |
904 | s->child.pidfd = safe_close(s->child.pidfd); | |
905 | } | |
906 | ||
15723a1d LP |
907 | if (s->destroy_callback) |
908 | s->destroy_callback(s->userdata); | |
ab93297c | 909 | |
356779df | 910 | free(s->description); |
fd38203a LP |
911 | free(s); |
912 | } | |
8c75fe17 | 913 | DEFINE_TRIVIAL_CLEANUP_FUNC(sd_event_source*, source_free); |
fd38203a LP |
914 | |
915 | static int source_set_pending(sd_event_source *s, bool b) { | |
916 | int r; | |
917 | ||
918 | assert(s); | |
6203e07a | 919 | assert(s->type != SOURCE_EXIT); |
fd38203a LP |
920 | |
921 | if (s->pending == b) | |
922 | return 0; | |
923 | ||
924 | s->pending = b; | |
925 | ||
926 | if (b) { | |
927 | s->pending_iteration = s->event->iteration; | |
928 | ||
929 | r = prioq_put(s->event->pending, s, &s->pending_index); | |
930 | if (r < 0) { | |
931 | s->pending = false; | |
932 | return r; | |
933 | } | |
934 | } else | |
935 | assert_se(prioq_remove(s->event->pending, s, &s->pending_index)); | |
936 | ||
e1951c16 MS |
937 | if (EVENT_SOURCE_IS_TIME(s->type)) |
938 | event_source_time_prioq_reshuffle(s); | |
2576a19e | 939 | |
9da4cb2b LP |
940 | if (s->type == SOURCE_SIGNAL && !b) { |
941 | struct signal_data *d; | |
942 | ||
943 | d = hashmap_get(s->event->signal_data, &s->priority); | |
944 | if (d && d->current == s) | |
945 | d->current = NULL; | |
946 | } | |
947 | ||
97ef5391 LP |
948 | if (s->type == SOURCE_INOTIFY) { |
949 | ||
950 | assert(s->inotify.inode_data); | |
951 | assert(s->inotify.inode_data->inotify_data); | |
952 | ||
953 | if (b) | |
954 | s->inotify.inode_data->inotify_data->n_pending ++; | |
955 | else { | |
956 | assert(s->inotify.inode_data->inotify_data->n_pending > 0); | |
957 | s->inotify.inode_data->inotify_data->n_pending --; | |
958 | } | |
959 | } | |
960 | ||
fd38203a LP |
961 | return 0; |
962 | } | |
963 | ||
a71fe8b8 | 964 | static sd_event_source *source_new(sd_event *e, bool floating, EventSourceType type) { |
fd38203a LP |
965 | sd_event_source *s; |
966 | ||
967 | assert(e); | |
968 | ||
d08eb1fa | 969 | s = new(sd_event_source, 1); |
fd38203a LP |
970 | if (!s) |
971 | return NULL; | |
972 | ||
d08eb1fa LP |
973 | *s = (struct sd_event_source) { |
974 | .n_ref = 1, | |
975 | .event = e, | |
976 | .floating = floating, | |
977 | .type = type, | |
978 | .pending_index = PRIOQ_IDX_NULL, | |
979 | .prepare_index = PRIOQ_IDX_NULL, | |
980 | }; | |
a71fe8b8 LP |
981 | |
982 | if (!floating) | |
983 | sd_event_ref(e); | |
fd38203a | 984 | |
a71fe8b8 | 985 | LIST_PREPEND(sources, e->sources, s); |
313cefa1 | 986 | e->n_sources++; |
15b38f93 | 987 | |
fd38203a LP |
988 | return s; |
989 | } | |
990 | ||
b9350e70 LP |
991 | static int io_exit_callback(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
992 | assert(s); | |
993 | ||
994 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
995 | } | |
996 | ||
f7262a9f | 997 | _public_ int sd_event_add_io( |
fd38203a | 998 | sd_event *e, |
151b9b96 | 999 | sd_event_source **ret, |
fd38203a LP |
1000 | int fd, |
1001 | uint32_t events, | |
718db961 | 1002 | sd_event_io_handler_t callback, |
151b9b96 | 1003 | void *userdata) { |
fd38203a | 1004 | |
ec766a51 | 1005 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1006 | int r; |
1007 | ||
305f78bf | 1008 | assert_return(e, -EINVAL); |
b937d761 | 1009 | assert_return(e = event_resolve(e), -ENOPKG); |
8ac43fee | 1010 | assert_return(fd >= 0, -EBADF); |
2a16a986 | 1011 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 1012 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1013 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1014 | |
b9350e70 LP |
1015 | if (!callback) |
1016 | callback = io_exit_callback; | |
1017 | ||
a71fe8b8 | 1018 | s = source_new(e, !ret, SOURCE_IO); |
fd38203a LP |
1019 | if (!s) |
1020 | return -ENOMEM; | |
1021 | ||
9da4cb2b | 1022 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1023 | s->io.fd = fd; |
1024 | s->io.events = events; | |
1025 | s->io.callback = callback; | |
1026 | s->userdata = userdata; | |
baf76283 | 1027 | s->enabled = SD_EVENT_ON; |
fd38203a | 1028 | |
baf76283 | 1029 | r = source_io_register(s, s->enabled, events); |
ec766a51 | 1030 | if (r < 0) |
050f74f2 | 1031 | return r; |
fd38203a | 1032 | |
a71fe8b8 LP |
1033 | if (ret) |
1034 | *ret = s; | |
ec766a51 | 1035 | TAKE_PTR(s); |
a71fe8b8 | 1036 | |
fd38203a LP |
1037 | return 0; |
1038 | } | |
1039 | ||
52444dc4 LP |
1040 | static void initialize_perturb(sd_event *e) { |
1041 | sd_id128_t bootid = {}; | |
1042 | ||
1043 | /* When we sleep for longer, we try to realign the wakeup to | |
f21f31b2 | 1044 | the same time within each minute/second/250ms, so that |
52444dc4 LP |
1045 | events all across the system can be coalesced into a single |
1046 | CPU wakeup. However, let's take some system-specific | |
1047 | randomness for this value, so that in a network of systems | |
1048 | with synced clocks timer events are distributed a | |
1049 | bit. Here, we calculate a perturbation usec offset from the | |
1050 | boot ID. */ | |
1051 | ||
3a43da28 | 1052 | if (_likely_(e->perturb != USEC_INFINITY)) |
52444dc4 LP |
1053 | return; |
1054 | ||
1055 | if (sd_id128_get_boot(&bootid) >= 0) | |
1056 | e->perturb = (bootid.qwords[0] ^ bootid.qwords[1]) % USEC_PER_MINUTE; | |
1057 | } | |
1058 | ||
fd38203a LP |
1059 | static int event_setup_timer_fd( |
1060 | sd_event *e, | |
6a0f1f6d LP |
1061 | struct clock_data *d, |
1062 | clockid_t clock) { | |
fd38203a | 1063 | |
fd38203a | 1064 | assert(e); |
6a0f1f6d | 1065 | assert(d); |
fd38203a | 1066 | |
6a0f1f6d | 1067 | if (_likely_(d->fd >= 0)) |
fd38203a LP |
1068 | return 0; |
1069 | ||
b44d87e2 | 1070 | _cleanup_close_ int fd = -1; |
b44d87e2 | 1071 | |
6a0f1f6d | 1072 | fd = timerfd_create(clock, TFD_NONBLOCK|TFD_CLOEXEC); |
fd38203a LP |
1073 | if (fd < 0) |
1074 | return -errno; | |
1075 | ||
7fe2903c LP |
1076 | fd = fd_move_above_stdio(fd); |
1077 | ||
1eac7948 | 1078 | struct epoll_event ev = { |
a82f89aa LP |
1079 | .events = EPOLLIN, |
1080 | .data.ptr = d, | |
1081 | }; | |
fd38203a | 1082 | |
15c689d7 | 1083 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) |
fd38203a | 1084 | return -errno; |
fd38203a | 1085 | |
b44d87e2 | 1086 | d->fd = TAKE_FD(fd); |
fd38203a LP |
1087 | return 0; |
1088 | } | |
1089 | ||
c4f1aff2 TG |
1090 | static int time_exit_callback(sd_event_source *s, uint64_t usec, void *userdata) { |
1091 | assert(s); | |
1092 | ||
1093 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1094 | } | |
1095 | ||
6a0f1f6d | 1096 | _public_ int sd_event_add_time( |
fd38203a | 1097 | sd_event *e, |
151b9b96 | 1098 | sd_event_source **ret, |
6a0f1f6d | 1099 | clockid_t clock, |
fd38203a | 1100 | uint64_t usec, |
c2ba3ad6 | 1101 | uint64_t accuracy, |
718db961 | 1102 | sd_event_time_handler_t callback, |
151b9b96 | 1103 | void *userdata) { |
fd38203a | 1104 | |
6a0f1f6d | 1105 | EventSourceType type; |
ec766a51 | 1106 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6a0f1f6d | 1107 | struct clock_data *d; |
fd38203a LP |
1108 | int r; |
1109 | ||
305f78bf | 1110 | assert_return(e, -EINVAL); |
b937d761 | 1111 | assert_return(e = event_resolve(e), -ENOPKG); |
305f78bf | 1112 | assert_return(accuracy != (uint64_t) -1, -EINVAL); |
da7e457c | 1113 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1114 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1115 | |
e475d10c LP |
1116 | if (!clock_supported(clock)) /* Checks whether the kernel supports the clock */ |
1117 | return -EOPNOTSUPP; | |
1118 | ||
1119 | type = clock_to_event_source_type(clock); /* checks whether sd-event supports this clock */ | |
1120 | if (type < 0) | |
3411372e LP |
1121 | return -EOPNOTSUPP; |
1122 | ||
c4f1aff2 TG |
1123 | if (!callback) |
1124 | callback = time_exit_callback; | |
1125 | ||
6a0f1f6d LP |
1126 | d = event_get_clock_data(e, type); |
1127 | assert(d); | |
c2ba3ad6 | 1128 | |
c983e776 EV |
1129 | r = prioq_ensure_allocated(&d->earliest, earliest_time_prioq_compare); |
1130 | if (r < 0) | |
1131 | return r; | |
fd38203a | 1132 | |
c983e776 EV |
1133 | r = prioq_ensure_allocated(&d->latest, latest_time_prioq_compare); |
1134 | if (r < 0) | |
1135 | return r; | |
fd38203a | 1136 | |
6a0f1f6d LP |
1137 | if (d->fd < 0) { |
1138 | r = event_setup_timer_fd(e, d, clock); | |
fd38203a LP |
1139 | if (r < 0) |
1140 | return r; | |
1141 | } | |
1142 | ||
a71fe8b8 | 1143 | s = source_new(e, !ret, type); |
fd38203a LP |
1144 | if (!s) |
1145 | return -ENOMEM; | |
1146 | ||
1147 | s->time.next = usec; | |
c2ba3ad6 | 1148 | s->time.accuracy = accuracy == 0 ? DEFAULT_ACCURACY_USEC : accuracy; |
fd38203a | 1149 | s->time.callback = callback; |
da7e457c | 1150 | s->time.earliest_index = s->time.latest_index = PRIOQ_IDX_NULL; |
fd38203a | 1151 | s->userdata = userdata; |
baf76283 | 1152 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1153 | |
e07bbb7c TG |
1154 | d->needs_rearm = true; |
1155 | ||
6a0f1f6d | 1156 | r = prioq_put(d->earliest, s, &s->time.earliest_index); |
c2ba3ad6 | 1157 | if (r < 0) |
ec766a51 | 1158 | return r; |
c2ba3ad6 | 1159 | |
6a0f1f6d | 1160 | r = prioq_put(d->latest, s, &s->time.latest_index); |
c2ba3ad6 | 1161 | if (r < 0) |
ec766a51 | 1162 | return r; |
fd38203a | 1163 | |
a71fe8b8 LP |
1164 | if (ret) |
1165 | *ret = s; | |
ec766a51 | 1166 | TAKE_PTR(s); |
a71fe8b8 | 1167 | |
fd38203a LP |
1168 | return 0; |
1169 | } | |
1170 | ||
d6a83dc4 LP |
1171 | _public_ int sd_event_add_time_relative( |
1172 | sd_event *e, | |
1173 | sd_event_source **ret, | |
1174 | clockid_t clock, | |
1175 | uint64_t usec, | |
1176 | uint64_t accuracy, | |
1177 | sd_event_time_handler_t callback, | |
1178 | void *userdata) { | |
1179 | ||
1180 | usec_t t; | |
1181 | int r; | |
1182 | ||
1183 | /* Same as sd_event_add_time() but operates relative to the event loop's current point in time, and | |
1184 | * checks for overflow. */ | |
1185 | ||
1186 | r = sd_event_now(e, clock, &t); | |
1187 | if (r < 0) | |
1188 | return r; | |
1189 | ||
1190 | if (usec >= USEC_INFINITY - t) | |
1191 | return -EOVERFLOW; | |
1192 | ||
1193 | return sd_event_add_time(e, ret, clock, t + usec, accuracy, callback, userdata); | |
1194 | } | |
1195 | ||
59bc1fd7 LP |
1196 | static int signal_exit_callback(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) { |
1197 | assert(s); | |
1198 | ||
1199 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1200 | } | |
1201 | ||
f7262a9f | 1202 | _public_ int sd_event_add_signal( |
305f78bf | 1203 | sd_event *e, |
151b9b96 | 1204 | sd_event_source **ret, |
305f78bf | 1205 | int sig, |
718db961 | 1206 | sd_event_signal_handler_t callback, |
151b9b96 | 1207 | void *userdata) { |
305f78bf | 1208 | |
ec766a51 | 1209 | _cleanup_(source_freep) sd_event_source *s = NULL; |
9da4cb2b | 1210 | struct signal_data *d; |
fd38203a LP |
1211 | int r; |
1212 | ||
305f78bf | 1213 | assert_return(e, -EINVAL); |
b937d761 | 1214 | assert_return(e = event_resolve(e), -ENOPKG); |
6eb7c172 | 1215 | assert_return(SIGNAL_VALID(sig), -EINVAL); |
da7e457c | 1216 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1217 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1218 | |
59bc1fd7 LP |
1219 | if (!callback) |
1220 | callback = signal_exit_callback; | |
1221 | ||
d1b75241 LP |
1222 | r = signal_is_blocked(sig); |
1223 | if (r < 0) | |
1224 | return r; | |
1225 | if (r == 0) | |
3022d74b LP |
1226 | return -EBUSY; |
1227 | ||
fd38203a LP |
1228 | if (!e->signal_sources) { |
1229 | e->signal_sources = new0(sd_event_source*, _NSIG); | |
1230 | if (!e->signal_sources) | |
1231 | return -ENOMEM; | |
1232 | } else if (e->signal_sources[sig]) | |
1233 | return -EBUSY; | |
1234 | ||
a71fe8b8 | 1235 | s = source_new(e, !ret, SOURCE_SIGNAL); |
fd38203a LP |
1236 | if (!s) |
1237 | return -ENOMEM; | |
1238 | ||
1239 | s->signal.sig = sig; | |
1240 | s->signal.callback = callback; | |
1241 | s->userdata = userdata; | |
baf76283 | 1242 | s->enabled = SD_EVENT_ON; |
fd38203a LP |
1243 | |
1244 | e->signal_sources[sig] = s; | |
fd38203a | 1245 | |
9da4cb2b | 1246 | r = event_make_signal_data(e, sig, &d); |
ec766a51 | 1247 | if (r < 0) |
9da4cb2b | 1248 | return r; |
fd38203a | 1249 | |
f1f00dbb LP |
1250 | /* Use the signal name as description for the event source by default */ |
1251 | (void) sd_event_source_set_description(s, signal_to_string(sig)); | |
1252 | ||
a71fe8b8 LP |
1253 | if (ret) |
1254 | *ret = s; | |
ec766a51 | 1255 | TAKE_PTR(s); |
a71fe8b8 | 1256 | |
fd38203a LP |
1257 | return 0; |
1258 | } | |
1259 | ||
b9350e70 LP |
1260 | static int child_exit_callback(sd_event_source *s, const siginfo_t *si, void *userdata) { |
1261 | assert(s); | |
1262 | ||
1263 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1264 | } | |
1265 | ||
f8f3f926 LP |
1266 | static bool shall_use_pidfd(void) { |
1267 | /* Mostly relevant for debugging, i.e. this is used in test-event.c to test the event loop once with and once without pidfd */ | |
1268 | return getenv_bool_secure("SYSTEMD_PIDFD") != 0; | |
1269 | } | |
1270 | ||
f7262a9f | 1271 | _public_ int sd_event_add_child( |
305f78bf | 1272 | sd_event *e, |
151b9b96 | 1273 | sd_event_source **ret, |
305f78bf LP |
1274 | pid_t pid, |
1275 | int options, | |
718db961 | 1276 | sd_event_child_handler_t callback, |
151b9b96 | 1277 | void *userdata) { |
305f78bf | 1278 | |
ec766a51 | 1279 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1280 | int r; |
1281 | ||
305f78bf | 1282 | assert_return(e, -EINVAL); |
b937d761 | 1283 | assert_return(e = event_resolve(e), -ENOPKG); |
305f78bf LP |
1284 | assert_return(pid > 1, -EINVAL); |
1285 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1286 | assert_return(options != 0, -EINVAL); | |
da7e457c | 1287 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1288 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1289 | |
