]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
fd38203a LP |
2 | |
3 | #include <sys/epoll.h> | |
4 | #include <sys/timerfd.h> | |
5 | #include <sys/wait.h> | |
6 | ||
cde93897 | 7 | #include "sd-daemon.h" |
07630cea LP |
8 | #include "sd-event.h" |
9 | #include "sd-id128.h" | |
10 | ||
b5efdb8a | 11 | #include "alloc-util.h" |
f8f3f926 | 12 | #include "env-util.h" |
a137a1c3 | 13 | #include "event-source.h" |
3ffd4af2 | 14 | #include "fd-util.h" |
97ef5391 | 15 | #include "fs-util.h" |
fd38203a | 16 | #include "hashmap.h" |
07630cea LP |
17 | #include "list.h" |
18 | #include "macro.h" | |
0a970718 | 19 | #include "memory-util.h" |
f5947a5e | 20 | #include "missing_syscall.h" |
07630cea | 21 | #include "prioq.h" |
4a0b58c4 | 22 | #include "process-util.h" |
6e9feda3 | 23 | #include "set.h" |
24882e06 | 24 | #include "signal-util.h" |
55cbfaa5 | 25 | #include "string-table.h" |
07630cea | 26 | #include "string-util.h" |
442ac269 | 27 | #include "strxcpyx.h" |
07630cea | 28 | #include "time-util.h" |
fd38203a | 29 | |
c2ba3ad6 | 30 | #define DEFAULT_ACCURACY_USEC (250 * USEC_PER_MSEC) |
fd38203a | 31 | |
f8f3f926 LP |
32 | static bool EVENT_SOURCE_WATCH_PIDFD(sd_event_source *s) { |
33 | /* Returns true if this is a PID event source and can be implemented by watching EPOLLIN */ | |
34 | return s && | |
35 | s->type == SOURCE_CHILD && | |
36 | s->child.pidfd >= 0 && | |
37 | s->child.options == WEXITED; | |
38 | } | |
39 | ||
55cbfaa5 DM |
40 | static const char* const event_source_type_table[_SOURCE_EVENT_SOURCE_TYPE_MAX] = { |
41 | [SOURCE_IO] = "io", | |
42 | [SOURCE_TIME_REALTIME] = "realtime", | |
43 | [SOURCE_TIME_BOOTTIME] = "bootime", | |
44 | [SOURCE_TIME_MONOTONIC] = "monotonic", | |
45 | [SOURCE_TIME_REALTIME_ALARM] = "realtime-alarm", | |
46 | [SOURCE_TIME_BOOTTIME_ALARM] = "boottime-alarm", | |
47 | [SOURCE_SIGNAL] = "signal", | |
48 | [SOURCE_CHILD] = "child", | |
49 | [SOURCE_DEFER] = "defer", | |
50 | [SOURCE_POST] = "post", | |
51 | [SOURCE_EXIT] = "exit", | |
52 | [SOURCE_WATCHDOG] = "watchdog", | |
97ef5391 | 53 | [SOURCE_INOTIFY] = "inotify", |
55cbfaa5 DM |
54 | }; |
55 | ||
56 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(event_source_type, int); | |
57 | ||
a8548816 | 58 | #define EVENT_SOURCE_IS_TIME(t) IN_SET((t), SOURCE_TIME_REALTIME, SOURCE_TIME_BOOTTIME, SOURCE_TIME_MONOTONIC, SOURCE_TIME_REALTIME_ALARM, SOURCE_TIME_BOOTTIME_ALARM) |
6a0f1f6d | 59 | |
fd38203a | 60 | struct sd_event { |
da7e457c | 61 | unsigned n_ref; |
fd38203a LP |
62 | |
63 | int epoll_fd; | |
cde93897 | 64 | int watchdog_fd; |
fd38203a LP |
65 | |
66 | Prioq *pending; | |
67 | Prioq *prepare; | |
c2ba3ad6 | 68 | |
a8548816 | 69 | /* timerfd_create() only supports these five clocks so far. We |
6a0f1f6d LP |
70 | * can add support for more clocks when the kernel learns to |
71 | * deal with them, too. */ | |
72 | struct clock_data realtime; | |
a8548816 | 73 | struct clock_data boottime; |
6a0f1f6d LP |
74 | struct clock_data monotonic; |
75 | struct clock_data realtime_alarm; | |
76 | struct clock_data boottime_alarm; | |
fd38203a | 77 | |
da7e457c LP |
78 | usec_t perturb; |
79 | ||
9da4cb2b LP |
80 | sd_event_source **signal_sources; /* indexed by signal number */ |
81 | Hashmap *signal_data; /* indexed by priority */ | |
fd38203a LP |
82 | |
83 | Hashmap *child_sources; | |
baf76283 | 84 | unsigned n_enabled_child_sources; |
fd38203a | 85 | |
6e9feda3 LP |
86 | Set *post_sources; |
87 | ||
6203e07a | 88 | Prioq *exit; |
fd38203a | 89 | |
97ef5391 LP |
90 | Hashmap *inotify_data; /* indexed by priority */ |
91 | ||
92 | /* A list of inode structures that still have an fd open, that we need to close before the next loop iteration */ | |
93 | LIST_HEAD(struct inode_data, inode_data_to_close); | |
94 | ||
95 | /* A list of inotify objects that already have events buffered which aren't processed yet */ | |
96 | LIST_HEAD(struct inotify_data, inotify_data_buffered); | |
97 | ||
da7e457c | 98 | pid_t original_pid; |
c2ba3ad6 | 99 | |
60a3b1e1 | 100 | uint64_t iteration; |
e475d10c | 101 | triple_timestamp timestamp; |
da7e457c | 102 | int state; |
eaa3cbef | 103 | |
6203e07a | 104 | bool exit_requested:1; |
da7e457c | 105 | bool need_process_child:1; |
cde93897 | 106 | bool watchdog:1; |
34b87517 | 107 | bool profile_delays:1; |
afc6adb5 | 108 | |
6203e07a LP |
109 | int exit_code; |
110 | ||
afc6adb5 LP |
111 | pid_t tid; |
112 | sd_event **default_event_ptr; | |
cde93897 LP |
113 | |
114 | usec_t watchdog_last, watchdog_period; | |
15b38f93 LP |
115 | |
116 | unsigned n_sources; | |
a71fe8b8 | 117 | |
5cddd924 LP |
118 | struct epoll_event *event_queue; |
119 | size_t event_queue_allocated; | |
120 | ||
a71fe8b8 | 121 | LIST_HEAD(sd_event_source, sources); |
34b87517 VC |
122 | |
123 | usec_t last_run, last_log; | |
124 | unsigned delays[sizeof(usec_t) * 8]; | |
fd38203a LP |
125 | }; |
126 | ||
b937d761 NM |
127 | static thread_local sd_event *default_event = NULL; |
128 | ||
a71fe8b8 | 129 | static void source_disconnect(sd_event_source *s); |
97ef5391 | 130 | static void event_gc_inode_data(sd_event *e, struct inode_data *d); |
a71fe8b8 | 131 | |
b937d761 NM |
132 | static sd_event *event_resolve(sd_event *e) { |
133 | return e == SD_EVENT_DEFAULT ? default_event : e; | |
134 | } | |
135 | ||
fd38203a LP |
136 | static int pending_prioq_compare(const void *a, const void *b) { |
137 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 138 | int r; |
fd38203a LP |
139 | |
140 | assert(x->pending); | |
141 | assert(y->pending); | |
142 | ||
baf76283 LP |
143 | /* Enabled ones first */ |
144 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
fd38203a | 145 | return -1; |
baf76283 | 146 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) |
fd38203a LP |
147 | return 1; |
148 | ||
149 | /* Lower priority values first */ | |
9c57a73b YW |
150 | r = CMP(x->priority, y->priority); |
151 | if (r != 0) | |
152 | return r; | |
fd38203a LP |
153 | |
154 | /* Older entries first */ | |
9c57a73b | 155 | return CMP(x->pending_iteration, y->pending_iteration); |
fd38203a LP |
156 | } |
157 | ||
158 | static int prepare_prioq_compare(const void *a, const void *b) { | |
159 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 160 | int r; |
fd38203a LP |
161 | |
162 | assert(x->prepare); | |
163 | assert(y->prepare); | |
164 | ||
8046c457 KK |
165 | /* Enabled ones first */ |
166 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
167 | return -1; | |
168 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) | |
169 | return 1; | |
170 | ||
fd38203a LP |
171 | /* Move most recently prepared ones last, so that we can stop |
172 | * preparing as soon as we hit one that has already been | |
173 | * prepared in the current iteration */ | |
9c57a73b YW |
174 | r = CMP(x->prepare_iteration, y->prepare_iteration); |
175 | if (r != 0) | |
176 | return r; | |
fd38203a | 177 | |
fd38203a | 178 | /* Lower priority values first */ |
9c57a73b | 179 | return CMP(x->priority, y->priority); |
fd38203a LP |
180 | } |
181 | ||
c2ba3ad6 | 182 | static int earliest_time_prioq_compare(const void *a, const void *b) { |
fd38203a LP |
183 | const sd_event_source *x = a, *y = b; |
184 | ||
6a0f1f6d LP |
185 | assert(EVENT_SOURCE_IS_TIME(x->type)); |
186 | assert(x->type == y->type); | |
fd38203a | 187 | |
baf76283 LP |
188 | /* Enabled ones first */ |
189 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
fd38203a | 190 | return -1; |
baf76283 | 191 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) |
fd38203a LP |
192 | return 1; |
193 | ||
194 | /* Move the pending ones to the end */ | |
195 | if (!x->pending && y->pending) | |
196 | return -1; | |
197 | if (x->pending && !y->pending) | |
198 | return 1; | |
199 | ||
200 | /* Order by time */ | |
9c57a73b | 201 | return CMP(x->time.next, y->time.next); |
fd38203a LP |
202 | } |
203 | ||
1bce0ffa LP |
204 | static usec_t time_event_source_latest(const sd_event_source *s) { |
205 | return usec_add(s->time.next, s->time.accuracy); | |
206 | } | |
207 | ||
c2ba3ad6 LP |
208 | static int latest_time_prioq_compare(const void *a, const void *b) { |
209 | const sd_event_source *x = a, *y = b; | |
210 | ||
6a0f1f6d LP |
211 | assert(EVENT_SOURCE_IS_TIME(x->type)); |
212 | assert(x->type == y->type); | |
c2ba3ad6 | 213 | |
baf76283 LP |
214 | /* Enabled ones first */ |
215 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
c2ba3ad6 | 216 | return -1; |
baf76283 | 217 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) |
c2ba3ad6 LP |
218 | return 1; |
219 | ||
220 | /* Move the pending ones to the end */ | |
221 | if (!x->pending && y->pending) | |
222 | return -1; | |
223 | if (x->pending && !y->pending) | |
224 | return 1; | |
225 | ||
226 | /* Order by time */ | |
9c57a73b | 227 | return CMP(time_event_source_latest(x), time_event_source_latest(y)); |
c2ba3ad6 LP |
228 | } |
229 | ||
6203e07a | 230 | static int exit_prioq_compare(const void *a, const void *b) { |
da7e457c LP |
231 | const sd_event_source *x = a, *y = b; |
232 | ||
6203e07a LP |
233 | assert(x->type == SOURCE_EXIT); |
234 | assert(y->type == SOURCE_EXIT); | |
da7e457c | 235 | |
baf76283 LP |
236 | /* Enabled ones first */ |
237 | if (x->enabled != SD_EVENT_OFF && y->enabled == SD_EVENT_OFF) | |
da7e457c | 238 | return -1; |
baf76283 | 239 | if (x->enabled == SD_EVENT_OFF && y->enabled != SD_EVENT_OFF) |
da7e457c LP |
240 | return 1; |
241 | ||
242 | /* Lower priority values first */ | |
6dd91b36 | 243 | return CMP(x->priority, y->priority); |
da7e457c LP |
244 | } |
245 | ||
6a0f1f6d LP |
246 | static void free_clock_data(struct clock_data *d) { |
247 | assert(d); | |
9da4cb2b | 248 | assert(d->wakeup == WAKEUP_CLOCK_DATA); |
6a0f1f6d LP |
249 | |
250 | safe_close(d->fd); | |
251 | prioq_free(d->earliest); | |
252 | prioq_free(d->latest); | |
253 | } | |
254 | ||
8301aa0b | 255 | static sd_event *event_free(sd_event *e) { |
a71fe8b8 LP |
256 | sd_event_source *s; |
257 | ||
fd38203a | 258 | assert(e); |
a71fe8b8 LP |
259 | |
260 | while ((s = e->sources)) { | |
261 | assert(s->floating); | |
262 | source_disconnect(s); | |
263 | sd_event_source_unref(s); | |
264 | } | |
265 | ||
15b38f93 | 266 | assert(e->n_sources == 0); |
fd38203a | 267 | |
afc6adb5 LP |
268 | if (e->default_event_ptr) |
269 | *(e->default_event_ptr) = NULL; | |
270 | ||
03e334a1 | 271 | safe_close(e->epoll_fd); |
03e334a1 | 272 | safe_close(e->watchdog_fd); |
cde93897 | 273 | |
6a0f1f6d | 274 | free_clock_data(&e->realtime); |
a8548816 | 275 | free_clock_data(&e->boottime); |
6a0f1f6d LP |
276 | free_clock_data(&e->monotonic); |
277 | free_clock_data(&e->realtime_alarm); | |
278 | free_clock_data(&e->boottime_alarm); | |
279 | ||
fd38203a LP |
280 | prioq_free(e->pending); |
281 | prioq_free(e->prepare); | |
6203e07a | 282 | prioq_free(e->exit); |
fd38203a LP |
283 | |
284 | free(e->signal_sources); | |
9da4cb2b | 285 | hashmap_free(e->signal_data); |
fd38203a | 286 | |
97ef5391 LP |
287 | hashmap_free(e->inotify_data); |
288 | ||
fd38203a | 289 | hashmap_free(e->child_sources); |
6e9feda3 | 290 | set_free(e->post_sources); |
8301aa0b | 291 | |
5cddd924 LP |
292 | free(e->event_queue); |
293 | ||
8301aa0b | 294 | return mfree(e); |
fd38203a LP |
295 | } |
296 | ||
f7262a9f | 297 | _public_ int sd_event_new(sd_event** ret) { |
fd38203a LP |
298 | sd_event *e; |
299 | int r; | |
300 | ||
305f78bf | 301 | assert_return(ret, -EINVAL); |
fd38203a | 302 | |
d08eb1fa | 303 | e = new(sd_event, 1); |
fd38203a LP |
304 | if (!e) |
305 | return -ENOMEM; | |
306 | ||
d08eb1fa LP |
307 | *e = (sd_event) { |
308 | .n_ref = 1, | |
309 | .epoll_fd = -1, | |
310 | .watchdog_fd = -1, | |
311 | .realtime.wakeup = WAKEUP_CLOCK_DATA, | |
312 | .realtime.fd = -1, | |
313 | .realtime.next = USEC_INFINITY, | |
314 | .boottime.wakeup = WAKEUP_CLOCK_DATA, | |
315 | .boottime.fd = -1, | |
316 | .boottime.next = USEC_INFINITY, | |
317 | .monotonic.wakeup = WAKEUP_CLOCK_DATA, | |
318 | .monotonic.fd = -1, | |
319 | .monotonic.next = USEC_INFINITY, | |
320 | .realtime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
321 | .realtime_alarm.fd = -1, | |
322 | .realtime_alarm.next = USEC_INFINITY, | |
323 | .boottime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
324 | .boottime_alarm.fd = -1, | |
325 | .boottime_alarm.next = USEC_INFINITY, | |
326 | .perturb = USEC_INFINITY, | |
327 | .original_pid = getpid_cached(), | |
328 | }; | |
fd38203a | 329 | |
c983e776 EV |
330 | r = prioq_ensure_allocated(&e->pending, pending_prioq_compare); |
331 | if (r < 0) | |
fd38203a | 332 | goto fail; |
fd38203a LP |
333 | |
334 | e->epoll_fd = epoll_create1(EPOLL_CLOEXEC); | |
335 | if (e->epoll_fd < 0) { | |
336 | r = -errno; | |
337 | goto fail; | |
338 | } | |
339 | ||
7fe2903c LP |
340 | e->epoll_fd = fd_move_above_stdio(e->epoll_fd); |
341 | ||
34b87517 | 342 | if (secure_getenv("SD_EVENT_PROFILE_DELAYS")) { |
34a6843d | 343 | log_debug("Event loop profiling enabled. Logarithmic histogram of event loop iterations in the range 2^0 ... 2^63 us will be logged every 5s."); |
34b87517 VC |
344 | e->profile_delays = true; |
345 | } | |
346 | ||
fd38203a LP |
347 | *ret = e; |
348 | return 0; | |
349 | ||
350 | fail: | |
351 | event_free(e); | |
352 | return r; | |
353 | } | |
354 | ||
8301aa0b | 355 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event, sd_event, event_free); |
fd38203a | 356 | |
afd15bbb ZJS |
357 | _public_ sd_event_source* sd_event_source_disable_unref(sd_event_source *s) { |
358 | if (s) | |
359 | (void) sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
360 | return sd_event_source_unref(s); | |
361 | } | |
362 | ||
eaa3cbef LP |
363 | static bool event_pid_changed(sd_event *e) { |
364 | assert(e); | |
365 | ||
a2360a46 | 366 | /* We don't support people creating an event loop and keeping |
eaa3cbef LP |
367 | * it around over a fork(). Let's complain. */ |
368 | ||
df0ff127 | 369 | return e->original_pid != getpid_cached(); |
eaa3cbef LP |
370 | } |
371 | ||
366e6411 | 372 | static void source_io_unregister(sd_event_source *s) { |
fd38203a LP |
373 | assert(s); |
374 | assert(s->type == SOURCE_IO); | |
375 | ||
f6806734 | 376 | if (event_pid_changed(s->event)) |
366e6411 | 377 | return; |
f6806734 | 378 | |
fd38203a | 379 | if (!s->io.registered) |
366e6411 | 380 | return; |
fd38203a | 381 | |
d1cf2023 | 382 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->io.fd, NULL) < 0) |
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)); | |
721 | prioq_reshuffle(d->earliest, s, &s->time.earliest_index); | |
722 | prioq_reshuffle(d->latest, s, &s->time.latest_index); | |
723 | d->needs_rearm = true; | |
724 | } | |
725 | ||
a71fe8b8 LP |
726 | static void source_disconnect(sd_event_source *s) { |
727 | sd_event *event; | |
728 | ||
fd38203a LP |
729 | assert(s); |
730 | ||
a71fe8b8 LP |
731 | if (!s->event) |
732 | return; | |
15b38f93 | 733 | |
a71fe8b8 | 734 | assert(s->event->n_sources > 0); |
fd38203a | 735 | |
a71fe8b8 | 736 | switch (s->type) { |
fd38203a | 737 | |
a71fe8b8 LP |
738 | case SOURCE_IO: |
739 | if (s->io.fd >= 0) | |
740 | source_io_unregister(s); | |
fd38203a | 741 | |
a71fe8b8 | 742 | break; |
6a0f1f6d | 743 | |
a71fe8b8 | 744 | case SOURCE_TIME_REALTIME: |
a8548816 | 745 | case SOURCE_TIME_BOOTTIME: |
a71fe8b8 LP |
746 | case SOURCE_TIME_MONOTONIC: |
747 | case SOURCE_TIME_REALTIME_ALARM: | |
748 | case SOURCE_TIME_BOOTTIME_ALARM: { | |
749 | struct clock_data *d; | |
fd38203a | 750 | |
a71fe8b8 LP |
751 | d = event_get_clock_data(s->event, s->type); |
752 | assert(d); | |
753 | ||
754 | prioq_remove(d->earliest, s, &s->time.earliest_index); | |
755 | prioq_remove(d->latest, s, &s->time.latest_index); | |
212bbb17 | 756 | d->needs_rearm = true; |
a71fe8b8 LP |
757 | break; |
758 | } | |
759 | ||
760 | case SOURCE_SIGNAL: | |
761 | if (s->signal.sig > 0) { | |
9da4cb2b | 762 | |
a71fe8b8 LP |
763 | if (s->event->signal_sources) |
764 | s->event->signal_sources[s->signal.sig] = NULL; | |
4807d2d0 | 765 | |
9da4cb2b | 766 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
6a0f1f6d | 767 | } |
fd38203a | 768 | |
a71fe8b8 | 769 | break; |
fd38203a | 770 | |
a71fe8b8 LP |
771 | case SOURCE_CHILD: |
772 | if (s->child.pid > 0) { | |
773 | if (s->enabled != SD_EVENT_OFF) { | |
774 | assert(s->event->n_enabled_child_sources > 0); | |
775 | s->event->n_enabled_child_sources--; | |
4807d2d0 | 776 | } |
fd38203a | 777 | |
4a0b58c4 | 778 | (void) hashmap_remove(s->event->child_sources, PID_TO_PTR(s->child.pid)); |
