]>
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", | |
c5ea1471 | 65 | [SOURCE_TIME_BOOTTIME] = "boottime", |
3f3548f8 | 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) | |
b3a9d980 | 1168 | s->inotify.inode_data->inotify_data->n_pending++; |
97ef5391 LP |
1169 | else { |
1170 | assert(s->inotify.inode_data->inotify_data->n_pending > 0); | |
b3a9d980 | 1171 | s->inotify.inode_data->inotify_data->n_pending--; |
97ef5391 LP |
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 | |
a19b6bd5 | 1206 | s = malloc0(size_table[type]); |
fd38203a LP |
1207 | if (!s) |
1208 | return NULL; | |
a19b6bd5 FS |
1209 | /* We use expand_to_usable() here to tell gcc that it should consider this an object of the full |
1210 | * size, even if we only allocate the initial part we need. */ | |
1211 | s = expand_to_usable(s, sizeof(sd_event_source)); | |
fd38203a | 1212 | |
a38cf9fb LP |
1213 | /* Note: we cannot use compound initialization here, because sizeof(sd_event_source) is likely larger |
1214 | * than what we allocated here. */ | |
1215 | s->n_ref = 1; | |
1216 | s->event = e; | |
1217 | s->floating = floating; | |
1218 | s->type = type; | |
1219 | s->pending_index = PRIOQ_IDX_NULL; | |
1220 | s->prepare_index = PRIOQ_IDX_NULL; | |
a71fe8b8 LP |
1221 | |
1222 | if (!floating) | |
1223 | sd_event_ref(e); | |
fd38203a | 1224 | |
a71fe8b8 | 1225 | LIST_PREPEND(sources, e->sources, s); |
313cefa1 | 1226 | e->n_sources++; |
15b38f93 | 1227 | |
fd38203a LP |
1228 | return s; |
1229 | } | |
1230 | ||
b9350e70 LP |
1231 | static int io_exit_callback(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
1232 | assert(s); | |
1233 | ||
1234 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1235 | } | |
1236 | ||
f7262a9f | 1237 | _public_ int sd_event_add_io( |
fd38203a | 1238 | sd_event *e, |
151b9b96 | 1239 | sd_event_source **ret, |
fd38203a LP |
1240 | int fd, |
1241 | uint32_t events, | |
718db961 | 1242 | sd_event_io_handler_t callback, |
151b9b96 | 1243 | void *userdata) { |
fd38203a | 1244 | |
ec766a51 | 1245 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1246 | int r; |
1247 | ||
305f78bf | 1248 | assert_return(e, -EINVAL); |
b937d761 | 1249 | assert_return(e = event_resolve(e), -ENOPKG); |
8ac43fee | 1250 | assert_return(fd >= 0, -EBADF); |
2a16a986 | 1251 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 1252 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 1253 | assert_return(!event_origin_changed(e), -ECHILD); |
fd38203a | 1254 | |
b9350e70 LP |
1255 | if (!callback) |
1256 | callback = io_exit_callback; | |
1257 | ||
a71fe8b8 | 1258 | s = source_new(e, !ret, SOURCE_IO); |
fd38203a LP |
1259 | if (!s) |
1260 | return -ENOMEM; | |
1261 | ||
9da4cb2b | 1262 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1263 | s->io.fd = fd; |
1264 | s->io.events = events; | |
1265 | s->io.callback = callback; | |
1266 | s->userdata = userdata; | |
baf76283 | 1267 | s->enabled = SD_EVENT_ON; |
fd38203a | 1268 | |
baf76283 | 1269 | r = source_io_register(s, s->enabled, events); |
ec766a51 | 1270 | if (r < 0) |
050f74f2 | 1271 | return r; |
fd38203a | 1272 | |
a71fe8b8 LP |
1273 | if (ret) |
1274 | *ret = s; | |
ec766a51 | 1275 | TAKE_PTR(s); |
a71fe8b8 | 1276 | |
fd38203a LP |
1277 | return 0; |
1278 | } | |
1279 | ||
52444dc4 | 1280 | static void initialize_perturb(sd_event *e) { |
6d2326e0 | 1281 | sd_id128_t id = {}; |
52444dc4 | 1282 | |
6d2326e0 YW |
1283 | /* When we sleep for longer, we try to realign the wakeup to the same time within each |
1284 | * minute/second/250ms, so that events all across the system can be coalesced into a single CPU | |
1285 | * wakeup. However, let's take some system-specific randomness for this value, so that in a network | |
1286 | * of systems with synced clocks timer events are distributed a bit. Here, we calculate a | |
1287 | * perturbation usec offset from the boot ID (or machine ID if failed, e.g. /proc is not mounted). */ | |
52444dc4 | 1288 | |
3a43da28 | 1289 | if (_likely_(e->perturb != USEC_INFINITY)) |
52444dc4 LP |
1290 | return; |
1291 | ||
1912f790 | 1292 | if (sd_id128_get_boot(&id) >= 0 || sd_id128_get_machine(&id) >= 0) |
6d2326e0 YW |
1293 | e->perturb = (id.qwords[0] ^ id.qwords[1]) % USEC_PER_MINUTE; |
1294 | else | |
1295 | e->perturb = 0; /* This is a super early process without /proc and /etc ?? */ | |
52444dc4 LP |
1296 | } |
1297 | ||
fd38203a LP |
1298 | static int event_setup_timer_fd( |
1299 | sd_event *e, | |
6a0f1f6d LP |
1300 | struct clock_data *d, |
1301 | clockid_t clock) { | |
fd38203a | 1302 | |
fd38203a | 1303 | assert(e); |
6a0f1f6d | 1304 | assert(d); |
fd38203a | 1305 | |
6a0f1f6d | 1306 | if (_likely_(d->fd >= 0)) |
fd38203a LP |
1307 | return 0; |
1308 | ||
254d1313 | 1309 | _cleanup_close_ int fd = -EBADF; |
b44d87e2 | 1310 | |
6a0f1f6d | 1311 | fd = timerfd_create(clock, TFD_NONBLOCK|TFD_CLOEXEC); |
fd38203a LP |
1312 | if (fd < 0) |
1313 | return -errno; | |
1314 | ||
7fe2903c LP |
1315 | fd = fd_move_above_stdio(fd); |
1316 | ||
1eac7948 | 1317 | struct epoll_event ev = { |
a82f89aa LP |
1318 | .events = EPOLLIN, |
1319 | .data.ptr = d, | |
1320 | }; | |
fd38203a | 1321 | |
15c689d7 | 1322 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) |
fd38203a | 1323 | return -errno; |
fd38203a | 1324 | |
b44d87e2 | 1325 | d->fd = TAKE_FD(fd); |
fd38203a LP |
1326 | return 0; |
1327 | } | |
1328 | ||
c4f1aff2 TG |
1329 | static int time_exit_callback(sd_event_source *s, uint64_t usec, void *userdata) { |
1330 | assert(s); | |
1331 | ||
1332 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1333 | } | |
1334 | ||
41c63f36 LP |
