]>
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 | ||
b6d5481b LP |
40 | static bool event_source_is_online(sd_event_source *s) { |
41 | assert(s); | |
42 | return s->enabled != SD_EVENT_OFF && !s->ratelimited; | |
43 | } | |
44 | ||
45 | static bool event_source_is_offline(sd_event_source *s) { | |
46 | assert(s); | |
47 | return s->enabled == SD_EVENT_OFF || s->ratelimited; | |
48 | } | |
49 | ||
55cbfaa5 DM |
50 | static const char* const event_source_type_table[_SOURCE_EVENT_SOURCE_TYPE_MAX] = { |
51 | [SOURCE_IO] = "io", | |
52 | [SOURCE_TIME_REALTIME] = "realtime", | |
53 | [SOURCE_TIME_BOOTTIME] = "bootime", | |
54 | [SOURCE_TIME_MONOTONIC] = "monotonic", | |
55 | [SOURCE_TIME_REALTIME_ALARM] = "realtime-alarm", | |
56 | [SOURCE_TIME_BOOTTIME_ALARM] = "boottime-alarm", | |
57 | [SOURCE_SIGNAL] = "signal", | |
58 | [SOURCE_CHILD] = "child", | |
59 | [SOURCE_DEFER] = "defer", | |
60 | [SOURCE_POST] = "post", | |
61 | [SOURCE_EXIT] = "exit", | |
62 | [SOURCE_WATCHDOG] = "watchdog", | |
97ef5391 | 63 | [SOURCE_INOTIFY] = "inotify", |
55cbfaa5 DM |
64 | }; |
65 | ||
66 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(event_source_type, int); | |
67 | ||
b6d5481b LP |
68 | #define EVENT_SOURCE_IS_TIME(t) \ |
69 | IN_SET((t), \ | |
70 | SOURCE_TIME_REALTIME, \ | |
71 | SOURCE_TIME_BOOTTIME, \ | |
72 | SOURCE_TIME_MONOTONIC, \ | |
73 | SOURCE_TIME_REALTIME_ALARM, \ | |
74 | SOURCE_TIME_BOOTTIME_ALARM) | |
75 | ||
76 | #define EVENT_SOURCE_CAN_RATE_LIMIT(t) \ | |
77 | IN_SET((t), \ | |
78 | SOURCE_IO, \ | |
79 | SOURCE_TIME_REALTIME, \ | |
80 | SOURCE_TIME_BOOTTIME, \ | |
81 | SOURCE_TIME_MONOTONIC, \ | |
82 | SOURCE_TIME_REALTIME_ALARM, \ | |
83 | SOURCE_TIME_BOOTTIME_ALARM, \ | |
84 | SOURCE_SIGNAL, \ | |
85 | SOURCE_DEFER, \ | |
86 | SOURCE_INOTIFY) | |
6a0f1f6d | 87 | |
fd38203a | 88 | struct sd_event { |
da7e457c | 89 | unsigned n_ref; |
fd38203a LP |
90 | |
91 | int epoll_fd; | |
cde93897 | 92 | int watchdog_fd; |
fd38203a LP |
93 | |
94 | Prioq *pending; | |
95 | Prioq *prepare; | |
c2ba3ad6 | 96 | |
a8548816 | 97 | /* timerfd_create() only supports these five clocks so far. We |
6a0f1f6d LP |
98 | * can add support for more clocks when the kernel learns to |
99 | * deal with them, too. */ | |
100 | struct clock_data realtime; | |
a8548816 | 101 | struct clock_data boottime; |
6a0f1f6d LP |
102 | struct clock_data monotonic; |
103 | struct clock_data realtime_alarm; | |
104 | struct clock_data boottime_alarm; | |
fd38203a | 105 | |
da7e457c LP |
106 | usec_t perturb; |
107 | ||
9da4cb2b LP |
108 | sd_event_source **signal_sources; /* indexed by signal number */ |
109 | Hashmap *signal_data; /* indexed by priority */ | |
fd38203a LP |
110 | |
111 | Hashmap *child_sources; | |
b6d5481b | 112 | unsigned n_online_child_sources; |
fd38203a | 113 | |
6e9feda3 LP |
114 | Set *post_sources; |
115 | ||
6203e07a | 116 | Prioq *exit; |
fd38203a | 117 | |
97ef5391 LP |
118 | Hashmap *inotify_data; /* indexed by priority */ |
119 | ||
120 | /* A list of inode structures that still have an fd open, that we need to close before the next loop iteration */ | |
121 | LIST_HEAD(struct inode_data, inode_data_to_close); | |
122 | ||
123 | /* A list of inotify objects that already have events buffered which aren't processed yet */ | |
124 | LIST_HEAD(struct inotify_data, inotify_data_buffered); | |
125 | ||
da7e457c | 126 | pid_t original_pid; |
c2ba3ad6 | 127 | |
60a3b1e1 | 128 | uint64_t iteration; |
e475d10c | 129 | triple_timestamp timestamp; |
da7e457c | 130 | int state; |
eaa3cbef | 131 | |
6203e07a | 132 | bool exit_requested:1; |
da7e457c | 133 | bool need_process_child:1; |
cde93897 | 134 | bool watchdog:1; |
34b87517 | 135 | bool profile_delays:1; |
afc6adb5 | 136 | |
6203e07a LP |
137 | int exit_code; |
138 | ||
afc6adb5 LP |
139 | pid_t tid; |
140 | sd_event **default_event_ptr; | |
cde93897 LP |
141 | |
142 | usec_t watchdog_last, watchdog_period; | |
15b38f93 LP |
143 | |
144 | unsigned n_sources; | |
a71fe8b8 | 145 | |
5cddd924 | 146 | struct epoll_event *event_queue; |
5cddd924 | 147 | |
a71fe8b8 | 148 | LIST_HEAD(sd_event_source, sources); |
34b87517 | 149 | |
e6a7bee5 | 150 | usec_t last_run_usec, last_log_usec; |
34b87517 | 151 | unsigned delays[sizeof(usec_t) * 8]; |
fd38203a LP |
152 | }; |
153 | ||
b937d761 NM |
154 | static thread_local sd_event *default_event = NULL; |
155 | ||
a71fe8b8 | 156 | static void source_disconnect(sd_event_source *s); |
97ef5391 | 157 | static void event_gc_inode_data(sd_event *e, struct inode_data *d); |
a71fe8b8 | 158 | |
b937d761 NM |
159 | static sd_event *event_resolve(sd_event *e) { |
160 | return e == SD_EVENT_DEFAULT ? default_event : e; | |
161 | } | |
162 | ||
fd38203a LP |
163 | static int pending_prioq_compare(const void *a, const void *b) { |
164 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 165 | int r; |
fd38203a LP |
166 | |
167 | assert(x->pending); | |
168 | assert(y->pending); | |
169 | ||
baf76283 LP |
170 | /* Enabled ones first */ |
171 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
fd38203a | 172 | return -1; |
baf76283 | 173 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) |
fd38203a LP |
174 | return 1; |
175 | ||
b6d5481b LP |
176 | /* Non rate-limited ones first. */ |
177 | r = CMP(!!x->ratelimited, !!y->ratelimited); | |
178 | if (r != 0) | |
179 | return r; | |
180 | ||
fd38203a | 181 | /* Lower priority values first */ |
9c57a73b YW |
182 | r = CMP(x->priority, y->priority); |
183 | if (r != 0) | |
184 | return r; | |
fd38203a LP |
185 | |
186 | /* Older entries first */ | |
9c57a73b | 187 | return CMP(x->pending_iteration, y->pending_iteration); |
fd38203a LP |
188 | } |
189 | ||
190 | static int prepare_prioq_compare(const void *a, const void *b) { | |
191 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 192 | int r; |
fd38203a LP |
193 | |
194 | assert(x->prepare); | |
195 | assert(y->prepare); | |
196 | ||
8046c457 KK |
197 | /* Enabled ones first */ |
198 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
199 | return -1; | |
200 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) | |
201 | return 1; | |
202 | ||
b6d5481b LP |
203 | /* Non rate-limited ones first. */ |
204 | r = CMP(!!x->ratelimited, !!y->ratelimited); | |
205 | if (r != 0) | |
206 | return r; | |
207 | ||
fd38203a LP |
208 | /* Move most recently prepared ones last, so that we can stop |
209 | * preparing as soon as we hit one that has already been | |
210 | * prepared in the current iteration */ | |
9c57a73b YW |
211 | r = CMP(x->prepare_iteration, y->prepare_iteration); |
212 | if (r != 0) | |
213 | return r; | |
fd38203a | 214 | |
fd38203a | 215 | /* Lower priority values first */ |
9c57a73b | 216 | return CMP(x->priority, y->priority); |
fd38203a LP |
217 | } |
218 | ||
b6d5481b LP |
219 | static usec_t time_event_source_next(const sd_event_source *s) { |
220 | assert(s); | |
221 | ||
222 | /* We have two kinds of event sources that have elapsation times associated with them: the actual | |
223 | * time based ones and the ones for which a ratelimit can be in effect (where we want to be notified | |
224 | * once the ratelimit time window ends). Let's return the next elapsing time depending on what we are | |
225 | * looking at here. */ | |
226 | ||
227 | if (s->ratelimited) { /* If rate-limited the next elapsation is when the ratelimit time window ends */ | |
228 | assert(s->rate_limit.begin != 0); | |
229 | assert(s->rate_limit.interval != 0); | |
230 | return usec_add(s->rate_limit.begin, s->rate_limit.interval); | |
231 | } | |
232 | ||
233 | /* Otherwise this must be a time event source, if not ratelimited */ | |
234 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
235 | return s->time.next; | |
236 | ||
237 | return USEC_INFINITY; | |
238 | } | |
239 | ||
1bce0ffa | 240 | static usec_t time_event_source_latest(const sd_event_source *s) { |
b6d5481b LP |
241 | assert(s); |
242 | ||
243 | if (s->ratelimited) { /* For ratelimited stuff the earliest and the latest time shall actually be the | |
244 | * same, as we should avoid adding additional inaccuracy on an inaccuracy time | |
245 | * window */ | |
246 | assert(s->rate_limit.begin != 0); | |
247 | assert(s->rate_limit.interval != 0); | |
248 | return usec_add(s->rate_limit.begin, s->rate_limit.interval); | |
249 | } | |
250 | ||
251 | /* Must be a time event source, if not ratelimited */ | |
252 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
253 | return usec_add(s->time.next, s->time.accuracy); | |
254 | ||
255 | return USEC_INFINITY; | |
1bce0ffa LP |
256 | } |
257 | ||
81107b84 LP |
258 | static bool event_source_timer_candidate(const sd_event_source *s) { |
259 | assert(s); | |
260 | ||
261 | /* Returns true for event sources that either are not pending yet (i.e. where it's worth to mark them pending) | |
262 | * or which are currently ratelimited (i.e. where it's worth leaving the ratelimited state) */ | |
263 | return !s->pending || s->ratelimited; | |
264 | } | |
265 | ||
266 | static int time_prioq_compare(const void *a, const void *b, usec_t (*time_func)(const sd_event_source *s)) { | |
c2ba3ad6 LP |
267 | const sd_event_source *x = a, *y = b; |
268 | ||
baf76283 LP |
269 | /* Enabled ones first */ |
270 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
c2ba3ad6 | 271 | return -1; |
baf76283 | 272 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) |
c2ba3ad6 LP |
273 | return 1; |
274 | ||
81107b84 LP |
275 | /* Order "non-pending OR ratelimited" before "pending AND not-ratelimited" */ |
276 | if (event_source_timer_candidate(x) && !event_source_timer_candidate(y)) | |
c2ba3ad6 | 277 | return -1; |
81107b84 | 278 | if (!event_source_timer_candidate(x) && event_source_timer_candidate(y)) |
c2ba3ad6 LP |
279 | return 1; |
280 | ||
281 | /* Order by time */ | |
81107b84 LP |
282 | return CMP(time_func(x), time_func(y)); |
283 | } | |
284 | ||
285 | static int earliest_time_prioq_compare(const void *a, const void *b) { | |
286 | return time_prioq_compare(a, b, time_event_source_next); | |
287 | } | |
288 | ||
289 | static int latest_time_prioq_compare(const void *a, const void *b) { | |
290 | return time_prioq_compare(a, b, time_event_source_latest); | |
c2ba3ad6 LP |
291 | } |
292 | ||
6203e07a | 293 | static int exit_prioq_compare(const void *a, const void *b) { |
da7e457c LP |
294 | const sd_event_source *x = a, *y = b; |
295 | ||
6203e07a LP |
296 | assert(x->type == SOURCE_EXIT); |
297 | assert(y->type == SOURCE_EXIT); | |
da7e457c | 298 | |
baf76283 LP |
299 | /* Enabled ones first */ |
300 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
da7e457c | 301 | return -1; |
baf76283 | 302 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) |
da7e457c LP |
303 | return 1; |
304 | ||
305 | /* Lower priority values first */ | |
6dd91b36 | 306 | return CMP(x->priority, y->priority); |
da7e457c LP |
307 | } |
308 | ||
6a0f1f6d LP |
309 | static void free_clock_data(struct clock_data *d) { |
310 | assert(d); | |
9da4cb2b | 311 | assert(d->wakeup == WAKEUP_CLOCK_DATA); |
6a0f1f6d LP |
312 | |
313 | safe_close(d->fd); | |
314 | prioq_free(d->earliest); | |
315 | prioq_free(d->latest); | |
316 | } | |
317 | ||
8301aa0b | 318 | static sd_event *event_free(sd_event *e) { |
a71fe8b8 LP |
319 | sd_event_source *s; |
320 | ||
fd38203a | 321 | assert(e); |
a71fe8b8 LP |
322 | |
323 | while ((s = e->sources)) { | |
324 | assert(s->floating); | |
325 | source_disconnect(s); | |
326 | sd_event_source_unref(s); | |
327 | } | |
328 | ||
15b38f93 | 329 | assert(e->n_sources == 0); |
fd38203a | 330 | |
afc6adb5 LP |
331 | if (e->default_event_ptr) |
332 | *(e->default_event_ptr) = NULL; | |
333 | ||
03e334a1 | 334 | safe_close(e->epoll_fd); |
03e334a1 | 335 | safe_close(e->watchdog_fd); |
cde93897 | 336 | |
6a0f1f6d | 337 | free_clock_data(&e->realtime); |
a8548816 | 338 | free_clock_data(&e->boottime); |
6a0f1f6d LP |
339 | free_clock_data(&e->monotonic); |
340 | free_clock_data(&e->realtime_alarm); | |
341 | free_clock_data(&e->boottime_alarm); | |
342 | ||
fd38203a LP |
343 | prioq_free(e->pending); |
344 | prioq_free(e->prepare); | |
6203e07a | 345 | prioq_free(e->exit); |
fd38203a LP |
346 | |
347 | free(e->signal_sources); | |
9da4cb2b | 348 | hashmap_free(e->signal_data); |
fd38203a | 349 | |
97ef5391 LP |
350 | hashmap_free(e->inotify_data); |
351 | ||
fd38203a | 352 | hashmap_free(e->child_sources); |
6e9feda3 | 353 | set_free(e->post_sources); |
8301aa0b | 354 | |
5cddd924 LP |
355 | free(e->event_queue); |
356 | ||
8301aa0b | 357 | return mfree(e); |
fd38203a LP |
358 | } |
359 | ||
f7262a9f | 360 | _public_ int sd_event_new(sd_event** ret) { |
fd38203a LP |
361 | sd_event *e; |
362 | int r; | |
363 | ||
305f78bf | 364 | assert_return(ret, -EINVAL); |
fd38203a | 365 | |
d08eb1fa | 366 | e = new(sd_event, 1); |
fd38203a LP |
367 | if (!e) |
368 | return -ENOMEM; | |
369 | ||
d08eb1fa LP |
370 | *e = (sd_event) { |
371 | .n_ref = 1, | |
372 | .epoll_fd = -1, | |
373 | .watchdog_fd = -1, | |
374 | .realtime.wakeup = WAKEUP_CLOCK_DATA, | |
375 | .realtime.fd = -1, | |
376 | .realtime.next = USEC_INFINITY, | |
377 | .boottime.wakeup = WAKEUP_CLOCK_DATA, | |
378 | .boottime.fd = -1, | |
379 | .boottime.next = USEC_INFINITY, | |
380 | .monotonic.wakeup = WAKEUP_CLOCK_DATA, | |
381 | .monotonic.fd = -1, | |
382 | .monotonic.next = USEC_INFINITY, | |
383 | .realtime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
384 | .realtime_alarm.fd = -1, | |
385 | .realtime_alarm.next = USEC_INFINITY, | |
386 | .boottime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
387 | .boottime_alarm.fd = -1, | |
388 | .boottime_alarm.next = USEC_INFINITY, | |
389 | .perturb = USEC_INFINITY, | |
390 | .original_pid = getpid_cached(), | |
391 | }; | |
fd38203a | 392 | |
c983e776 EV |
393 | r = prioq_ensure_allocated(&e->pending, pending_prioq_compare); |
394 | if (r < 0) | |
fd38203a | 395 | goto fail; |
fd38203a LP |
396 | |
397 | e->epoll_fd = epoll_create1(EPOLL_CLOEXEC); | |
398 | if (e->epoll_fd < 0) { | |
399 | r = -errno; | |
400 | goto fail; | |
401 | } | |
402 | ||
7fe2903c LP |
403 | e->epoll_fd = fd_move_above_stdio(e->epoll_fd); |
404 | ||
34b87517 | 405 | if (secure_getenv("SD_EVENT_PROFILE_DELAYS")) { |
1d3a473b | 406 | 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 |
407 | e->profile_delays = true; |
408 | } | |
409 | ||
fd38203a LP |
410 | *ret = e; |
411 | return 0; | |
412 | ||
413 | fail: | |
414 | event_free(e); | |
415 | return r; | |
416 | } | |
417 | ||
8301aa0b | 418 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event, sd_event, event_free); |
fd38203a | 419 | |
afd15bbb ZJS |
420 | _public_ sd_event_source* sd_event_source_disable_unref(sd_event_source *s) { |
421 | if (s) | |
422 | (void) sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
423 | return sd_event_source_unref(s); | |
424 | } | |
425 | ||
eaa3cbef LP |
426 | static bool event_pid_changed(sd_event *e) { |
427 | assert(e); | |
428 | ||
a2360a46 | 429 | /* We don't support people creating an event loop and keeping |
eaa3cbef LP |
430 | * it around over a fork(). Let's complain. */ |
431 | ||
df0ff127 | 432 | return e->original_pid != getpid_cached(); |
eaa3cbef LP |
433 | } |
434 | ||
366e6411 | 435 | static void source_io_unregister(sd_event_source *s) { |
fd38203a LP |
436 | assert(s); |
437 | assert(s->type == SOURCE_IO); | |
438 | ||
f6806734 | 439 | if (event_pid_changed(s->event)) |
366e6411 | 440 | return; |
f6806734 | 441 | |
fd38203a | 442 | if (!s->io.registered) |
366e6411 | 443 | return; |
fd38203a | 444 | |
d1cf2023 | 445 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->io.fd, NULL) < 0) |
f80a5d6a | 446 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll, ignoring: %m", |
55cbfaa5 | 447 | strna(s->description), event_source_type_to_string(s->type)); |
fd38203a LP |
448 | |
449 | s->io.registered = false; | |
fd38203a LP |
450 | } |
451 | ||
305f78bf LP |
452 | static int source_io_register( |
453 | sd_event_source *s, | |
454 | int enabled, | |
455 | uint32_t events) { | |
456 | ||
fd38203a LP |
457 | assert(s); |
458 | assert(s->type == SOURCE_IO); | |
baf76283 | 459 | assert(enabled != SD_EVENT_OFF); |
fd38203a | 460 | |
1eac7948 | 461 | struct epoll_event ev = { |
a82f89aa LP |
462 | .events = events | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0), |
463 | .data.ptr = s, | |
464 | }; | |
fd38203a | 465 | |
15c689d7 | 466 | if (epoll_ctl(s->event->epoll_fd, |
1eac7948 | 467 | s->io.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, |
55c540d3 | 468 | s->io.fd, &ev) < 0) |
fd38203a LP |
469 | return -errno; |
470 | ||
471 | s->io.registered = true; | |
472 | ||
473 | return 0; | |
474 | } | |
475 | ||
f8f3f926 LP |
476 | static void source_child_pidfd_unregister(sd_event_source *s) { |
477 | assert(s); | |
478 | assert(s->type == SOURCE_CHILD); | |
479 | ||
480 | if (event_pid_changed(s->event)) | |
481 | return; | |
482 | ||
483 | if (!s->child.registered) | |
484 | return; | |
485 | ||
486 | if (EVENT_SOURCE_WATCH_PIDFD(s)) | |
487 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->child.pidfd, NULL) < 0) | |
f80a5d6a | 488 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll, ignoring: %m", |
f8f3f926 LP |
489 | strna(s->description), event_source_type_to_string(s->type)); |
490 | ||
491 | s->child.registered = false; | |
492 | } | |
493 | ||
494 | static int source_child_pidfd_register(sd_event_source *s, int enabled) { | |
f8f3f926 LP |
495 | assert(s); |
496 | assert(s->type == SOURCE_CHILD); | |
497 | assert(enabled != SD_EVENT_OFF); | |
498 | ||
499 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { | |
1eac7948 | 500 | struct epoll_event ev = { |
f8f3f926 LP |
501 | .events = EPOLLIN | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0), |
502 | .data.ptr = s, | |
503 | }; | |
504 | ||
55c540d3 ZJS |
505 | if (epoll_ctl(s->event->epoll_fd, |
506 | s->child.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, | |
507 | s->child.pidfd, &ev) < 0) | |
f8f3f926 LP |
508 | return -errno; |
509 | } | |
510 | ||
511 | s->child.registered = true; | |
512 | return 0; | |
513 | } | |
514 | ||
6a0f1f6d LP |
515 | static clockid_t event_source_type_to_clock(EventSourceType t) { |
516 | ||
517 | switch (t) { | |
518 | ||
519 | case SOURCE_TIME_REALTIME: | |
520 | return CLOCK_REALTIME; | |
521 | ||
a8548816 TG |
522 | case SOURCE_TIME_BOOTTIME: |
523 | return CLOCK_BOOTTIME; | |
524 | ||
6a0f1f6d LP |
525 | case SOURCE_TIME_MONOTONIC: |
526 | return CLOCK_MONOTONIC; | |
527 | ||
528 | case SOURCE_TIME_REALTIME_ALARM: | |
529 | return CLOCK_REALTIME_ALARM; | |
530 | ||
531 | case SOURCE_TIME_BOOTTIME_ALARM: | |
532 | return CLOCK_BOOTTIME_ALARM; | |
533 | ||
534 | default: | |
535 | return (clockid_t) -1; | |
536 | } | |
537 | } | |
538 | ||
539 | static EventSourceType clock_to_event_source_type(clockid_t clock) { | |
540 | ||
541 | switch (clock) { | |
542 | ||
543 | case CLOCK_REALTIME: | |
544 | return SOURCE_TIME_REALTIME; | |
545 | ||
a8548816 TG |
546 | case CLOCK_BOOTTIME: |
547 | return SOURCE_TIME_BOOTTIME; | |
548 | ||
6a0f1f6d LP |
549 | case CLOCK_MONOTONIC: |
550 | return SOURCE_TIME_MONOTONIC; | |
551 | ||
552 | case CLOCK_REALTIME_ALARM: | |
553 | return SOURCE_TIME_REALTIME_ALARM; | |
554 | ||
555 | case CLOCK_BOOTTIME_ALARM: | |
556 | return SOURCE_TIME_BOOTTIME_ALARM; | |
557 | ||
558 | default: | |
559 | return _SOURCE_EVENT_SOURCE_TYPE_INVALID; | |
560 | } | |
561 | } | |
562 | ||
563 | static struct clock_data* event_get_clock_data(sd_event *e, EventSourceType t) { | |
564 | assert(e); | |
565 | ||
566 | switch (t) { | |
567 | ||
568 | case SOURCE_TIME_REALTIME: | |
569 | return &e->realtime; | |
570 | ||
a8548816 TG |
571 | case SOURCE_TIME_BOOTTIME: |
572 | return &e->boottime; | |
573 | ||
6a0f1f6d LP |
574 | case SOURCE_TIME_MONOTONIC: |
575 | return &e->monotonic; | |