a71fe8b8 | 779 | } |
fd38203a | 780 | |
f8f3f926 LP |
781 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
782 | source_child_pidfd_unregister(s); | |
783 | else | |
784 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
785 | ||
a71fe8b8 | 786 | break; |
fd38203a | 787 | |
a71fe8b8 LP |
788 | case SOURCE_DEFER: |
789 | /* nothing */ | |
790 | break; | |
fd38203a | 791 | |
a71fe8b8 LP |
792 | case SOURCE_POST: |
793 | set_remove(s->event->post_sources, s); | |
794 | break; | |
da7e457c | 795 | |
a71fe8b8 LP |
796 | case SOURCE_EXIT: |
797 | prioq_remove(s->event->exit, s, &s->exit.prioq_index); | |
798 | break; | |
0eb2e0e3 | 799 | |
97ef5391 LP |
800 | case SOURCE_INOTIFY: { |
801 | struct inode_data *inode_data; | |
802 | ||
803 | inode_data = s->inotify.inode_data; | |
804 | if (inode_data) { | |
805 | struct inotify_data *inotify_data; | |
806 | assert_se(inotify_data = inode_data->inotify_data); | |
807 | ||
808 | /* Detach this event source from the inode object */ | |
809 | LIST_REMOVE(inotify.by_inode_data, inode_data->event_sources, s); | |
810 | s->inotify.inode_data = NULL; | |
811 | ||
812 | if (s->pending) { | |
813 | assert(inotify_data->n_pending > 0); | |
814 | inotify_data->n_pending--; | |
815 | } | |
816 | ||
817 | /* Note that we don't reduce the inotify mask for the watch descriptor here if the inode is | |
818 | * continued to being watched. That's because inotify doesn't really have an API for that: we | |
819 | * can only change watch masks with access to the original inode either by fd or by path. But | |
820 | * paths aren't stable, and keeping an O_PATH fd open all the time would mean wasting an fd | |
f21f31b2 | 821 | * continuously and keeping the mount busy which we can't really do. We could reconstruct the |
97ef5391 LP |
822 | * original inode from /proc/self/fdinfo/$INOTIFY_FD (as all watch descriptors are listed |
823 | * there), but given the need for open_by_handle_at() which is privileged and not universally | |
824 | * available this would be quite an incomplete solution. Hence we go the other way, leave the | |
825 | * mask set, even if it is not minimized now, and ignore all events we aren't interested in | |
826 | * anymore after reception. Yes, this sucks, but … Linux … */ | |
827 | ||
828 | /* Maybe release the inode data (and its inotify) */ | |
829 | event_gc_inode_data(s->event, inode_data); | |
830 | } | |
831 | ||
832 | break; | |
833 | } | |
834 | ||
a71fe8b8 LP |
835 | default: |
836 | assert_not_reached("Wut? I shouldn't exist."); | |
837 | } | |
6e9feda3 | 838 | |
a71fe8b8 LP |
839 | if (s->pending) |
840 | prioq_remove(s->event->pending, s, &s->pending_index); | |
9d3e3aa5 | 841 | |
a71fe8b8 LP |
842 | if (s->prepare) |
843 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
fd38203a | 844 | |
e514aa1e | 845 | event = TAKE_PTR(s->event); |
a71fe8b8 LP |
846 | LIST_REMOVE(sources, event->sources, s); |
847 | event->n_sources--; | |
fd38203a | 848 | |
f5982559 LP |
849 | /* Note that we don't invalidate the type here, since we still need it in order to close the fd or |
850 | * pidfd associated with this event source, which we'll do only on source_free(). */ | |
851 | ||
a71fe8b8 LP |
852 | if (!s->floating) |
853 | sd_event_unref(event); | |
854 | } | |
855 | ||
856 | static void source_free(sd_event_source *s) { | |
857 | assert(s); | |
fd38203a | 858 | |
a71fe8b8 | 859 | source_disconnect(s); |
ab93297c NM |
860 | |
861 | if (s->type == SOURCE_IO && s->io.owned) | |
15723a1d LP |
862 | s->io.fd = safe_close(s->io.fd); |
863 | ||
f8f3f926 LP |
864 | if (s->type == SOURCE_CHILD) { |
865 | /* Eventually the kernel will do this automatically for us, but for now let's emulate this (unreliably) in userspace. */ | |
866 | ||
867 | if (s->child.process_owned) { | |
868 | ||
869 | if (!s->child.exited) { | |
870 | bool sent = false; | |
871 | ||
872 | if (s->child.pidfd >= 0) { | |
873 | if (pidfd_send_signal(s->child.pidfd, SIGKILL, NULL, 0) < 0) { | |
874 | if (errno == ESRCH) /* Already dead */ | |
875 | sent = true; | |
876 | else if (!ERRNO_IS_NOT_SUPPORTED(errno)) | |
877 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via pidfd_send_signal(), re-trying via kill(): %m", | |
878 | s->child.pid); | |
879 | } else | |
880 | sent = true; | |
881 | } | |
882 | ||
883 | if (!sent) | |
884 | if (kill(s->child.pid, SIGKILL) < 0) | |
885 | if (errno != ESRCH) /* Already dead */ | |
886 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via kill(), ignoring: %m", | |
887 | s->child.pid); | |
888 | } | |
889 | ||
890 | if (!s->child.waited) { | |
891 | siginfo_t si = {}; | |
892 | ||
893 | /* Reap the child if we can */ | |
894 | (void) waitid(P_PID, s->child.pid, &si, WEXITED); | |
895 | } | |
896 | } | |
897 | ||
898 | if (s->child.pidfd_owned) | |
899 | s->child.pidfd = safe_close(s->child.pidfd); | |
900 | } | |
901 | ||
15723a1d LP |
902 | if (s->destroy_callback) |
903 | s->destroy_callback(s->userdata); | |
ab93297c | 904 | |
356779df | 905 | free(s->description); |
fd38203a LP |
906 | free(s); |
907 | } | |
8c75fe17 | 908 | DEFINE_TRIVIAL_CLEANUP_FUNC(sd_event_source*, source_free); |
fd38203a LP |
909 | |
910 | static int source_set_pending(sd_event_source *s, bool b) { | |
911 | int r; | |
912 | ||
913 | assert(s); | |
6203e07a | 914 | assert(s->type != SOURCE_EXIT); |
fd38203a LP |
915 | |
916 | if (s->pending == b) | |
917 | return 0; | |
918 | ||
919 | s->pending = b; | |
920 | ||
921 | if (b) { | |
922 | s->pending_iteration = s->event->iteration; | |
923 | ||
924 | r = prioq_put(s->event->pending, s, &s->pending_index); | |
925 | if (r < 0) { | |
926 | s->pending = false; | |
927 | return r; | |
928 | } | |
929 | } else | |
930 | assert_se(prioq_remove(s->event->pending, s, &s->pending_index)); | |
931 | ||
e1951c16 MS |
932 | if (EVENT_SOURCE_IS_TIME(s->type)) |
933 | event_source_time_prioq_reshuffle(s); | |
2576a19e | 934 | |
9da4cb2b LP |
935 | if (s->type == SOURCE_SIGNAL && !b) { |
936 | struct signal_data *d; | |
937 | ||
938 | d = hashmap_get(s->event->signal_data, &s->priority); | |
939 | if (d && d->current == s) | |
940 | d->current = NULL; | |
941 | } | |
942 | ||
97ef5391 LP |
943 | if (s->type == SOURCE_INOTIFY) { |
944 | ||
945 | assert(s->inotify.inode_data); | |
946 | assert(s->inotify.inode_data->inotify_data); | |
947 | ||
948 | if (b) | |
949 | s->inotify.inode_data->inotify_data->n_pending ++; | |
950 | else { | |
951 | assert(s->inotify.inode_data->inotify_data->n_pending > 0); | |
952 | s->inotify.inode_data->inotify_data->n_pending --; | |
953 | } | |
954 | } | |
955 | ||
fd38203a LP |
956 | return 0; |
957 | } | |
958 | ||
a71fe8b8 | 959 | static sd_event_source *source_new(sd_event *e, bool floating, EventSourceType type) { |
fd38203a LP |
960 | sd_event_source *s; |
961 | ||
962 | assert(e); | |
963 | ||
d08eb1fa | 964 | s = new(sd_event_source, 1); |
fd38203a LP |
965 | if (!s) |
966 | return NULL; | |
967 | ||
d08eb1fa LP |
968 | *s = (struct sd_event_source) { |
969 | .n_ref = 1, | |
970 | .event = e, | |
971 | .floating = floating, | |
972 | .type = type, | |
973 | .pending_index = PRIOQ_IDX_NULL, | |
974 | .prepare_index = PRIOQ_IDX_NULL, | |
975 | }; | |
a71fe8b8 LP |
976 | |
977 | if (!floating) | |
978 | sd_event_ref(e); | |
fd38203a | 979 | |
a71fe8b8 | 980 | LIST_PREPEND(sources, e->sources, s); |
313cefa1 | 981 | e->n_sources++; |
15b38f93 | 982 | |
fd38203a LP |
983 | return s; |
984 | } | |
985 | ||
b9350e70 LP |
986 | static int io_exit_callback(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
987 | assert(s); | |
988 | ||
989 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
990 | } | |
991 | ||
f7262a9f | 992 | _public_ int sd_event_add_io( |
fd38203a | 993 | sd_event *e, |
151b9b96 | 994 | sd_event_source **ret, |
fd38203a LP |
995 | int fd, |
996 | uint32_t events, | |
718db961 | 997 | sd_event_io_handler_t callback, |
151b9b96 | 998 | void *userdata) { |
fd38203a | 999 | |
ec766a51 | 1000 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1001 | int r; |
1002 | ||
305f78bf | 1003 | assert_return(e, -EINVAL); |
b937d761 | 1004 | assert_return(e = event_resolve(e), -ENOPKG); |
8ac43fee | 1005 | assert_return(fd >= 0, -EBADF); |
2a16a986 | 1006 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 1007 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1008 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1009 | |
b9350e70 LP |
1010 | if (!callback) |
1011 | callback = io_exit_callback; | |
1012 | ||
a71fe8b8 | 1013 | s = source_new(e, !ret, SOURCE_IO); |
fd38203a LP |
1014 | if (!s) |
1015 | return -ENOMEM; | |
1016 | ||
9da4cb2b | 1017 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1018 | s->io.fd = fd; |
1019 | s->io.events = events; | |
1020 | s->io.callback = callback; | |
1021 | s->userdata = userdata; | |
baf76283 | 1022 | s->enabled = SD_EVENT_ON; |
fd38203a | 1023 | |
baf76283 | 1024 | r = source_io_register(s, s->enabled, events); |
ec766a51 | 1025 | if (r < 0) |
050f74f2 | 1026 | return r; |
fd38203a | 1027 | |
a71fe8b8 LP |
1028 | if (ret) |
1029 | *ret = s; | |
ec766a51 | 1030 | TAKE_PTR(s); |
a71fe8b8 | 1031 | |
fd38203a LP |
1032 | return 0; |
1033 | } | |
1034 | ||
52444dc4 LP |
1035 | static void initialize_perturb(sd_event *e) { |
1036 | sd_id128_t bootid = {}; | |
1037 | ||
1038 | /* When we sleep for longer, we try to realign the wakeup to | |
f21f31b2 | 1039 | the same time within each minute/second/250ms, so that |
52444dc4 LP |
1040 | events all across the system can be coalesced into a single |
1041 | CPU wakeup. However, let's take some system-specific | |
1042 | randomness for this value, so that in a network of systems | |
1043 | with synced clocks timer events are distributed a | |
1044 | bit. Here, we calculate a perturbation usec offset from the | |
1045 | boot ID. */ | |
1046 | ||
3a43da28 | 1047 | if (_likely_(e->perturb != USEC_INFINITY)) |
52444dc4 LP |
1048 | return; |
1049 | ||
1050 | if (sd_id128_get_boot(&bootid) >= 0) | |
1051 | e->perturb = (bootid.qwords[0] ^ bootid.qwords[1]) % USEC_PER_MINUTE; | |
1052 | } | |
1053 | ||
fd38203a LP |
1054 | static int event_setup_timer_fd( |
1055 | sd_event *e, | |
6a0f1f6d LP |
1056 | struct clock_data *d, |
1057 | clockid_t clock) { | |
fd38203a | 1058 | |
fd38203a | 1059 | assert(e); |
6a0f1f6d | 1060 | assert(d); |
fd38203a | 1061 | |
6a0f1f6d | 1062 | if (_likely_(d->fd >= 0)) |
fd38203a LP |
1063 | return 0; |
1064 | ||
b44d87e2 | 1065 | _cleanup_close_ int fd = -1; |
b44d87e2 | 1066 | |
6a0f1f6d | 1067 | fd = timerfd_create(clock, TFD_NONBLOCK|TFD_CLOEXEC); |
fd38203a LP |
1068 | if (fd < 0) |
1069 | return -errno; | |
1070 | ||
7fe2903c LP |
1071 | fd = fd_move_above_stdio(fd); |
1072 | ||
1eac7948 | 1073 | struct epoll_event ev = { |
a82f89aa LP |
1074 | .events = EPOLLIN, |
1075 | .data.ptr = d, | |
1076 | }; | |
fd38203a | 1077 | |
15c689d7 | 1078 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) |
fd38203a | 1079 | return -errno; |
fd38203a | 1080 | |
b44d87e2 | 1081 | d->fd = TAKE_FD(fd); |
fd38203a LP |
1082 | return 0; |
1083 | } | |
1084 | ||
c4f1aff2 TG |
1085 | static int time_exit_callback(sd_event_source *s, uint64_t usec, void *userdata) { |
1086 | assert(s); | |
1087 | ||
1088 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1089 | } | |
1090 | ||
41c63f36 LP |
1091 | static int setup_clock_data(sd_event *e, struct clock_data *d, clockid_t clock) { |
1092 | int r; | |
1093 | ||
1094 | assert(d); | |
1095 | ||
1096 | if (d->fd < 0) { | |
1097 | r = event_setup_timer_fd(e, d, clock); | |
1098 | if (r < 0) | |
1099 | return r; | |
1100 | } | |
1101 | ||
1102 | r = prioq_ensure_allocated(&d->earliest, earliest_time_prioq_compare); | |
1103 | if (r < 0) | |
1104 | return r; | |
1105 | ||
1106 | r = prioq_ensure_allocated(&d->latest, latest_time_prioq_compare); | |
1107 | if (r < 0) | |
1108 | return r; | |
1109 | ||
1110 | return 0; | |
1111 | } | |
1112 | ||
6a0f1f6d | 1113 | _public_ int sd_event_add_time( |
fd38203a | 1114 | sd_event *e, |
151b9b96 | 1115 | sd_event_source **ret, |
6a0f1f6d | 1116 | clockid_t clock, |
fd38203a | 1117 | uint64_t usec, |
c2ba3ad6 | 1118 | uint64_t accuracy, |
718db961 | 1119 | sd_event_time_handler_t callback, |
151b9b96 | 1120 | void *userdata) { |
fd38203a | 1121 | |
6a0f1f6d | 1122 | EventSourceType type; |
ec766a51 | 1123 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6a0f1f6d | 1124 | struct clock_data *d; |
fd38203a LP |
1125 | int r; |
1126 | ||
305f78bf | 1127 | assert_return(e, -EINVAL); |
b937d761 | 1128 | assert_return(e = event_resolve(e), -ENOPKG); |
305f78bf | 1129 | assert_return(accuracy != (uint64_t) -1, -EINVAL); |
da7e457c | 1130 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1131 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1132 | |
e475d10c LP |
1133 | if (!clock_supported(clock)) /* Checks whether the kernel supports the clock */ |
1134 | return -EOPNOTSUPP; | |
1135 | ||
1136 | type = clock_to_event_source_type(clock); /* checks whether sd-event supports this clock */ | |
1137 | if (type < 0) | |
3411372e LP |
1138 | return -EOPNOTSUPP; |
1139 | ||
c4f1aff2 TG |
1140 | if (!callback) |
1141 | callback = time_exit_callback; | |
1142 | ||
6a0f1f6d LP |
1143 | d = event_get_clock_data(e, type); |
1144 | assert(d); | |
c2ba3ad6 | 1145 | |
41c63f36 | 1146 | r = setup_clock_data(e, d, clock); |
c983e776 EV |
1147 | if (r < 0) |
1148 | return r; | |
fd38203a | 1149 | |
a71fe8b8 | 1150 | s = source_new(e, !ret, type); |
fd38203a LP |
1151 | if (!s) |
1152 | return -ENOMEM; | |
1153 | ||
1154 | s->time.next = usec; | |
c2ba3ad6 | 1155 | s->time.accuracy = accuracy == 0 ? DEFAULT_ACCURACY_USEC : accuracy; |
fd38203a | 1156 | s->time.callback = callback; |
da7e457c | 1157 | s->time.earliest_index = s->time.latest_index = PRIOQ_IDX_NULL; |
fd38203a | 1158 | s->userdata = userdata; |
baf76283 | 1159 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1160 | |
e07bbb7c TG |
1161 | d->needs_rearm = true; |
1162 | ||
6a0f1f6d | 1163 | r = prioq_put(d->earliest, s, &s->time.earliest_index); |
c2ba3ad6 | 1164 | if (r < 0) |
ec766a51 | 1165 | return r; |
c2ba3ad6 | 1166 | |
6a0f1f6d | 1167 | r = prioq_put(d->latest, s, &s->time.latest_index); |
c2ba3ad6 | 1168 | if (r < 0) |
ec766a51 | 1169 | return r; |
fd38203a | 1170 | |
a71fe8b8 LP |
1171 | if (ret) |
1172 | *ret = s; | |
ec766a51 | 1173 | TAKE_PTR(s); |
a71fe8b8 | 1174 | |
fd38203a LP |
1175 | return 0; |
1176 | } | |
1177 | ||
d6a83dc4 LP |
1178 | _public_ int sd_event_add_time_relative( |
1179 | sd_event *e, | |
1180 | sd_event_source **ret, | |
1181 | clockid_t clock, | |
1182 | uint64_t usec, | |
1183 | uint64_t accuracy, | |
1184 | sd_event_time_handler_t callback, | |
1185 | void *userdata) { | |
1186 | ||
1187 | usec_t t; | |
1188 | int r; | |
1189 | ||
1190 | /* Same as sd_event_add_time() but operates relative to the event loop's current point in time, and | |
1191 | * checks for overflow. */ | |
1192 | ||
1193 | r = sd_event_now(e, clock, &t); | |
1194 | if (r < 0) | |
1195 | return r; | |
1196 | ||
1197 | if (usec >= USEC_INFINITY - t) | |
1198 | return -EOVERFLOW; | |
1199 | ||
1200 | return sd_event_add_time(e, ret, clock, t + usec, accuracy, callback, userdata); | |
1201 | } | |
1202 | ||
59bc1fd7 LP |
1203 | static int signal_exit_callback(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) { |
1204 | assert(s); | |
1205 | ||
1206 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1207 | } | |
1208 | ||
f7262a9f | 1209 | _public_ int sd_event_add_signal( |
305f78bf | 1210 | sd_event *e, |
151b9b96 | 1211 | sd_event_source **ret, |
305f78bf | 1212 | int sig, |
718db961 | 1213 | sd_event_signal_handler_t callback, |
151b9b96 | 1214 | void *userdata) { |
305f78bf | 1215 | |
ec766a51 | 1216 | _cleanup_(source_freep) sd_event_source *s = NULL; |
9da4cb2b | 1217 | struct signal_data *d; |
fd38203a LP |
1218 | int r; |
1219 | ||
305f78bf | 1220 | assert_return(e, -EINVAL); |
b937d761 | 1221 | assert_return(e = event_resolve(e), -ENOPKG); |
6eb7c172 | 1222 | assert_return(SIGNAL_VALID(sig), -EINVAL); |
da7e457c | 1223 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1224 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1225 | |
59bc1fd7 LP |
1226 | if (!callback) |
1227 | callback = signal_exit_callback; | |
1228 | ||
d1b75241 LP |
1229 | r = signal_is_blocked(sig); |
1230 | if (r < 0) | |
1231 | return r; | |
1232 | if (r == 0) | |
3022d74b LP |
1233 | return -EBUSY; |
1234 | ||
