]>
Commit | Line | Data |
---|---|---|
53e1b683 | 1 | /* SPDX-License-Identifier: LGPL-2.1+ */ |
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) |
55cbfaa5 DM |
383 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll: %m", |
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 ZJS |
404 | s->io.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, |
405 | s->io.fd, | |
15c689d7 | 406 | &ev) < 0) |
fd38203a LP |
407 | return -errno; |
408 | ||
409 | s->io.registered = true; | |
410 | ||
411 | return 0; | |
412 | } | |
413 | ||
f8f3f926 LP |
414 | static void source_child_pidfd_unregister(sd_event_source *s) { |
415 | assert(s); | |
416 | assert(s->type == SOURCE_CHILD); | |
417 | ||
418 | if (event_pid_changed(s->event)) | |
419 | return; | |
420 | ||
421 | if (!s->child.registered) | |
422 | return; | |
423 | ||
424 | if (EVENT_SOURCE_WATCH_PIDFD(s)) | |
425 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->child.pidfd, NULL) < 0) | |
426 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll: %m", | |
427 | strna(s->description), event_source_type_to_string(s->type)); | |
428 | ||
429 | s->child.registered = false; | |
430 | } | |
431 | ||
432 | static int source_child_pidfd_register(sd_event_source *s, int enabled) { | |
433 | int r; | |
434 | ||
435 | assert(s); | |
436 | assert(s->type == SOURCE_CHILD); | |
437 | assert(enabled != SD_EVENT_OFF); | |
438 | ||
439 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { | |
1eac7948 | 440 | struct epoll_event ev = { |
f8f3f926 LP |
441 | .events = EPOLLIN | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0), |
442 | .data.ptr = s, | |
443 | }; | |
444 | ||
445 | if (s->child.registered) | |
446 | r = epoll_ctl(s->event->epoll_fd, EPOLL_CTL_MOD, s->child.pidfd, &ev); | |
447 | else | |
448 | r = epoll_ctl(s->event->epoll_fd, EPOLL_CTL_ADD, s->child.pidfd, &ev); | |
449 | if (r < 0) | |
450 | return -errno; | |
451 | } | |
452 | ||
453 | s->child.registered = true; | |
454 | return 0; | |
455 | } | |
456 | ||
6a0f1f6d LP |
457 | static clockid_t event_source_type_to_clock(EventSourceType t) { |
458 | ||
459 | switch (t) { | |
460 | ||
461 | case SOURCE_TIME_REALTIME: | |
462 | return CLOCK_REALTIME; | |
463 | ||
a8548816 TG |
464 | case SOURCE_TIME_BOOTTIME: |
465 | return CLOCK_BOOTTIME; | |
466 | ||
6a0f1f6d LP |
467 | case SOURCE_TIME_MONOTONIC: |
468 | return CLOCK_MONOTONIC; | |
469 | ||
470 | case SOURCE_TIME_REALTIME_ALARM: | |
471 | return CLOCK_REALTIME_ALARM; | |
472 | ||
473 | case SOURCE_TIME_BOOTTIME_ALARM: | |
474 | return CLOCK_BOOTTIME_ALARM; | |
475 | ||
476 | default: | |
477 | return (clockid_t) -1; | |
478 | } | |
479 | } | |
480 | ||
481 | static EventSourceType clock_to_event_source_type(clockid_t clock) { | |
482 | ||
483 | switch (clock) { | |
484 | ||
485 | case CLOCK_REALTIME: | |
486 | return SOURCE_TIME_REALTIME; | |
487 | ||
a8548816 TG |
488 | case CLOCK_BOOTTIME: |
489 | return SOURCE_TIME_BOOTTIME; | |
490 | ||
6a0f1f6d LP |
491 | case CLOCK_MONOTONIC: |
492 | return SOURCE_TIME_MONOTONIC; | |
493 | ||
494 | case CLOCK_REALTIME_ALARM: | |
495 | return SOURCE_TIME_REALTIME_ALARM; | |
496 | ||
497 | case CLOCK_BOOTTIME_ALARM: | |
498 | return SOURCE_TIME_BOOTTIME_ALARM; | |
499 | ||
500 | default: | |
501 | return _SOURCE_EVENT_SOURCE_TYPE_INVALID; | |
502 | } | |
503 | } | |
504 | ||
505 | static struct clock_data* event_get_clock_data(sd_event *e, EventSourceType t) { | |
506 | assert(e); | |
507 | ||
508 | switch (t) { | |
509 | ||
510 | case SOURCE_TIME_REALTIME: | |
511 | return &e->realtime; | |
512 | ||
a8548816 TG |
513 | case SOURCE_TIME_BOOTTIME: |
514 | return &e->boottime; | |
515 | ||
6a0f1f6d LP |
516 | case SOURCE_TIME_MONOTONIC: |
517 | return &e->monotonic; | |
518 | ||
519 | case SOURCE_TIME_REALTIME_ALARM: | |
520 | return &e->realtime_alarm; | |
521 | ||
522 | case SOURCE_TIME_BOOTTIME_ALARM: | |
523 | return &e->boottime_alarm; | |
524 | ||
525 | default: | |
526 | return NULL; | |
527 | } | |
528 | } | |
529 | ||
3e4eb8e7 YW |
530 | static void event_free_signal_data(sd_event *e, struct signal_data *d) { |
531 | assert(e); | |
532 | ||
533 | if (!d) | |
534 | return; | |
535 | ||
536 | hashmap_remove(e->signal_data, &d->priority); | |
537 | safe_close(d->fd); | |
538 | free(d); | |
539 | } | |
540 | ||
9da4cb2b LP |
541 | static int event_make_signal_data( |
542 | sd_event *e, | |
543 | int sig, | |
544 | struct signal_data **ret) { | |
4807d2d0 | 545 | |
9da4cb2b LP |
546 | struct signal_data *d; |
547 | bool added = false; | |
548 | sigset_t ss_copy; | |
549 | int64_t priority; | |
f95387cd ZJS |
550 | int r; |
551 | ||
552 | assert(e); | |
553 | ||
f6806734 | 554 | if (event_pid_changed(e)) |
9da4cb2b | 555 | return -ECHILD; |
f6806734 | 556 | |
9da4cb2b LP |
557 | if (e->signal_sources && e->signal_sources[sig]) |
558 | priority = e->signal_sources[sig]->priority; | |
559 | else | |
de05913d | 560 | priority = SD_EVENT_PRIORITY_NORMAL; |
f95387cd | 561 | |
9da4cb2b LP |
562 | d = hashmap_get(e->signal_data, &priority); |
563 | if (d) { | |
564 | if (sigismember(&d->sigset, sig) > 0) { | |
565 | if (ret) | |
566 | *ret = d; | |
567 | return 0; | |
568 | } | |
569 | } else { | |
570 | r = hashmap_ensure_allocated(&e->signal_data, &uint64_hash_ops); | |
571 | if (r < 0) | |
572 | return r; | |
573 | ||
d08eb1fa | 574 | d = new(struct signal_data, 1); |
9da4cb2b LP |
575 | if (!d) |
576 | return -ENOMEM; | |
577 | ||
d08eb1fa LP |
578 | *d = (struct signal_data) { |
579 | .wakeup = WAKEUP_SIGNAL_DATA, | |
580 | .fd = -1, | |
581 | .priority = priority, | |
582 | }; | |
9da4cb2b LP |
583 | |
584 | r = hashmap_put(e->signal_data, &d->priority, d); | |
90f604d1 ZJS |
585 | if (r < 0) { |
586 | free(d); | |
9da4cb2b | 587 | return r; |
90f604d1 | 588 | } |
f95387cd | 589 | |
9da4cb2b LP |
590 | added = true; |
591 | } | |
592 | ||
593 | ss_copy = d->sigset; | |
594 | assert_se(sigaddset(&ss_copy, sig) >= 0); | |
595 | ||
596 | r = signalfd(d->fd, &ss_copy, SFD_NONBLOCK|SFD_CLOEXEC); | |
597 | if (r < 0) { | |
598 | r = -errno; | |
599 | goto fail; | |
600 | } | |
601 | ||
602 | d->sigset = ss_copy; | |
f95387cd | 603 | |
9da4cb2b LP |
604 | if (d->fd >= 0) { |
605 | if (ret) | |
606 | *ret = d; | |
f95387cd | 607 | return 0; |
9da4cb2b LP |
608 | } |
609 | ||
7fe2903c | 610 | d->fd = fd_move_above_stdio(r); |
f95387cd | 611 | |
1eac7948 | 612 | struct epoll_event ev = { |
a82f89aa LP |
613 | .events = EPOLLIN, |
614 | .data.ptr = d, | |
615 | }; | |
f95387cd | 616 | |
15c689d7 | 617 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { |
9da4cb2b LP |
618 | r = -errno; |
619 | goto fail; | |
f95387cd ZJS |
620 | } |
621 | ||
9da4cb2b LP |
622 | if (ret) |
623 | *ret = d; | |
624 | ||
f95387cd | 625 | return 0; |
9da4cb2b LP |
626 | |
627 | fail: | |
3e4eb8e7 YW |
628 | if (added) |
629 | event_free_signal_data(e, d); | |
9da4cb2b LP |
630 | |
631 | return r; | |
632 | } | |
633 | ||
634 | static void event_unmask_signal_data(sd_event *e, struct signal_data *d, int sig) { | |
635 | assert(e); | |
636 | assert(d); | |
637 | ||
638 | /* Turns off the specified signal in the signal data | |
639 | * object. If the signal mask of the object becomes empty that | |
640 | * way removes it. */ | |
641 | ||
642 | if (sigismember(&d->sigset, sig) == 0) | |
643 | return; | |
644 | ||
645 | assert_se(sigdelset(&d->sigset, sig) >= 0); | |
646 | ||
647 | if (sigisemptyset(&d->sigset)) { | |
9da4cb2b | 648 | /* If all the mask is all-zero we can get rid of the structure */ |
3e4eb8e7 | 649 | event_free_signal_data(e, d); |
9da4cb2b LP |
650 | return; |
651 | } | |
652 | ||
653 | assert(d->fd >= 0); | |
654 | ||
655 | if (signalfd(d->fd, &d->sigset, SFD_NONBLOCK|SFD_CLOEXEC) < 0) | |
656 | log_debug_errno(errno, "Failed to unset signal bit, ignoring: %m"); | |
657 | } | |
658 | ||
659 | static void event_gc_signal_data(sd_event *e, const int64_t *priority, int sig) { | |
660 | struct signal_data *d; | |
661 | static const int64_t zero_priority = 0; | |
662 | ||
663 | assert(e); | |
664 | ||
f8f3f926 LP |
665 | /* Rechecks if the specified signal is still something we are interested in. If not, we'll unmask it, |
666 | * and possibly drop the signalfd for it. */ | |
9da4cb2b LP |
667 | |
668 | if (sig == SIGCHLD && | |
669 | e->n_enabled_child_sources > 0) | |
670 | return; | |
671 | ||
672 | if (e->signal_sources && | |
673 | e->signal_sources[sig] && | |
674 | e->signal_sources[sig]->enabled != SD_EVENT_OFF) | |
675 | return; | |
676 | ||
677 | /* | |
678 | * The specified signal might be enabled in three different queues: | |
679 | * | |
680 | * 1) the one that belongs to the priority passed (if it is non-NULL) | |
681 | * 2) the one that belongs to the priority of the event source of the signal (if there is one) | |
682 | * 3) the 0 priority (to cover the SIGCHLD case) | |
683 | * | |
684 | * Hence, let's remove it from all three here. | |
685 | */ | |
686 | ||
687 | if (priority) { | |
688 | d = hashmap_get(e->signal_data, priority); | |
689 | if (d) | |
690 | event_unmask_signal_data(e, d, sig); | |
691 | } | |
692 | ||
693 | if (e->signal_sources && e->signal_sources[sig]) { | |
694 | d = hashmap_get(e->signal_data, &e->signal_sources[sig]->priority); | |
695 | if (d) | |
696 | event_unmask_signal_data(e, d, sig); | |
697 | } | |
698 | ||
699 | d = hashmap_get(e->signal_data, &zero_priority); | |
700 | if (d) | |
701 | event_unmask_signal_data(e, d, sig); | |
f95387cd ZJS |
702 | } |
703 | ||
a71fe8b8 LP |
704 | static void source_disconnect(sd_event_source *s) { |
705 | sd_event *event; | |
706 | ||
fd38203a LP |
707 | assert(s); |
708 | ||
a71fe8b8 LP |
709 | if (!s->event) |
710 | return; | |
15b38f93 | 711 | |
a71fe8b8 | 712 | assert(s->event->n_sources > 0); |
fd38203a | 713 | |
a71fe8b8 | 714 | switch (s->type) { |
fd38203a | 715 | |
a71fe8b8 LP |
716 | case SOURCE_IO: |
717 | if (s->io.fd >= 0) | |
718 | source_io_unregister(s); | |
fd38203a | 719 | |
a71fe8b8 | 720 | break; |
6a0f1f6d | 721 | |
a71fe8b8 | 722 | case SOURCE_TIME_REALTIME: |
a8548816 | 723 | case SOURCE_TIME_BOOTTIME: |
a71fe8b8 LP |
724 | case SOURCE_TIME_MONOTONIC: |
725 | case SOURCE_TIME_REALTIME_ALARM: | |
726 | case SOURCE_TIME_BOOTTIME_ALARM: { | |
727 | struct clock_data *d; | |
fd38203a | 728 | |
a71fe8b8 LP |
729 | d = event_get_clock_data(s->event, s->type); |
730 | assert(d); | |
731 | ||
732 | prioq_remove(d->earliest, s, &s->time.earliest_index); | |
733 | prioq_remove(d->latest, s, &s->time.latest_index); | |
212bbb17 | 734 | d->needs_rearm = true; |
a71fe8b8 LP |
735 | break; |
736 | } | |
737 | ||
738 | case SOURCE_SIGNAL: | |
739 | if (s->signal.sig > 0) { | |
9da4cb2b | 740 | |
a71fe8b8 LP |
741 | if (s->event->signal_sources) |
742 | s->event->signal_sources[s->signal.sig] = NULL; | |
4807d2d0 | 743 | |
9da4cb2b | 744 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
6a0f1f6d | 745 | } |
fd38203a | 746 | |
a71fe8b8 | 747 | break; |
fd38203a | 748 | |
a71fe8b8 LP |
749 | case SOURCE_CHILD: |
750 | if (s->child.pid > 0) { | |
751 | if (s->enabled != SD_EVENT_OFF) { | |
752 | assert(s->event->n_enabled_child_sources > 0); | |
753 | s->event->n_enabled_child_sources--; | |
4807d2d0 | 754 | } |
fd38203a | 755 | |
4a0b58c4 | 756 | (void) hashmap_remove(s->event->child_sources, PID_TO_PTR(s->child.pid)); |
a71fe8b8 | 757 | } |
fd38203a | 758 | |
f8f3f926 LP |
759 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
760 | source_child_pidfd_unregister(s); | |
761 | else | |
762 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
763 | ||
a71fe8b8 | 764 | break; |
fd38203a | 765 | |
a71fe8b8 LP |
766 | case SOURCE_DEFER: |
767 | /* nothing */ | |
768 | break; | |
fd38203a | 769 | |
a71fe8b8 LP |
770 | case SOURCE_POST: |
771 | set_remove(s->event->post_sources, s); | |
772 | break; | |
da7e457c | 773 | |
a71fe8b8 LP |
774 | case SOURCE_EXIT: |
775 | prioq_remove(s->event->exit, s, &s->exit.prioq_index); | |
776 | break; | |
0eb2e0e3 | 777 | |
97ef5391 LP |
778 | case SOURCE_INOTIFY: { |
779 | struct inode_data *inode_data; | |
780 | ||
781 | inode_data = s->inotify.inode_data; | |
782 | if (inode_data) { | |
783 | struct inotify_data *inotify_data; | |
784 | assert_se(inotify_data = inode_data->inotify_data); | |
785 | ||
786 | /* Detach this event source from the inode object */ | |
787 | LIST_REMOVE(inotify.by_inode_data, inode_data->event_sources, s); | |
788 | s->inotify.inode_data = NULL; | |
789 | ||
790 | if (s->pending) { | |
791 | assert(inotify_data->n_pending > 0); | |
792 | inotify_data->n_pending--; | |
793 | } | |
794 | ||
795 | /* Note that we don't reduce the inotify mask for the watch descriptor here if the inode is | |
796 | * continued to being watched. That's because inotify doesn't really have an API for that: we | |
797 | * can only change watch masks with access to the original inode either by fd or by path. But | |
798 | * paths aren't stable, and keeping an O_PATH fd open all the time would mean wasting an fd | |
f21f31b2 | 799 | * continuously and keeping the mount busy which we can't really do. We could reconstruct the |
97ef5391 LP |
800 | * original inode from /proc/self/fdinfo/$INOTIFY_FD (as all watch descriptors are listed |
801 | * there), but given the need for open_by_handle_at() which is privileged and not universally | |
802 | * available this would be quite an incomplete solution. Hence we go the other way, leave the | |
803 | * mask set, even if it is not minimized now, and ignore all events we aren't interested in | |
804 | * anymore after reception. Yes, this sucks, but … Linux … */ | |
805 | ||
806 | /* Maybe release the inode data (and its inotify) */ | |
807 | event_gc_inode_data(s->event, inode_data); | |
808 | } | |
809 | ||
810 | break; | |
811 | } | |
812 | ||
a71fe8b8 LP |
813 | default: |
814 | assert_not_reached("Wut? I shouldn't exist."); | |
815 | } | |
6e9feda3 | 816 | |
a71fe8b8 LP |
817 | if (s->pending) |
818 | prioq_remove(s->event->pending, s, &s->pending_index); | |
9d3e3aa5 | 819 | |
a71fe8b8 LP |
820 | if (s->prepare) |
821 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
fd38203a | 822 | |
e514aa1e | 823 | event = TAKE_PTR(s->event); |
a71fe8b8 LP |
824 | LIST_REMOVE(sources, event->sources, s); |
825 | event->n_sources--; | |
fd38203a | 826 | |
f5982559 LP |
827 | /* Note that we don't invalidate the type here, since we still need it in order to close the fd or |
828 | * pidfd associated with this event source, which we'll do only on source_free(). */ | |
829 | ||
a71fe8b8 LP |
830 | if (!s->floating) |
831 | sd_event_unref(event); | |
832 | } | |
833 | ||
834 | static void source_free(sd_event_source *s) { | |
835 | assert(s); | |
fd38203a | 836 | |
a71fe8b8 | 837 | source_disconnect(s); |
ab93297c NM |
838 | |
839 | if (s->type == SOURCE_IO && s->io.owned) | |
15723a1d LP |
840 | s->io.fd = safe_close(s->io.fd); |
841 | ||
f8f3f926 LP |
842 | if (s->type == SOURCE_CHILD) { |
843 | /* Eventually the kernel will do this automatically for us, but for now let's emulate this (unreliably) in userspace. */ | |
844 | ||
845 | if (s->child.process_owned) { | |
846 | ||
847 | if (!s->child.exited) { | |
848 | bool sent = false; | |
849 | ||
850 | if (s->child.pidfd >= 0) { | |
851 | if (pidfd_send_signal(s->child.pidfd, SIGKILL, NULL, 0) < 0) { | |
852 | if (errno == ESRCH) /* Already dead */ | |
853 | sent = true; | |
854 | else if (!ERRNO_IS_NOT_SUPPORTED(errno)) | |
855 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via pidfd_send_signal(), re-trying via kill(): %m", | |
856 | s->child.pid); | |
857 | } else | |
858 | sent = true; | |
859 | } | |
860 | ||
861 | if (!sent) | |
862 | if (kill(s->child.pid, SIGKILL) < 0) | |
863 | if (errno != ESRCH) /* Already dead */ | |
864 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via kill(), ignoring: %m", | |
865 | s->child.pid); | |
866 | } | |
867 | ||
868 | if (!s->child.waited) { | |
869 | siginfo_t si = {}; | |
870 | ||
