]>
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, | |
254d1313 ZJS |
381 | .epoll_fd = -EBADF, |
382 | .watchdog_fd = -EBADF, | |
d08eb1fa | 383 | .realtime.wakeup = WAKEUP_CLOCK_DATA, |
254d1313 | 384 | .realtime.fd = -EBADF, |
d08eb1fa LP |
385 | .realtime.next = USEC_INFINITY, |
386 | .boottime.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 387 | .boottime.fd = -EBADF, |
d08eb1fa LP |
388 | .boottime.next = USEC_INFINITY, |
389 | .monotonic.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 390 | .monotonic.fd = -EBADF, |
d08eb1fa LP |
391 | .monotonic.next = USEC_INFINITY, |
392 | .realtime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 393 | .realtime_alarm.fd = -EBADF, |
d08eb1fa LP |
394 | .realtime_alarm.next = USEC_INFINITY, |
395 | .boottime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 396 | .boottime_alarm.fd = -EBADF, |
d08eb1fa LP |
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, | |
254d1313 | 646 | .fd = -EBADF, |
d08eb1fa LP |
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 | ||
cbff793f ZJS |
662 | r = signalfd(d->fd >= 0 ? d->fd : -1, /* the first arg must be -1 or a valid signalfd */ |
663 | &ss_copy, | |
664 | SFD_NONBLOCK|SFD_CLOEXEC); | |
9da4cb2b LP |
665 | if (r < 0) { |
666 | r = -errno; | |
667 | goto fail; | |
668 | } | |
669 | ||
670 | d->sigset = ss_copy; | |
f95387cd | 671 | |
9da4cb2b LP |
672 | if (d->fd >= 0) { |
673 | if (ret) | |
674 | *ret = d; | |
f95387cd | 675 | return 0; |
9da4cb2b LP |
676 | } |
677 | ||
7fe2903c | 678 | d->fd = fd_move_above_stdio(r); |
f95387cd | 679 | |
1eac7948 | 680 | struct epoll_event ev = { |
a82f89aa LP |
681 | .events = EPOLLIN, |
682 | .data.ptr = d, | |
683 | }; | |
f95387cd | 684 | |
15c689d7 | 685 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { |
9da4cb2b LP |
686 | r = -errno; |
687 | goto fail; | |
f95387cd ZJS |
688 | } |
689 | ||
9da4cb2b LP |
690 | if (ret) |
691 | *ret = d; | |
692 | ||
f95387cd | 693 | return 0; |
9da4cb2b LP |
694 | |
695 | fail: | |
3e4eb8e7 YW |
696 | if (added) |
697 | event_free_signal_data(e, d); | |
9da4cb2b LP |
698 | |
699 | return r; | |
700 | } | |
701 | ||
702 | static void event_unmask_signal_data(sd_event *e, struct signal_data *d, int sig) { | |
703 | assert(e); | |
704 | assert(d); | |
705 | ||
706 | /* Turns off the specified signal in the signal data | |
707 | * object. If the signal mask of the object becomes empty that | |
708 | * way removes it. */ | |
709 | ||
710 | if (sigismember(&d->sigset, sig) == 0) | |
711 | return; | |
712 | ||
713 | assert_se(sigdelset(&d->sigset, sig) >= 0); | |
714 | ||
715 | if (sigisemptyset(&d->sigset)) { | |
9da4cb2b | 716 | /* If all the mask is all-zero we can get rid of the structure */ |
3e4eb8e7 | 717 | event_free_signal_data(e, d); |
9da4cb2b LP |
718 | return; |
719 | } | |
720 | ||
01e6af73 YW |
721 | if (event_pid_changed(e)) |
722 | return; | |
723 | ||
9da4cb2b LP |
724 | assert(d->fd >= 0); |
725 | ||
726 | if (signalfd(d->fd, &d->sigset, SFD_NONBLOCK|SFD_CLOEXEC) < 0) | |
727 | log_debug_errno(errno, "Failed to unset signal bit, ignoring: %m"); | |
728 | } | |
729 | ||
730 | static void event_gc_signal_data(sd_event *e, const int64_t *priority, int sig) { | |
731 | struct signal_data *d; | |
732 | static const int64_t zero_priority = 0; | |
733 | ||
734 | assert(e); | |
735 | ||
f8f3f926 LP |
736 | /* Rechecks if the specified signal is still something we are interested in. If not, we'll unmask it, |
737 | * and possibly drop the signalfd for it. */ | |
9da4cb2b LP |
738 | |
739 | if (sig == SIGCHLD && | |
b6d5481b | 740 | e->n_online_child_sources > 0) |
9da4cb2b LP |
741 | return; |
742 | ||
743 | if (e->signal_sources && | |
744 | e->signal_sources[sig] && | |
b6d5481b | 745 | event_source_is_online(e->signal_sources[sig])) |
9da4cb2b LP |
746 | return; |
747 | ||
748 | /* | |
749 | * The specified signal might be enabled in three different queues: | |
750 | * | |
751 | * 1) the one that belongs to the priority passed (if it is non-NULL) | |
752 | * 2) the one that belongs to the priority of the event source of the signal (if there is one) | |
753 | * 3) the 0 priority (to cover the SIGCHLD case) | |
754 | * | |
755 | * Hence, let's remove it from all three here. | |
756 | */ | |
757 | ||
758 | if (priority) { | |
759 | d = hashmap_get(e->signal_data, priority); | |
760 | if (d) | |
761 | event_unmask_signal_data(e, d, sig); | |
762 | } | |
763 | ||
764 | if (e->signal_sources && e->signal_sources[sig]) { | |
765 | d = hashmap_get(e->signal_data, &e->signal_sources[sig]->priority); | |
766 | if (d) | |
767 | event_unmask_signal_data(e, d, sig); | |
768 | } | |
769 | ||
770 | d = hashmap_get(e->signal_data, &zero_priority); | |
771 | if (d) | |
772 | event_unmask_signal_data(e, d, sig); | |
f95387cd ZJS |
773 | } |
774 | ||
e1951c16 MS |
775 | static void event_source_pp_prioq_reshuffle(sd_event_source *s) { |
776 | assert(s); | |
777 | ||
778 | /* Reshuffles the pending + prepare prioqs. Called whenever the dispatch order changes, i.e. when | |
779 | * they are enabled/disabled or marked pending and such. */ | |
780 | ||
781 | if (s->pending) | |
782 | prioq_reshuffle(s->event->pending, s, &s->pending_index); | |
783 | ||
784 | if (s->prepare) | |
785 | prioq_reshuffle(s->event->prepare, s, &s->prepare_index); | |
786 | } | |
787 | ||
788 | static void event_source_time_prioq_reshuffle(sd_event_source *s) { | |
789 | struct clock_data *d; | |
790 | ||
791 | assert(s); | |
e1951c16 MS |
792 | |
793 | /* Called whenever the event source's timer ordering properties changed, i.e. time, accuracy, | |
5c08c7ab YW |
794 | * pending, enable state, and ratelimiting state. Makes sure the two prioq's are ordered |
795 | * properly again. */ | |
b6d5481b LP |
796 | |
797 | if (s->ratelimited) | |
798 | d = &s->event->monotonic; | |
5c08c7ab | 799 | else if (EVENT_SOURCE_IS_TIME(s->type)) |
b6d5481b | 800 | assert_se(d = event_get_clock_data(s->event, s->type)); |
5c08c7ab YW |
801 | else |
802 | return; /* no-op for an event source which is neither a timer nor ratelimited. */ | |
b6d5481b | 803 | |
f41315fc LP |
804 | prioq_reshuffle(d->earliest, s, &s->earliest_index); |
805 | prioq_reshuffle(d->latest, s, &s->latest_index); | |
e1951c16 MS |
806 | d->needs_rearm = true; |
807 | } | |
808 | ||
1e45e3fe LP |
809 | static void event_source_time_prioq_remove( |
810 | sd_event_source *s, | |
811 | struct clock_data *d) { | |
812 | ||
813 | assert(s); | |
814 | assert(d); | |
815 | ||
f41315fc LP |
816 | prioq_remove(d->earliest, s, &s->earliest_index); |
817 | prioq_remove(d->latest, s, &s->latest_index); | |
818 | s->earliest_index = s->latest_index = PRIOQ_IDX_NULL; | |
1e45e3fe LP |
819 | d->needs_rearm = true; |
820 | } | |
821 | ||
a71fe8b8 LP |
822 | static void source_disconnect(sd_event_source *s) { |
823 | sd_event *event; | |
897448bd | 824 | int r; |
a71fe8b8 | 825 | |
fd38203a LP |
826 | assert(s); |
827 | ||
a71fe8b8 LP |
828 | if (!s->event) |
829 | return; | |
15b38f93 | 830 | |
a71fe8b8 | 831 | assert(s->event->n_sources > 0); |
fd38203a | 832 | |
a71fe8b8 | 833 | switch (s->type) { |
fd38203a | 834 | |
a71fe8b8 LP |
835 | case SOURCE_IO: |
836 | if (s->io.fd >= 0) | |
837 | source_io_unregister(s); | |
fd38203a | 838 | |
a71fe8b8 | 839 | break; |
6a0f1f6d | 840 | |
a71fe8b8 | 841 | case SOURCE_TIME_REALTIME: |
a8548816 | 842 | case SOURCE_TIME_BOOTTIME: |
a71fe8b8 LP |
843 | case SOURCE_TIME_MONOTONIC: |
844 | case SOURCE_TIME_REALTIME_ALARM: | |
b6d5481b LP |
845 | case SOURCE_TIME_BOOTTIME_ALARM: |
846 | /* Only remove this event source from the time event source here if it is not ratelimited. If | |
847 | * it is ratelimited, we'll remove it below, separately. Why? Because the clock used might | |
848 | * differ: ratelimiting always uses CLOCK_MONOTONIC, but timer events might use any clock */ | |
849 | ||
850 | if (!s->ratelimited) { | |
851 | struct clock_data *d; | |
852 | assert_se(d = event_get_clock_data(s->event, s->type)); | |
853 | event_source_time_prioq_remove(s, d); | |
854 | } | |
855 | ||
a71fe8b8 | 856 | break; |
a71fe8b8 LP |
857 | |
858 | case SOURCE_SIGNAL: | |
859 | if (s->signal.sig > 0) { | |
9da4cb2b | 860 | |
a71fe8b8 LP |
861 | if (s->event->signal_sources) |
862 | s->event->signal_sources[s->signal.sig] = NULL; | |
4807d2d0 | 863 | |
9da4cb2b | 864 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
897448bd LP |
865 | |
866 | if (s->signal.unblock) { | |
867 | sigset_t new_ss; | |
868 | ||
869 | if (sigemptyset(&new_ss) < 0) | |
870 | log_debug_errno(errno, "Failed to reset signal set, ignoring: %m"); | |
871 | else if (sigaddset(&new_ss, s->signal.sig) < 0) | |
872 | log_debug_errno(errno, "Failed to add signal %i to signal mask, ignoring: %m", s->signal.sig); | |
873 | else { | |
874 | r = pthread_sigmask(SIG_UNBLOCK, &new_ss, NULL); | |
875 | if (r != 0) | |
876 | log_debug_errno(r, "Failed to unblock signal %i, ignoring: %m", s->signal.sig); | |
877 | } | |
878 | } | |
6a0f1f6d | 879 | } |
fd38203a | 880 | |
a71fe8b8 | 881 | break; |
fd38203a | 882 | |
a71fe8b8 | 883 | case SOURCE_CHILD: |
86587c93 YW |
884 | if (event_pid_changed(s->event)) |
885 | s->child.process_owned = false; | |
886 | ||
a71fe8b8 | 887 | if (s->child.pid > 0) { |
b6d5481b LP |
888 | if (event_source_is_online(s)) { |
889 | assert(s->event->n_online_child_sources > 0); | |
890 | s->event->n_online_child_sources--; | |
4807d2d0 | 891 | } |
fd38203a | 892 | |
4a0b58c4 | 893 | (void) hashmap_remove(s->event->child_sources, PID_TO_PTR(s->child.pid)); |
a71fe8b8 | 894 | } |
fd38203a | 895 | |
f8f3f926 LP |
896 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
897 | source_child_pidfd_unregister(s); | |
898 | else | |
899 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
900 | ||
a71fe8b8 | 901 | break; |
fd38203a | 902 | |
a71fe8b8 LP |
903 | case SOURCE_DEFER: |
904 | /* nothing */ | |
905 | break; | |
fd38203a | 906 | |
a71fe8b8 LP |
907 | case SOURCE_POST: |
908 | set_remove(s->event->post_sources, s); | |
909 | break; | |
da7e457c | 910 | |
a71fe8b8 LP |
911 | case SOURCE_EXIT: |
912 | prioq_remove(s->event->exit, s, &s->exit.prioq_index); | |
913 | break; | |
0eb2e0e3 | 914 | |
97ef5391 LP |
915 | case SOURCE_INOTIFY: { |
916 | struct inode_data *inode_data; | |
917 | ||
918 | inode_data = s->inotify.inode_data; | |
919 | if (inode_data) { | |
920 | struct inotify_data *inotify_data; | |
921 | assert_se(inotify_data = inode_data->inotify_data); | |
922 | ||
923 | /* Detach this event source from the inode object */ | |
924 | LIST_REMOVE(inotify.by_inode_data, inode_data->event_sources, s); | |
925 | s->inotify.inode_data = NULL; | |
926 | ||
927 | if (s->pending) { | |
928 | assert(inotify_data->n_pending > 0); | |
929 | inotify_data->n_pending--; | |
930 | } | |
931 | ||
932 | /* Note that we don't reduce the inotify mask for the watch descriptor here if the inode is | |
933 | * continued to being watched. That's because inotify doesn't really have an API for that: we | |
934 | * can only change watch masks with access to the original inode either by fd or by path. But | |
935 | * paths aren't stable, and keeping an O_PATH fd open all the time would mean wasting an fd | |
f21f31b2 | 936 | * continuously and keeping the mount busy which we can't really do. We could reconstruct the |
97ef5391 LP |
937 | * original inode from /proc/self/fdinfo/$INOTIFY_FD (as all watch descriptors are listed |
938 | * there), but given the need for open_by_handle_at() which is privileged and not universally | |
939 | * available this would be quite an incomplete solution. Hence we go the other way, leave the | |
940 | * mask set, even if it is not minimized now, and ignore all events we aren't interested in | |
941 | * anymore after reception. Yes, this sucks, but … Linux … */ | |
942 | ||
943 | /* Maybe release the inode data (and its inotify) */ | |
944 | event_gc_inode_data(s->event, inode_data); | |
945 | } | |
946 | ||
947 | break; | |
948 | } | |
949 | ||
a71fe8b8 | 950 | default: |
04499a70 | 951 | assert_not_reached(); |
a71fe8b8 | 952 | } |
6e9feda3 | 953 | |
a71fe8b8 LP |
954 | if (s->pending) |
955 | prioq_remove(s->event->pending, s, &s->pending_index); | |
9d3e3aa5 | 956 | |
a71fe8b8 LP |
957 | if (s->prepare) |
958 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
fd38203a | 959 | |
b6d5481b LP |
960 | if (s->ratelimited) |
961 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
962 | ||
e514aa1e | 963 | event = TAKE_PTR(s->event); |
a71fe8b8 LP |
964 | LIST_REMOVE(sources, event->sources, s); |
965 | event->n_sources--; | |
fd38203a | 966 | |
f5982559 LP |
967 | /* Note that we don't invalidate the type here, since we still need it in order to close the fd or |
968 | * pidfd associated with this event source, which we'll do only on source_free(). */ | |
969 | ||
a71fe8b8 LP |
970 | if (!s->floating) |
971 | sd_event_unref(event); | |
972 | } | |
973 | ||
75db809a | 974 | static sd_event_source* source_free(sd_event_source *s) { |
a71fe8b8 | 975 | assert(s); |
fd38203a | 976 | |
a71fe8b8 | 977 | source_disconnect(s); |
ab93297c NM |
978 | |
979 | if (s->type == SOURCE_IO && s->io.owned) | |
15723a1d LP |
980 | s->io.fd = safe_close(s->io.fd); |
981 | ||
f8f3f926 LP |
982 | if (s->type == SOURCE_CHILD) { |
983 | /* Eventually the kernel will do this automatically for us, but for now let's emulate this (unreliably) in userspace. */ | |
984 | ||
985 | if (s->child.process_owned) { | |
986 | ||
987 | if (!s->child.exited) { | |
988 | bool sent = false; | |
989 | ||
990 | if (s->child.pidfd >= 0) { | |
991 | if (pidfd_send_signal(s->child.pidfd, SIGKILL, NULL, 0) < 0) { | |
992 | if (errno == ESRCH) /* Already dead */ | |
993 | sent = true; | |
994 | else if (!ERRNO_IS_NOT_SUPPORTED(errno)) | |
995 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via pidfd_send_signal(), re-trying via kill(): %m", | |
996 | s->child.pid); | |
997 | } else | |
998 | sent = true; | |
999 | } | |
1000 | ||
1001 | if (!sent) | |
1002 | if (kill(s->child.pid, SIGKILL) < 0) | |
1003 | if (errno != ESRCH) /* Already dead */ | |
1004 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via kill(), ignoring: %m", | |
1005 | s->child.pid); | |
1006 | } | |
1007 | ||
1008 | if (!s->child.waited) { | |
1009 | siginfo_t si = {}; | |
1010 | ||
1011 | /* Reap the child if we can */ | |
1012 | (void) waitid(P_PID, s->child.pid, &si, WEXITED); | |
1013 | } | |
1014 | } | |
1015 | ||
1016 | if (s->child.pidfd_owned) | |
1017 | s->child.pidfd = safe_close(s->child.pidfd); | |
1018 | } | |
1019 | ||
15723a1d LP |
1020 | if (s->destroy_callback) |
1021 | s->destroy_callback(s->userdata); | |
ab93297c | 1022 | |
356779df | 1023 | free(s->description); |
75db809a | 1024 | return mfree(s); |
fd38203a | 1025 | } |
8c75fe17 | 1026 | DEFINE_TRIVIAL_CLEANUP_FUNC(sd_event_source*, source_free); |
fd38203a LP |
1027 | |
1028 | static int source_set_pending(sd_event_source *s, bool b) { | |
1029 | int r; | |
1030 | ||
1031 | assert(s); | |
6203e07a | 1032 | assert(s->type != SOURCE_EXIT); |
fd38203a LP |
1033 | |
1034 | if (s->pending == b) | |
1035 | return 0; | |
1036 | ||
1037 | s->pending = b; | |
1038 | ||
1039 | if (b) { | |
1040 | s->pending_iteration = s->event->iteration; | |
1041 | ||
1042 | r = prioq_put(s->event->pending, s, &s->pending_index); | |
1043 | if (r < 0) { | |
1044 | s->pending = false; | |
1045 | return r; | |
1046 | } | |
1047 | } else | |
1048 | assert_se(prioq_remove(s->event->pending, s, &s->pending_index)); | |
1049 | ||
e1951c16 MS |
1050 | if (EVENT_SOURCE_IS_TIME(s->type)) |
1051 | event_source_time_prioq_reshuffle(s); | |
2576a19e | 1052 | |
9da4cb2b LP |
1053 | if (s->type == SOURCE_SIGNAL && !b) { |
1054 | struct signal_data *d; | |
1055 | ||
1056 | d = hashmap_get(s->event->signal_data, &s->priority); | |
1057 | if (d && d->current == s) | |
1058 | d->current = NULL; | |
1059 | } | |
1060 | ||
97ef5391 LP |
1061 | if (s->type == SOURCE_INOTIFY) { |
1062 | ||
1063 | assert(s->inotify.inode_data); | |
1064 | assert(s->inotify.inode_data->inotify_data); | |
1065 | ||
1066 | if (b) | |
1067 | s->inotify.inode_data->inotify_data->n_pending ++; | |
1068 | else { | |
1069 | assert(s->inotify.inode_data->inotify_data->n_pending > 0); | |
1070 | s->inotify.inode_data->inotify_data->n_pending --; | |
1071 | } | |
1072 | } | |
1073 | ||
efd3be9d | 1074 | return 1; |
fd38203a LP |
1075 | } |
1076 | ||
a71fe8b8 | 1077 | static sd_event_source *source_new(sd_event *e, bool floating, EventSourceType type) { |
a38cf9fb LP |
1078 | |
1079 | /* Let's allocate exactly what we need. Note that the difference of the smallest event source | |
1080 | * structure to the largest is 144 bytes on x86-64 at the time of writing, i.e. more than two cache | |
1081 | * lines. */ | |
1082 | static const size_t size_table[_SOURCE_EVENT_SOURCE_TYPE_MAX] = { | |
1083 | [SOURCE_IO] = endoffsetof_field(sd_event_source, io), | |
1084 | [SOURCE_TIME_REALTIME] = endoffsetof_field(sd_event_source, time), | |
1085 | [SOURCE_TIME_BOOTTIME] = endoffsetof_field(sd_event_source, time), | |
1086 | [SOURCE_TIME_MONOTONIC] = endoffsetof_field(sd_event_source, time), | |
1087 | [SOURCE_TIME_REALTIME_ALARM] = endoffsetof_field(sd_event_source, time), | |
1088 | [SOURCE_TIME_BOOTTIME_ALARM] = endoffsetof_field(sd_event_source, time), | |
1089 | [SOURCE_SIGNAL] = endoffsetof_field(sd_event_source, signal), | |
1090 | [SOURCE_CHILD] = endoffsetof_field(sd_event_source, child), | |
1091 | [SOURCE_DEFER] = endoffsetof_field(sd_event_source, defer), | |
1092 | [SOURCE_POST] = endoffsetof_field(sd_event_source, post), | |
1093 | [SOURCE_EXIT] = endoffsetof_field(sd_event_source, exit), | |
1094 | [SOURCE_INOTIFY] = endoffsetof_field(sd_event_source, inotify), | |
1095 | }; | |
1096 | ||
fd38203a LP |
1097 | sd_event_source *s; |
1098 | ||
1099 | assert(e); | |
a38cf9fb LP |
1100 | assert(type >= 0); |
1101 | assert(type < _SOURCE_EVENT_SOURCE_TYPE_MAX); | |
1102 | assert(size_table[type] > 0); | |
fd38203a | 1103 | |
a38cf9fb LP |
1104 | /* We use expand_to_usable() here to tell gcc that it should consider this an object of the full |
1105 | * size, even if we only allocate the initial part we need. */ | |
1106 | s = expand_to_usable(malloc0(size_table[type]), sizeof(sd_event_source)); | |
fd38203a LP |
1107 | if (!s) |
1108 | return NULL; | |
1109 | ||
a38cf9fb LP |
1110 | /* Note: we cannot use compound initialization here, because sizeof(sd_event_source) is likely larger |
1111 | * than what we allocated here. */ | |
1112 | s->n_ref = 1; | |
1113 | s->event = e; | |
1114 | s->floating = floating; | |
1115 | s->type = type; | |
1116 | s->pending_index = PRIOQ_IDX_NULL; | |
1117 | s->prepare_index = PRIOQ_IDX_NULL; | |
a71fe8b8 LP |
1118 | |