1335 | static int setup_clock_data(sd_event *e, struct clock_data *d, clockid_t clock) { |
1336 | int r; | |
1337 | ||
1338 | assert(d); | |
1339 | ||
1340 | if (d->fd < 0) { | |
1341 | r = event_setup_timer_fd(e, d, clock); | |
1342 | if (r < 0) | |
1343 | return r; | |
1344 | } | |
1345 | ||
1346 | r = prioq_ensure_allocated(&d->earliest, earliest_time_prioq_compare); | |
1347 | if (r < 0) | |
1348 | return r; | |
1349 | ||
1350 | r = prioq_ensure_allocated(&d->latest, latest_time_prioq_compare); | |
1351 | if (r < 0) | |
1352 | return r; | |
1353 | ||
1354 | return 0; | |
1355 | } | |
1356 | ||
1e45e3fe LP |
1357 | static int event_source_time_prioq_put( |
1358 | sd_event_source *s, | |
1359 | struct clock_data *d) { | |
1360 | ||
1361 | int r; | |
1362 | ||
1363 | assert(s); | |
1364 | assert(d); | |
19947509 | 1365 | assert(EVENT_SOURCE_USES_TIME_PRIOQ(s->type)); |
1e45e3fe | 1366 | |
f41315fc | 1367 | r = prioq_put(d->earliest, s, &s->earliest_index); |
1e45e3fe LP |
1368 | if (r < 0) |
1369 | return r; | |
1370 | ||
f41315fc | 1371 | r = prioq_put(d->latest, s, &s->latest_index); |
1e45e3fe | 1372 | if (r < 0) { |
f41315fc LP |
1373 | assert_se(prioq_remove(d->earliest, s, &s->earliest_index) > 0); |
1374 | s->earliest_index = PRIOQ_IDX_NULL; | |
1e45e3fe LP |
1375 | return r; |
1376 | } | |
1377 | ||
1378 | d->needs_rearm = true; | |
1379 | return 0; | |
1380 | } | |
1381 | ||
6a0f1f6d | 1382 | _public_ int sd_event_add_time( |
fd38203a | 1383 | sd_event *e, |
151b9b96 | 1384 | sd_event_source **ret, |
6a0f1f6d | 1385 | clockid_t clock, |
fd38203a | 1386 | uint64_t usec, |
c2ba3ad6 | 1387 | uint64_t accuracy, |
718db961 | 1388 | sd_event_time_handler_t callback, |
151b9b96 | 1389 | void *userdata) { |
fd38203a | 1390 | |
6a0f1f6d | 1391 | EventSourceType type; |
ec766a51 | 1392 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6a0f1f6d | 1393 | struct clock_data *d; |
fd38203a LP |
1394 | int r; |
1395 | ||
305f78bf | 1396 | assert_return(e, -EINVAL); |
b937d761 | 1397 | assert_return(e = event_resolve(e), -ENOPKG); |
f5fbe71d | 1398 | assert_return(accuracy != UINT64_MAX, -EINVAL); |
da7e457c | 1399 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 1400 | assert_return(!event_origin_changed(e), -ECHILD); |
fd38203a | 1401 | |
e475d10c LP |
1402 | if (!clock_supported(clock)) /* Checks whether the kernel supports the clock */ |
1403 | return -EOPNOTSUPP; | |
1404 | ||
1405 | type = clock_to_event_source_type(clock); /* checks whether sd-event supports this clock */ | |
1406 | if (type < 0) | |
3411372e LP |
1407 | return -EOPNOTSUPP; |
1408 | ||
c4f1aff2 TG |
1409 | if (!callback) |
1410 | callback = time_exit_callback; | |
1411 | ||
1e45e3fe | 1412 | assert_se(d = event_get_clock_data(e, type)); |
c2ba3ad6 | 1413 | |
41c63f36 | 1414 | r = setup_clock_data(e, d, clock); |
c983e776 EV |
1415 | if (r < 0) |
1416 | return r; | |
fd38203a | 1417 | |
a71fe8b8 | 1418 | s = source_new(e, !ret, type); |
fd38203a LP |
1419 | if (!s) |
1420 | return -ENOMEM; | |
1421 | ||
1422 | s->time.next = usec; | |
c2ba3ad6 | 1423 | s->time.accuracy = accuracy == 0 ? DEFAULT_ACCURACY_USEC : accuracy; |
fd38203a | 1424 | s->time.callback = callback; |
f41315fc | 1425 | s->earliest_index = s->latest_index = PRIOQ_IDX_NULL; |
fd38203a | 1426 | s->userdata = userdata; |
baf76283 | 1427 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1428 | |
1e45e3fe | 1429 | r = event_source_time_prioq_put(s, d); |
c2ba3ad6 | 1430 | if (r < 0) |
ec766a51 | 1431 | return r; |
fd38203a | 1432 | |
a71fe8b8 LP |
1433 | if (ret) |
1434 | *ret = s; | |
ec766a51 | 1435 | TAKE_PTR(s); |
a71fe8b8 | 1436 | |
fd38203a LP |
1437 | return 0; |
1438 | } | |
1439 | ||
d6a83dc4 LP |
1440 | _public_ int sd_event_add_time_relative( |
1441 | sd_event *e, | |
1442 | sd_event_source **ret, | |
1443 | clockid_t clock, | |
1444 | uint64_t usec, | |
1445 | uint64_t accuracy, | |
1446 | sd_event_time_handler_t callback, | |
1447 | void *userdata) { | |
1448 | ||
1449 | usec_t t; | |
1450 | int r; | |
1451 | ||
1452 | /* Same as sd_event_add_time() but operates relative to the event loop's current point in time, and | |
1453 | * checks for overflow. */ | |
1454 | ||
1455 | r = sd_event_now(e, clock, &t); | |
1456 | if (r < 0) | |
1457 | return r; | |
1458 | ||
1459 | if (usec >= USEC_INFINITY - t) | |
1460 | return -EOVERFLOW; | |
1461 | ||
1462 | return sd_event_add_time(e, ret, clock, t + usec, accuracy, callback, userdata); | |
1463 | } | |
1464 | ||
59bc1fd7 LP |
1465 | static int signal_exit_callback(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) { |
1466 | assert(s); | |
1467 | ||
1468 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1469 | } | |
1470 | ||
f7262a9f | 1471 | _public_ int sd_event_add_signal( |
305f78bf | 1472 | sd_event *e, |
151b9b96 | 1473 | sd_event_source **ret, |
305f78bf | 1474 | int sig, |
718db961 | 1475 | sd_event_signal_handler_t callback, |
151b9b96 | 1476 | void *userdata) { |
305f78bf | 1477 | |
ec766a51 | 1478 | _cleanup_(source_freep) sd_event_source *s = NULL; |
9da4cb2b | 1479 | struct signal_data *d; |
897448bd LP |
1480 | sigset_t new_ss; |
1481 | bool block_it; | |
fd38203a LP |
1482 | int r; |
1483 | ||
305f78bf | 1484 | assert_return(e, -EINVAL); |
b937d761 | 1485 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1486 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 1487 | assert_return(!event_origin_changed(e), -ECHILD); |
fd38203a | 1488 | |
897448bd LP |
1489 | /* Let's make sure our special flag stays outside of the valid signal range */ |
1490 | assert_cc(_NSIG < SD_EVENT_SIGNAL_PROCMASK); | |
1491 | ||
1492 | if (sig & SD_EVENT_SIGNAL_PROCMASK) { | |
