]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
fd38203a LP |
2 | |
3 | #include <sys/epoll.h> | |
4 | #include <sys/timerfd.h> | |
5 | #include <sys/wait.h> | |
6 | ||
cde93897 | 7 | #include "sd-daemon.h" |
07630cea LP |
8 | #include "sd-event.h" |
9 | #include "sd-id128.h" | |
10 | ||
b5efdb8a | 11 | #include "alloc-util.h" |
f8f3f926 | 12 | #include "env-util.h" |
a137a1c3 | 13 | #include "event-source.h" |
3ffd4af2 | 14 | #include "fd-util.h" |
97ef5391 | 15 | #include "fs-util.h" |
fd38203a | 16 | #include "hashmap.h" |
07630cea LP |
17 | #include "list.h" |
18 | #include "macro.h" | |
0a970718 | 19 | #include "memory-util.h" |
f5947a5e | 20 | #include "missing_syscall.h" |
07630cea | 21 | #include "prioq.h" |
4a0b58c4 | 22 | #include "process-util.h" |
6e9feda3 | 23 | #include "set.h" |
24882e06 | 24 | #include "signal-util.h" |
55cbfaa5 | 25 | #include "string-table.h" |
07630cea | 26 | #include "string-util.h" |
442ac269 | 27 | #include "strxcpyx.h" |
07630cea | 28 | #include "time-util.h" |
fd38203a | 29 | |
c2ba3ad6 | 30 | #define DEFAULT_ACCURACY_USEC (250 * USEC_PER_MSEC) |
fd38203a | 31 | |
f8f3f926 LP |
32 | static bool EVENT_SOURCE_WATCH_PIDFD(sd_event_source *s) { |
33 | /* Returns true if this is a PID event source and can be implemented by watching EPOLLIN */ | |
34 | return s && | |
35 | s->type == SOURCE_CHILD && | |
36 | s->child.pidfd >= 0 && | |
37 | s->child.options == WEXITED; | |
38 | } | |
39 | ||
55cbfaa5 DM |
40 | static const char* const event_source_type_table[_SOURCE_EVENT_SOURCE_TYPE_MAX] = { |
41 | [SOURCE_IO] = "io", | |
42 | [SOURCE_TIME_REALTIME] = "realtime", | |
43 | [SOURCE_TIME_BOOTTIME] = "bootime", | |
44 | [SOURCE_TIME_MONOTONIC] = "monotonic", | |
45 | [SOURCE_TIME_REALTIME_ALARM] = "realtime-alarm", | |
46 | [SOURCE_TIME_BOOTTIME_ALARM] = "boottime-alarm", | |
47 | [SOURCE_SIGNAL] = "signal", | |
48 | [SOURCE_CHILD] = "child", | |
49 | [SOURCE_DEFER] = "defer", | |
50 | [SOURCE_POST] = "post", | |
51 | [SOURCE_EXIT] = "exit", | |
52 | [SOURCE_WATCHDOG] = "watchdog", | |
97ef5391 | 53 | [SOURCE_INOTIFY] = "inotify", |
55cbfaa5 DM |
54 | }; |
55 | ||
56 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(event_source_type, int); | |
57 | ||
a8548816 | 58 | #define EVENT_SOURCE_IS_TIME(t) IN_SET((t), SOURCE_TIME_REALTIME, SOURCE_TIME_BOOTTIME, SOURCE_TIME_MONOTONIC, SOURCE_TIME_REALTIME_ALARM, SOURCE_TIME_BOOTTIME_ALARM) |
6a0f1f6d | 59 | |
fd38203a | 60 | struct sd_event { |
da7e457c | 61 | unsigned n_ref; |
fd38203a LP |
62 | |
63 | int epoll_fd; | |
cde93897 | 64 | int watchdog_fd; |
fd38203a LP |
65 | |
66 | Prioq *pending; | |
67 | Prioq *prepare; | |
c2ba3ad6 | 68 | |
a8548816 | 69 | /* timerfd_create() only supports these five clocks so far. We |
6a0f1f6d LP |
70 | * can add support for more clocks when the kernel learns to |
71 | * deal with them, too. */ | |
72 | struct clock_data realtime; | |
a8548816 | 73 | struct clock_data boottime; |
6a0f1f6d LP |
74 | struct clock_data monotonic; |
75 | struct clock_data realtime_alarm; | |
76 | struct clock_data boottime_alarm; | |
fd38203a | 77 | |
da7e457c LP |
78 | usec_t perturb; |
79 | ||
9da4cb2b LP |
80 | sd_event_source **signal_sources; /* indexed by signal number */ |
81 | Hashmap *signal_data; /* indexed by priority */ | |
fd38203a LP |
82 | |
83 | Hashmap *child_sources; | |
baf76283 | 84 | unsigned n_enabled_child_sources; |
fd38203a | 85 | |
6e9feda3 LP |
86 | Set *post_sources; |
87 | ||
6203e07a | 88 | Prioq *exit; |
fd38203a | 89 | |
97ef5391 LP |
90 | Hashmap *inotify_data; /* indexed by priority */ |
91 | ||
92 | /* A list of inode structures that still have an fd open, that we need to close before the next loop iteration */ | |
93 | LIST_HEAD(struct inode_data, inode_data_to_close); | |
94 | ||
95 | /* A list of inotify objects that already have events buffered which aren't processed yet */ | |
96 | LIST_HEAD(struct inotify_data, inotify_data_buffered); | |
97 | ||
da7e457c | 98 | pid_t original_pid; |
c2ba3ad6 | 99 | |
60a3b1e1 | 100 | uint64_t iteration; |
e475d10c | 101 | triple_timestamp timestamp; |
da7e457c | 102 | int state; |
eaa3cbef | 103 | |
6203e07a | 104 | bool exit_requested:1; |
da7e457c | 105 | bool need_process_child:1; |
cde93897 | 106 | bool watchdog:1; |
34b87517 | 107 | bool profile_delays:1; |
afc6adb5 | 108 | |
6203e07a LP |
109 | int exit_code; |
110 | ||
afc6adb5 LP |
111 | pid_t tid; |
112 | sd_event **default_event_ptr; | |
cde93897 LP |
113 | |
114 | usec_t watchdog_last, watchdog_period; | |
15b38f93 LP |
115 | |
116 | unsigned n_sources; | |
a71fe8b8 | 117 | |
5cddd924 LP |
118 | struct epoll_event *event_queue; |
119 | size_t event_queue_allocated; | |
120 | ||
a71fe8b8 | 121 | LIST_HEAD(sd_event_source, sources); |
34b87517 | 122 | |
e6a7bee5 | 123 | usec_t last_run_usec, last_log_usec; |
34b87517 | 124 | unsigned delays[sizeof(usec_t) * 8]; |
fd38203a LP |
125 | }; |
126 | ||
b937d761 NM |
127 | static thread_local sd_event *default_event = NULL; |
128 | ||
a71fe8b8 | 129 | static void source_disconnect(sd_event_source *s); |
97ef5391 | 130 | static void event_gc_inode_data(sd_event *e, struct inode_data *d); |
a71fe8b8 | 131 | |
b937d761 NM |
132 | static sd_event *event_resolve(sd_event *e) { |
133 | return e == SD_EVENT_DEFAULT ? default_event : e; | |
134 | } | |
135 | ||
fd38203a LP |
136 | static int pending_prioq_compare(const void *a, const void *b) { |
137 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 138 | int r; |
fd38203a LP |
139 | |
140 | assert(x->pending); | |
141 | assert(y->pending); | |
142 | ||
baf76283 LP |
143 | /* Enabled ones first */ |
144 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
fd38203a | 145 | return -1; |
baf76283 | 146 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) |
fd38203a LP |
147 | return 1; |
148 | ||
149 | /* Lower priority values first */ | |
9c57a73b YW |
150 | r = CMP(x->priority, y->priority); |
151 | if (r != 0) | |
152 | return r; | |
fd38203a LP |
153 | |
154 | /* Older entries first */ | |
9c57a73b | 155 | return CMP(x->pending_iteration, y->pending_iteration); |
fd38203a LP |
156 | } |
157 | ||
158 | static int prepare_prioq_compare(const void *a, const void *b) { | |
159 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 160 | int r; |
fd38203a LP |
161 | |
162 | assert(x->prepare); | |
163 | assert(y->prepare); | |
164 | ||
8046c457 KK |
165 | /* Enabled ones first */ |
166 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
167 | return -1; | |
168 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) | |
169 | return 1; | |
170 | ||
fd38203a LP |
171 | /* Move most recently prepared ones last, so that we can stop |
172 | * preparing as soon as we hit one that has already been | |
173 | * prepared in the current iteration */ | |
9c57a73b YW |
174 | r = CMP(x->prepare_iteration, y->prepare_iteration); |
175 | if (r != 0) | |
176 | return r; | |
fd38203a | 177 | |
fd38203a | 178 | /* Lower priority values first */ |
9c57a73b | 179 | return CMP(x->priority, y->priority); |
fd38203a LP |
180 | } |
181 | ||
c2ba3ad6 | 182 | static int earliest_time_prioq_compare(const void *a, const void *b) { |
fd38203a LP |
183 | const sd_event_source *x = a, *y = b; |
184 | ||
6a0f1f6d LP |
185 | assert(EVENT_SOURCE_IS_TIME(x->type)); |
186 | assert(x->type == y->type); | |
fd38203a | 187 | |
baf76283 LP |
188 | /* Enabled ones first */ |
189 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
fd38203a | 190 | return -1; |
baf76283 | 191 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) |
fd38203a LP |
192 | return 1; |
193 | ||
194 | /* Move the pending ones to the end */ | |
195 | if (!x->pending && y->pending) | |
196 | return -1; | |
197 | if (x->pending && !y->pending) | |
198 | return 1; | |
199 | ||
200 | /* Order by time */ | |
9c57a73b | 201 | return CMP(x->time.next, y->time.next); |
fd38203a LP |
202 | } |
203 | ||
1bce0ffa LP |
204 | static usec_t time_event_source_latest(const sd_event_source *s) { |
205 | return usec_add(s->time.next, s->time.accuracy); | |
206 | } | |
207 | ||
c2ba3ad6 LP |
208 | static int latest_time_prioq_compare(const void *a, const void *b) { |
209 | const sd_event_source *x = a, *y = b; | |
210 | ||
6a0f1f6d LP |
211 | assert(EVENT_SOURCE_IS_TIME(x->type)); |
212 | assert(x->type == y->type); | |
c2ba3ad6 | 213 | |
baf76283 LP |
214 | /* Enabled ones first */ |
215 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
c2ba3ad6 | 216 | return -1; |
baf76283 | 217 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) |
c2ba3ad6 LP |
218 | return 1; |
219 | ||
220 | /* Move the pending ones to the end */ | |
221 | if (!x->pending && y->pending) | |
222 | return -1; | |
223 | if (x->pending && !y->pending) | |
224 | return 1; | |
225 | ||
226 | /* Order by time */ | |
9c57a73b | 227 | return CMP(time_event_source_latest(x), time_event_source_latest(y)); |
c2ba3ad6 LP |
228 | } |
229 | ||
6203e07a | 230 | static int exit_prioq_compare(const void *a, const void *b) { |
da7e457c LP |
231 | const sd_event_source *x = a, *y = b; |
232 | ||
6203e07a LP |
233 | assert(x->type == SOURCE_EXIT); |
234 | assert(y->type == SOURCE_EXIT); | |
da7e457c | 235 | |
baf76283 LP |
236 | /* Enabled ones first */ |
237 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
da7e457c | 238 | return -1; |
baf76283 | 239 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) |
da7e457c LP |
240 | return 1; |
241 | ||
242 | /* Lower priority values first */ | |
6dd91b36 | 243 | return CMP(x->priority, y->priority); |
da7e457c LP |
244 | } |
245 | ||
6a0f1f6d LP |
246 | static void free_clock_data(struct clock_data *d) { |
247 | assert(d); | |
9da4cb2b | 248 | assert(d->wakeup == WAKEUP_CLOCK_DATA); |
6a0f1f6d LP |
249 | |
250 | safe_close(d->fd); | |
251 | prioq_free(d->earliest); | |
252 | prioq_free(d->latest); | |
253 | } | |
254 | ||
8301aa0b | 255 | static sd_event *event_free(sd_event *e) { |
a71fe8b8 LP |
256 | sd_event_source *s; |
257 | ||
fd38203a | 258 | assert(e); |
a71fe8b8 LP |
259 | |
260 | while ((s = e->sources)) { | |
261 | assert(s->floating); | |
262 | source_disconnect(s); | |
263 | sd_event_source_unref(s); | |
264 | } | |
265 | ||
15b38f93 | 266 | assert(e->n_sources == 0); |
fd38203a | 267 | |
afc6adb5 LP |
268 | if (e->default_event_ptr) |
269 | *(e->default_event_ptr) = NULL; | |
270 | ||
03e334a1 | 271 | safe_close(e->epoll_fd); |
03e334a1 | 272 | safe_close(e->watchdog_fd); |
cde93897 | 273 | |
6a0f1f6d | 274 | free_clock_data(&e->realtime); |
a8548816 | 275 | free_clock_data(&e->boottime); |
6a0f1f6d LP |
276 | free_clock_data(&e->monotonic); |
277 | free_clock_data(&e->realtime_alarm); | |
278 | free_clock_data(&e->boottime_alarm); | |
279 | ||
fd38203a LP |
280 | prioq_free(e->pending); |
281 | prioq_free(e->prepare); | |
6203e07a | 282 | prioq_free(e->exit); |
fd38203a LP |
283 | |
284 | free(e->signal_sources); | |
9da4cb2b | 285 | hashmap_free(e->signal_data); |
fd38203a | 286 | |
97ef5391 LP |
287 | hashmap_free(e->inotify_data); |
288 | ||
fd38203a | 289 | hashmap_free(e->child_sources); |
6e9feda3 | 290 | set_free(e->post_sources); |
8301aa0b | 291 | |
5cddd924 LP |
292 | free(e->event_queue); |
293 | ||
8301aa0b | 294 | return mfree(e); |
fd38203a LP |
295 | } |
296 | ||
f7262a9f | 297 | _public_ int sd_event_new(sd_event** ret) { |
fd38203a LP |
298 | sd_event *e; |
299 | int r; | |
300 | ||
305f78bf | 301 | assert_return(ret, -EINVAL); |
fd38203a | 302 | |
d08eb1fa | 303 | e = new(sd_event, 1); |
fd38203a LP |
304 | if (!e) |
305 | return -ENOMEM; | |
306 | ||
d08eb1fa LP |
307 | *e = (sd_event) { |
308 | .n_ref = 1, | |
309 | .epoll_fd = -1, | |
310 | .watchdog_fd = -1, | |
311 | .realtime.wakeup = WAKEUP_CLOCK_DATA, | |
312 | .realtime.fd = -1, | |
313 | .realtime.next = USEC_INFINITY, | |
314 | .boottime.wakeup = WAKEUP_CLOCK_DATA, | |
315 | .boottime.fd = -1, | |
316 | .boottime.next = USEC_INFINITY, | |
317 | .monotonic.wakeup = WAKEUP_CLOCK_DATA, | |
318 | .monotonic.fd = -1, | |
319 | .monotonic.next = USEC_INFINITY, | |
320 | .realtime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
321 | .realtime_alarm.fd = -1, | |
322 | .realtime_alarm.next = USEC_INFINITY, | |
323 | .boottime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
324 | .boottime_alarm.fd = -1, | |
325 | .boottime_alarm.next = USEC_INFINITY, | |
326 | .perturb = USEC_INFINITY, | |
327 | .original_pid = getpid_cached(), | |
328 | }; | |
fd38203a | 329 | |
c983e776 EV |
330 | r = prioq_ensure_allocated(&e->pending, pending_prioq_compare); |
331 | if (r < 0) | |
fd38203a | 332 | goto fail; |
fd38203a LP |
333 | |
334 | e->epoll_fd = epoll_create1(EPOLL_CLOEXEC); | |
335 | if (e->epoll_fd < 0) { | |
336 | r = -errno; | |
337 | goto fail; | |
338 | } | |
339 | ||
7fe2903c LP |
340 | e->epoll_fd = fd_move_above_stdio(e->epoll_fd); |
341 | ||
34b87517 | 342 | if (secure_getenv("SD_EVENT_PROFILE_DELAYS")) { |
34a6843d | 343 | log_debug("Event loop profiling enabled. Logarithmic histogram of event loop iterations in the range 2^0 ... 2^63 us will be logged every 5s."); |
34b87517 VC |
344 | e->profile_delays = true; |
345 | } | |
346 | ||
fd38203a LP |
347 | *ret = e; |
348 | return 0; | |
349 | ||
350 | fail: | |
351 | event_free(e); | |
352 | return r; | |
353 | } | |
354 | ||
8301aa0b | 355 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event, sd_event, event_free); |
fd38203a | 356 | |
afd15bbb ZJS |
357 | _public_ sd_event_source* sd_event_source_disable_unref(sd_event_source *s) { |
358 | if (s) | |
359 | (void) sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
360 | return sd_event_source_unref(s); | |
361 | } | |
362 | ||
eaa3cbef LP |
363 | static bool event_pid_changed(sd_event *e) { |
364 | assert(e); | |
365 | ||
a2360a46 | 366 | /* We don't support people creating an event loop and keeping |
eaa3cbef LP |
367 | * it around over a fork(). Let's complain. */ |
368 | ||
df0ff127 | 369 | return e->original_pid != getpid_cached(); |
eaa3cbef LP |
370 | } |
371 | ||
366e6411 | 372 | static void source_io_unregister(sd_event_source *s) { |
fd38203a LP |
373 | assert(s); |
374 | assert(s->type == SOURCE_IO); | |
375 | ||
f6806734 | 376 | if (event_pid_changed(s->event)) |
366e6411 | 377 | return; |
f6806734 | 378 | |
fd38203a | 379 | if (!s->io.registered) |
366e6411 | 380 | return; |
fd38203a | 381 | |
d1cf2023 | 382 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->io.fd, NULL) < 0) |
f80a5d6a | 383 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll, ignoring: %m", |
55cbfaa5 | 384 | strna(s->description), event_source_type_to_string(s->type)); |
fd38203a LP |
385 | |
386 | s->io.registered = false; | |
fd38203a LP |
387 | } |
388 | ||
305f78bf LP |
389 | static int source_io_register( |
390 | sd_event_source *s, | |
391 | int enabled, | |
392 | uint32_t events) { | |
393 | ||
fd38203a LP |
394 | assert(s); |
395 | assert(s->type == SOURCE_IO); | |
baf76283 | 396 | assert(enabled != SD_EVENT_OFF); |
fd38203a | 397 | |
1eac7948 | 398 | struct epoll_event ev = { |
a82f89aa LP |
399 | .events = events | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0), |
400 | .data.ptr = s, | |
401 | }; | |
fd38203a | 402 | |
15c689d7 | 403 | if (epoll_ctl(s->event->epoll_fd, |
1eac7948 | 404 | s->io.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, |
55c540d3 | 405 | s->io.fd, &ev) < 0) |
fd38203a LP |
406 | return -errno; |
407 | ||
408 | s->io.registered = true; | |
409 | ||
410 | return 0; | |
411 | } | |
412 | ||
f8f3f926 LP |
413 | static void source_child_pidfd_unregister(sd_event_source *s) { |
414 | assert(s); | |
415 | assert(s->type == SOURCE_CHILD); | |
416 | ||
417 | if (event_pid_changed(s->event)) | |
418 | return; | |
419 | ||
420 | if (!s->child.registered) | |
421 | return; | |
422 | ||
423 | if (EVENT_SOURCE_WATCH_PIDFD(s)) | |
424 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->child.pidfd, NULL) < 0) | |
f80a5d6a | 425 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll, ignoring: %m", |
f8f3f926 LP |
426 | strna(s->description), event_source_type_to_string(s->type)); |
427 | ||
428 | s->child.registered = false; | |
429 | } | |
430 | ||
431 | static int source_child_pidfd_register(sd_event_source *s, int enabled) { | |
f8f3f926 LP |
432 | assert(s); |
433 | assert(s->type == SOURCE_CHILD); | |
434 | assert(enabled != SD_EVENT_OFF); | |
435 | ||
436 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { | |
1eac7948 | 437 | struct epoll_event ev = { |
f8f3f926 LP |
438 | .events = EPOLLIN | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0), |
439 | .data.ptr = s, | |
440 | }; | |
441 | ||
55c540d3 ZJS |
442 | if (epoll_ctl(s->event->epoll_fd, |
443 | s->child.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, | |
444 | s->child.pidfd, &ev) < 0) | |
f8f3f926 LP |
445 | return -errno; |
446 | } | |
447 | ||
448 | s->child.registered = true; | |
449 | return 0; | |
450 | } | |
451 | ||
6a0f1f6d LP |
452 | static clockid_t event_source_type_to_clock(EventSourceType t) { |
453 | ||
454 | switch (t) { | |
455 | ||
456 | case SOURCE_TIME_REALTIME: | |
457 | return CLOCK_REALTIME; | |
458 | ||
a8548816 TG |
459 | case SOURCE_TIME_BOOTTIME: |
460 | return CLOCK_BOOTTIME; | |
461 | ||
6a0f1f6d LP |
462 | case SOURCE_TIME_MONOTONIC: |
463 | return CLOCK_MONOTONIC; | |
464 | ||
465 | case SOURCE_TIME_REALTIME_ALARM: | |
466 | return CLOCK_REALTIME_ALARM; | |
467 | ||
468 | case SOURCE_TIME_BOOTTIME_ALARM: | |
469 | return CLOCK_BOOTTIME_ALARM; | |
470 | ||
471 | default: | |
472 | return (clockid_t) -1; | |
473 | } | |
474 | } | |
475 | ||
476 | static EventSourceType clock_to_event_source_type(clockid_t clock) { | |
477 | ||
478 | switch (clock) { | |
479 | ||
480 | case CLOCK_REALTIME: | |
481 | return SOURCE_TIME_REALTIME; | |
482 | ||
a8548816 TG |
483 | case CLOCK_BOOTTIME: |
484 | return SOURCE_TIME_BOOTTIME; | |
485 | ||
6a0f1f6d LP |
486 | case CLOCK_MONOTONIC: |
487 | return SOURCE_TIME_MONOTONIC; | |
488 | ||
489 | case CLOCK_REALTIME_ALARM: | |