1119 | if (!floating) | |
1120 | sd_event_ref(e); | |
fd38203a | 1121 | |
a71fe8b8 | 1122 | LIST_PREPEND(sources, e->sources, s); |
313cefa1 | 1123 | e->n_sources++; |
15b38f93 | 1124 | |
fd38203a LP |
1125 | return s; |
1126 | } | |
1127 | ||
b9350e70 LP |
1128 | static int io_exit_callback(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
1129 | assert(s); | |
1130 | ||
1131 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1132 | } | |
1133 | ||
f7262a9f | 1134 | _public_ int sd_event_add_io( |
fd38203a | 1135 | sd_event *e, |
151b9b96 | 1136 | sd_event_source **ret, |
fd38203a LP |
1137 | int fd, |
1138 | uint32_t events, | |
718db961 | 1139 | sd_event_io_handler_t callback, |
151b9b96 | 1140 | void *userdata) { |
fd38203a | 1141 | |
ec766a51 | 1142 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1143 | int r; |
1144 | ||
305f78bf | 1145 | assert_return(e, -EINVAL); |
b937d761 | 1146 | assert_return(e = event_resolve(e), -ENOPKG); |
8ac43fee | 1147 | assert_return(fd >= 0, -EBADF); |
2a16a986 | 1148 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 1149 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1150 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1151 | |
b9350e70 LP |
1152 | if (!callback) |
1153 | callback = io_exit_callback; | |
1154 | ||
a71fe8b8 | 1155 | s = source_new(e, !ret, SOURCE_IO); |
fd38203a LP |
1156 | if (!s) |
1157 | return -ENOMEM; | |
1158 | ||
9da4cb2b | 1159 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1160 | s->io.fd = fd; |
1161 | s->io.events = events; | |
1162 | s->io.callback = callback; | |
1163 | s->userdata = userdata; | |
baf76283 | 1164 | s->enabled = SD_EVENT_ON; |
fd38203a | 1165 | |
baf76283 | 1166 | r = source_io_register(s, s->enabled, events); |
ec766a51 | 1167 | if (r < 0) |
050f74f2 | 1168 | return r; |
fd38203a | 1169 | |
a71fe8b8 LP |
1170 | if (ret) |
1171 | *ret = s; | |
ec766a51 | 1172 | TAKE_PTR(s); |
a71fe8b8 | 1173 | |
fd38203a LP |
1174 | return 0; |
1175 | } | |
1176 | ||
52444dc4 LP |
1177 | static void initialize_perturb(sd_event *e) { |
1178 | sd_id128_t bootid = {}; | |
1179 | ||
1180 | /* When we sleep for longer, we try to realign the wakeup to | |
f21f31b2 | 1181 | the same time within each minute/second/250ms, so that |
52444dc4 LP |
1182 | events all across the system can be coalesced into a single |
1183 | CPU wakeup. However, let's take some system-specific | |
1184 | randomness for this value, so that in a network of systems | |
1185 | with synced clocks timer events are distributed a | |
1186 | bit. Here, we calculate a perturbation usec offset from the | |
1187 | boot ID. */ | |
1188 | ||
3a43da28 | 1189 | if (_likely_(e->perturb != USEC_INFINITY)) |
52444dc4 LP |
1190 | return; |
1191 | ||
1192 | if (sd_id128_get_boot(&bootid) >= 0) | |
1193 | e->perturb = (bootid.qwords[0] ^ bootid.qwords[1]) % USEC_PER_MINUTE; | |
1194 | } | |
1195 | ||
fd38203a LP |
1196 | static int event_setup_timer_fd( |
1197 | sd_event *e, | |
6a0f1f6d LP |
1198 | struct clock_data *d, |
1199 | clockid_t clock) { | |
fd38203a | 1200 | |
fd38203a | 1201 | assert(e); |
6a0f1f6d | 1202 | assert(d); |
fd38203a | 1203 | |
6a0f1f6d | 1204 | if (_likely_(d->fd >= 0)) |
fd38203a LP |
1205 | return 0; |
1206 | ||
254d1313 | 1207 | _cleanup_close_ int fd = -EBADF; |
b44d87e2 | 1208 | |
6a0f1f6d | 1209 | fd = timerfd_create(clock, TFD_NONBLOCK|TFD_CLOEXEC); |
fd38203a LP |
1210 | if (fd < 0) |
1211 | return -errno; | |
1212 | ||
7fe2903c LP |
1213 | fd = fd_move_above_stdio(fd); |
1214 | ||
1eac7948 | 1215 | struct epoll_event ev = { |
a82f89aa LP |
1216 | .events = EPOLLIN, |
1217 | .data.ptr = d, | |
1218 | }; | |
fd38203a | 1219 | |
15c689d7 | 1220 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) |
fd38203a | 1221 | return -errno; |
fd38203a | 1222 | |
b44d87e2 | 1223 | d->fd = TAKE_FD(fd); |
fd38203a LP |
1224 | return 0; |
1225 | } | |
1226 | ||
c4f1aff2 TG |
1227 | static int time_exit_callback(sd_event_source *s, uint64_t usec, void *userdata) { |
1228 | assert(s); | |
1229 | ||
1230 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1231 | } | |
1232 | ||
41c63f36 LP |
1233 | static int setup_clock_data(sd_event *e, struct clock_data *d, clockid_t clock) { |
1234 | int r; | |
1235 | ||
1236 | assert(d); | |
1237 | ||
1238 | if (d->fd < 0) { | |
1239 | r = event_setup_timer_fd(e, d, clock); | |
1240 | if (r < 0) | |
1241 | return r; | |
1242 | } | |
1243 | ||
1244 | r = prioq_ensure_allocated(&d->earliest, earliest_time_prioq_compare); | |
1245 | if (r < 0) | |
1246 | return r; | |
1247 | ||
1248 | r = prioq_ensure_allocated(&d->latest, latest_time_prioq_compare); | |
1249 | if (r < 0) | |
1250 | return r; | |
1251 | ||
1252 | return 0; | |
1253 | } | |
1254 | ||
1e45e3fe LP |
1255 | static int event_source_time_prioq_put( |
1256 | sd_event_source *s, | |
1257 | struct clock_data *d) { | |
1258 | ||
1259 | int r; | |
1260 | ||
1261 | assert(s); | |
1262 | assert(d); | |
19947509 | 1263 | assert(EVENT_SOURCE_USES_TIME_PRIOQ(s->type)); |
1e45e3fe | 1264 | |
f41315fc | 1265 | r = prioq_put(d->earliest, s, &s->earliest_index); |
1e45e3fe LP |
1266 | if (r < 0) |
1267 | return r; | |
1268 | ||
f41315fc | 1269 | r = prioq_put(d->latest, s, &s->latest_index); |
1e45e3fe | 1270 | if (r < 0) { |
f41315fc LP |
1271 | assert_se(prioq_remove(d->earliest, s, &s->earliest_index) > 0); |
1272 | s->earliest_index = PRIOQ_IDX_NULL; | |
1e45e3fe LP |
1273 | return r; |
1274 | } | |
1275 | ||
1276 | d->needs_rearm = true; | |
1277 | return 0; | |
1278 | } | |
1279 | ||
6a0f1f6d | 1280 | _public_ int sd_event_add_time( |
fd38203a | 1281 | sd_event *e, |
151b9b96 | 1282 | sd_event_source **ret, |
6a0f1f6d | 1283 | clockid_t clock, |
fd38203a | 1284 | uint64_t usec, |
c2ba3ad6 | 1285 | uint64_t accuracy, |
718db961 | 1286 | sd_event_time_handler_t callback, |
151b9b96 | 1287 | void *userdata) { |
fd38203a | 1288 | |
6a0f1f6d | 1289 | EventSourceType type; |
ec766a51 | 1290 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6a0f1f6d | 1291 | struct clock_data *d; |
fd38203a LP |
1292 | int r; |
1293 | ||
305f78bf | 1294 | assert_return(e, -EINVAL); |
b937d761 | 1295 | assert_return(e = event_resolve(e), -ENOPKG); |
f5fbe71d | 1296 | assert_return(accuracy != UINT64_MAX, -EINVAL); |
da7e457c | 1297 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1298 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1299 | |
e475d10c LP |
1300 | if (!clock_supported(clock)) /* Checks whether the kernel supports the clock */ |
1301 | return -EOPNOTSUPP; | |
1302 | ||
1303 | type = clock_to_event_source_type(clock); /* checks whether sd-event supports this clock */ | |
1304 | if (type < 0) | |
3411372e LP |
1305 | return -EOPNOTSUPP; |
1306 | ||
c4f1aff2 TG |
1307 | if (!callback) |
1308 | callback = time_exit_callback; | |
1309 | ||
1e45e3fe | 1310 | assert_se(d = event_get_clock_data(e, type)); |
c2ba3ad6 | 1311 | |
41c63f36 | 1312 | r = setup_clock_data(e, d, clock); |
c983e776 EV |
1313 | if (r < 0) |
1314 | return r; | |
fd38203a | 1315 | |
a71fe8b8 | 1316 | s = source_new(e, !ret, type); |
fd38203a LP |
1317 | if (!s) |
1318 | return -ENOMEM; | |
1319 | ||
1320 | s->time.next = usec; | |
c2ba3ad6 | 1321 | s->time.accuracy = accuracy == 0 ? DEFAULT_ACCURACY_USEC : accuracy; |
fd38203a | 1322 | s->time.callback = callback; |
f41315fc | 1323 | s->earliest_index = s->latest_index = PRIOQ_IDX_NULL; |
fd38203a | 1324 | s->userdata = userdata; |
baf76283 | 1325 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1326 | |
1e45e3fe | 1327 | r = event_source_time_prioq_put(s, d); |
c2ba3ad6 | 1328 | if (r < 0) |
ec766a51 | 1329 | return r; |
fd38203a | 1330 | |
a71fe8b8 LP |
1331 | if (ret) |
1332 | *ret = s; | |
ec766a51 | 1333 | TAKE_PTR(s); |
a71fe8b8 | 1334 | |
fd38203a LP |
1335 | return 0; |
1336 | } | |
1337 | ||
d6a83dc4 LP |
1338 | _public_ int sd_event_add_time_relative( |
1339 | sd_event *e, | |
1340 | sd_event_source **ret, | |
1341 | clockid_t clock, | |
1342 | uint64_t usec, | |
1343 | uint64_t accuracy, | |
1344 | sd_event_time_handler_t callback, | |
1345 | void *userdata) { | |
1346 | ||
1347 | usec_t t; | |
1348 | int r; | |
1349 | ||
1350 | /* Same as sd_event_add_time() but operates relative to the event loop's current point in time, and | |
1351 | * checks for overflow. */ | |
1352 | ||
1353 | r = sd_event_now(e, clock, &t); | |
1354 | if (r < 0) | |
1355 | return r; | |
1356 | ||
1357 | if (usec >= USEC_INFINITY - t) | |
1358 | return -EOVERFLOW; | |
1359 | ||
1360 | return sd_event_add_time(e, ret, clock, t + usec, accuracy, callback, userdata); | |
1361 | } | |
1362 | ||
59bc1fd7 LP |
1363 | static int signal_exit_callback(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) { |
1364 | assert(s); | |
1365 | ||
1366 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1367 | } | |
1368 | ||
f7262a9f | 1369 | _public_ int sd_event_add_signal( |
305f78bf | 1370 | sd_event *e, |
151b9b96 | 1371 | sd_event_source **ret, |
305f78bf | 1372 | int sig, |
718db961 | 1373 | sd_event_signal_handler_t callback, |
151b9b96 | 1374 | void *userdata) { |
305f78bf | 1375 | |
ec766a51 | 1376 | _cleanup_(source_freep) sd_event_source *s = NULL; |
9da4cb2b | 1377 | struct signal_data *d; |
897448bd LP |
1378 | sigset_t new_ss; |
1379 | bool block_it; | |
fd38203a LP |
1380 | int r; |
1381 | ||
305f78bf | 1382 | assert_return(e, -EINVAL); |
b937d761 | 1383 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1384 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1385 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1386 | |
897448bd LP |
1387 | /* Let's make sure our special flag stays outside of the valid signal range */ |
1388 | assert_cc(_NSIG < SD_EVENT_SIGNAL_PROCMASK); | |
1389 | ||
1390 | if (sig & SD_EVENT_SIGNAL_PROCMASK) { | |
1391 | sig &= ~SD_EVENT_SIGNAL_PROCMASK; | |
1392 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
1393 | ||
1394 | block_it = true; | |
1395 | } else { | |
1396 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
1397 | ||
1398 | r = signal_is_blocked(sig); | |
1399 | if (r < 0) | |
1400 | return r; | |
1401 | if (r == 0) | |
1402 | return -EBUSY; | |
1403 | ||
1404 | block_it = false; | |
1405 | } | |
1406 | ||
59bc1fd7 LP |
1407 | if (!callback) |
1408 | callback = signal_exit_callback; | |
1409 | ||
fd38203a LP |
1410 | if (!e->signal_sources) { |
1411 | e->signal_sources = new0(sd_event_source*, _NSIG); | |
1412 | if (!e->signal_sources) | |
1413 | return -ENOMEM; | |
1414 | } else if (e->signal_sources[sig]) | |
1415 | return -EBUSY; | |
1416 | ||
a71fe8b8 | 1417 | s = source_new(e, !ret, SOURCE_SIGNAL); |
fd38203a LP |
1418 | if (!s) |
1419 | return -ENOMEM; | |
1420 | ||
1421 | s->signal.sig = sig; | |
1422 | s->signal.callback = callback; | |
1423 | s->userdata = userdata; | |
baf76283 | 1424 | s->enabled = SD_EVENT_ON; |
fd38203a LP |
1425 | |
1426 | e->signal_sources[sig] = s; | |
fd38203a | 1427 | |
897448bd LP |
1428 | if (block_it) { |
1429 | sigset_t old_ss; | |
1430 | ||
1431 | if (sigemptyset(&new_ss) < 0) | |
1432 | return -errno; | |
1433 | ||
1434 | if (sigaddset(&new_ss, sig) < 0) | |
1435 | return -errno; | |
1436 | ||
1437 | r = pthread_sigmask(SIG_BLOCK, &new_ss, &old_ss); | |
1438 | if (r != 0) | |
1439 | return -r; | |
1440 | ||
1441 | r = sigismember(&old_ss, sig); | |
1442 | if (r < 0) | |
1443 | return -errno; | |
1444 | ||
1445 | s->signal.unblock = !r; | |
1446 | } else | |
1447 | s->signal.unblock = false; | |
1448 | ||
9da4cb2b | 1449 | r = event_make_signal_data(e, sig, &d); |
897448bd LP |
1450 | if (r < 0) { |
1451 | if (s->signal.unblock) | |
1452 | (void) pthread_sigmask(SIG_UNBLOCK, &new_ss, NULL); | |
1453 | ||
9da4cb2b | 1454 | return r; |
897448bd | 1455 | } |
fd38203a | 1456 | |
f1f00dbb LP |
1457 | /* Use the signal name as description for the event source by default */ |
1458 | (void) sd_event_source_set_description(s, signal_to_string(sig)); | |
1459 | ||
a71fe8b8 LP |
1460 | if (ret) |
1461 | *ret = s; | |
ec766a51 | 1462 | TAKE_PTR(s); |
a71fe8b8 | 1463 | |
fd38203a LP |
1464 | return 0; |
1465 | } | |
1466 | ||
b9350e70 LP |
1467 | static int child_exit_callback(sd_event_source *s, const siginfo_t *si, void *userdata) { |
1468 | assert(s); | |
1469 | ||
1470 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1471 | } | |
1472 | ||
f8f3f926 LP |
1473 | static bool shall_use_pidfd(void) { |
1474 | /* Mostly relevant for debugging, i.e. this is used in test-event.c to test the event loop once with and once without pidfd */ | |
1475 | return getenv_bool_secure("SYSTEMD_PIDFD") != 0; | |
1476 | } | |
1477 | ||
f7262a9f | 1478 | _public_ int sd_event_add_child( |
305f78bf | 1479 | sd_event *e, |
151b9b96 | 1480 | sd_event_source **ret, |
305f78bf LP |
1481 | pid_t pid, |
1482 | int options, | |
718db961 | 1483 | sd_event_child_handler_t callback, |
151b9b96 | 1484 | void *userdata) { |
305f78bf | 1485 | |
ec766a51 | 1486 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1487 | int r; |
1488 | ||
305f78bf | 1489 | assert_return(e, -EINVAL); |
b937d761 | 1490 | assert_return(e = event_resolve(e), -ENOPKG); |
305f78bf LP |
1491 | assert_return(pid > 1, -EINVAL); |
1492 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1493 | assert_return(options != 0, -EINVAL); | |
da7e457c | 1494 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1495 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1496 | |
b9350e70 LP |
1497 | if (!callback) |
1498 | callback = child_exit_callback; | |
1499 | ||
b6d5481b | 1500 | if (e->n_online_child_sources == 0) { |
ee880b37 LP |
1501 | /* Caller must block SIGCHLD before using us to watch children, even if pidfd is available, |
1502 | * for compatibility with pre-pidfd and because we don't want the reap the child processes | |
1503 | * ourselves, i.e. call waitid(), and don't want Linux' default internal logic for that to | |
1504 | * take effect. | |
1505 | * | |
1506 | * (As an optimization we only do this check on the first child event source created.) */ | |
1507 | r = signal_is_blocked(SIGCHLD); | |
1508 | if (r < 0) | |
1509 | return r; | |
1510 | if (r == 0) | |
1511 | return -EBUSY; | |
1512 | } | |
1513 | ||
d5099efc | 1514 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
fd38203a LP |
1515 | if (r < 0) |
1516 | return r; | |
1517 | ||
4a0b58c4 | 1518 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) |
fd38203a LP |
1519 | return -EBUSY; |
1520 | ||
a71fe8b8 | 1521 | s = source_new(e, !ret, SOURCE_CHILD); |
fd38203a LP |
1522 | if (!s) |
1523 | return -ENOMEM; | |
1524 | ||
f8f3f926 | 1525 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1526 | s->child.options = options; |
1527 | s->child.callback = callback; | |
1528 | s->userdata = userdata; | |
baf76283 | 1529 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1530 | |
f8f3f926 LP |
1531 | /* We always take a pidfd here if we can, even if we wait for anything else than WEXITED, so that we |
1532 | * pin the PID, and make regular waitid() handling race-free. */ | |
1533 | ||
1534 | if (shall_use_pidfd()) { | |
54988a27 | 1535 | s->child.pidfd = pidfd_open(pid, 0); |
f8f3f926 LP |
1536 | if (s->child.pidfd < 0) { |
1537 | /* Propagate errors unless the syscall is not supported or blocked */ | |
1538 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
1539 | return -errno; | |
1540 | } else | |
1541 | s->child.pidfd_owned = true; /* If we allocate the pidfd we own it by default */ | |
1542 | } else | |
254d1313 | 1543 | s->child.pidfd = -EBADF; |
f8f3f926 | 1544 | |
f8f3f926 LP |
1545 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1546 | /* We have a pidfd and we only want to watch for exit */ | |
f8f3f926 | 1547 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1548 | if (r < 0) |
f8f3f926 | 1549 | return r; |
ac9f2640 | 1550 | |
f8f3f926 LP |
1551 | } else { |
1552 | /* We have no pidfd or we shall wait for some other event than WEXITED */ | |
f8f3f926 | 1553 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1554 | if (r < 0) |
f8f3f926 | 1555 | return r; |
f8f3f926 LP |
1556 | |
1557 | e->need_process_child = true; | |
1558 | } | |
c2ba3ad6 | 1559 | |
54988a27 YW |
1560 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); |
1561 | if (r < 0) | |
1562 | return r; | |
1563 | ||
1564 | /* These must be done after everything succeeds. */ | |
1565 | s->child.pid = pid; | |
b6d5481b | 1566 | e->n_online_child_sources++; |
ac9f2640 | 1567 | |
a71fe8b8 LP |
1568 | if (ret) |
1569 | *ret = s; | |
ec766a51 | 1570 | TAKE_PTR(s); |
f8f3f926 LP |
1571 | return 0; |
1572 | } | |
1573 | ||
1574 | _public_ int sd_event_add_child_pidfd( | |
1575 | sd_event *e, | |
1576 | sd_event_source **ret, | |
1577 | int pidfd, | |
1578 | int options, | |
1579 | sd_event_child_handler_t callback, | |
1580 | void *userdata) { | |
1581 | ||
1582 | ||
1583 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
1584 | pid_t pid; | |
1585 | int r; | |
1586 | ||
1587 | assert_return(e, -EINVAL); | |
1588 | assert_return(e = event_resolve(e), -ENOPKG); | |
1589 | assert_return(pidfd >= 0, -EBADF); | |
1590 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1591 | assert_return(options != 0, -EINVAL); | |
f8f3f926 LP |
1592 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1593 | assert_return(!event_pid_changed(e), -ECHILD); | |
1594 | ||
b9350e70 LP |
1595 | if (!callback) |
1596 | callback = child_exit_callback; | |
1597 | ||
b6d5481b | 1598 | if (e->n_online_child_sources == 0) { |
ee880b37 LP |
1599 | r = signal_is_blocked(SIGCHLD); |
1600 | if (r < 0) | |
1601 | return r; | |
1602 | if (r == 0) | |
1603 | return -EBUSY; | |
1604 | } | |
1605 | ||
f8f3f926 LP |
1606 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
1607 | if (r < 0) | |
1608 | return r; | |
1609 | ||
1610 | r = pidfd_get_pid(pidfd, &pid); | |
1611 | if (r < 0) | |
1612 | return r; | |
1613 | ||
1614 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) | |
1615 | return -EBUSY; | |
1616 | ||
1617 | s = source_new(e, !ret, SOURCE_CHILD); | |
1618 | if (!s) | |
1619 | return -ENOMEM; | |
1620 | ||
1621 | s->wakeup = WAKEUP_EVENT_SOURCE; | |
1622 | s->child.pidfd = pidfd; | |
1623 | s->child.pid = pid; | |
1624 | s->child.options = options; | |
1625 | s->child.callback = callback; | |
1626 | 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) */ | |
1627 | s->userdata = userdata; | |
1628 | s->enabled = SD_EVENT_ONESHOT; | |
1629 | ||
1630 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); | |
1631 | if (r < 0) | |
1632 | return r; | |
1633 | ||
f8f3f926 LP |
1634 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1635 | /* We only want to watch for WEXITED */ | |
f8f3f926 | 1636 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1637 | if (r < 0) |
f8f3f926 | 1638 | return r; |