1493 | sig &= ~SD_EVENT_SIGNAL_PROCMASK; | |
1494 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
1495 | ||
1496 | block_it = true; | |
1497 | } else { | |
1498 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
1499 | ||
1500 | r = signal_is_blocked(sig); | |
1501 | if (r < 0) | |
1502 | return r; | |
1503 | if (r == 0) | |
1504 | return -EBUSY; | |
1505 | ||
1506 | block_it = false; | |
1507 | } | |
1508 | ||
59bc1fd7 LP |
1509 | if (!callback) |
1510 | callback = signal_exit_callback; | |
1511 | ||
fd38203a LP |
1512 | if (!e->signal_sources) { |
1513 | e->signal_sources = new0(sd_event_source*, _NSIG); | |
1514 | if (!e->signal_sources) | |
1515 | return -ENOMEM; | |
1516 | } else if (e->signal_sources[sig]) | |
1517 | return -EBUSY; | |
1518 | ||
a71fe8b8 | 1519 | s = source_new(e, !ret, SOURCE_SIGNAL); |
fd38203a LP |
1520 | if (!s) |
1521 | return -ENOMEM; | |
1522 | ||
1523 | s->signal.sig = sig; | |
1524 | s->signal.callback = callback; | |
1525 | s->userdata = userdata; | |
baf76283 | 1526 | s->enabled = SD_EVENT_ON; |
fd38203a LP |
1527 | |
1528 | e->signal_sources[sig] = s; | |
fd38203a | 1529 | |
897448bd LP |
1530 | if (block_it) { |
1531 | sigset_t old_ss; | |
1532 | ||
1533 | if (sigemptyset(&new_ss) < 0) | |
1534 | return -errno; | |
1535 | ||
1536 | if (sigaddset(&new_ss, sig) < 0) | |
1537 | return -errno; | |
1538 | ||
1539 | r = pthread_sigmask(SIG_BLOCK, &new_ss, &old_ss); | |
1540 | if (r != 0) | |
1541 | return -r; | |
1542 | ||
1543 | r = sigismember(&old_ss, sig); | |
1544 | if (r < 0) | |
1545 | return -errno; | |
1546 | ||
1547 | s->signal.unblock = !r; | |
1548 | } else | |
1549 | s->signal.unblock = false; | |
1550 | ||
9da4cb2b | 1551 | r = event_make_signal_data(e, sig, &d); |
897448bd LP |
1552 | if (r < 0) { |
1553 | if (s->signal.unblock) | |
1554 | (void) pthread_sigmask(SIG_UNBLOCK, &new_ss, NULL); | |
1555 | ||
9da4cb2b | 1556 | return r; |
897448bd | 1557 | } |
fd38203a | 1558 | |
f1f00dbb LP |
1559 | /* Use the signal name as description for the event source by default */ |
1560 | (void) sd_event_source_set_description(s, signal_to_string(sig)); | |
1561 | ||
a71fe8b8 LP |
1562 | if (ret) |
1563 | *ret = s; | |
ec766a51 | 1564 | TAKE_PTR(s); |
a71fe8b8 | 1565 | |
fd38203a LP |
1566 | return 0; |
1567 | } | |
1568 | ||
b9350e70 LP |
1569 | static int child_exit_callback(sd_event_source *s, const siginfo_t *si, void *userdata) { |
1570 | assert(s); | |
1571 | ||
1572 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1573 | } | |
1574 | ||
f8f3f926 LP |
1575 | static bool shall_use_pidfd(void) { |
1576 | /* Mostly relevant for debugging, i.e. this is used in test-event.c to test the event loop once with and once without pidfd */ | |
efb9b3ba | 1577 | return secure_getenv_bool("SYSTEMD_PIDFD") != 0; |
f8f3f926 LP |
1578 | } |
1579 | ||
f7262a9f | 1580 | _public_ int sd_event_add_child( |
305f78bf | 1581 | sd_event *e, |
151b9b96 | 1582 | sd_event_source **ret, |
305f78bf LP |
1583 | pid_t pid, |
1584 | int options, | |
718db961 | 1585 | sd_event_child_handler_t callback, |
151b9b96 | 1586 | void *userdata) { |
305f78bf | 1587 | |
ec766a51 | 1588 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1589 | int r; |
1590 | ||
305f78bf | 1591 | assert_return(e, -EINVAL); |
b937d761 | 1592 | assert_return(e = event_resolve(e), -ENOPKG); |
305f78bf LP |
1593 | assert_return(pid > 1, -EINVAL); |
1594 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1595 | assert_return(options != 0, -EINVAL); | |
da7e457c | 1596 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 1597 | assert_return(!event_origin_changed(e), -ECHILD); |
fd38203a | 1598 | |
b9350e70 LP |
1599 | if (!callback) |
1600 | callback = child_exit_callback; | |
1601 | ||
b6d5481b | 1602 | if (e->n_online_child_sources == 0) { |
ee880b37 LP |
1603 | /* Caller must block SIGCHLD before using us to watch children, even if pidfd is available, |
1604 | * for compatibility with pre-pidfd and because we don't want the reap the child processes | |
1605 | * ourselves, i.e. call waitid(), and don't want Linux' default internal logic for that to | |
1606 | * take effect. | |
1607 | * | |
1608 | * (As an optimization we only do this check on the first child event source created.) */ | |
1609 | r = signal_is_blocked(SIGCHLD); | |
1610 | if (r < 0) | |
1611 | return r; | |
1612 | if (r == 0) | |
1613 | return -EBUSY; | |
1614 | } | |
1615 | ||
d5099efc | 1616 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
fd38203a LP |
1617 | if (r < 0) |
1618 | return r; | |
1619 | ||
4a0b58c4 | 1620 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) |
fd38203a LP |
1621 | return -EBUSY; |
1622 | ||
a71fe8b8 | 1623 | s = source_new(e, !ret, SOURCE_CHILD); |
fd38203a LP |
1624 | if (!s) |
1625 | return -ENOMEM; | |
1626 | ||
f8f3f926 | 1627 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1628 | s->child.options = options; |
1629 | s->child.callback = callback; | |
1630 | s->userdata = userdata; | |
baf76283 | 1631 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1632 | |
f8f3f926 LP |
1633 | /* We always take a pidfd here if we can, even if we wait for anything else than WEXITED, so that we |
1634 | * pin the PID, and make regular waitid() handling race-free. */ | |
1635 | ||
1636 | if (shall_use_pidfd()) { | |
54988a27 | 1637 | s->child.pidfd = pidfd_open(pid, 0); |
f8f3f926 LP |
1638 | if (s->child.pidfd < 0) { |
1639 | /* Propagate errors unless the syscall is not supported or blocked */ | |
1640 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
1641 | return -errno; | |
1642 | } else | |
1643 | s->child.pidfd_owned = true; /* If we allocate the pidfd we own it by default */ | |