490 | return SOURCE_TIME_REALTIME_ALARM; | |
491 | ||
492 | case CLOCK_BOOTTIME_ALARM: | |
493 | return SOURCE_TIME_BOOTTIME_ALARM; | |
494 | ||
495 | default: | |
496 | return _SOURCE_EVENT_SOURCE_TYPE_INVALID; | |
497 | } | |
498 | } | |
499 | ||
500 | static struct clock_data* event_get_clock_data(sd_event *e, EventSourceType t) { | |
501 | assert(e); | |
502 | ||
503 | switch (t) { | |
504 | ||
505 | case SOURCE_TIME_REALTIME: | |
506 | return &e->realtime; | |
507 | ||
a8548816 TG |
508 | case SOURCE_TIME_BOOTTIME: |
509 | return &e->boottime; | |
510 | ||
6a0f1f6d LP |
511 | case SOURCE_TIME_MONOTONIC: |
512 | return &e->monotonic; | |
513 | ||
514 | case SOURCE_TIME_REALTIME_ALARM: | |
515 | return &e->realtime_alarm; | |
516 | ||
517 | case SOURCE_TIME_BOOTTIME_ALARM: | |
518 | return &e->boottime_alarm; | |
519 | ||
520 | default: | |
521 | return NULL; | |
522 | } | |
523 | } | |
524 | ||
3e4eb8e7 YW |
525 | static void event_free_signal_data(sd_event *e, struct signal_data *d) { |
526 | assert(e); | |
527 | ||
528 | if (!d) | |
529 | return; | |
530 | ||
531 | hashmap_remove(e->signal_data, &d->priority); | |
532 | safe_close(d->fd); | |
533 | free(d); | |
534 | } | |
535 | ||
9da4cb2b LP |
536 | static int event_make_signal_data( |
537 | sd_event *e, | |
538 | int sig, | |
539 | struct signal_data **ret) { | |
4807d2d0 | 540 | |
9da4cb2b LP |
541 | struct signal_data *d; |
542 | bool added = false; | |
543 | sigset_t ss_copy; | |
544 | int64_t priority; | |
f95387cd ZJS |
545 | int r; |
546 | ||
547 | assert(e); | |
548 | ||
f6806734 | 549 | if (event_pid_changed(e)) |
9da4cb2b | 550 | return -ECHILD; |
f6806734 | 551 | |
9da4cb2b LP |
552 | if (e->signal_sources && e->signal_sources[sig]) |
553 | priority = e->signal_sources[sig]->priority; | |
554 | else | |
de05913d | 555 | priority = SD_EVENT_PRIORITY_NORMAL; |
f95387cd | 556 | |
9da4cb2b LP |
557 | d = hashmap_get(e->signal_data, &priority); |
558 | if (d) { | |
559 | if (sigismember(&d->sigset, sig) > 0) { | |
560 | if (ret) | |
561 | *ret = d; | |
562 | return 0; | |
563 | } | |
564 | } else { | |
565 | r = hashmap_ensure_allocated(&e->signal_data, &uint64_hash_ops); | |
566 | if (r < 0) | |
567 | return r; | |
568 | ||
d08eb1fa | 569 | d = new(struct signal_data, 1); |
9da4cb2b LP |
570 | if (!d) |
571 | return -ENOMEM; | |
572 | ||
d08eb1fa LP |
573 | *d = (struct signal_data) { |
574 | .wakeup = WAKEUP_SIGNAL_DATA, | |
575 | .fd = -1, | |
576 | .priority = priority, | |
577 | }; | |
9da4cb2b LP |
578 | |
579 | r = hashmap_put(e->signal_data, &d->priority, d); | |
90f604d1 ZJS |
580 | if (r < 0) { |
581 | free(d); | |
9da4cb2b | 582 | return r; |
90f604d1 | 583 | } |
f95387cd | 584 | |
9da4cb2b LP |
585 | added = true; |
586 | } | |
587 | ||
588 | ss_copy = d->sigset; | |
589 | assert_se(sigaddset(&ss_copy, sig) >= 0); | |
590 | ||
591 | r = signalfd(d->fd, &ss_copy, SFD_NONBLOCK|SFD_CLOEXEC); | |
592 | if (r < 0) { | |
593 | r = -errno; | |
594 | goto fail; | |
595 | } | |
596 | ||
597 | d->sigset = ss_copy; | |
f95387cd | 598 | |
9da4cb2b LP |
599 | if (d->fd >= 0) { |
600 | if (ret) | |
601 | *ret = d; | |
f95387cd | 602 | return 0; |
9da4cb2b LP |
603 | } |
604 | ||
7fe2903c | 605 | d->fd = fd_move_above_stdio(r); |
f95387cd | 606 | |
1eac7948 | 607 | struct epoll_event ev = { |
a82f89aa LP |
608 | .events = EPOLLIN, |
609 | .data.ptr = d, | |
610 | }; | |
f95387cd | 611 | |
15c689d7 | 612 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { |
9da4cb2b LP |
613 | r = -errno; |
614 | goto fail; | |
f95387cd ZJS |
615 | } |
616 | ||
9da4cb2b LP |
617 | if (ret) |
618 | *ret = d; | |
619 | ||
f95387cd | 620 | return 0; |
9da4cb2b LP |
621 | |
622 | fail: | |
3e4eb8e7 YW |
623 | if (added) |
624 | event_free_signal_data(e, d); | |
9da4cb2b LP |
625 | |
626 | return r; | |
627 | } | |
628 | ||
629 | static void event_unmask_signal_data(sd_event *e, struct signal_data *d, int sig) { | |
630 | assert(e); | |
631 | assert(d); | |
632 | ||
633 | /* Turns off the specified signal in the signal data | |
634 | * object. If the signal mask of the object becomes empty that | |
635 | * way removes it. */ | |
636 | ||
637 | if (sigismember(&d->sigset, sig) == 0) | |
638 | return; | |
639 | ||
640 | assert_se(sigdelset(&d->sigset, sig) >= 0); | |
641 | ||
642 | if (sigisemptyset(&d->sigset)) { | |
9da4cb2b | 643 | /* If all the mask is all-zero we can get rid of the structure */ |
3e4eb8e7 | 644 | event_free_signal_data(e, d); |
9da4cb2b LP |
645 | return; |
646 | } | |
647 | ||
648 | assert(d->fd >= 0); | |
649 | ||
650 | if (signalfd(d->fd, &d->sigset, SFD_NONBLOCK|SFD_CLOEXEC) < 0) | |
651 | log_debug_errno(errno, "Failed to unset signal bit, ignoring: %m"); | |
652 | } | |
653 | ||
654 | static void event_gc_signal_data(sd_event *e, const int64_t *priority, int sig) { | |
655 | struct signal_data *d; | |
656 | static const int64_t zero_priority = 0; | |
657 | ||
658 | assert(e); | |
659 | ||
f8f3f926 LP |
660 | /* Rechecks if the specified signal is still something we are interested in. If not, we'll unmask it, |
661 | * and possibly drop the signalfd for it. */ | |
9da4cb2b LP |
662 | |
663 | if (sig == SIGCHLD && | |
664 | e->n_enabled_child_sources > 0) | |
665 | return; | |
666 | ||
667 | if (e->signal_sources && | |
668 | e->signal_sources[sig] && | |
669 | e->signal_sources[sig]->enabled != SD_EVENT_OFF) | |
670 | return; | |
671 | ||
672 | /* | |
673 | * The specified signal might be enabled in three different queues: | |
674 | * | |
675 | * 1) the one that belongs to the priority passed (if it is non-NULL) | |
676 | * 2) the one that belongs to the priority of the event source of the signal (if there is one) | |
677 | * 3) the 0 priority (to cover the SIGCHLD case) | |
678 | * | |
679 | * Hence, let's remove it from all three here. | |
680 | */ | |
681 | ||
682 | if (priority) { | |
683 | d = hashmap_get(e->signal_data, priority); | |
684 | if (d) | |
685 | event_unmask_signal_data(e, d, sig); | |
686 | } | |
687 | ||
688 | if (e->signal_sources && e->signal_sources[sig]) { | |
689 | d = hashmap_get(e->signal_data, &e->signal_sources[sig]->priority); | |
690 | if (d) | |
691 | event_unmask_signal_data(e, d, sig); | |
692 | } | |
693 | ||
694 | d = hashmap_get(e->signal_data, &zero_priority); | |
695 | if (d) | |
696 | event_unmask_signal_data(e, d, sig); | |
f95387cd ZJS |
697 | } |
698 | ||
e1951c16 MS |
699 | static void event_source_pp_prioq_reshuffle(sd_event_source *s) { |
700 | assert(s); | |
701 | ||
702 | /* Reshuffles the pending + prepare prioqs. Called whenever the dispatch order changes, i.e. when | |
703 | * they are enabled/disabled or marked pending and such. */ | |
704 | ||
705 | if (s->pending) | |
706 | prioq_reshuffle(s->event->pending, s, &s->pending_index); | |
707 | ||
708 | if (s->prepare) | |
709 | prioq_reshuffle(s->event->prepare, s, &s->prepare_index); | |
710 | } | |
711 | ||
712 | static void event_source_time_prioq_reshuffle(sd_event_source *s) { | |
713 | struct clock_data *d; | |
714 | ||
715 | assert(s); | |
716 | assert(EVENT_SOURCE_IS_TIME(s->type)); | |
717 | ||
718 | /* Called whenever the event source's timer ordering properties changed, i.e. time, accuracy, | |
719 | * pending, enable state. Makes sure the two prioq's are ordered properly again. */ | |
720 | assert_se(d = event_get_clock_data(s->event, s->type)); | |
f41315fc LP |
721 | prioq_reshuffle(d->earliest, s, &s->earliest_index); |
722 | prioq_reshuffle(d->latest, s, &s->latest_index); | |
e1951c16 MS |
723 | d->needs_rearm = true; |
724 | } | |
725 | ||
1e45e3fe LP |
726 | static void event_source_time_prioq_remove( |
727 | sd_event_source *s, | |
728 | struct clock_data *d) { | |
729 | ||
730 | assert(s); | |
731 | assert(d); | |
732 | ||
f41315fc LP |
733 | prioq_remove(d->earliest, s, &s->earliest_index); |
734 | prioq_remove(d->latest, s, &s->latest_index); | |
735 | s->earliest_index = s->latest_index = PRIOQ_IDX_NULL; | |
1e45e3fe LP |
736 | d->needs_rearm = true; |
737 | } | |
738 | ||
a71fe8b8 LP |
739 | static void source_disconnect(sd_event_source *s) { |
740 | sd_event *event; | |
741 | ||
fd38203a LP |
742 | assert(s); |
743 | ||
a71fe8b8 LP |
744 | if (!s->event) |
745 | return; | |
15b38f93 | 746 | |
a71fe8b8 | 747 | assert(s->event->n_sources > 0); |
fd38203a | 748 | |
a71fe8b8 | 749 | switch (s->type) { |
fd38203a | 750 | |
a71fe8b8 LP |
751 | case SOURCE_IO: |
752 | if (s->io.fd >= 0) | |
753 | source_io_unregister(s); | |
fd38203a | 754 | |
a71fe8b8 | 755 | break; |
6a0f1f6d | 756 | |
a71fe8b8 | 757 | case SOURCE_TIME_REALTIME: |
a8548816 | 758 | case SOURCE_TIME_BOOTTIME: |
a71fe8b8 LP |
759 | case SOURCE_TIME_MONOTONIC: |
760 | case SOURCE_TIME_REALTIME_ALARM: | |
761 | case SOURCE_TIME_BOOTTIME_ALARM: { | |
762 | struct clock_data *d; | |
1e45e3fe LP |
763 | assert_se(d = event_get_clock_data(s->event, s->type)); |
764 | event_source_time_prioq_remove(s, d); | |
a71fe8b8 LP |
765 | break; |
766 | } | |
767 | ||
768 | case SOURCE_SIGNAL: | |
769 | if (s->signal.sig > 0) { | |
9da4cb2b | 770 | |
a71fe8b8 LP |
771 | if (s->event->signal_sources) |
772 | s->event->signal_sources[s->signal.sig] = NULL; | |
4807d2d0 | 773 | |
9da4cb2b | 774 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
6a0f1f6d | 775 | } |
fd38203a | 776 | |
a71fe8b8 | 777 | break; |
fd38203a | 778 | |
a71fe8b8 LP |
779 | case SOURCE_CHILD: |
780 | if (s->child.pid > 0) { | |
781 | if (s->enabled != SD_EVENT_OFF) { | |
782 | assert(s->event->n_enabled_child_sources > 0); | |
783 | s->event->n_enabled_child_sources--; | |
4807d2d0 | 784 | } |
fd38203a | 785 | |
4a0b58c4 | 786 | (void) hashmap_remove(s->event->child_sources, PID_TO_PTR(s->child.pid)); |
a71fe8b8 | 787 | } |
fd38203a | 788 | |
f8f3f926 LP |
789 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
790 | source_child_pidfd_unregister(s); | |
791 | else | |
792 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
793 | ||
a71fe8b8 | 794 | break; |
fd38203a | 795 | |
a71fe8b8 LP |
796 | case SOURCE_DEFER: |
797 | /* nothing */ | |
798 | break; | |
fd38203a | 799 | |
a71fe8b8 LP |
800 | case SOURCE_POST: |
801 | set_remove(s->event->post_sources, s); | |
802 | break; | |
da7e457c | 803 | |
a71fe8b8 LP |
804 | case SOURCE_EXIT: |
805 | prioq_remove(s->event->exit, s, &s->exit.prioq_index); | |
806 | break; | |
0eb2e0e3 | 807 | |
97ef5391 LP |
808 | case SOURCE_INOTIFY: { |
809 | struct inode_data *inode_data; | |
810 | ||
811 | inode_data = s->inotify.inode_data; | |
812 | if (inode_data) { | |
813 | struct inotify_data *inotify_data; | |
814 | assert_se(inotify_data = inode_data->inotify_data); | |
815 | ||
816 | /* Detach this event source from the inode object */ | |
817 | LIST_REMOVE(inotify.by_inode_data, inode_data->event_sources, s); | |
818 | s->inotify.inode_data = NULL; | |
819 | ||
820 | if (s->pending) { | |
821 | assert(inotify_data->n_pending > 0); | |
822 | inotify_data->n_pending--; | |
823 | } | |
824 | ||
825 | /* Note that we don't reduce the inotify mask for the watch descriptor here if the inode is | |
826 | * continued to being watched. That's because inotify doesn't really have an API for that: we | |
827 | * can only change watch masks with access to the original inode either by fd or by path. But | |
828 | * paths aren't stable, and keeping an O_PATH fd open all the time would mean wasting an fd | |
f21f31b2 | 829 | * continuously and keeping the mount busy which we can't really do. We could reconstruct the |
97ef5391 LP |
830 | * original inode from /proc/self/fdinfo/$INOTIFY_FD (as all watch descriptors are listed |
831 | * there), but given the need for open_by_handle_at() which is privileged and not universally | |
832 | * available this would be quite an incomplete solution. Hence we go the other way, leave the | |
833 | * mask set, even if it is not minimized now, and ignore all events we aren't interested in | |
834 | * anymore after reception. Yes, this sucks, but … Linux … */ | |
835 | ||
836 | /* Maybe release the inode data (and its inotify) */ | |
837 | event_gc_inode_data(s->event, inode_data); | |
838 | } | |
839 | ||
840 | break; | |
841 | } | |
842 | ||
a71fe8b8 LP |
843 | default: |
844 | assert_not_reached("Wut? I shouldn't exist."); | |
845 | } | |
6e9feda3 | 846 | |
a71fe8b8 LP |
847 | if (s->pending) |
848 | prioq_remove(s->event->pending, s, &s->pending_index); | |
9d3e3aa5 | 849 | |
a71fe8b8 LP |
850 | if (s->prepare) |
851 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
fd38203a | 852 | |
e514aa1e | 853 | event = TAKE_PTR(s->event); |
a71fe8b8 LP |
854 | LIST_REMOVE(sources, event->sources, s); |
855 | event->n_sources--; | |
fd38203a | 856 | |
f5982559 LP |
857 | /* Note that we don't invalidate the type here, since we still need it in order to close the fd or |
858 | * pidfd associated with this event source, which we'll do only on source_free(). */ | |
859 | ||
a71fe8b8 LP |
860 | if (!s->floating) |
861 | sd_event_unref(event); | |
862 | } | |
863 | ||
864 | static void source_free(sd_event_source *s) { | |
865 | assert(s); | |
fd38203a | 866 | |
a71fe8b8 | 867 | source_disconnect(s); |
ab93297c NM |
868 | |
869 | if (s->type == SOURCE_IO && s->io.owned) | |
15723a1d LP |
870 | s->io.fd = safe_close(s->io.fd); |
871 | ||
f8f3f926 LP |
872 | if (s->type == SOURCE_CHILD) { |
873 | /* Eventually the kernel will do this automatically for us, but for now let's emulate this (unreliably) in userspace. */ | |
874 | ||
875 | if (s->child.process_owned) { | |
876 | ||
877 | if (!s->child.exited) { | |
878 | bool sent = false; | |
879 | ||
880 | if (s->child.pidfd >= 0) { | |
881 | if (pidfd_send_signal(s->child.pidfd, SIGKILL, NULL, 0) < 0) { | |
882 | if (errno == ESRCH) /* Already dead */ | |
883 | sent = true; | |
884 | else if (!ERRNO_IS_NOT_SUPPORTED(errno)) | |
885 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via pidfd_send_signal(), re-trying via kill(): %m", | |
886 | s->child.pid); | |
887 | } else | |
888 | sent = true; | |
889 | } | |
890 | ||
891 | if (!sent) | |
892 | if (kill(s->child.pid, SIGKILL) < 0) | |
893 | if (errno != ESRCH) /* Already dead */ | |
894 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via kill(), ignoring: %m", | |
895 | s->child.pid); | |
896 | } | |
897 | ||
898 | if (!s->child.waited) { | |
899 | siginfo_t si = {}; | |
900 | ||
901 | /* Reap the child if we can */ | |
902 | (void) waitid(P_PID, s->child.pid, &si, WEXITED); | |
903 | } | |
904 | } | |
905 | ||
906 | if (s->child.pidfd_owned) | |
907 | s->child.pidfd = safe_close(s->child.pidfd); | |
908 | } | |
909 | ||
15723a1d LP |
910 | if (s->destroy_callback) |
911 | s->destroy_callback(s->userdata); | |
ab93297c | 912 | |
356779df | 913 | free(s->description); |
fd38203a LP |
914 | free(s); |
915 | } | |
8c75fe17 | 916 | DEFINE_TRIVIAL_CLEANUP_FUNC(sd_event_source*, source_free); |
fd38203a LP |
917 | |
918 | static int source_set_pending(sd_event_source *s, bool b) { | |
919 | int r; | |
920 | ||
921 | assert(s); | |
6203e07a | 922 | assert(s->type != SOURCE_EXIT); |
fd38203a LP |
923 | |
924 | if (s->pending == b) | |
925 | return 0; | |
926 | ||
927 | s->pending = b; | |
928 | ||
929 | if (b) { | |
930 | s->pending_iteration = s->event->iteration; | |
931 | ||
932 | r = prioq_put(s->event->pending, s, &s->pending_index); | |
933 | if (r < 0) { | |
934 | s->pending = false; | |
935 | return r; | |
936 | } | |
937 | } else | |
938 | assert_se(prioq_remove(s->event->pending, s, &s->pending_index)); | |
939 | ||
e1951c16 MS |
940 | if (EVENT_SOURCE_IS_TIME(s->type)) |
941 | event_source_time_prioq_reshuffle(s); | |
2576a19e | 942 | |
9da4cb2b LP |
943 | if (s->type == SOURCE_SIGNAL && !b) { |
944 | struct signal_data *d; | |
945 | ||
946 | d = hashmap_get(s->event->signal_data, &s->priority); | |
947 | if (d && d->current == s) | |
948 | d->current = NULL; | |
949 | } | |
950 | ||
97ef5391 LP |
951 | if (s->type == SOURCE_INOTIFY) { |
952 | ||
953 | assert(s->inotify.inode_data); | |
954 | assert(s->inotify.inode_data->inotify_data); | |
955 | ||
956 | if (b) | |
957 | s->inotify.inode_data->inotify_data->n_pending ++; | |
958 | else { | |
959 | assert(s->inotify.inode_data->inotify_data->n_pending > 0); | |
960 | s->inotify.inode_data->inotify_data->n_pending --; | |
961 | } | |
962 | } | |
963 | ||
fd38203a LP |
964 | return 0; |
965 | } | |
966 | ||
a71fe8b8 | 967 | static sd_event_source *source_new(sd_event *e, bool floating, EventSourceType type) { |
fd38203a LP |
968 | sd_event_source *s; |
969 | ||
970 | assert(e); | |
971 | ||
d08eb1fa | 972 | s = new(sd_event_source, 1); |
fd38203a LP |
973 | if (!s) |
974 | return NULL; | |
975 | ||
d08eb1fa LP |
976 | *s = (struct sd_event_source) { |
977 | .n_ref = 1, | |
978 | .event = e, | |
979 | .floating = floating, | |
980 | .type = type, | |
981 | .pending_index = PRIOQ_IDX_NULL, | |
982 | .prepare_index = PRIOQ_IDX_NULL, | |
983 | }; | |
a71fe8b8 LP |
984 | |
985 | if (!floating) | |
986 | sd_event_ref(e); | |
fd38203a | 987 | |
a71fe8b8 | 988 | LIST_PREPEND(sources, e->sources, s); |
313cefa1 | 989 | e->n_sources++; |
15b38f93 | 990 | |
fd38203a LP |
991 | return s; |
992 | } | |
993 | ||
b9350e70 LP |
994 | static int io_exit_callback(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
995 | assert(s); | |
996 | ||
997 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
998 | } | |
999 | ||
f7262a9f | 1000 | _public_ int sd_event_add_io( |
fd38203a | 1001 | sd_event *e, |
151b9b96 | 1002 | sd_event_source **ret, |
fd38203a LP |
1003 | int fd, |
1004 | uint32_t events, | |
718db961 | 1005 | sd_event_io_handler_t callback, |
151b9b96 | 1006 | void *userdata) { |
fd38203a | 1007 | |
ec766a51 | 1008 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1009 | int r; |
1010 | ||
305f78bf | 1011 | assert_return(e, -EINVAL); |
b937d761 | 1012 | assert_return(e = event_resolve(e), -ENOPKG); |
8ac43fee | 1013 | assert_return(fd >= 0, -EBADF); |