f8f3f926 LP |
1639 | } else { |
1640 | /* We shall wait for some other event than WEXITED */ | |
f8f3f926 | 1641 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1642 | if (r < 0) |
f8f3f926 | 1643 | return r; |
a71fe8b8 | 1644 | |
f8f3f926 LP |
1645 | e->need_process_child = true; |
1646 | } | |
1647 | ||
b6d5481b | 1648 | e->n_online_child_sources++; |
ac9f2640 | 1649 | |
f8f3f926 LP |
1650 | if (ret) |
1651 | *ret = s; | |
f8f3f926 | 1652 | TAKE_PTR(s); |
fd38203a LP |
1653 | return 0; |
1654 | } | |
1655 | ||
b9350e70 LP |
1656 | static int generic_exit_callback(sd_event_source *s, void *userdata) { |
1657 | assert(s); | |
1658 | ||
1659 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1660 | } | |
1661 | ||
f7262a9f | 1662 | _public_ int sd_event_add_defer( |
305f78bf | 1663 | sd_event *e, |
151b9b96 | 1664 | sd_event_source **ret, |
718db961 | 1665 | sd_event_handler_t callback, |
151b9b96 | 1666 | void *userdata) { |
305f78bf | 1667 | |
ec766a51 | 1668 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1669 | int r; |
1670 | ||
305f78bf | 1671 | assert_return(e, -EINVAL); |
b937d761 | 1672 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1673 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1674 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1675 | |
b9350e70 LP |
1676 | if (!callback) |
1677 | callback = generic_exit_callback; | |
1678 | ||
a71fe8b8 | 1679 | s = source_new(e, !ret, SOURCE_DEFER); |
fd38203a LP |
1680 | if (!s) |
1681 | return -ENOMEM; | |
1682 | ||
1683 | s->defer.callback = callback; | |
1684 | s->userdata = userdata; | |
baf76283 | 1685 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a LP |
1686 | |
1687 | r = source_set_pending(s, true); | |
ec766a51 | 1688 | if (r < 0) |
fd38203a | 1689 | return r; |
fd38203a | 1690 | |
a71fe8b8 LP |
1691 | if (ret) |
1692 | *ret = s; | |
ec766a51 | 1693 | TAKE_PTR(s); |
a71fe8b8 | 1694 | |
fd38203a LP |
1695 | return 0; |
1696 | } | |
1697 | ||
6e9feda3 LP |
1698 | _public_ int sd_event_add_post( |
1699 | sd_event *e, | |
1700 | sd_event_source **ret, | |
1701 | sd_event_handler_t callback, | |
1702 | void *userdata) { | |
1703 | ||
ec766a51 | 1704 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6e9feda3 LP |
1705 | int r; |
1706 | ||
1707 | assert_return(e, -EINVAL); | |
b937d761 | 1708 | assert_return(e = event_resolve(e), -ENOPKG); |
6e9feda3 LP |
1709 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1710 | assert_return(!event_pid_changed(e), -ECHILD); | |
1711 | ||
b9350e70 LP |
1712 | if (!callback) |
1713 | callback = generic_exit_callback; | |
1714 | ||
a71fe8b8 | 1715 | s = source_new(e, !ret, SOURCE_POST); |
6e9feda3 LP |
1716 | if (!s) |
1717 | return -ENOMEM; | |
1718 | ||
1719 | s->post.callback = callback; | |
1720 | s->userdata = userdata; | |
1721 | s->enabled = SD_EVENT_ON; | |
1722 | ||
de7fef4b | 1723 | r = set_ensure_put(&e->post_sources, NULL, s); |
ec766a51 | 1724 | if (r < 0) |
6e9feda3 | 1725 | return r; |
de7fef4b | 1726 | assert(r > 0); |
6e9feda3 | 1727 | |
a71fe8b8 LP |
1728 | if (ret) |
1729 | *ret = s; | |
ec766a51 | 1730 | TAKE_PTR(s); |
a71fe8b8 | 1731 | |
6e9feda3 LP |
1732 | return 0; |
1733 | } | |
1734 | ||
6203e07a | 1735 | _public_ int sd_event_add_exit( |
305f78bf | 1736 | sd_event *e, |
151b9b96 | 1737 | sd_event_source **ret, |
718db961 | 1738 | sd_event_handler_t callback, |
151b9b96 | 1739 | void *userdata) { |
305f78bf | 1740 | |
ec766a51 | 1741 | _cleanup_(source_freep) sd_event_source *s = NULL; |
da7e457c LP |
1742 | int r; |
1743 | ||
1744 | assert_return(e, -EINVAL); | |
b937d761 | 1745 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
1746 | assert_return(callback, -EINVAL); |
1747 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
1748 | assert_return(!event_pid_changed(e), -ECHILD); | |
1749 | ||
c983e776 EV |
1750 | r = prioq_ensure_allocated(&e->exit, exit_prioq_compare); |
1751 | if (r < 0) | |
1752 | return r; | |
da7e457c | 1753 | |
a71fe8b8 | 1754 | s = source_new(e, !ret, SOURCE_EXIT); |
fd38203a | 1755 | if (!s) |
da7e457c | 1756 | return -ENOMEM; |
fd38203a | 1757 | |
6203e07a | 1758 | s->exit.callback = callback; |
da7e457c | 1759 | s->userdata = userdata; |
6203e07a | 1760 | s->exit.prioq_index = PRIOQ_IDX_NULL; |
baf76283 | 1761 | s->enabled = SD_EVENT_ONESHOT; |
da7e457c | 1762 | |
6203e07a | 1763 | r = prioq_put(s->event->exit, s, &s->exit.prioq_index); |
ec766a51 | 1764 | if (r < 0) |
da7e457c | 1765 | return r; |
da7e457c | 1766 | |
a71fe8b8 LP |
1767 | if (ret) |
1768 | *ret = s; | |
ec766a51 | 1769 | TAKE_PTR(s); |
a71fe8b8 | 1770 | |
da7e457c LP |
1771 | return 0; |
1772 | } | |
1773 | ||
97ef5391 LP |
1774 | static void event_free_inotify_data(sd_event *e, struct inotify_data *d) { |
1775 | assert(e); | |
1776 | ||
1777 | if (!d) | |
1778 | return; | |
1779 | ||
1780 | assert(hashmap_isempty(d->inodes)); | |
1781 | assert(hashmap_isempty(d->wd)); | |
1782 | ||
1783 | if (d->buffer_filled > 0) | |
0601b958 | 1784 | LIST_REMOVE(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
1785 | |
1786 | hashmap_free(d->inodes); | |
1787 | hashmap_free(d->wd); | |
1788 | ||
1789 | assert_se(hashmap_remove(e->inotify_data, &d->priority) == d); | |
1790 | ||
1791 | if (d->fd >= 0) { | |
fbae5090 YW |
1792 | if (!event_pid_changed(e) && |
1793 | epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, d->fd, NULL) < 0) | |
97ef5391 LP |
1794 | log_debug_errno(errno, "Failed to remove inotify fd from epoll, ignoring: %m"); |
1795 | ||
1796 | safe_close(d->fd); | |
1797 | } | |
1798 | free(d); | |
1799 | } | |
1800 | ||
1801 | static int event_make_inotify_data( | |
1802 | sd_event *e, | |
1803 | int64_t priority, | |
1804 | struct inotify_data **ret) { | |
1805 | ||
254d1313 | 1806 | _cleanup_close_ int fd = -EBADF; |
97ef5391 | 1807 | struct inotify_data *d; |
97ef5391 LP |
1808 | int r; |
1809 | ||
1810 | assert(e); | |
1811 | ||
1812 | d = hashmap_get(e->inotify_data, &priority); | |
1813 | if (d) { | |
1814 | if (ret) | |
1815 | *ret = d; | |
1816 | return 0; | |
1817 | } | |
1818 | ||
1819 | fd = inotify_init1(IN_NONBLOCK|O_CLOEXEC); | |
1820 | if (fd < 0) | |
1821 | return -errno; | |
1822 | ||
1823 | fd = fd_move_above_stdio(fd); | |
1824 | ||
97ef5391 LP |
1825 | d = new(struct inotify_data, 1); |
1826 | if (!d) | |
1827 | return -ENOMEM; | |
1828 | ||
1829 | *d = (struct inotify_data) { | |
1830 | .wakeup = WAKEUP_INOTIFY_DATA, | |
1831 | .fd = TAKE_FD(fd), | |
1832 | .priority = priority, | |
1833 | }; | |
1834 | ||
c2484a75 | 1835 | r = hashmap_ensure_put(&e->inotify_data, &uint64_hash_ops, &d->priority, d); |
97ef5391 LP |
1836 | if (r < 0) { |
1837 | d->fd = safe_close(d->fd); | |
1838 | free(d); | |
1839 | return r; | |
1840 | } | |
1841 | ||
1eac7948 | 1842 | struct epoll_event ev = { |
97ef5391 LP |
1843 | .events = EPOLLIN, |
1844 | .data.ptr = d, | |
1845 | }; | |
1846 | ||
1847 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { | |
1848 | r = -errno; | |
1849 | d->fd = safe_close(d->fd); /* let's close this ourselves, as event_free_inotify_data() would otherwise | |
1850 | * remove the fd from the epoll first, which we don't want as we couldn't | |
1851 | * add it in the first place. */ | |
1852 | event_free_inotify_data(e, d); | |
1853 | return r; | |
1854 | } | |
1855 | ||
1856 | if (ret) | |
1857 | *ret = d; | |
1858 | ||
1859 | return 1; | |
1860 | } | |
1861 | ||
7a08d314 | 1862 | static int inode_data_compare(const struct inode_data *x, const struct inode_data *y) { |
90c88092 | 1863 | int r; |
97ef5391 LP |
1864 | |
1865 | assert(x); | |
1866 | assert(y); | |
1867 | ||
90c88092 YW |
1868 | r = CMP(x->dev, y->dev); |
1869 | if (r != 0) | |
1870 | return r; | |
97ef5391 | 1871 | |
6dd91b36 | 1872 | return CMP(x->ino, y->ino); |
97ef5391 LP |
1873 | } |
1874 | ||
7a08d314 YW |
1875 | static void inode_data_hash_func(const struct inode_data *d, struct siphash *state) { |
1876 | assert(d); | |
97ef5391 LP |
1877 | |
1878 | siphash24_compress(&d->dev, sizeof(d->dev), state); | |
1879 | siphash24_compress(&d->ino, sizeof(d->ino), state); | |
1880 | } | |
1881 | ||
7a08d314 | 1882 | DEFINE_PRIVATE_HASH_OPS(inode_data_hash_ops, struct inode_data, inode_data_hash_func, inode_data_compare); |
97ef5391 LP |
1883 | |
1884 | static void event_free_inode_data( | |
1885 | sd_event *e, | |
1886 | struct inode_data *d) { | |
1887 | ||
1888 | assert(e); | |
1889 | ||
1890 | if (!d) | |
1891 | return; | |
1892 | ||
64903d18 | 1893 | assert(!d->event_sources); |
97ef5391 LP |
1894 | |
1895 | if (d->fd >= 0) { | |
ed828563 | 1896 | LIST_REMOVE(to_close, e->inode_data_to_close_list, d); |
97ef5391 LP |
1897 | safe_close(d->fd); |
1898 | } | |
1899 | ||
1900 | if (d->inotify_data) { | |
1901 | ||
1902 | if (d->wd >= 0) { | |
fbae5090 | 1903 | if (d->inotify_data->fd >= 0 && !event_pid_changed(e)) { |
97ef5391 LP |
1904 | /* So here's a problem. At the time this runs the watch descriptor might already be |
1905 | * invalidated, because an IN_IGNORED event might be queued right the moment we enter | |
1906 | * the syscall. Hence, whenever we get EINVAL, ignore it entirely, since it's a very | |
1907 | * likely case to happen. */ | |
1908 | ||
1909 | if (inotify_rm_watch(d->inotify_data->fd, d->wd) < 0 && errno != EINVAL) | |
1910 | log_debug_errno(errno, "Failed to remove watch descriptor %i from inotify, ignoring: %m", d->wd); | |
1911 | } | |
1912 | ||
1913 | assert_se(hashmap_remove(d->inotify_data->wd, INT_TO_PTR(d->wd)) == d); | |
1914 | } | |
1915 | ||
1916 | assert_se(hashmap_remove(d->inotify_data->inodes, d) == d); | |
1917 | } | |
1918 | ||
1919 | free(d); | |
1920 | } | |
1921 | ||
53baf2ef LP |
1922 | static void event_gc_inotify_data( |
1923 | sd_event *e, | |
1924 | struct inotify_data *d) { | |
1925 | ||
1926 | assert(e); | |
1927 | ||
1928 | /* GCs the inotify data object if we don't need it anymore. That's the case if we don't want to watch | |
1929 | * any inode with it anymore, which in turn happens if no event source of this priority is interested | |
1930 | * in any inode any longer. That said, we maintain an extra busy counter: if non-zero we'll delay GC | |
1931 | * (under the expectation that the GC is called again once the counter is decremented). */ | |
1932 | ||
1933 | if (!d) | |
1934 | return; | |
1935 | ||
1936 | if (!hashmap_isempty(d->inodes)) | |
1937 | return; | |
1938 | ||
1939 | if (d->n_busy > 0) | |
1940 | return; | |
1941 | ||
1942 | event_free_inotify_data(e, d); | |
1943 | } | |
1944 | ||
97ef5391 LP |
1945 | static void event_gc_inode_data( |
1946 | sd_event *e, | |
1947 | struct inode_data *d) { | |
1948 | ||
1949 | struct inotify_data *inotify_data; | |
1950 | ||
1951 | assert(e); | |
1952 | ||
1953 | if (!d) | |
1954 | return; | |
1955 | ||
64903d18 | 1956 | if (d->event_sources) |
97ef5391 LP |
1957 | return; |
1958 | ||
1959 | inotify_data = d->inotify_data; | |
1960 | event_free_inode_data(e, d); | |
1961 | ||
53baf2ef | 1962 | event_gc_inotify_data(e, inotify_data); |
97ef5391 LP |
1963 | } |
1964 | ||
1965 | static int event_make_inode_data( | |
1966 | sd_event *e, | |
1967 | struct inotify_data *inotify_data, | |
1968 | dev_t dev, | |
1969 | ino_t ino, | |
1970 | struct inode_data **ret) { | |
1971 | ||
1972 | struct inode_data *d, key; | |
1973 | int r; | |
1974 | ||
1975 | assert(e); | |
1976 | assert(inotify_data); | |
1977 | ||
1978 | key = (struct inode_data) { | |
1979 | .ino = ino, | |
1980 | .dev = dev, | |
1981 | }; | |
1982 | ||
1983 | d = hashmap_get(inotify_data->inodes, &key); | |
1984 | if (d) { | |
1985 | if (ret) | |
1986 | *ret = d; | |
1987 | ||
1988 | return 0; | |
1989 | } | |
1990 | ||
1991 | r = hashmap_ensure_allocated(&inotify_data->inodes, &inode_data_hash_ops); | |
1992 | if (r < 0) | |
1993 | return r; | |
1994 | ||
1995 | d = new(struct inode_data, 1); | |
1996 | if (!d) | |
1997 | return -ENOMEM; | |
1998 | ||
1999 | *d = (struct inode_data) { | |
2000 | .dev = dev, | |
2001 | .ino = ino, | |
2002 | .wd = -1, | |
254d1313 | 2003 | .fd = -EBADF, |
97ef5391 LP |
2004 | .inotify_data = inotify_data, |
2005 | }; | |
2006 | ||
2007 | r = hashmap_put(inotify_data->inodes, d, d); | |
2008 | if (r < 0) { | |
2009 | free(d); | |
2010 | return r; | |
2011 | } | |
2012 | ||
2013 | if (ret) | |
2014 | *ret = d; | |
2015 | ||
2016 | return 1; | |
2017 | } | |
2018 | ||
2019 | static uint32_t inode_data_determine_mask(struct inode_data *d) { | |
2020 | bool excl_unlink = true; | |
2021 | uint32_t combined = 0; | |
97ef5391 LP |
2022 | |
2023 | assert(d); | |
2024 | ||
2025 | /* Combines the watch masks of all event sources watching this inode. We generally just OR them together, but | |
2026 | * the IN_EXCL_UNLINK flag is ANDed instead. | |
2027 | * | |
2028 | * Note that we add all sources to the mask here, regardless whether enabled, disabled or oneshot. That's | |
2029 | * because we cannot change the mask anymore after the event source was created once, since the kernel has no | |
f21f31b2 | 2030 | * API for that. Hence we need to subscribe to the maximum mask we ever might be interested in, and suppress |
97ef5391 LP |
2031 | * events we don't care for client-side. */ |
2032 | ||
2033 | LIST_FOREACH(inotify.by_inode_data, s, d->event_sources) { | |
2034 | ||
2035 | if ((s->inotify.mask & IN_EXCL_UNLINK) == 0) | |
2036 | excl_unlink = false; | |
2037 | ||
2038 | combined |= s->inotify.mask; | |
2039 | } | |
2040 | ||
2041 | return (combined & ~(IN_ONESHOT|IN_DONT_FOLLOW|IN_ONLYDIR|IN_EXCL_UNLINK)) | (excl_unlink ? IN_EXCL_UNLINK : 0); | |
2042 | } | |
2043 | ||
2044 | static int inode_data_realize_watch(sd_event *e, struct inode_data *d) { | |
2045 | uint32_t combined_mask; | |
2046 | int wd, r; | |
2047 | ||
2048 | assert(d); | |
2049 | assert(d->fd >= 0); | |
2050 | ||
2051 | combined_mask = inode_data_determine_mask(d); | |
2052 | ||
2053 | if (d->wd >= 0 && combined_mask == d->combined_mask) | |
2054 | return 0; | |
2055 | ||
2056 | r = hashmap_ensure_allocated(&d->inotify_data->wd, NULL); | |
2057 | if (r < 0) | |
2058 | return r; | |
2059 | ||
2060 | wd = inotify_add_watch_fd(d->inotify_data->fd, d->fd, combined_mask); | |
2061 | if (wd < 0) | |
2062 | return -errno; | |
2063 | ||
2064 | if (d->wd < 0) { | |
2065 | r = hashmap_put(d->inotify_data->wd, INT_TO_PTR(wd), d); | |
2066 | if (r < 0) { | |
2067 | (void) inotify_rm_watch(d->inotify_data->fd, wd); | |
2068 | return r; | |
2069 | } | |
2070 | ||
2071 | d->wd = wd; | |
2072 | ||
2073 | } else if (d->wd != wd) { | |
2074 | ||
2075 | log_debug("Weird, the watch descriptor we already knew for this inode changed?"); | |
2076 | (void) inotify_rm_watch(d->fd, wd); | |
2077 | return -EINVAL; | |
2078 | } | |
2079 | ||
2080 | d->combined_mask = combined_mask; | |
2081 | return 1; | |
2082 | } | |
2083 | ||
b9350e70 LP |
2084 | static int inotify_exit_callback(sd_event_source *s, const struct inotify_event *event, void *userdata) { |
2085 | assert(s); | |
2086 | ||
2087 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
2088 | } | |
2089 | ||
e67d738a | 2090 | static int event_add_inotify_fd_internal( |
97ef5391 LP |
2091 | sd_event *e, |
2092 | sd_event_source **ret, | |
e67d738a LP |
2093 | int fd, |
2094 | bool donate, | |
97ef5391 LP |
2095 | uint32_t mask, |
2096 | sd_event_inotify_handler_t callback, | |
2097 | void *userdata) { | |
2098 | ||
5bb1d7fb | 2099 | _cleanup_close_ int donated_fd = donate ? fd : -EBADF; |
e67d738a | 2100 | _cleanup_(source_freep) sd_event_source *s = NULL; |
97ef5391 LP |
2101 | struct inotify_data *inotify_data = NULL; |
2102 | struct inode_data *inode_data = NULL; | |
97ef5391 LP |
2103 | struct stat st; |
2104 | int r; | |
2105 | ||
2106 | assert_return(e, -EINVAL); | |
2107 | assert_return(e = event_resolve(e), -ENOPKG); | |
e67d738a | 2108 | assert_return(fd >= 0, -EBADF); |
97ef5391 LP |
2109 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2110 | assert_return(!event_pid_changed(e), -ECHILD); | |
2111 | ||
b9350e70 LP |
2112 | if (!callback) |
2113 | callback = inotify_exit_callback; | |
2114 | ||
97ef5391 LP |
2115 | /* Refuse IN_MASK_ADD since we coalesce watches on the same inode, and hence really don't want to merge |
2116 | * masks. Or in other words, this whole code exists only to manage IN_MASK_ADD type operations for you, hence | |
2117 | * the user can't use them for us. */ | |
2118 | if (mask & IN_MASK_ADD) | |
2119 | return -EINVAL; | |
2120 | ||
97ef5391 LP |
2121 | if (fstat(fd, &st) < 0) |
2122 | return -errno; | |
2123 | ||
2124 | s = source_new(e, !ret, SOURCE_INOTIFY); | |
2125 | if (!s) | |
2126 | return -ENOMEM; | |
2127 | ||
2128 | s->enabled = mask & IN_ONESHOT ? SD_EVENT_ONESHOT : SD_EVENT_ON; | |
2129 | s->inotify.mask = mask; | |
2130 | s->inotify.callback = callback; | |
2131 | s->userdata = userdata; | |
2132 | ||
2133 | /* Allocate an inotify object for this priority, and an inode object within it */ | |
2134 | r = event_make_inotify_data(e, SD_EVENT_PRIORITY_NORMAL, &inotify_data); | |
2135 | if (r < 0) | |
8c75fe17 | 2136 | return r; |
97ef5391 LP |
2137 | |
2138 | r = event_make_inode_data(e, inotify_data, st.st_dev, st.st_ino, &inode_data); | |
8c75fe17 | 2139 | if (r < 0) { |
e67d738a | 2140 | event_gc_inotify_data(e, inotify_data); |
8c75fe17 ZJS |
2141 | return r; |
2142 | } | |
97ef5391 LP |
2143 | |
2144 | /* Keep the O_PATH fd around until the first iteration of the loop, so that we can still change the priority of | |
2145 | * the event source, until then, for which we need the original inode. */ | |
2146 | if (inode_data->fd < 0) { | |
e67d738a LP |
2147 | if (donated_fd >= 0) |
2148 | inode_data->fd = TAKE_FD(donated_fd); | |
2149 | else { | |
2150 | inode_data->fd = fcntl(fd, F_DUPFD_CLOEXEC, 3); | |
2151 | if (inode_data->fd < 0) { | |
2152 | r = -errno; | |
2153 | event_gc_inode_data(e, inode_data); | |
2154 | return r; | |
2155 | } | |
2156 | } | |
2157 | ||
ed828563 | 2158 | LIST_PREPEND(to_close, e->inode_data_to_close_list, inode_data); |
97ef5391 LP |
2159 | } |
2160 | ||
2161 | /* Link our event source to the inode data object */ | |
2162 | LIST_PREPEND(inotify.by_inode_data, inode_data->event_sources, s); | |
2163 | s->inotify.inode_data = inode_data; | |
2164 | ||
97ef5391 LP |
2165 | /* Actually realize the watch now */ |
2166 | r = inode_data_realize_watch(e, inode_data); | |
2167 | if (r < 0) | |
8c75fe17 | 2168 | return r; |
97ef5391 | 2169 | |
97ef5391 LP |
2170 | if (ret) |
2171 | *ret = s; | |
8c75fe17 | 2172 | TAKE_PTR(s); |
97ef5391 LP |
2173 | |
2174 | return 0; | |
97ef5391 LP |
2175 | } |
2176 | ||
e67d738a LP |
2177 | _public_ int sd_event_add_inotify_fd( |
2178 | sd_event *e, | |
2179 | sd_event_source **ret, | |
2180 | int fd, | |
2181 | uint32_t mask, | |
2182 | sd_event_inotify_handler_t callback, | |