1644 | } else | |
254d1313 | 1645 | s->child.pidfd = -EBADF; |
f8f3f926 | 1646 | |
f8f3f926 LP |
1647 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1648 | /* We have a pidfd and we only want to watch for exit */ | |
f8f3f926 | 1649 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1650 | if (r < 0) |
f8f3f926 | 1651 | return r; |
ac9f2640 | 1652 | |
f8f3f926 LP |
1653 | } else { |
1654 | /* We have no pidfd or we shall wait for some other event than WEXITED */ | |
f8f3f926 | 1655 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1656 | if (r < 0) |
f8f3f926 | 1657 | return r; |
f8f3f926 LP |
1658 | |
1659 | e->need_process_child = true; | |
1660 | } | |
c2ba3ad6 | 1661 | |
54988a27 YW |
1662 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); |
1663 | if (r < 0) | |
1664 | return r; | |
1665 | ||
1666 | /* These must be done after everything succeeds. */ | |
1667 | s->child.pid = pid; | |
b6d5481b | 1668 | e->n_online_child_sources++; |
ac9f2640 | 1669 | |
a71fe8b8 LP |
1670 | if (ret) |
1671 | *ret = s; | |
ec766a51 | 1672 | TAKE_PTR(s); |
f8f3f926 LP |
1673 | return 0; |
1674 | } | |
1675 | ||
1676 | _public_ int sd_event_add_child_pidfd( | |
1677 | sd_event *e, | |
1678 | sd_event_source **ret, | |
1679 | int pidfd, | |
1680 | int options, | |
1681 | sd_event_child_handler_t callback, | |
1682 | void *userdata) { | |
1683 | ||
1684 | ||
1685 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
1686 | pid_t pid; | |
1687 | int r; | |
1688 | ||
1689 | assert_return(e, -EINVAL); | |
1690 | assert_return(e = event_resolve(e), -ENOPKG); | |
1691 | assert_return(pidfd >= 0, -EBADF); | |
1692 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1693 | assert_return(options != 0, -EINVAL); | |
f8f3f926 | 1694 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 1695 | assert_return(!event_origin_changed(e), -ECHILD); |
f8f3f926 | 1696 | |
b9350e70 LP |
1697 | if (!callback) |
1698 | callback = child_exit_callback; | |
1699 | ||
b6d5481b | 1700 | if (e->n_online_child_sources == 0) { |
ee880b37 LP |
1701 | r = signal_is_blocked(SIGCHLD); |
1702 | if (r < 0) | |
1703 | return r; | |
1704 | if (r == 0) | |
1705 | return -EBUSY; | |
1706 | } | |
1707 | ||
f8f3f926 LP |
1708 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
1709 | if (r < 0) | |
1710 | return r; | |
1711 | ||
1712 | r = pidfd_get_pid(pidfd, &pid); | |
1713 | if (r < 0) | |
1714 | return r; | |
1715 | ||
1716 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) | |
1717 | return -EBUSY; | |
1718 | ||
1719 | s = source_new(e, !ret, SOURCE_CHILD); | |
1720 | if (!s) | |
1721 | return -ENOMEM; | |
1722 | ||
1723 | s->wakeup = WAKEUP_EVENT_SOURCE; | |
1724 | s->child.pidfd = pidfd; | |
1725 | s->child.pid = pid; | |
1726 | s->child.options = options; | |
1727 | s->child.callback = callback; | |
1728 | 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) */ | |
1729 | s->userdata = userdata; | |
1730 | s->enabled = SD_EVENT_ONESHOT; | |
1731 | ||
1732 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); | |
1733 | if (r < 0) | |
1734 | return r; | |
1735 | ||
f8f3f926 LP |
1736 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1737 | /* We only want to watch for WEXITED */ | |
f8f3f926 | 1738 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1739 | if (r < 0) |
f8f3f926 | 1740 | return r; |
f8f3f926 LP |
1741 | } else { |
1742 | /* We shall wait for some other event than WEXITED */ | |
f8f3f926 | 1743 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1744 | if (r < 0) |
f8f3f926 | 1745 | return r; |
a71fe8b8 | 1746 | |
f8f3f926 LP |
1747 | e->need_process_child = true; |
1748 | } | |
1749 | ||
b6d5481b | 1750 | e->n_online_child_sources++; |
ac9f2640 | 1751 | |
f8f3f926 LP |
1752 | if (ret) |
1753 | *ret = s; | |
f8f3f926 | 1754 | TAKE_PTR(s); |
fd38203a LP |
1755 | return 0; |
1756 | } | |
1757 | ||
b9350e70 LP |
1758 | static int generic_exit_callback(sd_event_source *s, void *userdata) { |
1759 | assert(s); | |
1760 | ||
1761 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1762 | } | |
1763 | ||
f7262a9f | 1764 | _public_ int sd_event_add_defer( |
305f78bf | 1765 | sd_event *e, |
151b9b96 | 1766 | sd_event_source **ret, |
718db961 | 1767 | sd_event_handler_t callback, |
151b9b96 | 1768 | void *userdata) { |
305f78bf | 1769 | |
ec766a51 | 1770 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1771 | int r; |
1772 | ||
305f78bf | 1773 | assert_return(e, -EINVAL); |
b937d761 | 1774 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1775 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 1776 | assert_return(!event_origin_changed(e), -ECHILD); |
fd38203a | 1777 | |
b9350e70 LP |
1778 | if (!callback) |
1779 | callback = generic_exit_callback; | |
1780 | ||
a71fe8b8 | 1781 | s = source_new(e, !ret, SOURCE_DEFER); |
fd38203a LP |
1782 | if (!s) |
1783 | return -ENOMEM; | |
1784 | ||
1785 | s->defer.callback = callback; | |
1786 | s->userdata = userdata; | |
baf76283 | 1787 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a LP |
1788 | |
1789 | r = source_set_pending(s, true); | |
ec766a51 | 1790 | if (r < 0) |
fd38203a | 1791 | return r; |
fd38203a | 1792 | |
a71fe8b8 LP |
1793 | if (ret) |
1794 | *ret = s; | |
ec766a51 | 1795 | TAKE_PTR(s); |
a71fe8b8 | 1796 | |
fd38203a LP |
1797 | return 0; |
1798 | } | |
1799 | ||
6e9feda3 LP |
1800 | _public_ int sd_event_add_post( |
1801 | sd_event *e, | |
1802 | sd_event_source **ret, | |
1803 | sd_event_handler_t callback, | |
1804 | void *userdata) { | |
1805 | ||
ec766a51 | 1806 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6e9feda3 LP |
1807 | int r; |
1808 | ||
