]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
fd38203a LP |
2 | |
3 | #include <sys/epoll.h> | |
4 | #include <sys/timerfd.h> | |
5 | #include <sys/wait.h> | |
6 | ||
cde93897 | 7 | #include "sd-daemon.h" |
07630cea LP |
8 | #include "sd-event.h" |
9 | #include "sd-id128.h" | |
158fe190 | 10 | #include "sd-messages.h" |
07630cea | 11 | |
b5efdb8a | 12 | #include "alloc-util.h" |
f8f3f926 | 13 | #include "env-util.h" |
a137a1c3 | 14 | #include "event-source.h" |
3ffd4af2 | 15 | #include "fd-util.h" |
97ef5391 | 16 | #include "fs-util.h" |
28e5e1e9 | 17 | #include "glyph-util.h" |
fd38203a | 18 | #include "hashmap.h" |
158fe190 | 19 | #include "hexdecoct.h" |
07630cea | 20 | #include "list.h" |
3ae6b3bf | 21 | #include "logarithm.h" |
07630cea | 22 | #include "macro.h" |
158fe190 | 23 | #include "mallinfo-util.h" |
0a970718 | 24 | #include "memory-util.h" |
158fe190 | 25 | #include "missing_magic.h" |
f5947a5e | 26 | #include "missing_syscall.h" |
5545f336 | 27 | #include "missing_threads.h" |
2eeff0f4 | 28 | #include "origin-id.h" |
158fe190 | 29 | #include "path-util.h" |
07630cea | 30 | #include "prioq.h" |
4a0b58c4 | 31 | #include "process-util.h" |
158fe190 | 32 | #include "psi-util.h" |
6e9feda3 | 33 | #include "set.h" |
24882e06 | 34 | #include "signal-util.h" |
158fe190 LP |
35 | #include "socket-util.h" |
36 | #include "stat-util.h" | |
55cbfaa5 | 37 | #include "string-table.h" |
07630cea | 38 | #include "string-util.h" |
442ac269 | 39 | #include "strxcpyx.h" |
07630cea | 40 | #include "time-util.h" |
fd38203a | 41 | |
c2ba3ad6 | 42 | #define DEFAULT_ACCURACY_USEC (250 * USEC_PER_MSEC) |
fd38203a | 43 | |
f8f3f926 LP |
44 | static bool EVENT_SOURCE_WATCH_PIDFD(sd_event_source *s) { |
45 | /* Returns true if this is a PID event source and can be implemented by watching EPOLLIN */ | |
46 | return s && | |
47 | s->type == SOURCE_CHILD && | |
48 | s->child.pidfd >= 0 && | |
49 | s->child.options == WEXITED; | |
50 | } | |
51 | ||
b6d5481b LP |
52 | static bool event_source_is_online(sd_event_source *s) { |
53 | assert(s); | |
54 | return s->enabled != SD_EVENT_OFF && !s->ratelimited; | |
55 | } | |
56 | ||
57 | static bool event_source_is_offline(sd_event_source *s) { | |
58 | assert(s); | |
59 | return s->enabled == SD_EVENT_OFF || s->ratelimited; | |
60 | } | |
61 | ||
55cbfaa5 | 62 | static const char* const event_source_type_table[_SOURCE_EVENT_SOURCE_TYPE_MAX] = { |
3f3548f8 ZJS |
63 | [SOURCE_IO] = "io", |
64 | [SOURCE_TIME_REALTIME] = "realtime", | |
65 | [SOURCE_TIME_BOOTTIME] = "bootime", | |
66 | [SOURCE_TIME_MONOTONIC] = "monotonic", | |
55cbfaa5 DM |
67 | [SOURCE_TIME_REALTIME_ALARM] = "realtime-alarm", |
68 | [SOURCE_TIME_BOOTTIME_ALARM] = "boottime-alarm", | |
3f3548f8 ZJS |
69 | [SOURCE_SIGNAL] = "signal", |
70 | [SOURCE_CHILD] = "child", | |
71 | [SOURCE_DEFER] = "defer", | |
72 | [SOURCE_POST] = "post", | |
73 | [SOURCE_EXIT] = "exit", | |
74 | [SOURCE_WATCHDOG] = "watchdog", | |
75 | [SOURCE_INOTIFY] = "inotify", | |
158fe190 | 76 | [SOURCE_MEMORY_PRESSURE] = "memory-pressure", |
55cbfaa5 DM |
77 | }; |
78 | ||
79 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(event_source_type, int); | |
80 | ||
b6d5481b LP |
81 | #define EVENT_SOURCE_IS_TIME(t) \ |
82 | IN_SET((t), \ | |
83 | SOURCE_TIME_REALTIME, \ | |
84 | SOURCE_TIME_BOOTTIME, \ | |
85 | SOURCE_TIME_MONOTONIC, \ | |
86 | SOURCE_TIME_REALTIME_ALARM, \ | |
87 | SOURCE_TIME_BOOTTIME_ALARM) | |
88 | ||
89 | #define EVENT_SOURCE_CAN_RATE_LIMIT(t) \ | |
90 | IN_SET((t), \ | |
91 | SOURCE_IO, \ | |
92 | SOURCE_TIME_REALTIME, \ | |
93 | SOURCE_TIME_BOOTTIME, \ | |
94 | SOURCE_TIME_MONOTONIC, \ | |
95 | SOURCE_TIME_REALTIME_ALARM, \ | |
96 | SOURCE_TIME_BOOTTIME_ALARM, \ | |
97 | SOURCE_SIGNAL, \ | |
98 | SOURCE_DEFER, \ | |
158fe190 LP |
99 | SOURCE_INOTIFY, \ |
100 | SOURCE_MEMORY_PRESSURE) | |
6a0f1f6d | 101 | |
19947509 ZJS |
102 | /* This is used to assert that we didn't pass an unexpected source type to event_source_time_prioq_put(). |
103 | * Time sources and ratelimited sources can be passed, so effectively this is the same as the | |
104 | * EVENT_SOURCE_CAN_RATE_LIMIT() macro. */ | |
105 | #define EVENT_SOURCE_USES_TIME_PRIOQ(t) EVENT_SOURCE_CAN_RATE_LIMIT(t) | |
106 | ||
fd38203a | 107 | struct sd_event { |
da7e457c | 108 | unsigned n_ref; |
fd38203a LP |
109 | |
110 | int epoll_fd; | |
cde93897 | 111 | int watchdog_fd; |
fd38203a LP |
112 | |
113 | Prioq *pending; | |
114 | Prioq *prepare; | |
c2ba3ad6 | 115 | |
a8548816 | 116 | /* timerfd_create() only supports these five clocks so far. We |
6a0f1f6d LP |
117 | * can add support for more clocks when the kernel learns to |
118 | * deal with them, too. */ | |
119 | struct clock_data realtime; | |
a8548816 | 120 | struct clock_data boottime; |
6a0f1f6d LP |
121 | struct clock_data monotonic; |
122 | struct clock_data realtime_alarm; | |
123 | struct clock_data boottime_alarm; | |
fd38203a | 124 | |
da7e457c LP |
125 | usec_t perturb; |
126 | ||
9da4cb2b LP |
127 | sd_event_source **signal_sources; /* indexed by signal number */ |
128 | Hashmap *signal_data; /* indexed by priority */ | |
fd38203a LP |
129 | |
130 | Hashmap *child_sources; | |
b6d5481b | 131 | unsigned n_online_child_sources; |
fd38203a | 132 | |
6e9feda3 LP |
133 | Set *post_sources; |
134 | ||
6203e07a | 135 | Prioq *exit; |
fd38203a | 136 | |
97ef5391 LP |
137 | Hashmap *inotify_data; /* indexed by priority */ |
138 | ||
139 | /* A list of inode structures that still have an fd open, that we need to close before the next loop iteration */ | |
ed828563 | 140 | LIST_HEAD(struct inode_data, inode_data_to_close_list); |
97ef5391 LP |
141 | |
142 | /* A list of inotify objects that already have events buffered which aren't processed yet */ | |
0601b958 | 143 | LIST_HEAD(struct inotify_data, buffered_inotify_data_list); |
97ef5391 | 144 | |
158fe190 LP |
145 | /* A list of memory pressure event sources that still need their subscription string written */ |
146 | LIST_HEAD(sd_event_source, memory_pressure_write_list); | |
147 | ||
2eeff0f4 | 148 | uint64_t origin_id; |
c2ba3ad6 | 149 | |
60a3b1e1 | 150 | uint64_t iteration; |
e475d10c | 151 | triple_timestamp timestamp; |
da7e457c | 152 | int state; |
eaa3cbef | 153 | |
6203e07a | 154 | bool exit_requested:1; |
da7e457c | 155 | bool need_process_child:1; |
cde93897 | 156 | bool watchdog:1; |
34b87517 | 157 | bool profile_delays:1; |
afc6adb5 | 158 | |
6203e07a LP |
159 | int exit_code; |
160 | ||
afc6adb5 LP |
161 | pid_t tid; |
162 | sd_event **default_event_ptr; | |
cde93897 LP |
163 | |
164 | usec_t watchdog_last, watchdog_period; | |
15b38f93 LP |
165 | |
166 | unsigned n_sources; | |
a71fe8b8 | 167 | |
5cddd924 | 168 | struct epoll_event *event_queue; |
5cddd924 | 169 | |
a71fe8b8 | 170 | LIST_HEAD(sd_event_source, sources); |
34b87517 | 171 | |
baf3fdec LP |
172 | sd_event_source *sigint_event_source, *sigterm_event_source; |
173 | ||
e6a7bee5 | 174 | usec_t last_run_usec, last_log_usec; |
34b87517 | 175 | unsigned delays[sizeof(usec_t) * 8]; |
fd38203a LP |
176 | }; |
177 | ||
2eeff0f4 LB |
178 | DEFINE_PRIVATE_ORIGIN_ID_HELPERS(sd_event, event); |
179 | ||
b937d761 NM |
180 | static thread_local sd_event *default_event = NULL; |
181 | ||
a71fe8b8 | 182 | static void source_disconnect(sd_event_source *s); |
97ef5391 | 183 | static void event_gc_inode_data(sd_event *e, struct inode_data *d); |
a71fe8b8 | 184 | |
b937d761 NM |
185 | static sd_event *event_resolve(sd_event *e) { |
186 | return e == SD_EVENT_DEFAULT ? default_event : e; | |
187 | } | |
188 | ||
fd38203a LP |
189 | static int pending_prioq_compare(const void *a, const void *b) { |
190 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 191 | int r; |
fd38203a LP |
192 | |
193 | assert(x->pending); | |
194 | assert(y->pending); | |
195 | ||
baf76283 | 196 | /* Enabled ones first */ |
06e13147 YW |
197 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
198 | if (r != 0) | |
199 | return r; | |
fd38203a | 200 | |
b6d5481b LP |
201 | /* Non rate-limited ones first. */ |
202 | r = CMP(!!x->ratelimited, !!y->ratelimited); | |
203 | if (r != 0) | |
204 | return r; | |
205 | ||
fd38203a | 206 | /* Lower priority values first */ |
9c57a73b YW |
207 | r = CMP(x->priority, y->priority); |
208 | if (r != 0) | |
209 | return r; | |
fd38203a LP |
210 | |
211 | /* Older entries first */ | |
9c57a73b | 212 | return CMP(x->pending_iteration, y->pending_iteration); |
fd38203a LP |
213 | } |
214 | ||
215 | static int prepare_prioq_compare(const void *a, const void *b) { | |
216 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 217 | int r; |
fd38203a LP |
218 | |
219 | assert(x->prepare); | |
220 | assert(y->prepare); | |
221 | ||
8046c457 | 222 | /* Enabled ones first */ |
06e13147 YW |
223 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
224 | if (r != 0) | |
225 | return r; | |
8046c457 | 226 | |
b6d5481b LP |
227 | /* Non rate-limited ones first. */ |
228 | r = CMP(!!x->ratelimited, !!y->ratelimited); | |
229 | if (r != 0) | |
230 | return r; | |
231 | ||
fd38203a LP |
232 | /* Move most recently prepared ones last, so that we can stop |
233 | * preparing as soon as we hit one that has already been | |
234 | * prepared in the current iteration */ | |
9c57a73b YW |
235 | r = CMP(x->prepare_iteration, y->prepare_iteration); |
236 | if (r != 0) | |
237 | return r; | |
fd38203a | 238 | |
fd38203a | 239 | /* Lower priority values first */ |
9c57a73b | 240 | return CMP(x->priority, y->priority); |
fd38203a LP |
241 | } |
242 | ||
b6d5481b LP |
243 | static usec_t time_event_source_next(const sd_event_source *s) { |
244 | assert(s); | |
245 | ||
246 | /* We have two kinds of event sources that have elapsation times associated with them: the actual | |
247 | * time based ones and the ones for which a ratelimit can be in effect (where we want to be notified | |
248 | * once the ratelimit time window ends). Let's return the next elapsing time depending on what we are | |
249 | * looking at here. */ | |
250 | ||
251 | if (s->ratelimited) { /* If rate-limited the next elapsation is when the ratelimit time window ends */ | |
252 | assert(s->rate_limit.begin != 0); | |
253 | assert(s->rate_limit.interval != 0); | |
254 | return usec_add(s->rate_limit.begin, s->rate_limit.interval); | |
255 | } | |
256 | ||
257 | /* Otherwise this must be a time event source, if not ratelimited */ | |
258 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
259 | return s->time.next; | |
260 | ||
261 | return USEC_INFINITY; | |
262 | } | |
263 | ||
1bce0ffa | 264 | static usec_t time_event_source_latest(const sd_event_source *s) { |
b6d5481b LP |
265 | assert(s); |
266 | ||
267 | if (s->ratelimited) { /* For ratelimited stuff the earliest and the latest time shall actually be the | |
268 | * same, as we should avoid adding additional inaccuracy on an inaccuracy time | |
269 | * window */ | |
270 | assert(s->rate_limit.begin != 0); | |
271 | assert(s->rate_limit.interval != 0); | |
272 | return usec_add(s->rate_limit.begin, s->rate_limit.interval); | |
273 | } | |
274 | ||
275 | /* Must be a time event source, if not ratelimited */ | |
276 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
277 | return usec_add(s->time.next, s->time.accuracy); | |
278 | ||
279 | return USEC_INFINITY; | |
1bce0ffa LP |
280 | } |
281 | ||
81107b84 LP |
282 | static bool event_source_timer_candidate(const sd_event_source *s) { |
283 | assert(s); | |
284 | ||
285 | /* Returns true for event sources that either are not pending yet (i.e. where it's worth to mark them pending) | |
286 | * or which are currently ratelimited (i.e. where it's worth leaving the ratelimited state) */ | |
287 | return !s->pending || s->ratelimited; | |
288 | } | |
289 | ||
290 | static int time_prioq_compare(const void *a, const void *b, usec_t (*time_func)(const sd_event_source *s)) { | |
c2ba3ad6 | 291 | const sd_event_source *x = a, *y = b; |
06e13147 | 292 | int r; |
c2ba3ad6 | 293 | |
baf76283 | 294 | /* Enabled ones first */ |
06e13147 YW |
295 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
296 | if (r != 0) | |
297 | return r; | |
c2ba3ad6 | 298 | |
81107b84 | 299 | /* Order "non-pending OR ratelimited" before "pending AND not-ratelimited" */ |
06e13147 YW |
300 | r = CMP(!event_source_timer_candidate(x), !event_source_timer_candidate(y)); |
301 | if (r != 0) | |
302 | return r; | |
c2ba3ad6 LP |
303 | |
304 | /* Order by time */ | |
81107b84 LP |
305 | return CMP(time_func(x), time_func(y)); |
306 | } | |
307 | ||
308 | static int earliest_time_prioq_compare(const void *a, const void *b) { | |
309 | return time_prioq_compare(a, b, time_event_source_next); | |
310 | } | |
311 | ||
312 | static int latest_time_prioq_compare(const void *a, const void *b) { | |
313 | return time_prioq_compare(a, b, time_event_source_latest); | |
c2ba3ad6 LP |
314 | } |
315 | ||
6203e07a | 316 | static int exit_prioq_compare(const void *a, const void *b) { |
da7e457c | 317 | const sd_event_source *x = a, *y = b; |
06e13147 | 318 | int r; |
da7e457c | 319 | |
6203e07a LP |
320 | assert(x->type == SOURCE_EXIT); |
321 | assert(y->type == SOURCE_EXIT); | |
da7e457c | 322 | |
baf76283 | 323 | /* Enabled ones first */ |
06e13147 YW |
324 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
325 | if (r != 0) | |
326 | return r; | |
da7e457c LP |
327 | |
328 | /* Lower priority values first */ | |
6dd91b36 | 329 | return CMP(x->priority, y->priority); |
da7e457c LP |
330 | } |
331 | ||
6a0f1f6d LP |
332 | static void free_clock_data(struct clock_data *d) { |
333 | assert(d); | |
9da4cb2b | 334 | assert(d->wakeup == WAKEUP_CLOCK_DATA); |
6a0f1f6d LP |
335 | |
336 | safe_close(d->fd); | |
337 | prioq_free(d->earliest); | |
338 | prioq_free(d->latest); | |
339 | } | |
340 | ||
8301aa0b | 341 | static sd_event *event_free(sd_event *e) { |
a71fe8b8 LP |
342 | sd_event_source *s; |
343 | ||
fd38203a | 344 | assert(e); |
a71fe8b8 | 345 | |
baf3fdec LP |
346 | e->sigterm_event_source = sd_event_source_unref(e->sigterm_event_source); |
347 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
348 | ||
a71fe8b8 LP |
349 | while ((s = e->sources)) { |
350 | assert(s->floating); | |
351 | source_disconnect(s); | |
352 | sd_event_source_unref(s); | |
353 | } | |
354 | ||
15b38f93 | 355 | assert(e->n_sources == 0); |
fd38203a | 356 | |
afc6adb5 LP |
357 | if (e->default_event_ptr) |
358 | *(e->default_event_ptr) = NULL; | |
359 | ||
03e334a1 | 360 | safe_close(e->epoll_fd); |
03e334a1 | 361 | safe_close(e->watchdog_fd); |
cde93897 | 362 | |
6a0f1f6d | 363 | free_clock_data(&e->realtime); |
a8548816 | 364 | free_clock_data(&e->boottime); |
6a0f1f6d LP |
365 | free_clock_data(&e->monotonic); |
366 | free_clock_data(&e->realtime_alarm); | |
367 | free_clock_data(&e->boottime_alarm); | |
368 | ||
fd38203a LP |
369 | prioq_free(e->pending); |
370 | prioq_free(e->prepare); | |
6203e07a | 371 | prioq_free(e->exit); |
fd38203a LP |
372 | |
373 | free(e->signal_sources); | |
9da4cb2b | 374 | hashmap_free(e->signal_data); |
fd38203a | 375 | |
97ef5391 LP |
376 | hashmap_free(e->inotify_data); |
377 | ||
fd38203a | 378 | hashmap_free(e->child_sources); |
6e9feda3 | 379 | set_free(e->post_sources); |
8301aa0b | 380 | |
5cddd924 LP |
381 | free(e->event_queue); |
382 | ||
8301aa0b | 383 | return mfree(e); |
fd38203a LP |
384 | } |
385 | ||
f7262a9f | 386 | _public_ int sd_event_new(sd_event** ret) { |
fd38203a LP |
387 | sd_event *e; |
388 | int r; | |
389 | ||
305f78bf | 390 | assert_return(ret, -EINVAL); |
fd38203a | 391 | |
d08eb1fa | 392 | e = new(sd_event, 1); |
fd38203a LP |
393 | if (!e) |
394 | return -ENOMEM; | |
395 | ||
d08eb1fa LP |
396 | *e = (sd_event) { |
397 | .n_ref = 1, | |
254d1313 ZJS |
398 | .epoll_fd = -EBADF, |
399 | .watchdog_fd = -EBADF, | |
d08eb1fa | 400 | .realtime.wakeup = WAKEUP_CLOCK_DATA, |
254d1313 | 401 | .realtime.fd = -EBADF, |
d08eb1fa LP |
402 | .realtime.next = USEC_INFINITY, |
403 | .boottime.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 404 | .boottime.fd = -EBADF, |
d08eb1fa LP |
405 | .boottime.next = USEC_INFINITY, |
406 | .monotonic.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 407 | .monotonic.fd = -EBADF, |
d08eb1fa LP |
408 | .monotonic.next = USEC_INFINITY, |
409 | .realtime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 410 | .realtime_alarm.fd = -EBADF, |
d08eb1fa LP |
411 | .realtime_alarm.next = USEC_INFINITY, |
412 | .boottime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 413 | .boottime_alarm.fd = -EBADF, |
d08eb1fa LP |
414 | .boottime_alarm.next = USEC_INFINITY, |
415 | .perturb = USEC_INFINITY, | |
2eeff0f4 | 416 | .origin_id = origin_id_query(), |
d08eb1fa | 417 | }; |
fd38203a | 418 | |
c983e776 EV |
419 | r = prioq_ensure_allocated(&e->pending, pending_prioq_compare); |
420 | if (r < 0) | |
fd38203a | 421 | goto fail; |
fd38203a LP |
422 | |
423 | e->epoll_fd = epoll_create1(EPOLL_CLOEXEC); | |
424 | if (e->epoll_fd < 0) { | |
425 | r = -errno; | |
426 | goto fail; | |
427 | } | |
428 | ||
7fe2903c LP |
429 | e->epoll_fd = fd_move_above_stdio(e->epoll_fd); |
430 | ||
34b87517 | 431 | if (secure_getenv("SD_EVENT_PROFILE_DELAYS")) { |
28e5e1e9 DT |
432 | log_debug("Event loop profiling enabled. Logarithmic histogram of event loop iterations in the range 2^0 %s 2^63 us will be logged every 5s.", |
433 | special_glyph(SPECIAL_GLYPH_ELLIPSIS)); | |
34b87517 VC |
434 | e->profile_delays = true; |
435 | } | |
436 | ||
fd38203a LP |
437 | *ret = e; |
438 | return 0; | |
439 | ||
440 | fail: | |
441 | event_free(e); | |
442 | return r; | |
443 | } | |
444 | ||
2eeff0f4 LB |
445 | /* Define manually so we can add the origin check */ |
446 | _public_ sd_event *sd_event_ref(sd_event *e) { | |
447 | if (!e) | |
448 | return NULL; | |
449 | if (event_origin_changed(e)) | |
450 | return NULL; | |
451 | ||
452 | e->n_ref++; | |
453 | ||
454 | return e; | |
455 | } | |
456 | ||
457 | _public_ sd_event* sd_event_unref(sd_event *e) { | |
458 | if (!e) | |
459 | return NULL; | |
460 | if (event_origin_changed(e)) | |
461 | return NULL; | |
462 | ||
463 | assert(e->n_ref > 0); | |
464 | if (--e->n_ref > 0) | |
465 | return NULL; | |
466 | ||
467 | return event_free(e); | |
468 | } | |
469 | ||
c8e9d15c YW |
470 | #define PROTECT_EVENT(e) \ |
471 | _unused_ _cleanup_(sd_event_unrefp) sd_event *_ref = sd_event_ref(e); | |
fd38203a | 472 | |
afd15bbb ZJS |
473 | _public_ sd_event_source* sd_event_source_disable_unref(sd_event_source *s) { |
474 | if (s) | |
475 | (void) sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
476 | return sd_event_source_unref(s); | |
477 | } | |
478 | ||
366e6411 | 479 | static void source_io_unregister(sd_event_source *s) { |
fd38203a LP |
480 | assert(s); |
481 | assert(s->type == SOURCE_IO); | |
482 | ||
2eeff0f4 | 483 | if (event_origin_changed(s->event)) |
366e6411 | 484 | return; |
f6806734 | 485 | |
fd38203a | 486 | if (!s->io.registered) |
366e6411 | 487 | return; |
fd38203a | 488 | |
d1cf2023 | 489 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->io.fd, NULL) < 0) |
f80a5d6a | 490 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll, ignoring: %m", |
55cbfaa5 | 491 | strna(s->description), event_source_type_to_string(s->type)); |
fd38203a LP |
492 | |
493 | s->io.registered = false; | |
fd38203a LP |
494 | } |
495 | ||
305f78bf LP |
496 | static int source_io_register( |
497 | sd_event_source *s, | |
498 | int enabled, | |
499 | uint32_t events) { | |
500 | ||
fd38203a LP |
501 | assert(s); |
502 | assert(s->type == SOURCE_IO); | |
baf76283 | 503 | assert(enabled != SD_EVENT_OFF); |
fd38203a | 504 | |
1eac7948 | 505 | struct epoll_event ev = { |
a82f89aa LP |
506 | .events = events | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0), |
507 | .data.ptr = s, | |
508 | }; | |
fd38203a | 509 | |
15c689d7 | 510 | if (epoll_ctl(s->event->epoll_fd, |
1eac7948 | 511 | s->io.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, |
55c540d3 | 512 | s->io.fd, &ev) < 0) |
fd38203a LP |
513 | return -errno; |
514 | ||
515 | s->io.registered = true; | |
516 | ||
517 | return 0; | |
518 | } | |
519 | ||
f8f3f926 LP |
520 | static void source_child_pidfd_unregister(sd_event_source *s) { |
521 | assert(s); | |
522 | assert(s->type == SOURCE_CHILD); | |
523 | ||
2eeff0f4 | 524 | if (event_origin_changed(s->event)) |
f8f3f926 LP |
525 | return; |
526 | ||
527 | if (!s->child.registered) | |
528 | return; | |
529 | ||
530 | if (EVENT_SOURCE_WATCH_PIDFD(s)) | |
531 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->child.pidfd, NULL) < 0) | |
f80a5d6a | 532 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll, ignoring: %m", |
f8f3f926 LP |
533 | strna(s->description), event_source_type_to_string(s->type)); |
534 | ||
535 | s->child.registered = false; | |
536 | } | |
537 | ||
538 | static int source_child_pidfd_register(sd_event_source *s, int enabled) { | |
f8f3f926 LP |
539 | assert(s); |
540 | assert(s->type == SOURCE_CHILD); | |
541 | assert(enabled != SD_EVENT_OFF); | |
542 | ||
543 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { | |
1eac7948 | 544 | struct epoll_event ev = { |
f8f3f926 LP |
545 | .events = EPOLLIN | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0), |
546 | .data.ptr = s, | |
547 | }; | |
548 | ||
55c540d3 ZJS |
549 | if (epoll_ctl(s->event->epoll_fd, |
550 | s->child.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, | |
551 | s->child.pidfd, &ev) < 0) | |
f8f3f926 LP |
552 | return -errno; |
553 | } | |
554 | ||
555 | s->child.registered = true; | |
556 | return 0; | |
557 | } | |
558 | ||
158fe190 LP |
559 | static void source_memory_pressure_unregister(sd_event_source *s) { |
560 | assert(s); | |
561 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
562 | ||
2eeff0f4 | 563 | if (event_origin_changed(s->event)) |
158fe190 LP |
564 | return; |
565 | ||
566 | if (!s->memory_pressure.registered) | |
567 | return; | |
568 | ||
569 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->memory_pressure.fd, NULL) < 0) | |
570 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll, ignoring: %m", | |
571 | strna(s->description), event_source_type_to_string(s->type)); | |
572 | ||
573 | s->memory_pressure.registered = false; | |
574 | } | |
575 | ||
576 | static int source_memory_pressure_register(sd_event_source *s, int enabled) { | |
577 | assert(s); | |
578 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
579 | assert(enabled != SD_EVENT_OFF); | |
580 | ||
581 | struct epoll_event ev = { | |
582 | .events = s->memory_pressure.write_buffer_size > 0 ? EPOLLOUT : | |
583 | (s->memory_pressure.events | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0)), | |
584 | .data.ptr = s, | |
585 | }; | |
586 | ||
587 | if (epoll_ctl(s->event->epoll_fd, | |
588 | s->memory_pressure.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, | |
589 | s->memory_pressure.fd, &ev) < 0) | |
590 | return -errno; | |
591 | ||
592 | s->memory_pressure.registered = true; | |
593 | return 0; | |
594 | } | |
595 | ||
596 | static void source_memory_pressure_add_to_write_list(sd_event_source *s) { | |
597 | assert(s); | |
598 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
599 | ||
600 | if (s->memory_pressure.in_write_list) | |
601 | return; | |
602 | ||
603 | LIST_PREPEND(memory_pressure.write_list, s->event->memory_pressure_write_list, s); | |
604 | s->memory_pressure.in_write_list = true; | |
605 | } | |
606 | ||
607 | static void source_memory_pressure_remove_from_write_list(sd_event_source *s) { | |
608 | assert(s); | |
609 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
610 | ||
611 | if (!s->memory_pressure.in_write_list) | |
612 | return; | |
613 | ||
614 | LIST_REMOVE(memory_pressure.write_list, s->event->memory_pressure_write_list, s); | |
615 | s->memory_pressure.in_write_list = false; | |
616 | } | |
617 | ||
6a0f1f6d LP |
618 | static clockid_t event_source_type_to_clock(EventSourceType t) { |
619 | ||
620 | switch (t) { | |
621 | ||
622 | case SOURCE_TIME_REALTIME: | |
623 | return CLOCK_REALTIME; | |
624 | ||
a8548816 TG |
625 | case SOURCE_TIME_BOOTTIME: |
626 | return CLOCK_BOOTTIME; | |
627 | ||
6a0f1f6d LP |
628 | case SOURCE_TIME_MONOTONIC: |
629 | return CLOCK_MONOTONIC; | |
630 | ||
631 | case SOURCE_TIME_REALTIME_ALARM: | |
632 | return CLOCK_REALTIME_ALARM; | |
633 | ||
634 | case SOURCE_TIME_BOOTTIME_ALARM: | |
635 | return CLOCK_BOOTTIME_ALARM; | |
636 | ||
637 | default: | |
638 | return (clockid_t) -1; | |
639 | } | |
640 | } | |
641 | ||
642 | static EventSourceType clock_to_event_source_type(clockid_t clock) { | |
643 | ||
644 | switch (clock) { | |
645 | ||
646 | case CLOCK_REALTIME: | |
647 | return SOURCE_TIME_REALTIME; | |
648 | ||
a8548816 TG |
649 | case CLOCK_BOOTTIME: |
650 | return SOURCE_TIME_BOOTTIME; | |
651 | ||
6a0f1f6d LP |
652 | case CLOCK_MONOTONIC: |
653 | return SOURCE_TIME_MONOTONIC; | |
654 | ||
655 | case CLOCK_REALTIME_ALARM: | |
656 | return SOURCE_TIME_REALTIME_ALARM; | |