871 | /* Reap the child if we can */ | |
872 | (void) waitid(P_PID, s->child.pid, &si, WEXITED); | |
873 | } | |
874 | } | |
875 | ||
876 | if (s->child.pidfd_owned) | |
877 | s->child.pidfd = safe_close(s->child.pidfd); | |
878 | } | |
879 | ||
15723a1d LP |
880 | if (s->destroy_callback) |
881 | s->destroy_callback(s->userdata); | |
ab93297c | 882 | |
356779df | 883 | free(s->description); |
fd38203a LP |
884 | free(s); |
885 | } | |
8c75fe17 | 886 | DEFINE_TRIVIAL_CLEANUP_FUNC(sd_event_source*, source_free); |
fd38203a LP |
887 | |
888 | static int source_set_pending(sd_event_source *s, bool b) { | |
889 | int r; | |
890 | ||
891 | assert(s); | |
6203e07a | 892 | assert(s->type != SOURCE_EXIT); |
fd38203a LP |
893 | |
894 | if (s->pending == b) | |
895 | return 0; | |
896 | ||
897 | s->pending = b; | |
898 | ||
899 | if (b) { | |
900 | s->pending_iteration = s->event->iteration; | |
901 | ||
902 | r = prioq_put(s->event->pending, s, &s->pending_index); | |
903 | if (r < 0) { | |
904 | s->pending = false; | |
905 | return r; | |
906 | } | |
907 | } else | |
908 | assert_se(prioq_remove(s->event->pending, s, &s->pending_index)); | |
909 | ||
6a0f1f6d LP |
910 | if (EVENT_SOURCE_IS_TIME(s->type)) { |
911 | struct clock_data *d; | |
912 | ||
913 | d = event_get_clock_data(s->event, s->type); | |
914 | assert(d); | |
915 | ||
916 | prioq_reshuffle(d->earliest, s, &s->time.earliest_index); | |
917 | prioq_reshuffle(d->latest, s, &s->time.latest_index); | |
212bbb17 | 918 | d->needs_rearm = true; |
2576a19e LP |
919 | } |
920 | ||
9da4cb2b LP |
921 | if (s->type == SOURCE_SIGNAL && !b) { |
922 | struct signal_data *d; | |
923 | ||
924 | d = hashmap_get(s->event->signal_data, &s->priority); | |
925 | if (d && d->current == s) | |
926 | d->current = NULL; | |
927 | } | |
928 | ||
97ef5391 LP |
929 | if (s->type == SOURCE_INOTIFY) { |
930 | ||
931 | assert(s->inotify.inode_data); | |
932 | assert(s->inotify.inode_data->inotify_data); | |
933 | ||
934 | if (b) | |
935 | s->inotify.inode_data->inotify_data->n_pending ++; | |
936 | else { | |
937 | assert(s->inotify.inode_data->inotify_data->n_pending > 0); | |
938 | s->inotify.inode_data->inotify_data->n_pending --; | |
939 | } | |
940 | } | |
941 | ||
fd38203a LP |
942 | return 0; |
943 | } | |
944 | ||
a71fe8b8 | 945 | static sd_event_source *source_new(sd_event *e, bool floating, EventSourceType type) { |
fd38203a LP |
946 | sd_event_source *s; |
947 | ||
948 | assert(e); | |
949 | ||
d08eb1fa | 950 | s = new(sd_event_source, 1); |
fd38203a LP |
951 | if (!s) |
952 | return NULL; | |
953 | ||
d08eb1fa LP |
954 | *s = (struct sd_event_source) { |
955 | .n_ref = 1, | |
956 | .event = e, | |
957 | .floating = floating, | |
958 | .type = type, | |
959 | .pending_index = PRIOQ_IDX_NULL, | |
960 | .prepare_index = PRIOQ_IDX_NULL, | |
961 | }; | |
a71fe8b8 LP |
962 | |
963 | if (!floating) | |
964 | sd_event_ref(e); | |
fd38203a | 965 | |
a71fe8b8 | 966 | LIST_PREPEND(sources, e->sources, s); |
313cefa1 | 967 | e->n_sources++; |
15b38f93 | 968 | |
fd38203a LP |
969 | return s; |
970 | } | |
971 | ||
b9350e70 LP |
972 | static int io_exit_callback(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
973 | assert(s); | |
974 | ||
975 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
976 | } | |
977 | ||
f7262a9f | 978 | _public_ int sd_event_add_io( |
fd38203a | 979 | sd_event *e, |
151b9b96 | 980 | sd_event_source **ret, |
fd38203a LP |
981 | int fd, |
982 | uint32_t events, | |
718db961 | 983 | sd_event_io_handler_t callback, |
151b9b96 | 984 | void *userdata) { |
fd38203a | 985 | |
ec766a51 | 986 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
987 | int r; |
988 | ||
305f78bf | 989 | assert_return(e, -EINVAL); |
b937d761 | 990 | assert_return(e = event_resolve(e), -ENOPKG); |
8ac43fee | 991 | assert_return(fd >= 0, -EBADF); |
2a16a986 | 992 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 993 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 994 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 995 | |
b9350e70 LP |
996 | if (!callback) |
997 | callback = io_exit_callback; | |
998 | ||
a71fe8b8 | 999 | s = source_new(e, !ret, SOURCE_IO); |
fd38203a LP |
1000 | if (!s) |
1001 | return -ENOMEM; | |
1002 | ||
9da4cb2b | 1003 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1004 | s->io.fd = fd; |
1005 | s->io.events = events; | |
1006 | s->io.callback = callback; | |
1007 | s->userdata = userdata; | |
baf76283 | 1008 | s->enabled = SD_EVENT_ON; |
fd38203a | 1009 | |
baf76283 | 1010 | r = source_io_register(s, s->enabled, events); |
ec766a51 | 1011 | if (r < 0) |
050f74f2 | 1012 | return r; |
fd38203a | 1013 | |
a71fe8b8 LP |
1014 | if (ret) |
1015 | *ret = s; | |
ec766a51 | 1016 | TAKE_PTR(s); |
a71fe8b8 | 1017 | |
fd38203a LP |
1018 | return 0; |
1019 | } | |
1020 | ||
52444dc4 LP |
1021 | static void initialize_perturb(sd_event *e) { |
1022 | sd_id128_t bootid = {}; | |
1023 | ||
1024 | /* When we sleep for longer, we try to realign the wakeup to | |
f21f31b2 | 1025 | the same time within each minute/second/250ms, so that |
52444dc4 LP |
1026 | events all across the system can be coalesced into a single |
1027 | CPU wakeup. However, let's take some system-specific | |
1028 | randomness for this value, so that in a network of systems | |
1029 | with synced clocks timer events are distributed a | |
1030 | bit. Here, we calculate a perturbation usec offset from the | |
1031 | boot ID. */ | |
1032 | ||
3a43da28 | 1033 | if (_likely_(e->perturb != USEC_INFINITY)) |
52444dc4 LP |
1034 | return; |
1035 | ||
1036 | if (sd_id128_get_boot(&bootid) >= 0) | |
1037 | e->perturb = (bootid.qwords[0] ^ bootid.qwords[1]) % USEC_PER_MINUTE; | |
1038 | } | |
1039 | ||
fd38203a LP |
1040 | static int event_setup_timer_fd( |
1041 | sd_event *e, | |
6a0f1f6d LP |
1042 | struct clock_data *d, |
1043 | clockid_t clock) { | |
fd38203a | 1044 | |
fd38203a | 1045 | assert(e); |
6a0f1f6d | 1046 | assert(d); |
fd38203a | 1047 | |
6a0f1f6d | 1048 | if (_likely_(d->fd >= 0)) |
fd38203a LP |
1049 | return 0; |
1050 | ||
b44d87e2 | 1051 | _cleanup_close_ int fd = -1; |
b44d87e2 | 1052 | |
6a0f1f6d | 1053 | fd = timerfd_create(clock, TFD_NONBLOCK|TFD_CLOEXEC); |
fd38203a LP |
1054 | if (fd < 0) |
1055 | return -errno; | |
1056 | ||
7fe2903c LP |
1057 | fd = fd_move_above_stdio(fd); |
1058 | ||
1eac7948 | 1059 | struct epoll_event ev = { |
a82f89aa LP |
1060 | .events = EPOLLIN, |
1061 | .data.ptr = d, | |
1062 | }; | |
fd38203a | 1063 | |
15c689d7 | 1064 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) |
fd38203a | 1065 | return -errno; |
fd38203a | 1066 | |
b44d87e2 | 1067 | d->fd = TAKE_FD(fd); |
fd38203a LP |
1068 | return 0; |
1069 | } | |
1070 | ||
c4f1aff2 TG |
1071 | static int time_exit_callback(sd_event_source *s, uint64_t usec, void *userdata) { |
1072 | assert(s); | |
1073 | ||
1074 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1075 | } | |
1076 | ||
6a0f1f6d | 1077 | _public_ int sd_event_add_time( |
fd38203a | 1078 | sd_event *e, |
151b9b96 | 1079 | sd_event_source **ret, |
6a0f1f6d | 1080 | clockid_t clock, |
fd38203a | 1081 | uint64_t usec, |
c2ba3ad6 | 1082 | uint64_t accuracy, |
718db961 | 1083 | sd_event_time_handler_t callback, |
151b9b96 | 1084 | void *userdata) { |
fd38203a | 1085 | |
6a0f1f6d | 1086 | EventSourceType type; |
ec766a51 | 1087 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6a0f1f6d | 1088 | struct clock_data *d; |
fd38203a LP |
1089 | int r; |
1090 | ||
305f78bf | 1091 | assert_return(e, -EINVAL); |
b937d761 | 1092 | assert_return(e = event_resolve(e), -ENOPKG); |
305f78bf | 1093 | assert_return(accuracy != (uint64_t) -1, -EINVAL); |
da7e457c | 1094 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1095 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1096 | |
e475d10c LP |
1097 | if (!clock_supported(clock)) /* Checks whether the kernel supports the clock */ |
1098 | return -EOPNOTSUPP; | |
1099 | ||
1100 | type = clock_to_event_source_type(clock); /* checks whether sd-event supports this clock */ | |
1101 | if (type < 0) | |
3411372e LP |
1102 | return -EOPNOTSUPP; |
1103 | ||
c4f1aff2 TG |
1104 | if (!callback) |
1105 | callback = time_exit_callback; | |
1106 | ||
6a0f1f6d LP |
1107 | d = event_get_clock_data(e, type); |
1108 | assert(d); | |
c2ba3ad6 | 1109 | |
c983e776 EV |
1110 | r = prioq_ensure_allocated(&d->earliest, earliest_time_prioq_compare); |
1111 | if (r < 0) | |
1112 | return r; | |
fd38203a | 1113 | |
c983e776 EV |
1114 | r = prioq_ensure_allocated(&d->latest, latest_time_prioq_compare); |
1115 | if (r < 0) | |
1116 | return r; | |
fd38203a | 1117 | |
6a0f1f6d LP |
1118 | if (d->fd < 0) { |
1119 | r = event_setup_timer_fd(e, d, clock); | |
fd38203a LP |
1120 | if (r < 0) |
1121 | return r; | |
1122 | } | |
1123 | ||
a71fe8b8 | 1124 | s = source_new(e, !ret, type); |
fd38203a LP |
1125 | if (!s) |
1126 | return -ENOMEM; | |
1127 | ||
1128 | s->time.next = usec; | |
c2ba3ad6 | 1129 | s->time.accuracy = accuracy == 0 ? DEFAULT_ACCURACY_USEC : accuracy; |
fd38203a | 1130 | s->time.callback = callback; |
da7e457c | 1131 | s->time.earliest_index = s->time.latest_index = PRIOQ_IDX_NULL; |
fd38203a | 1132 | s->userdata = userdata; |
baf76283 | 1133 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1134 | |
e07bbb7c TG |
1135 | d->needs_rearm = true; |
1136 | ||
6a0f1f6d | 1137 | r = prioq_put(d->earliest, s, &s->time.earliest_index); |
c2ba3ad6 | 1138 | if (r < 0) |
ec766a51 | 1139 | return r; |
c2ba3ad6 | 1140 | |
6a0f1f6d | 1141 | r = prioq_put(d->latest, s, &s->time.latest_index); |
c2ba3ad6 | 1142 | if (r < 0) |
ec766a51 | 1143 | return r; |
fd38203a | 1144 | |
a71fe8b8 LP |
1145 | if (ret) |
1146 | *ret = s; | |
ec766a51 | 1147 | TAKE_PTR(s); |
a71fe8b8 | 1148 | |
fd38203a LP |
1149 | return 0; |
1150 | } | |
1151 | ||
d6a83dc4 LP |
1152 | _public_ int sd_event_add_time_relative( |
1153 | sd_event *e, | |
1154 | sd_event_source **ret, | |
1155 | clockid_t clock, | |
1156 | uint64_t usec, | |
1157 | uint64_t accuracy, | |
1158 | sd_event_time_handler_t callback, | |
1159 | void *userdata) { | |
1160 | ||
1161 | usec_t t; | |
1162 | int r; | |
1163 | ||
1164 | /* Same as sd_event_add_time() but operates relative to the event loop's current point in time, and | |
1165 | * checks for overflow. */ | |
1166 | ||
1167 | r = sd_event_now(e, clock, &t); | |
1168 | if (r < 0) | |
1169 | return r; | |
1170 | ||
1171 | if (usec >= USEC_INFINITY - t) | |
1172 | return -EOVERFLOW; | |
1173 | ||
1174 | return sd_event_add_time(e, ret, clock, t + usec, accuracy, callback, userdata); | |
1175 | } | |
1176 | ||
59bc1fd7 LP |
1177 | static int signal_exit_callback(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) { |
1178 | assert(s); | |
1179 | ||
1180 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1181 | } | |
1182 | ||
f7262a9f | 1183 | _public_ int sd_event_add_signal( |
305f78bf | 1184 | sd_event *e, |
151b9b96 | 1185 | sd_event_source **ret, |
305f78bf | 1186 | int sig, |
718db961 | 1187 | sd_event_signal_handler_t callback, |
151b9b96 | 1188 | void *userdata) { |
305f78bf | 1189 | |
ec766a51 | 1190 | _cleanup_(source_freep) sd_event_source *s = NULL; |
9da4cb2b | 1191 | struct signal_data *d; |
fd38203a LP |
1192 | int r; |
1193 | ||
305f78bf | 1194 | assert_return(e, -EINVAL); |
b937d761 | 1195 | assert_return(e = event_resolve(e), -ENOPKG); |
6eb7c172 | 1196 | assert_return(SIGNAL_VALID(sig), -EINVAL); |
da7e457c | 1197 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1198 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1199 | |
59bc1fd7 LP |
1200 | if (!callback) |
1201 | callback = signal_exit_callback; | |
1202 | ||
d1b75241 LP |
1203 | r = signal_is_blocked(sig); |
1204 | if (r < 0) | |
1205 | return r; | |
1206 | if (r == 0) | |
3022d74b LP |
1207 | return -EBUSY; |
1208 | ||
fd38203a LP |
1209 | if (!e->signal_sources) { |
1210 | e->signal_sources = new0(sd_event_source*, _NSIG); | |
1211 | if (!e->signal_sources) | |
1212 | return -ENOMEM; | |
1213 | } else if (e->signal_sources[sig]) | |
1214 | return -EBUSY; | |
1215 | ||
a71fe8b8 | 1216 | s = source_new(e, !ret, SOURCE_SIGNAL); |
fd38203a LP |
1217 | if (!s) |
1218 | return -ENOMEM; | |
1219 | ||
1220 | s->signal.sig = sig; | |
1221 | s->signal.callback = callback; | |
1222 | s->userdata = userdata; | |
baf76283 | 1223 | s->enabled = SD_EVENT_ON; |
fd38203a LP |
1224 | |
1225 | e->signal_sources[sig] = s; | |
fd38203a | 1226 | |
9da4cb2b | 1227 | r = event_make_signal_data(e, sig, &d); |
ec766a51 | 1228 | if (r < 0) |
9da4cb2b | 1229 | return r; |
fd38203a | 1230 | |
f1f00dbb LP |
1231 | /* Use the signal name as description for the event source by default */ |
1232 | (void) sd_event_source_set_description(s, signal_to_string(sig)); | |
1233 | ||
a71fe8b8 LP |
1234 | if (ret) |
1235 | *ret = s; | |
ec766a51 | 1236 | TAKE_PTR(s); |
a71fe8b8 | 1237 | |
fd38203a LP |
1238 | return 0; |
1239 | } | |
1240 | ||
b9350e70 LP |
1241 | static int child_exit_callback(sd_event_source *s, const siginfo_t *si, void *userdata) { |
1242 | assert(s); | |
1243 | ||
1244 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1245 | } | |
1246 | ||
f8f3f926 LP |
1247 | static bool shall_use_pidfd(void) { |
1248 | /* Mostly relevant for debugging, i.e. this is used in test-event.c to test the event loop once with and once without pidfd */ | |
1249 | return getenv_bool_secure("SYSTEMD_PIDFD") != 0; | |
1250 | } | |
1251 | ||
f7262a9f | 1252 | _public_ int sd_event_add_child( |
305f78bf | 1253 | sd_event *e, |
151b9b96 | 1254 | sd_event_source **ret, |
305f78bf LP |
1255 | pid_t pid, |
1256 | int options, | |
718db961 | 1257 | sd_event_child_handler_t callback, |
151b9b96 | 1258 | void *userdata) { |
305f78bf | 1259 | |
ec766a51 | 1260 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1261 | int r; |
1262 | ||
305f78bf | 1263 | assert_return(e, -EINVAL); |
b937d761 | 1264 | assert_return(e = event_resolve(e), -ENOPKG); |
305f78bf LP |
1265 | assert_return(pid > 1, -EINVAL); |
1266 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1267 | assert_return(options != 0, -EINVAL); | |
da7e457c | 1268 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1269 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1270 | |
b9350e70 LP |
1271 | if (!callback) |
1272 | callback = child_exit_callback; | |
1273 | ||
ee880b37 LP |
1274 | if (e->n_enabled_child_sources == 0) { |
1275 | /* Caller must block SIGCHLD before using us to watch children, even if pidfd is available, | |
1276 | * for compatibility with pre-pidfd and because we don't want the reap the child processes | |
1277 | * ourselves, i.e. call waitid(), and don't want Linux' default internal logic for that to | |
1278 | * take effect. | |
1279 | * | |
1280 | * (As an optimization we only do this check on the first child event source created.) */ | |
1281 | r = signal_is_blocked(SIGCHLD); | |
1282 | if (r < 0) | |
1283 | return r; | |
1284 | if (r == 0) | |
1285 | return -EBUSY; | |
1286 | } | |
1287 | ||
d5099efc | 1288 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
fd38203a LP |
1289 | if (r < 0) |
1290 | return r; | |
1291 | ||
4a0b58c4 | 1292 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) |
fd38203a LP |
1293 | return -EBUSY; |
1294 | ||
a71fe8b8 | 1295 | s = source_new(e, !ret, SOURCE_CHILD); |
fd38203a LP |
1296 | if (!s) |
1297 | return -ENOMEM; | |
1298 | ||
f8f3f926 | 1299 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1300 | s->child.pid = pid; |
1301 | s->child.options = options; | |
1302 | s->child.callback = callback; | |
1303 | s->userdata = userdata; | |
baf76283 | 1304 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1305 | |
f8f3f926 LP |
1306 | /* We always take a pidfd here if we can, even if we wait for anything else than WEXITED, so that we |
1307 | * pin the PID, and make regular waitid() handling race-free. */ | |
1308 | ||
1309 | if (shall_use_pidfd()) { | |
1310 | s->child.pidfd = pidfd_open(s->child.pid, 0); | |