576 | ||
577 | case SOURCE_TIME_REALTIME_ALARM: | |
578 | return &e->realtime_alarm; | |
579 | ||
580 | case SOURCE_TIME_BOOTTIME_ALARM: | |
581 | return &e->boottime_alarm; | |
582 | ||
583 | default: | |
584 | return NULL; | |
585 | } | |
586 | } | |
587 | ||
3e4eb8e7 YW |
588 | static void event_free_signal_data(sd_event *e, struct signal_data *d) { |
589 | assert(e); | |
590 | ||
591 | if (!d) | |
592 | return; | |
593 | ||
594 | hashmap_remove(e->signal_data, &d->priority); | |
595 | safe_close(d->fd); | |
596 | free(d); | |
597 | } | |
598 | ||
9da4cb2b LP |
599 | static int event_make_signal_data( |
600 | sd_event *e, | |
601 | int sig, | |
602 | struct signal_data **ret) { | |
4807d2d0 | 603 | |
9da4cb2b LP |
604 | struct signal_data *d; |
605 | bool added = false; | |
606 | sigset_t ss_copy; | |
607 | int64_t priority; | |
f95387cd ZJS |
608 | int r; |
609 | ||
610 | assert(e); | |
611 | ||
f6806734 | 612 | if (event_pid_changed(e)) |
9da4cb2b | 613 | return -ECHILD; |
f6806734 | 614 | |
9da4cb2b LP |
615 | if (e->signal_sources && e->signal_sources[sig]) |
616 | priority = e->signal_sources[sig]->priority; | |
617 | else | |
de05913d | 618 | priority = SD_EVENT_PRIORITY_NORMAL; |
f95387cd | 619 | |
9da4cb2b LP |
620 | d = hashmap_get(e->signal_data, &priority); |
621 | if (d) { | |
622 | if (sigismember(&d->sigset, sig) > 0) { | |
623 | if (ret) | |
624 | *ret = d; | |
625 | return 0; | |
626 | } | |
627 | } else { | |
d08eb1fa | 628 | d = new(struct signal_data, 1); |
9da4cb2b LP |
629 | if (!d) |
630 | return -ENOMEM; | |
631 | ||
d08eb1fa LP |
632 | *d = (struct signal_data) { |
633 | .wakeup = WAKEUP_SIGNAL_DATA, | |
634 | .fd = -1, | |
635 | .priority = priority, | |
636 | }; | |
9da4cb2b | 637 | |
f656fdb6 | 638 | r = hashmap_ensure_put(&e->signal_data, &uint64_hash_ops, &d->priority, d); |
90f604d1 ZJS |
639 | if (r < 0) { |
640 | free(d); | |
9da4cb2b | 641 | return r; |
90f604d1 | 642 | } |
f95387cd | 643 | |
9da4cb2b LP |
644 | added = true; |
645 | } | |
646 | ||
647 | ss_copy = d->sigset; | |
648 | assert_se(sigaddset(&ss_copy, sig) >= 0); | |
649 | ||
650 | r = signalfd(d->fd, &ss_copy, SFD_NONBLOCK|SFD_CLOEXEC); | |
651 | if (r < 0) { | |
652 | r = -errno; | |
653 | goto fail; | |
654 | } | |
655 | ||
656 | d->sigset = ss_copy; | |
f95387cd | 657 | |
9da4cb2b LP |
658 | if (d->fd >= 0) { |
659 | if (ret) | |
660 | *ret = d; | |
f95387cd | 661 | return 0; |
9da4cb2b LP |
662 | } |
663 | ||
7fe2903c | 664 | d->fd = fd_move_above_stdio(r); |
f95387cd | 665 | |
1eac7948 | 666 | struct epoll_event ev = { |
a82f89aa LP |
667 | .events = EPOLLIN, |
668 | .data.ptr = d, | |
669 | }; | |
f95387cd | 670 | |
15c689d7 | 671 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { |
9da4cb2b LP |
672 | r = -errno; |
673 | goto fail; | |
f95387cd ZJS |
674 | } |
675 | ||
9da4cb2b LP |
676 | if (ret) |
677 | *ret = d; | |
678 | ||
f95387cd | 679 | return 0; |
9da4cb2b LP |
680 | |
681 | fail: | |
3e4eb8e7 YW |
682 | if (added) |
683 | event_free_signal_data(e, d); | |
9da4cb2b LP |
684 | |
685 | return r; | |
686 | } | |
687 | ||
688 | static void event_unmask_signal_data(sd_event *e, struct signal_data *d, int sig) { | |
689 | assert(e); | |
690 | assert(d); | |
691 | ||
692 | /* Turns off the specified signal in the signal data | |
693 | * object. If the signal mask of the object becomes empty that | |
694 | * way removes it. */ | |
695 | ||
696 | if (sigismember(&d->sigset, sig) == 0) | |
697 | return; | |
698 | ||
699 | assert_se(sigdelset(&d->sigset, sig) >= 0); | |
700 | ||
701 | if (sigisemptyset(&d->sigset)) { | |
9da4cb2b | 702 | /* If all the mask is all-zero we can get rid of the structure */ |
3e4eb8e7 | 703 | event_free_signal_data(e, d); |
9da4cb2b LP |
704 | return; |
705 | } | |
706 | ||
707 | assert(d->fd >= 0); | |
708 | ||
709 | if (signalfd(d->fd, &d->sigset, SFD_NONBLOCK|SFD_CLOEXEC) < 0) | |
710 | log_debug_errno(errno, "Failed to unset signal bit, ignoring: %m"); | |
711 | } | |
712 | ||
713 | static void event_gc_signal_data(sd_event *e, const int64_t *priority, int sig) { | |
714 | struct signal_data *d; | |
715 | static const int64_t zero_priority = 0; | |
716 | ||
717 | assert(e); | |
718 | ||
f8f3f926 LP |
719 | /* Rechecks if the specified signal is still something we are interested in. If not, we'll unmask it, |
720 | * and possibly drop the signalfd for it. */ | |
9da4cb2b LP |
721 | |
722 | if (sig == SIGCHLD && | |
b6d5481b | 723 | e->n_online_child_sources > 0) |
9da4cb2b LP |
724 | return; |
725 | ||
726 | if (e->signal_sources && | |
727 | e->signal_sources[sig] && | |
b6d5481b | 728 | event_source_is_online(e->signal_sources[sig])) |
9da4cb2b LP |
729 | return; |
730 | ||
731 | /* | |
732 | * The specified signal might be enabled in three different queues: | |
733 | * | |
734 | * 1) the one that belongs to the priority passed (if it is non-NULL) | |
735 | * 2) the one that belongs to the priority of the event source of the signal (if there is one) | |
736 | * 3) the 0 priority (to cover the SIGCHLD case) | |
737 | * | |
738 | * Hence, let's remove it from all three here. | |
739 | */ | |
740 | ||
741 | if (priority) { | |
742 | d = hashmap_get(e->signal_data, priority); | |
743 | if (d) | |
744 | event_unmask_signal_data(e, d, sig); | |
745 | } | |
746 | ||
747 | if (e->signal_sources && e->signal_sources[sig]) { | |
748 | d = hashmap_get(e->signal_data, &e->signal_sources[sig]->priority); | |
749 | if (d) | |
750 | event_unmask_signal_data(e, d, sig); | |
751 | } | |
752 | ||
753 | d = hashmap_get(e->signal_data, &zero_priority); | |
754 | if (d) | |
755 | event_unmask_signal_data(e, d, sig); | |
f95387cd ZJS |
756 | } |
757 | ||
e1951c16 MS |
758 | static void event_source_pp_prioq_reshuffle(sd_event_source *s) { |
759 | assert(s); | |
760 | ||
761 | /* Reshuffles the pending + prepare prioqs. Called whenever the dispatch order changes, i.e. when | |
762 | * they are enabled/disabled or marked pending and such. */ | |
763 | ||
764 | if (s->pending) | |
765 | prioq_reshuffle(s->event->pending, s, &s->pending_index); | |
766 | ||
767 | if (s->prepare) | |
768 | prioq_reshuffle(s->event->prepare, s, &s->prepare_index); | |
769 | } | |
770 | ||
771 | static void event_source_time_prioq_reshuffle(sd_event_source *s) { | |
772 | struct clock_data *d; | |
773 | ||
774 | assert(s); | |
e1951c16 MS |
775 | |
776 | /* Called whenever the event source's timer ordering properties changed, i.e. time, accuracy, | |
777 | * pending, enable state. Makes sure the two prioq's are ordered properly again. */ | |
b6d5481b LP |
778 | |
779 | if (s->ratelimited) | |
780 | d = &s->event->monotonic; | |
781 | else { | |
782 | assert(EVENT_SOURCE_IS_TIME(s->type)); | |
783 | assert_se(d = event_get_clock_data(s->event, s->type)); | |
784 | } | |
785 | ||
f41315fc LP |
786 | prioq_reshuffle(d->earliest, s, &s->earliest_index); |
787 | prioq_reshuffle(d->latest, s, &s->latest_index); | |
e1951c16 MS |
788 | d->needs_rearm = true; |
789 | } | |
790 | ||
1e45e3fe LP |
791 | static void event_source_time_prioq_remove( |
792 | sd_event_source *s, | |
793 | struct clock_data *d) { | |
794 | ||
795 | assert(s); | |
796 | assert(d); | |
797 | ||
f41315fc LP |
798 | prioq_remove(d->earliest, s, &s->earliest_index); |
799 | prioq_remove(d->latest, s, &s->latest_index); | |
800 | s->earliest_index = s->latest_index = PRIOQ_IDX_NULL; | |
1e45e3fe LP |
801 | d->needs_rearm = true; |
802 | } | |
803 | ||
a71fe8b8 LP |
804 | static void source_disconnect(sd_event_source *s) { |
805 | sd_event *event; | |
806 | ||
fd38203a LP |
807 | assert(s); |
808 | ||
a71fe8b8 LP |
809 | if (!s->event) |
810 | return; | |
15b38f93 | 811 | |
a71fe8b8 | 812 | assert(s->event->n_sources > 0); |
fd38203a | 813 | |
a71fe8b8 | 814 | switch (s->type) { |
fd38203a | 815 | |
a71fe8b8 LP |
816 | case SOURCE_IO: |
817 | if (s->io.fd >= 0) | |
818 | source_io_unregister(s); | |
fd38203a | 819 | |
a71fe8b8 | 820 | break; |
6a0f1f6d | 821 | |
a71fe8b8 | 822 | case SOURCE_TIME_REALTIME: |
a8548816 | 823 | case SOURCE_TIME_BOOTTIME: |
a71fe8b8 LP |
824 | case SOURCE_TIME_MONOTONIC: |
825 | case SOURCE_TIME_REALTIME_ALARM: | |
b6d5481b LP |
826 | case SOURCE_TIME_BOOTTIME_ALARM: |
827 | /* Only remove this event source from the time event source here if it is not ratelimited. If | |
828 | * it is ratelimited, we'll remove it below, separately. Why? Because the clock used might | |
829 | * differ: ratelimiting always uses CLOCK_MONOTONIC, but timer events might use any clock */ | |
830 | ||
831 | if (!s->ratelimited) { | |
832 | struct clock_data *d; | |
833 | assert_se(d = event_get_clock_data(s->event, s->type)); | |
834 | event_source_time_prioq_remove(s, d); | |
835 | } | |
836 | ||
a71fe8b8 | 837 | break; |
a71fe8b8 LP |
838 | |
839 | case SOURCE_SIGNAL: | |
840 | if (s->signal.sig > 0) { | |
9da4cb2b | 841 | |
a71fe8b8 LP |
842 | if (s->event->signal_sources) |
843 | s->event->signal_sources[s->signal.sig] = NULL; | |
4807d2d0 | 844 | |
9da4cb2b | 845 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
6a0f1f6d | 846 | } |
fd38203a | 847 | |
a71fe8b8 | 848 | break; |
fd38203a | 849 | |
a71fe8b8 LP |
850 | case SOURCE_CHILD: |
851 | if (s->child.pid > 0) { | |
b6d5481b LP |
852 | if (event_source_is_online(s)) { |
853 | assert(s->event->n_online_child_sources > 0); | |
854 | s->event->n_online_child_sources--; | |
4807d2d0 | 855 | } |
fd38203a | 856 | |
4a0b58c4 | 857 | (void) hashmap_remove(s->event->child_sources, PID_TO_PTR(s->child.pid)); |
a71fe8b8 | 858 | } |
fd38203a | 859 | |
f8f3f926 LP |
860 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
861 | source_child_pidfd_unregister(s); | |
862 | else | |
863 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
864 | ||
a71fe8b8 | 865 | break; |
fd38203a | 866 | |
a71fe8b8 LP |
867 | case SOURCE_DEFER: |
868 | /* nothing */ | |
869 | break; | |
fd38203a | 870 | |
a71fe8b8 LP |
871 | case SOURCE_POST: |
872 | set_remove(s->event->post_sources, s); | |
873 | break; | |
da7e457c | 874 | |
a71fe8b8 LP |
875 | case SOURCE_EXIT: |
876 | prioq_remove(s->event->exit, s, &s->exit.prioq_index); | |
877 | break; | |
0eb2e0e3 | 878 | |
97ef5391 LP |
879 | case SOURCE_INOTIFY: { |
880 | struct inode_data *inode_data; | |
881 | ||
882 | inode_data = s->inotify.inode_data; | |
883 | if (inode_data) { | |
884 | struct inotify_data *inotify_data; | |
885 | assert_se(inotify_data = inode_data->inotify_data); | |
886 | ||
887 | /* Detach this event source from the inode object */ | |
888 | LIST_REMOVE(inotify.by_inode_data, inode_data->event_sources, s); | |
889 | s->inotify.inode_data = NULL; | |
890 | ||
891 | if (s->pending) { | |
892 | assert(inotify_data->n_pending > 0); | |
893 | inotify_data->n_pending--; | |
894 | } | |
895 | ||
896 | /* Note that we don't reduce the inotify mask for the watch descriptor here if the inode is | |
897 | * continued to being watched. That's because inotify doesn't really have an API for that: we | |
898 | * can only change watch masks with access to the original inode either by fd or by path. But | |
899 | * paths aren't stable, and keeping an O_PATH fd open all the time would mean wasting an fd | |
f21f31b2 | 900 | * continuously and keeping the mount busy which we can't really do. We could reconstruct the |
97ef5391 LP |
901 | * original inode from /proc/self/fdinfo/$INOTIFY_FD (as all watch descriptors are listed |
902 | * there), but given the need for open_by_handle_at() which is privileged and not universally | |
903 | * available this would be quite an incomplete solution. Hence we go the other way, leave the | |
904 | * mask set, even if it is not minimized now, and ignore all events we aren't interested in | |
905 | * anymore after reception. Yes, this sucks, but … Linux … */ | |
906 | ||
907 | /* Maybe release the inode data (and its inotify) */ | |
908 | event_gc_inode_data(s->event, inode_data); | |
909 | } | |
910 | ||
911 | break; | |
912 | } | |
913 | ||
a71fe8b8 LP |
914 | default: |
915 | assert_not_reached("Wut? I shouldn't exist."); | |
916 | } | |
6e9feda3 | 917 | |
a71fe8b8 LP |
918 | if (s->pending) |
919 | prioq_remove(s->event->pending, s, &s->pending_index); | |
9d3e3aa5 | 920 | |
a71fe8b8 LP |
921 | if (s->prepare) |
922 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
fd38203a | 923 | |
b6d5481b LP |
924 | if (s->ratelimited) |
925 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
926 | ||
e514aa1e | 927 | event = TAKE_PTR(s->event); |
a71fe8b8 LP |
928 | LIST_REMOVE(sources, event->sources, s); |
929 | event->n_sources--; | |
fd38203a | 930 | |
f5982559 LP |
931 | /* Note that we don't invalidate the type here, since we still need it in order to close the fd or |
932 | * pidfd associated with this event source, which we'll do only on source_free(). */ | |
933 | ||
a71fe8b8 LP |
934 | if (!s->floating) |
935 | sd_event_unref(event); | |
936 | } | |
937 | ||
75db809a | 938 | static sd_event_source* source_free(sd_event_source *s) { |
a71fe8b8 | 939 | assert(s); |
fd38203a | 940 | |
a71fe8b8 | 941 | source_disconnect(s); |
ab93297c NM |
942 | |
943 | if (s->type == SOURCE_IO && s->io.owned) | |
15723a1d LP |
944 | s->io.fd = safe_close(s->io.fd); |
945 | ||
f8f3f926 LP |
946 | if (s->type == SOURCE_CHILD) { |
947 | /* Eventually the kernel will do this automatically for us, but for now let's emulate this (unreliably) in userspace. */ | |
948 | ||
949 | if (s->child.process_owned) { | |
950 | ||
951 | if (!s->child.exited) { | |
952 | bool sent = false; | |
953 | ||
954 | if (s->child.pidfd >= 0) { | |
955 | if (pidfd_send_signal(s->child.pidfd, SIGKILL, NULL, 0) < 0) { | |
956 | if (errno == ESRCH) /* Already dead */ | |
957 | sent = true; | |
958 | else if (!ERRNO_IS_NOT_SUPPORTED(errno)) | |
959 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via pidfd_send_signal(), re-trying via kill(): %m", | |
960 | s->child.pid); | |
961 | } else | |
962 | sent = true; | |
963 | } | |
964 | ||
965 | if (!sent) | |
966 | if (kill(s->child.pid, SIGKILL) < 0) | |
967 | if (errno != ESRCH) /* Already dead */ | |
968 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via kill(), ignoring: %m", | |
969 | s->child.pid); | |
970 | } | |
971 | ||
972 | if (!s->child.waited) { | |
973 | siginfo_t si = {}; | |
974 | ||
975 | /* Reap the child if we can */ | |
976 | (void) waitid(P_PID, s->child.pid, &si, WEXITED); | |
977 | } | |
978 | } | |
979 | ||
980 | if (s->child.pidfd_owned) | |
981 | s->child.pidfd = safe_close(s->child.pidfd); | |
982 | } | |
983 | ||
15723a1d LP |
984 | if (s->destroy_callback) |
985 | s->destroy_callback(s->userdata); | |
ab93297c | 986 | |
356779df | 987 | free(s->description); |
75db809a | 988 | return mfree(s); |
fd38203a | 989 | } |
8c75fe17 | 990 | DEFINE_TRIVIAL_CLEANUP_FUNC(sd_event_source*, source_free); |
fd38203a LP |
991 | |
992 | static int source_set_pending(sd_event_source *s, bool b) { | |
993 | int r; | |
994 | ||
995 | assert(s); | |
6203e07a | 996 | assert(s->type != SOURCE_EXIT); |
fd38203a LP |
997 | |
998 | if (s->pending == b) | |
999 | return 0; | |
1000 | ||
1001 | s->pending = b; | |
1002 | ||
1003 | if (b) { | |
1004 | s->pending_iteration = s->event->iteration; | |
1005 | ||
1006 | r = prioq_put(s->event->pending, s, &s->pending_index); | |
1007 | if (r < 0) { | |
1008 | s->pending = false; | |
1009 | return r; | |
1010 | } | |
1011 | } else | |
1012 | assert_se(prioq_remove(s->event->pending, s, &s->pending_index)); | |
1013 | ||
e1951c16 MS |
1014 | if (EVENT_SOURCE_IS_TIME(s->type)) |
1015 | event_source_time_prioq_reshuffle(s); | |
2576a19e | 1016 | |
9da4cb2b LP |
1017 | if (s->type == SOURCE_SIGNAL && !b) { |
1018 | struct signal_data *d; | |
1019 | ||
1020 | d = hashmap_get(s->event->signal_data, &s->priority); | |
1021 | if (d && d->current == s) | |
1022 | d->current = NULL; | |
1023 | } | |
1024 | ||
97ef5391 LP |
1025 | if (s->type == SOURCE_INOTIFY) { |
1026 | ||
1027 | assert(s->inotify.inode_data); | |
1028 | assert(s->inotify.inode_data->inotify_data); | |
1029 | ||
1030 | if (b) | |
1031 | s->inotify.inode_data->inotify_data->n_pending ++; | |
1032 | else { | |
1033 | assert(s->inotify.inode_data->inotify_data->n_pending > 0); | |
1034 | s->inotify.inode_data->inotify_data->n_pending --; | |
1035 | } | |
1036 | } | |
1037 | ||
efd3be9d | 1038 | return 1; |
fd38203a LP |
1039 | } |
1040 | ||
a71fe8b8 | 1041 | static sd_event_source *source_new(sd_event *e, bool floating, EventSourceType type) { |
fd38203a LP |
1042 | sd_event_source *s; |
1043 | ||
1044 | assert(e); | |
1045 | ||
d08eb1fa | 1046 | s = new(sd_event_source, 1); |
fd38203a LP |
1047 | if (!s) |
1048 | return NULL; | |
1049 | ||
d08eb1fa LP |
1050 | *s = (struct sd_event_source) { |
1051 | .n_ref = 1, | |
1052 | .event = e, | |
1053 | .floating = floating, | |
1054 | .type = type, | |
1055 | .pending_index = PRIOQ_IDX_NULL, | |
1056 | .prepare_index = PRIOQ_IDX_NULL, | |
1057 | }; | |
a71fe8b8 LP |
1058 | |
1059 | if (!floating) | |
1060 | sd_event_ref(e); | |
fd38203a | 1061 | |
a71fe8b8 | 1062 | LIST_PREPEND(sources, e->sources, s); |
313cefa1 | 1063 | e->n_sources++; |
15b38f93 | 1064 | |
fd38203a LP |
1065 | return s; |
1066 | } | |
1067 | ||
b9350e70 LP |
1068 | static int io_exit_callback(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
1069 | assert(s); | |
1070 | ||
1071 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1072 | } | |
1073 | ||
f7262a9f | 1074 | _public_ int sd_event_add_io( |
fd38203a | 1075 | sd_event *e, |
151b9b96 | 1076 | sd_event_source **ret, |
fd38203a LP |
1077 | int fd, |
1078 | uint32_t events, | |
718db961 | 1079 | sd_event_io_handler_t callback, |
151b9b96 | 1080 | void *userdata) { |
fd38203a | 1081 | |
ec766a51 | 1082 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1083 | int r; |
1084 | ||
305f78bf | 1085 | assert_return(e, -EINVAL); |
b937d761 | 1086 | assert_return(e = event_resolve(e), -ENOPKG); |
8ac43fee | 1087 | assert_return(fd >= 0, -EBADF); |
2a16a986 | 1088 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 1089 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1090 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1091 | |
b9350e70 LP |
1092 | if (!callback) |
1093 | callback = io_exit_callback; | |
1094 | ||
a71fe8b8 | 1095 | s = source_new(e, !ret, SOURCE_IO); |
fd38203a LP |
1096 | if (!s) |
1097 | return -ENOMEM; | |
1098 | ||
9da4cb2b | 1099 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1100 | s->io.fd = fd; |
1101 | s->io.events = events; | |
1102 | s->io.callback = callback; | |
1103 | s->userdata = userdata; | |
baf76283 | 1104 | s->enabled = SD_EVENT_ON; |
fd38203a | 1105 | |
baf76283 | 1106 | r = source_io_register(s, s->enabled, events); |
ec766a51 | 1107 | if (r < 0) |
050f74f2 | 1108 | return r; |
fd38203a | 1109 | |
a71fe8b8 LP |
1110 | if (ret) |
1111 | *ret = s; | |
ec766a51 | 1112 | TAKE_PTR(s); |
a71fe8b8 | 1113 | |
fd38203a LP |
1114 | return 0; |
1115 | } | |
1116 | ||
52444dc4 LP |
1117 | static void initialize_perturb(sd_event *e) { |
1118 | sd_id128_t bootid = {}; | |
1119 | ||
1120 | /* When we sleep for longer, we try to realign the wakeup to | |
f21f31b2 | 1121 | the same time within each minute/second/250ms, so that |
52444dc4 LP |
1122 | events all across the system can be coalesced into a single |
1123 | CPU wakeup. However, let's take some system-specific | |
1124 | randomness for this value, so that in a network of systems | |
1125 | with synced clocks timer events are distributed a | |
1126 | bit. Here, we calculate a perturbation usec offset from the | |
1127 | boot ID. */ | |
1128 | ||
3a43da28 | 1129 | if (_likely_(e->perturb != USEC_INFINITY)) |
52444dc4 LP |
1130 | return; |
1131 | ||
1132 | if (sd_id128_get_boot(&bootid) >= 0) | |
1133 | e->perturb = (bootid.qwords[0] ^ bootid.qwords[1]) % USEC_PER_MINUTE; | |
1134 | } | |
1135 | ||
fd38203a LP |