2a16a986 | 1014 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 1015 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1016 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1017 | |
b9350e70 LP |
1018 | if (!callback) |
1019 | callback = io_exit_callback; | |
1020 | ||
a71fe8b8 | 1021 | s = source_new(e, !ret, SOURCE_IO); |
fd38203a LP |
1022 | if (!s) |
1023 | return -ENOMEM; | |
1024 | ||
9da4cb2b | 1025 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1026 | s->io.fd = fd; |
1027 | s->io.events = events; | |
1028 | s->io.callback = callback; | |
1029 | s->userdata = userdata; | |
baf76283 | 1030 | s->enabled = SD_EVENT_ON; |
fd38203a | 1031 | |
baf76283 | 1032 | r = source_io_register(s, s->enabled, events); |
ec766a51 | 1033 | if (r < 0) |
050f74f2 | 1034 | return r; |
fd38203a | 1035 | |
a71fe8b8 LP |
1036 | if (ret) |
1037 | *ret = s; | |
ec766a51 | 1038 | TAKE_PTR(s); |
a71fe8b8 | 1039 | |
fd38203a LP |
1040 | return 0; |
1041 | } | |
1042 | ||
52444dc4 LP |
1043 | static void initialize_perturb(sd_event *e) { |
1044 | sd_id128_t bootid = {}; | |
1045 | ||
1046 | /* When we sleep for longer, we try to realign the wakeup to | |
f21f31b2 | 1047 | the same time within each minute/second/250ms, so that |
52444dc4 LP |
1048 | events all across the system can be coalesced into a single |
1049 | CPU wakeup. However, let's take some system-specific | |
1050 | randomness for this value, so that in a network of systems | |
1051 | with synced clocks timer events are distributed a | |
1052 | bit. Here, we calculate a perturbation usec offset from the | |
1053 | boot ID. */ | |
1054 | ||
3a43da28 | 1055 | if (_likely_(e->perturb != USEC_INFINITY)) |
52444dc4 LP |
1056 | return; |
1057 | ||
1058 | if (sd_id128_get_boot(&bootid) >= 0) | |
1059 | e->perturb = (bootid.qwords[0] ^ bootid.qwords[1]) % USEC_PER_MINUTE; | |
1060 | } | |
1061 | ||
fd38203a LP |
1062 | static int event_setup_timer_fd( |
1063 | sd_event *e, | |
6a0f1f6d LP |
1064 | struct clock_data *d, |
1065 | clockid_t clock) { | |
fd38203a | 1066 | |
fd38203a | 1067 | assert(e); |
6a0f1f6d | 1068 | assert(d); |
fd38203a | 1069 | |
6a0f1f6d | 1070 | if (_likely_(d->fd >= 0)) |
fd38203a LP |
1071 | return 0; |
1072 | ||
b44d87e2 | 1073 | _cleanup_close_ int fd = -1; |
b44d87e2 | 1074 | |
6a0f1f6d | 1075 | fd = timerfd_create(clock, TFD_NONBLOCK|TFD_CLOEXEC); |
fd38203a LP |
1076 | if (fd < 0) |
1077 | return -errno; | |
1078 | ||
7fe2903c LP |
1079 | fd = fd_move_above_stdio(fd); |
1080 | ||
1eac7948 | 1081 | struct epoll_event ev = { |
a82f89aa LP |
1082 | .events = EPOLLIN, |
1083 | .data.ptr = d, | |
1084 | }; | |
fd38203a | 1085 | |
15c689d7 | 1086 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) |
fd38203a | 1087 | return -errno; |
fd38203a | 1088 | |
b44d87e2 | 1089 | d->fd = TAKE_FD(fd); |
fd38203a LP |
1090 | return 0; |
1091 | } | |
1092 | ||
c4f1aff2 TG |
1093 | static int time_exit_callback(sd_event_source *s, uint64_t usec, void *userdata) { |
1094 | assert(s); | |
1095 | ||
1096 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1097 | } | |
1098 | ||
41c63f36 LP |
1099 | static int setup_clock_data(sd_event *e, struct clock_data *d, clockid_t clock) { |
1100 | int r; | |
1101 | ||
1102 | assert(d); | |
1103 | ||
1104 | if (d->fd < 0) { | |
1105 | r = event_setup_timer_fd(e, d, clock); | |
1106 | if (r < 0) | |
1107 | return r; | |
1108 | } | |
1109 | ||
1110 | r = prioq_ensure_allocated(&d->earliest, earliest_time_prioq_compare); | |
1111 | if (r < 0) | |
1112 | return r; | |
1113 | ||
1114 | r = prioq_ensure_allocated(&d->latest, latest_time_prioq_compare); | |
1115 | if (r < 0) | |
1116 | return r; | |
1117 | ||
1118 | return 0; | |
1119 | } | |
1120 | ||
1e45e3fe LP |
1121 | static int event_source_time_prioq_put( |
1122 | sd_event_source *s, | |
1123 | struct clock_data *d) { | |
1124 | ||
1125 | int r; | |
1126 | ||
1127 | assert(s); | |
1128 | assert(d); | |
1129 | ||
f41315fc | 1130 | r = prioq_put(d->earliest, s, &s->earliest_index); |
1e45e3fe LP |
1131 | if (r < 0) |
1132 | return r; | |
1133 | ||
f41315fc | 1134 | r = prioq_put(d->latest, s, &s->latest_index); |
1e45e3fe | 1135 | if (r < 0) { |
f41315fc LP |
1136 | assert_se(prioq_remove(d->earliest, s, &s->earliest_index) > 0); |
1137 | s->earliest_index = PRIOQ_IDX_NULL; | |
1e45e3fe LP |
1138 | return r; |
1139 | } | |
1140 | ||
1141 | d->needs_rearm = true; | |
1142 | return 0; | |
1143 | } | |
1144 | ||
6a0f1f6d | 1145 | _public_ int sd_event_add_time( |
fd38203a | 1146 | sd_event *e, |
151b9b96 | 1147 | sd_event_source **ret, |
6a0f1f6d | 1148 | clockid_t clock, |
fd38203a | 1149 | uint64_t usec, |
c2ba3ad6 | 1150 | uint64_t accuracy, |
718db961 | 1151 | sd_event_time_handler_t callback, |
151b9b96 | 1152 | void *userdata) { |
fd38203a | 1153 | |
6a0f1f6d | 1154 | EventSourceType type; |
ec766a51 | 1155 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6a0f1f6d | 1156 | struct clock_data *d; |
fd38203a LP |
1157 | int r; |
1158 | ||
305f78bf | 1159 | assert_return(e, -EINVAL); |
b937d761 | 1160 | assert_return(e = event_resolve(e), -ENOPKG); |
305f78bf | 1161 | assert_return(accuracy != (uint64_t) -1, -EINVAL); |
da7e457c | 1162 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1163 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1164 | |
e475d10c LP |
1165 | if (!clock_supported(clock)) /* Checks whether the kernel supports the clock */ |
1166 | return -EOPNOTSUPP; | |
1167 | ||
1168 | type = clock_to_event_source_type(clock); /* checks whether sd-event supports this clock */ | |
1169 | if (type < 0) | |
3411372e LP |
1170 | return -EOPNOTSUPP; |
1171 | ||
c4f1aff2 TG |
1172 | if (!callback) |
1173 | callback = time_exit_callback; | |
1174 | ||
1e45e3fe | 1175 | assert_se(d = event_get_clock_data(e, type)); |
c2ba3ad6 | 1176 | |
41c63f36 | 1177 | r = setup_clock_data(e, d, clock); |
c983e776 EV |
1178 | if (r < 0) |
1179 | return r; | |
fd38203a | 1180 | |
a71fe8b8 | 1181 | s = source_new(e, !ret, type); |
fd38203a LP |
1182 | if (!s) |
1183 | return -ENOMEM; | |
1184 | ||
1185 | s->time.next = usec; | |
c2ba3ad6 | 1186 | s->time.accuracy = accuracy == 0 ? DEFAULT_ACCURACY_USEC : accuracy; |
fd38203a | 1187 | s->time.callback = callback; |
f41315fc | 1188 | s->earliest_index = s->latest_index = PRIOQ_IDX_NULL; |
fd38203a | 1189 | s->userdata = userdata; |
baf76283 | 1190 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1191 | |
1e45e3fe | 1192 | r = event_source_time_prioq_put(s, d); |
c2ba3ad6 | 1193 | if (r < 0) |
ec766a51 | 1194 | return r; |
fd38203a | 1195 | |
a71fe8b8 LP |
1196 | if (ret) |
1197 | *ret = s; | |
ec766a51 | 1198 | TAKE_PTR(s); |
a71fe8b8 | 1199 | |
fd38203a LP |
1200 | return 0; |
1201 | } | |
1202 | ||
d6a83dc4 LP |
1203 | _public_ int sd_event_add_time_relative( |
1204 | sd_event *e, | |
1205 | sd_event_source **ret, | |
1206 | clockid_t clock, | |
1207 | uint64_t usec, | |
1208 | uint64_t accuracy, | |
1209 | sd_event_time_handler_t callback, | |
1210 | void *userdata) { | |
1211 | ||
1212 | usec_t t; | |
1213 | int r; | |
1214 | ||
1215 | /* Same as sd_event_add_time() but operates relative to the event loop's current point in time, and | |
1216 | * checks for overflow. */ | |
1217 | ||
1218 | r = sd_event_now(e, clock, &t); | |
1219 | if (r < 0) | |
1220 | return r; | |
1221 | ||
1222 | if (usec >= USEC_INFINITY - t) | |
1223 | return -EOVERFLOW; | |
1224 | ||
1225 | return sd_event_add_time(e, ret, clock, t + usec, accuracy, callback, userdata); | |
1226 | } | |
1227 | ||
59bc1fd7 LP |
1228 | static int signal_exit_callback(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) { |
1229 | assert(s); | |
1230 | ||
1231 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1232 | } | |
1233 | ||
f7262a9f | 1234 | _public_ int sd_event_add_signal( |
305f78bf | 1235 | sd_event *e, |
151b9b96 | 1236 | sd_event_source **ret, |
305f78bf | 1237 | int sig, |
718db961 | 1238 | sd_event_signal_handler_t callback, |
151b9b96 | 1239 | void *userdata) { |
305f78bf | 1240 | |
ec766a51 | 1241 | _cleanup_(source_freep) sd_event_source *s = NULL; |
9da4cb2b | 1242 | struct signal_data *d; |
fd38203a LP |
1243 | int r; |
1244 | ||
305f78bf | 1245 | assert_return(e, -EINVAL); |
b937d761 | 1246 | assert_return(e = event_resolve(e), -ENOPKG); |
6eb7c172 | 1247 | assert_return(SIGNAL_VALID(sig), -EINVAL); |
da7e457c | 1248 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1249 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1250 | |
59bc1fd7 LP |
1251 | if (!callback) |
1252 | callback = signal_exit_callback; | |
1253 | ||
d1b75241 LP |
1254 | r = signal_is_blocked(sig); |
1255 | if (r < 0) | |
1256 | return r; | |
1257 | if (r == 0) | |
3022d74b LP |
1258 | return -EBUSY; |
1259 | ||
fd38203a LP |
1260 | if (!e->signal_sources) { |
1261 | e->signal_sources = new0(sd_event_source*, _NSIG); | |
1262 | if (!e->signal_sources) | |
1263 | return -ENOMEM; | |
1264 | } else if (e->signal_sources[sig]) | |
1265 | return -EBUSY; | |
1266 | ||
a71fe8b8 | 1267 | s = source_new(e, !ret, SOURCE_SIGNAL); |
fd38203a LP |
1268 | if (!s) |
1269 | return -ENOMEM; | |
1270 | ||
1271 | s->signal.sig = sig; | |
1272 | s->signal.callback = callback; | |
1273 | s->userdata = userdata; | |
baf76283 | 1274 | s->enabled = SD_EVENT_ON; |
fd38203a LP |
1275 | |
1276 | e->signal_sources[sig] = s; | |
fd38203a | 1277 | |
9da4cb2b | 1278 | r = event_make_signal_data(e, sig, &d); |
ec766a51 | 1279 | if (r < 0) |
9da4cb2b | 1280 | return r; |
fd38203a | 1281 | |
f1f00dbb LP |
1282 | /* Use the signal name as description for the event source by default */ |
1283 | (void) sd_event_source_set_description(s, signal_to_string(sig)); | |
1284 | ||
a71fe8b8 LP |
1285 | if (ret) |
1286 | *ret = s; | |
ec766a51 | 1287 | TAKE_PTR(s); |
a71fe8b8 | 1288 | |
fd38203a LP |
1289 | return 0; |
1290 | } | |
1291 | ||
b9350e70 LP |
1292 | static int child_exit_callback(sd_event_source *s, const siginfo_t *si, void *userdata) { |
1293 | assert(s); | |
1294 | ||
1295 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1296 | } | |
1297 | ||
f8f3f926 LP |
1298 | static bool shall_use_pidfd(void) { |
1299 | /* Mostly relevant for debugging, i.e. this is used in test-event.c to test the event loop once with and once without pidfd */ | |
1300 | return getenv_bool_secure("SYSTEMD_PIDFD") != 0; | |
1301 | } | |
1302 | ||
f7262a9f | 1303 | _public_ int sd_event_add_child( |
305f78bf | 1304 | sd_event *e, |
151b9b96 | 1305 | sd_event_source **ret, |
305f78bf LP |
1306 | pid_t pid, |
1307 | int options, | |
718db961 | 1308 | sd_event_child_handler_t callback, |
151b9b96 | 1309 | void *userdata) { |
305f78bf | 1310 | |
ec766a51 | 1311 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1312 | int r; |
1313 | ||
305f78bf | 1314 | assert_return(e, -EINVAL); |
b937d761 | 1315 | assert_return(e = event_resolve(e), -ENOPKG); |
305f78bf LP |
1316 | assert_return(pid > 1, -EINVAL); |
1317 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1318 | assert_return(options != 0, -EINVAL); | |
da7e457c | 1319 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1320 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1321 | |
b9350e70 LP |
1322 | if (!callback) |
1323 | callback = child_exit_callback; | |
1324 | ||
ee880b37 LP |
1325 | if (e->n_enabled_child_sources == 0) { |
1326 | /* Caller must block SIGCHLD before using us to watch children, even if pidfd is available, | |
1327 | * for compatibility with pre-pidfd and because we don't want the reap the child processes | |
1328 | * ourselves, i.e. call waitid(), and don't want Linux' default internal logic for that to | |
1329 | * take effect. | |
1330 | * | |
1331 | * (As an optimization we only do this check on the first child event source created.) */ | |
1332 | r = signal_is_blocked(SIGCHLD); | |
1333 | if (r < 0) | |
1334 | return r; | |
1335 | if (r == 0) | |
1336 | return -EBUSY; | |
1337 | } | |
1338 | ||
d5099efc | 1339 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
fd38203a LP |
1340 | if (r < 0) |
1341 | return r; | |
1342 | ||
4a0b58c4 | 1343 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) |
fd38203a LP |
1344 | return -EBUSY; |
1345 | ||
a71fe8b8 | 1346 | s = source_new(e, !ret, SOURCE_CHILD); |
fd38203a LP |
1347 | if (!s) |
1348 | return -ENOMEM; | |
1349 | ||
f8f3f926 | 1350 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1351 | s->child.pid = pid; |
1352 | s->child.options = options; | |
1353 | s->child.callback = callback; | |
1354 | s->userdata = userdata; | |
baf76283 | 1355 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1356 | |
f8f3f926 LP |
1357 | /* We always take a pidfd here if we can, even if we wait for anything else than WEXITED, so that we |
1358 | * pin the PID, and make regular waitid() handling race-free. */ | |
1359 | ||
1360 | if (shall_use_pidfd()) { | |
1361 | s->child.pidfd = pidfd_open(s->child.pid, 0); | |
1362 | if (s->child.pidfd < 0) { | |
1363 | /* Propagate errors unless the syscall is not supported or blocked */ | |
1364 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
1365 | return -errno; | |
1366 | } else | |
1367 | s->child.pidfd_owned = true; /* If we allocate the pidfd we own it by default */ | |
1368 | } else | |
1369 | s->child.pidfd = -1; | |
1370 | ||
4a0b58c4 | 1371 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); |
ec766a51 | 1372 | if (r < 0) |
fd38203a | 1373 | return r; |
fd38203a | 1374 | |
f8f3f926 LP |
1375 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1376 | /* We have a pidfd and we only want to watch for exit */ | |
f8f3f926 | 1377 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1378 | if (r < 0) |
f8f3f926 | 1379 | return r; |
ac9f2640 | 1380 | |
f8f3f926 LP |
1381 | } else { |
1382 | /* We have no pidfd or we shall wait for some other event than WEXITED */ | |
f8f3f926 | 1383 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1384 | if (r < 0) |
f8f3f926 | 1385 | return r; |
f8f3f926 LP |
1386 | |
1387 | e->need_process_child = true; | |
1388 | } | |
c2ba3ad6 | 1389 | |
ac9f2640 ZJS |
1390 | e->n_enabled_child_sources++; |
1391 | ||
a71fe8b8 LP |
1392 | if (ret) |
1393 | *ret = s; | |
ec766a51 | 1394 | TAKE_PTR(s); |
f8f3f926 LP |
1395 | return 0; |
1396 | } | |
1397 | ||
1398 | _public_ int sd_event_add_child_pidfd( | |
1399 | sd_event *e, | |
1400 | sd_event_source **ret, | |
1401 | int pidfd, | |
1402 | int options, | |
1403 | sd_event_child_handler_t callback, | |
1404 | void *userdata) { | |
1405 | ||
1406 | ||
1407 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
1408 | pid_t pid; | |
1409 | int r; | |
1410 | ||
1411 | assert_return(e, -EINVAL); | |
1412 | assert_return(e = event_resolve(e), -ENOPKG); | |
1413 | assert_return(pidfd >= 0, -EBADF); | |
1414 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1415 | assert_return(options != 0, -EINVAL); | |
f8f3f926 LP |
1416 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1417 | assert_return(!event_pid_changed(e), -ECHILD); | |
1418 | ||
b9350e70 LP |
1419 | if (!callback) |
1420 | callback = child_exit_callback; | |
1421 | ||
ee880b37 LP |
1422 | if (e->n_enabled_child_sources == 0) { |
1423 | r = signal_is_blocked(SIGCHLD); | |
1424 | if (r < 0) | |
1425 | return r; | |
1426 | if (r == 0) | |
1427 | return -EBUSY; | |
1428 | } | |
1429 | ||
f8f3f926 LP |
1430 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
1431 | if (r < 0) | |
1432 | return r; | |
1433 | ||
1434 | r = pidfd_get_pid(pidfd, &pid); | |
1435 | if (r < 0) | |
1436 | return r; | |
1437 | ||
1438 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) | |
1439 | return -EBUSY; | |
1440 | ||
1441 | s = source_new(e, !ret, SOURCE_CHILD); | |
1442 | if (!s) | |
1443 | return -ENOMEM; | |
1444 | ||
1445 | s->wakeup = WAKEUP_EVENT_SOURCE; | |
1446 | s->child.pidfd = pidfd; | |
1447 | s->child.pid = pid; | |
1448 | s->child.options = options; | |
1449 | s->child.callback = callback; | |
1450 | 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) */ | |
1451 | s->userdata = userdata; | |
1452 | s->enabled = SD_EVENT_ONESHOT; | |
1453 | ||
1454 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); | |
1455 | if (r < 0) | |
1456 | return r; | |
1457 | ||
f8f3f926 LP |
1458 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1459 | /* We only want to watch for WEXITED */ | |
f8f3f926 | 1460 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1461 | if (r < 0) |
f8f3f926 | 1462 | return r; |
f8f3f926 LP |
1463 | } else { |
1464 | /* We shall wait for some other event than WEXITED */ | |
f8f3f926 | 1465 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1466 | if (r < 0) |
f8f3f926 | 1467 | return r; |
a71fe8b8 | 1468 | |
f8f3f926 LP |
1469 | e->need_process_child = true; |
1470 | } | |
1471 | ||
ac9f2640 ZJS |
1472 | e->n_enabled_child_sources++; |
1473 | ||
f8f3f926 LP |
1474 | if (ret) |
1475 | *ret = s; | |
f8f3f926 | 1476 | TAKE_PTR(s); |
fd38203a LP |
1477 | return 0; |
1478 | } | |
1479 | ||
b9350e70 LP |
1480 | static int generic_exit_callback(sd_event_source *s, void *userdata) { |
1481 | assert(s); | |
1482 | ||
1483 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1484 | } | |
1485 | ||
f7262a9f | 1486 | _public_ int sd_event_add_defer( |
305f78bf | 1487 | sd_event *e, |
151b9b96 | 1488 | sd_event_source **ret, |
718db961 | 1489 | sd_event_handler_t callback, |
151b9b96 | 1490 | void *userdata) { |
305f78bf | 1491 | |
ec766a51 | 1492 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1493 | int r; |
1494 | ||
305f78bf | 1495 | assert_return(e, -EINVAL); |
b937d761 | 1496 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1497 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1498 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1499 | |
b9350e70 LP |
1500 | if (!callback) |
1501 | callback = generic_exit_callback; | |
1502 | ||
a71fe8b8 | 1503 | s = source_new(e, !ret, SOURCE_DEFER); |
fd38203a LP |
1504 | if (!s) |
1505 | return -ENOMEM; | |
1506 | ||
1507 | s->defer.callback = callback; | |
1508 | s->userdata = userdata; | |
baf76283 | 1509 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a LP |