1311 | if (s->child.pidfd < 0) { | |
1312 | /* Propagate errors unless the syscall is not supported or blocked */ | |
1313 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
1314 | return -errno; | |
1315 | } else | |
1316 | s->child.pidfd_owned = true; /* If we allocate the pidfd we own it by default */ | |
1317 | } else | |
1318 | s->child.pidfd = -1; | |
1319 | ||
4a0b58c4 | 1320 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); |
ec766a51 | 1321 | if (r < 0) |
fd38203a | 1322 | return r; |
fd38203a | 1323 | |
313cefa1 | 1324 | e->n_enabled_child_sources++; |
fd38203a | 1325 | |
f8f3f926 LP |
1326 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1327 | /* We have a pidfd and we only want to watch for exit */ | |
1328 | ||
1329 | r = source_child_pidfd_register(s, s->enabled); | |
1330 | if (r < 0) { | |
1331 | e->n_enabled_child_sources--; | |
1332 | return r; | |
1333 | } | |
1334 | } else { | |
1335 | /* We have no pidfd or we shall wait for some other event than WEXITED */ | |
fd38203a | 1336 | |
f8f3f926 LP |
1337 | r = event_make_signal_data(e, SIGCHLD, NULL); |
1338 | if (r < 0) { | |
1339 | e->n_enabled_child_sources--; | |
1340 | return r; | |
1341 | } | |
1342 | ||
1343 | e->need_process_child = true; | |
1344 | } | |
c2ba3ad6 | 1345 | |
a71fe8b8 LP |
1346 | if (ret) |
1347 | *ret = s; | |
f8f3f926 | 1348 | |
ec766a51 | 1349 | TAKE_PTR(s); |
f8f3f926 LP |
1350 | return 0; |
1351 | } | |
1352 | ||
1353 | _public_ int sd_event_add_child_pidfd( | |
1354 | sd_event *e, | |
1355 | sd_event_source **ret, | |
1356 | int pidfd, | |
1357 | int options, | |
1358 | sd_event_child_handler_t callback, | |
1359 | void *userdata) { | |
1360 | ||
1361 | ||
1362 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
1363 | pid_t pid; | |
1364 | int r; | |
1365 | ||
1366 | assert_return(e, -EINVAL); | |
1367 | assert_return(e = event_resolve(e), -ENOPKG); | |
1368 | assert_return(pidfd >= 0, -EBADF); | |
1369 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1370 | assert_return(options != 0, -EINVAL); | |
f8f3f926 LP |
1371 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1372 | assert_return(!event_pid_changed(e), -ECHILD); | |
1373 | ||
b9350e70 LP |
1374 | if (!callback) |
1375 | callback = child_exit_callback; | |
1376 | ||
ee880b37 LP |
1377 | if (e->n_enabled_child_sources == 0) { |
1378 | r = signal_is_blocked(SIGCHLD); | |
1379 | if (r < 0) | |
1380 | return r; | |
1381 | if (r == 0) | |
1382 | return -EBUSY; | |
1383 | } | |
1384 | ||
f8f3f926 LP |
1385 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
1386 | if (r < 0) | |
1387 | return r; | |
1388 | ||
1389 | r = pidfd_get_pid(pidfd, &pid); | |
1390 | if (r < 0) | |
1391 | return r; | |
1392 | ||
1393 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) | |
1394 | return -EBUSY; | |
1395 | ||
1396 | s = source_new(e, !ret, SOURCE_CHILD); | |
1397 | if (!s) | |
1398 | return -ENOMEM; | |
1399 | ||
1400 | s->wakeup = WAKEUP_EVENT_SOURCE; | |
1401 | s->child.pidfd = pidfd; | |
1402 | s->child.pid = pid; | |
1403 | s->child.options = options; | |
1404 | s->child.callback = callback; | |
1405 | 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) */ | |
1406 | s->userdata = userdata; | |
1407 | s->enabled = SD_EVENT_ONESHOT; | |
1408 | ||
1409 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); | |
1410 | if (r < 0) | |
1411 | return r; | |
1412 | ||
1413 | e->n_enabled_child_sources++; | |
1414 | ||
1415 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { | |
1416 | /* We only want to watch for WEXITED */ | |
1417 | ||
1418 | r = source_child_pidfd_register(s, s->enabled); | |
1419 | if (r < 0) { | |
1420 | e->n_enabled_child_sources--; | |
1421 | return r; | |
1422 | } | |
1423 | } else { | |
1424 | /* We shall wait for some other event than WEXITED */ | |
1425 | ||
1426 | r = event_make_signal_data(e, SIGCHLD, NULL); | |
1427 | if (r < 0) { | |
1428 | e->n_enabled_child_sources--; | |
1429 | return r; | |
1430 | } | |
a71fe8b8 | 1431 | |
f8f3f926 LP |
1432 | e->need_process_child = true; |
1433 | } | |
1434 | ||
1435 | if (ret) | |
1436 | *ret = s; | |
1437 | ||
1438 | TAKE_PTR(s); | |
fd38203a LP |
1439 | return 0; |
1440 | } | |
1441 | ||
b9350e70 LP |
1442 | static int generic_exit_callback(sd_event_source *s, void *userdata) { |
1443 | assert(s); | |
1444 | ||
1445 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1446 | } | |
1447 | ||
f7262a9f | 1448 | _public_ int sd_event_add_defer( |
305f78bf | 1449 | sd_event *e, |
151b9b96 | 1450 | sd_event_source **ret, |
718db961 | 1451 | sd_event_handler_t callback, |
151b9b96 | 1452 | void *userdata) { |
305f78bf | 1453 | |
ec766a51 | 1454 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1455 | int r; |
1456 | ||
305f78bf | 1457 | assert_return(e, -EINVAL); |
b937d761 | 1458 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1459 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1460 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1461 | |
b9350e70 LP |
1462 | if (!callback) |
1463 | callback = generic_exit_callback; | |
1464 | ||
a71fe8b8 | 1465 | s = source_new(e, !ret, SOURCE_DEFER); |
fd38203a LP |
1466 | if (!s) |
1467 | return -ENOMEM; | |
1468 | ||
1469 | s->defer.callback = callback; | |
1470 | s->userdata = userdata; | |
baf76283 | 1471 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a LP |
1472 | |
1473 | r = source_set_pending(s, true); | |
ec766a51 | 1474 | if (r < 0) |
fd38203a | 1475 | return r; |
fd38203a | 1476 | |
a71fe8b8 LP |
1477 | if (ret) |
1478 | *ret = s; | |
ec766a51 | 1479 | TAKE_PTR(s); |
a71fe8b8 | 1480 | |
fd38203a LP |
1481 | return 0; |
1482 | } | |
1483 | ||
6e9feda3 LP |
1484 | _public_ int sd_event_add_post( |
1485 | sd_event *e, | |
1486 | sd_event_source **ret, | |
1487 | sd_event_handler_t callback, | |
1488 | void *userdata) { | |
1489 | ||
ec766a51 | 1490 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6e9feda3 LP |
1491 | int r; |
1492 | ||
1493 | assert_return(e, -EINVAL); | |
b937d761 | 1494 | assert_return(e = event_resolve(e), -ENOPKG); |
6e9feda3 LP |
1495 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1496 | assert_return(!event_pid_changed(e), -ECHILD); | |
1497 | ||
b9350e70 LP |
1498 | if (!callback) |
1499 | callback = generic_exit_callback; | |
1500 | ||
a71fe8b8 | 1501 | s = source_new(e, !ret, SOURCE_POST); |
6e9feda3 LP |
1502 | if (!s) |
1503 | return -ENOMEM; | |
1504 | ||
1505 | s->post.callback = callback; | |
1506 | s->userdata = userdata; | |
1507 | s->enabled = SD_EVENT_ON; | |
1508 | ||
de7fef4b | 1509 | r = set_ensure_put(&e->post_sources, NULL, s); |
ec766a51 | 1510 | if (r < 0) |
6e9feda3 | 1511 | return r; |
de7fef4b | 1512 | assert(r > 0); |
6e9feda3 | 1513 | |
a71fe8b8 LP |
1514 | if (ret) |
1515 | *ret = s; | |
ec766a51 | 1516 | TAKE_PTR(s); |
a71fe8b8 | 1517 | |
6e9feda3 LP |
1518 | return 0; |
1519 | } | |
1520 | ||
6203e07a | 1521 | _public_ int sd_event_add_exit( |
305f78bf | 1522 | sd_event *e, |
151b9b96 | 1523 | sd_event_source **ret, |
718db961 | 1524 | sd_event_handler_t callback, |
151b9b96 | 1525 | void *userdata) { |
305f78bf | 1526 | |
ec766a51 | 1527 | _cleanup_(source_freep) sd_event_source *s = NULL; |
da7e457c LP |
1528 | int r; |
1529 | ||
1530 | assert_return(e, -EINVAL); | |
b937d761 | 1531 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
1532 | assert_return(callback, -EINVAL); |
1533 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
1534 | assert_return(!event_pid_changed(e), -ECHILD); | |
1535 | ||
c983e776 EV |
1536 | r = prioq_ensure_allocated(&e->exit, exit_prioq_compare); |
1537 | if (r < 0) | |
1538 | return r; | |
da7e457c | 1539 | |
a71fe8b8 | 1540 | s = source_new(e, !ret, SOURCE_EXIT); |
fd38203a | 1541 | if (!s) |
da7e457c | 1542 | return -ENOMEM; |
fd38203a | 1543 | |
6203e07a | 1544 | s->exit.callback = callback; |
da7e457c | 1545 | s->userdata = userdata; |
6203e07a | 1546 | s->exit.prioq_index = PRIOQ_IDX_NULL; |
baf76283 | 1547 | s->enabled = SD_EVENT_ONESHOT; |
da7e457c | 1548 | |
6203e07a | 1549 | r = prioq_put(s->event->exit, s, &s->exit.prioq_index); |
ec766a51 | 1550 | if (r < 0) |
da7e457c | 1551 | return r; |
da7e457c | 1552 | |
a71fe8b8 LP |
1553 | if (ret) |
1554 | *ret = s; | |
ec766a51 | 1555 | TAKE_PTR(s); |
a71fe8b8 | 1556 | |
da7e457c LP |
1557 | return 0; |
1558 | } | |
1559 | ||
97ef5391 LP |
1560 | static void event_free_inotify_data(sd_event *e, struct inotify_data *d) { |
1561 | assert(e); | |
1562 | ||
1563 | if (!d) | |
1564 | return; | |
1565 | ||
1566 | assert(hashmap_isempty(d->inodes)); | |
1567 | assert(hashmap_isempty(d->wd)); | |
1568 | ||
1569 | if (d->buffer_filled > 0) | |
1570 | LIST_REMOVE(buffered, e->inotify_data_buffered, d); | |
1571 | ||
1572 | hashmap_free(d->inodes); | |
1573 | hashmap_free(d->wd); | |
1574 | ||
1575 | assert_se(hashmap_remove(e->inotify_data, &d->priority) == d); | |
1576 | ||
1577 | if (d->fd >= 0) { | |
1578 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, d->fd, NULL) < 0) | |
1579 | log_debug_errno(errno, "Failed to remove inotify fd from epoll, ignoring: %m"); | |
1580 | ||
1581 | safe_close(d->fd); | |
1582 | } | |
1583 | free(d); | |
1584 | } | |
1585 | ||
1586 | static int event_make_inotify_data( | |
1587 | sd_event *e, | |
1588 | int64_t priority, | |
1589 | struct inotify_data **ret) { | |
1590 | ||
1591 | _cleanup_close_ int fd = -1; | |
1592 | struct inotify_data *d; | |
97ef5391 LP |
1593 | int r; |
1594 | ||
1595 | assert(e); | |
1596 | ||
1597 | d = hashmap_get(e->inotify_data, &priority); | |
1598 | if (d) { | |
1599 | if (ret) | |
1600 | *ret = d; | |
1601 | return 0; | |
1602 | } | |
1603 | ||
1604 | fd = inotify_init1(IN_NONBLOCK|O_CLOEXEC); | |
1605 | if (fd < 0) | |
1606 | return -errno; | |
1607 | ||
1608 | fd = fd_move_above_stdio(fd); | |
1609 | ||
1610 | r = hashmap_ensure_allocated(&e->inotify_data, &uint64_hash_ops); | |
1611 | if (r < 0) | |
1612 | return r; | |
1613 | ||
1614 | d = new(struct inotify_data, 1); | |
1615 | if (!d) | |
1616 | return -ENOMEM; | |
1617 | ||
1618 | *d = (struct inotify_data) { | |
1619 | .wakeup = WAKEUP_INOTIFY_DATA, | |
1620 | .fd = TAKE_FD(fd), | |
1621 | .priority = priority, | |
1622 | }; | |
1623 | ||
1624 | r = hashmap_put(e->inotify_data, &d->priority, d); | |
1625 | if (r < 0) { | |
1626 | d->fd = safe_close(d->fd); | |
1627 | free(d); | |
1628 | return r; | |
1629 | } | |
1630 | ||
1eac7948 | 1631 | struct epoll_event ev = { |
97ef5391 LP |
1632 | .events = EPOLLIN, |
1633 | .data.ptr = d, | |
1634 | }; | |
1635 | ||
1636 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { | |
1637 | r = -errno; | |
1638 | d->fd = safe_close(d->fd); /* let's close this ourselves, as event_free_inotify_data() would otherwise | |
1639 | * remove the fd from the epoll first, which we don't want as we couldn't | |
1640 | * add it in the first place. */ | |
1641 | event_free_inotify_data(e, d); | |
1642 | return r; | |
1643 | } | |
1644 | ||
1645 | if (ret) | |
1646 | *ret = d; | |
1647 | ||
1648 | return 1; | |
1649 | } | |
1650 | ||
7a08d314 | 1651 | static int inode_data_compare(const struct inode_data *x, const struct inode_data *y) { |
90c88092 | 1652 | int r; |
97ef5391 LP |
1653 | |
1654 | assert(x); | |
1655 | assert(y); | |
1656 | ||
90c88092 YW |
1657 | r = CMP(x->dev, y->dev); |
1658 | if (r != 0) | |
1659 | return r; | |
97ef5391 | 1660 | |
6dd91b36 | 1661 | return CMP(x->ino, y->ino); |
97ef5391 LP |
1662 | } |
1663 | ||
7a08d314 YW |
1664 | static void inode_data_hash_func(const struct inode_data *d, struct siphash *state) { |
1665 | assert(d); | |
97ef5391 LP |
1666 | |
1667 | siphash24_compress(&d->dev, sizeof(d->dev), state); | |
1668 | siphash24_compress(&d->ino, sizeof(d->ino), state); | |
1669 | } | |
1670 | ||
7a08d314 | 1671 | DEFINE_PRIVATE_HASH_OPS(inode_data_hash_ops, struct inode_data, inode_data_hash_func, inode_data_compare); |
97ef5391 LP |
1672 | |
1673 | static void event_free_inode_data( | |
1674 | sd_event *e, | |
1675 | struct inode_data *d) { | |
1676 | ||
1677 | assert(e); | |
1678 | ||
1679 | if (!d) | |
1680 | return; | |
1681 | ||
1682 | assert(!d->event_sources); | |
1683 | ||
1684 | if (d->fd >= 0) { | |
1685 | LIST_REMOVE(to_close, e->inode_data_to_close, d); | |
1686 | safe_close(d->fd); | |
1687 | } | |
1688 | ||
1689 | if (d->inotify_data) { | |
1690 | ||
1691 | if (d->wd >= 0) { | |
1692 | if (d->inotify_data->fd >= 0) { | |
1693 | /* So here's a problem. At the time this runs the watch descriptor might already be | |
1694 | * invalidated, because an IN_IGNORED event might be queued right the moment we enter | |
1695 | * the syscall. Hence, whenever we get EINVAL, ignore it entirely, since it's a very | |
1696 | * likely case to happen. */ | |
1697 | ||
1698 | if (inotify_rm_watch(d->inotify_data->fd, d->wd) < 0 && errno != EINVAL) | |
1699 | log_debug_errno(errno, "Failed to remove watch descriptor %i from inotify, ignoring: %m", d->wd); | |
1700 | } | |
1701 | ||
1702 | assert_se(hashmap_remove(d->inotify_data->wd, INT_TO_PTR(d->wd)) == d); | |
1703 | } | |
1704 | ||
1705 | assert_se(hashmap_remove(d->inotify_data->inodes, d) == d); | |
1706 | } | |
1707 | ||
1708 | free(d); | |
1709 | } | |
1710 | ||
1711 | static void event_gc_inode_data( | |
1712 | sd_event *e, | |
1713 | struct inode_data *d) { | |
1714 | ||
1715 | struct inotify_data *inotify_data; | |
1716 | ||
1717 | assert(e); | |
1718 | ||
1719 | if (!d) | |
1720 | return; | |
1721 | ||
1722 | if (d->event_sources) | |
1723 | return; | |
1724 | ||
1725 | inotify_data = d->inotify_data; | |
1726 | event_free_inode_data(e, d); | |
1727 | ||
1728 | if (inotify_data && hashmap_isempty(inotify_data->inodes)) | |
1729 | event_free_inotify_data(e, inotify_data); | |
1730 | } | |
1731 | ||
1732 | static int event_make_inode_data( | |
1733 | sd_event *e, | |
1734 | struct inotify_data *inotify_data, | |
1735 | dev_t dev, | |
1736 | ino_t ino, | |
1737 | struct inode_data **ret) { | |
1738 | ||
1739 | struct inode_data *d, key; | |
1740 | int r; | |
1741 | ||
1742 | assert(e); | |
1743 | assert(inotify_data); | |
1744 | ||
1745 | key = (struct inode_data) { | |
1746 | .ino = ino, | |
1747 | .dev = dev, | |
1748 | }; | |
1749 | ||
1750 | d = hashmap_get(inotify_data->inodes, &key); | |
1751 | if (d) { | |
1752 | if (ret) | |
1753 | *ret = d; | |
1754 | ||
1755 | return 0; | |
1756 | } | |
1757 | ||
1758 | r = hashmap_ensure_allocated(&inotify_data->inodes, &inode_data_hash_ops); | |
1759 | if (r < 0) | |
1760 | return r; | |
1761 | ||
1762 | d = new(struct inode_data, 1); | |
1763 | if (!d) | |
1764 | return -ENOMEM; | |
1765 | ||
1766 | *d = (struct inode_data) { | |
1767 | .dev = dev, | |
1768 | .ino = ino, | |
1769 | .wd = -1, | |
1770 | .fd = -1, | |
1771 | .inotify_data = inotify_data, | |
1772 | }; | |
1773 | ||
1774 | r = hashmap_put(inotify_data->inodes, d, d); | |
1775 | if (r < 0) { | |
1776 | free(d); | |
1777 | return r; | |
1778 | } | |
1779 | ||
1780 | if (ret) | |
1781 | *ret = d; | |
1782 | ||
1783 | return 1; | |
1784 | } | |
1785 | ||
1786 | static uint32_t inode_data_determine_mask(struct inode_data *d) { | |
1787 | bool excl_unlink = true; | |
1788 | uint32_t combined = 0; | |
1789 | sd_event_source *s; | |
1790 | ||
1791 | assert(d); | |
1792 | ||
1793 | /* Combines the watch masks of all event sources watching this inode. We generally just OR them together, but | |
1794 | * the IN_EXCL_UNLINK flag is ANDed instead. | |
1795 | * | |
1796 | * Note that we add all sources to the mask here, regardless whether enabled, disabled or oneshot. That's | |
1797 | * because we cannot change the mask anymore after the event source was created once, since the kernel has no | |
f21f31b2 | 1798 | * API for that. Hence we need to subscribe to the maximum mask we ever might be interested in, and suppress |
97ef5391 LP |
1799 | * events we don't care for client-side. */ |
1800 | ||
1801 | LIST_FOREACH(inotify.by_inode_data, s, d->event_sources) { | |