b9350e70 LP |
1290 | if (!callback) |
1291 | callback = child_exit_callback; | |
1292 | ||
ee880b37 LP |
1293 | if (e->n_enabled_child_sources == 0) { |
1294 | /* Caller must block SIGCHLD before using us to watch children, even if pidfd is available, | |
1295 | * for compatibility with pre-pidfd and because we don't want the reap the child processes | |
1296 | * ourselves, i.e. call waitid(), and don't want Linux' default internal logic for that to | |
1297 | * take effect. | |
1298 | * | |
1299 | * (As an optimization we only do this check on the first child event source created.) */ | |
1300 | r = signal_is_blocked(SIGCHLD); | |
1301 | if (r < 0) | |
1302 | return r; | |
1303 | if (r == 0) | |
1304 | return -EBUSY; | |
1305 | } | |
1306 | ||
d5099efc | 1307 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
fd38203a LP |
1308 | if (r < 0) |
1309 | return r; | |
1310 | ||
4a0b58c4 | 1311 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) |
fd38203a LP |
1312 | return -EBUSY; |
1313 | ||
a71fe8b8 | 1314 | s = source_new(e, !ret, SOURCE_CHILD); |
fd38203a LP |
1315 | if (!s) |
1316 | return -ENOMEM; | |
1317 | ||
f8f3f926 | 1318 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1319 | s->child.pid = pid; |
1320 | s->child.options = options; | |
1321 | s->child.callback = callback; | |
1322 | s->userdata = userdata; | |
baf76283 | 1323 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1324 | |
f8f3f926 LP |
1325 | /* We always take a pidfd here if we can, even if we wait for anything else than WEXITED, so that we |
1326 | * pin the PID, and make regular waitid() handling race-free. */ | |
1327 | ||
1328 | if (shall_use_pidfd()) { | |
1329 | s->child.pidfd = pidfd_open(s->child.pid, 0); | |
1330 | if (s->child.pidfd < 0) { | |
1331 | /* Propagate errors unless the syscall is not supported or blocked */ | |
1332 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
1333 | return -errno; | |
1334 | } else | |
1335 | s->child.pidfd_owned = true; /* If we allocate the pidfd we own it by default */ | |
1336 | } else | |
1337 | s->child.pidfd = -1; | |
1338 | ||
4a0b58c4 | 1339 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); |
ec766a51 | 1340 | if (r < 0) |
fd38203a | 1341 | return r; |
fd38203a | 1342 | |
313cefa1 | 1343 | e->n_enabled_child_sources++; |
fd38203a | 1344 | |
f8f3f926 LP |
1345 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1346 | /* We have a pidfd and we only want to watch for exit */ | |
1347 | ||
1348 | r = source_child_pidfd_register(s, s->enabled); | |
1349 | if (r < 0) { | |
1350 | e->n_enabled_child_sources--; | |
1351 | return r; | |
1352 | } | |
1353 | } else { | |
1354 | /* We have no pidfd or we shall wait for some other event than WEXITED */ | |
fd38203a | 1355 | |
f8f3f926 LP |
1356 | r = event_make_signal_data(e, SIGCHLD, NULL); |
1357 | if (r < 0) { | |
1358 | e->n_enabled_child_sources--; | |
1359 | return r; | |
1360 | } | |
1361 | ||
1362 | e->need_process_child = true; | |
1363 | } | |
c2ba3ad6 | 1364 | |
a71fe8b8 LP |
1365 | if (ret) |
1366 | *ret = s; | |
f8f3f926 | 1367 | |
ec766a51 | 1368 | TAKE_PTR(s); |
f8f3f926 LP |
1369 | return 0; |
1370 | } | |
1371 | ||
1372 | _public_ int sd_event_add_child_pidfd( | |
1373 | sd_event *e, | |
1374 | sd_event_source **ret, | |
1375 | int pidfd, | |
1376 | int options, | |
1377 | sd_event_child_handler_t callback, | |
1378 | void *userdata) { | |
1379 | ||
1380 | ||
1381 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
1382 | pid_t pid; | |
1383 | int r; | |
1384 | ||
1385 | assert_return(e, -EINVAL); | |
1386 | assert_return(e = event_resolve(e), -ENOPKG); | |
1387 | assert_return(pidfd >= 0, -EBADF); | |
1388 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1389 | assert_return(options != 0, -EINVAL); | |
f8f3f926 LP |
1390 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1391 | assert_return(!event_pid_changed(e), -ECHILD); | |
1392 | ||
b9350e70 LP |
1393 | if (!callback) |
1394 | callback = child_exit_callback; | |
1395 | ||
ee880b37 LP |
1396 | if (e->n_enabled_child_sources == 0) { |
1397 | r = signal_is_blocked(SIGCHLD); | |
1398 | if (r < 0) | |
1399 | return r; | |
1400 | if (r == 0) | |
1401 | return -EBUSY; | |
1402 | } | |
1403 | ||
f8f3f926 LP |
1404 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
1405 | if (r < 0) | |
1406 | return r; | |
1407 | ||
1408 | r = pidfd_get_pid(pidfd, &pid); | |
1409 | if (r < 0) | |
1410 | return r; | |
1411 | ||
1412 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) | |
1413 | return -EBUSY; | |
1414 | ||
1415 | s = source_new(e, !ret, SOURCE_CHILD); | |
1416 | if (!s) | |
1417 | return -ENOMEM; | |
1418 | ||
1419 | s->wakeup = WAKEUP_EVENT_SOURCE; | |
1420 | s->child.pidfd = pidfd; | |
1421 | s->child.pid = pid; | |
1422 | s->child.options = options; | |
1423 | s->child.callback = callback; | |
1424 | s->child.pidfd_owned = false; /* If we got the pidfd passed in we don't own it by default (similar to the IO fd case) */ | |
1425 | s->userdata = userdata; | |
1426 | s->enabled = SD_EVENT_ONESHOT; | |
1427 | ||
1428 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); | |
1429 | if (r < 0) | |
1430 | return r; | |
1431 | ||
1432 | e->n_enabled_child_sources++; | |
1433 | ||
1434 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { | |
1435 | /* We only want to watch for WEXITED */ | |
1436 | ||
1437 | r = source_child_pidfd_register(s, s->enabled); | |
1438 | if (r < 0) { | |
1439 | e->n_enabled_child_sources--; | |
1440 | return r; | |
1441 | } | |
1442 | } else { | |
1443 | /* We shall wait for some other event than WEXITED */ | |
1444 | ||
1445 | r = event_make_signal_data(e, SIGCHLD, NULL); | |
1446 | if (r < 0) { | |
1447 | e->n_enabled_child_sources--; | |
1448 | return r; | |
1449 | } | |
a71fe8b8 | 1450 | |
f8f3f926 LP |
1451 | e->need_process_child = true; |
1452 | } | |
1453 | ||
1454 | if (ret) | |
1455 | *ret = s; | |
1456 | ||
1457 | TAKE_PTR(s); | |
fd38203a LP |
1458 | return 0; |
1459 | } | |
1460 | ||
b9350e70 LP |
1461 | static int generic_exit_callback(sd_event_source *s, void *userdata) { |
1462 | assert(s); | |
1463 | ||
1464 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1465 | } | |
1466 | ||
f7262a9f | 1467 | _public_ int sd_event_add_defer( |
305f78bf | 1468 | sd_event *e, |
151b9b96 | 1469 | sd_event_source **ret, |
718db961 | 1470 | sd_event_handler_t callback, |
151b9b96 | 1471 | void *userdata) { |
305f78bf | 1472 | |
ec766a51 | 1473 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1474 | int r; |
1475 | ||
305f78bf | 1476 | assert_return(e, -EINVAL); |
b937d761 | 1477 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1478 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1479 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1480 | |
b9350e70 LP |
1481 | if (!callback) |
1482 | callback = generic_exit_callback; | |
1483 | ||
a71fe8b8 | 1484 | s = source_new(e, !ret, SOURCE_DEFER); |
fd38203a LP |
1485 | if (!s) |
1486 | return -ENOMEM; | |
1487 | ||
1488 | s->defer.callback = callback; | |
1489 | s->userdata = userdata; | |
baf76283 | 1490 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a LP |
1491 | |
1492 | r = source_set_pending(s, true); | |
ec766a51 | 1493 | if (r < 0) |
fd38203a | 1494 | return r; |
fd38203a | 1495 | |
a71fe8b8 LP |
1496 | if (ret) |
1497 | *ret = s; | |
ec766a51 | 1498 | TAKE_PTR(s); |
a71fe8b8 | 1499 | |
fd38203a LP |
1500 | return 0; |
1501 | } | |
1502 | ||
6e9feda3 LP |
1503 | _public_ int sd_event_add_post( |
1504 | sd_event *e, | |
1505 | sd_event_source **ret, | |
1506 | sd_event_handler_t callback, | |
1507 | void *userdata) { | |
1508 | ||
ec766a51 | 1509 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6e9feda3 LP |
1510 | int r; |
1511 | ||
1512 | assert_return(e, -EINVAL); | |
b937d761 | 1513 | assert_return(e = event_resolve(e), -ENOPKG); |
6e9feda3 LP |
1514 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1515 | assert_return(!event_pid_changed(e), -ECHILD); | |
1516 | ||
b9350e70 LP |
1517 | if (!callback) |
1518 | callback = generic_exit_callback; | |
1519 | ||
a71fe8b8 | 1520 | s = source_new(e, !ret, SOURCE_POST); |
6e9feda3 LP |
1521 | if (!s) |
1522 | return -ENOMEM; | |
1523 | ||
1524 | s->post.callback = callback; | |
1525 | s->userdata = userdata; | |
1526 | s->enabled = SD_EVENT_ON; | |
1527 | ||
de7fef4b | 1528 | r = set_ensure_put(&e->post_sources, NULL, s); |
ec766a51 | 1529 | if (r < 0) |
6e9feda3 | 1530 | return r; |
de7fef4b | 1531 | assert(r > 0); |
6e9feda3 | 1532 | |
a71fe8b8 LP |
1533 | if (ret) |
1534 | *ret = s; | |
ec766a51 | 1535 | TAKE_PTR(s); |
a71fe8b8 | 1536 | |
6e9feda3 LP |
1537 | return 0; |
1538 | } | |
1539 | ||
6203e07a | 1540 | _public_ int sd_event_add_exit( |
305f78bf | 1541 | sd_event *e, |
151b9b96 | 1542 | sd_event_source **ret, |
718db961 | 1543 | sd_event_handler_t callback, |
151b9b96 | 1544 | void *userdata) { |
305f78bf | 1545 | |
ec766a51 | 1546 | _cleanup_(source_freep) sd_event_source *s = NULL; |
da7e457c LP |
1547 | int r; |
1548 | ||
1549 | assert_return(e, -EINVAL); | |
b937d761 | 1550 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
1551 | assert_return(callback, -EINVAL); |
1552 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
1553 | assert_return(!event_pid_changed(e), -ECHILD); | |
1554 | ||
c983e776 EV |
1555 | r = prioq_ensure_allocated(&e->exit, exit_prioq_compare); |
1556 | if (r < 0) | |
1557 | return r; | |
da7e457c | 1558 | |
a71fe8b8 | 1559 | s = source_new(e, !ret, SOURCE_EXIT); |
fd38203a | 1560 | if (!s) |
da7e457c | 1561 | return -ENOMEM; |
fd38203a | 1562 | |
6203e07a | 1563 | s->exit.callback = callback; |
da7e457c | 1564 | s->userdata = userdata; |
6203e07a | 1565 | s->exit.prioq_index = PRIOQ_IDX_NULL; |
baf76283 | 1566 | s->enabled = SD_EVENT_ONESHOT; |
da7e457c | 1567 | |
6203e07a | 1568 | r = prioq_put(s->event->exit, s, &s->exit.prioq_index); |
ec766a51 | 1569 | if (r < 0) |
da7e457c | 1570 | return r; |
da7e457c | 1571 | |
a71fe8b8 LP |
1572 | if (ret) |
1573 | *ret = s; | |
ec766a51 | 1574 | TAKE_PTR(s); |
a71fe8b8 | 1575 | |
da7e457c LP |
1576 | return 0; |
1577 | } | |
1578 | ||
97ef5391 LP |
1579 | static void event_free_inotify_data(sd_event *e, struct inotify_data *d) { |
1580 | assert(e); | |
1581 | ||
1582 | if (!d) | |
1583 | return; | |
1584 | ||
1585 | assert(hashmap_isempty(d->inodes)); | |
1586 | assert(hashmap_isempty(d->wd)); | |
1587 | ||
1588 | if (d->buffer_filled > 0) | |
1589 | LIST_REMOVE(buffered, e->inotify_data_buffered, d); | |
1590 | ||
1591 | hashmap_free(d->inodes); | |
1592 | hashmap_free(d->wd); | |
1593 | ||
1594 | assert_se(hashmap_remove(e->inotify_data, &d->priority) == d); | |
1595 | ||
1596 | if (d->fd >= 0) { | |
1597 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, d->fd, NULL) < 0) | |
1598 | log_debug_errno(errno, "Failed to remove inotify fd from epoll, ignoring: %m"); | |
1599 | ||
1600 | safe_close(d->fd); | |
1601 | } | |
1602 | free(d); | |
1603 | } | |
1604 | ||
1605 | static int event_make_inotify_data( | |
1606 | sd_event *e, | |
1607 | int64_t priority, | |
1608 | struct inotify_data **ret) { | |
1609 | ||
1610 | _cleanup_close_ int fd = -1; | |
1611 | struct inotify_data *d; | |
97ef5391 LP |
1612 | int r; |
1613 | ||
1614 | assert(e); | |
1615 | ||
1616 | d = hashmap_get(e->inotify_data, &priority); | |
1617 | if (d) { | |
1618 | if (ret) | |
1619 | *ret = d; | |
1620 | return 0; | |
1621 | } | |
1622 | ||
1623 | fd = inotify_init1(IN_NONBLOCK|O_CLOEXEC); | |
1624 | if (fd < 0) | |
1625 | return -errno; | |
1626 | ||
1627 | fd = fd_move_above_stdio(fd); | |
1628 | ||
1629 | r = hashmap_ensure_allocated(&e->inotify_data, &uint64_hash_ops); | |
1630 | if (r < 0) | |
1631 | return r; | |
1632 | ||
1633 | d = new(struct inotify_data, 1); | |
1634 | if (!d) | |
1635 | return -ENOMEM; | |
1636 | ||
1637 | *d = (struct inotify_data) { | |
1638 | .wakeup = WAKEUP_INOTIFY_DATA, | |
1639 | .fd = TAKE_FD(fd), | |
1640 | .priority = priority, | |
1641 | }; | |
1642 | ||
1643 | r = hashmap_put(e->inotify_data, &d->priority, d); | |
1644 | if (r < 0) { | |
1645 | d->fd = safe_close(d->fd); | |
1646 | free(d); | |
1647 | return r; | |
1648 | } | |
1649 | ||
1eac7948 | 1650 | struct epoll_event ev = { |
97ef5391 LP |
1651 | .events = EPOLLIN, |
1652 | .data.ptr = d, | |
1653 | }; | |
1654 | ||
1655 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { | |
1656 | r = -errno; | |
1657 | d->fd = safe_close(d->fd); /* let's close this ourselves, as event_free_inotify_data() would otherwise | |
1658 | * remove the fd from the epoll first, which we don't want as we couldn't | |
1659 | * add it in the first place. */ | |
1660 | event_free_inotify_data(e, d); | |
1661 | return r; | |
1662 | } | |
1663 | ||
1664 | if (ret) | |
1665 | *ret = d; | |
1666 | ||
1667 | return 1; | |
1668 | } | |
1669 | ||
7a08d314 | 1670 | static int inode_data_compare(const struct inode_data *x, const struct inode_data *y) { |
90c88092 | 1671 | int r; |
97ef5391 LP |
1672 | |
1673 | assert(x); | |
1674 | assert(y); | |
1675 | ||
90c88092 YW |
1676 | r = CMP(x->dev, y->dev); |
1677 | if (r != 0) | |
1678 | return r; | |
97ef5391 | 1679 | |
6dd91b36 | 1680 | return CMP(x->ino, y->ino); |
97ef5391 LP |
1681 | } |
1682 | ||
7a08d314 YW |
1683 | static void inode_data_hash_func(const struct inode_data *d, struct siphash *state) { |
1684 | assert(d); | |
97ef5391 LP |
1685 | |
1686 | siphash24_compress(&d->dev, sizeof(d->dev), state); | |
1687 | siphash24_compress(&d->ino, sizeof(d->ino), state); | |
1688 | } | |
1689 | ||
7a08d314 | 1690 | DEFINE_PRIVATE_HASH_OPS(inode_data_hash_ops, struct inode_data, inode_data_hash_func, inode_data_compare); |
97ef5391 LP |
1691 | |
1692 | static void event_free_inode_data( | |
1693 | sd_event *e, | |
1694 | struct inode_data *d) { | |