657 | ||
658 | case CLOCK_BOOTTIME_ALARM: | |
659 | return SOURCE_TIME_BOOTTIME_ALARM; | |
660 | ||
661 | default: | |
662 | return _SOURCE_EVENT_SOURCE_TYPE_INVALID; | |
663 | } | |
664 | } | |
665 | ||
666 | static struct clock_data* event_get_clock_data(sd_event *e, EventSourceType t) { | |
667 | assert(e); | |
668 | ||
669 | switch (t) { | |
670 | ||
671 | case SOURCE_TIME_REALTIME: | |
672 | return &e->realtime; | |
673 | ||
a8548816 TG |
674 | case SOURCE_TIME_BOOTTIME: |
675 | return &e->boottime; | |
676 | ||
6a0f1f6d LP |
677 | case SOURCE_TIME_MONOTONIC: |
678 | return &e->monotonic; | |
679 | ||
680 | case SOURCE_TIME_REALTIME_ALARM: | |
681 | return &e->realtime_alarm; | |
682 | ||
683 | case SOURCE_TIME_BOOTTIME_ALARM: | |
684 | return &e->boottime_alarm; | |
685 | ||
686 | default: | |
687 | return NULL; | |
688 | } | |
689 | } | |
690 | ||
3e4eb8e7 YW |
691 | static void event_free_signal_data(sd_event *e, struct signal_data *d) { |
692 | assert(e); | |
693 | ||
694 | if (!d) | |
695 | return; | |
696 | ||
697 | hashmap_remove(e->signal_data, &d->priority); | |
698 | safe_close(d->fd); | |
699 | free(d); | |
700 | } | |
701 | ||
9da4cb2b LP |
702 | static int event_make_signal_data( |
703 | sd_event *e, | |
704 | int sig, | |
705 | struct signal_data **ret) { | |
4807d2d0 | 706 | |
9da4cb2b LP |
707 | struct signal_data *d; |
708 | bool added = false; | |
709 | sigset_t ss_copy; | |
710 | int64_t priority; | |
f95387cd ZJS |
711 | int r; |
712 | ||
713 | assert(e); | |
714 | ||
2eeff0f4 | 715 | if (event_origin_changed(e)) |
9da4cb2b | 716 | return -ECHILD; |
f6806734 | 717 | |
9da4cb2b LP |
718 | if (e->signal_sources && e->signal_sources[sig]) |
719 | priority = e->signal_sources[sig]->priority; | |
720 | else | |
de05913d | 721 | priority = SD_EVENT_PRIORITY_NORMAL; |
f95387cd | 722 | |
9da4cb2b LP |
723 | d = hashmap_get(e->signal_data, &priority); |
724 | if (d) { | |
725 | if (sigismember(&d->sigset, sig) > 0) { | |
726 | if (ret) | |
727 | *ret = d; | |
728 | return 0; | |
729 | } | |
730 | } else { | |
d08eb1fa | 731 | d = new(struct signal_data, 1); |
9da4cb2b LP |
732 | if (!d) |
733 | return -ENOMEM; | |
734 | ||
d08eb1fa LP |
735 | *d = (struct signal_data) { |
736 | .wakeup = WAKEUP_SIGNAL_DATA, | |
254d1313 | 737 | .fd = -EBADF, |
d08eb1fa LP |
738 | .priority = priority, |
739 | }; | |
9da4cb2b | 740 | |
f656fdb6 | 741 | r = hashmap_ensure_put(&e->signal_data, &uint64_hash_ops, &d->priority, d); |
90f604d1 ZJS |
742 | if (r < 0) { |
743 | free(d); | |
9da4cb2b | 744 | return r; |
90f604d1 | 745 | } |
f95387cd | 746 | |
9da4cb2b LP |
747 | added = true; |
748 | } | |
749 | ||
750 | ss_copy = d->sigset; | |
751 | assert_se(sigaddset(&ss_copy, sig) >= 0); | |
752 | ||
cbff793f ZJS |
753 | r = signalfd(d->fd >= 0 ? d->fd : -1, /* the first arg must be -1 or a valid signalfd */ |
754 | &ss_copy, | |
755 | SFD_NONBLOCK|SFD_CLOEXEC); | |
9da4cb2b LP |
756 | if (r < 0) { |
757 | r = -errno; | |
758 | goto fail; | |
759 | } | |
760 | ||
761 | d->sigset = ss_copy; | |
f95387cd | 762 | |
9da4cb2b LP |
763 | if (d->fd >= 0) { |
764 | if (ret) | |
765 | *ret = d; | |
f95387cd | 766 | return 0; |
9da4cb2b LP |
767 | } |
768 | ||
7fe2903c | 769 | d->fd = fd_move_above_stdio(r); |
f95387cd | 770 | |
1eac7948 | 771 | struct epoll_event ev = { |
a82f89aa LP |
772 | .events = EPOLLIN, |
773 | .data.ptr = d, | |
774 | }; | |
f95387cd | 775 | |
15c689d7 | 776 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { |
9da4cb2b LP |
777 | r = -errno; |
778 | goto fail; | |
f95387cd ZJS |
779 | } |
780 | ||
9da4cb2b LP |
781 | if (ret) |
782 | *ret = d; | |
783 | ||
f95387cd | 784 | return 0; |
9da4cb2b LP |
785 | |
786 | fail: | |
3e4eb8e7 YW |
787 | if (added) |
788 | event_free_signal_data(e, d); | |
9da4cb2b LP |
789 | |
790 | return r; | |
791 | } | |
792 | ||
793 | static void event_unmask_signal_data(sd_event *e, struct signal_data *d, int sig) { | |
794 | assert(e); | |
795 | assert(d); | |
796 | ||
797 | /* Turns off the specified signal in the signal data | |
798 | * object. If the signal mask of the object becomes empty that | |
799 | * way removes it. */ | |
800 | ||
801 | if (sigismember(&d->sigset, sig) == 0) | |
802 | return; | |
803 | ||
804 | assert_se(sigdelset(&d->sigset, sig) >= 0); | |
805 | ||
806 | if (sigisemptyset(&d->sigset)) { | |
9da4cb2b | 807 | /* If all the mask is all-zero we can get rid of the structure */ |
3e4eb8e7 | 808 | event_free_signal_data(e, d); |
9da4cb2b LP |
809 | return; |
810 | } | |
811 | ||
2eeff0f4 | 812 | if (event_origin_changed(e)) |
01e6af73 YW |
813 | return; |
814 | ||
9da4cb2b LP |
815 | assert(d->fd >= 0); |
816 | ||
817 | if (signalfd(d->fd, &d->sigset, SFD_NONBLOCK|SFD_CLOEXEC) < 0) | |
818 | log_debug_errno(errno, "Failed to unset signal bit, ignoring: %m"); | |
819 | } | |
820 | ||
821 | static void event_gc_signal_data(sd_event *e, const int64_t *priority, int sig) { | |
822 | struct signal_data *d; | |
823 | static const int64_t zero_priority = 0; | |
824 | ||
825 | assert(e); | |
826 | ||
f8f3f926 LP |
827 | /* Rechecks if the specified signal is still something we are interested in. If not, we'll unmask it, |
828 | * and possibly drop the signalfd for it. */ | |
9da4cb2b LP |
829 | |
830 | if (sig == SIGCHLD && | |
b6d5481b | 831 | e->n_online_child_sources > 0) |
9da4cb2b LP |
832 | return; |
833 | ||
834 | if (e->signal_sources && | |
835 | e->signal_sources[sig] && | |
b6d5481b | 836 | event_source_is_online(e->signal_sources[sig])) |
9da4cb2b LP |
837 | return; |
838 | ||
839 | /* | |
840 | * The specified signal might be enabled in three different queues: | |
841 | * | |
842 | * 1) the one that belongs to the priority passed (if it is non-NULL) | |
843 | * 2) the one that belongs to the priority of the event source of the signal (if there is one) | |
844 | * 3) the 0 priority (to cover the SIGCHLD case) | |
845 | * | |
846 | * Hence, let's remove it from all three here. | |
847 | */ | |
848 | ||
849 | if (priority) { | |
850 | d = hashmap_get(e->signal_data, priority); | |
851 | if (d) | |
852 | event_unmask_signal_data(e, d, sig); | |
853 | } | |
854 | ||
855 | if (e->signal_sources && e->signal_sources[sig]) { | |
856 | d = hashmap_get(e->signal_data, &e->signal_sources[sig]->priority); | |
857 | if (d) | |
858 | event_unmask_signal_data(e, d, sig); | |
859 | } | |
860 | ||
861 | d = hashmap_get(e->signal_data, &zero_priority); | |
862 | if (d) | |
863 | event_unmask_signal_data(e, d, sig); | |
f95387cd ZJS |
864 | } |
865 | ||
e1951c16 MS |
866 | static void event_source_pp_prioq_reshuffle(sd_event_source *s) { |
867 | assert(s); | |
868 | ||
869 | /* Reshuffles the pending + prepare prioqs. Called whenever the dispatch order changes, i.e. when | |
870 | * they are enabled/disabled or marked pending and such. */ | |
871 | ||
872 | if (s->pending) | |
873 | prioq_reshuffle(s->event->pending, s, &s->pending_index); | |
874 | ||
875 | if (s->prepare) | |
876 | prioq_reshuffle(s->event->prepare, s, &s->prepare_index); | |
877 | } | |
878 | ||
879 | static void event_source_time_prioq_reshuffle(sd_event_source *s) { | |
880 | struct clock_data *d; | |
881 | ||
882 | assert(s); | |
e1951c16 MS |
883 | |
884 | /* Called whenever the event source's timer ordering properties changed, i.e. time, accuracy, | |
5c08c7ab YW |
885 | * pending, enable state, and ratelimiting state. Makes sure the two prioq's are ordered |
886 | * properly again. */ | |
b6d5481b LP |
887 | |
888 | if (s->ratelimited) | |
889 | d = &s->event->monotonic; | |
5c08c7ab | 890 | else if (EVENT_SOURCE_IS_TIME(s->type)) |
b6d5481b | 891 | assert_se(d = event_get_clock_data(s->event, s->type)); |
5c08c7ab YW |
892 | else |
893 | return; /* no-op for an event source which is neither a timer nor ratelimited. */ | |
b6d5481b | 894 | |
f41315fc LP |
895 | prioq_reshuffle(d->earliest, s, &s->earliest_index); |
896 | prioq_reshuffle(d->latest, s, &s->latest_index); | |
e1951c16 MS |
897 | d->needs_rearm = true; |
898 | } | |
899 | ||
1e45e3fe LP |
900 | static void event_source_time_prioq_remove( |
901 | sd_event_source *s, | |
902 | struct clock_data *d) { | |
903 | ||
904 | assert(s); | |
905 | assert(d); | |
906 | ||
f41315fc LP |
907 | prioq_remove(d->earliest, s, &s->earliest_index); |
908 | prioq_remove(d->latest, s, &s->latest_index); | |
909 | s->earliest_index = s->latest_index = PRIOQ_IDX_NULL; | |
1e45e3fe LP |
910 | d->needs_rearm = true; |
911 | } | |
912 | ||
a71fe8b8 LP |
913 | static void source_disconnect(sd_event_source *s) { |
914 | sd_event *event; | |
897448bd | 915 | int r; |
a71fe8b8 | 916 | |
fd38203a LP |
917 | assert(s); |
918 | ||
a71fe8b8 LP |
919 | if (!s->event) |
920 | return; | |
15b38f93 | 921 | |
a71fe8b8 | 922 | assert(s->event->n_sources > 0); |
fd38203a | 923 | |
a71fe8b8 | 924 | switch (s->type) { |
fd38203a | 925 | |
a71fe8b8 LP |
926 | case SOURCE_IO: |
927 | if (s->io.fd >= 0) | |
928 | source_io_unregister(s); | |
fd38203a | 929 | |
a71fe8b8 | 930 | break; |
6a0f1f6d | 931 | |
a71fe8b8 | 932 | case SOURCE_TIME_REALTIME: |
a8548816 | 933 | case SOURCE_TIME_BOOTTIME: |
a71fe8b8 LP |
934 | case SOURCE_TIME_MONOTONIC: |
935 | case SOURCE_TIME_REALTIME_ALARM: | |
b6d5481b LP |
936 | case SOURCE_TIME_BOOTTIME_ALARM: |
937 | /* Only remove this event source from the time event source here if it is not ratelimited. If | |
938 | * it is ratelimited, we'll remove it below, separately. Why? Because the clock used might | |
939 | * differ: ratelimiting always uses CLOCK_MONOTONIC, but timer events might use any clock */ | |
940 | ||
941 | if (!s->ratelimited) { | |
942 | struct clock_data *d; | |
943 | assert_se(d = event_get_clock_data(s->event, s->type)); | |
944 | event_source_time_prioq_remove(s, d); | |
945 | } | |
946 | ||
a71fe8b8 | 947 | break; |
a71fe8b8 LP |
948 | |
949 | case SOURCE_SIGNAL: | |
950 | if (s->signal.sig > 0) { | |
9da4cb2b | 951 | |
a71fe8b8 LP |
952 | if (s->event->signal_sources) |
953 | s->event->signal_sources[s->signal.sig] = NULL; | |
4807d2d0 | 954 | |
9da4cb2b | 955 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
897448bd LP |
956 | |
957 | if (s->signal.unblock) { | |
958 | sigset_t new_ss; | |
959 | ||
960 | if (sigemptyset(&new_ss) < 0) | |
961 | log_debug_errno(errno, "Failed to reset signal set, ignoring: %m"); | |
962 | else if (sigaddset(&new_ss, s->signal.sig) < 0) | |
963 | log_debug_errno(errno, "Failed to add signal %i to signal mask, ignoring: %m", s->signal.sig); | |
964 | else { | |
965 | r = pthread_sigmask(SIG_UNBLOCK, &new_ss, NULL); | |
966 | if (r != 0) | |
967 | log_debug_errno(r, "Failed to unblock signal %i, ignoring: %m", s->signal.sig); | |
968 | } | |
969 | } | |
6a0f1f6d | 970 | } |
fd38203a | 971 | |
a71fe8b8 | 972 | break; |
fd38203a | 973 | |
a71fe8b8 | 974 | case SOURCE_CHILD: |
2eeff0f4 | 975 | if (event_origin_changed(s->event)) |
86587c93 YW |
976 | s->child.process_owned = false; |
977 | ||
a71fe8b8 | 978 | if (s->child.pid > 0) { |
b6d5481b LP |
979 | if (event_source_is_online(s)) { |
980 | assert(s->event->n_online_child_sources > 0); | |
981 | s->event->n_online_child_sources--; | |
4807d2d0 | 982 | } |
fd38203a | 983 | |
4a0b58c4 | 984 | (void) hashmap_remove(s->event->child_sources, PID_TO_PTR(s->child.pid)); |
a71fe8b8 | 985 | } |
fd38203a | 986 | |
f8f3f926 LP |
987 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
988 | source_child_pidfd_unregister(s); | |
989 | else | |
990 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
991 | ||
a71fe8b8 | 992 | break; |
fd38203a | 993 | |
a71fe8b8 LP |
994 | case SOURCE_DEFER: |
995 | /* nothing */ | |
996 | break; | |
fd38203a | 997 | |
a71fe8b8 LP |
998 | case SOURCE_POST: |
999 | set_remove(s->event->post_sources, s); | |
1000 | break; | |
da7e457c | 1001 | |
a71fe8b8 LP |
1002 | case SOURCE_EXIT: |
1003 | prioq_remove(s->event->exit, s, &s->exit.prioq_index); | |
1004 | break; | |
0eb2e0e3 | 1005 | |
97ef5391 LP |
1006 | case SOURCE_INOTIFY: { |
1007 | struct inode_data *inode_data; | |
1008 | ||
1009 | inode_data = s->inotify.inode_data; | |
1010 | if (inode_data) { | |
1011 | struct inotify_data *inotify_data; | |
1012 | assert_se(inotify_data = inode_data->inotify_data); | |
1013 | ||
1014 | /* Detach this event source from the inode object */ | |
1015 | LIST_REMOVE(inotify.by_inode_data, inode_data->event_sources, s); | |
1016 | s->inotify.inode_data = NULL; | |
1017 | ||
1018 | if (s->pending) { | |
1019 | assert(inotify_data->n_pending > 0); | |
1020 | inotify_data->n_pending--; | |
1021 | } | |
1022 | ||
1023 | /* Note that we don't reduce the inotify mask for the watch descriptor here if the inode is | |
1024 | * continued to being watched. That's because inotify doesn't really have an API for that: we | |
1025 | * can only change watch masks with access to the original inode either by fd or by path. But | |
1026 | * paths aren't stable, and keeping an O_PATH fd open all the time would mean wasting an fd | |
f21f31b2 | 1027 | * continuously and keeping the mount busy which we can't really do. We could reconstruct the |
97ef5391 LP |
1028 | * original inode from /proc/self/fdinfo/$INOTIFY_FD (as all watch descriptors are listed |
1029 | * there), but given the need for open_by_handle_at() which is privileged and not universally | |
1030 | * available this would be quite an incomplete solution. Hence we go the other way, leave the | |
1031 | * mask set, even if it is not minimized now, and ignore all events we aren't interested in | |
1032 | * anymore after reception. Yes, this sucks, but … Linux … */ | |
1033 | ||
1034 | /* Maybe release the inode data (and its inotify) */ | |
1035 | event_gc_inode_data(s->event, inode_data); | |
1036 | } | |
1037 | ||
1038 | break; | |
1039 | } | |
1040 | ||
158fe190 LP |
1041 | case SOURCE_MEMORY_PRESSURE: |
1042 | source_memory_pressure_remove_from_write_list(s); | |
1043 | source_memory_pressure_unregister(s); | |
1044 | break; | |
1045 | ||
a71fe8b8 | 1046 | default: |
04499a70 | 1047 | assert_not_reached(); |
a71fe8b8 | 1048 | } |
6e9feda3 | 1049 | |
a71fe8b8 LP |
1050 | if (s->pending) |
1051 | prioq_remove(s->event->pending, s, &s->pending_index); | |
9d3e3aa5 | 1052 | |
a71fe8b8 LP |
1053 | if (s->prepare) |
1054 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
fd38203a | 1055 | |
b6d5481b LP |
1056 | if (s->ratelimited) |
1057 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
1058 | ||
e514aa1e | 1059 | event = TAKE_PTR(s->event); |
a71fe8b8 LP |
1060 | LIST_REMOVE(sources, event->sources, s); |
1061 | event->n_sources--; | |
fd38203a | 1062 | |
f5982559 LP |
1063 | /* Note that we don't invalidate the type here, since we still need it in order to close the fd or |
1064 | * pidfd associated with this event source, which we'll do only on source_free(). */ | |
1065 | ||
a71fe8b8 LP |
1066 | if (!s->floating) |
1067 | sd_event_unref(event); | |
1068 | } | |
1069 | ||
75db809a | 1070 | static sd_event_source* source_free(sd_event_source *s) { |
a71fe8b8 | 1071 | assert(s); |
fd38203a | 1072 | |
a71fe8b8 | 1073 | source_disconnect(s); |
ab93297c NM |
1074 | |
1075 | if (s->type == SOURCE_IO && s->io.owned) | |
15723a1d LP |
1076 | s->io.fd = safe_close(s->io.fd); |
1077 | ||
f8f3f926 LP |
1078 | if (s->type == SOURCE_CHILD) { |
1079 | /* Eventually the kernel will do this automatically for us, but for now let's emulate this (unreliably) in userspace. */ | |
1080 | ||
1081 | if (s->child.process_owned) { | |
1082 | ||
1083 | if (!s->child.exited) { | |
1084 | bool sent = false; | |
1085 | ||
1086 | if (s->child.pidfd >= 0) { | |
1087 | if (pidfd_send_signal(s->child.pidfd, SIGKILL, NULL, 0) < 0) { | |
1088 | if (errno == ESRCH) /* Already dead */ | |
1089 | sent = true; | |
1090 | else if (!ERRNO_IS_NOT_SUPPORTED(errno)) | |
1091 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via pidfd_send_signal(), re-trying via kill(): %m", | |
1092 | s->child.pid); | |
1093 | } else | |
1094 | sent = true; | |
1095 | } | |
1096 | ||
1097 | if (!sent) | |
1098 | if (kill(s->child.pid, SIGKILL) < 0) | |
1099 | if (errno != ESRCH) /* Already dead */ | |
1100 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via kill(), ignoring: %m", | |
1101 | s->child.pid); | |
1102 | } | |
1103 | ||
1104 | if (!s->child.waited) { | |
1105 | siginfo_t si = {}; | |
1106 | ||
1107 | /* Reap the child if we can */ | |
1108 | (void) waitid(P_PID, s->child.pid, &si, WEXITED); | |
1109 | } | |
1110 | } | |
1111 | ||
1112 | if (s->child.pidfd_owned) | |
1113 | s->child.pidfd = safe_close(s->child.pidfd); | |
1114 | } | |
1115 | ||
158fe190 LP |
1116 | if (s->type == SOURCE_MEMORY_PRESSURE) { |
1117 | s->memory_pressure.fd = safe_close(s->memory_pressure.fd); | |
1118 | s->memory_pressure.write_buffer = mfree(s->memory_pressure.write_buffer); | |
1119 | } | |
1120 | ||
15723a1d LP |
1121 | if (s->destroy_callback) |
1122 | s->destroy_callback(s->userdata); | |
ab93297c | 1123 | |
356779df | 1124 | free(s->description); |
75db809a | 1125 | return mfree(s); |
fd38203a | 1126 | } |
8c75fe17 | 1127 | DEFINE_TRIVIAL_CLEANUP_FUNC(sd_event_source*, source_free); |
fd38203a LP |
1128 | |
1129 | static int source_set_pending(sd_event_source *s, bool b) { | |
1130 | int r; | |
1131 | ||
1132 | assert(s); | |
6203e07a | 1133 | assert(s->type != SOURCE_EXIT); |
fd38203a LP |
1134 | |
1135 | if (s->pending == b) | |
1136 | return 0; | |
1137 | ||
1138 | s->pending = b; | |
1139 | ||
1140 | if (b) { | |
1141 | s->pending_iteration = s->event->iteration; | |
1142 | ||
1143 | r = prioq_put(s->event->pending, s, &s->pending_index); | |
1144 | if (r < 0) { | |
1145 | s->pending = false; | |
1146 | return r; | |
1147 | } | |
1148 | } else | |
1149 | assert_se(prioq_remove(s->event->pending, s, &s->pending_index)); | |
1150 | ||
e1951c16 MS |
1151 | if (EVENT_SOURCE_IS_TIME(s->type)) |
1152 | event_source_time_prioq_reshuffle(s); | |
2576a19e | 1153 | |
9da4cb2b LP |
1154 | if (s->type == SOURCE_SIGNAL && !b) { |
1155 | struct signal_data *d; | |
1156 | ||
1157 | d = hashmap_get(s->event->signal_data, &s->priority); | |
1158 | if (d && d->current == s) | |
1159 | d->current = NULL; | |
1160 | } | |
1161 | ||
97ef5391 LP |
1162 | if (s->type == SOURCE_INOTIFY) { |
1163 | ||
1164 | assert(s->inotify.inode_data); | |
1165 | assert(s->inotify.inode_data->inotify_data); | |
1166 | ||
1167 | if (b) | |
1168 | s->inotify.inode_data->inotify_data->n_pending ++; | |
1169 | else { | |
1170 | assert(s->inotify.inode_data->inotify_data->n_pending > 0); | |
1171 | s->inotify.inode_data->inotify_data->n_pending --; | |
1172 | } | |
1173 | } | |
1174 | ||
efd3be9d | 1175 | return 1; |
fd38203a LP |
1176 | } |
1177 | ||
a71fe8b8 | 1178 | static sd_event_source *source_new(sd_event *e, bool floating, EventSourceType type) { |
a38cf9fb LP |
1179 | |
1180 | /* Let's allocate exactly what we need. Note that the difference of the smallest event source | |
1181 | * structure to the largest is 144 bytes on x86-64 at the time of writing, i.e. more than two cache | |
1182 | * lines. */ | |
1183 | static const size_t size_table[_SOURCE_EVENT_SOURCE_TYPE_MAX] = { | |
1184 | [SOURCE_IO] = endoffsetof_field(sd_event_source, io), | |
1185 | [SOURCE_TIME_REALTIME] = endoffsetof_field(sd_event_source, time), | |
1186 | [SOURCE_TIME_BOOTTIME] = endoffsetof_field(sd_event_source, time), | |
1187 | [SOURCE_TIME_MONOTONIC] = endoffsetof_field(sd_event_source, time), | |
1188 | [SOURCE_TIME_REALTIME_ALARM] = endoffsetof_field(sd_event_source, time), | |
1189 | [SOURCE_TIME_BOOTTIME_ALARM] = endoffsetof_field(sd_event_source, time), | |
1190 | [SOURCE_SIGNAL] = endoffsetof_field(sd_event_source, signal), | |
1191 | [SOURCE_CHILD] = endoffsetof_field(sd_event_source, child), | |
1192 | [SOURCE_DEFER] = endoffsetof_field(sd_event_source, defer), | |
1193 | [SOURCE_POST] = endoffsetof_field(sd_event_source, post), | |
1194 | [SOURCE_EXIT] = endoffsetof_field(sd_event_source, exit), | |
1195 | [SOURCE_INOTIFY] = endoffsetof_field(sd_event_source, inotify), | |
158fe190 | 1196 | [SOURCE_MEMORY_PRESSURE] = endoffsetof_field(sd_event_source, memory_pressure), |
a38cf9fb LP |
1197 | }; |
1198 | ||
fd38203a LP |
1199 | sd_event_source *s; |
1200 | ||
1201 | assert(e); | |
a38cf9fb LP |
1202 | assert(type >= 0); |
1203 | assert(type < _SOURCE_EVENT_SOURCE_TYPE_MAX); | |
1204 | assert(size_table[type] > 0); | |
fd38203a | 1205 | |
a38cf9fb LP |
1206 | /* We use expand_to_usable() here to tell gcc that it should consider this an object of the full |
1207 | * size, even if we only allocate the initial part we need. */ | |
1208 | s = expand_to_usable(malloc0(size_table[type]), sizeof(sd_event_source)); | |
fd38203a LP |
1209 | if (!s) |
1210 | return NULL; | |
1211 | ||
a38cf9fb LP |
1212 | /* Note: we cannot use compound initialization here, because sizeof(sd_event_source) is likely larger |
1213 | * than what we allocated here. */ | |
1214 | s->n_ref = 1; | |
1215 | s->event = e; | |
1216 | s->floating = floating; | |
1217 | s->type = type; | |
1218 | s->pending_index = PRIOQ_IDX_NULL; | |
1219 | s->prepare_index = PRIOQ_IDX_NULL; | |
a71fe8b8 LP |
1220 | |
1221 | if (!floating) | |
1222 | sd_event_ref(e); | |
fd38203a | 1223 | |
a71fe8b8 | 1224 | LIST_PREPEND(sources, e->sources, s); |
313cefa1 | 1225 | e->n_sources++; |
15b38f93 | 1226 | |
fd38203a LP |
1227 | return s; |
1228 | } | |
1229 | ||
b9350e70 LP |
1230 | static int io_exit_callback(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
1231 | assert(s); | |
1232 | ||
1233 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1234 | } | |
1235 | ||
f7262a9f | 1236 | _public_ int sd_event_add_io( |
fd38203a | 1237 | sd_event *e, |
151b9b96 | 1238 | sd_event_source **ret, |
fd38203a LP |
1239 | int fd, |
1240 | uint32_t events, | |
718db961 | 1241 | sd_event_io_handler_t callback, |
151b9b96 | 1242 | void *userdata) { |
fd38203a | 1243 | |
ec766a51 | 1244 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1245 | int r; |
1246 | ||
305f78bf | 1247 | assert_return(e, -EINVAL); |
b937d761 | 1248 | assert_return(e = event_resolve(e), -ENOPKG); |
8ac43fee | 1249 | assert_return(fd >= 0, -EBADF); |
2a16a986 | 1250 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 1251 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 1252 | assert_return(!event_origin_changed(e), -ECHILD); |
fd38203a | 1253 | |
b9350e70 LP |
1254 | if (!callback) |
1255 | callback = io_exit_callback; | |
1256 | ||
a71fe8b8 | 1257 | s = source_new(e, !ret, SOURCE_IO); |
fd38203a LP |
1258 | if (!s) |
1259 | return -ENOMEM; | |
1260 | ||
9da4cb2b | 1261 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1262 | s->io.fd = fd; |
1263 | s->io.events = events; | |
1264 | s->io.callback = callback; | |
1265 | s->userdata = userdata; | |
baf76283 | 1266 | s->enabled = SD_EVENT_ON; |
fd38203a | 1267 | |
baf76283 | 1268 | r = source_io_register(s, s->enabled, events); |
ec766a51 | 1269 | if (r < 0) |
050f74f2 | 1270 | return r; |
fd38203a | 1271 | |
a71fe8b8 LP |
1272 | if (ret) |
1273 | *ret = s; | |
ec766a51 | 1274 | TAKE_PTR(s); |
a71fe8b8 | 1275 | |
fd38203a LP |
1276 | return 0; |
1277 | } | |
1278 | ||
52444dc4 | 1279 | static void initialize_perturb(sd_event *e) { |
6d2326e0 | 1280 | sd_id128_t id = {}; |
52444dc4 | 1281 | |
6d2326e0 YW |
1282 | /* When we sleep for longer, we try to realign the wakeup to the same time within each |
1283 | * minute/second/250ms, so that events all across the system can be coalesced into a single CPU | |
1284 | * wakeup. However, let's take some system-specific randomness for this value, so that in a network | |
1285 | * of systems with synced clocks timer events are distributed a bit. Here, we calculate a | |
1286 | * perturbation usec offset from the boot ID (or machine ID if failed, e.g. /proc is not mounted). */ | |
52444dc4 | 1287 | |
3a43da28 | 1288 | if (_likely_(e->perturb != USEC_INFINITY)) |
52444dc4 LP |
1289 | return; |
1290 | ||
1912f790 | 1291 | if (sd_id128_get_boot(&id) >= 0 || sd_id128_get_machine(&id) >= 0) |
6d2326e0 YW |
1292 | e->perturb = (id.qwords[0] ^ id.qwords[1]) % USEC_PER_MINUTE; |
1293 | else | |
1294 | e->perturb = 0; /* This is a super early process without /proc and /etc ?? */ | |
52444dc4 LP |
1295 | } |
1296 | ||
fd38203a LP |
1297 | static int event_setup_timer_fd( |
1298 | sd_event *e, | |
6a0f1f6d LP |
1299 | struct clock_data *d, |
1300 | clockid_t clock) { | |
fd38203a | 1301 | |
fd38203a | 1302 | assert(e); |
6a0f1f6d | 1303 | assert(d); |
fd38203a | 1304 | |
6a0f1f6d | 1305 | if (_likely_(d->fd >= 0)) |
fd38203a LP |
1306 | return 0; |
1307 | ||
254d1313 | 1308 | _cleanup_close_ int fd = -EBADF; |
b44d87e2 | 1309 | |
6a0f1f6d | 1310 | fd = timerfd_create(clock, TFD_NONBLOCK|TFD_CLOEXEC); |
fd38203a LP |
1311 | if (fd < 0) |
1312 | return -errno; | |
1313 | ||
7fe2903c LP |
1314 | fd = fd_move_above_stdio(fd); |
1315 | ||
1eac7948 | 1316 | struct epoll_event ev = { |
a82f89aa LP |