1802 | ||
1803 | if ((s->inotify.mask & IN_EXCL_UNLINK) == 0) | |
1804 | excl_unlink = false; | |
1805 | ||
1806 | combined |= s->inotify.mask; | |
1807 | } | |
1808 | ||
1809 | return (combined & ~(IN_ONESHOT|IN_DONT_FOLLOW|IN_ONLYDIR|IN_EXCL_UNLINK)) | (excl_unlink ? IN_EXCL_UNLINK : 0); | |
1810 | } | |
1811 | ||
1812 | static int inode_data_realize_watch(sd_event *e, struct inode_data *d) { | |
1813 | uint32_t combined_mask; | |
1814 | int wd, r; | |
1815 | ||
1816 | assert(d); | |
1817 | assert(d->fd >= 0); | |
1818 | ||
1819 | combined_mask = inode_data_determine_mask(d); | |
1820 | ||
1821 | if (d->wd >= 0 && combined_mask == d->combined_mask) | |
1822 | return 0; | |
1823 | ||
1824 | r = hashmap_ensure_allocated(&d->inotify_data->wd, NULL); | |
1825 | if (r < 0) | |
1826 | return r; | |
1827 | ||
1828 | wd = inotify_add_watch_fd(d->inotify_data->fd, d->fd, combined_mask); | |
1829 | if (wd < 0) | |
1830 | return -errno; | |
1831 | ||
1832 | if (d->wd < 0) { | |
1833 | r = hashmap_put(d->inotify_data->wd, INT_TO_PTR(wd), d); | |
1834 | if (r < 0) { | |
1835 | (void) inotify_rm_watch(d->inotify_data->fd, wd); | |
1836 | return r; | |
1837 | } | |
1838 | ||
1839 | d->wd = wd; | |
1840 | ||
1841 | } else if (d->wd != wd) { | |
1842 | ||
1843 | log_debug("Weird, the watch descriptor we already knew for this inode changed?"); | |
1844 | (void) inotify_rm_watch(d->fd, wd); | |
1845 | return -EINVAL; | |
1846 | } | |
1847 | ||
1848 | d->combined_mask = combined_mask; | |
1849 | return 1; | |
1850 | } | |
1851 | ||
b9350e70 LP |
1852 | static int inotify_exit_callback(sd_event_source *s, const struct inotify_event *event, void *userdata) { |
1853 | assert(s); | |
1854 | ||
1855 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1856 | } | |
1857 | ||
97ef5391 LP |
1858 | _public_ int sd_event_add_inotify( |
1859 | sd_event *e, | |
1860 | sd_event_source **ret, | |
1861 | const char *path, | |
1862 | uint32_t mask, | |
1863 | sd_event_inotify_handler_t callback, | |
1864 | void *userdata) { | |
1865 | ||
97ef5391 LP |
1866 | struct inotify_data *inotify_data = NULL; |
1867 | struct inode_data *inode_data = NULL; | |
1868 | _cleanup_close_ int fd = -1; | |
8c75fe17 | 1869 | _cleanup_(source_freep) sd_event_source *s = NULL; |
97ef5391 LP |
1870 | struct stat st; |
1871 | int r; | |
1872 | ||
1873 | assert_return(e, -EINVAL); | |
1874 | assert_return(e = event_resolve(e), -ENOPKG); | |
1875 | assert_return(path, -EINVAL); | |
97ef5391 LP |
1876 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1877 | assert_return(!event_pid_changed(e), -ECHILD); | |
1878 | ||
b9350e70 LP |
1879 | if (!callback) |
1880 | callback = inotify_exit_callback; | |
1881 | ||
97ef5391 LP |
1882 | /* Refuse IN_MASK_ADD since we coalesce watches on the same inode, and hence really don't want to merge |
1883 | * masks. Or in other words, this whole code exists only to manage IN_MASK_ADD type operations for you, hence | |
1884 | * the user can't use them for us. */ | |
1885 | if (mask & IN_MASK_ADD) | |
1886 | return -EINVAL; | |
1887 | ||
1888 | fd = open(path, O_PATH|O_CLOEXEC| | |
1889 | (mask & IN_ONLYDIR ? O_DIRECTORY : 0)| | |
1890 | (mask & IN_DONT_FOLLOW ? O_NOFOLLOW : 0)); | |
1891 | if (fd < 0) | |
1892 | return -errno; | |
1893 | ||
1894 | if (fstat(fd, &st) < 0) | |
1895 | return -errno; | |
1896 | ||
1897 | s = source_new(e, !ret, SOURCE_INOTIFY); | |
1898 | if (!s) | |
1899 | return -ENOMEM; | |
1900 | ||
1901 | s->enabled = mask & IN_ONESHOT ? SD_EVENT_ONESHOT : SD_EVENT_ON; | |
1902 | s->inotify.mask = mask; | |
1903 | s->inotify.callback = callback; | |
1904 | s->userdata = userdata; | |
1905 | ||
1906 | /* Allocate an inotify object for this priority, and an inode object within it */ | |
1907 | r = event_make_inotify_data(e, SD_EVENT_PRIORITY_NORMAL, &inotify_data); | |
1908 | if (r < 0) | |
8c75fe17 | 1909 | return r; |
97ef5391 LP |
1910 | |
1911 | r = event_make_inode_data(e, inotify_data, st.st_dev, st.st_ino, &inode_data); | |
8c75fe17 ZJS |
1912 | if (r < 0) { |
1913 | event_free_inotify_data(e, inotify_data); | |
1914 | return r; | |
1915 | } | |
97ef5391 LP |
1916 | |
1917 | /* Keep the O_PATH fd around until the first iteration of the loop, so that we can still change the priority of | |
1918 | * the event source, until then, for which we need the original inode. */ | |
1919 | if (inode_data->fd < 0) { | |
1920 | inode_data->fd = TAKE_FD(fd); | |
1921 | LIST_PREPEND(to_close, e->inode_data_to_close, inode_data); | |
1922 | } | |
1923 | ||
1924 | /* Link our event source to the inode data object */ | |
1925 | LIST_PREPEND(inotify.by_inode_data, inode_data->event_sources, s); | |
1926 | s->inotify.inode_data = inode_data; | |
1927 | ||
97ef5391 LP |
1928 | /* Actually realize the watch now */ |
1929 | r = inode_data_realize_watch(e, inode_data); | |
1930 | if (r < 0) | |
8c75fe17 | 1931 | return r; |
97ef5391 LP |
1932 | |
1933 | (void) sd_event_source_set_description(s, path); | |
1934 | ||
1935 | if (ret) | |
1936 | *ret = s; | |
8c75fe17 | 1937 | TAKE_PTR(s); |
97ef5391 LP |
1938 | |
1939 | return 0; | |
97ef5391 LP |
1940 | } |
1941 | ||
8301aa0b | 1942 | static sd_event_source* event_source_free(sd_event_source *s) { |
6680dd6b LP |
1943 | if (!s) |
1944 | return NULL; | |
da7e457c | 1945 | |
8301aa0b YW |
1946 | /* Here's a special hack: when we are called from a |
1947 | * dispatch handler we won't free the event source | |
1948 | * immediately, but we will detach the fd from the | |
1949 | * epoll. This way it is safe for the caller to unref | |
1950 | * the event source and immediately close the fd, but | |
1951 | * we still retain a valid event source object after | |
1952 | * the callback. */ | |
fd38203a | 1953 | |
8301aa0b YW |
1954 | if (s->dispatching) { |
1955 | if (s->type == SOURCE_IO) | |
1956 | source_io_unregister(s); | |
fd38203a | 1957 | |
8301aa0b YW |
1958 | source_disconnect(s); |
1959 | } else | |
1960 | source_free(s); | |
fd38203a LP |
1961 | |
1962 | return NULL; | |
1963 | } | |
1964 | ||
8301aa0b YW |
1965 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event_source, sd_event_source, event_source_free); |
1966 | ||
356779df | 1967 | _public_ int sd_event_source_set_description(sd_event_source *s, const char *description) { |
f7f53e9e | 1968 | assert_return(s, -EINVAL); |
f4b2933e | 1969 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 1970 | |
356779df | 1971 | return free_and_strdup(&s->description, description); |
f7f53e9e TG |
1972 | } |
1973 | ||
356779df | 1974 | _public_ int sd_event_source_get_description(sd_event_source *s, const char **description) { |
f7f53e9e | 1975 | assert_return(s, -EINVAL); |
356779df | 1976 | assert_return(description, -EINVAL); |
f4b2933e | 1977 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 1978 | |
7d92a1a4 ZJS |
1979 | if (!s->description) |
1980 | return -ENXIO; | |
1981 | ||
356779df | 1982 | *description = s->description; |
f7f53e9e TG |
1983 | return 0; |
1984 | } | |
1985 | ||
adcc4ca3 | 1986 | _public_ sd_event *sd_event_source_get_event(sd_event_source *s) { |
305f78bf | 1987 | assert_return(s, NULL); |
eaa3cbef LP |
1988 | |
1989 | return s->event; | |
1990 | } | |
1991 | ||
f7262a9f | 1992 | _public_ int sd_event_source_get_pending(sd_event_source *s) { |
305f78bf | 1993 | assert_return(s, -EINVAL); |
6203e07a | 1994 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c | 1995 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1996 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
1997 | |
1998 | return s->pending; | |
1999 | } | |
2000 | ||
f7262a9f | 2001 | _public_ int sd_event_source_get_io_fd(sd_event_source *s) { |
305f78bf LP |
2002 | assert_return(s, -EINVAL); |
2003 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2004 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2005 | |
2006 | return s->io.fd; | |
2007 | } | |
2008 | ||
30caf8f3 LP |
2009 | _public_ int sd_event_source_set_io_fd(sd_event_source *s, int fd) { |
2010 | int r; | |
2011 | ||
2012 | assert_return(s, -EINVAL); | |
8ac43fee | 2013 | assert_return(fd >= 0, -EBADF); |
30caf8f3 LP |
2014 | assert_return(s->type == SOURCE_IO, -EDOM); |
2015 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2016 | ||
2017 | if (s->io.fd == fd) | |
2018 | return 0; | |
2019 | ||
2020 | if (s->enabled == SD_EVENT_OFF) { | |
2021 | s->io.fd = fd; | |
2022 | s->io.registered = false; | |
2023 | } else { | |
2024 | int saved_fd; | |
2025 | ||
2026 | saved_fd = s->io.fd; | |
2027 | assert(s->io.registered); | |
2028 | ||
2029 | s->io.fd = fd; | |
2030 | s->io.registered = false; | |
2031 | ||
2032 | r = source_io_register(s, s->enabled, s->io.events); | |
2033 | if (r < 0) { | |
2034 | s->io.fd = saved_fd; | |
2035 | s->io.registered = true; | |
2036 | return r; | |
2037 | } | |
2038 | ||
5a795bff | 2039 | (void) epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, saved_fd, NULL); |
30caf8f3 LP |
2040 | } |
2041 | ||
2042 | return 0; | |
2043 | } | |
2044 | ||
ab93297c NM |
2045 | _public_ int sd_event_source_get_io_fd_own(sd_event_source *s) { |
2046 | assert_return(s, -EINVAL); | |
2047 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2048 | ||
2049 | return s->io.owned; | |
2050 | } | |
2051 | ||
2052 | _public_ int sd_event_source_set_io_fd_own(sd_event_source *s, int own) { | |
2053 | assert_return(s, -EINVAL); | |
2054 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2055 | ||
2056 | s->io.owned = own; | |
2057 | return 0; | |
2058 | } | |
2059 | ||
f7262a9f | 2060 | _public_ int sd_event_source_get_io_events(sd_event_source *s, uint32_t* events) { |
305f78bf LP |
2061 | assert_return(s, -EINVAL); |
2062 | assert_return(events, -EINVAL); | |
2063 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2064 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2065 | |
2066 | *events = s->io.events; | |
2067 | return 0; | |
2068 | } | |
2069 | ||
f7262a9f | 2070 | _public_ int sd_event_source_set_io_events(sd_event_source *s, uint32_t events) { |
fd38203a LP |
2071 | int r; |
2072 | ||
305f78bf LP |
2073 | assert_return(s, -EINVAL); |
2074 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2a16a986 | 2075 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 2076 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2077 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2078 | |
b63c8d4f DH |
2079 | /* edge-triggered updates are never skipped, so we can reset edges */ |
2080 | if (s->io.events == events && !(events & EPOLLET)) | |
fd38203a LP |
2081 | return 0; |
2082 | ||
2a0dc6cd LP |
2083 | r = source_set_pending(s, false); |
2084 | if (r < 0) | |
2085 | return r; | |
2086 | ||
baf76283 | 2087 | if (s->enabled != SD_EVENT_OFF) { |
e4715127 | 2088 | r = source_io_register(s, s->enabled, events); |
fd38203a LP |
2089 | if (r < 0) |
2090 | return r; | |
2091 | } | |
2092 | ||
2093 | s->io.events = events; | |
2094 | ||
2095 | return 0; | |
2096 | } | |
2097 | ||
f7262a9f | 2098 | _public_ int sd_event_source_get_io_revents(sd_event_source *s, uint32_t* revents) { |
305f78bf LP |
2099 | assert_return(s, -EINVAL); |
2100 | assert_return(revents, -EINVAL); | |
2101 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2102 | assert_return(s->pending, -ENODATA); | |
2103 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2104 | |
2105 | *revents = s->io.revents; | |
2106 | return 0; | |
2107 | } | |
2108 | ||
f7262a9f | 2109 | _public_ int sd_event_source_get_signal(sd_event_source *s) { |
305f78bf LP |
2110 | assert_return(s, -EINVAL); |
2111 | assert_return(s->type == SOURCE_SIGNAL, -EDOM); | |
2112 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2113 | |
2114 | return s->signal.sig; | |
2115 | } | |
2116 | ||
31927c16 | 2117 | _public_ int sd_event_source_get_priority(sd_event_source *s, int64_t *priority) { |
305f78bf LP |
2118 | assert_return(s, -EINVAL); |
2119 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a | 2120 | |
6680b8d1 ME |
2121 | *priority = s->priority; |
2122 | return 0; | |
fd38203a LP |
2123 | } |
2124 | ||
31927c16 | 2125 | _public_ int sd_event_source_set_priority(sd_event_source *s, int64_t priority) { |
97ef5391 LP |
2126 | bool rm_inotify = false, rm_inode = false; |
2127 | struct inotify_data *new_inotify_data = NULL; | |
2128 | struct inode_data *new_inode_data = NULL; | |
9da4cb2b LP |
2129 | int r; |
2130 | ||
305f78bf | 2131 | assert_return(s, -EINVAL); |
da7e457c | 2132 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2133 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2134 | |
2135 | if (s->priority == priority) | |
2136 | return 0; | |
2137 | ||
97ef5391 LP |
2138 | if (s->type == SOURCE_INOTIFY) { |
2139 | struct inode_data *old_inode_data; | |
2140 | ||
2141 | assert(s->inotify.inode_data); | |
2142 | old_inode_data = s->inotify.inode_data; | |
2143 | ||
2144 | /* We need the original fd to change the priority. If we don't have it we can't change the priority, | |
2145 | * anymore. Note that we close any fds when entering the next event loop iteration, i.e. for inotify | |
2146 | * events we allow priority changes only until the first following iteration. */ | |
2147 | if (old_inode_data->fd < 0) | |
2148 | return -EOPNOTSUPP; | |
2149 | ||
2150 | r = event_make_inotify_data(s->event, priority, &new_inotify_data); | |
2151 | if (r < 0) | |
2152 | return r; | |
2153 | rm_inotify = r > 0; | |
2154 | ||
2155 | r = event_make_inode_data(s->event, new_inotify_data, old_inode_data->dev, old_inode_data->ino, &new_inode_data); | |
2156 | if (r < 0) | |
2157 | goto fail; | |
2158 | rm_inode = r > 0; | |
2159 | ||
2160 | if (new_inode_data->fd < 0) { | |
2161 | /* Duplicate the fd for the new inode object if we don't have any yet */ | |
2162 | new_inode_data->fd = fcntl(old_inode_data->fd, F_DUPFD_CLOEXEC, 3); | |
2163 | if (new_inode_data->fd < 0) { | |
2164 | r = -errno; | |
2165 | goto fail; | |
2166 | } | |
2167 | ||
2168 | LIST_PREPEND(to_close, s->event->inode_data_to_close, new_inode_data); | |
2169 | } | |
2170 | ||
2171 | /* Move the event source to the new inode data structure */ | |
2172 | LIST_REMOVE(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2173 | LIST_PREPEND(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2174 | s->inotify.inode_data = new_inode_data; | |
2175 | ||
2176 | /* Now create the new watch */ | |
2177 | r = inode_data_realize_watch(s->event, new_inode_data); | |
2178 | if (r < 0) { | |
2179 | /* Move it back */ | |
2180 | LIST_REMOVE(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2181 | LIST_PREPEND(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2182 | s->inotify.inode_data = old_inode_data; | |
2183 | goto fail; | |
2184 | } | |
2185 | ||
2186 | s->priority = priority; | |
2187 | ||
2188 | event_gc_inode_data(s->event, old_inode_data); | |
2189 | ||
2190 | } else if (s->type == SOURCE_SIGNAL && s->enabled != SD_EVENT_OFF) { | |
9da4cb2b LP |
2191 | struct signal_data *old, *d; |
2192 | ||
2193 | /* Move us from the signalfd belonging to the old | |
2194 | * priority to the signalfd of the new priority */ | |
2195 | ||
2196 | assert_se(old = hashmap_get(s->event->signal_data, &s->priority)); | |
2197 | ||
2198 | s->priority = priority; | |
2199 | ||
2200 | r = event_make_signal_data(s->event, s->signal.sig, &d); | |
2201 | if (r < 0) { | |
2202 | s->priority = old->priority; | |
2203 | return r; | |
2204 | } | |
2205 | ||
2206 | event_unmask_signal_data(s->event, old, s->signal.sig); | |
2207 | } else | |
2208 | s->priority = priority; | |
fd38203a LP |
2209 | |
2210 | if (s->pending) | |
c2ba3ad6 | 2211 | prioq_reshuffle(s->event->pending, s, &s->pending_index); |
fd38203a LP |
2212 | |
2213 | if (s->prepare) | |
c2ba3ad6 | 2214 | prioq_reshuffle(s->event->prepare, s, &s->prepare_index); |
fd38203a | 2215 | |
6203e07a LP |
2216 | if (s->type == SOURCE_EXIT) |
2217 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
305f78bf | 2218 | |
fd38203a | 2219 | return 0; |