1136 | static int event_setup_timer_fd( |
1137 | sd_event *e, | |
6a0f1f6d LP |
1138 | struct clock_data *d, |
1139 | clockid_t clock) { | |
fd38203a | 1140 | |
fd38203a | 1141 | assert(e); |
6a0f1f6d | 1142 | assert(d); |
fd38203a | 1143 | |
6a0f1f6d | 1144 | if (_likely_(d->fd >= 0)) |
fd38203a LP |
1145 | return 0; |
1146 | ||
b44d87e2 | 1147 | _cleanup_close_ int fd = -1; |
b44d87e2 | 1148 | |
6a0f1f6d | 1149 | fd = timerfd_create(clock, TFD_NONBLOCK|TFD_CLOEXEC); |
fd38203a LP |
1150 | if (fd < 0) |
1151 | return -errno; | |
1152 | ||
7fe2903c LP |
1153 | fd = fd_move_above_stdio(fd); |
1154 | ||
1eac7948 | 1155 | struct epoll_event ev = { |
a82f89aa LP |
1156 | .events = EPOLLIN, |
1157 | .data.ptr = d, | |
1158 | }; | |
fd38203a | 1159 | |
15c689d7 | 1160 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) |
fd38203a | 1161 | return -errno; |
fd38203a | 1162 | |
b44d87e2 | 1163 | d->fd = TAKE_FD(fd); |
fd38203a LP |
1164 | return 0; |
1165 | } | |
1166 | ||
c4f1aff2 TG |
1167 | static int time_exit_callback(sd_event_source *s, uint64_t usec, void *userdata) { |
1168 | assert(s); | |
1169 | ||
1170 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1171 | } | |
1172 | ||
41c63f36 LP |
1173 | static int setup_clock_data(sd_event *e, struct clock_data *d, clockid_t clock) { |
1174 | int r; | |
1175 | ||
1176 | assert(d); | |
1177 | ||
1178 | if (d->fd < 0) { | |
1179 | r = event_setup_timer_fd(e, d, clock); | |
1180 | if (r < 0) | |
1181 | return r; | |
1182 | } | |
1183 | ||
1184 | r = prioq_ensure_allocated(&d->earliest, earliest_time_prioq_compare); | |
1185 | if (r < 0) | |
1186 | return r; | |
1187 | ||
1188 | r = prioq_ensure_allocated(&d->latest, latest_time_prioq_compare); | |
1189 | if (r < 0) | |
1190 | return r; | |
1191 | ||
1192 | return 0; | |
1193 | } | |
1194 | ||
1e45e3fe LP |
1195 | static int event_source_time_prioq_put( |
1196 | sd_event_source *s, | |
1197 | struct clock_data *d) { | |
1198 | ||
1199 | int r; | |
1200 | ||
1201 | assert(s); | |
1202 | assert(d); | |
1203 | ||
f41315fc | 1204 | r = prioq_put(d->earliest, s, &s->earliest_index); |
1e45e3fe LP |
1205 | if (r < 0) |
1206 | return r; | |
1207 | ||
f41315fc | 1208 | r = prioq_put(d->latest, s, &s->latest_index); |
1e45e3fe | 1209 | if (r < 0) { |
f41315fc LP |
1210 | assert_se(prioq_remove(d->earliest, s, &s->earliest_index) > 0); |
1211 | s->earliest_index = PRIOQ_IDX_NULL; | |
1e45e3fe LP |
1212 | return r; |
1213 | } | |
1214 | ||
1215 | d->needs_rearm = true; | |
1216 | return 0; | |
1217 | } | |
1218 | ||
6a0f1f6d | 1219 | _public_ int sd_event_add_time( |
fd38203a | 1220 | sd_event *e, |
151b9b96 | 1221 | sd_event_source **ret, |
6a0f1f6d | 1222 | clockid_t clock, |
fd38203a | 1223 | uint64_t usec, |
c2ba3ad6 | 1224 | uint64_t accuracy, |
718db961 | 1225 | sd_event_time_handler_t callback, |
151b9b96 | 1226 | void *userdata) { |
fd38203a | 1227 | |
6a0f1f6d | 1228 | EventSourceType type; |
ec766a51 | 1229 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6a0f1f6d | 1230 | struct clock_data *d; |
fd38203a LP |
1231 | int r; |
1232 | ||
305f78bf | 1233 | assert_return(e, -EINVAL); |
b937d761 | 1234 | assert_return(e = event_resolve(e), -ENOPKG); |
f5fbe71d | 1235 | assert_return(accuracy != UINT64_MAX, -EINVAL); |
da7e457c | 1236 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1237 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1238 | |
e475d10c LP |
1239 | if (!clock_supported(clock)) /* Checks whether the kernel supports the clock */ |
1240 | return -EOPNOTSUPP; | |
1241 | ||
1242 | type = clock_to_event_source_type(clock); /* checks whether sd-event supports this clock */ | |
1243 | if (type < 0) | |
3411372e LP |
1244 | return -EOPNOTSUPP; |
1245 | ||
c4f1aff2 TG |
1246 | if (!callback) |
1247 | callback = time_exit_callback; | |
1248 | ||
1e45e3fe | 1249 | assert_se(d = event_get_clock_data(e, type)); |
c2ba3ad6 | 1250 | |
41c63f36 | 1251 | r = setup_clock_data(e, d, clock); |
c983e776 EV |
1252 | if (r < 0) |
1253 | return r; | |
fd38203a | 1254 | |
a71fe8b8 | 1255 | s = source_new(e, !ret, type); |
fd38203a LP |
1256 | if (!s) |
1257 | return -ENOMEM; | |
1258 | ||
1259 | s->time.next = usec; | |
c2ba3ad6 | 1260 | s->time.accuracy = accuracy == 0 ? DEFAULT_ACCURACY_USEC : accuracy; |
fd38203a | 1261 | s->time.callback = callback; |
f41315fc | 1262 | s->earliest_index = s->latest_index = PRIOQ_IDX_NULL; |
fd38203a | 1263 | s->userdata = userdata; |
baf76283 | 1264 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1265 | |
1e45e3fe | 1266 | r = event_source_time_prioq_put(s, d); |
c2ba3ad6 | 1267 | if (r < 0) |
ec766a51 | 1268 | return r; |
fd38203a | 1269 | |
a71fe8b8 LP |
1270 | if (ret) |
1271 | *ret = s; | |
ec766a51 | 1272 | TAKE_PTR(s); |
a71fe8b8 | 1273 | |
fd38203a LP |
1274 | return 0; |
1275 | } | |
1276 | ||
d6a83dc4 LP |
1277 | _public_ int sd_event_add_time_relative( |
1278 | sd_event *e, | |
1279 | sd_event_source **ret, | |
1280 | clockid_t clock, | |
1281 | uint64_t usec, | |
1282 | uint64_t accuracy, | |
1283 | sd_event_time_handler_t callback, | |
1284 | void *userdata) { | |
1285 | ||
1286 | usec_t t; | |
1287 | int r; | |
1288 | ||
1289 | /* Same as sd_event_add_time() but operates relative to the event loop's current point in time, and | |
1290 | * checks for overflow. */ | |
1291 | ||
1292 | r = sd_event_now(e, clock, &t); | |
1293 | if (r < 0) | |
1294 | return r; | |
1295 | ||
1296 | if (usec >= USEC_INFINITY - t) | |
1297 | return -EOVERFLOW; | |
1298 | ||
1299 | return sd_event_add_time(e, ret, clock, t + usec, accuracy, callback, userdata); | |
1300 | } | |
1301 | ||
59bc1fd7 LP |
1302 | static int signal_exit_callback(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) { |
1303 | assert(s); | |
1304 | ||
1305 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1306 | } | |
1307 | ||
f7262a9f | 1308 | _public_ int sd_event_add_signal( |
305f78bf | 1309 | sd_event *e, |
151b9b96 | 1310 | sd_event_source **ret, |
305f78bf | 1311 | int sig, |
718db961 | 1312 | sd_event_signal_handler_t callback, |
151b9b96 | 1313 | void *userdata) { |
305f78bf | 1314 | |
ec766a51 | 1315 | _cleanup_(source_freep) sd_event_source *s = NULL; |
9da4cb2b | 1316 | struct signal_data *d; |
fd38203a LP |
1317 | int r; |
1318 | ||
305f78bf | 1319 | assert_return(e, -EINVAL); |
b937d761 | 1320 | assert_return(e = event_resolve(e), -ENOPKG); |
6eb7c172 | 1321 | assert_return(SIGNAL_VALID(sig), -EINVAL); |
da7e457c | 1322 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1323 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1324 | |
59bc1fd7 LP |
1325 | if (!callback) |
1326 | callback = signal_exit_callback; | |
1327 | ||
d1b75241 LP |
1328 | r = signal_is_blocked(sig); |
1329 | if (r < 0) | |
1330 | return r; | |
1331 | if (r == 0) | |
3022d74b LP |
1332 | return -EBUSY; |
1333 | ||
fd38203a LP |
1334 | if (!e->signal_sources) { |
1335 | e->signal_sources = new0(sd_event_source*, _NSIG); | |
1336 | if (!e->signal_sources) | |
1337 | return -ENOMEM; | |
1338 | } else if (e->signal_sources[sig]) | |
1339 | return -EBUSY; | |
1340 | ||
a71fe8b8 | 1341 | s = source_new(e, !ret, SOURCE_SIGNAL); |
fd38203a LP |
1342 | if (!s) |
1343 | return -ENOMEM; | |
1344 | ||
1345 | s->signal.sig = sig; | |
1346 | s->signal.callback = callback; | |
1347 | s->userdata = userdata; | |
baf76283 | 1348 | s->enabled = SD_EVENT_ON; |
fd38203a LP |
1349 | |
1350 | e->signal_sources[sig] = s; | |
fd38203a | 1351 | |
9da4cb2b | 1352 | r = event_make_signal_data(e, sig, &d); |
ec766a51 | 1353 | if (r < 0) |
9da4cb2b | 1354 | return r; |
fd38203a | 1355 | |
f1f00dbb LP |
1356 | /* Use the signal name as description for the event source by default */ |
1357 | (void) sd_event_source_set_description(s, signal_to_string(sig)); | |
1358 | ||
a71fe8b8 LP |
1359 | if (ret) |
1360 | *ret = s; | |
ec766a51 | 1361 | TAKE_PTR(s); |
a71fe8b8 | 1362 | |
fd38203a LP |
1363 | return 0; |
1364 | } | |
1365 | ||
b9350e70 LP |
1366 | static int child_exit_callback(sd_event_source *s, const siginfo_t *si, void *userdata) { |
1367 | assert(s); | |
1368 | ||
1369 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1370 | } | |
1371 | ||
f8f3f926 LP |
1372 | static bool shall_use_pidfd(void) { |
1373 | /* Mostly relevant for debugging, i.e. this is used in test-event.c to test the event loop once with and once without pidfd */ | |
1374 | return getenv_bool_secure("SYSTEMD_PIDFD") != 0; | |
1375 | } | |
1376 | ||
f7262a9f | 1377 | _public_ int sd_event_add_child( |
305f78bf | 1378 | sd_event *e, |
151b9b96 | 1379 | sd_event_source **ret, |
305f78bf LP |
1380 | pid_t pid, |
1381 | int options, | |
718db961 | 1382 | sd_event_child_handler_t callback, |
151b9b96 | 1383 | void *userdata) { |
305f78bf | 1384 | |
ec766a51 | 1385 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1386 | int r; |
1387 | ||
305f78bf | 1388 | assert_return(e, -EINVAL); |
b937d761 | 1389 | assert_return(e = event_resolve(e), -ENOPKG); |
305f78bf LP |
1390 | assert_return(pid > 1, -EINVAL); |
1391 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1392 | assert_return(options != 0, -EINVAL); | |
da7e457c | 1393 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1394 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1395 | |
b9350e70 LP |
1396 | if (!callback) |
1397 | callback = child_exit_callback; | |
1398 | ||
b6d5481b | 1399 | if (e->n_online_child_sources == 0) { |
ee880b37 LP |
1400 | /* Caller must block SIGCHLD before using us to watch children, even if pidfd is available, |
1401 | * for compatibility with pre-pidfd and because we don't want the reap the child processes | |
1402 | * ourselves, i.e. call waitid(), and don't want Linux' default internal logic for that to | |
1403 | * take effect. | |
1404 | * | |
1405 | * (As an optimization we only do this check on the first child event source created.) */ | |
1406 | r = signal_is_blocked(SIGCHLD); | |
1407 | if (r < 0) | |
1408 | return r; | |
1409 | if (r == 0) | |
1410 | return -EBUSY; | |
1411 | } | |
1412 | ||
d5099efc | 1413 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
fd38203a LP |
1414 | if (r < 0) |
1415 | return r; | |
1416 | ||
4a0b58c4 | 1417 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) |
fd38203a LP |
1418 | return -EBUSY; |
1419 | ||
a71fe8b8 | 1420 | s = source_new(e, !ret, SOURCE_CHILD); |
fd38203a LP |
1421 | if (!s) |
1422 | return -ENOMEM; | |
1423 | ||
f8f3f926 | 1424 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1425 | s->child.pid = pid; |
1426 | s->child.options = options; | |
1427 | s->child.callback = callback; | |
1428 | s->userdata = userdata; | |
baf76283 | 1429 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1430 | |
f8f3f926 LP |
1431 | /* We always take a pidfd here if we can, even if we wait for anything else than WEXITED, so that we |
1432 | * pin the PID, and make regular waitid() handling race-free. */ | |
1433 | ||
1434 | if (shall_use_pidfd()) { | |
1435 | s->child.pidfd = pidfd_open(s->child.pid, 0); | |
1436 | if (s->child.pidfd < 0) { | |
1437 | /* Propagate errors unless the syscall is not supported or blocked */ | |
1438 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
1439 | return -errno; | |
1440 | } else | |
1441 | s->child.pidfd_owned = true; /* If we allocate the pidfd we own it by default */ | |
1442 | } else | |
1443 | s->child.pidfd = -1; | |
1444 | ||
4a0b58c4 | 1445 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); |
ec766a51 | 1446 | if (r < 0) |
fd38203a | 1447 | return r; |
fd38203a | 1448 | |
f8f3f926 LP |
1449 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1450 | /* We have a pidfd and we only want to watch for exit */ | |
f8f3f926 | 1451 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1452 | if (r < 0) |
f8f3f926 | 1453 | return r; |
ac9f2640 | 1454 | |
f8f3f926 LP |
1455 | } else { |
1456 | /* We have no pidfd or we shall wait for some other event than WEXITED */ | |
f8f3f926 | 1457 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1458 | if (r < 0) |
f8f3f926 | 1459 | return r; |
f8f3f926 LP |
1460 | |
1461 | e->need_process_child = true; | |
1462 | } | |
c2ba3ad6 | 1463 | |
b6d5481b | 1464 | e->n_online_child_sources++; |
ac9f2640 | 1465 | |
a71fe8b8 LP |
1466 | if (ret) |
1467 | *ret = s; | |
ec766a51 | 1468 | TAKE_PTR(s); |
f8f3f926 LP |
1469 | return 0; |
1470 | } | |
1471 | ||
1472 | _public_ int sd_event_add_child_pidfd( | |
1473 | sd_event *e, | |
1474 | sd_event_source **ret, | |
1475 | int pidfd, | |
1476 | int options, | |
1477 | sd_event_child_handler_t callback, | |
1478 | void *userdata) { | |
1479 | ||
1480 | ||
1481 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
1482 | pid_t pid; | |
1483 | int r; | |
1484 | ||
1485 | assert_return(e, -EINVAL); | |
1486 | assert_return(e = event_resolve(e), -ENOPKG); | |
1487 | assert_return(pidfd >= 0, -EBADF); | |
1488 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1489 | assert_return(options != 0, -EINVAL); | |
f8f3f926 LP |
1490 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1491 | assert_return(!event_pid_changed(e), -ECHILD); | |
1492 | ||
b9350e70 LP |
1493 | if (!callback) |
1494 | callback = child_exit_callback; | |
1495 | ||
b6d5481b | 1496 | if (e->n_online_child_sources == 0) { |
ee880b37 LP |
1497 | r = signal_is_blocked(SIGCHLD); |
1498 | if (r < 0) | |
1499 | return r; | |
1500 | if (r == 0) | |
1501 | return -EBUSY; | |
1502 | } | |
1503 | ||
f8f3f926 LP |
1504 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
1505 | if (r < 0) | |
1506 | return r; | |
1507 | ||
1508 | r = pidfd_get_pid(pidfd, &pid); | |
1509 | if (r < 0) | |
1510 | return r; | |
1511 | ||
1512 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) | |
1513 | return -EBUSY; | |
1514 | ||
1515 | s = source_new(e, !ret, SOURCE_CHILD); | |
1516 | if (!s) | |
1517 | return -ENOMEM; | |
1518 | ||
1519 | s->wakeup = WAKEUP_EVENT_SOURCE; | |
1520 | s->child.pidfd = pidfd; | |
1521 | s->child.pid = pid; | |
1522 | s->child.options = options; | |
1523 | s->child.callback = callback; | |
1524 | 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) */ | |
1525 | s->userdata = userdata; | |
1526 | s->enabled = SD_EVENT_ONESHOT; | |
1527 | ||
1528 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); | |
1529 | if (r < 0) | |
1530 | return r; | |
1531 | ||
f8f3f926 LP |
1532 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1533 | /* We only want to watch for WEXITED */ | |
f8f3f926 | 1534 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1535 | if (r < 0) |
f8f3f926 | 1536 | return r; |
f8f3f926 LP |
1537 | } else { |
1538 | /* We shall wait for some other event than WEXITED */ | |
f8f3f926 | 1539 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1540 | if (r < 0) |
f8f3f926 | 1541 | return r; |
a71fe8b8 | 1542 | |
f8f3f926 LP |
1543 | e->need_process_child = true; |
1544 | } | |
1545 | ||
b6d5481b | 1546 | e->n_online_child_sources++; |
ac9f2640 | 1547 | |
f8f3f926 LP |
1548 | if (ret) |
1549 | *ret = s; | |
f8f3f926 | 1550 | TAKE_PTR(s); |
fd38203a LP |
1551 | return 0; |
1552 | } | |
1553 | ||
b9350e70 LP |
1554 | static int generic_exit_callback(sd_event_source *s, void *userdata) { |
1555 | assert(s); | |
1556 | ||
1557 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1558 | } | |
1559 | ||
f7262a9f | 1560 | _public_ int sd_event_add_defer( |
305f78bf | 1561 | sd_event *e, |
151b9b96 | 1562 | sd_event_source **ret, |
718db961 | 1563 | sd_event_handler_t callback, |
151b9b96 | 1564 | void *userdata) { |
305f78bf | 1565 | |
ec766a51 | 1566 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1567 | int r; |
1568 | ||
305f78bf | 1569 | assert_return(e, -EINVAL); |
b937d761 | 1570 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1571 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1572 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1573 | |
b9350e70 LP |
1574 | if (!callback) |
1575 | callback = generic_exit_callback; | |
1576 | ||
a71fe8b8 | 1577 | s = source_new(e, !ret, SOURCE_DEFER); |
fd38203a LP |
1578 | if (!s) |
1579 | return -ENOMEM; | |
1580 | ||
1581 | s->defer.callback = callback; | |
1582 | s->userdata = userdata; | |
baf76283 | 1583 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a LP |
1584 | |
1585 | r = source_set_pending(s, true); | |
ec766a51 | 1586 | if (r < 0) |
fd38203a | 1587 | return r; |
fd38203a | 1588 | |
a71fe8b8 LP |
1589 | if (ret) |
1590 | *ret = s; | |
ec766a51 | 1591 | TAKE_PTR(s); |
a71fe8b8 | 1592 | |
fd38203a LP |
1593 | return 0; |
1594 | } | |
1595 | ||
6e9feda3 LP |
1596 | _public_ int sd_event_add_post( |
1597 | sd_event *e, | |
1598 | sd_event_source **ret, | |
1599 | sd_event_handler_t callback, | |
1600 | void *userdata) { | |
1601 | ||
ec766a51 | 1602 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6e9feda3 LP |
1603 | int r; |
1604 | ||
1605 | assert_return(e, -EINVAL); | |
b937d761 | 1606 | assert_return(e = event_resolve(e), -ENOPKG); |
6e9feda3 LP |
1607 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1608 | assert_return(!event_pid_changed(e), -ECHILD); | |
1609 | ||
b9350e70 LP |
1610 | if (!callback) |
1611 | callback = generic_exit_callback; | |
1612 | ||
a71fe8b8 | 1613 | s = source_new(e, !ret, SOURCE_POST); |
6e9feda3 LP |
1614 | if (!s) |
1615 | return -ENOMEM; | |
1616 | ||
1617 | s->post.callback = callback; | |
1618 | s->userdata = userdata; | |
1619 | s->enabled = SD_EVENT_ON; | |
1620 | ||
de7fef4b | 1621 | r = set_ensure_put(&e->post_sources, NULL, s); |
ec766a51 | 1622 | if (r < 0) |
6e9feda3 | 1623 | return r; |
de7fef4b | 1624 | assert(r > 0); |
6e9feda3 | 1625 | |
a71fe8b8 LP |
1626 | if (ret) |
1627 | *ret = s; | |
ec766a51 | 1628 | TAKE_PTR(s); |
a71fe8b8 | 1629 | |
6e9feda3 LP |
1630 | return 0; |
1631 | } | |
1632 | ||
6203e07a | 1633 | _public_ int sd_event_add_exit( |
305f78bf | 1634 | sd_event *e, |
151b9b96 | 1635 | sd_event_source **ret, |
718db961 | 1636 | sd_event_handler_t callback, |
151b9b96 | 1637 | void *userdata) { |
305f78bf | 1638 | |
ec766a51 | 1639 | _cleanup_(source_freep) sd_event_source *s = NULL; |
da7e457c LP |
1640 | int r; |
1641 | ||
1642 | assert_return(e, -EINVAL); | |
b937d761 | 1643 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
1644 | assert_return(callback, -EINVAL); |
1645 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
1646 | assert_return(!event_pid_changed(e), -ECHILD); | |
1647 | ||
c983e776 EV |
1648 | r = prioq_ensure_allocated(&e->exit, exit_prioq_compare); |
1649 | if (r < 0) | |
1650 | return r; | |
da7e457c | 1651 | |
a71fe8b8 | 1652 | s = source_new(e, !ret, SOURCE_EXIT); |
fd38203a | 1653 | if (!s) |
da7e457c | 1654 | return -ENOMEM; |
fd38203a | 1655 | |
6203e07a | 1656 | s->exit.callback = callback; |
da7e457c | 1657 | s->userdata = userdata; |
6203e07a | 1658 | s->exit.prioq_index = PRIOQ_IDX_NULL; |
baf76283 | 1659 | s->enabled = SD_EVENT_ONESHOT; |
da7e457c | 1660 | |
6203e07a | 1661 | r = prioq_put(s->event->exit, s, &s->exit.prioq_index); |
ec766a51 | 1662 | if (r < 0) |
da7e457c | 1663 | return r; |
da7e457c | 1664 | |
a71fe8b8 LP |
1665 | if (ret) |
1666 | *ret = s; | |
ec766a51 | 1667 | TAKE_PTR(s); |
a71fe8b8 | 1668 | |
da7e457c LP |
1669 | return 0; |
1670 | } | |
1671 | ||
97ef5391 LP |
1672 | static void event_free_inotify_data(sd_event *e, struct inotify_data *d) { |
1673 | assert(e); | |
1674 | ||
1675 | if (!d) | |
1676 | return; | |
1677 | ||
1678 | assert(hashmap_isempty(d->inodes)); | |
1679 | assert(hashmap_isempty(d->wd)); | |
1680 | ||
1681 | if (d->buffer_filled > 0) | |
1682 | LIST_REMOVE(buffered, e->inotify_data_buffered, d); | |
1683 | ||
1684 | hashmap_free(d->inodes); | |
1685 | hashmap_free(d->wd); | |
1686 | ||
1687 | assert_se(hashmap_remove(e->inotify_data, &d->priority) == d); | |
1688 | ||
1689 | if (d->fd >= 0) { | |
1690 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, d->fd, NULL) < 0) | |