1317 | .events = EPOLLIN, |
1318 | .data.ptr = d, | |
1319 | }; | |
fd38203a | 1320 | |
15c689d7 | 1321 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) |
fd38203a | 1322 | return -errno; |
fd38203a | 1323 | |
b44d87e2 | 1324 | d->fd = TAKE_FD(fd); |
fd38203a LP |
1325 | return 0; |
1326 | } | |
1327 | ||
c4f1aff2 TG |
1328 | static int time_exit_callback(sd_event_source *s, uint64_t usec, void *userdata) { |
1329 | assert(s); | |
1330 | ||
1331 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1332 | } | |
1333 | ||
41c63f36 LP |
1334 | static int setup_clock_data(sd_event *e, struct clock_data *d, clockid_t clock) { |
1335 | int r; | |
1336 | ||
1337 | assert(d); | |
1338 | ||
1339 | if (d->fd < 0) { | |
1340 | r = event_setup_timer_fd(e, d, clock); | |
1341 | if (r < 0) | |
1342 | return r; | |
1343 | } | |
1344 | ||
1345 | r = prioq_ensure_allocated(&d->earliest, earliest_time_prioq_compare); | |
1346 | if (r < 0) | |
1347 | return r; | |
1348 | ||
1349 | r = prioq_ensure_allocated(&d->latest, latest_time_prioq_compare); | |
1350 | if (r < 0) | |
1351 | return r; | |
1352 | ||
1353 | return 0; | |
1354 | } | |
1355 | ||
1e45e3fe LP |
1356 | static int event_source_time_prioq_put( |
1357 | sd_event_source *s, | |
1358 | struct clock_data *d) { | |
1359 | ||
1360 | int r; | |
1361 | ||
1362 | assert(s); | |
1363 | assert(d); | |
19947509 | 1364 | assert(EVENT_SOURCE_USES_TIME_PRIOQ(s->type)); |
1e45e3fe | 1365 | |
f41315fc | 1366 | r = prioq_put(d->earliest, s, &s->earliest_index); |
1e45e3fe LP |
1367 | if (r < 0) |
1368 | return r; | |
1369 | ||
f41315fc | 1370 | r = prioq_put(d->latest, s, &s->latest_index); |
1e45e3fe | 1371 | if (r < 0) { |
f41315fc LP |
1372 | assert_se(prioq_remove(d->earliest, s, &s->earliest_index) > 0); |
1373 | s->earliest_index = PRIOQ_IDX_NULL; | |
1e45e3fe LP |
1374 | return r; |
1375 | } | |
1376 | ||
1377 | d->needs_rearm = true; | |
1378 | return 0; | |
1379 | } | |
1380 | ||
6a0f1f6d | 1381 | _public_ int sd_event_add_time( |
fd38203a | 1382 | sd_event *e, |
151b9b96 | 1383 | sd_event_source **ret, |
6a0f1f6d | 1384 | clockid_t clock, |
fd38203a | 1385 | uint64_t usec, |
c2ba3ad6 | 1386 | uint64_t accuracy, |
718db961 | 1387 | sd_event_time_handler_t callback, |
151b9b96 | 1388 | void *userdata) { |
fd38203a | 1389 | |
6a0f1f6d | 1390 | EventSourceType type; |
ec766a51 | 1391 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6a0f1f6d | 1392 | struct clock_data *d; |
fd38203a LP |
1393 | int r; |
1394 | ||
305f78bf | 1395 | assert_return(e, -EINVAL); |
b937d761 | 1396 | assert_return(e = event_resolve(e), -ENOPKG); |
f5fbe71d | 1397 | assert_return(accuracy != UINT64_MAX, -EINVAL); |
da7e457c | 1398 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 1399 | assert_return(!event_origin_changed(e), -ECHILD); |
fd38203a | 1400 | |
e475d10c LP |
1401 | if (!clock_supported(clock)) /* Checks whether the kernel supports the clock */ |
1402 | return -EOPNOTSUPP; | |
1403 | ||
1404 | type = clock_to_event_source_type(clock); /* checks whether sd-event supports this clock */ | |
1405 | if (type < 0) | |
3411372e LP |
1406 | return -EOPNOTSUPP; |
1407 | ||
c4f1aff2 TG |
1408 | if (!callback) |
1409 | callback = time_exit_callback; | |
1410 | ||
1e45e3fe | 1411 | assert_se(d = event_get_clock_data(e, type)); |
c2ba3ad6 | 1412 | |
41c63f36 | 1413 | r = setup_clock_data(e, d, clock); |
c983e776 EV |
1414 | if (r < 0) |
1415 | return r; | |
fd38203a | 1416 | |
a71fe8b8 | 1417 | s = source_new(e, !ret, type); |
fd38203a LP |
1418 | if (!s) |
1419 | return -ENOMEM; | |
1420 | ||
1421 | s->time.next = usec; | |
c2ba3ad6 | 1422 | s->time.accuracy = accuracy == 0 ? DEFAULT_ACCURACY_USEC : accuracy; |
fd38203a | 1423 | s->time.callback = callback; |
f41315fc | 1424 | s->earliest_index = s->latest_index = PRIOQ_IDX_NULL; |
fd38203a | 1425 | s->userdata = userdata; |
baf76283 | 1426 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1427 | |
1e45e3fe | 1428 | r = event_source_time_prioq_put(s, d); |
c2ba3ad6 | 1429 | if (r < 0) |
ec766a51 | 1430 | return r; |
fd38203a | 1431 | |
a71fe8b8 LP |
1432 | if (ret) |
1433 | *ret = s; | |
ec766a51 | 1434 | TAKE_PTR(s); |
a71fe8b8 | 1435 | |
fd38203a LP |
1436 | return 0; |
1437 | } | |
1438 | ||
d6a83dc4 LP |
1439 | _public_ int sd_event_add_time_relative( |
1440 | sd_event *e, | |
1441 | sd_event_source **ret, | |
1442 | clockid_t clock, | |
1443 | uint64_t usec, | |
1444 | uint64_t accuracy, | |
1445 | sd_event_time_handler_t callback, | |
1446 | void *userdata) { | |
1447 | ||
1448 | usec_t t; | |
1449 | int r; | |
1450 | ||
1451 | /* Same as sd_event_add_time() but operates relative to the event loop's current point in time, and | |
1452 | * checks for overflow. */ | |
1453 | ||
1454 | r = sd_event_now(e, clock, &t); | |
1455 | if (r < 0) | |
1456 | return r; | |
1457 | ||
1458 | if (usec >= USEC_INFINITY - t) | |
1459 | return -EOVERFLOW; | |
1460 | ||
1461 | return sd_event_add_time(e, ret, clock, t + usec, accuracy, callback, userdata); | |
1462 | } | |
1463 | ||
59bc1fd7 LP |
1464 | static int signal_exit_callback(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) { |
1465 | assert(s); | |
1466 | ||
1467 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1468 | } | |
1469 | ||
f7262a9f | 1470 | _public_ int sd_event_add_signal( |
305f78bf | 1471 | sd_event *e, |
151b9b96 | 1472 | sd_event_source **ret, |
305f78bf | 1473 | int sig, |
718db961 | 1474 | sd_event_signal_handler_t callback, |
151b9b96 | 1475 | void *userdata) { |
305f78bf | 1476 | |
ec766a51 | 1477 | _cleanup_(source_freep) sd_event_source *s = NULL; |
9da4cb2b | 1478 | struct signal_data *d; |
897448bd LP |
1479 | sigset_t new_ss; |
1480 | bool block_it; | |
fd38203a LP |
1481 | int r; |
1482 | ||
305f78bf | 1483 | assert_return(e, -EINVAL); |
b937d761 | 1484 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1485 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 1486 | assert_return(!event_origin_changed(e), -ECHILD); |
fd38203a | 1487 | |
897448bd LP |
1488 | /* Let's make sure our special flag stays outside of the valid signal range */ |
1489 | assert_cc(_NSIG < SD_EVENT_SIGNAL_PROCMASK); | |
1490 | ||
1491 | if (sig & SD_EVENT_SIGNAL_PROCMASK) { | |
1492 | sig &= ~SD_EVENT_SIGNAL_PROCMASK; | |
1493 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
1494 | ||
1495 | block_it = true; | |
1496 | } else { | |
1497 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
1498 | ||
1499 | r = signal_is_blocked(sig); | |
1500 | if (r < 0) | |
1501 | return r; | |
1502 | if (r == 0) | |
1503 | return -EBUSY; | |
1504 | ||
1505 | block_it = false; | |
1506 | } | |
1507 | ||
59bc1fd7 LP |
1508 | if (!callback) |
1509 | callback = signal_exit_callback; | |
1510 | ||
fd38203a LP |
1511 | if (!e->signal_sources) { |
1512 | e->signal_sources = new0(sd_event_source*, _NSIG); | |
1513 | if (!e->signal_sources) | |
1514 | return -ENOMEM; | |
1515 | } else if (e->signal_sources[sig]) | |
1516 | return -EBUSY; | |
1517 | ||
a71fe8b8 | 1518 | s = source_new(e, !ret, SOURCE_SIGNAL); |
fd38203a LP |
1519 | if (!s) |
1520 | return -ENOMEM; | |
1521 | ||
1522 | s->signal.sig = sig; | |
1523 | s->signal.callback = callback; | |
1524 | s->userdata = userdata; | |
baf76283 | 1525 | s->enabled = SD_EVENT_ON; |
fd38203a LP |
1526 | |
1527 | e->signal_sources[sig] = s; | |
fd38203a | 1528 | |
897448bd LP |
1529 | if (block_it) { |
1530 | sigset_t old_ss; | |
1531 | ||
1532 | if (sigemptyset(&new_ss) < 0) | |
1533 | return -errno; | |
1534 | ||
1535 | if (sigaddset(&new_ss, sig) < 0) | |
1536 | return -errno; | |
1537 | ||
1538 | r = pthread_sigmask(SIG_BLOCK, &new_ss, &old_ss); | |
1539 | if (r != 0) | |
1540 | return -r; | |
1541 | ||
1542 | r = sigismember(&old_ss, sig); | |
1543 | if (r < 0) | |
1544 | return -errno; | |
1545 | ||
1546 | s->signal.unblock = !r; | |
1547 | } else | |
1548 | s->signal.unblock = false; | |
1549 | ||
9da4cb2b | 1550 | r = event_make_signal_data(e, sig, &d); |
897448bd LP |
1551 | if (r < 0) { |
1552 | if (s->signal.unblock) | |
1553 | (void) pthread_sigmask(SIG_UNBLOCK, &new_ss, NULL); | |
1554 | ||
9da4cb2b | 1555 | return r; |
897448bd | 1556 | } |
fd38203a | 1557 | |
f1f00dbb LP |
1558 | /* Use the signal name as description for the event source by default */ |
1559 | (void) sd_event_source_set_description(s, signal_to_string(sig)); | |
1560 | ||
a71fe8b8 LP |
1561 | if (ret) |
1562 | *ret = s; | |
ec766a51 | 1563 | TAKE_PTR(s); |
a71fe8b8 | 1564 | |
fd38203a LP |
1565 | return 0; |
1566 | } | |
1567 | ||
b9350e70 LP |
1568 | static int child_exit_callback(sd_event_source *s, const siginfo_t *si, void *userdata) { |
1569 | assert(s); | |
1570 | ||
1571 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1572 | } | |
1573 | ||
f8f3f926 LP |
1574 | static bool shall_use_pidfd(void) { |
1575 | /* Mostly relevant for debugging, i.e. this is used in test-event.c to test the event loop once with and once without pidfd */ | |
1576 | return getenv_bool_secure("SYSTEMD_PIDFD") != 0; | |
1577 | } | |
1578 | ||
f7262a9f | 1579 | _public_ int sd_event_add_child( |
305f78bf | 1580 | sd_event *e, |
151b9b96 | 1581 | sd_event_source **ret, |
305f78bf LP |
1582 | pid_t pid, |
1583 | int options, | |
718db961 | 1584 | sd_event_child_handler_t callback, |
151b9b96 | 1585 | void *userdata) { |
305f78bf | 1586 | |
ec766a51 | 1587 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1588 | int r; |
1589 | ||
305f78bf | 1590 | assert_return(e, -EINVAL); |
b937d761 | 1591 | assert_return(e = event_resolve(e), -ENOPKG); |
305f78bf LP |
1592 | assert_return(pid > 1, -EINVAL); |
1593 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1594 | assert_return(options != 0, -EINVAL); | |
da7e457c | 1595 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 1596 | assert_return(!event_origin_changed(e), -ECHILD); |
fd38203a | 1597 | |
b9350e70 LP |
1598 | if (!callback) |
1599 | callback = child_exit_callback; | |
1600 | ||
b6d5481b | 1601 | if (e->n_online_child_sources == 0) { |
ee880b37 LP |
1602 | /* Caller must block SIGCHLD before using us to watch children, even if pidfd is available, |
1603 | * for compatibility with pre-pidfd and because we don't want the reap the child processes | |
1604 | * ourselves, i.e. call waitid(), and don't want Linux' default internal logic for that to | |
1605 | * take effect. | |
1606 | * | |
1607 | * (As an optimization we only do this check on the first child event source created.) */ | |
1608 | r = signal_is_blocked(SIGCHLD); | |
1609 | if (r < 0) | |
1610 | return r; | |
1611 | if (r == 0) | |
1612 | return -EBUSY; | |
1613 | } | |
1614 | ||
d5099efc | 1615 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
fd38203a LP |
1616 | if (r < 0) |
1617 | return r; | |
1618 | ||
4a0b58c4 | 1619 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) |
fd38203a LP |
1620 | return -EBUSY; |
1621 | ||
a71fe8b8 | 1622 | s = source_new(e, !ret, SOURCE_CHILD); |
fd38203a LP |
1623 | if (!s) |
1624 | return -ENOMEM; | |
1625 | ||
f8f3f926 | 1626 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1627 | s->child.options = options; |
1628 | s->child.callback = callback; | |
1629 | s->userdata = userdata; | |
baf76283 | 1630 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1631 | |
f8f3f926 LP |
1632 | /* We always take a pidfd here if we can, even if we wait for anything else than WEXITED, so that we |
1633 | * pin the PID, and make regular waitid() handling race-free. */ | |
1634 | ||
1635 | if (shall_use_pidfd()) { | |
54988a27 | 1636 | s->child.pidfd = pidfd_open(pid, 0); |
f8f3f926 LP |
1637 | if (s->child.pidfd < 0) { |
1638 | /* Propagate errors unless the syscall is not supported or blocked */ | |
1639 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
1640 | return -errno; | |
1641 | } else | |
1642 | s->child.pidfd_owned = true; /* If we allocate the pidfd we own it by default */ | |
1643 | } else | |
254d1313 | 1644 | s->child.pidfd = -EBADF; |
f8f3f926 | 1645 | |
f8f3f926 LP |
1646 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1647 | /* We have a pidfd and we only want to watch for exit */ | |
f8f3f926 | 1648 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1649 | if (r < 0) |
f8f3f926 | 1650 | return r; |
ac9f2640 | 1651 | |
f8f3f926 LP |
1652 | } else { |
1653 | /* We have no pidfd or we shall wait for some other event than WEXITED */ | |
f8f3f926 | 1654 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1655 | if (r < 0) |
f8f3f926 | 1656 | return r; |
f8f3f926 LP |
1657 | |
1658 | e->need_process_child = true; | |
1659 | } | |
c2ba3ad6 | 1660 | |
54988a27 YW |
1661 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); |
1662 | if (r < 0) | |
1663 | return r; | |
1664 | ||
1665 | /* These must be done after everything succeeds. */ | |
1666 | s->child.pid = pid; | |
b6d5481b | 1667 | e->n_online_child_sources++; |
ac9f2640 | 1668 | |
a71fe8b8 LP |
1669 | if (ret) |
1670 | *ret = s; | |
ec766a51 | 1671 | TAKE_PTR(s); |
f8f3f926 LP |
1672 | return 0; |
1673 | } | |
1674 | ||
1675 | _public_ int sd_event_add_child_pidfd( | |
1676 | sd_event *e, | |
1677 | sd_event_source **ret, | |
1678 | int pidfd, | |
1679 | int options, | |
1680 | sd_event_child_handler_t callback, | |
1681 | void *userdata) { | |
1682 | ||
1683 | ||
1684 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
1685 | pid_t pid; | |
1686 | int r; | |
1687 | ||
1688 | assert_return(e, -EINVAL); | |
1689 | assert_return(e = event_resolve(e), -ENOPKG); | |
1690 | assert_return(pidfd >= 0, -EBADF); | |
1691 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1692 | assert_return(options != 0, -EINVAL); | |
f8f3f926 | 1693 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 1694 | assert_return(!event_origin_changed(e), -ECHILD); |
f8f3f926 | 1695 | |
b9350e70 LP |
1696 | if (!callback) |
1697 | callback = child_exit_callback; | |
1698 | ||
b6d5481b | 1699 | if (e->n_online_child_sources == 0) { |
ee880b37 LP |
1700 | r = signal_is_blocked(SIGCHLD); |
1701 | if (r < 0) | |
1702 | return r; | |
1703 | if (r == 0) | |
1704 | return -EBUSY; | |
1705 | } | |
1706 | ||
f8f3f926 LP |
1707 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
1708 | if (r < 0) | |
1709 | return r; | |
1710 | ||
1711 | r = pidfd_get_pid(pidfd, &pid); | |
1712 | if (r < 0) | |
1713 | return r; | |
1714 | ||
1715 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) | |
1716 | return -EBUSY; | |
1717 | ||
1718 | s = source_new(e, !ret, SOURCE_CHILD); | |
1719 | if (!s) | |
1720 | return -ENOMEM; | |
1721 | ||
1722 | s->wakeup = WAKEUP_EVENT_SOURCE; | |
1723 | s->child.pidfd = pidfd; | |
1724 | s->child.pid = pid; | |
1725 | s->child.options = options; | |
1726 | s->child.callback = callback; | |
1727 | 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) */ | |
1728 | s->userdata = userdata; | |
1729 | s->enabled = SD_EVENT_ONESHOT; | |
1730 | ||
1731 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); | |
1732 | if (r < 0) | |
1733 | return r; | |
1734 | ||
f8f3f926 LP |
1735 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1736 | /* We only want to watch for WEXITED */ | |
f8f3f926 | 1737 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1738 | if (r < 0) |
f8f3f926 | 1739 | return r; |
f8f3f926 LP |
1740 | } else { |
1741 | /* We shall wait for some other event than WEXITED */ | |
f8f3f926 | 1742 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1743 | if (r < 0) |
f8f3f926 | 1744 | return r; |
a71fe8b8 | 1745 | |
f8f3f926 LP |
1746 | e->need_process_child = true; |
1747 | } | |
1748 | ||
b6d5481b | 1749 | e->n_online_child_sources++; |
ac9f2640 | 1750 | |
f8f3f926 LP |
1751 | if (ret) |
1752 | *ret = s; | |
f8f3f926 | 1753 | TAKE_PTR(s); |
fd38203a LP |
1754 | return 0; |
1755 | } | |
1756 | ||
b9350e70 LP |
1757 | static int generic_exit_callback(sd_event_source *s, void *userdata) { |
1758 | assert(s); | |
1759 | ||
1760 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1761 | } | |
1762 | ||
f7262a9f | 1763 | _public_ int sd_event_add_defer( |
305f78bf | 1764 | sd_event *e, |
151b9b96 | 1765 | sd_event_source **ret, |
718db961 | 1766 | sd_event_handler_t callback, |
151b9b96 | 1767 | void *userdata) { |
305f78bf | 1768 | |
ec766a51 | 1769 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1770 | int r; |
1771 | ||
305f78bf | 1772 | assert_return(e, -EINVAL); |
b937d761 | 1773 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1774 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 1775 | assert_return(!event_origin_changed(e), -ECHILD); |
fd38203a | 1776 | |
b9350e70 LP |
1777 | if (!callback) |
1778 | callback = generic_exit_callback; | |
1779 | ||
a71fe8b8 | 1780 | s = source_new(e, !ret, SOURCE_DEFER); |
fd38203a LP |
1781 | if (!s) |
1782 | return -ENOMEM; | |
1783 | ||
1784 | s->defer.callback = callback; | |
1785 | s->userdata = userdata; | |
baf76283 | 1786 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a LP |
1787 | |
1788 | r = source_set_pending(s, true); | |
ec766a51 | 1789 | if (r < 0) |
fd38203a | 1790 | return r; |
fd38203a | 1791 | |
a71fe8b8 LP |
1792 | if (ret) |
1793 | *ret = s; | |
ec766a51 | 1794 | TAKE_PTR(s); |
a71fe8b8 | 1795 | |
fd38203a LP |
1796 | return 0; |
1797 | } | |
1798 | ||
6e9feda3 LP |
1799 | _public_ int sd_event_add_post( |
1800 | sd_event *e, | |
1801 | sd_event_source **ret, | |
1802 | sd_event_handler_t callback, | |
1803 | void *userdata) { | |
1804 | ||
ec766a51 | 1805 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6e9feda3 LP |
1806 | int r; |
1807 | ||
1808 | assert_return(e, -EINVAL); | |
b937d761 | 1809 | assert_return(e = event_resolve(e), -ENOPKG); |
6e9feda3 | 1810 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 1811 | assert_return(!event_origin_changed(e), -ECHILD); |
6e9feda3 | 1812 | |
b9350e70 LP |
1813 | if (!callback) |
1814 | callback = generic_exit_callback; | |
1815 | ||
a71fe8b8 | 1816 | s = source_new(e, !ret, SOURCE_POST); |
6e9feda3 LP |
1817 | if (!s) |
1818 | return -ENOMEM; | |
1819 | ||
1820 | s->post.callback = callback; | |
1821 | s->userdata = userdata; | |
1822 | s->enabled = SD_EVENT_ON; | |
1823 | ||
de7fef4b | 1824 | r = set_ensure_put(&e->post_sources, NULL, s); |
ec766a51 | 1825 | if (r < 0) |
6e9feda3 | 1826 | return r; |
de7fef4b | 1827 | assert(r > 0); |
6e9feda3 | 1828 | |
a71fe8b8 LP |
1829 | if (ret) |
1830 | *ret = s; | |
ec766a51 | 1831 | TAKE_PTR(s); |
a71fe8b8 | 1832 | |
6e9feda3 LP |
1833 | return 0; |
1834 | } | |
1835 | ||
6203e07a | 1836 | _public_ int sd_event_add_exit( |
305f78bf | 1837 | sd_event *e, |
151b9b96 | 1838 | sd_event_source **ret, |
718db961 | 1839 | sd_event_handler_t callback, |
151b9b96 | 1840 | void *userdata) { |
305f78bf | 1841 | |
ec766a51 | 1842 | _cleanup_(source_freep) sd_event_source *s = NULL; |
da7e457c LP |
1843 | int r; |
1844 | ||
1845 | assert_return(e, -EINVAL); | |
b937d761 | 1846 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
1847 | assert_return(callback, -EINVAL); |
1848 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2eeff0f4 | 1849 | assert_return(!event_origin_changed(e), -ECHILD); |
da7e457c | 1850 | |
c983e776 EV |
1851 | r = prioq_ensure_allocated(&e->exit, exit_prioq_compare); |
1852 | if (r < 0) | |
1853 | return r; | |
da7e457c | 1854 | |
a71fe8b8 | 1855 | s = source_new(e, !ret, SOURCE_EXIT); |
fd38203a | 1856 | if (!s) |
da7e457c | 1857 | return -ENOMEM; |
fd38203a | 1858 | |
6203e07a | 1859 | s->exit.callback = callback; |
da7e457c | 1860 | s->userdata = userdata; |
6203e07a | 1861 | s->exit.prioq_index = PRIOQ_IDX_NULL; |
baf76283 | 1862 | s->enabled = SD_EVENT_ONESHOT; |
da7e457c | 1863 | |
6203e07a | 1864 | r = prioq_put(s->event->exit, s, &s->exit.prioq_index); |
ec766a51 | 1865 | if (r < 0) |
da7e457c | 1866 | return r; |
da7e457c | 1867 | |
a71fe8b8 LP |
1868 | if (ret) |
1869 | *ret = s; | |
ec766a51 | 1870 | TAKE_PTR(s); |
a71fe8b8 | 1871 | |
da7e457c LP |
1872 | return 0; |
1873 | } | |
1874 | ||
9857de4f | 1875 | _public_ int sd_event_trim_memory(void) { |
158fe190 LP |
1876 | int r; |
1877 | ||
1878 | /* A default implementation of a memory pressure callback. Simply releases our own allocation caches | |
1879 | * and glibc's. This is automatically used when people call sd_event_add_memory_pressure() with a | |
1880 | * NULL callback parameter. */ | |
1881 | ||
1882 | log_debug("Memory pressure event, trimming malloc() memory."); | |
1883 | ||
1884 | #if HAVE_GENERIC_MALLINFO | |
1885 | generic_mallinfo before_mallinfo = generic_mallinfo_get(); | |
1886 | #endif | |
1887 | ||
1888 | usec_t before_timestamp = now(CLOCK_MONOTONIC); | |
1889 | hashmap_trim_pools(); | |
1890 | r = malloc_trim(0); | |
1891 | usec_t after_timestamp = now(CLOCK_MONOTONIC); | |
1892 | ||
1893 | if (r > 0) | |
1894 | log_debug("Successfully trimmed some memory."); | |
1895 | else | |
1896 | log_debug("Couldn't trim any memory."); | |
1897 | ||
1898 | usec_t period = after_timestamp - before_timestamp; | |
1899 | ||
1900 | #if HAVE_GENERIC_MALLINFO | |
1901 | generic_mallinfo after_mallinfo = generic_mallinfo_get(); | |
1902 | size_t l = LESS_BY((size_t) before_mallinfo.hblkhd, (size_t) after_mallinfo.hblkhd) + | |
1903 | LESS_BY((size_t) before_mallinfo.arena, (size_t) after_mallinfo.arena); | |
1904 | log_struct(LOG_DEBUG, | |
1905 | LOG_MESSAGE("Memory trimming took %s, returned %s to OS.", | |
1906 | FORMAT_TIMESPAN(period, 0), | |
1907 | FORMAT_BYTES(l)), | |
1908 | "MESSAGE_ID=" SD_MESSAGE_MEMORY_TRIM_STR, | |
1909 | "TRIMMED_BYTES=%zu", l, | |
1910 | "TRIMMED_USEC=" USEC_FMT, period); | |
1911 | #else | |
1912 | log_struct(LOG_DEBUG, | |
1913 | LOG_MESSAGE("Memory trimming took %s.", | |
1914 | FORMAT_TIMESPAN(period, 0)), | |
1915 | "MESSAGE_ID=" SD_MESSAGE_MEMORY_TRIM_STR, | |
1916 | "TRIMMED_USEC=" USEC_FMT, period); | |
1917 | #endif | |
1918 | ||
1919 | return 0; | |
1920 | } | |
1921 | ||
1922 | static int memory_pressure_callback(sd_event_source *s, void *userdata) { | |
1923 | assert(s); | |
1924 | ||
1925 | sd_event_trim_memory(); | |
1926 | return 0; | |
1927 | } | |
1928 | ||
1929 | _public_ int sd_event_add_memory_pressure( | |
1930 | sd_event *e, | |
1931 | sd_event_source **ret, | |
1932 | sd_event_handler_t callback, | |
1933 | void *userdata) { | |
1934 | ||
1935 | _cleanup_free_ char *w = NULL; | |
1936 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
92651a7a | 1937 | _cleanup_close_ int path_fd = -EBADF, fd = -EBADF; |
158fe190 | 1938 | _cleanup_free_ void *write_buffer = NULL; |
40c5d5d2 | 1939 | const char *watch, *watch_fallback = NULL, *env; |
158fe190 LP |
1940 | size_t write_buffer_size = 0; |
1941 | struct stat st; | |
1942 | uint32_t events; | |
1943 | bool locked; | |
1944 | int r; | |
1945 | ||
1946 | assert_return(e, -EINVAL); | |
1947 | assert_return(e = event_resolve(e), -ENOPKG); | |
1948 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2eeff0f4 | 1949 | assert_return(!event_origin_changed(e), -ECHILD); |
158fe190 LP |
1950 | |
1951 | if (!callback) | |
1952 | callback = memory_pressure_callback; | |
1953 | ||
1954 | s = source_new(e, !ret, SOURCE_MEMORY_PRESSURE); | |
1955 | if (!s) | |
1956 | return -ENOMEM; | |
1957 | ||
1958 | s->wakeup = WAKEUP_EVENT_SOURCE; | |
1959 | s->memory_pressure.callback = callback; | |
1960 | s->userdata = userdata; | |
1961 | s->enabled = SD_EVENT_ON; | |
1962 | s->memory_pressure.fd = -EBADF; | |
1963 | ||
1964 | env = secure_getenv("MEMORY_PRESSURE_WATCH"); | |
1965 | if (env) { | |
1966 | if (isempty(env) || path_equal(env, "/dev/null")) | |
1967 | return log_debug_errno(SYNTHETIC_ERRNO(EHOSTDOWN), | |
1968 | "Memory pressure logic is explicitly disabled via $MEMORY_PRESSURE_WATCH."); | |
1969 | ||
1970 | if (!path_is_absolute(env) || !path_is_normalized(env)) | |
1971 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
1972 | "$MEMORY_PRESSURE_WATCH set to invalid path: %s", env); | |
1973 | ||
1974 | watch = env; | |
1975 | ||
1976 | env = secure_getenv("MEMORY_PRESSURE_WRITE"); | |
1977 | if (env) { | |
1978 | r = unbase64mem(env, SIZE_MAX, &write_buffer, &write_buffer_size); | |
1979 | if (r < 0) | |
1980 | return r; | |
1981 | } | |
1982 | ||
1983 | locked = true; | |
1984 | } else { | |
1985 | ||
1986 | r = is_pressure_supported(); | |
1987 | if (r < 0) | |
1988 | return r; | |
1989 | if (r == 0) | |
1990 | return -EOPNOTSUPP; | |
1991 | ||
1992 | /* By default we want to watch memory pressure on the local cgroup, but we'll fall back on | |
1993 | * the system wide pressure if for some reason we cannot (which could be: memory controller | |
1994 | * not delegated to us, or PSI simply not available in the kernel). On legacy cgroupv1 we'll | |
1995 | * only use the system-wide logic. */ | |
1996 | r = cg_all_unified(); | |
1997 | if (r < 0) | |
1998 | return r; | |
1999 | if (r == 0) | |
2000 | watch = "/proc/pressure/memory"; | |
2001 | else { | |
2002 | _cleanup_free_ char *cg = NULL; | |
2003 | ||
2004 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &cg); | |
