]>
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" |
28e5e1e9 | 16 | #include "glyph-util.h" |
fd38203a | 17 | #include "hashmap.h" |
07630cea | 18 | #include "list.h" |
3ae6b3bf | 19 | #include "logarithm.h" |
07630cea | 20 | #include "macro.h" |
0a970718 | 21 | #include "memory-util.h" |
f5947a5e | 22 | #include "missing_syscall.h" |
07630cea | 23 | #include "prioq.h" |
4a0b58c4 | 24 | #include "process-util.h" |
6e9feda3 | 25 | #include "set.h" |
24882e06 | 26 | #include "signal-util.h" |
55cbfaa5 | 27 | #include "string-table.h" |
07630cea | 28 | #include "string-util.h" |
442ac269 | 29 | #include "strxcpyx.h" |
07630cea | 30 | #include "time-util.h" |
fd38203a | 31 | |
c2ba3ad6 | 32 | #define DEFAULT_ACCURACY_USEC (250 * USEC_PER_MSEC) |
fd38203a | 33 | |
f8f3f926 LP |
34 | static bool EVENT_SOURCE_WATCH_PIDFD(sd_event_source *s) { |
35 | /* Returns true if this is a PID event source and can be implemented by watching EPOLLIN */ | |
36 | return s && | |
37 | s->type == SOURCE_CHILD && | |
38 | s->child.pidfd >= 0 && | |
39 | s->child.options == WEXITED; | |
40 | } | |
41 | ||
b6d5481b LP |
42 | static bool event_source_is_online(sd_event_source *s) { |
43 | assert(s); | |
44 | return s->enabled != SD_EVENT_OFF && !s->ratelimited; | |
45 | } | |
46 | ||
47 | static bool event_source_is_offline(sd_event_source *s) { | |
48 | assert(s); | |
49 | return s->enabled == SD_EVENT_OFF || s->ratelimited; | |
50 | } | |
51 | ||
55cbfaa5 | 52 | static const char* const event_source_type_table[_SOURCE_EVENT_SOURCE_TYPE_MAX] = { |
3f3548f8 ZJS |
53 | [SOURCE_IO] = "io", |
54 | [SOURCE_TIME_REALTIME] = "realtime", | |
55 | [SOURCE_TIME_BOOTTIME] = "bootime", | |
56 | [SOURCE_TIME_MONOTONIC] = "monotonic", | |
55cbfaa5 DM |
57 | [SOURCE_TIME_REALTIME_ALARM] = "realtime-alarm", |
58 | [SOURCE_TIME_BOOTTIME_ALARM] = "boottime-alarm", | |
3f3548f8 ZJS |
59 | [SOURCE_SIGNAL] = "signal", |
60 | [SOURCE_CHILD] = "child", | |
61 | [SOURCE_DEFER] = "defer", | |
62 | [SOURCE_POST] = "post", | |
63 | [SOURCE_EXIT] = "exit", | |
64 | [SOURCE_WATCHDOG] = "watchdog", | |
65 | [SOURCE_INOTIFY] = "inotify", | |
55cbfaa5 DM |
66 | }; |
67 | ||
68 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(event_source_type, int); | |
69 | ||
b6d5481b LP |
70 | #define EVENT_SOURCE_IS_TIME(t) \ |
71 | IN_SET((t), \ | |
72 | SOURCE_TIME_REALTIME, \ | |
73 | SOURCE_TIME_BOOTTIME, \ | |
74 | SOURCE_TIME_MONOTONIC, \ | |
75 | SOURCE_TIME_REALTIME_ALARM, \ | |
76 | SOURCE_TIME_BOOTTIME_ALARM) | |
77 | ||
78 | #define EVENT_SOURCE_CAN_RATE_LIMIT(t) \ | |
79 | IN_SET((t), \ | |
80 | SOURCE_IO, \ | |
81 | SOURCE_TIME_REALTIME, \ | |
82 | SOURCE_TIME_BOOTTIME, \ | |
83 | SOURCE_TIME_MONOTONIC, \ | |
84 | SOURCE_TIME_REALTIME_ALARM, \ | |
85 | SOURCE_TIME_BOOTTIME_ALARM, \ | |
86 | SOURCE_SIGNAL, \ | |
87 | SOURCE_DEFER, \ | |
88 | SOURCE_INOTIFY) | |
6a0f1f6d | 89 | |
19947509 ZJS |
90 | /* This is used to assert that we didn't pass an unexpected source type to event_source_time_prioq_put(). |
91 | * Time sources and ratelimited sources can be passed, so effectively this is the same as the | |
92 | * EVENT_SOURCE_CAN_RATE_LIMIT() macro. */ | |
93 | #define EVENT_SOURCE_USES_TIME_PRIOQ(t) EVENT_SOURCE_CAN_RATE_LIMIT(t) | |
94 | ||
fd38203a | 95 | struct sd_event { |
da7e457c | 96 | unsigned n_ref; |
fd38203a LP |
97 | |
98 | int epoll_fd; | |
cde93897 | 99 | int watchdog_fd; |
fd38203a LP |
100 | |
101 | Prioq *pending; | |
102 | Prioq *prepare; | |
c2ba3ad6 | 103 | |
a8548816 | 104 | /* timerfd_create() only supports these five clocks so far. We |
6a0f1f6d LP |
105 | * can add support for more clocks when the kernel learns to |
106 | * deal with them, too. */ | |
107 | struct clock_data realtime; | |
a8548816 | 108 | struct clock_data boottime; |
6a0f1f6d LP |
109 | struct clock_data monotonic; |
110 | struct clock_data realtime_alarm; | |
111 | struct clock_data boottime_alarm; | |
fd38203a | 112 | |
da7e457c LP |
113 | usec_t perturb; |
114 | ||
9da4cb2b LP |
115 | sd_event_source **signal_sources; /* indexed by signal number */ |
116 | Hashmap *signal_data; /* indexed by priority */ | |
fd38203a LP |
117 | |
118 | Hashmap *child_sources; | |
b6d5481b | 119 | unsigned n_online_child_sources; |
fd38203a | 120 | |
6e9feda3 LP |
121 | Set *post_sources; |
122 | ||
6203e07a | 123 | Prioq *exit; |
fd38203a | 124 | |
97ef5391 LP |
125 | Hashmap *inotify_data; /* indexed by priority */ |
126 | ||
127 | /* A list of inode structures that still have an fd open, that we need to close before the next loop iteration */ | |
ed828563 | 128 | LIST_HEAD(struct inode_data, inode_data_to_close_list); |
97ef5391 LP |
129 | |
130 | /* A list of inotify objects that already have events buffered which aren't processed yet */ | |
0601b958 | 131 | LIST_HEAD(struct inotify_data, buffered_inotify_data_list); |
97ef5391 | 132 | |
da7e457c | 133 | pid_t original_pid; |
c2ba3ad6 | 134 | |
60a3b1e1 | 135 | uint64_t iteration; |
e475d10c | 136 | triple_timestamp timestamp; |
da7e457c | 137 | int state; |
eaa3cbef | 138 | |
6203e07a | 139 | bool exit_requested:1; |
da7e457c | 140 | bool need_process_child:1; |
cde93897 | 141 | bool watchdog:1; |
34b87517 | 142 | bool profile_delays:1; |
afc6adb5 | 143 | |
6203e07a LP |
144 | int exit_code; |
145 | ||
afc6adb5 LP |
146 | pid_t tid; |
147 | sd_event **default_event_ptr; | |
cde93897 LP |
148 | |
149 | usec_t watchdog_last, watchdog_period; | |
15b38f93 LP |
150 | |
151 | unsigned n_sources; | |
a71fe8b8 | 152 | |
5cddd924 | 153 | struct epoll_event *event_queue; |
5cddd924 | 154 | |
a71fe8b8 | 155 | LIST_HEAD(sd_event_source, sources); |
34b87517 | 156 | |
baf3fdec LP |
157 | sd_event_source *sigint_event_source, *sigterm_event_source; |
158 | ||
e6a7bee5 | 159 | usec_t last_run_usec, last_log_usec; |
34b87517 | 160 | unsigned delays[sizeof(usec_t) * 8]; |
fd38203a LP |
161 | }; |
162 | ||
b937d761 NM |
163 | static thread_local sd_event *default_event = NULL; |
164 | ||
a71fe8b8 | 165 | static void source_disconnect(sd_event_source *s); |
97ef5391 | 166 | static void event_gc_inode_data(sd_event *e, struct inode_data *d); |
a71fe8b8 | 167 | |
b937d761 NM |
168 | static sd_event *event_resolve(sd_event *e) { |
169 | return e == SD_EVENT_DEFAULT ? default_event : e; | |
170 | } | |
171 | ||
fd38203a LP |
172 | static int pending_prioq_compare(const void *a, const void *b) { |
173 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 174 | int r; |
fd38203a LP |
175 | |
176 | assert(x->pending); | |
177 | assert(y->pending); | |
178 | ||
baf76283 | 179 | /* Enabled ones first */ |
06e13147 YW |
180 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
181 | if (r != 0) | |
182 | return r; | |
fd38203a | 183 | |
b6d5481b LP |
184 | /* Non rate-limited ones first. */ |
185 | r = CMP(!!x->ratelimited, !!y->ratelimited); | |
186 | if (r != 0) | |
187 | return r; | |
188 | ||
fd38203a | 189 | /* Lower priority values first */ |
9c57a73b YW |
190 | r = CMP(x->priority, y->priority); |
191 | if (r != 0) | |
192 | return r; | |
fd38203a LP |
193 | |
194 | /* Older entries first */ | |
9c57a73b | 195 | return CMP(x->pending_iteration, y->pending_iteration); |
fd38203a LP |
196 | } |
197 | ||
198 | static int prepare_prioq_compare(const void *a, const void *b) { | |
199 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 200 | int r; |
fd38203a LP |
201 | |
202 | assert(x->prepare); | |
203 | assert(y->prepare); | |
204 | ||
8046c457 | 205 | /* Enabled ones first */ |
06e13147 YW |
206 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
207 | if (r != 0) | |
208 | return r; | |
8046c457 | 209 | |
b6d5481b LP |
210 | /* Non rate-limited ones first. */ |
211 | r = CMP(!!x->ratelimited, !!y->ratelimited); | |
212 | if (r != 0) | |
213 | return r; | |
214 | ||
fd38203a LP |
215 | /* Move most recently prepared ones last, so that we can stop |
216 | * preparing as soon as we hit one that has already been | |
217 | * prepared in the current iteration */ | |
9c57a73b YW |
218 | r = CMP(x->prepare_iteration, y->prepare_iteration); |
219 | if (r != 0) | |
220 | return r; | |
fd38203a | 221 | |
fd38203a | 222 | /* Lower priority values first */ |
9c57a73b | 223 | return CMP(x->priority, y->priority); |
fd38203a LP |
224 | } |
225 | ||
b6d5481b LP |
226 | static usec_t time_event_source_next(const sd_event_source *s) { |
227 | assert(s); | |
228 | ||
229 | /* We have two kinds of event sources that have elapsation times associated with them: the actual | |
230 | * time based ones and the ones for which a ratelimit can be in effect (where we want to be notified | |
231 | * once the ratelimit time window ends). Let's return the next elapsing time depending on what we are | |
232 | * looking at here. */ | |
233 | ||
234 | if (s->ratelimited) { /* If rate-limited the next elapsation is when the ratelimit time window ends */ | |
235 | assert(s->rate_limit.begin != 0); | |
236 | assert(s->rate_limit.interval != 0); | |
237 | return usec_add(s->rate_limit.begin, s->rate_limit.interval); | |
238 | } | |
239 | ||
240 | /* Otherwise this must be a time event source, if not ratelimited */ | |
241 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
242 | return s->time.next; | |
243 | ||
244 | return USEC_INFINITY; | |
245 | } | |
246 | ||
1bce0ffa | 247 | static usec_t time_event_source_latest(const sd_event_source *s) { |
b6d5481b LP |
248 | assert(s); |
249 | ||
250 | if (s->ratelimited) { /* For ratelimited stuff the earliest and the latest time shall actually be the | |
251 | * same, as we should avoid adding additional inaccuracy on an inaccuracy time | |
252 | * window */ | |
253 | assert(s->rate_limit.begin != 0); | |
254 | assert(s->rate_limit.interval != 0); | |
255 | return usec_add(s->rate_limit.begin, s->rate_limit.interval); | |
256 | } | |
257 | ||
258 | /* Must be a time event source, if not ratelimited */ | |
259 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
260 | return usec_add(s->time.next, s->time.accuracy); | |
261 | ||
262 | return USEC_INFINITY; | |
1bce0ffa LP |
263 | } |
264 | ||
81107b84 LP |
265 | static bool event_source_timer_candidate(const sd_event_source *s) { |
266 | assert(s); | |
267 | ||
268 | /* Returns true for event sources that either are not pending yet (i.e. where it's worth to mark them pending) | |
269 | * or which are currently ratelimited (i.e. where it's worth leaving the ratelimited state) */ | |
270 | return !s->pending || s->ratelimited; | |
271 | } | |
272 | ||
273 | static int time_prioq_compare(const void *a, const void *b, usec_t (*time_func)(const sd_event_source *s)) { | |
c2ba3ad6 | 274 | const sd_event_source *x = a, *y = b; |
06e13147 | 275 | int r; |
c2ba3ad6 | 276 | |
baf76283 | 277 | /* Enabled ones first */ |
06e13147 YW |
278 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
279 | if (r != 0) | |
280 | return r; | |
c2ba3ad6 | 281 | |
81107b84 | 282 | /* Order "non-pending OR ratelimited" before "pending AND not-ratelimited" */ |
06e13147 YW |
283 | r = CMP(!event_source_timer_candidate(x), !event_source_timer_candidate(y)); |
284 | if (r != 0) | |
285 | return r; | |
c2ba3ad6 LP |
286 | |
287 | /* Order by time */ | |
81107b84 LP |
288 | return CMP(time_func(x), time_func(y)); |
289 | } | |
290 | ||
291 | static int earliest_time_prioq_compare(const void *a, const void *b) { | |
292 | return time_prioq_compare(a, b, time_event_source_next); | |
293 | } | |
294 | ||
295 | static int latest_time_prioq_compare(const void *a, const void *b) { | |
296 | return time_prioq_compare(a, b, time_event_source_latest); | |
c2ba3ad6 LP |
297 | } |
298 | ||
6203e07a | 299 | static int exit_prioq_compare(const void *a, const void *b) { |
da7e457c | 300 | const sd_event_source *x = a, *y = b; |
06e13147 | 301 | int r; |
da7e457c | 302 | |
6203e07a LP |
303 | assert(x->type == SOURCE_EXIT); |
304 | assert(y->type == SOURCE_EXIT); | |
da7e457c | 305 | |
baf76283 | 306 | /* Enabled ones first */ |
06e13147 YW |
307 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
308 | if (r != 0) | |
309 | return r; | |
da7e457c LP |
310 | |
311 | /* Lower priority values first */ | |
6dd91b36 | 312 | return CMP(x->priority, y->priority); |
da7e457c LP |
313 | } |
314 | ||
6a0f1f6d LP |
315 | static void free_clock_data(struct clock_data *d) { |
316 | assert(d); | |
9da4cb2b | 317 | assert(d->wakeup == WAKEUP_CLOCK_DATA); |
6a0f1f6d LP |
318 | |
319 | safe_close(d->fd); | |
320 | prioq_free(d->earliest); | |
321 | prioq_free(d->latest); | |
322 | } | |
323 | ||
8301aa0b | 324 | static sd_event *event_free(sd_event *e) { |
a71fe8b8 LP |
325 | sd_event_source *s; |
326 | ||
fd38203a | 327 | assert(e); |
a71fe8b8 | 328 | |
baf3fdec LP |
329 | e->sigterm_event_source = sd_event_source_unref(e->sigterm_event_source); |
330 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
331 | ||
a71fe8b8 LP |
332 | while ((s = e->sources)) { |
333 | assert(s->floating); | |
334 | source_disconnect(s); | |
335 | sd_event_source_unref(s); | |
336 | } | |
337 | ||
15b38f93 | 338 | assert(e->n_sources == 0); |
fd38203a | 339 | |
afc6adb5 LP |
340 | if (e->default_event_ptr) |
341 | *(e->default_event_ptr) = NULL; | |
342 | ||
03e334a1 | 343 | safe_close(e->epoll_fd); |
03e334a1 | 344 | safe_close(e->watchdog_fd); |
cde93897 | 345 | |
6a0f1f6d | 346 | free_clock_data(&e->realtime); |
a8548816 | 347 | free_clock_data(&e->boottime); |
6a0f1f6d LP |
348 | free_clock_data(&e->monotonic); |
349 | free_clock_data(&e->realtime_alarm); | |
350 | free_clock_data(&e->boottime_alarm); | |
351 | ||
fd38203a LP |
352 | prioq_free(e->pending); |
353 | prioq_free(e->prepare); | |
6203e07a | 354 | prioq_free(e->exit); |
fd38203a LP |
355 | |
356 | free(e->signal_sources); | |
9da4cb2b | 357 | hashmap_free(e->signal_data); |
fd38203a | 358 | |
97ef5391 LP |
359 | hashmap_free(e->inotify_data); |
360 | ||
fd38203a | 361 | hashmap_free(e->child_sources); |
6e9feda3 | 362 | set_free(e->post_sources); |
8301aa0b | 363 | |
5cddd924 LP |
364 | free(e->event_queue); |
365 | ||
8301aa0b | 366 | return mfree(e); |
fd38203a LP |
367 | } |
368 | ||
f7262a9f | 369 | _public_ int sd_event_new(sd_event** ret) { |
fd38203a LP |
370 | sd_event *e; |
371 | int r; | |
372 | ||
305f78bf | 373 | assert_return(ret, -EINVAL); |
fd38203a | 374 | |
d08eb1fa | 375 | e = new(sd_event, 1); |
fd38203a LP |
376 | if (!e) |
377 | return -ENOMEM; | |
378 | ||
d08eb1fa LP |
379 | *e = (sd_event) { |
380 | .n_ref = 1, | |
381 | .epoll_fd = -1, | |
382 | .watchdog_fd = -1, | |
383 | .realtime.wakeup = WAKEUP_CLOCK_DATA, | |
384 | .realtime.fd = -1, | |
385 | .realtime.next = USEC_INFINITY, | |
386 | .boottime.wakeup = WAKEUP_CLOCK_DATA, | |
387 | .boottime.fd = -1, | |
388 | .boottime.next = USEC_INFINITY, | |
389 | .monotonic.wakeup = WAKEUP_CLOCK_DATA, | |
390 | .monotonic.fd = -1, | |
391 | .monotonic.next = USEC_INFINITY, | |
392 | .realtime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
393 | .realtime_alarm.fd = -1, | |
394 | .realtime_alarm.next = USEC_INFINITY, | |
395 | .boottime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
396 | .boottime_alarm.fd = -1, | |
397 | .boottime_alarm.next = USEC_INFINITY, | |
398 | .perturb = USEC_INFINITY, | |
399 | .original_pid = getpid_cached(), | |
400 | }; | |
fd38203a | 401 | |
c983e776 EV |
402 | r = prioq_ensure_allocated(&e->pending, pending_prioq_compare); |
403 | if (r < 0) | |
fd38203a | 404 | goto fail; |
fd38203a LP |
405 | |
406 | e->epoll_fd = epoll_create1(EPOLL_CLOEXEC); | |
407 | if (e->epoll_fd < 0) { | |
408 | r = -errno; | |
409 | goto fail; | |
410 | } | |
411 | ||
7fe2903c LP |
412 | e->epoll_fd = fd_move_above_stdio(e->epoll_fd); |
413 | ||
34b87517 | 414 | if (secure_getenv("SD_EVENT_PROFILE_DELAYS")) { |
28e5e1e9 DT |
415 | log_debug("Event loop profiling enabled. Logarithmic histogram of event loop iterations in the range 2^0 %s 2^63 us will be logged every 5s.", |
416 | special_glyph(SPECIAL_GLYPH_ELLIPSIS)); | |
34b87517 VC |
417 | e->profile_delays = true; |
418 | } | |
419 | ||
fd38203a LP |
420 | *ret = e; |
421 | return 0; | |
422 | ||
423 | fail: | |
424 | event_free(e); | |
425 | return r; | |
426 | } | |
427 | ||
8301aa0b | 428 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event, sd_event, event_free); |
c8e9d15c YW |
429 | #define PROTECT_EVENT(e) \ |
430 | _unused_ _cleanup_(sd_event_unrefp) sd_event *_ref = sd_event_ref(e); | |
fd38203a | 431 | |
afd15bbb ZJS |
432 | _public_ sd_event_source* sd_event_source_disable_unref(sd_event_source *s) { |
433 | if (s) | |
434 | (void) sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
435 | return sd_event_source_unref(s); | |
436 | } | |
437 | ||
eaa3cbef LP |
438 | static bool event_pid_changed(sd_event *e) { |
439 | assert(e); | |
440 | ||
a2360a46 | 441 | /* We don't support people creating an event loop and keeping |
eaa3cbef LP |
442 | * it around over a fork(). Let's complain. */ |
443 | ||
df0ff127 | 444 | return e->original_pid != getpid_cached(); |
eaa3cbef LP |
445 | } |
446 | ||
366e6411 | 447 | static void source_io_unregister(sd_event_source *s) { |
fd38203a LP |
448 | assert(s); |
449 | assert(s->type == SOURCE_IO); | |
450 | ||
f6806734 | 451 | if (event_pid_changed(s->event)) |
366e6411 | 452 | return; |
f6806734 | 453 | |
fd38203a | 454 | if (!s->io.registered) |
366e6411 | 455 | return; |
fd38203a | 456 | |
d1cf2023 | 457 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->io.fd, NULL) < 0) |
f80a5d6a | 458 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll, ignoring: %m", |
55cbfaa5 | 459 | strna(s->description), event_source_type_to_string(s->type)); |
fd38203a LP |
460 | |
461 | s->io.registered = false; | |
fd38203a LP |
462 | } |
463 | ||
305f78bf LP |
464 | static int source_io_register( |
465 | sd_event_source *s, | |
466 | int enabled, | |
467 | uint32_t events) { | |
468 | ||
fd38203a LP |
469 | assert(s); |
470 | assert(s->type == SOURCE_IO); | |
baf76283 | 471 | assert(enabled != SD_EVENT_OFF); |
fd38203a | 472 | |
1eac7948 | 473 | struct epoll_event ev = { |
a82f89aa LP |
474 | .events = events | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0), |
475 | .data.ptr = s, | |
476 | }; | |
fd38203a | 477 | |
15c689d7 | 478 | if (epoll_ctl(s->event->epoll_fd, |
1eac7948 | 479 | s->io.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, |
55c540d3 | 480 | s->io.fd, &ev) < 0) |
fd38203a LP |
481 | return -errno; |
482 | ||
483 | s->io.registered = true; | |
484 | ||
485 | return 0; | |
486 | } | |
487 | ||
f8f3f926 LP |
488 | static void source_child_pidfd_unregister(sd_event_source *s) { |
489 | assert(s); | |
490 | assert(s->type == SOURCE_CHILD); | |
491 | ||
492 | if (event_pid_changed(s->event)) | |
493 | return; | |
494 | ||
495 | if (!s->child.registered) | |
496 | return; | |
497 | ||
498 | if (EVENT_SOURCE_WATCH_PIDFD(s)) | |
499 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->child.pidfd, NULL) < 0) | |
f80a5d6a | 500 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll, ignoring: %m", |
f8f3f926 LP |
501 | strna(s->description), event_source_type_to_string(s->type)); |
502 | ||
503 | s->child.registered = false; | |
504 | } | |
505 | ||
506 | static int source_child_pidfd_register(sd_event_source *s, int enabled) { | |
f8f3f926 LP |
507 | assert(s); |
508 | assert(s->type == SOURCE_CHILD); | |
509 | assert(enabled != SD_EVENT_OFF); | |
510 | ||
511 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { | |
1eac7948 | 512 | struct epoll_event ev = { |
f8f3f926 LP |
513 | .events = EPOLLIN | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0), |
514 | .data.ptr = s, | |
515 | }; | |
516 | ||
55c540d3 ZJS |
517 | if (epoll_ctl(s->event->epoll_fd, |
518 | s->child.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, | |
519 | s->child.pidfd, &ev) < 0) | |
f8f3f926 LP |
520 | return -errno; |
521 | } | |
522 | ||
523 | s->child.registered = true; | |
524 | return 0; | |
525 | } | |
526 | ||
6a0f1f6d LP |
527 | static clockid_t event_source_type_to_clock(EventSourceType t) { |
528 | ||
529 | switch (t) { | |
530 | ||
531 | case SOURCE_TIME_REALTIME: | |
532 | return CLOCK_REALTIME; | |
533 | ||
a8548816 TG |
534 | case SOURCE_TIME_BOOTTIME: |
535 | return CLOCK_BOOTTIME; | |
536 | ||
6a0f1f6d LP |
537 | case SOURCE_TIME_MONOTONIC: |
538 | return CLOCK_MONOTONIC; | |
539 | ||
540 | case SOURCE_TIME_REALTIME_ALARM: | |
541 | return CLOCK_REALTIME_ALARM; | |
542 | ||
543 | case SOURCE_TIME_BOOTTIME_ALARM: | |
544 | return CLOCK_BOOTTIME_ALARM; | |
545 | ||
546 | default: | |
547 | return (clockid_t) -1; | |
548 | } | |
549 | } | |
550 | ||
551 | static EventSourceType clock_to_event_source_type(clockid_t clock) { | |
552 | ||
553 | switch (clock) { | |
554 | ||
555 | case CLOCK_REALTIME: | |
556 | return SOURCE_TIME_REALTIME; | |
557 | ||
a8548816 TG |
558 | case CLOCK_BOOTTIME: |
559 | return SOURCE_TIME_BOOTTIME; | |
560 | ||
6a0f1f6d LP |
561 | case CLOCK_MONOTONIC: |
562 | return SOURCE_TIME_MONOTONIC; | |
563 | ||
564 | case CLOCK_REALTIME_ALARM: | |
565 | return SOURCE_TIME_REALTIME_ALARM; | |
566 | ||
567 | case CLOCK_BOOTTIME_ALARM: | |
568 | return SOURCE_TIME_BOOTTIME_ALARM; | |
569 | ||
570 | default: | |
571 | return _SOURCE_EVENT_SOURCE_TYPE_INVALID; | |
572 | } | |
573 | } | |
574 | ||
575 | static struct clock_data* event_get_clock_data(sd_event *e, EventSourceType t) { | |