1510 | |
1511 | r = source_set_pending(s, true); | |
ec766a51 | 1512 | if (r < 0) |
fd38203a | 1513 | return r; |
fd38203a | 1514 | |
a71fe8b8 LP |
1515 | if (ret) |
1516 | *ret = s; | |
ec766a51 | 1517 | TAKE_PTR(s); |
a71fe8b8 | 1518 | |
fd38203a LP |
1519 | return 0; |
1520 | } | |
1521 | ||
6e9feda3 LP |
1522 | _public_ int sd_event_add_post( |
1523 | sd_event *e, | |
1524 | sd_event_source **ret, | |
1525 | sd_event_handler_t callback, | |
1526 | void *userdata) { | |
1527 | ||
ec766a51 | 1528 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6e9feda3 LP |
1529 | int r; |
1530 | ||
1531 | assert_return(e, -EINVAL); | |
b937d761 | 1532 | assert_return(e = event_resolve(e), -ENOPKG); |
6e9feda3 LP |
1533 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1534 | assert_return(!event_pid_changed(e), -ECHILD); | |
1535 | ||
b9350e70 LP |
1536 | if (!callback) |
1537 | callback = generic_exit_callback; | |
1538 | ||
a71fe8b8 | 1539 | s = source_new(e, !ret, SOURCE_POST); |
6e9feda3 LP |
1540 | if (!s) |
1541 | return -ENOMEM; | |
1542 | ||
1543 | s->post.callback = callback; | |
1544 | s->userdata = userdata; | |
1545 | s->enabled = SD_EVENT_ON; | |
1546 | ||
de7fef4b | 1547 | r = set_ensure_put(&e->post_sources, NULL, s); |
ec766a51 | 1548 | if (r < 0) |
6e9feda3 | 1549 | return r; |
de7fef4b | 1550 | assert(r > 0); |
6e9feda3 | 1551 | |
a71fe8b8 LP |
1552 | if (ret) |
1553 | *ret = s; | |
ec766a51 | 1554 | TAKE_PTR(s); |
a71fe8b8 | 1555 | |
6e9feda3 LP |
1556 | return 0; |
1557 | } | |
1558 | ||
6203e07a | 1559 | _public_ int sd_event_add_exit( |
305f78bf | 1560 | sd_event *e, |
151b9b96 | 1561 | sd_event_source **ret, |
718db961 | 1562 | sd_event_handler_t callback, |
151b9b96 | 1563 | void *userdata) { |
305f78bf | 1564 | |
ec766a51 | 1565 | _cleanup_(source_freep) sd_event_source *s = NULL; |
da7e457c LP |
1566 | int r; |
1567 | ||
1568 | assert_return(e, -EINVAL); | |
b937d761 | 1569 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
1570 | assert_return(callback, -EINVAL); |
1571 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
1572 | assert_return(!event_pid_changed(e), -ECHILD); | |
1573 | ||
c983e776 EV |
1574 | r = prioq_ensure_allocated(&e->exit, exit_prioq_compare); |
1575 | if (r < 0) | |
1576 | return r; | |
da7e457c | 1577 | |
a71fe8b8 | 1578 | s = source_new(e, !ret, SOURCE_EXIT); |
fd38203a | 1579 | if (!s) |
da7e457c | 1580 | return -ENOMEM; |
fd38203a | 1581 | |
6203e07a | 1582 | s->exit.callback = callback; |
da7e457c | 1583 | s->userdata = userdata; |
6203e07a | 1584 | s->exit.prioq_index = PRIOQ_IDX_NULL; |
baf76283 | 1585 | s->enabled = SD_EVENT_ONESHOT; |
da7e457c | 1586 | |
6203e07a | 1587 | r = prioq_put(s->event->exit, s, &s->exit.prioq_index); |
ec766a51 | 1588 | if (r < 0) |
da7e457c | 1589 | return r; |
da7e457c | 1590 | |
a71fe8b8 LP |
1591 | if (ret) |
1592 | *ret = s; | |
ec766a51 | 1593 | TAKE_PTR(s); |
a71fe8b8 | 1594 | |
da7e457c LP |
1595 | return 0; |
1596 | } | |
1597 | ||
97ef5391 LP |
1598 | static void event_free_inotify_data(sd_event *e, struct inotify_data *d) { |
1599 | assert(e); | |
1600 | ||
1601 | if (!d) | |
1602 | return; | |
1603 | ||
1604 | assert(hashmap_isempty(d->inodes)); | |
1605 | assert(hashmap_isempty(d->wd)); | |
1606 | ||
1607 | if (d->buffer_filled > 0) | |
1608 | LIST_REMOVE(buffered, e->inotify_data_buffered, d); | |
1609 | ||
1610 | hashmap_free(d->inodes); | |
1611 | hashmap_free(d->wd); | |
1612 | ||
1613 | assert_se(hashmap_remove(e->inotify_data, &d->priority) == d); | |
1614 | ||
1615 | if (d->fd >= 0) { | |
1616 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, d->fd, NULL) < 0) | |
1617 | log_debug_errno(errno, "Failed to remove inotify fd from epoll, ignoring: %m"); | |
1618 | ||
1619 | safe_close(d->fd); | |
1620 | } | |
1621 | free(d); | |
1622 | } | |
1623 | ||
1624 | static int event_make_inotify_data( | |
1625 | sd_event *e, | |
1626 | int64_t priority, | |
1627 | struct inotify_data **ret) { | |
1628 | ||
1629 | _cleanup_close_ int fd = -1; | |
1630 | struct inotify_data *d; | |
97ef5391 LP |
1631 | int r; |
1632 | ||
1633 | assert(e); | |
1634 | ||
1635 | d = hashmap_get(e->inotify_data, &priority); | |
1636 | if (d) { | |
1637 | if (ret) | |
1638 | *ret = d; | |
1639 | return 0; | |
1640 | } | |
1641 | ||
1642 | fd = inotify_init1(IN_NONBLOCK|O_CLOEXEC); | |
1643 | if (fd < 0) | |
1644 | return -errno; | |
1645 | ||
1646 | fd = fd_move_above_stdio(fd); | |
1647 | ||
1648 | r = hashmap_ensure_allocated(&e->inotify_data, &uint64_hash_ops); | |
1649 | if (r < 0) | |
1650 | return r; | |
1651 | ||
1652 | d = new(struct inotify_data, 1); | |
1653 | if (!d) | |
1654 | return -ENOMEM; | |
1655 | ||
1656 | *d = (struct inotify_data) { | |
1657 | .wakeup = WAKEUP_INOTIFY_DATA, | |
1658 | .fd = TAKE_FD(fd), | |
1659 | .priority = priority, | |
1660 | }; | |
1661 | ||
1662 | r = hashmap_put(e->inotify_data, &d->priority, d); | |
1663 | if (r < 0) { | |
1664 | d->fd = safe_close(d->fd); | |
1665 | free(d); | |
1666 | return r; | |
1667 | } | |
1668 | ||
1eac7948 | 1669 | struct epoll_event ev = { |
97ef5391 LP |
1670 | .events = EPOLLIN, |
1671 | .data.ptr = d, | |
1672 | }; | |
1673 | ||
1674 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { | |
1675 | r = -errno; | |
1676 | d->fd = safe_close(d->fd); /* let's close this ourselves, as event_free_inotify_data() would otherwise | |
1677 | * remove the fd from the epoll first, which we don't want as we couldn't | |
1678 | * add it in the first place. */ | |
1679 | event_free_inotify_data(e, d); | |
1680 | return r; | |
1681 | } | |
1682 | ||
1683 | if (ret) | |
1684 | *ret = d; | |
1685 | ||
1686 | return 1; | |
1687 | } | |
1688 | ||
7a08d314 | 1689 | static int inode_data_compare(const struct inode_data *x, const struct inode_data *y) { |
90c88092 | 1690 | int r; |
97ef5391 LP |
1691 | |
1692 | assert(x); | |
1693 | assert(y); | |
1694 | ||
90c88092 YW |
1695 | r = CMP(x->dev, y->dev); |
1696 | if (r != 0) | |
1697 | return r; | |
97ef5391 | 1698 | |
6dd91b36 | 1699 | return CMP(x->ino, y->ino); |
97ef5391 LP |
1700 | } |
1701 | ||
7a08d314 YW |
1702 | static void inode_data_hash_func(const struct inode_data *d, struct siphash *state) { |
1703 | assert(d); | |
97ef5391 LP |
1704 | |
1705 | siphash24_compress(&d->dev, sizeof(d->dev), state); | |
1706 | siphash24_compress(&d->ino, sizeof(d->ino), state); | |
1707 | } | |
1708 | ||
7a08d314 | 1709 | DEFINE_PRIVATE_HASH_OPS(inode_data_hash_ops, struct inode_data, inode_data_hash_func, inode_data_compare); |
97ef5391 LP |
1710 | |
1711 | static void event_free_inode_data( | |
1712 | sd_event *e, | |
1713 | struct inode_data *d) { | |
1714 | ||
1715 | assert(e); | |
1716 | ||
1717 | if (!d) | |
1718 | return; | |
1719 | ||
1720 | assert(!d->event_sources); | |
1721 | ||
1722 | if (d->fd >= 0) { | |
1723 | LIST_REMOVE(to_close, e->inode_data_to_close, d); | |
1724 | safe_close(d->fd); | |
1725 | } | |
1726 | ||
1727 | if (d->inotify_data) { | |
1728 | ||
1729 | if (d->wd >= 0) { | |
1730 | if (d->inotify_data->fd >= 0) { | |
1731 | /* So here's a problem. At the time this runs the watch descriptor might already be | |
1732 | * invalidated, because an IN_IGNORED event might be queued right the moment we enter | |
1733 | * the syscall. Hence, whenever we get EINVAL, ignore it entirely, since it's a very | |
1734 | * likely case to happen. */ | |
1735 | ||
1736 | if (inotify_rm_watch(d->inotify_data->fd, d->wd) < 0 && errno != EINVAL) | |
1737 | log_debug_errno(errno, "Failed to remove watch descriptor %i from inotify, ignoring: %m", d->wd); | |
1738 | } | |
1739 | ||
1740 | assert_se(hashmap_remove(d->inotify_data->wd, INT_TO_PTR(d->wd)) == d); | |
1741 | } | |
1742 | ||
1743 | assert_se(hashmap_remove(d->inotify_data->inodes, d) == d); | |
1744 | } | |
1745 | ||
1746 | free(d); | |
1747 | } | |
1748 | ||
1749 | static void event_gc_inode_data( | |
1750 | sd_event *e, | |
1751 | struct inode_data *d) { | |
1752 | ||
1753 | struct inotify_data *inotify_data; | |
1754 | ||
1755 | assert(e); | |
1756 | ||
1757 | if (!d) | |
1758 | return; | |
1759 | ||
1760 | if (d->event_sources) | |
1761 | return; | |
1762 | ||
1763 | inotify_data = d->inotify_data; | |
1764 | event_free_inode_data(e, d); | |
1765 | ||
1766 | if (inotify_data && hashmap_isempty(inotify_data->inodes)) | |
1767 | event_free_inotify_data(e, inotify_data); | |
1768 | } | |
1769 | ||
1770 | static int event_make_inode_data( | |
1771 | sd_event *e, | |
1772 | struct inotify_data *inotify_data, | |
1773 | dev_t dev, | |
1774 | ino_t ino, | |
1775 | struct inode_data **ret) { | |
1776 | ||
1777 | struct inode_data *d, key; | |
1778 | int r; | |
1779 | ||
1780 | assert(e); | |
1781 | assert(inotify_data); | |
1782 | ||
1783 | key = (struct inode_data) { | |
1784 | .ino = ino, | |
1785 | .dev = dev, | |
1786 | }; | |
1787 | ||
1788 | d = hashmap_get(inotify_data->inodes, &key); | |
1789 | if (d) { | |
1790 | if (ret) | |
1791 | *ret = d; | |
1792 | ||
1793 | return 0; | |
1794 | } | |
1795 | ||
1796 | r = hashmap_ensure_allocated(&inotify_data->inodes, &inode_data_hash_ops); | |
1797 | if (r < 0) | |
1798 | return r; | |
1799 | ||
1800 | d = new(struct inode_data, 1); | |
1801 | if (!d) | |
1802 | return -ENOMEM; | |
1803 | ||
1804 | *d = (struct inode_data) { | |
1805 | .dev = dev, | |
1806 | .ino = ino, | |
1807 | .wd = -1, | |
1808 | .fd = -1, | |
1809 | .inotify_data = inotify_data, | |
1810 | }; | |
1811 | ||
1812 | r = hashmap_put(inotify_data->inodes, d, d); | |
1813 | if (r < 0) { | |
1814 | free(d); | |
1815 | return r; | |
1816 | } | |
1817 | ||
1818 | if (ret) | |
1819 | *ret = d; | |
1820 | ||
1821 | return 1; | |
1822 | } | |
1823 | ||
1824 | static uint32_t inode_data_determine_mask(struct inode_data *d) { | |
1825 | bool excl_unlink = true; | |
1826 | uint32_t combined = 0; | |
1827 | sd_event_source *s; | |
1828 | ||
1829 | assert(d); | |
1830 | ||
1831 | /* Combines the watch masks of all event sources watching this inode. We generally just OR them together, but | |
1832 | * the IN_EXCL_UNLINK flag is ANDed instead. | |
1833 | * | |
1834 | * Note that we add all sources to the mask here, regardless whether enabled, disabled or oneshot. That's | |
1835 | * because we cannot change the mask anymore after the event source was created once, since the kernel has no | |
f21f31b2 | 1836 | * API for that. Hence we need to subscribe to the maximum mask we ever might be interested in, and suppress |
97ef5391 LP |
1837 | * events we don't care for client-side. */ |
1838 | ||
1839 | LIST_FOREACH(inotify.by_inode_data, s, d->event_sources) { | |
1840 | ||
1841 | if ((s->inotify.mask & IN_EXCL_UNLINK) == 0) | |
1842 | excl_unlink = false; | |
1843 | ||
1844 | combined |= s->inotify.mask; | |
1845 | } | |
1846 | ||
1847 | return (combined & ~(IN_ONESHOT|IN_DONT_FOLLOW|IN_ONLYDIR|IN_EXCL_UNLINK)) | (excl_unlink ? IN_EXCL_UNLINK : 0); | |
1848 | } | |
1849 | ||
1850 | static int inode_data_realize_watch(sd_event *e, struct inode_data *d) { | |
1851 | uint32_t combined_mask; | |
1852 | int wd, r; | |
1853 | ||
1854 | assert(d); | |
1855 | assert(d->fd >= 0); | |
1856 | ||
1857 | combined_mask = inode_data_determine_mask(d); | |
1858 | ||
1859 | if (d->wd >= 0 && combined_mask == d->combined_mask) | |
1860 | return 0; | |
1861 | ||
1862 | r = hashmap_ensure_allocated(&d->inotify_data->wd, NULL); | |
1863 | if (r < 0) | |
1864 | return r; | |
1865 | ||
1866 | wd = inotify_add_watch_fd(d->inotify_data->fd, d->fd, combined_mask); | |
1867 | if (wd < 0) | |
1868 | return -errno; | |
1869 | ||
1870 | if (d->wd < 0) { | |
1871 | r = hashmap_put(d->inotify_data->wd, INT_TO_PTR(wd), d); | |
1872 | if (r < 0) { | |
1873 | (void) inotify_rm_watch(d->inotify_data->fd, wd); | |
1874 | return r; | |
1875 | } | |
1876 | ||
1877 | d->wd = wd; | |
1878 | ||
1879 | } else if (d->wd != wd) { | |
1880 | ||
1881 | log_debug("Weird, the watch descriptor we already knew for this inode changed?"); | |
1882 | (void) inotify_rm_watch(d->fd, wd); | |
1883 | return -EINVAL; | |
1884 | } | |
1885 | ||
1886 | d->combined_mask = combined_mask; | |
1887 | return 1; | |
1888 | } | |
1889 | ||
b9350e70 LP |
1890 | static int inotify_exit_callback(sd_event_source *s, const struct inotify_event *event, void *userdata) { |
1891 | assert(s); | |
1892 | ||
1893 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1894 | } | |
1895 | ||
97ef5391 LP |
1896 | _public_ int sd_event_add_inotify( |
1897 | sd_event *e, | |
1898 | sd_event_source **ret, | |
1899 | const char *path, | |
1900 | uint32_t mask, | |
1901 | sd_event_inotify_handler_t callback, | |
1902 | void *userdata) { | |
1903 | ||
97ef5391 LP |
1904 | struct inotify_data *inotify_data = NULL; |
1905 | struct inode_data *inode_data = NULL; | |
1906 | _cleanup_close_ int fd = -1; | |
8c75fe17 | 1907 | _cleanup_(source_freep) sd_event_source *s = NULL; |
97ef5391 LP |
1908 | struct stat st; |
1909 | int r; | |
1910 | ||
1911 | assert_return(e, -EINVAL); | |
1912 | assert_return(e = event_resolve(e), -ENOPKG); | |
1913 | assert_return(path, -EINVAL); | |
97ef5391 LP |
1914 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1915 | assert_return(!event_pid_changed(e), -ECHILD); | |
1916 | ||
b9350e70 LP |
1917 | if (!callback) |
1918 | callback = inotify_exit_callback; | |
1919 | ||
97ef5391 LP |
1920 | /* Refuse IN_MASK_ADD since we coalesce watches on the same inode, and hence really don't want to merge |
1921 | * masks. Or in other words, this whole code exists only to manage IN_MASK_ADD type operations for you, hence | |
1922 | * the user can't use them for us. */ | |
1923 | if (mask & IN_MASK_ADD) | |
1924 | return -EINVAL; | |
1925 | ||
1926 | fd = open(path, O_PATH|O_CLOEXEC| | |
1927 | (mask & IN_ONLYDIR ? O_DIRECTORY : 0)| | |
1928 | (mask & IN_DONT_FOLLOW ? O_NOFOLLOW : 0)); | |
1929 | if (fd < 0) | |
1930 | return -errno; | |
1931 | ||
1932 | if (fstat(fd, &st) < 0) | |
1933 | return -errno; | |
1934 | ||
1935 | s = source_new(e, !ret, SOURCE_INOTIFY); | |
1936 | if (!s) | |
1937 | return -ENOMEM; | |
1938 | ||
1939 | s->enabled = mask & IN_ONESHOT ? SD_EVENT_ONESHOT : SD_EVENT_ON; | |
1940 | s->inotify.mask = mask; | |
1941 | s->inotify.callback = callback; | |
1942 | s->userdata = userdata; | |
1943 | ||
1944 | /* Allocate an inotify object for this priority, and an inode object within it */ | |
1945 | r = event_make_inotify_data(e, SD_EVENT_PRIORITY_NORMAL, &inotify_data); | |
1946 | if (r < 0) | |
8c75fe17 | 1947 | return r; |
97ef5391 LP |
1948 | |
1949 | r = event_make_inode_data(e, inotify_data, st.st_dev, st.st_ino, &inode_data); | |
8c75fe17 ZJS |
1950 | if (r < 0) { |
1951 | event_free_inotify_data(e, inotify_data); | |
1952 | return r; | |
1953 | } | |
97ef5391 LP |
1954 | |
1955 | /* Keep the O_PATH fd around until the first iteration of the loop, so that we can still change the priority of | |
1956 | * the event source, until then, for which we need the original inode. */ | |
1957 | if (inode_data->fd < 0) { | |
1958 | inode_data->fd = TAKE_FD(fd); | |
1959 | LIST_PREPEND(to_close, e->inode_data_to_close, inode_data); | |
1960 | } | |
1961 | ||
1962 | /* Link our event source to the inode data object */ | |
1963 | LIST_PREPEND(inotify.by_inode_data, inode_data->event_sources, s); | |
1964 | s->inotify.inode_data = inode_data; | |
1965 | ||
97ef5391 LP |
1966 | /* Actually realize the watch now */ |
1967 | r = inode_data_realize_watch(e, inode_data); | |
1968 | if (r < 0) | |
8c75fe17 | 1969 | return r; |
97ef5391 LP |
1970 | |
1971 | (void) sd_event_source_set_description(s, path); | |
1972 | ||
1973 | if (ret) | |
1974 | *ret = s; | |
8c75fe17 | 1975 | TAKE_PTR(s); |
97ef5391 LP |
1976 | |
1977 | return 0; | |
97ef5391 LP |
1978 | } |
1979 | ||
8301aa0b | 1980 | static sd_event_source* event_source_free(sd_event_source *s) { |
6680dd6b LP |
1981 | if (!s) |
1982 | return NULL; | |
da7e457c | 1983 | |
8301aa0b YW |
1984 | /* Here's a special hack: when we are called from a |
1985 | * dispatch handler we won't free the event source | |
1986 | * immediately, but we will detach the fd from the | |
1987 | * epoll. This way it is safe for the caller to unref | |
1988 | * the event source and immediately close the fd, but | |
1989 | * we still retain a valid event source object after | |
1990 | * the callback. */ | |
fd38203a | 1991 | |
8301aa0b YW |
1992 | if (s->dispatching) { |
1993 | if (s->type == SOURCE_IO) | |
1994 | source_io_unregister(s); | |
fd38203a | 1995 | |
8301aa0b YW |
1996 | source_disconnect(s); |
1997 | } else | |
1998 | source_free(s); | |
fd38203a LP |
1999 | |
2000 | return NULL; | |
2001 | } | |
2002 | ||
8301aa0b YW |
2003 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event_source, sd_event_source, event_source_free); |
2004 | ||
356779df | 2005 | _public_ int sd_event_source_set_description(sd_event_source *s, const char *description) { |
f7f53e9e | 2006 | assert_return(s, -EINVAL); |
f4b2933e | 2007 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 2008 | |
356779df | 2009 | return free_and_strdup(&s->description, description); |
f7f53e9e TG |
2010 | } |
2011 | ||
356779df | 2012 | _public_ int sd_event_source_get_description(sd_event_source *s, const char **description) { |
f7f53e9e | 2013 | assert_return(s, -EINVAL); |
356779df | 2014 | assert_return(description, -EINVAL); |
f4b2933e | 2015 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 2016 | |
7d92a1a4 ZJS |
2017 | if (!s->description) |
2018 | return -ENXIO; | |
2019 | ||
356779df | 2020 | *description = s->description; |
f7f53e9e TG |
2021 | return 0; |
2022 | } | |
2023 | ||
adcc4ca3 | 2024 | _public_ sd_event *sd_event_source_get_event(sd_event_source *s) { |
305f78bf | 2025 | assert_return(s, NULL); |
eaa3cbef LP |
2026 | |
2027 | return s->event; | |
2028 | } | |
2029 | ||
f7262a9f | 2030 | _public_ int sd_event_source_get_pending(sd_event_source *s) { |
305f78bf | 2031 | assert_return(s, -EINVAL); |