fd38203a LP |
1235 | if (!e->signal_sources) { |
1236 | e->signal_sources = new0(sd_event_source*, _NSIG); | |
1237 | if (!e->signal_sources) | |
1238 | return -ENOMEM; | |
1239 | } else if (e->signal_sources[sig]) | |
1240 | return -EBUSY; | |
1241 | ||
a71fe8b8 | 1242 | s = source_new(e, !ret, SOURCE_SIGNAL); |
fd38203a LP |
1243 | if (!s) |
1244 | return -ENOMEM; | |
1245 | ||
1246 | s->signal.sig = sig; | |
1247 | s->signal.callback = callback; | |
1248 | s->userdata = userdata; | |
baf76283 | 1249 | s->enabled = SD_EVENT_ON; |
fd38203a LP |
1250 | |
1251 | e->signal_sources[sig] = s; | |
fd38203a | 1252 | |
9da4cb2b | 1253 | r = event_make_signal_data(e, sig, &d); |
ec766a51 | 1254 | if (r < 0) |
9da4cb2b | 1255 | return r; |
fd38203a | 1256 | |
f1f00dbb LP |
1257 | /* Use the signal name as description for the event source by default */ |
1258 | (void) sd_event_source_set_description(s, signal_to_string(sig)); | |
1259 | ||
a71fe8b8 LP |
1260 | if (ret) |
1261 | *ret = s; | |
ec766a51 | 1262 | TAKE_PTR(s); |
a71fe8b8 | 1263 | |
fd38203a LP |
1264 | return 0; |
1265 | } | |
1266 | ||
b9350e70 LP |
1267 | static int child_exit_callback(sd_event_source *s, const siginfo_t *si, void *userdata) { |
1268 | assert(s); | |
1269 | ||
1270 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1271 | } | |
1272 | ||
f8f3f926 LP |
1273 | static bool shall_use_pidfd(void) { |
1274 | /* Mostly relevant for debugging, i.e. this is used in test-event.c to test the event loop once with and once without pidfd */ | |
1275 | return getenv_bool_secure("SYSTEMD_PIDFD") != 0; | |
1276 | } | |
1277 | ||
f7262a9f | 1278 | _public_ int sd_event_add_child( |
305f78bf | 1279 | sd_event *e, |
151b9b96 | 1280 | sd_event_source **ret, |
305f78bf LP |
1281 | pid_t pid, |
1282 | int options, | |
718db961 | 1283 | sd_event_child_handler_t callback, |
151b9b96 | 1284 | void *userdata) { |
305f78bf | 1285 | |
ec766a51 | 1286 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1287 | int r; |
1288 | ||
305f78bf | 1289 | assert_return(e, -EINVAL); |
b937d761 | 1290 | assert_return(e = event_resolve(e), -ENOPKG); |
305f78bf LP |
1291 | assert_return(pid > 1, -EINVAL); |
1292 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1293 | assert_return(options != 0, -EINVAL); | |
da7e457c | 1294 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1295 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1296 | |
b9350e70 LP |
1297 | if (!callback) |
1298 | callback = child_exit_callback; | |
1299 | ||
ee880b37 LP |
1300 | if (e->n_enabled_child_sources == 0) { |
1301 | /* Caller must block SIGCHLD before using us to watch children, even if pidfd is available, | |
1302 | * for compatibility with pre-pidfd and because we don't want the reap the child processes | |
1303 | * ourselves, i.e. call waitid(), and don't want Linux' default internal logic for that to | |
1304 | * take effect. | |
1305 | * | |
1306 | * (As an optimization we only do this check on the first child event source created.) */ | |
1307 | r = signal_is_blocked(SIGCHLD); | |
1308 | if (r < 0) | |
1309 | return r; | |
1310 | if (r == 0) | |
1311 | return -EBUSY; | |
1312 | } | |
1313 | ||
d5099efc | 1314 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
fd38203a LP |
1315 | if (r < 0) |
1316 | return r; | |
1317 | ||
4a0b58c4 | 1318 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) |
fd38203a LP |
1319 | return -EBUSY; |
1320 | ||
a71fe8b8 | 1321 | s = source_new(e, !ret, SOURCE_CHILD); |
fd38203a LP |
1322 | if (!s) |
1323 | return -ENOMEM; | |
1324 | ||
f8f3f926 | 1325 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1326 | s->child.pid = pid; |
1327 | s->child.options = options; | |
1328 | s->child.callback = callback; | |
1329 | s->userdata = userdata; | |
baf76283 | 1330 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1331 | |
f8f3f926 LP |
1332 | /* We always take a pidfd here if we can, even if we wait for anything else than WEXITED, so that we |
1333 | * pin the PID, and make regular waitid() handling race-free. */ | |
1334 | ||
1335 | if (shall_use_pidfd()) { | |
1336 | s->child.pidfd = pidfd_open(s->child.pid, 0); | |
1337 | if (s->child.pidfd < 0) { | |
1338 | /* Propagate errors unless the syscall is not supported or blocked */ | |
1339 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
1340 | return -errno; | |
1341 | } else | |
1342 | s->child.pidfd_owned = true; /* If we allocate the pidfd we own it by default */ | |
1343 | } else | |
1344 | s->child.pidfd = -1; | |
1345 | ||
4a0b58c4 | 1346 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); |
ec766a51 | 1347 | if (r < 0) |
fd38203a | 1348 | return r; |
fd38203a | 1349 | |
f8f3f926 LP |
1350 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1351 | /* We have a pidfd and we only want to watch for exit */ | |
f8f3f926 | 1352 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1353 | if (r < 0) |
f8f3f926 | 1354 | return r; |
ac9f2640 | 1355 | |
f8f3f926 LP |
1356 | } else { |
1357 | /* We have no pidfd or we shall wait for some other event than WEXITED */ | |
f8f3f926 | 1358 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1359 | if (r < 0) |
f8f3f926 | 1360 | return r; |
f8f3f926 LP |
1361 | |
1362 | e->need_process_child = true; | |
1363 | } | |
c2ba3ad6 | 1364 | |
ac9f2640 ZJS |
1365 | e->n_enabled_child_sources++; |
1366 | ||
a71fe8b8 LP |
1367 | if (ret) |
1368 | *ret = s; | |
ec766a51 | 1369 | TAKE_PTR(s); |
f8f3f926 LP |
1370 | return 0; |
1371 | } | |
1372 | ||
1373 | _public_ int sd_event_add_child_pidfd( | |
1374 | sd_event *e, | |
1375 | sd_event_source **ret, | |
1376 | int pidfd, | |
1377 | int options, | |
1378 | sd_event_child_handler_t callback, | |
1379 | void *userdata) { | |
1380 | ||
1381 | ||
1382 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
1383 | pid_t pid; | |
1384 | int r; | |
1385 | ||
1386 | assert_return(e, -EINVAL); | |
1387 | assert_return(e = event_resolve(e), -ENOPKG); | |
1388 | assert_return(pidfd >= 0, -EBADF); | |
1389 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1390 | assert_return(options != 0, -EINVAL); | |
f8f3f926 LP |
1391 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1392 | assert_return(!event_pid_changed(e), -ECHILD); | |
1393 | ||
b9350e70 LP |
1394 | if (!callback) |
1395 | callback = child_exit_callback; | |
1396 | ||
ee880b37 LP |
1397 | if (e->n_enabled_child_sources == 0) { |
1398 | r = signal_is_blocked(SIGCHLD); | |
1399 | if (r < 0) | |
1400 | return r; | |
1401 | if (r == 0) | |
1402 | return -EBUSY; | |
1403 | } | |
1404 | ||
f8f3f926 LP |
1405 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
1406 | if (r < 0) | |
1407 | return r; | |
1408 | ||
1409 | r = pidfd_get_pid(pidfd, &pid); | |
1410 | if (r < 0) | |
1411 | return r; | |
1412 | ||
1413 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) | |
1414 | return -EBUSY; | |
1415 | ||
1416 | s = source_new(e, !ret, SOURCE_CHILD); | |
1417 | if (!s) | |
1418 | return -ENOMEM; | |
1419 | ||
1420 | s->wakeup = WAKEUP_EVENT_SOURCE; | |
1421 | s->child.pidfd = pidfd; | |
1422 | s->child.pid = pid; | |
1423 | s->child.options = options; | |
1424 | s->child.callback = callback; | |
1425 | 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) */ | |
1426 | s->userdata = userdata; | |
1427 | s->enabled = SD_EVENT_ONESHOT; | |
1428 | ||
1429 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); | |
1430 | if (r < 0) | |
1431 | return r; | |
1432 | ||
f8f3f926 LP |
1433 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1434 | /* We only want to watch for WEXITED */ | |
f8f3f926 | 1435 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1436 | if (r < 0) |
f8f3f926 | 1437 | return r; |
f8f3f926 LP |
1438 | } else { |
1439 | /* We shall wait for some other event than WEXITED */ | |
f8f3f926 | 1440 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1441 | if (r < 0) |
f8f3f926 | 1442 | return r; |
a71fe8b8 | 1443 | |
f8f3f926 LP |
1444 | e->need_process_child = true; |
1445 | } | |
1446 | ||
ac9f2640 ZJS |
1447 | e->n_enabled_child_sources++; |
1448 | ||
f8f3f926 LP |
1449 | if (ret) |
1450 | *ret = s; | |
f8f3f926 | 1451 | TAKE_PTR(s); |
fd38203a LP |
1452 | return 0; |
1453 | } | |
1454 | ||
b9350e70 LP |
1455 | static int generic_exit_callback(sd_event_source *s, void *userdata) { |
1456 | assert(s); | |
1457 | ||
1458 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1459 | } | |
1460 | ||
f7262a9f | 1461 | _public_ int sd_event_add_defer( |
305f78bf | 1462 | sd_event *e, |
151b9b96 | 1463 | sd_event_source **ret, |
718db961 | 1464 | sd_event_handler_t callback, |
151b9b96 | 1465 | void *userdata) { |
305f78bf | 1466 | |
ec766a51 | 1467 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1468 | int r; |
1469 | ||
305f78bf | 1470 | assert_return(e, -EINVAL); |
b937d761 | 1471 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1472 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1473 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1474 | |
b9350e70 LP |
1475 | if (!callback) |
1476 | callback = generic_exit_callback; | |
1477 | ||
a71fe8b8 | 1478 | s = source_new(e, !ret, SOURCE_DEFER); |
fd38203a LP |
1479 | if (!s) |
1480 | return -ENOMEM; | |
1481 | ||
1482 | s->defer.callback = callback; | |
1483 | s->userdata = userdata; | |
baf76283 | 1484 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a LP |
1485 | |
1486 | r = source_set_pending(s, true); | |
ec766a51 | 1487 | if (r < 0) |
fd38203a | 1488 | return r; |
fd38203a | 1489 | |
a71fe8b8 LP |
1490 | if (ret) |
1491 | *ret = s; | |
ec766a51 | 1492 | TAKE_PTR(s); |
a71fe8b8 | 1493 | |
fd38203a LP |
1494 | return 0; |
1495 | } | |
1496 | ||
6e9feda3 LP |
1497 | _public_ int sd_event_add_post( |
1498 | sd_event *e, | |
1499 | sd_event_source **ret, | |
1500 | sd_event_handler_t callback, | |
1501 | void *userdata) { | |
1502 | ||
ec766a51 | 1503 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6e9feda3 LP |
1504 | int r; |
1505 | ||
1506 | assert_return(e, -EINVAL); | |
b937d761 | 1507 | assert_return(e = event_resolve(e), -ENOPKG); |
6e9feda3 LP |
1508 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1509 | assert_return(!event_pid_changed(e), -ECHILD); | |
1510 | ||
b9350e70 LP |
1511 | if (!callback) |
1512 | callback = generic_exit_callback; | |
1513 | ||
a71fe8b8 | 1514 | s = source_new(e, !ret, SOURCE_POST); |
6e9feda3 LP |
1515 | if (!s) |
1516 | return -ENOMEM; | |
1517 | ||
1518 | s->post.callback = callback; | |
1519 | s->userdata = userdata; | |
1520 | s->enabled = SD_EVENT_ON; | |
1521 | ||
de7fef4b | 1522 | r = set_ensure_put(&e->post_sources, NULL, s); |
ec766a51 | 1523 | if (r < 0) |
6e9feda3 | 1524 | return r; |
de7fef4b | 1525 | assert(r > 0); |
6e9feda3 | 1526 | |
a71fe8b8 LP |
1527 | if (ret) |
1528 | *ret = s; | |
ec766a51 | 1529 | TAKE_PTR(s); |
a71fe8b8 | 1530 | |
6e9feda3 LP |
1531 | return 0; |
1532 | } | |
1533 | ||
6203e07a | 1534 | _public_ int sd_event_add_exit( |
305f78bf | 1535 | sd_event *e, |
151b9b96 | 1536 | sd_event_source **ret, |
718db961 | 1537 | sd_event_handler_t callback, |
151b9b96 | 1538 | void *userdata) { |
305f78bf | 1539 | |
ec766a51 | 1540 | _cleanup_(source_freep) sd_event_source *s = NULL; |
da7e457c LP |
1541 | int r; |
1542 | ||
1543 | assert_return(e, -EINVAL); | |
b937d761 | 1544 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
1545 | assert_return(callback, -EINVAL); |
1546 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
1547 | assert_return(!event_pid_changed(e), -ECHILD); | |
1548 | ||
c983e776 EV |
1549 | r = prioq_ensure_allocated(&e->exit, exit_prioq_compare); |
1550 | if (r < 0) | |
1551 | return r; | |
da7e457c | 1552 | |
a71fe8b8 | 1553 | s = source_new(e, !ret, SOURCE_EXIT); |
fd38203a | 1554 | if (!s) |
da7e457c | 1555 | return -ENOMEM; |
fd38203a | 1556 | |
6203e07a | 1557 | s->exit.callback = callback; |
da7e457c | 1558 | s->userdata = userdata; |
6203e07a | 1559 | s->exit.prioq_index = PRIOQ_IDX_NULL; |
baf76283 | 1560 | s->enabled = SD_EVENT_ONESHOT; |
da7e457c | 1561 | |
6203e07a | 1562 | r = prioq_put(s->event->exit, s, &s->exit.prioq_index); |
ec766a51 | 1563 | if (r < 0) |
da7e457c | 1564 | return r; |
da7e457c | 1565 | |
a71fe8b8 LP |
1566 | if (ret) |
1567 | *ret = s; | |
ec766a51 | 1568 | TAKE_PTR(s); |
a71fe8b8 | 1569 | |
da7e457c LP |
1570 | return 0; |
1571 | } | |
1572 | ||
97ef5391 LP |
1573 | static void event_free_inotify_data(sd_event *e, struct inotify_data *d) { |
1574 | assert(e); | |
1575 | ||
1576 | if (!d) | |
1577 | return; | |
1578 | ||
1579 | assert(hashmap_isempty(d->inodes)); | |
1580 | assert(hashmap_isempty(d->wd)); | |
1581 | ||
1582 | if (d->buffer_filled > 0) | |
1583 | LIST_REMOVE(buffered, e->inotify_data_buffered, d); | |
1584 | ||
1585 | hashmap_free(d->inodes); | |
1586 | hashmap_free(d->wd); | |
1587 | ||
1588 | assert_se(hashmap_remove(e->inotify_data, &d->priority) == d); | |
1589 | ||
1590 | if (d->fd >= 0) { | |
1591 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, d->fd, NULL) < 0) | |
1592 | log_debug_errno(errno, "Failed to remove inotify fd from epoll, ignoring: %m"); | |
1593 | ||
1594 | safe_close(d->fd); | |
1595 | } | |
1596 | free(d); | |
1597 | } | |
1598 | ||
1599 | static int event_make_inotify_data( | |
1600 | sd_event *e, | |
1601 | int64_t priority, | |
1602 | struct inotify_data **ret) { | |
1603 | ||
1604 | _cleanup_close_ int fd = -1; | |
1605 | struct inotify_data *d; | |
97ef5391 LP |
1606 | int r; |
1607 | ||
1608 | assert(e); | |
1609 | ||
1610 | d = hashmap_get(e->inotify_data, &priority); | |
1611 | if (d) { | |
1612 | if (ret) | |
1613 | *ret = d; | |
1614 | return 0; | |
1615 | } | |
1616 | ||
1617 | fd = inotify_init1(IN_NONBLOCK|O_CLOEXEC); | |
1618 | if (fd < 0) | |
1619 | return -errno; | |
1620 | ||
1621 | fd = fd_move_above_stdio(fd); | |
1622 | ||
1623 | r = hashmap_ensure_allocated(&e->inotify_data, &uint64_hash_ops); | |
1624 | if (r < 0) | |
1625 | return r; | |
1626 | ||
1627 | d = new(struct inotify_data, 1); | |
1628 | if (!d) | |
1629 | return -ENOMEM; | |
1630 | ||
1631 | *d = (struct inotify_data) { | |
1632 | .wakeup = WAKEUP_INOTIFY_DATA, | |
1633 | .fd = TAKE_FD(fd), | |
1634 | .priority = priority, | |
1635 | }; | |
1636 | ||
1637 | r = hashmap_put(e->inotify_data, &d->priority, d); | |
1638 | if (r < 0) { | |
1639 | d->fd = safe_close(d->fd); | |
1640 | free(d); | |
1641 | return r; | |
1642 | } | |
1643 | ||
1eac7948 | 1644 | struct epoll_event ev = { |
97ef5391 LP |
1645 | .events = EPOLLIN, |
1646 | .data.ptr = d, | |
1647 | }; | |
1648 | ||
1649 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { | |
1650 | r = -errno; | |
1651 | d->fd = safe_close(d->fd); /* let's close this ourselves, as event_free_inotify_data() would otherwise | |
1652 | * remove the fd from the epoll first, which we don't want as we couldn't | |
1653 | * add it in the first place. */ | |
1654 | event_free_inotify_data(e, d); | |
1655 | return r; | |
1656 | } | |
1657 | ||
1658 | if (ret) | |
1659 | *ret = d; | |
1660 | ||
1661 | return 1; | |
1662 | } | |
1663 | ||
7a08d314 | 1664 | static int inode_data_compare(const struct inode_data *x, const struct inode_data *y) { |
90c88092 | 1665 | int r; |
97ef5391 LP |
1666 | |
1667 | assert(x); | |
1668 | assert(y); | |
1669 | ||
90c88092 YW |
1670 | r = CMP(x->dev, y->dev); |
1671 | if (r != 0) | |
1672 | return r; | |
97ef5391 | 1673 | |
6dd91b36 | 1674 | return CMP(x->ino, y->ino); |
97ef5391 LP |
1675 | } |
1676 | ||
7a08d314 YW |
1677 | static void inode_data_hash_func(const struct inode_data *d, struct siphash *state) { |
1678 | assert(d); | |
97ef5391 LP |
1679 | |
1680 | siphash24_compress(&d->dev, sizeof(d->dev), state); | |
1681 | siphash24_compress(&d->ino, sizeof(d->ino), state); | |
1682 | } | |
1683 | ||
7a08d314 | 1684 | DEFINE_PRIVATE_HASH_OPS(inode_data_hash_ops, struct inode_data, inode_data_hash_func, inode_data_compare); |
97ef5391 LP |
1685 | |
1686 | static void event_free_inode_data( | |
1687 | sd_event *e, | |
1688 | struct inode_data *d) { | |
1689 | ||
1690 | assert(e); | |
1691 | ||
1692 | if (!d) | |
1693 | return; | |
1694 | ||
1695 | assert(!d->event_sources); | |
1696 | ||
1697 | if (d->fd >= 0) { | |
1698 | LIST_REMOVE(to_close, e->inode_data_to_close, d); | |
1699 | safe_close(d->fd); | |
1700 | } | |
1701 | ||
1702 | if (d->inotify_data) { | |
1703 | ||
1704 | if (d->wd >= 0) { | |
1705 | if (d->inotify_data->fd >= 0) { | |
1706 | /* So here's a problem. At the time this runs the watch descriptor might already be | |