576 | assert(e); | |
577 | ||
578 | switch (t) { | |
579 | ||
580 | case SOURCE_TIME_REALTIME: | |
581 | return &e->realtime; | |
582 | ||
a8548816 TG |
583 | case SOURCE_TIME_BOOTTIME: |
584 | return &e->boottime; | |
585 | ||
6a0f1f6d LP |
586 | case SOURCE_TIME_MONOTONIC: |
587 | return &e->monotonic; | |
588 | ||
589 | case SOURCE_TIME_REALTIME_ALARM: | |
590 | return &e->realtime_alarm; | |
591 | ||
592 | case SOURCE_TIME_BOOTTIME_ALARM: | |
593 | return &e->boottime_alarm; | |
594 | ||
595 | default: | |
596 | return NULL; | |
597 | } | |
598 | } | |
599 | ||
3e4eb8e7 YW |
600 | static void event_free_signal_data(sd_event *e, struct signal_data *d) { |
601 | assert(e); | |
602 | ||
603 | if (!d) | |
604 | return; | |
605 | ||
606 | hashmap_remove(e->signal_data, &d->priority); | |
607 | safe_close(d->fd); | |
608 | free(d); | |
609 | } | |
610 | ||
9da4cb2b LP |
611 | static int event_make_signal_data( |
612 | sd_event *e, | |
613 | int sig, | |
614 | struct signal_data **ret) { | |
4807d2d0 | 615 | |
9da4cb2b LP |
616 | struct signal_data *d; |
617 | bool added = false; | |
618 | sigset_t ss_copy; | |
619 | int64_t priority; | |
f95387cd ZJS |
620 | int r; |
621 | ||
622 | assert(e); | |
623 | ||
f6806734 | 624 | if (event_pid_changed(e)) |
9da4cb2b | 625 | return -ECHILD; |
f6806734 | 626 | |
9da4cb2b LP |
627 | if (e->signal_sources && e->signal_sources[sig]) |
628 | priority = e->signal_sources[sig]->priority; | |
629 | else | |
de05913d | 630 | priority = SD_EVENT_PRIORITY_NORMAL; |
f95387cd | 631 | |
9da4cb2b LP |
632 | d = hashmap_get(e->signal_data, &priority); |
633 | if (d) { | |
634 | if (sigismember(&d->sigset, sig) > 0) { | |
635 | if (ret) | |
636 | *ret = d; | |
637 | return 0; | |
638 | } | |
639 | } else { | |
d08eb1fa | 640 | d = new(struct signal_data, 1); |
9da4cb2b LP |
641 | if (!d) |
642 | return -ENOMEM; | |
643 | ||
d08eb1fa LP |
644 | *d = (struct signal_data) { |
645 | .wakeup = WAKEUP_SIGNAL_DATA, | |
646 | .fd = -1, | |
647 | .priority = priority, | |
648 | }; | |
9da4cb2b | 649 | |
f656fdb6 | 650 | r = hashmap_ensure_put(&e->signal_data, &uint64_hash_ops, &d->priority, d); |
90f604d1 ZJS |
651 | if (r < 0) { |
652 | free(d); | |
9da4cb2b | 653 | return r; |
90f604d1 | 654 | } |
f95387cd | 655 | |
9da4cb2b LP |
656 | added = true; |
657 | } | |
658 | ||
659 | ss_copy = d->sigset; | |
660 | assert_se(sigaddset(&ss_copy, sig) >= 0); | |
661 | ||
662 | r = signalfd(d->fd, &ss_copy, SFD_NONBLOCK|SFD_CLOEXEC); | |
663 | if (r < 0) { | |
664 | r = -errno; | |
665 | goto fail; | |
666 | } | |
667 | ||
668 | d->sigset = ss_copy; | |
f95387cd | 669 | |
9da4cb2b LP |
670 | if (d->fd >= 0) { |
671 | if (ret) | |
672 | *ret = d; | |
f95387cd | 673 | return 0; |
9da4cb2b LP |
674 | } |
675 | ||
7fe2903c | 676 | d->fd = fd_move_above_stdio(r); |
f95387cd | 677 | |
1eac7948 | 678 | struct epoll_event ev = { |
a82f89aa LP |
679 | .events = EPOLLIN, |
680 | .data.ptr = d, | |
681 | }; | |
f95387cd | 682 | |
15c689d7 | 683 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { |
9da4cb2b LP |
684 | r = -errno; |
685 | goto fail; | |
f95387cd ZJS |
686 | } |
687 | ||
9da4cb2b LP |
688 | if (ret) |
689 | *ret = d; | |
690 | ||
f95387cd | 691 | return 0; |
9da4cb2b LP |
692 | |
693 | fail: | |
3e4eb8e7 YW |
694 | if (added) |
695 | event_free_signal_data(e, d); | |
9da4cb2b LP |
696 | |
697 | return r; | |
698 | } | |
699 | ||
700 | static void event_unmask_signal_data(sd_event *e, struct signal_data *d, int sig) { | |
701 | assert(e); | |
702 | assert(d); | |
703 | ||
704 | /* Turns off the specified signal in the signal data | |
705 | * object. If the signal mask of the object becomes empty that | |
706 | * way removes it. */ | |
707 | ||
708 | if (sigismember(&d->sigset, sig) == 0) | |
709 | return; | |
710 | ||
711 | assert_se(sigdelset(&d->sigset, sig) >= 0); | |
712 | ||
713 | if (sigisemptyset(&d->sigset)) { | |
9da4cb2b | 714 | /* If all the mask is all-zero we can get rid of the structure */ |
3e4eb8e7 | 715 | event_free_signal_data(e, d); |
9da4cb2b LP |
716 | return; |
717 | } | |
718 | ||
01e6af73 YW |
719 | if (event_pid_changed(e)) |
720 | return; | |
721 | ||
9da4cb2b LP |
722 | assert(d->fd >= 0); |
723 | ||
724 | if (signalfd(d->fd, &d->sigset, SFD_NONBLOCK|SFD_CLOEXEC) < 0) | |
725 | log_debug_errno(errno, "Failed to unset signal bit, ignoring: %m"); | |
726 | } | |
727 | ||
728 | static void event_gc_signal_data(sd_event *e, const int64_t *priority, int sig) { | |
729 | struct signal_data *d; | |
730 | static const int64_t zero_priority = 0; | |
731 | ||
732 | assert(e); | |
733 | ||
f8f3f926 LP |
734 | /* Rechecks if the specified signal is still something we are interested in. If not, we'll unmask it, |
735 | * and possibly drop the signalfd for it. */ | |
9da4cb2b LP |
736 | |
737 | if (sig == SIGCHLD && | |
b6d5481b | 738 | e->n_online_child_sources > 0) |
9da4cb2b LP |
739 | return; |
740 | ||
741 | if (e->signal_sources && | |
742 | e->signal_sources[sig] && | |
b6d5481b | 743 | event_source_is_online(e->signal_sources[sig])) |
9da4cb2b LP |
744 | return; |
745 | ||
746 | /* | |
747 | * The specified signal might be enabled in three different queues: | |
748 | * | |
749 | * 1) the one that belongs to the priority passed (if it is non-NULL) | |
750 | * 2) the one that belongs to the priority of the event source of the signal (if there is one) | |
751 | * 3) the 0 priority (to cover the SIGCHLD case) | |
752 | * | |
753 | * Hence, let's remove it from all three here. | |
754 | */ | |
755 | ||
756 | if (priority) { | |
757 | d = hashmap_get(e->signal_data, priority); | |
758 | if (d) | |
759 | event_unmask_signal_data(e, d, sig); | |
760 | } | |
761 | ||
762 | if (e->signal_sources && e->signal_sources[sig]) { | |
763 | d = hashmap_get(e->signal_data, &e->signal_sources[sig]->priority); | |
764 | if (d) | |
765 | event_unmask_signal_data(e, d, sig); | |
766 | } | |
767 | ||
768 | d = hashmap_get(e->signal_data, &zero_priority); | |
769 | if (d) | |
770 | event_unmask_signal_data(e, d, sig); | |
f95387cd ZJS |
771 | } |
772 | ||
e1951c16 MS |
773 | static void event_source_pp_prioq_reshuffle(sd_event_source *s) { |
774 | assert(s); | |
775 | ||
776 | /* Reshuffles the pending + prepare prioqs. Called whenever the dispatch order changes, i.e. when | |
777 | * they are enabled/disabled or marked pending and such. */ | |
778 | ||
779 | if (s->pending) | |
780 | prioq_reshuffle(s->event->pending, s, &s->pending_index); | |
781 | ||
782 | if (s->prepare) | |
783 | prioq_reshuffle(s->event->prepare, s, &s->prepare_index); | |
784 | } | |
785 | ||
786 | static void event_source_time_prioq_reshuffle(sd_event_source *s) { | |
787 | struct clock_data *d; | |
788 | ||
789 | assert(s); | |
e1951c16 MS |
790 | |
791 | /* Called whenever the event source's timer ordering properties changed, i.e. time, accuracy, | |
5c08c7ab YW |
792 | * pending, enable state, and ratelimiting state. Makes sure the two prioq's are ordered |
793 | * properly again. */ | |
b6d5481b LP |
794 | |
795 | if (s->ratelimited) | |
796 | d = &s->event->monotonic; | |
5c08c7ab | 797 | else if (EVENT_SOURCE_IS_TIME(s->type)) |
b6d5481b | 798 | assert_se(d = event_get_clock_data(s->event, s->type)); |
5c08c7ab YW |
799 | else |
800 | return; /* no-op for an event source which is neither a timer nor ratelimited. */ | |
b6d5481b | 801 | |
f41315fc LP |
802 | prioq_reshuffle(d->earliest, s, &s->earliest_index); |
803 | prioq_reshuffle(d->latest, s, &s->latest_index); | |
e1951c16 MS |
804 | d->needs_rearm = true; |
805 | } | |
806 | ||
1e45e3fe LP |
807 | static void event_source_time_prioq_remove( |
808 | sd_event_source *s, | |
809 | struct clock_data *d) { | |
810 | ||
811 | assert(s); | |
812 | assert(d); | |
813 | ||
f41315fc LP |
814 | prioq_remove(d->earliest, s, &s->earliest_index); |
815 | prioq_remove(d->latest, s, &s->latest_index); | |
816 | s->earliest_index = s->latest_index = PRIOQ_IDX_NULL; | |
1e45e3fe LP |
817 | d->needs_rearm = true; |
818 | } | |
819 | ||
a71fe8b8 LP |
820 | static void source_disconnect(sd_event_source *s) { |
821 | sd_event *event; | |
897448bd | 822 | int r; |
a71fe8b8 | 823 | |
fd38203a LP |
824 | assert(s); |
825 | ||
a71fe8b8 LP |
826 | if (!s->event) |
827 | return; | |
15b38f93 | 828 | |
a71fe8b8 | 829 | assert(s->event->n_sources > 0); |
fd38203a | 830 | |
a71fe8b8 | 831 | switch (s->type) { |
fd38203a | 832 | |
a71fe8b8 LP |
833 | case SOURCE_IO: |
834 | if (s->io.fd >= 0) | |
835 | source_io_unregister(s); | |
fd38203a | 836 | |
a71fe8b8 | 837 | break; |
6a0f1f6d | 838 | |
a71fe8b8 | 839 | case SOURCE_TIME_REALTIME: |
a8548816 | 840 | case SOURCE_TIME_BOOTTIME: |
a71fe8b8 LP |
841 | case SOURCE_TIME_MONOTONIC: |
842 | case SOURCE_TIME_REALTIME_ALARM: | |
b6d5481b LP |
843 | case SOURCE_TIME_BOOTTIME_ALARM: |
844 | /* Only remove this event source from the time event source here if it is not ratelimited. If | |
845 | * it is ratelimited, we'll remove it below, separately. Why? Because the clock used might | |
846 | * differ: ratelimiting always uses CLOCK_MONOTONIC, but timer events might use any clock */ | |
847 | ||
848 | if (!s->ratelimited) { | |
849 | struct clock_data *d; | |
850 | assert_se(d = event_get_clock_data(s->event, s->type)); | |
851 | event_source_time_prioq_remove(s, d); | |
852 | } | |
853 | ||
a71fe8b8 | 854 | break; |
a71fe8b8 LP |
855 | |
856 | case SOURCE_SIGNAL: | |
857 | if (s->signal.sig > 0) { | |
9da4cb2b | 858 | |
a71fe8b8 LP |
859 | if (s->event->signal_sources) |
860 | s->event->signal_sources[s->signal.sig] = NULL; | |
4807d2d0 | 861 | |
9da4cb2b | 862 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
897448bd LP |
863 | |
864 | if (s->signal.unblock) { | |
865 | sigset_t new_ss; | |
866 | ||
867 | if (sigemptyset(&new_ss) < 0) | |
868 | log_debug_errno(errno, "Failed to reset signal set, ignoring: %m"); | |
869 | else if (sigaddset(&new_ss, s->signal.sig) < 0) | |
870 | log_debug_errno(errno, "Failed to add signal %i to signal mask, ignoring: %m", s->signal.sig); | |
871 | else { | |
872 | r = pthread_sigmask(SIG_UNBLOCK, &new_ss, NULL); | |
873 | if (r != 0) | |
874 | log_debug_errno(r, "Failed to unblock signal %i, ignoring: %m", s->signal.sig); | |
875 | } | |
876 | } | |
6a0f1f6d | 877 | } |
fd38203a | 878 | |
a71fe8b8 | 879 | break; |
fd38203a | 880 | |
a71fe8b8 | 881 | case SOURCE_CHILD: |
86587c93 YW |
882 | if (event_pid_changed(s->event)) |
883 | s->child.process_owned = false; | |
884 | ||
a71fe8b8 | 885 | if (s->child.pid > 0) { |
b6d5481b LP |
886 | if (event_source_is_online(s)) { |
887 | assert(s->event->n_online_child_sources > 0); | |
888 | s->event->n_online_child_sources--; | |
4807d2d0 | 889 | } |
fd38203a | 890 | |
4a0b58c4 | 891 | (void) hashmap_remove(s->event->child_sources, PID_TO_PTR(s->child.pid)); |
a71fe8b8 | 892 | } |
fd38203a | 893 | |
f8f3f926 LP |
894 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
895 | source_child_pidfd_unregister(s); | |
896 | else | |
897 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
898 | ||
a71fe8b8 | 899 | break; |
fd38203a | 900 | |
a71fe8b8 LP |
901 | case SOURCE_DEFER: |
902 | /* nothing */ | |
903 | break; | |
fd38203a | 904 | |
a71fe8b8 LP |
905 | case SOURCE_POST: |
906 | set_remove(s->event->post_sources, s); | |
907 | break; | |
da7e457c | 908 | |
a71fe8b8 LP |
909 | case SOURCE_EXIT: |
910 | prioq_remove(s->event->exit, s, &s->exit.prioq_index); | |
911 | break; | |
0eb2e0e3 | 912 | |
97ef5391 LP |
913 | case SOURCE_INOTIFY: { |
914 | struct inode_data *inode_data; | |
915 | ||
916 | inode_data = s->inotify.inode_data; | |
917 | if (inode_data) { | |
918 | struct inotify_data *inotify_data; | |
919 | assert_se(inotify_data = inode_data->inotify_data); | |
920 | ||
921 | /* Detach this event source from the inode object */ | |
922 | LIST_REMOVE(inotify.by_inode_data, inode_data->event_sources, s); | |
923 | s->inotify.inode_data = NULL; | |
924 | ||
925 | if (s->pending) { | |
926 | assert(inotify_data->n_pending > 0); | |
927 | inotify_data->n_pending--; | |
928 | } | |
929 | ||
930 | /* Note that we don't reduce the inotify mask for the watch descriptor here if the inode is | |
931 | * continued to being watched. That's because inotify doesn't really have an API for that: we | |
932 | * can only change watch masks with access to the original inode either by fd or by path. But | |
933 | * paths aren't stable, and keeping an O_PATH fd open all the time would mean wasting an fd | |
f21f31b2 | 934 | * continuously and keeping the mount busy which we can't really do. We could reconstruct the |
97ef5391 LP |
935 | * original inode from /proc/self/fdinfo/$INOTIFY_FD (as all watch descriptors are listed |
936 | * there), but given the need for open_by_handle_at() which is privileged and not universally | |
937 | * available this would be quite an incomplete solution. Hence we go the other way, leave the | |
938 | * mask set, even if it is not minimized now, and ignore all events we aren't interested in | |
939 | * anymore after reception. Yes, this sucks, but … Linux … */ | |
940 | ||
941 | /* Maybe release the inode data (and its inotify) */ | |
942 | event_gc_inode_data(s->event, inode_data); | |
943 | } | |
944 | ||
945 | break; | |
946 | } | |
947 | ||
a71fe8b8 | 948 | default: |
04499a70 | 949 | assert_not_reached(); |
a71fe8b8 | 950 | } |
6e9feda3 | 951 | |
a71fe8b8 LP |
952 | if (s->pending) |
953 | prioq_remove(s->event->pending, s, &s->pending_index); | |
9d3e3aa5 | 954 | |
a71fe8b8 LP |
955 | if (s->prepare) |
956 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
fd38203a | 957 | |
b6d5481b LP |
958 | if (s->ratelimited) |
959 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
960 | ||
e514aa1e | 961 | event = TAKE_PTR(s->event); |
a71fe8b8 LP |
962 | LIST_REMOVE(sources, event->sources, s); |
963 | event->n_sources--; | |
fd38203a | 964 | |
f5982559 LP |
965 | /* Note that we don't invalidate the type here, since we still need it in order to close the fd or |
966 | * pidfd associated with this event source, which we'll do only on source_free(). */ | |
967 | ||
a71fe8b8 LP |
968 | if (!s->floating) |
969 | sd_event_unref(event); | |
970 | } | |
971 | ||
75db809a | 972 | static sd_event_source* source_free(sd_event_source *s) { |
a71fe8b8 | 973 | assert(s); |
fd38203a | 974 | |
a71fe8b8 | 975 | source_disconnect(s); |
ab93297c NM |
976 | |
977 | if (s->type == SOURCE_IO && s->io.owned) | |
15723a1d LP |
978 | s->io.fd = safe_close(s->io.fd); |
979 | ||
f8f3f926 LP |
980 | if (s->type == SOURCE_CHILD) { |
981 | /* Eventually the kernel will do this automatically for us, but for now let's emulate this (unreliably) in userspace. */ | |
982 | ||
983 | if (s->child.process_owned) { | |
984 | ||
985 | if (!s->child.exited) { | |
986 | bool sent = false; | |
987 | ||
988 | if (s->child.pidfd >= 0) { | |
989 | if (pidfd_send_signal(s->child.pidfd, SIGKILL, NULL, 0) < 0) { | |
990 | if (errno == ESRCH) /* Already dead */ | |
991 | sent = true; | |
992 | else if (!ERRNO_IS_NOT_SUPPORTED(errno)) | |
993 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via pidfd_send_signal(), re-trying via kill(): %m", | |
994 | s->child.pid); | |
995 | } else | |
996 | sent = true; | |
997 | } | |
998 | ||
999 | if (!sent) | |
1000 | if (kill(s->child.pid, SIGKILL) < 0) | |
1001 | if (errno != ESRCH) /* Already dead */ | |
1002 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via kill(), ignoring: %m", | |
1003 | s->child.pid); | |
1004 | } | |
1005 | ||
1006 | if (!s->child.waited) { | |
1007 | siginfo_t si = {}; | |
1008 | ||
1009 | /* Reap the child if we can */ | |
1010 | (void) waitid(P_PID, s->child.pid, &si, WEXITED); | |
1011 | } | |
1012 | } | |
1013 | ||
1014 | if (s->child.pidfd_owned) | |
1015 | s->child.pidfd = safe_close(s->child.pidfd); | |
1016 | } | |
1017 | ||
15723a1d LP |
1018 | if (s->destroy_callback) |
1019 | s->destroy_callback(s->userdata); | |
ab93297c | 1020 | |
356779df | 1021 | free(s->description); |
75db809a | 1022 | return mfree(s); |
fd38203a | 1023 | } |
8c75fe17 | 1024 | DEFINE_TRIVIAL_CLEANUP_FUNC(sd_event_source*, source_free); |
fd38203a LP |
1025 | |
1026 | static int source_set_pending(sd_event_source *s, bool b) { | |
1027 | int r; | |
1028 | ||
1029 | assert(s); | |
6203e07a | 1030 | assert(s->type != SOURCE_EXIT); |
fd38203a LP |
1031 | |
1032 | if (s->pending == b) | |
1033 | return 0; | |
1034 | ||
1035 | s->pending = b; | |
1036 | ||
1037 | if (b) { | |
1038 | s->pending_iteration = s->event->iteration; | |
1039 | ||
1040 | r = prioq_put(s->event->pending, s, &s->pending_index); | |
1041 | if (r < 0) { | |
1042 | s->pending = false; | |
1043 | return r; | |
1044 | } | |
1045 | } else | |
1046 | assert_se(prioq_remove(s->event->pending, s, &s->pending_index)); | |
1047 | ||
e1951c16 MS |
1048 | if (EVENT_SOURCE_IS_TIME(s->type)) |
1049 | event_source_time_prioq_reshuffle(s); | |
2576a19e | 1050 | |
9da4cb2b LP |
1051 | if (s->type == SOURCE_SIGNAL && !b) { |
1052 | struct signal_data *d; | |
1053 | ||
1054 | d = hashmap_get(s->event->signal_data, &s->priority); | |
1055 | if (d && d->current == s) | |
1056 | d->current = NULL; | |
1057 | } | |
1058 | ||
97ef5391 LP |
1059 | if (s->type == SOURCE_INOTIFY) { |
1060 | ||
1061 | assert(s->inotify.inode_data); | |
1062 | assert(s->inotify.inode_data->inotify_data); | |
1063 | ||
1064 | if (b) | |
1065 | s->inotify.inode_data->inotify_data->n_pending ++; | |
1066 | else { | |
1067 | assert(s->inotify.inode_data->inotify_data->n_pending > 0); | |
1068 | s->inotify.inode_data->inotify_data->n_pending --; | |
1069 | } | |
1070 | } | |
1071 | ||
efd3be9d | 1072 | return 1; |
fd38203a LP |
1073 | } |
1074 | ||
a71fe8b8 | 1075 | static sd_event_source *source_new(sd_event *e, bool floating, EventSourceType type) { |
fd38203a LP |
1076 | sd_event_source *s; |
1077 | ||
1078 | assert(e); | |
1079 | ||
d08eb1fa | 1080 | s = new(sd_event_source, 1); |
fd38203a LP |
1081 | if (!s) |
1082 | return NULL; | |
1083 | ||
d08eb1fa LP |
1084 | *s = (struct sd_event_source) { |
1085 | .n_ref = 1, | |
1086 | .event = e, | |
1087 | .floating = floating, | |
1088 | .type = type, | |
1089 | .pending_index = PRIOQ_IDX_NULL, | |
1090 | .prepare_index = PRIOQ_IDX_NULL, | |
1091 | }; | |
a71fe8b8 LP |
1092 | |
1093 | if (!floating) | |
1094 | sd_event_ref(e); | |
fd38203a | 1095 | |
a71fe8b8 | 1096 | LIST_PREPEND(sources, e->sources, s); |
313cefa1 | 1097 | e->n_sources++; |
15b38f93 | 1098 | |
fd38203a LP |
1099 | return s; |
1100 | } | |
1101 | ||
b9350e70 LP |
1102 | static int io_exit_callback(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
1103 | assert(s); | |
1104 | ||
1105 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1106 | } | |
1107 | ||
f7262a9f | 1108 | _public_ int sd_event_add_io( |
fd38203a | 1109 | sd_event *e, |
151b9b96 | 1110 | sd_event_source **ret, |
fd38203a LP |
1111 | int fd, |
1112 | uint32_t events, | |
718db961 | 1113 | sd_event_io_handler_t callback, |
151b9b96 | 1114 | void *userdata) { |
fd38203a | 1115 | |
ec766a51 | 1116 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1117 | int r; |
1118 | ||
305f78bf | 1119 | assert_return(e, -EINVAL); |
b937d761 | 1120 | assert_return(e = event_resolve(e), -ENOPKG); |
8ac43fee | 1121 | assert_return(fd >= 0, -EBADF); |
2a16a986 | 1122 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 1123 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1124 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1125 | |
b9350e70 LP |
1126 | if (!callback) |
1127 | callback = io_exit_callback; | |
1128 | ||
a71fe8b8 | 1129 | s = source_new(e, !ret, SOURCE_IO); |
fd38203a LP |
1130 | if (!s) |
1131 | return -ENOMEM; | |
1132 | ||
9da4cb2b | 1133 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1134 | s->io.fd = fd; |
1135 | s->io.events = events; | |
1136 | s->io.callback = callback; | |
1137 | s->userdata = userdata; | |
baf76283 | 1138 | s->enabled = SD_EVENT_ON; |
fd38203a | 1139 | |
baf76283 | 1140 | r = source_io_register(s, s->enabled, events); |
ec766a51 | 1141 | if (r < 0) |
050f74f2 | 1142 | return r; |
fd38203a | 1143 | |
a71fe8b8 LP |
1144 | if (ret) |
1145 | *ret = s; | |
ec766a51 | 1146 | TAKE_PTR(s); |
a71fe8b8 | 1147 | |
fd38203a LP |
1148 | return 0; |
1149 | } | |
1150 | ||
52444dc4 LP |
1151 | static void initialize_perturb(sd_event *e) { |
1152 | sd_id128_t bootid = {}; | |
1153 | ||
1154 | /* When we sleep for longer, we try to realign the wakeup to | |
f21f31b2 | 1155 | the same time within each minute/second/250ms, so that |
52444dc4 LP |
1156 | events all across the system can be coalesced into a single |
1157 | CPU wakeup. However, let's take some system-specific | |
1158 | randomness for this value, so that in a network of systems | |
1159 | with synced clocks timer events are distributed a | |
1160 | bit. Here, we calculate a perturbation usec offset from the | |
1161 | boot ID. */ | |
1162 | ||
3a43da28 | 1163 | if (_likely_(e->perturb != USEC_INFINITY)) |
52444dc4 LP |
1164 | return; |
1165 | ||
1166 | if (sd_id128_get_boot(&bootid) >= 0) | |
1167 | e->perturb = (bootid.qwords[0] ^ bootid.qwords[1]) % USEC_PER_MINUTE; | |
1168 | } | |
1169 | ||
fd38203a LP |
1170 | static int event_setup_timer_fd( |
1171 | sd_event *e, | |
6a0f1f6d LP |
1172 | struct clock_data *d, |