2005 | if (r < 0) | |
2006 | return r; | |
2007 | ||
2008 | w = path_join("/sys/fs/cgroup", cg, "memory.pressure"); | |
2009 | if (!w) | |
2010 | return -ENOMEM; | |
2011 | ||
2012 | watch = w; | |
2013 | watch_fallback = "/proc/pressure/memory"; | |
2014 | } | |
2015 | ||
2016 | /* Android uses three levels in its userspace low memory killer logic: | |
2017 | * some 70000 1000000 | |
2018 | * some 100000 1000000 | |
2019 | * full 70000 1000000 | |
2020 | * | |
2021 | * GNOME's low memory monitor uses: | |
2022 | * some 70000 1000000 | |
2023 | * some 100000 1000000 | |
2024 | * full 100000 1000000 | |
2025 | * | |
a6170074 LP |
2026 | * We'll default to the middle level that both agree on. Except we do it on a 2s window |
2027 | * (i.e. 200ms per 2s, rather than 100ms per 1s), because that's the window duration the | |
2028 | * kernel will allow us to do unprivileged, also in the future. */ | |
158fe190 LP |
2029 | if (asprintf((char**) &write_buffer, |
2030 | "%s " USEC_FMT " " USEC_FMT, | |
2031 | MEMORY_PRESSURE_DEFAULT_TYPE, | |
2032 | MEMORY_PRESSURE_DEFAULT_THRESHOLD_USEC, | |
2033 | MEMORY_PRESSURE_DEFAULT_WINDOW_USEC) < 0) | |
2034 | return -ENOMEM; | |
2035 | ||
2036 | write_buffer_size = strlen(write_buffer) + 1; | |
2037 | locked = false; | |
2038 | } | |
2039 | ||
2040 | path_fd = open(watch, O_PATH|O_CLOEXEC); | |
2041 | if (path_fd < 0) { | |
2042 | if (errno != ENOENT) | |
2043 | return -errno; | |
2044 | ||
2045 | /* We got ENOENT. Three options now: try the fallback if we have one, or return the error as | |
2046 | * is (if based on user/env config), or return -EOPNOTSUPP (because we picked the path, and | |
2047 | * the PSI service apparently is not supported) */ | |
2048 | if (!watch_fallback) | |
2049 | return locked ? -ENOENT : -EOPNOTSUPP; | |
2050 | ||
2051 | path_fd = open(watch_fallback, O_PATH|O_CLOEXEC); | |
63b1e67e YW |
2052 | if (path_fd < 0) { |
2053 | if (errno == ENOENT) /* PSI is not available in the kernel even under the fallback path? */ | |
2054 | return -EOPNOTSUPP; | |
158fe190 | 2055 | return -errno; |
63b1e67e | 2056 | } |
158fe190 LP |
2057 | } |
2058 | ||
2059 | if (fstat(path_fd, &st) < 0) | |
2060 | return -errno; | |
2061 | ||
2062 | if (S_ISSOCK(st.st_mode)) { | |
2063 | fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); | |
2064 | if (fd < 0) | |
2065 | return -errno; | |
2066 | ||
2067 | r = connect_unix_path(fd, path_fd, NULL); | |
2068 | if (r < 0) | |
2069 | return r; | |
2070 | ||
2071 | events = EPOLLIN; | |
2072 | ||
2073 | } else if (S_ISREG(st.st_mode) || S_ISFIFO(st.st_mode) || S_ISCHR(st.st_mode)) { | |
2074 | fd = fd_reopen(path_fd, (write_buffer_size > 0 ? O_RDWR : O_RDONLY) |O_CLOEXEC|O_NONBLOCK|O_NOCTTY); | |
2075 | if (fd < 0) | |
2076 | return fd; | |
2077 | ||
2078 | if (S_ISREG(st.st_mode)) { | |
2079 | struct statfs sfs; | |
2080 | ||
2081 | /* If this is a regular file validate this is a procfs or cgroupfs file, where we look for EPOLLPRI */ | |
2082 | ||
2083 | if (fstatfs(fd, &sfs) < 0) | |
2084 | return -errno; | |
2085 | ||
2086 | if (!is_fs_type(&sfs, PROC_SUPER_MAGIC) && | |
2087 | !is_fs_type(&sfs, CGROUP2_SUPER_MAGIC)) | |
2088 | return -ENOTTY; | |
2089 | ||
2090 | events = EPOLLPRI; | |
2091 | } else | |
2092 | /* For fifos and char devices just watch for EPOLLIN */ | |
2093 | events = EPOLLIN; | |
2094 | ||
2095 | } else if (S_ISDIR(st.st_mode)) | |
2096 | return -EISDIR; | |
2097 | else | |
2098 | return -EBADF; | |
2099 | ||
2100 | s->memory_pressure.fd = TAKE_FD(fd); | |
2101 | s->memory_pressure.write_buffer = TAKE_PTR(write_buffer); | |
2102 | s->memory_pressure.write_buffer_size = write_buffer_size; | |
2103 | s->memory_pressure.events = events; | |
2104 | s->memory_pressure.locked = locked; | |
2105 | ||
2106 | /* So here's the thing: if we are talking to PSI we need to write the watch string before adding the | |
2107 | * fd to epoll (if we ignore this, then the watch won't work). Hence we'll not actually register the | |
2108 | * fd with the epoll right-away. Instead, we just add the event source to a list of memory pressure | |
2109 | * event sources on which writes must be executed before the first event loop iteration is | |
2110 | * executed. (We could also write the data here, right away, but we want to give the caller the | |
2111 | * freedom to call sd_event_source_set_memory_pressure_type() and | |
2112 | * sd_event_source_set_memory_pressure_rate() before we write it. */ | |
2113 | ||
2114 | if (s->memory_pressure.write_buffer_size > 0) | |
2115 | source_memory_pressure_add_to_write_list(s); | |
2116 | else { | |
2117 | r = source_memory_pressure_register(s, s->enabled); | |
2118 | if (r < 0) | |
2119 | return r; | |
2120 | } | |
2121 | ||
2122 | if (ret) | |
2123 | *ret = s; | |
2124 | TAKE_PTR(s); | |
2125 | ||
2126 | return 0; | |
2127 | } | |
2128 | ||
97ef5391 LP |
2129 | static void event_free_inotify_data(sd_event *e, struct inotify_data *d) { |
2130 | assert(e); | |
2131 | ||
2132 | if (!d) | |
2133 | return; | |
2134 | ||
2135 | assert(hashmap_isempty(d->inodes)); | |
2136 | assert(hashmap_isempty(d->wd)); | |
2137 | ||
2138 | if (d->buffer_filled > 0) | |
0601b958 | 2139 | LIST_REMOVE(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
2140 | |
2141 | hashmap_free(d->inodes); | |
2142 | hashmap_free(d->wd); | |
2143 | ||
2144 | assert_se(hashmap_remove(e->inotify_data, &d->priority) == d); | |
2145 | ||
2146 | if (d->fd >= 0) { | |
2eeff0f4 | 2147 | if (!event_origin_changed(e) && |
fbae5090 | 2148 | epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, d->fd, NULL) < 0) |
97ef5391 LP |
2149 | log_debug_errno(errno, "Failed to remove inotify fd from epoll, ignoring: %m"); |
2150 | ||
2151 | safe_close(d->fd); | |
2152 | } | |
2153 | free(d); | |
2154 | } | |
2155 | ||
2156 | static int event_make_inotify_data( | |
2157 | sd_event *e, | |
2158 | int64_t priority, | |
2159 | struct inotify_data **ret) { | |
2160 | ||
254d1313 | 2161 | _cleanup_close_ int fd = -EBADF; |
97ef5391 | 2162 | struct inotify_data *d; |
97ef5391 LP |
2163 | int r; |
2164 | ||
2165 | assert(e); | |
2166 | ||
2167 | d = hashmap_get(e->inotify_data, &priority); | |
2168 | if (d) { | |
2169 | if (ret) | |
2170 | *ret = d; | |
2171 | return 0; | |
2172 | } | |
2173 | ||
2174 | fd = inotify_init1(IN_NONBLOCK|O_CLOEXEC); | |
2175 | if (fd < 0) | |
2176 | return -errno; | |
2177 | ||
2178 | fd = fd_move_above_stdio(fd); | |
2179 | ||
97ef5391 LP |
2180 | d = new(struct inotify_data, 1); |
2181 | if (!d) | |
2182 | return -ENOMEM; | |
2183 | ||
2184 | *d = (struct inotify_data) { | |
2185 | .wakeup = WAKEUP_INOTIFY_DATA, | |
2186 | .fd = TAKE_FD(fd), | |
2187 | .priority = priority, | |
2188 | }; | |
2189 | ||
c2484a75 | 2190 | r = hashmap_ensure_put(&e->inotify_data, &uint64_hash_ops, &d->priority, d); |
97ef5391 LP |
2191 | if (r < 0) { |
2192 | d->fd = safe_close(d->fd); | |
2193 | free(d); | |
2194 | return r; | |
2195 | } | |
2196 | ||
1eac7948 | 2197 | struct epoll_event ev = { |
97ef5391 LP |
2198 | .events = EPOLLIN, |
2199 | .data.ptr = d, | |
2200 | }; | |
2201 | ||
2202 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { | |
2203 | r = -errno; | |
2204 | d->fd = safe_close(d->fd); /* let's close this ourselves, as event_free_inotify_data() would otherwise | |
2205 | * remove the fd from the epoll first, which we don't want as we couldn't | |
2206 | * add it in the first place. */ | |
2207 | event_free_inotify_data(e, d); | |
2208 | return r; | |
2209 | } | |
2210 | ||
2211 | if (ret) | |
2212 | *ret = d; | |
2213 | ||
2214 | return 1; | |
2215 | } | |
2216 | ||
7a08d314 | 2217 | static int inode_data_compare(const struct inode_data *x, const struct inode_data *y) { |
90c88092 | 2218 | int r; |
97ef5391 LP |
2219 | |
2220 | assert(x); | |
2221 | assert(y); | |
2222 | ||
90c88092 YW |
2223 | r = CMP(x->dev, y->dev); |
2224 | if (r != 0) | |
2225 | return r; | |
97ef5391 | 2226 | |
6dd91b36 | 2227 | return CMP(x->ino, y->ino); |
97ef5391 LP |
2228 | } |
2229 | ||
7a08d314 YW |
2230 | static void inode_data_hash_func(const struct inode_data *d, struct siphash *state) { |
2231 | assert(d); | |
97ef5391 LP |
2232 | |
2233 | siphash24_compress(&d->dev, sizeof(d->dev), state); | |
2234 | siphash24_compress(&d->ino, sizeof(d->ino), state); | |
2235 | } | |
2236 | ||
7a08d314 | 2237 | DEFINE_PRIVATE_HASH_OPS(inode_data_hash_ops, struct inode_data, inode_data_hash_func, inode_data_compare); |
97ef5391 LP |
2238 | |
2239 | static void event_free_inode_data( | |
2240 | sd_event *e, | |
2241 | struct inode_data *d) { | |
2242 | ||
2243 | assert(e); | |
2244 | ||
2245 | if (!d) | |
2246 | return; | |
2247 | ||
64903d18 | 2248 | assert(!d->event_sources); |
97ef5391 LP |
2249 | |
2250 | if (d->fd >= 0) { | |
ed828563 | 2251 | LIST_REMOVE(to_close, e->inode_data_to_close_list, d); |
97ef5391 LP |
2252 | safe_close(d->fd); |
2253 | } | |
2254 | ||
2255 | if (d->inotify_data) { | |
2256 | ||
2257 | if (d->wd >= 0) { | |
2eeff0f4 | 2258 | if (d->inotify_data->fd >= 0 && !event_origin_changed(e)) { |
97ef5391 LP |
2259 | /* So here's a problem. At the time this runs the watch descriptor might already be |
2260 | * invalidated, because an IN_IGNORED event might be queued right the moment we enter | |
2261 | * the syscall. Hence, whenever we get EINVAL, ignore it entirely, since it's a very | |
2262 | * likely case to happen. */ | |
2263 | ||
2264 | if (inotify_rm_watch(d->inotify_data->fd, d->wd) < 0 && errno != EINVAL) | |
2265 | log_debug_errno(errno, "Failed to remove watch descriptor %i from inotify, ignoring: %m", d->wd); | |
2266 | } | |
2267 | ||
2268 | assert_se(hashmap_remove(d->inotify_data->wd, INT_TO_PTR(d->wd)) == d); | |
2269 | } | |
2270 | ||
2271 | assert_se(hashmap_remove(d->inotify_data->inodes, d) == d); | |
2272 | } | |
2273 | ||
2274 | free(d); | |
2275 | } | |
2276 | ||
53baf2ef LP |
2277 | static void event_gc_inotify_data( |
2278 | sd_event *e, | |
2279 | struct inotify_data *d) { | |
2280 | ||
2281 | assert(e); | |
2282 | ||
2283 | /* GCs the inotify data object if we don't need it anymore. That's the case if we don't want to watch | |
2284 | * any inode with it anymore, which in turn happens if no event source of this priority is interested | |
2285 | * in any inode any longer. That said, we maintain an extra busy counter: if non-zero we'll delay GC | |
2286 | * (under the expectation that the GC is called again once the counter is decremented). */ | |
2287 | ||
2288 | if (!d) | |
2289 | return; | |
2290 | ||
2291 | if (!hashmap_isempty(d->inodes)) | |
2292 | return; | |
2293 | ||
2294 | if (d->n_busy > 0) | |
2295 | return; | |
2296 | ||
2297 | event_free_inotify_data(e, d); | |
2298 | } | |
2299 | ||
97ef5391 LP |
2300 | static void event_gc_inode_data( |
2301 | sd_event *e, | |
2302 | struct inode_data *d) { | |
2303 | ||
2304 | struct inotify_data *inotify_data; | |
2305 | ||
2306 | assert(e); | |
2307 | ||
2308 | if (!d) | |
2309 | return; | |
2310 | ||
64903d18 | 2311 | if (d->event_sources) |
97ef5391 LP |
2312 | return; |
2313 | ||
2314 | inotify_data = d->inotify_data; | |
2315 | event_free_inode_data(e, d); | |
2316 | ||
53baf2ef | 2317 | event_gc_inotify_data(e, inotify_data); |
97ef5391 LP |
2318 | } |
2319 | ||
2320 | static int event_make_inode_data( | |
2321 | sd_event *e, | |
2322 | struct inotify_data *inotify_data, | |
2323 | dev_t dev, | |
2324 | ino_t ino, | |
2325 | struct inode_data **ret) { | |
2326 | ||
2327 | struct inode_data *d, key; | |
2328 | int r; | |
2329 | ||
2330 | assert(e); | |
2331 | assert(inotify_data); | |
2332 | ||
2333 | key = (struct inode_data) { | |
2334 | .ino = ino, | |
2335 | .dev = dev, | |
2336 | }; | |
2337 | ||
2338 | d = hashmap_get(inotify_data->inodes, &key); | |
2339 | if (d) { | |
2340 | if (ret) | |
2341 | *ret = d; | |
2342 | ||
2343 | return 0; | |
2344 | } | |
2345 | ||
2346 | r = hashmap_ensure_allocated(&inotify_data->inodes, &inode_data_hash_ops); | |
2347 | if (r < 0) | |
2348 | return r; | |
2349 | ||
2350 | d = new(struct inode_data, 1); | |
2351 | if (!d) | |
2352 | return -ENOMEM; | |
2353 | ||
2354 | *d = (struct inode_data) { | |
2355 | .dev = dev, | |
2356 | .ino = ino, | |
2357 | .wd = -1, | |
254d1313 | 2358 | .fd = -EBADF, |
97ef5391 LP |
2359 | .inotify_data = inotify_data, |
2360 | }; | |
2361 | ||
2362 | r = hashmap_put(inotify_data->inodes, d, d); | |
2363 | if (r < 0) { | |
2364 | free(d); | |
2365 | return r; | |
2366 | } | |
2367 | ||
2368 | if (ret) | |
2369 | *ret = d; | |
2370 | ||
2371 | return 1; | |
2372 | } | |
2373 | ||
2374 | static uint32_t inode_data_determine_mask(struct inode_data *d) { | |
2375 | bool excl_unlink = true; | |
2376 | uint32_t combined = 0; | |
97ef5391 LP |
2377 | |
2378 | assert(d); | |
2379 | ||
2380 | /* Combines the watch masks of all event sources watching this inode. We generally just OR them together, but | |
2381 | * the IN_EXCL_UNLINK flag is ANDed instead. | |
2382 | * | |
2383 | * Note that we add all sources to the mask here, regardless whether enabled, disabled or oneshot. That's | |
2384 | * because we cannot change the mask anymore after the event source was created once, since the kernel has no | |
f21f31b2 | 2385 | * API for that. Hence we need to subscribe to the maximum mask we ever might be interested in, and suppress |
97ef5391 LP |
2386 | * events we don't care for client-side. */ |
2387 | ||
2388 | LIST_FOREACH(inotify.by_inode_data, s, d->event_sources) { | |
2389 | ||
2390 | if ((s->inotify.mask & IN_EXCL_UNLINK) == 0) | |
2391 | excl_unlink = false; | |
2392 | ||
2393 | combined |= s->inotify.mask; | |
2394 | } | |
2395 | ||
2396 | return (combined & ~(IN_ONESHOT|IN_DONT_FOLLOW|IN_ONLYDIR|IN_EXCL_UNLINK)) | (excl_unlink ? IN_EXCL_UNLINK : 0); | |
2397 | } | |
2398 | ||
2399 | static int inode_data_realize_watch(sd_event *e, struct inode_data *d) { | |
2400 | uint32_t combined_mask; | |
2401 | int wd, r; | |
2402 | ||
2403 | assert(d); | |
2404 | assert(d->fd >= 0); | |
2405 | ||
2406 | combined_mask = inode_data_determine_mask(d); | |
2407 | ||
2408 | if (d->wd >= 0 && combined_mask == d->combined_mask) | |
2409 | return 0; | |
2410 | ||
2411 | r = hashmap_ensure_allocated(&d->inotify_data->wd, NULL); | |
2412 | if (r < 0) | |
2413 | return r; | |
2414 | ||
2415 | wd = inotify_add_watch_fd(d->inotify_data->fd, d->fd, combined_mask); | |
2416 | if (wd < 0) | |
2417 | return -errno; | |
2418 | ||
2419 | if (d->wd < 0) { | |
2420 | r = hashmap_put(d->inotify_data->wd, INT_TO_PTR(wd), d); | |
2421 | if (r < 0) { | |
2422 | (void) inotify_rm_watch(d->inotify_data->fd, wd); | |
2423 | return r; | |
2424 | } | |
2425 | ||
2426 | d->wd = wd; | |
2427 | ||
2428 | } else if (d->wd != wd) { | |
2429 | ||
2430 | log_debug("Weird, the watch descriptor we already knew for this inode changed?"); | |
2431 | (void) inotify_rm_watch(d->fd, wd); | |
2432 | return -EINVAL; | |
2433 | } | |
2434 | ||
2435 | d->combined_mask = combined_mask; | |
2436 | return 1; | |
2437 | } | |
2438 | ||
b9350e70 LP |
2439 | static int inotify_exit_callback(sd_event_source *s, const struct inotify_event *event, void *userdata) { |
2440 | assert(s); | |
2441 | ||
2442 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
2443 | } | |
2444 | ||
e67d738a | 2445 | static int event_add_inotify_fd_internal( |
97ef5391 LP |
2446 | sd_event *e, |
2447 | sd_event_source **ret, | |
e67d738a LP |
2448 | int fd, |
2449 | bool donate, | |
97ef5391 LP |
2450 | uint32_t mask, |
2451 | sd_event_inotify_handler_t callback, | |
2452 | void *userdata) { | |
2453 | ||
5bb1d7fb | 2454 | _cleanup_close_ int donated_fd = donate ? fd : -EBADF; |
e67d738a | 2455 | _cleanup_(source_freep) sd_event_source *s = NULL; |
97ef5391 LP |
2456 | struct inotify_data *inotify_data = NULL; |
2457 | struct inode_data *inode_data = NULL; | |
97ef5391 LP |
2458 | struct stat st; |
2459 | int r; | |
2460 | ||
2461 | assert_return(e, -EINVAL); | |
2462 | assert_return(e = event_resolve(e), -ENOPKG); | |
e67d738a | 2463 | assert_return(fd >= 0, -EBADF); |
97ef5391 | 2464 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 2465 | assert_return(!event_origin_changed(e), -ECHILD); |
97ef5391 | 2466 | |
b9350e70 LP |
2467 | if (!callback) |
2468 | callback = inotify_exit_callback; | |
2469 | ||
97ef5391 LP |
2470 | /* Refuse IN_MASK_ADD since we coalesce watches on the same inode, and hence really don't want to merge |
2471 | * masks. Or in other words, this whole code exists only to manage IN_MASK_ADD type operations for you, hence | |
2472 | * the user can't use them for us. */ | |
2473 | if (mask & IN_MASK_ADD) | |
2474 | return -EINVAL; | |
2475 | ||
97ef5391 LP |
2476 | if (fstat(fd, &st) < 0) |
2477 | return -errno; | |
2478 | ||
2479 | s = source_new(e, !ret, SOURCE_INOTIFY); | |
2480 | if (!s) | |
2481 | return -ENOMEM; | |
2482 | ||
2483 | s->enabled = mask & IN_ONESHOT ? SD_EVENT_ONESHOT : SD_EVENT_ON; | |
2484 | s->inotify.mask = mask; | |
2485 | s->inotify.callback = callback; | |
2486 | s->userdata = userdata; | |
2487 | ||
2488 | /* Allocate an inotify object for this priority, and an inode object within it */ | |
2489 | r = event_make_inotify_data(e, SD_EVENT_PRIORITY_NORMAL, &inotify_data); | |
2490 | if (r < 0) | |
8c75fe17 | 2491 | return r; |
97ef5391 LP |
2492 | |
2493 | r = event_make_inode_data(e, inotify_data, st.st_dev, st.st_ino, &inode_data); | |
8c75fe17 | 2494 | if (r < 0) { |
e67d738a | 2495 | event_gc_inotify_data(e, inotify_data); |
8c75fe17 ZJS |
2496 | return r; |
2497 | } | |
97ef5391 LP |
2498 | |
2499 | /* Keep the O_PATH fd around until the first iteration of the loop, so that we can still change the priority of | |
2500 | * the event source, until then, for which we need the original inode. */ | |
2501 | if (inode_data->fd < 0) { | |
e67d738a LP |
2502 | if (donated_fd >= 0) |
2503 | inode_data->fd = TAKE_FD(donated_fd); | |
2504 | else { | |
2505 | inode_data->fd = fcntl(fd, F_DUPFD_CLOEXEC, 3); | |
2506 | if (inode_data->fd < 0) { | |
2507 | r = -errno; | |
2508 | event_gc_inode_data(e, inode_data); | |
2509 | return r; | |
2510 | } | |
2511 | } | |
2512 | ||
ed828563 | 2513 | LIST_PREPEND(to_close, e->inode_data_to_close_list, inode_data); |
97ef5391 LP |
2514 | } |
2515 | ||
2516 | /* Link our event source to the inode data object */ | |
2517 | LIST_PREPEND(inotify.by_inode_data, inode_data->event_sources, s); | |
2518 | s->inotify.inode_data = inode_data; | |
2519 | ||
97ef5391 LP |
2520 | /* Actually realize the watch now */ |
2521 | r = inode_data_realize_watch(e, inode_data); | |
2522 | if (r < 0) | |
8c75fe17 | 2523 | return r; |
97ef5391 | 2524 | |
97ef5391 LP |
2525 | if (ret) |
2526 | *ret = s; | |
8c75fe17 | 2527 | TAKE_PTR(s); |
97ef5391 LP |
2528 | |
2529 | return 0; | |
97ef5391 LP |
2530 | } |
2531 | ||
e67d738a LP |
2532 | _public_ int sd_event_add_inotify_fd( |
2533 | sd_event *e, | |
2534 | sd_event_source **ret, | |
2535 | int fd, | |
2536 | uint32_t mask, | |
2537 | sd_event_inotify_handler_t callback, | |
2538 | void *userdata) { | |
2539 | ||
2540 | return event_add_inotify_fd_internal(e, ret, fd, /* donate= */ false, mask, callback, userdata); | |
2541 | } | |
2542 | ||
2543 | _public_ int sd_event_add_inotify( | |
2544 | sd_event *e, | |
2545 | sd_event_source **ret, | |
2546 | const char *path, | |
2547 | uint32_t mask, | |
2548 | sd_event_inotify_handler_t callback, | |
2549 | void *userdata) { | |
2550 | ||
2091c779 | 2551 | sd_event_source *s = NULL; /* avoid false maybe-uninitialized warning */ |
e67d738a LP |
2552 | int fd, r; |
2553 | ||
2554 | assert_return(path, -EINVAL); | |
2555 | ||
586c8cee ZJS |
2556 | fd = open(path, O_PATH | O_CLOEXEC | |
2557 | (mask & IN_ONLYDIR ? O_DIRECTORY : 0) | | |
2558 | (mask & IN_DONT_FOLLOW ? O_NOFOLLOW : 0)); | |
e67d738a LP |
2559 | if (fd < 0) |
2560 | return -errno; | |
2561 | ||
2562 | r = event_add_inotify_fd_internal(e, &s, fd, /* donate= */ true, mask, callback, userdata); | |
2563 | if (r < 0) | |
2564 | return r; | |
2565 | ||
2566 | (void) sd_event_source_set_description(s, path); | |
2567 | ||
2568 | if (ret) | |
2569 | *ret = s; | |
2570 | ||
2571 | return r; | |
2572 | } | |
2573 | ||
8301aa0b | 2574 | static sd_event_source* event_source_free(sd_event_source *s) { |
6680dd6b LP |
2575 | if (!s) |
2576 | return NULL; | |
da7e457c | 2577 | |
8301aa0b YW |
2578 | /* Here's a special hack: when we are called from a |
2579 | * dispatch handler we won't free the event source | |
2580 | * immediately, but we will detach the fd from the | |
2581 | * epoll. This way it is safe for the caller to unref | |
2582 | * the event source and immediately close the fd, but | |
2583 | * we still retain a valid event source object after | |
2584 | * the callback. */ | |
fd38203a | 2585 | |
76d04c3a | 2586 | if (s->dispatching) |
8301aa0b | 2587 | source_disconnect(s); |
76d04c3a | 2588 | else |
8301aa0b | 2589 | source_free(s); |
fd38203a LP |
2590 | |
2591 | return NULL; | |
2592 | } | |
2593 | ||
8301aa0b YW |
2594 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event_source, sd_event_source, event_source_free); |
2595 | ||
356779df | 2596 | _public_ int sd_event_source_set_description(sd_event_source *s, const char *description) { |
f7f53e9e | 2597 | assert_return(s, -EINVAL); |
2eeff0f4 | 2598 | assert_return(!event_origin_changed(s->event), -ECHILD); |
f7f53e9e | 2599 | |
356779df | 2600 | return free_and_strdup(&s->description, description); |
f7f53e9e TG |
2601 | } |
2602 | ||
356779df | 2603 | _public_ int sd_event_source_get_description(sd_event_source *s, const char **description) { |
f7f53e9e | 2604 | assert_return(s, -EINVAL); |
356779df | 2605 | assert_return(description, -EINVAL); |
f7f53e9e | 2606 | |
7d92a1a4 ZJS |
2607 | if (!s->description) |
2608 | return -ENXIO; | |
2609 | ||
356779df | 2610 | *description = s->description; |
f7f53e9e TG |
2611 | return 0; |
2612 | } | |
2613 | ||
adcc4ca3 | 2614 | _public_ sd_event *sd_event_source_get_event(sd_event_source *s) { |
305f78bf | 2615 | assert_return(s, NULL); |
2eeff0f4 | 2616 | assert_return(!event_origin_changed(s->event), NULL); |
eaa3cbef LP |
2617 | |
2618 | return s->event; | |
2619 | } | |
2620 | ||
f7262a9f | 2621 | _public_ int sd_event_source_get_pending(sd_event_source *s) { |
305f78bf | 2622 | assert_return(s, -EINVAL); |
6203e07a | 2623 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c | 2624 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 2625 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
2626 | |
2627 | return s->pending; | |
2628 | } | |
2629 | ||
f7262a9f | 2630 | _public_ int sd_event_source_get_io_fd(sd_event_source *s) { |
305f78bf LP |
2631 | assert_return(s, -EINVAL); |
2632 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2eeff0f4 | 2633 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
2634 | |
2635 | return s->io.fd; | |
2636 | } | |
2637 | ||
30caf8f3 LP |
2638 | _public_ int sd_event_source_set_io_fd(sd_event_source *s, int fd) { |
2639 | int r; | |
2640 | ||
2641 | assert_return(s, -EINVAL); | |
8ac43fee | 2642 | assert_return(fd >= 0, -EBADF); |
30caf8f3 | 2643 | assert_return(s->type == SOURCE_IO, -EDOM); |
2eeff0f4 | 2644 | assert_return(!event_origin_changed(s->event), -ECHILD); |
30caf8f3 LP |
2645 | |
2646 | if (s->io.fd == fd) | |
2647 | return 0; | |
2648 | ||
b6d5481b | 2649 | if (event_source_is_offline(s)) { |
30caf8f3 LP |
2650 | s->io.fd = fd; |
2651 | s->io.registered = false; | |
2652 | } else { | |
2653 | int saved_fd; | |
2654 | ||
2655 | saved_fd = s->io.fd; | |
2656 | assert(s->io.registered); | |
2657 | ||
2658 | s->io.fd = fd; | |
2659 | s->io.registered = false; | |
2660 | ||
2661 | r = source_io_register(s, s->enabled, s->io.events); | |
2662 | if (r < 0) { | |
2663 | s->io.fd = saved_fd; | |
2664 | s->io.registered = true; | |
2665 | return r; | |
2666 | } | |
2667 | ||
5a795bff | 2668 | (void) epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, saved_fd, NULL); |
30caf8f3 LP |
2669 | } |
2670 | ||
2671 | return 0; | |
2672 | } | |
2673 | ||
ab93297c NM |
2674 | _public_ int sd_event_source_get_io_fd_own(sd_event_source *s) { |
2675 | assert_return(s, -EINVAL); | |
2676 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2eeff0f4 | 2677 | assert_return(!event_origin_changed(s->event), -ECHILD); |
ab93297c NM |
2678 | |
2679 | return s->io.owned; | |
2680 | } | |
2681 | ||
2682 | _public_ int sd_event_source_set_io_fd_own(sd_event_source *s, int own) { | |
2683 | assert_return(s, -EINVAL); | |
2684 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2eeff0f4 | 2685 | assert_return(!event_origin_changed(s->event), -ECHILD); |
ab93297c NM |
2686 | |
2687 | s->io.owned = own; | |
2688 | return 0; | |
2689 | } | |
2690 | ||
f7262a9f | 2691 | _public_ int sd_event_source_get_io_events(sd_event_source *s, uint32_t* events) { |
305f78bf LP |
2692 | assert_return(s, -EINVAL); |
2693 | assert_return(events, -EINVAL); | |