1809 | assert_return(e, -EINVAL); | |
b937d761 | 1810 | assert_return(e = event_resolve(e), -ENOPKG); |
6e9feda3 | 1811 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 1812 | assert_return(!event_origin_changed(e), -ECHILD); |
6e9feda3 | 1813 | |
b9350e70 LP |
1814 | if (!callback) |
1815 | callback = generic_exit_callback; | |
1816 | ||
a71fe8b8 | 1817 | s = source_new(e, !ret, SOURCE_POST); |
6e9feda3 LP |
1818 | if (!s) |
1819 | return -ENOMEM; | |
1820 | ||
1821 | s->post.callback = callback; | |
1822 | s->userdata = userdata; | |
1823 | s->enabled = SD_EVENT_ON; | |
1824 | ||
de7fef4b | 1825 | r = set_ensure_put(&e->post_sources, NULL, s); |
ec766a51 | 1826 | if (r < 0) |
6e9feda3 | 1827 | return r; |
de7fef4b | 1828 | assert(r > 0); |
6e9feda3 | 1829 | |
a71fe8b8 LP |
1830 | if (ret) |
1831 | *ret = s; | |
ec766a51 | 1832 | TAKE_PTR(s); |
a71fe8b8 | 1833 | |
6e9feda3 LP |
1834 | return 0; |
1835 | } | |
1836 | ||
6203e07a | 1837 | _public_ int sd_event_add_exit( |
305f78bf | 1838 | sd_event *e, |
151b9b96 | 1839 | sd_event_source **ret, |
718db961 | 1840 | sd_event_handler_t callback, |
151b9b96 | 1841 | void *userdata) { |
305f78bf | 1842 | |
ec766a51 | 1843 | _cleanup_(source_freep) sd_event_source *s = NULL; |
da7e457c LP |
1844 | int r; |
1845 | ||
1846 | assert_return(e, -EINVAL); | |
b937d761 | 1847 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
1848 | assert_return(callback, -EINVAL); |
1849 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2eeff0f4 | 1850 | assert_return(!event_origin_changed(e), -ECHILD); |
da7e457c | 1851 | |
c983e776 EV |
1852 | r = prioq_ensure_allocated(&e->exit, exit_prioq_compare); |
1853 | if (r < 0) | |
1854 | return r; | |
da7e457c | 1855 | |
a71fe8b8 | 1856 | s = source_new(e, !ret, SOURCE_EXIT); |
fd38203a | 1857 | if (!s) |
da7e457c | 1858 | return -ENOMEM; |
fd38203a | 1859 | |
6203e07a | 1860 | s->exit.callback = callback; |
da7e457c | 1861 | s->userdata = userdata; |
6203e07a | 1862 | s->exit.prioq_index = PRIOQ_IDX_NULL; |
baf76283 | 1863 | s->enabled = SD_EVENT_ONESHOT; |
da7e457c | 1864 | |
6203e07a | 1865 | r = prioq_put(s->event->exit, s, &s->exit.prioq_index); |
ec766a51 | 1866 | if (r < 0) |
da7e457c | 1867 | return r; |
da7e457c | 1868 | |
a71fe8b8 LP |
1869 | if (ret) |
1870 | *ret = s; | |
ec766a51 | 1871 | TAKE_PTR(s); |
a71fe8b8 | 1872 | |
da7e457c LP |
1873 | return 0; |
1874 | } | |
1875 | ||
9857de4f | 1876 | _public_ int sd_event_trim_memory(void) { |
158fe190 LP |
1877 | int r; |
1878 | ||
1879 | /* A default implementation of a memory pressure callback. Simply releases our own allocation caches | |
1880 | * and glibc's. This is automatically used when people call sd_event_add_memory_pressure() with a | |
1881 | * NULL callback parameter. */ | |
1882 | ||
1883 | log_debug("Memory pressure event, trimming malloc() memory."); | |
1884 | ||
1885 | #if HAVE_GENERIC_MALLINFO | |
1886 | generic_mallinfo before_mallinfo = generic_mallinfo_get(); | |
1887 | #endif | |
1888 | ||
1889 | usec_t before_timestamp = now(CLOCK_MONOTONIC); | |
1890 | hashmap_trim_pools(); | |
1891 | r = malloc_trim(0); | |
1892 | usec_t after_timestamp = now(CLOCK_MONOTONIC); | |
1893 | ||
1894 | if (r > 0) | |
1895 | log_debug("Successfully trimmed some memory."); | |
1896 | else | |
1897 | log_debug("Couldn't trim any memory."); | |
1898 | ||
1899 | usec_t period = after_timestamp - before_timestamp; | |
1900 | ||
1901 | #if HAVE_GENERIC_MALLINFO | |
1902 | generic_mallinfo after_mallinfo = generic_mallinfo_get(); | |
1903 | size_t l = LESS_BY((size_t) before_mallinfo.hblkhd, (size_t) after_mallinfo.hblkhd) + | |
1904 | LESS_BY((size_t) before_mallinfo.arena, (size_t) after_mallinfo.arena); | |
1905 | log_struct(LOG_DEBUG, | |
1906 | LOG_MESSAGE("Memory trimming took %s, returned %s to OS.", | |
1907 | FORMAT_TIMESPAN(period, 0), | |
1908 | FORMAT_BYTES(l)), | |
1909 | "MESSAGE_ID=" SD_MESSAGE_MEMORY_TRIM_STR, | |
1910 | "TRIMMED_BYTES=%zu", l, | |
1911 | "TRIMMED_USEC=" USEC_FMT, period); | |
1912 | #else | |
1913 | log_struct(LOG_DEBUG, | |
1914 | LOG_MESSAGE("Memory trimming took %s.", | |
1915 | FORMAT_TIMESPAN(period, 0)), | |
1916 | "MESSAGE_ID=" SD_MESSAGE_MEMORY_TRIM_STR, | |
1917 | "TRIMMED_USEC=" USEC_FMT, period); | |
1918 | #endif | |
1919 | ||
1920 | return 0; | |
1921 | } | |
1922 | ||
1923 | static int memory_pressure_callback(sd_event_source *s, void *userdata) { | |
1924 | assert(s); | |
1925 | ||
1926 | sd_event_trim_memory(); | |
1927 | return 0; | |
1928 | } | |
1929 | ||
1930 | _public_ int sd_event_add_memory_pressure( | |
1931 | sd_event *e, | |
1932 | sd_event_source **ret, | |
1933 | sd_event_handler_t callback, | |
1934 | void *userdata) { | |
1935 | ||
1936 | _cleanup_free_ char *w = NULL; | |
1937 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
92651a7a | 1938 | _cleanup_close_ int path_fd = -EBADF, fd = -EBADF; |
158fe190 | 1939 | _cleanup_free_ void *write_buffer = NULL; |
40c5d5d2 | 1940 | const char *watch, *watch_fallback = NULL, *env; |
158fe190 LP |
1941 | size_t write_buffer_size = 0; |
1942 | struct stat st; | |
1943 | uint32_t events; | |
1944 | bool locked; | |
1945 | int r; | |
1946 | ||
1947 | assert_return(e, -EINVAL); | |
1948 | assert_return(e = event_resolve(e), -ENOPKG); | |
1949 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2eeff0f4 | 1950 | assert_return(!event_origin_changed(e), -ECHILD); |
158fe190 LP |
1951 | |
1952 | if (!callback) | |
1953 | callback = memory_pressure_callback; | |
1954 | ||
1955 | s = source_new(e, !ret, SOURCE_MEMORY_PRESSURE); | |
1956 | if (!s) | |
1957 | return -ENOMEM; | |
1958 | ||
1959 | s->wakeup = WAKEUP_EVENT_SOURCE; | |