1173 | clockid_t clock) { | |
fd38203a | 1174 | |
fd38203a | 1175 | assert(e); |
6a0f1f6d | 1176 | assert(d); |
fd38203a | 1177 | |
6a0f1f6d | 1178 | if (_likely_(d->fd >= 0)) |
fd38203a LP |
1179 | return 0; |
1180 | ||
b44d87e2 | 1181 | _cleanup_close_ int fd = -1; |
b44d87e2 | 1182 | |
6a0f1f6d | 1183 | fd = timerfd_create(clock, TFD_NONBLOCK|TFD_CLOEXEC); |
fd38203a LP |
1184 | if (fd < 0) |
1185 | return -errno; | |
1186 | ||
7fe2903c LP |
1187 | fd = fd_move_above_stdio(fd); |
1188 | ||
1eac7948 | 1189 | struct epoll_event ev = { |
a82f89aa LP |
1190 | .events = EPOLLIN, |
1191 | .data.ptr = d, | |
1192 | }; | |
fd38203a | 1193 | |
15c689d7 | 1194 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) |
fd38203a | 1195 | return -errno; |
fd38203a | 1196 | |
b44d87e2 | 1197 | d->fd = TAKE_FD(fd); |
fd38203a LP |
1198 | return 0; |
1199 | } | |
1200 | ||
c4f1aff2 TG |
1201 | static int time_exit_callback(sd_event_source *s, uint64_t usec, void *userdata) { |
1202 | assert(s); | |
1203 | ||
1204 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1205 | } | |
1206 | ||
41c63f36 LP |
1207 | static int setup_clock_data(sd_event *e, struct clock_data *d, clockid_t clock) { |
1208 | int r; | |
1209 | ||
1210 | assert(d); | |
1211 | ||
1212 | if (d->fd < 0) { | |
1213 | r = event_setup_timer_fd(e, d, clock); | |
1214 | if (r < 0) | |
1215 | return r; | |
1216 | } | |
1217 | ||
1218 | r = prioq_ensure_allocated(&d->earliest, earliest_time_prioq_compare); | |
1219 | if (r < 0) | |
1220 | return r; | |
1221 | ||
1222 | r = prioq_ensure_allocated(&d->latest, latest_time_prioq_compare); | |
1223 | if (r < 0) | |
1224 | return r; | |
1225 | ||
1226 | return 0; | |
1227 | } | |
1228 | ||
1e45e3fe LP |
1229 | static int event_source_time_prioq_put( |
1230 | sd_event_source *s, | |
1231 | struct clock_data *d) { | |
1232 | ||
1233 | int r; | |
1234 | ||
1235 | assert(s); | |
1236 | assert(d); | |
19947509 | 1237 | assert(EVENT_SOURCE_USES_TIME_PRIOQ(s->type)); |
1e45e3fe | 1238 | |
f41315fc | 1239 | r = prioq_put(d->earliest, s, &s->earliest_index); |
1e45e3fe LP |
1240 | if (r < 0) |
1241 | return r; | |
1242 | ||
f41315fc | 1243 | r = prioq_put(d->latest, s, &s->latest_index); |
1e45e3fe | 1244 | if (r < 0) { |
f41315fc LP |
1245 | assert_se(prioq_remove(d->earliest, s, &s->earliest_index) > 0); |
1246 | s->earliest_index = PRIOQ_IDX_NULL; | |
1e45e3fe LP |
1247 | return r; |
1248 | } | |
1249 | ||
1250 | d->needs_rearm = true; | |
1251 | return 0; | |
1252 | } | |
1253 | ||
6a0f1f6d | 1254 | _public_ int sd_event_add_time( |
fd38203a | 1255 | sd_event *e, |
151b9b96 | 1256 | sd_event_source **ret, |
6a0f1f6d | 1257 | clockid_t clock, |
fd38203a | 1258 | uint64_t usec, |
c2ba3ad6 | 1259 | uint64_t accuracy, |
718db961 | 1260 | sd_event_time_handler_t callback, |
151b9b96 | 1261 | void *userdata) { |
fd38203a | 1262 | |
6a0f1f6d | 1263 | EventSourceType type; |
ec766a51 | 1264 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6a0f1f6d | 1265 | struct clock_data *d; |
fd38203a LP |
1266 | int r; |
1267 | ||
305f78bf | 1268 | assert_return(e, -EINVAL); |
b937d761 | 1269 | assert_return(e = event_resolve(e), -ENOPKG); |
f5fbe71d | 1270 | assert_return(accuracy != UINT64_MAX, -EINVAL); |
da7e457c | 1271 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1272 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1273 | |
e475d10c LP |
1274 | if (!clock_supported(clock)) /* Checks whether the kernel supports the clock */ |
1275 | return -EOPNOTSUPP; | |
1276 | ||
1277 | type = clock_to_event_source_type(clock); /* checks whether sd-event supports this clock */ | |
1278 | if (type < 0) | |
3411372e LP |
1279 | return -EOPNOTSUPP; |
1280 | ||
c4f1aff2 TG |
1281 | if (!callback) |
1282 | callback = time_exit_callback; | |
1283 | ||
1e45e3fe | 1284 | assert_se(d = event_get_clock_data(e, type)); |
c2ba3ad6 | 1285 | |
41c63f36 | 1286 | r = setup_clock_data(e, d, clock); |
c983e776 EV |
1287 | if (r < 0) |
1288 | return r; | |
fd38203a | 1289 | |
a71fe8b8 | 1290 | s = source_new(e, !ret, type); |
fd38203a LP |
1291 | if (!s) |
1292 | return -ENOMEM; | |
1293 | ||
1294 | s->time.next = usec; | |
c2ba3ad6 | 1295 | s->time.accuracy = accuracy == 0 ? DEFAULT_ACCURACY_USEC : accuracy; |
fd38203a | 1296 | s->time.callback = callback; |
f41315fc | 1297 | s->earliest_index = s->latest_index = PRIOQ_IDX_NULL; |
fd38203a | 1298 | s->userdata = userdata; |
baf76283 | 1299 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1300 | |
1e45e3fe | 1301 | r = event_source_time_prioq_put(s, d); |
c2ba3ad6 | 1302 | if (r < 0) |
ec766a51 | 1303 | return r; |
fd38203a | 1304 | |
a71fe8b8 LP |
1305 | if (ret) |
1306 | *ret = s; | |
ec766a51 | 1307 | TAKE_PTR(s); |
a71fe8b8 | 1308 | |
fd38203a LP |
1309 | return 0; |
1310 | } | |
1311 | ||
d6a83dc4 LP |
1312 | _public_ int sd_event_add_time_relative( |
1313 | sd_event *e, | |
1314 | sd_event_source **ret, | |
1315 | clockid_t clock, | |
1316 | uint64_t usec, | |
1317 | uint64_t accuracy, | |
1318 | sd_event_time_handler_t callback, | |
1319 | void *userdata) { | |
1320 | ||
1321 | usec_t t; | |
1322 | int r; | |
1323 | ||
1324 | /* Same as sd_event_add_time() but operates relative to the event loop's current point in time, and | |
1325 | * checks for overflow. */ | |
1326 | ||
1327 | r = sd_event_now(e, clock, &t); | |
1328 | if (r < 0) | |
1329 | return r; | |
1330 | ||
1331 | if (usec >= USEC_INFINITY - t) | |
1332 | return -EOVERFLOW; | |
1333 | ||
1334 | return sd_event_add_time(e, ret, clock, t + usec, accuracy, callback, userdata); | |
1335 | } | |
1336 | ||
59bc1fd7 LP |
1337 | static int signal_exit_callback(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) { |
1338 | assert(s); | |
1339 | ||
1340 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1341 | } | |
1342 | ||
f7262a9f | 1343 | _public_ int sd_event_add_signal( |
305f78bf | 1344 | sd_event *e, |
151b9b96 | 1345 | sd_event_source **ret, |
305f78bf | 1346 | int sig, |
718db961 | 1347 | sd_event_signal_handler_t callback, |
151b9b96 | 1348 | void *userdata) { |
305f78bf | 1349 | |
ec766a51 | 1350 | _cleanup_(source_freep) sd_event_source *s = NULL; |
9da4cb2b | 1351 | struct signal_data *d; |
897448bd LP |
1352 | sigset_t new_ss; |
1353 | bool block_it; | |
fd38203a LP |
1354 | int r; |
1355 | ||
305f78bf | 1356 | assert_return(e, -EINVAL); |
b937d761 | 1357 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1358 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1359 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1360 | |
897448bd LP |
1361 | /* Let's make sure our special flag stays outside of the valid signal range */ |
1362 | assert_cc(_NSIG < SD_EVENT_SIGNAL_PROCMASK); | |
1363 | ||
1364 | if (sig & SD_EVENT_SIGNAL_PROCMASK) { | |
1365 | sig &= ~SD_EVENT_SIGNAL_PROCMASK; | |
1366 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
1367 | ||
1368 | block_it = true; | |
1369 | } else { | |
1370 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
1371 | ||
1372 | r = signal_is_blocked(sig); | |
1373 | if (r < 0) | |
1374 | return r; | |
1375 | if (r == 0) | |
1376 | return -EBUSY; | |
1377 | ||
1378 | block_it = false; | |
1379 | } | |
1380 | ||
59bc1fd7 LP |
1381 | if (!callback) |
1382 | callback = signal_exit_callback; | |
1383 | ||
fd38203a LP |
1384 | if (!e->signal_sources) { |
1385 | e->signal_sources = new0(sd_event_source*, _NSIG); | |
1386 | if (!e->signal_sources) | |
1387 | return -ENOMEM; | |
1388 | } else if (e->signal_sources[sig]) | |
1389 | return -EBUSY; | |
1390 | ||
a71fe8b8 | 1391 | s = source_new(e, !ret, SOURCE_SIGNAL); |
fd38203a LP |
1392 | if (!s) |
1393 | return -ENOMEM; | |
1394 | ||
1395 | s->signal.sig = sig; | |
1396 | s->signal.callback = callback; | |
1397 | s->userdata = userdata; | |
baf76283 | 1398 | s->enabled = SD_EVENT_ON; |
fd38203a LP |
1399 | |
1400 | e->signal_sources[sig] = s; | |
fd38203a | 1401 | |
897448bd LP |
1402 | if (block_it) { |
1403 | sigset_t old_ss; | |
1404 | ||
1405 | if (sigemptyset(&new_ss) < 0) | |
1406 | return -errno; | |
1407 | ||
1408 | if (sigaddset(&new_ss, sig) < 0) | |
1409 | return -errno; | |
1410 | ||
1411 | r = pthread_sigmask(SIG_BLOCK, &new_ss, &old_ss); | |
1412 | if (r != 0) | |
1413 | return -r; | |
1414 | ||
1415 | r = sigismember(&old_ss, sig); | |
1416 | if (r < 0) | |
1417 | return -errno; | |
1418 | ||
1419 | s->signal.unblock = !r; | |
1420 | } else | |
1421 | s->signal.unblock = false; | |
1422 | ||
9da4cb2b | 1423 | r = event_make_signal_data(e, sig, &d); |
897448bd LP |
1424 | if (r < 0) { |
1425 | if (s->signal.unblock) | |
1426 | (void) pthread_sigmask(SIG_UNBLOCK, &new_ss, NULL); | |
1427 | ||
9da4cb2b | 1428 | return r; |
897448bd | 1429 | } |
fd38203a | 1430 | |
f1f00dbb LP |
1431 | /* Use the signal name as description for the event source by default */ |
1432 | (void) sd_event_source_set_description(s, signal_to_string(sig)); | |
1433 | ||
a71fe8b8 LP |
1434 | if (ret) |
1435 | *ret = s; | |
ec766a51 | 1436 | TAKE_PTR(s); |
a71fe8b8 | 1437 | |
fd38203a LP |
1438 | return 0; |
1439 | } | |
1440 | ||
b9350e70 LP |
1441 | static int child_exit_callback(sd_event_source *s, const siginfo_t *si, void *userdata) { |
1442 | assert(s); | |
1443 | ||
1444 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1445 | } | |
1446 | ||
f8f3f926 LP |
1447 | static bool shall_use_pidfd(void) { |
1448 | /* Mostly relevant for debugging, i.e. this is used in test-event.c to test the event loop once with and once without pidfd */ | |
1449 | return getenv_bool_secure("SYSTEMD_PIDFD") != 0; | |
1450 | } | |
1451 | ||
f7262a9f | 1452 | _public_ int sd_event_add_child( |
305f78bf | 1453 | sd_event *e, |
151b9b96 | 1454 | sd_event_source **ret, |
305f78bf LP |
1455 | pid_t pid, |
1456 | int options, | |
718db961 | 1457 | sd_event_child_handler_t callback, |
151b9b96 | 1458 | void *userdata) { |
305f78bf | 1459 | |
ec766a51 | 1460 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1461 | int r; |
1462 | ||
305f78bf | 1463 | assert_return(e, -EINVAL); |
b937d761 | 1464 | assert_return(e = event_resolve(e), -ENOPKG); |
305f78bf LP |
1465 | assert_return(pid > 1, -EINVAL); |
1466 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1467 | assert_return(options != 0, -EINVAL); | |
da7e457c | 1468 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1469 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1470 | |
b9350e70 LP |
1471 | if (!callback) |
1472 | callback = child_exit_callback; | |
1473 | ||
b6d5481b | 1474 | if (e->n_online_child_sources == 0) { |
ee880b37 LP |
1475 | /* Caller must block SIGCHLD before using us to watch children, even if pidfd is available, |
1476 | * for compatibility with pre-pidfd and because we don't want the reap the child processes | |
1477 | * ourselves, i.e. call waitid(), and don't want Linux' default internal logic for that to | |
1478 | * take effect. | |
1479 | * | |
1480 | * (As an optimization we only do this check on the first child event source created.) */ | |
1481 | r = signal_is_blocked(SIGCHLD); | |
1482 | if (r < 0) | |
1483 | return r; | |
1484 | if (r == 0) | |
1485 | return -EBUSY; | |
1486 | } | |
1487 | ||
d5099efc | 1488 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
fd38203a LP |
1489 | if (r < 0) |
1490 | return r; | |
1491 | ||
4a0b58c4 | 1492 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) |
fd38203a LP |
1493 | return -EBUSY; |
1494 | ||
a71fe8b8 | 1495 | s = source_new(e, !ret, SOURCE_CHILD); |
fd38203a LP |
1496 | if (!s) |
1497 | return -ENOMEM; | |
1498 | ||
f8f3f926 | 1499 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1500 | s->child.options = options; |
1501 | s->child.callback = callback; | |
1502 | s->userdata = userdata; | |
baf76283 | 1503 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1504 | |
f8f3f926 LP |
1505 | /* We always take a pidfd here if we can, even if we wait for anything else than WEXITED, so that we |
1506 | * pin the PID, and make regular waitid() handling race-free. */ | |
1507 | ||
1508 | if (shall_use_pidfd()) { | |
54988a27 | 1509 | s->child.pidfd = pidfd_open(pid, 0); |
f8f3f926 LP |
1510 | if (s->child.pidfd < 0) { |
1511 | /* Propagate errors unless the syscall is not supported or blocked */ | |
1512 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
1513 | return -errno; | |
1514 | } else | |
1515 | s->child.pidfd_owned = true; /* If we allocate the pidfd we own it by default */ | |
1516 | } else | |
1517 | s->child.pidfd = -1; | |
1518 | ||
f8f3f926 LP |
1519 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1520 | /* We have a pidfd and we only want to watch for exit */ | |
f8f3f926 | 1521 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1522 | if (r < 0) |
f8f3f926 | 1523 | return r; |
ac9f2640 | 1524 | |
f8f3f926 LP |
1525 | } else { |
1526 | /* We have no pidfd or we shall wait for some other event than WEXITED */ | |
f8f3f926 | 1527 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1528 | if (r < 0) |
f8f3f926 | 1529 | return r; |
f8f3f926 LP |
1530 | |
1531 | e->need_process_child = true; | |
1532 | } | |
c2ba3ad6 | 1533 | |
54988a27 YW |
1534 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); |
1535 | if (r < 0) | |
1536 | return r; | |
1537 | ||
1538 | /* These must be done after everything succeeds. */ | |
1539 | s->child.pid = pid; | |
b6d5481b | 1540 | e->n_online_child_sources++; |
ac9f2640 | 1541 | |
a71fe8b8 LP |
1542 | if (ret) |
1543 | *ret = s; | |
ec766a51 | 1544 | TAKE_PTR(s); |
f8f3f926 LP |
1545 | return 0; |
1546 | } | |
1547 | ||
1548 | _public_ int sd_event_add_child_pidfd( | |
1549 | sd_event *e, | |
1550 | sd_event_source **ret, | |
1551 | int pidfd, | |
1552 | int options, | |
1553 | sd_event_child_handler_t callback, | |
1554 | void *userdata) { | |
1555 | ||
1556 | ||
1557 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
1558 | pid_t pid; | |
1559 | int r; | |
1560 | ||
1561 | assert_return(e, -EINVAL); | |
1562 | assert_return(e = event_resolve(e), -ENOPKG); | |
1563 | assert_return(pidfd >= 0, -EBADF); | |
1564 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1565 | assert_return(options != 0, -EINVAL); | |
f8f3f926 LP |
1566 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1567 | assert_return(!event_pid_changed(e), -ECHILD); | |
1568 | ||
b9350e70 LP |
1569 | if (!callback) |
1570 | callback = child_exit_callback; | |
1571 | ||
b6d5481b | 1572 | if (e->n_online_child_sources == 0) { |
ee880b37 LP |
1573 | r = signal_is_blocked(SIGCHLD); |
1574 | if (r < 0) | |
1575 | return r; | |
1576 | if (r == 0) | |
1577 | return -EBUSY; | |
1578 | } | |
1579 | ||
f8f3f926 LP |
1580 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
1581 | if (r < 0) | |
1582 | return r; | |
1583 | ||
1584 | r = pidfd_get_pid(pidfd, &pid); | |
1585 | if (r < 0) | |
1586 | return r; | |
1587 | ||
1588 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) | |
1589 | return -EBUSY; | |
1590 | ||
1591 | s = source_new(e, !ret, SOURCE_CHILD); | |
1592 | if (!s) | |
1593 | return -ENOMEM; | |
1594 | ||
1595 | s->wakeup = WAKEUP_EVENT_SOURCE; | |
1596 | s->child.pidfd = pidfd; | |
1597 | s->child.pid = pid; | |
1598 | s->child.options = options; | |
1599 | s->child.callback = callback; | |
1600 | 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) */ | |
1601 | s->userdata = userdata; | |
1602 | s->enabled = SD_EVENT_ONESHOT; | |
1603 | ||
1604 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); | |
1605 | if (r < 0) | |
1606 | return r; | |
1607 | ||
f8f3f926 LP |
1608 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1609 | /* We only want to watch for WEXITED */ | |
f8f3f926 | 1610 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1611 | if (r < 0) |
f8f3f926 | 1612 | return r; |
f8f3f926 LP |
1613 | } else { |
1614 | /* We shall wait for some other event than WEXITED */ | |
f8f3f926 | 1615 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1616 | if (r < 0) |
f8f3f926 | 1617 | return r; |
a71fe8b8 | 1618 | |
f8f3f926 LP |
1619 | e->need_process_child = true; |
1620 | } | |
1621 | ||
b6d5481b | 1622 | e->n_online_child_sources++; |
ac9f2640 | 1623 | |
f8f3f926 LP |
1624 | if (ret) |
1625 | *ret = s; | |
f8f3f926 | 1626 | TAKE_PTR(s); |
fd38203a LP |
1627 | return 0; |
1628 | } | |
1629 | ||
b9350e70 LP |
1630 | static int generic_exit_callback(sd_event_source *s, void *userdata) { |
1631 | assert(s); | |
1632 | ||
1633 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1634 | } | |
1635 | ||
f7262a9f | 1636 | _public_ int sd_event_add_defer( |
305f78bf | 1637 | sd_event *e, |
151b9b96 | 1638 | sd_event_source **ret, |
718db961 | 1639 | sd_event_handler_t callback, |
151b9b96 | 1640 | void *userdata) { |
305f78bf | 1641 | |
ec766a51 | 1642 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1643 | int r; |
1644 | ||
305f78bf | 1645 | assert_return(e, -EINVAL); |
b937d761 | 1646 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1647 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1648 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1649 | |
b9350e70 LP |
1650 | if (!callback) |
1651 | callback = generic_exit_callback; | |
1652 | ||
a71fe8b8 | 1653 | s = source_new(e, !ret, SOURCE_DEFER); |
fd38203a LP |
1654 | if (!s) |
1655 | return -ENOMEM; | |
1656 | ||
1657 | s->defer.callback = callback; | |
1658 | s->userdata = userdata; | |
baf76283 | 1659 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a LP |
1660 | |
1661 | r = source_set_pending(s, true); | |
ec766a51 | 1662 | if (r < 0) |
fd38203a | 1663 | return r; |
fd38203a | 1664 | |
a71fe8b8 LP |
1665 | if (ret) |
1666 | *ret = s; | |
ec766a51 | 1667 | TAKE_PTR(s); |
a71fe8b8 | 1668 | |
fd38203a LP |
1669 | return 0; |
1670 | } | |
1671 | ||
6e9feda3 LP |
1672 | _public_ int sd_event_add_post( |
1673 | sd_event *e, | |
1674 | sd_event_source **ret, | |
1675 | sd_event_handler_t callback, | |
1676 | void *userdata) { | |
1677 | ||
ec766a51 | 1678 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6e9feda3 LP |
1679 | int r; |
1680 | ||
1681 | assert_return(e, -EINVAL); | |
b937d761 | 1682 | assert_return(e = event_resolve(e), -ENOPKG); |
6e9feda3 LP |
1683 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1684 | assert_return(!event_pid_changed(e), -ECHILD); | |
1685 | ||
b9350e70 LP |
1686 | if (!callback) |
1687 | callback = generic_exit_callback; | |
1688 | ||
a71fe8b8 | 1689 | s = source_new(e, !ret, SOURCE_POST); |
6e9feda3 LP |
1690 | if (!s) |
1691 | return -ENOMEM; | |
1692 | ||
1693 | s->post.callback = callback; | |
1694 | s->userdata = userdata; | |
1695 | s->enabled = SD_EVENT_ON; | |
1696 | ||
de7fef4b | 1697 | r = set_ensure_put(&e->post_sources, NULL, s); |
ec766a51 | 1698 | if (r < 0) |
6e9feda3 | 1699 | return r; |
de7fef4b | 1700 | assert(r > 0); |
6e9feda3 | 1701 | |
a71fe8b8 LP |
1702 | if (ret) |
1703 | *ret = s; | |
ec766a51 | 1704 | TAKE_PTR(s); |
a71fe8b8 | 1705 | |
6e9feda3 LP |
1706 | return 0; |
1707 | } | |
1708 | ||
6203e07a | 1709 | _public_ int sd_event_add_exit( |
305f78bf | 1710 | sd_event *e, |
151b9b96 | 1711 | sd_event_source **ret, |
718db961 | 1712 | sd_event_handler_t callback, |
151b9b96 | 1713 | void *userdata) { |
305f78bf | 1714 | |
ec766a51 | 1715 | _cleanup_(source_freep) sd_event_source *s = NULL; |
da7e457c LP |
1716 | int r; |
1717 | ||
1718 | assert_return(e, -EINVAL); | |
b937d761 | 1719 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
1720 | assert_return(callback, -EINVAL); |
1721 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
1722 | assert_return(!event_pid_changed(e), -ECHILD); | |
1723 | ||
c983e776 EV |
1724 | r = prioq_ensure_allocated(&e->exit, exit_prioq_compare); |
1725 | if (r < 0) | |
1726 | return r; | |
da7e457c | 1727 | |
a71fe8b8 | 1728 | s = source_new(e, !ret, SOURCE_EXIT); |
fd38203a | 1729 | if (!s) |
da7e457c | 1730 | return -ENOMEM; |
fd38203a | 1731 | |
6203e07a | 1732 | s->exit.callback = callback; |
da7e457c | 1733 | s->userdata = userdata; |
6203e07a | 1734 | s->exit.prioq_index = PRIOQ_IDX_NULL; |
baf76283 | 1735 | s->enabled = SD_EVENT_ONESHOT; |
da7e457c | 1736 | |
6203e07a | 1737 | r = prioq_put(s->event->exit, s, &s->exit.prioq_index); |
ec766a51 | 1738 | if (r < 0) |
da7e457c | 1739 | return r; |
da7e457c | 1740 | |
a71fe8b8 LP |
1741 | if (ret) |
1742 | *ret = s; | |
ec766a51 | 1743 | TAKE_PTR(s); |
a71fe8b8 | 1744 | |
da7e457c LP |
1745 | return 0; |
1746 | } | |
1747 | ||
97ef5391 LP |
1748 | static void event_free_inotify_data(sd_event *e, struct inotify_data *d) { |
1749 | assert(e); | |
1750 | ||
1751 | if (!d) | |
1752 | return; | |
1753 | ||
1754 | assert(hashmap_isempty(d->inodes)); | |
1755 | assert(hashmap_isempty(d->wd)); | |
1756 | ||
1757 | if (d->buffer_filled > 0) | |
0601b958 | 1758 | LIST_REMOVE(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
1759 | |
1760 | hashmap_free(d->inodes); | |
1761 | hashmap_free(d->wd); | |
1762 | ||
1763 | assert_se(hashmap_remove(e->inotify_data, &d->priority) == d); | |
1764 | ||
1765 | if (d->fd >= 0) { | |
fbae5090 YW |
1766 | if (!event_pid_changed(e) && |
1767 | epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, d->fd, NULL) < 0) | |
97ef5391 LP |
1768 | log_debug_errno(errno, "Failed to remove inotify fd from epoll, ignoring: %m"); |
1769 | ||
1770 | safe_close(d->fd); | |
1771 | } | |
1772 | free(d); | |
1773 | } | |