97ef5391 LP |
2220 | |
2221 | fail: | |
2222 | if (rm_inode) | |
2223 | event_free_inode_data(s->event, new_inode_data); | |
2224 | ||
2225 | if (rm_inotify) | |
2226 | event_free_inotify_data(s->event, new_inotify_data); | |
2227 | ||
2228 | return r; | |
fd38203a LP |
2229 | } |
2230 | ||
f7262a9f | 2231 | _public_ int sd_event_source_get_enabled(sd_event_source *s, int *m) { |
305f78bf | 2232 | assert_return(s, -EINVAL); |
305f78bf | 2233 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2234 | |
08c1eb0e ZJS |
2235 | if (m) |
2236 | *m = s->enabled; | |
2237 | return s->enabled != SD_EVENT_OFF; | |
fd38203a LP |
2238 | } |
2239 | ||
f7262a9f | 2240 | _public_ int sd_event_source_set_enabled(sd_event_source *s, int m) { |
fd38203a LP |
2241 | int r; |
2242 | ||
305f78bf | 2243 | assert_return(s, -EINVAL); |
945c2931 | 2244 | assert_return(IN_SET(m, SD_EVENT_OFF, SD_EVENT_ON, SD_EVENT_ONESHOT), -EINVAL); |
305f78bf | 2245 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2246 | |
cc567911 LP |
2247 | /* If we are dead anyway, we are fine with turning off |
2248 | * sources, but everything else needs to fail. */ | |
2249 | if (s->event->state == SD_EVENT_FINISHED) | |
2250 | return m == SD_EVENT_OFF ? 0 : -ESTALE; | |
2251 | ||
baf76283 | 2252 | if (s->enabled == m) |
fd38203a LP |
2253 | return 0; |
2254 | ||
baf76283 | 2255 | if (m == SD_EVENT_OFF) { |
fd38203a | 2256 | |
ac989a78 LP |
2257 | /* Unset the pending flag when this event source is disabled */ |
2258 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
2259 | r = source_set_pending(s, false); | |
2260 | if (r < 0) | |
2261 | return r; | |
2262 | } | |
2263 | ||
fd38203a LP |
2264 | switch (s->type) { |
2265 | ||
2266 | case SOURCE_IO: | |
366e6411 | 2267 | source_io_unregister(s); |
baf76283 | 2268 | s->enabled = m; |
fd38203a LP |
2269 | break; |
2270 | ||
6a0f1f6d | 2271 | case SOURCE_TIME_REALTIME: |
a8548816 | 2272 | case SOURCE_TIME_BOOTTIME: |
6a0f1f6d LP |
2273 | case SOURCE_TIME_MONOTONIC: |
2274 | case SOURCE_TIME_REALTIME_ALARM: | |
2275 | case SOURCE_TIME_BOOTTIME_ALARM: { | |
2276 | struct clock_data *d; | |
fd38203a | 2277 | |
baf76283 | 2278 | s->enabled = m; |
6a0f1f6d LP |
2279 | d = event_get_clock_data(s->event, s->type); |
2280 | assert(d); | |
2281 | ||
2282 | prioq_reshuffle(d->earliest, s, &s->time.earliest_index); | |
2283 | prioq_reshuffle(d->latest, s, &s->time.latest_index); | |
212bbb17 | 2284 | d->needs_rearm = true; |
fd38203a | 2285 | break; |
6a0f1f6d | 2286 | } |
fd38203a LP |
2287 | |
2288 | case SOURCE_SIGNAL: | |
baf76283 | 2289 | s->enabled = m; |
4807d2d0 | 2290 | |
9da4cb2b | 2291 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
fd38203a LP |
2292 | break; |
2293 | ||
2294 | case SOURCE_CHILD: | |
baf76283 | 2295 | s->enabled = m; |
fd38203a | 2296 | |
baf76283 LP |
2297 | assert(s->event->n_enabled_child_sources > 0); |
2298 | s->event->n_enabled_child_sources--; | |
fd38203a | 2299 | |
f8f3f926 LP |
2300 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
2301 | source_child_pidfd_unregister(s); | |
2302 | else | |
2303 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
2304 | ||
fd38203a LP |
2305 | break; |
2306 | ||
6203e07a | 2307 | case SOURCE_EXIT: |
305f78bf | 2308 | s->enabled = m; |
6203e07a | 2309 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); |
305f78bf LP |
2310 | break; |
2311 | ||
2312 | case SOURCE_DEFER: | |
6e9feda3 | 2313 | case SOURCE_POST: |
97ef5391 | 2314 | case SOURCE_INOTIFY: |
baf76283 | 2315 | s->enabled = m; |
fd38203a | 2316 | break; |
9d3e3aa5 | 2317 | |
6a0f1f6d | 2318 | default: |
9d3e3aa5 | 2319 | assert_not_reached("Wut? I shouldn't exist."); |
fd38203a LP |
2320 | } |
2321 | ||
2322 | } else { | |
ac989a78 LP |
2323 | |
2324 | /* Unset the pending flag when this event source is enabled */ | |
2325 | if (s->enabled == SD_EVENT_OFF && !IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
2326 | r = source_set_pending(s, false); | |
2327 | if (r < 0) | |
2328 | return r; | |
2329 | } | |
2330 | ||
fd38203a LP |
2331 | switch (s->type) { |
2332 | ||
2333 | case SOURCE_IO: | |
2334 | r = source_io_register(s, m, s->io.events); | |
2335 | if (r < 0) | |
2336 | return r; | |
2337 | ||
baf76283 | 2338 | s->enabled = m; |
fd38203a LP |
2339 | break; |
2340 | ||
6a0f1f6d | 2341 | case SOURCE_TIME_REALTIME: |
a8548816 | 2342 | case SOURCE_TIME_BOOTTIME: |
6a0f1f6d LP |
2343 | case SOURCE_TIME_MONOTONIC: |
2344 | case SOURCE_TIME_REALTIME_ALARM: | |
2345 | case SOURCE_TIME_BOOTTIME_ALARM: { | |
2346 | struct clock_data *d; | |
fd38203a | 2347 | |
baf76283 | 2348 | s->enabled = m; |
6a0f1f6d LP |
2349 | d = event_get_clock_data(s->event, s->type); |
2350 | assert(d); | |
2351 | ||
2352 | prioq_reshuffle(d->earliest, s, &s->time.earliest_index); | |
2353 | prioq_reshuffle(d->latest, s, &s->time.latest_index); | |
212bbb17 | 2354 | d->needs_rearm = true; |
fd38203a | 2355 | break; |
6a0f1f6d | 2356 | } |
fd38203a LP |
2357 | |
2358 | case SOURCE_SIGNAL: | |
4807d2d0 ZJS |
2359 | |
2360 | s->enabled = m; | |
9da4cb2b LP |
2361 | |
2362 | r = event_make_signal_data(s->event, s->signal.sig, NULL); | |
2363 | if (r < 0) { | |
2364 | s->enabled = SD_EVENT_OFF; | |
2365 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); | |
2366 | return r; | |
2367 | } | |
2368 | ||
fd38203a LP |
2369 | break; |
2370 | ||
2371 | case SOURCE_CHILD: | |
4807d2d0 | 2372 | |
9da4cb2b | 2373 | if (s->enabled == SD_EVENT_OFF) |
4807d2d0 | 2374 | s->event->n_enabled_child_sources++; |
7a0d4a3d DH |
2375 | |
2376 | s->enabled = m; | |
9da4cb2b | 2377 | |
f8f3f926 LP |
2378 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
2379 | /* yes, we have pidfd */ | |
2380 | ||
2381 | r = source_child_pidfd_register(s, s->enabled); | |
2382 | if (r < 0) { | |
2383 | s->enabled = SD_EVENT_OFF; | |
2384 | s->event->n_enabled_child_sources--; | |
2385 | return r; | |
2386 | } | |
2387 | } else { | |
2388 | /* no pidfd, or something other to watch for than WEXITED */ | |
2389 | ||
2390 | r = event_make_signal_data(s->event, SIGCHLD, NULL); | |
2391 | if (r < 0) { | |
2392 | s->enabled = SD_EVENT_OFF; | |
2393 | s->event->n_enabled_child_sources--; | |
2394 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
2395 | return r; | |
2396 | } | |
9da4cb2b LP |
2397 | } |
2398 | ||
fd38203a LP |
2399 | break; |
2400 | ||
6203e07a | 2401 | case SOURCE_EXIT: |
305f78bf | 2402 | s->enabled = m; |
6203e07a | 2403 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); |
305f78bf LP |
2404 | break; |
2405 | ||
2406 | case SOURCE_DEFER: | |
6e9feda3 | 2407 | case SOURCE_POST: |
97ef5391 | 2408 | case SOURCE_INOTIFY: |
baf76283 | 2409 | s->enabled = m; |
fd38203a | 2410 | break; |
9d3e3aa5 | 2411 | |
6a0f1f6d | 2412 | default: |
9d3e3aa5 | 2413 | assert_not_reached("Wut? I shouldn't exist."); |
fd38203a LP |
2414 | } |
2415 | } | |
2416 | ||
2417 | if (s->pending) | |
2418 | prioq_reshuffle(s->event->pending, s, &s->pending_index); | |
2419 | ||
2420 | if (s->prepare) | |
2421 | prioq_reshuffle(s->event->prepare, s, &s->prepare_index); | |
2422 | ||
2423 | return 0; | |
2424 | } | |
2425 | ||
f7262a9f | 2426 | _public_ int sd_event_source_get_time(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
2427 | assert_return(s, -EINVAL); |
2428 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 2429 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf | 2430 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2431 | |
2432 | *usec = s->time.next; | |
2433 | return 0; | |
2434 | } | |
2435 | ||
f7262a9f | 2436 | _public_ int sd_event_source_set_time(sd_event_source *s, uint64_t usec) { |
6a0f1f6d | 2437 | struct clock_data *d; |
2a0dc6cd | 2438 | int r; |
6a0f1f6d | 2439 | |
305f78bf | 2440 | assert_return(s, -EINVAL); |
6a0f1f6d | 2441 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 2442 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2443 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2444 | |
2a0dc6cd LP |
2445 | r = source_set_pending(s, false); |
2446 | if (r < 0) | |
2447 | return r; | |
2576a19e | 2448 | |
2a0dc6cd | 2449 | s->time.next = usec; |
fd38203a | 2450 | |
6a0f1f6d LP |
2451 | d = event_get_clock_data(s->event, s->type); |
2452 | assert(d); | |
2453 | ||
2454 | prioq_reshuffle(d->earliest, s, &s->time.earliest_index); | |
2455 | prioq_reshuffle(d->latest, s, &s->time.latest_index); | |
212bbb17 | 2456 | d->needs_rearm = true; |
fd38203a LP |
2457 | |
2458 | return 0; | |
2459 | } | |
2460 | ||
d6a83dc4 LP |
2461 | _public_ int sd_event_source_set_time_relative(sd_event_source *s, uint64_t usec) { |
2462 | usec_t t; | |
2463 | int r; | |
2464 | ||
2465 | assert_return(s, -EINVAL); | |
2466 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2467 | ||
2468 | r = sd_event_now(s->event, event_source_type_to_clock(s->type), &t); | |
2469 | if (r < 0) | |
2470 | return r; | |
2471 | ||
2472 | if (usec >= USEC_INFINITY - t) | |
2473 | return -EOVERFLOW; | |
2474 | ||
2475 | return sd_event_source_set_time(s, t + usec); | |
2476 | } | |
2477 | ||
f7262a9f | 2478 | _public_ int sd_event_source_get_time_accuracy(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
2479 | assert_return(s, -EINVAL); |
2480 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 2481 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf LP |
2482 | assert_return(!event_pid_changed(s->event), -ECHILD); |
2483 | ||
2484 | *usec = s->time.accuracy; | |
2485 | return 0; | |
2486 | } | |
2487 | ||
f7262a9f | 2488 | _public_ int sd_event_source_set_time_accuracy(sd_event_source *s, uint64_t usec) { |
6a0f1f6d | 2489 | struct clock_data *d; |
2a0dc6cd | 2490 | int r; |
6a0f1f6d | 2491 | |
305f78bf LP |
2492 | assert_return(s, -EINVAL); |
2493 | assert_return(usec != (uint64_t) -1, -EINVAL); | |
6a0f1f6d | 2494 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 2495 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2496 | assert_return(!event_pid_changed(s->event), -ECHILD); |
eaa3cbef | 2497 | |
2a0dc6cd LP |
2498 | r = source_set_pending(s, false); |
2499 | if (r < 0) | |
2500 | return r; | |
2501 | ||
eaa3cbef LP |
2502 | if (usec == 0) |
2503 | usec = DEFAULT_ACCURACY_USEC; | |
2504 | ||
eaa3cbef LP |
2505 | s->time.accuracy = usec; |
2506 | ||
6a0f1f6d LP |
2507 | d = event_get_clock_data(s->event, s->type); |
2508 | assert(d); | |
2509 | ||
2510 | prioq_reshuffle(d->latest, s, &s->time.latest_index); | |
212bbb17 | 2511 | d->needs_rearm = true; |
6a0f1f6d LP |
2512 | |
2513 | return 0; | |
2514 | } | |
2515 | ||
2516 | _public_ int sd_event_source_get_time_clock(sd_event_source *s, clockid_t *clock) { | |
2517 | assert_return(s, -EINVAL); | |
2518 | assert_return(clock, -EINVAL); | |
2519 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2520 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
eaa3cbef | 2521 | |
6a0f1f6d | 2522 | *clock = event_source_type_to_clock(s->type); |
eaa3cbef LP |
2523 | return 0; |
2524 | } | |
2525 | ||
f7262a9f | 2526 | _public_ int sd_event_source_get_child_pid(sd_event_source *s, pid_t *pid) { |
4bee8012 LP |
2527 | assert_return(s, -EINVAL); |
2528 | assert_return(pid, -EINVAL); | |
2529 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2530 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2531 | ||
2532 | *pid = s->child.pid; | |
2533 | return 0; | |
2534 | } | |
2535 | ||
f8f3f926 LP |
2536 | _public_ int sd_event_source_get_child_pidfd(sd_event_source *s) { |
2537 | assert_return(s, -EINVAL); | |
2538 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2539 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2540 | ||
2541 | if (s->child.pidfd < 0) | |
2542 | return -EOPNOTSUPP; | |
2543 | ||
2544 | return s->child.pidfd; | |
2545 | } | |
2546 | ||
2547 | _public_ int sd_event_source_send_child_signal(sd_event_source *s, int sig, const siginfo_t *si, unsigned flags) { | |
2548 | assert_return(s, -EINVAL); | |
2549 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2550 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2551 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
2552 | ||
2553 | /* If we already have seen indication the process exited refuse sending a signal early. This way we | |
2554 | * can be sure we don't accidentally kill the wrong process on PID reuse when pidfds are not | |
2555 | * available. */ | |
2556 | if (s->child.exited) | |
2557 | return -ESRCH; | |
2558 | ||
2559 | if (s->child.pidfd >= 0) { | |
2560 | siginfo_t copy; | |
2561 | ||
2562 | /* pidfd_send_signal() changes the siginfo_t argument. This is weird, let's hence copy the | |
2563 | * structure here */ | |
2564 | if (si) | |
2565 | copy = *si; | |
2566 | ||
2567 | if (pidfd_send_signal(s->child.pidfd, sig, si ? © : NULL, 0) < 0) { | |
2568 | /* Let's propagate the error only if the system call is not implemented or prohibited */ | |
2569 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
2570 | return -errno; | |
2571 | } else | |
2572 | return 0; | |
2573 | } | |
2574 | ||
2575 | /* Flags are only supported for pidfd_send_signal(), not for rt_sigqueueinfo(), hence let's refuse | |
2576 | * this here. */ | |
2577 | if (flags != 0) | |
2578 | return -EOPNOTSUPP; | |
2579 | ||
2580 | if (si) { | |
2581 | /* We use rt_sigqueueinfo() only if siginfo_t is specified. */ | |
2582 | siginfo_t copy = *si; | |
2583 | ||
2584 | if (rt_sigqueueinfo(s->child.pid, sig, ©) < 0) | |
2585 | return -errno; | |
2586 | } else if (kill(s->child.pid, sig) < 0) | |
2587 | return -errno; | |
2588 | ||
2589 | return 0; | |
2590 | } | |
2591 | ||
2592 | _public_ int sd_event_source_get_child_pidfd_own(sd_event_source *s) { | |
2593 | assert_return(s, -EINVAL); | |
2594 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2595 | ||
2596 | if (s->child.pidfd < 0) | |
2597 | return -EOPNOTSUPP; | |
2598 | ||
2599 | return s->child.pidfd_owned; | |
2600 | } | |
2601 | ||
2602 | _public_ int sd_event_source_set_child_pidfd_own(sd_event_source *s, int own) { | |
2603 | assert_return(s, -EINVAL); | |
2604 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2605 | ||
2606 | if (s->child.pidfd < 0) | |
2607 | return -EOPNOTSUPP; | |
2608 | ||
2609 | s->child.pidfd_owned = own; | |
2610 | return 0; | |
2611 | } | |
2612 | ||
2613 | _public_ int sd_event_source_get_child_process_own(sd_event_source *s) { | |
2614 | assert_return(s, -EINVAL); | |
2615 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2616 | ||
2617 | return s->child.process_owned; | |
2618 | } | |
2619 | ||
2620 | _public_ int sd_event_source_set_child_process_own(sd_event_source *s, int own) { | |
2621 | assert_return(s, -EINVAL); | |
2622 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2623 | ||
2624 | s->child.process_owned = own; | |
2625 | return 0; | |
2626 | } | |
2627 | ||
97ef5391 LP |
2628 | _public_ int sd_event_source_get_inotify_mask(sd_event_source *s, uint32_t *mask) { |
2629 | assert_return(s, -EINVAL); | |
2630 | assert_return(mask, -EINVAL); | |
2631 | assert_return(s->type == SOURCE_INOTIFY, -EDOM); | |
2632 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2633 | ||
2634 | *mask = s->inotify.mask; | |
2635 | return 0; | |
2636 | } | |
2637 | ||
718db961 | 2638 | _public_ int sd_event_source_set_prepare(sd_event_source *s, sd_event_handler_t callback) { |
fd38203a LP |
2639 | int r; |
2640 | ||
da7e457c | 2641 | assert_return(s, -EINVAL); |
6203e07a | 2642 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c LP |
2643 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2644 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2645 | |
2646 | if (s->prepare == callback) | |
2647 | return 0; | |
2648 | ||
2649 | if (callback && s->prepare) { | |
2650 | s->prepare = callback; | |
2651 | return 0; | |
2652 | } | |
2653 | ||
2654 | r = prioq_ensure_allocated(&s->event->prepare, prepare_prioq_compare); | |
2655 | if (r < 0) | |
2656 | return r; | |
2657 | ||
2658 | s->prepare = callback; | |
2659 | ||
2660 | if (callback) { | |