2694 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2eeff0f4 | 2695 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
2696 | |
2697 | *events = s->io.events; | |
2698 | return 0; | |
2699 | } | |
2700 | ||
f7262a9f | 2701 | _public_ int sd_event_source_set_io_events(sd_event_source *s, uint32_t events) { |
fd38203a LP |
2702 | int r; |
2703 | ||
305f78bf LP |
2704 | assert_return(s, -EINVAL); |
2705 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2a16a986 | 2706 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 2707 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 2708 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a | 2709 | |
b63c8d4f DH |
2710 | /* edge-triggered updates are never skipped, so we can reset edges */ |
2711 | if (s->io.events == events && !(events & EPOLLET)) | |
fd38203a LP |
2712 | return 0; |
2713 | ||
2a0dc6cd LP |
2714 | r = source_set_pending(s, false); |
2715 | if (r < 0) | |
2716 | return r; | |
2717 | ||
b6d5481b | 2718 | if (event_source_is_online(s)) { |
e4715127 | 2719 | r = source_io_register(s, s->enabled, events); |
fd38203a LP |
2720 | if (r < 0) |
2721 | return r; | |
2722 | } | |
2723 | ||
2724 | s->io.events = events; | |
2725 | ||
2726 | return 0; | |
2727 | } | |
2728 | ||
f7262a9f | 2729 | _public_ int sd_event_source_get_io_revents(sd_event_source *s, uint32_t* revents) { |
305f78bf LP |
2730 | assert_return(s, -EINVAL); |
2731 | assert_return(revents, -EINVAL); | |
2732 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2733 | assert_return(s->pending, -ENODATA); | |
2eeff0f4 | 2734 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
2735 | |
2736 | *revents = s->io.revents; | |
2737 | return 0; | |
2738 | } | |
2739 | ||
f7262a9f | 2740 | _public_ int sd_event_source_get_signal(sd_event_source *s) { |
305f78bf LP |
2741 | assert_return(s, -EINVAL); |
2742 | assert_return(s->type == SOURCE_SIGNAL, -EDOM); | |
2eeff0f4 | 2743 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
2744 | |
2745 | return s->signal.sig; | |
2746 | } | |
2747 | ||
31927c16 | 2748 | _public_ int sd_event_source_get_priority(sd_event_source *s, int64_t *priority) { |
305f78bf | 2749 | assert_return(s, -EINVAL); |
2eeff0f4 | 2750 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a | 2751 | |
6680b8d1 ME |
2752 | *priority = s->priority; |
2753 | return 0; | |
fd38203a LP |
2754 | } |
2755 | ||
31927c16 | 2756 | _public_ int sd_event_source_set_priority(sd_event_source *s, int64_t priority) { |
97ef5391 LP |
2757 | bool rm_inotify = false, rm_inode = false; |
2758 | struct inotify_data *new_inotify_data = NULL; | |
2759 | struct inode_data *new_inode_data = NULL; | |
9da4cb2b LP |
2760 | int r; |
2761 | ||
305f78bf | 2762 | assert_return(s, -EINVAL); |
da7e457c | 2763 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 2764 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
2765 | |
2766 | if (s->priority == priority) | |
2767 | return 0; | |
2768 | ||
97ef5391 LP |
2769 | if (s->type == SOURCE_INOTIFY) { |
2770 | struct inode_data *old_inode_data; | |
2771 | ||
2772 | assert(s->inotify.inode_data); | |
2773 | old_inode_data = s->inotify.inode_data; | |
2774 | ||
2775 | /* We need the original fd to change the priority. If we don't have it we can't change the priority, | |
2776 | * anymore. Note that we close any fds when entering the next event loop iteration, i.e. for inotify | |
2777 | * events we allow priority changes only until the first following iteration. */ | |
2778 | if (old_inode_data->fd < 0) | |
2779 | return -EOPNOTSUPP; | |
2780 | ||
2781 | r = event_make_inotify_data(s->event, priority, &new_inotify_data); | |
2782 | if (r < 0) | |
2783 | return r; | |
2784 | rm_inotify = r > 0; | |
2785 | ||
2786 | r = event_make_inode_data(s->event, new_inotify_data, old_inode_data->dev, old_inode_data->ino, &new_inode_data); | |
2787 | if (r < 0) | |
2788 | goto fail; | |
2789 | rm_inode = r > 0; | |
2790 | ||
2791 | if (new_inode_data->fd < 0) { | |
2792 | /* Duplicate the fd for the new inode object if we don't have any yet */ | |
2793 | new_inode_data->fd = fcntl(old_inode_data->fd, F_DUPFD_CLOEXEC, 3); | |
2794 | if (new_inode_data->fd < 0) { | |
2795 | r = -errno; | |
2796 | goto fail; | |
2797 | } | |
2798 | ||
ed828563 | 2799 | LIST_PREPEND(to_close, s->event->inode_data_to_close_list, new_inode_data); |
97ef5391 LP |
2800 | } |
2801 | ||
2802 | /* Move the event source to the new inode data structure */ | |
2803 | LIST_REMOVE(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2804 | LIST_PREPEND(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2805 | s->inotify.inode_data = new_inode_data; | |
2806 | ||
2807 | /* Now create the new watch */ | |
2808 | r = inode_data_realize_watch(s->event, new_inode_data); | |
2809 | if (r < 0) { | |
2810 | /* Move it back */ | |
2811 | LIST_REMOVE(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2812 | LIST_PREPEND(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2813 | s->inotify.inode_data = old_inode_data; | |
2814 | goto fail; | |
2815 | } | |
2816 | ||
2817 | s->priority = priority; | |
2818 | ||
2819 | event_gc_inode_data(s->event, old_inode_data); | |
2820 | ||
b6d5481b | 2821 | } else if (s->type == SOURCE_SIGNAL && event_source_is_online(s)) { |
9da4cb2b LP |
2822 | struct signal_data *old, *d; |
2823 | ||
2824 | /* Move us from the signalfd belonging to the old | |
2825 | * priority to the signalfd of the new priority */ | |
2826 | ||
2827 | assert_se(old = hashmap_get(s->event->signal_data, &s->priority)); | |
2828 | ||
2829 | s->priority = priority; | |
2830 | ||
2831 | r = event_make_signal_data(s->event, s->signal.sig, &d); | |
2832 | if (r < 0) { | |
2833 | s->priority = old->priority; | |
2834 | return r; | |
2835 | } | |
2836 | ||
2837 | event_unmask_signal_data(s->event, old, s->signal.sig); | |
2838 | } else | |
2839 | s->priority = priority; | |
fd38203a | 2840 | |
e1951c16 | 2841 | event_source_pp_prioq_reshuffle(s); |
fd38203a | 2842 | |
6203e07a LP |
2843 | if (s->type == SOURCE_EXIT) |
2844 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
305f78bf | 2845 | |
fd38203a | 2846 | return 0; |
97ef5391 LP |
2847 | |
2848 | fail: | |
2849 | if (rm_inode) | |
2850 | event_free_inode_data(s->event, new_inode_data); | |
2851 | ||
2852 | if (rm_inotify) | |
2853 | event_free_inotify_data(s->event, new_inotify_data); | |
2854 | ||
2855 | return r; | |
fd38203a LP |
2856 | } |
2857 | ||
cad143a8 | 2858 | _public_ int sd_event_source_get_enabled(sd_event_source *s, int *ret) { |
71193c0b ZJS |
2859 | /* Quick mode: the event source doesn't exist and we only want to query boolean enablement state. */ |
2860 | if (!s && !ret) | |
2861 | return false; | |
2862 | ||
305f78bf | 2863 | assert_return(s, -EINVAL); |
2eeff0f4 | 2864 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a | 2865 | |
cad143a8 LP |
2866 | if (ret) |
2867 | *ret = s->enabled; | |
2868 | ||
08c1eb0e | 2869 | return s->enabled != SD_EVENT_OFF; |
fd38203a LP |
2870 | } |
2871 | ||
b6d5481b LP |
2872 | static int event_source_offline( |
2873 | sd_event_source *s, | |
2874 | int enabled, | |
2875 | bool ratelimited) { | |
2876 | ||
2877 | bool was_offline; | |
fd38203a LP |
2878 | int r; |
2879 | ||
ddfde737 | 2880 | assert(s); |
b6d5481b | 2881 | assert(enabled == SD_EVENT_OFF || ratelimited); |
fd38203a | 2882 | |
ddfde737 | 2883 | /* Unset the pending flag when this event source is disabled */ |
b6d5481b LP |
2884 | if (s->enabled != SD_EVENT_OFF && |
2885 | enabled == SD_EVENT_OFF && | |
2886 | !IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
ddfde737 LP |
2887 | r = source_set_pending(s, false); |
2888 | if (r < 0) | |
2889 | return r; | |
2890 | } | |
cc567911 | 2891 | |
b6d5481b LP |
2892 | was_offline = event_source_is_offline(s); |
2893 | s->enabled = enabled; | |
2894 | s->ratelimited = ratelimited; | |
fd38203a | 2895 | |
ddfde737 | 2896 | switch (s->type) { |
fd38203a | 2897 | |
ddfde737 LP |
2898 | case SOURCE_IO: |
2899 | source_io_unregister(s); | |
2900 | break; | |
ac989a78 | 2901 | |
ddfde737 LP |
2902 | case SOURCE_SIGNAL: |
2903 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); | |
2904 | break; | |
fd38203a | 2905 | |
ddfde737 | 2906 | case SOURCE_CHILD: |
b6d5481b LP |
2907 | if (!was_offline) { |
2908 | assert(s->event->n_online_child_sources > 0); | |
2909 | s->event->n_online_child_sources--; | |
2910 | } | |
fd38203a | 2911 | |
ddfde737 LP |
2912 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
2913 | source_child_pidfd_unregister(s); | |
2914 | else | |
2915 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
2916 | break; | |
4807d2d0 | 2917 | |
ddfde737 LP |
2918 | case SOURCE_EXIT: |
2919 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
2920 | break; | |
fd38203a | 2921 | |
158fe190 LP |
2922 | case SOURCE_MEMORY_PRESSURE: |
2923 | source_memory_pressure_unregister(s); | |
2924 | break; | |
2925 | ||
2115b9b6 YW |
2926 | case SOURCE_TIME_REALTIME: |
2927 | case SOURCE_TIME_BOOTTIME: | |
2928 | case SOURCE_TIME_MONOTONIC: | |
2929 | case SOURCE_TIME_REALTIME_ALARM: | |
2930 | case SOURCE_TIME_BOOTTIME_ALARM: | |
ddfde737 LP |
2931 | case SOURCE_DEFER: |
2932 | case SOURCE_POST: | |
2933 | case SOURCE_INOTIFY: | |
2934 | break; | |
fd38203a | 2935 | |
ddfde737 | 2936 | default: |
04499a70 | 2937 | assert_not_reached(); |
ddfde737 | 2938 | } |
fd38203a | 2939 | |
2115b9b6 YW |
2940 | /* Always reshuffle time prioq, as the ratelimited flag may be changed. */ |
2941 | event_source_time_prioq_reshuffle(s); | |
2942 | ||
b6d5481b | 2943 | return 1; |
ddfde737 | 2944 | } |
f8f3f926 | 2945 | |
b6d5481b LP |
2946 | static int event_source_online( |
2947 | sd_event_source *s, | |
2948 | int enabled, | |
2949 | bool ratelimited) { | |
2950 | ||
2951 | bool was_online; | |
ddfde737 | 2952 | int r; |
fd38203a | 2953 | |
ddfde737 | 2954 | assert(s); |
b6d5481b | 2955 | assert(enabled != SD_EVENT_OFF || !ratelimited); |
305f78bf | 2956 | |
ddfde737 | 2957 | /* Unset the pending flag when this event source is enabled */ |
b6d5481b LP |
2958 | if (s->enabled == SD_EVENT_OFF && |
2959 | enabled != SD_EVENT_OFF && | |
2960 | !IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
ddfde737 LP |
2961 | r = source_set_pending(s, false); |
2962 | if (r < 0) | |
2963 | return r; | |
2964 | } | |
9d3e3aa5 | 2965 | |
b6d5481b LP |
2966 | /* Are we really ready for onlining? */ |
2967 | if (enabled == SD_EVENT_OFF || ratelimited) { | |
2968 | /* Nope, we are not ready for onlining, then just update the precise state and exit */ | |
2969 | s->enabled = enabled; | |
2970 | s->ratelimited = ratelimited; | |
2971 | return 0; | |
2972 | } | |
2973 | ||
2974 | was_online = event_source_is_online(s); | |
2975 | ||
ddfde737 | 2976 | switch (s->type) { |
ddfde737 | 2977 | case SOURCE_IO: |
b6d5481b | 2978 | r = source_io_register(s, enabled, s->io.events); |
d2eafe61 | 2979 | if (r < 0) |
ddfde737 | 2980 | return r; |
ddfde737 | 2981 | break; |
fd38203a | 2982 | |
ddfde737 LP |
2983 | case SOURCE_SIGNAL: |
2984 | r = event_make_signal_data(s->event, s->signal.sig, NULL); | |
2985 | if (r < 0) { | |
ddfde737 LP |
2986 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
2987 | return r; | |
2988 | } | |
fd38203a | 2989 | |
ddfde737 | 2990 | break; |
fd38203a | 2991 | |
ddfde737 | 2992 | case SOURCE_CHILD: |
ddfde737 LP |
2993 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
2994 | /* yes, we have pidfd */ | |
9da4cb2b | 2995 | |
b6d5481b | 2996 | r = source_child_pidfd_register(s, enabled); |
ac9f2640 | 2997 | if (r < 0) |
9da4cb2b | 2998 | return r; |
ddfde737 LP |
2999 | } else { |
3000 | /* no pidfd, or something other to watch for than WEXITED */ | |
9da4cb2b | 3001 | |
ddfde737 LP |
3002 | r = event_make_signal_data(s->event, SIGCHLD, NULL); |
3003 | if (r < 0) { | |
ddfde737 LP |
3004 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); |
3005 | return r; | |
3006 | } | |
3007 | } | |
fd38203a | 3008 | |
b6d5481b LP |
3009 | if (!was_online) |
3010 | s->event->n_online_child_sources++; | |
ddfde737 | 3011 | break; |
4807d2d0 | 3012 | |
158fe190 LP |
3013 | case SOURCE_MEMORY_PRESSURE: |
3014 | r = source_memory_pressure_register(s, enabled); | |
3015 | if (r < 0) | |
3016 | return r; | |
3017 | ||
3018 | break; | |
3019 | ||
d2eafe61 ZJS |
3020 | case SOURCE_TIME_REALTIME: |
3021 | case SOURCE_TIME_BOOTTIME: | |
3022 | case SOURCE_TIME_MONOTONIC: | |
3023 | case SOURCE_TIME_REALTIME_ALARM: | |
3024 | case SOURCE_TIME_BOOTTIME_ALARM: | |
ddfde737 | 3025 | case SOURCE_EXIT: |
ddfde737 LP |
3026 | case SOURCE_DEFER: |
3027 | case SOURCE_POST: | |
3028 | case SOURCE_INOTIFY: | |
3029 | break; | |
9da4cb2b | 3030 | |
ddfde737 | 3031 | default: |
04499a70 | 3032 | assert_not_reached(); |
ddfde737 | 3033 | } |
f8f3f926 | 3034 | |
b6d5481b LP |
3035 | s->enabled = enabled; |
3036 | s->ratelimited = ratelimited; | |
d2eafe61 ZJS |
3037 | |
3038 | /* Non-failing operations below */ | |
2115b9b6 | 3039 | if (s->type == SOURCE_EXIT) |
d2eafe61 | 3040 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); |
d2eafe61 | 3041 | |
2115b9b6 YW |
3042 | /* Always reshuffle time prioq, as the ratelimited flag may be changed. */ |
3043 | event_source_time_prioq_reshuffle(s); | |
d2eafe61 | 3044 | |
b6d5481b | 3045 | return 1; |
ddfde737 LP |
3046 | } |
3047 | ||
3048 | _public_ int sd_event_source_set_enabled(sd_event_source *s, int m) { | |
3049 | int r; | |
9da4cb2b | 3050 | |
ddfde737 | 3051 | assert_return(IN_SET(m, SD_EVENT_OFF, SD_EVENT_ON, SD_EVENT_ONESHOT), -EINVAL); |
7e922b05 ZJS |
3052 | |
3053 | /* Quick mode: if the source doesn't exist, SD_EVENT_OFF is a noop. */ | |
3054 | if (m == SD_EVENT_OFF && !s) | |
3055 | return 0; | |
3056 | ||
3057 | assert_return(s, -EINVAL); | |
2eeff0f4 | 3058 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a | 3059 | |
ddfde737 LP |
3060 | /* If we are dead anyway, we are fine with turning off sources, but everything else needs to fail. */ |
3061 | if (s->event->state == SD_EVENT_FINISHED) | |
3062 | return m == SD_EVENT_OFF ? 0 : -ESTALE; | |
305f78bf | 3063 | |
ddfde737 LP |
3064 | if (s->enabled == m) /* No change? */ |
3065 | return 0; | |
9d3e3aa5 | 3066 | |
ddfde737 | 3067 | if (m == SD_EVENT_OFF) |
b6d5481b | 3068 | r = event_source_offline(s, m, s->ratelimited); |
ddfde737 LP |
3069 | else { |
3070 | if (s->enabled != SD_EVENT_OFF) { | |
3071 | /* Switching from "on" to "oneshot" or back? If that's the case, we can take a shortcut, the | |
3072 | * event source is already enabled after all. */ | |
3073 | s->enabled = m; | |
3074 | return 0; | |
fd38203a | 3075 | } |
ddfde737 | 3076 | |
b6d5481b | 3077 | r = event_source_online(s, m, s->ratelimited); |
fd38203a | 3078 | } |
ddfde737 LP |
3079 | if (r < 0) |
3080 | return r; | |
fd38203a | 3081 | |
e1951c16 | 3082 | event_source_pp_prioq_reshuffle(s); |
fd38203a LP |
3083 | return 0; |
3084 | } | |
3085 | ||
f7262a9f | 3086 | _public_ int sd_event_source_get_time(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
3087 | assert_return(s, -EINVAL); |
3088 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 3089 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
2eeff0f4 | 3090 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
3091 | |
3092 | *usec = s->time.next; | |
3093 | return 0; | |
3094 | } | |
3095 | ||
f7262a9f | 3096 | _public_ int sd_event_source_set_time(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 3097 | int r; |
6a0f1f6d | 3098 | |
305f78bf | 3099 | assert_return(s, -EINVAL); |
6a0f1f6d | 3100 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 3101 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 3102 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a | 3103 | |
2a0dc6cd LP |
3104 | r = source_set_pending(s, false); |
3105 | if (r < 0) | |
3106 | return r; | |
2576a19e | 3107 | |
2a0dc6cd | 3108 | s->time.next = usec; |
fd38203a | 3109 | |
e1951c16 | 3110 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
3111 | return 0; |
3112 | } | |
3113 | ||
d6a83dc4 LP |
3114 | _public_ int sd_event_source_set_time_relative(sd_event_source *s, uint64_t usec) { |
3115 | usec_t t; | |
3116 | int r; | |
3117 | ||
3118 | assert_return(s, -EINVAL); | |
3119 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2eeff0f4 | 3120 | assert_return(!event_origin_changed(s->event), -ECHILD); |
d6a83dc4 | 3121 | |
ef859195 LP |
3122 | if (usec == USEC_INFINITY) |
3123 | return sd_event_source_set_time(s, USEC_INFINITY); | |
3124 | ||
d6a83dc4 LP |
3125 | r = sd_event_now(s->event, event_source_type_to_clock(s->type), &t); |
3126 | if (r < 0) | |
3127 | return r; | |
3128 | ||
496db330 YW |
3129 | usec = usec_add(t, usec); |
3130 | if (usec == USEC_INFINITY) | |
d6a83dc4 LP |
3131 | return -EOVERFLOW; |
3132 | ||
496db330 | 3133 | return sd_event_source_set_time(s, usec); |
d6a83dc4 LP |
3134 | } |
3135 | ||
f7262a9f | 3136 | _public_ int sd_event_source_get_time_accuracy(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
3137 | assert_return(s, -EINVAL); |
3138 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 3139 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
2eeff0f4 | 3140 | assert_return(!event_origin_changed(s->event), -ECHILD); |
305f78bf LP |
3141 | |
3142 | *usec = s->time.accuracy; | |
3143 | return 0; | |
3144 | } | |
3145 | ||
f7262a9f | 3146 | _public_ int sd_event_source_set_time_accuracy(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 3147 | int r; |
6a0f1f6d | 3148 | |
305f78bf | 3149 | assert_return(s, -EINVAL); |
f5fbe71d | 3150 | assert_return(usec != UINT64_MAX, -EINVAL); |
6a0f1f6d | 3151 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 3152 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 3153 | assert_return(!event_origin_changed(s->event), -ECHILD); |
eaa3cbef | 3154 | |
2a0dc6cd LP |
3155 | r = source_set_pending(s, false); |
3156 | if (r < 0) | |
3157 | return r; | |
3158 | ||
eaa3cbef LP |
3159 | if (usec == 0) |
3160 | usec = DEFAULT_ACCURACY_USEC; | |
3161 | ||
eaa3cbef LP |
3162 | s->time.accuracy = usec; |
3163 | ||
e1951c16 | 3164 | event_source_time_prioq_reshuffle(s); |
6a0f1f6d LP |
3165 | return 0; |
3166 | } | |
3167 | ||
3168 | _public_ int sd_event_source_get_time_clock(sd_event_source *s, clockid_t *clock) { | |
3169 | assert_return(s, -EINVAL); | |
3170 | assert_return(clock, -EINVAL); | |
3171 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2eeff0f4 | 3172 | assert_return(!event_origin_changed(s->event), -ECHILD); |
eaa3cbef | 3173 | |
6a0f1f6d | 3174 | *clock = event_source_type_to_clock(s->type); |
eaa3cbef LP |
3175 | return 0; |
3176 | } | |
3177 | ||
f7262a9f | 3178 | _public_ int sd_event_source_get_child_pid(sd_event_source *s, pid_t *pid) { |
4bee8012 LP |
3179 | assert_return(s, -EINVAL); |
3180 | assert_return(pid, -EINVAL); | |
3181 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2eeff0f4 | 3182 | assert_return(!event_origin_changed(s->event), -ECHILD); |
4bee8012 LP |
3183 | |
3184 | *pid = s->child.pid; | |
3185 | return 0; | |
3186 | } | |
3187 | ||
f8f3f926 LP |
3188 | _public_ int sd_event_source_get_child_pidfd(sd_event_source *s) { |
3189 | assert_return(s, -EINVAL); | |
3190 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2eeff0f4 | 3191 | assert_return(!event_origin_changed(s->event), -ECHILD); |
f8f3f926 LP |
3192 | |
3193 | if (s->child.pidfd < 0) | |
3194 | return -EOPNOTSUPP; | |
3195 | ||
3196 | return s->child.pidfd; | |
3197 | } | |
3198 | ||
3199 | _public_ int sd_event_source_send_child_signal(sd_event_source *s, int sig, const siginfo_t *si, unsigned flags) { | |
3200 | assert_return(s, -EINVAL); | |
3201 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2eeff0f4 | 3202 | assert_return(!event_origin_changed(s->event), -ECHILD); |
f8f3f926 LP |
3203 | assert_return(SIGNAL_VALID(sig), -EINVAL); |
3204 | ||
3205 | /* If we already have seen indication the process exited refuse sending a signal early. This way we | |
3206 | * can be sure we don't accidentally kill the wrong process on PID reuse when pidfds are not | |
3207 | * available. */ | |
3208 | if (s->child.exited) | |
3209 | return -ESRCH; | |
3210 | ||
3211 | if (s->child.pidfd >= 0) { | |
3212 | siginfo_t copy; | |
3213 | ||
3214 | /* pidfd_send_signal() changes the siginfo_t argument. This is weird, let's hence copy the | |
3215 | * structure here */ | |
3216 | if (si) | |
3217 | copy = *si; | |
3218 | ||
3219 | if (pidfd_send_signal(s->child.pidfd, sig, si ? © : NULL, 0) < 0) { | |
3220 | /* Let's propagate the error only if the system call is not implemented or prohibited */ | |
3221 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
3222 | return -errno; | |
3223 | } else | |
3224 | return 0; | |
3225 | } | |
3226 | ||
3227 | /* Flags are only supported for pidfd_send_signal(), not for rt_sigqueueinfo(), hence let's refuse | |
3228 | * this here. */ | |
3229 | if (flags != 0) | |
3230 | return -EOPNOTSUPP; | |
3231 | ||
3232 | if (si) { | |
3233 | /* We use rt_sigqueueinfo() only if siginfo_t is specified. */ | |
3234 | siginfo_t copy = *si; | |
3235 | ||
3236 | if (rt_sigqueueinfo(s->child.pid, sig, ©) < 0) | |
3237 | return -errno; | |
3238 | } else if (kill(s->child.pid, sig) < 0) | |
3239 | return -errno; | |
3240 | ||
3241 | return 0; | |
3242 | } | |
3243 | ||
3244 | _public_ int sd_event_source_get_child_pidfd_own(sd_event_source *s) { | |
3245 | assert_return(s, -EINVAL); | |
3246 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2eeff0f4 | 3247 | assert_return(!event_origin_changed(s->event), -ECHILD); |
f8f3f926 LP |
3248 | |
3249 | if (s->child.pidfd < 0) | |
3250 | return -EOPNOTSUPP; | |
3251 | ||
3252 | return s->child.pidfd_owned; | |
3253 | } | |
3254 | ||
3255 | _public_ int sd_event_source_set_child_pidfd_own(sd_event_source *s, int own) { | |
3256 | assert_return(s, -EINVAL); | |
3257 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2eeff0f4 | 3258 | assert_return(!event_origin_changed(s->event), -ECHILD); |
f8f3f926 LP |
3259 | |
3260 | if (s->child.pidfd < 0) | |
3261 | return -EOPNOTSUPP; | |
3262 | ||
3263 | s->child.pidfd_owned = own; | |
3264 | return 0; | |
3265 | } | |
3266 | ||
3267 | _public_ int sd_event_source_get_child_process_own(sd_event_source *s) { | |
3268 | assert_return(s, -EINVAL); | |
3269 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2eeff0f4 | 3270 | assert_return(!event_origin_changed(s->event), -ECHILD); |
f8f3f926 LP |
3271 | |
3272 | return s->child.process_owned; | |
3273 | } | |
3274 | ||
3275 | _public_ int sd_event_source_set_child_process_own(sd_event_source *s, int own) { | |
3276 | assert_return(s, -EINVAL); | |
3277 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2eeff0f4 | 3278 | assert_return(!event_origin_changed(s->event), -ECHILD); |
f8f3f926 LP |
3279 | |
3280 | s->child.process_owned = own; | |
3281 | return 0; | |
3282 | } | |
3283 | ||
97ef5391 LP |
3284 | _public_ int sd_event_source_get_inotify_mask(sd_event_source *s, uint32_t *mask) { |
3285 | assert_return(s, -EINVAL); | |
3286 | assert_return(mask, -EINVAL); | |
3287 | assert_return(s->type == SOURCE_INOTIFY, -EDOM); | |
2eeff0f4 | 3288 | assert_return(!event_origin_changed(s->event), -ECHILD); |
97ef5391 LP |
3289 | |
3290 | *mask = s->inotify.mask; | |
3291 | return 0; | |
3292 | } | |
3293 | ||
718db961 | 3294 | _public_ int sd_event_source_set_prepare(sd_event_source *s, sd_event_handler_t callback) { |
fd38203a LP |
3295 | int r; |
3296 | ||
da7e457c | 3297 | assert_return(s, -EINVAL); |
6203e07a | 3298 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c | 3299 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 3300 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
3301 | |
3302 | if (s->prepare == callback) | |
3303 | return 0; | |
3304 | ||
3305 | if (callback && s->prepare) { | |
3306 | s->prepare = callback; | |
3307 | return 0; | |
3308 | } | |
3309 | ||
3310 | r = prioq_ensure_allocated(&s->event->prepare, prepare_prioq_compare); | |
3311 | if (r < 0) | |
3312 | return r; | |
3313 | ||
3314 | s->prepare = callback; | |
3315 | ||
3316 | if (callback) { | |
3317 | r = prioq_put(s->event->prepare, s, &s->prepare_index); | |
3318 | if (r < 0) | |
3319 | return r; | |
3320 | } else | |
3321 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
3322 | ||
3323 | return 0; | |
3324 | } | |
3325 | ||
f7262a9f | 3326 | _public_ void* sd_event_source_get_userdata(sd_event_source *s) { |
da7e457c | 3327 | assert_return(s, NULL); |
2eeff0f4 | 3328 | assert_return(!event_origin_changed(s->event), NULL); |
fd38203a LP |
3329 | |
3330 | return s->userdata; | |
3331 | } | |
3332 | ||
8f726607 LP |
3333 | _public_ void *sd_event_source_set_userdata(sd_event_source *s, void *userdata) { |
3334 | void *ret; | |
3335 | ||
3336 | assert_return(s, NULL); | |
2eeff0f4 | 3337 | assert_return(!event_origin_changed(s->event), NULL); |
8f726607 LP |
3338 | |
3339 | ret = s->userdata; | |
3340 | s->userdata = userdata; | |
3341 | ||