1774 | ||
1775 | static int event_make_inotify_data( | |
1776 | sd_event *e, | |
1777 | int64_t priority, | |
1778 | struct inotify_data **ret) { | |
1779 | ||
1780 | _cleanup_close_ int fd = -1; | |
1781 | struct inotify_data *d; | |
97ef5391 LP |
1782 | int r; |
1783 | ||
1784 | assert(e); | |
1785 | ||
1786 | d = hashmap_get(e->inotify_data, &priority); | |
1787 | if (d) { | |
1788 | if (ret) | |
1789 | *ret = d; | |
1790 | return 0; | |
1791 | } | |
1792 | ||
1793 | fd = inotify_init1(IN_NONBLOCK|O_CLOEXEC); | |
1794 | if (fd < 0) | |
1795 | return -errno; | |
1796 | ||
1797 | fd = fd_move_above_stdio(fd); | |
1798 | ||
97ef5391 LP |
1799 | d = new(struct inotify_data, 1); |
1800 | if (!d) | |
1801 | return -ENOMEM; | |
1802 | ||
1803 | *d = (struct inotify_data) { | |
1804 | .wakeup = WAKEUP_INOTIFY_DATA, | |
1805 | .fd = TAKE_FD(fd), | |
1806 | .priority = priority, | |
1807 | }; | |
1808 | ||
c2484a75 | 1809 | r = hashmap_ensure_put(&e->inotify_data, &uint64_hash_ops, &d->priority, d); |
97ef5391 LP |
1810 | if (r < 0) { |
1811 | d->fd = safe_close(d->fd); | |
1812 | free(d); | |
1813 | return r; | |
1814 | } | |
1815 | ||
1eac7948 | 1816 | struct epoll_event ev = { |
97ef5391 LP |
1817 | .events = EPOLLIN, |
1818 | .data.ptr = d, | |
1819 | }; | |
1820 | ||
1821 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { | |
1822 | r = -errno; | |
1823 | d->fd = safe_close(d->fd); /* let's close this ourselves, as event_free_inotify_data() would otherwise | |
1824 | * remove the fd from the epoll first, which we don't want as we couldn't | |
1825 | * add it in the first place. */ | |
1826 | event_free_inotify_data(e, d); | |
1827 | return r; | |
1828 | } | |
1829 | ||
1830 | if (ret) | |
1831 | *ret = d; | |
1832 | ||
1833 | return 1; | |
1834 | } | |
1835 | ||
7a08d314 | 1836 | static int inode_data_compare(const struct inode_data *x, const struct inode_data *y) { |
90c88092 | 1837 | int r; |
97ef5391 LP |
1838 | |
1839 | assert(x); | |
1840 | assert(y); | |
1841 | ||
90c88092 YW |
1842 | r = CMP(x->dev, y->dev); |
1843 | if (r != 0) | |
1844 | return r; | |
97ef5391 | 1845 | |
6dd91b36 | 1846 | return CMP(x->ino, y->ino); |
97ef5391 LP |
1847 | } |
1848 | ||
7a08d314 YW |
1849 | static void inode_data_hash_func(const struct inode_data *d, struct siphash *state) { |
1850 | assert(d); | |
97ef5391 LP |
1851 | |
1852 | siphash24_compress(&d->dev, sizeof(d->dev), state); | |
1853 | siphash24_compress(&d->ino, sizeof(d->ino), state); | |
1854 | } | |
1855 | ||
7a08d314 | 1856 | DEFINE_PRIVATE_HASH_OPS(inode_data_hash_ops, struct inode_data, inode_data_hash_func, inode_data_compare); |
97ef5391 LP |
1857 | |
1858 | static void event_free_inode_data( | |
1859 | sd_event *e, | |
1860 | struct inode_data *d) { | |
1861 | ||
1862 | assert(e); | |
1863 | ||
1864 | if (!d) | |
1865 | return; | |
1866 | ||
64903d18 | 1867 | assert(!d->event_sources); |
97ef5391 LP |
1868 | |
1869 | if (d->fd >= 0) { | |
ed828563 | 1870 | LIST_REMOVE(to_close, e->inode_data_to_close_list, d); |
97ef5391 LP |
1871 | safe_close(d->fd); |
1872 | } | |
1873 | ||
1874 | if (d->inotify_data) { | |
1875 | ||
1876 | if (d->wd >= 0) { | |
fbae5090 | 1877 | if (d->inotify_data->fd >= 0 && !event_pid_changed(e)) { |
97ef5391 LP |
1878 | /* So here's a problem. At the time this runs the watch descriptor might already be |
1879 | * invalidated, because an IN_IGNORED event might be queued right the moment we enter | |
1880 | * the syscall. Hence, whenever we get EINVAL, ignore it entirely, since it's a very | |
1881 | * likely case to happen. */ | |
1882 | ||
1883 | if (inotify_rm_watch(d->inotify_data->fd, d->wd) < 0 && errno != EINVAL) | |
1884 | log_debug_errno(errno, "Failed to remove watch descriptor %i from inotify, ignoring: %m", d->wd); | |
1885 | } | |
1886 | ||
1887 | assert_se(hashmap_remove(d->inotify_data->wd, INT_TO_PTR(d->wd)) == d); | |
1888 | } | |
1889 | ||
1890 | assert_se(hashmap_remove(d->inotify_data->inodes, d) == d); | |
1891 | } | |
1892 | ||
1893 | free(d); | |
1894 | } | |
1895 | ||
53baf2ef LP |
1896 | static void event_gc_inotify_data( |
1897 | sd_event *e, | |
1898 | struct inotify_data *d) { | |
1899 | ||
1900 | assert(e); | |
1901 | ||
1902 | /* GCs the inotify data object if we don't need it anymore. That's the case if we don't want to watch | |
1903 | * any inode with it anymore, which in turn happens if no event source of this priority is interested | |
1904 | * in any inode any longer. That said, we maintain an extra busy counter: if non-zero we'll delay GC | |
1905 | * (under the expectation that the GC is called again once the counter is decremented). */ | |
1906 | ||
1907 | if (!d) | |
1908 | return; | |
1909 | ||
1910 | if (!hashmap_isempty(d->inodes)) | |
1911 | return; | |
1912 | ||
1913 | if (d->n_busy > 0) | |
1914 | return; | |
1915 | ||
1916 | event_free_inotify_data(e, d); | |
1917 | } | |
1918 | ||
97ef5391 LP |
1919 | static void event_gc_inode_data( |
1920 | sd_event *e, | |
1921 | struct inode_data *d) { | |
1922 | ||
1923 | struct inotify_data *inotify_data; | |
1924 | ||
1925 | assert(e); | |
1926 | ||
1927 | if (!d) | |
1928 | return; | |
1929 | ||
64903d18 | 1930 | if (d->event_sources) |
97ef5391 LP |
1931 | return; |
1932 | ||
1933 | inotify_data = d->inotify_data; | |
1934 | event_free_inode_data(e, d); | |
1935 | ||
53baf2ef | 1936 | event_gc_inotify_data(e, inotify_data); |
97ef5391 LP |
1937 | } |
1938 | ||
1939 | static int event_make_inode_data( | |
1940 | sd_event *e, | |
1941 | struct inotify_data *inotify_data, | |
1942 | dev_t dev, | |
1943 | ino_t ino, | |
1944 | struct inode_data **ret) { | |
1945 | ||
1946 | struct inode_data *d, key; | |
1947 | int r; | |
1948 | ||
1949 | assert(e); | |
1950 | assert(inotify_data); | |
1951 | ||
1952 | key = (struct inode_data) { | |
1953 | .ino = ino, | |
1954 | .dev = dev, | |
1955 | }; | |
1956 | ||
1957 | d = hashmap_get(inotify_data->inodes, &key); | |
1958 | if (d) { | |
1959 | if (ret) | |
1960 | *ret = d; | |
1961 | ||
1962 | return 0; | |
1963 | } | |
1964 | ||
1965 | r = hashmap_ensure_allocated(&inotify_data->inodes, &inode_data_hash_ops); | |
1966 | if (r < 0) | |
1967 | return r; | |
1968 | ||
1969 | d = new(struct inode_data, 1); | |
1970 | if (!d) | |
1971 | return -ENOMEM; | |
1972 | ||
1973 | *d = (struct inode_data) { | |
1974 | .dev = dev, | |
1975 | .ino = ino, | |
1976 | .wd = -1, | |
1977 | .fd = -1, | |
1978 | .inotify_data = inotify_data, | |
1979 | }; | |
1980 | ||
1981 | r = hashmap_put(inotify_data->inodes, d, d); | |
1982 | if (r < 0) { | |
1983 | free(d); | |
1984 | return r; | |
1985 | } | |
1986 | ||
1987 | if (ret) | |
1988 | *ret = d; | |
1989 | ||
1990 | return 1; | |
1991 | } | |
1992 | ||
1993 | static uint32_t inode_data_determine_mask(struct inode_data *d) { | |
1994 | bool excl_unlink = true; | |
1995 | uint32_t combined = 0; | |
97ef5391 LP |
1996 | |
1997 | assert(d); | |
1998 | ||
1999 | /* Combines the watch masks of all event sources watching this inode. We generally just OR them together, but | |
2000 | * the IN_EXCL_UNLINK flag is ANDed instead. | |
2001 | * | |
2002 | * Note that we add all sources to the mask here, regardless whether enabled, disabled or oneshot. That's | |
2003 | * because we cannot change the mask anymore after the event source was created once, since the kernel has no | |
f21f31b2 | 2004 | * API for that. Hence we need to subscribe to the maximum mask we ever might be interested in, and suppress |
97ef5391 LP |
2005 | * events we don't care for client-side. */ |
2006 | ||
2007 | LIST_FOREACH(inotify.by_inode_data, s, d->event_sources) { | |
2008 | ||
2009 | if ((s->inotify.mask & IN_EXCL_UNLINK) == 0) | |
2010 | excl_unlink = false; | |
2011 | ||
2012 | combined |= s->inotify.mask; | |
2013 | } | |
2014 | ||
2015 | return (combined & ~(IN_ONESHOT|IN_DONT_FOLLOW|IN_ONLYDIR|IN_EXCL_UNLINK)) | (excl_unlink ? IN_EXCL_UNLINK : 0); | |
2016 | } | |
2017 | ||
2018 | static int inode_data_realize_watch(sd_event *e, struct inode_data *d) { | |
2019 | uint32_t combined_mask; | |
2020 | int wd, r; | |
2021 | ||
2022 | assert(d); | |
2023 | assert(d->fd >= 0); | |
2024 | ||
2025 | combined_mask = inode_data_determine_mask(d); | |
2026 | ||
2027 | if (d->wd >= 0 && combined_mask == d->combined_mask) | |
2028 | return 0; | |
2029 | ||
2030 | r = hashmap_ensure_allocated(&d->inotify_data->wd, NULL); | |
2031 | if (r < 0) | |
2032 | return r; | |
2033 | ||
2034 | wd = inotify_add_watch_fd(d->inotify_data->fd, d->fd, combined_mask); | |
2035 | if (wd < 0) | |
2036 | return -errno; | |
2037 | ||
2038 | if (d->wd < 0) { | |
2039 | r = hashmap_put(d->inotify_data->wd, INT_TO_PTR(wd), d); | |
2040 | if (r < 0) { | |
2041 | (void) inotify_rm_watch(d->inotify_data->fd, wd); | |
2042 | return r; | |
2043 | } | |
2044 | ||
2045 | d->wd = wd; | |
2046 | ||
2047 | } else if (d->wd != wd) { | |
2048 | ||
2049 | log_debug("Weird, the watch descriptor we already knew for this inode changed?"); | |
2050 | (void) inotify_rm_watch(d->fd, wd); | |
2051 | return -EINVAL; | |
2052 | } | |
2053 | ||
2054 | d->combined_mask = combined_mask; | |
2055 | return 1; | |
2056 | } | |
2057 | ||
b9350e70 LP |
2058 | static int inotify_exit_callback(sd_event_source *s, const struct inotify_event *event, void *userdata) { |
2059 | assert(s); | |
2060 | ||
2061 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
2062 | } | |
2063 | ||
e67d738a | 2064 | static int event_add_inotify_fd_internal( |
97ef5391 LP |
2065 | sd_event *e, |
2066 | sd_event_source **ret, | |
e67d738a LP |
2067 | int fd, |
2068 | bool donate, | |
97ef5391 LP |
2069 | uint32_t mask, |
2070 | sd_event_inotify_handler_t callback, | |
2071 | void *userdata) { | |
2072 | ||
e67d738a LP |
2073 | _cleanup_close_ int donated_fd = donate ? fd : -1; |
2074 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
97ef5391 LP |
2075 | struct inotify_data *inotify_data = NULL; |
2076 | struct inode_data *inode_data = NULL; | |
97ef5391 LP |
2077 | struct stat st; |
2078 | int r; | |
2079 | ||
2080 | assert_return(e, -EINVAL); | |
2081 | assert_return(e = event_resolve(e), -ENOPKG); | |
e67d738a | 2082 | assert_return(fd >= 0, -EBADF); |
97ef5391 LP |
2083 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2084 | assert_return(!event_pid_changed(e), -ECHILD); | |
2085 | ||
b9350e70 LP |
2086 | if (!callback) |
2087 | callback = inotify_exit_callback; | |
2088 | ||
97ef5391 LP |
2089 | /* Refuse IN_MASK_ADD since we coalesce watches on the same inode, and hence really don't want to merge |
2090 | * masks. Or in other words, this whole code exists only to manage IN_MASK_ADD type operations for you, hence | |
2091 | * the user can't use them for us. */ | |
2092 | if (mask & IN_MASK_ADD) | |
2093 | return -EINVAL; | |
2094 | ||
97ef5391 LP |
2095 | if (fstat(fd, &st) < 0) |
2096 | return -errno; | |
2097 | ||
2098 | s = source_new(e, !ret, SOURCE_INOTIFY); | |
2099 | if (!s) | |
2100 | return -ENOMEM; | |
2101 | ||
2102 | s->enabled = mask & IN_ONESHOT ? SD_EVENT_ONESHOT : SD_EVENT_ON; | |
2103 | s->inotify.mask = mask; | |
2104 | s->inotify.callback = callback; | |
2105 | s->userdata = userdata; | |
2106 | ||
2107 | /* Allocate an inotify object for this priority, and an inode object within it */ | |
2108 | r = event_make_inotify_data(e, SD_EVENT_PRIORITY_NORMAL, &inotify_data); | |
2109 | if (r < 0) | |
8c75fe17 | 2110 | return r; |
97ef5391 LP |
2111 | |
2112 | r = event_make_inode_data(e, inotify_data, st.st_dev, st.st_ino, &inode_data); | |
8c75fe17 | 2113 | if (r < 0) { |
e67d738a | 2114 | event_gc_inotify_data(e, inotify_data); |
8c75fe17 ZJS |
2115 | return r; |
2116 | } | |
97ef5391 LP |
2117 | |
2118 | /* Keep the O_PATH fd around until the first iteration of the loop, so that we can still change the priority of | |
2119 | * the event source, until then, for which we need the original inode. */ | |
2120 | if (inode_data->fd < 0) { | |
e67d738a LP |
2121 | if (donated_fd >= 0) |
2122 | inode_data->fd = TAKE_FD(donated_fd); | |
2123 | else { | |
2124 | inode_data->fd = fcntl(fd, F_DUPFD_CLOEXEC, 3); | |
2125 | if (inode_data->fd < 0) { | |
2126 | r = -errno; | |
2127 | event_gc_inode_data(e, inode_data); | |
2128 | return r; | |
2129 | } | |
2130 | } | |
2131 | ||
ed828563 | 2132 | LIST_PREPEND(to_close, e->inode_data_to_close_list, inode_data); |
97ef5391 LP |
2133 | } |
2134 | ||
2135 | /* Link our event source to the inode data object */ | |
2136 | LIST_PREPEND(inotify.by_inode_data, inode_data->event_sources, s); | |
2137 | s->inotify.inode_data = inode_data; | |
2138 | ||
97ef5391 LP |
2139 | /* Actually realize the watch now */ |
2140 | r = inode_data_realize_watch(e, inode_data); | |
2141 | if (r < 0) | |
8c75fe17 | 2142 | return r; |
97ef5391 | 2143 | |
97ef5391 LP |
2144 | if (ret) |
2145 | *ret = s; | |
8c75fe17 | 2146 | TAKE_PTR(s); |
97ef5391 LP |
2147 | |
2148 | return 0; | |
97ef5391 LP |
2149 | } |
2150 | ||
e67d738a LP |
2151 | _public_ int sd_event_add_inotify_fd( |
2152 | sd_event *e, | |
2153 | sd_event_source **ret, | |
2154 | int fd, | |
2155 | uint32_t mask, | |
2156 | sd_event_inotify_handler_t callback, | |
2157 | void *userdata) { | |
2158 | ||
2159 | return event_add_inotify_fd_internal(e, ret, fd, /* donate= */ false, mask, callback, userdata); | |
2160 | } | |
2161 | ||
2162 | _public_ int sd_event_add_inotify( | |
2163 | sd_event *e, | |
2164 | sd_event_source **ret, | |
2165 | const char *path, | |
2166 | uint32_t mask, | |
2167 | sd_event_inotify_handler_t callback, | |
2168 | void *userdata) { | |
2169 | ||
2091c779 | 2170 | sd_event_source *s = NULL; /* avoid false maybe-uninitialized warning */ |
e67d738a LP |
2171 | int fd, r; |
2172 | ||
2173 | assert_return(path, -EINVAL); | |
2174 | ||
586c8cee ZJS |
2175 | fd = open(path, O_PATH | O_CLOEXEC | |
2176 | (mask & IN_ONLYDIR ? O_DIRECTORY : 0) | | |
2177 | (mask & IN_DONT_FOLLOW ? O_NOFOLLOW : 0)); | |
e67d738a LP |
2178 | if (fd < 0) |
2179 | return -errno; | |
2180 | ||
2181 | r = event_add_inotify_fd_internal(e, &s, fd, /* donate= */ true, mask, callback, userdata); | |
2182 | if (r < 0) | |
2183 | return r; | |
2184 | ||
2185 | (void) sd_event_source_set_description(s, path); | |
2186 | ||
2187 | if (ret) | |
2188 | *ret = s; | |
2189 | ||
2190 | return r; | |
2191 | } | |
2192 | ||
8301aa0b | 2193 | static sd_event_source* event_source_free(sd_event_source *s) { |
6680dd6b LP |
2194 | if (!s) |
2195 | return NULL; | |
da7e457c | 2196 | |
8301aa0b YW |
2197 | /* Here's a special hack: when we are called from a |
2198 | * dispatch handler we won't free the event source | |
2199 | * immediately, but we will detach the fd from the | |
2200 | * epoll. This way it is safe for the caller to unref | |
2201 | * the event source and immediately close the fd, but | |
2202 | * we still retain a valid event source object after | |
2203 | * the callback. */ | |
fd38203a | 2204 | |
76d04c3a | 2205 | if (s->dispatching) |
8301aa0b | 2206 | source_disconnect(s); |
76d04c3a | 2207 | else |
8301aa0b | 2208 | source_free(s); |
fd38203a LP |
2209 | |
2210 | return NULL; | |
2211 | } | |
2212 | ||
8301aa0b YW |
2213 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event_source, sd_event_source, event_source_free); |
2214 | ||
356779df | 2215 | _public_ int sd_event_source_set_description(sd_event_source *s, const char *description) { |
f7f53e9e | 2216 | assert_return(s, -EINVAL); |
f4b2933e | 2217 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 2218 | |
356779df | 2219 | return free_and_strdup(&s->description, description); |
f7f53e9e TG |
2220 | } |
2221 | ||
356779df | 2222 | _public_ int sd_event_source_get_description(sd_event_source *s, const char **description) { |
f7f53e9e | 2223 | assert_return(s, -EINVAL); |
356779df | 2224 | assert_return(description, -EINVAL); |
f4b2933e | 2225 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 2226 | |
7d92a1a4 ZJS |
2227 | if (!s->description) |
2228 | return -ENXIO; | |
2229 | ||
356779df | 2230 | *description = s->description; |
f7f53e9e TG |
2231 | return 0; |
2232 | } | |
2233 | ||
adcc4ca3 | 2234 | _public_ sd_event *sd_event_source_get_event(sd_event_source *s) { |
305f78bf | 2235 | assert_return(s, NULL); |
eaa3cbef LP |
2236 | |
2237 | return s->event; | |
2238 | } | |
2239 | ||
f7262a9f | 2240 | _public_ int sd_event_source_get_pending(sd_event_source *s) { |
305f78bf | 2241 | assert_return(s, -EINVAL); |
6203e07a | 2242 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c | 2243 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2244 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2245 | |
2246 | return s->pending; | |
2247 | } | |
2248 | ||
f7262a9f | 2249 | _public_ int sd_event_source_get_io_fd(sd_event_source *s) { |
305f78bf LP |
2250 | assert_return(s, -EINVAL); |
2251 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2252 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2253 | |
2254 | return s->io.fd; | |
2255 | } | |
2256 | ||
30caf8f3 LP |
2257 | _public_ int sd_event_source_set_io_fd(sd_event_source *s, int fd) { |
2258 | int r; | |
2259 | ||
2260 | assert_return(s, -EINVAL); | |
8ac43fee | 2261 | assert_return(fd >= 0, -EBADF); |
30caf8f3 LP |
2262 | assert_return(s->type == SOURCE_IO, -EDOM); |
2263 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2264 | ||
2265 | if (s->io.fd == fd) | |
2266 | return 0; | |
2267 | ||
b6d5481b | 2268 | if (event_source_is_offline(s)) { |
30caf8f3 LP |
2269 | s->io.fd = fd; |
2270 | s->io.registered = false; | |
2271 | } else { | |
2272 | int saved_fd; | |
2273 | ||
2274 | saved_fd = s->io.fd; | |
2275 | assert(s->io.registered); | |
2276 | ||
2277 | s->io.fd = fd; | |
2278 | s->io.registered = false; | |
2279 | ||
2280 | r = source_io_register(s, s->enabled, s->io.events); | |
2281 | if (r < 0) { | |
2282 | s->io.fd = saved_fd; | |
2283 | s->io.registered = true; | |
2284 | return r; | |
2285 | } | |
2286 | ||
5a795bff | 2287 | (void) epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, saved_fd, NULL); |
30caf8f3 LP |
2288 | } |
2289 | ||
2290 | return 0; | |
2291 | } | |
2292 | ||
ab93297c NM |
2293 | _public_ int sd_event_source_get_io_fd_own(sd_event_source *s) { |
2294 | assert_return(s, -EINVAL); | |
2295 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2296 | ||
2297 | return s->io.owned; | |
2298 | } | |
2299 | ||
2300 | _public_ int sd_event_source_set_io_fd_own(sd_event_source *s, int own) { | |
2301 | assert_return(s, -EINVAL); | |
2302 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2303 | ||
2304 | s->io.owned = own; | |
2305 | return 0; | |
2306 | } | |
2307 | ||
f7262a9f | 2308 | _public_ int sd_event_source_get_io_events(sd_event_source *s, uint32_t* events) { |
305f78bf LP |
2309 | assert_return(s, -EINVAL); |
2310 | assert_return(events, -EINVAL); | |
2311 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2312 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2313 | |
2314 | *events = s->io.events; | |
2315 | return 0; | |
2316 | } | |
2317 | ||
f7262a9f | 2318 | _public_ int sd_event_source_set_io_events(sd_event_source *s, uint32_t events) { |
fd38203a LP |
2319 | int r; |
2320 | ||
305f78bf LP |
2321 | assert_return(s, -EINVAL); |
2322 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2a16a986 | 2323 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 2324 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2325 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2326 | |
b63c8d4f DH |
2327 | /* edge-triggered updates are never skipped, so we can reset edges */ |
2328 | if (s->io.events == events && !(events & EPOLLET)) | |
fd38203a LP |
2329 | return 0; |
2330 | ||
2a0dc6cd LP |
2331 | r = source_set_pending(s, false); |
2332 | if (r < 0) | |
2333 | return r; | |
2334 | ||
b6d5481b | 2335 | if (event_source_is_online(s)) { |
e4715127 | 2336 | r = source_io_register(s, s->enabled, events); |
fd38203a LP |
2337 | if (r < 0) |
2338 | return r; | |
2339 | } | |
2340 | ||
2341 | s->io.events = events; | |
2342 | ||
2343 | return 0; | |
2344 | } | |
2345 | ||
f7262a9f | 2346 | _public_ int sd_event_source_get_io_revents(sd_event_source *s, uint32_t* revents) { |
305f78bf LP |
2347 | assert_return(s, -EINVAL); |
2348 | assert_return(revents, -EINVAL); | |
2349 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2350 | assert_return(s->pending, -ENODATA); | |
2351 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2352 | |
2353 | *revents = s->io.revents; | |
2354 | return 0; | |
2355 | } | |
2356 | ||
f7262a9f | 2357 | _public_ int sd_event_source_get_signal(sd_event_source *s) { |
305f78bf LP |
2358 | assert_return(s, -EINVAL); |
2359 | assert_return(s->type == SOURCE_SIGNAL, -EDOM); | |
2360 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2361 | |
2362 | return s->signal.sig; | |
2363 | } | |
2364 | ||
31927c16 | 2365 | _public_ int sd_event_source_get_priority(sd_event_source *s, int64_t *priority) { |
305f78bf LP |
2366 | assert_return(s, -EINVAL); |
2367 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a | 2368 | |
6680b8d1 ME |
2369 | *priority = s->priority; |
2370 | return 0; | |
fd38203a LP |
2371 | } |
2372 | ||
31927c16 | 2373 | _public_ int sd_event_source_set_priority(sd_event_source *s, int64_t priority) { |
97ef5391 LP |
2374 | bool rm_inotify = false, rm_inode = false; |