6203e07a | 2032 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c | 2033 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2034 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2035 | |
2036 | return s->pending; | |
2037 | } | |
2038 | ||
f7262a9f | 2039 | _public_ int sd_event_source_get_io_fd(sd_event_source *s) { |
305f78bf LP |
2040 | assert_return(s, -EINVAL); |
2041 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2042 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2043 | |
2044 | return s->io.fd; | |
2045 | } | |
2046 | ||
30caf8f3 LP |
2047 | _public_ int sd_event_source_set_io_fd(sd_event_source *s, int fd) { |
2048 | int r; | |
2049 | ||
2050 | assert_return(s, -EINVAL); | |
8ac43fee | 2051 | assert_return(fd >= 0, -EBADF); |
30caf8f3 LP |
2052 | assert_return(s->type == SOURCE_IO, -EDOM); |
2053 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2054 | ||
2055 | if (s->io.fd == fd) | |
2056 | return 0; | |
2057 | ||
2058 | if (s->enabled == SD_EVENT_OFF) { | |
2059 | s->io.fd = fd; | |
2060 | s->io.registered = false; | |
2061 | } else { | |
2062 | int saved_fd; | |
2063 | ||
2064 | saved_fd = s->io.fd; | |
2065 | assert(s->io.registered); | |
2066 | ||
2067 | s->io.fd = fd; | |
2068 | s->io.registered = false; | |
2069 | ||
2070 | r = source_io_register(s, s->enabled, s->io.events); | |
2071 | if (r < 0) { | |
2072 | s->io.fd = saved_fd; | |
2073 | s->io.registered = true; | |
2074 | return r; | |
2075 | } | |
2076 | ||
5a795bff | 2077 | (void) epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, saved_fd, NULL); |
30caf8f3 LP |
2078 | } |
2079 | ||
2080 | return 0; | |
2081 | } | |
2082 | ||
ab93297c NM |
2083 | _public_ int sd_event_source_get_io_fd_own(sd_event_source *s) { |
2084 | assert_return(s, -EINVAL); | |
2085 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2086 | ||
2087 | return s->io.owned; | |
2088 | } | |
2089 | ||
2090 | _public_ int sd_event_source_set_io_fd_own(sd_event_source *s, int own) { | |
2091 | assert_return(s, -EINVAL); | |
2092 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2093 | ||
2094 | s->io.owned = own; | |
2095 | return 0; | |
2096 | } | |
2097 | ||
f7262a9f | 2098 | _public_ int sd_event_source_get_io_events(sd_event_source *s, uint32_t* events) { |
305f78bf LP |
2099 | assert_return(s, -EINVAL); |
2100 | assert_return(events, -EINVAL); | |
2101 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2102 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2103 | |
2104 | *events = s->io.events; | |
2105 | return 0; | |
2106 | } | |
2107 | ||
f7262a9f | 2108 | _public_ int sd_event_source_set_io_events(sd_event_source *s, uint32_t events) { |
fd38203a LP |
2109 | int r; |
2110 | ||
305f78bf LP |
2111 | assert_return(s, -EINVAL); |
2112 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2a16a986 | 2113 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 2114 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2115 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2116 | |
b63c8d4f DH |
2117 | /* edge-triggered updates are never skipped, so we can reset edges */ |
2118 | if (s->io.events == events && !(events & EPOLLET)) | |
fd38203a LP |
2119 | return 0; |
2120 | ||
2a0dc6cd LP |
2121 | r = source_set_pending(s, false); |
2122 | if (r < 0) | |
2123 | return r; | |
2124 | ||
baf76283 | 2125 | if (s->enabled != SD_EVENT_OFF) { |
e4715127 | 2126 | r = source_io_register(s, s->enabled, events); |
fd38203a LP |
2127 | if (r < 0) |
2128 | return r; | |
2129 | } | |
2130 | ||
2131 | s->io.events = events; | |
2132 | ||
2133 | return 0; | |
2134 | } | |
2135 | ||
f7262a9f | 2136 | _public_ int sd_event_source_get_io_revents(sd_event_source *s, uint32_t* revents) { |
305f78bf LP |
2137 | assert_return(s, -EINVAL); |
2138 | assert_return(revents, -EINVAL); | |
2139 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2140 | assert_return(s->pending, -ENODATA); | |
2141 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2142 | |
2143 | *revents = s->io.revents; | |
2144 | return 0; | |
2145 | } | |
2146 | ||
f7262a9f | 2147 | _public_ int sd_event_source_get_signal(sd_event_source *s) { |
305f78bf LP |
2148 | assert_return(s, -EINVAL); |
2149 | assert_return(s->type == SOURCE_SIGNAL, -EDOM); | |
2150 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2151 | |
2152 | return s->signal.sig; | |
2153 | } | |
2154 | ||
31927c16 | 2155 | _public_ int sd_event_source_get_priority(sd_event_source *s, int64_t *priority) { |
305f78bf LP |
2156 | assert_return(s, -EINVAL); |
2157 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a | 2158 | |
6680b8d1 ME |
2159 | *priority = s->priority; |
2160 | return 0; | |
fd38203a LP |
2161 | } |
2162 | ||
31927c16 | 2163 | _public_ int sd_event_source_set_priority(sd_event_source *s, int64_t priority) { |
97ef5391 LP |
2164 | bool rm_inotify = false, rm_inode = false; |
2165 | struct inotify_data *new_inotify_data = NULL; | |
2166 | struct inode_data *new_inode_data = NULL; | |
9da4cb2b LP |
2167 | int r; |
2168 | ||
305f78bf | 2169 | assert_return(s, -EINVAL); |
da7e457c | 2170 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2171 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2172 | |
2173 | if (s->priority == priority) | |
2174 | return 0; | |
2175 | ||
97ef5391 LP |
2176 | if (s->type == SOURCE_INOTIFY) { |
2177 | struct inode_data *old_inode_data; | |
2178 | ||
2179 | assert(s->inotify.inode_data); | |
2180 | old_inode_data = s->inotify.inode_data; | |
2181 | ||
2182 | /* We need the original fd to change the priority. If we don't have it we can't change the priority, | |
2183 | * anymore. Note that we close any fds when entering the next event loop iteration, i.e. for inotify | |
2184 | * events we allow priority changes only until the first following iteration. */ | |
2185 | if (old_inode_data->fd < 0) | |
2186 | return -EOPNOTSUPP; | |
2187 | ||
2188 | r = event_make_inotify_data(s->event, priority, &new_inotify_data); | |
2189 | if (r < 0) | |
2190 | return r; | |
2191 | rm_inotify = r > 0; | |
2192 | ||
2193 | r = event_make_inode_data(s->event, new_inotify_data, old_inode_data->dev, old_inode_data->ino, &new_inode_data); | |
2194 | if (r < 0) | |
2195 | goto fail; | |
2196 | rm_inode = r > 0; | |
2197 | ||
2198 | if (new_inode_data->fd < 0) { | |
2199 | /* Duplicate the fd for the new inode object if we don't have any yet */ | |
2200 | new_inode_data->fd = fcntl(old_inode_data->fd, F_DUPFD_CLOEXEC, 3); | |
2201 | if (new_inode_data->fd < 0) { | |
2202 | r = -errno; | |
2203 | goto fail; | |
2204 | } | |
2205 | ||
2206 | LIST_PREPEND(to_close, s->event->inode_data_to_close, new_inode_data); | |
2207 | } | |
2208 | ||
2209 | /* Move the event source to the new inode data structure */ | |
2210 | LIST_REMOVE(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2211 | LIST_PREPEND(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2212 | s->inotify.inode_data = new_inode_data; | |
2213 | ||
2214 | /* Now create the new watch */ | |
2215 | r = inode_data_realize_watch(s->event, new_inode_data); | |
2216 | if (r < 0) { | |
2217 | /* Move it back */ | |
2218 | LIST_REMOVE(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2219 | LIST_PREPEND(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2220 | s->inotify.inode_data = old_inode_data; | |
2221 | goto fail; | |
2222 | } | |
2223 | ||
2224 | s->priority = priority; | |
2225 | ||
2226 | event_gc_inode_data(s->event, old_inode_data); | |
2227 | ||
2228 | } else if (s->type == SOURCE_SIGNAL && s->enabled != SD_EVENT_OFF) { | |
9da4cb2b LP |
2229 | struct signal_data *old, *d; |
2230 | ||
2231 | /* Move us from the signalfd belonging to the old | |
2232 | * priority to the signalfd of the new priority */ | |
2233 | ||
2234 | assert_se(old = hashmap_get(s->event->signal_data, &s->priority)); | |
2235 | ||
2236 | s->priority = priority; | |
2237 | ||
2238 | r = event_make_signal_data(s->event, s->signal.sig, &d); | |
2239 | if (r < 0) { | |
2240 | s->priority = old->priority; | |
2241 | return r; | |
2242 | } | |
2243 | ||
2244 | event_unmask_signal_data(s->event, old, s->signal.sig); | |
2245 | } else | |
2246 | s->priority = priority; | |
fd38203a | 2247 | |
e1951c16 | 2248 | event_source_pp_prioq_reshuffle(s); |
fd38203a | 2249 | |
6203e07a LP |
2250 | if (s->type == SOURCE_EXIT) |
2251 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
305f78bf | 2252 | |
fd38203a | 2253 | return 0; |
97ef5391 LP |
2254 | |
2255 | fail: | |
2256 | if (rm_inode) | |
2257 | event_free_inode_data(s->event, new_inode_data); | |
2258 | ||
2259 | if (rm_inotify) | |
2260 | event_free_inotify_data(s->event, new_inotify_data); | |
2261 | ||
2262 | return r; | |
fd38203a LP |
2263 | } |
2264 | ||
cad143a8 | 2265 | _public_ int sd_event_source_get_enabled(sd_event_source *s, int *ret) { |
305f78bf | 2266 | assert_return(s, -EINVAL); |
305f78bf | 2267 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2268 | |
cad143a8 LP |
2269 | if (ret) |
2270 | *ret = s->enabled; | |
2271 | ||
08c1eb0e | 2272 | return s->enabled != SD_EVENT_OFF; |
fd38203a LP |
2273 | } |
2274 | ||
ddfde737 | 2275 | static int event_source_disable(sd_event_source *s) { |
fd38203a LP |
2276 | int r; |
2277 | ||
ddfde737 LP |
2278 | assert(s); |
2279 | assert(s->enabled != SD_EVENT_OFF); | |
fd38203a | 2280 | |
ddfde737 LP |
2281 | /* Unset the pending flag when this event source is disabled */ |
2282 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
2283 | r = source_set_pending(s, false); | |
2284 | if (r < 0) | |
2285 | return r; | |
2286 | } | |
cc567911 | 2287 | |
ddfde737 | 2288 | s->enabled = SD_EVENT_OFF; |
fd38203a | 2289 | |
ddfde737 | 2290 | switch (s->type) { |
fd38203a | 2291 | |
ddfde737 LP |
2292 | case SOURCE_IO: |
2293 | source_io_unregister(s); | |
2294 | break; | |
ac989a78 | 2295 | |
ddfde737 LP |
2296 | case SOURCE_TIME_REALTIME: |
2297 | case SOURCE_TIME_BOOTTIME: | |
2298 | case SOURCE_TIME_MONOTONIC: | |
2299 | case SOURCE_TIME_REALTIME_ALARM: | |
2300 | case SOURCE_TIME_BOOTTIME_ALARM: | |
2301 | event_source_time_prioq_reshuffle(s); | |
2302 | break; | |
fd38203a | 2303 | |
ddfde737 LP |
2304 | case SOURCE_SIGNAL: |
2305 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); | |
2306 | break; | |
fd38203a | 2307 | |
ddfde737 LP |
2308 | case SOURCE_CHILD: |
2309 | assert(s->event->n_enabled_child_sources > 0); | |
2310 | s->event->n_enabled_child_sources--; | |
fd38203a | 2311 | |
ddfde737 LP |
2312 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
2313 | source_child_pidfd_unregister(s); | |
2314 | else | |
2315 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
2316 | break; | |
4807d2d0 | 2317 | |
ddfde737 LP |
2318 | case SOURCE_EXIT: |
2319 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
2320 | break; | |
fd38203a | 2321 | |
ddfde737 LP |
2322 | case SOURCE_DEFER: |
2323 | case SOURCE_POST: | |
2324 | case SOURCE_INOTIFY: | |
2325 | break; | |
fd38203a | 2326 | |
ddfde737 LP |
2327 | default: |
2328 | assert_not_reached("Wut? I shouldn't exist."); | |
2329 | } | |
fd38203a | 2330 | |
ddfde737 LP |
2331 | return 0; |
2332 | } | |
f8f3f926 | 2333 | |
d2eafe61 | 2334 | static int event_source_enable(sd_event_source *s, int enable) { |
ddfde737 | 2335 | int r; |
fd38203a | 2336 | |
ddfde737 | 2337 | assert(s); |
d2eafe61 | 2338 | assert(IN_SET(enable, SD_EVENT_ON, SD_EVENT_ONESHOT)); |
ddfde737 | 2339 | assert(s->enabled == SD_EVENT_OFF); |
305f78bf | 2340 | |
ddfde737 LP |
2341 | /* Unset the pending flag when this event source is enabled */ |
2342 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
2343 | r = source_set_pending(s, false); | |
2344 | if (r < 0) | |
2345 | return r; | |
2346 | } | |
9d3e3aa5 | 2347 | |
ddfde737 | 2348 | switch (s->type) { |
ddfde737 | 2349 | case SOURCE_IO: |
d2eafe61 ZJS |
2350 | r = source_io_register(s, enable, s->io.events); |
2351 | if (r < 0) | |
ddfde737 | 2352 | return r; |
ddfde737 | 2353 | break; |
fd38203a | 2354 | |
ddfde737 LP |
2355 | case SOURCE_SIGNAL: |
2356 | r = event_make_signal_data(s->event, s->signal.sig, NULL); | |
2357 | if (r < 0) { | |
ddfde737 LP |
2358 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
2359 | return r; | |
2360 | } | |
fd38203a | 2361 | |
ddfde737 | 2362 | break; |
fd38203a | 2363 | |
ddfde737 | 2364 | case SOURCE_CHILD: |
ddfde737 LP |
2365 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
2366 | /* yes, we have pidfd */ | |
9da4cb2b | 2367 | |
d2eafe61 | 2368 | r = source_child_pidfd_register(s, enable); |
ac9f2640 | 2369 | if (r < 0) |
9da4cb2b | 2370 | return r; |
ddfde737 LP |
2371 | } else { |
2372 | /* no pidfd, or something other to watch for than WEXITED */ | |
9da4cb2b | 2373 | |
ddfde737 LP |
2374 | r = event_make_signal_data(s->event, SIGCHLD, NULL); |
2375 | if (r < 0) { | |
ddfde737 LP |
2376 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); |
2377 | return r; | |
2378 | } | |
2379 | } | |
fd38203a | 2380 | |
ac9f2640 ZJS |
2381 | s->event->n_enabled_child_sources++; |
2382 | ||
ddfde737 | 2383 | break; |
4807d2d0 | 2384 | |
d2eafe61 ZJS |
2385 | case SOURCE_TIME_REALTIME: |
2386 | case SOURCE_TIME_BOOTTIME: | |
2387 | case SOURCE_TIME_MONOTONIC: | |
2388 | case SOURCE_TIME_REALTIME_ALARM: | |
2389 | case SOURCE_TIME_BOOTTIME_ALARM: | |
ddfde737 | 2390 | case SOURCE_EXIT: |
ddfde737 LP |
2391 | case SOURCE_DEFER: |
2392 | case SOURCE_POST: | |
2393 | case SOURCE_INOTIFY: | |
2394 | break; | |
9da4cb2b | 2395 | |
ddfde737 LP |
2396 | default: |
2397 | assert_not_reached("Wut? I shouldn't exist."); | |
2398 | } | |
f8f3f926 | 2399 | |
d2eafe61 ZJS |
2400 | s->enabled = enable; |
2401 | ||
2402 | /* Non-failing operations below */ | |
2403 | switch (s->type) { | |
2404 | case SOURCE_TIME_REALTIME: | |
2405 | case SOURCE_TIME_BOOTTIME: | |
2406 | case SOURCE_TIME_MONOTONIC: | |
2407 | case SOURCE_TIME_REALTIME_ALARM: | |
2408 | case SOURCE_TIME_BOOTTIME_ALARM: | |
2409 | event_source_time_prioq_reshuffle(s); | |
2410 | break; | |
2411 | ||
2412 | case SOURCE_EXIT: | |
2413 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
2414 | break; | |
2415 | ||
2416 | default: | |
2417 | break; | |
2418 | } | |
2419 | ||
ddfde737 LP |
2420 | return 0; |
2421 | } | |
2422 | ||
2423 | _public_ int sd_event_source_set_enabled(sd_event_source *s, int m) { | |
2424 | int r; | |
9da4cb2b | 2425 | |
ddfde737 LP |
2426 | assert_return(s, -EINVAL); |
2427 | assert_return(IN_SET(m, SD_EVENT_OFF, SD_EVENT_ON, SD_EVENT_ONESHOT), -EINVAL); | |
2428 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a | 2429 | |
ddfde737 LP |
2430 | /* If we are dead anyway, we are fine with turning off sources, but everything else needs to fail. */ |
2431 | if (s->event->state == SD_EVENT_FINISHED) | |
2432 | return m == SD_EVENT_OFF ? 0 : -ESTALE; | |
305f78bf | 2433 | |
ddfde737 LP |
2434 | if (s->enabled == m) /* No change? */ |
2435 | return 0; | |
9d3e3aa5 | 2436 | |
ddfde737 LP |
2437 | if (m == SD_EVENT_OFF) |
2438 | r = event_source_disable(s); | |
2439 | else { | |
2440 | if (s->enabled != SD_EVENT_OFF) { | |
2441 | /* Switching from "on" to "oneshot" or back? If that's the case, we can take a shortcut, the | |
2442 | * event source is already enabled after all. */ | |
2443 | s->enabled = m; | |
2444 | return 0; | |
fd38203a | 2445 | } |
ddfde737 LP |
2446 | |
2447 | r = event_source_enable(s, m); | |
fd38203a | 2448 | } |
ddfde737 LP |
2449 | if (r < 0) |
2450 | return r; | |
fd38203a | 2451 | |
e1951c16 | 2452 | event_source_pp_prioq_reshuffle(s); |
fd38203a LP |
2453 | return 0; |
2454 | } | |
2455 | ||
f7262a9f | 2456 | _public_ int sd_event_source_get_time(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
2457 | assert_return(s, -EINVAL); |
2458 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 2459 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf | 2460 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2461 | |
2462 | *usec = s->time.next; | |
2463 | return 0; | |
2464 | } | |
2465 | ||
f7262a9f | 2466 | _public_ int sd_event_source_set_time(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 2467 | int r; |
6a0f1f6d | 2468 | |
305f78bf | 2469 | assert_return(s, -EINVAL); |
6a0f1f6d | 2470 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 2471 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2472 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2473 | |
2a0dc6cd LP |
2474 | r = source_set_pending(s, false); |
2475 | if (r < 0) | |
2476 | return r; | |
2576a19e | 2477 | |
2a0dc6cd | 2478 | s->time.next = usec; |
fd38203a | 2479 | |
e1951c16 | 2480 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
2481 | return 0; |
2482 | } | |
2483 | ||
d6a83dc4 LP |
2484 | _public_ int sd_event_source_set_time_relative(sd_event_source *s, uint64_t usec) { |
2485 | usec_t t; | |
2486 | int r; | |
2487 | ||
2488 | assert_return(s, -EINVAL); | |
2489 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2490 | ||
2491 | r = sd_event_now(s->event, event_source_type_to_clock(s->type), &t); | |
2492 | if (r < 0) | |
2493 | return r; | |
2494 | ||
2495 | if (usec >= USEC_INFINITY - t) | |
2496 | return -EOVERFLOW; | |
2497 | ||
2498 | return sd_event_source_set_time(s, t + usec); | |
2499 | } | |
2500 | ||
f7262a9f | 2501 | _public_ int sd_event_source_get_time_accuracy(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
2502 | assert_return(s, -EINVAL); |
2503 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 2504 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf LP |
2505 | assert_return(!event_pid_changed(s->event), -ECHILD); |
2506 | ||
2507 | *usec = s->time.accuracy; | |
2508 | return 0; | |
2509 | } | |
2510 | ||
f7262a9f | 2511 | _public_ int sd_event_source_set_time_accuracy(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 2512 | int r; |
6a0f1f6d | 2513 | |
305f78bf LP |
2514 | assert_return(s, -EINVAL); |
2515 | assert_return(usec != (uint64_t) -1, -EINVAL); | |
6a0f1f6d | 2516 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 2517 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2518 | assert_return(!event_pid_changed(s->event), -ECHILD); |