2661 | r = prioq_put(s->event->prepare, s, &s->prepare_index); | |
2662 | if (r < 0) | |
2663 | return r; | |
2664 | } else | |
2665 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
2666 | ||
2667 | return 0; | |
2668 | } | |
2669 | ||
f7262a9f | 2670 | _public_ void* sd_event_source_get_userdata(sd_event_source *s) { |
da7e457c | 2671 | assert_return(s, NULL); |
fd38203a LP |
2672 | |
2673 | return s->userdata; | |
2674 | } | |
2675 | ||
8f726607 LP |
2676 | _public_ void *sd_event_source_set_userdata(sd_event_source *s, void *userdata) { |
2677 | void *ret; | |
2678 | ||
2679 | assert_return(s, NULL); | |
2680 | ||
2681 | ret = s->userdata; | |
2682 | s->userdata = userdata; | |
2683 | ||
2684 | return ret; | |
2685 | } | |
2686 | ||
c2ba3ad6 LP |
2687 | static usec_t sleep_between(sd_event *e, usec_t a, usec_t b) { |
2688 | usec_t c; | |
2689 | assert(e); | |
2690 | assert(a <= b); | |
2691 | ||
2692 | if (a <= 0) | |
2693 | return 0; | |
393003e1 LP |
2694 | if (a >= USEC_INFINITY) |
2695 | return USEC_INFINITY; | |
c2ba3ad6 LP |
2696 | |
2697 | if (b <= a + 1) | |
2698 | return a; | |
2699 | ||
52444dc4 LP |
2700 | initialize_perturb(e); |
2701 | ||
c2ba3ad6 LP |
2702 | /* |
2703 | Find a good time to wake up again between times a and b. We | |
2704 | have two goals here: | |
2705 | ||
2706 | a) We want to wake up as seldom as possible, hence prefer | |
2707 | later times over earlier times. | |
2708 | ||
2709 | b) But if we have to wake up, then let's make sure to | |
2710 | dispatch as much as possible on the entire system. | |
2711 | ||
2712 | We implement this by waking up everywhere at the same time | |
850516e0 | 2713 | within any given minute if we can, synchronised via the |
c2ba3ad6 | 2714 | perturbation value determined from the boot ID. If we can't, |
ba276c81 LP |
2715 | then we try to find the same spot in every 10s, then 1s and |
2716 | then 250ms step. Otherwise, we pick the last possible time | |
2717 | to wake up. | |
c2ba3ad6 LP |
2718 | */ |
2719 | ||
850516e0 LP |
2720 | c = (b / USEC_PER_MINUTE) * USEC_PER_MINUTE + e->perturb; |
2721 | if (c >= b) { | |
2722 | if (_unlikely_(c < USEC_PER_MINUTE)) | |
2723 | return b; | |
2724 | ||
2725 | c -= USEC_PER_MINUTE; | |
2726 | } | |
2727 | ||
ba276c81 LP |
2728 | if (c >= a) |
2729 | return c; | |
2730 | ||
2731 | c = (b / (USEC_PER_SEC*10)) * (USEC_PER_SEC*10) + (e->perturb % (USEC_PER_SEC*10)); | |
2732 | if (c >= b) { | |
2733 | if (_unlikely_(c < USEC_PER_SEC*10)) | |
2734 | return b; | |
2735 | ||
2736 | c -= USEC_PER_SEC*10; | |
2737 | } | |
2738 | ||
850516e0 LP |
2739 | if (c >= a) |
2740 | return c; | |
2741 | ||
2742 | c = (b / USEC_PER_SEC) * USEC_PER_SEC + (e->perturb % USEC_PER_SEC); | |
c2ba3ad6 LP |
2743 | if (c >= b) { |
2744 | if (_unlikely_(c < USEC_PER_SEC)) | |
2745 | return b; | |
2746 | ||
2747 | c -= USEC_PER_SEC; | |
2748 | } | |
2749 | ||
2750 | if (c >= a) | |
2751 | return c; | |
2752 | ||
2753 | c = (b / (USEC_PER_MSEC*250)) * (USEC_PER_MSEC*250) + (e->perturb % (USEC_PER_MSEC*250)); | |
2754 | if (c >= b) { | |
2755 | if (_unlikely_(c < USEC_PER_MSEC*250)) | |
2756 | return b; | |
2757 | ||
2758 | c -= USEC_PER_MSEC*250; | |
2759 | } | |
2760 | ||
2761 | if (c >= a) | |
2762 | return c; | |
2763 | ||
2764 | return b; | |
2765 | } | |
2766 | ||
fd38203a LP |
2767 | static int event_arm_timer( |
2768 | sd_event *e, | |
6a0f1f6d | 2769 | struct clock_data *d) { |
fd38203a LP |
2770 | |
2771 | struct itimerspec its = {}; | |
c2ba3ad6 LP |
2772 | sd_event_source *a, *b; |
2773 | usec_t t; | |
fd38203a | 2774 | |
cde93897 | 2775 | assert(e); |
6a0f1f6d | 2776 | assert(d); |
fd38203a | 2777 | |
d06441da | 2778 | if (!d->needs_rearm) |
212bbb17 TG |
2779 | return 0; |
2780 | else | |
2781 | d->needs_rearm = false; | |
2782 | ||
6a0f1f6d | 2783 | a = prioq_peek(d->earliest); |
393003e1 | 2784 | if (!a || a->enabled == SD_EVENT_OFF || a->time.next == USEC_INFINITY) { |
72aedc1e | 2785 | |
6a0f1f6d | 2786 | if (d->fd < 0) |
c57b5ca3 LP |
2787 | return 0; |
2788 | ||
3a43da28 | 2789 | if (d->next == USEC_INFINITY) |
72aedc1e LP |
2790 | return 0; |
2791 | ||
2792 | /* disarm */ | |
15c689d7 LP |
2793 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
2794 | return -errno; | |
72aedc1e | 2795 | |
3a43da28 | 2796 | d->next = USEC_INFINITY; |
fd38203a | 2797 | return 0; |
72aedc1e | 2798 | } |
fd38203a | 2799 | |
6a0f1f6d | 2800 | b = prioq_peek(d->latest); |
baf76283 | 2801 | assert_se(b && b->enabled != SD_EVENT_OFF); |
c2ba3ad6 | 2802 | |
1bce0ffa | 2803 | t = sleep_between(e, a->time.next, time_event_source_latest(b)); |
6a0f1f6d | 2804 | if (d->next == t) |
fd38203a LP |
2805 | return 0; |
2806 | ||
6a0f1f6d | 2807 | assert_se(d->fd >= 0); |
fd38203a | 2808 | |
c2ba3ad6 | 2809 | if (t == 0) { |
fd38203a LP |
2810 | /* We don' want to disarm here, just mean some time looooong ago. */ |
2811 | its.it_value.tv_sec = 0; | |
2812 | its.it_value.tv_nsec = 1; | |
2813 | } else | |
c2ba3ad6 | 2814 | timespec_store(&its.it_value, t); |
fd38203a | 2815 | |
15c689d7 | 2816 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 | 2817 | return -errno; |
fd38203a | 2818 | |
6a0f1f6d | 2819 | d->next = t; |
fd38203a LP |
2820 | return 0; |
2821 | } | |
2822 | ||
9a800b56 | 2823 | static int process_io(sd_event *e, sd_event_source *s, uint32_t revents) { |
fd38203a LP |
2824 | assert(e); |
2825 | assert(s); | |
2826 | assert(s->type == SOURCE_IO); | |
2827 | ||
9a800b56 LP |
2828 | /* If the event source was already pending, we just OR in the |
2829 | * new revents, otherwise we reset the value. The ORing is | |
2830 | * necessary to handle EPOLLONESHOT events properly where | |
2831 | * readability might happen independently of writability, and | |
2832 | * we need to keep track of both */ | |
2833 | ||
2834 | if (s->pending) | |
2835 | s->io.revents |= revents; | |
2836 | else | |
2837 | s->io.revents = revents; | |
fd38203a | 2838 | |
fd38203a LP |
2839 | return source_set_pending(s, true); |
2840 | } | |
2841 | ||
72aedc1e | 2842 | static int flush_timer(sd_event *e, int fd, uint32_t events, usec_t *next) { |
fd38203a LP |
2843 | uint64_t x; |
2844 | ssize_t ss; | |
2845 | ||
2846 | assert(e); | |
da7e457c | 2847 | assert(fd >= 0); |
72aedc1e | 2848 | |
305f78bf | 2849 | assert_return(events == EPOLLIN, -EIO); |
fd38203a LP |
2850 | |
2851 | ss = read(fd, &x, sizeof(x)); | |
2852 | if (ss < 0) { | |
945c2931 | 2853 | if (IN_SET(errno, EAGAIN, EINTR)) |
fd38203a LP |
2854 | return 0; |
2855 | ||
2856 | return -errno; | |
2857 | } | |
2858 | ||
8d35dae7 | 2859 | if (_unlikely_(ss != sizeof(x))) |
fd38203a LP |
2860 | return -EIO; |
2861 | ||
cde93897 | 2862 | if (next) |
3a43da28 | 2863 | *next = USEC_INFINITY; |
72aedc1e | 2864 | |
fd38203a LP |
2865 | return 0; |
2866 | } | |
2867 | ||
305f78bf LP |
2868 | static int process_timer( |
2869 | sd_event *e, | |
2870 | usec_t n, | |
6a0f1f6d | 2871 | struct clock_data *d) { |
305f78bf | 2872 | |
fd38203a LP |
2873 | sd_event_source *s; |
2874 | int r; | |
2875 | ||
2876 | assert(e); | |
6a0f1f6d | 2877 | assert(d); |
fd38203a LP |
2878 | |
2879 | for (;;) { | |
6a0f1f6d | 2880 | s = prioq_peek(d->earliest); |
fd38203a LP |
2881 | if (!s || |
2882 | s->time.next > n || | |
baf76283 | 2883 | s->enabled == SD_EVENT_OFF || |
fd38203a LP |
2884 | s->pending) |
2885 | break; | |
2886 | ||
2887 | r = source_set_pending(s, true); | |
2888 | if (r < 0) | |
2889 | return r; | |
2890 | ||
6a0f1f6d LP |
2891 | prioq_reshuffle(d->earliest, s, &s->time.earliest_index); |
2892 | prioq_reshuffle(d->latest, s, &s->time.latest_index); | |
e07bbb7c | 2893 | d->needs_rearm = true; |
fd38203a LP |
2894 | } |
2895 | ||
2896 | return 0; | |
2897 | } | |
2898 | ||
2899 | static int process_child(sd_event *e) { | |
2900 | sd_event_source *s; | |
fd38203a LP |
2901 | int r; |
2902 | ||
2903 | assert(e); | |
2904 | ||
c2ba3ad6 LP |
2905 | e->need_process_child = false; |
2906 | ||
fd38203a LP |
2907 | /* |
2908 | So, this is ugly. We iteratively invoke waitid() with P_PID | |
2909 | + WNOHANG for each PID we wait for, instead of using | |
2910 | P_ALL. This is because we only want to get child | |
2911 | information of very specific child processes, and not all | |
2912 | of them. We might not have processed the SIGCHLD even of a | |
2913 | previous invocation and we don't want to maintain a | |
2914 | unbounded *per-child* event queue, hence we really don't | |
2915 | want anything flushed out of the kernel's queue that we | |
2916 | don't care about. Since this is O(n) this means that if you | |
2917 | have a lot of processes you probably want to handle SIGCHLD | |
2918 | yourself. | |
08cd1552 LP |
2919 | |
2920 | We do not reap the children here (by using WNOWAIT), this | |
2921 | is only done after the event source is dispatched so that | |
2922 | the callback still sees the process as a zombie. | |
fd38203a LP |
2923 | */ |
2924 | ||
90e74a66 | 2925 | HASHMAP_FOREACH(s, e->child_sources) { |
fd38203a LP |
2926 | assert(s->type == SOURCE_CHILD); |
2927 | ||
2928 | if (s->pending) | |
2929 | continue; | |
2930 | ||
baf76283 | 2931 | if (s->enabled == SD_EVENT_OFF) |
fd38203a LP |
2932 | continue; |
2933 | ||
f8f3f926 LP |
2934 | if (s->child.exited) |
2935 | continue; | |
2936 | ||
2937 | if (EVENT_SOURCE_WATCH_PIDFD(s)) /* There's a usable pidfd known for this event source? then don't waitid() for it here */ | |
2938 | continue; | |
2939 | ||
fd38203a | 2940 | zero(s->child.siginfo); |
15c689d7 LP |
2941 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, |
2942 | WNOHANG | (s->child.options & WEXITED ? WNOWAIT : 0) | s->child.options) < 0) | |
fd38203a LP |
2943 | return -errno; |
2944 | ||
2945 | if (s->child.siginfo.si_pid != 0) { | |
945c2931 | 2946 | bool zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 2947 | |
f8f3f926 LP |
2948 | if (zombie) |
2949 | s->child.exited = true; | |
2950 | ||
08cd1552 LP |
2951 | if (!zombie && (s->child.options & WEXITED)) { |
2952 | /* If the child isn't dead then let's | |
2953 | * immediately remove the state change | |
2954 | * from the queue, since there's no | |
2955 | * benefit in leaving it queued */ | |
2956 | ||
2957 | assert(s->child.options & (WSTOPPED|WCONTINUED)); | |
a5d27871 | 2958 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|(s->child.options & (WSTOPPED|WCONTINUED))); |
08cd1552 LP |
2959 | } |
2960 | ||
fd38203a LP |
2961 | r = source_set_pending(s, true); |
2962 | if (r < 0) | |
2963 | return r; | |
2964 | } | |
2965 | } | |
2966 | ||
fd38203a LP |
2967 | return 0; |
2968 | } | |
2969 | ||
f8f3f926 LP |
2970 | static int process_pidfd(sd_event *e, sd_event_source *s, uint32_t revents) { |
2971 | assert(e); | |
2972 | assert(s); | |
2973 | assert(s->type == SOURCE_CHILD); | |
2974 | ||
2975 | if (s->pending) | |
2976 | return 0; | |
2977 | ||
2978 | if (s->enabled == SD_EVENT_OFF) | |
2979 | return 0; | |
2980 | ||
2981 | if (!EVENT_SOURCE_WATCH_PIDFD(s)) | |
2982 | return 0; | |
2983 | ||
2984 | zero(s->child.siginfo); | |
2985 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG | WNOWAIT | s->child.options) < 0) | |
2986 | return -errno; | |
2987 | ||
2988 | if (s->child.siginfo.si_pid == 0) | |
2989 | return 0; | |
2990 | ||
2991 | if (IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) | |
2992 | s->child.exited = true; | |
2993 | ||
2994 | return source_set_pending(s, true); | |
2995 | } | |
2996 | ||
9da4cb2b | 2997 | static int process_signal(sd_event *e, struct signal_data *d, uint32_t events) { |
fd38203a | 2998 | bool read_one = false; |
fd38203a LP |
2999 | int r; |
3000 | ||
da7e457c | 3001 | assert(e); |
97ef5391 | 3002 | assert(d); |
305f78bf | 3003 | assert_return(events == EPOLLIN, -EIO); |
fd38203a | 3004 | |
9da4cb2b LP |
3005 | /* If there's a signal queued on this priority and SIGCHLD is |
3006 | on this priority too, then make sure to recheck the | |
3007 | children we watch. This is because we only ever dequeue | |
3008 | the first signal per priority, and if we dequeue one, and | |
3009 | SIGCHLD might be enqueued later we wouldn't know, but we | |
3010 | might have higher priority children we care about hence we | |
3011 | need to check that explicitly. */ | |
3012 | ||
3013 | if (sigismember(&d->sigset, SIGCHLD)) | |
3014 | e->need_process_child = true; | |
3015 | ||
3016 | /* If there's already an event source pending for this | |
3017 | * priority we don't read another */ | |
3018 | if (d->current) | |
3019 | return 0; | |
3020 | ||
fd38203a | 3021 | for (;;) { |
0eb2e0e3 | 3022 | struct signalfd_siginfo si; |
7057bd99 | 3023 | ssize_t n; |
92daebc0 | 3024 | sd_event_source *s = NULL; |
fd38203a | 3025 | |
9da4cb2b | 3026 | n = read(d->fd, &si, sizeof(si)); |
7057bd99 | 3027 | if (n < 0) { |
945c2931 | 3028 | if (IN_SET(errno, EAGAIN, EINTR)) |
fd38203a LP |
3029 | return read_one; |
3030 | ||
3031 | return -errno; | |
3032 | } | |
3033 | ||
7057bd99 | 3034 | if (_unlikely_(n != sizeof(si))) |
fd38203a LP |
3035 | return -EIO; |
3036 | ||
6eb7c172 | 3037 | assert(SIGNAL_VALID(si.ssi_signo)); |
7057bd99 | 3038 | |
fd38203a LP |
3039 | read_one = true; |
3040 | ||
92daebc0 LP |
3041 | if (e->signal_sources) |
3042 | s = e->signal_sources[si.ssi_signo]; | |
92daebc0 LP |
3043 | if (!s) |
3044 | continue; | |
9da4cb2b LP |
3045 | if (s->pending) |
3046 | continue; | |
fd38203a LP |
3047 | |
3048 | s->signal.siginfo = si; | |
9da4cb2b LP |
3049 | d->current = s; |
3050 | ||
fd38203a LP |
3051 | r = source_set_pending(s, true); |
3052 | if (r < 0) | |
3053 | return r; | |
9da4cb2b LP |
3054 | |
3055 | return 1; | |
fd38203a | 3056 | } |
fd38203a LP |
3057 | } |
3058 | ||
97ef5391 LP |
3059 | static int event_inotify_data_read(sd_event *e, struct inotify_data *d, uint32_t revents) { |
3060 | ssize_t n; | |
3061 | ||
3062 | assert(e); | |
3063 | assert(d); | |
3064 | ||
3065 | assert_return(revents == EPOLLIN, -EIO); | |
3066 | ||
3067 | /* If there's already an event source pending for this priority, don't read another */ | |
3068 | if (d->n_pending > 0) | |
3069 | return 0; | |
3070 | ||
3071 | /* Is the read buffer non-empty? If so, let's not read more */ | |
3072 | if (d->buffer_filled > 0) | |
3073 | return 0; | |
3074 | ||
3075 | n = read(d->fd, &d->buffer, sizeof(d->buffer)); | |
3076 | if (n < 0) { | |
3077 | if (IN_SET(errno, EAGAIN, EINTR)) | |
3078 | return 0; | |
3079 | ||
3080 | return -errno; | |
3081 | } | |
3082 | ||
3083 | assert(n > 0); | |
3084 | d->buffer_filled = (size_t) n; | |
3085 | LIST_PREPEND(buffered, e->inotify_data_buffered, d); | |
3086 | ||
3087 | return 1; | |
3088 | } | |
3089 | ||
3090 | static void event_inotify_data_drop(sd_event *e, struct inotify_data *d, size_t sz) { | |
3091 | assert(e); | |
3092 | assert(d); | |
3093 | assert(sz <= d->buffer_filled); | |
3094 | ||
3095 | if (sz == 0) | |
3096 | return; | |
3097 | ||
3098 | /* Move the rest to the buffer to the front, in order to get things properly aligned again */ | |
3099 | memmove(d->buffer.raw, d->buffer.raw + sz, d->buffer_filled - sz); | |
3100 | d->buffer_filled -= sz; | |
3101 | ||
3102 | if (d->buffer_filled == 0) | |
3103 | LIST_REMOVE(buffered, e->inotify_data_buffered, d); | |
3104 | } | |
3105 | ||
3106 | static int event_inotify_data_process(sd_event *e, struct inotify_data *d) { | |
3107 | int r; | |
3108 | ||
3109 | assert(e); | |
3110 | assert(d); | |
3111 | ||
3112 | /* If there's already an event source pending for this priority, don't read another */ | |
3113 | if (d->n_pending > 0) | |
3114 | return 0; | |
3115 | ||
3116 | while (d->buffer_filled > 0) { | |
3117 | size_t sz; | |
3118 | ||
3119 | /* Let's validate that the event structures are complete */ | |
3120 | if (d->buffer_filled < offsetof(struct inotify_event, name)) | |
3121 | return -EIO; | |
3122 | ||
3123 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3124 | if (d->buffer_filled < sz) | |
3125 | return -EIO; | |
3126 | ||
3127 | if (d->buffer.ev.mask & IN_Q_OVERFLOW) { | |