2183 | void *userdata) { | |
2184 | ||
2185 | return event_add_inotify_fd_internal(e, ret, fd, /* donate= */ false, mask, callback, userdata); | |
2186 | } | |
2187 | ||
2188 | _public_ int sd_event_add_inotify( | |
2189 | sd_event *e, | |
2190 | sd_event_source **ret, | |
2191 | const char *path, | |
2192 | uint32_t mask, | |
2193 | sd_event_inotify_handler_t callback, | |
2194 | void *userdata) { | |
2195 | ||
2091c779 | 2196 | sd_event_source *s = NULL; /* avoid false maybe-uninitialized warning */ |
e67d738a LP |
2197 | int fd, r; |
2198 | ||
2199 | assert_return(path, -EINVAL); | |
2200 | ||
586c8cee ZJS |
2201 | fd = open(path, O_PATH | O_CLOEXEC | |
2202 | (mask & IN_ONLYDIR ? O_DIRECTORY : 0) | | |
2203 | (mask & IN_DONT_FOLLOW ? O_NOFOLLOW : 0)); | |
e67d738a LP |
2204 | if (fd < 0) |
2205 | return -errno; | |
2206 | ||
2207 | r = event_add_inotify_fd_internal(e, &s, fd, /* donate= */ true, mask, callback, userdata); | |
2208 | if (r < 0) | |
2209 | return r; | |
2210 | ||
2211 | (void) sd_event_source_set_description(s, path); | |
2212 | ||
2213 | if (ret) | |
2214 | *ret = s; | |
2215 | ||
2216 | return r; | |
2217 | } | |
2218 | ||
8301aa0b | 2219 | static sd_event_source* event_source_free(sd_event_source *s) { |
6680dd6b LP |
2220 | if (!s) |
2221 | return NULL; | |
da7e457c | 2222 | |
8301aa0b YW |
2223 | /* Here's a special hack: when we are called from a |
2224 | * dispatch handler we won't free the event source | |
2225 | * immediately, but we will detach the fd from the | |
2226 | * epoll. This way it is safe for the caller to unref | |
2227 | * the event source and immediately close the fd, but | |
2228 | * we still retain a valid event source object after | |
2229 | * the callback. */ | |
fd38203a | 2230 | |
76d04c3a | 2231 | if (s->dispatching) |
8301aa0b | 2232 | source_disconnect(s); |
76d04c3a | 2233 | else |
8301aa0b | 2234 | source_free(s); |
fd38203a LP |
2235 | |
2236 | return NULL; | |
2237 | } | |
2238 | ||
8301aa0b YW |
2239 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event_source, sd_event_source, event_source_free); |
2240 | ||
356779df | 2241 | _public_ int sd_event_source_set_description(sd_event_source *s, const char *description) { |
f7f53e9e | 2242 | assert_return(s, -EINVAL); |
f4b2933e | 2243 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 2244 | |
356779df | 2245 | return free_and_strdup(&s->description, description); |
f7f53e9e TG |
2246 | } |
2247 | ||
356779df | 2248 | _public_ int sd_event_source_get_description(sd_event_source *s, const char **description) { |
f7f53e9e | 2249 | assert_return(s, -EINVAL); |
356779df | 2250 | assert_return(description, -EINVAL); |
f4b2933e | 2251 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 2252 | |
7d92a1a4 ZJS |
2253 | if (!s->description) |
2254 | return -ENXIO; | |
2255 | ||
356779df | 2256 | *description = s->description; |
f7f53e9e TG |
2257 | return 0; |
2258 | } | |
2259 | ||
adcc4ca3 | 2260 | _public_ sd_event *sd_event_source_get_event(sd_event_source *s) { |
305f78bf | 2261 | assert_return(s, NULL); |
eaa3cbef LP |
2262 | |
2263 | return s->event; | |
2264 | } | |
2265 | ||
f7262a9f | 2266 | _public_ int sd_event_source_get_pending(sd_event_source *s) { |
305f78bf | 2267 | assert_return(s, -EINVAL); |
6203e07a | 2268 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c | 2269 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2270 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2271 | |
2272 | return s->pending; | |
2273 | } | |
2274 | ||
f7262a9f | 2275 | _public_ int sd_event_source_get_io_fd(sd_event_source *s) { |
305f78bf LP |
2276 | assert_return(s, -EINVAL); |
2277 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2278 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2279 | |
2280 | return s->io.fd; | |
2281 | } | |
2282 | ||
30caf8f3 LP |
2283 | _public_ int sd_event_source_set_io_fd(sd_event_source *s, int fd) { |
2284 | int r; | |
2285 | ||
2286 | assert_return(s, -EINVAL); | |
8ac43fee | 2287 | assert_return(fd >= 0, -EBADF); |
30caf8f3 LP |
2288 | assert_return(s->type == SOURCE_IO, -EDOM); |
2289 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2290 | ||
2291 | if (s->io.fd == fd) | |
2292 | return 0; | |
2293 | ||
b6d5481b | 2294 | if (event_source_is_offline(s)) { |
30caf8f3 LP |
2295 | s->io.fd = fd; |
2296 | s->io.registered = false; | |
2297 | } else { | |
2298 | int saved_fd; | |
2299 | ||
2300 | saved_fd = s->io.fd; | |
2301 | assert(s->io.registered); | |
2302 | ||
2303 | s->io.fd = fd; | |
2304 | s->io.registered = false; | |
2305 | ||
2306 | r = source_io_register(s, s->enabled, s->io.events); | |
2307 | if (r < 0) { | |
2308 | s->io.fd = saved_fd; | |
2309 | s->io.registered = true; | |
2310 | return r; | |
2311 | } | |
2312 | ||
5a795bff | 2313 | (void) epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, saved_fd, NULL); |
30caf8f3 LP |
2314 | } |
2315 | ||
2316 | return 0; | |
2317 | } | |
2318 | ||
ab93297c NM |
2319 | _public_ int sd_event_source_get_io_fd_own(sd_event_source *s) { |
2320 | assert_return(s, -EINVAL); | |
2321 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2322 | ||
2323 | return s->io.owned; | |
2324 | } | |
2325 | ||
2326 | _public_ int sd_event_source_set_io_fd_own(sd_event_source *s, int own) { | |
2327 | assert_return(s, -EINVAL); | |
2328 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2329 | ||
2330 | s->io.owned = own; | |
2331 | return 0; | |
2332 | } | |
2333 | ||
f7262a9f | 2334 | _public_ int sd_event_source_get_io_events(sd_event_source *s, uint32_t* events) { |
305f78bf LP |
2335 | assert_return(s, -EINVAL); |
2336 | assert_return(events, -EINVAL); | |
2337 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2338 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2339 | |
2340 | *events = s->io.events; | |
2341 | return 0; | |
2342 | } | |
2343 | ||
f7262a9f | 2344 | _public_ int sd_event_source_set_io_events(sd_event_source *s, uint32_t events) { |
fd38203a LP |
2345 | int r; |
2346 | ||
305f78bf LP |
2347 | assert_return(s, -EINVAL); |
2348 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2a16a986 | 2349 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 2350 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2351 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2352 | |
b63c8d4f DH |
2353 | /* edge-triggered updates are never skipped, so we can reset edges */ |
2354 | if (s->io.events == events && !(events & EPOLLET)) | |
fd38203a LP |
2355 | return 0; |
2356 | ||
2a0dc6cd LP |
2357 | r = source_set_pending(s, false); |
2358 | if (r < 0) | |
2359 | return r; | |
2360 | ||
b6d5481b | 2361 | if (event_source_is_online(s)) { |
e4715127 | 2362 | r = source_io_register(s, s->enabled, events); |
fd38203a LP |
2363 | if (r < 0) |
2364 | return r; | |
2365 | } | |
2366 | ||
2367 | s->io.events = events; | |
2368 | ||
2369 | return 0; | |
2370 | } | |
2371 | ||
f7262a9f | 2372 | _public_ int sd_event_source_get_io_revents(sd_event_source *s, uint32_t* revents) { |
305f78bf LP |
2373 | assert_return(s, -EINVAL); |
2374 | assert_return(revents, -EINVAL); | |
2375 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2376 | assert_return(s->pending, -ENODATA); | |
2377 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2378 | |
2379 | *revents = s->io.revents; | |
2380 | return 0; | |
2381 | } | |
2382 | ||
f7262a9f | 2383 | _public_ int sd_event_source_get_signal(sd_event_source *s) { |
305f78bf LP |
2384 | assert_return(s, -EINVAL); |
2385 | assert_return(s->type == SOURCE_SIGNAL, -EDOM); | |
2386 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2387 | |
2388 | return s->signal.sig; | |
2389 | } | |
2390 | ||
31927c16 | 2391 | _public_ int sd_event_source_get_priority(sd_event_source *s, int64_t *priority) { |
305f78bf LP |
2392 | assert_return(s, -EINVAL); |
2393 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a | 2394 | |
6680b8d1 ME |
2395 | *priority = s->priority; |
2396 | return 0; | |
fd38203a LP |
2397 | } |
2398 | ||
31927c16 | 2399 | _public_ int sd_event_source_set_priority(sd_event_source *s, int64_t priority) { |
97ef5391 LP |
2400 | bool rm_inotify = false, rm_inode = false; |
2401 | struct inotify_data *new_inotify_data = NULL; | |
2402 | struct inode_data *new_inode_data = NULL; | |
9da4cb2b LP |
2403 | int r; |
2404 | ||
305f78bf | 2405 | assert_return(s, -EINVAL); |
da7e457c | 2406 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2407 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2408 | |
2409 | if (s->priority == priority) | |
2410 | return 0; | |
2411 | ||
97ef5391 LP |
2412 | if (s->type == SOURCE_INOTIFY) { |
2413 | struct inode_data *old_inode_data; | |
2414 | ||
2415 | assert(s->inotify.inode_data); | |
2416 | old_inode_data = s->inotify.inode_data; | |
2417 | ||
2418 | /* We need the original fd to change the priority. If we don't have it we can't change the priority, | |
2419 | * anymore. Note that we close any fds when entering the next event loop iteration, i.e. for inotify | |
2420 | * events we allow priority changes only until the first following iteration. */ | |
2421 | if (old_inode_data->fd < 0) | |
2422 | return -EOPNOTSUPP; | |
2423 | ||
2424 | r = event_make_inotify_data(s->event, priority, &new_inotify_data); | |
2425 | if (r < 0) | |
2426 | return r; | |
2427 | rm_inotify = r > 0; | |
2428 | ||
2429 | r = event_make_inode_data(s->event, new_inotify_data, old_inode_data->dev, old_inode_data->ino, &new_inode_data); | |
2430 | if (r < 0) | |
2431 | goto fail; | |
2432 | rm_inode = r > 0; | |
2433 | ||
2434 | if (new_inode_data->fd < 0) { | |
2435 | /* Duplicate the fd for the new inode object if we don't have any yet */ | |
2436 | new_inode_data->fd = fcntl(old_inode_data->fd, F_DUPFD_CLOEXEC, 3); | |
2437 | if (new_inode_data->fd < 0) { | |
2438 | r = -errno; | |
2439 | goto fail; | |
2440 | } | |
2441 | ||
ed828563 | 2442 | LIST_PREPEND(to_close, s->event->inode_data_to_close_list, new_inode_data); |
97ef5391 LP |
2443 | } |
2444 | ||
2445 | /* Move the event source to the new inode data structure */ | |
2446 | LIST_REMOVE(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2447 | LIST_PREPEND(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2448 | s->inotify.inode_data = new_inode_data; | |
2449 | ||
2450 | /* Now create the new watch */ | |
2451 | r = inode_data_realize_watch(s->event, new_inode_data); | |
2452 | if (r < 0) { | |
2453 | /* Move it back */ | |
2454 | LIST_REMOVE(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2455 | LIST_PREPEND(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2456 | s->inotify.inode_data = old_inode_data; | |
2457 | goto fail; | |
2458 | } | |
2459 | ||
2460 | s->priority = priority; | |
2461 | ||
2462 | event_gc_inode_data(s->event, old_inode_data); | |
2463 | ||
b6d5481b | 2464 | } else if (s->type == SOURCE_SIGNAL && event_source_is_online(s)) { |
9da4cb2b LP |
2465 | struct signal_data *old, *d; |
2466 | ||
2467 | /* Move us from the signalfd belonging to the old | |
2468 | * priority to the signalfd of the new priority */ | |
2469 | ||
2470 | assert_se(old = hashmap_get(s->event->signal_data, &s->priority)); | |
2471 | ||
2472 | s->priority = priority; | |
2473 | ||
2474 | r = event_make_signal_data(s->event, s->signal.sig, &d); | |
2475 | if (r < 0) { | |
2476 | s->priority = old->priority; | |
2477 | return r; | |
2478 | } | |
2479 | ||
2480 | event_unmask_signal_data(s->event, old, s->signal.sig); | |
2481 | } else | |
2482 | s->priority = priority; | |
fd38203a | 2483 | |
e1951c16 | 2484 | event_source_pp_prioq_reshuffle(s); |
fd38203a | 2485 | |
6203e07a LP |
2486 | if (s->type == SOURCE_EXIT) |
2487 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
305f78bf | 2488 | |
fd38203a | 2489 | return 0; |
97ef5391 LP |
2490 | |
2491 | fail: | |
2492 | if (rm_inode) | |
2493 | event_free_inode_data(s->event, new_inode_data); | |
2494 | ||
2495 | if (rm_inotify) | |
2496 | event_free_inotify_data(s->event, new_inotify_data); | |
2497 | ||
2498 | return r; | |
fd38203a LP |
2499 | } |
2500 | ||
cad143a8 | 2501 | _public_ int sd_event_source_get_enabled(sd_event_source *s, int *ret) { |
71193c0b ZJS |
2502 | /* Quick mode: the event source doesn't exist and we only want to query boolean enablement state. */ |
2503 | if (!s && !ret) | |
2504 | return false; | |
2505 | ||
305f78bf | 2506 | assert_return(s, -EINVAL); |
305f78bf | 2507 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2508 | |
cad143a8 LP |
2509 | if (ret) |
2510 | *ret = s->enabled; | |
2511 | ||
08c1eb0e | 2512 | return s->enabled != SD_EVENT_OFF; |
fd38203a LP |
2513 | } |
2514 | ||
b6d5481b LP |
2515 | static int event_source_offline( |
2516 | sd_event_source *s, | |
2517 | int enabled, | |
2518 | bool ratelimited) { | |
2519 | ||
2520 | bool was_offline; | |
fd38203a LP |
2521 | int r; |
2522 | ||
ddfde737 | 2523 | assert(s); |
b6d5481b | 2524 | assert(enabled == SD_EVENT_OFF || ratelimited); |
fd38203a | 2525 | |
ddfde737 | 2526 | /* Unset the pending flag when this event source is disabled */ |
b6d5481b LP |
2527 | if (s->enabled != SD_EVENT_OFF && |
2528 | enabled == SD_EVENT_OFF && | |
2529 | !IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
ddfde737 LP |
2530 | r = source_set_pending(s, false); |
2531 | if (r < 0) | |
2532 | return r; | |
2533 | } | |
cc567911 | 2534 | |
b6d5481b LP |
2535 | was_offline = event_source_is_offline(s); |
2536 | s->enabled = enabled; | |
2537 | s->ratelimited = ratelimited; | |
fd38203a | 2538 | |
ddfde737 | 2539 | switch (s->type) { |
fd38203a | 2540 | |
ddfde737 LP |
2541 | case SOURCE_IO: |
2542 | source_io_unregister(s); | |
2543 | break; | |
ac989a78 | 2544 | |
ddfde737 LP |
2545 | case SOURCE_SIGNAL: |
2546 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); | |
2547 | break; | |
fd38203a | 2548 | |
ddfde737 | 2549 | case SOURCE_CHILD: |
b6d5481b LP |
2550 | if (!was_offline) { |
2551 | assert(s->event->n_online_child_sources > 0); | |
2552 | s->event->n_online_child_sources--; | |
2553 | } | |
fd38203a | 2554 | |
ddfde737 LP |
2555 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
2556 | source_child_pidfd_unregister(s); | |
2557 | else | |
2558 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
2559 | break; | |
4807d2d0 | 2560 | |
ddfde737 LP |
2561 | case SOURCE_EXIT: |
2562 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
2563 | break; | |
fd38203a | 2564 | |
2115b9b6 YW |
2565 | case SOURCE_TIME_REALTIME: |
2566 | case SOURCE_TIME_BOOTTIME: | |
2567 | case SOURCE_TIME_MONOTONIC: | |
2568 | case SOURCE_TIME_REALTIME_ALARM: | |
2569 | case SOURCE_TIME_BOOTTIME_ALARM: | |
ddfde737 LP |
2570 | case SOURCE_DEFER: |
2571 | case SOURCE_POST: | |
2572 | case SOURCE_INOTIFY: | |
2573 | break; | |
fd38203a | 2574 | |
ddfde737 | 2575 | default: |
04499a70 | 2576 | assert_not_reached(); |
ddfde737 | 2577 | } |
fd38203a | 2578 | |
2115b9b6 YW |
2579 | /* Always reshuffle time prioq, as the ratelimited flag may be changed. */ |
2580 | event_source_time_prioq_reshuffle(s); | |
2581 | ||
b6d5481b | 2582 | return 1; |
ddfde737 | 2583 | } |
f8f3f926 | 2584 | |
b6d5481b LP |
2585 | static int event_source_online( |
2586 | sd_event_source *s, | |
2587 | int enabled, | |
2588 | bool ratelimited) { | |
2589 | ||
2590 | bool was_online; | |
ddfde737 | 2591 | int r; |
fd38203a | 2592 | |
ddfde737 | 2593 | assert(s); |
b6d5481b | 2594 | assert(enabled != SD_EVENT_OFF || !ratelimited); |
305f78bf | 2595 | |
ddfde737 | 2596 | /* Unset the pending flag when this event source is enabled */ |
b6d5481b LP |
2597 | if (s->enabled == SD_EVENT_OFF && |
2598 | enabled != SD_EVENT_OFF && | |
2599 | !IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
ddfde737 LP |
2600 | r = source_set_pending(s, false); |
2601 | if (r < 0) | |
2602 | return r; | |
2603 | } | |
9d3e3aa5 | 2604 | |
b6d5481b LP |
2605 | /* Are we really ready for onlining? */ |
2606 | if (enabled == SD_EVENT_OFF || ratelimited) { | |
2607 | /* Nope, we are not ready for onlining, then just update the precise state and exit */ | |
2608 | s->enabled = enabled; | |
2609 | s->ratelimited = ratelimited; | |
2610 | return 0; | |
2611 | } | |
2612 | ||
2613 | was_online = event_source_is_online(s); | |
2614 | ||
ddfde737 | 2615 | switch (s->type) { |
ddfde737 | 2616 | case SOURCE_IO: |
b6d5481b | 2617 | r = source_io_register(s, enabled, s->io.events); |
d2eafe61 | 2618 | if (r < 0) |
ddfde737 | 2619 | return r; |
ddfde737 | 2620 | break; |
fd38203a | 2621 | |
ddfde737 LP |
2622 | case SOURCE_SIGNAL: |
2623 | r = event_make_signal_data(s->event, s->signal.sig, NULL); | |
2624 | if (r < 0) { | |
ddfde737 LP |
2625 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
2626 | return r; | |
2627 | } | |
fd38203a | 2628 | |
ddfde737 | 2629 | break; |
fd38203a | 2630 | |
ddfde737 | 2631 | case SOURCE_CHILD: |
ddfde737 LP |
2632 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
2633 | /* yes, we have pidfd */ | |
9da4cb2b | 2634 | |
b6d5481b | 2635 | r = source_child_pidfd_register(s, enabled); |
ac9f2640 | 2636 | if (r < 0) |
9da4cb2b | 2637 | return r; |
ddfde737 LP |
2638 | } else { |
2639 | /* no pidfd, or something other to watch for than WEXITED */ | |
9da4cb2b | 2640 | |
ddfde737 LP |
2641 | r = event_make_signal_data(s->event, SIGCHLD, NULL); |
2642 | if (r < 0) { | |
ddfde737 LP |
2643 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); |
2644 | return r; | |
2645 | } | |
2646 | } | |
fd38203a | 2647 | |
b6d5481b LP |
2648 | if (!was_online) |
2649 | s->event->n_online_child_sources++; | |
ddfde737 | 2650 | break; |
4807d2d0 | 2651 | |
d2eafe61 ZJS |
2652 | case SOURCE_TIME_REALTIME: |
2653 | case SOURCE_TIME_BOOTTIME: | |
2654 | case SOURCE_TIME_MONOTONIC: | |
2655 | case SOURCE_TIME_REALTIME_ALARM: | |
2656 | case SOURCE_TIME_BOOTTIME_ALARM: | |
ddfde737 | 2657 | case SOURCE_EXIT: |
ddfde737 LP |
2658 | case SOURCE_DEFER: |
2659 | case SOURCE_POST: | |
2660 | case SOURCE_INOTIFY: | |
2661 | break; | |
9da4cb2b | 2662 | |
ddfde737 | 2663 | default: |
04499a70 | 2664 | assert_not_reached(); |
ddfde737 | 2665 | } |
f8f3f926 | 2666 | |
b6d5481b LP |
2667 | s->enabled = enabled; |
2668 | s->ratelimited = ratelimited; | |
d2eafe61 ZJS |
2669 | |
2670 | /* Non-failing operations below */ | |
2115b9b6 | 2671 | if (s->type == SOURCE_EXIT) |
d2eafe61 | 2672 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); |
d2eafe61 | 2673 | |
2115b9b6 YW |
2674 | /* Always reshuffle time prioq, as the ratelimited flag may be changed. */ |
2675 | event_source_time_prioq_reshuffle(s); | |
d2eafe61 | 2676 | |
b6d5481b | 2677 | return 1; |
ddfde737 LP |
2678 | } |
2679 | ||
2680 | _public_ int sd_event_source_set_enabled(sd_event_source *s, int m) { | |
2681 | int r; | |
9da4cb2b | 2682 | |