1695 | ||
1696 | assert(e); | |
1697 | ||
1698 | if (!d) | |
1699 | return; | |
1700 | ||
1701 | assert(!d->event_sources); | |
1702 | ||
1703 | if (d->fd >= 0) { | |
1704 | LIST_REMOVE(to_close, e->inode_data_to_close, d); | |
1705 | safe_close(d->fd); | |
1706 | } | |
1707 | ||
1708 | if (d->inotify_data) { | |
1709 | ||
1710 | if (d->wd >= 0) { | |
1711 | if (d->inotify_data->fd >= 0) { | |
1712 | /* So here's a problem. At the time this runs the watch descriptor might already be | |
1713 | * invalidated, because an IN_IGNORED event might be queued right the moment we enter | |
1714 | * the syscall. Hence, whenever we get EINVAL, ignore it entirely, since it's a very | |
1715 | * likely case to happen. */ | |
1716 | ||
1717 | if (inotify_rm_watch(d->inotify_data->fd, d->wd) < 0 && errno != EINVAL) | |
1718 | log_debug_errno(errno, "Failed to remove watch descriptor %i from inotify, ignoring: %m", d->wd); | |
1719 | } | |
1720 | ||
1721 | assert_se(hashmap_remove(d->inotify_data->wd, INT_TO_PTR(d->wd)) == d); | |
1722 | } | |
1723 | ||
1724 | assert_se(hashmap_remove(d->inotify_data->inodes, d) == d); | |
1725 | } | |
1726 | ||
1727 | free(d); | |
1728 | } | |
1729 | ||
1730 | static void event_gc_inode_data( | |
1731 | sd_event *e, | |
1732 | struct inode_data *d) { | |
1733 | ||
1734 | struct inotify_data *inotify_data; | |
1735 | ||
1736 | assert(e); | |
1737 | ||
1738 | if (!d) | |
1739 | return; | |
1740 | ||
1741 | if (d->event_sources) | |
1742 | return; | |
1743 | ||
1744 | inotify_data = d->inotify_data; | |
1745 | event_free_inode_data(e, d); | |
1746 | ||
1747 | if (inotify_data && hashmap_isempty(inotify_data->inodes)) | |
1748 | event_free_inotify_data(e, inotify_data); | |
1749 | } | |
1750 | ||
1751 | static int event_make_inode_data( | |
1752 | sd_event *e, | |
1753 | struct inotify_data *inotify_data, | |
1754 | dev_t dev, | |
1755 | ino_t ino, | |
1756 | struct inode_data **ret) { | |
1757 | ||
1758 | struct inode_data *d, key; | |
1759 | int r; | |
1760 | ||
1761 | assert(e); | |
1762 | assert(inotify_data); | |
1763 | ||
1764 | key = (struct inode_data) { | |
1765 | .ino = ino, | |
1766 | .dev = dev, | |
1767 | }; | |
1768 | ||
1769 | d = hashmap_get(inotify_data->inodes, &key); | |
1770 | if (d) { | |
1771 | if (ret) | |
1772 | *ret = d; | |
1773 | ||
1774 | return 0; | |
1775 | } | |
1776 | ||
1777 | r = hashmap_ensure_allocated(&inotify_data->inodes, &inode_data_hash_ops); | |
1778 | if (r < 0) | |
1779 | return r; | |
1780 | ||
1781 | d = new(struct inode_data, 1); | |
1782 | if (!d) | |
1783 | return -ENOMEM; | |
1784 | ||
1785 | *d = (struct inode_data) { | |
1786 | .dev = dev, | |
1787 | .ino = ino, | |
1788 | .wd = -1, | |
1789 | .fd = -1, | |
1790 | .inotify_data = inotify_data, | |
1791 | }; | |
1792 | ||
1793 | r = hashmap_put(inotify_data->inodes, d, d); | |
1794 | if (r < 0) { | |
1795 | free(d); | |
1796 | return r; | |
1797 | } | |
1798 | ||
1799 | if (ret) | |
1800 | *ret = d; | |
1801 | ||
1802 | return 1; | |
1803 | } | |
1804 | ||
1805 | static uint32_t inode_data_determine_mask(struct inode_data *d) { | |
1806 | bool excl_unlink = true; | |
1807 | uint32_t combined = 0; | |
1808 | sd_event_source *s; | |
1809 | ||
1810 | assert(d); | |
1811 | ||
1812 | /* Combines the watch masks of all event sources watching this inode. We generally just OR them together, but | |
1813 | * the IN_EXCL_UNLINK flag is ANDed instead. | |
1814 | * | |
1815 | * Note that we add all sources to the mask here, regardless whether enabled, disabled or oneshot. That's | |
1816 | * because we cannot change the mask anymore after the event source was created once, since the kernel has no | |
f21f31b2 | 1817 | * API for that. Hence we need to subscribe to the maximum mask we ever might be interested in, and suppress |
97ef5391 LP |
1818 | * events we don't care for client-side. */ |
1819 | ||
1820 | LIST_FOREACH(inotify.by_inode_data, s, d->event_sources) { | |
1821 | ||
1822 | if ((s->inotify.mask & IN_EXCL_UNLINK) == 0) | |
1823 | excl_unlink = false; | |
1824 | ||
1825 | combined |= s->inotify.mask; | |
1826 | } | |
1827 | ||
1828 | return (combined & ~(IN_ONESHOT|IN_DONT_FOLLOW|IN_ONLYDIR|IN_EXCL_UNLINK)) | (excl_unlink ? IN_EXCL_UNLINK : 0); | |
1829 | } | |
1830 | ||
1831 | static int inode_data_realize_watch(sd_event *e, struct inode_data *d) { | |
1832 | uint32_t combined_mask; | |
1833 | int wd, r; | |
1834 | ||
1835 | assert(d); | |
1836 | assert(d->fd >= 0); | |
1837 | ||
1838 | combined_mask = inode_data_determine_mask(d); | |
1839 | ||
1840 | if (d->wd >= 0 && combined_mask == d->combined_mask) | |
1841 | return 0; | |
1842 | ||
1843 | r = hashmap_ensure_allocated(&d->inotify_data->wd, NULL); | |
1844 | if (r < 0) | |
1845 | return r; | |
1846 | ||
1847 | wd = inotify_add_watch_fd(d->inotify_data->fd, d->fd, combined_mask); | |
1848 | if (wd < 0) | |
1849 | return -errno; | |
1850 | ||
1851 | if (d->wd < 0) { | |
1852 | r = hashmap_put(d->inotify_data->wd, INT_TO_PTR(wd), d); | |
1853 | if (r < 0) { | |
1854 | (void) inotify_rm_watch(d->inotify_data->fd, wd); | |
1855 | return r; | |
1856 | } | |
1857 | ||
1858 | d->wd = wd; | |
1859 | ||
1860 | } else if (d->wd != wd) { | |
1861 | ||
1862 | log_debug("Weird, the watch descriptor we already knew for this inode changed?"); | |
1863 | (void) inotify_rm_watch(d->fd, wd); | |
1864 | return -EINVAL; | |
1865 | } | |
1866 | ||
1867 | d->combined_mask = combined_mask; | |
1868 | return 1; | |
1869 | } | |
1870 | ||
b9350e70 LP |
1871 | static int inotify_exit_callback(sd_event_source *s, const struct inotify_event *event, void *userdata) { |
1872 | assert(s); | |
1873 | ||
1874 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1875 | } | |
1876 | ||
97ef5391 LP |
1877 | _public_ int sd_event_add_inotify( |
1878 | sd_event *e, | |
1879 | sd_event_source **ret, | |
1880 | const char *path, | |
1881 | uint32_t mask, | |
1882 | sd_event_inotify_handler_t callback, | |
1883 | void *userdata) { | |
1884 | ||
97ef5391 LP |
1885 | struct inotify_data *inotify_data = NULL; |
1886 | struct inode_data *inode_data = NULL; | |
1887 | _cleanup_close_ int fd = -1; | |
8c75fe17 | 1888 | _cleanup_(source_freep) sd_event_source *s = NULL; |
97ef5391 LP |
1889 | struct stat st; |
1890 | int r; | |
1891 | ||
1892 | assert_return(e, -EINVAL); | |
1893 | assert_return(e = event_resolve(e), -ENOPKG); | |
1894 | assert_return(path, -EINVAL); | |
97ef5391 LP |
1895 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1896 | assert_return(!event_pid_changed(e), -ECHILD); | |
1897 | ||
b9350e70 LP |
1898 | if (!callback) |
1899 | callback = inotify_exit_callback; | |
1900 | ||
97ef5391 LP |
1901 | /* Refuse IN_MASK_ADD since we coalesce watches on the same inode, and hence really don't want to merge |
1902 | * masks. Or in other words, this whole code exists only to manage IN_MASK_ADD type operations for you, hence | |
1903 | * the user can't use them for us. */ | |
1904 | if (mask & IN_MASK_ADD) | |
1905 | return -EINVAL; | |
1906 | ||
1907 | fd = open(path, O_PATH|O_CLOEXEC| | |
1908 | (mask & IN_ONLYDIR ? O_DIRECTORY : 0)| | |
1909 | (mask & IN_DONT_FOLLOW ? O_NOFOLLOW : 0)); | |
1910 | if (fd < 0) | |
1911 | return -errno; | |
1912 | ||
1913 | if (fstat(fd, &st) < 0) | |
1914 | return -errno; | |
1915 | ||
1916 | s = source_new(e, !ret, SOURCE_INOTIFY); | |
1917 | if (!s) | |
1918 | return -ENOMEM; | |
1919 | ||
1920 | s->enabled = mask & IN_ONESHOT ? SD_EVENT_ONESHOT : SD_EVENT_ON; | |
1921 | s->inotify.mask = mask; | |
1922 | s->inotify.callback = callback; | |
1923 | s->userdata = userdata; | |
1924 | ||
1925 | /* Allocate an inotify object for this priority, and an inode object within it */ | |
1926 | r = event_make_inotify_data(e, SD_EVENT_PRIORITY_NORMAL, &inotify_data); | |
1927 | if (r < 0) | |
8c75fe17 | 1928 | return r; |
97ef5391 LP |
1929 | |
1930 | r = event_make_inode_data(e, inotify_data, st.st_dev, st.st_ino, &inode_data); | |
8c75fe17 ZJS |
1931 | if (r < 0) { |
1932 | event_free_inotify_data(e, inotify_data); | |
1933 | return r; | |
1934 | } | |
97ef5391 LP |
1935 | |
1936 | /* Keep the O_PATH fd around until the first iteration of the loop, so that we can still change the priority of | |
1937 | * the event source, until then, for which we need the original inode. */ | |
1938 | if (inode_data->fd < 0) { | |
1939 | inode_data->fd = TAKE_FD(fd); | |
1940 | LIST_PREPEND(to_close, e->inode_data_to_close, inode_data); | |
1941 | } | |
1942 | ||
1943 | /* Link our event source to the inode data object */ | |
1944 | LIST_PREPEND(inotify.by_inode_data, inode_data->event_sources, s); | |
1945 | s->inotify.inode_data = inode_data; | |
1946 | ||
97ef5391 LP |
1947 | /* Actually realize the watch now */ |
1948 | r = inode_data_realize_watch(e, inode_data); | |
1949 | if (r < 0) | |
8c75fe17 | 1950 | return r; |
97ef5391 LP |
1951 | |
1952 | (void) sd_event_source_set_description(s, path); | |
1953 | ||
1954 | if (ret) | |
1955 | *ret = s; | |
8c75fe17 | 1956 | TAKE_PTR(s); |
97ef5391 LP |
1957 | |
1958 | return 0; | |
97ef5391 LP |
1959 | } |
1960 | ||
8301aa0b | 1961 | static sd_event_source* event_source_free(sd_event_source *s) { |
6680dd6b LP |
1962 | if (!s) |
1963 | return NULL; | |
da7e457c | 1964 | |
8301aa0b YW |
1965 | /* Here's a special hack: when we are called from a |
1966 | * dispatch handler we won't free the event source | |
1967 | * immediately, but we will detach the fd from the | |
1968 | * epoll. This way it is safe for the caller to unref | |
1969 | * the event source and immediately close the fd, but | |
1970 | * we still retain a valid event source object after | |
1971 | * the callback. */ | |
fd38203a | 1972 | |
8301aa0b YW |
1973 | if (s->dispatching) { |
1974 | if (s->type == SOURCE_IO) | |
1975 | source_io_unregister(s); | |
fd38203a | 1976 | |
8301aa0b YW |
1977 | source_disconnect(s); |
1978 | } else | |
1979 | source_free(s); | |
fd38203a LP |
1980 | |
1981 | return NULL; | |
1982 | } | |
1983 | ||
8301aa0b YW |
1984 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event_source, sd_event_source, event_source_free); |
1985 | ||
356779df | 1986 | _public_ int sd_event_source_set_description(sd_event_source *s, const char *description) { |
f7f53e9e | 1987 | assert_return(s, -EINVAL); |
f4b2933e | 1988 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 1989 | |
356779df | 1990 | return free_and_strdup(&s->description, description); |
f7f53e9e TG |
1991 | } |
1992 | ||
356779df | 1993 | _public_ int sd_event_source_get_description(sd_event_source *s, const char **description) { |
f7f53e9e | 1994 | assert_return(s, -EINVAL); |
356779df | 1995 | assert_return(description, -EINVAL); |
f4b2933e | 1996 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 1997 | |
7d92a1a4 ZJS |
1998 | if (!s->description) |
1999 | return -ENXIO; | |
2000 | ||
356779df | 2001 | *description = s->description; |
f7f53e9e TG |
2002 | return 0; |
2003 | } | |
2004 | ||
adcc4ca3 | 2005 | _public_ sd_event *sd_event_source_get_event(sd_event_source *s) { |
305f78bf | 2006 | assert_return(s, NULL); |
eaa3cbef LP |
2007 | |
2008 | return s->event; | |
2009 | } | |
2010 | ||
f7262a9f | 2011 | _public_ int sd_event_source_get_pending(sd_event_source *s) { |
305f78bf | 2012 | assert_return(s, -EINVAL); |
6203e07a | 2013 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c | 2014 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2015 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2016 | |
2017 | return s->pending; | |
2018 | } | |
2019 | ||
f7262a9f | 2020 | _public_ int sd_event_source_get_io_fd(sd_event_source *s) { |
305f78bf LP |
2021 | assert_return(s, -EINVAL); |
2022 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2023 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2024 | |
2025 | return s->io.fd; | |
2026 | } | |
2027 | ||
30caf8f3 LP |
2028 | _public_ int sd_event_source_set_io_fd(sd_event_source *s, int fd) { |
2029 | int r; | |
2030 | ||
2031 | assert_return(s, -EINVAL); | |
8ac43fee | 2032 | assert_return(fd >= 0, -EBADF); |
30caf8f3 LP |
2033 | assert_return(s->type == SOURCE_IO, -EDOM); |
2034 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2035 | ||
2036 | if (s->io.fd == fd) | |
2037 | return 0; | |
2038 | ||
2039 | if (s->enabled == SD_EVENT_OFF) { | |
2040 | s->io.fd = fd; | |
2041 | s->io.registered = false; | |
2042 | } else { | |
2043 | int saved_fd; | |
2044 | ||
2045 | saved_fd = s->io.fd; | |
2046 | assert(s->io.registered); | |
2047 | ||
2048 | s->io.fd = fd; | |
2049 | s->io.registered = false; | |
2050 | ||
2051 | r = source_io_register(s, s->enabled, s->io.events); | |
2052 | if (r < 0) { | |
2053 | s->io.fd = saved_fd; | |
2054 | s->io.registered = true; | |
2055 | return r; | |
2056 | } | |
2057 | ||
5a795bff | 2058 | (void) epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, saved_fd, NULL); |
30caf8f3 LP |
2059 | } |
2060 | ||
2061 | return 0; | |
2062 | } | |
2063 | ||
ab93297c NM |
2064 | _public_ int sd_event_source_get_io_fd_own(sd_event_source *s) { |
2065 | assert_return(s, -EINVAL); | |
2066 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2067 | ||
2068 | return s->io.owned; | |
2069 | } | |
2070 | ||
2071 | _public_ int sd_event_source_set_io_fd_own(sd_event_source *s, int own) { | |
2072 | assert_return(s, -EINVAL); | |
2073 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2074 | ||
2075 | s->io.owned = own; | |
2076 | return 0; | |
2077 | } | |
2078 | ||
f7262a9f | 2079 | _public_ int sd_event_source_get_io_events(sd_event_source *s, uint32_t* events) { |
305f78bf LP |
2080 | assert_return(s, -EINVAL); |
2081 | assert_return(events, -EINVAL); | |
2082 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2083 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2084 | |
2085 | *events = s->io.events; | |