2375 | struct inotify_data *new_inotify_data = NULL; | |
2376 | struct inode_data *new_inode_data = NULL; | |
9da4cb2b LP |
2377 | int r; |
2378 | ||
305f78bf | 2379 | assert_return(s, -EINVAL); |
da7e457c | 2380 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2381 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2382 | |
2383 | if (s->priority == priority) | |
2384 | return 0; | |
2385 | ||
97ef5391 LP |
2386 | if (s->type == SOURCE_INOTIFY) { |
2387 | struct inode_data *old_inode_data; | |
2388 | ||
2389 | assert(s->inotify.inode_data); | |
2390 | old_inode_data = s->inotify.inode_data; | |
2391 | ||
2392 | /* We need the original fd to change the priority. If we don't have it we can't change the priority, | |
2393 | * anymore. Note that we close any fds when entering the next event loop iteration, i.e. for inotify | |
2394 | * events we allow priority changes only until the first following iteration. */ | |
2395 | if (old_inode_data->fd < 0) | |
2396 | return -EOPNOTSUPP; | |
2397 | ||
2398 | r = event_make_inotify_data(s->event, priority, &new_inotify_data); | |
2399 | if (r < 0) | |
2400 | return r; | |
2401 | rm_inotify = r > 0; | |
2402 | ||
2403 | r = event_make_inode_data(s->event, new_inotify_data, old_inode_data->dev, old_inode_data->ino, &new_inode_data); | |
2404 | if (r < 0) | |
2405 | goto fail; | |
2406 | rm_inode = r > 0; | |
2407 | ||
2408 | if (new_inode_data->fd < 0) { | |
2409 | /* Duplicate the fd for the new inode object if we don't have any yet */ | |
2410 | new_inode_data->fd = fcntl(old_inode_data->fd, F_DUPFD_CLOEXEC, 3); | |
2411 | if (new_inode_data->fd < 0) { | |
2412 | r = -errno; | |
2413 | goto fail; | |
2414 | } | |
2415 | ||
ed828563 | 2416 | LIST_PREPEND(to_close, s->event->inode_data_to_close_list, new_inode_data); |
97ef5391 LP |
2417 | } |
2418 | ||
2419 | /* Move the event source to the new inode data structure */ | |
2420 | LIST_REMOVE(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2421 | LIST_PREPEND(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2422 | s->inotify.inode_data = new_inode_data; | |
2423 | ||
2424 | /* Now create the new watch */ | |
2425 | r = inode_data_realize_watch(s->event, new_inode_data); | |
2426 | if (r < 0) { | |
2427 | /* Move it back */ | |
2428 | LIST_REMOVE(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2429 | LIST_PREPEND(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2430 | s->inotify.inode_data = old_inode_data; | |
2431 | goto fail; | |
2432 | } | |
2433 | ||
2434 | s->priority = priority; | |
2435 | ||
2436 | event_gc_inode_data(s->event, old_inode_data); | |
2437 | ||
b6d5481b | 2438 | } else if (s->type == SOURCE_SIGNAL && event_source_is_online(s)) { |
9da4cb2b LP |
2439 | struct signal_data *old, *d; |
2440 | ||
2441 | /* Move us from the signalfd belonging to the old | |
2442 | * priority to the signalfd of the new priority */ | |
2443 | ||
2444 | assert_se(old = hashmap_get(s->event->signal_data, &s->priority)); | |
2445 | ||
2446 | s->priority = priority; | |
2447 | ||
2448 | r = event_make_signal_data(s->event, s->signal.sig, &d); | |
2449 | if (r < 0) { | |
2450 | s->priority = old->priority; | |
2451 | return r; | |
2452 | } | |
2453 | ||
2454 | event_unmask_signal_data(s->event, old, s->signal.sig); | |
2455 | } else | |
2456 | s->priority = priority; | |
fd38203a | 2457 | |
e1951c16 | 2458 | event_source_pp_prioq_reshuffle(s); |
fd38203a | 2459 | |
6203e07a LP |
2460 | if (s->type == SOURCE_EXIT) |
2461 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
305f78bf | 2462 | |
fd38203a | 2463 | return 0; |
97ef5391 LP |
2464 | |
2465 | fail: | |
2466 | if (rm_inode) | |
2467 | event_free_inode_data(s->event, new_inode_data); | |
2468 | ||
2469 | if (rm_inotify) | |
2470 | event_free_inotify_data(s->event, new_inotify_data); | |
2471 | ||
2472 | return r; | |
fd38203a LP |
2473 | } |
2474 | ||
cad143a8 | 2475 | _public_ int sd_event_source_get_enabled(sd_event_source *s, int *ret) { |
71193c0b ZJS |
2476 | /* Quick mode: the event source doesn't exist and we only want to query boolean enablement state. */ |
2477 | if (!s && !ret) | |
2478 | return false; | |
2479 | ||
305f78bf | 2480 | assert_return(s, -EINVAL); |
305f78bf | 2481 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2482 | |
cad143a8 LP |
2483 | if (ret) |
2484 | *ret = s->enabled; | |
2485 | ||
08c1eb0e | 2486 | return s->enabled != SD_EVENT_OFF; |
fd38203a LP |
2487 | } |
2488 | ||
b6d5481b LP |
2489 | static int event_source_offline( |
2490 | sd_event_source *s, | |
2491 | int enabled, | |
2492 | bool ratelimited) { | |
2493 | ||
2494 | bool was_offline; | |
fd38203a LP |
2495 | int r; |
2496 | ||
ddfde737 | 2497 | assert(s); |
b6d5481b | 2498 | assert(enabled == SD_EVENT_OFF || ratelimited); |
fd38203a | 2499 | |
ddfde737 | 2500 | /* Unset the pending flag when this event source is disabled */ |
b6d5481b LP |
2501 | if (s->enabled != SD_EVENT_OFF && |
2502 | enabled == SD_EVENT_OFF && | |
2503 | !IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
ddfde737 LP |
2504 | r = source_set_pending(s, false); |
2505 | if (r < 0) | |
2506 | return r; | |
2507 | } | |
cc567911 | 2508 | |
b6d5481b LP |
2509 | was_offline = event_source_is_offline(s); |
2510 | s->enabled = enabled; | |
2511 | s->ratelimited = ratelimited; | |
fd38203a | 2512 | |
ddfde737 | 2513 | switch (s->type) { |
fd38203a | 2514 | |
ddfde737 LP |
2515 | case SOURCE_IO: |
2516 | source_io_unregister(s); | |
2517 | break; | |
ac989a78 | 2518 | |
ddfde737 LP |
2519 | case SOURCE_SIGNAL: |
2520 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); | |
2521 | break; | |
fd38203a | 2522 | |
ddfde737 | 2523 | case SOURCE_CHILD: |
b6d5481b LP |
2524 | if (!was_offline) { |
2525 | assert(s->event->n_online_child_sources > 0); | |
2526 | s->event->n_online_child_sources--; | |
2527 | } | |
fd38203a | 2528 | |
ddfde737 LP |
2529 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
2530 | source_child_pidfd_unregister(s); | |
2531 | else | |
2532 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
2533 | break; | |
4807d2d0 | 2534 | |
ddfde737 LP |
2535 | case SOURCE_EXIT: |
2536 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
2537 | break; | |
fd38203a | 2538 | |
2115b9b6 YW |
2539 | case SOURCE_TIME_REALTIME: |
2540 | case SOURCE_TIME_BOOTTIME: | |
2541 | case SOURCE_TIME_MONOTONIC: | |
2542 | case SOURCE_TIME_REALTIME_ALARM: | |
2543 | case SOURCE_TIME_BOOTTIME_ALARM: | |
ddfde737 LP |
2544 | case SOURCE_DEFER: |
2545 | case SOURCE_POST: | |
2546 | case SOURCE_INOTIFY: | |
2547 | break; | |
fd38203a | 2548 | |
ddfde737 | 2549 | default: |
04499a70 | 2550 | assert_not_reached(); |
ddfde737 | 2551 | } |
fd38203a | 2552 | |
2115b9b6 YW |
2553 | /* Always reshuffle time prioq, as the ratelimited flag may be changed. */ |
2554 | event_source_time_prioq_reshuffle(s); | |
2555 | ||
b6d5481b | 2556 | return 1; |
ddfde737 | 2557 | } |
f8f3f926 | 2558 | |
b6d5481b LP |
2559 | static int event_source_online( |
2560 | sd_event_source *s, | |
2561 | int enabled, | |
2562 | bool ratelimited) { | |
2563 | ||
2564 | bool was_online; | |
ddfde737 | 2565 | int r; |
fd38203a | 2566 | |
ddfde737 | 2567 | assert(s); |
b6d5481b | 2568 | assert(enabled != SD_EVENT_OFF || !ratelimited); |
305f78bf | 2569 | |
ddfde737 | 2570 | /* Unset the pending flag when this event source is enabled */ |
b6d5481b LP |
2571 | if (s->enabled == SD_EVENT_OFF && |
2572 | enabled != SD_EVENT_OFF && | |
2573 | !IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
ddfde737 LP |
2574 | r = source_set_pending(s, false); |
2575 | if (r < 0) | |
2576 | return r; | |
2577 | } | |
9d3e3aa5 | 2578 | |
b6d5481b LP |
2579 | /* Are we really ready for onlining? */ |
2580 | if (enabled == SD_EVENT_OFF || ratelimited) { | |
2581 | /* Nope, we are not ready for onlining, then just update the precise state and exit */ | |
2582 | s->enabled = enabled; | |
2583 | s->ratelimited = ratelimited; | |
2584 | return 0; | |
2585 | } | |
2586 | ||
2587 | was_online = event_source_is_online(s); | |
2588 | ||
ddfde737 | 2589 | switch (s->type) { |
ddfde737 | 2590 | case SOURCE_IO: |
b6d5481b | 2591 | r = source_io_register(s, enabled, s->io.events); |
d2eafe61 | 2592 | if (r < 0) |
ddfde737 | 2593 | return r; |
ddfde737 | 2594 | break; |
fd38203a | 2595 | |
ddfde737 LP |
2596 | case SOURCE_SIGNAL: |
2597 | r = event_make_signal_data(s->event, s->signal.sig, NULL); | |
2598 | if (r < 0) { | |
ddfde737 LP |
2599 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
2600 | return r; | |
2601 | } | |
fd38203a | 2602 | |
ddfde737 | 2603 | break; |
fd38203a | 2604 | |
ddfde737 | 2605 | case SOURCE_CHILD: |
ddfde737 LP |
2606 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
2607 | /* yes, we have pidfd */ | |
9da4cb2b | 2608 | |
b6d5481b | 2609 | r = source_child_pidfd_register(s, enabled); |
ac9f2640 | 2610 | if (r < 0) |
9da4cb2b | 2611 | return r; |
ddfde737 LP |
2612 | } else { |
2613 | /* no pidfd, or something other to watch for than WEXITED */ | |
9da4cb2b | 2614 | |
ddfde737 LP |
2615 | r = event_make_signal_data(s->event, SIGCHLD, NULL); |
2616 | if (r < 0) { | |
ddfde737 LP |
2617 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); |
2618 | return r; | |
2619 | } | |
2620 | } | |
fd38203a | 2621 | |
b6d5481b LP |
2622 | if (!was_online) |
2623 | s->event->n_online_child_sources++; | |
ddfde737 | 2624 | break; |
4807d2d0 | 2625 | |
d2eafe61 ZJS |
2626 | case SOURCE_TIME_REALTIME: |
2627 | case SOURCE_TIME_BOOTTIME: | |
2628 | case SOURCE_TIME_MONOTONIC: | |
2629 | case SOURCE_TIME_REALTIME_ALARM: | |
2630 | case SOURCE_TIME_BOOTTIME_ALARM: | |
ddfde737 | 2631 | case SOURCE_EXIT: |
ddfde737 LP |
2632 | case SOURCE_DEFER: |
2633 | case SOURCE_POST: | |
2634 | case SOURCE_INOTIFY: | |
2635 | break; | |
9da4cb2b | 2636 | |
ddfde737 | 2637 | default: |
04499a70 | 2638 | assert_not_reached(); |
ddfde737 | 2639 | } |
f8f3f926 | 2640 | |
b6d5481b LP |
2641 | s->enabled = enabled; |
2642 | s->ratelimited = ratelimited; | |
d2eafe61 ZJS |
2643 | |
2644 | /* Non-failing operations below */ | |
2115b9b6 | 2645 | if (s->type == SOURCE_EXIT) |
d2eafe61 | 2646 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); |
d2eafe61 | 2647 | |
2115b9b6 YW |
2648 | /* Always reshuffle time prioq, as the ratelimited flag may be changed. */ |
2649 | event_source_time_prioq_reshuffle(s); | |
d2eafe61 | 2650 | |
b6d5481b | 2651 | return 1; |
ddfde737 LP |
2652 | } |
2653 | ||
2654 | _public_ int sd_event_source_set_enabled(sd_event_source *s, int m) { | |
2655 | int r; | |
9da4cb2b | 2656 | |
ddfde737 | 2657 | assert_return(IN_SET(m, SD_EVENT_OFF, SD_EVENT_ON, SD_EVENT_ONESHOT), -EINVAL); |
7e922b05 ZJS |
2658 | |
2659 | /* Quick mode: if the source doesn't exist, SD_EVENT_OFF is a noop. */ | |
2660 | if (m == SD_EVENT_OFF && !s) | |
2661 | return 0; | |
2662 | ||
2663 | assert_return(s, -EINVAL); | |
ddfde737 | 2664 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2665 | |
ddfde737 LP |
2666 | /* If we are dead anyway, we are fine with turning off sources, but everything else needs to fail. */ |
2667 | if (s->event->state == SD_EVENT_FINISHED) | |
2668 | return m == SD_EVENT_OFF ? 0 : -ESTALE; | |
305f78bf | 2669 | |
ddfde737 LP |
2670 | if (s->enabled == m) /* No change? */ |
2671 | return 0; | |
9d3e3aa5 | 2672 | |
ddfde737 | 2673 | if (m == SD_EVENT_OFF) |
b6d5481b | 2674 | r = event_source_offline(s, m, s->ratelimited); |
ddfde737 LP |
2675 | else { |
2676 | if (s->enabled != SD_EVENT_OFF) { | |
2677 | /* Switching from "on" to "oneshot" or back? If that's the case, we can take a shortcut, the | |
2678 | * event source is already enabled after all. */ | |
2679 | s->enabled = m; | |
2680 | return 0; | |
fd38203a | 2681 | } |
ddfde737 | 2682 | |
b6d5481b | 2683 | r = event_source_online(s, m, s->ratelimited); |
fd38203a | 2684 | } |
ddfde737 LP |
2685 | if (r < 0) |
2686 | return r; | |
fd38203a | 2687 | |
e1951c16 | 2688 | event_source_pp_prioq_reshuffle(s); |
fd38203a LP |
2689 | return 0; |
2690 | } | |
2691 | ||
f7262a9f | 2692 | _public_ int sd_event_source_get_time(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
2693 | assert_return(s, -EINVAL); |
2694 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 2695 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf | 2696 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2697 | |
2698 | *usec = s->time.next; | |
2699 | return 0; | |
2700 | } | |
2701 | ||
f7262a9f | 2702 | _public_ int sd_event_source_set_time(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 2703 | int r; |
6a0f1f6d | 2704 | |
305f78bf | 2705 | assert_return(s, -EINVAL); |
6a0f1f6d | 2706 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 2707 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2708 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2709 | |
2a0dc6cd LP |
2710 | r = source_set_pending(s, false); |
2711 | if (r < 0) | |
2712 | return r; | |
2576a19e | 2713 | |
2a0dc6cd | 2714 | s->time.next = usec; |
fd38203a | 2715 | |
e1951c16 | 2716 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
2717 | return 0; |
2718 | } | |
2719 | ||
d6a83dc4 LP |
2720 | _public_ int sd_event_source_set_time_relative(sd_event_source *s, uint64_t usec) { |
2721 | usec_t t; | |
2722 | int r; | |
2723 | ||
2724 | assert_return(s, -EINVAL); | |
2725 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2726 | ||
2727 | r = sd_event_now(s->event, event_source_type_to_clock(s->type), &t); | |
2728 | if (r < 0) | |
2729 | return r; | |
2730 | ||
496db330 YW |
2731 | usec = usec_add(t, usec); |
2732 | if (usec == USEC_INFINITY) | |
d6a83dc4 LP |
2733 | return -EOVERFLOW; |
2734 | ||
496db330 | 2735 | return sd_event_source_set_time(s, usec); |
d6a83dc4 LP |
2736 | } |
2737 | ||
f7262a9f | 2738 | _public_ int sd_event_source_get_time_accuracy(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
2739 | assert_return(s, -EINVAL); |
2740 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 2741 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf LP |
2742 | assert_return(!event_pid_changed(s->event), -ECHILD); |
2743 | ||
2744 | *usec = s->time.accuracy; | |
2745 | return 0; | |
2746 | } | |
2747 | ||
f7262a9f | 2748 | _public_ int sd_event_source_set_time_accuracy(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 2749 | int r; |
6a0f1f6d | 2750 | |
305f78bf | 2751 | assert_return(s, -EINVAL); |
f5fbe71d | 2752 | assert_return(usec != UINT64_MAX, -EINVAL); |
6a0f1f6d | 2753 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 2754 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2755 | assert_return(!event_pid_changed(s->event), -ECHILD); |
eaa3cbef | 2756 | |
2a0dc6cd LP |
2757 | r = source_set_pending(s, false); |
2758 | if (r < 0) | |
2759 | return r; | |
2760 | ||
eaa3cbef LP |
2761 | if (usec == 0) |
2762 | usec = DEFAULT_ACCURACY_USEC; | |
2763 | ||
eaa3cbef LP |
2764 | s->time.accuracy = usec; |
2765 | ||
e1951c16 | 2766 | event_source_time_prioq_reshuffle(s); |
6a0f1f6d LP |
2767 | return 0; |
2768 | } | |
2769 | ||
2770 | _public_ int sd_event_source_get_time_clock(sd_event_source *s, clockid_t *clock) { | |
2771 | assert_return(s, -EINVAL); | |
2772 | assert_return(clock, -EINVAL); | |
2773 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2774 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
eaa3cbef | 2775 | |
6a0f1f6d | 2776 | *clock = event_source_type_to_clock(s->type); |
eaa3cbef LP |
2777 | return 0; |
2778 | } | |
2779 | ||
f7262a9f | 2780 | _public_ int sd_event_source_get_child_pid(sd_event_source *s, pid_t *pid) { |
4bee8012 LP |
2781 | assert_return(s, -EINVAL); |
2782 | assert_return(pid, -EINVAL); | |
2783 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2784 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2785 | ||
2786 | *pid = s->child.pid; | |
2787 | return 0; | |
2788 | } | |
2789 | ||
f8f3f926 LP |
2790 | _public_ int sd_event_source_get_child_pidfd(sd_event_source *s) { |
2791 | assert_return(s, -EINVAL); | |
2792 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2793 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2794 | ||
2795 | if (s->child.pidfd < 0) | |
2796 | return -EOPNOTSUPP; | |
2797 | ||
2798 | return s->child.pidfd; | |
2799 | } | |
2800 | ||
2801 | _public_ int sd_event_source_send_child_signal(sd_event_source *s, int sig, const siginfo_t *si, unsigned flags) { | |
2802 | assert_return(s, -EINVAL); | |
2803 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2804 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2805 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
2806 | ||
2807 | /* If we already have seen indication the process exited refuse sending a signal early. This way we | |
2808 | * can be sure we don't accidentally kill the wrong process on PID reuse when pidfds are not | |
2809 | * available. */ | |
2810 | if (s->child.exited) | |
2811 | return -ESRCH; | |
2812 | ||
2813 | if (s->child.pidfd >= 0) { | |
2814 | siginfo_t copy; | |
2815 | ||
2816 | /* pidfd_send_signal() changes the siginfo_t argument. This is weird, let's hence copy the | |
2817 | * structure here */ | |
2818 | if (si) | |
2819 | copy = *si; | |
2820 | ||
2821 | if (pidfd_send_signal(s->child.pidfd, sig, si ? © : NULL, 0) < 0) { | |
2822 | /* Let's propagate the error only if the system call is not implemented or prohibited */ | |
2823 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
2824 | return -errno; | |
2825 | } else | |
2826 | return 0; | |
2827 | } | |
2828 | ||
2829 | /* Flags are only supported for pidfd_send_signal(), not for rt_sigqueueinfo(), hence let's refuse | |
2830 | * this here. */ | |
2831 | if (flags != 0) | |
2832 | return -EOPNOTSUPP; | |
2833 | ||
2834 | if (si) { | |
2835 | /* We use rt_sigqueueinfo() only if siginfo_t is specified. */ | |
2836 | siginfo_t copy = *si; | |
2837 | ||
2838 | if (rt_sigqueueinfo(s->child.pid, sig, ©) < 0) | |
2839 | return -errno; | |
2840 | } else if (kill(s->child.pid, sig) < 0) | |
2841 | return -errno; | |
2842 | ||
2843 | return 0; | |
2844 | } | |
2845 | ||
2846 | _public_ int sd_event_source_get_child_pidfd_own(sd_event_source *s) { | |
2847 | assert_return(s, -EINVAL); | |
2848 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2849 | ||
2850 | if (s->child.pidfd < 0) | |
2851 | return -EOPNOTSUPP; | |
2852 | ||
2853 | return s->child.pidfd_owned; | |
2854 | } | |
2855 | ||
2856 | _public_ int sd_event_source_set_child_pidfd_own(sd_event_source *s, int own) { | |
2857 | assert_return(s, -EINVAL); | |
2858 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2859 | ||
2860 | if (s->child.pidfd < 0) | |
2861 | return -EOPNOTSUPP; | |
2862 | ||
2863 | s->child.pidfd_owned = own; | |
2864 | return 0; | |
2865 | } | |
2866 | ||
2867 | _public_ int sd_event_source_get_child_process_own(sd_event_source *s) { | |
2868 | assert_return(s, -EINVAL); | |
2869 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2870 | ||
2871 | return s->child.process_owned; | |
2872 | } | |
2873 | ||
2874 | _public_ int sd_event_source_set_child_process_own(sd_event_source *s, int own) { | |
2875 | assert_return(s, -EINVAL); | |
2876 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2877 | ||
2878 | s->child.process_owned = own; | |
2879 | return 0; | |
2880 | } | |
2881 | ||
97ef5391 LP |
2882 | _public_ int sd_event_source_get_inotify_mask(sd_event_source *s, uint32_t *mask) { |
2883 | assert_return(s, -EINVAL); | |
2884 | assert_return(mask, -EINVAL); | |
2885 | assert_return(s->type == SOURCE_INOTIFY, -EDOM); | |
2886 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2887 | ||
2888 | *mask = s->inotify.mask; | |
2889 | return 0; | |
2890 | } | |
2891 | ||
718db961 | 2892 | _public_ int sd_event_source_set_prepare(sd_event_source *s, sd_event_handler_t callback) { |
fd38203a LP |
2893 | int r; |
2894 | ||
da7e457c | 2895 | assert_return(s, -EINVAL); |
6203e07a | 2896 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c LP |
2897 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2898 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2899 | |
2900 | if (s->prepare == callback) | |
2901 | return 0; | |
2902 | ||
2903 | if (callback && s->prepare) { | |
2904 | s->prepare = callback; | |
2905 | return 0; | |
2906 | } | |
2907 | ||
2908 | r = prioq_ensure_allocated(&s->event->prepare, prepare_prioq_compare); | |
2909 | if (r < 0) | |
2910 | return r; | |
2911 | ||
2912 | s->prepare = callback; | |
2913 | ||
2914 | if (callback) { | |
2915 | r = prioq_put(s->event->prepare, s, &s->prepare_index); | |
2916 | if (r < 0) | |
2917 | return r; | |
2918 | } else | |
2919 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
2920 | ||
2921 | return 0; | |
2922 | } | |
2923 | ||
f7262a9f | 2924 | _public_ void* sd_event_source_get_userdata(sd_event_source *s) { |
da7e457c | 2925 | assert_return(s, NULL); |
fd38203a LP |
2926 | |
2927 | return s->userdata; | |
2928 | } | |
2929 | ||
8f726607 LP |
2930 | _public_ void *sd_event_source_set_userdata(sd_event_source *s, void *userdata) { |
2931 | void *ret; | |
2932 | ||
2933 | assert_return(s, NULL); | |
2934 | ||
2935 | ret = s->userdata; | |
2936 | s->userdata = userdata; | |
2937 | ||
2938 | return ret; | |
2939 | } | |
2940 | ||
b6d5481b LP |
2941 | static int event_source_enter_ratelimited(sd_event_source *s) { |
2942 | int r; | |