ddfde737 | 2683 | assert_return(IN_SET(m, SD_EVENT_OFF, SD_EVENT_ON, SD_EVENT_ONESHOT), -EINVAL); |
7e922b05 ZJS |
2684 | |
2685 | /* Quick mode: if the source doesn't exist, SD_EVENT_OFF is a noop. */ | |
2686 | if (m == SD_EVENT_OFF && !s) | |
2687 | return 0; | |
2688 | ||
2689 | assert_return(s, -EINVAL); | |
ddfde737 | 2690 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2691 | |
ddfde737 LP |
2692 | /* If we are dead anyway, we are fine with turning off sources, but everything else needs to fail. */ |
2693 | if (s->event->state == SD_EVENT_FINISHED) | |
2694 | return m == SD_EVENT_OFF ? 0 : -ESTALE; | |
305f78bf | 2695 | |
ddfde737 LP |
2696 | if (s->enabled == m) /* No change? */ |
2697 | return 0; | |
9d3e3aa5 | 2698 | |
ddfde737 | 2699 | if (m == SD_EVENT_OFF) |
b6d5481b | 2700 | r = event_source_offline(s, m, s->ratelimited); |
ddfde737 LP |
2701 | else { |
2702 | if (s->enabled != SD_EVENT_OFF) { | |
2703 | /* Switching from "on" to "oneshot" or back? If that's the case, we can take a shortcut, the | |
2704 | * event source is already enabled after all. */ | |
2705 | s->enabled = m; | |
2706 | return 0; | |
fd38203a | 2707 | } |
ddfde737 | 2708 | |
b6d5481b | 2709 | r = event_source_online(s, m, s->ratelimited); |
fd38203a | 2710 | } |
ddfde737 LP |
2711 | if (r < 0) |
2712 | return r; | |
fd38203a | 2713 | |
e1951c16 | 2714 | event_source_pp_prioq_reshuffle(s); |
fd38203a LP |
2715 | return 0; |
2716 | } | |
2717 | ||
f7262a9f | 2718 | _public_ int sd_event_source_get_time(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
2719 | assert_return(s, -EINVAL); |
2720 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 2721 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf | 2722 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2723 | |
2724 | *usec = s->time.next; | |
2725 | return 0; | |
2726 | } | |
2727 | ||
f7262a9f | 2728 | _public_ int sd_event_source_set_time(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 2729 | int r; |
6a0f1f6d | 2730 | |
305f78bf | 2731 | assert_return(s, -EINVAL); |
6a0f1f6d | 2732 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 2733 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2734 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2735 | |
2a0dc6cd LP |
2736 | r = source_set_pending(s, false); |
2737 | if (r < 0) | |
2738 | return r; | |
2576a19e | 2739 | |
2a0dc6cd | 2740 | s->time.next = usec; |
fd38203a | 2741 | |
e1951c16 | 2742 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
2743 | return 0; |
2744 | } | |
2745 | ||
d6a83dc4 LP |
2746 | _public_ int sd_event_source_set_time_relative(sd_event_source *s, uint64_t usec) { |
2747 | usec_t t; | |
2748 | int r; | |
2749 | ||
2750 | assert_return(s, -EINVAL); | |
2751 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2752 | ||
ef859195 LP |
2753 | if (usec == USEC_INFINITY) |
2754 | return sd_event_source_set_time(s, USEC_INFINITY); | |
2755 | ||
d6a83dc4 LP |
2756 | r = sd_event_now(s->event, event_source_type_to_clock(s->type), &t); |
2757 | if (r < 0) | |
2758 | return r; | |
2759 | ||
496db330 YW |
2760 | usec = usec_add(t, usec); |
2761 | if (usec == USEC_INFINITY) | |
d6a83dc4 LP |
2762 | return -EOVERFLOW; |
2763 | ||
496db330 | 2764 | return sd_event_source_set_time(s, usec); |
d6a83dc4 LP |
2765 | } |
2766 | ||
f7262a9f | 2767 | _public_ int sd_event_source_get_time_accuracy(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
2768 | assert_return(s, -EINVAL); |
2769 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 2770 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf LP |
2771 | assert_return(!event_pid_changed(s->event), -ECHILD); |
2772 | ||
2773 | *usec = s->time.accuracy; | |
2774 | return 0; | |
2775 | } | |
2776 | ||
f7262a9f | 2777 | _public_ int sd_event_source_set_time_accuracy(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 2778 | int r; |
6a0f1f6d | 2779 | |
305f78bf | 2780 | assert_return(s, -EINVAL); |
f5fbe71d | 2781 | assert_return(usec != UINT64_MAX, -EINVAL); |
6a0f1f6d | 2782 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 2783 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2784 | assert_return(!event_pid_changed(s->event), -ECHILD); |
eaa3cbef | 2785 | |
2a0dc6cd LP |
2786 | r = source_set_pending(s, false); |
2787 | if (r < 0) | |
2788 | return r; | |
2789 | ||
eaa3cbef LP |
2790 | if (usec == 0) |
2791 | usec = DEFAULT_ACCURACY_USEC; | |
2792 | ||
eaa3cbef LP |
2793 | s->time.accuracy = usec; |
2794 | ||
e1951c16 | 2795 | event_source_time_prioq_reshuffle(s); |
6a0f1f6d LP |
2796 | return 0; |
2797 | } | |
2798 | ||
2799 | _public_ int sd_event_source_get_time_clock(sd_event_source *s, clockid_t *clock) { | |
2800 | assert_return(s, -EINVAL); | |
2801 | assert_return(clock, -EINVAL); | |
2802 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2803 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
eaa3cbef | 2804 | |
6a0f1f6d | 2805 | *clock = event_source_type_to_clock(s->type); |
eaa3cbef LP |
2806 | return 0; |
2807 | } | |
2808 | ||
f7262a9f | 2809 | _public_ int sd_event_source_get_child_pid(sd_event_source *s, pid_t *pid) { |
4bee8012 LP |
2810 | assert_return(s, -EINVAL); |
2811 | assert_return(pid, -EINVAL); | |
2812 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2813 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2814 | ||
2815 | *pid = s->child.pid; | |
2816 | return 0; | |
2817 | } | |
2818 | ||
f8f3f926 LP |
2819 | _public_ int sd_event_source_get_child_pidfd(sd_event_source *s) { |
2820 | assert_return(s, -EINVAL); | |
2821 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2822 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2823 | ||
2824 | if (s->child.pidfd < 0) | |
2825 | return -EOPNOTSUPP; | |
2826 | ||
2827 | return s->child.pidfd; | |
2828 | } | |
2829 | ||
2830 | _public_ int sd_event_source_send_child_signal(sd_event_source *s, int sig, const siginfo_t *si, unsigned flags) { | |
2831 | assert_return(s, -EINVAL); | |
2832 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2833 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2834 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
2835 | ||
2836 | /* If we already have seen indication the process exited refuse sending a signal early. This way we | |
2837 | * can be sure we don't accidentally kill the wrong process on PID reuse when pidfds are not | |
2838 | * available. */ | |
2839 | if (s->child.exited) | |
2840 | return -ESRCH; | |
2841 | ||
2842 | if (s->child.pidfd >= 0) { | |
2843 | siginfo_t copy; | |
2844 | ||
2845 | /* pidfd_send_signal() changes the siginfo_t argument. This is weird, let's hence copy the | |
2846 | * structure here */ | |
2847 | if (si) | |
2848 | copy = *si; | |
2849 | ||
2850 | if (pidfd_send_signal(s->child.pidfd, sig, si ? © : NULL, 0) < 0) { | |
2851 | /* Let's propagate the error only if the system call is not implemented or prohibited */ | |
2852 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
2853 | return -errno; | |
2854 | } else | |
2855 | return 0; | |
2856 | } | |
2857 | ||
2858 | /* Flags are only supported for pidfd_send_signal(), not for rt_sigqueueinfo(), hence let's refuse | |
2859 | * this here. */ | |
2860 | if (flags != 0) | |
2861 | return -EOPNOTSUPP; | |
2862 | ||
2863 | if (si) { | |
2864 | /* We use rt_sigqueueinfo() only if siginfo_t is specified. */ | |
2865 | siginfo_t copy = *si; | |
2866 | ||
2867 | if (rt_sigqueueinfo(s->child.pid, sig, ©) < 0) | |
2868 | return -errno; | |
2869 | } else if (kill(s->child.pid, sig) < 0) | |
2870 | return -errno; | |
2871 | ||
2872 | return 0; | |
2873 | } | |
2874 | ||
2875 | _public_ int sd_event_source_get_child_pidfd_own(sd_event_source *s) { | |
2876 | assert_return(s, -EINVAL); | |
2877 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2878 | ||
2879 | if (s->child.pidfd < 0) | |
2880 | return -EOPNOTSUPP; | |
2881 | ||
2882 | return s->child.pidfd_owned; | |
2883 | } | |
2884 | ||
2885 | _public_ int sd_event_source_set_child_pidfd_own(sd_event_source *s, int own) { | |
2886 | assert_return(s, -EINVAL); | |
2887 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2888 | ||
2889 | if (s->child.pidfd < 0) | |
2890 | return -EOPNOTSUPP; | |
2891 | ||
2892 | s->child.pidfd_owned = own; | |
2893 | return 0; | |
2894 | } | |
2895 | ||
2896 | _public_ int sd_event_source_get_child_process_own(sd_event_source *s) { | |
2897 | assert_return(s, -EINVAL); | |
2898 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2899 | ||
2900 | return s->child.process_owned; | |
2901 | } | |
2902 | ||
2903 | _public_ int sd_event_source_set_child_process_own(sd_event_source *s, int own) { | |
2904 | assert_return(s, -EINVAL); | |
2905 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2906 | ||
2907 | s->child.process_owned = own; | |
2908 | return 0; | |
2909 | } | |
2910 | ||
97ef5391 LP |
2911 | _public_ int sd_event_source_get_inotify_mask(sd_event_source *s, uint32_t *mask) { |
2912 | assert_return(s, -EINVAL); | |
2913 | assert_return(mask, -EINVAL); | |
2914 | assert_return(s->type == SOURCE_INOTIFY, -EDOM); | |
2915 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2916 | ||
2917 | *mask = s->inotify.mask; | |
2918 | return 0; | |
2919 | } | |
2920 | ||
718db961 | 2921 | _public_ int sd_event_source_set_prepare(sd_event_source *s, sd_event_handler_t callback) { |
fd38203a LP |
2922 | int r; |
2923 | ||
da7e457c | 2924 | assert_return(s, -EINVAL); |
6203e07a | 2925 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c LP |
2926 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2927 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2928 | |
2929 | if (s->prepare == callback) | |
2930 | return 0; | |
2931 | ||
2932 | if (callback && s->prepare) { | |
2933 | s->prepare = callback; | |
2934 | return 0; | |
2935 | } | |
2936 | ||
2937 | r = prioq_ensure_allocated(&s->event->prepare, prepare_prioq_compare); | |
2938 | if (r < 0) | |
2939 | return r; | |
2940 | ||
2941 | s->prepare = callback; | |
2942 | ||
2943 | if (callback) { | |
2944 | r = prioq_put(s->event->prepare, s, &s->prepare_index); | |
2945 | if (r < 0) | |
2946 | return r; | |
2947 | } else | |
2948 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
2949 | ||
2950 | return 0; | |
2951 | } | |
2952 | ||
f7262a9f | 2953 | _public_ void* sd_event_source_get_userdata(sd_event_source *s) { |
da7e457c | 2954 | assert_return(s, NULL); |
fd38203a LP |
2955 | |
2956 | return s->userdata; | |
2957 | } | |
2958 | ||
8f726607 LP |
2959 | _public_ void *sd_event_source_set_userdata(sd_event_source *s, void *userdata) { |
2960 | void *ret; | |
2961 | ||
2962 | assert_return(s, NULL); | |
2963 | ||
2964 | ret = s->userdata; | |
2965 | s->userdata = userdata; | |
2966 | ||
2967 | return ret; | |
2968 | } | |
2969 | ||
b6d5481b LP |
2970 | static int event_source_enter_ratelimited(sd_event_source *s) { |
2971 | int r; | |
2972 | ||
2973 | assert(s); | |
2974 | ||
2975 | /* When an event source becomes ratelimited, we place it in the CLOCK_MONOTONIC priority queue, with | |
2976 | * the end of the rate limit time window, much as if it was a timer event source. */ | |
2977 | ||
2978 | if (s->ratelimited) | |
2979 | return 0; /* Already ratelimited, this is a NOP hence */ | |
2980 | ||
2981 | /* Make sure we can install a CLOCK_MONOTONIC event further down. */ | |
2982 | r = setup_clock_data(s->event, &s->event->monotonic, CLOCK_MONOTONIC); | |
2983 | if (r < 0) | |
2984 | return r; | |
2985 | ||
2986 | /* Timer event sources are already using the earliest/latest queues for the timer scheduling. Let's | |
2987 | * first remove them from the prioq appropriate for their own clock, so that we can use the prioq | |
2988 | * fields of the event source then for adding it to the CLOCK_MONOTONIC prioq instead. */ | |
2989 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
2990 | event_source_time_prioq_remove(s, event_get_clock_data(s->event, s->type)); | |
2991 | ||
2992 | /* Now, let's add the event source to the monotonic clock instead */ | |
2993 | r = event_source_time_prioq_put(s, &s->event->monotonic); | |
2994 | if (r < 0) | |
2995 | goto fail; | |
2996 | ||
2997 | /* And let's take the event source officially offline */ | |
2998 | r = event_source_offline(s, s->enabled, /* ratelimited= */ true); | |
2999 | if (r < 0) { | |
3000 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
3001 | goto fail; | |
3002 | } | |
3003 | ||
3004 | event_source_pp_prioq_reshuffle(s); | |
3005 | ||
3006 | log_debug("Event source %p (%s) entered rate limit state.", s, strna(s->description)); | |
3007 | return 0; | |
3008 | ||
3009 | fail: | |
3010 | /* Reinstall time event sources in the priority queue as before. This shouldn't fail, since the queue | |
3011 | * space for it should already be allocated. */ | |
3012 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
3013 | assert_se(event_source_time_prioq_put(s, event_get_clock_data(s->event, s->type)) >= 0); | |
3014 | ||
3015 | return r; | |
3016 | } | |
3017 | ||
fd69f224 | 3018 | static int event_source_leave_ratelimit(sd_event_source *s, bool run_callback) { |
b6d5481b LP |
3019 | int r; |
3020 | ||
3021 | assert(s); | |
3022 | ||
3023 | if (!s->ratelimited) | |
3024 | return 0; | |
3025 | ||
3026 | /* Let's take the event source out of the monotonic prioq first. */ | |
3027 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
3028 | ||
3029 | /* Let's then add the event source to its native clock prioq again — if this is a timer event source */ | |
3030 | if (EVENT_SOURCE_IS_TIME(s->type)) { | |
3031 | r = event_source_time_prioq_put(s, event_get_clock_data(s->event, s->type)); | |
3032 | if (r < 0) | |
3033 | goto fail; | |
3034 | } | |
3035 | ||
3036 | /* Let's try to take it online again. */ | |
3037 | r = event_source_online(s, s->enabled, /* ratelimited= */ false); | |
3038 | if (r < 0) { | |
3039 | /* Do something roughly sensible when this failed: undo the two prioq ops above */ | |
3040 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
3041 | event_source_time_prioq_remove(s, event_get_clock_data(s->event, s->type)); | |
3042 | ||
3043 | goto fail; | |
3044 | } | |
3045 | ||
3046 | event_source_pp_prioq_reshuffle(s); | |
3047 | ratelimit_reset(&s->rate_limit); | |
3048 | ||
3049 | log_debug("Event source %p (%s) left rate limit state.", s, strna(s->description)); | |
fd69f224 MS |
3050 | |
3051 | if (run_callback && s->ratelimit_expire_callback) { | |
3052 | s->dispatching = true; | |
3053 | r = s->ratelimit_expire_callback(s, s->userdata); | |
3054 | s->dispatching = false; | |
3055 | ||
3056 | if (r < 0) { | |
3057 | log_debug_errno(r, "Ratelimit expiry callback of event source %s (type %s) returned error, %s: %m", | |
3058 | strna(s->description), | |
3059 | event_source_type_to_string(s->type), | |
3060 | s->exit_on_failure ? "exiting" : "disabling"); | |
3061 | ||
3062 | if (s->exit_on_failure) | |
3063 | (void) sd_event_exit(s->event, r); | |
3064 | } | |
3065 | ||
3066 | if (s->n_ref == 0) | |
3067 | source_free(s); | |
3068 | else if (r < 0) | |
0a040e64 | 3069 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
fd69f224 MS |
3070 | |
3071 | return 1; | |
3072 | } | |
3073 | ||
b6d5481b LP |
3074 | return 0; |
3075 | ||
3076 | fail: | |
3077 | /* Do something somewhat reasonable when we cannot move an event sources out of ratelimited mode: | |
3078 | * simply put it back in it, maybe we can then process it more successfully next iteration. */ | |
3079 | assert_se(event_source_time_prioq_put(s, &s->event->monotonic) >= 0); | |
3080 | ||
3081 | return r; | |
3082 | } | |
3083 | ||
c2ba3ad6 LP |
3084 | static usec_t sleep_between(sd_event *e, usec_t a, usec_t b) { |
3085 | usec_t c; | |
3086 | assert(e); | |
3087 | assert(a <= b); | |
3088 | ||
3089 | if (a <= 0) | |
3090 | return 0; | |
393003e1 LP |
3091 | if (a >= USEC_INFINITY) |
3092 | return USEC_INFINITY; | |
c2ba3ad6 LP |
3093 | |
3094 | if (b <= a + 1) | |
3095 | return a; | |
3096 | ||
52444dc4 LP |
3097 | initialize_perturb(e); |
3098 | ||
c2ba3ad6 LP |
3099 | /* |
3100 | Find a good time to wake up again between times a and b. We | |
3101 | have two goals here: | |
3102 | ||
3103 | a) We want to wake up as seldom as possible, hence prefer | |
3104 | later times over earlier times. | |
3105 | ||
3106 | b) But if we have to wake up, then let's make sure to | |
3107 | dispatch as much as possible on the entire system. | |
3108 | ||
3109 | We implement this by waking up everywhere at the same time | |
850516e0 | 3110 | within any given minute if we can, synchronised via the |
c2ba3ad6 | 3111 | perturbation value determined from the boot ID. If we can't, |
ba276c81 LP |
3112 | then we try to find the same spot in every 10s, then 1s and |
3113 | then 250ms step. Otherwise, we pick the last possible time | |
3114 | to wake up. | |
c2ba3ad6 LP |
3115 | */ |
3116 | ||
850516e0 LP |
3117 | c = (b / USEC_PER_MINUTE) * USEC_PER_MINUTE + e->perturb; |
3118 | if (c >= b) { | |
3119 | if (_unlikely_(c < USEC_PER_MINUTE)) | |
3120 | return b; | |
3121 | ||
3122 | c -= USEC_PER_MINUTE; | |
3123 | } | |
3124 | ||
ba276c81 LP |
3125 | if (c >= a) |
3126 | return c; | |
3127 | ||
3128 | c = (b / (USEC_PER_SEC*10)) * (USEC_PER_SEC*10) + (e->perturb % (USEC_PER_SEC*10)); | |
3129 | if (c >= b) { | |
3130 | if (_unlikely_(c < USEC_PER_SEC*10)) | |
3131 | return b; | |
3132 | ||
3133 | c -= USEC_PER_SEC*10; | |
3134 | } | |
3135 | ||
850516e0 LP |
3136 | if (c >= a) |
3137 | return c; | |
3138 | ||
3139 | c = (b / USEC_PER_SEC) * USEC_PER_SEC + (e->perturb % USEC_PER_SEC); | |
c2ba3ad6 LP |
3140 | if (c >= b) { |
3141 | if (_unlikely_(c < USEC_PER_SEC)) | |
3142 | return b; | |
3143 | ||
3144 | c -= USEC_PER_SEC; | |
3145 | } | |
3146 | ||
3147 | if (c >= a) | |
3148 | return c; | |
3149 | ||
3150 | c = (b / (USEC_PER_MSEC*250)) * (USEC_PER_MSEC*250) + (e->perturb % (USEC_PER_MSEC*250)); | |
3151 | if (c >= b) { | |
3152 | if (_unlikely_(c < USEC_PER_MSEC*250)) | |
3153 | return b; | |
3154 | ||
3155 | c -= USEC_PER_MSEC*250; | |
3156 | } | |
3157 | ||
3158 | if (c >= a) | |
3159 | return c; | |
3160 | ||
3161 | return b; | |
3162 | } | |
3163 | ||
fd38203a LP |
3164 | static int event_arm_timer( |
3165 | sd_event *e, | |
6a0f1f6d | 3166 | struct clock_data *d) { |
fd38203a LP |
3167 | |
3168 | struct itimerspec its = {}; | |
c2ba3ad6 LP |
3169 | sd_event_source *a, *b; |
3170 | usec_t t; | |
fd38203a | 3171 | |
cde93897 | 3172 | assert(e); |
6a0f1f6d | 3173 | assert(d); |
fd38203a | 3174 | |
d06441da | 3175 | if (!d->needs_rearm) |
212bbb17 | 3176 | return 0; |
7e2bf71c YW |
3177 | |
3178 | d->needs_rearm = false; | |
212bbb17 | 3179 | |
6a0f1f6d | 3180 | a = prioq_peek(d->earliest); |
19947509 | 3181 | assert(!a || EVENT_SOURCE_USES_TIME_PRIOQ(a->type)); |