1691 | log_debug_errno(errno, "Failed to remove inotify fd from epoll, ignoring: %m"); | |
1692 | ||
1693 | safe_close(d->fd); | |
1694 | } | |
1695 | free(d); | |
1696 | } | |
1697 | ||
1698 | static int event_make_inotify_data( | |
1699 | sd_event *e, | |
1700 | int64_t priority, | |
1701 | struct inotify_data **ret) { | |
1702 | ||
1703 | _cleanup_close_ int fd = -1; | |
1704 | struct inotify_data *d; | |
97ef5391 LP |
1705 | int r; |
1706 | ||
1707 | assert(e); | |
1708 | ||
1709 | d = hashmap_get(e->inotify_data, &priority); | |
1710 | if (d) { | |
1711 | if (ret) | |
1712 | *ret = d; | |
1713 | return 0; | |
1714 | } | |
1715 | ||
1716 | fd = inotify_init1(IN_NONBLOCK|O_CLOEXEC); | |
1717 | if (fd < 0) | |
1718 | return -errno; | |
1719 | ||
1720 | fd = fd_move_above_stdio(fd); | |
1721 | ||
97ef5391 LP |
1722 | d = new(struct inotify_data, 1); |
1723 | if (!d) | |
1724 | return -ENOMEM; | |
1725 | ||
1726 | *d = (struct inotify_data) { | |
1727 | .wakeup = WAKEUP_INOTIFY_DATA, | |
1728 | .fd = TAKE_FD(fd), | |
1729 | .priority = priority, | |
1730 | }; | |
1731 | ||
c2484a75 | 1732 | r = hashmap_ensure_put(&e->inotify_data, &uint64_hash_ops, &d->priority, d); |
97ef5391 LP |
1733 | if (r < 0) { |
1734 | d->fd = safe_close(d->fd); | |
1735 | free(d); | |
1736 | return r; | |
1737 | } | |
1738 | ||
1eac7948 | 1739 | struct epoll_event ev = { |
97ef5391 LP |
1740 | .events = EPOLLIN, |
1741 | .data.ptr = d, | |
1742 | }; | |
1743 | ||
1744 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { | |
1745 | r = -errno; | |
1746 | d->fd = safe_close(d->fd); /* let's close this ourselves, as event_free_inotify_data() would otherwise | |
1747 | * remove the fd from the epoll first, which we don't want as we couldn't | |
1748 | * add it in the first place. */ | |
1749 | event_free_inotify_data(e, d); | |
1750 | return r; | |
1751 | } | |
1752 | ||
1753 | if (ret) | |
1754 | *ret = d; | |
1755 | ||
1756 | return 1; | |
1757 | } | |
1758 | ||
7a08d314 | 1759 | static int inode_data_compare(const struct inode_data *x, const struct inode_data *y) { |
90c88092 | 1760 | int r; |
97ef5391 LP |
1761 | |
1762 | assert(x); | |
1763 | assert(y); | |
1764 | ||
90c88092 YW |
1765 | r = CMP(x->dev, y->dev); |
1766 | if (r != 0) | |
1767 | return r; | |
97ef5391 | 1768 | |
6dd91b36 | 1769 | return CMP(x->ino, y->ino); |
97ef5391 LP |
1770 | } |
1771 | ||
7a08d314 YW |
1772 | static void inode_data_hash_func(const struct inode_data *d, struct siphash *state) { |
1773 | assert(d); | |
97ef5391 LP |
1774 | |
1775 | siphash24_compress(&d->dev, sizeof(d->dev), state); | |
1776 | siphash24_compress(&d->ino, sizeof(d->ino), state); | |
1777 | } | |
1778 | ||
7a08d314 | 1779 | DEFINE_PRIVATE_HASH_OPS(inode_data_hash_ops, struct inode_data, inode_data_hash_func, inode_data_compare); |
97ef5391 LP |
1780 | |
1781 | static void event_free_inode_data( | |
1782 | sd_event *e, | |
1783 | struct inode_data *d) { | |
1784 | ||
1785 | assert(e); | |
1786 | ||
1787 | if (!d) | |
1788 | return; | |
1789 | ||
1790 | assert(!d->event_sources); | |
1791 | ||
1792 | if (d->fd >= 0) { | |
1793 | LIST_REMOVE(to_close, e->inode_data_to_close, d); | |
1794 | safe_close(d->fd); | |
1795 | } | |
1796 | ||
1797 | if (d->inotify_data) { | |
1798 | ||
1799 | if (d->wd >= 0) { | |
1800 | if (d->inotify_data->fd >= 0) { | |
1801 | /* So here's a problem. At the time this runs the watch descriptor might already be | |
1802 | * invalidated, because an IN_IGNORED event might be queued right the moment we enter | |
1803 | * the syscall. Hence, whenever we get EINVAL, ignore it entirely, since it's a very | |
1804 | * likely case to happen. */ | |
1805 | ||
1806 | if (inotify_rm_watch(d->inotify_data->fd, d->wd) < 0 && errno != EINVAL) | |
1807 | log_debug_errno(errno, "Failed to remove watch descriptor %i from inotify, ignoring: %m", d->wd); | |
1808 | } | |
1809 | ||
1810 | assert_se(hashmap_remove(d->inotify_data->wd, INT_TO_PTR(d->wd)) == d); | |
1811 | } | |
1812 | ||
1813 | assert_se(hashmap_remove(d->inotify_data->inodes, d) == d); | |
1814 | } | |
1815 | ||
1816 | free(d); | |
1817 | } | |
1818 | ||
1819 | static void event_gc_inode_data( | |
1820 | sd_event *e, | |
1821 | struct inode_data *d) { | |
1822 | ||
1823 | struct inotify_data *inotify_data; | |
1824 | ||
1825 | assert(e); | |
1826 | ||
1827 | if (!d) | |
1828 | return; | |
1829 | ||
1830 | if (d->event_sources) | |
1831 | return; | |
1832 | ||
1833 | inotify_data = d->inotify_data; | |
1834 | event_free_inode_data(e, d); | |
1835 | ||
1836 | if (inotify_data && hashmap_isempty(inotify_data->inodes)) | |
1837 | event_free_inotify_data(e, inotify_data); | |
1838 | } | |
1839 | ||
1840 | static int event_make_inode_data( | |
1841 | sd_event *e, | |
1842 | struct inotify_data *inotify_data, | |
1843 | dev_t dev, | |
1844 | ino_t ino, | |
1845 | struct inode_data **ret) { | |
1846 | ||
1847 | struct inode_data *d, key; | |
1848 | int r; | |
1849 | ||
1850 | assert(e); | |
1851 | assert(inotify_data); | |
1852 | ||
1853 | key = (struct inode_data) { | |
1854 | .ino = ino, | |
1855 | .dev = dev, | |
1856 | }; | |
1857 | ||
1858 | d = hashmap_get(inotify_data->inodes, &key); | |
1859 | if (d) { | |
1860 | if (ret) | |
1861 | *ret = d; | |
1862 | ||
1863 | return 0; | |
1864 | } | |
1865 | ||
1866 | r = hashmap_ensure_allocated(&inotify_data->inodes, &inode_data_hash_ops); | |
1867 | if (r < 0) | |
1868 | return r; | |
1869 | ||
1870 | d = new(struct inode_data, 1); | |
1871 | if (!d) | |
1872 | return -ENOMEM; | |
1873 | ||
1874 | *d = (struct inode_data) { | |
1875 | .dev = dev, | |
1876 | .ino = ino, | |
1877 | .wd = -1, | |
1878 | .fd = -1, | |
1879 | .inotify_data = inotify_data, | |
1880 | }; | |
1881 | ||
1882 | r = hashmap_put(inotify_data->inodes, d, d); | |
1883 | if (r < 0) { | |
1884 | free(d); | |
1885 | return r; | |
1886 | } | |
1887 | ||
1888 | if (ret) | |
1889 | *ret = d; | |
1890 | ||
1891 | return 1; | |
1892 | } | |
1893 | ||
1894 | static uint32_t inode_data_determine_mask(struct inode_data *d) { | |
1895 | bool excl_unlink = true; | |
1896 | uint32_t combined = 0; | |
1897 | sd_event_source *s; | |
1898 | ||
1899 | assert(d); | |
1900 | ||
1901 | /* Combines the watch masks of all event sources watching this inode. We generally just OR them together, but | |
1902 | * the IN_EXCL_UNLINK flag is ANDed instead. | |
1903 | * | |
1904 | * Note that we add all sources to the mask here, regardless whether enabled, disabled or oneshot. That's | |
1905 | * because we cannot change the mask anymore after the event source was created once, since the kernel has no | |
f21f31b2 | 1906 | * API for that. Hence we need to subscribe to the maximum mask we ever might be interested in, and suppress |
97ef5391 LP |
1907 | * events we don't care for client-side. */ |
1908 | ||
1909 | LIST_FOREACH(inotify.by_inode_data, s, d->event_sources) { | |
1910 | ||
1911 | if ((s->inotify.mask & IN_EXCL_UNLINK) == 0) | |
1912 | excl_unlink = false; | |
1913 | ||
1914 | combined |= s->inotify.mask; | |
1915 | } | |
1916 | ||
1917 | return (combined & ~(IN_ONESHOT|IN_DONT_FOLLOW|IN_ONLYDIR|IN_EXCL_UNLINK)) | (excl_unlink ? IN_EXCL_UNLINK : 0); | |
1918 | } | |
1919 | ||
1920 | static int inode_data_realize_watch(sd_event *e, struct inode_data *d) { | |
1921 | uint32_t combined_mask; | |
1922 | int wd, r; | |
1923 | ||
1924 | assert(d); | |
1925 | assert(d->fd >= 0); | |
1926 | ||
1927 | combined_mask = inode_data_determine_mask(d); | |
1928 | ||
1929 | if (d->wd >= 0 && combined_mask == d->combined_mask) | |
1930 | return 0; | |
1931 | ||
1932 | r = hashmap_ensure_allocated(&d->inotify_data->wd, NULL); | |
1933 | if (r < 0) | |
1934 | return r; | |
1935 | ||
1936 | wd = inotify_add_watch_fd(d->inotify_data->fd, d->fd, combined_mask); | |
1937 | if (wd < 0) | |
1938 | return -errno; | |
1939 | ||
1940 | if (d->wd < 0) { | |
1941 | r = hashmap_put(d->inotify_data->wd, INT_TO_PTR(wd), d); | |
1942 | if (r < 0) { | |
1943 | (void) inotify_rm_watch(d->inotify_data->fd, wd); | |
1944 | return r; | |
1945 | } | |
1946 | ||
1947 | d->wd = wd; | |
1948 | ||
1949 | } else if (d->wd != wd) { | |
1950 | ||
1951 | log_debug("Weird, the watch descriptor we already knew for this inode changed?"); | |
1952 | (void) inotify_rm_watch(d->fd, wd); | |
1953 | return -EINVAL; | |
1954 | } | |
1955 | ||
1956 | d->combined_mask = combined_mask; | |
1957 | return 1; | |
1958 | } | |
1959 | ||
b9350e70 LP |
1960 | static int inotify_exit_callback(sd_event_source *s, const struct inotify_event *event, void *userdata) { |
1961 | assert(s); | |
1962 | ||
1963 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1964 | } | |
1965 | ||
97ef5391 LP |
1966 | _public_ int sd_event_add_inotify( |
1967 | sd_event *e, | |
1968 | sd_event_source **ret, | |
1969 | const char *path, | |
1970 | uint32_t mask, | |
1971 | sd_event_inotify_handler_t callback, | |
1972 | void *userdata) { | |
1973 | ||
97ef5391 LP |
1974 | struct inotify_data *inotify_data = NULL; |
1975 | struct inode_data *inode_data = NULL; | |
1976 | _cleanup_close_ int fd = -1; | |
8c75fe17 | 1977 | _cleanup_(source_freep) sd_event_source *s = NULL; |
97ef5391 LP |
1978 | struct stat st; |
1979 | int r; | |
1980 | ||
1981 | assert_return(e, -EINVAL); | |
1982 | assert_return(e = event_resolve(e), -ENOPKG); | |
1983 | assert_return(path, -EINVAL); | |
97ef5391 LP |
1984 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1985 | assert_return(!event_pid_changed(e), -ECHILD); | |
1986 | ||
b9350e70 LP |
1987 | if (!callback) |
1988 | callback = inotify_exit_callback; | |
1989 | ||
97ef5391 LP |
1990 | /* Refuse IN_MASK_ADD since we coalesce watches on the same inode, and hence really don't want to merge |
1991 | * masks. Or in other words, this whole code exists only to manage IN_MASK_ADD type operations for you, hence | |
1992 | * the user can't use them for us. */ | |
1993 | if (mask & IN_MASK_ADD) | |
1994 | return -EINVAL; | |
1995 | ||
1996 | fd = open(path, O_PATH|O_CLOEXEC| | |
1997 | (mask & IN_ONLYDIR ? O_DIRECTORY : 0)| | |
1998 | (mask & IN_DONT_FOLLOW ? O_NOFOLLOW : 0)); | |
1999 | if (fd < 0) | |
2000 | return -errno; | |
2001 | ||
2002 | if (fstat(fd, &st) < 0) | |
2003 | return -errno; | |
2004 | ||
2005 | s = source_new(e, !ret, SOURCE_INOTIFY); | |
2006 | if (!s) | |
2007 | return -ENOMEM; | |
2008 | ||
2009 | s->enabled = mask & IN_ONESHOT ? SD_EVENT_ONESHOT : SD_EVENT_ON; | |
2010 | s->inotify.mask = mask; | |
2011 | s->inotify.callback = callback; | |
2012 | s->userdata = userdata; | |
2013 | ||
2014 | /* Allocate an inotify object for this priority, and an inode object within it */ | |
2015 | r = event_make_inotify_data(e, SD_EVENT_PRIORITY_NORMAL, &inotify_data); | |
2016 | if (r < 0) | |
8c75fe17 | 2017 | return r; |
97ef5391 LP |
2018 | |
2019 | r = event_make_inode_data(e, inotify_data, st.st_dev, st.st_ino, &inode_data); | |
8c75fe17 ZJS |
2020 | if (r < 0) { |
2021 | event_free_inotify_data(e, inotify_data); | |
2022 | return r; | |
2023 | } | |
97ef5391 LP |
2024 | |
2025 | /* Keep the O_PATH fd around until the first iteration of the loop, so that we can still change the priority of | |
2026 | * the event source, until then, for which we need the original inode. */ | |
2027 | if (inode_data->fd < 0) { | |
2028 | inode_data->fd = TAKE_FD(fd); | |
2029 | LIST_PREPEND(to_close, e->inode_data_to_close, inode_data); | |
2030 | } | |
2031 | ||
2032 | /* Link our event source to the inode data object */ | |
2033 | LIST_PREPEND(inotify.by_inode_data, inode_data->event_sources, s); | |
2034 | s->inotify.inode_data = inode_data; | |
2035 | ||
97ef5391 LP |
2036 | /* Actually realize the watch now */ |
2037 | r = inode_data_realize_watch(e, inode_data); | |
2038 | if (r < 0) | |
8c75fe17 | 2039 | return r; |
97ef5391 LP |
2040 | |
2041 | (void) sd_event_source_set_description(s, path); | |
2042 | ||
2043 | if (ret) | |
2044 | *ret = s; | |
8c75fe17 | 2045 | TAKE_PTR(s); |
97ef5391 LP |
2046 | |
2047 | return 0; | |
97ef5391 LP |
2048 | } |
2049 | ||
8301aa0b | 2050 | static sd_event_source* event_source_free(sd_event_source *s) { |
6680dd6b LP |
2051 | if (!s) |
2052 | return NULL; | |
da7e457c | 2053 | |
8301aa0b YW |
2054 | /* Here's a special hack: when we are called from a |
2055 | * dispatch handler we won't free the event source | |
2056 | * immediately, but we will detach the fd from the | |
2057 | * epoll. This way it is safe for the caller to unref | |
2058 | * the event source and immediately close the fd, but | |
2059 | * we still retain a valid event source object after | |
2060 | * the callback. */ | |
fd38203a | 2061 | |
8301aa0b YW |
2062 | if (s->dispatching) { |
2063 | if (s->type == SOURCE_IO) | |
2064 | source_io_unregister(s); | |
fd38203a | 2065 | |
8301aa0b YW |
2066 | source_disconnect(s); |
2067 | } else | |
2068 | source_free(s); | |
fd38203a LP |
2069 | |
2070 | return NULL; | |
2071 | } | |
2072 | ||
8301aa0b YW |
2073 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event_source, sd_event_source, event_source_free); |
2074 | ||
356779df | 2075 | _public_ int sd_event_source_set_description(sd_event_source *s, const char *description) { |
f7f53e9e | 2076 | assert_return(s, -EINVAL); |
f4b2933e | 2077 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 2078 | |
356779df | 2079 | return free_and_strdup(&s->description, description); |
f7f53e9e TG |
2080 | } |
2081 | ||
356779df | 2082 | _public_ int sd_event_source_get_description(sd_event_source *s, const char **description) { |
f7f53e9e | 2083 | assert_return(s, -EINVAL); |
356779df | 2084 | assert_return(description, -EINVAL); |
f4b2933e | 2085 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 2086 | |
7d92a1a4 ZJS |
2087 | if (!s->description) |
2088 | return -ENXIO; | |
2089 | ||
356779df | 2090 | *description = s->description; |
f7f53e9e TG |
2091 | return 0; |
2092 | } | |
2093 | ||
adcc4ca3 | 2094 | _public_ sd_event *sd_event_source_get_event(sd_event_source *s) { |
305f78bf | 2095 | assert_return(s, NULL); |
eaa3cbef LP |
2096 | |
2097 | return s->event; | |
2098 | } | |
2099 | ||
f7262a9f | 2100 | _public_ int sd_event_source_get_pending(sd_event_source *s) { |
305f78bf | 2101 | assert_return(s, -EINVAL); |
6203e07a | 2102 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c | 2103 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2104 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2105 | |
2106 | return s->pending; | |
2107 | } | |
2108 | ||
f7262a9f | 2109 | _public_ int sd_event_source_get_io_fd(sd_event_source *s) { |
305f78bf LP |
2110 | assert_return(s, -EINVAL); |
2111 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2112 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2113 | |
2114 | return s->io.fd; | |
2115 | } | |
2116 | ||
30caf8f3 LP |
2117 | _public_ int sd_event_source_set_io_fd(sd_event_source *s, int fd) { |
2118 | int r; | |
2119 | ||
2120 | assert_return(s, -EINVAL); | |
8ac43fee | 2121 | assert_return(fd >= 0, -EBADF); |
30caf8f3 LP |
2122 | assert_return(s->type == SOURCE_IO, -EDOM); |
2123 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2124 | ||
2125 | if (s->io.fd == fd) | |
2126 | return 0; | |
2127 | ||
b6d5481b | 2128 | if (event_source_is_offline(s)) { |
30caf8f3 LP |
2129 | s->io.fd = fd; |
2130 | s->io.registered = false; | |
2131 | } else { | |
2132 | int saved_fd; | |
2133 | ||
2134 | saved_fd = s->io.fd; | |
2135 | assert(s->io.registered); | |
2136 | ||
2137 | s->io.fd = fd; | |
2138 | s->io.registered = false; | |
2139 | ||
2140 | r = source_io_register(s, s->enabled, s->io.events); | |
2141 | if (r < 0) { | |
2142 | s->io.fd = saved_fd; | |
2143 | s->io.registered = true; | |
2144 | return r; | |
2145 | } | |
2146 | ||
5a795bff | 2147 | (void) epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, saved_fd, NULL); |
30caf8f3 LP |
2148 | } |
2149 | ||
2150 | return 0; | |
2151 | } | |
2152 | ||
ab93297c NM |
2153 | _public_ int sd_event_source_get_io_fd_own(sd_event_source *s) { |
2154 | assert_return(s, -EINVAL); | |
2155 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2156 | ||
2157 | return s->io.owned; | |
2158 | } | |
2159 | ||
2160 | _public_ int sd_event_source_set_io_fd_own(sd_event_source *s, int own) { | |
2161 | assert_return(s, -EINVAL); | |
2162 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2163 | ||
2164 | s->io.owned = own; | |
2165 | return 0; | |
2166 | } | |
2167 | ||
f7262a9f | 2168 | _public_ int sd_event_source_get_io_events(sd_event_source *s, uint32_t* events) { |
305f78bf LP |
2169 | assert_return(s, -EINVAL); |
2170 | assert_return(events, -EINVAL); | |
2171 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2172 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2173 | |
2174 | *events = s->io.events; | |
2175 | return 0; | |
2176 | } | |
2177 | ||
f7262a9f | 2178 | _public_ int sd_event_source_set_io_events(sd_event_source *s, uint32_t events) { |
fd38203a LP |
2179 | int r; |
2180 | ||
305f78bf LP |
2181 | assert_return(s, -EINVAL); |
2182 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2a16a986 | 2183 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 2184 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2185 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2186 | |
b63c8d4f DH |
2187 | /* edge-triggered updates are never skipped, so we can reset edges */ |
2188 | if (s->io.events == events && !(events & EPOLLET)) | |
fd38203a LP |
2189 | return 0; |
2190 | ||
2a0dc6cd LP |
2191 | r = source_set_pending(s, false); |
2192 | if (r < 0) | |
2193 | return r; | |
2194 | ||
b6d5481b | 2195 | if (event_source_is_online(s)) { |
e4715127 | 2196 | r = source_io_register(s, s->enabled, events); |
fd38203a LP |
2197 | if (r < 0) |
2198 | return r; | |
2199 | } | |
2200 | ||
2201 | s->io.events = events; | |
2202 | ||
2203 | return 0; | |
2204 | } | |
2205 | ||
f7262a9f | 2206 | _public_ int sd_event_source_get_io_revents(sd_event_source *s, uint32_t* revents) { |
305f78bf LP |
2207 | assert_return(s, -EINVAL); |
2208 | assert_return(revents, -EINVAL); | |
2209 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2210 | assert_return(s->pending, -ENODATA); | |
2211 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2212 | |
2213 | *revents = s->io.revents; | |
2214 | return 0; | |
2215 | } | |
2216 | ||
f7262a9f | 2217 | _public_ int sd_event_source_get_signal(sd_event_source *s) { |
305f78bf LP |
2218 | assert_return(s, -EINVAL); |
2219 | assert_return(s->type == SOURCE_SIGNAL, -EDOM); | |
2220 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2221 | |
2222 | return s->signal.sig; | |
2223 | } | |
2224 | ||
31927c16 | 2225 | _public_ int sd_event_source_get_priority(sd_event_source *s, int64_t *priority) { |
305f78bf LP |
2226 | assert_return(s, -EINVAL); |
2227 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a | 2228 | |
6680b8d1 ME |
2229 | *priority = s->priority; |
2230 | return 0; | |
fd38203a LP |
2231 | } |
2232 | ||
31927c16 | 2233 | _public_ int sd_event_source_set_priority(sd_event_source *s, int64_t priority) { |
97ef5391 LP |
2234 | bool rm_inotify = false, rm_inode = false; |
2235 | struct inotify_data *new_inotify_data = NULL; | |
2236 | struct inode_data *new_inode_data = NULL; | |
9da4cb2b LP |
2237 | int r; |
2238 | ||
305f78bf | 2239 | assert_return(s, -EINVAL); |
da7e457c | 2240 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2241 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2242 | |
2243 | if (s->priority == priority) | |
2244 | return 0; | |
2245 | ||
97ef5391 LP |
2246 | if (s->type == SOURCE_INOTIFY) { |
2247 | struct inode_data *old_inode_data; | |
2248 | ||
2249 | assert(s->inotify.inode_data); | |
2250 | old_inode_data = s->inotify.inode_data; | |
2251 | ||
2252 | /* We need the original fd to change the priority. If we don't have it we can't change the priority, | |