1707 | * invalidated, because an IN_IGNORED event might be queued right the moment we enter | |
1708 | * the syscall. Hence, whenever we get EINVAL, ignore it entirely, since it's a very | |
1709 | * likely case to happen. */ | |
1710 | ||
1711 | if (inotify_rm_watch(d->inotify_data->fd, d->wd) < 0 && errno != EINVAL) | |
1712 | log_debug_errno(errno, "Failed to remove watch descriptor %i from inotify, ignoring: %m", d->wd); | |
1713 | } | |
1714 | ||
1715 | assert_se(hashmap_remove(d->inotify_data->wd, INT_TO_PTR(d->wd)) == d); | |
1716 | } | |
1717 | ||
1718 | assert_se(hashmap_remove(d->inotify_data->inodes, d) == d); | |
1719 | } | |
1720 | ||
1721 | free(d); | |
1722 | } | |
1723 | ||
1724 | static void event_gc_inode_data( | |
1725 | sd_event *e, | |
1726 | struct inode_data *d) { | |
1727 | ||
1728 | struct inotify_data *inotify_data; | |
1729 | ||
1730 | assert(e); | |
1731 | ||
1732 | if (!d) | |
1733 | return; | |
1734 | ||
1735 | if (d->event_sources) | |
1736 | return; | |
1737 | ||
1738 | inotify_data = d->inotify_data; | |
1739 | event_free_inode_data(e, d); | |
1740 | ||
1741 | if (inotify_data && hashmap_isempty(inotify_data->inodes)) | |
1742 | event_free_inotify_data(e, inotify_data); | |
1743 | } | |
1744 | ||
1745 | static int event_make_inode_data( | |
1746 | sd_event *e, | |
1747 | struct inotify_data *inotify_data, | |
1748 | dev_t dev, | |
1749 | ino_t ino, | |
1750 | struct inode_data **ret) { | |
1751 | ||
1752 | struct inode_data *d, key; | |
1753 | int r; | |
1754 | ||
1755 | assert(e); | |
1756 | assert(inotify_data); | |
1757 | ||
1758 | key = (struct inode_data) { | |
1759 | .ino = ino, | |
1760 | .dev = dev, | |
1761 | }; | |
1762 | ||
1763 | d = hashmap_get(inotify_data->inodes, &key); | |
1764 | if (d) { | |
1765 | if (ret) | |
1766 | *ret = d; | |
1767 | ||
1768 | return 0; | |
1769 | } | |
1770 | ||
1771 | r = hashmap_ensure_allocated(&inotify_data->inodes, &inode_data_hash_ops); | |
1772 | if (r < 0) | |
1773 | return r; | |
1774 | ||
1775 | d = new(struct inode_data, 1); | |
1776 | if (!d) | |
1777 | return -ENOMEM; | |
1778 | ||
1779 | *d = (struct inode_data) { | |
1780 | .dev = dev, | |
1781 | .ino = ino, | |
1782 | .wd = -1, | |
1783 | .fd = -1, | |
1784 | .inotify_data = inotify_data, | |
1785 | }; | |
1786 | ||
1787 | r = hashmap_put(inotify_data->inodes, d, d); | |
1788 | if (r < 0) { | |
1789 | free(d); | |
1790 | return r; | |
1791 | } | |
1792 | ||
1793 | if (ret) | |
1794 | *ret = d; | |
1795 | ||
1796 | return 1; | |
1797 | } | |
1798 | ||
1799 | static uint32_t inode_data_determine_mask(struct inode_data *d) { | |
1800 | bool excl_unlink = true; | |
1801 | uint32_t combined = 0; | |
1802 | sd_event_source *s; | |
1803 | ||
1804 | assert(d); | |
1805 | ||
1806 | /* Combines the watch masks of all event sources watching this inode. We generally just OR them together, but | |
1807 | * the IN_EXCL_UNLINK flag is ANDed instead. | |
1808 | * | |
1809 | * Note that we add all sources to the mask here, regardless whether enabled, disabled or oneshot. That's | |
1810 | * because we cannot change the mask anymore after the event source was created once, since the kernel has no | |
f21f31b2 | 1811 | * API for that. Hence we need to subscribe to the maximum mask we ever might be interested in, and suppress |
97ef5391 LP |
1812 | * events we don't care for client-side. */ |
1813 | ||
1814 | LIST_FOREACH(inotify.by_inode_data, s, d->event_sources) { | |
1815 | ||
1816 | if ((s->inotify.mask & IN_EXCL_UNLINK) == 0) | |
1817 | excl_unlink = false; | |
1818 | ||
1819 | combined |= s->inotify.mask; | |
1820 | } | |
1821 | ||
1822 | return (combined & ~(IN_ONESHOT|IN_DONT_FOLLOW|IN_ONLYDIR|IN_EXCL_UNLINK)) | (excl_unlink ? IN_EXCL_UNLINK : 0); | |
1823 | } | |
1824 | ||
1825 | static int inode_data_realize_watch(sd_event *e, struct inode_data *d) { | |
1826 | uint32_t combined_mask; | |
1827 | int wd, r; | |
1828 | ||
1829 | assert(d); | |
1830 | assert(d->fd >= 0); | |
1831 | ||
1832 | combined_mask = inode_data_determine_mask(d); | |
1833 | ||
1834 | if (d->wd >= 0 && combined_mask == d->combined_mask) | |
1835 | return 0; | |
1836 | ||
1837 | r = hashmap_ensure_allocated(&d->inotify_data->wd, NULL); | |
1838 | if (r < 0) | |
1839 | return r; | |
1840 | ||
1841 | wd = inotify_add_watch_fd(d->inotify_data->fd, d->fd, combined_mask); | |
1842 | if (wd < 0) | |
1843 | return -errno; | |
1844 | ||
1845 | if (d->wd < 0) { | |
1846 | r = hashmap_put(d->inotify_data->wd, INT_TO_PTR(wd), d); | |
1847 | if (r < 0) { | |
1848 | (void) inotify_rm_watch(d->inotify_data->fd, wd); | |
1849 | return r; | |
1850 | } | |
1851 | ||
1852 | d->wd = wd; | |
1853 | ||
1854 | } else if (d->wd != wd) { | |
1855 | ||
1856 | log_debug("Weird, the watch descriptor we already knew for this inode changed?"); | |
1857 | (void) inotify_rm_watch(d->fd, wd); | |
1858 | return -EINVAL; | |
1859 | } | |
1860 | ||
1861 | d->combined_mask = combined_mask; | |
1862 | return 1; | |
1863 | } | |
1864 | ||
b9350e70 LP |
1865 | static int inotify_exit_callback(sd_event_source *s, const struct inotify_event *event, void *userdata) { |
1866 | assert(s); | |
1867 | ||
1868 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1869 | } | |
1870 | ||
97ef5391 LP |
1871 | _public_ int sd_event_add_inotify( |
1872 | sd_event *e, | |
1873 | sd_event_source **ret, | |
1874 | const char *path, | |
1875 | uint32_t mask, | |
1876 | sd_event_inotify_handler_t callback, | |
1877 | void *userdata) { | |
1878 | ||
97ef5391 LP |
1879 | struct inotify_data *inotify_data = NULL; |
1880 | struct inode_data *inode_data = NULL; | |
1881 | _cleanup_close_ int fd = -1; | |
8c75fe17 | 1882 | _cleanup_(source_freep) sd_event_source *s = NULL; |
97ef5391 LP |
1883 | struct stat st; |
1884 | int r; | |
1885 | ||
1886 | assert_return(e, -EINVAL); | |
1887 | assert_return(e = event_resolve(e), -ENOPKG); | |
1888 | assert_return(path, -EINVAL); | |
97ef5391 LP |
1889 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1890 | assert_return(!event_pid_changed(e), -ECHILD); | |
1891 | ||
b9350e70 LP |
1892 | if (!callback) |
1893 | callback = inotify_exit_callback; | |
1894 | ||
97ef5391 LP |
1895 | /* Refuse IN_MASK_ADD since we coalesce watches on the same inode, and hence really don't want to merge |
1896 | * masks. Or in other words, this whole code exists only to manage IN_MASK_ADD type operations for you, hence | |
1897 | * the user can't use them for us. */ | |
1898 | if (mask & IN_MASK_ADD) | |
1899 | return -EINVAL; | |
1900 | ||
1901 | fd = open(path, O_PATH|O_CLOEXEC| | |
1902 | (mask & IN_ONLYDIR ? O_DIRECTORY : 0)| | |
1903 | (mask & IN_DONT_FOLLOW ? O_NOFOLLOW : 0)); | |
1904 | if (fd < 0) | |
1905 | return -errno; | |
1906 | ||
1907 | if (fstat(fd, &st) < 0) | |
1908 | return -errno; | |
1909 | ||
1910 | s = source_new(e, !ret, SOURCE_INOTIFY); | |
1911 | if (!s) | |
1912 | return -ENOMEM; | |
1913 | ||
1914 | s->enabled = mask & IN_ONESHOT ? SD_EVENT_ONESHOT : SD_EVENT_ON; | |
1915 | s->inotify.mask = mask; | |
1916 | s->inotify.callback = callback; | |
1917 | s->userdata = userdata; | |
1918 | ||
1919 | /* Allocate an inotify object for this priority, and an inode object within it */ | |
1920 | r = event_make_inotify_data(e, SD_EVENT_PRIORITY_NORMAL, &inotify_data); | |
1921 | if (r < 0) | |
8c75fe17 | 1922 | return r; |
97ef5391 LP |
1923 | |
1924 | r = event_make_inode_data(e, inotify_data, st.st_dev, st.st_ino, &inode_data); | |
8c75fe17 ZJS |
1925 | if (r < 0) { |
1926 | event_free_inotify_data(e, inotify_data); | |
1927 | return r; | |
1928 | } | |
97ef5391 LP |
1929 | |
1930 | /* Keep the O_PATH fd around until the first iteration of the loop, so that we can still change the priority of | |
1931 | * the event source, until then, for which we need the original inode. */ | |
1932 | if (inode_data->fd < 0) { | |
1933 | inode_data->fd = TAKE_FD(fd); | |
1934 | LIST_PREPEND(to_close, e->inode_data_to_close, inode_data); | |
1935 | } | |
1936 | ||
1937 | /* Link our event source to the inode data object */ | |
1938 | LIST_PREPEND(inotify.by_inode_data, inode_data->event_sources, s); | |
1939 | s->inotify.inode_data = inode_data; | |
1940 | ||
97ef5391 LP |
1941 | /* Actually realize the watch now */ |
1942 | r = inode_data_realize_watch(e, inode_data); | |
1943 | if (r < 0) | |
8c75fe17 | 1944 | return r; |
97ef5391 LP |
1945 | |
1946 | (void) sd_event_source_set_description(s, path); | |
1947 | ||
1948 | if (ret) | |
1949 | *ret = s; | |
8c75fe17 | 1950 | TAKE_PTR(s); |
97ef5391 LP |
1951 | |
1952 | return 0; | |
97ef5391 LP |
1953 | } |
1954 | ||
8301aa0b | 1955 | static sd_event_source* event_source_free(sd_event_source *s) { |
6680dd6b LP |
1956 | if (!s) |
1957 | return NULL; | |
da7e457c | 1958 | |
8301aa0b YW |
1959 | /* Here's a special hack: when we are called from a |
1960 | * dispatch handler we won't free the event source | |
1961 | * immediately, but we will detach the fd from the | |
1962 | * epoll. This way it is safe for the caller to unref | |
1963 | * the event source and immediately close the fd, but | |
1964 | * we still retain a valid event source object after | |
1965 | * the callback. */ | |
fd38203a | 1966 | |
8301aa0b YW |
1967 | if (s->dispatching) { |
1968 | if (s->type == SOURCE_IO) | |
1969 | source_io_unregister(s); | |
fd38203a | 1970 | |
8301aa0b YW |
1971 | source_disconnect(s); |
1972 | } else | |
1973 | source_free(s); | |
fd38203a LP |
1974 | |
1975 | return NULL; | |
1976 | } | |
1977 | ||
8301aa0b YW |
1978 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event_source, sd_event_source, event_source_free); |
1979 | ||
356779df | 1980 | _public_ int sd_event_source_set_description(sd_event_source *s, const char *description) { |
f7f53e9e | 1981 | assert_return(s, -EINVAL); |
f4b2933e | 1982 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 1983 | |
356779df | 1984 | return free_and_strdup(&s->description, description); |
f7f53e9e TG |
1985 | } |
1986 | ||
356779df | 1987 | _public_ int sd_event_source_get_description(sd_event_source *s, const char **description) { |
f7f53e9e | 1988 | assert_return(s, -EINVAL); |
356779df | 1989 | assert_return(description, -EINVAL); |
f4b2933e | 1990 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 1991 | |
7d92a1a4 ZJS |
1992 | if (!s->description) |
1993 | return -ENXIO; | |
1994 | ||
356779df | 1995 | *description = s->description; |
f7f53e9e TG |
1996 | return 0; |
1997 | } | |
1998 | ||
adcc4ca3 | 1999 | _public_ sd_event *sd_event_source_get_event(sd_event_source *s) { |
305f78bf | 2000 | assert_return(s, NULL); |
eaa3cbef LP |
2001 | |
2002 | return s->event; | |
2003 | } | |
2004 | ||
f7262a9f | 2005 | _public_ int sd_event_source_get_pending(sd_event_source *s) { |
305f78bf | 2006 | assert_return(s, -EINVAL); |
6203e07a | 2007 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c | 2008 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2009 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2010 | |
2011 | return s->pending; | |
2012 | } | |
2013 | ||
f7262a9f | 2014 | _public_ int sd_event_source_get_io_fd(sd_event_source *s) { |
305f78bf LP |
2015 | assert_return(s, -EINVAL); |
2016 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2017 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2018 | |
2019 | return s->io.fd; | |
2020 | } | |
2021 | ||
30caf8f3 LP |
2022 | _public_ int sd_event_source_set_io_fd(sd_event_source *s, int fd) { |
2023 | int r; | |
2024 | ||
2025 | assert_return(s, -EINVAL); | |
8ac43fee | 2026 | assert_return(fd >= 0, -EBADF); |
30caf8f3 LP |
2027 | assert_return(s->type == SOURCE_IO, -EDOM); |
2028 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2029 | ||
2030 | if (s->io.fd == fd) | |
2031 | return 0; | |
2032 | ||
2033 | if (s->enabled == SD_EVENT_OFF) { | |
2034 | s->io.fd = fd; | |
2035 | s->io.registered = false; | |
2036 | } else { | |
2037 | int saved_fd; | |
2038 | ||
2039 | saved_fd = s->io.fd; | |
2040 | assert(s->io.registered); | |
2041 | ||
2042 | s->io.fd = fd; | |
2043 | s->io.registered = false; | |
2044 | ||
2045 | r = source_io_register(s, s->enabled, s->io.events); | |
2046 | if (r < 0) { | |
2047 | s->io.fd = saved_fd; | |
2048 | s->io.registered = true; | |
2049 | return r; | |
2050 | } | |
2051 | ||
5a795bff | 2052 | (void) epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, saved_fd, NULL); |
30caf8f3 LP |
2053 | } |
2054 | ||
2055 | return 0; | |
2056 | } | |
2057 | ||
ab93297c NM |
2058 | _public_ int sd_event_source_get_io_fd_own(sd_event_source *s) { |
2059 | assert_return(s, -EINVAL); | |
2060 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2061 | ||
2062 | return s->io.owned; | |
2063 | } | |
2064 | ||
2065 | _public_ int sd_event_source_set_io_fd_own(sd_event_source *s, int own) { | |
2066 | assert_return(s, -EINVAL); | |
2067 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2068 | ||
2069 | s->io.owned = own; | |
2070 | return 0; | |
2071 | } | |
2072 | ||
f7262a9f | 2073 | _public_ int sd_event_source_get_io_events(sd_event_source *s, uint32_t* events) { |
305f78bf LP |
2074 | assert_return(s, -EINVAL); |
2075 | assert_return(events, -EINVAL); | |
2076 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2077 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2078 | |
2079 | *events = s->io.events; | |
2080 | return 0; | |
2081 | } | |
2082 | ||
f7262a9f | 2083 | _public_ int sd_event_source_set_io_events(sd_event_source *s, uint32_t events) { |
fd38203a LP |
2084 | int r; |
2085 | ||
305f78bf LP |
2086 | assert_return(s, -EINVAL); |
2087 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2a16a986 | 2088 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 2089 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2090 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2091 | |
b63c8d4f DH |
2092 | /* edge-triggered updates are never skipped, so we can reset edges */ |
2093 | if (s->io.events == events && !(events & EPOLLET)) | |
fd38203a LP |
2094 | return 0; |
2095 | ||
2a0dc6cd LP |
2096 | r = source_set_pending(s, false); |
2097 | if (r < 0) | |
2098 | return r; | |
2099 | ||
baf76283 | 2100 | if (s->enabled != SD_EVENT_OFF) { |
e4715127 | 2101 | r = source_io_register(s, s->enabled, events); |
fd38203a LP |
2102 | if (r < 0) |
2103 | return r; | |
2104 | } | |
2105 | ||
2106 | s->io.events = events; | |
2107 | ||
2108 | return 0; | |
2109 | } | |
2110 | ||
f7262a9f | 2111 | _public_ int sd_event_source_get_io_revents(sd_event_source *s, uint32_t* revents) { |
305f78bf LP |
2112 | assert_return(s, -EINVAL); |
2113 | assert_return(revents, -EINVAL); | |
2114 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2115 | assert_return(s->pending, -ENODATA); | |
2116 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2117 | |
2118 | *revents = s->io.revents; | |
2119 | return 0; | |
2120 | } | |
2121 | ||
f7262a9f | 2122 | _public_ int sd_event_source_get_signal(sd_event_source *s) { |
305f78bf LP |
2123 | assert_return(s, -EINVAL); |
2124 | assert_return(s->type == SOURCE_SIGNAL, -EDOM); | |
2125 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2126 | |
2127 | return s->signal.sig; | |
2128 | } | |
2129 | ||
31927c16 | 2130 | _public_ int sd_event_source_get_priority(sd_event_source *s, int64_t *priority) { |
305f78bf LP |
2131 | assert_return(s, -EINVAL); |
2132 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a | 2133 | |
6680b8d1 ME |
2134 | *priority = s->priority; |
2135 | return 0; | |
fd38203a LP |
2136 | } |
2137 | ||
31927c16 | 2138 | _public_ int sd_event_source_set_priority(sd_event_source *s, int64_t priority) { |
97ef5391 LP |
2139 | bool rm_inotify = false, rm_inode = false; |
2140 | struct inotify_data *new_inotify_data = NULL; | |
2141 | struct inode_data *new_inode_data = NULL; | |
9da4cb2b LP |
2142 | int r; |
2143 | ||
305f78bf | 2144 | assert_return(s, -EINVAL); |
da7e457c | 2145 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2146 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2147 | |
2148 | if (s->priority == priority) | |
2149 | return 0; | |
2150 | ||
97ef5391 LP |
2151 | if (s->type == SOURCE_INOTIFY) { |
2152 | struct inode_data *old_inode_data; | |
2153 | ||
2154 | assert(s->inotify.inode_data); | |
2155 | old_inode_data = s->inotify.inode_data; | |
2156 | ||
2157 | /* We need the original fd to change the priority. If we don't have it we can't change the priority, | |
2158 | * anymore. Note that we close any fds when entering the next event loop iteration, i.e. for inotify | |