2943 | ||
2944 | assert(s); | |
2945 | ||
2946 | /* When an event source becomes ratelimited, we place it in the CLOCK_MONOTONIC priority queue, with | |
2947 | * the end of the rate limit time window, much as if it was a timer event source. */ | |
2948 | ||
2949 | if (s->ratelimited) | |
2950 | return 0; /* Already ratelimited, this is a NOP hence */ | |
2951 | ||
2952 | /* Make sure we can install a CLOCK_MONOTONIC event further down. */ | |
2953 | r = setup_clock_data(s->event, &s->event->monotonic, CLOCK_MONOTONIC); | |
2954 | if (r < 0) | |
2955 | return r; | |
2956 | ||
2957 | /* Timer event sources are already using the earliest/latest queues for the timer scheduling. Let's | |
2958 | * first remove them from the prioq appropriate for their own clock, so that we can use the prioq | |
2959 | * fields of the event source then for adding it to the CLOCK_MONOTONIC prioq instead. */ | |
2960 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
2961 | event_source_time_prioq_remove(s, event_get_clock_data(s->event, s->type)); | |
2962 | ||
2963 | /* Now, let's add the event source to the monotonic clock instead */ | |
2964 | r = event_source_time_prioq_put(s, &s->event->monotonic); | |
2965 | if (r < 0) | |
2966 | goto fail; | |
2967 | ||
2968 | /* And let's take the event source officially offline */ | |
2969 | r = event_source_offline(s, s->enabled, /* ratelimited= */ true); | |
2970 | if (r < 0) { | |
2971 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
2972 | goto fail; | |
2973 | } | |
2974 | ||
2975 | event_source_pp_prioq_reshuffle(s); | |
2976 | ||
2977 | log_debug("Event source %p (%s) entered rate limit state.", s, strna(s->description)); | |
2978 | return 0; | |
2979 | ||
2980 | fail: | |
2981 | /* Reinstall time event sources in the priority queue as before. This shouldn't fail, since the queue | |
2982 | * space for it should already be allocated. */ | |
2983 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
2984 | assert_se(event_source_time_prioq_put(s, event_get_clock_data(s->event, s->type)) >= 0); | |
2985 | ||
2986 | return r; | |
2987 | } | |
2988 | ||
fd69f224 | 2989 | static int event_source_leave_ratelimit(sd_event_source *s, bool run_callback) { |
b6d5481b LP |
2990 | int r; |
2991 | ||
2992 | assert(s); | |
2993 | ||
2994 | if (!s->ratelimited) | |
2995 | return 0; | |
2996 | ||
2997 | /* Let's take the event source out of the monotonic prioq first. */ | |
2998 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
2999 | ||
3000 | /* Let's then add the event source to its native clock prioq again — if this is a timer event source */ | |
3001 | if (EVENT_SOURCE_IS_TIME(s->type)) { | |
3002 | r = event_source_time_prioq_put(s, event_get_clock_data(s->event, s->type)); | |
3003 | if (r < 0) | |
3004 | goto fail; | |
3005 | } | |
3006 | ||
3007 | /* Let's try to take it online again. */ | |
3008 | r = event_source_online(s, s->enabled, /* ratelimited= */ false); | |
3009 | if (r < 0) { | |
3010 | /* Do something roughly sensible when this failed: undo the two prioq ops above */ | |
3011 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
3012 | event_source_time_prioq_remove(s, event_get_clock_data(s->event, s->type)); | |
3013 | ||
3014 | goto fail; | |
3015 | } | |
3016 | ||
3017 | event_source_pp_prioq_reshuffle(s); | |
3018 | ratelimit_reset(&s->rate_limit); | |
3019 | ||
3020 | log_debug("Event source %p (%s) left rate limit state.", s, strna(s->description)); | |
fd69f224 MS |
3021 | |
3022 | if (run_callback && s->ratelimit_expire_callback) { | |
3023 | s->dispatching = true; | |
3024 | r = s->ratelimit_expire_callback(s, s->userdata); | |
3025 | s->dispatching = false; | |
3026 | ||
3027 | if (r < 0) { | |
3028 | log_debug_errno(r, "Ratelimit expiry callback of event source %s (type %s) returned error, %s: %m", | |
3029 | strna(s->description), | |
3030 | event_source_type_to_string(s->type), | |
3031 | s->exit_on_failure ? "exiting" : "disabling"); | |
3032 | ||
3033 | if (s->exit_on_failure) | |
3034 | (void) sd_event_exit(s->event, r); | |
3035 | } | |
3036 | ||
3037 | if (s->n_ref == 0) | |
3038 | source_free(s); | |
3039 | else if (r < 0) | |
0a040e64 | 3040 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
fd69f224 MS |
3041 | |
3042 | return 1; | |
3043 | } | |
3044 | ||
b6d5481b LP |
3045 | return 0; |
3046 | ||
3047 | fail: | |
3048 | /* Do something somewhat reasonable when we cannot move an event sources out of ratelimited mode: | |
3049 | * simply put it back in it, maybe we can then process it more successfully next iteration. */ | |
3050 | assert_se(event_source_time_prioq_put(s, &s->event->monotonic) >= 0); | |
3051 | ||
3052 | return r; | |
3053 | } | |
3054 | ||
c2ba3ad6 LP |
3055 | static usec_t sleep_between(sd_event *e, usec_t a, usec_t b) { |
3056 | usec_t c; | |
3057 | assert(e); | |
3058 | assert(a <= b); | |
3059 | ||
3060 | if (a <= 0) | |
3061 | return 0; | |
393003e1 LP |
3062 | if (a >= USEC_INFINITY) |
3063 | return USEC_INFINITY; | |
c2ba3ad6 LP |
3064 | |
3065 | if (b <= a + 1) | |
3066 | return a; | |
3067 | ||
52444dc4 LP |
3068 | initialize_perturb(e); |
3069 | ||
c2ba3ad6 LP |
3070 | /* |
3071 | Find a good time to wake up again between times a and b. We | |
3072 | have two goals here: | |
3073 | ||
3074 | a) We want to wake up as seldom as possible, hence prefer | |
3075 | later times over earlier times. | |
3076 | ||
3077 | b) But if we have to wake up, then let's make sure to | |
3078 | dispatch as much as possible on the entire system. | |
3079 | ||
3080 | We implement this by waking up everywhere at the same time | |
850516e0 | 3081 | within any given minute if we can, synchronised via the |
c2ba3ad6 | 3082 | perturbation value determined from the boot ID. If we can't, |
ba276c81 LP |
3083 | then we try to find the same spot in every 10s, then 1s and |
3084 | then 250ms step. Otherwise, we pick the last possible time | |
3085 | to wake up. | |
c2ba3ad6 LP |
3086 | */ |
3087 | ||
850516e0 LP |
3088 | c = (b / USEC_PER_MINUTE) * USEC_PER_MINUTE + e->perturb; |
3089 | if (c >= b) { | |
3090 | if (_unlikely_(c < USEC_PER_MINUTE)) | |
3091 | return b; | |
3092 | ||
3093 | c -= USEC_PER_MINUTE; | |
3094 | } | |
3095 | ||
ba276c81 LP |
3096 | if (c >= a) |
3097 | return c; | |
3098 | ||
3099 | c = (b / (USEC_PER_SEC*10)) * (USEC_PER_SEC*10) + (e->perturb % (USEC_PER_SEC*10)); | |
3100 | if (c >= b) { | |
3101 | if (_unlikely_(c < USEC_PER_SEC*10)) | |
3102 | return b; | |
3103 | ||
3104 | c -= USEC_PER_SEC*10; | |
3105 | } | |
3106 | ||
850516e0 LP |
3107 | if (c >= a) |
3108 | return c; | |
3109 | ||
3110 | c = (b / USEC_PER_SEC) * USEC_PER_SEC + (e->perturb % USEC_PER_SEC); | |
c2ba3ad6 LP |
3111 | if (c >= b) { |
3112 | if (_unlikely_(c < USEC_PER_SEC)) | |
3113 | return b; | |
3114 | ||
3115 | c -= USEC_PER_SEC; | |
3116 | } | |
3117 | ||
3118 | if (c >= a) | |
3119 | return c; | |
3120 | ||
3121 | c = (b / (USEC_PER_MSEC*250)) * (USEC_PER_MSEC*250) + (e->perturb % (USEC_PER_MSEC*250)); | |
3122 | if (c >= b) { | |
3123 | if (_unlikely_(c < USEC_PER_MSEC*250)) | |
3124 | return b; | |
3125 | ||
3126 | c -= USEC_PER_MSEC*250; | |
3127 | } | |
3128 | ||
3129 | if (c >= a) | |
3130 | return c; | |
3131 | ||
3132 | return b; | |
3133 | } | |
3134 | ||
fd38203a LP |
3135 | static int event_arm_timer( |
3136 | sd_event *e, | |
6a0f1f6d | 3137 | struct clock_data *d) { |
fd38203a LP |
3138 | |
3139 | struct itimerspec its = {}; | |
c2ba3ad6 LP |
3140 | sd_event_source *a, *b; |
3141 | usec_t t; | |
fd38203a | 3142 | |
cde93897 | 3143 | assert(e); |
6a0f1f6d | 3144 | assert(d); |
fd38203a | 3145 | |
d06441da | 3146 | if (!d->needs_rearm) |
212bbb17 | 3147 | return 0; |
7e2bf71c YW |
3148 | |
3149 | d->needs_rearm = false; | |
212bbb17 | 3150 | |
6a0f1f6d | 3151 | a = prioq_peek(d->earliest); |
19947509 | 3152 | assert(!a || EVENT_SOURCE_USES_TIME_PRIOQ(a->type)); |
b6d5481b | 3153 | if (!a || a->enabled == SD_EVENT_OFF || time_event_source_next(a) == USEC_INFINITY) { |
72aedc1e | 3154 | |
6a0f1f6d | 3155 | if (d->fd < 0) |
c57b5ca3 LP |
3156 | return 0; |
3157 | ||
3a43da28 | 3158 | if (d->next == USEC_INFINITY) |
72aedc1e LP |
3159 | return 0; |
3160 | ||
3161 | /* disarm */ | |
15c689d7 LP |
3162 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
3163 | return -errno; | |
72aedc1e | 3164 | |
3a43da28 | 3165 | d->next = USEC_INFINITY; |
fd38203a | 3166 | return 0; |
72aedc1e | 3167 | } |
fd38203a | 3168 | |
6a0f1f6d | 3169 | b = prioq_peek(d->latest); |
19947509 ZJS |
3170 | assert(!b || EVENT_SOURCE_USES_TIME_PRIOQ(b->type)); |
3171 | assert(b && b->enabled != SD_EVENT_OFF); | |
c2ba3ad6 | 3172 | |
b6d5481b | 3173 | t = sleep_between(e, time_event_source_next(a), time_event_source_latest(b)); |
6a0f1f6d | 3174 | if (d->next == t) |
fd38203a LP |
3175 | return 0; |
3176 | ||
6a0f1f6d | 3177 | assert_se(d->fd >= 0); |
fd38203a | 3178 | |
c2ba3ad6 | 3179 | if (t == 0) { |
fd38203a LP |
3180 | /* We don' want to disarm here, just mean some time looooong ago. */ |
3181 | its.it_value.tv_sec = 0; | |
3182 | its.it_value.tv_nsec = 1; | |
3183 | } else | |
c2ba3ad6 | 3184 | timespec_store(&its.it_value, t); |
fd38203a | 3185 | |
15c689d7 | 3186 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 | 3187 | return -errno; |
fd38203a | 3188 | |
6a0f1f6d | 3189 | d->next = t; |
fd38203a LP |
3190 | return 0; |
3191 | } | |
3192 | ||
9a800b56 | 3193 | static int process_io(sd_event *e, sd_event_source *s, uint32_t revents) { |
fd38203a LP |
3194 | assert(e); |
3195 | assert(s); | |
3196 | assert(s->type == SOURCE_IO); | |
3197 | ||
9a800b56 LP |
3198 | /* If the event source was already pending, we just OR in the |
3199 | * new revents, otherwise we reset the value. The ORing is | |
3200 | * necessary to handle EPOLLONESHOT events properly where | |
3201 | * readability might happen independently of writability, and | |
3202 | * we need to keep track of both */ | |
3203 | ||
3204 | if (s->pending) | |
3205 | s->io.revents |= revents; | |
3206 | else | |
3207 | s->io.revents = revents; | |
fd38203a | 3208 | |
fd38203a LP |
3209 | return source_set_pending(s, true); |
3210 | } | |
3211 | ||
72aedc1e | 3212 | static int flush_timer(sd_event *e, int fd, uint32_t events, usec_t *next) { |
fd38203a LP |
3213 | uint64_t x; |
3214 | ssize_t ss; | |
3215 | ||
3216 | assert(e); | |
da7e457c | 3217 | assert(fd >= 0); |
72aedc1e | 3218 | |
305f78bf | 3219 | assert_return(events == EPOLLIN, -EIO); |
fd38203a LP |
3220 | |
3221 | ss = read(fd, &x, sizeof(x)); | |
3222 | if (ss < 0) { | |
8add30a0 | 3223 | if (ERRNO_IS_TRANSIENT(errno)) |
fd38203a LP |
3224 | return 0; |
3225 | ||
3226 | return -errno; | |
3227 | } | |
3228 | ||
8d35dae7 | 3229 | if (_unlikely_(ss != sizeof(x))) |
fd38203a LP |
3230 | return -EIO; |
3231 | ||
cde93897 | 3232 | if (next) |
3a43da28 | 3233 | *next = USEC_INFINITY; |
72aedc1e | 3234 | |
fd38203a LP |
3235 | return 0; |
3236 | } | |
3237 | ||
305f78bf LP |
3238 | static int process_timer( |
3239 | sd_event *e, | |
3240 | usec_t n, | |
6a0f1f6d | 3241 | struct clock_data *d) { |
305f78bf | 3242 | |
fd38203a | 3243 | sd_event_source *s; |
fd69f224 | 3244 | bool callback_invoked = false; |
fd38203a LP |
3245 | int r; |
3246 | ||
3247 | assert(e); | |
6a0f1f6d | 3248 | assert(d); |
fd38203a LP |
3249 | |
3250 | for (;;) { | |
6a0f1f6d | 3251 | s = prioq_peek(d->earliest); |
19947509 ZJS |
3252 | assert(!s || EVENT_SOURCE_USES_TIME_PRIOQ(s->type)); |
3253 | ||
b6d5481b LP |
3254 | if (!s || time_event_source_next(s) > n) |
3255 | break; | |
3256 | ||
3257 | if (s->ratelimited) { | |
3258 | /* This is an event sources whose ratelimit window has ended. Let's turn it on | |
3259 | * again. */ | |
3260 | assert(s->ratelimited); | |
3261 | ||
fd69f224 | 3262 | r = event_source_leave_ratelimit(s, /* run_callback */ true); |
b6d5481b LP |
3263 | if (r < 0) |
3264 | return r; | |
fd69f224 MS |
3265 | else if (r == 1) |
3266 | callback_invoked = true; | |
b6d5481b LP |
3267 | |
3268 | continue; | |
3269 | } | |
3270 | ||
3271 | if (s->enabled == SD_EVENT_OFF || s->pending) | |
fd38203a LP |
3272 | break; |
3273 | ||
3274 | r = source_set_pending(s, true); | |
3275 | if (r < 0) | |
3276 | return r; | |
3277 | ||
e1951c16 | 3278 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
3279 | } |
3280 | ||
fd69f224 | 3281 | return callback_invoked; |
fd38203a LP |
3282 | } |
3283 | ||
efd3be9d YW |
3284 | static int process_child(sd_event *e, int64_t threshold, int64_t *ret_min_priority) { |
3285 | int64_t min_priority = threshold; | |
3286 | bool something_new = false; | |
fd38203a | 3287 | sd_event_source *s; |
fd38203a LP |
3288 | int r; |
3289 | ||
3290 | assert(e); | |
efd3be9d YW |
3291 | assert(ret_min_priority); |
3292 | ||
3293 | if (!e->need_process_child) { | |
3294 | *ret_min_priority = min_priority; | |
3295 | return 0; | |
3296 | } | |
fd38203a | 3297 | |
c2ba3ad6 LP |
3298 | e->need_process_child = false; |
3299 | ||
91c70071 YW |
3300 | /* So, this is ugly. We iteratively invoke waitid() with P_PID + WNOHANG for each PID we wait |
3301 | * for, instead of using P_ALL. This is because we only want to get child information of very | |
3302 | * specific child processes, and not all of them. We might not have processed the SIGCHLD event | |
3303 | * of a previous invocation and we don't want to maintain a unbounded *per-child* event queue, | |
3304 | * hence we really don't want anything flushed out of the kernel's queue that we don't care | |
3305 | * about. Since this is O(n) this means that if you have a lot of processes you probably want | |
3306 | * to handle SIGCHLD yourself. | |
3307 | * | |
3308 | * We do not reap the children here (by using WNOWAIT), this is only done after the event | |
3309 | * source is dispatched so that the callback still sees the process as a zombie. */ | |
fd38203a | 3310 | |
90e74a66 | 3311 | HASHMAP_FOREACH(s, e->child_sources) { |
fd38203a LP |
3312 | assert(s->type == SOURCE_CHILD); |
3313 | ||
efd3be9d YW |
3314 | if (s->priority > threshold) |
3315 | continue; | |
3316 | ||
fd38203a LP |
3317 | if (s->pending) |
3318 | continue; | |
3319 | ||
b6d5481b | 3320 | if (event_source_is_offline(s)) |
fd38203a LP |
3321 | continue; |
3322 | ||
f8f3f926 LP |
3323 | if (s->child.exited) |
3324 | continue; | |
3325 | ||
91c70071 YW |
3326 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
3327 | /* There's a usable pidfd known for this event source? Then don't waitid() for | |
3328 | * it here */ | |
f8f3f926 LP |
3329 | continue; |
3330 | ||
fd38203a | 3331 | zero(s->child.siginfo); |
15c689d7 LP |
3332 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, |
3333 | WNOHANG | (s->child.options & WEXITED ? WNOWAIT : 0) | s->child.options) < 0) | |
bfd9bfcc | 3334 | return negative_errno(); |
fd38203a LP |
3335 | |
3336 | if (s->child.siginfo.si_pid != 0) { | |
945c2931 | 3337 | bool zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 3338 | |
f8f3f926 LP |
3339 | if (zombie) |
3340 | s->child.exited = true; | |
3341 | ||
08cd1552 | 3342 | if (!zombie && (s->child.options & WEXITED)) { |
91c70071 YW |
3343 | /* If the child isn't dead then let's immediately remove the state |
3344 | * change from the queue, since there's no benefit in leaving it | |
3345 | * queued. */ | |
08cd1552 LP |
3346 | |
3347 | assert(s->child.options & (WSTOPPED|WCONTINUED)); | |
a5d27871 | 3348 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|(s->child.options & (WSTOPPED|WCONTINUED))); |
08cd1552 LP |
3349 | } |
3350 | ||
fd38203a LP |
3351 | r = source_set_pending(s, true); |
3352 | if (r < 0) | |
3353 | return r; | |
efd3be9d YW |
3354 | if (r > 0) { |
3355 | something_new = true; | |
3356 | min_priority = MIN(min_priority, s->priority); | |
3357 | } | |
fd38203a LP |
3358 | } |
3359 | } | |
3360 | ||
efd3be9d YW |
3361 | *ret_min_priority = min_priority; |
3362 | return something_new; | |
fd38203a LP |
3363 | } |
3364 | ||
f8f3f926 LP |
3365 | static int process_pidfd(sd_event *e, sd_event_source *s, uint32_t revents) { |
3366 | assert(e); | |
3367 | assert(s); | |
3368 | assert(s->type == SOURCE_CHILD); | |
3369 | ||
3370 | if (s->pending) | |
3371 | return 0; | |
3372 | ||
b6d5481b | 3373 | if (event_source_is_offline(s)) |
f8f3f926 LP |
3374 | return 0; |
3375 | ||
3376 | if (!EVENT_SOURCE_WATCH_PIDFD(s)) | |
3377 | return 0; | |
3378 | ||
3379 | zero(s->child.siginfo); | |
3380 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG | WNOWAIT | s->child.options) < 0) | |
3381 | return -errno; | |
3382 | ||
3383 | if (s->child.siginfo.si_pid == 0) | |
3384 | return 0; | |
3385 | ||
3386 | if (IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) | |
3387 | s->child.exited = true; | |
3388 | ||
3389 | return source_set_pending(s, true); | |
3390 | } | |
3391 | ||
efd3be9d | 3392 | static int process_signal(sd_event *e, struct signal_data *d, uint32_t events, int64_t *min_priority) { |
fd38203a LP |
3393 | int r; |
3394 | ||
da7e457c | 3395 | assert(e); |
97ef5391 | 3396 | assert(d); |
305f78bf | 3397 | assert_return(events == EPOLLIN, -EIO); |
efd3be9d | 3398 | assert(min_priority); |
fd38203a | 3399 | |
91c70071 YW |
3400 | /* If there's a signal queued on this priority and SIGCHLD is on this priority too, then make |
3401 | * sure to recheck the children we watch. This is because we only ever dequeue the first signal | |
3402 | * per priority, and if we dequeue one, and SIGCHLD might be enqueued later we wouldn't know, | |
3403 | * but we might have higher priority children we care about hence we need to check that | |
3404 | * explicitly. */ | |
9da4cb2b LP |
3405 | |
3406 | if (sigismember(&d->sigset, SIGCHLD)) | |
3407 | e->need_process_child = true; | |
3408 | ||
91c70071 | 3409 | /* If there's already an event source pending for this priority we don't read another */ |
9da4cb2b LP |
3410 | if (d->current) |
3411 | return 0; | |
3412 | ||
fd38203a | 3413 | for (;;) { |
0eb2e0e3 | 3414 | struct signalfd_siginfo si; |
7057bd99 | 3415 | ssize_t n; |
92daebc0 | 3416 | sd_event_source *s = NULL; |
fd38203a | 3417 | |
9da4cb2b | 3418 | n = read(d->fd, &si, sizeof(si)); |
7057bd99 | 3419 | if (n < 0) { |
8add30a0 | 3420 | if (ERRNO_IS_TRANSIENT(errno)) |
efd3be9d | 3421 | return 0; |
fd38203a LP |
3422 | |
3423 | return -errno; | |
3424 | } | |
3425 | ||
7057bd99 | 3426 | if (_unlikely_(n != sizeof(si))) |
fd38203a LP |
3427 | return -EIO; |
3428 | ||
6eb7c172 | 3429 | assert(SIGNAL_VALID(si.ssi_signo)); |
7057bd99 | 3430 | |
92daebc0 LP |
3431 | if (e->signal_sources) |
3432 | s = e->signal_sources[si.ssi_signo]; | |
92daebc0 LP |
3433 | if (!s) |
3434 | continue; | |
9da4cb2b LP |
3435 | if (s->pending) |
3436 | continue; | |
fd38203a LP |
3437 | |
3438 | s->signal.siginfo = si; | |
9da4cb2b LP |
3439 | d->current = s; |
3440 | ||
fd38203a LP |
3441 | r = source_set_pending(s, true); |
3442 | if (r < 0) | |
3443 | return r; | |
efd3be9d YW |
3444 | if (r > 0 && *min_priority >= s->priority) { |
3445 | *min_priority = s->priority; | |
3446 | return 1; /* an event source with smaller priority is queued. */ | |
3447 | } | |
9da4cb2b | 3448 | |
efd3be9d | 3449 | return 0; |
fd38203a | 3450 | } |
fd38203a LP |
3451 | } |
3452 | ||
efd3be9d | 3453 | static int event_inotify_data_read(sd_event *e, struct inotify_data *d, uint32_t revents, int64_t threshold) { |
97ef5391 LP |
3454 | ssize_t n; |
3455 | ||
3456 | assert(e); | |
3457 | assert(d); | |
3458 | ||
3459 | assert_return(revents == EPOLLIN, -EIO); | |
3460 | ||
3461 | /* If there's already an event source pending for this priority, don't read another */ | |
3462 | if (d->n_pending > 0) | |
3463 | return 0; | |
3464 | ||
3465 | /* Is the read buffer non-empty? If so, let's not read more */ | |
3466 | if (d->buffer_filled > 0) | |
3467 | return 0; | |
3468 | ||
efd3be9d YW |
3469 | if (d->priority > threshold) |
3470 | return 0; | |
3471 | ||
97ef5391 LP |
3472 | n = read(d->fd, &d->buffer, sizeof(d->buffer)); |
3473 | if (n < 0) { | |
8add30a0 | 3474 | if (ERRNO_IS_TRANSIENT(errno)) |
97ef5391 LP |
3475 | return 0; |
3476 | ||
3477 | return -errno; | |
3478 | } | |
3479 | ||
3480 | assert(n > 0); | |
3481 | d->buffer_filled = (size_t) n; | |
0601b958 | 3482 | LIST_PREPEND(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
3483 | |
3484 | return 1; | |
3485 | } | |
3486 | ||
3487 | static void event_inotify_data_drop(sd_event *e, struct inotify_data *d, size_t sz) { | |
3488 | assert(e); | |
3489 | assert(d); | |
3490 | assert(sz <= d->buffer_filled); | |
3491 | ||
3492 | if (sz == 0) | |
3493 | return; | |
3494 | ||
3495 | /* Move the rest to the buffer to the front, in order to get things properly aligned again */ | |
3496 | memmove(d->buffer.raw, d->buffer.raw + sz, d->buffer_filled - sz); | |
3497 | d->buffer_filled -= sz; | |
3498 | ||
3499 | if (d->buffer_filled == 0) | |
0601b958 | 3500 | LIST_REMOVE(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
3501 | } |
3502 | ||
3503 | static int event_inotify_data_process(sd_event *e, struct inotify_data *d) { | |
3504 | int r; | |
3505 | ||
3506 | assert(e); | |
3507 | assert(d); | |
3508 | ||
3509 | /* If there's already an event source pending for this priority, don't read another */ | |
3510 | if (d->n_pending > 0) | |
3511 | return 0; | |
3512 | ||
3513 | while (d->buffer_filled > 0) { | |