2253 | * anymore. Note that we close any fds when entering the next event loop iteration, i.e. for inotify | |
2254 | * events we allow priority changes only until the first following iteration. */ | |
2255 | if (old_inode_data->fd < 0) | |
2256 | return -EOPNOTSUPP; | |
2257 | ||
2258 | r = event_make_inotify_data(s->event, priority, &new_inotify_data); | |
2259 | if (r < 0) | |
2260 | return r; | |
2261 | rm_inotify = r > 0; | |
2262 | ||
2263 | r = event_make_inode_data(s->event, new_inotify_data, old_inode_data->dev, old_inode_data->ino, &new_inode_data); | |
2264 | if (r < 0) | |
2265 | goto fail; | |
2266 | rm_inode = r > 0; | |
2267 | ||
2268 | if (new_inode_data->fd < 0) { | |
2269 | /* Duplicate the fd for the new inode object if we don't have any yet */ | |
2270 | new_inode_data->fd = fcntl(old_inode_data->fd, F_DUPFD_CLOEXEC, 3); | |
2271 | if (new_inode_data->fd < 0) { | |
2272 | r = -errno; | |
2273 | goto fail; | |
2274 | } | |
2275 | ||
2276 | LIST_PREPEND(to_close, s->event->inode_data_to_close, new_inode_data); | |
2277 | } | |
2278 | ||
2279 | /* Move the event source to the new inode data structure */ | |
2280 | LIST_REMOVE(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2281 | LIST_PREPEND(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2282 | s->inotify.inode_data = new_inode_data; | |
2283 | ||
2284 | /* Now create the new watch */ | |
2285 | r = inode_data_realize_watch(s->event, new_inode_data); | |
2286 | if (r < 0) { | |
2287 | /* Move it back */ | |
2288 | LIST_REMOVE(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2289 | LIST_PREPEND(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2290 | s->inotify.inode_data = old_inode_data; | |
2291 | goto fail; | |
2292 | } | |
2293 | ||
2294 | s->priority = priority; | |
2295 | ||
2296 | event_gc_inode_data(s->event, old_inode_data); | |
2297 | ||
b6d5481b | 2298 | } else if (s->type == SOURCE_SIGNAL && event_source_is_online(s)) { |
9da4cb2b LP |
2299 | struct signal_data *old, *d; |
2300 | ||
2301 | /* Move us from the signalfd belonging to the old | |
2302 | * priority to the signalfd of the new priority */ | |
2303 | ||
2304 | assert_se(old = hashmap_get(s->event->signal_data, &s->priority)); | |
2305 | ||
2306 | s->priority = priority; | |
2307 | ||
2308 | r = event_make_signal_data(s->event, s->signal.sig, &d); | |
2309 | if (r < 0) { | |
2310 | s->priority = old->priority; | |
2311 | return r; | |
2312 | } | |
2313 | ||
2314 | event_unmask_signal_data(s->event, old, s->signal.sig); | |
2315 | } else | |
2316 | s->priority = priority; | |
fd38203a | 2317 | |
e1951c16 | 2318 | event_source_pp_prioq_reshuffle(s); |
fd38203a | 2319 | |
6203e07a LP |
2320 | if (s->type == SOURCE_EXIT) |
2321 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
305f78bf | 2322 | |
fd38203a | 2323 | return 0; |
97ef5391 LP |
2324 | |
2325 | fail: | |
2326 | if (rm_inode) | |
2327 | event_free_inode_data(s->event, new_inode_data); | |
2328 | ||
2329 | if (rm_inotify) | |
2330 | event_free_inotify_data(s->event, new_inotify_data); | |
2331 | ||
2332 | return r; | |
fd38203a LP |
2333 | } |
2334 | ||
cad143a8 | 2335 | _public_ int sd_event_source_get_enabled(sd_event_source *s, int *ret) { |
305f78bf | 2336 | assert_return(s, -EINVAL); |
305f78bf | 2337 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2338 | |
cad143a8 LP |
2339 | if (ret) |
2340 | *ret = s->enabled; | |
2341 | ||
08c1eb0e | 2342 | return s->enabled != SD_EVENT_OFF; |
fd38203a LP |
2343 | } |
2344 | ||
b6d5481b LP |
2345 | static int event_source_offline( |
2346 | sd_event_source *s, | |
2347 | int enabled, | |
2348 | bool ratelimited) { | |
2349 | ||
2350 | bool was_offline; | |
fd38203a LP |
2351 | int r; |
2352 | ||
ddfde737 | 2353 | assert(s); |
b6d5481b | 2354 | assert(enabled == SD_EVENT_OFF || ratelimited); |
fd38203a | 2355 | |
ddfde737 | 2356 | /* Unset the pending flag when this event source is disabled */ |
b6d5481b LP |
2357 | if (s->enabled != SD_EVENT_OFF && |
2358 | enabled == SD_EVENT_OFF && | |
2359 | !IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
ddfde737 LP |
2360 | r = source_set_pending(s, false); |
2361 | if (r < 0) | |
2362 | return r; | |
2363 | } | |
cc567911 | 2364 | |
b6d5481b LP |
2365 | was_offline = event_source_is_offline(s); |
2366 | s->enabled = enabled; | |
2367 | s->ratelimited = ratelimited; | |
fd38203a | 2368 | |
ddfde737 | 2369 | switch (s->type) { |
fd38203a | 2370 | |
ddfde737 LP |
2371 | case SOURCE_IO: |
2372 | source_io_unregister(s); | |
2373 | break; | |
ac989a78 | 2374 | |
ddfde737 LP |
2375 | case SOURCE_TIME_REALTIME: |
2376 | case SOURCE_TIME_BOOTTIME: | |
2377 | case SOURCE_TIME_MONOTONIC: | |
2378 | case SOURCE_TIME_REALTIME_ALARM: | |
2379 | case SOURCE_TIME_BOOTTIME_ALARM: | |
2380 | event_source_time_prioq_reshuffle(s); | |
2381 | break; | |
fd38203a | 2382 | |
ddfde737 LP |
2383 | case SOURCE_SIGNAL: |
2384 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); | |
2385 | break; | |
fd38203a | 2386 | |
ddfde737 | 2387 | case SOURCE_CHILD: |
b6d5481b LP |
2388 | if (!was_offline) { |
2389 | assert(s->event->n_online_child_sources > 0); | |
2390 | s->event->n_online_child_sources--; | |
2391 | } | |
fd38203a | 2392 | |
ddfde737 LP |
2393 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
2394 | source_child_pidfd_unregister(s); | |
2395 | else | |
2396 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
2397 | break; | |
4807d2d0 | 2398 | |
ddfde737 LP |
2399 | case SOURCE_EXIT: |
2400 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
2401 | break; | |
fd38203a | 2402 | |
ddfde737 LP |
2403 | case SOURCE_DEFER: |
2404 | case SOURCE_POST: | |
2405 | case SOURCE_INOTIFY: | |
2406 | break; | |
fd38203a | 2407 | |
ddfde737 LP |
2408 | default: |
2409 | assert_not_reached("Wut? I shouldn't exist."); | |
2410 | } | |
fd38203a | 2411 | |
b6d5481b | 2412 | return 1; |
ddfde737 | 2413 | } |
f8f3f926 | 2414 | |
b6d5481b LP |
2415 | static int event_source_online( |
2416 | sd_event_source *s, | |
2417 | int enabled, | |
2418 | bool ratelimited) { | |
2419 | ||
2420 | bool was_online; | |
ddfde737 | 2421 | int r; |
fd38203a | 2422 | |
ddfde737 | 2423 | assert(s); |
b6d5481b | 2424 | assert(enabled != SD_EVENT_OFF || !ratelimited); |
305f78bf | 2425 | |
ddfde737 | 2426 | /* Unset the pending flag when this event source is enabled */ |
b6d5481b LP |
2427 | if (s->enabled == SD_EVENT_OFF && |
2428 | enabled != SD_EVENT_OFF && | |
2429 | !IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
ddfde737 LP |
2430 | r = source_set_pending(s, false); |
2431 | if (r < 0) | |
2432 | return r; | |
2433 | } | |
9d3e3aa5 | 2434 | |
b6d5481b LP |
2435 | /* Are we really ready for onlining? */ |
2436 | if (enabled == SD_EVENT_OFF || ratelimited) { | |
2437 | /* Nope, we are not ready for onlining, then just update the precise state and exit */ | |
2438 | s->enabled = enabled; | |
2439 | s->ratelimited = ratelimited; | |
2440 | return 0; | |
2441 | } | |
2442 | ||
2443 | was_online = event_source_is_online(s); | |
2444 | ||
ddfde737 | 2445 | switch (s->type) { |
ddfde737 | 2446 | case SOURCE_IO: |
b6d5481b | 2447 | r = source_io_register(s, enabled, s->io.events); |
d2eafe61 | 2448 | if (r < 0) |
ddfde737 | 2449 | return r; |
ddfde737 | 2450 | break; |
fd38203a | 2451 | |
ddfde737 LP |
2452 | case SOURCE_SIGNAL: |
2453 | r = event_make_signal_data(s->event, s->signal.sig, NULL); | |
2454 | if (r < 0) { | |
ddfde737 LP |
2455 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
2456 | return r; | |
2457 | } | |
fd38203a | 2458 | |
ddfde737 | 2459 | break; |
fd38203a | 2460 | |
ddfde737 | 2461 | case SOURCE_CHILD: |
ddfde737 LP |
2462 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
2463 | /* yes, we have pidfd */ | |
9da4cb2b | 2464 | |
b6d5481b | 2465 | r = source_child_pidfd_register(s, enabled); |
ac9f2640 | 2466 | if (r < 0) |
9da4cb2b | 2467 | return r; |
ddfde737 LP |
2468 | } else { |
2469 | /* no pidfd, or something other to watch for than WEXITED */ | |
9da4cb2b | 2470 | |
ddfde737 LP |
2471 | r = event_make_signal_data(s->event, SIGCHLD, NULL); |
2472 | if (r < 0) { | |
ddfde737 LP |
2473 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); |
2474 | return r; | |
2475 | } | |
2476 | } | |
fd38203a | 2477 | |
b6d5481b LP |
2478 | if (!was_online) |
2479 | s->event->n_online_child_sources++; | |
ddfde737 | 2480 | break; |
4807d2d0 | 2481 | |
d2eafe61 ZJS |
2482 | case SOURCE_TIME_REALTIME: |
2483 | case SOURCE_TIME_BOOTTIME: | |
2484 | case SOURCE_TIME_MONOTONIC: | |
2485 | case SOURCE_TIME_REALTIME_ALARM: | |
2486 | case SOURCE_TIME_BOOTTIME_ALARM: | |
ddfde737 | 2487 | case SOURCE_EXIT: |
ddfde737 LP |
2488 | case SOURCE_DEFER: |
2489 | case SOURCE_POST: | |
2490 | case SOURCE_INOTIFY: | |
2491 | break; | |
9da4cb2b | 2492 | |
ddfde737 LP |
2493 | default: |
2494 | assert_not_reached("Wut? I shouldn't exist."); | |
2495 | } | |
f8f3f926 | 2496 | |
b6d5481b LP |
2497 | s->enabled = enabled; |
2498 | s->ratelimited = ratelimited; | |
d2eafe61 ZJS |
2499 | |
2500 | /* Non-failing operations below */ | |
2501 | switch (s->type) { | |
2502 | case SOURCE_TIME_REALTIME: | |
2503 | case SOURCE_TIME_BOOTTIME: | |
2504 | case SOURCE_TIME_MONOTONIC: | |
2505 | case SOURCE_TIME_REALTIME_ALARM: | |
2506 | case SOURCE_TIME_BOOTTIME_ALARM: | |
2507 | event_source_time_prioq_reshuffle(s); | |
2508 | break; | |
2509 | ||
2510 | case SOURCE_EXIT: | |
2511 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
2512 | break; | |
2513 | ||
2514 | default: | |
2515 | break; | |
2516 | } | |
2517 | ||
b6d5481b | 2518 | return 1; |
ddfde737 LP |
2519 | } |
2520 | ||
2521 | _public_ int sd_event_source_set_enabled(sd_event_source *s, int m) { | |
2522 | int r; | |
9da4cb2b | 2523 | |
ddfde737 LP |
2524 | assert_return(s, -EINVAL); |
2525 | assert_return(IN_SET(m, SD_EVENT_OFF, SD_EVENT_ON, SD_EVENT_ONESHOT), -EINVAL); | |
2526 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a | 2527 | |
ddfde737 LP |
2528 | /* If we are dead anyway, we are fine with turning off sources, but everything else needs to fail. */ |
2529 | if (s->event->state == SD_EVENT_FINISHED) | |
2530 | return m == SD_EVENT_OFF ? 0 : -ESTALE; | |
305f78bf | 2531 | |
ddfde737 LP |
2532 | if (s->enabled == m) /* No change? */ |
2533 | return 0; | |
9d3e3aa5 | 2534 | |
ddfde737 | 2535 | if (m == SD_EVENT_OFF) |
b6d5481b | 2536 | r = event_source_offline(s, m, s->ratelimited); |
ddfde737 LP |
2537 | else { |
2538 | if (s->enabled != SD_EVENT_OFF) { | |
2539 | /* Switching from "on" to "oneshot" or back? If that's the case, we can take a shortcut, the | |
2540 | * event source is already enabled after all. */ | |
2541 | s->enabled = m; | |
2542 | return 0; | |
fd38203a | 2543 | } |
ddfde737 | 2544 | |
b6d5481b | 2545 | r = event_source_online(s, m, s->ratelimited); |
fd38203a | 2546 | } |
ddfde737 LP |
2547 | if (r < 0) |
2548 | return r; | |
fd38203a | 2549 | |
e1951c16 | 2550 | event_source_pp_prioq_reshuffle(s); |
fd38203a LP |
2551 | return 0; |
2552 | } | |
2553 | ||
f7262a9f | 2554 | _public_ int sd_event_source_get_time(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
2555 | assert_return(s, -EINVAL); |
2556 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 2557 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf | 2558 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2559 | |
2560 | *usec = s->time.next; | |
2561 | return 0; | |
2562 | } | |
2563 | ||
f7262a9f | 2564 | _public_ int sd_event_source_set_time(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 2565 | int r; |
6a0f1f6d | 2566 | |
305f78bf | 2567 | assert_return(s, -EINVAL); |
6a0f1f6d | 2568 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 2569 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2570 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2571 | |
2a0dc6cd LP |
2572 | r = source_set_pending(s, false); |
2573 | if (r < 0) | |
2574 | return r; | |
2576a19e | 2575 | |
2a0dc6cd | 2576 | s->time.next = usec; |
fd38203a | 2577 | |
e1951c16 | 2578 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
2579 | return 0; |
2580 | } | |
2581 | ||
d6a83dc4 LP |
2582 | _public_ int sd_event_source_set_time_relative(sd_event_source *s, uint64_t usec) { |
2583 | usec_t t; | |
2584 | int r; | |
2585 | ||
2586 | assert_return(s, -EINVAL); | |
2587 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2588 | ||
2589 | r = sd_event_now(s->event, event_source_type_to_clock(s->type), &t); | |
2590 | if (r < 0) | |
2591 | return r; | |
2592 | ||
496db330 YW |
2593 | usec = usec_add(t, usec); |
2594 | if (usec == USEC_INFINITY) | |
d6a83dc4 LP |
2595 | return -EOVERFLOW; |
2596 | ||
496db330 | 2597 | return sd_event_source_set_time(s, usec); |
d6a83dc4 LP |
2598 | } |
2599 | ||
f7262a9f | 2600 | _public_ int sd_event_source_get_time_accuracy(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
2601 | assert_return(s, -EINVAL); |
2602 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 2603 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf LP |
2604 | assert_return(!event_pid_changed(s->event), -ECHILD); |
2605 | ||
2606 | *usec = s->time.accuracy; | |
2607 | return 0; | |
2608 | } | |
2609 | ||
f7262a9f | 2610 | _public_ int sd_event_source_set_time_accuracy(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 2611 | int r; |
6a0f1f6d | 2612 | |
305f78bf | 2613 | assert_return(s, -EINVAL); |
f5fbe71d | 2614 | assert_return(usec != UINT64_MAX, -EINVAL); |
6a0f1f6d | 2615 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 2616 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2617 | assert_return(!event_pid_changed(s->event), -ECHILD); |
eaa3cbef | 2618 | |
2a0dc6cd LP |
2619 | r = source_set_pending(s, false); |
2620 | if (r < 0) | |
2621 | return r; | |
2622 | ||
eaa3cbef LP |
2623 | if (usec == 0) |
2624 | usec = DEFAULT_ACCURACY_USEC; | |
2625 | ||
eaa3cbef LP |
2626 | s->time.accuracy = usec; |
2627 | ||
e1951c16 | 2628 | event_source_time_prioq_reshuffle(s); |
6a0f1f6d LP |
2629 | return 0; |
2630 | } | |
2631 | ||
2632 | _public_ int sd_event_source_get_time_clock(sd_event_source *s, clockid_t *clock) { | |
2633 | assert_return(s, -EINVAL); | |
2634 | assert_return(clock, -EINVAL); | |
2635 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2636 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
eaa3cbef | 2637 | |
6a0f1f6d | 2638 | *clock = event_source_type_to_clock(s->type); |
eaa3cbef LP |
2639 | return 0; |
2640 | } | |
2641 | ||
f7262a9f | 2642 | _public_ int sd_event_source_get_child_pid(sd_event_source *s, pid_t *pid) { |
4bee8012 LP |
2643 | assert_return(s, -EINVAL); |
2644 | assert_return(pid, -EINVAL); | |
2645 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2646 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2647 | ||
2648 | *pid = s->child.pid; | |
2649 | return 0; | |
2650 | } | |
2651 | ||
f8f3f926 LP |
2652 | _public_ int sd_event_source_get_child_pidfd(sd_event_source *s) { |
2653 | assert_return(s, -EINVAL); | |
2654 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2655 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2656 | ||
2657 | if (s->child.pidfd < 0) | |
2658 | return -EOPNOTSUPP; | |
2659 | ||
2660 | return s->child.pidfd; | |
2661 | } | |
2662 | ||
2663 | _public_ int sd_event_source_send_child_signal(sd_event_source *s, int sig, const siginfo_t *si, unsigned flags) { | |
2664 | assert_return(s, -EINVAL); | |
2665 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2666 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2667 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
2668 | ||
2669 | /* If we already have seen indication the process exited refuse sending a signal early. This way we | |
2670 | * can be sure we don't accidentally kill the wrong process on PID reuse when pidfds are not | |
2671 | * available. */ | |
2672 | if (s->child.exited) | |
2673 | return -ESRCH; | |
2674 | ||
2675 | if (s->child.pidfd >= 0) { | |
2676 | siginfo_t copy; | |
2677 | ||
2678 | /* pidfd_send_signal() changes the siginfo_t argument. This is weird, let's hence copy the | |
2679 | * structure here */ | |
2680 | if (si) | |
2681 | copy = *si; | |
2682 | ||
2683 | if (pidfd_send_signal(s->child.pidfd, sig, si ? © : NULL, 0) < 0) { | |
2684 | /* Let's propagate the error only if the system call is not implemented or prohibited */ | |
2685 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
2686 | return -errno; | |
2687 | } else | |
2688 | return 0; | |
2689 | } | |
2690 | ||
2691 | /* Flags are only supported for pidfd_send_signal(), not for rt_sigqueueinfo(), hence let's refuse | |
2692 | * this here. */ | |
2693 | if (flags != 0) | |
2694 | return -EOPNOTSUPP; | |
2695 | ||
2696 | if (si) { | |
2697 | /* We use rt_sigqueueinfo() only if siginfo_t is specified. */ | |
2698 | siginfo_t copy = *si; | |
2699 | ||
2700 | if (rt_sigqueueinfo(s->child.pid, sig, ©) < 0) | |
2701 | return -errno; | |
2702 | } else if (kill(s->child.pid, sig) < 0) | |
2703 | return -errno; | |
2704 | ||
2705 | return 0; | |
2706 | } | |
2707 | ||
2708 | _public_ int sd_event_source_get_child_pidfd_own(sd_event_source *s) { | |
2709 | assert_return(s, -EINVAL); | |
2710 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2711 | ||
2712 | if (s->child.pidfd < 0) | |
2713 | return -EOPNOTSUPP; | |
2714 | ||
2715 | return s->child.pidfd_owned; | |
2716 | } | |
2717 | ||
2718 | _public_ int sd_event_source_set_child_pidfd_own(sd_event_source *s, int own) { | |
2719 | assert_return(s, -EINVAL); | |
2720 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2721 | ||
2722 | if (s->child.pidfd < 0) | |
2723 | return -EOPNOTSUPP; | |
2724 | ||
2725 | s->child.pidfd_owned = own; | |
2726 | return 0; | |
2727 | } | |
2728 | ||
2729 | _public_ int sd_event_source_get_child_process_own(sd_event_source *s) { | |
2730 | assert_return(s, -EINVAL); | |
2731 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2732 | ||
2733 | return s->child.process_owned; | |
2734 | } | |
2735 | ||
2736 | _public_ int sd_event_source_set_child_process_own(sd_event_source *s, int own) { | |
2737 | assert_return(s, -EINVAL); | |
2738 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2739 | ||
2740 | s->child.process_owned = own; | |
2741 | return 0; | |
2742 | } | |
2743 | ||
97ef5391 LP |
2744 | _public_ int sd_event_source_get_inotify_mask(sd_event_source *s, uint32_t *mask) { |
2745 | assert_return(s, -EINVAL); | |
2746 | assert_return(mask, -EINVAL); | |
2747 | assert_return(s->type == SOURCE_INOTIFY, -EDOM); | |
2748 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2749 | ||
2750 | *mask = s->inotify.mask; | |
2751 | return 0; | |
2752 | } | |
2753 | ||
718db961 | 2754 | _public_ int sd_event_source_set_prepare(sd_event_source *s, sd_event_handler_t callback) { |
fd38203a LP |
2755 | int r; |
2756 | ||
da7e457c | 2757 | assert_return(s, -EINVAL); |
6203e07a | 2758 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c LP |
2759 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2760 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2761 | |
2762 | if (s->prepare == callback) | |
2763 | return 0; | |
2764 | ||
2765 | if (callback && s->prepare) { | |
2766 | s->prepare = callback; | |
2767 | return 0; | |
2768 | } | |
2769 | ||
2770 | r = prioq_ensure_allocated(&s->event->prepare, prepare_prioq_compare); | |
2771 | if (r < 0) | |
2772 | return r; | |
2773 | ||
2774 | s->prepare = callback; | |
2775 | ||
2776 | if (callback) { | |
2777 | r = prioq_put(s->event->prepare, s, &s->prepare_index); | |
2778 | if (r < 0) | |
2779 | return r; | |
2780 | } else | |
2781 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
2782 | ||
2783 | return 0; | |
2784 | } | |
2785 | ||
f7262a9f | 2786 | _public_ void* sd_event_source_get_userdata(sd_event_source *s) { |
da7e457c | 2787 | assert_return(s, NULL); |
fd38203a LP |
2788 | |
2789 | return s->userdata; | |
2790 | } | |
2791 | ||
8f726607 LP |