1960 | s->memory_pressure.callback = callback; | |
1961 | s->userdata = userdata; | |
1962 | s->enabled = SD_EVENT_ON; | |
1963 | s->memory_pressure.fd = -EBADF; | |
1964 | ||
1965 | env = secure_getenv("MEMORY_PRESSURE_WATCH"); | |
1966 | if (env) { | |
1967 | if (isempty(env) || path_equal(env, "/dev/null")) | |
1968 | return log_debug_errno(SYNTHETIC_ERRNO(EHOSTDOWN), | |
1969 | "Memory pressure logic is explicitly disabled via $MEMORY_PRESSURE_WATCH."); | |
1970 | ||
1971 | if (!path_is_absolute(env) || !path_is_normalized(env)) | |
1972 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
1973 | "$MEMORY_PRESSURE_WATCH set to invalid path: %s", env); | |
1974 | ||
1975 | watch = env; | |
1976 | ||
1977 | env = secure_getenv("MEMORY_PRESSURE_WRITE"); | |
1978 | if (env) { | |
bdd2036e | 1979 | r = unbase64mem(env, &write_buffer, &write_buffer_size); |
158fe190 LP |
1980 | if (r < 0) |
1981 | return r; | |
1982 | } | |
1983 | ||
1984 | locked = true; | |
1985 | } else { | |
1986 | ||
1987 | r = is_pressure_supported(); | |
1988 | if (r < 0) | |
1989 | return r; | |
1990 | if (r == 0) | |
1991 | return -EOPNOTSUPP; | |
1992 | ||
1993 | /* By default we want to watch memory pressure on the local cgroup, but we'll fall back on | |
1994 | * the system wide pressure if for some reason we cannot (which could be: memory controller | |
1995 | * not delegated to us, or PSI simply not available in the kernel). On legacy cgroupv1 we'll | |
1996 | * only use the system-wide logic. */ | |
1997 | r = cg_all_unified(); | |
1998 | if (r < 0) | |
1999 | return r; | |
2000 | if (r == 0) | |
2001 | watch = "/proc/pressure/memory"; | |
2002 | else { | |
2003 | _cleanup_free_ char *cg = NULL; | |
2004 | ||
2005 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &cg); | |
2006 | if (r < 0) | |
2007 | return r; | |
2008 | ||
2009 | w = path_join("/sys/fs/cgroup", cg, "memory.pressure"); | |
2010 | if (!w) | |
2011 | return -ENOMEM; | |
2012 | ||
2013 | watch = w; | |
2014 | watch_fallback = "/proc/pressure/memory"; | |
2015 | } | |
2016 | ||
2017 | /* Android uses three levels in its userspace low memory killer logic: | |
2018 | * some 70000 1000000 | |
2019 | * some 100000 1000000 | |
2020 | * full 70000 1000000 | |
2021 | * | |
2022 | * GNOME's low memory monitor uses: | |
2023 | * some 70000 1000000 | |
2024 | * some 100000 1000000 | |
2025 | * full 100000 1000000 | |
2026 | * | |
a6170074 LP |
2027 | * We'll default to the middle level that both agree on. Except we do it on a 2s window |
2028 | * (i.e. 200ms per 2s, rather than 100ms per 1s), because that's the window duration the | |
2029 | * kernel will allow us to do unprivileged, also in the future. */ | |
158fe190 LP |
2030 | if (asprintf((char**) &write_buffer, |
2031 | "%s " USEC_FMT " " USEC_FMT, | |
2032 | MEMORY_PRESSURE_DEFAULT_TYPE, | |
2033 | MEMORY_PRESSURE_DEFAULT_THRESHOLD_USEC, | |
2034 | MEMORY_PRESSURE_DEFAULT_WINDOW_USEC) < 0) | |
2035 | return -ENOMEM; | |
2036 | ||
2037 | write_buffer_size = strlen(write_buffer) + 1; | |
2038 | locked = false; | |
2039 | } | |
2040 | ||
2041 | path_fd = open(watch, O_PATH|O_CLOEXEC); | |
2042 | if (path_fd < 0) { | |
2043 | if (errno != ENOENT) | |
2044 | return -errno; | |
2045 | ||
2046 | /* We got ENOENT. Three options now: try the fallback if we have one, or return the error as | |
2047 | * is (if based on user/env config), or return -EOPNOTSUPP (because we picked the path, and | |
2048 | * the PSI service apparently is not supported) */ | |
2049 | if (!watch_fallback) | |
2050 | return locked ? -ENOENT : -EOPNOTSUPP; | |
2051 | ||
2052 | path_fd = open(watch_fallback, O_PATH|O_CLOEXEC); | |
63b1e67e YW |
2053 | if (path_fd < 0) { |
2054 | if (errno == ENOENT) /* PSI is not available in the kernel even under the fallback path? */ | |
2055 | return -EOPNOTSUPP; | |
158fe190 | 2056 | return -errno; |
63b1e67e | 2057 | } |
158fe190 LP |
2058 | } |
2059 | ||
2060 | if (fstat(path_fd, &st) < 0) | |
2061 | return -errno; | |
2062 | ||
2063 | if (S_ISSOCK(st.st_mode)) { | |
2064 | fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); | |
2065 | if (fd < 0) | |
2066 | return -errno; | |
2067 | ||
2068 | r = connect_unix_path(fd, path_fd, NULL); | |
2069 | if (r < 0) | |
2070 | return r; | |
2071 | ||
2072 | events = EPOLLIN; | |
2073 | ||
2074 | } else if (S_ISREG(st.st_mode) || S_ISFIFO(st.st_mode) || S_ISCHR(st.st_mode)) { | |
2075 | fd = fd_reopen(path_fd, (write_buffer_size > 0 ? O_RDWR : O_RDONLY) |O_CLOEXEC|O_NONBLOCK|O_NOCTTY); | |
2076 | if (fd < 0) | |
2077 | return fd; | |
2078 | ||
2079 | if (S_ISREG(st.st_mode)) { | |
2080 | struct statfs sfs; | |
2081 | ||
2082 | /* If this is a regular file validate this is a procfs or cgroupfs file, where we look for EPOLLPRI */ | |
2083 | ||
2084 | if (fstatfs(fd, &sfs) < 0) | |
2085 | return -errno; | |
2086 | ||
2087 | if (!is_fs_type(&sfs, PROC_SUPER_MAGIC) && | |
2088 | !is_fs_type(&sfs, CGROUP2_SUPER_MAGIC)) | |
2089 | return -ENOTTY; | |
2090 | ||
2091 | events = EPOLLPRI; | |
2092 | } else | |
2093 | /* For fifos and char devices just watch for EPOLLIN */ | |
2094 | events = EPOLLIN; | |
2095 | ||
2096 | } else if (S_ISDIR(st.st_mode)) | |
2097 | return -EISDIR; | |
2098 | else | |
2099 | return -EBADF; | |
2100 | ||
2101 | s->memory_pressure.fd = TAKE_FD(fd); | |
2102 | s->memory_pressure.write_buffer = TAKE_PTR(write_buffer); | |
2103 | s->memory_pressure.write_buffer_size = write_buffer_size; | |
2104 | s->memory_pressure.events = events; | |
2105 | s->memory_pressure.locked = locked; | |
2106 | ||
2107 | /* So here's the thing: if we are talking to PSI we need to write the watch string before adding the | |