eaa3cbef | 2519 | |
2a0dc6cd LP |
2520 | r = source_set_pending(s, false); |
2521 | if (r < 0) | |
2522 | return r; | |
2523 | ||
eaa3cbef LP |
2524 | if (usec == 0) |
2525 | usec = DEFAULT_ACCURACY_USEC; | |
2526 | ||
eaa3cbef LP |
2527 | s->time.accuracy = usec; |
2528 | ||
e1951c16 | 2529 | event_source_time_prioq_reshuffle(s); |
6a0f1f6d LP |
2530 | return 0; |
2531 | } | |
2532 | ||
2533 | _public_ int sd_event_source_get_time_clock(sd_event_source *s, clockid_t *clock) { | |
2534 | assert_return(s, -EINVAL); | |
2535 | assert_return(clock, -EINVAL); | |
2536 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2537 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
eaa3cbef | 2538 | |
6a0f1f6d | 2539 | *clock = event_source_type_to_clock(s->type); |
eaa3cbef LP |
2540 | return 0; |
2541 | } | |
2542 | ||
f7262a9f | 2543 | _public_ int sd_event_source_get_child_pid(sd_event_source *s, pid_t *pid) { |
4bee8012 LP |
2544 | assert_return(s, -EINVAL); |
2545 | assert_return(pid, -EINVAL); | |
2546 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2547 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2548 | ||
2549 | *pid = s->child.pid; | |
2550 | return 0; | |
2551 | } | |
2552 | ||
f8f3f926 LP |
2553 | _public_ int sd_event_source_get_child_pidfd(sd_event_source *s) { |
2554 | assert_return(s, -EINVAL); | |
2555 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2556 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2557 | ||
2558 | if (s->child.pidfd < 0) | |
2559 | return -EOPNOTSUPP; | |
2560 | ||
2561 | return s->child.pidfd; | |
2562 | } | |
2563 | ||
2564 | _public_ int sd_event_source_send_child_signal(sd_event_source *s, int sig, const siginfo_t *si, unsigned flags) { | |
2565 | assert_return(s, -EINVAL); | |
2566 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2567 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2568 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
2569 | ||
2570 | /* If we already have seen indication the process exited refuse sending a signal early. This way we | |
2571 | * can be sure we don't accidentally kill the wrong process on PID reuse when pidfds are not | |
2572 | * available. */ | |
2573 | if (s->child.exited) | |
2574 | return -ESRCH; | |
2575 | ||
2576 | if (s->child.pidfd >= 0) { | |
2577 | siginfo_t copy; | |
2578 | ||
2579 | /* pidfd_send_signal() changes the siginfo_t argument. This is weird, let's hence copy the | |
2580 | * structure here */ | |
2581 | if (si) | |
2582 | copy = *si; | |
2583 | ||
2584 | if (pidfd_send_signal(s->child.pidfd, sig, si ? © : NULL, 0) < 0) { | |
2585 | /* Let's propagate the error only if the system call is not implemented or prohibited */ | |
2586 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
2587 | return -errno; | |
2588 | } else | |
2589 | return 0; | |
2590 | } | |
2591 | ||
2592 | /* Flags are only supported for pidfd_send_signal(), not for rt_sigqueueinfo(), hence let's refuse | |
2593 | * this here. */ | |
2594 | if (flags != 0) | |
2595 | return -EOPNOTSUPP; | |
2596 | ||
2597 | if (si) { | |
2598 | /* We use rt_sigqueueinfo() only if siginfo_t is specified. */ | |
2599 | siginfo_t copy = *si; | |
2600 | ||
2601 | if (rt_sigqueueinfo(s->child.pid, sig, ©) < 0) | |
2602 | return -errno; | |
2603 | } else if (kill(s->child.pid, sig) < 0) | |
2604 | return -errno; | |
2605 | ||
2606 | return 0; | |
2607 | } | |
2608 | ||
2609 | _public_ int sd_event_source_get_child_pidfd_own(sd_event_source *s) { | |
2610 | assert_return(s, -EINVAL); | |
2611 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2612 | ||
2613 | if (s->child.pidfd < 0) | |
2614 | return -EOPNOTSUPP; | |
2615 | ||
2616 | return s->child.pidfd_owned; | |
2617 | } | |
2618 | ||
2619 | _public_ int sd_event_source_set_child_pidfd_own(sd_event_source *s, int own) { | |
2620 | assert_return(s, -EINVAL); | |
2621 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2622 | ||
2623 | if (s->child.pidfd < 0) | |
2624 | return -EOPNOTSUPP; | |
2625 | ||
2626 | s->child.pidfd_owned = own; | |
2627 | return 0; | |
2628 | } | |
2629 | ||
2630 | _public_ int sd_event_source_get_child_process_own(sd_event_source *s) { | |
2631 | assert_return(s, -EINVAL); | |
2632 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2633 | ||
2634 | return s->child.process_owned; | |
2635 | } | |
2636 | ||
2637 | _public_ int sd_event_source_set_child_process_own(sd_event_source *s, int own) { | |
2638 | assert_return(s, -EINVAL); | |
2639 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2640 | ||
2641 | s->child.process_owned = own; | |
2642 | return 0; | |
2643 | } | |
2644 | ||
97ef5391 LP |
2645 | _public_ int sd_event_source_get_inotify_mask(sd_event_source *s, uint32_t *mask) { |
2646 | assert_return(s, -EINVAL); | |
2647 | assert_return(mask, -EINVAL); | |
2648 | assert_return(s->type == SOURCE_INOTIFY, -EDOM); | |
2649 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2650 | ||
2651 | *mask = s->inotify.mask; | |
2652 | return 0; | |
2653 | } | |
2654 | ||
718db961 | 2655 | _public_ int sd_event_source_set_prepare(sd_event_source *s, sd_event_handler_t callback) { |
fd38203a LP |
2656 | int r; |
2657 | ||
da7e457c | 2658 | assert_return(s, -EINVAL); |
6203e07a | 2659 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c LP |
2660 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2661 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2662 | |
2663 | if (s->prepare == callback) | |
2664 | return 0; | |
2665 | ||
2666 | if (callback && s->prepare) { | |
2667 | s->prepare = callback; | |
2668 | return 0; | |
2669 | } | |
2670 | ||
2671 | r = prioq_ensure_allocated(&s->event->prepare, prepare_prioq_compare); | |
2672 | if (r < 0) | |
2673 | return r; | |
2674 | ||
2675 | s->prepare = callback; | |
2676 | ||
2677 | if (callback) { | |
2678 | r = prioq_put(s->event->prepare, s, &s->prepare_index); | |
2679 | if (r < 0) | |
2680 | return r; | |
2681 | } else | |
2682 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
2683 | ||
2684 | return 0; | |
2685 | } | |
2686 | ||
f7262a9f | 2687 | _public_ void* sd_event_source_get_userdata(sd_event_source *s) { |
da7e457c | 2688 | assert_return(s, NULL); |
fd38203a LP |
2689 | |
2690 | return s->userdata; | |
2691 | } | |
2692 | ||
8f726607 LP |
2693 | _public_ void *sd_event_source_set_userdata(sd_event_source *s, void *userdata) { |
2694 | void *ret; | |
2695 | ||
2696 | assert_return(s, NULL); | |
2697 | ||
2698 | ret = s->userdata; | |
2699 | s->userdata = userdata; | |
2700 | ||
2701 | return ret; | |
2702 | } | |
2703 | ||
c2ba3ad6 LP |
2704 | static usec_t sleep_between(sd_event *e, usec_t a, usec_t b) { |
2705 | usec_t c; | |
2706 | assert(e); | |
2707 | assert(a <= b); | |
2708 | ||
2709 | if (a <= 0) | |
2710 | return 0; | |
393003e1 LP |
2711 | if (a >= USEC_INFINITY) |
2712 | return USEC_INFINITY; | |
c2ba3ad6 LP |
2713 | |
2714 | if (b <= a + 1) | |
2715 | return a; | |
2716 | ||
52444dc4 LP |
2717 | initialize_perturb(e); |
2718 | ||
c2ba3ad6 LP |
2719 | /* |
2720 | Find a good time to wake up again between times a and b. We | |
2721 | have two goals here: | |
2722 | ||
2723 | a) We want to wake up as seldom as possible, hence prefer | |
2724 | later times over earlier times. | |
2725 | ||
2726 | b) But if we have to wake up, then let's make sure to | |
2727 | dispatch as much as possible on the entire system. | |
2728 | ||
2729 | We implement this by waking up everywhere at the same time | |
850516e0 | 2730 | within any given minute if we can, synchronised via the |
c2ba3ad6 | 2731 | perturbation value determined from the boot ID. If we can't, |
ba276c81 LP |
2732 | then we try to find the same spot in every 10s, then 1s and |
2733 | then 250ms step. Otherwise, we pick the last possible time | |
2734 | to wake up. | |
c2ba3ad6 LP |
2735 | */ |
2736 | ||
850516e0 LP |
2737 | c = (b / USEC_PER_MINUTE) * USEC_PER_MINUTE + e->perturb; |
2738 | if (c >= b) { | |
2739 | if (_unlikely_(c < USEC_PER_MINUTE)) | |
2740 | return b; | |
2741 | ||
2742 | c -= USEC_PER_MINUTE; | |
2743 | } | |
2744 | ||
ba276c81 LP |
2745 | if (c >= a) |
2746 | return c; | |
2747 | ||
2748 | c = (b / (USEC_PER_SEC*10)) * (USEC_PER_SEC*10) + (e->perturb % (USEC_PER_SEC*10)); | |
2749 | if (c >= b) { | |
2750 | if (_unlikely_(c < USEC_PER_SEC*10)) | |
2751 | return b; | |
2752 | ||
2753 | c -= USEC_PER_SEC*10; | |
2754 | } | |
2755 | ||
850516e0 LP |
2756 | if (c >= a) |
2757 | return c; | |
2758 | ||
2759 | c = (b / USEC_PER_SEC) * USEC_PER_SEC + (e->perturb % USEC_PER_SEC); | |
c2ba3ad6 LP |
2760 | if (c >= b) { |
2761 | if (_unlikely_(c < USEC_PER_SEC)) | |
2762 | return b; | |
2763 | ||
2764 | c -= USEC_PER_SEC; | |
2765 | } | |
2766 | ||
2767 | if (c >= a) | |
2768 | return c; | |
2769 | ||
2770 | c = (b / (USEC_PER_MSEC*250)) * (USEC_PER_MSEC*250) + (e->perturb % (USEC_PER_MSEC*250)); | |
2771 | if (c >= b) { | |
2772 | if (_unlikely_(c < USEC_PER_MSEC*250)) | |
2773 | return b; | |
2774 | ||
2775 | c -= USEC_PER_MSEC*250; | |
2776 | } | |
2777 | ||
2778 | if (c >= a) | |
2779 | return c; | |
2780 | ||
2781 | return b; | |
2782 | } | |
2783 | ||
fd38203a LP |
2784 | static int event_arm_timer( |
2785 | sd_event *e, | |
6a0f1f6d | 2786 | struct clock_data *d) { |
fd38203a LP |
2787 | |
2788 | struct itimerspec its = {}; | |
c2ba3ad6 LP |
2789 | sd_event_source *a, *b; |
2790 | usec_t t; | |
fd38203a | 2791 | |
cde93897 | 2792 | assert(e); |
6a0f1f6d | 2793 | assert(d); |
fd38203a | 2794 | |
d06441da | 2795 | if (!d->needs_rearm) |
212bbb17 TG |
2796 | return 0; |
2797 | else | |
2798 | d->needs_rearm = false; | |
2799 | ||
6a0f1f6d | 2800 | a = prioq_peek(d->earliest); |
393003e1 | 2801 | if (!a || a->enabled == SD_EVENT_OFF || a->time.next == USEC_INFINITY) { |
72aedc1e | 2802 | |
6a0f1f6d | 2803 | if (d->fd < 0) |
c57b5ca3 LP |
2804 | return 0; |
2805 | ||
3a43da28 | 2806 | if (d->next == USEC_INFINITY) |
72aedc1e LP |
2807 | return 0; |
2808 | ||
2809 | /* disarm */ | |
15c689d7 LP |
2810 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
2811 | return -errno; | |
72aedc1e | 2812 | |
3a43da28 | 2813 | d->next = USEC_INFINITY; |
fd38203a | 2814 | return 0; |
72aedc1e | 2815 | } |
fd38203a | 2816 | |
6a0f1f6d | 2817 | b = prioq_peek(d->latest); |
baf76283 | 2818 | assert_se(b && b->enabled != SD_EVENT_OFF); |
c2ba3ad6 | 2819 | |
1bce0ffa | 2820 | t = sleep_between(e, a->time.next, time_event_source_latest(b)); |
6a0f1f6d | 2821 | if (d->next == t) |
fd38203a LP |
2822 | return 0; |
2823 | ||
6a0f1f6d | 2824 | assert_se(d->fd >= 0); |
fd38203a | 2825 | |
c2ba3ad6 | 2826 | if (t == 0) { |
fd38203a LP |
2827 | /* We don' want to disarm here, just mean some time looooong ago. */ |
2828 | its.it_value.tv_sec = 0; | |
2829 | its.it_value.tv_nsec = 1; | |
2830 | } else | |
c2ba3ad6 | 2831 | timespec_store(&its.it_value, t); |
fd38203a | 2832 | |
15c689d7 | 2833 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 | 2834 | return -errno; |
fd38203a | 2835 | |
6a0f1f6d | 2836 | d->next = t; |
fd38203a LP |
2837 | return 0; |
2838 | } | |
2839 | ||
9a800b56 | 2840 | static int process_io(sd_event *e, sd_event_source *s, uint32_t revents) { |
fd38203a LP |
2841 | assert(e); |
2842 | assert(s); | |
2843 | assert(s->type == SOURCE_IO); | |
2844 | ||
9a800b56 LP |
2845 | /* If the event source was already pending, we just OR in the |
2846 | * new revents, otherwise we reset the value. The ORing is | |
2847 | * necessary to handle EPOLLONESHOT events properly where | |
2848 | * readability might happen independently of writability, and | |
2849 | * we need to keep track of both */ | |
2850 | ||
2851 | if (s->pending) | |
2852 | s->io.revents |= revents; | |
2853 | else | |
2854 | s->io.revents = revents; | |
fd38203a | 2855 | |
fd38203a LP |
2856 | return source_set_pending(s, true); |
2857 | } | |
2858 | ||
72aedc1e | 2859 | static int flush_timer(sd_event *e, int fd, uint32_t events, usec_t *next) { |
fd38203a LP |
2860 | uint64_t x; |
2861 | ssize_t ss; | |
2862 | ||
2863 | assert(e); | |
da7e457c | 2864 | assert(fd >= 0); |
72aedc1e | 2865 | |
305f78bf | 2866 | assert_return(events == EPOLLIN, -EIO); |
fd38203a LP |
2867 | |
2868 | ss = read(fd, &x, sizeof(x)); | |
2869 | if (ss < 0) { | |
945c2931 | 2870 | if (IN_SET(errno, EAGAIN, EINTR)) |
fd38203a LP |
2871 | return 0; |
2872 | ||
2873 | return -errno; | |
2874 | } | |
2875 | ||
8d35dae7 | 2876 | if (_unlikely_(ss != sizeof(x))) |
fd38203a LP |
2877 | return -EIO; |
2878 | ||
cde93897 | 2879 | if (next) |
3a43da28 | 2880 | *next = USEC_INFINITY; |
72aedc1e | 2881 | |
fd38203a LP |
2882 | return 0; |
2883 | } | |
2884 | ||
305f78bf LP |
2885 | static int process_timer( |
2886 | sd_event *e, | |
2887 | usec_t n, | |
6a0f1f6d | 2888 | struct clock_data *d) { |
305f78bf | 2889 | |
fd38203a LP |
2890 | sd_event_source *s; |
2891 | int r; | |
2892 | ||
2893 | assert(e); | |
6a0f1f6d | 2894 | assert(d); |
fd38203a LP |
2895 | |
2896 | for (;;) { | |
6a0f1f6d | 2897 | s = prioq_peek(d->earliest); |
fd38203a LP |
2898 | if (!s || |
2899 | s->time.next > n || | |
baf76283 | 2900 | s->enabled == SD_EVENT_OFF || |
fd38203a LP |
2901 | s->pending) |
2902 | break; | |
2903 | ||
2904 | r = source_set_pending(s, true); | |
2905 | if (r < 0) | |
2906 | return r; | |
2907 | ||
e1951c16 | 2908 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
2909 | } |
2910 | ||
2911 | return 0; | |
2912 | } | |
2913 | ||
2914 | static int process_child(sd_event *e) { | |
2915 | sd_event_source *s; | |
fd38203a LP |
2916 | int r; |
2917 | ||
2918 | assert(e); | |
2919 | ||
c2ba3ad6 LP |
2920 | e->need_process_child = false; |
2921 | ||
fd38203a LP |
2922 | /* |
2923 | So, this is ugly. We iteratively invoke waitid() with P_PID | |
2924 | + WNOHANG for each PID we wait for, instead of using | |
2925 | P_ALL. This is because we only want to get child | |
2926 | information of very specific child processes, and not all | |
2927 | of them. We might not have processed the SIGCHLD even of a | |
2928 | previous invocation and we don't want to maintain a | |
2929 | unbounded *per-child* event queue, hence we really don't | |
2930 | want anything flushed out of the kernel's queue that we | |
2931 | don't care about. Since this is O(n) this means that if you | |
2932 | have a lot of processes you probably want to handle SIGCHLD | |
2933 | yourself. | |
08cd1552 LP |
2934 | |
2935 | We do not reap the children here (by using WNOWAIT), this | |
2936 | is only done after the event source is dispatched so that | |
2937 | the callback still sees the process as a zombie. | |
fd38203a LP |
2938 | */ |
2939 | ||
90e74a66 | 2940 | HASHMAP_FOREACH(s, e->child_sources) { |
fd38203a LP |
2941 | assert(s->type == SOURCE_CHILD); |
2942 | ||
2943 | if (s->pending) | |
2944 | continue; | |
2945 | ||
baf76283 | 2946 | if (s->enabled == SD_EVENT_OFF) |
fd38203a LP |
2947 | continue; |
2948 | ||
f8f3f926 LP |
2949 | if (s->child.exited) |
2950 | continue; | |
2951 | ||
2952 | if (EVENT_SOURCE_WATCH_PIDFD(s)) /* There's a usable pidfd known for this event source? then don't waitid() for it here */ | |
2953 | continue; | |
2954 | ||
fd38203a | 2955 | zero(s->child.siginfo); |
15c689d7 LP |
2956 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, |
2957 | WNOHANG | (s->child.options & WEXITED ? WNOWAIT : 0) | s->child.options) < 0) | |
fd38203a LP |
2958 | return -errno; |
2959 | ||
2960 | if (s->child.siginfo.si_pid != 0) { | |
945c2931 | 2961 | bool zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 2962 | |
f8f3f926 LP |
2963 | if (zombie) |
2964 | s->child.exited = true; | |
2965 | ||
08cd1552 LP |
2966 | if (!zombie && (s->child.options & WEXITED)) { |
2967 | /* If the child isn't dead then let's | |
2968 | * immediately remove the state change | |
2969 | * from the queue, since there's no | |
2970 | * benefit in leaving it queued */ | |
2971 | ||
2972 | assert(s->child.options & (WSTOPPED|WCONTINUED)); | |
a5d27871 | 2973 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|(s->child.options & (WSTOPPED|WCONTINUED))); |
08cd1552 LP |
2974 | } |
2975 | ||
fd38203a LP |
2976 | r = source_set_pending(s, true); |
2977 | if (r < 0) | |
2978 | return r; | |
2979 | } | |
2980 | } | |
2981 | ||
fd38203a LP |
2982 | return 0; |
2983 | } | |
2984 | ||
f8f3f926 LP |
2985 | static int process_pidfd(sd_event *e, sd_event_source *s, uint32_t revents) { |
2986 | assert(e); | |
2987 | assert(s); | |
2988 | assert(s->type == SOURCE_CHILD); | |
2989 | ||
2990 | if (s->pending) | |
2991 | return 0; | |
2992 | ||
2993 | if (s->enabled == SD_EVENT_OFF) | |
2994 | return 0; | |
2995 | ||
2996 | if (!EVENT_SOURCE_WATCH_PIDFD(s)) | |
2997 | return 0; | |
2998 | ||
2999 | zero(s->child.siginfo); | |
3000 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG | WNOWAIT | s->child.options) < 0) | |
3001 | return -errno; | |
3002 | ||
3003 | if (s->child.siginfo.si_pid == 0) | |
3004 | return 0; | |
3005 | ||
3006 | if (IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) | |
3007 | s->child.exited = true; | |
3008 | ||
3009 | return source_set_pending(s, true); | |
3010 | } | |
3011 | ||
9da4cb2b | 3012 | static int process_signal(sd_event *e, struct signal_data *d, uint32_t events) { |
fd38203a | 3013 | bool read_one = false; |
fd38203a LP |
3014 | int r; |
3015 | ||
da7e457c | 3016 | assert(e); |
97ef5391 | 3017 | assert(d); |
305f78bf | 3018 | assert_return(events == EPOLLIN, -EIO); |
fd38203a | 3019 | |
9da4cb2b LP |
3020 | /* If there's a signal queued on this priority and SIGCHLD is |
3021 | on this priority too, then make sure to recheck the | |
3022 | children we watch. This is because we only ever dequeue | |
3023 | the first signal per priority, and if we dequeue one, and | |
3024 | SIGCHLD might be enqueued later we wouldn't know, but we | |
3025 | might have higher priority children we care about hence we | |
3026 | need to check that explicitly. */ | |
3027 | ||
3028 | if (sigismember(&d->sigset, SIGCHLD)) | |