3342 | return ret; | |
3343 | } | |
3344 | ||
b6d5481b LP |
3345 | static int event_source_enter_ratelimited(sd_event_source *s) { |
3346 | int r; | |
3347 | ||
3348 | assert(s); | |
3349 | ||
3350 | /* When an event source becomes ratelimited, we place it in the CLOCK_MONOTONIC priority queue, with | |
3351 | * the end of the rate limit time window, much as if it was a timer event source. */ | |
3352 | ||
3353 | if (s->ratelimited) | |
3354 | return 0; /* Already ratelimited, this is a NOP hence */ | |
3355 | ||
3356 | /* Make sure we can install a CLOCK_MONOTONIC event further down. */ | |
3357 | r = setup_clock_data(s->event, &s->event->monotonic, CLOCK_MONOTONIC); | |
3358 | if (r < 0) | |
3359 | return r; | |
3360 | ||
3361 | /* Timer event sources are already using the earliest/latest queues for the timer scheduling. Let's | |
3362 | * first remove them from the prioq appropriate for their own clock, so that we can use the prioq | |
3363 | * fields of the event source then for adding it to the CLOCK_MONOTONIC prioq instead. */ | |
3364 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
3365 | event_source_time_prioq_remove(s, event_get_clock_data(s->event, s->type)); | |
3366 | ||
3367 | /* Now, let's add the event source to the monotonic clock instead */ | |
3368 | r = event_source_time_prioq_put(s, &s->event->monotonic); | |
3369 | if (r < 0) | |
3370 | goto fail; | |
3371 | ||
3372 | /* And let's take the event source officially offline */ | |
3373 | r = event_source_offline(s, s->enabled, /* ratelimited= */ true); | |
3374 | if (r < 0) { | |
3375 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
3376 | goto fail; | |
3377 | } | |
3378 | ||
3379 | event_source_pp_prioq_reshuffle(s); | |
3380 | ||
3381 | log_debug("Event source %p (%s) entered rate limit state.", s, strna(s->description)); | |
3382 | return 0; | |
3383 | ||
3384 | fail: | |
3385 | /* Reinstall time event sources in the priority queue as before. This shouldn't fail, since the queue | |
3386 | * space for it should already be allocated. */ | |
3387 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
3388 | assert_se(event_source_time_prioq_put(s, event_get_clock_data(s->event, s->type)) >= 0); | |
3389 | ||
3390 | return r; | |
3391 | } | |
3392 | ||
fd69f224 | 3393 | static int event_source_leave_ratelimit(sd_event_source *s, bool run_callback) { |
b6d5481b LP |
3394 | int r; |
3395 | ||
3396 | assert(s); | |
3397 | ||
3398 | if (!s->ratelimited) | |
3399 | return 0; | |
3400 | ||
3401 | /* Let's take the event source out of the monotonic prioq first. */ | |
3402 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
3403 | ||
3404 | /* Let's then add the event source to its native clock prioq again — if this is a timer event source */ | |
3405 | if (EVENT_SOURCE_IS_TIME(s->type)) { | |
3406 | r = event_source_time_prioq_put(s, event_get_clock_data(s->event, s->type)); | |
3407 | if (r < 0) | |
3408 | goto fail; | |
3409 | } | |
3410 | ||
3411 | /* Let's try to take it online again. */ | |
3412 | r = event_source_online(s, s->enabled, /* ratelimited= */ false); | |
3413 | if (r < 0) { | |
3414 | /* Do something roughly sensible when this failed: undo the two prioq ops above */ | |
3415 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
3416 | event_source_time_prioq_remove(s, event_get_clock_data(s->event, s->type)); | |
3417 | ||
3418 | goto fail; | |
3419 | } | |
3420 | ||
3421 | event_source_pp_prioq_reshuffle(s); | |
3422 | ratelimit_reset(&s->rate_limit); | |
3423 | ||
3424 | log_debug("Event source %p (%s) left rate limit state.", s, strna(s->description)); | |
fd69f224 MS |
3425 | |
3426 | if (run_callback && s->ratelimit_expire_callback) { | |
3427 | s->dispatching = true; | |
3428 | r = s->ratelimit_expire_callback(s, s->userdata); | |
3429 | s->dispatching = false; | |
3430 | ||
3431 | if (r < 0) { | |
3432 | log_debug_errno(r, "Ratelimit expiry callback of event source %s (type %s) returned error, %s: %m", | |
3433 | strna(s->description), | |
3434 | event_source_type_to_string(s->type), | |
3435 | s->exit_on_failure ? "exiting" : "disabling"); | |
3436 | ||
3437 | if (s->exit_on_failure) | |
3438 | (void) sd_event_exit(s->event, r); | |
3439 | } | |
3440 | ||
3441 | if (s->n_ref == 0) | |
3442 | source_free(s); | |
3443 | else if (r < 0) | |
0a040e64 | 3444 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
fd69f224 MS |
3445 | |
3446 | return 1; | |
3447 | } | |
3448 | ||
b6d5481b LP |
3449 | return 0; |
3450 | ||
3451 | fail: | |
3452 | /* Do something somewhat reasonable when we cannot move an event sources out of ratelimited mode: | |
3453 | * simply put it back in it, maybe we can then process it more successfully next iteration. */ | |
3454 | assert_se(event_source_time_prioq_put(s, &s->event->monotonic) >= 0); | |
3455 | ||
3456 | return r; | |
3457 | } | |
3458 | ||
c2ba3ad6 LP |
3459 | static usec_t sleep_between(sd_event *e, usec_t a, usec_t b) { |
3460 | usec_t c; | |
3461 | assert(e); | |
3462 | assert(a <= b); | |
3463 | ||
3464 | if (a <= 0) | |
3465 | return 0; | |
393003e1 LP |
3466 | if (a >= USEC_INFINITY) |
3467 | return USEC_INFINITY; | |
c2ba3ad6 LP |
3468 | |
3469 | if (b <= a + 1) | |
3470 | return a; | |
3471 | ||
52444dc4 LP |
3472 | initialize_perturb(e); |
3473 | ||
c2ba3ad6 LP |
3474 | /* |
3475 | Find a good time to wake up again between times a and b. We | |
3476 | have two goals here: | |
3477 | ||
3478 | a) We want to wake up as seldom as possible, hence prefer | |
3479 | later times over earlier times. | |
3480 | ||
3481 | b) But if we have to wake up, then let's make sure to | |
3482 | dispatch as much as possible on the entire system. | |
3483 | ||
3484 | We implement this by waking up everywhere at the same time | |
850516e0 | 3485 | within any given minute if we can, synchronised via the |
c2ba3ad6 | 3486 | perturbation value determined from the boot ID. If we can't, |
ba276c81 LP |
3487 | then we try to find the same spot in every 10s, then 1s and |
3488 | then 250ms step. Otherwise, we pick the last possible time | |
3489 | to wake up. | |
c2ba3ad6 LP |
3490 | */ |
3491 | ||
850516e0 LP |
3492 | c = (b / USEC_PER_MINUTE) * USEC_PER_MINUTE + e->perturb; |
3493 | if (c >= b) { | |
3494 | if (_unlikely_(c < USEC_PER_MINUTE)) | |
3495 | return b; | |
3496 | ||
3497 | c -= USEC_PER_MINUTE; | |
3498 | } | |
3499 | ||
ba276c81 LP |
3500 | if (c >= a) |
3501 | return c; | |
3502 | ||
3503 | c = (b / (USEC_PER_SEC*10)) * (USEC_PER_SEC*10) + (e->perturb % (USEC_PER_SEC*10)); | |
3504 | if (c >= b) { | |
3505 | if (_unlikely_(c < USEC_PER_SEC*10)) | |
3506 | return b; | |
3507 | ||
3508 | c -= USEC_PER_SEC*10; | |
3509 | } | |
3510 | ||
850516e0 LP |
3511 | if (c >= a) |
3512 | return c; | |
3513 | ||
3514 | c = (b / USEC_PER_SEC) * USEC_PER_SEC + (e->perturb % USEC_PER_SEC); | |
c2ba3ad6 LP |
3515 | if (c >= b) { |
3516 | if (_unlikely_(c < USEC_PER_SEC)) | |
3517 | return b; | |
3518 | ||
3519 | c -= USEC_PER_SEC; | |
3520 | } | |
3521 | ||
3522 | if (c >= a) | |
3523 | return c; | |
3524 | ||
3525 | c = (b / (USEC_PER_MSEC*250)) * (USEC_PER_MSEC*250) + (e->perturb % (USEC_PER_MSEC*250)); | |
3526 | if (c >= b) { | |
3527 | if (_unlikely_(c < USEC_PER_MSEC*250)) | |
3528 | return b; | |
3529 | ||
3530 | c -= USEC_PER_MSEC*250; | |
3531 | } | |
3532 | ||
3533 | if (c >= a) | |
3534 | return c; | |
3535 | ||
3536 | return b; | |
3537 | } | |
3538 | ||
fd38203a LP |
3539 | static int event_arm_timer( |
3540 | sd_event *e, | |
6a0f1f6d | 3541 | struct clock_data *d) { |
fd38203a LP |
3542 | |
3543 | struct itimerspec its = {}; | |
c2ba3ad6 LP |
3544 | sd_event_source *a, *b; |
3545 | usec_t t; | |
fd38203a | 3546 | |
cde93897 | 3547 | assert(e); |
6a0f1f6d | 3548 | assert(d); |
fd38203a | 3549 | |
d06441da | 3550 | if (!d->needs_rearm) |
212bbb17 | 3551 | return 0; |
7e2bf71c YW |
3552 | |
3553 | d->needs_rearm = false; | |
212bbb17 | 3554 | |
6a0f1f6d | 3555 | a = prioq_peek(d->earliest); |
19947509 | 3556 | assert(!a || EVENT_SOURCE_USES_TIME_PRIOQ(a->type)); |
b6d5481b | 3557 | if (!a || a->enabled == SD_EVENT_OFF || time_event_source_next(a) == USEC_INFINITY) { |
72aedc1e | 3558 | |
6a0f1f6d | 3559 | if (d->fd < 0) |
c57b5ca3 LP |
3560 | return 0; |
3561 | ||
3a43da28 | 3562 | if (d->next == USEC_INFINITY) |
72aedc1e LP |
3563 | return 0; |
3564 | ||
3565 | /* disarm */ | |
15c689d7 LP |
3566 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
3567 | return -errno; | |
72aedc1e | 3568 | |
3a43da28 | 3569 | d->next = USEC_INFINITY; |
fd38203a | 3570 | return 0; |
72aedc1e | 3571 | } |
fd38203a | 3572 | |
6a0f1f6d | 3573 | b = prioq_peek(d->latest); |
19947509 ZJS |
3574 | assert(!b || EVENT_SOURCE_USES_TIME_PRIOQ(b->type)); |
3575 | assert(b && b->enabled != SD_EVENT_OFF); | |
c2ba3ad6 | 3576 | |
b6d5481b | 3577 | t = sleep_between(e, time_event_source_next(a), time_event_source_latest(b)); |
6a0f1f6d | 3578 | if (d->next == t) |
fd38203a LP |
3579 | return 0; |
3580 | ||
6a0f1f6d | 3581 | assert_se(d->fd >= 0); |
fd38203a | 3582 | |
c2ba3ad6 | 3583 | if (t == 0) { |
1751bdde | 3584 | /* We don't want to disarm here, just mean some time looooong ago. */ |
fd38203a LP |
3585 | its.it_value.tv_sec = 0; |
3586 | its.it_value.tv_nsec = 1; | |
3587 | } else | |
c2ba3ad6 | 3588 | timespec_store(&its.it_value, t); |
fd38203a | 3589 | |
15c689d7 | 3590 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 | 3591 | return -errno; |
fd38203a | 3592 | |
6a0f1f6d | 3593 | d->next = t; |
fd38203a LP |
3594 | return 0; |
3595 | } | |
3596 | ||
9a800b56 | 3597 | static int process_io(sd_event *e, sd_event_source *s, uint32_t revents) { |
fd38203a LP |
3598 | assert(e); |
3599 | assert(s); | |
3600 | assert(s->type == SOURCE_IO); | |
3601 | ||
9a800b56 LP |
3602 | /* If the event source was already pending, we just OR in the |
3603 | * new revents, otherwise we reset the value. The ORing is | |
3604 | * necessary to handle EPOLLONESHOT events properly where | |
3605 | * readability might happen independently of writability, and | |
3606 | * we need to keep track of both */ | |
3607 | ||
3608 | if (s->pending) | |
3609 | s->io.revents |= revents; | |
3610 | else | |
3611 | s->io.revents = revents; | |
fd38203a | 3612 | |
fd38203a LP |
3613 | return source_set_pending(s, true); |
3614 | } | |
3615 | ||
72aedc1e | 3616 | static int flush_timer(sd_event *e, int fd, uint32_t events, usec_t *next) { |
fd38203a LP |
3617 | uint64_t x; |
3618 | ssize_t ss; | |
3619 | ||
3620 | assert(e); | |
da7e457c | 3621 | assert(fd >= 0); |
72aedc1e | 3622 | |
305f78bf | 3623 | assert_return(events == EPOLLIN, -EIO); |
fd38203a LP |
3624 | |
3625 | ss = read(fd, &x, sizeof(x)); | |
3626 | if (ss < 0) { | |
8add30a0 | 3627 | if (ERRNO_IS_TRANSIENT(errno)) |
fd38203a LP |
3628 | return 0; |
3629 | ||
3630 | return -errno; | |
3631 | } | |
3632 | ||
8d35dae7 | 3633 | if (_unlikely_(ss != sizeof(x))) |
fd38203a LP |
3634 | return -EIO; |
3635 | ||
cde93897 | 3636 | if (next) |
3a43da28 | 3637 | *next = USEC_INFINITY; |
72aedc1e | 3638 | |
fd38203a LP |
3639 | return 0; |
3640 | } | |
3641 | ||
305f78bf LP |
3642 | static int process_timer( |
3643 | sd_event *e, | |
3644 | usec_t n, | |
6a0f1f6d | 3645 | struct clock_data *d) { |
305f78bf | 3646 | |
fd38203a | 3647 | sd_event_source *s; |
fd69f224 | 3648 | bool callback_invoked = false; |
fd38203a LP |
3649 | int r; |
3650 | ||
3651 | assert(e); | |
6a0f1f6d | 3652 | assert(d); |
fd38203a LP |
3653 | |
3654 | for (;;) { | |
6a0f1f6d | 3655 | s = prioq_peek(d->earliest); |
19947509 ZJS |
3656 | assert(!s || EVENT_SOURCE_USES_TIME_PRIOQ(s->type)); |
3657 | ||
b6d5481b LP |
3658 | if (!s || time_event_source_next(s) > n) |
3659 | break; | |
3660 | ||
3661 | if (s->ratelimited) { | |
3662 | /* This is an event sources whose ratelimit window has ended. Let's turn it on | |
3663 | * again. */ | |
3664 | assert(s->ratelimited); | |
3665 | ||
fd69f224 | 3666 | r = event_source_leave_ratelimit(s, /* run_callback */ true); |
b6d5481b LP |
3667 | if (r < 0) |
3668 | return r; | |
fd69f224 MS |
3669 | else if (r == 1) |
3670 | callback_invoked = true; | |
b6d5481b LP |
3671 | |
3672 | continue; | |
3673 | } | |
3674 | ||
3675 | if (s->enabled == SD_EVENT_OFF || s->pending) | |
fd38203a LP |
3676 | break; |
3677 | ||
3678 | r = source_set_pending(s, true); | |
3679 | if (r < 0) | |
3680 | return r; | |
3681 | ||
e1951c16 | 3682 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
3683 | } |
3684 | ||
fd69f224 | 3685 | return callback_invoked; |
fd38203a LP |
3686 | } |
3687 | ||
efd3be9d YW |
3688 | static int process_child(sd_event *e, int64_t threshold, int64_t *ret_min_priority) { |
3689 | int64_t min_priority = threshold; | |
3690 | bool something_new = false; | |
fd38203a | 3691 | sd_event_source *s; |
fd38203a LP |
3692 | int r; |
3693 | ||
3694 | assert(e); | |
efd3be9d YW |
3695 | assert(ret_min_priority); |
3696 | ||
3697 | if (!e->need_process_child) { | |
3698 | *ret_min_priority = min_priority; | |
3699 | return 0; | |
3700 | } | |
fd38203a | 3701 | |
c2ba3ad6 LP |
3702 | e->need_process_child = false; |
3703 | ||
91c70071 YW |
3704 | /* So, this is ugly. We iteratively invoke waitid() with P_PID + WNOHANG for each PID we wait |
3705 | * for, instead of using P_ALL. This is because we only want to get child information of very | |
3706 | * specific child processes, and not all of them. We might not have processed the SIGCHLD event | |
3707 | * of a previous invocation and we don't want to maintain a unbounded *per-child* event queue, | |
3708 | * hence we really don't want anything flushed out of the kernel's queue that we don't care | |
3709 | * about. Since this is O(n) this means that if you have a lot of processes you probably want | |
3710 | * to handle SIGCHLD yourself. | |
3711 | * | |
3712 | * We do not reap the children here (by using WNOWAIT), this is only done after the event | |
3713 | * source is dispatched so that the callback still sees the process as a zombie. */ | |
fd38203a | 3714 | |
90e74a66 | 3715 | HASHMAP_FOREACH(s, e->child_sources) { |
fd38203a LP |
3716 | assert(s->type == SOURCE_CHILD); |
3717 | ||
efd3be9d YW |
3718 | if (s->priority > threshold) |
3719 | continue; | |
3720 | ||
fd38203a LP |
3721 | if (s->pending) |
3722 | continue; | |
3723 | ||
b6d5481b | 3724 | if (event_source_is_offline(s)) |
fd38203a LP |
3725 | continue; |
3726 | ||
f8f3f926 LP |
3727 | if (s->child.exited) |
3728 | continue; | |
3729 | ||
91c70071 YW |
3730 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
3731 | /* There's a usable pidfd known for this event source? Then don't waitid() for | |
3732 | * it here */ | |
f8f3f926 LP |
3733 | continue; |
3734 | ||
fd38203a | 3735 | zero(s->child.siginfo); |
15c689d7 LP |
3736 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, |
3737 | WNOHANG | (s->child.options & WEXITED ? WNOWAIT : 0) | s->child.options) < 0) | |
bfd9bfcc | 3738 | return negative_errno(); |
fd38203a LP |
3739 | |
3740 | if (s->child.siginfo.si_pid != 0) { | |
945c2931 | 3741 | bool zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 3742 | |
f8f3f926 LP |
3743 | if (zombie) |
3744 | s->child.exited = true; | |
3745 | ||
08cd1552 | 3746 | if (!zombie && (s->child.options & WEXITED)) { |
91c70071 YW |
3747 | /* If the child isn't dead then let's immediately remove the state |
3748 | * change from the queue, since there's no benefit in leaving it | |
3749 | * queued. */ | |
08cd1552 LP |
3750 | |
3751 | assert(s->child.options & (WSTOPPED|WCONTINUED)); | |
a5d27871 | 3752 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|(s->child.options & (WSTOPPED|WCONTINUED))); |
08cd1552 LP |
3753 | } |
3754 | ||
fd38203a LP |
3755 | r = source_set_pending(s, true); |
3756 | if (r < 0) | |
3757 | return r; | |
efd3be9d YW |
3758 | if (r > 0) { |
3759 | something_new = true; | |
3760 | min_priority = MIN(min_priority, s->priority); | |
3761 | } | |
fd38203a LP |
3762 | } |
3763 | } | |
3764 | ||
efd3be9d YW |
3765 | *ret_min_priority = min_priority; |
3766 | return something_new; | |
fd38203a LP |
3767 | } |
3768 | ||
f8f3f926 LP |
3769 | static int process_pidfd(sd_event *e, sd_event_source *s, uint32_t revents) { |
3770 | assert(e); | |
3771 | assert(s); | |
3772 | assert(s->type == SOURCE_CHILD); | |
3773 | ||
3774 | if (s->pending) | |
3775 | return 0; | |
3776 | ||
b6d5481b | 3777 | if (event_source_is_offline(s)) |
f8f3f926 LP |
3778 | return 0; |
3779 | ||
3780 | if (!EVENT_SOURCE_WATCH_PIDFD(s)) | |
3781 | return 0; | |
3782 | ||
3783 | zero(s->child.siginfo); | |
3784 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG | WNOWAIT | s->child.options) < 0) | |
3785 | return -errno; | |
3786 | ||
3787 | if (s->child.siginfo.si_pid == 0) | |
3788 | return 0; | |
3789 | ||
3790 | if (IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) | |
3791 | s->child.exited = true; | |
3792 | ||
3793 | return source_set_pending(s, true); | |
3794 | } | |
3795 | ||
efd3be9d | 3796 | static int process_signal(sd_event *e, struct signal_data *d, uint32_t events, int64_t *min_priority) { |
fd38203a LP |
3797 | int r; |
3798 | ||
da7e457c | 3799 | assert(e); |
97ef5391 | 3800 | assert(d); |
305f78bf | 3801 | assert_return(events == EPOLLIN, -EIO); |
efd3be9d | 3802 | assert(min_priority); |
fd38203a | 3803 | |
91c70071 YW |
3804 | /* If there's a signal queued on this priority and SIGCHLD is on this priority too, then make |
3805 | * sure to recheck the children we watch. This is because we only ever dequeue the first signal | |
3806 | * per priority, and if we dequeue one, and SIGCHLD might be enqueued later we wouldn't know, | |
3807 | * but we might have higher priority children we care about hence we need to check that | |
3808 | * explicitly. */ | |
9da4cb2b LP |
3809 | |
3810 | if (sigismember(&d->sigset, SIGCHLD)) | |
3811 | e->need_process_child = true; | |
3812 | ||
91c70071 | 3813 | /* If there's already an event source pending for this priority we don't read another */ |
9da4cb2b LP |
3814 | if (d->current) |
3815 | return 0; | |
3816 | ||
fd38203a | 3817 | for (;;) { |
0eb2e0e3 | 3818 | struct signalfd_siginfo si; |
7057bd99 | 3819 | ssize_t n; |
92daebc0 | 3820 | sd_event_source *s = NULL; |
fd38203a | 3821 | |
9da4cb2b | 3822 | n = read(d->fd, &si, sizeof(si)); |
7057bd99 | 3823 | if (n < 0) { |
8add30a0 | 3824 | if (ERRNO_IS_TRANSIENT(errno)) |
efd3be9d | 3825 | return 0; |
fd38203a LP |
3826 | |
3827 | return -errno; | |
3828 | } | |
3829 | ||
7057bd99 | 3830 | if (_unlikely_(n != sizeof(si))) |
fd38203a LP |
3831 | return -EIO; |
3832 | ||
6eb7c172 | 3833 | assert(SIGNAL_VALID(si.ssi_signo)); |
7057bd99 | 3834 | |
92daebc0 LP |
3835 | if (e->signal_sources) |
3836 | s = e->signal_sources[si.ssi_signo]; | |
92daebc0 LP |
3837 | if (!s) |
3838 | continue; | |
9da4cb2b LP |
3839 | if (s->pending) |
3840 | continue; | |
fd38203a LP |
3841 | |
3842 | s->signal.siginfo = si; | |
9da4cb2b LP |
3843 | d->current = s; |
3844 | ||
fd38203a LP |
3845 | r = source_set_pending(s, true); |
3846 | if (r < 0) | |
3847 | return r; | |
efd3be9d YW |
3848 | if (r > 0 && *min_priority >= s->priority) { |
3849 | *min_priority = s->priority; | |
3850 | return 1; /* an event source with smaller priority is queued. */ | |
3851 | } | |
9da4cb2b | 3852 | |
efd3be9d | 3853 | return 0; |
fd38203a | 3854 | } |
fd38203a LP |
3855 | } |
3856 | ||
efd3be9d | 3857 | static int event_inotify_data_read(sd_event *e, struct inotify_data *d, uint32_t revents, int64_t threshold) { |
97ef5391 LP |
3858 | ssize_t n; |
3859 | ||
3860 | assert(e); | |
3861 | assert(d); | |
3862 | ||
3863 | assert_return(revents == EPOLLIN, -EIO); | |
3864 | ||
3865 | /* If there's already an event source pending for this priority, don't read another */ | |
3866 | if (d->n_pending > 0) | |
3867 | return 0; | |
3868 | ||
3869 | /* Is the read buffer non-empty? If so, let's not read more */ | |
3870 | if (d->buffer_filled > 0) | |
3871 | return 0; | |
3872 | ||
efd3be9d YW |
3873 | if (d->priority > threshold) |
3874 | return 0; | |
3875 | ||
97ef5391 LP |
3876 | n = read(d->fd, &d->buffer, sizeof(d->buffer)); |
3877 | if (n < 0) { | |
8add30a0 | 3878 | if (ERRNO_IS_TRANSIENT(errno)) |
97ef5391 LP |
3879 | return 0; |
3880 | ||
3881 | return -errno; | |
3882 | } | |
3883 | ||
3884 | assert(n > 0); | |
3885 | d->buffer_filled = (size_t) n; | |
0601b958 | 3886 | LIST_PREPEND(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
3887 | |
3888 | return 1; | |
3889 | } | |
3890 | ||
3891 | static void event_inotify_data_drop(sd_event *e, struct inotify_data *d, size_t sz) { | |
3892 | assert(e); | |
3893 | assert(d); | |
3894 | assert(sz <= d->buffer_filled); | |
3895 | ||
3896 | if (sz == 0) | |
3897 | return; | |
3898 | ||
3899 | /* Move the rest to the buffer to the front, in order to get things properly aligned again */ | |
3900 | memmove(d->buffer.raw, d->buffer.raw + sz, d->buffer_filled - sz); | |
3901 | d->buffer_filled -= sz; | |
3902 | ||
3903 | if (d->buffer_filled == 0) | |
0601b958 | 3904 | LIST_REMOVE(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
3905 | } |
3906 | ||
3907 | static int event_inotify_data_process(sd_event *e, struct inotify_data *d) { | |
3908 | int r; | |
3909 | ||
3910 | assert(e); | |
3911 | assert(d); | |
3912 | ||
3913 | /* If there's already an event source pending for this priority, don't read another */ | |
3914 | if (d->n_pending > 0) | |
3915 | return 0; | |
3916 | ||
3917 | while (d->buffer_filled > 0) { | |
3918 | size_t sz; | |
3919 | ||
3920 | /* Let's validate that the event structures are complete */ | |
3921 | if (d->buffer_filled < offsetof(struct inotify_event, name)) | |
3922 | return -EIO; | |
3923 | ||
3924 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3925 | if (d->buffer_filled < sz) | |
3926 | return -EIO; | |
3927 | ||
3928 | if (d->buffer.ev.mask & IN_Q_OVERFLOW) { | |
3929 | struct inode_data *inode_data; | |
97ef5391 LP |
3930 | |
3931 | /* The queue overran, let's pass this event to all event sources connected to this inotify | |
3932 | * object */ | |
3933 | ||
03677889 | 3934 | HASHMAP_FOREACH(inode_data, d->inodes) |
97ef5391 LP |
3935 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { |
3936 | ||
b6d5481b | 3937 | if (event_source_is_offline(s)) |
97ef5391 LP |
3938 | continue; |
3939 | ||
3940 | r = source_set_pending(s, true); | |
3941 | if (r < 0) | |
3942 | return r; | |
3943 | } | |
97ef5391 LP |
3944 | } else { |
3945 | struct inode_data *inode_data; | |
97ef5391 LP |
3946 | |
3947 | /* Find the inode object for this watch descriptor. If IN_IGNORED is set we also remove it from | |
3948 | * our watch descriptor table. */ | |
3949 | if (d->buffer.ev.mask & IN_IGNORED) { | |
3950 | ||
3951 | inode_data = hashmap_remove(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3952 | if (!inode_data) { | |
3953 | event_inotify_data_drop(e, d, sz); | |
3954 | continue; | |
3955 | } | |
3956 | ||
3957 | /* The watch descriptor was removed by the kernel, let's drop it here too */ | |
3958 | inode_data->wd = -1; | |
3959 | } else { | |
3960 | inode_data = hashmap_get(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3961 | if (!inode_data) { | |
3962 | event_inotify_data_drop(e, d, sz); | |
3963 | continue; | |
3964 | } | |
3965 | } | |
3966 | ||
3967 | /* Trigger all event sources that are interested in these events. Also trigger all event | |
3968 | * sources if IN_IGNORED or IN_UNMOUNT is set. */ | |
3969 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
3970 | ||
b6d5481b | 3971 | if (event_source_is_offline(s)) |
97ef5391 LP |
3972 | continue; |
3973 | ||
3974 | if ((d->buffer.ev.mask & (IN_IGNORED|IN_UNMOUNT)) == 0 && | |
3975 | (s->inotify.mask & d->buffer.ev.mask & IN_ALL_EVENTS) == 0) | |
3976 | continue; | |
3977 | ||
3978 | r = source_set_pending(s, true); | |
3979 | if (r < 0) | |
3980 | return r; | |
3981 | } | |
3982 | } | |
3983 | ||
3984 | /* Something pending now? If so, let's finish, otherwise let's read more. */ | |
3985 | if (d->n_pending > 0) | |
3986 | return 1; | |
3987 | } | |
3988 | ||
3989 | return 0; | |
3990 | } | |
3991 | ||
3992 | static int process_inotify(sd_event *e) { | |
97ef5391 LP |
3993 | int r, done = 0; |
3994 | ||
3995 | assert(e); | |
3996 | ||
0601b958 | 3997 | LIST_FOREACH(buffered, d, e->buffered_inotify_data_list) { |
97ef5391 LP |
3998 | r = event_inotify_data_process(e, d); |
3999 | if (r < 0) | |
4000 | return r; | |
4001 | if (r > 0) | |
4002 | done ++; | |
4003 | } | |
4004 | ||
4005 | return done; | |
4006 | } | |
4007 | ||
158fe190 LP |
4008 | static int process_memory_pressure(sd_event_source *s, uint32_t revents) { |
4009 | assert(s); | |
4010 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
4011 | ||
4012 | if (s->pending) | |
4013 | s->memory_pressure.revents |= revents; | |
4014 | else | |
4015 | s->memory_pressure.revents = revents; | |
4016 | ||
4017 | return source_set_pending(s, true); | |
4018 | } | |
4019 | ||