2159 | * events we allow priority changes only until the first following iteration. */ | |
2160 | if (old_inode_data->fd < 0) | |
2161 | return -EOPNOTSUPP; | |
2162 | ||
2163 | r = event_make_inotify_data(s->event, priority, &new_inotify_data); | |
2164 | if (r < 0) | |
2165 | return r; | |
2166 | rm_inotify = r > 0; | |
2167 | ||
2168 | r = event_make_inode_data(s->event, new_inotify_data, old_inode_data->dev, old_inode_data->ino, &new_inode_data); | |
2169 | if (r < 0) | |
2170 | goto fail; | |
2171 | rm_inode = r > 0; | |
2172 | ||
2173 | if (new_inode_data->fd < 0) { | |
2174 | /* Duplicate the fd for the new inode object if we don't have any yet */ | |
2175 | new_inode_data->fd = fcntl(old_inode_data->fd, F_DUPFD_CLOEXEC, 3); | |
2176 | if (new_inode_data->fd < 0) { | |
2177 | r = -errno; | |
2178 | goto fail; | |
2179 | } | |
2180 | ||
2181 | LIST_PREPEND(to_close, s->event->inode_data_to_close, new_inode_data); | |
2182 | } | |
2183 | ||
2184 | /* Move the event source to the new inode data structure */ | |
2185 | LIST_REMOVE(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2186 | LIST_PREPEND(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2187 | s->inotify.inode_data = new_inode_data; | |
2188 | ||
2189 | /* Now create the new watch */ | |
2190 | r = inode_data_realize_watch(s->event, new_inode_data); | |
2191 | if (r < 0) { | |
2192 | /* Move it back */ | |
2193 | LIST_REMOVE(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2194 | LIST_PREPEND(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2195 | s->inotify.inode_data = old_inode_data; | |
2196 | goto fail; | |
2197 | } | |
2198 | ||
2199 | s->priority = priority; | |
2200 | ||
2201 | event_gc_inode_data(s->event, old_inode_data); | |
2202 | ||
2203 | } else if (s->type == SOURCE_SIGNAL && s->enabled != SD_EVENT_OFF) { | |
9da4cb2b LP |
2204 | struct signal_data *old, *d; |
2205 | ||
2206 | /* Move us from the signalfd belonging to the old | |
2207 | * priority to the signalfd of the new priority */ | |
2208 | ||
2209 | assert_se(old = hashmap_get(s->event->signal_data, &s->priority)); | |
2210 | ||
2211 | s->priority = priority; | |
2212 | ||
2213 | r = event_make_signal_data(s->event, s->signal.sig, &d); | |
2214 | if (r < 0) { | |
2215 | s->priority = old->priority; | |
2216 | return r; | |
2217 | } | |
2218 | ||
2219 | event_unmask_signal_data(s->event, old, s->signal.sig); | |
2220 | } else | |
2221 | s->priority = priority; | |
fd38203a | 2222 | |
e1951c16 | 2223 | event_source_pp_prioq_reshuffle(s); |
fd38203a | 2224 | |
6203e07a LP |
2225 | if (s->type == SOURCE_EXIT) |
2226 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
305f78bf | 2227 | |
fd38203a | 2228 | return 0; |
97ef5391 LP |
2229 | |
2230 | fail: | |
2231 | if (rm_inode) | |
2232 | event_free_inode_data(s->event, new_inode_data); | |
2233 | ||
2234 | if (rm_inotify) | |
2235 | event_free_inotify_data(s->event, new_inotify_data); | |
2236 | ||
2237 | return r; | |
fd38203a LP |
2238 | } |
2239 | ||
f7262a9f | 2240 | _public_ int sd_event_source_get_enabled(sd_event_source *s, int *m) { |
305f78bf | 2241 | assert_return(s, -EINVAL); |
305f78bf | 2242 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2243 | |
08c1eb0e ZJS |
2244 | if (m) |
2245 | *m = s->enabled; | |
2246 | return s->enabled != SD_EVENT_OFF; | |
fd38203a LP |
2247 | } |
2248 | ||
ddfde737 | 2249 | static int event_source_disable(sd_event_source *s) { |
fd38203a LP |
2250 | int r; |
2251 | ||
ddfde737 LP |
2252 | assert(s); |
2253 | assert(s->enabled != SD_EVENT_OFF); | |
fd38203a | 2254 | |
ddfde737 LP |
2255 | /* Unset the pending flag when this event source is disabled */ |
2256 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
2257 | r = source_set_pending(s, false); | |
2258 | if (r < 0) | |
2259 | return r; | |
2260 | } | |
cc567911 | 2261 | |
ddfde737 | 2262 | s->enabled = SD_EVENT_OFF; |
fd38203a | 2263 | |
ddfde737 | 2264 | switch (s->type) { |
fd38203a | 2265 | |
ddfde737 LP |
2266 | case SOURCE_IO: |
2267 | source_io_unregister(s); | |
2268 | break; | |
ac989a78 | 2269 | |
ddfde737 LP |
2270 | case SOURCE_TIME_REALTIME: |
2271 | case SOURCE_TIME_BOOTTIME: | |
2272 | case SOURCE_TIME_MONOTONIC: | |
2273 | case SOURCE_TIME_REALTIME_ALARM: | |
2274 | case SOURCE_TIME_BOOTTIME_ALARM: | |
2275 | event_source_time_prioq_reshuffle(s); | |
2276 | break; | |
fd38203a | 2277 | |
ddfde737 LP |
2278 | case SOURCE_SIGNAL: |
2279 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); | |
2280 | break; | |
fd38203a | 2281 | |
ddfde737 LP |
2282 | case SOURCE_CHILD: |
2283 | assert(s->event->n_enabled_child_sources > 0); | |
2284 | s->event->n_enabled_child_sources--; | |
fd38203a | 2285 | |
ddfde737 LP |
2286 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
2287 | source_child_pidfd_unregister(s); | |
2288 | else | |
2289 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
2290 | break; | |
4807d2d0 | 2291 | |
ddfde737 LP |
2292 | case SOURCE_EXIT: |
2293 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
2294 | break; | |
fd38203a | 2295 | |
ddfde737 LP |
2296 | case SOURCE_DEFER: |
2297 | case SOURCE_POST: | |
2298 | case SOURCE_INOTIFY: | |
2299 | break; | |
fd38203a | 2300 | |
ddfde737 LP |
2301 | default: |
2302 | assert_not_reached("Wut? I shouldn't exist."); | |
2303 | } | |
fd38203a | 2304 | |
ddfde737 LP |
2305 | return 0; |
2306 | } | |
f8f3f926 | 2307 | |
d2eafe61 | 2308 | static int event_source_enable(sd_event_source *s, int enable) { |
ddfde737 | 2309 | int r; |
fd38203a | 2310 | |
ddfde737 | 2311 | assert(s); |
d2eafe61 | 2312 | assert(IN_SET(enable, SD_EVENT_ON, SD_EVENT_ONESHOT)); |
ddfde737 | 2313 | assert(s->enabled == SD_EVENT_OFF); |
305f78bf | 2314 | |
ddfde737 LP |
2315 | /* Unset the pending flag when this event source is enabled */ |
2316 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
2317 | r = source_set_pending(s, false); | |
2318 | if (r < 0) | |
2319 | return r; | |
2320 | } | |
9d3e3aa5 | 2321 | |
ddfde737 | 2322 | switch (s->type) { |
ddfde737 | 2323 | case SOURCE_IO: |
d2eafe61 ZJS |
2324 | r = source_io_register(s, enable, s->io.events); |
2325 | if (r < 0) | |
ddfde737 | 2326 | return r; |
ddfde737 | 2327 | break; |
fd38203a | 2328 | |
ddfde737 LP |
2329 | case SOURCE_SIGNAL: |
2330 | r = event_make_signal_data(s->event, s->signal.sig, NULL); | |
2331 | if (r < 0) { | |
ddfde737 LP |
2332 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
2333 | return r; | |
2334 | } | |
fd38203a | 2335 | |
ddfde737 | 2336 | break; |
fd38203a | 2337 | |
ddfde737 | 2338 | case SOURCE_CHILD: |
ddfde737 LP |
2339 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
2340 | /* yes, we have pidfd */ | |
9da4cb2b | 2341 | |
d2eafe61 | 2342 | r = source_child_pidfd_register(s, enable); |
ac9f2640 | 2343 | if (r < 0) |
9da4cb2b | 2344 | return r; |
ddfde737 LP |
2345 | } else { |
2346 | /* no pidfd, or something other to watch for than WEXITED */ | |
9da4cb2b | 2347 | |
ddfde737 LP |
2348 | r = event_make_signal_data(s->event, SIGCHLD, NULL); |
2349 | if (r < 0) { | |
ddfde737 LP |
2350 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); |
2351 | return r; | |
2352 | } | |
2353 | } | |
fd38203a | 2354 | |
ac9f2640 ZJS |
2355 | s->event->n_enabled_child_sources++; |
2356 | ||
ddfde737 | 2357 | break; |
4807d2d0 | 2358 | |
d2eafe61 ZJS |
2359 | case SOURCE_TIME_REALTIME: |
2360 | case SOURCE_TIME_BOOTTIME: | |
2361 | case SOURCE_TIME_MONOTONIC: | |
2362 | case SOURCE_TIME_REALTIME_ALARM: | |
2363 | case SOURCE_TIME_BOOTTIME_ALARM: | |
ddfde737 | 2364 | case SOURCE_EXIT: |
ddfde737 LP |
2365 | case SOURCE_DEFER: |
2366 | case SOURCE_POST: | |
2367 | case SOURCE_INOTIFY: | |
2368 | break; | |
9da4cb2b | 2369 | |
ddfde737 LP |
2370 | default: |
2371 | assert_not_reached("Wut? I shouldn't exist."); | |
2372 | } | |
f8f3f926 | 2373 | |
d2eafe61 ZJS |
2374 | s->enabled = enable; |
2375 | ||
2376 | /* Non-failing operations below */ | |
2377 | switch (s->type) { | |
2378 | case SOURCE_TIME_REALTIME: | |
2379 | case SOURCE_TIME_BOOTTIME: | |
2380 | case SOURCE_TIME_MONOTONIC: | |
2381 | case SOURCE_TIME_REALTIME_ALARM: | |
2382 | case SOURCE_TIME_BOOTTIME_ALARM: | |
2383 | event_source_time_prioq_reshuffle(s); | |
2384 | break; | |
2385 | ||
2386 | case SOURCE_EXIT: | |
2387 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
2388 | break; | |
2389 | ||
2390 | default: | |
2391 | break; | |
2392 | } | |
2393 | ||
ddfde737 LP |
2394 | return 0; |
2395 | } | |
2396 | ||
2397 | _public_ int sd_event_source_set_enabled(sd_event_source *s, int m) { | |
2398 | int r; | |
9da4cb2b | 2399 | |
ddfde737 LP |
2400 | assert_return(s, -EINVAL); |
2401 | assert_return(IN_SET(m, SD_EVENT_OFF, SD_EVENT_ON, SD_EVENT_ONESHOT), -EINVAL); | |
2402 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a | 2403 | |
ddfde737 LP |
2404 | /* If we are dead anyway, we are fine with turning off sources, but everything else needs to fail. */ |
2405 | if (s->event->state == SD_EVENT_FINISHED) | |
2406 | return m == SD_EVENT_OFF ? 0 : -ESTALE; | |
305f78bf | 2407 | |
ddfde737 LP |
2408 | if (s->enabled == m) /* No change? */ |
2409 | return 0; | |
9d3e3aa5 | 2410 | |
ddfde737 LP |
2411 | if (m == SD_EVENT_OFF) |
2412 | r = event_source_disable(s); | |
2413 | else { | |
2414 | if (s->enabled != SD_EVENT_OFF) { | |
2415 | /* Switching from "on" to "oneshot" or back? If that's the case, we can take a shortcut, the | |
2416 | * event source is already enabled after all. */ | |
2417 | s->enabled = m; | |
2418 | return 0; | |
fd38203a | 2419 | } |
ddfde737 LP |
2420 | |
2421 | r = event_source_enable(s, m); | |
fd38203a | 2422 | } |
ddfde737 LP |
2423 | if (r < 0) |
2424 | return r; | |
fd38203a | 2425 | |
e1951c16 | 2426 | event_source_pp_prioq_reshuffle(s); |
fd38203a LP |
2427 | return 0; |
2428 | } | |
2429 | ||
f7262a9f | 2430 | _public_ int sd_event_source_get_time(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
2431 | assert_return(s, -EINVAL); |
2432 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 2433 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf | 2434 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2435 | |
2436 | *usec = s->time.next; | |
2437 | return 0; | |
2438 | } | |
2439 | ||
f7262a9f | 2440 | _public_ int sd_event_source_set_time(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 2441 | int r; |
6a0f1f6d | 2442 | |
305f78bf | 2443 | assert_return(s, -EINVAL); |
6a0f1f6d | 2444 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 2445 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2446 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2447 | |
2a0dc6cd LP |
2448 | r = source_set_pending(s, false); |
2449 | if (r < 0) | |
2450 | return r; | |
2576a19e | 2451 | |
2a0dc6cd | 2452 | s->time.next = usec; |
fd38203a | 2453 | |
e1951c16 | 2454 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
2455 | return 0; |
2456 | } | |
2457 | ||
d6a83dc4 LP |
2458 | _public_ int sd_event_source_set_time_relative(sd_event_source *s, uint64_t usec) { |
2459 | usec_t t; | |
2460 | int r; | |
2461 | ||
2462 | assert_return(s, -EINVAL); | |
2463 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2464 | ||
2465 | r = sd_event_now(s->event, event_source_type_to_clock(s->type), &t); | |
2466 | if (r < 0) | |
2467 | return r; | |
2468 | ||
2469 | if (usec >= USEC_INFINITY - t) | |
2470 | return -EOVERFLOW; | |
2471 | ||
2472 | return sd_event_source_set_time(s, t + usec); | |
2473 | } | |
2474 | ||
f7262a9f | 2475 | _public_ int sd_event_source_get_time_accuracy(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
2476 | assert_return(s, -EINVAL); |
2477 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 2478 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf LP |
2479 | assert_return(!event_pid_changed(s->event), -ECHILD); |
2480 | ||
2481 | *usec = s->time.accuracy; | |
2482 | return 0; | |
2483 | } | |
2484 | ||
f7262a9f | 2485 | _public_ int sd_event_source_set_time_accuracy(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 2486 | int r; |
6a0f1f6d | 2487 | |
305f78bf LP |
2488 | assert_return(s, -EINVAL); |
2489 | assert_return(usec != (uint64_t) -1, -EINVAL); | |
6a0f1f6d | 2490 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 2491 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2492 | assert_return(!event_pid_changed(s->event), -ECHILD); |
eaa3cbef | 2493 | |
2a0dc6cd LP |
2494 | r = source_set_pending(s, false); |
2495 | if (r < 0) | |
2496 | return r; | |
2497 | ||
eaa3cbef LP |
2498 | if (usec == 0) |
2499 | usec = DEFAULT_ACCURACY_USEC; | |
2500 | ||
eaa3cbef LP |
2501 | s->time.accuracy = usec; |
2502 | ||
e1951c16 | 2503 | event_source_time_prioq_reshuffle(s); |
6a0f1f6d LP |
2504 | return 0; |
2505 | } | |
2506 | ||
2507 | _public_ int sd_event_source_get_time_clock(sd_event_source *s, clockid_t *clock) { | |
2508 | assert_return(s, -EINVAL); | |
2509 | assert_return(clock, -EINVAL); | |
2510 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2511 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
eaa3cbef | 2512 | |
6a0f1f6d | 2513 | *clock = event_source_type_to_clock(s->type); |
eaa3cbef LP |
2514 | return 0; |
2515 | } | |
2516 | ||
f7262a9f | 2517 | _public_ int sd_event_source_get_child_pid(sd_event_source *s, pid_t *pid) { |
4bee8012 LP |
2518 | assert_return(s, -EINVAL); |
2519 | assert_return(pid, -EINVAL); | |
2520 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2521 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2522 | ||
2523 | *pid = s->child.pid; | |
2524 | return 0; | |
2525 | } | |
2526 | ||
f8f3f926 LP |
2527 | _public_ int sd_event_source_get_child_pidfd(sd_event_source *s) { |
2528 | assert_return(s, -EINVAL); | |
2529 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2530 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2531 | ||
2532 | if (s->child.pidfd < 0) | |
2533 | return -EOPNOTSUPP; | |
2534 | ||
2535 | return s->child.pidfd; | |
2536 | } | |
2537 | ||
2538 | _public_ int sd_event_source_send_child_signal(sd_event_source *s, int sig, const siginfo_t *si, unsigned flags) { | |
2539 | assert_return(s, -EINVAL); | |
2540 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2541 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2542 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
2543 | ||
2544 | /* If we already have seen indication the process exited refuse sending a signal early. This way we | |
2545 | * can be sure we don't accidentally kill the wrong process on PID reuse when pidfds are not | |
2546 | * available. */ | |
2547 | if (s->child.exited) | |
2548 | return -ESRCH; | |
2549 | ||
2550 | if (s->child.pidfd >= 0) { | |
2551 | siginfo_t copy; | |
2552 | ||
2553 | /* pidfd_send_signal() changes the siginfo_t argument. This is weird, let's hence copy the | |
2554 | * structure here */ | |
2555 | if (si) | |
2556 | copy = *si; | |
2557 | ||
2558 | if (pidfd_send_signal(s->child.pidfd, sig, si ? © : NULL, 0) < 0) { | |
2559 | /* Let's propagate the error only if the system call is not implemented or prohibited */ | |
2560 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
2561 | return -errno; | |
2562 | } else | |
2563 | return 0; | |
2564 | } | |
2565 | ||
2566 | /* Flags are only supported for pidfd_send_signal(), not for rt_sigqueueinfo(), hence let's refuse | |
2567 | * this here. */ | |
2568 | if (flags != 0) | |
2569 | return -EOPNOTSUPP; | |
2570 | ||
2571 | if (si) { | |
2572 | /* We use rt_sigqueueinfo() only if siginfo_t is specified. */ | |
2573 | siginfo_t copy = *si; | |
2574 | ||
2575 | if (rt_sigqueueinfo(s->child.pid, sig, ©) < 0) | |
2576 | return -errno; | |
2577 | } else if (kill(s->child.pid, sig) < 0) | |
2578 | return -errno; | |
2579 | ||
2580 | return 0; | |
2581 | } | |
2582 | ||
2583 | _public_ int sd_event_source_get_child_pidfd_own(sd_event_source *s) { | |
2584 | assert_return(s, -EINVAL); | |
2585 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2586 | ||
2587 | if (s->child.pidfd < 0) | |
2588 | return -EOPNOTSUPP; | |
2589 | ||
2590 | return s->child.pidfd_owned; | |
2591 | } | |
2592 | ||
2593 | _public_ int sd_event_source_set_child_pidfd_own(sd_event_source *s, int own) { | |
2594 | assert_return(s, -EINVAL); | |
2595 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2596 | ||
2597 | if (s->child.pidfd < 0) | |
2598 | return -EOPNOTSUPP; | |
2599 | ||
2600 | s->child.pidfd_owned = own; | |
2601 | return 0; | |
2602 | } | |
2603 | ||