2108 | * fd to epoll (if we ignore this, then the watch won't work). Hence we'll not actually register the | |
2109 | * fd with the epoll right-away. Instead, we just add the event source to a list of memory pressure | |
2110 | * event sources on which writes must be executed before the first event loop iteration is | |
2111 | * executed. (We could also write the data here, right away, but we want to give the caller the | |
2112 | * freedom to call sd_event_source_set_memory_pressure_type() and | |
2113 | * sd_event_source_set_memory_pressure_rate() before we write it. */ | |
2114 | ||
2115 | if (s->memory_pressure.write_buffer_size > 0) | |
2116 | source_memory_pressure_add_to_write_list(s); | |
2117 | else { | |
2118 | r = source_memory_pressure_register(s, s->enabled); | |
2119 | if (r < 0) | |
2120 | return r; | |
2121 | } | |
2122 | ||
2123 | if (ret) | |
2124 | *ret = s; | |
2125 | TAKE_PTR(s); | |
2126 | ||
2127 | return 0; | |
2128 | } | |
2129 | ||
97ef5391 LP |
2130 | static void event_free_inotify_data(sd_event *e, struct inotify_data *d) { |
2131 | assert(e); | |
2132 | ||
2133 | if (!d) | |
2134 | return; | |
2135 | ||
2136 | assert(hashmap_isempty(d->inodes)); | |
2137 | assert(hashmap_isempty(d->wd)); | |
2138 | ||
2139 | if (d->buffer_filled > 0) | |
0601b958 | 2140 | LIST_REMOVE(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
2141 | |
2142 | hashmap_free(d->inodes); | |
2143 | hashmap_free(d->wd); | |
2144 | ||
2145 | assert_se(hashmap_remove(e->inotify_data, &d->priority) == d); | |
2146 | ||
2147 | if (d->fd >= 0) { | |
2eeff0f4 | 2148 | if (!event_origin_changed(e) && |
fbae5090 | 2149 | epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, d->fd, NULL) < 0) |
97ef5391 LP |
2150 | log_debug_errno(errno, "Failed to remove inotify fd from epoll, ignoring: %m"); |
2151 | ||
2152 | safe_close(d->fd); | |
2153 | } | |
2154 | free(d); | |
2155 | } | |
2156 | ||
2157 | static int event_make_inotify_data( | |
2158 | sd_event *e, | |
2159 | int64_t priority, | |
2160 | struct inotify_data **ret) { | |
2161 | ||
254d1313 | 2162 | _cleanup_close_ int fd = -EBADF; |
97ef5391 | 2163 | struct inotify_data *d; |
97ef5391 LP |
2164 | int r; |
2165 | ||
2166 | assert(e); | |
2167 | ||
2168 | d = hashmap_get(e->inotify_data, &priority); | |
2169 | if (d) { | |
2170 | if (ret) | |
2171 | *ret = d; | |
2172 | return 0; | |
2173 | } | |
2174 | ||
2175 | fd = inotify_init1(IN_NONBLOCK|O_CLOEXEC); | |
2176 | if (fd < 0) | |
2177 | return -errno; | |
2178 | ||
2179 | fd = fd_move_above_stdio(fd); | |
2180 | ||
97ef5391 LP |
2181 | d = new(struct inotify_data, 1); |
2182 | if (!d) | |
2183 | return -ENOMEM; | |
2184 | ||
2185 | *d = (struct inotify_data) { | |
2186 | .wakeup = WAKEUP_INOTIFY_DATA, | |
2187 | .fd = TAKE_FD(fd), | |
2188 | .priority = priority, | |
2189 | }; | |
2190 | ||
c2484a75 | 2191 | r = hashmap_ensure_put(&e->inotify_data, &uint64_hash_ops, &d->priority, d); |
97ef5391 LP |
2192 | if (r < 0) { |
2193 | d->fd = safe_close(d->fd); | |
2194 | free(d); | |
2195 | return r; | |
2196 | } | |
2197 | ||
1eac7948 | 2198 | struct epoll_event ev = { |
97ef5391 LP |
2199 | .events = EPOLLIN, |
2200 | .data.ptr = d, | |
2201 | }; | |
2202 | ||
2203 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { | |
2204 | r = -errno; | |
2205 | d->fd = safe_close(d->fd); /* let's close this ourselves, as event_free_inotify_data() would otherwise | |
2206 | * remove the fd from the epoll first, which we don't want as we couldn't | |
2207 | * add it in the first place. */ | |
2208 | event_free_inotify_data(e, d); | |
2209 | return r; | |
2210 | } | |
2211 | ||
2212 | if (ret) | |
2213 | *ret = d; | |
2214 | ||
2215 | return 1; | |
2216 | } | |
2217 | ||
7a08d314 | 2218 | static int inode_data_compare(const struct inode_data *x, const struct inode_data *y) { |
90c88092 | 2219 | int r; |
97ef5391 LP |
2220 | |
2221 | assert(x); | |
2222 | assert(y); | |
2223 | ||
90c88092 YW |
2224 | r = CMP(x->dev, y->dev); |
2225 | if (r != 0) | |
2226 | return r; | |
97ef5391 | 2227 | |
6dd91b36 | 2228 | return CMP(x->ino, y->ino); |
97ef5391 LP |
2229 | } |
2230 | ||
7a08d314 YW |
2231 | static void inode_data_hash_func(const struct inode_data *d, struct siphash *state) { |
2232 | assert(d); | |
97ef5391 | 2233 | |
c01a5c05 YW |
2234 | siphash24_compress_typesafe(d->dev, state); |
2235 | siphash24_compress_typesafe(d->ino, state); | |
97ef5391 LP |
2236 | } |
2237 | ||
7a08d314 | 2238 | DEFINE_PRIVATE_HASH_OPS(inode_data_hash_ops, struct inode_data, inode_data_hash_func, inode_data_compare); |
97ef5391 LP |
2239 | |
2240 | static void event_free_inode_data( | |
2241 | sd_event *e, | |
2242 | struct inode_data *d) { | |
2243 | ||
2244 | assert(e); | |
2245 | ||
2246 | if (!d) | |
2247 | return; | |
2248 | ||
64903d18 | 2249 | assert(!d->event_sources); |
97ef5391 LP |
2250 | |
2251 | if (d->fd >= 0) { | |
ed828563 | 2252 | LIST_REMOVE(to_close, e->inode_data_to_close_list, d); |
97ef5391 LP |
2253 | safe_close(d->fd); |
2254 | } | |
2255 | ||
2256 | if (d->inotify_data) { | |
2257 | ||
2258 | if (d->wd >= 0) { | |
2eeff0f4 | 2259 | if (d->inotify_data->fd >= 0 && !event_origin_changed(e)) { |
97ef5391 LP |
2260 | /* So here's a problem. At the time this runs the watch descriptor might already be |
2261 | * invalidated, because an IN_IGNORED event might be queued right the moment we enter | |
2262 | * the syscall. Hence, whenever we get EINVAL, ignore it entirely, since it's a very | |
2263 | * likely case to happen. */ | |
2264 | ||
2265 | if (inotify_rm_watch(d->inotify_data->fd, d->wd) < 0 && errno != EINVAL) | |