2086 | return 0; | |
2087 | } | |
2088 | ||
f7262a9f | 2089 | _public_ int sd_event_source_set_io_events(sd_event_source *s, uint32_t events) { |
fd38203a LP |
2090 | int r; |
2091 | ||
305f78bf LP |
2092 | assert_return(s, -EINVAL); |
2093 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2a16a986 | 2094 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 2095 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2096 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2097 | |
b63c8d4f DH |
2098 | /* edge-triggered updates are never skipped, so we can reset edges */ |
2099 | if (s->io.events == events && !(events & EPOLLET)) | |
fd38203a LP |
2100 | return 0; |
2101 | ||
2a0dc6cd LP |
2102 | r = source_set_pending(s, false); |
2103 | if (r < 0) | |
2104 | return r; | |
2105 | ||
baf76283 | 2106 | if (s->enabled != SD_EVENT_OFF) { |
e4715127 | 2107 | r = source_io_register(s, s->enabled, events); |
fd38203a LP |
2108 | if (r < 0) |
2109 | return r; | |
2110 | } | |
2111 | ||
2112 | s->io.events = events; | |
2113 | ||
2114 | return 0; | |
2115 | } | |
2116 | ||
f7262a9f | 2117 | _public_ int sd_event_source_get_io_revents(sd_event_source *s, uint32_t* revents) { |
305f78bf LP |
2118 | assert_return(s, -EINVAL); |
2119 | assert_return(revents, -EINVAL); | |
2120 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2121 | assert_return(s->pending, -ENODATA); | |
2122 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2123 | |
2124 | *revents = s->io.revents; | |
2125 | return 0; | |
2126 | } | |
2127 | ||
f7262a9f | 2128 | _public_ int sd_event_source_get_signal(sd_event_source *s) { |
305f78bf LP |
2129 | assert_return(s, -EINVAL); |
2130 | assert_return(s->type == SOURCE_SIGNAL, -EDOM); | |
2131 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2132 | |
2133 | return s->signal.sig; | |
2134 | } | |
2135 | ||
31927c16 | 2136 | _public_ int sd_event_source_get_priority(sd_event_source *s, int64_t *priority) { |
305f78bf LP |
2137 | assert_return(s, -EINVAL); |
2138 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a | 2139 | |
6680b8d1 ME |
2140 | *priority = s->priority; |
2141 | return 0; | |
fd38203a LP |
2142 | } |
2143 | ||
31927c16 | 2144 | _public_ int sd_event_source_set_priority(sd_event_source *s, int64_t priority) { |
97ef5391 LP |
2145 | bool rm_inotify = false, rm_inode = false; |
2146 | struct inotify_data *new_inotify_data = NULL; | |
2147 | struct inode_data *new_inode_data = NULL; | |
9da4cb2b LP |
2148 | int r; |
2149 | ||
305f78bf | 2150 | assert_return(s, -EINVAL); |
da7e457c | 2151 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2152 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2153 | |
2154 | if (s->priority == priority) | |
2155 | return 0; | |
2156 | ||
97ef5391 LP |
2157 | if (s->type == SOURCE_INOTIFY) { |
2158 | struct inode_data *old_inode_data; | |
2159 | ||
2160 | assert(s->inotify.inode_data); | |
2161 | old_inode_data = s->inotify.inode_data; | |
2162 | ||
2163 | /* We need the original fd to change the priority. If we don't have it we can't change the priority, | |
2164 | * anymore. Note that we close any fds when entering the next event loop iteration, i.e. for inotify | |
2165 | * events we allow priority changes only until the first following iteration. */ | |
2166 | if (old_inode_data->fd < 0) | |
2167 | return -EOPNOTSUPP; | |
2168 | ||
2169 | r = event_make_inotify_data(s->event, priority, &new_inotify_data); | |
2170 | if (r < 0) | |
2171 | return r; | |
2172 | rm_inotify = r > 0; | |
2173 | ||
2174 | r = event_make_inode_data(s->event, new_inotify_data, old_inode_data->dev, old_inode_data->ino, &new_inode_data); | |
2175 | if (r < 0) | |
2176 | goto fail; | |
2177 | rm_inode = r > 0; | |
2178 | ||
2179 | if (new_inode_data->fd < 0) { | |
2180 | /* Duplicate the fd for the new inode object if we don't have any yet */ | |
2181 | new_inode_data->fd = fcntl(old_inode_data->fd, F_DUPFD_CLOEXEC, 3); | |
2182 | if (new_inode_data->fd < 0) { | |
2183 | r = -errno; | |
2184 | goto fail; | |
2185 | } | |
2186 | ||
2187 | LIST_PREPEND(to_close, s->event->inode_data_to_close, new_inode_data); | |
2188 | } | |
2189 | ||
2190 | /* Move the event source to the new inode data structure */ | |
2191 | LIST_REMOVE(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2192 | LIST_PREPEND(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2193 | s->inotify.inode_data = new_inode_data; | |
2194 | ||
2195 | /* Now create the new watch */ | |
2196 | r = inode_data_realize_watch(s->event, new_inode_data); | |
2197 | if (r < 0) { | |
2198 | /* Move it back */ | |
2199 | LIST_REMOVE(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2200 | LIST_PREPEND(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2201 | s->inotify.inode_data = old_inode_data; | |
2202 | goto fail; | |
2203 | } | |
2204 | ||
2205 | s->priority = priority; | |
2206 | ||
2207 | event_gc_inode_data(s->event, old_inode_data); | |
2208 | ||
2209 | } else if (s->type == SOURCE_SIGNAL && s->enabled != SD_EVENT_OFF) { | |
9da4cb2b LP |
2210 | struct signal_data *old, *d; |
2211 | ||
2212 | /* Move us from the signalfd belonging to the old | |
2213 | * priority to the signalfd of the new priority */ | |
2214 | ||
2215 | assert_se(old = hashmap_get(s->event->signal_data, &s->priority)); | |
2216 | ||
2217 | s->priority = priority; | |
2218 | ||
2219 | r = event_make_signal_data(s->event, s->signal.sig, &d); | |
2220 | if (r < 0) { | |
2221 | s->priority = old->priority; | |
2222 | return r; | |
2223 | } | |
2224 | ||
2225 | event_unmask_signal_data(s->event, old, s->signal.sig); | |
2226 | } else | |
2227 | s->priority = priority; | |
fd38203a | 2228 | |
e1951c16 | 2229 | event_source_pp_prioq_reshuffle(s); |
fd38203a | 2230 | |
6203e07a LP |
2231 | if (s->type == SOURCE_EXIT) |
2232 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
305f78bf | 2233 | |
fd38203a | 2234 | return 0; |
97ef5391 LP |
2235 | |
2236 | fail: | |
2237 | if (rm_inode) | |
2238 | event_free_inode_data(s->event, new_inode_data); | |
2239 | ||
2240 | if (rm_inotify) | |
2241 | event_free_inotify_data(s->event, new_inotify_data); | |
2242 | ||
2243 | return r; | |
fd38203a LP |
2244 | } |
2245 | ||
f7262a9f | 2246 | _public_ int sd_event_source_get_enabled(sd_event_source *s, int *m) { |
305f78bf | 2247 | assert_return(s, -EINVAL); |
305f78bf | 2248 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2249 | |
08c1eb0e ZJS |
2250 | if (m) |
2251 | *m = s->enabled; | |
2252 | return s->enabled != SD_EVENT_OFF; | |
fd38203a LP |
2253 | } |
2254 | ||
ddfde737 | 2255 | static int event_source_disable(sd_event_source *s) { |
fd38203a LP |
2256 | int r; |
2257 | ||
ddfde737 LP |
2258 | assert(s); |
2259 | assert(s->enabled != SD_EVENT_OFF); | |
fd38203a | 2260 | |
ddfde737 LP |
2261 | /* Unset the pending flag when this event source is disabled */ |
2262 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
2263 | r = source_set_pending(s, false); | |
2264 | if (r < 0) | |
2265 | return r; | |
2266 | } | |
cc567911 | 2267 | |
ddfde737 | 2268 | s->enabled = SD_EVENT_OFF; |
fd38203a | 2269 | |
ddfde737 | 2270 | switch (s->type) { |
fd38203a | 2271 | |
ddfde737 LP |
2272 | case SOURCE_IO: |
2273 | source_io_unregister(s); | |
2274 | break; | |
ac989a78 | 2275 | |
ddfde737 LP |
2276 | case SOURCE_TIME_REALTIME: |
2277 | case SOURCE_TIME_BOOTTIME: | |
2278 | case SOURCE_TIME_MONOTONIC: | |
2279 | case SOURCE_TIME_REALTIME_ALARM: | |
2280 | case SOURCE_TIME_BOOTTIME_ALARM: | |
2281 | event_source_time_prioq_reshuffle(s); | |
2282 | break; | |
fd38203a | 2283 | |
ddfde737 LP |
2284 | case SOURCE_SIGNAL: |
2285 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); | |
2286 | break; | |
fd38203a | 2287 | |
ddfde737 LP |
2288 | case SOURCE_CHILD: |
2289 | assert(s->event->n_enabled_child_sources > 0); | |
2290 | s->event->n_enabled_child_sources--; | |
fd38203a | 2291 | |
ddfde737 LP |
2292 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
2293 | source_child_pidfd_unregister(s); | |
2294 | else | |
2295 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
2296 | break; | |
4807d2d0 | 2297 | |
ddfde737 LP |
2298 | case SOURCE_EXIT: |
2299 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
2300 | break; | |
fd38203a | 2301 | |
ddfde737 LP |
2302 | case SOURCE_DEFER: |
2303 | case SOURCE_POST: | |
2304 | case SOURCE_INOTIFY: | |
2305 | break; | |
fd38203a | 2306 | |
ddfde737 LP |
2307 | default: |
2308 | assert_not_reached("Wut? I shouldn't exist."); | |
2309 | } | |
fd38203a | 2310 | |
ddfde737 LP |
2311 | return 0; |
2312 | } | |
f8f3f926 | 2313 | |
ddfde737 LP |
2314 | static int event_source_enable(sd_event_source *s, int m) { |
2315 | int r; | |
fd38203a | 2316 | |
ddfde737 LP |
2317 | assert(s); |
2318 | assert(IN_SET(m, SD_EVENT_ON, SD_EVENT_ONESHOT)); | |
2319 | assert(s->enabled == SD_EVENT_OFF); | |
305f78bf | 2320 | |
ddfde737 LP |
2321 | /* Unset the pending flag when this event source is enabled */ |
2322 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
2323 | r = source_set_pending(s, false); | |
2324 | if (r < 0) | |
2325 | return r; | |
2326 | } | |
9d3e3aa5 | 2327 | |
ddfde737 | 2328 | s->enabled = m; |
fd38203a | 2329 | |
ddfde737 | 2330 | switch (s->type) { |
ac989a78 | 2331 | |
ddfde737 LP |
2332 | case SOURCE_IO: |
2333 | r = source_io_register(s, m, s->io.events); | |
2334 | if (r < 0) { | |
2335 | s->enabled = SD_EVENT_OFF; | |
2336 | return r; | |
ac989a78 LP |
2337 | } |
2338 | ||
ddfde737 | 2339 | break; |
fd38203a | 2340 | |
ddfde737 LP |
2341 | case SOURCE_TIME_REALTIME: |
2342 | case SOURCE_TIME_BOOTTIME: | |
2343 | case SOURCE_TIME_MONOTONIC: | |
2344 | case SOURCE_TIME_REALTIME_ALARM: | |
2345 | case SOURCE_TIME_BOOTTIME_ALARM: | |
2346 | event_source_time_prioq_reshuffle(s); | |
2347 | break; | |
fd38203a | 2348 | |
ddfde737 LP |
2349 | case SOURCE_SIGNAL: |
2350 | r = event_make_signal_data(s->event, s->signal.sig, NULL); | |
2351 | if (r < 0) { | |
2352 | s->enabled = SD_EVENT_OFF; | |
2353 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); | |
2354 | return r; | |
2355 | } | |
fd38203a | 2356 | |
ddfde737 | 2357 | break; |
fd38203a | 2358 | |
ddfde737 LP |
2359 | case SOURCE_CHILD: |
2360 | s->event->n_enabled_child_sources++; | |
4807d2d0 | 2361 | |
ddfde737 LP |
2362 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
2363 | /* yes, we have pidfd */ | |
9da4cb2b | 2364 | |
ddfde737 | 2365 | r = source_child_pidfd_register(s, s->enabled); |
9da4cb2b LP |
2366 | if (r < 0) { |
2367 | s->enabled = SD_EVENT_OFF; | |
ddfde737 | 2368 | s->event->n_enabled_child_sources--; |
9da4cb2b LP |
2369 | return r; |
2370 | } | |
ddfde737 LP |
2371 | } else { |
2372 | /* no pidfd, or something other to watch for than WEXITED */ | |
9da4cb2b | 2373 | |
ddfde737 LP |
2374 | r = event_make_signal_data(s->event, SIGCHLD, NULL); |
2375 | if (r < 0) { | |
2376 | s->enabled = SD_EVENT_OFF; | |
2377 | s->event->n_enabled_child_sources--; | |
2378 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
2379 | return r; | |
2380 | } | |
2381 | } | |
fd38203a | 2382 | |
ddfde737 | 2383 | break; |
4807d2d0 | 2384 | |
ddfde737 LP |
2385 | case SOURCE_EXIT: |
2386 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
2387 | break; | |
7a0d4a3d | 2388 | |
ddfde737 LP |
2389 | case SOURCE_DEFER: |
2390 | case SOURCE_POST: | |
2391 | case SOURCE_INOTIFY: | |
2392 | break; | |
9da4cb2b | 2393 | |
ddfde737 LP |
2394 | default: |
2395 | assert_not_reached("Wut? I shouldn't exist."); | |
2396 | } | |
f8f3f926 | 2397 | |
ddfde737 LP |
2398 | return 0; |
2399 | } | |
2400 | ||
2401 | _public_ int sd_event_source_set_enabled(sd_event_source *s, int m) { | |
2402 | int r; | |
9da4cb2b | 2403 | |
ddfde737 LP |
2404 | assert_return(s, -EINVAL); |
2405 | assert_return(IN_SET(m, SD_EVENT_OFF, SD_EVENT_ON, SD_EVENT_ONESHOT), -EINVAL); | |
2406 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a | 2407 | |
ddfde737 LP |
2408 | /* If we are dead anyway, we are fine with turning off sources, but everything else needs to fail. */ |
2409 | if (s->event->state == SD_EVENT_FINISHED) | |
2410 | return m == SD_EVENT_OFF ? 0 : -ESTALE; | |
305f78bf | 2411 | |
ddfde737 LP |
2412 | if (s->enabled == m) /* No change? */ |
2413 | return 0; | |
9d3e3aa5 | 2414 | |
ddfde737 LP |
2415 | if (m == SD_EVENT_OFF) |
2416 | r = event_source_disable(s); | |
2417 | else { | |
2418 | if (s->enabled != SD_EVENT_OFF) { | |
2419 | /* Switching from "on" to "oneshot" or back? If that's the case, we can take a shortcut, the | |
2420 | * event source is already enabled after all. */ | |
2421 | s->enabled = m; | |
2422 | return 0; | |
fd38203a | 2423 | } |
ddfde737 LP |
2424 | |
2425 | r = event_source_enable(s, m); | |
fd38203a | 2426 | } |
ddfde737 LP |
2427 | if (r < 0) |
2428 | return r; | |
fd38203a | 2429 | |
e1951c16 | 2430 | event_source_pp_prioq_reshuffle(s); |
fd38203a LP |
2431 | return 0; |
2432 | } | |
2433 | ||
f7262a9f | 2434 | _public_ int sd_event_source_get_time(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
2435 | assert_return(s, -EINVAL); |
2436 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 2437 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf | 2438 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2439 | |
2440 | *usec = s->time.next; | |
2441 | return 0; | |
2442 | } | |
2443 | ||
f7262a9f | 2444 | _public_ int sd_event_source_set_time(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 2445 | int r; |
6a0f1f6d | 2446 | |
305f78bf | 2447 | assert_return(s, -EINVAL); |
6a0f1f6d | 2448 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 2449 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2450 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2451 | |
2a0dc6cd LP |
2452 | r = source_set_pending(s, false); |
2453 | if (r < 0) | |