b6d5481b | 3182 | if (!a || a->enabled == SD_EVENT_OFF || time_event_source_next(a) == USEC_INFINITY) { |
72aedc1e | 3183 | |
6a0f1f6d | 3184 | if (d->fd < 0) |
c57b5ca3 LP |
3185 | return 0; |
3186 | ||
3a43da28 | 3187 | if (d->next == USEC_INFINITY) |
72aedc1e LP |
3188 | return 0; |
3189 | ||
3190 | /* disarm */ | |
15c689d7 LP |
3191 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
3192 | return -errno; | |
72aedc1e | 3193 | |
3a43da28 | 3194 | d->next = USEC_INFINITY; |
fd38203a | 3195 | return 0; |
72aedc1e | 3196 | } |
fd38203a | 3197 | |
6a0f1f6d | 3198 | b = prioq_peek(d->latest); |
19947509 ZJS |
3199 | assert(!b || EVENT_SOURCE_USES_TIME_PRIOQ(b->type)); |
3200 | assert(b && b->enabled != SD_EVENT_OFF); | |
c2ba3ad6 | 3201 | |
b6d5481b | 3202 | t = sleep_between(e, time_event_source_next(a), time_event_source_latest(b)); |
6a0f1f6d | 3203 | if (d->next == t) |
fd38203a LP |
3204 | return 0; |
3205 | ||
6a0f1f6d | 3206 | assert_se(d->fd >= 0); |
fd38203a | 3207 | |
c2ba3ad6 | 3208 | if (t == 0) { |
1751bdde | 3209 | /* We don't want to disarm here, just mean some time looooong ago. */ |
fd38203a LP |
3210 | its.it_value.tv_sec = 0; |
3211 | its.it_value.tv_nsec = 1; | |
3212 | } else | |
c2ba3ad6 | 3213 | timespec_store(&its.it_value, t); |
fd38203a | 3214 | |
15c689d7 | 3215 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 | 3216 | return -errno; |
fd38203a | 3217 | |
6a0f1f6d | 3218 | d->next = t; |
fd38203a LP |
3219 | return 0; |
3220 | } | |
3221 | ||
9a800b56 | 3222 | static int process_io(sd_event *e, sd_event_source *s, uint32_t revents) { |
fd38203a LP |
3223 | assert(e); |
3224 | assert(s); | |
3225 | assert(s->type == SOURCE_IO); | |
3226 | ||
9a800b56 LP |
3227 | /* If the event source was already pending, we just OR in the |
3228 | * new revents, otherwise we reset the value. The ORing is | |
3229 | * necessary to handle EPOLLONESHOT events properly where | |
3230 | * readability might happen independently of writability, and | |
3231 | * we need to keep track of both */ | |
3232 | ||
3233 | if (s->pending) | |
3234 | s->io.revents |= revents; | |
3235 | else | |
3236 | s->io.revents = revents; | |
fd38203a | 3237 | |
fd38203a LP |
3238 | return source_set_pending(s, true); |
3239 | } | |
3240 | ||
72aedc1e | 3241 | static int flush_timer(sd_event *e, int fd, uint32_t events, usec_t *next) { |
fd38203a LP |
3242 | uint64_t x; |
3243 | ssize_t ss; | |
3244 | ||
3245 | assert(e); | |
da7e457c | 3246 | assert(fd >= 0); |
72aedc1e | 3247 | |
305f78bf | 3248 | assert_return(events == EPOLLIN, -EIO); |
fd38203a LP |
3249 | |
3250 | ss = read(fd, &x, sizeof(x)); | |
3251 | if (ss < 0) { | |
8add30a0 | 3252 | if (ERRNO_IS_TRANSIENT(errno)) |
fd38203a LP |
3253 | return 0; |
3254 | ||
3255 | return -errno; | |
3256 | } | |
3257 | ||
8d35dae7 | 3258 | if (_unlikely_(ss != sizeof(x))) |
fd38203a LP |
3259 | return -EIO; |
3260 | ||
cde93897 | 3261 | if (next) |
3a43da28 | 3262 | *next = USEC_INFINITY; |
72aedc1e | 3263 | |
fd38203a LP |
3264 | return 0; |
3265 | } | |
3266 | ||
305f78bf LP |
3267 | static int process_timer( |
3268 | sd_event *e, | |
3269 | usec_t n, | |
6a0f1f6d | 3270 | struct clock_data *d) { |
305f78bf | 3271 | |
fd38203a | 3272 | sd_event_source *s; |
fd69f224 | 3273 | bool callback_invoked = false; |
fd38203a LP |
3274 | int r; |
3275 | ||
3276 | assert(e); | |
6a0f1f6d | 3277 | assert(d); |
fd38203a LP |
3278 | |
3279 | for (;;) { | |
6a0f1f6d | 3280 | s = prioq_peek(d->earliest); |
19947509 ZJS |
3281 | assert(!s || EVENT_SOURCE_USES_TIME_PRIOQ(s->type)); |
3282 | ||
b6d5481b LP |
3283 | if (!s || time_event_source_next(s) > n) |
3284 | break; | |
3285 | ||
3286 | if (s->ratelimited) { | |
3287 | /* This is an event sources whose ratelimit window has ended. Let's turn it on | |
3288 | * again. */ | |
3289 | assert(s->ratelimited); | |
3290 | ||
fd69f224 | 3291 | r = event_source_leave_ratelimit(s, /* run_callback */ true); |
b6d5481b LP |
3292 | if (r < 0) |
3293 | return r; | |
fd69f224 MS |
3294 | else if (r == 1) |
3295 | callback_invoked = true; | |
b6d5481b LP |
3296 | |
3297 | continue; | |
3298 | } | |
3299 | ||
3300 | if (s->enabled == SD_EVENT_OFF || s->pending) | |
fd38203a LP |
3301 | break; |
3302 | ||
3303 | r = source_set_pending(s, true); | |
3304 | if (r < 0) | |
3305 | return r; | |
3306 | ||
e1951c16 | 3307 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
3308 | } |
3309 | ||
fd69f224 | 3310 | return callback_invoked; |
fd38203a LP |
3311 | } |
3312 | ||
efd3be9d YW |
3313 | static int process_child(sd_event *e, int64_t threshold, int64_t *ret_min_priority) { |
3314 | int64_t min_priority = threshold; | |
3315 | bool something_new = false; | |
fd38203a | 3316 | sd_event_source *s; |
fd38203a LP |
3317 | int r; |
3318 | ||
3319 | assert(e); | |
efd3be9d YW |
3320 | assert(ret_min_priority); |
3321 | ||
3322 | if (!e->need_process_child) { | |
3323 | *ret_min_priority = min_priority; | |
3324 | return 0; | |
3325 | } | |
fd38203a | 3326 | |
c2ba3ad6 LP |
3327 | e->need_process_child = false; |
3328 | ||
91c70071 YW |
3329 | /* So, this is ugly. We iteratively invoke waitid() with P_PID + WNOHANG for each PID we wait |
3330 | * for, instead of using P_ALL. This is because we only want to get child information of very | |
3331 | * specific child processes, and not all of them. We might not have processed the SIGCHLD event | |
3332 | * of a previous invocation and we don't want to maintain a unbounded *per-child* event queue, | |
3333 | * hence we really don't want anything flushed out of the kernel's queue that we don't care | |
3334 | * about. Since this is O(n) this means that if you have a lot of processes you probably want | |
3335 | * to handle SIGCHLD yourself. | |
3336 | * | |
3337 | * We do not reap the children here (by using WNOWAIT), this is only done after the event | |
3338 | * source is dispatched so that the callback still sees the process as a zombie. */ | |
fd38203a | 3339 | |
90e74a66 | 3340 | HASHMAP_FOREACH(s, e->child_sources) { |
fd38203a LP |
3341 | assert(s->type == SOURCE_CHILD); |
3342 | ||
efd3be9d YW |
3343 | if (s->priority > threshold) |
3344 | continue; | |
3345 | ||
fd38203a LP |
3346 | if (s->pending) |
3347 | continue; | |
3348 | ||
b6d5481b | 3349 | if (event_source_is_offline(s)) |
fd38203a LP |
3350 | continue; |
3351 | ||
f8f3f926 LP |
3352 | if (s->child.exited) |
3353 | continue; | |
3354 | ||
91c70071 YW |
3355 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
3356 | /* There's a usable pidfd known for this event source? Then don't waitid() for | |
3357 | * it here */ | |
f8f3f926 LP |
3358 | continue; |
3359 | ||
fd38203a | 3360 | zero(s->child.siginfo); |
15c689d7 LP |
3361 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, |
3362 | WNOHANG | (s->child.options & WEXITED ? WNOWAIT : 0) | s->child.options) < 0) | |
bfd9bfcc | 3363 | return negative_errno(); |
fd38203a LP |
3364 | |
3365 | if (s->child.siginfo.si_pid != 0) { | |
945c2931 | 3366 | bool zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 3367 | |
f8f3f926 LP |
3368 | if (zombie) |
3369 | s->child.exited = true; | |
3370 | ||
08cd1552 | 3371 | if (!zombie && (s->child.options & WEXITED)) { |
91c70071 YW |
3372 | /* If the child isn't dead then let's immediately remove the state |
3373 | * change from the queue, since there's no benefit in leaving it | |
3374 | * queued. */ | |
08cd1552 LP |
3375 | |
3376 | assert(s->child.options & (WSTOPPED|WCONTINUED)); | |
a5d27871 | 3377 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|(s->child.options & (WSTOPPED|WCONTINUED))); |
08cd1552 LP |
3378 | } |
3379 | ||
fd38203a LP |
3380 | r = source_set_pending(s, true); |
3381 | if (r < 0) | |
3382 | return r; | |
efd3be9d YW |
3383 | if (r > 0) { |
3384 | something_new = true; | |
3385 | min_priority = MIN(min_priority, s->priority); | |
3386 | } | |
fd38203a LP |
3387 | } |
3388 | } | |
3389 | ||
efd3be9d YW |
3390 | *ret_min_priority = min_priority; |
3391 | return something_new; | |
fd38203a LP |
3392 | } |
3393 | ||
f8f3f926 LP |
3394 | static int process_pidfd(sd_event *e, sd_event_source *s, uint32_t revents) { |
3395 | assert(e); | |
3396 | assert(s); | |
3397 | assert(s->type == SOURCE_CHILD); | |
3398 | ||
3399 | if (s->pending) | |
3400 | return 0; | |
3401 | ||
b6d5481b | 3402 | if (event_source_is_offline(s)) |
f8f3f926 LP |
3403 | return 0; |
3404 | ||
3405 | if (!EVENT_SOURCE_WATCH_PIDFD(s)) | |
3406 | return 0; | |
3407 | ||
3408 | zero(s->child.siginfo); | |
3409 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG | WNOWAIT | s->child.options) < 0) | |
3410 | return -errno; | |
3411 | ||
3412 | if (s->child.siginfo.si_pid == 0) | |
3413 | return 0; | |
3414 | ||
3415 | if (IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) | |
3416 | s->child.exited = true; | |
3417 | ||
3418 | return source_set_pending(s, true); | |
3419 | } | |
3420 | ||
efd3be9d | 3421 | static int process_signal(sd_event *e, struct signal_data *d, uint32_t events, int64_t *min_priority) { |
fd38203a LP |
3422 | int r; |
3423 | ||
da7e457c | 3424 | assert(e); |
97ef5391 | 3425 | assert(d); |
305f78bf | 3426 | assert_return(events == EPOLLIN, -EIO); |
efd3be9d | 3427 | assert(min_priority); |
fd38203a | 3428 | |
91c70071 YW |
3429 | /* If there's a signal queued on this priority and SIGCHLD is on this priority too, then make |
3430 | * sure to recheck the children we watch. This is because we only ever dequeue the first signal | |
3431 | * per priority, and if we dequeue one, and SIGCHLD might be enqueued later we wouldn't know, | |
3432 | * but we might have higher priority children we care about hence we need to check that | |
3433 | * explicitly. */ | |
9da4cb2b LP |
3434 | |
3435 | if (sigismember(&d->sigset, SIGCHLD)) | |
3436 | e->need_process_child = true; | |
3437 | ||
91c70071 | 3438 | /* If there's already an event source pending for this priority we don't read another */ |
9da4cb2b LP |
3439 | if (d->current) |
3440 | return 0; | |
3441 | ||
fd38203a | 3442 | for (;;) { |
0eb2e0e3 | 3443 | struct signalfd_siginfo si; |
7057bd99 | 3444 | ssize_t n; |
92daebc0 | 3445 | sd_event_source *s = NULL; |
fd38203a | 3446 | |
9da4cb2b | 3447 | n = read(d->fd, &si, sizeof(si)); |
7057bd99 | 3448 | if (n < 0) { |
8add30a0 | 3449 | if (ERRNO_IS_TRANSIENT(errno)) |
efd3be9d | 3450 | return 0; |
fd38203a LP |
3451 | |
3452 | return -errno; | |
3453 | } | |
3454 | ||
7057bd99 | 3455 | if (_unlikely_(n != sizeof(si))) |
fd38203a LP |
3456 | return -EIO; |
3457 | ||
6eb7c172 | 3458 | assert(SIGNAL_VALID(si.ssi_signo)); |
7057bd99 | 3459 | |
92daebc0 LP |
3460 | if (e->signal_sources) |
3461 | s = e->signal_sources[si.ssi_signo]; | |
92daebc0 LP |
3462 | if (!s) |
3463 | continue; | |
9da4cb2b LP |
3464 | if (s->pending) |
3465 | continue; | |
fd38203a LP |
3466 | |
3467 | s->signal.siginfo = si; | |
9da4cb2b LP |
3468 | d->current = s; |
3469 | ||
fd38203a LP |
3470 | r = source_set_pending(s, true); |
3471 | if (r < 0) | |
3472 | return r; | |
efd3be9d YW |
3473 | if (r > 0 && *min_priority >= s->priority) { |
3474 | *min_priority = s->priority; | |
3475 | return 1; /* an event source with smaller priority is queued. */ | |
3476 | } | |
9da4cb2b | 3477 | |
efd3be9d | 3478 | return 0; |
fd38203a | 3479 | } |
fd38203a LP |
3480 | } |
3481 | ||
efd3be9d | 3482 | static int event_inotify_data_read(sd_event *e, struct inotify_data *d, uint32_t revents, int64_t threshold) { |
97ef5391 LP |
3483 | ssize_t n; |
3484 | ||
3485 | assert(e); | |
3486 | assert(d); | |
3487 | ||
3488 | assert_return(revents == EPOLLIN, -EIO); | |
3489 | ||
3490 | /* If there's already an event source pending for this priority, don't read another */ | |
3491 | if (d->n_pending > 0) | |
3492 | return 0; | |
3493 | ||
3494 | /* Is the read buffer non-empty? If so, let's not read more */ | |
3495 | if (d->buffer_filled > 0) | |
3496 | return 0; | |
3497 | ||
efd3be9d YW |
3498 | if (d->priority > threshold) |
3499 | return 0; | |
3500 | ||
97ef5391 LP |
3501 | n = read(d->fd, &d->buffer, sizeof(d->buffer)); |
3502 | if (n < 0) { | |
8add30a0 | 3503 | if (ERRNO_IS_TRANSIENT(errno)) |
97ef5391 LP |
3504 | return 0; |
3505 | ||
3506 | return -errno; | |
3507 | } | |
3508 | ||
3509 | assert(n > 0); | |
3510 | d->buffer_filled = (size_t) n; | |
0601b958 | 3511 | LIST_PREPEND(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
3512 | |
3513 | return 1; | |
3514 | } | |
3515 | ||
3516 | static void event_inotify_data_drop(sd_event *e, struct inotify_data *d, size_t sz) { | |
3517 | assert(e); | |
3518 | assert(d); | |
3519 | assert(sz <= d->buffer_filled); | |
3520 | ||
3521 | if (sz == 0) | |
3522 | return; | |
3523 | ||
3524 | /* Move the rest to the buffer to the front, in order to get things properly aligned again */ | |
3525 | memmove(d->buffer.raw, d->buffer.raw + sz, d->buffer_filled - sz); | |
3526 | d->buffer_filled -= sz; | |
3527 | ||
3528 | if (d->buffer_filled == 0) | |
0601b958 | 3529 | LIST_REMOVE(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
3530 | } |
3531 | ||
3532 | static int event_inotify_data_process(sd_event *e, struct inotify_data *d) { | |
3533 | int r; | |
3534 | ||
3535 | assert(e); | |
3536 | assert(d); | |
3537 | ||
3538 | /* If there's already an event source pending for this priority, don't read another */ | |
3539 | if (d->n_pending > 0) | |
3540 | return 0; | |
3541 | ||
3542 | while (d->buffer_filled > 0) { | |
3543 | size_t sz; | |
3544 | ||
3545 | /* Let's validate that the event structures are complete */ | |
3546 | if (d->buffer_filled < offsetof(struct inotify_event, name)) | |
3547 | return -EIO; | |
3548 | ||
3549 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3550 | if (d->buffer_filled < sz) | |
3551 | return -EIO; | |
3552 | ||
3553 | if (d->buffer.ev.mask & IN_Q_OVERFLOW) { | |
3554 | struct inode_data *inode_data; | |
97ef5391 LP |
3555 | |
3556 | /* The queue overran, let's pass this event to all event sources connected to this inotify | |
3557 | * object */ | |
3558 | ||
03677889 | 3559 | HASHMAP_FOREACH(inode_data, d->inodes) |
97ef5391 LP |
3560 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { |
3561 | ||
b6d5481b | 3562 | if (event_source_is_offline(s)) |
97ef5391 LP |
3563 | continue; |
3564 | ||
3565 | r = source_set_pending(s, true); | |
3566 | if (r < 0) | |
3567 | return r; | |
3568 | } | |
97ef5391 LP |
3569 | } else { |
3570 | struct inode_data *inode_data; | |
97ef5391 LP |
3571 | |
3572 | /* Find the inode object for this watch descriptor. If IN_IGNORED is set we also remove it from | |
3573 | * our watch descriptor table. */ | |
3574 | if (d->buffer.ev.mask & IN_IGNORED) { | |
3575 | ||
3576 | inode_data = hashmap_remove(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3577 | if (!inode_data) { | |
3578 | event_inotify_data_drop(e, d, sz); | |
3579 | continue; | |
3580 | } | |
3581 | ||
3582 | /* The watch descriptor was removed by the kernel, let's drop it here too */ | |
3583 | inode_data->wd = -1; | |
3584 | } else { | |
3585 | inode_data = hashmap_get(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3586 | if (!inode_data) { | |
3587 | event_inotify_data_drop(e, d, sz); | |
3588 | continue; | |
3589 | } | |
3590 | } | |
3591 | ||
3592 | /* Trigger all event sources that are interested in these events. Also trigger all event | |
3593 | * sources if IN_IGNORED or IN_UNMOUNT is set. */ | |
3594 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
3595 | ||
b6d5481b | 3596 | if (event_source_is_offline(s)) |
97ef5391 LP |
3597 | continue; |
3598 | ||
3599 | if ((d->buffer.ev.mask & (IN_IGNORED|IN_UNMOUNT)) == 0 && | |
3600 | (s->inotify.mask & d->buffer.ev.mask & IN_ALL_EVENTS) == 0) | |
3601 | continue; | |
3602 | ||
3603 | r = source_set_pending(s, true); | |
3604 | if (r < 0) | |
3605 | return r; | |
3606 | } | |
3607 | } | |
3608 | ||
3609 | /* Something pending now? If so, let's finish, otherwise let's read more. */ | |
3610 | if (d->n_pending > 0) | |
3611 | return 1; | |
3612 | } | |
3613 | ||
3614 | return 0; | |
3615 | } | |
3616 | ||
3617 | static int process_inotify(sd_event *e) { | |
97ef5391 LP |
3618 | int r, done = 0; |
3619 | ||
3620 | assert(e); | |
3621 | ||
0601b958 | 3622 | LIST_FOREACH(buffered, d, e->buffered_inotify_data_list) { |
97ef5391 LP |
3623 | r = event_inotify_data_process(e, d); |
3624 | if (r < 0) | |
3625 | return r; | |
3626 | if (r > 0) | |
3627 | done ++; | |
3628 | } | |
3629 | ||
3630 | return done; | |
3631 | } | |
3632 | ||
fd38203a | 3633 | static int source_dispatch(sd_event_source *s) { |
8f5c235d | 3634 | EventSourceType saved_type; |
c8e9d15c | 3635 | sd_event *saved_event; |
fe8245eb | 3636 | int r = 0; |
fd38203a LP |
3637 | |
3638 | assert(s); | |
6203e07a | 3639 | assert(s->pending || s->type == SOURCE_EXIT); |
fd38203a | 3640 | |
b778cba4 LP |
3641 | /* Save the event source type, here, so that we still know it after the event callback which might |
3642 | * invalidate the event. */ | |
8f5c235d LP |
3643 | saved_type = s->type; |
3644 | ||
de02634c | 3645 | /* Similarly, store a reference to the event loop object, so that we can still access it after the |
b778cba4 | 3646 | * callback might have invalidated/disconnected the event source. */ |
c8e9d15c YW |
3647 | saved_event = s->event; |
3648 | PROTECT_EVENT(saved_event); | |
b778cba4 | 3649 | |
de02634c | 3650 | /* Check if we hit the ratelimit for this event source, and if so, let's disable it. */ |
b6d5481b LP |
3651 | assert(!s->ratelimited); |
3652 | if (!ratelimit_below(&s->rate_limit)) { | |
3653 | r = event_source_enter_ratelimited(s); | |
3654 | if (r < 0) | |
3655 | return r; | |
3656 | ||
3657 | return 1; | |
3658 | } | |
3659 | ||
945c2931 | 3660 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { |
da7e457c LP |
3661 | r = source_set_pending(s, false); |
3662 | if (r < 0) | |
3663 | return r; | |
3664 | } | |
fd38203a | 3665 | |
6e9feda3 LP |
3666 | if (s->type != SOURCE_POST) { |
3667 | sd_event_source *z; | |
6e9feda3 | 3668 | |
de02634c | 3669 | /* If we execute a non-post source, let's mark all post sources as pending. */ |
6e9feda3 | 3670 | |
90e74a66 | 3671 | SET_FOREACH(z, s->event->post_sources) { |
b6d5481b | 3672 | if (event_source_is_offline(z)) |
6e9feda3 LP |
3673 | continue; |
3674 | ||
3675 | r = source_set_pending(z, true); | |
3676 | if (r < 0) | |
3677 | return r; | |
3678 | } | |
3679 | } | |
3680 | ||
baf76283 LP |
3681 | if (s->enabled == SD_EVENT_ONESHOT) { |
3682 | r = sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
3683 | if (r < 0) |
3684 | return r; | |
3685 | } | |
3686 | ||
12179984 | 3687 | s->dispatching = true; |
b7484e2a | 3688 | |
fd38203a LP |
3689 | switch (s->type) { |
3690 | ||
3691 | case SOURCE_IO: | |
3692 | r = s->io.callback(s, s->io.fd, s->io.revents, s->userdata); | |