3128 | struct inode_data *inode_data; | |
97ef5391 LP |
3129 | |
3130 | /* The queue overran, let's pass this event to all event sources connected to this inotify | |
3131 | * object */ | |
3132 | ||
90e74a66 | 3133 | HASHMAP_FOREACH(inode_data, d->inodes) { |
97ef5391 LP |
3134 | sd_event_source *s; |
3135 | ||
3136 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
3137 | ||
3138 | if (s->enabled == SD_EVENT_OFF) | |
3139 | continue; | |
3140 | ||
3141 | r = source_set_pending(s, true); | |
3142 | if (r < 0) | |
3143 | return r; | |
3144 | } | |
3145 | } | |
3146 | } else { | |
3147 | struct inode_data *inode_data; | |
3148 | sd_event_source *s; | |
3149 | ||
3150 | /* Find the inode object for this watch descriptor. If IN_IGNORED is set we also remove it from | |
3151 | * our watch descriptor table. */ | |
3152 | if (d->buffer.ev.mask & IN_IGNORED) { | |
3153 | ||
3154 | inode_data = hashmap_remove(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3155 | if (!inode_data) { | |
3156 | event_inotify_data_drop(e, d, sz); | |
3157 | continue; | |
3158 | } | |
3159 | ||
3160 | /* The watch descriptor was removed by the kernel, let's drop it here too */ | |
3161 | inode_data->wd = -1; | |
3162 | } else { | |
3163 | inode_data = hashmap_get(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3164 | if (!inode_data) { | |
3165 | event_inotify_data_drop(e, d, sz); | |
3166 | continue; | |
3167 | } | |
3168 | } | |
3169 | ||
3170 | /* Trigger all event sources that are interested in these events. Also trigger all event | |
3171 | * sources if IN_IGNORED or IN_UNMOUNT is set. */ | |
3172 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
3173 | ||
3174 | if (s->enabled == SD_EVENT_OFF) | |
3175 | continue; | |
3176 | ||
3177 | if ((d->buffer.ev.mask & (IN_IGNORED|IN_UNMOUNT)) == 0 && | |
3178 | (s->inotify.mask & d->buffer.ev.mask & IN_ALL_EVENTS) == 0) | |
3179 | continue; | |
3180 | ||
3181 | r = source_set_pending(s, true); | |
3182 | if (r < 0) | |
3183 | return r; | |
3184 | } | |
3185 | } | |
3186 | ||
3187 | /* Something pending now? If so, let's finish, otherwise let's read more. */ | |
3188 | if (d->n_pending > 0) | |
3189 | return 1; | |
3190 | } | |
3191 | ||
3192 | return 0; | |
3193 | } | |
3194 | ||
3195 | static int process_inotify(sd_event *e) { | |
3196 | struct inotify_data *d; | |
3197 | int r, done = 0; | |
3198 | ||
3199 | assert(e); | |
3200 | ||
3201 | LIST_FOREACH(buffered, d, e->inotify_data_buffered) { | |
3202 | r = event_inotify_data_process(e, d); | |
3203 | if (r < 0) | |
3204 | return r; | |
3205 | if (r > 0) | |
3206 | done ++; | |
3207 | } | |
3208 | ||
3209 | return done; | |
3210 | } | |
3211 | ||
fd38203a | 3212 | static int source_dispatch(sd_event_source *s) { |
b778cba4 | 3213 | _cleanup_(sd_event_unrefp) sd_event *saved_event = NULL; |
8f5c235d | 3214 | EventSourceType saved_type; |
fe8245eb | 3215 | int r = 0; |
fd38203a LP |
3216 | |
3217 | assert(s); | |
6203e07a | 3218 | assert(s->pending || s->type == SOURCE_EXIT); |
fd38203a | 3219 | |
b778cba4 LP |
3220 | /* Save the event source type, here, so that we still know it after the event callback which might |
3221 | * invalidate the event. */ | |
8f5c235d LP |
3222 | saved_type = s->type; |
3223 | ||
b778cba4 LP |
3224 | /* Similar, store a reference to the event loop object, so that we can still access it after the |
3225 | * callback might have invalidated/disconnected the event source. */ | |
3226 | saved_event = sd_event_ref(s->event); | |
3227 | ||
945c2931 | 3228 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { |
da7e457c LP |
3229 | r = source_set_pending(s, false); |
3230 | if (r < 0) | |
3231 | return r; | |
3232 | } | |
fd38203a | 3233 | |
6e9feda3 LP |
3234 | if (s->type != SOURCE_POST) { |
3235 | sd_event_source *z; | |
6e9feda3 LP |
3236 | |
3237 | /* If we execute a non-post source, let's mark all | |
3238 | * post sources as pending */ | |
3239 | ||
90e74a66 | 3240 | SET_FOREACH(z, s->event->post_sources) { |
6e9feda3 LP |
3241 | if (z->enabled == SD_EVENT_OFF) |
3242 | continue; | |
3243 | ||
3244 | r = source_set_pending(z, true); | |
3245 | if (r < 0) | |
3246 | return r; | |
3247 | } | |
3248 | } | |
3249 | ||
baf76283 LP |
3250 | if (s->enabled == SD_EVENT_ONESHOT) { |
3251 | r = sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
3252 | if (r < 0) |
3253 | return r; | |
3254 | } | |
3255 | ||
12179984 | 3256 | s->dispatching = true; |
b7484e2a | 3257 | |
fd38203a LP |
3258 | switch (s->type) { |
3259 | ||
3260 | case SOURCE_IO: | |
3261 | r = s->io.callback(s, s->io.fd, s->io.revents, s->userdata); | |
3262 | break; | |
3263 | ||
6a0f1f6d | 3264 | case SOURCE_TIME_REALTIME: |
a8548816 | 3265 | case SOURCE_TIME_BOOTTIME: |
6a0f1f6d LP |
3266 | case SOURCE_TIME_MONOTONIC: |
3267 | case SOURCE_TIME_REALTIME_ALARM: | |
3268 | case SOURCE_TIME_BOOTTIME_ALARM: | |
fd38203a LP |
3269 | r = s->time.callback(s, s->time.next, s->userdata); |
3270 | break; | |
3271 | ||
3272 | case SOURCE_SIGNAL: | |
3273 | r = s->signal.callback(s, &s->signal.siginfo, s->userdata); | |
3274 | break; | |
3275 | ||
08cd1552 LP |
3276 | case SOURCE_CHILD: { |
3277 | bool zombie; | |
3278 | ||
945c2931 | 3279 | zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 3280 | |
fd38203a | 3281 | r = s->child.callback(s, &s->child.siginfo, s->userdata); |
08cd1552 LP |
3282 | |
3283 | /* Now, reap the PID for good. */ | |
f8f3f926 | 3284 | if (zombie) { |
cc59d290 | 3285 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|WEXITED); |
f8f3f926 LP |
3286 | s->child.waited = true; |
3287 | } | |
08cd1552 | 3288 | |
fd38203a | 3289 | break; |
08cd1552 | 3290 | } |
fd38203a LP |
3291 | |
3292 | case SOURCE_DEFER: | |
3293 | r = s->defer.callback(s, s->userdata); | |
3294 | break; | |
da7e457c | 3295 | |
6e9feda3 LP |
3296 | case SOURCE_POST: |
3297 | r = s->post.callback(s, s->userdata); | |
3298 | break; | |
3299 | ||
6203e07a LP |
3300 | case SOURCE_EXIT: |
3301 | r = s->exit.callback(s, s->userdata); | |
da7e457c | 3302 | break; |
9d3e3aa5 | 3303 | |
97ef5391 LP |
3304 | case SOURCE_INOTIFY: { |
3305 | struct sd_event *e = s->event; | |
3306 | struct inotify_data *d; | |
3307 | size_t sz; | |
3308 | ||
3309 | assert(s->inotify.inode_data); | |
3310 | assert_se(d = s->inotify.inode_data->inotify_data); | |
3311 | ||
3312 | assert(d->buffer_filled >= offsetof(struct inotify_event, name)); | |
3313 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3314 | assert(d->buffer_filled >= sz); | |
3315 | ||
3316 | r = s->inotify.callback(s, &d->buffer.ev, s->userdata); | |
3317 | ||
3318 | /* When no event is pending anymore on this inotify object, then let's drop the event from the | |
3319 | * buffer. */ | |
3320 | if (d->n_pending == 0) | |
3321 | event_inotify_data_drop(e, d, sz); | |
3322 | ||
3323 | break; | |
3324 | } | |
3325 | ||
9d3e3aa5 | 3326 | case SOURCE_WATCHDOG: |
a71fe8b8 | 3327 | case _SOURCE_EVENT_SOURCE_TYPE_MAX: |
9f2a50a3 | 3328 | case _SOURCE_EVENT_SOURCE_TYPE_INVALID: |
9d3e3aa5 | 3329 | assert_not_reached("Wut? I shouldn't exist."); |
fd38203a LP |
3330 | } |
3331 | ||
12179984 LP |
3332 | s->dispatching = false; |
3333 | ||
b778cba4 LP |
3334 | if (r < 0) { |
3335 | log_debug_errno(r, "Event source %s (type %s) returned error, %s: %m", | |
3336 | strna(s->description), | |
3337 | event_source_type_to_string(saved_type), | |
3338 | s->exit_on_failure ? "exiting" : "disabling"); | |
3339 | ||
3340 | if (s->exit_on_failure) | |
3341 | (void) sd_event_exit(saved_event, r); | |
3342 | } | |
12179984 LP |
3343 | |
3344 | if (s->n_ref == 0) | |
3345 | source_free(s); | |
3346 | else if (r < 0) | |
6203e07a | 3347 | sd_event_source_set_enabled(s, SD_EVENT_OFF); |
b7484e2a | 3348 | |
6203e07a | 3349 | return 1; |
fd38203a LP |
3350 | } |
3351 | ||
3352 | static int event_prepare(sd_event *e) { | |
3353 | int r; | |
3354 | ||
3355 | assert(e); | |
3356 | ||
3357 | for (;;) { | |
3358 | sd_event_source *s; | |
3359 | ||
3360 | s = prioq_peek(e->prepare); | |
baf76283 | 3361 | if (!s || s->prepare_iteration == e->iteration || s->enabled == SD_EVENT_OFF) |
fd38203a LP |
3362 | break; |
3363 | ||
3364 | s->prepare_iteration = e->iteration; | |
3365 | r = prioq_reshuffle(e->prepare, s, &s->prepare_index); | |
3366 | if (r < 0) | |
3367 | return r; | |
3368 | ||
3369 | assert(s->prepare); | |
12179984 LP |
3370 | |
3371 | s->dispatching = true; | |
fd38203a | 3372 | r = s->prepare(s, s->userdata); |
12179984 LP |
3373 | s->dispatching = false; |
3374 | ||
b778cba4 LP |
3375 | if (r < 0) { |
3376 | log_debug_errno(r, "Prepare callback of event source %s (type %s) returned error, %s: %m", | |
3377 | strna(s->description), | |
3378 | event_source_type_to_string(s->type), | |
3379 | s->exit_on_failure ? "exiting" : "disabling"); | |
3380 | ||
3381 | if (s->exit_on_failure) | |
3382 | (void) sd_event_exit(e, r); | |
3383 | } | |
fd38203a | 3384 | |
12179984 LP |
3385 | if (s->n_ref == 0) |
3386 | source_free(s); | |
3387 | else if (r < 0) | |
3388 | sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
3389 | } |
3390 | ||
3391 | return 0; | |
3392 | } | |
3393 | ||
6203e07a | 3394 | static int dispatch_exit(sd_event *e) { |
da7e457c | 3395 | sd_event_source *p; |
30dd293c | 3396 | _cleanup_(sd_event_unrefp) sd_event *ref = NULL; |
da7e457c LP |
3397 | int r; |
3398 | ||
3399 | assert(e); | |
3400 | ||
6203e07a | 3401 | p = prioq_peek(e->exit); |
baf76283 | 3402 | if (!p || p->enabled == SD_EVENT_OFF) { |
da7e457c LP |
3403 | e->state = SD_EVENT_FINISHED; |
3404 | return 0; | |
3405 | } | |
3406 | ||
30dd293c | 3407 | ref = sd_event_ref(e); |
da7e457c | 3408 | e->iteration++; |
6203e07a | 3409 | e->state = SD_EVENT_EXITING; |
da7e457c | 3410 | r = source_dispatch(p); |
2b0c9ef7 | 3411 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
3412 | return r; |
3413 | } | |
3414 | ||
c2ba3ad6 LP |
3415 | static sd_event_source* event_next_pending(sd_event *e) { |
3416 | sd_event_source *p; | |
3417 | ||
da7e457c LP |
3418 | assert(e); |
3419 | ||
c2ba3ad6 LP |
3420 | p = prioq_peek(e->pending); |
3421 | if (!p) | |
3422 | return NULL; | |
3423 | ||
baf76283 | 3424 | if (p->enabled == SD_EVENT_OFF) |
c2ba3ad6 LP |
3425 | return NULL; |
3426 | ||
3427 | return p; | |
3428 | } | |
3429 | ||
cde93897 LP |
3430 | static int arm_watchdog(sd_event *e) { |
3431 | struct itimerspec its = {}; | |
3432 | usec_t t; | |
cde93897 LP |
3433 | |
3434 | assert(e); | |
3435 | assert(e->watchdog_fd >= 0); | |
3436 | ||
3437 | t = sleep_between(e, | |
3438 | e->watchdog_last + (e->watchdog_period / 2), | |
3439 | e->watchdog_last + (e->watchdog_period * 3 / 4)); | |
3440 | ||
3441 | timespec_store(&its.it_value, t); | |
3442 | ||
75145780 LP |
3443 | /* Make sure we never set the watchdog to 0, which tells the |
3444 | * kernel to disable it. */ | |
3445 | if (its.it_value.tv_sec == 0 && its.it_value.tv_nsec == 0) | |
3446 | its.it_value.tv_nsec = 1; | |
3447 | ||
15c689d7 | 3448 | if (timerfd_settime(e->watchdog_fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 LP |
3449 | return -errno; |
3450 | ||
3451 | return 0; | |
3452 | } | |
3453 | ||
3454 | static int process_watchdog(sd_event *e) { | |
3455 | assert(e); | |
3456 | ||
3457 | if (!e->watchdog) | |
3458 | return 0; | |
3459 | ||
3460 | /* Don't notify watchdog too often */ | |
3461 | if (e->watchdog_last + e->watchdog_period / 4 > e->timestamp.monotonic) | |
3462 | return 0; | |
3463 | ||
3464 | sd_notify(false, "WATCHDOG=1"); | |
3465 | e->watchdog_last = e->timestamp.monotonic; | |
3466 | ||
3467 | return arm_watchdog(e); | |
3468 | } | |
3469 | ||
97ef5391 LP |
3470 | static void event_close_inode_data_fds(sd_event *e) { |
3471 | struct inode_data *d; | |
3472 | ||
3473 | assert(e); | |
3474 | ||
3475 | /* Close the fds pointing to the inodes to watch now. We need to close them as they might otherwise pin | |
3476 | * filesystems. But we can't close them right-away as we need them as long as the user still wants to make | |
5238e957 | 3477 | * adjustments to the even source, such as changing the priority (which requires us to remove and re-add a watch |
97ef5391 LP |
3478 | * for the inode). Hence, let's close them when entering the first iteration after they were added, as a |
3479 | * compromise. */ | |
3480 | ||
3481 | while ((d = e->inode_data_to_close)) { | |
3482 | assert(d->fd >= 0); | |
3483 | d->fd = safe_close(d->fd); | |
3484 | ||
3485 | LIST_REMOVE(to_close, e->inode_data_to_close, d); | |
3486 | } | |
3487 | } | |
3488 | ||
c45a5a74 TG |
3489 | _public_ int sd_event_prepare(sd_event *e) { |
3490 | int r; | |
fd38203a | 3491 | |
da7e457c | 3492 | assert_return(e, -EINVAL); |
b937d761 | 3493 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
3494 | assert_return(!event_pid_changed(e), -ECHILD); |
3495 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 3496 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 3497 | |
e5446015 LP |
3498 | /* Let's check that if we are a default event loop we are executed in the correct thread. We only do |
3499 | * this check here once, since gettid() is typically not cached, and thus want to minimize | |
3500 | * syscalls */ | |
3501 | assert_return(!e->default_event_ptr || e->tid == gettid(), -EREMOTEIO); | |
3502 | ||
6203e07a | 3503 | if (e->exit_requested) |
c45a5a74 | 3504 | goto pending; |
fd38203a LP |
3505 | |
3506 | e->iteration++; | |
3507 | ||
0be6c2f6 | 3508 | e->state = SD_EVENT_PREPARING; |
fd38203a | 3509 | r = event_prepare(e); |
0be6c2f6 | 3510 | e->state = SD_EVENT_INITIAL; |
fd38203a | 3511 | if (r < 0) |
c45a5a74 | 3512 | return r; |
fd38203a | 3513 | |
6a0f1f6d LP |
3514 | r = event_arm_timer(e, &e->realtime); |
3515 | if (r < 0) | |
c45a5a74 | 3516 | return r; |
6a0f1f6d | 3517 | |
a8548816 TG |
3518 | r = event_arm_timer(e, &e->boottime); |
3519 | if (r < 0) | |
c45a5a74 | 3520 | return r; |
a8548816 | 3521 | |
6a0f1f6d LP |
3522 | r = event_arm_timer(e, &e->monotonic); |
3523 | if (r < 0) | |
c45a5a74 | 3524 | return r; |
6a0f1f6d LP |
3525 | |
3526 | r = event_arm_timer(e, &e->realtime_alarm); | |
1b5995b0 | 3527 | if (r < 0) |
c45a5a74 | 3528 | return r; |
fd38203a | 3529 | |
6a0f1f6d | 3530 | r = event_arm_timer(e, &e->boottime_alarm); |
1b5995b0 | 3531 | if (r < 0) |
c45a5a74 | 3532 | return r; |
fd38203a | 3533 | |
97ef5391 LP |
3534 | event_close_inode_data_fds(e); |
3535 | ||
1b5995b0 | 3536 | if (event_next_pending(e) || e->need_process_child) |
c45a5a74 TG |
3537 | goto pending; |
3538 | ||
2b0c9ef7 | 3539 | e->state = SD_EVENT_ARMED; |
c45a5a74 TG |
3540 | |
3541 | return 0; | |
3542 | ||
3543 | pending: | |
2b0c9ef7 | 3544 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
3545 | r = sd_event_wait(e, 0); |
3546 | if (r == 0) | |
2b0c9ef7 | 3547 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
3548 | |
3549 | return r; | |
c45a5a74 TG |
3550 | } |
3551 | ||
3552 | _public_ int sd_event_wait(sd_event *e, uint64_t timeout) { | |
5cddd924 | 3553 | size_t event_queue_max; |
c45a5a74 TG |
3554 | int r, m, i; |
3555 | ||
3556 | assert_return(e, -EINVAL); | |
b937d761 | 3557 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
3558 | assert_return(!event_pid_changed(e), -ECHILD); |
3559 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 3560 | assert_return(e->state == SD_EVENT_ARMED, -EBUSY); |
c45a5a74 TG |
3561 | |
3562 | if (e->exit_requested) { | |
3563 | e->state = SD_EVENT_PENDING; | |
3564 | return 1; | |
3565 | } | |
6a0f1f6d | 3566 | |
5cddd924 LP |
3567 | event_queue_max = MAX(e->n_sources, 1u); |
3568 | if (!GREEDY_REALLOC(e->event_queue, e->event_queue_allocated, event_queue_max)) | |
3569 | return -ENOMEM; | |
fd38203a | 3570 | |
97ef5391 LP |
3571 | /* If we still have inotify data buffered, then query the other fds, but don't wait on it */ |
3572 | if (e->inotify_data_buffered) | |
3573 | timeout = 0; | |
3574 | ||
5cddd924 | 3575 | m = epoll_wait(e->epoll_fd, e->event_queue, event_queue_max, |