2604 | _public_ int sd_event_source_get_child_process_own(sd_event_source *s) { | |
2605 | assert_return(s, -EINVAL); | |
2606 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2607 | ||
2608 | return s->child.process_owned; | |
2609 | } | |
2610 | ||
2611 | _public_ int sd_event_source_set_child_process_own(sd_event_source *s, int own) { | |
2612 | assert_return(s, -EINVAL); | |
2613 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2614 | ||
2615 | s->child.process_owned = own; | |
2616 | return 0; | |
2617 | } | |
2618 | ||
97ef5391 LP |
2619 | _public_ int sd_event_source_get_inotify_mask(sd_event_source *s, uint32_t *mask) { |
2620 | assert_return(s, -EINVAL); | |
2621 | assert_return(mask, -EINVAL); | |
2622 | assert_return(s->type == SOURCE_INOTIFY, -EDOM); | |
2623 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2624 | ||
2625 | *mask = s->inotify.mask; | |
2626 | return 0; | |
2627 | } | |
2628 | ||
718db961 | 2629 | _public_ int sd_event_source_set_prepare(sd_event_source *s, sd_event_handler_t callback) { |
fd38203a LP |
2630 | int r; |
2631 | ||
da7e457c | 2632 | assert_return(s, -EINVAL); |
6203e07a | 2633 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c LP |
2634 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2635 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2636 | |
2637 | if (s->prepare == callback) | |
2638 | return 0; | |
2639 | ||
2640 | if (callback && s->prepare) { | |
2641 | s->prepare = callback; | |
2642 | return 0; | |
2643 | } | |
2644 | ||
2645 | r = prioq_ensure_allocated(&s->event->prepare, prepare_prioq_compare); | |
2646 | if (r < 0) | |
2647 | return r; | |
2648 | ||
2649 | s->prepare = callback; | |
2650 | ||
2651 | if (callback) { | |
2652 | r = prioq_put(s->event->prepare, s, &s->prepare_index); | |
2653 | if (r < 0) | |
2654 | return r; | |
2655 | } else | |
2656 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
2657 | ||
2658 | return 0; | |
2659 | } | |
2660 | ||
f7262a9f | 2661 | _public_ void* sd_event_source_get_userdata(sd_event_source *s) { |
da7e457c | 2662 | assert_return(s, NULL); |
fd38203a LP |
2663 | |
2664 | return s->userdata; | |
2665 | } | |
2666 | ||
8f726607 LP |
2667 | _public_ void *sd_event_source_set_userdata(sd_event_source *s, void *userdata) { |
2668 | void *ret; | |
2669 | ||
2670 | assert_return(s, NULL); | |
2671 | ||
2672 | ret = s->userdata; | |
2673 | s->userdata = userdata; | |
2674 | ||
2675 | return ret; | |
2676 | } | |
2677 | ||
c2ba3ad6 LP |
2678 | static usec_t sleep_between(sd_event *e, usec_t a, usec_t b) { |
2679 | usec_t c; | |
2680 | assert(e); | |
2681 | assert(a <= b); | |
2682 | ||
2683 | if (a <= 0) | |
2684 | return 0; | |
393003e1 LP |
2685 | if (a >= USEC_INFINITY) |
2686 | return USEC_INFINITY; | |
c2ba3ad6 LP |
2687 | |
2688 | if (b <= a + 1) | |
2689 | return a; | |
2690 | ||
52444dc4 LP |
2691 | initialize_perturb(e); |
2692 | ||
c2ba3ad6 LP |
2693 | /* |
2694 | Find a good time to wake up again between times a and b. We | |
2695 | have two goals here: | |
2696 | ||
2697 | a) We want to wake up as seldom as possible, hence prefer | |
2698 | later times over earlier times. | |
2699 | ||
2700 | b) But if we have to wake up, then let's make sure to | |
2701 | dispatch as much as possible on the entire system. | |
2702 | ||
2703 | We implement this by waking up everywhere at the same time | |
850516e0 | 2704 | within any given minute if we can, synchronised via the |
c2ba3ad6 | 2705 | perturbation value determined from the boot ID. If we can't, |
ba276c81 LP |
2706 | then we try to find the same spot in every 10s, then 1s and |
2707 | then 250ms step. Otherwise, we pick the last possible time | |
2708 | to wake up. | |
c2ba3ad6 LP |
2709 | */ |
2710 | ||
850516e0 LP |
2711 | c = (b / USEC_PER_MINUTE) * USEC_PER_MINUTE + e->perturb; |
2712 | if (c >= b) { | |
2713 | if (_unlikely_(c < USEC_PER_MINUTE)) | |
2714 | return b; | |
2715 | ||
2716 | c -= USEC_PER_MINUTE; | |
2717 | } | |
2718 | ||
ba276c81 LP |
2719 | if (c >= a) |
2720 | return c; | |
2721 | ||
2722 | c = (b / (USEC_PER_SEC*10)) * (USEC_PER_SEC*10) + (e->perturb % (USEC_PER_SEC*10)); | |
2723 | if (c >= b) { | |
2724 | if (_unlikely_(c < USEC_PER_SEC*10)) | |
2725 | return b; | |
2726 | ||
2727 | c -= USEC_PER_SEC*10; | |
2728 | } | |
2729 | ||
850516e0 LP |
2730 | if (c >= a) |
2731 | return c; | |
2732 | ||
2733 | c = (b / USEC_PER_SEC) * USEC_PER_SEC + (e->perturb % USEC_PER_SEC); | |
c2ba3ad6 LP |
2734 | if (c >= b) { |
2735 | if (_unlikely_(c < USEC_PER_SEC)) | |
2736 | return b; | |
2737 | ||
2738 | c -= USEC_PER_SEC; | |
2739 | } | |
2740 | ||
2741 | if (c >= a) | |
2742 | return c; | |
2743 | ||
2744 | c = (b / (USEC_PER_MSEC*250)) * (USEC_PER_MSEC*250) + (e->perturb % (USEC_PER_MSEC*250)); | |
2745 | if (c >= b) { | |
2746 | if (_unlikely_(c < USEC_PER_MSEC*250)) | |
2747 | return b; | |
2748 | ||
2749 | c -= USEC_PER_MSEC*250; | |
2750 | } | |
2751 | ||
2752 | if (c >= a) | |
2753 | return c; | |
2754 | ||
2755 | return b; | |
2756 | } | |
2757 | ||
fd38203a LP |
2758 | static int event_arm_timer( |
2759 | sd_event *e, | |
6a0f1f6d | 2760 | struct clock_data *d) { |
fd38203a LP |
2761 | |
2762 | struct itimerspec its = {}; | |
c2ba3ad6 LP |
2763 | sd_event_source *a, *b; |
2764 | usec_t t; | |
fd38203a | 2765 | |
cde93897 | 2766 | assert(e); |
6a0f1f6d | 2767 | assert(d); |
fd38203a | 2768 | |
d06441da | 2769 | if (!d->needs_rearm) |
212bbb17 TG |
2770 | return 0; |
2771 | else | |
2772 | d->needs_rearm = false; | |
2773 | ||
6a0f1f6d | 2774 | a = prioq_peek(d->earliest); |
393003e1 | 2775 | if (!a || a->enabled == SD_EVENT_OFF || a->time.next == USEC_INFINITY) { |
72aedc1e | 2776 | |
6a0f1f6d | 2777 | if (d->fd < 0) |
c57b5ca3 LP |
2778 | return 0; |
2779 | ||
3a43da28 | 2780 | if (d->next == USEC_INFINITY) |
72aedc1e LP |
2781 | return 0; |
2782 | ||
2783 | /* disarm */ | |
15c689d7 LP |
2784 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
2785 | return -errno; | |
72aedc1e | 2786 | |
3a43da28 | 2787 | d->next = USEC_INFINITY; |
fd38203a | 2788 | return 0; |
72aedc1e | 2789 | } |
fd38203a | 2790 | |
6a0f1f6d | 2791 | b = prioq_peek(d->latest); |
baf76283 | 2792 | assert_se(b && b->enabled != SD_EVENT_OFF); |
c2ba3ad6 | 2793 | |
1bce0ffa | 2794 | t = sleep_between(e, a->time.next, time_event_source_latest(b)); |
6a0f1f6d | 2795 | if (d->next == t) |
fd38203a LP |
2796 | return 0; |
2797 | ||
6a0f1f6d | 2798 | assert_se(d->fd >= 0); |
fd38203a | 2799 | |
c2ba3ad6 | 2800 | if (t == 0) { |
fd38203a LP |
2801 | /* We don' want to disarm here, just mean some time looooong ago. */ |
2802 | its.it_value.tv_sec = 0; | |
2803 | its.it_value.tv_nsec = 1; | |
2804 | } else | |
c2ba3ad6 | 2805 | timespec_store(&its.it_value, t); |
fd38203a | 2806 | |
15c689d7 | 2807 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 | 2808 | return -errno; |
fd38203a | 2809 | |
6a0f1f6d | 2810 | d->next = t; |
fd38203a LP |
2811 | return 0; |
2812 | } | |
2813 | ||
9a800b56 | 2814 | static int process_io(sd_event *e, sd_event_source *s, uint32_t revents) { |
fd38203a LP |
2815 | assert(e); |
2816 | assert(s); | |
2817 | assert(s->type == SOURCE_IO); | |
2818 | ||
9a800b56 LP |
2819 | /* If the event source was already pending, we just OR in the |
2820 | * new revents, otherwise we reset the value. The ORing is | |
2821 | * necessary to handle EPOLLONESHOT events properly where | |
2822 | * readability might happen independently of writability, and | |
2823 | * we need to keep track of both */ | |
2824 | ||
2825 | if (s->pending) | |
2826 | s->io.revents |= revents; | |
2827 | else | |
2828 | s->io.revents = revents; | |
fd38203a | 2829 | |
fd38203a LP |
2830 | return source_set_pending(s, true); |
2831 | } | |
2832 | ||
72aedc1e | 2833 | static int flush_timer(sd_event *e, int fd, uint32_t events, usec_t *next) { |
fd38203a LP |
2834 | uint64_t x; |
2835 | ssize_t ss; | |
2836 | ||
2837 | assert(e); | |
da7e457c | 2838 | assert(fd >= 0); |
72aedc1e | 2839 | |
305f78bf | 2840 | assert_return(events == EPOLLIN, -EIO); |
fd38203a LP |
2841 | |
2842 | ss = read(fd, &x, sizeof(x)); | |
2843 | if (ss < 0) { | |
945c2931 | 2844 | if (IN_SET(errno, EAGAIN, EINTR)) |
fd38203a LP |
2845 | return 0; |
2846 | ||
2847 | return -errno; | |
2848 | } | |
2849 | ||
8d35dae7 | 2850 | if (_unlikely_(ss != sizeof(x))) |
fd38203a LP |
2851 | return -EIO; |
2852 | ||
cde93897 | 2853 | if (next) |
3a43da28 | 2854 | *next = USEC_INFINITY; |
72aedc1e | 2855 | |
fd38203a LP |
2856 | return 0; |
2857 | } | |
2858 | ||
305f78bf LP |
2859 | static int process_timer( |
2860 | sd_event *e, | |
2861 | usec_t n, | |
6a0f1f6d | 2862 | struct clock_data *d) { |
305f78bf | 2863 | |
fd38203a LP |
2864 | sd_event_source *s; |
2865 | int r; | |
2866 | ||
2867 | assert(e); | |
6a0f1f6d | 2868 | assert(d); |
fd38203a LP |
2869 | |
2870 | for (;;) { | |
6a0f1f6d | 2871 | s = prioq_peek(d->earliest); |
fd38203a LP |
2872 | if (!s || |
2873 | s->time.next > n || | |
baf76283 | 2874 | s->enabled == SD_EVENT_OFF || |
fd38203a LP |
2875 | s->pending) |
2876 | break; | |
2877 | ||
2878 | r = source_set_pending(s, true); | |
2879 | if (r < 0) | |
2880 | return r; | |
2881 | ||
e1951c16 | 2882 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
2883 | } |
2884 | ||
2885 | return 0; | |
2886 | } | |
2887 | ||
2888 | static int process_child(sd_event *e) { | |
2889 | sd_event_source *s; | |
fd38203a LP |
2890 | int r; |
2891 | ||
2892 | assert(e); | |
2893 | ||
c2ba3ad6 LP |
2894 | e->need_process_child = false; |
2895 | ||
fd38203a LP |
2896 | /* |
2897 | So, this is ugly. We iteratively invoke waitid() with P_PID | |
2898 | + WNOHANG for each PID we wait for, instead of using | |
2899 | P_ALL. This is because we only want to get child | |
2900 | information of very specific child processes, and not all | |
2901 | of them. We might not have processed the SIGCHLD even of a | |
2902 | previous invocation and we don't want to maintain a | |
2903 | unbounded *per-child* event queue, hence we really don't | |
2904 | want anything flushed out of the kernel's queue that we | |
2905 | don't care about. Since this is O(n) this means that if you | |
2906 | have a lot of processes you probably want to handle SIGCHLD | |
2907 | yourself. | |
08cd1552 LP |
2908 | |
2909 | We do not reap the children here (by using WNOWAIT), this | |
2910 | is only done after the event source is dispatched so that | |
2911 | the callback still sees the process as a zombie. | |
fd38203a LP |
2912 | */ |
2913 | ||
90e74a66 | 2914 | HASHMAP_FOREACH(s, e->child_sources) { |
fd38203a LP |
2915 | assert(s->type == SOURCE_CHILD); |
2916 | ||
2917 | if (s->pending) | |
2918 | continue; | |
2919 | ||
baf76283 | 2920 | if (s->enabled == SD_EVENT_OFF) |
fd38203a LP |
2921 | continue; |
2922 | ||
f8f3f926 LP |
2923 | if (s->child.exited) |
2924 | continue; | |
2925 | ||
2926 | if (EVENT_SOURCE_WATCH_PIDFD(s)) /* There's a usable pidfd known for this event source? then don't waitid() for it here */ | |
2927 | continue; | |
2928 | ||
fd38203a | 2929 | zero(s->child.siginfo); |
15c689d7 LP |
2930 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, |
2931 | WNOHANG | (s->child.options & WEXITED ? WNOWAIT : 0) | s->child.options) < 0) | |
fd38203a LP |
2932 | return -errno; |
2933 | ||
2934 | if (s->child.siginfo.si_pid != 0) { | |
945c2931 | 2935 | bool zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 2936 | |
f8f3f926 LP |
2937 | if (zombie) |
2938 | s->child.exited = true; | |
2939 | ||
08cd1552 LP |
2940 | if (!zombie && (s->child.options & WEXITED)) { |
2941 | /* If the child isn't dead then let's | |
2942 | * immediately remove the state change | |
2943 | * from the queue, since there's no | |
2944 | * benefit in leaving it queued */ | |
2945 | ||
2946 | assert(s->child.options & (WSTOPPED|WCONTINUED)); | |
a5d27871 | 2947 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|(s->child.options & (WSTOPPED|WCONTINUED))); |
08cd1552 LP |
2948 | } |
2949 | ||
fd38203a LP |
2950 | r = source_set_pending(s, true); |
2951 | if (r < 0) | |
2952 | return r; | |
2953 | } | |
2954 | } | |
2955 | ||
fd38203a LP |
2956 | return 0; |
2957 | } | |
2958 | ||
f8f3f926 LP |
2959 | static int process_pidfd(sd_event *e, sd_event_source *s, uint32_t revents) { |
2960 | assert(e); | |
2961 | assert(s); | |
2962 | assert(s->type == SOURCE_CHILD); | |
2963 | ||
2964 | if (s->pending) | |
2965 | return 0; | |
2966 | ||
2967 | if (s->enabled == SD_EVENT_OFF) | |
2968 | return 0; | |
2969 | ||
2970 | if (!EVENT_SOURCE_WATCH_PIDFD(s)) | |
2971 | return 0; | |
2972 | ||
2973 | zero(s->child.siginfo); | |
2974 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG | WNOWAIT | s->child.options) < 0) | |
2975 | return -errno; | |
2976 | ||
2977 | if (s->child.siginfo.si_pid == 0) | |
2978 | return 0; | |
2979 | ||
2980 | if (IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) | |
2981 | s->child.exited = true; | |
2982 | ||
2983 | return source_set_pending(s, true); | |
2984 | } | |
2985 | ||
9da4cb2b | 2986 | static int process_signal(sd_event *e, struct signal_data *d, uint32_t events) { |
fd38203a | 2987 | bool read_one = false; |
fd38203a LP |
2988 | int r; |
2989 | ||
da7e457c | 2990 | assert(e); |
97ef5391 | 2991 | assert(d); |
305f78bf | 2992 | assert_return(events == EPOLLIN, -EIO); |
fd38203a | 2993 | |
9da4cb2b LP |
2994 | /* If there's a signal queued on this priority and SIGCHLD is |
2995 | on this priority too, then make sure to recheck the | |
2996 | children we watch. This is because we only ever dequeue | |
2997 | the first signal per priority, and if we dequeue one, and | |
2998 | SIGCHLD might be enqueued later we wouldn't know, but we | |
2999 | might have higher priority children we care about hence we | |
3000 | need to check that explicitly. */ | |
3001 | ||
3002 | if (sigismember(&d->sigset, SIGCHLD)) | |
3003 | e->need_process_child = true; | |
3004 | ||
3005 | /* If there's already an event source pending for this | |
3006 | * priority we don't read another */ | |
3007 | if (d->current) | |
3008 | return 0; | |
3009 | ||
fd38203a | 3010 | for (;;) { |
0eb2e0e3 | 3011 | struct signalfd_siginfo si; |
7057bd99 | 3012 | ssize_t n; |
92daebc0 | 3013 | sd_event_source *s = NULL; |
fd38203a | 3014 | |
9da4cb2b | 3015 | n = read(d->fd, &si, sizeof(si)); |
7057bd99 | 3016 | if (n < 0) { |
945c2931 | 3017 | if (IN_SET(errno, EAGAIN, EINTR)) |
fd38203a LP |
3018 | return read_one; |
3019 | ||
3020 | return -errno; | |
3021 | } | |
3022 | ||
7057bd99 | 3023 | if (_unlikely_(n != sizeof(si))) |
fd38203a LP |
3024 | return -EIO; |
3025 | ||
6eb7c172 | 3026 | assert(SIGNAL_VALID(si.ssi_signo)); |
7057bd99 | 3027 | |
fd38203a LP |
3028 | read_one = true; |
3029 | ||
92daebc0 LP |
3030 | if (e->signal_sources) |
3031 | s = e->signal_sources[si.ssi_signo]; | |
92daebc0 LP |
3032 | if (!s) |
3033 | continue; | |
9da4cb2b LP |
3034 | if (s->pending) |
3035 | continue; | |
fd38203a LP |
3036 | |
3037 | s->signal.siginfo = si; | |
9da4cb2b LP |
3038 | d->current = s; |
3039 | ||
fd38203a LP |
3040 | r = source_set_pending(s, true); |
3041 | if (r < 0) | |
3042 | return r; | |
9da4cb2b LP |
3043 | |
3044 | return 1; | |
fd38203a | 3045 | } |
fd38203a LP |
3046 | } |
3047 | ||
97ef5391 LP |
3048 | static int event_inotify_data_read(sd_event *e, struct inotify_data *d, uint32_t revents) { |
3049 | ssize_t n; | |
3050 | ||
3051 | assert(e); | |
3052 | assert(d); | |
3053 | ||
3054 | assert_return(revents == EPOLLIN, -EIO); | |
3055 | ||
3056 | /* If there's already an event source pending for this priority, don't read another */ | |
3057 | if (d->n_pending > 0) | |
3058 | return 0; | |
3059 | ||
3060 | /* Is the read buffer non-empty? If so, let's not read more */ | |
3061 | if (d->buffer_filled > 0) | |
3062 | return 0; | |
3063 | ||
3064 | n = read(d->fd, &d->buffer, sizeof(d->buffer)); | |
3065 | if (n < 0) { | |
3066 | if (IN_SET(errno, EAGAIN, EINTR)) | |
3067 | return 0; | |
3068 | ||
3069 | return -errno; | |
3070 | } | |
3071 | ||
3072 | assert(n > 0); | |
3073 | d->buffer_filled = (size_t) n; | |
3074 | LIST_PREPEND(buffered, e->inotify_data_buffered, d); | |
3075 | ||
3076 | return 1; | |
3077 | } | |
3078 | ||
3079 | static void event_inotify_data_drop(sd_event *e, struct inotify_data *d, size_t sz) { | |
3080 | assert(e); | |
3081 | assert(d); | |
3082 | assert(sz <= d->buffer_filled); | |
3083 | ||
3084 | if (sz == 0) | |
3085 | return; | |
3086 | ||
3087 | /* Move the rest to the buffer to the front, in order to get things properly aligned again */ | |