4020 | static int source_memory_pressure_write(sd_event_source *s) { | |
4021 | ssize_t n; | |
4022 | int r; | |
4023 | ||
4024 | assert(s); | |
4025 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
4026 | ||
4027 | /* once we start writing, the buffer is locked, we allow no further changes. */ | |
4028 | s->memory_pressure.locked = true; | |
4029 | ||
4030 | if (s->memory_pressure.write_buffer_size > 0) { | |
4031 | n = write(s->memory_pressure.fd, s->memory_pressure.write_buffer, s->memory_pressure.write_buffer_size); | |
4032 | if (n < 0) { | |
9897f5dd LP |
4033 | if (!ERRNO_IS_TRANSIENT(errno)) { |
4034 | /* If kernel is built with CONFIG_PSI_DEFAULT_DISABLED it will expose PSI | |
4035 | * files, but then generates EOPNOSUPP on read() and write() (instead of on | |
4036 | * open()!). This sucks hard, since we can only detect this kind of failure | |
4037 | * so late. Let's make the best of it, and turn off the event source like we | |
4038 | * do for failed event source handlers. */ | |
4039 | ||
4040 | log_debug_errno(errno, "Writing memory pressure settings to kernel failed, disabling memory pressure event source: %m"); | |
4041 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); | |
4042 | return 0; | |
4043 | } | |
158fe190 LP |
4044 | |
4045 | n = 0; | |
4046 | } | |
4047 | } else | |
4048 | n = 0; | |
4049 | ||
4050 | assert(n >= 0); | |
4051 | ||
4052 | if ((size_t) n == s->memory_pressure.write_buffer_size) { | |
4053 | s->memory_pressure.write_buffer = mfree(s->memory_pressure.write_buffer); | |
4054 | ||
4055 | if (n > 0) { | |
4056 | s->memory_pressure.write_buffer_size = 0; | |
4057 | ||
4058 | /* Update epoll events mask, since we have now written everything and don't care for EPOLLOUT anymore */ | |
4059 | r = source_memory_pressure_register(s, s->enabled); | |
4060 | if (r < 0) | |
4061 | return r; | |
4062 | } | |
4063 | } else if (n > 0) { | |
4064 | _cleanup_free_ void *c = NULL; | |
4065 | ||
4066 | assert((size_t) n < s->memory_pressure.write_buffer_size); | |
4067 | ||
4068 | c = memdup((uint8_t*) s->memory_pressure.write_buffer + n, s->memory_pressure.write_buffer_size - n); | |
4069 | if (!c) | |
4070 | return -ENOMEM; | |
4071 | ||
4072 | free_and_replace(s->memory_pressure.write_buffer, c); | |
4073 | s->memory_pressure.write_buffer_size -= n; | |
4074 | return 1; | |
4075 | } | |
4076 | ||
4077 | return 0; | |
4078 | } | |
4079 | ||
4080 | static int source_memory_pressure_initiate_dispatch(sd_event_source *s) { | |
4081 | int r; | |
4082 | ||
4083 | assert(s); | |
4084 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
4085 | ||
4086 | r = source_memory_pressure_write(s); | |
4087 | if (r < 0) | |
4088 | return r; | |
4089 | if (r > 0) | |
4090 | return 1; /* if we wrote something, then don't continue with dispatching user dispatch | |
4091 | * function. Instead, shortcut it so that we wait for next EPOLLOUT immediately. */ | |
4092 | ||
4093 | /* No pending incoming IO? Then let's not continue further */ | |
4094 | if ((s->memory_pressure.revents & (EPOLLIN|EPOLLPRI)) == 0) { | |
4095 | ||
4096 | /* Treat IO errors on the notifier the same ways errors returned from a callback */ | |
4097 | if ((s->memory_pressure.revents & (EPOLLHUP|EPOLLERR|EPOLLRDHUP)) != 0) | |
4098 | return -EIO; | |
4099 | ||
4100 | return 1; /* leave dispatch, we already processed everything */ | |
4101 | } | |
4102 | ||
4103 | if (s->memory_pressure.revents & EPOLLIN) { | |
4104 | uint8_t pipe_buf[PIPE_BUF]; | |
4105 | ssize_t n; | |
4106 | ||
4107 | /* If the fd is readable, then flush out anything that might be queued */ | |
4108 | ||
4109 | n = read(s->memory_pressure.fd, pipe_buf, sizeof(pipe_buf)); | |
4110 | if (n < 0 && !ERRNO_IS_TRANSIENT(errno)) | |
4111 | return -errno; | |
4112 | } | |
4113 | ||
4114 | return 0; /* go on, dispatch to user callback */ | |
4115 | } | |
4116 | ||
fd38203a | 4117 | static int source_dispatch(sd_event_source *s) { |
8f5c235d | 4118 | EventSourceType saved_type; |
c8e9d15c | 4119 | sd_event *saved_event; |
fe8245eb | 4120 | int r = 0; |
fd38203a LP |
4121 | |
4122 | assert(s); | |
6203e07a | 4123 | assert(s->pending || s->type == SOURCE_EXIT); |
fd38203a | 4124 | |
b778cba4 LP |
4125 | /* Save the event source type, here, so that we still know it after the event callback which might |
4126 | * invalidate the event. */ | |
8f5c235d LP |
4127 | saved_type = s->type; |
4128 | ||
de02634c | 4129 | /* Similarly, store a reference to the event loop object, so that we can still access it after the |
b778cba4 | 4130 | * callback might have invalidated/disconnected the event source. */ |
c8e9d15c YW |
4131 | saved_event = s->event; |
4132 | PROTECT_EVENT(saved_event); | |
b778cba4 | 4133 | |
de02634c | 4134 | /* Check if we hit the ratelimit for this event source, and if so, let's disable it. */ |
b6d5481b LP |
4135 | assert(!s->ratelimited); |
4136 | if (!ratelimit_below(&s->rate_limit)) { | |
4137 | r = event_source_enter_ratelimited(s); | |
4138 | if (r < 0) | |
4139 | return r; | |
4140 | ||
4141 | return 1; | |
4142 | } | |
4143 | ||
945c2931 | 4144 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { |
da7e457c LP |
4145 | r = source_set_pending(s, false); |
4146 | if (r < 0) | |
4147 | return r; | |
4148 | } | |
fd38203a | 4149 | |
6e9feda3 LP |
4150 | if (s->type != SOURCE_POST) { |
4151 | sd_event_source *z; | |
6e9feda3 | 4152 | |
de02634c | 4153 | /* If we execute a non-post source, let's mark all post sources as pending. */ |
6e9feda3 | 4154 | |
90e74a66 | 4155 | SET_FOREACH(z, s->event->post_sources) { |
b6d5481b | 4156 | if (event_source_is_offline(z)) |
6e9feda3 LP |
4157 | continue; |
4158 | ||
4159 | r = source_set_pending(z, true); | |
4160 | if (r < 0) | |
4161 | return r; | |
4162 | } | |
4163 | } | |
4164 | ||
158fe190 LP |
4165 | if (s->type == SOURCE_MEMORY_PRESSURE) { |
4166 | r = source_memory_pressure_initiate_dispatch(s); | |
4167 | if (r == -EIO) /* handle EIO errors similar to callback errors */ | |
4168 | goto finish; | |
4169 | if (r < 0) | |
4170 | return r; | |
4171 | if (r > 0) /* already handled */ | |
4172 | return 1; | |
4173 | } | |
4174 | ||
baf76283 LP |
4175 | if (s->enabled == SD_EVENT_ONESHOT) { |
4176 | r = sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
4177 | if (r < 0) |
4178 | return r; | |
4179 | } | |
4180 | ||
12179984 | 4181 | s->dispatching = true; |
b7484e2a | 4182 | |
fd38203a LP |
4183 | switch (s->type) { |
4184 | ||
4185 | case SOURCE_IO: | |
4186 | r = s->io.callback(s, s->io.fd, s->io.revents, s->userdata); | |
4187 | break; | |
4188 | ||
6a0f1f6d | 4189 | case SOURCE_TIME_REALTIME: |
a8548816 | 4190 | case SOURCE_TIME_BOOTTIME: |
6a0f1f6d LP |
4191 | case SOURCE_TIME_MONOTONIC: |
4192 | case SOURCE_TIME_REALTIME_ALARM: | |
4193 | case SOURCE_TIME_BOOTTIME_ALARM: | |
fd38203a LP |
4194 | r = s->time.callback(s, s->time.next, s->userdata); |
4195 | break; | |
4196 | ||
4197 | case SOURCE_SIGNAL: | |
4198 | r = s->signal.callback(s, &s->signal.siginfo, s->userdata); | |
4199 | break; | |
4200 | ||
08cd1552 LP |
4201 | case SOURCE_CHILD: { |
4202 | bool zombie; | |
4203 | ||
945c2931 | 4204 | zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 4205 | |
fd38203a | 4206 | r = s->child.callback(s, &s->child.siginfo, s->userdata); |
08cd1552 LP |
4207 | |
4208 | /* Now, reap the PID for good. */ | |
f8f3f926 | 4209 | if (zombie) { |
cc59d290 | 4210 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|WEXITED); |
f8f3f926 LP |
4211 | s->child.waited = true; |
4212 | } | |
08cd1552 | 4213 | |
fd38203a | 4214 | break; |
08cd1552 | 4215 | } |
fd38203a LP |
4216 | |
4217 | case SOURCE_DEFER: | |
4218 | r = s->defer.callback(s, s->userdata); | |
4219 | break; | |
da7e457c | 4220 | |
6e9feda3 LP |
4221 | case SOURCE_POST: |
4222 | r = s->post.callback(s, s->userdata); | |
4223 | break; | |
4224 | ||
6203e07a LP |
4225 | case SOURCE_EXIT: |
4226 | r = s->exit.callback(s, s->userdata); | |
da7e457c | 4227 | break; |
9d3e3aa5 | 4228 | |
97ef5391 LP |
4229 | case SOURCE_INOTIFY: { |
4230 | struct sd_event *e = s->event; | |
4231 | struct inotify_data *d; | |
4232 | size_t sz; | |
4233 | ||
4234 | assert(s->inotify.inode_data); | |
4235 | assert_se(d = s->inotify.inode_data->inotify_data); | |
4236 | ||
4237 | assert(d->buffer_filled >= offsetof(struct inotify_event, name)); | |
4238 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
4239 | assert(d->buffer_filled >= sz); | |
4240 | ||
53baf2ef LP |
4241 | /* If the inotify callback destroys the event source then this likely means we don't need to |
4242 | * watch the inode anymore, and thus also won't need the inotify object anymore. But if we'd | |
4243 | * free it immediately, then we couldn't drop the event from the inotify event queue without | |
4244 | * memory corruption anymore, as below. Hence, let's not free it immediately, but mark it | |
4245 | * "busy" with a counter (which will ensure it's not GC'ed away prematurely). Let's then | |
4246 | * explicitly GC it after we are done dropping the inotify event from the buffer. */ | |
4247 | d->n_busy++; | |
97ef5391 | 4248 | r = s->inotify.callback(s, &d->buffer.ev, s->userdata); |
53baf2ef | 4249 | d->n_busy--; |
97ef5391 | 4250 | |
53baf2ef LP |
4251 | /* When no event is pending anymore on this inotify object, then let's drop the event from |
4252 | * the inotify event queue buffer. */ | |
97ef5391 LP |
4253 | if (d->n_pending == 0) |
4254 | event_inotify_data_drop(e, d, sz); | |
4255 | ||
53baf2ef LP |
4256 | /* Now we don't want to access 'd' anymore, it's OK to GC now. */ |
4257 | event_gc_inotify_data(e, d); | |
97ef5391 LP |
4258 | break; |
4259 | } | |
4260 | ||
158fe190 LP |
4261 | case SOURCE_MEMORY_PRESSURE: |
4262 | r = s->memory_pressure.callback(s, s->userdata); | |
4263 | break; | |
4264 | ||
9d3e3aa5 | 4265 | case SOURCE_WATCHDOG: |
a71fe8b8 | 4266 | case _SOURCE_EVENT_SOURCE_TYPE_MAX: |
9f2a50a3 | 4267 | case _SOURCE_EVENT_SOURCE_TYPE_INVALID: |
04499a70 | 4268 | assert_not_reached(); |
fd38203a LP |
4269 | } |
4270 | ||
12179984 LP |
4271 | s->dispatching = false; |
4272 | ||
158fe190 | 4273 | finish: |
b778cba4 LP |
4274 | if (r < 0) { |
4275 | log_debug_errno(r, "Event source %s (type %s) returned error, %s: %m", | |
4276 | strna(s->description), | |
4277 | event_source_type_to_string(saved_type), | |
4278 | s->exit_on_failure ? "exiting" : "disabling"); | |
4279 | ||
4280 | if (s->exit_on_failure) | |
4281 | (void) sd_event_exit(saved_event, r); | |
4282 | } | |
12179984 LP |
4283 | |
4284 | if (s->n_ref == 0) | |
4285 | source_free(s); | |
4286 | else if (r < 0) | |
c3c50474 | 4287 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
b7484e2a | 4288 | |
6203e07a | 4289 | return 1; |
fd38203a LP |
4290 | } |
4291 | ||
4292 | static int event_prepare(sd_event *e) { | |
4293 | int r; | |
4294 | ||
4295 | assert(e); | |
4296 | ||
4297 | for (;;) { | |
4298 | sd_event_source *s; | |
4299 | ||
4300 | s = prioq_peek(e->prepare); | |
b6d5481b | 4301 | if (!s || s->prepare_iteration == e->iteration || event_source_is_offline(s)) |
fd38203a LP |
4302 | break; |
4303 | ||
4304 | s->prepare_iteration = e->iteration; | |
8656f4a6 | 4305 | prioq_reshuffle(e->prepare, s, &s->prepare_index); |
fd38203a LP |
4306 | |
4307 | assert(s->prepare); | |
12179984 | 4308 | s->dispatching = true; |
fd38203a | 4309 | r = s->prepare(s, s->userdata); |
12179984 LP |
4310 | s->dispatching = false; |
4311 | ||
b778cba4 LP |
4312 | if (r < 0) { |
4313 | log_debug_errno(r, "Prepare callback of event source %s (type %s) returned error, %s: %m", | |
4314 | strna(s->description), | |
4315 | event_source_type_to_string(s->type), | |
4316 | s->exit_on_failure ? "exiting" : "disabling"); | |
4317 | ||
4318 | if (s->exit_on_failure) | |
4319 | (void) sd_event_exit(e, r); | |
4320 | } | |
fd38203a | 4321 | |
12179984 LP |
4322 | if (s->n_ref == 0) |
4323 | source_free(s); | |
4324 | else if (r < 0) | |
c3c50474 | 4325 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
fd38203a LP |
4326 | } |
4327 | ||
4328 | return 0; | |
4329 | } | |
4330 | ||
6203e07a | 4331 | static int dispatch_exit(sd_event *e) { |
da7e457c LP |
4332 | sd_event_source *p; |
4333 | int r; | |
4334 | ||
4335 | assert(e); | |
4336 | ||
6203e07a | 4337 | p = prioq_peek(e->exit); |
19947509 ZJS |
4338 | assert(!p || p->type == SOURCE_EXIT); |
4339 | ||
b6d5481b | 4340 | if (!p || event_source_is_offline(p)) { |
da7e457c LP |
4341 | e->state = SD_EVENT_FINISHED; |
4342 | return 0; | |
4343 | } | |
4344 | ||
c8e9d15c | 4345 | PROTECT_EVENT(e); |
da7e457c | 4346 | e->iteration++; |
6203e07a | 4347 | e->state = SD_EVENT_EXITING; |
da7e457c | 4348 | r = source_dispatch(p); |
2b0c9ef7 | 4349 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
4350 | return r; |
4351 | } | |
4352 | ||
c2ba3ad6 LP |
4353 | static sd_event_source* event_next_pending(sd_event *e) { |
4354 | sd_event_source *p; | |
4355 | ||
da7e457c LP |
4356 | assert(e); |
4357 | ||
c2ba3ad6 LP |
4358 | p = prioq_peek(e->pending); |
4359 | if (!p) | |
4360 | return NULL; | |
4361 | ||
b6d5481b | 4362 | if (event_source_is_offline(p)) |
c2ba3ad6 LP |
4363 | return NULL; |
4364 | ||
4365 | return p; | |
4366 | } | |
4367 | ||
cde93897 LP |
4368 | static int arm_watchdog(sd_event *e) { |
4369 | struct itimerspec its = {}; | |
4370 | usec_t t; | |
cde93897 LP |
4371 | |
4372 | assert(e); | |
4373 | assert(e->watchdog_fd >= 0); | |
4374 | ||
4375 | t = sleep_between(e, | |
a595fb5c YW |
4376 | usec_add(e->watchdog_last, (e->watchdog_period / 2)), |
4377 | usec_add(e->watchdog_last, (e->watchdog_period * 3 / 4))); | |
cde93897 LP |
4378 | |
4379 | timespec_store(&its.it_value, t); | |
4380 | ||
75145780 LP |
4381 | /* Make sure we never set the watchdog to 0, which tells the |
4382 | * kernel to disable it. */ | |
4383 | if (its.it_value.tv_sec == 0 && its.it_value.tv_nsec == 0) | |
4384 | its.it_value.tv_nsec = 1; | |
4385 | ||
7c248223 | 4386 | return RET_NERRNO(timerfd_settime(e->watchdog_fd, TFD_TIMER_ABSTIME, &its, NULL)); |
cde93897 LP |
4387 | } |
4388 | ||
4389 | static int process_watchdog(sd_event *e) { | |
4390 | assert(e); | |
4391 | ||
4392 | if (!e->watchdog) | |
4393 | return 0; | |
4394 | ||
4395 | /* Don't notify watchdog too often */ | |
4396 | if (e->watchdog_last + e->watchdog_period / 4 > e->timestamp.monotonic) | |
4397 | return 0; | |
4398 | ||
4399 | sd_notify(false, "WATCHDOG=1"); | |
4400 | e->watchdog_last = e->timestamp.monotonic; | |
4401 | ||
4402 | return arm_watchdog(e); | |
4403 | } | |
4404 | ||
97ef5391 LP |
4405 | static void event_close_inode_data_fds(sd_event *e) { |
4406 | struct inode_data *d; | |
4407 | ||
4408 | assert(e); | |
4409 | ||
4410 | /* Close the fds pointing to the inodes to watch now. We need to close them as they might otherwise pin | |
4411 | * filesystems. But we can't close them right-away as we need them as long as the user still wants to make | |
365c2885 | 4412 | * adjustments to the event source, such as changing the priority (which requires us to remove and re-add a watch |
97ef5391 LP |
4413 | * for the inode). Hence, let's close them when entering the first iteration after they were added, as a |
4414 | * compromise. */ | |
4415 | ||
ed828563 | 4416 | while ((d = e->inode_data_to_close_list)) { |
97ef5391 LP |
4417 | assert(d->fd >= 0); |
4418 | d->fd = safe_close(d->fd); | |
4419 | ||
ed828563 | 4420 | LIST_REMOVE(to_close, e->inode_data_to_close_list, d); |
97ef5391 LP |
4421 | } |
4422 | } | |
4423 | ||
158fe190 LP |
4424 | static int event_memory_pressure_write_list(sd_event *e) { |
4425 | int r; | |
4426 | ||
4427 | assert(e); | |
4428 | ||
4429 | for (;;) { | |
4430 | sd_event_source *s; | |
4431 | ||
4432 | s = LIST_POP(memory_pressure.write_list, e->memory_pressure_write_list); | |
4433 | if (!s) | |
4434 | break; | |
4435 | ||
4436 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
4437 | assert(s->memory_pressure.write_buffer_size > 0); | |
4438 | s->memory_pressure.in_write_list = false; | |
4439 | ||
4440 | r = source_memory_pressure_write(s); | |
4441 | if (r < 0) | |
4442 | return r; | |
4443 | } | |
4444 | ||
4445 | return 0; | |
4446 | } | |
4447 | ||
c45a5a74 TG |
4448 | _public_ int sd_event_prepare(sd_event *e) { |
4449 | int r; | |
fd38203a | 4450 | |
da7e457c | 4451 | assert_return(e, -EINVAL); |
b937d761 | 4452 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 4453 | assert_return(!event_origin_changed(e), -ECHILD); |
da7e457c | 4454 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2b0c9ef7 | 4455 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 4456 | |
e5446015 LP |
4457 | /* Let's check that if we are a default event loop we are executed in the correct thread. We only do |
4458 | * this check here once, since gettid() is typically not cached, and thus want to minimize | |
4459 | * syscalls */ | |
4460 | assert_return(!e->default_event_ptr || e->tid == gettid(), -EREMOTEIO); | |
4461 | ||
f814c871 | 4462 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
c8e9d15c | 4463 | PROTECT_EVENT(e); |
f814c871 | 4464 | |
6203e07a | 4465 | if (e->exit_requested) |
c45a5a74 | 4466 | goto pending; |
fd38203a LP |
4467 | |
4468 | e->iteration++; | |
4469 | ||
0be6c2f6 | 4470 | e->state = SD_EVENT_PREPARING; |
fd38203a | 4471 | r = event_prepare(e); |
0be6c2f6 | 4472 | e->state = SD_EVENT_INITIAL; |
fd38203a | 4473 | if (r < 0) |
c45a5a74 | 4474 | return r; |
fd38203a | 4475 | |
158fe190 LP |
4476 | r = event_memory_pressure_write_list(e); |
4477 | if (r < 0) | |
4478 | return r; | |
4479 | ||
6a0f1f6d LP |
4480 | r = event_arm_timer(e, &e->realtime); |
4481 | if (r < 0) | |
c45a5a74 | 4482 | return r; |
6a0f1f6d | 4483 | |
a8548816 TG |
4484 | r = event_arm_timer(e, &e->boottime); |
4485 | if (r < 0) | |
c45a5a74 | 4486 | return r; |
a8548816 | 4487 | |
6a0f1f6d LP |
4488 | r = event_arm_timer(e, &e->monotonic); |
4489 | if (r < 0) | |
c45a5a74 | 4490 | return r; |
6a0f1f6d LP |
4491 | |
4492 | r = event_arm_timer(e, &e->realtime_alarm); | |
1b5995b0 | 4493 | if (r < 0) |
c45a5a74 | 4494 | return r; |
fd38203a | 4495 | |
6a0f1f6d | 4496 | r = event_arm_timer(e, &e->boottime_alarm); |
1b5995b0 | 4497 | if (r < 0) |
c45a5a74 | 4498 | return r; |
fd38203a | 4499 | |
97ef5391 LP |
4500 | event_close_inode_data_fds(e); |
4501 | ||
0601b958 | 4502 | if (event_next_pending(e) || e->need_process_child || e->buffered_inotify_data_list) |
c45a5a74 TG |
4503 | goto pending; |
4504 | ||
2b0c9ef7 | 4505 | e->state = SD_EVENT_ARMED; |
c45a5a74 TG |
4506 | |
4507 | return 0; | |
4508 | ||
4509 | pending: | |
2b0c9ef7 | 4510 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
4511 | r = sd_event_wait(e, 0); |
4512 | if (r == 0) | |
2b0c9ef7 | 4513 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
4514 | |
4515 | return r; | |
c45a5a74 TG |
4516 | } |
4517 | ||
798445ab LP |
4518 | static int epoll_wait_usec( |
4519 | int fd, | |
4520 | struct epoll_event *events, | |
4521 | int maxevents, | |
4522 | usec_t timeout) { | |
4523 | ||
7c248223 | 4524 | int msec; |
0c14c45e LP |
4525 | /* A wrapper that uses epoll_pwait2() if available, and falls back to epoll_wait() if not. */ |
4526 | ||
4527 | #if HAVE_EPOLL_PWAIT2 | |
39f756d3 | 4528 | static bool epoll_pwait2_absent = false; |
52bb308c | 4529 | int r; |
798445ab | 4530 | |
0c14c45e LP |
4531 | /* epoll_pwait2() was added to Linux 5.11 (2021-02-14) and to glibc in 2.35 (2022-02-03). In contrast |
4532 | * to other syscalls we don't bother with our own fallback syscall wrappers on old libcs, since this | |
4533 | * is not that obvious to implement given the libc and kernel definitions differ in the last | |
4534 | * argument. Moreover, the only reason to use it is the more accurate time-outs (which is not a | |
4535 | * biggie), let's hence rely on glibc's definitions, and fallback to epoll_pwait() when that's | |
4536 | * missing. */ | |
798445ab LP |
4537 | |
4538 | if (!epoll_pwait2_absent && timeout != USEC_INFINITY) { | |
798445ab LP |
4539 | r = epoll_pwait2(fd, |
4540 | events, | |
4541 | maxevents, | |
52bb308c | 4542 | TIMESPEC_STORE(timeout), |
798445ab LP |
4543 | NULL); |
4544 | if (r >= 0) | |
4545 | return r; | |
7cb45dbf | 4546 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) |
798445ab LP |
4547 | return -errno; /* Only fallback to old epoll_wait() if the syscall is masked or not |
4548 | * supported. */ | |
4549 | ||
4550 | epoll_pwait2_absent = true; | |
4551 | } | |
39f756d3 | 4552 | #endif |
798445ab LP |
4553 | |
4554 | if (timeout == USEC_INFINITY) | |
4555 | msec = -1; | |
4556 | else { | |
4557 | usec_t k; | |
4558 | ||
4559 | k = DIV_ROUND_UP(timeout, USEC_PER_MSEC); | |
4560 | if (k >= INT_MAX) | |
4561 | msec = INT_MAX; /* Saturate */ | |
4562 | else | |
4563 | msec = (int) k; | |
4564 | } | |
4565 | ||
7c248223 | 4566 | return RET_NERRNO(epoll_wait(fd, events, maxevents, msec)); |
798445ab LP |
4567 | } |
4568 | ||
efd3be9d | 4569 | static int process_epoll(sd_event *e, usec_t timeout, int64_t threshold, int64_t *ret_min_priority) { |
319a4f4b | 4570 | size_t n_event_queue, m, n_event_max; |
efd3be9d YW |
4571 | int64_t min_priority = threshold; |
4572 | bool something_new = false; | |
798445ab | 4573 | int r; |
c45a5a74 | 4574 | |
efd3be9d YW |
4575 | assert(e); |
4576 | assert(ret_min_priority); | |
6a0f1f6d | 4577 | |
8b9708d1 | 4578 | n_event_queue = MAX(e->n_sources, 1u); |
319a4f4b | 4579 | if (!GREEDY_REALLOC(e->event_queue, n_event_queue)) |
5cddd924 | 4580 | return -ENOMEM; |
fd38203a | 4581 | |
319a4f4b LP |
4582 | n_event_max = MALLOC_ELEMENTSOF(e->event_queue); |
4583 | ||
97ef5391 | 4584 | /* If we still have inotify data buffered, then query the other fds, but don't wait on it */ |
0601b958 | 4585 | if (e->buffered_inotify_data_list) |
798445ab | 4586 | timeout = 0; |
97ef5391 | 4587 | |
8b9708d1 | 4588 | for (;;) { |
319a4f4b LP |
4589 | r = epoll_wait_usec( |
4590 | e->epoll_fd, | |
4591 | e->event_queue, | |
4592 | n_event_max, | |
4593 | timeout); | |
798445ab | 4594 | if (r < 0) |
efd3be9d | 4595 | return r; |
c45a5a74 | 4596 | |
8b9708d1 YW |
4597 | m = (size_t) r; |
4598 | ||
319a4f4b | 4599 | if (m < n_event_max) |
8b9708d1 YW |
4600 | break; |
4601 | ||
319a4f4b | 4602 | if (n_event_max >= n_event_queue * 10) |
8b9708d1 YW |
4603 | break; |
4604 | ||
319a4f4b | 4605 | if (!GREEDY_REALLOC(e->event_queue, n_event_max + n_event_queue)) |
8b9708d1 YW |
4606 | return -ENOMEM; |
4607 | ||
319a4f4b | 4608 | n_event_max = MALLOC_ELEMENTSOF(e->event_queue); |
798445ab | 4609 | timeout = 0; |
da7e457c | 4610 | } |
fd38203a | 4611 | |
efd3be9d YW |
4612 | /* Set timestamp only when this is called first time. */ |
4613 | if (threshold == INT64_MAX) | |
4614 | triple_timestamp_get(&e->timestamp); | |
fd38203a | 4615 | |
8b9708d1 | 4616 | for (size_t i = 0; i < m; i++) { |
fd38203a | 4617 | |
5cddd924 LP |
4618 | if (e->event_queue[i].data.ptr == INT_TO_PTR(SOURCE_WATCHDOG)) |
4619 | r = flush_timer(e, e->watchdog_fd, e->event_queue[i].events, NULL); | |
9da4cb2b | 4620 | else { |
5cddd924 | 4621 | WakeupType *t = e->event_queue[i].data.ptr; |
9da4cb2b LP |
4622 | |
4623 | switch (*t) { | |
4624 | ||
f8f3f926 | 4625 | case WAKEUP_EVENT_SOURCE: { |
5cddd924 | 4626 | sd_event_source *s = e->event_queue[i].data.ptr; |
f8f3f926 LP |
4627 | |
4628 | assert(s); | |
4629 | ||
efd3be9d YW |
4630 | if (s->priority > threshold) |
4631 | continue; | |
4632 | ||
4633 | min_priority = MIN(min_priority, s->priority); | |
4634 | ||
f8f3f926 LP |
4635 | switch (s->type) { |
4636 | ||
4637 | case SOURCE_IO: | |
5cddd924 | 4638 | r = process_io(e, s, e->event_queue[i].events); |
f8f3f926 LP |
4639 | break; |
4640 | ||
4641 | case SOURCE_CHILD: | |
5cddd924 | 4642 | r = process_pidfd(e, s, e->event_queue[i].events); |
f8f3f926 LP |
4643 | break; |
4644 | ||
158fe190 LP |
4645 | case SOURCE_MEMORY_PRESSURE: |
4646 | r = process_memory_pressure(s, e->event_queue[i].events); | |
4647 | break; | |
4648 | ||
f8f3f926 | 4649 | default: |
04499a70 | 4650 | assert_not_reached(); |
f8f3f926 LP |
4651 | } |
4652 | ||
9da4cb2b | 4653 | break; |
f8f3f926 | 4654 | } |
fd38203a | 4655 | |
9da4cb2b | 4656 | case WAKEUP_CLOCK_DATA: { |
5cddd924 | 4657 | struct clock_data *d = e->event_queue[i].data.ptr; |
f8f3f926 LP |
4658 | |
4659 | assert(d); | |
4660 | ||
5cddd924 | 4661 | r = flush_timer(e, d->fd, e->event_queue[i].events, &d->next); |
9da4cb2b LP |
4662 | break; |
4663 | } | |
4664 | ||
4665 | case WAKEUP_SIGNAL_DATA: | |
efd3be9d | 4666 | r = process_signal(e, e->event_queue[i].data.ptr, e->event_queue[i].events, &min_priority); |
9da4cb2b LP |
4667 | break; |
4668 | ||
97ef5391 | 4669 | case WAKEUP_INOTIFY_DATA: |
efd3be9d | 4670 | r = event_inotify_data_read(e, e->event_queue[i].data.ptr, e->event_queue[i].events, threshold); |
97ef5391 LP |
4671 | break; |
4672 | ||
9da4cb2b | 4673 | default: |
04499a70 | 4674 | assert_not_reached(); |
9da4cb2b LP |
4675 | } |
4676 | } | |
efd3be9d YW |