bab4820e | 3576 | timeout == (uint64_t) -1 ? -1 : (int) DIV_ROUND_UP(timeout, USEC_PER_MSEC)); |
da7e457c | 3577 | if (m < 0) { |
c45a5a74 TG |
3578 | if (errno == EINTR) { |
3579 | e->state = SD_EVENT_PENDING; | |
3580 | return 1; | |
3581 | } | |
3582 | ||
3583 | r = -errno; | |
da7e457c LP |
3584 | goto finish; |
3585 | } | |
fd38203a | 3586 | |
e475d10c | 3587 | triple_timestamp_get(&e->timestamp); |
fd38203a LP |
3588 | |
3589 | for (i = 0; i < m; i++) { | |
3590 | ||
5cddd924 LP |
3591 | if (e->event_queue[i].data.ptr == INT_TO_PTR(SOURCE_WATCHDOG)) |
3592 | r = flush_timer(e, e->watchdog_fd, e->event_queue[i].events, NULL); | |
9da4cb2b | 3593 | else { |
5cddd924 | 3594 | WakeupType *t = e->event_queue[i].data.ptr; |
9da4cb2b LP |
3595 | |
3596 | switch (*t) { | |
3597 | ||
f8f3f926 | 3598 | case WAKEUP_EVENT_SOURCE: { |
5cddd924 | 3599 | sd_event_source *s = e->event_queue[i].data.ptr; |
f8f3f926 LP |
3600 | |
3601 | assert(s); | |
3602 | ||
3603 | switch (s->type) { | |
3604 | ||
3605 | case SOURCE_IO: | |
5cddd924 | 3606 | r = process_io(e, s, e->event_queue[i].events); |
f8f3f926 LP |
3607 | break; |
3608 | ||
3609 | case SOURCE_CHILD: | |
5cddd924 | 3610 | r = process_pidfd(e, s, e->event_queue[i].events); |
f8f3f926 LP |
3611 | break; |
3612 | ||
3613 | default: | |
3614 | assert_not_reached("Unexpected event source type"); | |
3615 | } | |
3616 | ||
9da4cb2b | 3617 | break; |
f8f3f926 | 3618 | } |
fd38203a | 3619 | |
9da4cb2b | 3620 | case WAKEUP_CLOCK_DATA: { |
5cddd924 | 3621 | struct clock_data *d = e->event_queue[i].data.ptr; |
f8f3f926 LP |
3622 | |
3623 | assert(d); | |
3624 | ||
5cddd924 | 3625 | r = flush_timer(e, d->fd, e->event_queue[i].events, &d->next); |
9da4cb2b LP |
3626 | break; |
3627 | } | |
3628 | ||
3629 | case WAKEUP_SIGNAL_DATA: | |
5cddd924 | 3630 | r = process_signal(e, e->event_queue[i].data.ptr, e->event_queue[i].events); |
9da4cb2b LP |
3631 | break; |
3632 | ||
97ef5391 | 3633 | case WAKEUP_INOTIFY_DATA: |
5cddd924 | 3634 | r = event_inotify_data_read(e, e->event_queue[i].data.ptr, e->event_queue[i].events); |
97ef5391 LP |
3635 | break; |
3636 | ||
9da4cb2b LP |
3637 | default: |
3638 | assert_not_reached("Invalid wake-up pointer"); | |
3639 | } | |
3640 | } | |
fd38203a | 3641 | if (r < 0) |
da7e457c | 3642 | goto finish; |
fd38203a LP |
3643 | } |
3644 | ||
cde93897 LP |
3645 | r = process_watchdog(e); |
3646 | if (r < 0) | |
3647 | goto finish; | |
3648 | ||
6a0f1f6d LP |
3649 | r = process_timer(e, e->timestamp.realtime, &e->realtime); |
3650 | if (r < 0) | |
3651 | goto finish; | |
3652 | ||
e475d10c | 3653 | r = process_timer(e, e->timestamp.boottime, &e->boottime); |
a8548816 TG |
3654 | if (r < 0) |
3655 | goto finish; | |
3656 | ||
6a0f1f6d LP |
3657 | r = process_timer(e, e->timestamp.monotonic, &e->monotonic); |
3658 | if (r < 0) | |
3659 | goto finish; | |
3660 | ||
3661 | r = process_timer(e, e->timestamp.realtime, &e->realtime_alarm); | |
fd38203a | 3662 | if (r < 0) |
da7e457c | 3663 | goto finish; |
fd38203a | 3664 | |
e475d10c | 3665 | r = process_timer(e, e->timestamp.boottime, &e->boottime_alarm); |
fd38203a | 3666 | if (r < 0) |
da7e457c | 3667 | goto finish; |
fd38203a | 3668 | |
c2ba3ad6 | 3669 | if (e->need_process_child) { |
fd38203a LP |
3670 | r = process_child(e); |
3671 | if (r < 0) | |
da7e457c | 3672 | goto finish; |
fd38203a LP |
3673 | } |
3674 | ||
97ef5391 LP |
3675 | r = process_inotify(e); |
3676 | if (r < 0) | |
3677 | goto finish; | |
3678 | ||
c45a5a74 TG |
3679 | if (event_next_pending(e)) { |
3680 | e->state = SD_EVENT_PENDING; | |
3681 | ||
3682 | return 1; | |
da7e457c LP |
3683 | } |
3684 | ||
c45a5a74 | 3685 | r = 0; |
fd38203a | 3686 | |
da7e457c | 3687 | finish: |
2b0c9ef7 | 3688 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
3689 | |
3690 | return r; | |
fd38203a LP |
3691 | } |
3692 | ||
c45a5a74 TG |
3693 | _public_ int sd_event_dispatch(sd_event *e) { |
3694 | sd_event_source *p; | |
3695 | int r; | |
3696 | ||
3697 | assert_return(e, -EINVAL); | |
b937d761 | 3698 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
3699 | assert_return(!event_pid_changed(e), -ECHILD); |
3700 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
3701 | assert_return(e->state == SD_EVENT_PENDING, -EBUSY); | |
3702 | ||
3703 | if (e->exit_requested) | |
3704 | return dispatch_exit(e); | |
3705 | ||
3706 | p = event_next_pending(e); | |
3707 | if (p) { | |
30dd293c | 3708 | _cleanup_(sd_event_unrefp) sd_event *ref = NULL; |
c45a5a74 | 3709 | |
30dd293c | 3710 | ref = sd_event_ref(e); |
c45a5a74 TG |
3711 | e->state = SD_EVENT_RUNNING; |
3712 | r = source_dispatch(p); | |
2b0c9ef7 | 3713 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
3714 | return r; |
3715 | } | |
3716 | ||
2b0c9ef7 | 3717 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
3718 | |
3719 | return 1; | |
3720 | } | |
3721 | ||
34b87517 | 3722 | static void event_log_delays(sd_event *e) { |
442ac269 YW |
3723 | char b[ELEMENTSOF(e->delays) * DECIMAL_STR_MAX(unsigned) + 1], *p; |
3724 | size_t l, i; | |
34b87517 | 3725 | |
442ac269 YW |
3726 | p = b; |
3727 | l = sizeof(b); | |
3728 | for (i = 0; i < ELEMENTSOF(e->delays); i++) { | |
3729 | l = strpcpyf(&p, l, "%u ", e->delays[i]); | |
34b87517 VC |
3730 | e->delays[i] = 0; |
3731 | } | |
442ac269 | 3732 | log_debug("Event loop iterations: %s", b); |
34b87517 VC |
3733 | } |
3734 | ||
c45a5a74 TG |
3735 | _public_ int sd_event_run(sd_event *e, uint64_t timeout) { |
3736 | int r; | |
3737 | ||
3738 | assert_return(e, -EINVAL); | |
b937d761 | 3739 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
3740 | assert_return(!event_pid_changed(e), -ECHILD); |
3741 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 3742 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
c45a5a74 | 3743 | |
34b87517 VC |
3744 | if (e->profile_delays && e->last_run) { |
3745 | usec_t this_run; | |
3746 | unsigned l; | |
3747 | ||
3748 | this_run = now(CLOCK_MONOTONIC); | |
3749 | ||
3750 | l = u64log2(this_run - e->last_run); | |
3751 | assert(l < sizeof(e->delays)); | |
3752 | e->delays[l]++; | |
3753 | ||
3754 | if (this_run - e->last_log >= 5*USEC_PER_SEC) { | |
3755 | event_log_delays(e); | |
3756 | e->last_log = this_run; | |
3757 | } | |
3758 | } | |
3759 | ||
c45a5a74 | 3760 | r = sd_event_prepare(e); |
53bac4e0 LP |
3761 | if (r == 0) |
3762 | /* There was nothing? Then wait... */ | |
3763 | r = sd_event_wait(e, timeout); | |
c45a5a74 | 3764 | |
34b87517 VC |
3765 | if (e->profile_delays) |
3766 | e->last_run = now(CLOCK_MONOTONIC); | |
3767 | ||
02d30981 | 3768 | if (r > 0) { |
53bac4e0 | 3769 | /* There's something now, then let's dispatch it */ |
02d30981 TG |
3770 | r = sd_event_dispatch(e); |
3771 | if (r < 0) | |
3772 | return r; | |
53bac4e0 LP |
3773 | |
3774 | return 1; | |
3775 | } | |
3776 | ||
3777 | return r; | |
c45a5a74 TG |
3778 | } |
3779 | ||
f7262a9f | 3780 | _public_ int sd_event_loop(sd_event *e) { |
30dd293c | 3781 | _cleanup_(sd_event_unrefp) sd_event *ref = NULL; |
fd38203a LP |
3782 | int r; |
3783 | ||
da7e457c | 3784 | assert_return(e, -EINVAL); |
b937d761 | 3785 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 3786 | assert_return(!event_pid_changed(e), -ECHILD); |
2b0c9ef7 | 3787 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 3788 | |
30dd293c | 3789 | ref = sd_event_ref(e); |
fd38203a | 3790 | |
da7e457c | 3791 | while (e->state != SD_EVENT_FINISHED) { |
fd38203a LP |
3792 | r = sd_event_run(e, (uint64_t) -1); |
3793 | if (r < 0) | |
30dd293c | 3794 | return r; |
fd38203a LP |
3795 | } |
3796 | ||
30dd293c | 3797 | return e->exit_code; |
fd38203a LP |
3798 | } |
3799 | ||
9b364545 | 3800 | _public_ int sd_event_get_fd(sd_event *e) { |
9b364545 | 3801 | assert_return(e, -EINVAL); |
b937d761 | 3802 | assert_return(e = event_resolve(e), -ENOPKG); |
9b364545 TG |
3803 | assert_return(!event_pid_changed(e), -ECHILD); |
3804 | ||
3805 | return e->epoll_fd; | |
3806 | } | |
3807 | ||
f7262a9f | 3808 | _public_ int sd_event_get_state(sd_event *e) { |
da7e457c | 3809 | assert_return(e, -EINVAL); |
b937d761 | 3810 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
3811 | assert_return(!event_pid_changed(e), -ECHILD); |
3812 | ||
3813 | return e->state; | |
3814 | } | |
3815 | ||
6203e07a | 3816 | _public_ int sd_event_get_exit_code(sd_event *e, int *code) { |
da7e457c | 3817 | assert_return(e, -EINVAL); |
b937d761 | 3818 | assert_return(e = event_resolve(e), -ENOPKG); |
6203e07a | 3819 | assert_return(code, -EINVAL); |
da7e457c | 3820 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 3821 | |
6203e07a LP |
3822 | if (!e->exit_requested) |
3823 | return -ENODATA; | |
3824 | ||
3825 | *code = e->exit_code; | |
3826 | return 0; | |
fd38203a LP |
3827 | } |
3828 | ||
6203e07a | 3829 | _public_ int sd_event_exit(sd_event *e, int code) { |
da7e457c | 3830 | assert_return(e, -EINVAL); |
b937d761 | 3831 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
3832 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
3833 | assert_return(!event_pid_changed(e), -ECHILD); | |
fd38203a | 3834 | |
6203e07a LP |
3835 | e->exit_requested = true; |
3836 | e->exit_code = code; | |
3837 | ||
fd38203a LP |
3838 | return 0; |
3839 | } | |
46e8c825 | 3840 | |
6a0f1f6d | 3841 | _public_ int sd_event_now(sd_event *e, clockid_t clock, uint64_t *usec) { |
46e8c825 | 3842 | assert_return(e, -EINVAL); |
b937d761 | 3843 | assert_return(e = event_resolve(e), -ENOPKG); |
46e8c825 | 3844 | assert_return(usec, -EINVAL); |
46e8c825 LP |
3845 | assert_return(!event_pid_changed(e), -ECHILD); |
3846 | ||
e475d10c LP |
3847 | if (!TRIPLE_TIMESTAMP_HAS_CLOCK(clock)) |
3848 | return -EOPNOTSUPP; | |
3849 | ||
3850 | /* Generate a clean error in case CLOCK_BOOTTIME is not available. Note that don't use clock_supported() here, | |
3851 | * for a reason: there are systems where CLOCK_BOOTTIME is supported, but CLOCK_BOOTTIME_ALARM is not, but for | |
3852 | * the purpose of getting the time this doesn't matter. */ | |
3411372e LP |
3853 | if (IN_SET(clock, CLOCK_BOOTTIME, CLOCK_BOOTTIME_ALARM) && !clock_boottime_supported()) |
3854 | return -EOPNOTSUPP; | |
3855 | ||
e475d10c | 3856 | if (!triple_timestamp_is_set(&e->timestamp)) { |
15c689d7 | 3857 | /* Implicitly fall back to now() if we never ran before and thus have no cached time. */ |
38a03f06 LP |
3858 | *usec = now(clock); |
3859 | return 1; | |
3860 | } | |
46e8c825 | 3861 | |
e475d10c | 3862 | *usec = triple_timestamp_by_clock(&e->timestamp, clock); |
46e8c825 LP |
3863 | return 0; |
3864 | } | |
afc6adb5 LP |
3865 | |
3866 | _public_ int sd_event_default(sd_event **ret) { | |
39883f62 | 3867 | sd_event *e = NULL; |
afc6adb5 LP |
3868 | int r; |
3869 | ||
3870 | if (!ret) | |
3871 | return !!default_event; | |
3872 | ||
3873 | if (default_event) { | |
3874 | *ret = sd_event_ref(default_event); | |
3875 | return 0; | |
3876 | } | |
3877 | ||
3878 | r = sd_event_new(&e); | |
3879 | if (r < 0) | |
3880 | return r; | |
3881 | ||
3882 | e->default_event_ptr = &default_event; | |
3883 | e->tid = gettid(); | |
3884 | default_event = e; | |
3885 | ||
3886 | *ret = e; | |
3887 | return 1; | |
3888 | } | |
3889 | ||
3890 | _public_ int sd_event_get_tid(sd_event *e, pid_t *tid) { | |
3891 | assert_return(e, -EINVAL); | |
b937d761 | 3892 | assert_return(e = event_resolve(e), -ENOPKG); |
afc6adb5 | 3893 | assert_return(tid, -EINVAL); |
76b54375 | 3894 | assert_return(!event_pid_changed(e), -ECHILD); |
afc6adb5 | 3895 | |
76b54375 LP |
3896 | if (e->tid != 0) { |
3897 | *tid = e->tid; | |
3898 | return 0; | |
3899 | } | |
3900 | ||
3901 | return -ENXIO; | |
afc6adb5 | 3902 | } |
cde93897 LP |
3903 | |
3904 | _public_ int sd_event_set_watchdog(sd_event *e, int b) { | |
3905 | int r; | |
3906 | ||
3907 | assert_return(e, -EINVAL); | |
b937d761 | 3908 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 | 3909 | assert_return(!event_pid_changed(e), -ECHILD); |
cde93897 LP |
3910 | |
3911 | if (e->watchdog == !!b) | |
3912 | return e->watchdog; | |
3913 | ||
3914 | if (b) { | |
09812eb7 LP |
3915 | r = sd_watchdog_enabled(false, &e->watchdog_period); |
3916 | if (r <= 0) | |
cde93897 | 3917 | return r; |
cde93897 LP |
3918 | |
3919 | /* Issue first ping immediately */ | |
3920 | sd_notify(false, "WATCHDOG=1"); | |
3921 | e->watchdog_last = now(CLOCK_MONOTONIC); | |
3922 | ||
3923 | e->watchdog_fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC); | |
3924 | if (e->watchdog_fd < 0) | |
3925 | return -errno; | |
3926 | ||
3927 | r = arm_watchdog(e); | |
3928 | if (r < 0) | |
3929 | goto fail; | |
3930 | ||
1eac7948 | 3931 | struct epoll_event ev = { |
a82f89aa LP |
3932 | .events = EPOLLIN, |
3933 | .data.ptr = INT_TO_PTR(SOURCE_WATCHDOG), | |
3934 | }; | |
cde93897 | 3935 | |
15c689d7 | 3936 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, e->watchdog_fd, &ev) < 0) { |
cde93897 LP |
3937 | r = -errno; |
3938 | goto fail; | |
3939 | } | |
3940 | ||
3941 | } else { | |
3942 | if (e->watchdog_fd >= 0) { | |
5a795bff | 3943 | (void) epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, e->watchdog_fd, NULL); |
03e334a1 | 3944 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
3945 | } |
3946 | } | |
3947 | ||
3948 | e->watchdog = !!b; | |
3949 | return e->watchdog; | |
3950 | ||
3951 | fail: | |
03e334a1 | 3952 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
3953 | return r; |
3954 | } | |
8f726607 LP |
3955 | |
3956 | _public_ int sd_event_get_watchdog(sd_event *e) { | |
3957 | assert_return(e, -EINVAL); | |
b937d761 | 3958 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 LP |
3959 | assert_return(!event_pid_changed(e), -ECHILD); |
3960 | ||
3961 | return e->watchdog; | |
3962 | } | |
60a3b1e1 LP |
3963 | |
3964 | _public_ int sd_event_get_iteration(sd_event *e, uint64_t *ret) { | |
3965 | assert_return(e, -EINVAL); | |
b937d761 | 3966 | assert_return(e = event_resolve(e), -ENOPKG); |
60a3b1e1 LP |
3967 | assert_return(!event_pid_changed(e), -ECHILD); |
3968 | ||
3969 | *ret = e->iteration; | |
3970 | return 0; | |
3971 | } | |
15723a1d LP |
3972 | |
3973 | _public_ int sd_event_source_set_destroy_callback(sd_event_source *s, sd_event_destroy_t callback) { | |
3974 | assert_return(s, -EINVAL); | |
3975 | ||
3976 | s->destroy_callback = callback; | |
3977 | return 0; | |
3978 | } | |
3979 | ||
3980 | _public_ int sd_event_source_get_destroy_callback(sd_event_source *s, sd_event_destroy_t *ret) { | |
3981 | assert_return(s, -EINVAL); | |
3982 | ||
3983 | if (ret) | |
3984 | *ret = s->destroy_callback; | |
3985 | ||
3986 | return !!s->destroy_callback; | |
3987 | } | |
2382c936 YW |
3988 | |
3989 | _public_ int sd_event_source_get_floating(sd_event_source *s) { | |
3990 | assert_return(s, -EINVAL); | |
3991 | ||
3992 | return s->floating; | |
3993 | } | |
3994 | ||
3995 | _public_ int sd_event_source_set_floating(sd_event_source *s, int b) { | |
3996 | assert_return(s, -EINVAL); | |
3997 | ||
3998 | if (s->floating == !!b) | |
3999 | return 0; | |
4000 | ||
4001 | if (!s->event) /* Already disconnected */ | |
4002 | return -ESTALE; | |
4003 | ||
4004 | s->floating = b; | |
4005 | ||
4006 | if (b) { | |
4007 | sd_event_source_ref(s); | |
4008 | sd_event_unref(s->event); | |
4009 | } else { | |
4010 | sd_event_ref(s->event); | |
4011 | sd_event_source_unref(s); | |
4012 | } | |
4013 | ||
4014 | return 1; | |
4015 | } | |
b778cba4 LP |
4016 | |
4017 | _public_ int sd_event_source_get_exit_on_failure(sd_event_source *s) { | |
4018 | assert_return(s, -EINVAL); | |
4019 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
4020 | ||
4021 | return s->exit_on_failure; | |
4022 | } | |
4023 | ||
4024 | _public_ int sd_event_source_set_exit_on_failure(sd_event_source *s, int b) { | |
4025 | assert_return(s, -EINVAL); | |
4026 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
4027 | ||
4028 | if (s->exit_on_failure == !!b) | |
4029 | return 0; | |
4030 | ||
4031 | s->exit_on_failure = b; | |
4032 | return 1; | |
4033 | } |