2454 | return r; | |
2576a19e | 2455 | |
2a0dc6cd | 2456 | s->time.next = usec; |
fd38203a | 2457 | |
e1951c16 | 2458 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
2459 | return 0; |
2460 | } | |
2461 | ||
d6a83dc4 LP |
2462 | _public_ int sd_event_source_set_time_relative(sd_event_source *s, uint64_t usec) { |
2463 | usec_t t; | |
2464 | int r; | |
2465 | ||
2466 | assert_return(s, -EINVAL); | |
2467 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2468 | ||
2469 | r = sd_event_now(s->event, event_source_type_to_clock(s->type), &t); | |
2470 | if (r < 0) | |
2471 | return r; | |
2472 | ||
2473 | if (usec >= USEC_INFINITY - t) | |
2474 | return -EOVERFLOW; | |
2475 | ||
2476 | return sd_event_source_set_time(s, t + usec); | |
2477 | } | |
2478 | ||
f7262a9f | 2479 | _public_ int sd_event_source_get_time_accuracy(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
2480 | assert_return(s, -EINVAL); |
2481 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 2482 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf LP |
2483 | assert_return(!event_pid_changed(s->event), -ECHILD); |
2484 | ||
2485 | *usec = s->time.accuracy; | |
2486 | return 0; | |
2487 | } | |
2488 | ||
f7262a9f | 2489 | _public_ int sd_event_source_set_time_accuracy(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 2490 | int r; |
6a0f1f6d | 2491 | |
305f78bf LP |
2492 | assert_return(s, -EINVAL); |
2493 | assert_return(usec != (uint64_t) -1, -EINVAL); | |
6a0f1f6d | 2494 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 2495 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2496 | assert_return(!event_pid_changed(s->event), -ECHILD); |
eaa3cbef | 2497 | |
2a0dc6cd LP |
2498 | r = source_set_pending(s, false); |
2499 | if (r < 0) | |
2500 | return r; | |
2501 | ||
eaa3cbef LP |
2502 | if (usec == 0) |
2503 | usec = DEFAULT_ACCURACY_USEC; | |
2504 | ||
eaa3cbef LP |
2505 | s->time.accuracy = usec; |
2506 | ||
e1951c16 | 2507 | event_source_time_prioq_reshuffle(s); |
6a0f1f6d LP |
2508 | return 0; |
2509 | } | |
2510 | ||
2511 | _public_ int sd_event_source_get_time_clock(sd_event_source *s, clockid_t *clock) { | |
2512 | assert_return(s, -EINVAL); | |
2513 | assert_return(clock, -EINVAL); | |
2514 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2515 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
eaa3cbef | 2516 | |
6a0f1f6d | 2517 | *clock = event_source_type_to_clock(s->type); |
eaa3cbef LP |
2518 | return 0; |
2519 | } | |
2520 | ||
f7262a9f | 2521 | _public_ int sd_event_source_get_child_pid(sd_event_source *s, pid_t *pid) { |
4bee8012 LP |
2522 | assert_return(s, -EINVAL); |
2523 | assert_return(pid, -EINVAL); | |
2524 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2525 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2526 | ||
2527 | *pid = s->child.pid; | |
2528 | return 0; | |
2529 | } | |
2530 | ||
f8f3f926 LP |
2531 | _public_ int sd_event_source_get_child_pidfd(sd_event_source *s) { |
2532 | assert_return(s, -EINVAL); | |
2533 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2534 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2535 | ||
2536 | if (s->child.pidfd < 0) | |
2537 | return -EOPNOTSUPP; | |
2538 | ||
2539 | return s->child.pidfd; | |
2540 | } | |
2541 | ||
2542 | _public_ int sd_event_source_send_child_signal(sd_event_source *s, int sig, const siginfo_t *si, unsigned flags) { | |
2543 | assert_return(s, -EINVAL); | |
2544 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2545 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2546 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
2547 | ||
2548 | /* If we already have seen indication the process exited refuse sending a signal early. This way we | |
2549 | * can be sure we don't accidentally kill the wrong process on PID reuse when pidfds are not | |
2550 | * available. */ | |
2551 | if (s->child.exited) | |
2552 | return -ESRCH; | |
2553 | ||
2554 | if (s->child.pidfd >= 0) { | |
2555 | siginfo_t copy; | |
2556 | ||
2557 | /* pidfd_send_signal() changes the siginfo_t argument. This is weird, let's hence copy the | |
2558 | * structure here */ | |
2559 | if (si) | |
2560 | copy = *si; | |
2561 | ||
2562 | if (pidfd_send_signal(s->child.pidfd, sig, si ? © : NULL, 0) < 0) { | |
2563 | /* Let's propagate the error only if the system call is not implemented or prohibited */ | |
2564 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
2565 | return -errno; | |
2566 | } else | |
2567 | return 0; | |
2568 | } | |
2569 | ||
2570 | /* Flags are only supported for pidfd_send_signal(), not for rt_sigqueueinfo(), hence let's refuse | |
2571 | * this here. */ | |
2572 | if (flags != 0) | |
2573 | return -EOPNOTSUPP; | |
2574 | ||
2575 | if (si) { | |
2576 | /* We use rt_sigqueueinfo() only if siginfo_t is specified. */ | |
2577 | siginfo_t copy = *si; | |
2578 | ||
2579 | if (rt_sigqueueinfo(s->child.pid, sig, ©) < 0) | |
2580 | return -errno; | |
2581 | } else if (kill(s->child.pid, sig) < 0) | |
2582 | return -errno; | |
2583 | ||
2584 | return 0; | |
2585 | } | |
2586 | ||
2587 | _public_ int sd_event_source_get_child_pidfd_own(sd_event_source *s) { | |
2588 | assert_return(s, -EINVAL); | |
2589 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2590 | ||
2591 | if (s->child.pidfd < 0) | |
2592 | return -EOPNOTSUPP; | |
2593 | ||
2594 | return s->child.pidfd_owned; | |
2595 | } | |
2596 | ||
2597 | _public_ int sd_event_source_set_child_pidfd_own(sd_event_source *s, int own) { | |
2598 | assert_return(s, -EINVAL); | |
2599 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2600 | ||
2601 | if (s->child.pidfd < 0) | |
2602 | return -EOPNOTSUPP; | |
2603 | ||
2604 | s->child.pidfd_owned = own; | |
2605 | return 0; | |
2606 | } | |
2607 | ||
2608 | _public_ int sd_event_source_get_child_process_own(sd_event_source *s) { | |
2609 | assert_return(s, -EINVAL); | |
2610 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2611 | ||
2612 | return s->child.process_owned; | |
2613 | } | |
2614 | ||
2615 | _public_ int sd_event_source_set_child_process_own(sd_event_source *s, int own) { | |
2616 | assert_return(s, -EINVAL); | |
2617 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2618 | ||
2619 | s->child.process_owned = own; | |
2620 | return 0; | |
2621 | } | |
2622 | ||
97ef5391 LP |
2623 | _public_ int sd_event_source_get_inotify_mask(sd_event_source *s, uint32_t *mask) { |
2624 | assert_return(s, -EINVAL); | |
2625 | assert_return(mask, -EINVAL); | |
2626 | assert_return(s->type == SOURCE_INOTIFY, -EDOM); | |
2627 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2628 | ||
2629 | *mask = s->inotify.mask; | |
2630 | return 0; | |
2631 | } | |
2632 | ||
718db961 | 2633 | _public_ int sd_event_source_set_prepare(sd_event_source *s, sd_event_handler_t callback) { |
fd38203a LP |
2634 | int r; |
2635 | ||
da7e457c | 2636 | assert_return(s, -EINVAL); |
6203e07a | 2637 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c LP |
2638 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2639 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2640 | |
2641 | if (s->prepare == callback) | |
2642 | return 0; | |
2643 | ||
2644 | if (callback && s->prepare) { | |
2645 | s->prepare = callback; | |
2646 | return 0; | |
2647 | } | |
2648 | ||
2649 | r = prioq_ensure_allocated(&s->event->prepare, prepare_prioq_compare); | |
2650 | if (r < 0) | |
2651 | return r; | |
2652 | ||
2653 | s->prepare = callback; | |
2654 | ||
2655 | if (callback) { | |
2656 | r = prioq_put(s->event->prepare, s, &s->prepare_index); | |
2657 | if (r < 0) | |
2658 | return r; | |
2659 | } else | |
2660 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
2661 | ||
2662 | return 0; | |
2663 | } | |
2664 | ||
f7262a9f | 2665 | _public_ void* sd_event_source_get_userdata(sd_event_source *s) { |
da7e457c | 2666 | assert_return(s, NULL); |
fd38203a LP |
2667 | |
2668 | return s->userdata; | |
2669 | } | |
2670 | ||
8f726607 LP |
2671 | _public_ void *sd_event_source_set_userdata(sd_event_source *s, void *userdata) { |
2672 | void *ret; | |
2673 | ||
2674 | assert_return(s, NULL); | |
2675 | ||
2676 | ret = s->userdata; | |
2677 | s->userdata = userdata; | |
2678 | ||
2679 | return ret; | |
2680 | } | |
2681 | ||
c2ba3ad6 LP |
2682 | static usec_t sleep_between(sd_event *e, usec_t a, usec_t b) { |
2683 | usec_t c; | |
2684 | assert(e); | |
2685 | assert(a <= b); | |
2686 | ||
2687 | if (a <= 0) | |
2688 | return 0; | |
393003e1 LP |
2689 | if (a >= USEC_INFINITY) |
2690 | return USEC_INFINITY; | |
c2ba3ad6 LP |
2691 | |
2692 | if (b <= a + 1) | |
2693 | return a; | |
2694 | ||
52444dc4 LP |
2695 | initialize_perturb(e); |
2696 | ||
c2ba3ad6 LP |
2697 | /* |
2698 | Find a good time to wake up again between times a and b. We | |
2699 | have two goals here: | |
2700 | ||
2701 | a) We want to wake up as seldom as possible, hence prefer | |
2702 | later times over earlier times. | |
2703 | ||
2704 | b) But if we have to wake up, then let's make sure to | |
2705 | dispatch as much as possible on the entire system. | |
2706 | ||
2707 | We implement this by waking up everywhere at the same time | |
850516e0 | 2708 | within any given minute if we can, synchronised via the |
c2ba3ad6 | 2709 | perturbation value determined from the boot ID. If we can't, |
ba276c81 LP |
2710 | then we try to find the same spot in every 10s, then 1s and |
2711 | then 250ms step. Otherwise, we pick the last possible time | |
2712 | to wake up. | |
c2ba3ad6 LP |
2713 | */ |
2714 | ||
850516e0 LP |
2715 | c = (b / USEC_PER_MINUTE) * USEC_PER_MINUTE + e->perturb; |
2716 | if (c >= b) { | |
2717 | if (_unlikely_(c < USEC_PER_MINUTE)) | |
2718 | return b; | |
2719 | ||
2720 | c -= USEC_PER_MINUTE; | |
2721 | } | |
2722 | ||
ba276c81 LP |
2723 | if (c >= a) |
2724 | return c; | |
2725 | ||
2726 | c = (b / (USEC_PER_SEC*10)) * (USEC_PER_SEC*10) + (e->perturb % (USEC_PER_SEC*10)); | |
2727 | if (c >= b) { | |
2728 | if (_unlikely_(c < USEC_PER_SEC*10)) | |
2729 | return b; | |
2730 | ||
2731 | c -= USEC_PER_SEC*10; | |
2732 | } | |
2733 | ||
850516e0 LP |
2734 | if (c >= a) |
2735 | return c; | |
2736 | ||
2737 | c = (b / USEC_PER_SEC) * USEC_PER_SEC + (e->perturb % USEC_PER_SEC); | |
c2ba3ad6 LP |
2738 | if (c >= b) { |
2739 | if (_unlikely_(c < USEC_PER_SEC)) | |
2740 | return b; | |
2741 | ||
2742 | c -= USEC_PER_SEC; | |
2743 | } | |
2744 | ||
2745 | if (c >= a) | |
2746 | return c; | |
2747 | ||
2748 | c = (b / (USEC_PER_MSEC*250)) * (USEC_PER_MSEC*250) + (e->perturb % (USEC_PER_MSEC*250)); | |
2749 | if (c >= b) { | |
2750 | if (_unlikely_(c < USEC_PER_MSEC*250)) | |
2751 | return b; | |
2752 | ||
2753 | c -= USEC_PER_MSEC*250; | |
2754 | } | |
2755 | ||
2756 | if (c >= a) | |
2757 | return c; | |
2758 | ||
2759 | return b; | |
2760 | } | |
2761 | ||
fd38203a LP |
2762 | static int event_arm_timer( |
2763 | sd_event *e, | |
6a0f1f6d | 2764 | struct clock_data *d) { |
fd38203a LP |
2765 | |
2766 | struct itimerspec its = {}; | |
c2ba3ad6 LP |
2767 | sd_event_source *a, *b; |
2768 | usec_t t; | |
fd38203a | 2769 | |
cde93897 | 2770 | assert(e); |
6a0f1f6d | 2771 | assert(d); |
fd38203a | 2772 | |
d06441da | 2773 | if (!d->needs_rearm) |
212bbb17 TG |
2774 | return 0; |
2775 | else | |
2776 | d->needs_rearm = false; | |
2777 | ||
6a0f1f6d | 2778 | a = prioq_peek(d->earliest); |
393003e1 | 2779 | if (!a || a->enabled == SD_EVENT_OFF || a->time.next == USEC_INFINITY) { |
72aedc1e | 2780 | |
6a0f1f6d | 2781 | if (d->fd < 0) |
c57b5ca3 LP |
2782 | return 0; |
2783 | ||
3a43da28 | 2784 | if (d->next == USEC_INFINITY) |
72aedc1e LP |
2785 | return 0; |
2786 | ||
2787 | /* disarm */ | |
15c689d7 LP |
2788 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
2789 | return -errno; | |
72aedc1e | 2790 | |
3a43da28 | 2791 | d->next = USEC_INFINITY; |
fd38203a | 2792 | return 0; |
72aedc1e | 2793 | } |
fd38203a | 2794 | |
6a0f1f6d | 2795 | b = prioq_peek(d->latest); |
baf76283 | 2796 | assert_se(b && b->enabled != SD_EVENT_OFF); |
c2ba3ad6 | 2797 | |
1bce0ffa | 2798 | t = sleep_between(e, a->time.next, time_event_source_latest(b)); |
6a0f1f6d | 2799 | if (d->next == t) |
fd38203a LP |
2800 | return 0; |
2801 | ||
6a0f1f6d | 2802 | assert_se(d->fd >= 0); |
fd38203a | 2803 | |
c2ba3ad6 | 2804 | if (t == 0) { |
fd38203a LP |
2805 | /* We don' want to disarm here, just mean some time looooong ago. */ |
2806 | its.it_value.tv_sec = 0; | |
2807 | its.it_value.tv_nsec = 1; | |
2808 | } else | |
c2ba3ad6 | 2809 | timespec_store(&its.it_value, t); |
fd38203a | 2810 | |
15c689d7 | 2811 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 | 2812 | return -errno; |
fd38203a | 2813 | |
6a0f1f6d | 2814 | d->next = t; |
fd38203a LP |
2815 | return 0; |
2816 | } | |
2817 | ||
9a800b56 | 2818 | static int process_io(sd_event *e, sd_event_source *s, uint32_t revents) { |
fd38203a LP |
2819 | assert(e); |
2820 | assert(s); | |
2821 | assert(s->type == SOURCE_IO); | |
2822 | ||
9a800b56 LP |
2823 | /* If the event source was already pending, we just OR in the |
2824 | * new revents, otherwise we reset the value. The ORing is | |
2825 | * necessary to handle EPOLLONESHOT events properly where | |
2826 | * readability might happen independently of writability, and | |
2827 | * we need to keep track of both */ | |
2828 | ||
2829 | if (s->pending) | |
2830 | s->io.revents |= revents; | |
2831 | else | |
2832 | s->io.revents = revents; | |
fd38203a | 2833 | |
fd38203a LP |
2834 | return source_set_pending(s, true); |
2835 | } | |
2836 | ||
72aedc1e | 2837 | static int flush_timer(sd_event *e, int fd, uint32_t events, usec_t *next) { |
fd38203a LP |
2838 | uint64_t x; |
2839 | ssize_t ss; | |
2840 | ||
2841 | assert(e); | |
da7e457c | 2842 | assert(fd >= 0); |
72aedc1e | 2843 | |
305f78bf | 2844 | assert_return(events == EPOLLIN, -EIO); |
fd38203a LP |