3029 | e->need_process_child = true; | |
3030 | ||
3031 | /* If there's already an event source pending for this | |
3032 | * priority we don't read another */ | |
3033 | if (d->current) | |
3034 | return 0; | |
3035 | ||
fd38203a | 3036 | for (;;) { |
0eb2e0e3 | 3037 | struct signalfd_siginfo si; |
7057bd99 | 3038 | ssize_t n; |
92daebc0 | 3039 | sd_event_source *s = NULL; |
fd38203a | 3040 | |
9da4cb2b | 3041 | n = read(d->fd, &si, sizeof(si)); |
7057bd99 | 3042 | if (n < 0) { |
945c2931 | 3043 | if (IN_SET(errno, EAGAIN, EINTR)) |
fd38203a LP |
3044 | return read_one; |
3045 | ||
3046 | return -errno; | |
3047 | } | |
3048 | ||
7057bd99 | 3049 | if (_unlikely_(n != sizeof(si))) |
fd38203a LP |
3050 | return -EIO; |
3051 | ||
6eb7c172 | 3052 | assert(SIGNAL_VALID(si.ssi_signo)); |
7057bd99 | 3053 | |
fd38203a LP |
3054 | read_one = true; |
3055 | ||
92daebc0 LP |
3056 | if (e->signal_sources) |
3057 | s = e->signal_sources[si.ssi_signo]; | |
92daebc0 LP |
3058 | if (!s) |
3059 | continue; | |
9da4cb2b LP |
3060 | if (s->pending) |
3061 | continue; | |
fd38203a LP |
3062 | |
3063 | s->signal.siginfo = si; | |
9da4cb2b LP |
3064 | d->current = s; |
3065 | ||
fd38203a LP |
3066 | r = source_set_pending(s, true); |
3067 | if (r < 0) | |
3068 | return r; | |
9da4cb2b LP |
3069 | |
3070 | return 1; | |
fd38203a | 3071 | } |
fd38203a LP |
3072 | } |
3073 | ||
97ef5391 LP |
3074 | static int event_inotify_data_read(sd_event *e, struct inotify_data *d, uint32_t revents) { |
3075 | ssize_t n; | |
3076 | ||
3077 | assert(e); | |
3078 | assert(d); | |
3079 | ||
3080 | assert_return(revents == EPOLLIN, -EIO); | |
3081 | ||
3082 | /* If there's already an event source pending for this priority, don't read another */ | |
3083 | if (d->n_pending > 0) | |
3084 | return 0; | |
3085 | ||
3086 | /* Is the read buffer non-empty? If so, let's not read more */ | |
3087 | if (d->buffer_filled > 0) | |
3088 | return 0; | |
3089 | ||
3090 | n = read(d->fd, &d->buffer, sizeof(d->buffer)); | |
3091 | if (n < 0) { | |
3092 | if (IN_SET(errno, EAGAIN, EINTR)) | |
3093 | return 0; | |
3094 | ||
3095 | return -errno; | |
3096 | } | |
3097 | ||
3098 | assert(n > 0); | |
3099 | d->buffer_filled = (size_t) n; | |
3100 | LIST_PREPEND(buffered, e->inotify_data_buffered, d); | |
3101 | ||
3102 | return 1; | |
3103 | } | |
3104 | ||
3105 | static void event_inotify_data_drop(sd_event *e, struct inotify_data *d, size_t sz) { | |
3106 | assert(e); | |
3107 | assert(d); | |
3108 | assert(sz <= d->buffer_filled); | |
3109 | ||
3110 | if (sz == 0) | |
3111 | return; | |
3112 | ||
3113 | /* Move the rest to the buffer to the front, in order to get things properly aligned again */ | |
3114 | memmove(d->buffer.raw, d->buffer.raw + sz, d->buffer_filled - sz); | |
3115 | d->buffer_filled -= sz; | |
3116 | ||
3117 | if (d->buffer_filled == 0) | |
3118 | LIST_REMOVE(buffered, e->inotify_data_buffered, d); | |
3119 | } | |
3120 | ||
3121 | static int event_inotify_data_process(sd_event *e, struct inotify_data *d) { | |
3122 | int r; | |
3123 | ||
3124 | assert(e); | |
3125 | assert(d); | |
3126 | ||
3127 | /* If there's already an event source pending for this priority, don't read another */ | |
3128 | if (d->n_pending > 0) | |
3129 | return 0; | |
3130 | ||
3131 | while (d->buffer_filled > 0) { | |
3132 | size_t sz; | |
3133 | ||
3134 | /* Let's validate that the event structures are complete */ | |
3135 | if (d->buffer_filled < offsetof(struct inotify_event, name)) | |
3136 | return -EIO; | |
3137 | ||
3138 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3139 | if (d->buffer_filled < sz) | |
3140 | return -EIO; | |
3141 | ||
3142 | if (d->buffer.ev.mask & IN_Q_OVERFLOW) { | |
3143 | struct inode_data *inode_data; | |
97ef5391 LP |
3144 | |
3145 | /* The queue overran, let's pass this event to all event sources connected to this inotify | |
3146 | * object */ | |
3147 | ||
90e74a66 | 3148 | HASHMAP_FOREACH(inode_data, d->inodes) { |
97ef5391 LP |
3149 | sd_event_source *s; |
3150 | ||
3151 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
3152 | ||
3153 | if (s->enabled == SD_EVENT_OFF) | |
3154 | continue; | |
3155 | ||
3156 | r = source_set_pending(s, true); | |
3157 | if (r < 0) | |
3158 | return r; | |
3159 | } | |
3160 | } | |
3161 | } else { | |
3162 | struct inode_data *inode_data; | |
3163 | sd_event_source *s; | |
3164 | ||
3165 | /* Find the inode object for this watch descriptor. If IN_IGNORED is set we also remove it from | |
3166 | * our watch descriptor table. */ | |
3167 | if (d->buffer.ev.mask & IN_IGNORED) { | |
3168 | ||
3169 | inode_data = hashmap_remove(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3170 | if (!inode_data) { | |
3171 | event_inotify_data_drop(e, d, sz); | |
3172 | continue; | |
3173 | } | |
3174 | ||
3175 | /* The watch descriptor was removed by the kernel, let's drop it here too */ | |
3176 | inode_data->wd = -1; | |
3177 | } else { | |
3178 | inode_data = hashmap_get(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3179 | if (!inode_data) { | |
3180 | event_inotify_data_drop(e, d, sz); | |
3181 | continue; | |
3182 | } | |
3183 | } | |
3184 | ||
3185 | /* Trigger all event sources that are interested in these events. Also trigger all event | |
3186 | * sources if IN_IGNORED or IN_UNMOUNT is set. */ | |
3187 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
3188 | ||
3189 | if (s->enabled == SD_EVENT_OFF) | |
3190 | continue; | |
3191 | ||
3192 | if ((d->buffer.ev.mask & (IN_IGNORED|IN_UNMOUNT)) == 0 && | |
3193 | (s->inotify.mask & d->buffer.ev.mask & IN_ALL_EVENTS) == 0) | |
3194 | continue; | |
3195 | ||
3196 | r = source_set_pending(s, true); | |
3197 | if (r < 0) | |
3198 | return r; | |
3199 | } | |
3200 | } | |
3201 | ||
3202 | /* Something pending now? If so, let's finish, otherwise let's read more. */ | |
3203 | if (d->n_pending > 0) | |
3204 | return 1; | |
3205 | } | |
3206 | ||
3207 | return 0; | |
3208 | } | |
3209 | ||
3210 | static int process_inotify(sd_event *e) { | |
3211 | struct inotify_data *d; | |
3212 | int r, done = 0; | |
3213 | ||
3214 | assert(e); | |
3215 | ||
3216 | LIST_FOREACH(buffered, d, e->inotify_data_buffered) { | |
3217 | r = event_inotify_data_process(e, d); | |
3218 | if (r < 0) | |
3219 | return r; | |
3220 | if (r > 0) | |
3221 | done ++; | |
3222 | } | |
3223 | ||
3224 | return done; | |
3225 | } | |
3226 | ||
fd38203a | 3227 | static int source_dispatch(sd_event_source *s) { |
b778cba4 | 3228 | _cleanup_(sd_event_unrefp) sd_event *saved_event = NULL; |
8f5c235d | 3229 | EventSourceType saved_type; |
fe8245eb | 3230 | int r = 0; |
fd38203a LP |
3231 | |
3232 | assert(s); | |
6203e07a | 3233 | assert(s->pending || s->type == SOURCE_EXIT); |
fd38203a | 3234 | |
b778cba4 LP |
3235 | /* Save the event source type, here, so that we still know it after the event callback which might |
3236 | * invalidate the event. */ | |
8f5c235d LP |
3237 | saved_type = s->type; |
3238 | ||
b778cba4 LP |
3239 | /* Similar, store a reference to the event loop object, so that we can still access it after the |
3240 | * callback might have invalidated/disconnected the event source. */ | |
3241 | saved_event = sd_event_ref(s->event); | |
3242 | ||
945c2931 | 3243 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { |
da7e457c LP |
3244 | r = source_set_pending(s, false); |
3245 | if (r < 0) | |
3246 | return r; | |
3247 | } | |
fd38203a | 3248 | |
6e9feda3 LP |
3249 | if (s->type != SOURCE_POST) { |
3250 | sd_event_source *z; | |
6e9feda3 LP |
3251 | |
3252 | /* If we execute a non-post source, let's mark all | |
3253 | * post sources as pending */ | |
3254 | ||
90e74a66 | 3255 | SET_FOREACH(z, s->event->post_sources) { |
6e9feda3 LP |
3256 | if (z->enabled == SD_EVENT_OFF) |
3257 | continue; | |
3258 | ||
3259 | r = source_set_pending(z, true); | |
3260 | if (r < 0) | |
3261 | return r; | |
3262 | } | |
3263 | } | |
3264 | ||
baf76283 LP |
3265 | if (s->enabled == SD_EVENT_ONESHOT) { |
3266 | r = sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
3267 | if (r < 0) |
3268 | return r; | |
3269 | } | |
3270 | ||
12179984 | 3271 | s->dispatching = true; |
b7484e2a | 3272 | |
fd38203a LP |
3273 | switch (s->type) { |
3274 | ||
3275 | case SOURCE_IO: | |
3276 | r = s->io.callback(s, s->io.fd, s->io.revents, s->userdata); | |
3277 | break; | |
3278 | ||
6a0f1f6d | 3279 | case SOURCE_TIME_REALTIME: |
a8548816 | 3280 | case SOURCE_TIME_BOOTTIME: |
6a0f1f6d LP |
3281 | case SOURCE_TIME_MONOTONIC: |
3282 | case SOURCE_TIME_REALTIME_ALARM: | |
3283 | case SOURCE_TIME_BOOTTIME_ALARM: | |
fd38203a LP |
3284 | r = s->time.callback(s, s->time.next, s->userdata); |
3285 | break; | |
3286 | ||
3287 | case SOURCE_SIGNAL: | |
3288 | r = s->signal.callback(s, &s->signal.siginfo, s->userdata); | |
3289 | break; | |
3290 | ||
08cd1552 LP |
3291 | case SOURCE_CHILD: { |
3292 | bool zombie; | |
3293 | ||
945c2931 | 3294 | zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 3295 | |
fd38203a | 3296 | r = s->child.callback(s, &s->child.siginfo, s->userdata); |
08cd1552 LP |
3297 | |
3298 | /* Now, reap the PID for good. */ | |
f8f3f926 | 3299 | if (zombie) { |
cc59d290 | 3300 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|WEXITED); |
f8f3f926 LP |
3301 | s->child.waited = true; |
3302 | } | |
08cd1552 | 3303 | |
fd38203a | 3304 | break; |
08cd1552 | 3305 | } |
fd38203a LP |
3306 | |
3307 | case SOURCE_DEFER: | |
3308 | r = s->defer.callback(s, s->userdata); | |
3309 | break; | |
da7e457c | 3310 | |
6e9feda3 LP |
3311 | case SOURCE_POST: |
3312 | r = s->post.callback(s, s->userdata); | |
3313 | break; | |
3314 | ||
6203e07a LP |
3315 | case SOURCE_EXIT: |
3316 | r = s->exit.callback(s, s->userdata); | |
da7e457c | 3317 | break; |
9d3e3aa5 | 3318 | |
97ef5391 LP |
3319 | case SOURCE_INOTIFY: { |
3320 | struct sd_event *e = s->event; | |
3321 | struct inotify_data *d; | |
3322 | size_t sz; | |
3323 | ||
3324 | assert(s->inotify.inode_data); | |
3325 | assert_se(d = s->inotify.inode_data->inotify_data); | |
3326 | ||
3327 | assert(d->buffer_filled >= offsetof(struct inotify_event, name)); | |
3328 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3329 | assert(d->buffer_filled >= sz); | |
3330 | ||
3331 | r = s->inotify.callback(s, &d->buffer.ev, s->userdata); | |
3332 | ||
3333 | /* When no event is pending anymore on this inotify object, then let's drop the event from the | |
3334 | * buffer. */ | |
3335 | if (d->n_pending == 0) | |
3336 | event_inotify_data_drop(e, d, sz); | |
3337 | ||
3338 | break; | |
3339 | } | |
3340 | ||
9d3e3aa5 | 3341 | case SOURCE_WATCHDOG: |
a71fe8b8 | 3342 | case _SOURCE_EVENT_SOURCE_TYPE_MAX: |
9f2a50a3 | 3343 | case _SOURCE_EVENT_SOURCE_TYPE_INVALID: |
9d3e3aa5 | 3344 | assert_not_reached("Wut? I shouldn't exist."); |
fd38203a LP |
3345 | } |
3346 | ||
12179984 LP |
3347 | s->dispatching = false; |
3348 | ||
b778cba4 LP |
3349 | if (r < 0) { |
3350 | log_debug_errno(r, "Event source %s (type %s) returned error, %s: %m", | |
3351 | strna(s->description), | |
3352 | event_source_type_to_string(saved_type), | |
3353 | s->exit_on_failure ? "exiting" : "disabling"); | |
3354 | ||
3355 | if (s->exit_on_failure) | |
3356 | (void) sd_event_exit(saved_event, r); | |
3357 | } | |
12179984 LP |
3358 | |
3359 | if (s->n_ref == 0) | |
3360 | source_free(s); | |
3361 | else if (r < 0) | |
6203e07a | 3362 | sd_event_source_set_enabled(s, SD_EVENT_OFF); |
b7484e2a | 3363 | |
6203e07a | 3364 | return 1; |
fd38203a LP |
3365 | } |
3366 | ||
3367 | static int event_prepare(sd_event *e) { | |
3368 | int r; | |
3369 | ||
3370 | assert(e); | |
3371 | ||
3372 | for (;;) { | |
3373 | sd_event_source *s; | |
3374 | ||
3375 | s = prioq_peek(e->prepare); | |
baf76283 | 3376 | if (!s || s->prepare_iteration == e->iteration || s->enabled == SD_EVENT_OFF) |
fd38203a LP |
3377 | break; |
3378 | ||
3379 | s->prepare_iteration = e->iteration; | |
3380 | r = prioq_reshuffle(e->prepare, s, &s->prepare_index); | |
3381 | if (r < 0) | |
3382 | return r; | |
3383 | ||
3384 | assert(s->prepare); | |
12179984 LP |
3385 | |
3386 | s->dispatching = true; | |
fd38203a | 3387 | r = s->prepare(s, s->userdata); |
12179984 LP |
3388 | s->dispatching = false; |
3389 | ||
b778cba4 LP |
3390 | if (r < 0) { |
3391 | log_debug_errno(r, "Prepare callback of event source %s (type %s) returned error, %s: %m", | |
3392 | strna(s->description), | |
3393 | event_source_type_to_string(s->type), | |
3394 | s->exit_on_failure ? "exiting" : "disabling"); | |
3395 | ||
3396 | if (s->exit_on_failure) | |
3397 | (void) sd_event_exit(e, r); | |
3398 | } | |
fd38203a | 3399 | |
12179984 LP |
3400 | if (s->n_ref == 0) |
3401 | source_free(s); | |
3402 | else if (r < 0) | |
3403 | sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
3404 | } |
3405 | ||
3406 | return 0; | |
3407 | } | |
3408 | ||
6203e07a | 3409 | static int dispatch_exit(sd_event *e) { |
da7e457c LP |
3410 | sd_event_source *p; |
3411 | int r; | |
3412 | ||
3413 | assert(e); | |
3414 | ||
6203e07a | 3415 | p = prioq_peek(e->exit); |
baf76283 | 3416 | if (!p || p->enabled == SD_EVENT_OFF) { |
da7e457c LP |
3417 | e->state = SD_EVENT_FINISHED; |
3418 | return 0; | |
3419 | } | |
3420 | ||
f814c871 | 3421 | _unused_ _cleanup_(sd_event_unrefp) sd_event *ref = sd_event_ref(e); |
da7e457c | 3422 | e->iteration++; |
6203e07a | 3423 | e->state = SD_EVENT_EXITING; |
da7e457c | 3424 | r = source_dispatch(p); |
2b0c9ef7 | 3425 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
3426 | return r; |
3427 | } | |
3428 | ||
c2ba3ad6 LP |
3429 | static sd_event_source* event_next_pending(sd_event *e) { |
3430 | sd_event_source *p; | |
3431 | ||
da7e457c LP |
3432 | assert(e); |
3433 | ||
c2ba3ad6 LP |
3434 | p = prioq_peek(e->pending); |
3435 | if (!p) | |
3436 | return NULL; | |
3437 | ||
baf76283 | 3438 | if (p->enabled == SD_EVENT_OFF) |
c2ba3ad6 LP |
3439 | return NULL; |
3440 | ||
3441 | return p; | |
3442 | } | |
3443 | ||
cde93897 LP |
3444 | static int arm_watchdog(sd_event *e) { |
3445 | struct itimerspec its = {}; | |
3446 | usec_t t; | |
cde93897 LP |
3447 | |
3448 | assert(e); | |
3449 | assert(e->watchdog_fd >= 0); | |
3450 | ||
3451 | t = sleep_between(e, | |
3452 | e->watchdog_last + (e->watchdog_period / 2), | |
3453 | e->watchdog_last + (e->watchdog_period * 3 / 4)); | |
3454 | ||
3455 | timespec_store(&its.it_value, t); | |
3456 | ||
75145780 LP |
3457 | /* Make sure we never set the watchdog to 0, which tells the |
3458 | * kernel to disable it. */ | |
3459 | if (its.it_value.tv_sec == 0 && its.it_value.tv_nsec == 0) | |
3460 | its.it_value.tv_nsec = 1; | |
3461 | ||
15c689d7 | 3462 | if (timerfd_settime(e->watchdog_fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 LP |
3463 | return -errno; |
3464 | ||
3465 | return 0; | |
3466 | } | |
3467 | ||
3468 | static int process_watchdog(sd_event *e) { | |
3469 | assert(e); | |
3470 | ||
3471 | if (!e->watchdog) | |
3472 | return 0; | |
3473 | ||
3474 | /* Don't notify watchdog too often */ | |
3475 | if (e->watchdog_last + e->watchdog_period / 4 > e->timestamp.monotonic) | |
3476 | return 0; | |
3477 | ||
3478 | sd_notify(false, "WATCHDOG=1"); | |
3479 | e->watchdog_last = e->timestamp.monotonic; | |
3480 | ||
3481 | return arm_watchdog(e); | |
3482 | } | |
3483 | ||
97ef5391 LP |
3484 | static void event_close_inode_data_fds(sd_event *e) { |
3485 | struct inode_data *d; | |
3486 | ||
3487 | assert(e); | |
3488 | ||
3489 | /* Close the fds pointing to the inodes to watch now. We need to close them as they might otherwise pin | |
3490 | * filesystems. But we can't close them right-away as we need them as long as the user still wants to make | |
5238e957 | 3491 | * adjustments to the even source, such as changing the priority (which requires us to remove and re-add a watch |
97ef5391 LP |
3492 | * for the inode). Hence, let's close them when entering the first iteration after they were added, as a |
3493 | * compromise. */ | |
3494 | ||
3495 | while ((d = e->inode_data_to_close)) { | |
3496 | assert(d->fd >= 0); | |
3497 | d->fd = safe_close(d->fd); | |
3498 | ||
3499 | LIST_REMOVE(to_close, e->inode_data_to_close, d); | |
3500 | } | |
3501 | } | |
3502 | ||
c45a5a74 TG |
3503 | _public_ int sd_event_prepare(sd_event *e) { |
3504 | int r; | |
fd38203a | 3505 | |
da7e457c | 3506 | assert_return(e, -EINVAL); |
b937d761 | 3507 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
3508 | assert_return(!event_pid_changed(e), -ECHILD); |
3509 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 3510 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 3511 | |
e5446015 LP |
3512 | /* Let's check that if we are a default event loop we are executed in the correct thread. We only do |
3513 | * this check here once, since gettid() is typically not cached, and thus want to minimize | |
3514 | * syscalls */ | |
3515 | assert_return(!e->default_event_ptr || e->tid == gettid(), -EREMOTEIO); | |
3516 | ||
f814c871 LP |
3517 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
3518 | _unused_ _cleanup_(sd_event_unrefp) sd_event *ref = sd_event_ref(e); | |
3519 | ||
6203e07a | 3520 | if (e->exit_requested) |
c45a5a74 | 3521 | goto pending; |
fd38203a LP |
3522 | |
3523 | e->iteration++; | |
3524 | ||
0be6c2f6 | 3525 | e->state = SD_EVENT_PREPARING; |
fd38203a | 3526 | r = event_prepare(e); |
0be6c2f6 | 3527 | e->state = SD_EVENT_INITIAL; |
fd38203a | 3528 | if (r < 0) |
c45a5a74 | 3529 | return r; |
fd38203a | 3530 | |
6a0f1f6d LP |
3531 | r = event_arm_timer(e, &e->realtime); |
3532 | if (r < 0) | |
c45a5a74 | 3533 | return r; |
6a0f1f6d | 3534 | |
a8548816 TG |
3535 | r = event_arm_timer(e, &e->boottime); |
3536 | if (r < 0) | |
c45a5a74 | 3537 | return r; |
a8548816 | 3538 | |
6a0f1f6d LP |
3539 | r = event_arm_timer(e, &e->monotonic); |
3540 | if (r < 0) | |
c45a5a74 | 3541 | return r; |
6a0f1f6d LP |
3542 | |
3543 | r = event_arm_timer(e, &e->realtime_alarm); | |
1b5995b0 | 3544 | if (r < 0) |
c45a5a74 | 3545 | return r; |