4677 | if (r < 0) |
4678 | return r; | |
4679 | if (r > 0) | |
4680 | something_new = true; | |
4681 | } | |
4682 | ||
4683 | *ret_min_priority = min_priority; | |
4684 | return something_new; | |
4685 | } | |
4686 | ||
4687 | _public_ int sd_event_wait(sd_event *e, uint64_t timeout) { | |
4688 | int r; | |
4689 | ||
4690 | assert_return(e, -EINVAL); | |
4691 | assert_return(e = event_resolve(e), -ENOPKG); | |
2eeff0f4 | 4692 | assert_return(!event_origin_changed(e), -ECHILD); |
efd3be9d YW |
4693 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
4694 | assert_return(e->state == SD_EVENT_ARMED, -EBUSY); | |
4695 | ||
4696 | if (e->exit_requested) { | |
4697 | e->state = SD_EVENT_PENDING; | |
4698 | return 1; | |
4699 | } | |
4700 | ||
4701 | for (int64_t threshold = INT64_MAX; ; threshold--) { | |
4702 | int64_t epoll_min_priority, child_min_priority; | |
4703 | ||
4704 | /* There may be a possibility that new epoll (especially IO) and child events are | |
4705 | * triggered just after process_epoll() call but before process_child(), and the new IO | |
4706 | * events may have higher priority than the child events. To salvage these events, | |
4707 | * let's call epoll_wait() again, but accepts only events with higher priority than the | |
4708 | * previous. See issue https://github.com/systemd/systemd/issues/18190 and comments | |
4709 | * https://github.com/systemd/systemd/pull/18750#issuecomment-785801085 | |
4710 | * https://github.com/systemd/systemd/pull/18922#issuecomment-792825226 */ | |
4711 | ||
4712 | r = process_epoll(e, timeout, threshold, &epoll_min_priority); | |
4713 | if (r == -EINTR) { | |
4714 | e->state = SD_EVENT_PENDING; | |
4715 | return 1; | |
4716 | } | |
4717 | if (r < 0) | |
4718 | goto finish; | |
4719 | if (r == 0 && threshold < INT64_MAX) | |
4720 | /* No new epoll event. */ | |
4721 | break; | |
4722 | ||
4723 | r = process_child(e, threshold, &child_min_priority); | |
fd38203a | 4724 | if (r < 0) |
da7e457c | 4725 | goto finish; |
efd3be9d YW |
4726 | if (r == 0) |
4727 | /* No new child event. */ | |
4728 | break; | |
4729 | ||
4730 | threshold = MIN(epoll_min_priority, child_min_priority); | |
4731 | if (threshold == INT64_MIN) | |
4732 | break; | |
4733 | ||
4734 | timeout = 0; | |
fd38203a LP |
4735 | } |
4736 | ||
cde93897 LP |
4737 | r = process_watchdog(e); |
4738 | if (r < 0) | |
4739 | goto finish; | |
4740 | ||
fd69f224 | 4741 | r = process_inotify(e); |
6a0f1f6d LP |
4742 | if (r < 0) |
4743 | goto finish; | |
4744 | ||
fd69f224 | 4745 | r = process_timer(e, e->timestamp.realtime, &e->realtime); |
a8548816 TG |
4746 | if (r < 0) |
4747 | goto finish; | |
4748 | ||
fd69f224 | 4749 | r = process_timer(e, e->timestamp.boottime, &e->boottime); |
6a0f1f6d LP |
4750 | if (r < 0) |
4751 | goto finish; | |
4752 | ||
4753 | r = process_timer(e, e->timestamp.realtime, &e->realtime_alarm); | |
fd38203a | 4754 | if (r < 0) |
da7e457c | 4755 | goto finish; |
fd38203a | 4756 | |
e475d10c | 4757 | r = process_timer(e, e->timestamp.boottime, &e->boottime_alarm); |
fd38203a | 4758 | if (r < 0) |
da7e457c | 4759 | goto finish; |
fd38203a | 4760 | |
fd69f224 | 4761 | r = process_timer(e, e->timestamp.monotonic, &e->monotonic); |
97ef5391 LP |
4762 | if (r < 0) |
4763 | goto finish; | |
fd69f224 MS |
4764 | else if (r == 1) { |
4765 | /* Ratelimit expiry callback was called. Let's postpone processing pending sources and | |
4766 | * put loop in the initial state in order to evaluate (in the next iteration) also sources | |
4767 | * there were potentially re-enabled by the callback. | |
4768 | * | |
4769 | * Wondering why we treat only this invocation of process_timer() differently? Once event | |
4770 | * source is ratelimited we essentially transform it into CLOCK_MONOTONIC timer hence | |
4771 | * ratelimit expiry callback is never called for any other timer type. */ | |
4772 | r = 0; | |
4773 | goto finish; | |
4774 | } | |
97ef5391 | 4775 | |
c45a5a74 TG |
4776 | if (event_next_pending(e)) { |
4777 | e->state = SD_EVENT_PENDING; | |
c45a5a74 | 4778 | return 1; |
da7e457c LP |
4779 | } |
4780 | ||
c45a5a74 | 4781 | r = 0; |
fd38203a | 4782 | |
da7e457c | 4783 | finish: |
2b0c9ef7 | 4784 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
4785 | |
4786 | return r; | |
fd38203a LP |
4787 | } |
4788 | ||
c45a5a74 TG |
4789 | _public_ int sd_event_dispatch(sd_event *e) { |
4790 | sd_event_source *p; | |
4791 | int r; | |
4792 | ||
4793 | assert_return(e, -EINVAL); | |
b937d761 | 4794 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 4795 | assert_return(!event_origin_changed(e), -ECHILD); |
c45a5a74 TG |
4796 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
4797 | assert_return(e->state == SD_EVENT_PENDING, -EBUSY); | |
4798 | ||
4799 | if (e->exit_requested) | |
4800 | return dispatch_exit(e); | |
4801 | ||
4802 | p = event_next_pending(e); | |
4803 | if (p) { | |
c8e9d15c | 4804 | PROTECT_EVENT(e); |
c45a5a74 TG |
4805 | |
4806 | e->state = SD_EVENT_RUNNING; | |
4807 | r = source_dispatch(p); | |
2b0c9ef7 | 4808 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
4809 | return r; |
4810 | } | |
4811 | ||
2b0c9ef7 | 4812 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
4813 | |
4814 | return 1; | |
4815 | } | |
4816 | ||
34b87517 | 4817 | static void event_log_delays(sd_event *e) { |
442ac269 YW |
4818 | char b[ELEMENTSOF(e->delays) * DECIMAL_STR_MAX(unsigned) + 1], *p; |
4819 | size_t l, i; | |
34b87517 | 4820 | |
442ac269 YW |
4821 | p = b; |
4822 | l = sizeof(b); | |
4823 | for (i = 0; i < ELEMENTSOF(e->delays); i++) { | |
4824 | l = strpcpyf(&p, l, "%u ", e->delays[i]); | |
34b87517 VC |
4825 | e->delays[i] = 0; |
4826 | } | |
442ac269 | 4827 | log_debug("Event loop iterations: %s", b); |
34b87517 VC |
4828 | } |
4829 | ||
c45a5a74 TG |
4830 | _public_ int sd_event_run(sd_event *e, uint64_t timeout) { |
4831 | int r; | |
4832 | ||
4833 | assert_return(e, -EINVAL); | |
b937d761 | 4834 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 4835 | assert_return(!event_origin_changed(e), -ECHILD); |
c45a5a74 | 4836 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2b0c9ef7 | 4837 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
c45a5a74 | 4838 | |
e6a7bee5 | 4839 | if (e->profile_delays && e->last_run_usec != 0) { |
34b87517 VC |
4840 | usec_t this_run; |
4841 | unsigned l; | |
4842 | ||
4843 | this_run = now(CLOCK_MONOTONIC); | |
4844 | ||
58c34be8 | 4845 | l = log2u64(this_run - e->last_run_usec); |
cb9d621e | 4846 | assert(l < ELEMENTSOF(e->delays)); |
34b87517 VC |
4847 | e->delays[l]++; |
4848 | ||
e6a7bee5 | 4849 | if (this_run - e->last_log_usec >= 5*USEC_PER_SEC) { |
34b87517 | 4850 | event_log_delays(e); |
e6a7bee5 | 4851 | e->last_log_usec = this_run; |
34b87517 VC |
4852 | } |
4853 | } | |
4854 | ||
f814c871 | 4855 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
c8e9d15c | 4856 | PROTECT_EVENT(e); |
f814c871 | 4857 | |
c45a5a74 | 4858 | r = sd_event_prepare(e); |
53bac4e0 LP |
4859 | if (r == 0) |
4860 | /* There was nothing? Then wait... */ | |
4861 | r = sd_event_wait(e, timeout); | |
c45a5a74 | 4862 | |
34b87517 | 4863 | if (e->profile_delays) |
e6a7bee5 | 4864 | e->last_run_usec = now(CLOCK_MONOTONIC); |
34b87517 | 4865 | |
02d30981 | 4866 | if (r > 0) { |
53bac4e0 | 4867 | /* There's something now, then let's dispatch it */ |
02d30981 TG |
4868 | r = sd_event_dispatch(e); |
4869 | if (r < 0) | |
4870 | return r; | |
53bac4e0 LP |
4871 | |
4872 | return 1; | |
4873 | } | |
4874 | ||
4875 | return r; | |
c45a5a74 TG |
4876 | } |
4877 | ||
f7262a9f | 4878 | _public_ int sd_event_loop(sd_event *e) { |
fd38203a LP |
4879 | int r; |
4880 | ||
da7e457c | 4881 | assert_return(e, -EINVAL); |
b937d761 | 4882 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 4883 | assert_return(!event_origin_changed(e), -ECHILD); |
2b0c9ef7 | 4884 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 4885 | |
2eeff0f4 | 4886 | |
c8e9d15c | 4887 | PROTECT_EVENT(e); |
fd38203a | 4888 | |
da7e457c | 4889 | while (e->state != SD_EVENT_FINISHED) { |
f5fbe71d | 4890 | r = sd_event_run(e, UINT64_MAX); |
fd38203a | 4891 | if (r < 0) |
30dd293c | 4892 | return r; |
fd38203a LP |
4893 | } |
4894 | ||
30dd293c | 4895 | return e->exit_code; |
fd38203a LP |
4896 | } |
4897 | ||
9b364545 | 4898 | _public_ int sd_event_get_fd(sd_event *e) { |
9b364545 | 4899 | assert_return(e, -EINVAL); |
b937d761 | 4900 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 4901 | assert_return(!event_origin_changed(e), -ECHILD); |
9b364545 TG |
4902 | |
4903 | return e->epoll_fd; | |
4904 | } | |
4905 | ||
f7262a9f | 4906 | _public_ int sd_event_get_state(sd_event *e) { |
da7e457c | 4907 | assert_return(e, -EINVAL); |
b937d761 | 4908 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 4909 | assert_return(!event_origin_changed(e), -ECHILD); |
da7e457c LP |
4910 | |
4911 | return e->state; | |
4912 | } | |
4913 | ||
6203e07a | 4914 | _public_ int sd_event_get_exit_code(sd_event *e, int *code) { |
da7e457c | 4915 | assert_return(e, -EINVAL); |
b937d761 | 4916 | assert_return(e = event_resolve(e), -ENOPKG); |
6203e07a | 4917 | assert_return(code, -EINVAL); |
2eeff0f4 | 4918 | assert_return(!event_origin_changed(e), -ECHILD); |
fd38203a | 4919 | |
6203e07a LP |
4920 | if (!e->exit_requested) |
4921 | return -ENODATA; | |
4922 | ||
4923 | *code = e->exit_code; | |
4924 | return 0; | |
fd38203a LP |
4925 | } |
4926 | ||
6203e07a | 4927 | _public_ int sd_event_exit(sd_event *e, int code) { |
da7e457c | 4928 | assert_return(e, -EINVAL); |
b937d761 | 4929 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 4930 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 4931 | assert_return(!event_origin_changed(e), -ECHILD); |
fd38203a | 4932 | |
6203e07a LP |
4933 | e->exit_requested = true; |
4934 | e->exit_code = code; | |
4935 | ||
fd38203a LP |
4936 | return 0; |
4937 | } | |
46e8c825 | 4938 | |
6a0f1f6d | 4939 | _public_ int sd_event_now(sd_event *e, clockid_t clock, uint64_t *usec) { |
46e8c825 | 4940 | assert_return(e, -EINVAL); |
b937d761 | 4941 | assert_return(e = event_resolve(e), -ENOPKG); |
46e8c825 | 4942 | assert_return(usec, -EINVAL); |
2eeff0f4 | 4943 | assert_return(!event_origin_changed(e), -ECHILD); |
46e8c825 | 4944 | |
e475d10c LP |
4945 | if (!TRIPLE_TIMESTAMP_HAS_CLOCK(clock)) |
4946 | return -EOPNOTSUPP; | |
4947 | ||
e475d10c | 4948 | if (!triple_timestamp_is_set(&e->timestamp)) { |
15c689d7 | 4949 | /* Implicitly fall back to now() if we never ran before and thus have no cached time. */ |
38a03f06 LP |
4950 | *usec = now(clock); |
4951 | return 1; | |
4952 | } | |
46e8c825 | 4953 | |
e475d10c | 4954 | *usec = triple_timestamp_by_clock(&e->timestamp, clock); |
46e8c825 LP |
4955 | return 0; |
4956 | } | |
afc6adb5 LP |
4957 | |
4958 | _public_ int sd_event_default(sd_event **ret) { | |
39883f62 | 4959 | sd_event *e = NULL; |
afc6adb5 LP |
4960 | int r; |
4961 | ||
4962 | if (!ret) | |
4963 | return !!default_event; | |
4964 | ||
4965 | if (default_event) { | |
4966 | *ret = sd_event_ref(default_event); | |
4967 | return 0; | |
4968 | } | |
4969 | ||
4970 | r = sd_event_new(&e); | |
4971 | if (r < 0) | |
4972 | return r; | |
4973 | ||
4974 | e->default_event_ptr = &default_event; | |
4975 | e->tid = gettid(); | |
4976 | default_event = e; | |
4977 | ||
4978 | *ret = e; | |
4979 | return 1; | |
4980 | } | |
4981 | ||
4982 | _public_ int sd_event_get_tid(sd_event *e, pid_t *tid) { | |
4983 | assert_return(e, -EINVAL); | |
b937d761 | 4984 | assert_return(e = event_resolve(e), -ENOPKG); |
afc6adb5 | 4985 | assert_return(tid, -EINVAL); |
2eeff0f4 | 4986 | assert_return(!event_origin_changed(e), -ECHILD); |
afc6adb5 | 4987 | |
76b54375 LP |
4988 | if (e->tid != 0) { |
4989 | *tid = e->tid; | |
4990 | return 0; | |
4991 | } | |
4992 | ||
4993 | return -ENXIO; | |
afc6adb5 | 4994 | } |
cde93897 LP |
4995 | |
4996 | _public_ int sd_event_set_watchdog(sd_event *e, int b) { | |
4997 | int r; | |
4998 | ||
4999 | assert_return(e, -EINVAL); | |
b937d761 | 5000 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 5001 | assert_return(!event_origin_changed(e), -ECHILD); |
cde93897 LP |
5002 | |
5003 | if (e->watchdog == !!b) | |
5004 | return e->watchdog; | |
5005 | ||
5006 | if (b) { | |
09812eb7 LP |
5007 | r = sd_watchdog_enabled(false, &e->watchdog_period); |
5008 | if (r <= 0) | |
cde93897 | 5009 | return r; |
cde93897 LP |
5010 | |
5011 | /* Issue first ping immediately */ | |
5012 | sd_notify(false, "WATCHDOG=1"); | |
5013 | e->watchdog_last = now(CLOCK_MONOTONIC); | |
5014 | ||
5015 | e->watchdog_fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC); | |
5016 | if (e->watchdog_fd < 0) | |
5017 | return -errno; | |
5018 | ||
5019 | r = arm_watchdog(e); | |
5020 | if (r < 0) | |
5021 | goto fail; | |
5022 | ||
1eac7948 | 5023 | struct epoll_event ev = { |
a82f89aa LP |
5024 | .events = EPOLLIN, |
5025 | .data.ptr = INT_TO_PTR(SOURCE_WATCHDOG), | |
5026 | }; | |
cde93897 | 5027 | |
15c689d7 | 5028 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, e->watchdog_fd, &ev) < 0) { |
cde93897 LP |
5029 | r = -errno; |
5030 | goto fail; | |
5031 | } | |
5032 | ||
5033 | } else { | |
5034 | if (e->watchdog_fd >= 0) { | |
5a795bff | 5035 | (void) epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, e->watchdog_fd, NULL); |
03e334a1 | 5036 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
5037 | } |
5038 | } | |
5039 | ||
5040 | e->watchdog = !!b; | |
5041 | return e->watchdog; | |
5042 | ||
5043 | fail: | |
03e334a1 | 5044 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
5045 | return r; |
5046 | } | |
8f726607 LP |
5047 | |
5048 | _public_ int sd_event_get_watchdog(sd_event *e) { | |
5049 | assert_return(e, -EINVAL); | |
b937d761 | 5050 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 5051 | assert_return(!event_origin_changed(e), -ECHILD); |
8f726607 LP |
5052 | |
5053 | return e->watchdog; | |
5054 | } | |
60a3b1e1 LP |
5055 | |
5056 | _public_ int sd_event_get_iteration(sd_event *e, uint64_t *ret) { | |
5057 | assert_return(e, -EINVAL); | |
b937d761 | 5058 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 5059 | assert_return(!event_origin_changed(e), -ECHILD); |
60a3b1e1 LP |
5060 | |
5061 | *ret = e->iteration; | |
5062 | return 0; | |
5063 | } | |
15723a1d LP |
5064 | |
5065 | _public_ int sd_event_source_set_destroy_callback(sd_event_source *s, sd_event_destroy_t callback) { | |
5066 | assert_return(s, -EINVAL); | |
2eeff0f4 LB |
5067 | assert_return(s->event, -EINVAL); |
5068 | assert_return(!event_origin_changed(s->event), -ECHILD); | |
15723a1d LP |
5069 | |
5070 | s->destroy_callback = callback; | |
5071 | return 0; | |
5072 | } | |
5073 | ||
5074 | _public_ int sd_event_source_get_destroy_callback(sd_event_source *s, sd_event_destroy_t *ret) { | |
5075 | assert_return(s, -EINVAL); | |
2eeff0f4 | 5076 | assert_return(!event_origin_changed(s->event), -ECHILD); |
15723a1d LP |
5077 | |
5078 | if (ret) | |
5079 | *ret = s->destroy_callback; | |
5080 | ||
5081 | return !!s->destroy_callback; | |
5082 | } | |
2382c936 YW |
5083 | |
5084 | _public_ int sd_event_source_get_floating(sd_event_source *s) { | |
5085 | assert_return(s, -EINVAL); | |
2eeff0f4 | 5086 | assert_return(!event_origin_changed(s->event), -ECHILD); |
2382c936 YW |
5087 | |
5088 | return s->floating; | |
5089 | } | |
5090 | ||
5091 | _public_ int sd_event_source_set_floating(sd_event_source *s, int b) { | |
5092 | assert_return(s, -EINVAL); | |
2eeff0f4 | 5093 | assert_return(!event_origin_changed(s->event), -ECHILD); |
2382c936 YW |
5094 | |
5095 | if (s->floating == !!b) | |
5096 | return 0; | |
5097 | ||
5098 | if (!s->event) /* Already disconnected */ | |
5099 | return -ESTALE; | |
5100 | ||
5101 | s->floating = b; | |
5102 | ||
5103 | if (b) { | |
5104 | sd_event_source_ref(s); | |
5105 | sd_event_unref(s->event); | |
5106 | } else { | |
5107 | sd_event_ref(s->event); | |
5108 | sd_event_source_unref(s); | |
5109 | } | |
5110 | ||
5111 | return 1; | |
5112 | } | |
b778cba4 LP |
5113 | |
5114 | _public_ int sd_event_source_get_exit_on_failure(sd_event_source *s) { | |
5115 | assert_return(s, -EINVAL); | |
5116 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
2eeff0f4 | 5117 | assert_return(!event_origin_changed(s->event), -ECHILD); |
b778cba4 LP |
5118 | |
5119 | return s->exit_on_failure; | |
5120 | } | |
5121 | ||
5122 | _public_ int sd_event_source_set_exit_on_failure(sd_event_source *s, int b) { | |
5123 | assert_return(s, -EINVAL); | |
5124 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
2eeff0f4 | 5125 | assert_return(!event_origin_changed(s->event), -ECHILD); |
b778cba4 LP |
5126 | |
5127 | if (s->exit_on_failure == !!b) | |
5128 | return 0; | |
5129 | ||
5130 | s->exit_on_failure = b; | |
5131 | return 1; | |
5132 | } | |
b6d5481b LP |
5133 | |
5134 | _public_ int sd_event_source_set_ratelimit(sd_event_source *s, uint64_t interval, unsigned burst) { | |
5135 | int r; | |
5136 | ||
5137 | assert_return(s, -EINVAL); | |
2eeff0f4 | 5138 | assert_return(!event_origin_changed(s->event), -ECHILD); |
b6d5481b LP |
5139 | |
5140 | /* Turning on ratelimiting on event source types that don't support it, is a loggable offense. Doing | |
5141 | * so is a programming error. */ | |
5142 | assert_return(EVENT_SOURCE_CAN_RATE_LIMIT(s->type), -EDOM); | |
5143 | ||
5144 | /* When ratelimiting is configured we'll always reset the rate limit state first and start fresh, | |
5145 | * non-ratelimited. */ | |
fd69f224 | 5146 | r = event_source_leave_ratelimit(s, /* run_callback */ false); |
b6d5481b LP |
5147 | if (r < 0) |
5148 | return r; | |
5149 | ||
5150 | s->rate_limit = (RateLimit) { interval, burst }; | |
5151 | return 0; | |
fd69f224 MS |
5152 | } |
5153 | ||
5154 | _public_ int sd_event_source_set_ratelimit_expire_callback(sd_event_source *s, sd_event_handler_t callback) { | |
5155 | assert_return(s, -EINVAL); | |
2eeff0f4 | 5156 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd69f224 MS |
5157 | |
5158 | s->ratelimit_expire_callback = callback; | |
5159 | return 0; | |
b6d5481b LP |
5160 | } |
5161 | ||
5162 | _public_ int sd_event_source_get_ratelimit(sd_event_source *s, uint64_t *ret_interval, unsigned *ret_burst) { | |
5163 | assert_return(s, -EINVAL); | |
2eeff0f4 | 5164 | assert_return(!event_origin_changed(s->event), -ECHILD); |
b6d5481b | 5165 | |
6dd3b818 YW |
5166 | /* Querying whether an event source has ratelimiting configured is not a loggable offense, hence |
5167 | * don't use assert_return(). Unlike turning on ratelimiting it's not really a programming error. */ | |
b6d5481b LP |
5168 | if (!EVENT_SOURCE_CAN_RATE_LIMIT(s->type)) |
5169 | return -EDOM; | |
5170 | ||
5171 | if (!ratelimit_configured(&s->rate_limit)) | |
5172 | return -ENOEXEC; | |
5173 | ||
5174 | if (ret_interval) | |
5175 | *ret_interval = s->rate_limit.interval; | |
5176 | if (ret_burst) | |
5177 | *ret_burst = s->rate_limit.burst; | |
5178 | ||
5179 | return 0; | |
5180 | } | |
5181 | ||
5182 | _public_ int sd_event_source_is_ratelimited(sd_event_source *s) { | |
5183 | assert_return(s, -EINVAL); | |
2eeff0f4 | 5184 | assert_return(!event_origin_changed(s->event), -ECHILD); |
b6d5481b LP |
5185 | |
5186 | if (!EVENT_SOURCE_CAN_RATE_LIMIT(s->type)) | |
5187 | return false; | |
5188 | ||
5189 | if (!ratelimit_configured(&s->rate_limit)) | |
5190 | return false; | |
5191 | ||
5192 | return s->ratelimited; | |
5193 | } | |
baf3fdec LP |
5194 | |
5195 | _public_ int sd_event_set_signal_exit(sd_event *e, int b) { | |
5196 | bool change = false; | |
5197 | int r; | |
5198 | ||
5199 | assert_return(e, -EINVAL); | |
5200 | ||
5201 | if (b) { | |
5202 | /* We want to maintain pointers to these event sources, so that we can destroy them when told | |
5203 | * so. But we also don't want them to pin the event loop itself. Hence we mark them as | |
5204 | * floating after creation (and undo this before deleting them again). */ | |
5205 | ||
5206 | if (!e->sigint_event_source) { | |
5207 | r = sd_event_add_signal(e, &e->sigint_event_source, SIGINT | SD_EVENT_SIGNAL_PROCMASK, NULL, NULL); | |
5208 | if (r < 0) | |
5209 | return r; | |
5210 | ||
5211 | assert(sd_event_source_set_floating(e->sigint_event_source, true) >= 0); | |
5212 | change = true; | |
5213 | } | |
5214 | ||
5215 | if (!e->sigterm_event_source) { | |
5216 | r = sd_event_add_signal(e, &e->sigterm_event_source, SIGTERM | SD_EVENT_SIGNAL_PROCMASK, NULL, NULL); | |
5217 | if (r < 0) { | |
5218 | if (change) { | |
5219 | assert(sd_event_source_set_floating(e->sigint_event_source, false) >= 0); | |
5220 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
5221 | } | |
5222 | ||
5223 | return r; | |
5224 | } | |
5225 | ||
5226 | assert(sd_event_source_set_floating(e->sigterm_event_source, true) >= 0); | |
5227 | change = true; | |
5228 | } | |
5229 | ||
5230 | } else { | |
5231 | if (e->sigint_event_source) { | |
5232 | assert(sd_event_source_set_floating(e->sigint_event_source, false) >= 0); | |
5233 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
5234 | change = true; | |
5235 | } | |
5236 | ||
5237 | if (e->sigterm_event_source) { | |
5238 | assert(sd_event_source_set_floating(e->sigterm_event_source, false) >= 0); | |
5239 | e->sigterm_event_source = sd_event_source_unref(e->sigterm_event_source); | |
5240 | change = true; | |
5241 | } | |
5242 | } | |
5243 | ||
5244 | return change; | |
5245 | } | |
158fe190 LP |
5246 | |
5247 | _public_ int sd_event_source_set_memory_pressure_type(sd_event_source *s, const char *ty) { | |
5248 | _cleanup_free_ char *b = NULL; | |
5249 | _cleanup_free_ void *w = NULL; | |
5250 | ||
5251 | assert_return(s, -EINVAL); | |
5252 | assert_return(s->type == SOURCE_MEMORY_PRESSURE, -EDOM); | |
5253 | assert_return(ty, -EINVAL); | |
2eeff0f4 | 5254 | assert_return(!event_origin_changed(s->event), -ECHILD); |
158fe190 LP |
5255 | |
5256 | if (!STR_IN_SET(ty, "some", "full")) | |
5257 | return -EINVAL; | |
5258 | ||
5259 | if (s->memory_pressure.locked) /* Refuse adjusting parameters, if caller told us how to watch for events */ | |
5260 | return -EBUSY; | |
5261 | ||
5262 | char* space = memchr(s->memory_pressure.write_buffer, ' ', s->memory_pressure.write_buffer_size); | |
5263 | if (!space) | |
5264 | return -EINVAL; | |
5265 | ||
5266 | size_t l = (char*) space - (char*) s->memory_pressure.write_buffer; | |
5267 | b = memdup_suffix0(s->memory_pressure.write_buffer, l); | |
5268 | if (!b) | |
5269 | return -ENOMEM; | |
5270 | if (!STR_IN_SET(b, "some", "full")) | |
5271 | return -EINVAL; | |
5272 | ||
5273 | if (streq(b, ty)) | |
5274 | return 0; | |
5275 | ||
5276 | size_t nl = strlen(ty) + (s->memory_pressure.write_buffer_size - l); | |
5277 | w = new(char, nl); | |
5278 | if (!w) | |
5279 | return -ENOMEM; | |
5280 | ||
5281 | memcpy(stpcpy(w, ty), space, (s->memory_pressure.write_buffer_size - l)); | |
5282 | ||
5283 | free_and_replace(s->memory_pressure.write_buffer, w); | |
5284 | s->memory_pressure.write_buffer_size = nl; | |
5285 | s->memory_pressure.locked = false; | |
5286 | ||
5287 | return 1; | |
5288 | } | |
5289 | ||
5290 | _public_ int sd_event_source_set_memory_pressure_period(sd_event_source *s, uint64_t threshold_usec, uint64_t window_usec) { | |
5291 | _cleanup_free_ char *b = NULL; | |
5292 | _cleanup_free_ void *w = NULL; | |
5293 | ||
5294 | assert_return(s, -EINVAL); | |
5295 | assert_return(s->type == SOURCE_MEMORY_PRESSURE, -EDOM); | |
2eeff0f4 | 5296 | assert_return(!event_origin_changed(s->event), -ECHILD); |
158fe190 LP |
5297 | |
5298 | if (threshold_usec <= 0 || threshold_usec >= UINT64_MAX) | |
5299 | return -ERANGE; | |
5300 | if (window_usec <= 0 || window_usec >= UINT64_MAX) | |
5301 | return -ERANGE; | |
5302 | if (threshold_usec > window_usec) | |
5303 | return -EINVAL; | |
5304 | ||
5305 | if (s->memory_pressure.locked) /* Refuse adjusting parameters, if caller told us how to watch for events */ | |
5306 | return -EBUSY; | |
5307 | ||
5308 | char* space = memchr(s->memory_pressure.write_buffer, ' ', s->memory_pressure.write_buffer_size); | |
5309 | if (!space) | |
5310 | return -EINVAL; | |
5311 | ||
5312 | size_t l = (char*) space - (char*) s->memory_pressure.write_buffer; | |
5313 | b = memdup_suffix0(s->memory_pressure.write_buffer, l); | |
5314 | if (!b) | |
5315 | return -ENOMEM; | |
5316 | if (!STR_IN_SET(b, "some", "full")) | |
5317 | return -EINVAL; | |
5318 | ||
5319 | if (asprintf((char**) &w, | |
5320 | "%s " USEC_FMT " " USEC_FMT "", | |
5321 | b, | |
5322 | threshold_usec, | |
5323 | window_usec) < 0) | |
5324 | return -EINVAL; | |
5325 | ||
5326 | l = strlen(w) + 1; | |
5327 | if (memcmp_nn(s->memory_pressure.write_buffer, s->memory_pressure.write_buffer_size, w, l) == 0) | |
5328 | return 0; | |
5329 | ||
5330 | free_and_replace(s->memory_pressure.write_buffer, w); | |
5331 | s->memory_pressure.write_buffer_size = l; | |
5332 | s->memory_pressure.locked = false; | |
5333 | ||
5334 | return 1; | |
5335 | } |