3088 | memmove(d->buffer.raw, d->buffer.raw + sz, d->buffer_filled - sz); | |
3089 | d->buffer_filled -= sz; | |
3090 | ||
3091 | if (d->buffer_filled == 0) | |
3092 | LIST_REMOVE(buffered, e->inotify_data_buffered, d); | |
3093 | } | |
3094 | ||
3095 | static int event_inotify_data_process(sd_event *e, struct inotify_data *d) { | |
3096 | int r; | |
3097 | ||
3098 | assert(e); | |
3099 | assert(d); | |
3100 | ||
3101 | /* If there's already an event source pending for this priority, don't read another */ | |
3102 | if (d->n_pending > 0) | |
3103 | return 0; | |
3104 | ||
3105 | while (d->buffer_filled > 0) { | |
3106 | size_t sz; | |
3107 | ||
3108 | /* Let's validate that the event structures are complete */ | |
3109 | if (d->buffer_filled < offsetof(struct inotify_event, name)) | |
3110 | return -EIO; | |
3111 | ||
3112 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3113 | if (d->buffer_filled < sz) | |
3114 | return -EIO; | |
3115 | ||
3116 | if (d->buffer.ev.mask & IN_Q_OVERFLOW) { | |
3117 | struct inode_data *inode_data; | |
97ef5391 LP |
3118 | |
3119 | /* The queue overran, let's pass this event to all event sources connected to this inotify | |
3120 | * object */ | |
3121 | ||
90e74a66 | 3122 | HASHMAP_FOREACH(inode_data, d->inodes) { |
97ef5391 LP |
3123 | sd_event_source *s; |
3124 | ||
3125 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
3126 | ||
3127 | if (s->enabled == SD_EVENT_OFF) | |
3128 | continue; | |
3129 | ||
3130 | r = source_set_pending(s, true); | |
3131 | if (r < 0) | |
3132 | return r; | |
3133 | } | |
3134 | } | |
3135 | } else { | |
3136 | struct inode_data *inode_data; | |
3137 | sd_event_source *s; | |
3138 | ||
3139 | /* Find the inode object for this watch descriptor. If IN_IGNORED is set we also remove it from | |
3140 | * our watch descriptor table. */ | |
3141 | if (d->buffer.ev.mask & IN_IGNORED) { | |
3142 | ||
3143 | inode_data = hashmap_remove(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3144 | if (!inode_data) { | |
3145 | event_inotify_data_drop(e, d, sz); | |
3146 | continue; | |
3147 | } | |
3148 | ||
3149 | /* The watch descriptor was removed by the kernel, let's drop it here too */ | |
3150 | inode_data->wd = -1; | |
3151 | } else { | |
3152 | inode_data = hashmap_get(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3153 | if (!inode_data) { | |
3154 | event_inotify_data_drop(e, d, sz); | |
3155 | continue; | |
3156 | } | |
3157 | } | |
3158 | ||
3159 | /* Trigger all event sources that are interested in these events. Also trigger all event | |
3160 | * sources if IN_IGNORED or IN_UNMOUNT is set. */ | |
3161 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
3162 | ||
3163 | if (s->enabled == SD_EVENT_OFF) | |
3164 | continue; | |
3165 | ||
3166 | if ((d->buffer.ev.mask & (IN_IGNORED|IN_UNMOUNT)) == 0 && | |
3167 | (s->inotify.mask & d->buffer.ev.mask & IN_ALL_EVENTS) == 0) | |
3168 | continue; | |
3169 | ||
3170 | r = source_set_pending(s, true); | |
3171 | if (r < 0) | |
3172 | return r; | |
3173 | } | |
3174 | } | |
3175 | ||
3176 | /* Something pending now? If so, let's finish, otherwise let's read more. */ | |
3177 | if (d->n_pending > 0) | |
3178 | return 1; | |
3179 | } | |
3180 | ||
3181 | return 0; | |
3182 | } | |
3183 | ||
3184 | static int process_inotify(sd_event *e) { | |
3185 | struct inotify_data *d; | |
3186 | int r, done = 0; | |
3187 | ||
3188 | assert(e); | |
3189 | ||
3190 | LIST_FOREACH(buffered, d, e->inotify_data_buffered) { | |
3191 | r = event_inotify_data_process(e, d); | |
3192 | if (r < 0) | |
3193 | return r; | |
3194 | if (r > 0) | |
3195 | done ++; | |
3196 | } | |
3197 | ||
3198 | return done; | |
3199 | } | |
3200 | ||
fd38203a | 3201 | static int source_dispatch(sd_event_source *s) { |
b778cba4 | 3202 | _cleanup_(sd_event_unrefp) sd_event *saved_event = NULL; |
8f5c235d | 3203 | EventSourceType saved_type; |
fe8245eb | 3204 | int r = 0; |
fd38203a LP |
3205 | |
3206 | assert(s); | |
6203e07a | 3207 | assert(s->pending || s->type == SOURCE_EXIT); |
fd38203a | 3208 | |
b778cba4 LP |
3209 | /* Save the event source type, here, so that we still know it after the event callback which might |
3210 | * invalidate the event. */ | |
8f5c235d LP |
3211 | saved_type = s->type; |
3212 | ||
b778cba4 LP |
3213 | /* Similar, store a reference to the event loop object, so that we can still access it after the |
3214 | * callback might have invalidated/disconnected the event source. */ | |
3215 | saved_event = sd_event_ref(s->event); | |
3216 | ||
945c2931 | 3217 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { |
da7e457c LP |
3218 | r = source_set_pending(s, false); |
3219 | if (r < 0) | |
3220 | return r; | |
3221 | } | |
fd38203a | 3222 | |
6e9feda3 LP |
3223 | if (s->type != SOURCE_POST) { |
3224 | sd_event_source *z; | |
6e9feda3 LP |
3225 | |
3226 | /* If we execute a non-post source, let's mark all | |
3227 | * post sources as pending */ | |
3228 | ||
90e74a66 | 3229 | SET_FOREACH(z, s->event->post_sources) { |
6e9feda3 LP |
3230 | if (z->enabled == SD_EVENT_OFF) |
3231 | continue; | |
3232 | ||
3233 | r = source_set_pending(z, true); | |
3234 | if (r < 0) | |
3235 | return r; | |
3236 | } | |
3237 | } | |
3238 | ||
baf76283 LP |
3239 | if (s->enabled == SD_EVENT_ONESHOT) { |
3240 | r = sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
3241 | if (r < 0) |
3242 | return r; | |
3243 | } | |
3244 | ||
12179984 | 3245 | s->dispatching = true; |
b7484e2a | 3246 | |
fd38203a LP |
3247 | switch (s->type) { |
3248 | ||
3249 | case SOURCE_IO: | |
3250 | r = s->io.callback(s, s->io.fd, s->io.revents, s->userdata); | |
3251 | break; | |
3252 | ||
6a0f1f6d | 3253 | case SOURCE_TIME_REALTIME: |
a8548816 | 3254 | case SOURCE_TIME_BOOTTIME: |
6a0f1f6d LP |
3255 | case SOURCE_TIME_MONOTONIC: |
3256 | case SOURCE_TIME_REALTIME_ALARM: | |
3257 | case SOURCE_TIME_BOOTTIME_ALARM: | |
fd38203a LP |
3258 | r = s->time.callback(s, s->time.next, s->userdata); |
3259 | break; | |
3260 | ||
3261 | case SOURCE_SIGNAL: | |
3262 | r = s->signal.callback(s, &s->signal.siginfo, s->userdata); | |
3263 | break; | |
3264 | ||
08cd1552 LP |
3265 | case SOURCE_CHILD: { |
3266 | bool zombie; | |
3267 | ||
945c2931 | 3268 | zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 3269 | |
fd38203a | 3270 | r = s->child.callback(s, &s->child.siginfo, s->userdata); |
08cd1552 LP |
3271 | |
3272 | /* Now, reap the PID for good. */ | |
f8f3f926 | 3273 | if (zombie) { |
cc59d290 | 3274 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|WEXITED); |
f8f3f926 LP |
3275 | s->child.waited = true; |
3276 | } | |
08cd1552 | 3277 | |
fd38203a | 3278 | break; |
08cd1552 | 3279 | } |
fd38203a LP |
3280 | |
3281 | case SOURCE_DEFER: | |
3282 | r = s->defer.callback(s, s->userdata); | |
3283 | break; | |
da7e457c | 3284 | |
6e9feda3 LP |
3285 | case SOURCE_POST: |
3286 | r = s->post.callback(s, s->userdata); | |
3287 | break; | |
3288 | ||
6203e07a LP |
3289 | case SOURCE_EXIT: |
3290 | r = s->exit.callback(s, s->userdata); | |
da7e457c | 3291 | break; |
9d3e3aa5 | 3292 | |
97ef5391 LP |
3293 | case SOURCE_INOTIFY: { |
3294 | struct sd_event *e = s->event; | |
3295 | struct inotify_data *d; | |
3296 | size_t sz; | |
3297 | ||
3298 | assert(s->inotify.inode_data); | |
3299 | assert_se(d = s->inotify.inode_data->inotify_data); | |
3300 | ||
3301 | assert(d->buffer_filled >= offsetof(struct inotify_event, name)); | |
3302 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3303 | assert(d->buffer_filled >= sz); | |
3304 | ||
3305 | r = s->inotify.callback(s, &d->buffer.ev, s->userdata); | |
3306 | ||
3307 | /* When no event is pending anymore on this inotify object, then let's drop the event from the | |
3308 | * buffer. */ | |
3309 | if (d->n_pending == 0) | |
3310 | event_inotify_data_drop(e, d, sz); | |
3311 | ||
3312 | break; | |
3313 | } | |
3314 | ||
9d3e3aa5 | 3315 | case SOURCE_WATCHDOG: |
a71fe8b8 | 3316 | case _SOURCE_EVENT_SOURCE_TYPE_MAX: |
9f2a50a3 | 3317 | case _SOURCE_EVENT_SOURCE_TYPE_INVALID: |
9d3e3aa5 | 3318 | assert_not_reached("Wut? I shouldn't exist."); |
fd38203a LP |
3319 | } |
3320 | ||
12179984 LP |
3321 | s->dispatching = false; |
3322 | ||
b778cba4 LP |
3323 | if (r < 0) { |
3324 | log_debug_errno(r, "Event source %s (type %s) returned error, %s: %m", | |
3325 | strna(s->description), | |
3326 | event_source_type_to_string(saved_type), | |
3327 | s->exit_on_failure ? "exiting" : "disabling"); | |
3328 | ||
3329 | if (s->exit_on_failure) | |
3330 | (void) sd_event_exit(saved_event, r); | |
3331 | } | |
12179984 LP |
3332 | |
3333 | if (s->n_ref == 0) | |
3334 | source_free(s); | |
3335 | else if (r < 0) | |
6203e07a | 3336 | sd_event_source_set_enabled(s, SD_EVENT_OFF); |
b7484e2a | 3337 | |
6203e07a | 3338 | return 1; |
fd38203a LP |
3339 | } |
3340 | ||
3341 | static int event_prepare(sd_event *e) { | |
3342 | int r; | |
3343 | ||
3344 | assert(e); | |
3345 | ||
3346 | for (;;) { | |
3347 | sd_event_source *s; | |
3348 | ||
3349 | s = prioq_peek(e->prepare); | |
baf76283 | 3350 | if (!s || s->prepare_iteration == e->iteration || s->enabled == SD_EVENT_OFF) |
fd38203a LP |
3351 | break; |
3352 | ||
3353 | s->prepare_iteration = e->iteration; | |
3354 | r = prioq_reshuffle(e->prepare, s, &s->prepare_index); | |
3355 | if (r < 0) | |
3356 | return r; | |
3357 | ||
3358 | assert(s->prepare); | |
12179984 LP |
3359 | |
3360 | s->dispatching = true; | |
fd38203a | 3361 | r = s->prepare(s, s->userdata); |
12179984 LP |
3362 | s->dispatching = false; |
3363 | ||
b778cba4 LP |
3364 | if (r < 0) { |
3365 | log_debug_errno(r, "Prepare callback of event source %s (type %s) returned error, %s: %m", | |
3366 | strna(s->description), | |
3367 | event_source_type_to_string(s->type), | |
3368 | s->exit_on_failure ? "exiting" : "disabling"); | |
3369 | ||
3370 | if (s->exit_on_failure) | |
3371 | (void) sd_event_exit(e, r); | |
3372 | } | |
fd38203a | 3373 | |
12179984 LP |
3374 | if (s->n_ref == 0) |
3375 | source_free(s); | |
3376 | else if (r < 0) | |
3377 | sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
3378 | } |
3379 | ||
3380 | return 0; | |
3381 | } | |
3382 | ||
6203e07a | 3383 | static int dispatch_exit(sd_event *e) { |
da7e457c | 3384 | sd_event_source *p; |
30dd293c | 3385 | _cleanup_(sd_event_unrefp) sd_event *ref = NULL; |
da7e457c LP |
3386 | int r; |
3387 | ||
3388 | assert(e); | |
3389 | ||
6203e07a | 3390 | p = prioq_peek(e->exit); |
baf76283 | 3391 | if (!p || p->enabled == SD_EVENT_OFF) { |
da7e457c LP |
3392 | e->state = SD_EVENT_FINISHED; |
3393 | return 0; | |
3394 | } | |
3395 | ||
30dd293c | 3396 | ref = sd_event_ref(e); |
da7e457c | 3397 | e->iteration++; |
6203e07a | 3398 | e->state = SD_EVENT_EXITING; |
da7e457c | 3399 | r = source_dispatch(p); |
2b0c9ef7 | 3400 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
3401 | return r; |
3402 | } | |
3403 | ||
c2ba3ad6 LP |
3404 | static sd_event_source* event_next_pending(sd_event *e) { |
3405 | sd_event_source *p; | |
3406 | ||
da7e457c LP |
3407 | assert(e); |
3408 | ||
c2ba3ad6 LP |
3409 | p = prioq_peek(e->pending); |
3410 | if (!p) | |
3411 | return NULL; | |
3412 | ||
baf76283 | 3413 | if (p->enabled == SD_EVENT_OFF) |
c2ba3ad6 LP |
3414 | return NULL; |
3415 | ||
3416 | return p; | |
3417 | } | |
3418 | ||
cde93897 LP |
3419 | static int arm_watchdog(sd_event *e) { |
3420 | struct itimerspec its = {}; | |
3421 | usec_t t; | |
cde93897 LP |
3422 | |
3423 | assert(e); | |
3424 | assert(e->watchdog_fd >= 0); | |
3425 | ||
3426 | t = sleep_between(e, | |
3427 | e->watchdog_last + (e->watchdog_period / 2), | |
3428 | e->watchdog_last + (e->watchdog_period * 3 / 4)); | |
3429 | ||
3430 | timespec_store(&its.it_value, t); | |
3431 | ||
75145780 LP |
3432 | /* Make sure we never set the watchdog to 0, which tells the |
3433 | * kernel to disable it. */ | |
3434 | if (its.it_value.tv_sec == 0 && its.it_value.tv_nsec == 0) | |
3435 | its.it_value.tv_nsec = 1; | |
3436 | ||
15c689d7 | 3437 | if (timerfd_settime(e->watchdog_fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 LP |
3438 | return -errno; |
3439 | ||
3440 | return 0; | |
3441 | } | |
3442 | ||
3443 | static int process_watchdog(sd_event *e) { | |
3444 | assert(e); | |
3445 | ||
3446 | if (!e->watchdog) | |
3447 | return 0; | |
3448 | ||
3449 | /* Don't notify watchdog too often */ | |
3450 | if (e->watchdog_last + e->watchdog_period / 4 > e->timestamp.monotonic) | |
3451 | return 0; | |
3452 | ||
3453 | sd_notify(false, "WATCHDOG=1"); | |
3454 | e->watchdog_last = e->timestamp.monotonic; | |
3455 | ||
3456 | return arm_watchdog(e); | |
3457 | } | |
3458 | ||
97ef5391 LP |
3459 | static void event_close_inode_data_fds(sd_event *e) { |
3460 | struct inode_data *d; | |
3461 | ||
3462 | assert(e); | |
3463 | ||
3464 | /* Close the fds pointing to the inodes to watch now. We need to close them as they might otherwise pin | |
3465 | * filesystems. But we can't close them right-away as we need them as long as the user still wants to make | |
5238e957 | 3466 | * adjustments to the even source, such as changing the priority (which requires us to remove and re-add a watch |
97ef5391 LP |
3467 | * for the inode). Hence, let's close them when entering the first iteration after they were added, as a |
3468 | * compromise. */ | |
3469 | ||
3470 | while ((d = e->inode_data_to_close)) { | |
3471 | assert(d->fd >= 0); | |
3472 | d->fd = safe_close(d->fd); | |
3473 | ||
3474 | LIST_REMOVE(to_close, e->inode_data_to_close, d); | |
3475 | } | |
3476 | } | |
3477 | ||
c45a5a74 TG |
3478 | _public_ int sd_event_prepare(sd_event *e) { |
3479 | int r; | |
fd38203a | 3480 | |
da7e457c | 3481 | assert_return(e, -EINVAL); |
b937d761 | 3482 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
3483 | assert_return(!event_pid_changed(e), -ECHILD); |
3484 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 3485 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 3486 | |
e5446015 LP |
3487 | /* Let's check that if we are a default event loop we are executed in the correct thread. We only do |
3488 | * this check here once, since gettid() is typically not cached, and thus want to minimize | |
3489 | * syscalls */ | |
3490 | assert_return(!e->default_event_ptr || e->tid == gettid(), -EREMOTEIO); | |
3491 | ||
6203e07a | 3492 | if (e->exit_requested) |
c45a5a74 | 3493 | goto pending; |
fd38203a LP |
3494 | |
3495 | e->iteration++; | |
3496 | ||
0be6c2f6 | 3497 | e->state = SD_EVENT_PREPARING; |
fd38203a | 3498 | r = event_prepare(e); |
0be6c2f6 | 3499 | e->state = SD_EVENT_INITIAL; |
fd38203a | 3500 | if (r < 0) |
c45a5a74 | 3501 | return r; |
fd38203a | 3502 | |
6a0f1f6d LP |
3503 | r = event_arm_timer(e, &e->realtime); |
3504 | if (r < 0) | |
c45a5a74 | 3505 | return r; |
6a0f1f6d | 3506 | |
a8548816 TG |
3507 | r = event_arm_timer(e, &e->boottime); |
3508 | if (r < 0) | |
c45a5a74 | 3509 | return r; |
a8548816 | 3510 | |
6a0f1f6d LP |
3511 | r = event_arm_timer(e, &e->monotonic); |
3512 | if (r < 0) | |
c45a5a74 | 3513 | return r; |
6a0f1f6d LP |
3514 | |
3515 | r = event_arm_timer(e, &e->realtime_alarm); | |
1b5995b0 | 3516 | if (r < 0) |
c45a5a74 | 3517 | return r; |
fd38203a | 3518 | |
6a0f1f6d | 3519 | r = event_arm_timer(e, &e->boottime_alarm); |
1b5995b0 | 3520 | if (r < 0) |
c45a5a74 | 3521 | return r; |
fd38203a | 3522 | |
97ef5391 LP |
3523 | event_close_inode_data_fds(e); |
3524 | ||
1b5995b0 | 3525 | if (event_next_pending(e) || e->need_process_child) |
c45a5a74 TG |
3526 | goto pending; |
3527 | ||
2b0c9ef7 | 3528 | e->state = SD_EVENT_ARMED; |
c45a5a74 TG |
3529 | |
3530 | return 0; | |
3531 | ||
3532 | pending: | |
2b0c9ef7 | 3533 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
3534 | r = sd_event_wait(e, 0); |
3535 | if (r == 0) | |
2b0c9ef7 | 3536 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
3537 | |
3538 | return r; | |
c45a5a74 TG |
3539 | } |
3540 | ||
3541 | _public_ int sd_event_wait(sd_event *e, uint64_t timeout) { | |
5cddd924 | 3542 | size_t event_queue_max; |
c45a5a74 TG |
3543 | int r, m, i; |
3544 | ||
3545 | assert_return(e, -EINVAL); | |
b937d761 | 3546 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
3547 | assert_return(!event_pid_changed(e), -ECHILD); |
3548 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 3549 | assert_return(e->state == SD_EVENT_ARMED, -EBUSY); |
c45a5a74 TG |
3550 | |
3551 | if (e->exit_requested) { | |
3552 | e->state = SD_EVENT_PENDING; | |
3553 | return 1; | |
3554 | } | |