3693 | break; | |
3694 | ||
6a0f1f6d | 3695 | case SOURCE_TIME_REALTIME: |
a8548816 | 3696 | case SOURCE_TIME_BOOTTIME: |
6a0f1f6d LP |
3697 | case SOURCE_TIME_MONOTONIC: |
3698 | case SOURCE_TIME_REALTIME_ALARM: | |
3699 | case SOURCE_TIME_BOOTTIME_ALARM: | |
fd38203a LP |
3700 | r = s->time.callback(s, s->time.next, s->userdata); |
3701 | break; | |
3702 | ||
3703 | case SOURCE_SIGNAL: | |
3704 | r = s->signal.callback(s, &s->signal.siginfo, s->userdata); | |
3705 | break; | |
3706 | ||
08cd1552 LP |
3707 | case SOURCE_CHILD: { |
3708 | bool zombie; | |
3709 | ||
945c2931 | 3710 | zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 3711 | |
fd38203a | 3712 | r = s->child.callback(s, &s->child.siginfo, s->userdata); |
08cd1552 LP |
3713 | |
3714 | /* Now, reap the PID for good. */ | |
f8f3f926 | 3715 | if (zombie) { |
cc59d290 | 3716 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|WEXITED); |
f8f3f926 LP |
3717 | s->child.waited = true; |
3718 | } | |
08cd1552 | 3719 | |
fd38203a | 3720 | break; |
08cd1552 | 3721 | } |
fd38203a LP |
3722 | |
3723 | case SOURCE_DEFER: | |
3724 | r = s->defer.callback(s, s->userdata); | |
3725 | break; | |
da7e457c | 3726 | |
6e9feda3 LP |
3727 | case SOURCE_POST: |
3728 | r = s->post.callback(s, s->userdata); | |
3729 | break; | |
3730 | ||
6203e07a LP |
3731 | case SOURCE_EXIT: |
3732 | r = s->exit.callback(s, s->userdata); | |
da7e457c | 3733 | break; |
9d3e3aa5 | 3734 | |
97ef5391 LP |
3735 | case SOURCE_INOTIFY: { |
3736 | struct sd_event *e = s->event; | |
3737 | struct inotify_data *d; | |
3738 | size_t sz; | |
3739 | ||
3740 | assert(s->inotify.inode_data); | |
3741 | assert_se(d = s->inotify.inode_data->inotify_data); | |
3742 | ||
3743 | assert(d->buffer_filled >= offsetof(struct inotify_event, name)); | |
3744 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3745 | assert(d->buffer_filled >= sz); | |
3746 | ||
53baf2ef LP |
3747 | /* If the inotify callback destroys the event source then this likely means we don't need to |
3748 | * watch the inode anymore, and thus also won't need the inotify object anymore. But if we'd | |
3749 | * free it immediately, then we couldn't drop the event from the inotify event queue without | |
3750 | * memory corruption anymore, as below. Hence, let's not free it immediately, but mark it | |
3751 | * "busy" with a counter (which will ensure it's not GC'ed away prematurely). Let's then | |
3752 | * explicitly GC it after we are done dropping the inotify event from the buffer. */ | |
3753 | d->n_busy++; | |
97ef5391 | 3754 | r = s->inotify.callback(s, &d->buffer.ev, s->userdata); |
53baf2ef | 3755 | d->n_busy--; |
97ef5391 | 3756 | |
53baf2ef LP |
3757 | /* When no event is pending anymore on this inotify object, then let's drop the event from |
3758 | * the inotify event queue buffer. */ | |
97ef5391 LP |
3759 | if (d->n_pending == 0) |
3760 | event_inotify_data_drop(e, d, sz); | |
3761 | ||
53baf2ef LP |
3762 | /* Now we don't want to access 'd' anymore, it's OK to GC now. */ |
3763 | event_gc_inotify_data(e, d); | |
97ef5391 LP |
3764 | break; |
3765 | } | |
3766 | ||
9d3e3aa5 | 3767 | case SOURCE_WATCHDOG: |
a71fe8b8 | 3768 | case _SOURCE_EVENT_SOURCE_TYPE_MAX: |
9f2a50a3 | 3769 | case _SOURCE_EVENT_SOURCE_TYPE_INVALID: |
04499a70 | 3770 | assert_not_reached(); |
fd38203a LP |
3771 | } |
3772 | ||
12179984 LP |
3773 | s->dispatching = false; |
3774 | ||
b778cba4 LP |
3775 | if (r < 0) { |
3776 | log_debug_errno(r, "Event source %s (type %s) returned error, %s: %m", | |
3777 | strna(s->description), | |
3778 | event_source_type_to_string(saved_type), | |
3779 | s->exit_on_failure ? "exiting" : "disabling"); | |
3780 | ||
3781 | if (s->exit_on_failure) | |
3782 | (void) sd_event_exit(saved_event, r); | |
3783 | } | |
12179984 LP |
3784 | |
3785 | if (s->n_ref == 0) | |
3786 | source_free(s); | |
3787 | else if (r < 0) | |
c3c50474 | 3788 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
b7484e2a | 3789 | |
6203e07a | 3790 | return 1; |
fd38203a LP |
3791 | } |
3792 | ||
3793 | static int event_prepare(sd_event *e) { | |
3794 | int r; | |
3795 | ||
3796 | assert(e); | |
3797 | ||
3798 | for (;;) { | |
3799 | sd_event_source *s; | |
3800 | ||
3801 | s = prioq_peek(e->prepare); | |
b6d5481b | 3802 | if (!s || s->prepare_iteration == e->iteration || event_source_is_offline(s)) |
fd38203a LP |
3803 | break; |
3804 | ||
3805 | s->prepare_iteration = e->iteration; | |
8656f4a6 | 3806 | prioq_reshuffle(e->prepare, s, &s->prepare_index); |
fd38203a LP |
3807 | |
3808 | assert(s->prepare); | |
12179984 | 3809 | s->dispatching = true; |
fd38203a | 3810 | r = s->prepare(s, s->userdata); |
12179984 LP |
3811 | s->dispatching = false; |
3812 | ||
b778cba4 LP |
3813 | if (r < 0) { |
3814 | log_debug_errno(r, "Prepare callback of event source %s (type %s) returned error, %s: %m", | |
3815 | strna(s->description), | |
3816 | event_source_type_to_string(s->type), | |
3817 | s->exit_on_failure ? "exiting" : "disabling"); | |
3818 | ||
3819 | if (s->exit_on_failure) | |
3820 | (void) sd_event_exit(e, r); | |
3821 | } | |
fd38203a | 3822 | |
12179984 LP |
3823 | if (s->n_ref == 0) |
3824 | source_free(s); | |
3825 | else if (r < 0) | |
c3c50474 | 3826 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
fd38203a LP |
3827 | } |
3828 | ||
3829 | return 0; | |
3830 | } | |
3831 | ||
6203e07a | 3832 | static int dispatch_exit(sd_event *e) { |
da7e457c LP |
3833 | sd_event_source *p; |
3834 | int r; | |
3835 | ||
3836 | assert(e); | |
3837 | ||
6203e07a | 3838 | p = prioq_peek(e->exit); |
19947509 ZJS |
3839 | assert(!p || p->type == SOURCE_EXIT); |
3840 | ||
b6d5481b | 3841 | if (!p || event_source_is_offline(p)) { |
da7e457c LP |
3842 | e->state = SD_EVENT_FINISHED; |
3843 | return 0; | |
3844 | } | |
3845 | ||
c8e9d15c | 3846 | PROTECT_EVENT(e); |
da7e457c | 3847 | e->iteration++; |
6203e07a | 3848 | e->state = SD_EVENT_EXITING; |
da7e457c | 3849 | r = source_dispatch(p); |
2b0c9ef7 | 3850 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
3851 | return r; |
3852 | } | |
3853 | ||
c2ba3ad6 LP |
3854 | static sd_event_source* event_next_pending(sd_event *e) { |
3855 | sd_event_source *p; | |
3856 | ||
da7e457c LP |
3857 | assert(e); |
3858 | ||
c2ba3ad6 LP |
3859 | p = prioq_peek(e->pending); |
3860 | if (!p) | |
3861 | return NULL; | |
3862 | ||
b6d5481b | 3863 | if (event_source_is_offline(p)) |
c2ba3ad6 LP |
3864 | return NULL; |
3865 | ||
3866 | return p; | |
3867 | } | |
3868 | ||
cde93897 LP |
3869 | static int arm_watchdog(sd_event *e) { |
3870 | struct itimerspec its = {}; | |
3871 | usec_t t; | |
cde93897 LP |
3872 | |
3873 | assert(e); | |
3874 | assert(e->watchdog_fd >= 0); | |
3875 | ||
3876 | t = sleep_between(e, | |
a595fb5c YW |
3877 | usec_add(e->watchdog_last, (e->watchdog_period / 2)), |
3878 | usec_add(e->watchdog_last, (e->watchdog_period * 3 / 4))); | |
cde93897 LP |
3879 | |
3880 | timespec_store(&its.it_value, t); | |
3881 | ||
75145780 LP |
3882 | /* Make sure we never set the watchdog to 0, which tells the |
3883 | * kernel to disable it. */ | |
3884 | if (its.it_value.tv_sec == 0 && its.it_value.tv_nsec == 0) | |
3885 | its.it_value.tv_nsec = 1; | |
3886 | ||
7c248223 | 3887 | return RET_NERRNO(timerfd_settime(e->watchdog_fd, TFD_TIMER_ABSTIME, &its, NULL)); |
cde93897 LP |
3888 | } |
3889 | ||
3890 | static int process_watchdog(sd_event *e) { | |
3891 | assert(e); | |
3892 | ||
3893 | if (!e->watchdog) | |
3894 | return 0; | |
3895 | ||
3896 | /* Don't notify watchdog too often */ | |
3897 | if (e->watchdog_last + e->watchdog_period / 4 > e->timestamp.monotonic) | |
3898 | return 0; | |
3899 | ||
3900 | sd_notify(false, "WATCHDOG=1"); | |
3901 | e->watchdog_last = e->timestamp.monotonic; | |
3902 | ||
3903 | return arm_watchdog(e); | |
3904 | } | |
3905 | ||
97ef5391 LP |
3906 | static void event_close_inode_data_fds(sd_event *e) { |
3907 | struct inode_data *d; | |
3908 | ||
3909 | assert(e); | |
3910 | ||
3911 | /* Close the fds pointing to the inodes to watch now. We need to close them as they might otherwise pin | |
3912 | * filesystems. But we can't close them right-away as we need them as long as the user still wants to make | |
365c2885 | 3913 | * adjustments to the event source, such as changing the priority (which requires us to remove and re-add a watch |
97ef5391 LP |
3914 | * for the inode). Hence, let's close them when entering the first iteration after they were added, as a |
3915 | * compromise. */ | |
3916 | ||
ed828563 | 3917 | while ((d = e->inode_data_to_close_list)) { |
97ef5391 LP |
3918 | assert(d->fd >= 0); |
3919 | d->fd = safe_close(d->fd); | |
3920 | ||
ed828563 | 3921 | LIST_REMOVE(to_close, e->inode_data_to_close_list, d); |
97ef5391 LP |
3922 | } |
3923 | } | |
3924 | ||
c45a5a74 TG |
3925 | _public_ int sd_event_prepare(sd_event *e) { |
3926 | int r; | |
fd38203a | 3927 | |
da7e457c | 3928 | assert_return(e, -EINVAL); |
b937d761 | 3929 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
3930 | assert_return(!event_pid_changed(e), -ECHILD); |
3931 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 3932 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 3933 | |
e5446015 LP |
3934 | /* Let's check that if we are a default event loop we are executed in the correct thread. We only do |
3935 | * this check here once, since gettid() is typically not cached, and thus want to minimize | |
3936 | * syscalls */ | |
3937 | assert_return(!e->default_event_ptr || e->tid == gettid(), -EREMOTEIO); | |
3938 | ||
f814c871 | 3939 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
c8e9d15c | 3940 | PROTECT_EVENT(e); |
f814c871 | 3941 | |
6203e07a | 3942 | if (e->exit_requested) |
c45a5a74 | 3943 | goto pending; |
fd38203a LP |
3944 | |
3945 | e->iteration++; | |
3946 | ||
0be6c2f6 | 3947 | e->state = SD_EVENT_PREPARING; |
fd38203a | 3948 | r = event_prepare(e); |
0be6c2f6 | 3949 | e->state = SD_EVENT_INITIAL; |
fd38203a | 3950 | if (r < 0) |
c45a5a74 | 3951 | return r; |
fd38203a | 3952 | |
6a0f1f6d LP |
3953 | r = event_arm_timer(e, &e->realtime); |
3954 | if (r < 0) | |
c45a5a74 | 3955 | return r; |
6a0f1f6d | 3956 | |
a8548816 TG |
3957 | r = event_arm_timer(e, &e->boottime); |
3958 | if (r < 0) | |
c45a5a74 | 3959 | return r; |
a8548816 | 3960 | |
6a0f1f6d LP |
3961 | r = event_arm_timer(e, &e->monotonic); |
3962 | if (r < 0) | |
c45a5a74 | 3963 | return r; |
6a0f1f6d LP |
3964 | |
3965 | r = event_arm_timer(e, &e->realtime_alarm); | |
1b5995b0 | 3966 | if (r < 0) |
c45a5a74 | 3967 | return r; |
fd38203a | 3968 | |
6a0f1f6d | 3969 | r = event_arm_timer(e, &e->boottime_alarm); |
1b5995b0 | 3970 | if (r < 0) |
c45a5a74 | 3971 | return r; |
fd38203a | 3972 | |
97ef5391 LP |
3973 | event_close_inode_data_fds(e); |
3974 | ||
0601b958 | 3975 | if (event_next_pending(e) || e->need_process_child || e->buffered_inotify_data_list) |
c45a5a74 TG |
3976 | goto pending; |
3977 | ||
2b0c9ef7 | 3978 | e->state = SD_EVENT_ARMED; |
c45a5a74 TG |
3979 | |
3980 | return 0; | |
3981 | ||
3982 | pending: | |
2b0c9ef7 | 3983 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
3984 | r = sd_event_wait(e, 0); |
3985 | if (r == 0) | |
2b0c9ef7 | 3986 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
3987 | |
3988 | return r; | |
c45a5a74 TG |
3989 | } |
3990 | ||
798445ab LP |
3991 | static int epoll_wait_usec( |
3992 | int fd, | |
3993 | struct epoll_event *events, | |
3994 | int maxevents, | |
3995 | usec_t timeout) { | |
3996 | ||
7c248223 | 3997 | int msec; |
0c14c45e LP |
3998 | /* A wrapper that uses epoll_pwait2() if available, and falls back to epoll_wait() if not. */ |
3999 | ||
4000 | #if HAVE_EPOLL_PWAIT2 | |
39f756d3 | 4001 | static bool epoll_pwait2_absent = false; |
52bb308c | 4002 | int r; |
798445ab | 4003 | |
0c14c45e LP |
4004 | /* epoll_pwait2() was added to Linux 5.11 (2021-02-14) and to glibc in 2.35 (2022-02-03). In contrast |
4005 | * to other syscalls we don't bother with our own fallback syscall wrappers on old libcs, since this | |
4006 | * is not that obvious to implement given the libc and kernel definitions differ in the last | |
4007 | * argument. Moreover, the only reason to use it is the more accurate time-outs (which is not a | |
4008 | * biggie), let's hence rely on glibc's definitions, and fallback to epoll_pwait() when that's | |
4009 | * missing. */ | |
798445ab LP |
4010 | |
4011 | if (!epoll_pwait2_absent && timeout != USEC_INFINITY) { | |
798445ab LP |
4012 | r = epoll_pwait2(fd, |
4013 | events, | |
4014 | maxevents, | |
52bb308c | 4015 | TIMESPEC_STORE(timeout), |
798445ab LP |
4016 | NULL); |
4017 | if (r >= 0) | |
4018 | return r; | |
7cb45dbf | 4019 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) |
798445ab LP |
4020 | return -errno; /* Only fallback to old epoll_wait() if the syscall is masked or not |
4021 | * supported. */ | |
4022 | ||
4023 | epoll_pwait2_absent = true; | |
4024 | } | |
39f756d3 | 4025 | #endif |
798445ab LP |
4026 | |
4027 | if (timeout == USEC_INFINITY) | |
4028 | msec = -1; | |
4029 | else { | |
4030 | usec_t k; | |
4031 | ||
4032 | k = DIV_ROUND_UP(timeout, USEC_PER_MSEC); | |
4033 | if (k >= INT_MAX) | |
4034 | msec = INT_MAX; /* Saturate */ | |
4035 | else | |
4036 | msec = (int) k; | |
4037 | } | |
4038 | ||
7c248223 | 4039 | return RET_NERRNO(epoll_wait(fd, events, maxevents, msec)); |
798445ab LP |
4040 | } |
4041 | ||
efd3be9d | 4042 | static int process_epoll(sd_event *e, usec_t timeout, int64_t threshold, int64_t *ret_min_priority) { |
319a4f4b | 4043 | size_t n_event_queue, m, n_event_max; |
efd3be9d YW |
4044 | int64_t min_priority = threshold; |
4045 | bool something_new = false; | |
798445ab | 4046 | int r; |
c45a5a74 | 4047 | |
efd3be9d YW |
4048 | assert(e); |
4049 | assert(ret_min_priority); | |
6a0f1f6d | 4050 | |
8b9708d1 | 4051 | n_event_queue = MAX(e->n_sources, 1u); |
319a4f4b | 4052 | if (!GREEDY_REALLOC(e->event_queue, n_event_queue)) |
5cddd924 | 4053 | return -ENOMEM; |
fd38203a | 4054 | |
319a4f4b LP |
4055 | n_event_max = MALLOC_ELEMENTSOF(e->event_queue); |
4056 | ||
97ef5391 | 4057 | /* If we still have inotify data buffered, then query the other fds, but don't wait on it */ |
0601b958 | 4058 | if (e->buffered_inotify_data_list) |
798445ab | 4059 | timeout = 0; |
97ef5391 | 4060 | |
8b9708d1 | 4061 | for (;;) { |
319a4f4b LP |
4062 | r = epoll_wait_usec( |
4063 | e->epoll_fd, | |
4064 | e->event_queue, | |
4065 | n_event_max, | |
4066 | timeout); | |
798445ab | 4067 | if (r < 0) |
efd3be9d | 4068 | return r; |
c45a5a74 | 4069 | |
8b9708d1 YW |
4070 | m = (size_t) r; |
4071 | ||
319a4f4b | 4072 | if (m < n_event_max) |
8b9708d1 YW |
4073 | break; |
4074 | ||
319a4f4b | 4075 | if (n_event_max >= n_event_queue * 10) |
8b9708d1 YW |
4076 | break; |
4077 | ||
319a4f4b | 4078 | if (!GREEDY_REALLOC(e->event_queue, n_event_max + n_event_queue)) |
8b9708d1 YW |
4079 | return -ENOMEM; |
4080 | ||
319a4f4b | 4081 | n_event_max = MALLOC_ELEMENTSOF(e->event_queue); |
798445ab | 4082 | timeout = 0; |
da7e457c | 4083 | } |
fd38203a | 4084 | |
efd3be9d YW |
4085 | /* Set timestamp only when this is called first time. */ |
4086 | if (threshold == INT64_MAX) | |
4087 | triple_timestamp_get(&e->timestamp); | |
fd38203a | 4088 | |
8b9708d1 | 4089 | for (size_t i = 0; i < m; i++) { |
fd38203a | 4090 | |
5cddd924 LP |
4091 | if (e->event_queue[i].data.ptr == INT_TO_PTR(SOURCE_WATCHDOG)) |
4092 | r = flush_timer(e, e->watchdog_fd, e->event_queue[i].events, NULL); | |
9da4cb2b | 4093 | else { |
5cddd924 | 4094 | WakeupType *t = e->event_queue[i].data.ptr; |
9da4cb2b LP |
4095 | |
4096 | switch (*t) { | |
4097 | ||
f8f3f926 | 4098 | case WAKEUP_EVENT_SOURCE: { |
5cddd924 | 4099 | sd_event_source *s = e->event_queue[i].data.ptr; |
f8f3f926 LP |
4100 | |
4101 | assert(s); | |
4102 | ||
efd3be9d YW |
4103 | if (s->priority > threshold) |
4104 | continue; | |
4105 | ||
4106 | min_priority = MIN(min_priority, s->priority); | |
4107 | ||
f8f3f926 LP |
4108 | switch (s->type) { |
4109 | ||
4110 | case SOURCE_IO: | |
5cddd924 | 4111 | r = process_io(e, s, e->event_queue[i].events); |
f8f3f926 LP |
4112 | break; |
4113 | ||
4114 | case SOURCE_CHILD: | |
5cddd924 | 4115 | r = process_pidfd(e, s, e->event_queue[i].events); |
f8f3f926 LP |
4116 | break; |
4117 | ||
4118 | default: | |
04499a70 | 4119 | assert_not_reached(); |
f8f3f926 LP |
4120 | } |
4121 | ||
9da4cb2b | 4122 | break; |
f8f3f926 | 4123 | } |
fd38203a | 4124 | |
9da4cb2b | 4125 | case WAKEUP_CLOCK_DATA: { |
5cddd924 | 4126 | struct clock_data *d = e->event_queue[i].data.ptr; |
f8f3f926 LP |
4127 | |
4128 | assert(d); | |
4129 | ||
5cddd924 | 4130 | r = flush_timer(e, d->fd, e->event_queue[i].events, &d->next); |
9da4cb2b LP |
4131 | break; |
4132 | } | |
4133 | ||
4134 | case WAKEUP_SIGNAL_DATA: | |
efd3be9d | 4135 | r = process_signal(e, e->event_queue[i].data.ptr, e->event_queue[i].events, &min_priority); |
9da4cb2b LP |
4136 | break; |
4137 | ||
97ef5391 | 4138 | case WAKEUP_INOTIFY_DATA: |
efd3be9d | 4139 | r = event_inotify_data_read(e, e->event_queue[i].data.ptr, e->event_queue[i].events, threshold); |
97ef5391 LP |
4140 | break; |
4141 | ||
9da4cb2b | 4142 | default: |
04499a70 | 4143 | assert_not_reached(); |
9da4cb2b LP |
4144 | } |
4145 | } | |
efd3be9d YW |
4146 | if (r < 0) |
4147 | return r; | |
4148 | if (r > 0) | |
4149 | something_new = true; | |
4150 | } | |
4151 | ||
4152 | *ret_min_priority = min_priority; | |
4153 | return something_new; | |
4154 | } | |
4155 | ||
4156 | _public_ int sd_event_wait(sd_event *e, uint64_t timeout) { | |
4157 | int r; | |
4158 | ||
4159 | assert_return(e, -EINVAL); | |
4160 | assert_return(e = event_resolve(e), -ENOPKG); | |
4161 | assert_return(!event_pid_changed(e), -ECHILD); | |
4162 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
4163 | assert_return(e->state == SD_EVENT_ARMED, -EBUSY); | |
4164 | ||
4165 | if (e->exit_requested) { | |
4166 | e->state = SD_EVENT_PENDING; | |
4167 | return 1; | |
4168 | } | |
4169 | ||
4170 | for (int64_t threshold = INT64_MAX; ; threshold--) { | |
4171 | int64_t epoll_min_priority, child_min_priority; | |
4172 | ||
4173 | /* There may be a possibility that new epoll (especially IO) and child events are | |
4174 | * triggered just after process_epoll() call but before process_child(), and the new IO | |
4175 | * events may have higher priority than the child events. To salvage these events, | |
4176 | * let's call epoll_wait() again, but accepts only events with higher priority than the | |
4177 | * previous. See issue https://github.com/systemd/systemd/issues/18190 and comments | |
4178 | * https://github.com/systemd/systemd/pull/18750#issuecomment-785801085 | |
4179 | * https://github.com/systemd/systemd/pull/18922#issuecomment-792825226 */ | |
4180 | ||