2792 | _public_ void *sd_event_source_set_userdata(sd_event_source *s, void *userdata) { |
2793 | void *ret; | |
2794 | ||
2795 | assert_return(s, NULL); | |
2796 | ||
2797 | ret = s->userdata; | |
2798 | s->userdata = userdata; | |
2799 | ||
2800 | return ret; | |
2801 | } | |
2802 | ||
b6d5481b LP |
2803 | static int event_source_enter_ratelimited(sd_event_source *s) { |
2804 | int r; | |
2805 | ||
2806 | assert(s); | |
2807 | ||
2808 | /* When an event source becomes ratelimited, we place it in the CLOCK_MONOTONIC priority queue, with | |
2809 | * the end of the rate limit time window, much as if it was a timer event source. */ | |
2810 | ||
2811 | if (s->ratelimited) | |
2812 | return 0; /* Already ratelimited, this is a NOP hence */ | |
2813 | ||
2814 | /* Make sure we can install a CLOCK_MONOTONIC event further down. */ | |
2815 | r = setup_clock_data(s->event, &s->event->monotonic, CLOCK_MONOTONIC); | |
2816 | if (r < 0) | |
2817 | return r; | |
2818 | ||
2819 | /* Timer event sources are already using the earliest/latest queues for the timer scheduling. Let's | |
2820 | * first remove them from the prioq appropriate for their own clock, so that we can use the prioq | |
2821 | * fields of the event source then for adding it to the CLOCK_MONOTONIC prioq instead. */ | |
2822 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
2823 | event_source_time_prioq_remove(s, event_get_clock_data(s->event, s->type)); | |
2824 | ||
2825 | /* Now, let's add the event source to the monotonic clock instead */ | |
2826 | r = event_source_time_prioq_put(s, &s->event->monotonic); | |
2827 | if (r < 0) | |
2828 | goto fail; | |
2829 | ||
2830 | /* And let's take the event source officially offline */ | |
2831 | r = event_source_offline(s, s->enabled, /* ratelimited= */ true); | |
2832 | if (r < 0) { | |
2833 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
2834 | goto fail; | |
2835 | } | |
2836 | ||
2837 | event_source_pp_prioq_reshuffle(s); | |
2838 | ||
2839 | log_debug("Event source %p (%s) entered rate limit state.", s, strna(s->description)); | |
2840 | return 0; | |
2841 | ||
2842 | fail: | |
2843 | /* Reinstall time event sources in the priority queue as before. This shouldn't fail, since the queue | |
2844 | * space for it should already be allocated. */ | |
2845 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
2846 | assert_se(event_source_time_prioq_put(s, event_get_clock_data(s->event, s->type)) >= 0); | |
2847 | ||
2848 | return r; | |
2849 | } | |
2850 | ||
2851 | static int event_source_leave_ratelimit(sd_event_source *s) { | |
2852 | int r; | |
2853 | ||
2854 | assert(s); | |
2855 | ||
2856 | if (!s->ratelimited) | |
2857 | return 0; | |
2858 | ||
2859 | /* Let's take the event source out of the monotonic prioq first. */ | |
2860 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
2861 | ||
2862 | /* Let's then add the event source to its native clock prioq again — if this is a timer event source */ | |
2863 | if (EVENT_SOURCE_IS_TIME(s->type)) { | |
2864 | r = event_source_time_prioq_put(s, event_get_clock_data(s->event, s->type)); | |
2865 | if (r < 0) | |
2866 | goto fail; | |
2867 | } | |
2868 | ||
2869 | /* Let's try to take it online again. */ | |
2870 | r = event_source_online(s, s->enabled, /* ratelimited= */ false); | |
2871 | if (r < 0) { | |
2872 | /* Do something roughly sensible when this failed: undo the two prioq ops above */ | |
2873 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
2874 | event_source_time_prioq_remove(s, event_get_clock_data(s->event, s->type)); | |
2875 | ||
2876 | goto fail; | |
2877 | } | |
2878 | ||
2879 | event_source_pp_prioq_reshuffle(s); | |
2880 | ratelimit_reset(&s->rate_limit); | |
2881 | ||
2882 | log_debug("Event source %p (%s) left rate limit state.", s, strna(s->description)); | |
2883 | return 0; | |
2884 | ||
2885 | fail: | |
2886 | /* Do something somewhat reasonable when we cannot move an event sources out of ratelimited mode: | |
2887 | * simply put it back in it, maybe we can then process it more successfully next iteration. */ | |
2888 | assert_se(event_source_time_prioq_put(s, &s->event->monotonic) >= 0); | |
2889 | ||
2890 | return r; | |
2891 | } | |
2892 | ||
c2ba3ad6 LP |
2893 | static usec_t sleep_between(sd_event *e, usec_t a, usec_t b) { |
2894 | usec_t c; | |
2895 | assert(e); | |
2896 | assert(a <= b); | |
2897 | ||
2898 | if (a <= 0) | |
2899 | return 0; | |
393003e1 LP |
2900 | if (a >= USEC_INFINITY) |
2901 | return USEC_INFINITY; | |
c2ba3ad6 LP |
2902 | |
2903 | if (b <= a + 1) | |
2904 | return a; | |
2905 | ||
52444dc4 LP |
2906 | initialize_perturb(e); |
2907 | ||
c2ba3ad6 LP |
2908 | /* |
2909 | Find a good time to wake up again between times a and b. We | |
2910 | have two goals here: | |
2911 | ||
2912 | a) We want to wake up as seldom as possible, hence prefer | |
2913 | later times over earlier times. | |
2914 | ||
2915 | b) But if we have to wake up, then let's make sure to | |
2916 | dispatch as much as possible on the entire system. | |
2917 | ||
2918 | We implement this by waking up everywhere at the same time | |
850516e0 | 2919 | within any given minute if we can, synchronised via the |
c2ba3ad6 | 2920 | perturbation value determined from the boot ID. If we can't, |
ba276c81 LP |
2921 | then we try to find the same spot in every 10s, then 1s and |
2922 | then 250ms step. Otherwise, we pick the last possible time | |
2923 | to wake up. | |
c2ba3ad6 LP |
2924 | */ |
2925 | ||
850516e0 LP |
2926 | c = (b / USEC_PER_MINUTE) * USEC_PER_MINUTE + e->perturb; |
2927 | if (c >= b) { | |
2928 | if (_unlikely_(c < USEC_PER_MINUTE)) | |
2929 | return b; | |
2930 | ||
2931 | c -= USEC_PER_MINUTE; | |
2932 | } | |
2933 | ||
ba276c81 LP |
2934 | if (c >= a) |
2935 | return c; | |
2936 | ||
2937 | c = (b / (USEC_PER_SEC*10)) * (USEC_PER_SEC*10) + (e->perturb % (USEC_PER_SEC*10)); | |
2938 | if (c >= b) { | |
2939 | if (_unlikely_(c < USEC_PER_SEC*10)) | |
2940 | return b; | |
2941 | ||
2942 | c -= USEC_PER_SEC*10; | |
2943 | } | |
2944 | ||
850516e0 LP |
2945 | if (c >= a) |
2946 | return c; | |
2947 | ||
2948 | c = (b / USEC_PER_SEC) * USEC_PER_SEC + (e->perturb % USEC_PER_SEC); | |
c2ba3ad6 LP |
2949 | if (c >= b) { |
2950 | if (_unlikely_(c < USEC_PER_SEC)) | |
2951 | return b; | |
2952 | ||
2953 | c -= USEC_PER_SEC; | |
2954 | } | |
2955 | ||
2956 | if (c >= a) | |
2957 | return c; | |
2958 | ||
2959 | c = (b / (USEC_PER_MSEC*250)) * (USEC_PER_MSEC*250) + (e->perturb % (USEC_PER_MSEC*250)); | |
2960 | if (c >= b) { | |
2961 | if (_unlikely_(c < USEC_PER_MSEC*250)) | |
2962 | return b; | |
2963 | ||
2964 | c -= USEC_PER_MSEC*250; | |
2965 | } | |
2966 | ||
2967 | if (c >= a) | |
2968 | return c; | |
2969 | ||
2970 | return b; | |
2971 | } | |
2972 | ||
fd38203a LP |
2973 | static int event_arm_timer( |
2974 | sd_event *e, | |
6a0f1f6d | 2975 | struct clock_data *d) { |
fd38203a LP |
2976 | |
2977 | struct itimerspec its = {}; | |
c2ba3ad6 LP |
2978 | sd_event_source *a, *b; |
2979 | usec_t t; | |
fd38203a | 2980 | |
cde93897 | 2981 | assert(e); |
6a0f1f6d | 2982 | assert(d); |
fd38203a | 2983 | |
d06441da | 2984 | if (!d->needs_rearm) |
212bbb17 TG |
2985 | return 0; |
2986 | else | |
2987 | d->needs_rearm = false; | |
2988 | ||
6a0f1f6d | 2989 | a = prioq_peek(d->earliest); |
b6d5481b | 2990 | if (!a || a->enabled == SD_EVENT_OFF || time_event_source_next(a) == USEC_INFINITY) { |
72aedc1e | 2991 | |
6a0f1f6d | 2992 | if (d->fd < 0) |
c57b5ca3 LP |
2993 | return 0; |
2994 | ||
3a43da28 | 2995 | if (d->next == USEC_INFINITY) |
72aedc1e LP |
2996 | return 0; |
2997 | ||
2998 | /* disarm */ | |
15c689d7 LP |
2999 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
3000 | return -errno; | |
72aedc1e | 3001 | |
3a43da28 | 3002 | d->next = USEC_INFINITY; |
fd38203a | 3003 | return 0; |
72aedc1e | 3004 | } |
fd38203a | 3005 | |
6a0f1f6d | 3006 | b = prioq_peek(d->latest); |
baf76283 | 3007 | assert_se(b && b->enabled != SD_EVENT_OFF); |
c2ba3ad6 | 3008 | |
b6d5481b | 3009 | t = sleep_between(e, time_event_source_next(a), time_event_source_latest(b)); |
6a0f1f6d | 3010 | if (d->next == t) |
fd38203a LP |
3011 | return 0; |
3012 | ||
6a0f1f6d | 3013 | assert_se(d->fd >= 0); |
fd38203a | 3014 | |
c2ba3ad6 | 3015 | if (t == 0) { |
fd38203a LP |
3016 | /* We don' want to disarm here, just mean some time looooong ago. */ |
3017 | its.it_value.tv_sec = 0; | |
3018 | its.it_value.tv_nsec = 1; | |
3019 | } else | |
c2ba3ad6 | 3020 | timespec_store(&its.it_value, t); |
fd38203a | 3021 | |
15c689d7 | 3022 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 | 3023 | return -errno; |
fd38203a | 3024 | |
6a0f1f6d | 3025 | d->next = t; |
fd38203a LP |
3026 | return 0; |
3027 | } | |
3028 | ||
9a800b56 | 3029 | static int process_io(sd_event *e, sd_event_source *s, uint32_t revents) { |
fd38203a LP |
3030 | assert(e); |
3031 | assert(s); | |
3032 | assert(s->type == SOURCE_IO); | |
3033 | ||
9a800b56 LP |
3034 | /* If the event source was already pending, we just OR in the |
3035 | * new revents, otherwise we reset the value. The ORing is | |
3036 | * necessary to handle EPOLLONESHOT events properly where | |
3037 | * readability might happen independently of writability, and | |
3038 | * we need to keep track of both */ | |
3039 | ||
3040 | if (s->pending) | |
3041 | s->io.revents |= revents; | |
3042 | else | |
3043 | s->io.revents = revents; | |
fd38203a | 3044 | |
fd38203a LP |
3045 | return source_set_pending(s, true); |
3046 | } | |
3047 | ||
72aedc1e | 3048 | static int flush_timer(sd_event *e, int fd, uint32_t events, usec_t *next) { |
fd38203a LP |
3049 | uint64_t x; |
3050 | ssize_t ss; | |
3051 | ||
3052 | assert(e); | |
da7e457c | 3053 | assert(fd >= 0); |
72aedc1e | 3054 | |
305f78bf | 3055 | assert_return(events == EPOLLIN, -EIO); |
fd38203a LP |
3056 | |
3057 | ss = read(fd, &x, sizeof(x)); | |
3058 | if (ss < 0) { | |
945c2931 | 3059 | if (IN_SET(errno, EAGAIN, EINTR)) |
fd38203a LP |
3060 | return 0; |
3061 | ||
3062 | return -errno; | |
3063 | } | |
3064 | ||
8d35dae7 | 3065 | if (_unlikely_(ss != sizeof(x))) |
fd38203a LP |
3066 | return -EIO; |
3067 | ||
cde93897 | 3068 | if (next) |
3a43da28 | 3069 | *next = USEC_INFINITY; |
72aedc1e | 3070 | |
fd38203a LP |
3071 | return 0; |
3072 | } | |
3073 | ||
305f78bf LP |
3074 | static int process_timer( |
3075 | sd_event *e, | |
3076 | usec_t n, | |
6a0f1f6d | 3077 | struct clock_data *d) { |
305f78bf | 3078 | |
fd38203a LP |
3079 | sd_event_source *s; |
3080 | int r; | |
3081 | ||
3082 | assert(e); | |
6a0f1f6d | 3083 | assert(d); |
fd38203a LP |
3084 | |
3085 | for (;;) { | |
6a0f1f6d | 3086 | s = prioq_peek(d->earliest); |
b6d5481b LP |
3087 | if (!s || time_event_source_next(s) > n) |
3088 | break; | |
3089 | ||
3090 | if (s->ratelimited) { | |
3091 | /* This is an event sources whose ratelimit window has ended. Let's turn it on | |
3092 | * again. */ | |
3093 | assert(s->ratelimited); | |
3094 | ||
3095 | r = event_source_leave_ratelimit(s); | |
3096 | if (r < 0) | |
3097 | return r; | |
3098 | ||
3099 | continue; | |
3100 | } | |
3101 | ||
3102 | if (s->enabled == SD_EVENT_OFF || s->pending) | |
fd38203a LP |
3103 | break; |
3104 | ||
3105 | r = source_set_pending(s, true); | |
3106 | if (r < 0) | |
3107 | return r; | |
3108 | ||
e1951c16 | 3109 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
3110 | } |
3111 | ||
3112 | return 0; | |
3113 | } | |
3114 | ||
efd3be9d YW |
3115 | static int process_child(sd_event *e, int64_t threshold, int64_t *ret_min_priority) { |
3116 | int64_t min_priority = threshold; | |
3117 | bool something_new = false; | |
fd38203a | 3118 | sd_event_source *s; |
fd38203a LP |
3119 | int r; |
3120 | ||
3121 | assert(e); | |
efd3be9d YW |
3122 | assert(ret_min_priority); |
3123 | ||
3124 | if (!e->need_process_child) { | |
3125 | *ret_min_priority = min_priority; | |
3126 | return 0; | |
3127 | } | |
fd38203a | 3128 | |
c2ba3ad6 LP |
3129 | e->need_process_child = false; |
3130 | ||
fd38203a LP |
3131 | /* |
3132 | So, this is ugly. We iteratively invoke waitid() with P_PID | |
3133 | + WNOHANG for each PID we wait for, instead of using | |
3134 | P_ALL. This is because we only want to get child | |
3135 | information of very specific child processes, and not all | |
3136 | of them. We might not have processed the SIGCHLD even of a | |
3137 | previous invocation and we don't want to maintain a | |
3138 | unbounded *per-child* event queue, hence we really don't | |
3139 | want anything flushed out of the kernel's queue that we | |
3140 | don't care about. Since this is O(n) this means that if you | |
3141 | have a lot of processes you probably want to handle SIGCHLD | |
3142 | yourself. | |
08cd1552 LP |
3143 | |
3144 | We do not reap the children here (by using WNOWAIT), this | |
3145 | is only done after the event source is dispatched so that | |
3146 | the callback still sees the process as a zombie. | |
fd38203a LP |
3147 | */ |
3148 | ||
90e74a66 | 3149 | HASHMAP_FOREACH(s, e->child_sources) { |
fd38203a LP |
3150 | assert(s->type == SOURCE_CHILD); |
3151 | ||
efd3be9d YW |
3152 | if (s->priority > threshold) |
3153 | continue; | |
3154 | ||
fd38203a LP |
3155 | if (s->pending) |
3156 | continue; | |
3157 | ||
b6d5481b | 3158 | if (event_source_is_offline(s)) |
fd38203a LP |
3159 | continue; |
3160 | ||
f8f3f926 LP |
3161 | if (s->child.exited) |
3162 | continue; | |
3163 | ||
3164 | if (EVENT_SOURCE_WATCH_PIDFD(s)) /* There's a usable pidfd known for this event source? then don't waitid() for it here */ | |
3165 | continue; | |
3166 | ||
fd38203a | 3167 | zero(s->child.siginfo); |
15c689d7 LP |
3168 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, |
3169 | WNOHANG | (s->child.options & WEXITED ? WNOWAIT : 0) | s->child.options) < 0) | |
bfd9bfcc | 3170 | return negative_errno(); |
fd38203a LP |
3171 | |
3172 | if (s->child.siginfo.si_pid != 0) { | |
945c2931 | 3173 | bool zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 3174 | |
f8f3f926 LP |
3175 | if (zombie) |
3176 | s->child.exited = true; | |
3177 | ||
08cd1552 LP |
3178 | if (!zombie && (s->child.options & WEXITED)) { |
3179 | /* If the child isn't dead then let's | |
3180 | * immediately remove the state change | |
3181 | * from the queue, since there's no | |
3182 | * benefit in leaving it queued */ | |
3183 | ||
3184 | assert(s->child.options & (WSTOPPED|WCONTINUED)); | |
a5d27871 | 3185 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|(s->child.options & (WSTOPPED|WCONTINUED))); |
08cd1552 LP |
3186 | } |
3187 | ||
fd38203a LP |
3188 | r = source_set_pending(s, true); |
3189 | if (r < 0) | |
3190 | return r; | |
efd3be9d YW |
3191 | if (r > 0) { |
3192 | something_new = true; | |
3193 | min_priority = MIN(min_priority, s->priority); | |
3194 | } | |
fd38203a LP |
3195 | } |
3196 | } | |
3197 | ||
efd3be9d YW |
3198 | *ret_min_priority = min_priority; |
3199 | return something_new; | |
fd38203a LP |
3200 | } |
3201 | ||
f8f3f926 LP |
3202 | static int process_pidfd(sd_event *e, sd_event_source *s, uint32_t revents) { |
3203 | assert(e); | |
3204 | assert(s); | |
3205 | assert(s->type == SOURCE_CHILD); | |
3206 | ||
3207 | if (s->pending) | |
3208 | return 0; | |
3209 | ||
b6d5481b | 3210 | if (event_source_is_offline(s)) |
f8f3f926 LP |
3211 | return 0; |
3212 | ||
3213 | if (!EVENT_SOURCE_WATCH_PIDFD(s)) | |
3214 | return 0; | |
3215 | ||
3216 | zero(s->child.siginfo); | |
3217 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG | WNOWAIT | s->child.options) < 0) | |
3218 | return -errno; | |
3219 | ||
3220 | if (s->child.siginfo.si_pid == 0) | |
3221 | return 0; | |
3222 | ||
3223 | if (IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) | |
3224 | s->child.exited = true; | |
3225 | ||
3226 | return source_set_pending(s, true); | |
3227 | } | |
3228 | ||
efd3be9d | 3229 | static int process_signal(sd_event *e, struct signal_data *d, uint32_t events, int64_t *min_priority) { |
fd38203a LP |
3230 | int r; |
3231 | ||
da7e457c | 3232 | assert(e); |
97ef5391 | 3233 | assert(d); |
305f78bf | 3234 | assert_return(events == EPOLLIN, -EIO); |
efd3be9d | 3235 | assert(min_priority); |
fd38203a | 3236 | |
9da4cb2b LP |
3237 | /* If there's a signal queued on this priority and SIGCHLD is |
3238 | on this priority too, then make sure to recheck the | |
3239 | children we watch. This is because we only ever dequeue | |
3240 | the first signal per priority, and if we dequeue one, and | |
3241 | SIGCHLD might be enqueued later we wouldn't know, but we | |
3242 | might have higher priority children we care about hence we | |
3243 | need to check that explicitly. */ | |
3244 | ||
3245 | if (sigismember(&d->sigset, SIGCHLD)) | |
3246 | e->need_process_child = true; | |
3247 | ||
3248 | /* If there's already an event source pending for this | |
3249 | * priority we don't read another */ | |
3250 | if (d->current) | |
3251 | return 0; | |
3252 | ||
fd38203a | 3253 | for (;;) { |
0eb2e0e3 | 3254 | struct signalfd_siginfo si; |
7057bd99 | 3255 | ssize_t n; |
92daebc0 | 3256 | sd_event_source *s = NULL; |
fd38203a | 3257 | |
9da4cb2b | 3258 | n = read(d->fd, &si, sizeof(si)); |
7057bd99 | 3259 | if (n < 0) { |
945c2931 | 3260 | if (IN_SET(errno, EAGAIN, EINTR)) |
efd3be9d | 3261 | return 0; |
fd38203a LP |
3262 | |
3263 | return -errno; | |
3264 | } | |
3265 | ||
7057bd99 | 3266 | if (_unlikely_(n != sizeof(si))) |
fd38203a LP |
3267 | return -EIO; |
3268 | ||
6eb7c172 | 3269 | assert(SIGNAL_VALID(si.ssi_signo)); |
7057bd99 | 3270 | |
92daebc0 LP |
3271 | if (e->signal_sources) |
3272 | s = e->signal_sources[si.ssi_signo]; | |
92daebc0 LP |
3273 | if (!s) |
3274 | continue; | |
9da4cb2b LP |
3275 | if (s->pending) |
3276 | continue; | |
fd38203a LP |
3277 | |
3278 | s->signal.siginfo = si; | |
9da4cb2b LP |
3279 | d->current = s; |
3280 | ||
fd38203a LP |
3281 | r = source_set_pending(s, true); |
3282 | if (r < 0) | |
3283 | return r; | |
efd3be9d YW |
3284 | if (r > 0 && *min_priority >= s->priority) { |
3285 | *min_priority = s->priority; | |
3286 | return 1; /* an event source with smaller priority is queued. */ | |
3287 | } | |
9da4cb2b | 3288 | |
efd3be9d | 3289 | return 0; |
fd38203a | 3290 | } |
fd38203a LP |
3291 | } |
3292 | ||
efd3be9d | 3293 | static int event_inotify_data_read(sd_event *e, struct inotify_data *d, uint32_t revents, int64_t threshold) { |
97ef5391 LP |
3294 | ssize_t n; |
3295 | ||
3296 | assert(e); | |
3297 | assert(d); | |
3298 | ||
3299 | assert_return(revents == EPOLLIN, -EIO); | |
3300 | ||
3301 | /* If there's already an event source pending for this priority, don't read another */ | |
3302 | if (d->n_pending > 0) | |
3303 | return 0; | |
3304 | ||
3305 | /* Is the read buffer non-empty? If so, let's not read more */ | |
3306 | if (d->buffer_filled > 0) | |
3307 | return 0; | |
3308 | ||
efd3be9d YW |
3309 | if (d->priority > threshold) |
3310 | return 0; | |
3311 | ||
97ef5391 LP |
3312 | n = read(d->fd, &d->buffer, sizeof(d->buffer)); |
3313 | if (n < 0) { | |
3314 | if (IN_SET(errno, EAGAIN, EINTR)) | |
3315 | return 0; | |
3316 | ||
3317 | return -errno; | |
3318 | } | |
3319 | ||
3320 | assert(n > 0); | |
3321 | d->buffer_filled = (size_t) n; | |
3322 | LIST_PREPEND(buffered, e->inotify_data_buffered, d); | |
3323 | ||
3324 | return 1; | |
3325 | } | |
3326 | ||
3327 | static void event_inotify_data_drop(sd_event *e, struct inotify_data *d, size_t sz) { | |
3328 | assert(e); | |
3329 | assert(d); | |
3330 | assert(sz <= d->buffer_filled); | |
3331 | ||
3332 | if (sz == 0) | |
3333 | return; | |
3334 | ||
3335 | /* Move the rest to the buffer to the front, in order to get things properly aligned again */ | |
3336 | memmove(d->buffer.raw, d->buffer.raw + sz, d->buffer_filled - sz); | |
3337 | d->buffer_filled -= sz; | |
3338 | ||
3339 | if (d->buffer_filled == 0) | |
3340 | LIST_REMOVE(buffered, e->inotify_data_buffered, d); | |