2845 | |
2846 | ss = read(fd, &x, sizeof(x)); | |
2847 | if (ss < 0) { | |
945c2931 | 2848 | if (IN_SET(errno, EAGAIN, EINTR)) |
fd38203a LP |
2849 | return 0; |
2850 | ||
2851 | return -errno; | |
2852 | } | |
2853 | ||
8d35dae7 | 2854 | if (_unlikely_(ss != sizeof(x))) |
fd38203a LP |
2855 | return -EIO; |
2856 | ||
cde93897 | 2857 | if (next) |
3a43da28 | 2858 | *next = USEC_INFINITY; |
72aedc1e | 2859 | |
fd38203a LP |
2860 | return 0; |
2861 | } | |
2862 | ||
305f78bf LP |
2863 | static int process_timer( |
2864 | sd_event *e, | |
2865 | usec_t n, | |
6a0f1f6d | 2866 | struct clock_data *d) { |
305f78bf | 2867 | |
fd38203a LP |
2868 | sd_event_source *s; |
2869 | int r; | |
2870 | ||
2871 | assert(e); | |
6a0f1f6d | 2872 | assert(d); |
fd38203a LP |
2873 | |
2874 | for (;;) { | |
6a0f1f6d | 2875 | s = prioq_peek(d->earliest); |
fd38203a LP |
2876 | if (!s || |
2877 | s->time.next > n || | |
baf76283 | 2878 | s->enabled == SD_EVENT_OFF || |
fd38203a LP |
2879 | s->pending) |
2880 | break; | |
2881 | ||
2882 | r = source_set_pending(s, true); | |
2883 | if (r < 0) | |
2884 | return r; | |
2885 | ||
e1951c16 | 2886 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
2887 | } |
2888 | ||
2889 | return 0; | |
2890 | } | |
2891 | ||
2892 | static int process_child(sd_event *e) { | |
2893 | sd_event_source *s; | |
fd38203a LP |
2894 | int r; |
2895 | ||
2896 | assert(e); | |
2897 | ||
c2ba3ad6 LP |
2898 | e->need_process_child = false; |
2899 | ||
fd38203a LP |
2900 | /* |
2901 | So, this is ugly. We iteratively invoke waitid() with P_PID | |
2902 | + WNOHANG for each PID we wait for, instead of using | |
2903 | P_ALL. This is because we only want to get child | |
2904 | information of very specific child processes, and not all | |
2905 | of them. We might not have processed the SIGCHLD even of a | |
2906 | previous invocation and we don't want to maintain a | |
2907 | unbounded *per-child* event queue, hence we really don't | |
2908 | want anything flushed out of the kernel's queue that we | |
2909 | don't care about. Since this is O(n) this means that if you | |
2910 | have a lot of processes you probably want to handle SIGCHLD | |
2911 | yourself. | |
08cd1552 LP |
2912 | |
2913 | We do not reap the children here (by using WNOWAIT), this | |
2914 | is only done after the event source is dispatched so that | |
2915 | the callback still sees the process as a zombie. | |
fd38203a LP |
2916 | */ |
2917 | ||
90e74a66 | 2918 | HASHMAP_FOREACH(s, e->child_sources) { |
fd38203a LP |
2919 | assert(s->type == SOURCE_CHILD); |
2920 | ||
2921 | if (s->pending) | |
2922 | continue; | |
2923 | ||
baf76283 | 2924 | if (s->enabled == SD_EVENT_OFF) |
fd38203a LP |
2925 | continue; |
2926 | ||
f8f3f926 LP |
2927 | if (s->child.exited) |
2928 | continue; | |
2929 | ||
2930 | if (EVENT_SOURCE_WATCH_PIDFD(s)) /* There's a usable pidfd known for this event source? then don't waitid() for it here */ | |
2931 | continue; | |
2932 | ||
fd38203a | 2933 | zero(s->child.siginfo); |
15c689d7 LP |
2934 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, |
2935 | WNOHANG | (s->child.options & WEXITED ? WNOWAIT : 0) | s->child.options) < 0) | |
fd38203a LP |
2936 | return -errno; |
2937 | ||
2938 | if (s->child.siginfo.si_pid != 0) { | |
945c2931 | 2939 | bool zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 2940 | |
f8f3f926 LP |
2941 | if (zombie) |
2942 | s->child.exited = true; | |
2943 | ||
08cd1552 LP |
2944 | if (!zombie && (s->child.options & WEXITED)) { |
2945 | /* If the child isn't dead then let's | |
2946 | * immediately remove the state change | |
2947 | * from the queue, since there's no | |
2948 | * benefit in leaving it queued */ | |
2949 | ||
2950 | assert(s->child.options & (WSTOPPED|WCONTINUED)); | |
a5d27871 | 2951 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|(s->child.options & (WSTOPPED|WCONTINUED))); |
08cd1552 LP |
2952 | } |
2953 | ||
fd38203a LP |
2954 | r = source_set_pending(s, true); |
2955 | if (r < 0) | |
2956 | return r; | |
2957 | } | |
2958 | } | |
2959 | ||
fd38203a LP |
2960 | return 0; |
2961 | } | |
2962 | ||
f8f3f926 LP |
2963 | static int process_pidfd(sd_event *e, sd_event_source *s, uint32_t revents) { |
2964 | assert(e); | |
2965 | assert(s); | |
2966 | assert(s->type == SOURCE_CHILD); | |
2967 | ||
2968 | if (s->pending) | |
2969 | return 0; | |
2970 | ||
2971 | if (s->enabled == SD_EVENT_OFF) | |
2972 | return 0; | |
2973 | ||
2974 | if (!EVENT_SOURCE_WATCH_PIDFD(s)) | |
2975 | return 0; | |
2976 | ||
2977 | zero(s->child.siginfo); | |
2978 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG | WNOWAIT | s->child.options) < 0) | |
2979 | return -errno; | |
2980 | ||
2981 | if (s->child.siginfo.si_pid == 0) | |
2982 | return 0; | |
2983 | ||
2984 | if (IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) | |
2985 | s->child.exited = true; | |
2986 | ||
2987 | return source_set_pending(s, true); | |
2988 | } | |
2989 | ||
9da4cb2b | 2990 | static int process_signal(sd_event *e, struct signal_data *d, uint32_t events) { |
fd38203a | 2991 | bool read_one = false; |
fd38203a LP |
2992 | int r; |
2993 | ||
da7e457c | 2994 | assert(e); |
97ef5391 | 2995 | assert(d); |
305f78bf | 2996 | assert_return(events == EPOLLIN, -EIO); |
fd38203a | 2997 | |
9da4cb2b LP |
2998 | /* If there's a signal queued on this priority and SIGCHLD is |
2999 | on this priority too, then make sure to recheck the | |
3000 | children we watch. This is because we only ever dequeue | |
3001 | the first signal per priority, and if we dequeue one, and | |
3002 | SIGCHLD might be enqueued later we wouldn't know, but we | |
3003 | might have higher priority children we care about hence we | |
3004 | need to check that explicitly. */ | |
3005 | ||
3006 | if (sigismember(&d->sigset, SIGCHLD)) | |
3007 | e->need_process_child = true; | |
3008 | ||
3009 | /* If there's already an event source pending for this | |
3010 | * priority we don't read another */ | |
3011 | if (d->current) | |
3012 | return 0; | |
3013 | ||
fd38203a | 3014 | for (;;) { |
0eb2e0e3 | 3015 | struct signalfd_siginfo si; |
7057bd99 | 3016 | ssize_t n; |
92daebc0 | 3017 | sd_event_source *s = NULL; |
fd38203a | 3018 | |
9da4cb2b | 3019 | n = read(d->fd, &si, sizeof(si)); |
7057bd99 | 3020 | if (n < 0) { |
945c2931 | 3021 | if (IN_SET(errno, EAGAIN, EINTR)) |
fd38203a LP |
3022 | return read_one; |
3023 | ||
3024 | return -errno; | |
3025 | } | |
3026 | ||
7057bd99 | 3027 | if (_unlikely_(n != sizeof(si))) |
fd38203a LP |
3028 | return -EIO; |
3029 | ||
6eb7c172 | 3030 | assert(SIGNAL_VALID(si.ssi_signo)); |
7057bd99 | 3031 | |
fd38203a LP |
3032 | read_one = true; |
3033 | ||
92daebc0 LP |
3034 | if (e->signal_sources) |
3035 | s = e->signal_sources[si.ssi_signo]; | |
92daebc0 LP |
3036 | if (!s) |
3037 | continue; | |
9da4cb2b LP |
3038 | if (s->pending) |
3039 | continue; | |
fd38203a LP |
3040 | |
3041 | s->signal.siginfo = si; | |
9da4cb2b LP |
3042 | d->current = s; |
3043 | ||
fd38203a LP |
3044 | r = source_set_pending(s, true); |
3045 | if (r < 0) | |
3046 | return r; | |
9da4cb2b LP |
3047 | |
3048 | return 1; | |
fd38203a | 3049 | } |
fd38203a LP |
3050 | } |
3051 | ||
97ef5391 LP |
3052 | static int event_inotify_data_read(sd_event *e, struct inotify_data *d, uint32_t revents) { |
3053 | ssize_t n; | |
3054 | ||
3055 | assert(e); | |
3056 | assert(d); | |
3057 | ||
3058 | assert_return(revents == EPOLLIN, -EIO); | |
3059 | ||
3060 | /* If there's already an event source pending for this priority, don't read another */ | |
3061 | if (d->n_pending > 0) | |
3062 | return 0; | |
3063 | ||
3064 | /* Is the read buffer non-empty? If so, let's not read more */ | |
3065 | if (d->buffer_filled > 0) | |
3066 | return 0; | |
3067 | ||
3068 | n = read(d->fd, &d->buffer, sizeof(d->buffer)); | |
3069 | if (n < 0) { | |
3070 | if (IN_SET(errno, EAGAIN, EINTR)) | |
3071 | return 0; | |
3072 | ||
3073 | return -errno; | |
3074 | } | |
3075 | ||
3076 | assert(n > 0); | |
3077 | d->buffer_filled = (size_t) n; | |
3078 | LIST_PREPEND(buffered, e->inotify_data_buffered, d); | |
3079 | ||
3080 | return 1; | |
3081 | } | |
3082 | ||
3083 | static void event_inotify_data_drop(sd_event *e, struct inotify_data *d, size_t sz) { | |
3084 | assert(e); | |
3085 | assert(d); | |
3086 | assert(sz <= d->buffer_filled); | |
3087 | ||
3088 | if (sz == 0) | |
3089 | return; | |
3090 | ||
3091 | /* Move the rest to the buffer to the front, in order to get things properly aligned again */ | |
3092 | memmove(d->buffer.raw, d->buffer.raw + sz, d->buffer_filled - sz); | |
3093 | d->buffer_filled -= sz; | |
3094 | ||
3095 | if (d->buffer_filled == 0) | |
3096 | LIST_REMOVE(buffered, e->inotify_data_buffered, d); | |
3097 | } | |
3098 | ||
3099 | static int event_inotify_data_process(sd_event *e, struct inotify_data *d) { | |
3100 | int r; | |
3101 | ||
3102 | assert(e); | |
3103 | assert(d); | |
3104 | ||
3105 | /* If there's already an event source pending for this priority, don't read another */ | |
3106 | if (d->n_pending > 0) | |
3107 | return 0; | |
3108 | ||
3109 | while (d->buffer_filled > 0) { | |
3110 | size_t sz; | |
3111 | ||
3112 | /* Let's validate that the event structures are complete */ | |
3113 | if (d->buffer_filled < offsetof(struct inotify_event, name)) | |
3114 | return -EIO; | |
3115 | ||
3116 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3117 | if (d->buffer_filled < sz) | |
3118 | return -EIO; | |
3119 | ||
3120 | if (d->buffer.ev.mask & IN_Q_OVERFLOW) { | |
3121 | struct inode_data *inode_data; | |
97ef5391 LP |
3122 | |
3123 | /* The queue overran, let's pass this event to all event sources connected to this inotify | |
3124 | * object */ | |
3125 | ||
90e74a66 | 3126 | HASHMAP_FOREACH(inode_data, d->inodes) { |
97ef5391 LP |
3127 | sd_event_source *s; |
3128 | ||
3129 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
3130 | ||
3131 | if (s->enabled == SD_EVENT_OFF) | |
3132 | continue; | |
3133 | ||
3134 | r = source_set_pending(s, true); | |
3135 | if (r < 0) | |
3136 | return r; | |
3137 | } | |
3138 | } | |
3139 | } else { | |
3140 | struct inode_data *inode_data; | |
3141 | sd_event_source *s; | |
3142 | ||
3143 | /* Find the inode object for this watch descriptor. If IN_IGNORED is set we also remove it from | |
3144 | * our watch descriptor table. */ | |
3145 | if (d->buffer.ev.mask & IN_IGNORED) { | |
3146 | ||
3147 | inode_data = hashmap_remove(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3148 | if (!inode_data) { | |
3149 | event_inotify_data_drop(e, d, sz); | |
3150 | continue; | |
3151 | } | |
3152 | ||
3153 | /* The watch descriptor was removed by the kernel, let's drop it here too */ | |
3154 | inode_data->wd = -1; | |
3155 | } else { | |
3156 | inode_data = hashmap_get(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3157 | if (!inode_data) { | |
3158 | event_inotify_data_drop(e, d, sz); | |
3159 | continue; | |
3160 | } | |
3161 | } | |
3162 | ||
3163 | /* Trigger all event sources that are interested in these events. Also trigger all event | |
3164 | * sources if IN_IGNORED or IN_UNMOUNT is set. */ | |
3165 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
3166 | ||
3167 | if (s->enabled == SD_EVENT_OFF) | |
3168 | continue; | |
3169 | ||
3170 | if ((d->buffer.ev.mask & (IN_IGNORED|IN_UNMOUNT)) == 0 && | |
3171 | (s->inotify.mask & d->buffer.ev.mask & IN_ALL_EVENTS) == 0) | |
3172 | continue; | |
3173 | ||
3174 | r = source_set_pending(s, true); | |
3175 | if (r < 0) | |
3176 | return r; | |
3177 | } | |
3178 | } | |
3179 | ||
3180 | /* Something pending now? If so, let's finish, otherwise let's read more. */ | |
3181 | if (d->n_pending > 0) | |
3182 | return 1; | |
3183 | } | |
3184 | ||
3185 | return 0; | |
3186 | } | |
3187 | ||
3188 | static int process_inotify(sd_event *e) { | |
3189 | struct inotify_data *d; | |
3190 | int r, done = 0; | |
3191 | ||
3192 | assert(e); | |
3193 | ||
3194 | LIST_FOREACH(buffered, d, e->inotify_data_buffered) { | |
3195 | r = event_inotify_data_process(e, d); | |
3196 | if (r < 0) | |
3197 | return r; | |
3198 | if (r > 0) | |
3199 | done ++; | |
3200 | } | |
3201 | ||
3202 | return done; | |
3203 | } | |
3204 | ||
fd38203a | 3205 | static int source_dispatch(sd_event_source *s) { |
b778cba4 | 3206 | _cleanup_(sd_event_unrefp) sd_event *saved_event = NULL; |
8f5c235d | 3207 | EventSourceType saved_type; |
fe8245eb | 3208 | int r = 0; |
fd38203a LP |
3209 | |
3210 | assert(s); | |
6203e07a | 3211 | assert(s->pending || s->type == SOURCE_EXIT); |
fd38203a | 3212 | |
b778cba4 LP |
3213 | /* Save the event source type, here, so that we still know it after the event callback which might |
3214 | * invalidate the event. */ | |
8f5c235d LP |
3215 | saved_type = s->type; |
3216 | ||
b778cba4 LP |
3217 | /* Similar, store a reference to the event loop object, so that we can still access it after the |
3218 | * callback might have invalidated/disconnected the event source. */ | |
3219 | saved_event = sd_event_ref(s->event); | |
3220 | ||
945c2931 | 3221 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { |
da7e457c LP |
3222 | r = source_set_pending(s, false); |
3223 | if (r < 0) | |
3224 | return r; | |
3225 | } | |
fd38203a | 3226 | |
6e9feda3 LP |
3227 | if (s->type != SOURCE_POST) { |
3228 | sd_event_source *z; | |
6e9feda3 LP |
3229 | |
3230 | /* If we execute a non-post source, let's mark all | |
3231 | * post sources as pending */ | |
3232 | ||
90e74a66 | 3233 | SET_FOREACH(z, s->event->post_sources) { |
6e9feda3 LP |
3234 | if (z->enabled == SD_EVENT_OFF) |
3235 | continue; | |
3236 | ||
3237 | r = source_set_pending(z, true); | |
3238 | if (r < 0) | |
3239 | return r; | |
3240 | } | |
3241 | } | |
3242 | ||
baf76283 LP |
3243 | if (s->enabled == SD_EVENT_ONESHOT) { |