6a0f1f6d | 3555 | |
5cddd924 LP |
3556 | event_queue_max = MAX(e->n_sources, 1u); |
3557 | if (!GREEDY_REALLOC(e->event_queue, e->event_queue_allocated, event_queue_max)) | |
3558 | return -ENOMEM; | |
fd38203a | 3559 | |
97ef5391 LP |
3560 | /* If we still have inotify data buffered, then query the other fds, but don't wait on it */ |
3561 | if (e->inotify_data_buffered) | |
3562 | timeout = 0; | |
3563 | ||
5cddd924 | 3564 | m = epoll_wait(e->epoll_fd, e->event_queue, event_queue_max, |
bab4820e | 3565 | timeout == (uint64_t) -1 ? -1 : (int) DIV_ROUND_UP(timeout, USEC_PER_MSEC)); |
da7e457c | 3566 | if (m < 0) { |
c45a5a74 TG |
3567 | if (errno == EINTR) { |
3568 | e->state = SD_EVENT_PENDING; | |
3569 | return 1; | |
3570 | } | |
3571 | ||
3572 | r = -errno; | |
da7e457c LP |
3573 | goto finish; |
3574 | } | |
fd38203a | 3575 | |
e475d10c | 3576 | triple_timestamp_get(&e->timestamp); |
fd38203a LP |
3577 | |
3578 | for (i = 0; i < m; i++) { | |
3579 | ||
5cddd924 LP |
3580 | if (e->event_queue[i].data.ptr == INT_TO_PTR(SOURCE_WATCHDOG)) |
3581 | r = flush_timer(e, e->watchdog_fd, e->event_queue[i].events, NULL); | |
9da4cb2b | 3582 | else { |
5cddd924 | 3583 | WakeupType *t = e->event_queue[i].data.ptr; |
9da4cb2b LP |
3584 | |
3585 | switch (*t) { | |
3586 | ||
f8f3f926 | 3587 | case WAKEUP_EVENT_SOURCE: { |
5cddd924 | 3588 | sd_event_source *s = e->event_queue[i].data.ptr; |
f8f3f926 LP |
3589 | |
3590 | assert(s); | |
3591 | ||
3592 | switch (s->type) { | |
3593 | ||
3594 | case SOURCE_IO: | |
5cddd924 | 3595 | r = process_io(e, s, e->event_queue[i].events); |
f8f3f926 LP |
3596 | break; |
3597 | ||
3598 | case SOURCE_CHILD: | |
5cddd924 | 3599 | r = process_pidfd(e, s, e->event_queue[i].events); |
f8f3f926 LP |
3600 | break; |
3601 | ||
3602 | default: | |
3603 | assert_not_reached("Unexpected event source type"); | |
3604 | } | |
3605 | ||
9da4cb2b | 3606 | break; |
f8f3f926 | 3607 | } |
fd38203a | 3608 | |
9da4cb2b | 3609 | case WAKEUP_CLOCK_DATA: { |
5cddd924 | 3610 | struct clock_data *d = e->event_queue[i].data.ptr; |
f8f3f926 LP |
3611 | |
3612 | assert(d); | |
3613 | ||
5cddd924 | 3614 | r = flush_timer(e, d->fd, e->event_queue[i].events, &d->next); |
9da4cb2b LP |
3615 | break; |
3616 | } | |
3617 | ||
3618 | case WAKEUP_SIGNAL_DATA: | |
5cddd924 | 3619 | r = process_signal(e, e->event_queue[i].data.ptr, e->event_queue[i].events); |
9da4cb2b LP |
3620 | break; |
3621 | ||
97ef5391 | 3622 | case WAKEUP_INOTIFY_DATA: |
5cddd924 | 3623 | r = event_inotify_data_read(e, e->event_queue[i].data.ptr, e->event_queue[i].events); |
97ef5391 LP |
3624 | break; |
3625 | ||
9da4cb2b LP |
3626 | default: |
3627 | assert_not_reached("Invalid wake-up pointer"); | |
3628 | } | |
3629 | } | |
fd38203a | 3630 | if (r < 0) |
da7e457c | 3631 | goto finish; |
fd38203a LP |
3632 | } |
3633 | ||
cde93897 LP |
3634 | r = process_watchdog(e); |
3635 | if (r < 0) | |
3636 | goto finish; | |
3637 | ||
6a0f1f6d LP |
3638 | r = process_timer(e, e->timestamp.realtime, &e->realtime); |
3639 | if (r < 0) | |
3640 | goto finish; | |
3641 | ||
e475d10c | 3642 | r = process_timer(e, e->timestamp.boottime, &e->boottime); |
a8548816 TG |
3643 | if (r < 0) |
3644 | goto finish; | |
3645 | ||
6a0f1f6d LP |
3646 | r = process_timer(e, e->timestamp.monotonic, &e->monotonic); |
3647 | if (r < 0) | |
3648 | goto finish; | |
3649 | ||
3650 | r = process_timer(e, e->timestamp.realtime, &e->realtime_alarm); | |
fd38203a | 3651 | if (r < 0) |
da7e457c | 3652 | goto finish; |
fd38203a | 3653 | |
e475d10c | 3654 | r = process_timer(e, e->timestamp.boottime, &e->boottime_alarm); |
fd38203a | 3655 | if (r < 0) |
da7e457c | 3656 | goto finish; |
fd38203a | 3657 | |
c2ba3ad6 | 3658 | if (e->need_process_child) { |
fd38203a LP |
3659 | r = process_child(e); |
3660 | if (r < 0) | |
da7e457c | 3661 | goto finish; |
fd38203a LP |
3662 | } |
3663 | ||
97ef5391 LP |
3664 | r = process_inotify(e); |
3665 | if (r < 0) | |
3666 | goto finish; | |
3667 | ||
c45a5a74 TG |
3668 | if (event_next_pending(e)) { |
3669 | e->state = SD_EVENT_PENDING; | |
3670 | ||
3671 | return 1; | |
da7e457c LP |
3672 | } |
3673 | ||
c45a5a74 | 3674 | r = 0; |
fd38203a | 3675 | |
da7e457c | 3676 | finish: |
2b0c9ef7 | 3677 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
3678 | |
3679 | return r; | |
fd38203a LP |
3680 | } |
3681 | ||
c45a5a74 TG |
3682 | _public_ int sd_event_dispatch(sd_event *e) { |
3683 | sd_event_source *p; | |
3684 | int r; | |
3685 | ||
3686 | assert_return(e, -EINVAL); | |
b937d761 | 3687 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
3688 | assert_return(!event_pid_changed(e), -ECHILD); |
3689 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
3690 | assert_return(e->state == SD_EVENT_PENDING, -EBUSY); | |
3691 | ||
3692 | if (e->exit_requested) | |
3693 | return dispatch_exit(e); | |
3694 | ||
3695 | p = event_next_pending(e); | |
3696 | if (p) { | |
30dd293c | 3697 | _cleanup_(sd_event_unrefp) sd_event *ref = NULL; |
c45a5a74 | 3698 | |
30dd293c | 3699 | ref = sd_event_ref(e); |
c45a5a74 TG |
3700 | e->state = SD_EVENT_RUNNING; |
3701 | r = source_dispatch(p); | |
2b0c9ef7 | 3702 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
3703 | return r; |
3704 | } | |
3705 | ||
2b0c9ef7 | 3706 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
3707 | |
3708 | return 1; | |
3709 | } | |
3710 | ||
34b87517 | 3711 | static void event_log_delays(sd_event *e) { |
442ac269 YW |
3712 | char b[ELEMENTSOF(e->delays) * DECIMAL_STR_MAX(unsigned) + 1], *p; |
3713 | size_t l, i; | |
34b87517 | 3714 | |
442ac269 YW |
3715 | p = b; |
3716 | l = sizeof(b); | |
3717 | for (i = 0; i < ELEMENTSOF(e->delays); i++) { | |
3718 | l = strpcpyf(&p, l, "%u ", e->delays[i]); | |
34b87517 VC |
3719 | e->delays[i] = 0; |
3720 | } | |
442ac269 | 3721 | log_debug("Event loop iterations: %s", b); |
34b87517 VC |
3722 | } |
3723 | ||
c45a5a74 TG |
3724 | _public_ int sd_event_run(sd_event *e, uint64_t timeout) { |
3725 | int r; | |
3726 | ||
3727 | assert_return(e, -EINVAL); | |
b937d761 | 3728 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
3729 | assert_return(!event_pid_changed(e), -ECHILD); |
3730 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 3731 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
c45a5a74 | 3732 | |
34b87517 VC |
3733 | if (e->profile_delays && e->last_run) { |
3734 | usec_t this_run; | |
3735 | unsigned l; | |
3736 | ||
3737 | this_run = now(CLOCK_MONOTONIC); | |
3738 | ||
3739 | l = u64log2(this_run - e->last_run); | |
cb9d621e | 3740 | assert(l < ELEMENTSOF(e->delays)); |
34b87517 VC |
3741 | e->delays[l]++; |
3742 | ||
3743 | if (this_run - e->last_log >= 5*USEC_PER_SEC) { | |
3744 | event_log_delays(e); | |
3745 | e->last_log = this_run; | |
3746 | } | |
3747 | } | |
3748 | ||
c45a5a74 | 3749 | r = sd_event_prepare(e); |
53bac4e0 LP |
3750 | if (r == 0) |
3751 | /* There was nothing? Then wait... */ | |
3752 | r = sd_event_wait(e, timeout); | |
c45a5a74 | 3753 | |
34b87517 VC |
3754 | if (e->profile_delays) |
3755 | e->last_run = now(CLOCK_MONOTONIC); | |
3756 | ||
02d30981 | 3757 | if (r > 0) { |
53bac4e0 | 3758 | /* There's something now, then let's dispatch it */ |
02d30981 TG |
3759 | r = sd_event_dispatch(e); |
3760 | if (r < 0) | |
3761 | return r; | |
53bac4e0 LP |
3762 | |
3763 | return 1; | |
3764 | } | |
3765 | ||
3766 | return r; | |
c45a5a74 TG |
3767 | } |
3768 | ||
f7262a9f | 3769 | _public_ int sd_event_loop(sd_event *e) { |
30dd293c | 3770 | _cleanup_(sd_event_unrefp) sd_event *ref = NULL; |
fd38203a LP |
3771 | int r; |
3772 | ||
da7e457c | 3773 | assert_return(e, -EINVAL); |
b937d761 | 3774 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 3775 | assert_return(!event_pid_changed(e), -ECHILD); |
2b0c9ef7 | 3776 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 3777 | |
30dd293c | 3778 | ref = sd_event_ref(e); |
fd38203a | 3779 | |
da7e457c | 3780 | while (e->state != SD_EVENT_FINISHED) { |
fd38203a LP |
3781 | r = sd_event_run(e, (uint64_t) -1); |
3782 | if (r < 0) | |
30dd293c | 3783 | return r; |
fd38203a LP |
3784 | } |
3785 | ||
30dd293c | 3786 | return e->exit_code; |
fd38203a LP |
3787 | } |
3788 | ||
9b364545 | 3789 | _public_ int sd_event_get_fd(sd_event *e) { |
9b364545 | 3790 | assert_return(e, -EINVAL); |
b937d761 | 3791 | assert_return(e = event_resolve(e), -ENOPKG); |
9b364545 TG |
3792 | assert_return(!event_pid_changed(e), -ECHILD); |
3793 | ||
3794 | return e->epoll_fd; | |
3795 | } | |
3796 | ||
f7262a9f | 3797 | _public_ int sd_event_get_state(sd_event *e) { |
da7e457c | 3798 | assert_return(e, -EINVAL); |
b937d761 | 3799 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
3800 | assert_return(!event_pid_changed(e), -ECHILD); |
3801 | ||
3802 | return e->state; | |
3803 | } | |
3804 | ||
6203e07a | 3805 | _public_ int sd_event_get_exit_code(sd_event *e, int *code) { |
da7e457c | 3806 | assert_return(e, -EINVAL); |
b937d761 | 3807 | assert_return(e = event_resolve(e), -ENOPKG); |
6203e07a | 3808 | assert_return(code, -EINVAL); |
da7e457c | 3809 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 3810 | |
6203e07a LP |
3811 | if (!e->exit_requested) |
3812 | return -ENODATA; | |
3813 | ||
3814 | *code = e->exit_code; | |
3815 | return 0; | |
fd38203a LP |
3816 | } |
3817 | ||
6203e07a | 3818 | _public_ int sd_event_exit(sd_event *e, int code) { |
da7e457c | 3819 | assert_return(e, -EINVAL); |
b937d761 | 3820 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
3821 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
3822 | assert_return(!event_pid_changed(e), -ECHILD); | |
fd38203a | 3823 | |
6203e07a LP |
3824 | e->exit_requested = true; |
3825 | e->exit_code = code; | |
3826 | ||
fd38203a LP |
3827 | return 0; |
3828 | } | |
46e8c825 | 3829 | |
6a0f1f6d | 3830 | _public_ int sd_event_now(sd_event *e, clockid_t clock, uint64_t *usec) { |
46e8c825 | 3831 | assert_return(e, -EINVAL); |
b937d761 | 3832 | assert_return(e = event_resolve(e), -ENOPKG); |
46e8c825 | 3833 | assert_return(usec, -EINVAL); |
46e8c825 LP |
3834 | assert_return(!event_pid_changed(e), -ECHILD); |
3835 | ||
e475d10c LP |
3836 | if (!TRIPLE_TIMESTAMP_HAS_CLOCK(clock)) |
3837 | return -EOPNOTSUPP; | |
3838 | ||
3839 | /* Generate a clean error in case CLOCK_BOOTTIME is not available. Note that don't use clock_supported() here, | |
3840 | * for a reason: there are systems where CLOCK_BOOTTIME is supported, but CLOCK_BOOTTIME_ALARM is not, but for | |
3841 | * the purpose of getting the time this doesn't matter. */ | |
3411372e LP |
3842 | if (IN_SET(clock, CLOCK_BOOTTIME, CLOCK_BOOTTIME_ALARM) && !clock_boottime_supported()) |
3843 | return -EOPNOTSUPP; | |
3844 | ||
e475d10c | 3845 | if (!triple_timestamp_is_set(&e->timestamp)) { |
15c689d7 | 3846 | /* Implicitly fall back to now() if we never ran before and thus have no cached time. */ |
38a03f06 LP |
3847 | *usec = now(clock); |
3848 | return 1; | |
3849 | } | |
46e8c825 | 3850 | |
e475d10c | 3851 | *usec = triple_timestamp_by_clock(&e->timestamp, clock); |
46e8c825 LP |
3852 | return 0; |
3853 | } | |
afc6adb5 LP |
3854 | |
3855 | _public_ int sd_event_default(sd_event **ret) { | |
39883f62 | 3856 | sd_event *e = NULL; |
afc6adb5 LP |
3857 | int r; |
3858 | ||
3859 | if (!ret) | |
3860 | return !!default_event; | |
3861 | ||
3862 | if (default_event) { | |
3863 | *ret = sd_event_ref(default_event); | |
3864 | return 0; | |
3865 | } | |
3866 | ||
3867 | r = sd_event_new(&e); | |
3868 | if (r < 0) | |
3869 | return r; | |
3870 | ||
3871 | e->default_event_ptr = &default_event; | |
3872 | e->tid = gettid(); | |
3873 | default_event = e; | |
3874 | ||
3875 | *ret = e; | |
3876 | return 1; | |
3877 | } | |
3878 | ||
3879 | _public_ int sd_event_get_tid(sd_event *e, pid_t *tid) { | |
3880 | assert_return(e, -EINVAL); | |
b937d761 | 3881 | assert_return(e = event_resolve(e), -ENOPKG); |
afc6adb5 | 3882 | assert_return(tid, -EINVAL); |
76b54375 | 3883 | assert_return(!event_pid_changed(e), -ECHILD); |
afc6adb5 | 3884 | |
76b54375 LP |
3885 | if (e->tid != 0) { |
3886 | *tid = e->tid; | |
3887 | return 0; | |
3888 | } | |
3889 | ||
3890 | return -ENXIO; | |
afc6adb5 | 3891 | } |
cde93897 LP |
3892 | |
3893 | _public_ int sd_event_set_watchdog(sd_event *e, int b) { | |
3894 | int r; | |
3895 | ||
3896 | assert_return(e, -EINVAL); | |
b937d761 | 3897 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 | 3898 | assert_return(!event_pid_changed(e), -ECHILD); |
cde93897 LP |
3899 | |
3900 | if (e->watchdog == !!b) | |
3901 | return e->watchdog; | |
3902 | ||
3903 | if (b) { | |
09812eb7 LP |
3904 | r = sd_watchdog_enabled(false, &e->watchdog_period); |
3905 | if (r <= 0) | |
cde93897 | 3906 | return r; |
cde93897 LP |
3907 | |
3908 | /* Issue first ping immediately */ | |
3909 | sd_notify(false, "WATCHDOG=1"); | |
3910 | e->watchdog_last = now(CLOCK_MONOTONIC); | |
3911 | ||
3912 | e->watchdog_fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC); | |
3913 | if (e->watchdog_fd < 0) | |
3914 | return -errno; | |
3915 | ||
3916 | r = arm_watchdog(e); | |
3917 | if (r < 0) | |
3918 | goto fail; | |
3919 | ||
1eac7948 | 3920 | struct epoll_event ev = { |
a82f89aa LP |
3921 | .events = EPOLLIN, |
3922 | .data.ptr = INT_TO_PTR(SOURCE_WATCHDOG), | |
3923 | }; | |
cde93897 | 3924 | |
15c689d7 | 3925 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, e->watchdog_fd, &ev) < 0) { |
cde93897 LP |
3926 | r = -errno; |
3927 | goto fail; | |
3928 | } | |
3929 | ||
3930 | } else { | |
3931 | if (e->watchdog_fd >= 0) { | |
5a795bff | 3932 | (void) epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, e->watchdog_fd, NULL); |
03e334a1 | 3933 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
3934 | } |
3935 | } | |
3936 | ||
3937 | e->watchdog = !!b; | |
3938 | return e->watchdog; | |
3939 | ||
3940 | fail: | |
03e334a1 | 3941 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
3942 | return r; |
3943 | } | |
8f726607 LP |
3944 | |
3945 | _public_ int sd_event_get_watchdog(sd_event *e) { | |
3946 | assert_return(e, -EINVAL); | |
b937d761 | 3947 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 LP |
3948 | assert_return(!event_pid_changed(e), -ECHILD); |
3949 | ||
3950 | return e->watchdog; | |
3951 | } | |
60a3b1e1 LP |
3952 | |
3953 | _public_ int sd_event_get_iteration(sd_event *e, uint64_t *ret) { | |
3954 | assert_return(e, -EINVAL); | |
b937d761 | 3955 | assert_return(e = event_resolve(e), -ENOPKG); |
60a3b1e1 LP |
3956 | assert_return(!event_pid_changed(e), -ECHILD); |
3957 | ||
3958 | *ret = e->iteration; | |
3959 | return 0; | |
3960 | } | |
15723a1d LP |
3961 | |
3962 | _public_ int sd_event_source_set_destroy_callback(sd_event_source *s, sd_event_destroy_t callback) { | |
3963 | assert_return(s, -EINVAL); | |
3964 | ||
3965 | s->destroy_callback = callback; | |
3966 | return 0; | |
3967 | } | |
3968 | ||
3969 | _public_ int sd_event_source_get_destroy_callback(sd_event_source *s, sd_event_destroy_t *ret) { | |
3970 | assert_return(s, -EINVAL); | |
3971 | ||
3972 | if (ret) | |
3973 | *ret = s->destroy_callback; | |
3974 | ||
3975 | return !!s->destroy_callback; | |
3976 | } | |
2382c936 YW |
3977 | |
3978 | _public_ int sd_event_source_get_floating(sd_event_source *s) { | |
3979 | assert_return(s, -EINVAL); | |
3980 | ||
3981 | return s->floating; | |
3982 | } | |
3983 | ||
3984 | _public_ int sd_event_source_set_floating(sd_event_source *s, int b) { | |
3985 | assert_return(s, -EINVAL); | |
3986 | ||
3987 | if (s->floating == !!b) | |
3988 | return 0; | |
3989 | ||
3990 | if (!s->event) /* Already disconnected */ | |
3991 | return -ESTALE; | |
3992 | ||
3993 | s->floating = b; | |
3994 | ||
3995 | if (b) { | |
3996 | sd_event_source_ref(s); | |
3997 | sd_event_unref(s->event); | |
3998 | } else { | |
3999 | sd_event_ref(s->event); | |
4000 | sd_event_source_unref(s); | |
4001 | } | |
4002 | ||
4003 | return 1; | |
4004 | } | |
b778cba4 LP |
4005 | |
4006 | _public_ int sd_event_source_get_exit_on_failure(sd_event_source *s) { | |
4007 | assert_return(s, -EINVAL); | |
4008 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
4009 | ||
4010 | return s->exit_on_failure; | |
4011 | } | |
4012 | ||
4013 | _public_ int sd_event_source_set_exit_on_failure(sd_event_source *s, int b) { | |
4014 | assert_return(s, -EINVAL); | |
4015 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
4016 | ||
4017 | if (s->exit_on_failure == !!b) | |
4018 | return 0; | |
4019 | ||
4020 | s->exit_on_failure = b; | |
4021 | return 1; | |
4022 | } |