4181 | r = process_epoll(e, timeout, threshold, &epoll_min_priority); | |
4182 | if (r == -EINTR) { | |
4183 | e->state = SD_EVENT_PENDING; | |
4184 | return 1; | |
4185 | } | |
4186 | if (r < 0) | |
4187 | goto finish; | |
4188 | if (r == 0 && threshold < INT64_MAX) | |
4189 | /* No new epoll event. */ | |
4190 | break; | |
4191 | ||
4192 | r = process_child(e, threshold, &child_min_priority); | |
fd38203a | 4193 | if (r < 0) |
da7e457c | 4194 | goto finish; |
efd3be9d YW |
4195 | if (r == 0) |
4196 | /* No new child event. */ | |
4197 | break; | |
4198 | ||
4199 | threshold = MIN(epoll_min_priority, child_min_priority); | |
4200 | if (threshold == INT64_MIN) | |
4201 | break; | |
4202 | ||
4203 | timeout = 0; | |
fd38203a LP |
4204 | } |
4205 | ||
cde93897 LP |
4206 | r = process_watchdog(e); |
4207 | if (r < 0) | |
4208 | goto finish; | |
4209 | ||
fd69f224 | 4210 | r = process_inotify(e); |
6a0f1f6d LP |
4211 | if (r < 0) |
4212 | goto finish; | |
4213 | ||
fd69f224 | 4214 | r = process_timer(e, e->timestamp.realtime, &e->realtime); |
a8548816 TG |
4215 | if (r < 0) |
4216 | goto finish; | |
4217 | ||
fd69f224 | 4218 | r = process_timer(e, e->timestamp.boottime, &e->boottime); |
6a0f1f6d LP |
4219 | if (r < 0) |
4220 | goto finish; | |
4221 | ||
4222 | r = process_timer(e, e->timestamp.realtime, &e->realtime_alarm); | |
fd38203a | 4223 | if (r < 0) |
da7e457c | 4224 | goto finish; |
fd38203a | 4225 | |
e475d10c | 4226 | r = process_timer(e, e->timestamp.boottime, &e->boottime_alarm); |
fd38203a | 4227 | if (r < 0) |
da7e457c | 4228 | goto finish; |
fd38203a | 4229 | |
fd69f224 | 4230 | r = process_timer(e, e->timestamp.monotonic, &e->monotonic); |
97ef5391 LP |
4231 | if (r < 0) |
4232 | goto finish; | |
fd69f224 MS |
4233 | else if (r == 1) { |
4234 | /* Ratelimit expiry callback was called. Let's postpone processing pending sources and | |
4235 | * put loop in the initial state in order to evaluate (in the next iteration) also sources | |
4236 | * there were potentially re-enabled by the callback. | |
4237 | * | |
4238 | * Wondering why we treat only this invocation of process_timer() differently? Once event | |
4239 | * source is ratelimited we essentially transform it into CLOCK_MONOTONIC timer hence | |
4240 | * ratelimit expiry callback is never called for any other timer type. */ | |
4241 | r = 0; | |
4242 | goto finish; | |
4243 | } | |
97ef5391 | 4244 | |
c45a5a74 TG |
4245 | if (event_next_pending(e)) { |
4246 | e->state = SD_EVENT_PENDING; | |
c45a5a74 | 4247 | return 1; |
da7e457c LP |
4248 | } |
4249 | ||
c45a5a74 | 4250 | r = 0; |
fd38203a | 4251 | |
da7e457c | 4252 | finish: |
2b0c9ef7 | 4253 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
4254 | |
4255 | return r; | |
fd38203a LP |
4256 | } |
4257 | ||
c45a5a74 TG |
4258 | _public_ int sd_event_dispatch(sd_event *e) { |
4259 | sd_event_source *p; | |
4260 | int r; | |
4261 | ||
4262 | assert_return(e, -EINVAL); | |
b937d761 | 4263 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
4264 | assert_return(!event_pid_changed(e), -ECHILD); |
4265 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
4266 | assert_return(e->state == SD_EVENT_PENDING, -EBUSY); | |
4267 | ||
4268 | if (e->exit_requested) | |
4269 | return dispatch_exit(e); | |
4270 | ||
4271 | p = event_next_pending(e); | |
4272 | if (p) { | |
c8e9d15c | 4273 | PROTECT_EVENT(e); |
c45a5a74 TG |
4274 | |
4275 | e->state = SD_EVENT_RUNNING; | |
4276 | r = source_dispatch(p); | |
2b0c9ef7 | 4277 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
4278 | return r; |
4279 | } | |
4280 | ||
2b0c9ef7 | 4281 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
4282 | |
4283 | return 1; | |
4284 | } | |
4285 | ||
34b87517 | 4286 | static void event_log_delays(sd_event *e) { |
442ac269 YW |
4287 | char b[ELEMENTSOF(e->delays) * DECIMAL_STR_MAX(unsigned) + 1], *p; |
4288 | size_t l, i; | |
34b87517 | 4289 | |
442ac269 YW |
4290 | p = b; |
4291 | l = sizeof(b); | |
4292 | for (i = 0; i < ELEMENTSOF(e->delays); i++) { | |
4293 | l = strpcpyf(&p, l, "%u ", e->delays[i]); | |
34b87517 VC |
4294 | e->delays[i] = 0; |
4295 | } | |
442ac269 | 4296 | log_debug("Event loop iterations: %s", b); |
34b87517 VC |
4297 | } |
4298 | ||
c45a5a74 TG |
4299 | _public_ int sd_event_run(sd_event *e, uint64_t timeout) { |
4300 | int r; | |
4301 | ||
4302 | assert_return(e, -EINVAL); | |
b937d761 | 4303 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
4304 | assert_return(!event_pid_changed(e), -ECHILD); |
4305 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 4306 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
c45a5a74 | 4307 | |
e6a7bee5 | 4308 | if (e->profile_delays && e->last_run_usec != 0) { |
34b87517 VC |
4309 | usec_t this_run; |
4310 | unsigned l; | |
4311 | ||
4312 | this_run = now(CLOCK_MONOTONIC); | |
4313 | ||
58c34be8 | 4314 | l = log2u64(this_run - e->last_run_usec); |
cb9d621e | 4315 | assert(l < ELEMENTSOF(e->delays)); |
34b87517 VC |
4316 | e->delays[l]++; |
4317 | ||
e6a7bee5 | 4318 | if (this_run - e->last_log_usec >= 5*USEC_PER_SEC) { |
34b87517 | 4319 | event_log_delays(e); |
e6a7bee5 | 4320 | e->last_log_usec = this_run; |
34b87517 VC |
4321 | } |
4322 | } | |
4323 | ||
f814c871 | 4324 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
c8e9d15c | 4325 | PROTECT_EVENT(e); |
f814c871 | 4326 | |
c45a5a74 | 4327 | r = sd_event_prepare(e); |
53bac4e0 LP |
4328 | if (r == 0) |
4329 | /* There was nothing? Then wait... */ | |
4330 | r = sd_event_wait(e, timeout); | |
c45a5a74 | 4331 | |
34b87517 | 4332 | if (e->profile_delays) |
e6a7bee5 | 4333 | e->last_run_usec = now(CLOCK_MONOTONIC); |
34b87517 | 4334 | |
02d30981 | 4335 | if (r > 0) { |
53bac4e0 | 4336 | /* There's something now, then let's dispatch it */ |
02d30981 TG |
4337 | r = sd_event_dispatch(e); |
4338 | if (r < 0) | |
4339 | return r; | |
53bac4e0 LP |
4340 | |
4341 | return 1; | |
4342 | } | |
4343 | ||
4344 | return r; | |
c45a5a74 TG |
4345 | } |
4346 | ||
f7262a9f | 4347 | _public_ int sd_event_loop(sd_event *e) { |
fd38203a LP |
4348 | int r; |
4349 | ||
da7e457c | 4350 | assert_return(e, -EINVAL); |
b937d761 | 4351 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 4352 | assert_return(!event_pid_changed(e), -ECHILD); |
2b0c9ef7 | 4353 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 4354 | |
c8e9d15c | 4355 | PROTECT_EVENT(e); |
fd38203a | 4356 | |
da7e457c | 4357 | while (e->state != SD_EVENT_FINISHED) { |
f5fbe71d | 4358 | r = sd_event_run(e, UINT64_MAX); |
fd38203a | 4359 | if (r < 0) |
30dd293c | 4360 | return r; |
fd38203a LP |
4361 | } |
4362 | ||
30dd293c | 4363 | return e->exit_code; |
fd38203a LP |
4364 | } |
4365 | ||
9b364545 | 4366 | _public_ int sd_event_get_fd(sd_event *e) { |
9b364545 | 4367 | assert_return(e, -EINVAL); |
b937d761 | 4368 | assert_return(e = event_resolve(e), -ENOPKG); |
9b364545 TG |
4369 | assert_return(!event_pid_changed(e), -ECHILD); |
4370 | ||
4371 | return e->epoll_fd; | |
4372 | } | |
4373 | ||
f7262a9f | 4374 | _public_ int sd_event_get_state(sd_event *e) { |
da7e457c | 4375 | assert_return(e, -EINVAL); |
b937d761 | 4376 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
4377 | assert_return(!event_pid_changed(e), -ECHILD); |
4378 | ||
4379 | return e->state; | |
4380 | } | |
4381 | ||
6203e07a | 4382 | _public_ int sd_event_get_exit_code(sd_event *e, int *code) { |
da7e457c | 4383 | assert_return(e, -EINVAL); |
b937d761 | 4384 | assert_return(e = event_resolve(e), -ENOPKG); |
6203e07a | 4385 | assert_return(code, -EINVAL); |
da7e457c | 4386 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 4387 | |
6203e07a LP |
4388 | if (!e->exit_requested) |
4389 | return -ENODATA; | |
4390 | ||
4391 | *code = e->exit_code; | |
4392 | return 0; | |
fd38203a LP |
4393 | } |
4394 | ||
6203e07a | 4395 | _public_ int sd_event_exit(sd_event *e, int code) { |
da7e457c | 4396 | assert_return(e, -EINVAL); |
b937d761 | 4397 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
4398 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
4399 | assert_return(!event_pid_changed(e), -ECHILD); | |
fd38203a | 4400 | |
6203e07a LP |
4401 | e->exit_requested = true; |
4402 | e->exit_code = code; | |
4403 | ||
fd38203a LP |
4404 | return 0; |
4405 | } | |
46e8c825 | 4406 | |
6a0f1f6d | 4407 | _public_ int sd_event_now(sd_event *e, clockid_t clock, uint64_t *usec) { |
46e8c825 | 4408 | assert_return(e, -EINVAL); |
b937d761 | 4409 | assert_return(e = event_resolve(e), -ENOPKG); |
46e8c825 | 4410 | assert_return(usec, -EINVAL); |
46e8c825 LP |
4411 | assert_return(!event_pid_changed(e), -ECHILD); |
4412 | ||
e475d10c LP |
4413 | if (!TRIPLE_TIMESTAMP_HAS_CLOCK(clock)) |
4414 | return -EOPNOTSUPP; | |
4415 | ||
e475d10c | 4416 | if (!triple_timestamp_is_set(&e->timestamp)) { |
15c689d7 | 4417 | /* Implicitly fall back to now() if we never ran before and thus have no cached time. */ |
38a03f06 LP |
4418 | *usec = now(clock); |
4419 | return 1; | |
4420 | } | |
46e8c825 | 4421 | |
e475d10c | 4422 | *usec = triple_timestamp_by_clock(&e->timestamp, clock); |
46e8c825 LP |
4423 | return 0; |
4424 | } | |
afc6adb5 LP |
4425 | |
4426 | _public_ int sd_event_default(sd_event **ret) { | |
39883f62 | 4427 | sd_event *e = NULL; |
afc6adb5 LP |
4428 | int r; |
4429 | ||
4430 | if (!ret) | |
4431 | return !!default_event; | |
4432 | ||
4433 | if (default_event) { | |
4434 | *ret = sd_event_ref(default_event); | |
4435 | return 0; | |
4436 | } | |
4437 | ||
4438 | r = sd_event_new(&e); | |
4439 | if (r < 0) | |
4440 | return r; | |
4441 | ||
4442 | e->default_event_ptr = &default_event; | |
4443 | e->tid = gettid(); | |
4444 | default_event = e; | |
4445 | ||
4446 | *ret = e; | |
4447 | return 1; | |
4448 | } | |
4449 | ||
4450 | _public_ int sd_event_get_tid(sd_event *e, pid_t *tid) { | |
4451 | assert_return(e, -EINVAL); | |
b937d761 | 4452 | assert_return(e = event_resolve(e), -ENOPKG); |
afc6adb5 | 4453 | assert_return(tid, -EINVAL); |
76b54375 | 4454 | assert_return(!event_pid_changed(e), -ECHILD); |
afc6adb5 | 4455 | |
76b54375 LP |
4456 | if (e->tid != 0) { |
4457 | *tid = e->tid; | |
4458 | return 0; | |
4459 | } | |
4460 | ||
4461 | return -ENXIO; | |
afc6adb5 | 4462 | } |
cde93897 LP |
4463 | |
4464 | _public_ int sd_event_set_watchdog(sd_event *e, int b) { | |
4465 | int r; | |
4466 | ||
4467 | assert_return(e, -EINVAL); | |
b937d761 | 4468 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 | 4469 | assert_return(!event_pid_changed(e), -ECHILD); |
cde93897 LP |
4470 | |
4471 | if (e->watchdog == !!b) | |
4472 | return e->watchdog; | |
4473 | ||
4474 | if (b) { | |
09812eb7 LP |
4475 | r = sd_watchdog_enabled(false, &e->watchdog_period); |
4476 | if (r <= 0) | |
cde93897 | 4477 | return r; |
cde93897 LP |
4478 | |
4479 | /* Issue first ping immediately */ | |
4480 | sd_notify(false, "WATCHDOG=1"); | |
4481 | e->watchdog_last = now(CLOCK_MONOTONIC); | |
4482 | ||
4483 | e->watchdog_fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC); | |
4484 | if (e->watchdog_fd < 0) | |
4485 | return -errno; | |
4486 | ||
4487 | r = arm_watchdog(e); | |
4488 | if (r < 0) | |
4489 | goto fail; | |
4490 | ||
1eac7948 | 4491 | struct epoll_event ev = { |
a82f89aa LP |
4492 | .events = EPOLLIN, |
4493 | .data.ptr = INT_TO_PTR(SOURCE_WATCHDOG), | |
4494 | }; | |
cde93897 | 4495 | |
15c689d7 | 4496 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, e->watchdog_fd, &ev) < 0) { |
cde93897 LP |
4497 | r = -errno; |
4498 | goto fail; | |
4499 | } | |
4500 | ||
4501 | } else { | |
4502 | if (e->watchdog_fd >= 0) { | |
5a795bff | 4503 | (void) epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, e->watchdog_fd, NULL); |
03e334a1 | 4504 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
4505 | } |
4506 | } | |
4507 | ||
4508 | e->watchdog = !!b; | |
4509 | return e->watchdog; | |
4510 | ||
4511 | fail: | |
03e334a1 | 4512 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
4513 | return r; |
4514 | } | |
8f726607 LP |
4515 | |
4516 | _public_ int sd_event_get_watchdog(sd_event *e) { | |
4517 | assert_return(e, -EINVAL); | |
b937d761 | 4518 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 LP |
4519 | assert_return(!event_pid_changed(e), -ECHILD); |
4520 | ||
4521 | return e->watchdog; | |
4522 | } | |
60a3b1e1 LP |
4523 | |
4524 | _public_ int sd_event_get_iteration(sd_event *e, uint64_t *ret) { | |
4525 | assert_return(e, -EINVAL); | |
b937d761 | 4526 | assert_return(e = event_resolve(e), -ENOPKG); |
60a3b1e1 LP |
4527 | assert_return(!event_pid_changed(e), -ECHILD); |
4528 | ||
4529 | *ret = e->iteration; | |
4530 | return 0; | |
4531 | } | |
15723a1d LP |
4532 | |
4533 | _public_ int sd_event_source_set_destroy_callback(sd_event_source *s, sd_event_destroy_t callback) { | |
4534 | assert_return(s, -EINVAL); | |
4535 | ||
4536 | s->destroy_callback = callback; | |
4537 | return 0; | |
4538 | } | |
4539 | ||
4540 | _public_ int sd_event_source_get_destroy_callback(sd_event_source *s, sd_event_destroy_t *ret) { | |
4541 | assert_return(s, -EINVAL); | |
4542 | ||
4543 | if (ret) | |
4544 | *ret = s->destroy_callback; | |
4545 | ||
4546 | return !!s->destroy_callback; | |
4547 | } | |
2382c936 YW |
4548 | |
4549 | _public_ int sd_event_source_get_floating(sd_event_source *s) { | |
4550 | assert_return(s, -EINVAL); | |
4551 | ||
4552 | return s->floating; | |
4553 | } | |
4554 | ||
4555 | _public_ int sd_event_source_set_floating(sd_event_source *s, int b) { | |
4556 | assert_return(s, -EINVAL); | |
4557 | ||
4558 | if (s->floating == !!b) | |
4559 | return 0; | |
4560 | ||
4561 | if (!s->event) /* Already disconnected */ | |
4562 | return -ESTALE; | |
4563 | ||
4564 | s->floating = b; | |
4565 | ||
4566 | if (b) { | |
4567 | sd_event_source_ref(s); | |
4568 | sd_event_unref(s->event); | |
4569 | } else { | |
4570 | sd_event_ref(s->event); | |
4571 | sd_event_source_unref(s); | |
4572 | } | |
4573 | ||
4574 | return 1; | |
4575 | } | |
b778cba4 LP |
4576 | |
4577 | _public_ int sd_event_source_get_exit_on_failure(sd_event_source *s) { | |
4578 | assert_return(s, -EINVAL); | |
4579 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
4580 | ||
4581 | return s->exit_on_failure; | |
4582 | } | |
4583 | ||
4584 | _public_ int sd_event_source_set_exit_on_failure(sd_event_source *s, int b) { | |
4585 | assert_return(s, -EINVAL); | |
4586 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
4587 | ||
4588 | if (s->exit_on_failure == !!b) | |
4589 | return 0; | |
4590 | ||
4591 | s->exit_on_failure = b; | |
4592 | return 1; | |
4593 | } | |
b6d5481b LP |
4594 | |
4595 | _public_ int sd_event_source_set_ratelimit(sd_event_source *s, uint64_t interval, unsigned burst) { | |
4596 | int r; | |
4597 | ||
4598 | assert_return(s, -EINVAL); | |
4599 | ||
4600 | /* Turning on ratelimiting on event source types that don't support it, is a loggable offense. Doing | |
4601 | * so is a programming error. */ | |
4602 | assert_return(EVENT_SOURCE_CAN_RATE_LIMIT(s->type), -EDOM); | |
4603 | ||
4604 | /* When ratelimiting is configured we'll always reset the rate limit state first and start fresh, | |
4605 | * non-ratelimited. */ | |
fd69f224 | 4606 | r = event_source_leave_ratelimit(s, /* run_callback */ false); |
b6d5481b LP |
4607 | if (r < 0) |
4608 | return r; | |
4609 | ||
4610 | s->rate_limit = (RateLimit) { interval, burst }; | |
4611 | return 0; | |
fd69f224 MS |
4612 | } |
4613 | ||
4614 | _public_ int sd_event_source_set_ratelimit_expire_callback(sd_event_source *s, sd_event_handler_t callback) { | |
4615 | assert_return(s, -EINVAL); | |
4616 | ||
4617 | s->ratelimit_expire_callback = callback; | |
4618 | return 0; | |
b6d5481b LP |
4619 | } |
4620 | ||
4621 | _public_ int sd_event_source_get_ratelimit(sd_event_source *s, uint64_t *ret_interval, unsigned *ret_burst) { | |
4622 | assert_return(s, -EINVAL); | |
4623 | ||
6dd3b818 YW |
4624 | /* Querying whether an event source has ratelimiting configured is not a loggable offense, hence |
4625 | * don't use assert_return(). Unlike turning on ratelimiting it's not really a programming error. */ | |
b6d5481b LP |
4626 | if (!EVENT_SOURCE_CAN_RATE_LIMIT(s->type)) |
4627 | return -EDOM; | |
4628 | ||
4629 | if (!ratelimit_configured(&s->rate_limit)) | |
4630 | return -ENOEXEC; | |
4631 | ||
4632 | if (ret_interval) | |
4633 | *ret_interval = s->rate_limit.interval; | |
4634 | if (ret_burst) | |
4635 | *ret_burst = s->rate_limit.burst; | |
4636 | ||
4637 | return 0; | |
4638 | } | |
4639 | ||
4640 | _public_ int sd_event_source_is_ratelimited(sd_event_source *s) { | |
4641 | assert_return(s, -EINVAL); | |
4642 | ||
4643 | if (!EVENT_SOURCE_CAN_RATE_LIMIT(s->type)) | |
4644 | return false; | |
4645 | ||
4646 | if (!ratelimit_configured(&s->rate_limit)) | |
4647 | return false; | |
4648 | ||
4649 | return s->ratelimited; | |
4650 | } | |
baf3fdec LP |
4651 | |
4652 | _public_ int sd_event_set_signal_exit(sd_event *e, int b) { | |
4653 | bool change = false; | |
4654 | int r; | |
4655 | ||
4656 | assert_return(e, -EINVAL); | |
4657 | ||
4658 | if (b) { | |
4659 | /* We want to maintain pointers to these event sources, so that we can destroy them when told | |
4660 | * so. But we also don't want them to pin the event loop itself. Hence we mark them as | |
4661 | * floating after creation (and undo this before deleting them again). */ | |
4662 | ||
4663 | if (!e->sigint_event_source) { | |
4664 | r = sd_event_add_signal(e, &e->sigint_event_source, SIGINT | SD_EVENT_SIGNAL_PROCMASK, NULL, NULL); | |
4665 | if (r < 0) | |
4666 | return r; | |
4667 | ||
4668 | assert(sd_event_source_set_floating(e->sigint_event_source, true) >= 0); | |
4669 | change = true; | |
4670 | } | |
4671 | ||
4672 | if (!e->sigterm_event_source) { | |
4673 | r = sd_event_add_signal(e, &e->sigterm_event_source, SIGTERM | SD_EVENT_SIGNAL_PROCMASK, NULL, NULL); | |
4674 | if (r < 0) { | |
4675 | if (change) { | |
4676 | assert(sd_event_source_set_floating(e->sigint_event_source, false) >= 0); | |
4677 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
4678 | } | |
4679 | ||
4680 | return r; | |
4681 | } | |
4682 | ||
4683 | assert(sd_event_source_set_floating(e->sigterm_event_source, true) >= 0); | |
4684 | change = true; | |
4685 | } | |
4686 | ||
4687 | } else { | |
4688 | if (e->sigint_event_source) { | |
4689 | assert(sd_event_source_set_floating(e->sigint_event_source, false) >= 0); | |
4690 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
4691 | change = true; | |
4692 | } | |
4693 | ||
4694 | if (e->sigterm_event_source) { | |
4695 | assert(sd_event_source_set_floating(e->sigterm_event_source, false) >= 0); | |
4696 | e->sigterm_event_source = sd_event_source_unref(e->sigterm_event_source); | |
4697 | change = true; | |
4698 | } | |
4699 | } | |
4700 | ||
4701 | return change; | |
4702 | } |