3514 | size_t sz; | |
3515 | ||
3516 | /* Let's validate that the event structures are complete */ | |
3517 | if (d->buffer_filled < offsetof(struct inotify_event, name)) | |
3518 | return -EIO; | |
3519 | ||
3520 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3521 | if (d->buffer_filled < sz) | |
3522 | return -EIO; | |
3523 | ||
3524 | if (d->buffer.ev.mask & IN_Q_OVERFLOW) { | |
3525 | struct inode_data *inode_data; | |
97ef5391 LP |
3526 | |
3527 | /* The queue overran, let's pass this event to all event sources connected to this inotify | |
3528 | * object */ | |
3529 | ||
03677889 | 3530 | HASHMAP_FOREACH(inode_data, d->inodes) |
97ef5391 LP |
3531 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { |
3532 | ||
b6d5481b | 3533 | if (event_source_is_offline(s)) |
97ef5391 LP |
3534 | continue; |
3535 | ||
3536 | r = source_set_pending(s, true); | |
3537 | if (r < 0) | |
3538 | return r; | |
3539 | } | |
97ef5391 LP |
3540 | } else { |
3541 | struct inode_data *inode_data; | |
97ef5391 LP |
3542 | |
3543 | /* Find the inode object for this watch descriptor. If IN_IGNORED is set we also remove it from | |
3544 | * our watch descriptor table. */ | |
3545 | if (d->buffer.ev.mask & IN_IGNORED) { | |
3546 | ||
3547 | inode_data = hashmap_remove(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3548 | if (!inode_data) { | |
3549 | event_inotify_data_drop(e, d, sz); | |
3550 | continue; | |
3551 | } | |
3552 | ||
3553 | /* The watch descriptor was removed by the kernel, let's drop it here too */ | |
3554 | inode_data->wd = -1; | |
3555 | } else { | |
3556 | inode_data = hashmap_get(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3557 | if (!inode_data) { | |
3558 | event_inotify_data_drop(e, d, sz); | |
3559 | continue; | |
3560 | } | |
3561 | } | |
3562 | ||
3563 | /* Trigger all event sources that are interested in these events. Also trigger all event | |
3564 | * sources if IN_IGNORED or IN_UNMOUNT is set. */ | |
3565 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
3566 | ||
b6d5481b | 3567 | if (event_source_is_offline(s)) |
97ef5391 LP |
3568 | continue; |
3569 | ||
3570 | if ((d->buffer.ev.mask & (IN_IGNORED|IN_UNMOUNT)) == 0 && | |
3571 | (s->inotify.mask & d->buffer.ev.mask & IN_ALL_EVENTS) == 0) | |
3572 | continue; | |
3573 | ||
3574 | r = source_set_pending(s, true); | |
3575 | if (r < 0) | |
3576 | return r; | |
3577 | } | |
3578 | } | |
3579 | ||
3580 | /* Something pending now? If so, let's finish, otherwise let's read more. */ | |
3581 | if (d->n_pending > 0) | |
3582 | return 1; | |
3583 | } | |
3584 | ||
3585 | return 0; | |
3586 | } | |
3587 | ||
3588 | static int process_inotify(sd_event *e) { | |
97ef5391 LP |
3589 | int r, done = 0; |
3590 | ||
3591 | assert(e); | |
3592 | ||
0601b958 | 3593 | LIST_FOREACH(buffered, d, e->buffered_inotify_data_list) { |
97ef5391 LP |
3594 | r = event_inotify_data_process(e, d); |
3595 | if (r < 0) | |
3596 | return r; | |
3597 | if (r > 0) | |
3598 | done ++; | |
3599 | } | |
3600 | ||
3601 | return done; | |
3602 | } | |
3603 | ||
fd38203a | 3604 | static int source_dispatch(sd_event_source *s) { |
8f5c235d | 3605 | EventSourceType saved_type; |
c8e9d15c | 3606 | sd_event *saved_event; |
fe8245eb | 3607 | int r = 0; |
fd38203a LP |
3608 | |
3609 | assert(s); | |
6203e07a | 3610 | assert(s->pending || s->type == SOURCE_EXIT); |
fd38203a | 3611 | |
b778cba4 LP |
3612 | /* Save the event source type, here, so that we still know it after the event callback which might |
3613 | * invalidate the event. */ | |
8f5c235d LP |
3614 | saved_type = s->type; |
3615 | ||
de02634c | 3616 | /* Similarly, store a reference to the event loop object, so that we can still access it after the |
b778cba4 | 3617 | * callback might have invalidated/disconnected the event source. */ |
c8e9d15c YW |
3618 | saved_event = s->event; |
3619 | PROTECT_EVENT(saved_event); | |
b778cba4 | 3620 | |
de02634c | 3621 | /* Check if we hit the ratelimit for this event source, and if so, let's disable it. */ |
b6d5481b LP |
3622 | assert(!s->ratelimited); |
3623 | if (!ratelimit_below(&s->rate_limit)) { | |
3624 | r = event_source_enter_ratelimited(s); | |
3625 | if (r < 0) | |
3626 | return r; | |
3627 | ||
3628 | return 1; | |
3629 | } | |
3630 | ||
945c2931 | 3631 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { |
da7e457c LP |
3632 | r = source_set_pending(s, false); |
3633 | if (r < 0) | |
3634 | return r; | |
3635 | } | |
fd38203a | 3636 | |
6e9feda3 LP |
3637 | if (s->type != SOURCE_POST) { |
3638 | sd_event_source *z; | |
6e9feda3 | 3639 | |
de02634c | 3640 | /* If we execute a non-post source, let's mark all post sources as pending. */ |
6e9feda3 | 3641 | |
90e74a66 | 3642 | SET_FOREACH(z, s->event->post_sources) { |
b6d5481b | 3643 | if (event_source_is_offline(z)) |
6e9feda3 LP |
3644 | continue; |
3645 | ||
3646 | r = source_set_pending(z, true); | |
3647 | if (r < 0) | |
3648 | return r; | |
3649 | } | |
3650 | } | |
3651 | ||
baf76283 LP |
3652 | if (s->enabled == SD_EVENT_ONESHOT) { |
3653 | r = sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
3654 | if (r < 0) |
3655 | return r; | |
3656 | } | |
3657 | ||
12179984 | 3658 | s->dispatching = true; |
b7484e2a | 3659 | |
fd38203a LP |
3660 | switch (s->type) { |
3661 | ||
3662 | case SOURCE_IO: | |
3663 | r = s->io.callback(s, s->io.fd, s->io.revents, s->userdata); | |
3664 | break; | |
3665 | ||
6a0f1f6d | 3666 | case SOURCE_TIME_REALTIME: |
a8548816 | 3667 | case SOURCE_TIME_BOOTTIME: |
6a0f1f6d LP |
3668 | case SOURCE_TIME_MONOTONIC: |
3669 | case SOURCE_TIME_REALTIME_ALARM: | |
3670 | case SOURCE_TIME_BOOTTIME_ALARM: | |
fd38203a LP |
3671 | r = s->time.callback(s, s->time.next, s->userdata); |
3672 | break; | |
3673 | ||
3674 | case SOURCE_SIGNAL: | |
3675 | r = s->signal.callback(s, &s->signal.siginfo, s->userdata); | |
3676 | break; | |
3677 | ||
08cd1552 LP |
3678 | case SOURCE_CHILD: { |
3679 | bool zombie; | |
3680 | ||
945c2931 | 3681 | zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 3682 | |
fd38203a | 3683 | r = s->child.callback(s, &s->child.siginfo, s->userdata); |
08cd1552 LP |
3684 | |
3685 | /* Now, reap the PID for good. */ | |
f8f3f926 | 3686 | if (zombie) { |
cc59d290 | 3687 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|WEXITED); |
f8f3f926 LP |
3688 | s->child.waited = true; |
3689 | } | |
08cd1552 | 3690 | |
fd38203a | 3691 | break; |
08cd1552 | 3692 | } |
fd38203a LP |
3693 | |
3694 | case SOURCE_DEFER: | |
3695 | r = s->defer.callback(s, s->userdata); | |
3696 | break; | |
da7e457c | 3697 | |
6e9feda3 LP |
3698 | case SOURCE_POST: |
3699 | r = s->post.callback(s, s->userdata); | |
3700 | break; | |
3701 | ||
6203e07a LP |
3702 | case SOURCE_EXIT: |
3703 | r = s->exit.callback(s, s->userdata); | |
da7e457c | 3704 | break; |
9d3e3aa5 | 3705 | |
97ef5391 LP |
3706 | case SOURCE_INOTIFY: { |
3707 | struct sd_event *e = s->event; | |
3708 | struct inotify_data *d; | |
3709 | size_t sz; | |
3710 | ||
3711 | assert(s->inotify.inode_data); | |
3712 | assert_se(d = s->inotify.inode_data->inotify_data); | |
3713 | ||
3714 | assert(d->buffer_filled >= offsetof(struct inotify_event, name)); | |
3715 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3716 | assert(d->buffer_filled >= sz); | |
3717 | ||
53baf2ef LP |
3718 | /* If the inotify callback destroys the event source then this likely means we don't need to |
3719 | * watch the inode anymore, and thus also won't need the inotify object anymore. But if we'd | |
3720 | * free it immediately, then we couldn't drop the event from the inotify event queue without | |
3721 | * memory corruption anymore, as below. Hence, let's not free it immediately, but mark it | |
3722 | * "busy" with a counter (which will ensure it's not GC'ed away prematurely). Let's then | |
3723 | * explicitly GC it after we are done dropping the inotify event from the buffer. */ | |
3724 | d->n_busy++; | |
97ef5391 | 3725 | r = s->inotify.callback(s, &d->buffer.ev, s->userdata); |
53baf2ef | 3726 | d->n_busy--; |
97ef5391 | 3727 | |
53baf2ef LP |
3728 | /* When no event is pending anymore on this inotify object, then let's drop the event from |
3729 | * the inotify event queue buffer. */ | |
97ef5391 LP |
3730 | if (d->n_pending == 0) |
3731 | event_inotify_data_drop(e, d, sz); | |
3732 | ||
53baf2ef LP |
3733 | /* Now we don't want to access 'd' anymore, it's OK to GC now. */ |
3734 | event_gc_inotify_data(e, d); | |
97ef5391 LP |
3735 | break; |
3736 | } | |
3737 | ||
9d3e3aa5 | 3738 | case SOURCE_WATCHDOG: |
a71fe8b8 | 3739 | case _SOURCE_EVENT_SOURCE_TYPE_MAX: |
9f2a50a3 | 3740 | case _SOURCE_EVENT_SOURCE_TYPE_INVALID: |
04499a70 | 3741 | assert_not_reached(); |
fd38203a LP |
3742 | } |
3743 | ||
12179984 LP |
3744 | s->dispatching = false; |
3745 | ||
b778cba4 LP |
3746 | if (r < 0) { |
3747 | log_debug_errno(r, "Event source %s (type %s) returned error, %s: %m", | |
3748 | strna(s->description), | |
3749 | event_source_type_to_string(saved_type), | |
3750 | s->exit_on_failure ? "exiting" : "disabling"); | |
3751 | ||
3752 | if (s->exit_on_failure) | |
3753 | (void) sd_event_exit(saved_event, r); | |
3754 | } | |
12179984 LP |
3755 | |
3756 | if (s->n_ref == 0) | |
3757 | source_free(s); | |
3758 | else if (r < 0) | |
c3c50474 | 3759 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
b7484e2a | 3760 | |
6203e07a | 3761 | return 1; |
fd38203a LP |
3762 | } |
3763 | ||
3764 | static int event_prepare(sd_event *e) { | |
3765 | int r; | |
3766 | ||
3767 | assert(e); | |
3768 | ||
3769 | for (;;) { | |
3770 | sd_event_source *s; | |
3771 | ||
3772 | s = prioq_peek(e->prepare); | |
b6d5481b | 3773 | if (!s || s->prepare_iteration == e->iteration || event_source_is_offline(s)) |
fd38203a LP |
3774 | break; |
3775 | ||
3776 | s->prepare_iteration = e->iteration; | |
3777 | r = prioq_reshuffle(e->prepare, s, &s->prepare_index); | |
3778 | if (r < 0) | |
3779 | return r; | |
3780 | ||
3781 | assert(s->prepare); | |
12179984 LP |
3782 | |
3783 | s->dispatching = true; | |
fd38203a | 3784 | r = s->prepare(s, s->userdata); |
12179984 LP |
3785 | s->dispatching = false; |
3786 | ||
b778cba4 LP |
3787 | if (r < 0) { |
3788 | log_debug_errno(r, "Prepare callback of event source %s (type %s) returned error, %s: %m", | |
3789 | strna(s->description), | |
3790 | event_source_type_to_string(s->type), | |
3791 | s->exit_on_failure ? "exiting" : "disabling"); | |
3792 | ||
3793 | if (s->exit_on_failure) | |
3794 | (void) sd_event_exit(e, r); | |
3795 | } | |
fd38203a | 3796 | |
12179984 LP |
3797 | if (s->n_ref == 0) |
3798 | source_free(s); | |
3799 | else if (r < 0) | |
c3c50474 | 3800 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
fd38203a LP |
3801 | } |
3802 | ||
3803 | return 0; | |
3804 | } | |
3805 | ||
6203e07a | 3806 | static int dispatch_exit(sd_event *e) { |
da7e457c LP |
3807 | sd_event_source *p; |
3808 | int r; | |
3809 | ||
3810 | assert(e); | |
3811 | ||
6203e07a | 3812 | p = prioq_peek(e->exit); |
19947509 ZJS |
3813 | assert(!p || p->type == SOURCE_EXIT); |
3814 | ||
b6d5481b | 3815 | if (!p || event_source_is_offline(p)) { |
da7e457c LP |
3816 | e->state = SD_EVENT_FINISHED; |
3817 | return 0; | |
3818 | } | |
3819 | ||
c8e9d15c | 3820 | PROTECT_EVENT(e); |
da7e457c | 3821 | e->iteration++; |
6203e07a | 3822 | e->state = SD_EVENT_EXITING; |
da7e457c | 3823 | r = source_dispatch(p); |
2b0c9ef7 | 3824 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
3825 | return r; |
3826 | } | |
3827 | ||
c2ba3ad6 LP |
3828 | static sd_event_source* event_next_pending(sd_event *e) { |
3829 | sd_event_source *p; | |
3830 | ||
da7e457c LP |
3831 | assert(e); |
3832 | ||
c2ba3ad6 LP |
3833 | p = prioq_peek(e->pending); |
3834 | if (!p) | |
3835 | return NULL; | |
3836 | ||
b6d5481b | 3837 | if (event_source_is_offline(p)) |
c2ba3ad6 LP |
3838 | return NULL; |
3839 | ||
3840 | return p; | |
3841 | } | |
3842 | ||
cde93897 LP |
3843 | static int arm_watchdog(sd_event *e) { |
3844 | struct itimerspec its = {}; | |
3845 | usec_t t; | |
cde93897 LP |
3846 | |
3847 | assert(e); | |
3848 | assert(e->watchdog_fd >= 0); | |
3849 | ||
3850 | t = sleep_between(e, | |
a595fb5c YW |
3851 | usec_add(e->watchdog_last, (e->watchdog_period / 2)), |
3852 | usec_add(e->watchdog_last, (e->watchdog_period * 3 / 4))); | |
cde93897 LP |
3853 | |
3854 | timespec_store(&its.it_value, t); | |
3855 | ||
75145780 LP |
3856 | /* Make sure we never set the watchdog to 0, which tells the |
3857 | * kernel to disable it. */ | |
3858 | if (its.it_value.tv_sec == 0 && its.it_value.tv_nsec == 0) | |
3859 | its.it_value.tv_nsec = 1; | |
3860 | ||
7c248223 | 3861 | return RET_NERRNO(timerfd_settime(e->watchdog_fd, TFD_TIMER_ABSTIME, &its, NULL)); |
cde93897 LP |
3862 | } |
3863 | ||
3864 | static int process_watchdog(sd_event *e) { | |
3865 | assert(e); | |
3866 | ||
3867 | if (!e->watchdog) | |
3868 | return 0; | |
3869 | ||
3870 | /* Don't notify watchdog too often */ | |
3871 | if (e->watchdog_last + e->watchdog_period / 4 > e->timestamp.monotonic) | |
3872 | return 0; | |
3873 | ||
3874 | sd_notify(false, "WATCHDOG=1"); | |
3875 | e->watchdog_last = e->timestamp.monotonic; | |
3876 | ||
3877 | return arm_watchdog(e); | |
3878 | } | |
3879 | ||
97ef5391 LP |
3880 | static void event_close_inode_data_fds(sd_event *e) { |
3881 | struct inode_data *d; | |
3882 | ||
3883 | assert(e); | |
3884 | ||
3885 | /* Close the fds pointing to the inodes to watch now. We need to close them as they might otherwise pin | |
3886 | * filesystems. But we can't close them right-away as we need them as long as the user still wants to make | |
365c2885 | 3887 | * adjustments to the event source, such as changing the priority (which requires us to remove and re-add a watch |
97ef5391 LP |
3888 | * for the inode). Hence, let's close them when entering the first iteration after they were added, as a |
3889 | * compromise. */ | |
3890 | ||
ed828563 | 3891 | while ((d = e->inode_data_to_close_list)) { |
97ef5391 LP |
3892 | assert(d->fd >= 0); |
3893 | d->fd = safe_close(d->fd); | |
3894 | ||
ed828563 | 3895 | LIST_REMOVE(to_close, e->inode_data_to_close_list, d); |
97ef5391 LP |
3896 | } |
3897 | } | |
3898 | ||
c45a5a74 TG |
3899 | _public_ int sd_event_prepare(sd_event *e) { |
3900 | int r; | |
fd38203a | 3901 | |
da7e457c | 3902 | assert_return(e, -EINVAL); |
b937d761 | 3903 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
3904 | assert_return(!event_pid_changed(e), -ECHILD); |
3905 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 3906 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 3907 | |
e5446015 LP |
3908 | /* Let's check that if we are a default event loop we are executed in the correct thread. We only do |
3909 | * this check here once, since gettid() is typically not cached, and thus want to minimize | |
3910 | * syscalls */ | |
3911 | assert_return(!e->default_event_ptr || e->tid == gettid(), -EREMOTEIO); | |
3912 | ||
f814c871 | 3913 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
c8e9d15c | 3914 | PROTECT_EVENT(e); |
f814c871 | 3915 | |
6203e07a | 3916 | if (e->exit_requested) |
c45a5a74 | 3917 | goto pending; |
fd38203a LP |
3918 | |
3919 | e->iteration++; | |
3920 | ||
0be6c2f6 | 3921 | e->state = SD_EVENT_PREPARING; |
fd38203a | 3922 | r = event_prepare(e); |
0be6c2f6 | 3923 | e->state = SD_EVENT_INITIAL; |
fd38203a | 3924 | if (r < 0) |
c45a5a74 | 3925 | return r; |
fd38203a | 3926 | |
6a0f1f6d LP |
3927 | r = event_arm_timer(e, &e->realtime); |
3928 | if (r < 0) | |
c45a5a74 | 3929 | return r; |
6a0f1f6d | 3930 | |
a8548816 TG |
3931 | r = event_arm_timer(e, &e->boottime); |
3932 | if (r < 0) | |
c45a5a74 | 3933 | return r; |
a8548816 | 3934 | |
6a0f1f6d LP |
3935 | r = event_arm_timer(e, &e->monotonic); |
3936 | if (r < 0) | |
c45a5a74 | 3937 | return r; |
6a0f1f6d LP |
3938 | |
3939 | r = event_arm_timer(e, &e->realtime_alarm); | |
1b5995b0 | 3940 | if (r < 0) |
c45a5a74 | 3941 | return r; |
fd38203a | 3942 | |
6a0f1f6d | 3943 | r = event_arm_timer(e, &e->boottime_alarm); |
1b5995b0 | 3944 | if (r < 0) |
c45a5a74 | 3945 | return r; |
fd38203a | 3946 | |
97ef5391 LP |
3947 | event_close_inode_data_fds(e); |
3948 | ||
0601b958 | 3949 | if (event_next_pending(e) || e->need_process_child || e->buffered_inotify_data_list) |
c45a5a74 TG |
3950 | goto pending; |
3951 | ||
2b0c9ef7 | 3952 | e->state = SD_EVENT_ARMED; |
c45a5a74 TG |
3953 | |
3954 | return 0; | |
3955 | ||
3956 | pending: | |
2b0c9ef7 | 3957 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
3958 | r = sd_event_wait(e, 0); |
3959 | if (r == 0) | |
2b0c9ef7 | 3960 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
3961 | |
3962 | return r; | |
c45a5a74 TG |
3963 | } |
3964 | ||
798445ab LP |
3965 | static int epoll_wait_usec( |
3966 | int fd, | |
3967 | struct epoll_event *events, | |
3968 | int maxevents, | |
3969 | usec_t timeout) { | |
3970 | ||
7c248223 | 3971 | int msec; |
39f756d3 ZJS |
3972 | #if 0 |
3973 | static bool epoll_pwait2_absent = false; | |
52bb308c | 3974 | int r; |
798445ab | 3975 | |
39f756d3 ZJS |
3976 | /* A wrapper that uses epoll_pwait2() if available, and falls back to epoll_wait() if not. |
3977 | * | |
3978 | * FIXME: this is temporarily disabled until epoll_pwait2() becomes more widely available. | |
3979 | * See https://github.com/systemd/systemd/pull/18973 and | |
3980 | * https://github.com/systemd/systemd/issues/19052. */ | |
798445ab LP |
3981 | |
3982 | if (!epoll_pwait2_absent && timeout != USEC_INFINITY) { | |
798445ab LP |
3983 | r = epoll_pwait2(fd, |
3984 | events, | |
3985 | maxevents, | |
52bb308c | 3986 | TIMESPEC_STORE(timeout), |
798445ab LP |
3987 | NULL); |
3988 | if (r >= 0) | |
3989 | return r; | |
7cb45dbf | 3990 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) |
798445ab LP |
3991 | return -errno; /* Only fallback to old epoll_wait() if the syscall is masked or not |
3992 | * supported. */ | |
3993 | ||
3994 | epoll_pwait2_absent = true; | |
3995 | } | |
39f756d3 | 3996 | #endif |
798445ab LP |
3997 | |
3998 | if (timeout == USEC_INFINITY) | |
3999 | msec = -1; | |
4000 | else { | |
4001 | usec_t k; | |
4002 | ||
4003 | k = DIV_ROUND_UP(timeout, USEC_PER_MSEC); | |
4004 | if (k >= INT_MAX) | |
4005 | msec = INT_MAX; /* Saturate */ | |
4006 | else | |
4007 | msec = (int) k; | |
4008 | } | |
4009 | ||
7c248223 | 4010 | return RET_NERRNO(epoll_wait(fd, events, maxevents, msec)); |
798445ab LP |
4011 | } |
4012 | ||
efd3be9d | 4013 | static int process_epoll(sd_event *e, usec_t timeout, int64_t threshold, int64_t *ret_min_priority) { |
319a4f4b | 4014 | size_t n_event_queue, m, n_event_max; |
efd3be9d YW |
4015 | int64_t min_priority = threshold; |
4016 | bool something_new = false; | |
798445ab | 4017 | int r; |
c45a5a74 | 4018 | |
efd3be9d YW |
4019 | assert(e); |
4020 | assert(ret_min_priority); | |
6a0f1f6d | 4021 | |
8b9708d1 | 4022 | n_event_queue = MAX(e->n_sources, 1u); |
319a4f4b | 4023 | if (!GREEDY_REALLOC(e->event_queue, n_event_queue)) |
5cddd924 | 4024 | return -ENOMEM; |
fd38203a | 4025 | |
319a4f4b LP |
4026 | n_event_max = MALLOC_ELEMENTSOF(e->event_queue); |
4027 | ||
97ef5391 | 4028 | /* If we still have inotify data buffered, then query the other fds, but don't wait on it */ |
0601b958 | 4029 | if (e->buffered_inotify_data_list) |
798445ab | 4030 | timeout = 0; |
97ef5391 | 4031 | |
8b9708d1 | 4032 | for (;;) { |
319a4f4b LP |
4033 | r = epoll_wait_usec( |
4034 | e->epoll_fd, | |
4035 | e->event_queue, | |
4036 | n_event_max, | |
4037 | timeout); | |
798445ab | 4038 | if (r < 0) |
efd3be9d | 4039 | return r; |
c45a5a74 | 4040 | |
8b9708d1 YW |
4041 | m = (size_t) r; |
4042 | ||
319a4f4b | 4043 | if (m < n_event_max) |
8b9708d1 YW |
4044 | break; |
4045 | ||
319a4f4b | 4046 | if (n_event_max >= n_event_queue * 10) |
8b9708d1 YW |
4047 | break; |
4048 | ||
319a4f4b | 4049 | if (!GREEDY_REALLOC(e->event_queue, n_event_max + n_event_queue)) |
8b9708d1 YW |
4050 | return -ENOMEM; |
4051 | ||
319a4f4b | 4052 | n_event_max = MALLOC_ELEMENTSOF(e->event_queue); |
798445ab | 4053 | timeout = 0; |
da7e457c | 4054 | } |
fd38203a | 4055 | |
efd3be9d YW |
4056 | /* Set timestamp only when this is called first time. */ |
4057 | if (threshold == INT64_MAX) | |
4058 | triple_timestamp_get(&e->timestamp); | |
fd38203a | 4059 | |
8b9708d1 | 4060 | for (size_t i = 0; i < m; i++) { |
fd38203a | 4061 | |
5cddd924 LP |
4062 | if (e->event_queue[i].data.ptr == INT_TO_PTR(SOURCE_WATCHDOG)) |
4063 | r = flush_timer(e, e->watchdog_fd, e->event_queue[i].events, NULL); | |
9da4cb2b | 4064 | else { |
5cddd924 | 4065 | WakeupType *t = e->event_queue[i].data.ptr; |
9da4cb2b LP |
4066 | |
4067 | switch (*t) { | |
4068 | ||
f8f3f926 | 4069 | case WAKEUP_EVENT_SOURCE: { |
5cddd924 | 4070 | sd_event_source *s = e->event_queue[i].data.ptr; |
f8f3f926 LP |
4071 | |
4072 | assert(s); | |
4073 | ||
efd3be9d YW |
4074 | if (s->priority > threshold) |
4075 | continue; | |
4076 | ||
4077 | min_priority = MIN(min_priority, s->priority); | |
4078 | ||
f8f3f926 LP |
4079 | switch (s->type) { |
4080 | ||
4081 | case SOURCE_IO: | |
5cddd924 | 4082 | r = process_io(e, s, e->event_queue[i].events); |
f8f3f926 LP |
4083 | break; |
4084 | ||
4085 | case SOURCE_CHILD: | |
5cddd924 | 4086 | r = process_pidfd(e, s, e->event_queue[i].events); |
f8f3f926 LP |