fd38203a | 3546 | |
6a0f1f6d | 3547 | r = event_arm_timer(e, &e->boottime_alarm); |
1b5995b0 | 3548 | if (r < 0) |
c45a5a74 | 3549 | return r; |
fd38203a | 3550 | |
97ef5391 LP |
3551 | event_close_inode_data_fds(e); |
3552 | ||
1b5995b0 | 3553 | if (event_next_pending(e) || e->need_process_child) |
c45a5a74 TG |
3554 | goto pending; |
3555 | ||
2b0c9ef7 | 3556 | e->state = SD_EVENT_ARMED; |
c45a5a74 TG |
3557 | |
3558 | return 0; | |
3559 | ||
3560 | pending: | |
2b0c9ef7 | 3561 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
3562 | r = sd_event_wait(e, 0); |
3563 | if (r == 0) | |
2b0c9ef7 | 3564 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
3565 | |
3566 | return r; | |
c45a5a74 TG |
3567 | } |
3568 | ||
3569 | _public_ int sd_event_wait(sd_event *e, uint64_t timeout) { | |
5cddd924 | 3570 | size_t event_queue_max; |
c45a5a74 TG |
3571 | int r, m, i; |
3572 | ||
3573 | assert_return(e, -EINVAL); | |
b937d761 | 3574 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
3575 | assert_return(!event_pid_changed(e), -ECHILD); |
3576 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 3577 | assert_return(e->state == SD_EVENT_ARMED, -EBUSY); |
c45a5a74 TG |
3578 | |
3579 | if (e->exit_requested) { | |
3580 | e->state = SD_EVENT_PENDING; | |
3581 | return 1; | |
3582 | } | |
6a0f1f6d | 3583 | |
5cddd924 LP |
3584 | event_queue_max = MAX(e->n_sources, 1u); |
3585 | if (!GREEDY_REALLOC(e->event_queue, e->event_queue_allocated, event_queue_max)) | |
3586 | return -ENOMEM; | |
fd38203a | 3587 | |
97ef5391 LP |
3588 | /* If we still have inotify data buffered, then query the other fds, but don't wait on it */ |
3589 | if (e->inotify_data_buffered) | |
3590 | timeout = 0; | |
3591 | ||
5cddd924 | 3592 | m = epoll_wait(e->epoll_fd, e->event_queue, event_queue_max, |
bab4820e | 3593 | timeout == (uint64_t) -1 ? -1 : (int) DIV_ROUND_UP(timeout, USEC_PER_MSEC)); |
da7e457c | 3594 | if (m < 0) { |
c45a5a74 TG |
3595 | if (errno == EINTR) { |
3596 | e->state = SD_EVENT_PENDING; | |
3597 | return 1; | |
3598 | } | |
3599 | ||
3600 | r = -errno; | |
da7e457c LP |
3601 | goto finish; |
3602 | } | |
fd38203a | 3603 | |
e475d10c | 3604 | triple_timestamp_get(&e->timestamp); |
fd38203a LP |
3605 | |
3606 | for (i = 0; i < m; i++) { | |
3607 | ||
5cddd924 LP |
3608 | if (e->event_queue[i].data.ptr == INT_TO_PTR(SOURCE_WATCHDOG)) |
3609 | r = flush_timer(e, e->watchdog_fd, e->event_queue[i].events, NULL); | |
9da4cb2b | 3610 | else { |
5cddd924 | 3611 | WakeupType *t = e->event_queue[i].data.ptr; |
9da4cb2b LP |
3612 | |
3613 | switch (*t) { | |
3614 | ||
f8f3f926 | 3615 | case WAKEUP_EVENT_SOURCE: { |
5cddd924 | 3616 | sd_event_source *s = e->event_queue[i].data.ptr; |
f8f3f926 LP |
3617 | |
3618 | assert(s); | |
3619 | ||
3620 | switch (s->type) { | |
3621 | ||
3622 | case SOURCE_IO: | |
5cddd924 | 3623 | r = process_io(e, s, e->event_queue[i].events); |
f8f3f926 LP |
3624 | break; |
3625 | ||
3626 | case SOURCE_CHILD: | |
5cddd924 | 3627 | r = process_pidfd(e, s, e->event_queue[i].events); |
f8f3f926 LP |
3628 | break; |
3629 | ||
3630 | default: | |
3631 | assert_not_reached("Unexpected event source type"); | |
3632 | } | |
3633 | ||
9da4cb2b | 3634 | break; |
f8f3f926 | 3635 | } |
fd38203a | 3636 | |
9da4cb2b | 3637 | case WAKEUP_CLOCK_DATA: { |
5cddd924 | 3638 | struct clock_data *d = e->event_queue[i].data.ptr; |
f8f3f926 LP |
3639 | |
3640 | assert(d); | |
3641 | ||
5cddd924 | 3642 | r = flush_timer(e, d->fd, e->event_queue[i].events, &d->next); |
9da4cb2b LP |
3643 | break; |
3644 | } | |
3645 | ||
3646 | case WAKEUP_SIGNAL_DATA: | |
5cddd924 | 3647 | r = process_signal(e, e->event_queue[i].data.ptr, e->event_queue[i].events); |
9da4cb2b LP |
3648 | break; |
3649 | ||
97ef5391 | 3650 | case WAKEUP_INOTIFY_DATA: |
5cddd924 | 3651 | r = event_inotify_data_read(e, e->event_queue[i].data.ptr, e->event_queue[i].events); |
97ef5391 LP |
3652 | break; |
3653 | ||
9da4cb2b LP |
3654 | default: |
3655 | assert_not_reached("Invalid wake-up pointer"); | |
3656 | } | |
3657 | } | |
fd38203a | 3658 | if (r < 0) |
da7e457c | 3659 | goto finish; |
fd38203a LP |
3660 | } |
3661 | ||
cde93897 LP |
3662 | r = process_watchdog(e); |
3663 | if (r < 0) | |
3664 | goto finish; | |
3665 | ||
6a0f1f6d LP |
3666 | r = process_timer(e, e->timestamp.realtime, &e->realtime); |
3667 | if (r < 0) | |
3668 | goto finish; | |
3669 | ||
e475d10c | 3670 | r = process_timer(e, e->timestamp.boottime, &e->boottime); |
a8548816 TG |
3671 | if (r < 0) |
3672 | goto finish; | |
3673 | ||
6a0f1f6d LP |
3674 | r = process_timer(e, e->timestamp.monotonic, &e->monotonic); |
3675 | if (r < 0) | |
3676 | goto finish; | |
3677 | ||
3678 | r = process_timer(e, e->timestamp.realtime, &e->realtime_alarm); | |
fd38203a | 3679 | if (r < 0) |
da7e457c | 3680 | goto finish; |
fd38203a | 3681 | |
e475d10c | 3682 | r = process_timer(e, e->timestamp.boottime, &e->boottime_alarm); |
fd38203a | 3683 | if (r < 0) |
da7e457c | 3684 | goto finish; |
fd38203a | 3685 | |
c2ba3ad6 | 3686 | if (e->need_process_child) { |
fd38203a LP |
3687 | r = process_child(e); |
3688 | if (r < 0) | |
da7e457c | 3689 | goto finish; |
fd38203a LP |
3690 | } |
3691 | ||
97ef5391 LP |
3692 | r = process_inotify(e); |
3693 | if (r < 0) | |
3694 | goto finish; | |
3695 | ||
c45a5a74 TG |
3696 | if (event_next_pending(e)) { |
3697 | e->state = SD_EVENT_PENDING; | |
3698 | ||
3699 | return 1; | |
da7e457c LP |
3700 | } |
3701 | ||
c45a5a74 | 3702 | r = 0; |
fd38203a | 3703 | |
da7e457c | 3704 | finish: |
2b0c9ef7 | 3705 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
3706 | |
3707 | return r; | |
fd38203a LP |
3708 | } |
3709 | ||
c45a5a74 TG |
3710 | _public_ int sd_event_dispatch(sd_event *e) { |
3711 | sd_event_source *p; | |
3712 | int r; | |
3713 | ||
3714 | assert_return(e, -EINVAL); | |
b937d761 | 3715 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
3716 | assert_return(!event_pid_changed(e), -ECHILD); |
3717 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
3718 | assert_return(e->state == SD_EVENT_PENDING, -EBUSY); | |
3719 | ||
3720 | if (e->exit_requested) | |
3721 | return dispatch_exit(e); | |
3722 | ||
3723 | p = event_next_pending(e); | |
3724 | if (p) { | |
f814c871 | 3725 | _unused_ _cleanup_(sd_event_unrefp) sd_event *ref = sd_event_ref(e); |
c45a5a74 TG |
3726 | |
3727 | e->state = SD_EVENT_RUNNING; | |
3728 | r = source_dispatch(p); | |
2b0c9ef7 | 3729 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
3730 | return r; |
3731 | } | |
3732 | ||
2b0c9ef7 | 3733 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
3734 | |
3735 | return 1; | |
3736 | } | |
3737 | ||
34b87517 | 3738 | static void event_log_delays(sd_event *e) { |
442ac269 YW |
3739 | char b[ELEMENTSOF(e->delays) * DECIMAL_STR_MAX(unsigned) + 1], *p; |
3740 | size_t l, i; | |
34b87517 | 3741 | |
442ac269 YW |
3742 | p = b; |
3743 | l = sizeof(b); | |
3744 | for (i = 0; i < ELEMENTSOF(e->delays); i++) { | |
3745 | l = strpcpyf(&p, l, "%u ", e->delays[i]); | |
34b87517 VC |
3746 | e->delays[i] = 0; |
3747 | } | |
442ac269 | 3748 | log_debug("Event loop iterations: %s", b); |
34b87517 VC |
3749 | } |
3750 | ||
c45a5a74 TG |
3751 | _public_ int sd_event_run(sd_event *e, uint64_t timeout) { |
3752 | int r; | |
3753 | ||
3754 | assert_return(e, -EINVAL); | |
b937d761 | 3755 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
3756 | assert_return(!event_pid_changed(e), -ECHILD); |
3757 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 3758 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
c45a5a74 | 3759 | |
e6a7bee5 | 3760 | if (e->profile_delays && e->last_run_usec != 0) { |
34b87517 VC |
3761 | usec_t this_run; |
3762 | unsigned l; | |
3763 | ||
3764 | this_run = now(CLOCK_MONOTONIC); | |
3765 | ||
e6a7bee5 | 3766 | l = u64log2(this_run - e->last_run_usec); |
cb9d621e | 3767 | assert(l < ELEMENTSOF(e->delays)); |
34b87517 VC |
3768 | e->delays[l]++; |
3769 | ||
e6a7bee5 | 3770 | if (this_run - e->last_log_usec >= 5*USEC_PER_SEC) { |
34b87517 | 3771 | event_log_delays(e); |
e6a7bee5 | 3772 | e->last_log_usec = this_run; |
34b87517 VC |
3773 | } |
3774 | } | |
3775 | ||
f814c871 LP |
3776 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
3777 | _unused_ _cleanup_(sd_event_unrefp) sd_event *ref = sd_event_ref(e); | |
3778 | ||
c45a5a74 | 3779 | r = sd_event_prepare(e); |
53bac4e0 LP |
3780 | if (r == 0) |
3781 | /* There was nothing? Then wait... */ | |
3782 | r = sd_event_wait(e, timeout); | |
c45a5a74 | 3783 | |
34b87517 | 3784 | if (e->profile_delays) |
e6a7bee5 | 3785 | e->last_run_usec = now(CLOCK_MONOTONIC); |
34b87517 | 3786 | |
02d30981 | 3787 | if (r > 0) { |
53bac4e0 | 3788 | /* There's something now, then let's dispatch it */ |
02d30981 TG |
3789 | r = sd_event_dispatch(e); |
3790 | if (r < 0) | |
3791 | return r; | |
53bac4e0 LP |
3792 | |
3793 | return 1; | |
3794 | } | |
3795 | ||
3796 | return r; | |
c45a5a74 TG |
3797 | } |
3798 | ||
f7262a9f | 3799 | _public_ int sd_event_loop(sd_event *e) { |
fd38203a LP |
3800 | int r; |
3801 | ||
da7e457c | 3802 | assert_return(e, -EINVAL); |
b937d761 | 3803 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 3804 | assert_return(!event_pid_changed(e), -ECHILD); |
2b0c9ef7 | 3805 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 3806 | |
f814c871 | 3807 | _unused_ _cleanup_(sd_event_unrefp) sd_event *ref = NULL; |
fd38203a | 3808 | |
da7e457c | 3809 | while (e->state != SD_EVENT_FINISHED) { |
fd38203a LP |
3810 | r = sd_event_run(e, (uint64_t) -1); |
3811 | if (r < 0) | |
30dd293c | 3812 | return r; |
fd38203a LP |
3813 | } |
3814 | ||
30dd293c | 3815 | return e->exit_code; |
fd38203a LP |
3816 | } |
3817 | ||
9b364545 | 3818 | _public_ int sd_event_get_fd(sd_event *e) { |
9b364545 | 3819 | assert_return(e, -EINVAL); |
b937d761 | 3820 | assert_return(e = event_resolve(e), -ENOPKG); |
9b364545 TG |
3821 | assert_return(!event_pid_changed(e), -ECHILD); |
3822 | ||
3823 | return e->epoll_fd; | |
3824 | } | |
3825 | ||
f7262a9f | 3826 | _public_ int sd_event_get_state(sd_event *e) { |
da7e457c | 3827 | assert_return(e, -EINVAL); |
b937d761 | 3828 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
3829 | assert_return(!event_pid_changed(e), -ECHILD); |
3830 | ||
3831 | return e->state; | |
3832 | } | |
3833 | ||
6203e07a | 3834 | _public_ int sd_event_get_exit_code(sd_event *e, int *code) { |
da7e457c | 3835 | assert_return(e, -EINVAL); |
b937d761 | 3836 | assert_return(e = event_resolve(e), -ENOPKG); |
6203e07a | 3837 | assert_return(code, -EINVAL); |
da7e457c | 3838 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 3839 | |
6203e07a LP |
3840 | if (!e->exit_requested) |
3841 | return -ENODATA; | |
3842 | ||
3843 | *code = e->exit_code; | |
3844 | return 0; | |
fd38203a LP |
3845 | } |
3846 | ||
6203e07a | 3847 | _public_ int sd_event_exit(sd_event *e, int code) { |
da7e457c | 3848 | assert_return(e, -EINVAL); |
b937d761 | 3849 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
3850 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
3851 | assert_return(!event_pid_changed(e), -ECHILD); | |
fd38203a | 3852 | |
6203e07a LP |
3853 | e->exit_requested = true; |
3854 | e->exit_code = code; | |
3855 | ||
fd38203a LP |
3856 | return 0; |
3857 | } | |
46e8c825 | 3858 | |
6a0f1f6d | 3859 | _public_ int sd_event_now(sd_event *e, clockid_t clock, uint64_t *usec) { |
46e8c825 | 3860 | assert_return(e, -EINVAL); |
b937d761 | 3861 | assert_return(e = event_resolve(e), -ENOPKG); |
46e8c825 | 3862 | assert_return(usec, -EINVAL); |
46e8c825 LP |
3863 | assert_return(!event_pid_changed(e), -ECHILD); |
3864 | ||
e475d10c LP |
3865 | if (!TRIPLE_TIMESTAMP_HAS_CLOCK(clock)) |
3866 | return -EOPNOTSUPP; | |
3867 | ||
3868 | /* Generate a clean error in case CLOCK_BOOTTIME is not available. Note that don't use clock_supported() here, | |
3869 | * for a reason: there are systems where CLOCK_BOOTTIME is supported, but CLOCK_BOOTTIME_ALARM is not, but for | |
3870 | * the purpose of getting the time this doesn't matter. */ | |
3411372e LP |
3871 | if (IN_SET(clock, CLOCK_BOOTTIME, CLOCK_BOOTTIME_ALARM) && !clock_boottime_supported()) |
3872 | return -EOPNOTSUPP; | |
3873 | ||
e475d10c | 3874 | if (!triple_timestamp_is_set(&e->timestamp)) { |
15c689d7 | 3875 | /* Implicitly fall back to now() if we never ran before and thus have no cached time. */ |
38a03f06 LP |
3876 | *usec = now(clock); |
3877 | return 1; | |
3878 | } | |
46e8c825 | 3879 | |
e475d10c | 3880 | *usec = triple_timestamp_by_clock(&e->timestamp, clock); |
46e8c825 LP |
3881 | return 0; |
3882 | } | |
afc6adb5 LP |
3883 | |
3884 | _public_ int sd_event_default(sd_event **ret) { | |
39883f62 | 3885 | sd_event *e = NULL; |
afc6adb5 LP |
3886 | int r; |
3887 | ||
3888 | if (!ret) | |
3889 | return !!default_event; | |
3890 | ||
3891 | if (default_event) { | |
3892 | *ret = sd_event_ref(default_event); | |
3893 | return 0; | |
3894 | } | |
3895 | ||
3896 | r = sd_event_new(&e); | |
3897 | if (r < 0) | |
3898 | return r; | |
3899 | ||
3900 | e->default_event_ptr = &default_event; | |
3901 | e->tid = gettid(); | |
3902 | default_event = e; | |
3903 | ||
3904 | *ret = e; | |
3905 | return 1; | |
3906 | } | |
3907 | ||
3908 | _public_ int sd_event_get_tid(sd_event *e, pid_t *tid) { | |
3909 | assert_return(e, -EINVAL); | |
b937d761 | 3910 | assert_return(e = event_resolve(e), -ENOPKG); |
afc6adb5 | 3911 | assert_return(tid, -EINVAL); |
76b54375 | 3912 | assert_return(!event_pid_changed(e), -ECHILD); |
afc6adb5 | 3913 | |
76b54375 LP |
3914 | if (e->tid != 0) { |
3915 | *tid = e->tid; | |
3916 | return 0; | |
3917 | } | |
3918 | ||
3919 | return -ENXIO; | |
afc6adb5 | 3920 | } |
cde93897 LP |
3921 | |
3922 | _public_ int sd_event_set_watchdog(sd_event *e, int b) { | |
3923 | int r; | |
3924 | ||
3925 | assert_return(e, -EINVAL); | |
b937d761 | 3926 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 | 3927 | assert_return(!event_pid_changed(e), -ECHILD); |
cde93897 LP |
3928 | |
3929 | if (e->watchdog == !!b) | |
3930 | return e->watchdog; | |
3931 | ||
3932 | if (b) { | |
09812eb7 LP |
3933 | r = sd_watchdog_enabled(false, &e->watchdog_period); |
3934 | if (r <= 0) | |
cde93897 | 3935 | return r; |
cde93897 LP |
3936 | |
3937 | /* Issue first ping immediately */ | |
3938 | sd_notify(false, "WATCHDOG=1"); | |
3939 | e->watchdog_last = now(CLOCK_MONOTONIC); | |
3940 | ||
3941 | e->watchdog_fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC); | |
3942 | if (e->watchdog_fd < 0) | |
3943 | return -errno; | |
3944 | ||
3945 | r = arm_watchdog(e); | |
3946 | if (r < 0) | |
3947 | goto fail; | |
3948 | ||
1eac7948 | 3949 | struct epoll_event ev = { |
a82f89aa LP |
3950 | .events = EPOLLIN, |
3951 | .data.ptr = INT_TO_PTR(SOURCE_WATCHDOG), | |
3952 | }; | |
cde93897 | 3953 | |
15c689d7 | 3954 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, e->watchdog_fd, &ev) < 0) { |
cde93897 LP |
3955 | r = -errno; |
3956 | goto fail; | |
3957 | } | |
3958 | ||
3959 | } else { | |
3960 | if (e->watchdog_fd >= 0) { | |
5a795bff | 3961 | (void) epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, e->watchdog_fd, NULL); |
03e334a1 | 3962 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
3963 | } |
3964 | } | |
3965 | ||
3966 | e->watchdog = !!b; | |
3967 | return e->watchdog; | |
3968 | ||
3969 | fail: | |
03e334a1 | 3970 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
3971 | return r; |
3972 | } | |
8f726607 LP |
3973 | |
3974 | _public_ int sd_event_get_watchdog(sd_event *e) { | |
3975 | assert_return(e, -EINVAL); | |
b937d761 | 3976 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 LP |
3977 | assert_return(!event_pid_changed(e), -ECHILD); |
3978 | ||
3979 | return e->watchdog; | |
3980 | } | |
60a3b1e1 LP |
3981 | |
3982 | _public_ int sd_event_get_iteration(sd_event *e, uint64_t *ret) { | |
3983 | assert_return(e, -EINVAL); | |
b937d761 | 3984 | assert_return(e = event_resolve(e), -ENOPKG); |
60a3b1e1 LP |
3985 | assert_return(!event_pid_changed(e), -ECHILD); |
3986 | ||
3987 | *ret = e->iteration; | |
3988 | return 0; | |
3989 | } | |
15723a1d LP |
3990 | |
3991 | _public_ int sd_event_source_set_destroy_callback(sd_event_source *s, sd_event_destroy_t callback) { | |
3992 | assert_return(s, -EINVAL); | |
3993 | ||
3994 | s->destroy_callback = callback; | |
3995 | return 0; | |
3996 | } | |
3997 | ||
3998 | _public_ int sd_event_source_get_destroy_callback(sd_event_source *s, sd_event_destroy_t *ret) { | |
3999 | assert_return(s, -EINVAL); | |
4000 | ||
4001 | if (ret) | |
4002 | *ret = s->destroy_callback; | |
4003 | ||
4004 | return !!s->destroy_callback; | |
4005 | } | |
2382c936 YW |
4006 | |
4007 | _public_ int sd_event_source_get_floating(sd_event_source *s) { | |
4008 | assert_return(s, -EINVAL); | |
4009 | ||
4010 | return s->floating; | |
4011 | } | |
4012 | ||
4013 | _public_ int sd_event_source_set_floating(sd_event_source *s, int b) { | |
4014 | assert_return(s, -EINVAL); | |
4015 | ||
4016 | if (s->floating == !!b) | |
4017 | return 0; | |
4018 | ||
4019 | if (!s->event) /* Already disconnected */ | |
4020 | return -ESTALE; | |
4021 | ||
4022 | s->floating = b; | |
4023 | ||
4024 | if (b) { | |
4025 | sd_event_source_ref(s); | |
4026 | sd_event_unref(s->event); | |
4027 | } else { | |
4028 | sd_event_ref(s->event); | |
4029 | sd_event_source_unref(s); | |
4030 | } | |
4031 | ||
4032 | return 1; | |
4033 | } | |
b778cba4 LP |
4034 | |
4035 | _public_ int sd_event_source_get_exit_on_failure(sd_event_source *s) { | |
4036 | assert_return(s, -EINVAL); | |
4037 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
4038 | ||
4039 | return s->exit_on_failure; | |
4040 | } | |
4041 | ||
4042 | _public_ int sd_event_source_set_exit_on_failure(sd_event_source *s, int b) { | |
4043 | assert_return(s, -EINVAL); | |
4044 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
4045 | ||
4046 | if (s->exit_on_failure == !!b) | |
4047 | return 0; | |
4048 | ||
4049 | s->exit_on_failure = b; | |
4050 | return 1; | |
4051 | } |