3341 | } | |
3342 | ||
3343 | static int event_inotify_data_process(sd_event *e, struct inotify_data *d) { | |
3344 | int r; | |
3345 | ||
3346 | assert(e); | |
3347 | assert(d); | |
3348 | ||
3349 | /* If there's already an event source pending for this priority, don't read another */ | |
3350 | if (d->n_pending > 0) | |
3351 | return 0; | |
3352 | ||
3353 | while (d->buffer_filled > 0) { | |
3354 | size_t sz; | |
3355 | ||
3356 | /* Let's validate that the event structures are complete */ | |
3357 | if (d->buffer_filled < offsetof(struct inotify_event, name)) | |
3358 | return -EIO; | |
3359 | ||
3360 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3361 | if (d->buffer_filled < sz) | |
3362 | return -EIO; | |
3363 | ||
3364 | if (d->buffer.ev.mask & IN_Q_OVERFLOW) { | |
3365 | struct inode_data *inode_data; | |
97ef5391 LP |
3366 | |
3367 | /* The queue overran, let's pass this event to all event sources connected to this inotify | |
3368 | * object */ | |
3369 | ||
90e74a66 | 3370 | HASHMAP_FOREACH(inode_data, d->inodes) { |
97ef5391 LP |
3371 | sd_event_source *s; |
3372 | ||
3373 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
3374 | ||
b6d5481b | 3375 | if (event_source_is_offline(s)) |
97ef5391 LP |
3376 | continue; |
3377 | ||
3378 | r = source_set_pending(s, true); | |
3379 | if (r < 0) | |
3380 | return r; | |
3381 | } | |
3382 | } | |
3383 | } else { | |
3384 | struct inode_data *inode_data; | |
3385 | sd_event_source *s; | |
3386 | ||
3387 | /* Find the inode object for this watch descriptor. If IN_IGNORED is set we also remove it from | |
3388 | * our watch descriptor table. */ | |
3389 | if (d->buffer.ev.mask & IN_IGNORED) { | |
3390 | ||
3391 | inode_data = hashmap_remove(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3392 | if (!inode_data) { | |
3393 | event_inotify_data_drop(e, d, sz); | |
3394 | continue; | |
3395 | } | |
3396 | ||
3397 | /* The watch descriptor was removed by the kernel, let's drop it here too */ | |
3398 | inode_data->wd = -1; | |
3399 | } else { | |
3400 | inode_data = hashmap_get(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3401 | if (!inode_data) { | |
3402 | event_inotify_data_drop(e, d, sz); | |
3403 | continue; | |
3404 | } | |
3405 | } | |
3406 | ||
3407 | /* Trigger all event sources that are interested in these events. Also trigger all event | |
3408 | * sources if IN_IGNORED or IN_UNMOUNT is set. */ | |
3409 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
3410 | ||
b6d5481b | 3411 | if (event_source_is_offline(s)) |
97ef5391 LP |
3412 | continue; |
3413 | ||
3414 | if ((d->buffer.ev.mask & (IN_IGNORED|IN_UNMOUNT)) == 0 && | |
3415 | (s->inotify.mask & d->buffer.ev.mask & IN_ALL_EVENTS) == 0) | |
3416 | continue; | |
3417 | ||
3418 | r = source_set_pending(s, true); | |
3419 | if (r < 0) | |
3420 | return r; | |
3421 | } | |
3422 | } | |
3423 | ||
3424 | /* Something pending now? If so, let's finish, otherwise let's read more. */ | |
3425 | if (d->n_pending > 0) | |
3426 | return 1; | |
3427 | } | |
3428 | ||
3429 | return 0; | |
3430 | } | |
3431 | ||
3432 | static int process_inotify(sd_event *e) { | |
3433 | struct inotify_data *d; | |
3434 | int r, done = 0; | |
3435 | ||
3436 | assert(e); | |
3437 | ||
3438 | LIST_FOREACH(buffered, d, e->inotify_data_buffered) { | |
3439 | r = event_inotify_data_process(e, d); | |
3440 | if (r < 0) | |
3441 | return r; | |
3442 | if (r > 0) | |
3443 | done ++; | |
3444 | } | |
3445 | ||
3446 | return done; | |
3447 | } | |
3448 | ||
fd38203a | 3449 | static int source_dispatch(sd_event_source *s) { |
b778cba4 | 3450 | _cleanup_(sd_event_unrefp) sd_event *saved_event = NULL; |
8f5c235d | 3451 | EventSourceType saved_type; |
fe8245eb | 3452 | int r = 0; |
fd38203a LP |
3453 | |
3454 | assert(s); | |
6203e07a | 3455 | assert(s->pending || s->type == SOURCE_EXIT); |
fd38203a | 3456 | |
b778cba4 LP |
3457 | /* Save the event source type, here, so that we still know it after the event callback which might |
3458 | * invalidate the event. */ | |
8f5c235d LP |
3459 | saved_type = s->type; |
3460 | ||
de02634c | 3461 | /* Similarly, store a reference to the event loop object, so that we can still access it after the |
b778cba4 LP |
3462 | * callback might have invalidated/disconnected the event source. */ |
3463 | saved_event = sd_event_ref(s->event); | |
3464 | ||
de02634c | 3465 | /* Check if we hit the ratelimit for this event source, and if so, let's disable it. */ |
b6d5481b LP |
3466 | assert(!s->ratelimited); |
3467 | if (!ratelimit_below(&s->rate_limit)) { | |
3468 | r = event_source_enter_ratelimited(s); | |
3469 | if (r < 0) | |
3470 | return r; | |
3471 | ||
3472 | return 1; | |
3473 | } | |
3474 | ||
945c2931 | 3475 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { |
da7e457c LP |
3476 | r = source_set_pending(s, false); |
3477 | if (r < 0) | |
3478 | return r; | |
3479 | } | |
fd38203a | 3480 | |
6e9feda3 LP |
3481 | if (s->type != SOURCE_POST) { |
3482 | sd_event_source *z; | |
6e9feda3 | 3483 | |
de02634c | 3484 | /* If we execute a non-post source, let's mark all post sources as pending. */ |
6e9feda3 | 3485 | |
90e74a66 | 3486 | SET_FOREACH(z, s->event->post_sources) { |
b6d5481b | 3487 | if (event_source_is_offline(z)) |
6e9feda3 LP |
3488 | continue; |
3489 | ||
3490 | r = source_set_pending(z, true); | |
3491 | if (r < 0) | |
3492 | return r; | |
3493 | } | |
3494 | } | |
3495 | ||
baf76283 LP |
3496 | if (s->enabled == SD_EVENT_ONESHOT) { |
3497 | r = sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
3498 | if (r < 0) |
3499 | return r; | |
3500 | } | |
3501 | ||
12179984 | 3502 | s->dispatching = true; |
b7484e2a | 3503 | |
fd38203a LP |
3504 | switch (s->type) { |
3505 | ||
3506 | case SOURCE_IO: | |
3507 | r = s->io.callback(s, s->io.fd, s->io.revents, s->userdata); | |
3508 | break; | |
3509 | ||
6a0f1f6d | 3510 | case SOURCE_TIME_REALTIME: |
a8548816 | 3511 | case SOURCE_TIME_BOOTTIME: |
6a0f1f6d LP |
3512 | case SOURCE_TIME_MONOTONIC: |
3513 | case SOURCE_TIME_REALTIME_ALARM: | |
3514 | case SOURCE_TIME_BOOTTIME_ALARM: | |
fd38203a LP |
3515 | r = s->time.callback(s, s->time.next, s->userdata); |
3516 | break; | |
3517 | ||
3518 | case SOURCE_SIGNAL: | |
3519 | r = s->signal.callback(s, &s->signal.siginfo, s->userdata); | |
3520 | break; | |
3521 | ||
08cd1552 LP |
3522 | case SOURCE_CHILD: { |
3523 | bool zombie; | |
3524 | ||
945c2931 | 3525 | zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 3526 | |
fd38203a | 3527 | r = s->child.callback(s, &s->child.siginfo, s->userdata); |
08cd1552 LP |
3528 | |
3529 | /* Now, reap the PID for good. */ | |
f8f3f926 | 3530 | if (zombie) { |
cc59d290 | 3531 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|WEXITED); |
f8f3f926 LP |
3532 | s->child.waited = true; |
3533 | } | |
08cd1552 | 3534 | |
fd38203a | 3535 | break; |
08cd1552 | 3536 | } |
fd38203a LP |
3537 | |
3538 | case SOURCE_DEFER: | |
3539 | r = s->defer.callback(s, s->userdata); | |
3540 | break; | |
da7e457c | 3541 | |
6e9feda3 LP |
3542 | case SOURCE_POST: |
3543 | r = s->post.callback(s, s->userdata); | |
3544 | break; | |
3545 | ||
6203e07a LP |
3546 | case SOURCE_EXIT: |
3547 | r = s->exit.callback(s, s->userdata); | |
da7e457c | 3548 | break; |
9d3e3aa5 | 3549 | |
97ef5391 LP |
3550 | case SOURCE_INOTIFY: { |
3551 | struct sd_event *e = s->event; | |
3552 | struct inotify_data *d; | |
3553 | size_t sz; | |
3554 | ||
3555 | assert(s->inotify.inode_data); | |
3556 | assert_se(d = s->inotify.inode_data->inotify_data); | |
3557 | ||
3558 | assert(d->buffer_filled >= offsetof(struct inotify_event, name)); | |
3559 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3560 | assert(d->buffer_filled >= sz); | |
3561 | ||
3562 | r = s->inotify.callback(s, &d->buffer.ev, s->userdata); | |
3563 | ||
3564 | /* When no event is pending anymore on this inotify object, then let's drop the event from the | |
3565 | * buffer. */ | |
3566 | if (d->n_pending == 0) | |
3567 | event_inotify_data_drop(e, d, sz); | |
3568 | ||
3569 | break; | |
3570 | } | |
3571 | ||
9d3e3aa5 | 3572 | case SOURCE_WATCHDOG: |
a71fe8b8 | 3573 | case _SOURCE_EVENT_SOURCE_TYPE_MAX: |
9f2a50a3 | 3574 | case _SOURCE_EVENT_SOURCE_TYPE_INVALID: |
9d3e3aa5 | 3575 | assert_not_reached("Wut? I shouldn't exist."); |
fd38203a LP |
3576 | } |
3577 | ||
12179984 LP |
3578 | s->dispatching = false; |
3579 | ||
b778cba4 LP |
3580 | if (r < 0) { |
3581 | log_debug_errno(r, "Event source %s (type %s) returned error, %s: %m", | |
3582 | strna(s->description), | |
3583 | event_source_type_to_string(saved_type), | |
3584 | s->exit_on_failure ? "exiting" : "disabling"); | |
3585 | ||
3586 | if (s->exit_on_failure) | |
3587 | (void) sd_event_exit(saved_event, r); | |
3588 | } | |
12179984 LP |
3589 | |
3590 | if (s->n_ref == 0) | |
3591 | source_free(s); | |
3592 | else if (r < 0) | |
6203e07a | 3593 | sd_event_source_set_enabled(s, SD_EVENT_OFF); |
b7484e2a | 3594 | |
6203e07a | 3595 | return 1; |
fd38203a LP |
3596 | } |
3597 | ||
3598 | static int event_prepare(sd_event *e) { | |
3599 | int r; | |
3600 | ||
3601 | assert(e); | |
3602 | ||
3603 | for (;;) { | |
3604 | sd_event_source *s; | |
3605 | ||
3606 | s = prioq_peek(e->prepare); | |
b6d5481b | 3607 | if (!s || s->prepare_iteration == e->iteration || event_source_is_offline(s)) |
fd38203a LP |
3608 | break; |
3609 | ||
3610 | s->prepare_iteration = e->iteration; | |
3611 | r = prioq_reshuffle(e->prepare, s, &s->prepare_index); | |
3612 | if (r < 0) | |
3613 | return r; | |
3614 | ||
3615 | assert(s->prepare); | |
12179984 LP |
3616 | |
3617 | s->dispatching = true; | |
fd38203a | 3618 | r = s->prepare(s, s->userdata); |
12179984 LP |
3619 | s->dispatching = false; |
3620 | ||
b778cba4 LP |
3621 | if (r < 0) { |
3622 | log_debug_errno(r, "Prepare callback of event source %s (type %s) returned error, %s: %m", | |
3623 | strna(s->description), | |
3624 | event_source_type_to_string(s->type), | |
3625 | s->exit_on_failure ? "exiting" : "disabling"); | |
3626 | ||
3627 | if (s->exit_on_failure) | |
3628 | (void) sd_event_exit(e, r); | |
3629 | } | |
fd38203a | 3630 | |
12179984 LP |
3631 | if (s->n_ref == 0) |
3632 | source_free(s); | |
3633 | else if (r < 0) | |
3634 | sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
3635 | } |
3636 | ||
3637 | return 0; | |
3638 | } | |
3639 | ||
6203e07a | 3640 | static int dispatch_exit(sd_event *e) { |
da7e457c LP |
3641 | sd_event_source *p; |
3642 | int r; | |
3643 | ||
3644 | assert(e); | |
3645 | ||
6203e07a | 3646 | p = prioq_peek(e->exit); |
b6d5481b | 3647 | if (!p || event_source_is_offline(p)) { |
da7e457c LP |
3648 | e->state = SD_EVENT_FINISHED; |
3649 | return 0; | |
3650 | } | |
3651 | ||
f814c871 | 3652 | _unused_ _cleanup_(sd_event_unrefp) sd_event *ref = sd_event_ref(e); |
da7e457c | 3653 | e->iteration++; |
6203e07a | 3654 | e->state = SD_EVENT_EXITING; |
da7e457c | 3655 | r = source_dispatch(p); |
2b0c9ef7 | 3656 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
3657 | return r; |
3658 | } | |
3659 | ||
c2ba3ad6 LP |
3660 | static sd_event_source* event_next_pending(sd_event *e) { |
3661 | sd_event_source *p; | |
3662 | ||
da7e457c LP |
3663 | assert(e); |
3664 | ||
c2ba3ad6 LP |
3665 | p = prioq_peek(e->pending); |
3666 | if (!p) | |
3667 | return NULL; | |
3668 | ||
b6d5481b | 3669 | if (event_source_is_offline(p)) |
c2ba3ad6 LP |
3670 | return NULL; |
3671 | ||
3672 | return p; | |
3673 | } | |
3674 | ||
cde93897 LP |
3675 | static int arm_watchdog(sd_event *e) { |
3676 | struct itimerspec its = {}; | |
3677 | usec_t t; | |
cde93897 LP |
3678 | |
3679 | assert(e); | |
3680 | assert(e->watchdog_fd >= 0); | |
3681 | ||
3682 | t = sleep_between(e, | |
3683 | e->watchdog_last + (e->watchdog_period / 2), | |
3684 | e->watchdog_last + (e->watchdog_period * 3 / 4)); | |
3685 | ||
3686 | timespec_store(&its.it_value, t); | |
3687 | ||
75145780 LP |
3688 | /* Make sure we never set the watchdog to 0, which tells the |
3689 | * kernel to disable it. */ | |
3690 | if (its.it_value.tv_sec == 0 && its.it_value.tv_nsec == 0) | |
3691 | its.it_value.tv_nsec = 1; | |
3692 | ||
15c689d7 | 3693 | if (timerfd_settime(e->watchdog_fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 LP |
3694 | return -errno; |
3695 | ||
3696 | return 0; | |
3697 | } | |
3698 | ||
3699 | static int process_watchdog(sd_event *e) { | |
3700 | assert(e); | |
3701 | ||
3702 | if (!e->watchdog) | |
3703 | return 0; | |
3704 | ||
3705 | /* Don't notify watchdog too often */ | |
3706 | if (e->watchdog_last + e->watchdog_period / 4 > e->timestamp.monotonic) | |
3707 | return 0; | |
3708 | ||
3709 | sd_notify(false, "WATCHDOG=1"); | |
3710 | e->watchdog_last = e->timestamp.monotonic; | |
3711 | ||
3712 | return arm_watchdog(e); | |
3713 | } | |
3714 | ||
97ef5391 LP |
3715 | static void event_close_inode_data_fds(sd_event *e) { |
3716 | struct inode_data *d; | |
3717 | ||
3718 | assert(e); | |
3719 | ||
3720 | /* Close the fds pointing to the inodes to watch now. We need to close them as they might otherwise pin | |
3721 | * filesystems. But we can't close them right-away as we need them as long as the user still wants to make | |
5238e957 | 3722 | * adjustments to the even source, such as changing the priority (which requires us to remove and re-add a watch |
97ef5391 LP |
3723 | * for the inode). Hence, let's close them when entering the first iteration after they were added, as a |
3724 | * compromise. */ | |
3725 | ||
3726 | while ((d = e->inode_data_to_close)) { | |
3727 | assert(d->fd >= 0); | |
3728 | d->fd = safe_close(d->fd); | |
3729 | ||
3730 | LIST_REMOVE(to_close, e->inode_data_to_close, d); | |
3731 | } | |
3732 | } | |
3733 | ||
c45a5a74 TG |
3734 | _public_ int sd_event_prepare(sd_event *e) { |
3735 | int r; | |
fd38203a | 3736 | |
da7e457c | 3737 | assert_return(e, -EINVAL); |
b937d761 | 3738 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
3739 | assert_return(!event_pid_changed(e), -ECHILD); |
3740 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 3741 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 3742 | |
e5446015 LP |
3743 | /* Let's check that if we are a default event loop we are executed in the correct thread. We only do |
3744 | * this check here once, since gettid() is typically not cached, and thus want to minimize | |
3745 | * syscalls */ | |
3746 | assert_return(!e->default_event_ptr || e->tid == gettid(), -EREMOTEIO); | |
3747 | ||
f814c871 LP |
3748 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
3749 | _unused_ _cleanup_(sd_event_unrefp) sd_event *ref = sd_event_ref(e); | |
3750 | ||
6203e07a | 3751 | if (e->exit_requested) |
c45a5a74 | 3752 | goto pending; |
fd38203a LP |
3753 | |
3754 | e->iteration++; | |
3755 | ||
0be6c2f6 | 3756 | e->state = SD_EVENT_PREPARING; |
fd38203a | 3757 | r = event_prepare(e); |
0be6c2f6 | 3758 | e->state = SD_EVENT_INITIAL; |
fd38203a | 3759 | if (r < 0) |
c45a5a74 | 3760 | return r; |
fd38203a | 3761 | |
6a0f1f6d LP |
3762 | r = event_arm_timer(e, &e->realtime); |
3763 | if (r < 0) | |
c45a5a74 | 3764 | return r; |
6a0f1f6d | 3765 | |
a8548816 TG |
3766 | r = event_arm_timer(e, &e->boottime); |
3767 | if (r < 0) | |
c45a5a74 | 3768 | return r; |
a8548816 | 3769 | |
6a0f1f6d LP |
3770 | r = event_arm_timer(e, &e->monotonic); |
3771 | if (r < 0) | |
c45a5a74 | 3772 | return r; |
6a0f1f6d LP |
3773 | |
3774 | r = event_arm_timer(e, &e->realtime_alarm); | |
1b5995b0 | 3775 | if (r < 0) |
c45a5a74 | 3776 | return r; |
fd38203a | 3777 | |
6a0f1f6d | 3778 | r = event_arm_timer(e, &e->boottime_alarm); |
1b5995b0 | 3779 | if (r < 0) |
c45a5a74 | 3780 | return r; |
fd38203a | 3781 | |
97ef5391 LP |
3782 | event_close_inode_data_fds(e); |
3783 | ||
1b5995b0 | 3784 | if (event_next_pending(e) || e->need_process_child) |
c45a5a74 TG |
3785 | goto pending; |
3786 | ||
2b0c9ef7 | 3787 | e->state = SD_EVENT_ARMED; |
c45a5a74 TG |
3788 | |
3789 | return 0; | |
3790 | ||
3791 | pending: | |
2b0c9ef7 | 3792 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
3793 | r = sd_event_wait(e, 0); |
3794 | if (r == 0) | |
2b0c9ef7 | 3795 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
3796 | |
3797 | return r; | |
c45a5a74 TG |
3798 | } |
3799 | ||
798445ab LP |
3800 | static int epoll_wait_usec( |
3801 | int fd, | |
3802 | struct epoll_event *events, | |
3803 | int maxevents, | |
3804 | usec_t timeout) { | |
3805 | ||
798445ab | 3806 | int r, msec; |
39f756d3 ZJS |
3807 | #if 0 |
3808 | static bool epoll_pwait2_absent = false; | |
798445ab | 3809 | |
39f756d3 ZJS |
3810 | /* A wrapper that uses epoll_pwait2() if available, and falls back to epoll_wait() if not. |
3811 | * | |
3812 | * FIXME: this is temporarily disabled until epoll_pwait2() becomes more widely available. | |
3813 | * See https://github.com/systemd/systemd/pull/18973 and | |
3814 | * https://github.com/systemd/systemd/issues/19052. */ | |
798445ab LP |
3815 | |
3816 | if (!epoll_pwait2_absent && timeout != USEC_INFINITY) { | |
3817 | struct timespec ts; | |
3818 | ||
3819 | r = epoll_pwait2(fd, | |
3820 | events, | |
3821 | maxevents, | |
3822 | timespec_store(&ts, timeout), | |
3823 | NULL); | |
3824 | if (r >= 0) | |
3825 | return r; | |
7cb45dbf | 3826 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) |
798445ab LP |
3827 | return -errno; /* Only fallback to old epoll_wait() if the syscall is masked or not |
3828 | * supported. */ | |
3829 | ||
3830 | epoll_pwait2_absent = true; | |
3831 | } | |
39f756d3 | 3832 | #endif |
798445ab LP |
3833 | |
3834 | if (timeout == USEC_INFINITY) | |
3835 | msec = -1; | |
3836 | else { | |
3837 | usec_t k; | |
3838 | ||
3839 | k = DIV_ROUND_UP(timeout, USEC_PER_MSEC); | |
3840 | if (k >= INT_MAX) | |
3841 | msec = INT_MAX; /* Saturate */ | |
3842 | else | |
3843 | msec = (int) k; | |
3844 | } | |
3845 | ||
3846 | r = epoll_wait(fd, | |
3847 | events, | |
3848 | maxevents, | |
3849 | msec); | |
3850 | if (r < 0) | |
3851 | return -errno; | |
3852 | ||
3853 | return r; | |
3854 | } | |
3855 | ||
efd3be9d | 3856 | static int process_epoll(sd_event *e, usec_t timeout, int64_t threshold, int64_t *ret_min_priority) { |
319a4f4b | 3857 | size_t n_event_queue, m, n_event_max; |
efd3be9d YW |
3858 | int64_t min_priority = threshold; |
3859 | bool something_new = false; | |
798445ab | 3860 | int r; |
c45a5a74 | 3861 | |
efd3be9d YW |
3862 | assert(e); |
3863 | assert(ret_min_priority); | |
6a0f1f6d | 3864 | |
8b9708d1 | 3865 | n_event_queue = MAX(e->n_sources, 1u); |
319a4f4b | 3866 | if (!GREEDY_REALLOC(e->event_queue, n_event_queue)) |
5cddd924 | 3867 | return -ENOMEM; |
fd38203a | 3868 | |
319a4f4b LP |
3869 | n_event_max = MALLOC_ELEMENTSOF(e->event_queue); |
3870 | ||
97ef5391 LP |
3871 | /* If we still have inotify data buffered, then query the other fds, but don't wait on it */ |
3872 | if (e->inotify_data_buffered) | |
798445ab | 3873 | timeout = 0; |
97ef5391 | 3874 | |
8b9708d1 | 3875 | for (;;) { |
319a4f4b LP |
3876 | r = epoll_wait_usec( |
3877 | e->epoll_fd, | |
3878 | e->event_queue, | |
3879 | n_event_max, | |
3880 | timeout); | |
798445ab | 3881 | if (r < 0) |
efd3be9d | 3882 | return r; |
c45a5a74 | 3883 | |
8b9708d1 YW |
3884 | m = (size_t) r; |
3885 | ||
319a4f4b | 3886 | if (m < n_event_max) |
8b9708d1 YW |
3887 | break; |
3888 | ||
319a4f4b | 3889 | if (n_event_max >= n_event_queue * 10) |
8b9708d1 YW |
3890 | break; |
3891 | ||
319a4f4b | 3892 | if (!GREEDY_REALLOC(e->event_queue, n_event_max + n_event_queue)) |
8b9708d1 YW |
3893 | return -ENOMEM; |
3894 | ||
319a4f4b | 3895 | n_event_max = MALLOC_ELEMENTSOF(e->event_queue); |
798445ab | 3896 | timeout = 0; |
da7e457c | 3897 | } |
fd38203a | 3898 | |
efd3be9d YW |
3899 | /* Set timestamp only when this is called first time. */ |
3900 | if (threshold == INT64_MAX) | |
3901 | triple_timestamp_get(&e->timestamp); | |
fd38203a | 3902 | |
8b9708d1 | 3903 | for (size_t i = 0; i < m; i++) { |