4087 | break; |
4088 | ||
4089 | default: | |
04499a70 | 4090 | assert_not_reached(); |
f8f3f926 LP |
4091 | } |
4092 | ||
9da4cb2b | 4093 | break; |
f8f3f926 | 4094 | } |
fd38203a | 4095 | |
9da4cb2b | 4096 | case WAKEUP_CLOCK_DATA: { |
5cddd924 | 4097 | struct clock_data *d = e->event_queue[i].data.ptr; |
f8f3f926 LP |
4098 | |
4099 | assert(d); | |
4100 | ||
5cddd924 | 4101 | r = flush_timer(e, d->fd, e->event_queue[i].events, &d->next); |
9da4cb2b LP |
4102 | break; |
4103 | } | |
4104 | ||
4105 | case WAKEUP_SIGNAL_DATA: | |
efd3be9d | 4106 | r = process_signal(e, e->event_queue[i].data.ptr, e->event_queue[i].events, &min_priority); |
9da4cb2b LP |
4107 | break; |
4108 | ||
97ef5391 | 4109 | case WAKEUP_INOTIFY_DATA: |
efd3be9d | 4110 | r = event_inotify_data_read(e, e->event_queue[i].data.ptr, e->event_queue[i].events, threshold); |
97ef5391 LP |
4111 | break; |
4112 | ||
9da4cb2b | 4113 | default: |
04499a70 | 4114 | assert_not_reached(); |
9da4cb2b LP |
4115 | } |
4116 | } | |
efd3be9d YW |
4117 | if (r < 0) |
4118 | return r; | |
4119 | if (r > 0) | |
4120 | something_new = true; | |
4121 | } | |
4122 | ||
4123 | *ret_min_priority = min_priority; | |
4124 | return something_new; | |
4125 | } | |
4126 | ||
4127 | _public_ int sd_event_wait(sd_event *e, uint64_t timeout) { | |
4128 | int r; | |
4129 | ||
4130 | assert_return(e, -EINVAL); | |
4131 | assert_return(e = event_resolve(e), -ENOPKG); | |
4132 | assert_return(!event_pid_changed(e), -ECHILD); | |
4133 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
4134 | assert_return(e->state == SD_EVENT_ARMED, -EBUSY); | |
4135 | ||
4136 | if (e->exit_requested) { | |
4137 | e->state = SD_EVENT_PENDING; | |
4138 | return 1; | |
4139 | } | |
4140 | ||
4141 | for (int64_t threshold = INT64_MAX; ; threshold--) { | |
4142 | int64_t epoll_min_priority, child_min_priority; | |
4143 | ||
4144 | /* There may be a possibility that new epoll (especially IO) and child events are | |
4145 | * triggered just after process_epoll() call but before process_child(), and the new IO | |
4146 | * events may have higher priority than the child events. To salvage these events, | |
4147 | * let's call epoll_wait() again, but accepts only events with higher priority than the | |
4148 | * previous. See issue https://github.com/systemd/systemd/issues/18190 and comments | |
4149 | * https://github.com/systemd/systemd/pull/18750#issuecomment-785801085 | |
4150 | * https://github.com/systemd/systemd/pull/18922#issuecomment-792825226 */ | |
4151 | ||
4152 | r = process_epoll(e, timeout, threshold, &epoll_min_priority); | |
4153 | if (r == -EINTR) { | |
4154 | e->state = SD_EVENT_PENDING; | |
4155 | return 1; | |
4156 | } | |
4157 | if (r < 0) | |
4158 | goto finish; | |
4159 | if (r == 0 && threshold < INT64_MAX) | |
4160 | /* No new epoll event. */ | |
4161 | break; | |
4162 | ||
4163 | r = process_child(e, threshold, &child_min_priority); | |
fd38203a | 4164 | if (r < 0) |
da7e457c | 4165 | goto finish; |
efd3be9d YW |
4166 | if (r == 0) |
4167 | /* No new child event. */ | |
4168 | break; | |
4169 | ||
4170 | threshold = MIN(epoll_min_priority, child_min_priority); | |
4171 | if (threshold == INT64_MIN) | |
4172 | break; | |
4173 | ||
4174 | timeout = 0; | |
fd38203a LP |
4175 | } |
4176 | ||
cde93897 LP |
4177 | r = process_watchdog(e); |
4178 | if (r < 0) | |
4179 | goto finish; | |
4180 | ||
fd69f224 | 4181 | r = process_inotify(e); |
6a0f1f6d LP |
4182 | if (r < 0) |
4183 | goto finish; | |
4184 | ||
fd69f224 | 4185 | r = process_timer(e, e->timestamp.realtime, &e->realtime); |
a8548816 TG |
4186 | if (r < 0) |
4187 | goto finish; | |
4188 | ||
fd69f224 | 4189 | r = process_timer(e, e->timestamp.boottime, &e->boottime); |
6a0f1f6d LP |
4190 | if (r < 0) |
4191 | goto finish; | |
4192 | ||
4193 | r = process_timer(e, e->timestamp.realtime, &e->realtime_alarm); | |
fd38203a | 4194 | if (r < 0) |
da7e457c | 4195 | goto finish; |
fd38203a | 4196 | |
e475d10c | 4197 | r = process_timer(e, e->timestamp.boottime, &e->boottime_alarm); |
fd38203a | 4198 | if (r < 0) |
da7e457c | 4199 | goto finish; |
fd38203a | 4200 | |
fd69f224 | 4201 | r = process_timer(e, e->timestamp.monotonic, &e->monotonic); |
97ef5391 LP |
4202 | if (r < 0) |
4203 | goto finish; | |
fd69f224 MS |
4204 | else if (r == 1) { |
4205 | /* Ratelimit expiry callback was called. Let's postpone processing pending sources and | |
4206 | * put loop in the initial state in order to evaluate (in the next iteration) also sources | |
4207 | * there were potentially re-enabled by the callback. | |
4208 | * | |
4209 | * Wondering why we treat only this invocation of process_timer() differently? Once event | |
4210 | * source is ratelimited we essentially transform it into CLOCK_MONOTONIC timer hence | |
4211 | * ratelimit expiry callback is never called for any other timer type. */ | |
4212 | r = 0; | |
4213 | goto finish; | |
4214 | } | |
97ef5391 | 4215 | |
c45a5a74 TG |
4216 | if (event_next_pending(e)) { |
4217 | e->state = SD_EVENT_PENDING; | |
c45a5a74 | 4218 | return 1; |
da7e457c LP |
4219 | } |
4220 | ||
c45a5a74 | 4221 | r = 0; |
fd38203a | 4222 | |
da7e457c | 4223 | finish: |
2b0c9ef7 | 4224 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
4225 | |
4226 | return r; | |
fd38203a LP |
4227 | } |
4228 | ||
c45a5a74 TG |
4229 | _public_ int sd_event_dispatch(sd_event *e) { |
4230 | sd_event_source *p; | |
4231 | int r; | |
4232 | ||
4233 | assert_return(e, -EINVAL); | |
b937d761 | 4234 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
4235 | assert_return(!event_pid_changed(e), -ECHILD); |
4236 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
4237 | assert_return(e->state == SD_EVENT_PENDING, -EBUSY); | |
4238 | ||
4239 | if (e->exit_requested) | |
4240 | return dispatch_exit(e); | |
4241 | ||
4242 | p = event_next_pending(e); | |
4243 | if (p) { | |
c8e9d15c | 4244 | PROTECT_EVENT(e); |
c45a5a74 TG |
4245 | |
4246 | e->state = SD_EVENT_RUNNING; | |
4247 | r = source_dispatch(p); | |
2b0c9ef7 | 4248 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
4249 | return r; |
4250 | } | |
4251 | ||
2b0c9ef7 | 4252 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
4253 | |
4254 | return 1; | |
4255 | } | |
4256 | ||
34b87517 | 4257 | static void event_log_delays(sd_event *e) { |
442ac269 YW |
4258 | char b[ELEMENTSOF(e->delays) * DECIMAL_STR_MAX(unsigned) + 1], *p; |
4259 | size_t l, i; | |
34b87517 | 4260 | |
442ac269 YW |
4261 | p = b; |
4262 | l = sizeof(b); | |
4263 | for (i = 0; i < ELEMENTSOF(e->delays); i++) { | |
4264 | l = strpcpyf(&p, l, "%u ", e->delays[i]); | |
34b87517 VC |
4265 | e->delays[i] = 0; |
4266 | } | |
442ac269 | 4267 | log_debug("Event loop iterations: %s", b); |
34b87517 VC |
4268 | } |
4269 | ||
c45a5a74 TG |
4270 | _public_ int sd_event_run(sd_event *e, uint64_t timeout) { |
4271 | int r; | |
4272 | ||
4273 | assert_return(e, -EINVAL); | |
b937d761 | 4274 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
4275 | assert_return(!event_pid_changed(e), -ECHILD); |
4276 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 4277 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
c45a5a74 | 4278 | |
e6a7bee5 | 4279 | if (e->profile_delays && e->last_run_usec != 0) { |
34b87517 VC |
4280 | usec_t this_run; |
4281 | unsigned l; | |
4282 | ||
4283 | this_run = now(CLOCK_MONOTONIC); | |
4284 | ||
58c34be8 | 4285 | l = log2u64(this_run - e->last_run_usec); |
cb9d621e | 4286 | assert(l < ELEMENTSOF(e->delays)); |
34b87517 VC |
4287 | e->delays[l]++; |
4288 | ||
e6a7bee5 | 4289 | if (this_run - e->last_log_usec >= 5*USEC_PER_SEC) { |
34b87517 | 4290 | event_log_delays(e); |
e6a7bee5 | 4291 | e->last_log_usec = this_run; |
34b87517 VC |
4292 | } |
4293 | } | |
4294 | ||
f814c871 | 4295 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
c8e9d15c | 4296 | PROTECT_EVENT(e); |
f814c871 | 4297 | |
c45a5a74 | 4298 | r = sd_event_prepare(e); |
53bac4e0 LP |
4299 | if (r == 0) |
4300 | /* There was nothing? Then wait... */ | |
4301 | r = sd_event_wait(e, timeout); | |
c45a5a74 | 4302 | |
34b87517 | 4303 | if (e->profile_delays) |
e6a7bee5 | 4304 | e->last_run_usec = now(CLOCK_MONOTONIC); |
34b87517 | 4305 | |
02d30981 | 4306 | if (r > 0) { |
53bac4e0 | 4307 | /* There's something now, then let's dispatch it */ |
02d30981 TG |
4308 | r = sd_event_dispatch(e); |
4309 | if (r < 0) | |
4310 | return r; | |
53bac4e0 LP |
4311 | |
4312 | return 1; | |
4313 | } | |
4314 | ||
4315 | return r; | |
c45a5a74 TG |
4316 | } |
4317 | ||
f7262a9f | 4318 | _public_ int sd_event_loop(sd_event *e) { |
fd38203a LP |
4319 | int r; |
4320 | ||
da7e457c | 4321 | assert_return(e, -EINVAL); |
b937d761 | 4322 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 4323 | assert_return(!event_pid_changed(e), -ECHILD); |
2b0c9ef7 | 4324 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 4325 | |
c8e9d15c | 4326 | PROTECT_EVENT(e); |
fd38203a | 4327 | |
da7e457c | 4328 | while (e->state != SD_EVENT_FINISHED) { |
f5fbe71d | 4329 | r = sd_event_run(e, UINT64_MAX); |
fd38203a | 4330 | if (r < 0) |
30dd293c | 4331 | return r; |
fd38203a LP |
4332 | } |
4333 | ||
30dd293c | 4334 | return e->exit_code; |
fd38203a LP |
4335 | } |
4336 | ||
9b364545 | 4337 | _public_ int sd_event_get_fd(sd_event *e) { |
9b364545 | 4338 | assert_return(e, -EINVAL); |
b937d761 | 4339 | assert_return(e = event_resolve(e), -ENOPKG); |
9b364545 TG |
4340 | assert_return(!event_pid_changed(e), -ECHILD); |
4341 | ||
4342 | return e->epoll_fd; | |
4343 | } | |
4344 | ||
f7262a9f | 4345 | _public_ int sd_event_get_state(sd_event *e) { |
da7e457c | 4346 | assert_return(e, -EINVAL); |
b937d761 | 4347 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
4348 | assert_return(!event_pid_changed(e), -ECHILD); |
4349 | ||
4350 | return e->state; | |
4351 | } | |
4352 | ||
6203e07a | 4353 | _public_ int sd_event_get_exit_code(sd_event *e, int *code) { |
da7e457c | 4354 | assert_return(e, -EINVAL); |
b937d761 | 4355 | assert_return(e = event_resolve(e), -ENOPKG); |
6203e07a | 4356 | assert_return(code, -EINVAL); |
da7e457c | 4357 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 4358 | |
6203e07a LP |
4359 | if (!e->exit_requested) |
4360 | return -ENODATA; | |
4361 | ||
4362 | *code = e->exit_code; | |
4363 | return 0; | |
fd38203a LP |
4364 | } |
4365 | ||
6203e07a | 4366 | _public_ int sd_event_exit(sd_event *e, int code) { |
da7e457c | 4367 | assert_return(e, -EINVAL); |
b937d761 | 4368 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
4369 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
4370 | assert_return(!event_pid_changed(e), -ECHILD); | |
fd38203a | 4371 | |
6203e07a LP |
4372 | e->exit_requested = true; |
4373 | e->exit_code = code; | |
4374 | ||
fd38203a LP |
4375 | return 0; |
4376 | } | |
46e8c825 | 4377 | |
6a0f1f6d | 4378 | _public_ int sd_event_now(sd_event *e, clockid_t clock, uint64_t *usec) { |
46e8c825 | 4379 | assert_return(e, -EINVAL); |
b937d761 | 4380 | assert_return(e = event_resolve(e), -ENOPKG); |
46e8c825 | 4381 | assert_return(usec, -EINVAL); |
46e8c825 LP |
4382 | assert_return(!event_pid_changed(e), -ECHILD); |
4383 | ||
e475d10c LP |
4384 | if (!TRIPLE_TIMESTAMP_HAS_CLOCK(clock)) |
4385 | return -EOPNOTSUPP; | |
4386 | ||
e475d10c | 4387 | if (!triple_timestamp_is_set(&e->timestamp)) { |
15c689d7 | 4388 | /* Implicitly fall back to now() if we never ran before and thus have no cached time. */ |
38a03f06 LP |
4389 | *usec = now(clock); |
4390 | return 1; | |
4391 | } | |
46e8c825 | 4392 | |
e475d10c | 4393 | *usec = triple_timestamp_by_clock(&e->timestamp, clock); |
46e8c825 LP |
4394 | return 0; |
4395 | } | |
afc6adb5 LP |
4396 | |
4397 | _public_ int sd_event_default(sd_event **ret) { | |
39883f62 | 4398 | sd_event *e = NULL; |
afc6adb5 LP |
4399 | int r; |
4400 | ||
4401 | if (!ret) | |
4402 | return !!default_event; | |
4403 | ||
4404 | if (default_event) { | |
4405 | *ret = sd_event_ref(default_event); | |
4406 | return 0; | |
4407 | } | |
4408 | ||
4409 | r = sd_event_new(&e); | |
4410 | if (r < 0) | |
4411 | return r; | |
4412 | ||
4413 | e->default_event_ptr = &default_event; | |
4414 | e->tid = gettid(); | |
4415 | default_event = e; | |
4416 | ||
4417 | *ret = e; | |
4418 | return 1; | |
4419 | } | |
4420 | ||
4421 | _public_ int sd_event_get_tid(sd_event *e, pid_t *tid) { | |
4422 | assert_return(e, -EINVAL); | |
b937d761 | 4423 | assert_return(e = event_resolve(e), -ENOPKG); |
afc6adb5 | 4424 | assert_return(tid, -EINVAL); |
76b54375 | 4425 | assert_return(!event_pid_changed(e), -ECHILD); |
afc6adb5 | 4426 | |
76b54375 LP |
4427 | if (e->tid != 0) { |
4428 | *tid = e->tid; | |
4429 | return 0; | |
4430 | } | |
4431 | ||
4432 | return -ENXIO; | |
afc6adb5 | 4433 | } |
cde93897 LP |
4434 | |
4435 | _public_ int sd_event_set_watchdog(sd_event *e, int b) { | |
4436 | int r; | |
4437 | ||
4438 | assert_return(e, -EINVAL); | |
b937d761 | 4439 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 | 4440 | assert_return(!event_pid_changed(e), -ECHILD); |
cde93897 LP |
4441 | |
4442 | if (e->watchdog == !!b) | |
4443 | return e->watchdog; | |
4444 | ||
4445 | if (b) { | |
09812eb7 LP |
4446 | r = sd_watchdog_enabled(false, &e->watchdog_period); |
4447 | if (r <= 0) | |
cde93897 | 4448 | return r; |
cde93897 LP |
4449 | |
4450 | /* Issue first ping immediately */ | |
4451 | sd_notify(false, "WATCHDOG=1"); | |
4452 | e->watchdog_last = now(CLOCK_MONOTONIC); | |
4453 | ||
4454 | e->watchdog_fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC); | |
4455 | if (e->watchdog_fd < 0) | |
4456 | return -errno; | |
4457 | ||
4458 | r = arm_watchdog(e); | |
4459 | if (r < 0) | |
4460 | goto fail; | |
4461 | ||
1eac7948 | 4462 | struct epoll_event ev = { |
a82f89aa LP |
4463 | .events = EPOLLIN, |
4464 | .data.ptr = INT_TO_PTR(SOURCE_WATCHDOG), | |
4465 | }; | |
cde93897 | 4466 | |
15c689d7 | 4467 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, e->watchdog_fd, &ev) < 0) { |
cde93897 LP |
4468 | r = -errno; |
4469 | goto fail; | |
4470 | } | |
4471 | ||
4472 | } else { | |
4473 | if (e->watchdog_fd >= 0) { | |
5a795bff | 4474 | (void) epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, e->watchdog_fd, NULL); |
03e334a1 | 4475 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
4476 | } |
4477 | } | |
4478 | ||
4479 | e->watchdog = !!b; | |
4480 | return e->watchdog; | |
4481 | ||
4482 | fail: | |
03e334a1 | 4483 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
4484 | return r; |
4485 | } | |
8f726607 LP |
4486 | |
4487 | _public_ int sd_event_get_watchdog(sd_event *e) { | |
4488 | assert_return(e, -EINVAL); | |
b937d761 | 4489 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 LP |
4490 | assert_return(!event_pid_changed(e), -ECHILD); |
4491 | ||
4492 | return e->watchdog; | |
4493 | } | |
60a3b1e1 LP |
4494 | |
4495 | _public_ int sd_event_get_iteration(sd_event *e, uint64_t *ret) { | |
4496 | assert_return(e, -EINVAL); | |
b937d761 | 4497 | assert_return(e = event_resolve(e), -ENOPKG); |
60a3b1e1 LP |
4498 | assert_return(!event_pid_changed(e), -ECHILD); |
4499 | ||
4500 | *ret = e->iteration; | |
4501 | return 0; | |
4502 | } | |
15723a1d LP |
4503 | |
4504 | _public_ int sd_event_source_set_destroy_callback(sd_event_source *s, sd_event_destroy_t callback) { | |
4505 | assert_return(s, -EINVAL); | |
4506 | ||
4507 | s->destroy_callback = callback; | |
4508 | return 0; | |
4509 | } | |
4510 | ||
4511 | _public_ int sd_event_source_get_destroy_callback(sd_event_source *s, sd_event_destroy_t *ret) { | |
4512 | assert_return(s, -EINVAL); | |
4513 | ||
4514 | if (ret) | |
4515 | *ret = s->destroy_callback; | |
4516 | ||
4517 | return !!s->destroy_callback; | |
4518 | } | |
2382c936 YW |
4519 | |
4520 | _public_ int sd_event_source_get_floating(sd_event_source *s) { | |
4521 | assert_return(s, -EINVAL); | |
4522 | ||
4523 | return s->floating; | |
4524 | } | |
4525 | ||
4526 | _public_ int sd_event_source_set_floating(sd_event_source *s, int b) { | |
4527 | assert_return(s, -EINVAL); | |
4528 | ||
4529 | if (s->floating == !!b) | |
4530 | return 0; | |
4531 | ||
4532 | if (!s->event) /* Already disconnected */ | |
4533 | return -ESTALE; | |
4534 | ||
4535 | s->floating = b; | |
4536 | ||
4537 | if (b) { | |
4538 | sd_event_source_ref(s); | |
4539 | sd_event_unref(s->event); | |
4540 | } else { | |
4541 | sd_event_ref(s->event); | |
4542 | sd_event_source_unref(s); | |
4543 | } | |
4544 | ||
4545 | return 1; | |
4546 | } | |
b778cba4 LP |
4547 | |
4548 | _public_ int sd_event_source_get_exit_on_failure(sd_event_source *s) { | |
4549 | assert_return(s, -EINVAL); | |
4550 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
4551 | ||
4552 | return s->exit_on_failure; | |
4553 | } | |
4554 | ||
4555 | _public_ int sd_event_source_set_exit_on_failure(sd_event_source *s, int b) { | |
4556 | assert_return(s, -EINVAL); | |
4557 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
4558 | ||
4559 | if (s->exit_on_failure == !!b) | |
4560 | return 0; | |
4561 | ||
4562 | s->exit_on_failure = b; | |
4563 | return 1; | |
4564 | } | |
b6d5481b LP |
4565 | |
4566 | _public_ int sd_event_source_set_ratelimit(sd_event_source *s, uint64_t interval, unsigned burst) { | |
4567 | int r; | |
4568 | ||
4569 | assert_return(s, -EINVAL); | |
4570 | ||
4571 | /* Turning on ratelimiting on event source types that don't support it, is a loggable offense. Doing | |
4572 | * so is a programming error. */ | |
4573 | assert_return(EVENT_SOURCE_CAN_RATE_LIMIT(s->type), -EDOM); | |
4574 | ||
4575 | /* When ratelimiting is configured we'll always reset the rate limit state first and start fresh, | |
4576 | * non-ratelimited. */ | |
fd69f224 | 4577 | r = event_source_leave_ratelimit(s, /* run_callback */ false); |
b6d5481b LP |
4578 | if (r < 0) |
4579 | return r; | |
4580 | ||
4581 | s->rate_limit = (RateLimit) { interval, burst }; | |
4582 | return 0; | |
fd69f224 MS |
4583 | } |
4584 | ||
4585 | _public_ int sd_event_source_set_ratelimit_expire_callback(sd_event_source *s, sd_event_handler_t callback) { | |
4586 | assert_return(s, -EINVAL); | |
4587 | ||
4588 | s->ratelimit_expire_callback = callback; | |
4589 | return 0; | |
b6d5481b LP |
4590 | } |
4591 | ||
4592 | _public_ int sd_event_source_get_ratelimit(sd_event_source *s, uint64_t *ret_interval, unsigned *ret_burst) { | |
4593 | assert_return(s, -EINVAL); | |
4594 | ||
6dd3b818 YW |
4595 | /* Querying whether an event source has ratelimiting configured is not a loggable offense, hence |
4596 | * don't use assert_return(). Unlike turning on ratelimiting it's not really a programming error. */ | |
b6d5481b LP |
4597 | if (!EVENT_SOURCE_CAN_RATE_LIMIT(s->type)) |
4598 | return -EDOM; | |
4599 | ||
4600 | if (!ratelimit_configured(&s->rate_limit)) | |
4601 | return -ENOEXEC; | |
4602 | ||
4603 | if (ret_interval) | |
4604 | *ret_interval = s->rate_limit.interval; | |
4605 | if (ret_burst) | |
4606 | *ret_burst = s->rate_limit.burst; | |
4607 | ||
4608 | return 0; | |
4609 | } | |
4610 | ||
4611 | _public_ int sd_event_source_is_ratelimited(sd_event_source *s) { | |
4612 | assert_return(s, -EINVAL); | |
4613 | ||
4614 | if (!EVENT_SOURCE_CAN_RATE_LIMIT(s->type)) | |
4615 | return false; | |
4616 | ||
4617 | if (!ratelimit_configured(&s->rate_limit)) | |
4618 | return false; | |
4619 | ||
4620 | return s->ratelimited; | |
4621 | } | |
baf3fdec LP |
4622 | |
4623 | _public_ int sd_event_set_signal_exit(sd_event *e, int b) { | |
4624 | bool change = false; | |
4625 | int r; | |
4626 | ||
4627 | assert_return(e, -EINVAL); | |
4628 | ||
4629 | if (b) { | |
4630 | /* We want to maintain pointers to these event sources, so that we can destroy them when told | |
4631 | * so. But we also don't want them to pin the event loop itself. Hence we mark them as | |
4632 | * floating after creation (and undo this before deleting them again). */ | |
4633 | ||
4634 | if (!e->sigint_event_source) { | |
4635 | r = sd_event_add_signal(e, &e->sigint_event_source, SIGINT | SD_EVENT_SIGNAL_PROCMASK, NULL, NULL); | |
4636 | if (r < 0) | |
4637 | return r; | |
4638 | ||
4639 | assert(sd_event_source_set_floating(e->sigint_event_source, true) >= 0); | |
4640 | change = true; | |
4641 | } | |
4642 | ||
4643 | if (!e->sigterm_event_source) { | |
4644 | r = sd_event_add_signal(e, &e->sigterm_event_source, SIGTERM | SD_EVENT_SIGNAL_PROCMASK, NULL, NULL); | |
4645 | if (r < 0) { | |
4646 | if (change) { | |
4647 | assert(sd_event_source_set_floating(e->sigint_event_source, false) >= 0); | |
4648 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
4649 | } | |
4650 | ||
4651 | return r; | |
4652 | } | |
4653 | ||
4654 | assert(sd_event_source_set_floating(e->sigterm_event_source, true) >= 0); | |
4655 | change = true; | |
4656 | } | |
4657 | ||
4658 | } else { | |
4659 | if (e->sigint_event_source) { | |
4660 | assert(sd_event_source_set_floating(e->sigint_event_source, false) >= 0); | |
4661 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
4662 | change = true; | |
4663 | } | |
4664 | ||
4665 | if (e->sigterm_event_source) { | |
4666 | assert(sd_event_source_set_floating(e->sigterm_event_source, false) >= 0); | |
4667 | e->sigterm_event_source = sd_event_source_unref(e->sigterm_event_source); | |
4668 | change = true; | |
4669 | } | |
4670 | } | |
4671 | ||
4672 | return change; | |
4673 | } |