3244 | r = sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
3245 | if (r < 0) |
3246 | return r; | |
3247 | } | |
3248 | ||
12179984 | 3249 | s->dispatching = true; |
b7484e2a | 3250 | |
fd38203a LP |
3251 | switch (s->type) { |
3252 | ||
3253 | case SOURCE_IO: | |
3254 | r = s->io.callback(s, s->io.fd, s->io.revents, s->userdata); | |
3255 | break; | |
3256 | ||
6a0f1f6d | 3257 | case SOURCE_TIME_REALTIME: |
a8548816 | 3258 | case SOURCE_TIME_BOOTTIME: |
6a0f1f6d LP |
3259 | case SOURCE_TIME_MONOTONIC: |
3260 | case SOURCE_TIME_REALTIME_ALARM: | |
3261 | case SOURCE_TIME_BOOTTIME_ALARM: | |
fd38203a LP |
3262 | r = s->time.callback(s, s->time.next, s->userdata); |
3263 | break; | |
3264 | ||
3265 | case SOURCE_SIGNAL: | |
3266 | r = s->signal.callback(s, &s->signal.siginfo, s->userdata); | |
3267 | break; | |
3268 | ||
08cd1552 LP |
3269 | case SOURCE_CHILD: { |
3270 | bool zombie; | |
3271 | ||
945c2931 | 3272 | zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 3273 | |
fd38203a | 3274 | r = s->child.callback(s, &s->child.siginfo, s->userdata); |
08cd1552 LP |
3275 | |
3276 | /* Now, reap the PID for good. */ | |
f8f3f926 | 3277 | if (zombie) { |
cc59d290 | 3278 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|WEXITED); |
f8f3f926 LP |
3279 | s->child.waited = true; |
3280 | } | |
08cd1552 | 3281 | |
fd38203a | 3282 | break; |
08cd1552 | 3283 | } |
fd38203a LP |
3284 | |
3285 | case SOURCE_DEFER: | |
3286 | r = s->defer.callback(s, s->userdata); | |
3287 | break; | |
da7e457c | 3288 | |
6e9feda3 LP |
3289 | case SOURCE_POST: |
3290 | r = s->post.callback(s, s->userdata); | |
3291 | break; | |
3292 | ||
6203e07a LP |
3293 | case SOURCE_EXIT: |
3294 | r = s->exit.callback(s, s->userdata); | |
da7e457c | 3295 | break; |
9d3e3aa5 | 3296 | |
97ef5391 LP |
3297 | case SOURCE_INOTIFY: { |
3298 | struct sd_event *e = s->event; | |
3299 | struct inotify_data *d; | |
3300 | size_t sz; | |
3301 | ||
3302 | assert(s->inotify.inode_data); | |
3303 | assert_se(d = s->inotify.inode_data->inotify_data); | |
3304 | ||
3305 | assert(d->buffer_filled >= offsetof(struct inotify_event, name)); | |
3306 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3307 | assert(d->buffer_filled >= sz); | |
3308 | ||
3309 | r = s->inotify.callback(s, &d->buffer.ev, s->userdata); | |
3310 | ||
3311 | /* When no event is pending anymore on this inotify object, then let's drop the event from the | |
3312 | * buffer. */ | |
3313 | if (d->n_pending == 0) | |
3314 | event_inotify_data_drop(e, d, sz); | |
3315 | ||
3316 | break; | |
3317 | } | |
3318 | ||
9d3e3aa5 | 3319 | case SOURCE_WATCHDOG: |
a71fe8b8 | 3320 | case _SOURCE_EVENT_SOURCE_TYPE_MAX: |
9f2a50a3 | 3321 | case _SOURCE_EVENT_SOURCE_TYPE_INVALID: |
9d3e3aa5 | 3322 | assert_not_reached("Wut? I shouldn't exist."); |
fd38203a LP |
3323 | } |
3324 | ||
12179984 LP |
3325 | s->dispatching = false; |
3326 | ||
b778cba4 LP |
3327 | if (r < 0) { |
3328 | log_debug_errno(r, "Event source %s (type %s) returned error, %s: %m", | |
3329 | strna(s->description), | |
3330 | event_source_type_to_string(saved_type), | |
3331 | s->exit_on_failure ? "exiting" : "disabling"); | |
3332 | ||
3333 | if (s->exit_on_failure) | |
3334 | (void) sd_event_exit(saved_event, r); | |
3335 | } | |
12179984 LP |
3336 | |
3337 | if (s->n_ref == 0) | |
3338 | source_free(s); | |
3339 | else if (r < 0) | |
6203e07a | 3340 | sd_event_source_set_enabled(s, SD_EVENT_OFF); |
b7484e2a | 3341 | |
6203e07a | 3342 | return 1; |
fd38203a LP |
3343 | } |
3344 | ||
3345 | static int event_prepare(sd_event *e) { | |
3346 | int r; | |
3347 | ||
3348 | assert(e); | |
3349 | ||
3350 | for (;;) { | |
3351 | sd_event_source *s; | |
3352 | ||
3353 | s = prioq_peek(e->prepare); | |
baf76283 | 3354 | if (!s || s->prepare_iteration == e->iteration || s->enabled == SD_EVENT_OFF) |
fd38203a LP |
3355 | break; |
3356 | ||
3357 | s->prepare_iteration = e->iteration; | |
3358 | r = prioq_reshuffle(e->prepare, s, &s->prepare_index); | |
3359 | if (r < 0) | |
3360 | return r; | |
3361 | ||
3362 | assert(s->prepare); | |
12179984 LP |
3363 | |
3364 | s->dispatching = true; | |
fd38203a | 3365 | r = s->prepare(s, s->userdata); |
12179984 LP |
3366 | s->dispatching = false; |
3367 | ||
b778cba4 LP |
3368 | if (r < 0) { |
3369 | log_debug_errno(r, "Prepare callback of event source %s (type %s) returned error, %s: %m", | |
3370 | strna(s->description), | |
3371 | event_source_type_to_string(s->type), | |
3372 | s->exit_on_failure ? "exiting" : "disabling"); | |
3373 | ||
3374 | if (s->exit_on_failure) | |
3375 | (void) sd_event_exit(e, r); | |
3376 | } | |
fd38203a | 3377 | |
12179984 LP |
3378 | if (s->n_ref == 0) |
3379 | source_free(s); | |
3380 | else if (r < 0) | |
3381 | sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
3382 | } |
3383 | ||
3384 | return 0; | |
3385 | } | |
3386 | ||
6203e07a | 3387 | static int dispatch_exit(sd_event *e) { |
da7e457c | 3388 | sd_event_source *p; |
30dd293c | 3389 | _cleanup_(sd_event_unrefp) sd_event *ref = NULL; |
da7e457c LP |
3390 | int r; |
3391 | ||
3392 | assert(e); | |
3393 | ||
6203e07a | 3394 | p = prioq_peek(e->exit); |
baf76283 | 3395 | if (!p || p->enabled == SD_EVENT_OFF) { |
da7e457c LP |
3396 | e->state = SD_EVENT_FINISHED; |
3397 | return 0; | |
3398 | } | |
3399 | ||
30dd293c | 3400 | ref = sd_event_ref(e); |
da7e457c | 3401 | e->iteration++; |
6203e07a | 3402 | e->state = SD_EVENT_EXITING; |
da7e457c | 3403 | r = source_dispatch(p); |
2b0c9ef7 | 3404 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
3405 | return r; |
3406 | } | |
3407 | ||
c2ba3ad6 LP |
3408 | static sd_event_source* event_next_pending(sd_event *e) { |
3409 | sd_event_source *p; | |
3410 | ||
da7e457c LP |
3411 | assert(e); |
3412 | ||
c2ba3ad6 LP |
3413 | p = prioq_peek(e->pending); |
3414 | if (!p) | |
3415 | return NULL; | |
3416 | ||
baf76283 | 3417 | if (p->enabled == SD_EVENT_OFF) |
c2ba3ad6 LP |
3418 | return NULL; |
3419 | ||
3420 | return p; | |
3421 | } | |
3422 | ||
cde93897 LP |
3423 | static int arm_watchdog(sd_event *e) { |
3424 | struct itimerspec its = {}; | |
3425 | usec_t t; | |
cde93897 LP |
3426 | |
3427 | assert(e); | |
3428 | assert(e->watchdog_fd >= 0); | |
3429 | ||
3430 | t = sleep_between(e, | |
3431 | e->watchdog_last + (e->watchdog_period / 2), | |
3432 | e->watchdog_last + (e->watchdog_period * 3 / 4)); | |
3433 | ||
3434 | timespec_store(&its.it_value, t); | |
3435 | ||
75145780 LP |
3436 | /* Make sure we never set the watchdog to 0, which tells the |
3437 | * kernel to disable it. */ | |
3438 | if (its.it_value.tv_sec == 0 && its.it_value.tv_nsec == 0) | |
3439 | its.it_value.tv_nsec = 1; | |
3440 | ||
15c689d7 | 3441 | if (timerfd_settime(e->watchdog_fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 LP |
3442 | return -errno; |
3443 | ||
3444 | return 0; | |
3445 | } | |
3446 | ||
3447 | static int process_watchdog(sd_event *e) { | |
3448 | assert(e); | |
3449 | ||
3450 | if (!e->watchdog) | |
3451 | return 0; | |
3452 | ||
3453 | /* Don't notify watchdog too often */ | |
3454 | if (e->watchdog_last + e->watchdog_period / 4 > e->timestamp.monotonic) | |
3455 | return 0; | |
3456 | ||
3457 | sd_notify(false, "WATCHDOG=1"); | |
3458 | e->watchdog_last = e->timestamp.monotonic; | |
3459 | ||
3460 | return arm_watchdog(e); | |
3461 | } | |
3462 | ||
97ef5391 LP |
3463 | static void event_close_inode_data_fds(sd_event *e) { |
3464 | struct inode_data *d; | |
3465 | ||
3466 | assert(e); | |
3467 | ||
3468 | /* Close the fds pointing to the inodes to watch now. We need to close them as they might otherwise pin | |
3469 | * filesystems. But we can't close them right-away as we need them as long as the user still wants to make | |
5238e957 | 3470 | * adjustments to the even source, such as changing the priority (which requires us to remove and re-add a watch |
97ef5391 LP |
3471 | * for the inode). Hence, let's close them when entering the first iteration after they were added, as a |
3472 | * compromise. */ | |
3473 | ||
3474 | while ((d = e->inode_data_to_close)) { | |
3475 | assert(d->fd >= 0); | |
3476 | d->fd = safe_close(d->fd); | |
3477 | ||
3478 | LIST_REMOVE(to_close, e->inode_data_to_close, d); | |
3479 | } | |
3480 | } | |
3481 | ||
c45a5a74 TG |
3482 | _public_ int sd_event_prepare(sd_event *e) { |
3483 | int r; | |
fd38203a | 3484 | |
da7e457c | 3485 | assert_return(e, -EINVAL); |
b937d761 | 3486 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
3487 | assert_return(!event_pid_changed(e), -ECHILD); |
3488 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 3489 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 3490 | |
e5446015 LP |
3491 | /* Let's check that if we are a default event loop we are executed in the correct thread. We only do |
3492 | * this check here once, since gettid() is typically not cached, and thus want to minimize | |
3493 | * syscalls */ | |
3494 | assert_return(!e->default_event_ptr || e->tid == gettid(), -EREMOTEIO); | |
3495 | ||
6203e07a | 3496 | if (e->exit_requested) |
c45a5a74 | 3497 | goto pending; |
fd38203a LP |
3498 | |
3499 | e->iteration++; | |
3500 | ||
0be6c2f6 | 3501 | e->state = SD_EVENT_PREPARING; |
fd38203a | 3502 | r = event_prepare(e); |
0be6c2f6 | 3503 | e->state = SD_EVENT_INITIAL; |
fd38203a | 3504 | if (r < 0) |
c45a5a74 | 3505 | return r; |
fd38203a | 3506 | |
6a0f1f6d LP |
3507 | r = event_arm_timer(e, &e->realtime); |
3508 | if (r < 0) | |
c45a5a74 | 3509 | return r; |
6a0f1f6d | 3510 | |
a8548816 TG |
3511 | r = event_arm_timer(e, &e->boottime); |
3512 | if (r < 0) | |
c45a5a74 | 3513 | return r; |
a8548816 | 3514 | |
6a0f1f6d LP |
3515 | r = event_arm_timer(e, &e->monotonic); |
3516 | if (r < 0) | |
c45a5a74 | 3517 | return r; |
6a0f1f6d LP |
3518 | |
3519 | r = event_arm_timer(e, &e->realtime_alarm); | |
1b5995b0 | 3520 | if (r < 0) |
c45a5a74 | 3521 | return r; |
fd38203a | 3522 | |
6a0f1f6d | 3523 | r = event_arm_timer(e, &e->boottime_alarm); |
1b5995b0 | 3524 | if (r < 0) |
c45a5a74 | 3525 | return r; |
fd38203a | 3526 | |
97ef5391 LP |
3527 | event_close_inode_data_fds(e); |
3528 | ||
1b5995b0 | 3529 | if (event_next_pending(e) || e->need_process_child) |
c45a5a74 TG |
3530 | goto pending; |
3531 | ||
2b0c9ef7 | 3532 | e->state = SD_EVENT_ARMED; |
c45a5a74 TG |
3533 | |
3534 | return 0; | |
3535 | ||
3536 | pending: | |
2b0c9ef7 | 3537 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
3538 | r = sd_event_wait(e, 0); |
3539 | if (r == 0) | |
2b0c9ef7 | 3540 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
3541 | |
3542 | return r; | |
c45a5a74 TG |
3543 | } |
3544 | ||
3545 | _public_ int sd_event_wait(sd_event *e, uint64_t timeout) { | |
5cddd924 | 3546 | size_t event_queue_max; |
c45a5a74 TG |
3547 | int r, m, i; |
3548 | ||
3549 | assert_return(e, -EINVAL); | |
b937d761 | 3550 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
3551 | assert_return(!event_pid_changed(e), -ECHILD); |
3552 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 3553 | assert_return(e->state == SD_EVENT_ARMED, -EBUSY); |
c45a5a74 TG |
3554 | |
3555 | if (e->exit_requested) { | |
3556 | e->state = SD_EVENT_PENDING; | |
3557 | return 1; | |
3558 | } | |
6a0f1f6d | 3559 | |
5cddd924 LP |
3560 | event_queue_max = MAX(e->n_sources, 1u); |
3561 | if (!GREEDY_REALLOC(e->event_queue, e->event_queue_allocated, event_queue_max)) | |
3562 | return -ENOMEM; | |
fd38203a | 3563 | |
97ef5391 LP |
3564 | /* If we still have inotify data buffered, then query the other fds, but don't wait on it */ |
3565 | if (e->inotify_data_buffered) | |
3566 | timeout = 0; | |
3567 | ||
5cddd924 | 3568 | m = epoll_wait(e->epoll_fd, e->event_queue, event_queue_max, |
bab4820e | 3569 | timeout == (uint64_t) -1 ? -1 : (int) DIV_ROUND_UP(timeout, USEC_PER_MSEC)); |
da7e457c | 3570 | if (m < 0) { |
c45a5a74 TG |
3571 | if (errno == EINTR) { |
3572 | e->state = SD_EVENT_PENDING; | |
3573 | return 1; | |
3574 | } | |
3575 | ||
3576 | r = -errno; | |
da7e457c LP |
3577 | goto finish; |
3578 | } | |
fd38203a | 3579 | |
e475d10c | 3580 | triple_timestamp_get(&e->timestamp); |
fd38203a LP |
3581 | |
3582 | for (i = 0; i < m; i++) { | |
3583 | ||
5cddd924 LP |
3584 | if (e->event_queue[i].data.ptr == INT_TO_PTR(SOURCE_WATCHDOG)) |
3585 | r = flush_timer(e, e->watchdog_fd, e->event_queue[i].events, NULL); | |
9da4cb2b | 3586 | else { |
5cddd924 | 3587 | WakeupType *t = e->event_queue[i].data.ptr; |
9da4cb2b LP |
3588 | |
3589 | switch (*t) { | |
3590 | ||
f8f3f926 | 3591 | case WAKEUP_EVENT_SOURCE: { |
5cddd924 | 3592 | sd_event_source *s = e->event_queue[i].data.ptr; |
f8f3f926 LP |
3593 | |
3594 | assert(s); | |
3595 | ||
3596 | switch (s->type) { | |
3597 | ||
3598 | case SOURCE_IO: | |
5cddd924 | 3599 | r = process_io(e, s, e->event_queue[i].events); |
f8f3f926 LP |
3600 | break; |
3601 | ||
3602 | case SOURCE_CHILD: | |
5cddd924 | 3603 | r = process_pidfd(e, s, e->event_queue[i].events); |
f8f3f926 LP |
3604 | break; |
3605 | ||
3606 | default: | |
3607 | assert_not_reached("Unexpected event source type"); | |
3608 | } | |
3609 | ||
9da4cb2b | 3610 | break; |
f8f3f926 | 3611 | } |
fd38203a | 3612 | |
9da4cb2b | 3613 | case WAKEUP_CLOCK_DATA: { |
5cddd924 | 3614 | struct clock_data *d = e->event_queue[i].data.ptr; |
f8f3f926 LP |
3615 | |
3616 | assert(d); | |
3617 | ||
5cddd924 | 3618 | r = flush_timer(e, d->fd, e->event_queue[i].events, &d->next); |
9da4cb2b LP |
3619 | break; |
3620 | } | |
3621 | ||
3622 | case WAKEUP_SIGNAL_DATA: | |
5cddd924 | 3623 | r = process_signal(e, e->event_queue[i].data.ptr, e->event_queue[i].events); |
9da4cb2b LP |
3624 | break; |
3625 | ||
97ef5391 | 3626 | case WAKEUP_INOTIFY_DATA: |
5cddd924 | 3627 | r = event_inotify_data_read(e, e->event_queue[i].data.ptr, e->event_queue[i].events); |
97ef5391 LP |
3628 | break; |
3629 | ||
9da4cb2b LP |
3630 | default: |