fd38203a | 3904 | |
5cddd924 LP |
3905 | if (e->event_queue[i].data.ptr == INT_TO_PTR(SOURCE_WATCHDOG)) |
3906 | r = flush_timer(e, e->watchdog_fd, e->event_queue[i].events, NULL); | |
9da4cb2b | 3907 | else { |
5cddd924 | 3908 | WakeupType *t = e->event_queue[i].data.ptr; |
9da4cb2b LP |
3909 | |
3910 | switch (*t) { | |
3911 | ||
f8f3f926 | 3912 | case WAKEUP_EVENT_SOURCE: { |
5cddd924 | 3913 | sd_event_source *s = e->event_queue[i].data.ptr; |
f8f3f926 LP |
3914 | |
3915 | assert(s); | |
3916 | ||
efd3be9d YW |
3917 | if (s->priority > threshold) |
3918 | continue; | |
3919 | ||
3920 | min_priority = MIN(min_priority, s->priority); | |
3921 | ||
f8f3f926 LP |
3922 | switch (s->type) { |
3923 | ||
3924 | case SOURCE_IO: | |
5cddd924 | 3925 | r = process_io(e, s, e->event_queue[i].events); |
f8f3f926 LP |
3926 | break; |
3927 | ||
3928 | case SOURCE_CHILD: | |
5cddd924 | 3929 | r = process_pidfd(e, s, e->event_queue[i].events); |
f8f3f926 LP |
3930 | break; |
3931 | ||
3932 | default: | |
3933 | assert_not_reached("Unexpected event source type"); | |
3934 | } | |
3935 | ||
9da4cb2b | 3936 | break; |
f8f3f926 | 3937 | } |
fd38203a | 3938 | |
9da4cb2b | 3939 | case WAKEUP_CLOCK_DATA: { |
5cddd924 | 3940 | struct clock_data *d = e->event_queue[i].data.ptr; |
f8f3f926 LP |
3941 | |
3942 | assert(d); | |
3943 | ||
5cddd924 | 3944 | r = flush_timer(e, d->fd, e->event_queue[i].events, &d->next); |
9da4cb2b LP |
3945 | break; |
3946 | } | |
3947 | ||
3948 | case WAKEUP_SIGNAL_DATA: | |
efd3be9d | 3949 | r = process_signal(e, e->event_queue[i].data.ptr, e->event_queue[i].events, &min_priority); |
9da4cb2b LP |
3950 | break; |
3951 | ||
97ef5391 | 3952 | case WAKEUP_INOTIFY_DATA: |
efd3be9d | 3953 | r = event_inotify_data_read(e, e->event_queue[i].data.ptr, e->event_queue[i].events, threshold); |
97ef5391 LP |
3954 | break; |
3955 | ||
9da4cb2b LP |
3956 | default: |
3957 | assert_not_reached("Invalid wake-up pointer"); | |
3958 | } | |
3959 | } | |
efd3be9d YW |
3960 | if (r < 0) |
3961 | return r; | |
3962 | if (r > 0) | |
3963 | something_new = true; | |
3964 | } | |
3965 | ||
3966 | *ret_min_priority = min_priority; | |
3967 | return something_new; | |
3968 | } | |
3969 | ||
3970 | _public_ int sd_event_wait(sd_event *e, uint64_t timeout) { | |
3971 | int r; | |
3972 | ||
3973 | assert_return(e, -EINVAL); | |
3974 | assert_return(e = event_resolve(e), -ENOPKG); | |
3975 | assert_return(!event_pid_changed(e), -ECHILD); | |
3976 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
3977 | assert_return(e->state == SD_EVENT_ARMED, -EBUSY); | |
3978 | ||
3979 | if (e->exit_requested) { | |
3980 | e->state = SD_EVENT_PENDING; | |
3981 | return 1; | |
3982 | } | |
3983 | ||
3984 | for (int64_t threshold = INT64_MAX; ; threshold--) { | |
3985 | int64_t epoll_min_priority, child_min_priority; | |
3986 | ||
3987 | /* There may be a possibility that new epoll (especially IO) and child events are | |
3988 | * triggered just after process_epoll() call but before process_child(), and the new IO | |
3989 | * events may have higher priority than the child events. To salvage these events, | |
3990 | * let's call epoll_wait() again, but accepts only events with higher priority than the | |
3991 | * previous. See issue https://github.com/systemd/systemd/issues/18190 and comments | |
3992 | * https://github.com/systemd/systemd/pull/18750#issuecomment-785801085 | |
3993 | * https://github.com/systemd/systemd/pull/18922#issuecomment-792825226 */ | |
3994 | ||
3995 | r = process_epoll(e, timeout, threshold, &epoll_min_priority); | |
3996 | if (r == -EINTR) { | |
3997 | e->state = SD_EVENT_PENDING; | |
3998 | return 1; | |
3999 | } | |
4000 | if (r < 0) | |
4001 | goto finish; | |
4002 | if (r == 0 && threshold < INT64_MAX) | |
4003 | /* No new epoll event. */ | |
4004 | break; | |
4005 | ||
4006 | r = process_child(e, threshold, &child_min_priority); | |
fd38203a | 4007 | if (r < 0) |
da7e457c | 4008 | goto finish; |
efd3be9d YW |
4009 | if (r == 0) |
4010 | /* No new child event. */ | |
4011 | break; | |
4012 | ||
4013 | threshold = MIN(epoll_min_priority, child_min_priority); | |
4014 | if (threshold == INT64_MIN) | |
4015 | break; | |
4016 | ||
4017 | timeout = 0; | |
fd38203a LP |
4018 | } |
4019 | ||
cde93897 LP |
4020 | r = process_watchdog(e); |
4021 | if (r < 0) | |
4022 | goto finish; | |
4023 | ||
6a0f1f6d LP |
4024 | r = process_timer(e, e->timestamp.realtime, &e->realtime); |
4025 | if (r < 0) | |
4026 | goto finish; | |
4027 | ||
e475d10c | 4028 | r = process_timer(e, e->timestamp.boottime, &e->boottime); |
a8548816 TG |
4029 | if (r < 0) |
4030 | goto finish; | |
4031 | ||
6a0f1f6d LP |
4032 | r = process_timer(e, e->timestamp.monotonic, &e->monotonic); |
4033 | if (r < 0) | |
4034 | goto finish; | |
4035 | ||
4036 | r = process_timer(e, e->timestamp.realtime, &e->realtime_alarm); | |
fd38203a | 4037 | if (r < 0) |
da7e457c | 4038 | goto finish; |
fd38203a | 4039 | |
e475d10c | 4040 | r = process_timer(e, e->timestamp.boottime, &e->boottime_alarm); |
fd38203a | 4041 | if (r < 0) |
da7e457c | 4042 | goto finish; |
fd38203a | 4043 | |
97ef5391 LP |
4044 | r = process_inotify(e); |
4045 | if (r < 0) | |
4046 | goto finish; | |
4047 | ||
c45a5a74 TG |
4048 | if (event_next_pending(e)) { |
4049 | e->state = SD_EVENT_PENDING; | |
c45a5a74 | 4050 | return 1; |
da7e457c LP |
4051 | } |
4052 | ||
c45a5a74 | 4053 | r = 0; |
fd38203a | 4054 | |
da7e457c | 4055 | finish: |
2b0c9ef7 | 4056 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
4057 | |
4058 | return r; | |
fd38203a LP |
4059 | } |
4060 | ||
c45a5a74 TG |
4061 | _public_ int sd_event_dispatch(sd_event *e) { |
4062 | sd_event_source *p; | |
4063 | int r; | |
4064 | ||
4065 | assert_return(e, -EINVAL); | |
b937d761 | 4066 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
4067 | assert_return(!event_pid_changed(e), -ECHILD); |
4068 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
4069 | assert_return(e->state == SD_EVENT_PENDING, -EBUSY); | |
4070 | ||
4071 | if (e->exit_requested) | |
4072 | return dispatch_exit(e); | |
4073 | ||
4074 | p = event_next_pending(e); | |
4075 | if (p) { | |
f814c871 | 4076 | _unused_ _cleanup_(sd_event_unrefp) sd_event *ref = sd_event_ref(e); |
c45a5a74 TG |
4077 | |
4078 | e->state = SD_EVENT_RUNNING; | |
4079 | r = source_dispatch(p); | |
2b0c9ef7 | 4080 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
4081 | return r; |
4082 | } | |
4083 | ||
2b0c9ef7 | 4084 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
4085 | |
4086 | return 1; | |
4087 | } | |
4088 | ||
34b87517 | 4089 | static void event_log_delays(sd_event *e) { |
442ac269 YW |
4090 | char b[ELEMENTSOF(e->delays) * DECIMAL_STR_MAX(unsigned) + 1], *p; |
4091 | size_t l, i; | |
34b87517 | 4092 | |
442ac269 YW |
4093 | p = b; |
4094 | l = sizeof(b); | |
4095 | for (i = 0; i < ELEMENTSOF(e->delays); i++) { | |
4096 | l = strpcpyf(&p, l, "%u ", e->delays[i]); | |
34b87517 VC |
4097 | e->delays[i] = 0; |
4098 | } | |
442ac269 | 4099 | log_debug("Event loop iterations: %s", b); |
34b87517 VC |
4100 | } |
4101 | ||
c45a5a74 TG |
4102 | _public_ int sd_event_run(sd_event *e, uint64_t timeout) { |
4103 | int r; | |
4104 | ||
4105 | assert_return(e, -EINVAL); | |
b937d761 | 4106 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
4107 | assert_return(!event_pid_changed(e), -ECHILD); |
4108 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 4109 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
c45a5a74 | 4110 | |
e6a7bee5 | 4111 | if (e->profile_delays && e->last_run_usec != 0) { |
34b87517 VC |
4112 | usec_t this_run; |
4113 | unsigned l; | |
4114 | ||
4115 | this_run = now(CLOCK_MONOTONIC); | |
4116 | ||
e6a7bee5 | 4117 | l = u64log2(this_run - e->last_run_usec); |
cb9d621e | 4118 | assert(l < ELEMENTSOF(e->delays)); |
34b87517 VC |
4119 | e->delays[l]++; |
4120 | ||
e6a7bee5 | 4121 | if (this_run - e->last_log_usec >= 5*USEC_PER_SEC) { |
34b87517 | 4122 | event_log_delays(e); |
e6a7bee5 | 4123 | e->last_log_usec = this_run; |
34b87517 VC |
4124 | } |
4125 | } | |
4126 | ||
f814c871 LP |
4127 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
4128 | _unused_ _cleanup_(sd_event_unrefp) sd_event *ref = sd_event_ref(e); | |
4129 | ||
c45a5a74 | 4130 | r = sd_event_prepare(e); |
53bac4e0 LP |
4131 | if (r == 0) |
4132 | /* There was nothing? Then wait... */ | |
4133 | r = sd_event_wait(e, timeout); | |
c45a5a74 | 4134 | |
34b87517 | 4135 | if (e->profile_delays) |
e6a7bee5 | 4136 | e->last_run_usec = now(CLOCK_MONOTONIC); |
34b87517 | 4137 | |
02d30981 | 4138 | if (r > 0) { |
53bac4e0 | 4139 | /* There's something now, then let's dispatch it */ |
02d30981 TG |
4140 | r = sd_event_dispatch(e); |
4141 | if (r < 0) | |
4142 | return r; | |
53bac4e0 LP |
4143 | |
4144 | return 1; | |
4145 | } | |
4146 | ||
4147 | return r; | |
c45a5a74 TG |
4148 | } |
4149 | ||
f7262a9f | 4150 | _public_ int sd_event_loop(sd_event *e) { |
fd38203a LP |
4151 | int r; |
4152 | ||
da7e457c | 4153 | assert_return(e, -EINVAL); |
b937d761 | 4154 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 4155 | assert_return(!event_pid_changed(e), -ECHILD); |
2b0c9ef7 | 4156 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 4157 | |
f814c871 | 4158 | _unused_ _cleanup_(sd_event_unrefp) sd_event *ref = NULL; |
fd38203a | 4159 | |
da7e457c | 4160 | while (e->state != SD_EVENT_FINISHED) { |
f5fbe71d | 4161 | r = sd_event_run(e, UINT64_MAX); |
fd38203a | 4162 | if (r < 0) |
30dd293c | 4163 | return r; |
fd38203a LP |
4164 | } |
4165 | ||
30dd293c | 4166 | return e->exit_code; |
fd38203a LP |
4167 | } |
4168 | ||
9b364545 | 4169 | _public_ int sd_event_get_fd(sd_event *e) { |
9b364545 | 4170 | assert_return(e, -EINVAL); |
b937d761 | 4171 | assert_return(e = event_resolve(e), -ENOPKG); |
9b364545 TG |
4172 | assert_return(!event_pid_changed(e), -ECHILD); |
4173 | ||
4174 | return e->epoll_fd; | |
4175 | } | |
4176 | ||
f7262a9f | 4177 | _public_ int sd_event_get_state(sd_event *e) { |
da7e457c | 4178 | assert_return(e, -EINVAL); |
b937d761 | 4179 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
4180 | assert_return(!event_pid_changed(e), -ECHILD); |
4181 | ||
4182 | return e->state; | |
4183 | } | |
4184 | ||
6203e07a | 4185 | _public_ int sd_event_get_exit_code(sd_event *e, int *code) { |
da7e457c | 4186 | assert_return(e, -EINVAL); |
b937d761 | 4187 | assert_return(e = event_resolve(e), -ENOPKG); |
6203e07a | 4188 | assert_return(code, -EINVAL); |
da7e457c | 4189 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 4190 | |
6203e07a LP |
4191 | if (!e->exit_requested) |
4192 | return -ENODATA; | |
4193 | ||
4194 | *code = e->exit_code; | |
4195 | return 0; | |
fd38203a LP |
4196 | } |
4197 | ||
6203e07a | 4198 | _public_ int sd_event_exit(sd_event *e, int code) { |
da7e457c | 4199 | assert_return(e, -EINVAL); |
b937d761 | 4200 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
4201 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
4202 | assert_return(!event_pid_changed(e), -ECHILD); | |
fd38203a | 4203 | |
6203e07a LP |
4204 | e->exit_requested = true; |
4205 | e->exit_code = code; | |
4206 | ||
fd38203a LP |
4207 | return 0; |
4208 | } | |
46e8c825 | 4209 | |
6a0f1f6d | 4210 | _public_ int sd_event_now(sd_event *e, clockid_t clock, uint64_t *usec) { |
46e8c825 | 4211 | assert_return(e, -EINVAL); |
b937d761 | 4212 | assert_return(e = event_resolve(e), -ENOPKG); |
46e8c825 | 4213 | assert_return(usec, -EINVAL); |
46e8c825 LP |
4214 | assert_return(!event_pid_changed(e), -ECHILD); |
4215 | ||
e475d10c LP |
4216 | if (!TRIPLE_TIMESTAMP_HAS_CLOCK(clock)) |
4217 | return -EOPNOTSUPP; | |
4218 | ||
4219 | /* Generate a clean error in case CLOCK_BOOTTIME is not available. Note that don't use clock_supported() here, | |
4220 | * for a reason: there are systems where CLOCK_BOOTTIME is supported, but CLOCK_BOOTTIME_ALARM is not, but for | |
4221 | * the purpose of getting the time this doesn't matter. */ | |
3411372e LP |
4222 | if (IN_SET(clock, CLOCK_BOOTTIME, CLOCK_BOOTTIME_ALARM) && !clock_boottime_supported()) |
4223 | return -EOPNOTSUPP; | |
4224 | ||
e475d10c | 4225 | if (!triple_timestamp_is_set(&e->timestamp)) { |
15c689d7 | 4226 | /* Implicitly fall back to now() if we never ran before and thus have no cached time. */ |
38a03f06 LP |
4227 | *usec = now(clock); |
4228 | return 1; | |
4229 | } | |
46e8c825 | 4230 | |
e475d10c | 4231 | *usec = triple_timestamp_by_clock(&e->timestamp, clock); |
46e8c825 LP |
4232 | return 0; |
4233 | } | |
afc6adb5 LP |
4234 | |
4235 | _public_ int sd_event_default(sd_event **ret) { | |
39883f62 | 4236 | sd_event *e = NULL; |
afc6adb5 LP |
4237 | int r; |
4238 | ||
4239 | if (!ret) | |
4240 | return !!default_event; | |
4241 | ||
4242 | if (default_event) { | |
4243 | *ret = sd_event_ref(default_event); | |
4244 | return 0; | |
4245 | } | |
4246 | ||
4247 | r = sd_event_new(&e); | |
4248 | if (r < 0) | |
4249 | return r; | |
4250 | ||
4251 | e->default_event_ptr = &default_event; | |
4252 | e->tid = gettid(); | |
4253 | default_event = e; | |
4254 | ||
4255 | *ret = e; | |
4256 | return 1; | |
4257 | } | |
4258 | ||
4259 | _public_ int sd_event_get_tid(sd_event *e, pid_t *tid) { | |
4260 | assert_return(e, -EINVAL); | |
b937d761 | 4261 | assert_return(e = event_resolve(e), -ENOPKG); |
afc6adb5 | 4262 | assert_return(tid, -EINVAL); |
76b54375 | 4263 | assert_return(!event_pid_changed(e), -ECHILD); |
afc6adb5 | 4264 | |
76b54375 LP |
4265 | if (e->tid != 0) { |
4266 | *tid = e->tid; | |
4267 | return 0; | |
4268 | } | |
4269 | ||
4270 | return -ENXIO; | |
afc6adb5 | 4271 | } |
cde93897 LP |
4272 | |
4273 | _public_ int sd_event_set_watchdog(sd_event *e, int b) { | |
4274 | int r; | |
4275 | ||
4276 | assert_return(e, -EINVAL); | |
b937d761 | 4277 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 | 4278 | assert_return(!event_pid_changed(e), -ECHILD); |
cde93897 LP |
4279 | |
4280 | if (e->watchdog == !!b) | |
4281 | return e->watchdog; | |
4282 | ||
4283 | if (b) { | |
09812eb7 LP |
4284 | r = sd_watchdog_enabled(false, &e->watchdog_period); |
4285 | if (r <= 0) | |
cde93897 | 4286 | return r; |
cde93897 LP |
4287 | |
4288 | /* Issue first ping immediately */ | |
4289 | sd_notify(false, "WATCHDOG=1"); | |
4290 | e->watchdog_last = now(CLOCK_MONOTONIC); | |
4291 | ||
4292 | e->watchdog_fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC); | |
4293 | if (e->watchdog_fd < 0) | |
4294 | return -errno; | |
4295 | ||
4296 | r = arm_watchdog(e); | |
4297 | if (r < 0) | |
4298 | goto fail; | |
4299 | ||
1eac7948 | 4300 | struct epoll_event ev = { |
a82f89aa LP |
4301 | .events = EPOLLIN, |
4302 | .data.ptr = INT_TO_PTR(SOURCE_WATCHDOG), | |
4303 | }; | |
cde93897 | 4304 | |
15c689d7 | 4305 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, e->watchdog_fd, &ev) < 0) { |
cde93897 LP |
4306 | r = -errno; |
4307 | goto fail; | |
4308 | } | |
4309 | ||
4310 | } else { | |
4311 | if (e->watchdog_fd >= 0) { | |
5a795bff | 4312 | (void) epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, e->watchdog_fd, NULL); |
03e334a1 | 4313 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
4314 | } |
4315 | } | |
4316 | ||
4317 | e->watchdog = !!b; | |
4318 | return e->watchdog; | |
4319 | ||
4320 | fail: | |
03e334a1 | 4321 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
4322 | return r; |
4323 | } | |
8f726607 LP |
4324 | |
4325 | _public_ int sd_event_get_watchdog(sd_event *e) { | |
4326 | assert_return(e, -EINVAL); | |
b937d761 | 4327 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 LP |
4328 | assert_return(!event_pid_changed(e), -ECHILD); |
4329 | ||
4330 | return e->watchdog; | |
4331 | } | |
60a3b1e1 LP |
4332 | |
4333 | _public_ int sd_event_get_iteration(sd_event *e, uint64_t *ret) { | |
4334 | assert_return(e, -EINVAL); | |
b937d761 | 4335 | assert_return(e = event_resolve(e), -ENOPKG); |
60a3b1e1 LP |
4336 | assert_return(!event_pid_changed(e), -ECHILD); |
4337 | ||
4338 | *ret = e->iteration; | |
4339 | return 0; | |
4340 | } | |
15723a1d LP |
4341 | |
4342 | _public_ int sd_event_source_set_destroy_callback(sd_event_source *s, sd_event_destroy_t callback) { | |
4343 | assert_return(s, -EINVAL); | |
4344 | ||
4345 | s->destroy_callback = callback; | |
4346 | return 0; | |
4347 | } | |
4348 | ||
4349 | _public_ int sd_event_source_get_destroy_callback(sd_event_source *s, sd_event_destroy_t *ret) { | |
4350 | assert_return(s, -EINVAL); | |
4351 | ||
4352 | if (ret) | |
4353 | *ret = s->destroy_callback; | |
4354 | ||
4355 | return !!s->destroy_callback; | |
4356 | } | |
2382c936 YW |
4357 | |
4358 | _public_ int sd_event_source_get_floating(sd_event_source *s) { | |
4359 | assert_return(s, -EINVAL); | |
4360 | ||
4361 | return s->floating; | |
4362 | } | |
4363 | ||
4364 | _public_ int sd_event_source_set_floating(sd_event_source *s, int b) { | |
4365 | assert_return(s, -EINVAL); | |
4366 | ||
4367 | if (s->floating == !!b) | |
4368 | return 0; | |
4369 | ||
4370 | if (!s->event) /* Already disconnected */ | |
4371 | return -ESTALE; | |
4372 | ||
4373 | s->floating = b; | |
4374 | ||
4375 | if (b) { | |
4376 | sd_event_source_ref(s); | |
4377 | sd_event_unref(s->event); | |
4378 | } else { | |
4379 | sd_event_ref(s->event); | |
4380 | sd_event_source_unref(s); | |
4381 | } | |
4382 | ||
4383 | return 1; | |
4384 | } | |
b778cba4 LP |
4385 | |
4386 | _public_ int sd_event_source_get_exit_on_failure(sd_event_source *s) { | |
4387 | assert_return(s, -EINVAL); | |
4388 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
4389 | ||
4390 | return s->exit_on_failure; | |
4391 | } | |
4392 | ||
4393 | _public_ int sd_event_source_set_exit_on_failure(sd_event_source *s, int b) { | |
4394 | assert_return(s, -EINVAL); | |
4395 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
4396 | ||
4397 | if (s->exit_on_failure == !!b) | |
4398 | return 0; | |
4399 | ||
4400 | s->exit_on_failure = b; | |
4401 | return 1; | |
4402 | } | |
b6d5481b LP |
4403 | |
4404 | _public_ int sd_event_source_set_ratelimit(sd_event_source *s, uint64_t interval, unsigned burst) { | |
4405 | int r; | |
4406 | ||
4407 | assert_return(s, -EINVAL); | |
4408 | ||
4409 | /* Turning on ratelimiting on event source types that don't support it, is a loggable offense. Doing | |
4410 | * so is a programming error. */ | |
4411 | assert_return(EVENT_SOURCE_CAN_RATE_LIMIT(s->type), -EDOM); | |
4412 | ||
4413 | /* When ratelimiting is configured we'll always reset the rate limit state first and start fresh, | |
4414 | * non-ratelimited. */ | |
4415 | r = event_source_leave_ratelimit(s); | |
4416 | if (r < 0) | |
4417 | return r; | |
4418 | ||
4419 | s->rate_limit = (RateLimit) { interval, burst }; | |
4420 | return 0; | |
4421 | } | |
4422 | ||
4423 | _public_ int sd_event_source_get_ratelimit(sd_event_source *s, uint64_t *ret_interval, unsigned *ret_burst) { | |
4424 | assert_return(s, -EINVAL); | |
4425 | ||
4426 | /* Querying whether an event source has ratelimiting configured is not a loggable offsense, hence | |
4427 | * don't use assert_return(). Unlike turning on ratelimiting it's not really a programming error */ | |
4428 | if (!EVENT_SOURCE_CAN_RATE_LIMIT(s->type)) | |
4429 | return -EDOM; | |
4430 | ||
4431 | if (!ratelimit_configured(&s->rate_limit)) | |
4432 | return -ENOEXEC; | |
4433 | ||
4434 | if (ret_interval) | |
4435 | *ret_interval = s->rate_limit.interval; | |
4436 | if (ret_burst) | |
4437 | *ret_burst = s->rate_limit.burst; | |
4438 | ||
4439 | return 0; | |
4440 | } | |
4441 | ||
4442 | _public_ int sd_event_source_is_ratelimited(sd_event_source *s) { | |
4443 | assert_return(s, -EINVAL); | |
4444 | ||
4445 | if (!EVENT_SOURCE_CAN_RATE_LIMIT(s->type)) | |
4446 | return false; | |
4447 | ||
4448 | if (!ratelimit_configured(&s->rate_limit)) | |
4449 | return false; | |
4450 | ||
4451 | return s->ratelimited; | |
4452 | } |