3631 | assert_not_reached("Invalid wake-up pointer"); | |
3632 | } | |
3633 | } | |
fd38203a | 3634 | if (r < 0) |
da7e457c | 3635 | goto finish; |
fd38203a LP |
3636 | } |
3637 | ||
cde93897 LP |
3638 | r = process_watchdog(e); |
3639 | if (r < 0) | |
3640 | goto finish; | |
3641 | ||
6a0f1f6d LP |
3642 | r = process_timer(e, e->timestamp.realtime, &e->realtime); |
3643 | if (r < 0) | |
3644 | goto finish; | |
3645 | ||
e475d10c | 3646 | r = process_timer(e, e->timestamp.boottime, &e->boottime); |
a8548816 TG |
3647 | if (r < 0) |
3648 | goto finish; | |
3649 | ||
6a0f1f6d LP |
3650 | r = process_timer(e, e->timestamp.monotonic, &e->monotonic); |
3651 | if (r < 0) | |
3652 | goto finish; | |
3653 | ||
3654 | r = process_timer(e, e->timestamp.realtime, &e->realtime_alarm); | |
fd38203a | 3655 | if (r < 0) |
da7e457c | 3656 | goto finish; |
fd38203a | 3657 | |
e475d10c | 3658 | r = process_timer(e, e->timestamp.boottime, &e->boottime_alarm); |
fd38203a | 3659 | if (r < 0) |
da7e457c | 3660 | goto finish; |
fd38203a | 3661 | |
c2ba3ad6 | 3662 | if (e->need_process_child) { |
fd38203a LP |
3663 | r = process_child(e); |
3664 | if (r < 0) | |
da7e457c | 3665 | goto finish; |
fd38203a LP |
3666 | } |
3667 | ||
97ef5391 LP |
3668 | r = process_inotify(e); |
3669 | if (r < 0) | |
3670 | goto finish; | |
3671 | ||
c45a5a74 TG |
3672 | if (event_next_pending(e)) { |
3673 | e->state = SD_EVENT_PENDING; | |
3674 | ||
3675 | return 1; | |
da7e457c LP |
3676 | } |
3677 | ||
c45a5a74 | 3678 | r = 0; |
fd38203a | 3679 | |
da7e457c | 3680 | finish: |
2b0c9ef7 | 3681 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
3682 | |
3683 | return r; | |
fd38203a LP |
3684 | } |
3685 | ||
c45a5a74 TG |
3686 | _public_ int sd_event_dispatch(sd_event *e) { |
3687 | sd_event_source *p; | |
3688 | int r; | |
3689 | ||
3690 | assert_return(e, -EINVAL); | |
b937d761 | 3691 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
3692 | assert_return(!event_pid_changed(e), -ECHILD); |
3693 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
3694 | assert_return(e->state == SD_EVENT_PENDING, -EBUSY); | |
3695 | ||
3696 | if (e->exit_requested) | |
3697 | return dispatch_exit(e); | |
3698 | ||
3699 | p = event_next_pending(e); | |
3700 | if (p) { | |
30dd293c | 3701 | _cleanup_(sd_event_unrefp) sd_event *ref = NULL; |
c45a5a74 | 3702 | |
30dd293c | 3703 | ref = sd_event_ref(e); |
c45a5a74 TG |
3704 | e->state = SD_EVENT_RUNNING; |
3705 | r = source_dispatch(p); | |
2b0c9ef7 | 3706 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
3707 | return r; |
3708 | } | |
3709 | ||
2b0c9ef7 | 3710 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
3711 | |
3712 | return 1; | |
3713 | } | |
3714 | ||
34b87517 | 3715 | static void event_log_delays(sd_event *e) { |
442ac269 YW |
3716 | char b[ELEMENTSOF(e->delays) * DECIMAL_STR_MAX(unsigned) + 1], *p; |
3717 | size_t l, i; | |
34b87517 | 3718 | |
442ac269 YW |
3719 | p = b; |
3720 | l = sizeof(b); | |
3721 | for (i = 0; i < ELEMENTSOF(e->delays); i++) { | |
3722 | l = strpcpyf(&p, l, "%u ", e->delays[i]); | |
34b87517 VC |
3723 | e->delays[i] = 0; |
3724 | } | |
442ac269 | 3725 | log_debug("Event loop iterations: %s", b); |
34b87517 VC |
3726 | } |
3727 | ||
c45a5a74 TG |
3728 | _public_ int sd_event_run(sd_event *e, uint64_t timeout) { |
3729 | int r; | |
3730 | ||
3731 | assert_return(e, -EINVAL); | |
b937d761 | 3732 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
3733 | assert_return(!event_pid_changed(e), -ECHILD); |
3734 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 3735 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
c45a5a74 | 3736 | |
34b87517 VC |
3737 | if (e->profile_delays && e->last_run) { |
3738 | usec_t this_run; | |
3739 | unsigned l; | |
3740 | ||
3741 | this_run = now(CLOCK_MONOTONIC); | |
3742 | ||
3743 | l = u64log2(this_run - e->last_run); | |
3744 | assert(l < sizeof(e->delays)); | |
3745 | e->delays[l]++; | |
3746 | ||
3747 | if (this_run - e->last_log >= 5*USEC_PER_SEC) { | |
3748 | event_log_delays(e); | |
3749 | e->last_log = this_run; | |
3750 | } | |
3751 | } | |
3752 | ||
c45a5a74 | 3753 | r = sd_event_prepare(e); |
53bac4e0 LP |
3754 | if (r == 0) |
3755 | /* There was nothing? Then wait... */ | |
3756 | r = sd_event_wait(e, timeout); | |
c45a5a74 | 3757 | |
34b87517 VC |
3758 | if (e->profile_delays) |
3759 | e->last_run = now(CLOCK_MONOTONIC); | |
3760 | ||
02d30981 | 3761 | if (r > 0) { |
53bac4e0 | 3762 | /* There's something now, then let's dispatch it */ |
02d30981 TG |
3763 | r = sd_event_dispatch(e); |
3764 | if (r < 0) | |
3765 | return r; | |
53bac4e0 LP |
3766 | |
3767 | return 1; | |
3768 | } | |
3769 | ||
3770 | return r; | |
c45a5a74 TG |
3771 | } |
3772 | ||
f7262a9f | 3773 | _public_ int sd_event_loop(sd_event *e) { |
30dd293c | 3774 | _cleanup_(sd_event_unrefp) sd_event *ref = NULL; |
fd38203a LP |
3775 | int r; |
3776 | ||
da7e457c | 3777 | assert_return(e, -EINVAL); |
b937d761 | 3778 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 3779 | assert_return(!event_pid_changed(e), -ECHILD); |
2b0c9ef7 | 3780 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 3781 | |
30dd293c | 3782 | ref = sd_event_ref(e); |
fd38203a | 3783 | |
da7e457c | 3784 | while (e->state != SD_EVENT_FINISHED) { |
fd38203a LP |
3785 | r = sd_event_run(e, (uint64_t) -1); |
3786 | if (r < 0) | |
30dd293c | 3787 | return r; |
fd38203a LP |
3788 | } |
3789 | ||
30dd293c | 3790 | return e->exit_code; |
fd38203a LP |
3791 | } |
3792 | ||
9b364545 | 3793 | _public_ int sd_event_get_fd(sd_event *e) { |
9b364545 | 3794 | assert_return(e, -EINVAL); |
b937d761 | 3795 | assert_return(e = event_resolve(e), -ENOPKG); |
9b364545 TG |
3796 | assert_return(!event_pid_changed(e), -ECHILD); |
3797 | ||
3798 | return e->epoll_fd; | |
3799 | } | |
3800 | ||
f7262a9f | 3801 | _public_ int sd_event_get_state(sd_event *e) { |
da7e457c | 3802 | assert_return(e, -EINVAL); |
b937d761 | 3803 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
3804 | assert_return(!event_pid_changed(e), -ECHILD); |
3805 | ||
3806 | return e->state; | |
3807 | } | |
3808 | ||
6203e07a | 3809 | _public_ int sd_event_get_exit_code(sd_event *e, int *code) { |
da7e457c | 3810 | assert_return(e, -EINVAL); |
b937d761 | 3811 | assert_return(e = event_resolve(e), -ENOPKG); |
6203e07a | 3812 | assert_return(code, -EINVAL); |
da7e457c | 3813 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 3814 | |
6203e07a LP |
3815 | if (!e->exit_requested) |
3816 | return -ENODATA; | |
3817 | ||
3818 | *code = e->exit_code; | |
3819 | return 0; | |
fd38203a LP |
3820 | } |
3821 | ||
6203e07a | 3822 | _public_ int sd_event_exit(sd_event *e, int code) { |
da7e457c | 3823 | assert_return(e, -EINVAL); |
b937d761 | 3824 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
3825 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
3826 | assert_return(!event_pid_changed(e), -ECHILD); | |
fd38203a | 3827 | |
6203e07a LP |
3828 | e->exit_requested = true; |
3829 | e->exit_code = code; | |
3830 | ||
fd38203a LP |
3831 | return 0; |
3832 | } | |
46e8c825 | 3833 | |
6a0f1f6d | 3834 | _public_ int sd_event_now(sd_event *e, clockid_t clock, uint64_t *usec) { |
46e8c825 | 3835 | assert_return(e, -EINVAL); |
b937d761 | 3836 | assert_return(e = event_resolve(e), -ENOPKG); |
46e8c825 | 3837 | assert_return(usec, -EINVAL); |
46e8c825 LP |
3838 | assert_return(!event_pid_changed(e), -ECHILD); |
3839 | ||
e475d10c LP |
3840 | if (!TRIPLE_TIMESTAMP_HAS_CLOCK(clock)) |
3841 | return -EOPNOTSUPP; | |
3842 | ||
3843 | /* Generate a clean error in case CLOCK_BOOTTIME is not available. Note that don't use clock_supported() here, | |
3844 | * for a reason: there are systems where CLOCK_BOOTTIME is supported, but CLOCK_BOOTTIME_ALARM is not, but for | |
3845 | * the purpose of getting the time this doesn't matter. */ | |
3411372e LP |
3846 | if (IN_SET(clock, CLOCK_BOOTTIME, CLOCK_BOOTTIME_ALARM) && !clock_boottime_supported()) |
3847 | return -EOPNOTSUPP; | |
3848 | ||
e475d10c | 3849 | if (!triple_timestamp_is_set(&e->timestamp)) { |
15c689d7 | 3850 | /* Implicitly fall back to now() if we never ran before and thus have no cached time. */ |
38a03f06 LP |
3851 | *usec = now(clock); |
3852 | return 1; | |
3853 | } | |
46e8c825 | 3854 | |
e475d10c | 3855 | *usec = triple_timestamp_by_clock(&e->timestamp, clock); |
46e8c825 LP |
3856 | return 0; |
3857 | } | |
afc6adb5 LP |
3858 | |
3859 | _public_ int sd_event_default(sd_event **ret) { | |
39883f62 | 3860 | sd_event *e = NULL; |
afc6adb5 LP |
3861 | int r; |
3862 | ||
3863 | if (!ret) | |
3864 | return !!default_event; | |
3865 | ||
3866 | if (default_event) { | |
3867 | *ret = sd_event_ref(default_event); | |
3868 | return 0; | |
3869 | } | |
3870 | ||
3871 | r = sd_event_new(&e); | |
3872 | if (r < 0) | |
3873 | return r; | |
3874 | ||
3875 | e->default_event_ptr = &default_event; | |
3876 | e->tid = gettid(); | |
3877 | default_event = e; | |
3878 | ||
3879 | *ret = e; | |
3880 | return 1; | |
3881 | } | |
3882 | ||
3883 | _public_ int sd_event_get_tid(sd_event *e, pid_t *tid) { | |
3884 | assert_return(e, -EINVAL); | |
b937d761 | 3885 | assert_return(e = event_resolve(e), -ENOPKG); |
afc6adb5 | 3886 | assert_return(tid, -EINVAL); |
76b54375 | 3887 | assert_return(!event_pid_changed(e), -ECHILD); |
afc6adb5 | 3888 | |
76b54375 LP |
3889 | if (e->tid != 0) { |
3890 | *tid = e->tid; | |
3891 | return 0; | |
3892 | } | |
3893 | ||
3894 | return -ENXIO; | |
afc6adb5 | 3895 | } |
cde93897 LP |
3896 | |
3897 | _public_ int sd_event_set_watchdog(sd_event *e, int b) { | |
3898 | int r; | |
3899 | ||
3900 | assert_return(e, -EINVAL); | |
b937d761 | 3901 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 | 3902 | assert_return(!event_pid_changed(e), -ECHILD); |
cde93897 LP |
3903 | |
3904 | if (e->watchdog == !!b) | |
3905 | return e->watchdog; | |
3906 | ||
3907 | if (b) { | |
09812eb7 LP |
3908 | r = sd_watchdog_enabled(false, &e->watchdog_period); |
3909 | if (r <= 0) | |
cde93897 | 3910 | return r; |
cde93897 LP |
3911 | |
3912 | /* Issue first ping immediately */ | |
3913 | sd_notify(false, "WATCHDOG=1"); | |
3914 | e->watchdog_last = now(CLOCK_MONOTONIC); | |
3915 | ||
3916 | e->watchdog_fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC); | |
3917 | if (e->watchdog_fd < 0) | |
3918 | return -errno; | |
3919 | ||
3920 | r = arm_watchdog(e); | |
3921 | if (r < 0) | |
3922 | goto fail; | |
3923 | ||
1eac7948 | 3924 | struct epoll_event ev = { |
a82f89aa LP |
3925 | .events = EPOLLIN, |
3926 | .data.ptr = INT_TO_PTR(SOURCE_WATCHDOG), | |
3927 | }; | |
cde93897 | 3928 | |
15c689d7 | 3929 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, e->watchdog_fd, &ev) < 0) { |
cde93897 LP |
3930 | r = -errno; |
3931 | goto fail; | |
3932 | } | |
3933 | ||
3934 | } else { | |
3935 | if (e->watchdog_fd >= 0) { | |
5a795bff | 3936 | (void) epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, e->watchdog_fd, NULL); |
03e334a1 | 3937 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
3938 | } |
3939 | } | |
3940 | ||
3941 | e->watchdog = !!b; | |
3942 | return e->watchdog; | |
3943 | ||
3944 | fail: | |
03e334a1 | 3945 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
3946 | return r; |
3947 | } | |
8f726607 LP |
3948 | |
3949 | _public_ int sd_event_get_watchdog(sd_event *e) { | |
3950 | assert_return(e, -EINVAL); | |
b937d761 | 3951 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 LP |
3952 | assert_return(!event_pid_changed(e), -ECHILD); |
3953 | ||
3954 | return e->watchdog; | |
3955 | } | |
60a3b1e1 LP |
3956 | |
3957 | _public_ int sd_event_get_iteration(sd_event *e, uint64_t *ret) { | |
3958 | assert_return(e, -EINVAL); | |
b937d761 | 3959 | assert_return(e = event_resolve(e), -ENOPKG); |
60a3b1e1 LP |
3960 | assert_return(!event_pid_changed(e), -ECHILD); |
3961 | ||
3962 | *ret = e->iteration; | |
3963 | return 0; | |
3964 | } | |
15723a1d LP |
3965 | |
3966 | _public_ int sd_event_source_set_destroy_callback(sd_event_source *s, sd_event_destroy_t callback) { | |
3967 | assert_return(s, -EINVAL); | |
3968 | ||
3969 | s->destroy_callback = callback; | |
3970 | return 0; | |
3971 | } | |
3972 | ||
3973 | _public_ int sd_event_source_get_destroy_callback(sd_event_source *s, sd_event_destroy_t *ret) { | |
3974 | assert_return(s, -EINVAL); | |
3975 | ||
3976 | if (ret) | |
3977 | *ret = s->destroy_callback; | |
3978 | ||
3979 | return !!s->destroy_callback; | |
3980 | } | |
2382c936 YW |
3981 | |
3982 | _public_ int sd_event_source_get_floating(sd_event_source *s) { | |
3983 | assert_return(s, -EINVAL); | |
3984 | ||
3985 | return s->floating; | |
3986 | } | |
3987 | ||
3988 | _public_ int sd_event_source_set_floating(sd_event_source *s, int b) { | |
3989 | assert_return(s, -EINVAL); | |
3990 | ||
3991 | if (s->floating == !!b) | |
3992 | return 0; | |
3993 | ||
3994 | if (!s->event) /* Already disconnected */ | |
3995 | return -ESTALE; | |
3996 | ||
3997 | s->floating = b; | |
3998 | ||
3999 | if (b) { | |
4000 | sd_event_source_ref(s); | |
4001 | sd_event_unref(s->event); | |
4002 | } else { | |
4003 | sd_event_ref(s->event); | |
4004 | sd_event_source_unref(s); | |
4005 | } | |
4006 | ||
4007 | return 1; | |
4008 | } | |
b778cba4 LP |
4009 | |
4010 | _public_ int sd_event_source_get_exit_on_failure(sd_event_source *s) { | |
4011 | assert_return(s, -EINVAL); | |
4012 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
4013 | ||
4014 | return s->exit_on_failure; | |
4015 | } | |
4016 | ||
4017 | _public_ int sd_event_source_set_exit_on_failure(sd_event_source *s, int b) { | |
4018 | assert_return(s, -EINVAL); | |
4019 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
4020 | ||
4021 | if (s->exit_on_failure == !!b) | |
4022 | return 0; | |
4023 | ||
4024 | s->exit_on_failure = b; | |
4025 | return 1; | |
4026 | } |