2266 | log_debug_errno(errno, "Failed to remove watch descriptor %i from inotify, ignoring: %m", d->wd); | |
2267 | } | |
2268 | ||
2269 | assert_se(hashmap_remove(d->inotify_data->wd, INT_TO_PTR(d->wd)) == d); | |
2270 | } | |
2271 | ||
2272 | assert_se(hashmap_remove(d->inotify_data->inodes, d) == d); | |
2273 | } | |
2274 | ||
4a7cd0ca | 2275 | free(d->path); |
97ef5391 LP |
2276 | free(d); |
2277 | } | |
2278 | ||
53baf2ef LP |
2279 | static void event_gc_inotify_data( |
2280 | sd_event *e, | |
2281 | struct inotify_data *d) { | |
2282 | ||
2283 | assert(e); | |
2284 | ||
2285 | /* GCs the inotify data object if we don't need it anymore. That's the case if we don't want to watch | |
2286 | * any inode with it anymore, which in turn happens if no event source of this priority is interested | |
2287 | * in any inode any longer. That said, we maintain an extra busy counter: if non-zero we'll delay GC | |
2288 | * (under the expectation that the GC is called again once the counter is decremented). */ | |
2289 | ||
2290 | if (!d) | |
2291 | return; | |
2292 | ||
2293 | if (!hashmap_isempty(d->inodes)) | |
2294 | return; | |
2295 | ||
2296 | if (d->n_busy > 0) | |
2297 | return; | |
2298 | ||
2299 | event_free_inotify_data(e, d); | |
2300 | } | |
2301 | ||
97ef5391 LP |
2302 | static void event_gc_inode_data( |
2303 | sd_event *e, | |
2304 | struct inode_data *d) { | |
2305 | ||
2306 | struct inotify_data *inotify_data; | |
2307 | ||
2308 | assert(e); | |
2309 | ||
2310 | if (!d) | |
2311 | return; | |
2312 | ||
64903d18 | 2313 | if (d->event_sources) |
97ef5391 LP |
2314 | return; |
2315 | ||
2316 | inotify_data = d->inotify_data; | |
2317 | event_free_inode_data(e, d); | |
2318 | ||
53baf2ef | 2319 | event_gc_inotify_data(e, inotify_data); |
97ef5391 LP |
2320 | } |
2321 | ||
2322 | static int event_make_inode_data( | |
2323 | sd_event *e, | |
2324 | struct inotify_data *inotify_data, | |
2325 | dev_t dev, | |
2326 | ino_t ino, | |
2327 | struct inode_data **ret) { | |
2328 | ||
2329 | struct inode_data *d, key; | |
2330 | int r; | |
2331 | ||
2332 | assert(e); | |
2333 | assert(inotify_data); | |
2334 | ||
2335 | key = (struct inode_data) { | |
2336 | .ino = ino, | |
2337 | .dev = dev, | |
2338 | }; | |
2339 | ||
2340 | d = hashmap_get(inotify_data->inodes, &key); | |
2341 | if (d) { | |
2342 | if (ret) | |
2343 | *ret = d; | |
2344 | ||
2345 | return 0; | |
2346 | } | |
2347 | ||
2348 | r = hashmap_ensure_allocated(&inotify_data->inodes, &inode_data_hash_ops); | |
2349 | if (r < 0) | |
2350 | return r; | |
2351 | ||
2352 | d = new(struct inode_data, 1); | |
2353 | if (!d) | |
2354 | return -ENOMEM; | |
2355 | ||
2356 | *d = (struct inode_data) { | |
2357 | .dev = dev, | |
2358 | .ino = ino, | |
2359 | .wd = -1, | |
254d1313 | 2360 | .fd = -EBADF, |
97ef5391 LP |
2361 | .inotify_data = inotify_data, |
2362 | }; | |
2363 | ||
2364 | r = hashmap_put(inotify_data->inodes, d, d); | |
2365 | if (r < 0) { | |
2366 | free(d); | |
2367 | return r; | |
2368 | } | |
2369 | ||
2370 | if (ret) | |
2371 | *ret = d; | |
2372 | ||
2373 | return 1; | |
2374 | } | |
2375 | ||
2376 | static uint32_t inode_data_determine_mask(struct inode_data *d) { | |
2377 | bool excl_unlink = true; | |
2378 | uint32_t combined = 0; | |
97ef5391 LP |
2379 | |
2380 | assert(d); | |
2381 | ||
2382 | /* Combines the watch masks of all event sources watching this inode. We generally just OR them together, but | |
2383 | * the IN_EXCL_UNLINK flag is ANDed instead. | |
2384 | * | |
2385 | * Note that we add all sources to the mask here, regardless whether enabled, disabled or oneshot. That's | |
2386 | * because we cannot change the mask anymore after the event source was created once, since the kernel has no | |
f21f31b2 | 2387 | * API for that. Hence we need to subscribe to the maximum mask we ever might be interested in, and suppress |
97ef5391 LP |
2388 | * events we don't care for client-side. */ |
2389 | ||
2390 | LIST_FOREACH(inotify.by_inode_data, s, d->event_sources) { | |
2391 | ||
2392 | if ((s->inotify.mask & IN_EXCL_UNLINK) == 0) | |
2393 | excl_unlink = false; | |
2394 | ||
2395 | combined |= s->inotify.mask; | |
2396 | } | |
2397 | ||
2398 | return (combined & ~(IN_ONESHOT|IN_DONT_FOLLOW|IN_ONLYDIR|IN_EXCL_UNLINK)) | (excl_unlink ? IN_EXCL_UNLINK : 0); | |
2399 | } | |
2400 | ||
2401 | static int inode_data_realize_watch(sd_event *e, struct inode_data *d) { | |
2402 | uint32_t combined_mask; | |
2403 | int wd, r; | |
2404 | ||
2405 | assert(d); | |
2406 | assert(d->fd >= 0); | |
2407 | ||
2408 | combined_mask = inode_data_determine_mask(d); | |
2409 | ||
2410 | if (d->wd >= 0 && combined_mask == d->combined_mask) | |
2411 | return 0; | |
2412 | ||
2413 | r = hashmap_ensure_allocated(&d->inotify_data->wd, NULL); | |
2414 | if (r < 0) | |
2415 | return r; | |
2416 | ||
2417 | wd = inotify_add_watch_fd(d->inotify_data->fd, d->fd, combined_mask); | |
2418 | if (wd < 0) | |
d5f24a0e | 2419 | return wd; |
97ef5391 LP |
2420 | |
2421 | if (d->wd < 0) { | |
2422 | r = hashmap_put(d->inotify_data->wd, INT_TO_PTR(wd), d); | |
2423 | if (r < 0) { | |
2424 | (void) inotify_rm_watch(d->inotify_data->fd, wd); | |
2425 | return r; | |
2426 | } | |
2427 | ||
2428 | d->wd = wd; | |
2429 | ||
2430 | } else if (d->wd != wd) { | |
2431 | ||
2432 | log_debug("Weird, the watch descriptor we already knew for this inode changed?"); | |
2433 | (void) inotify_rm_watch(d->fd, wd); | |
2434 | return -EINVAL; | |
2435 | } | |
2436 | ||
2437 | d->combined_mask = combined_mask; | |
2438 | return 1; | |
2439 | } | |
2440 | ||
b9350e70 LP |
2441 | static int inotify_exit_callback(sd_event_source *s, const struct inotify_event *event, void *userdata) { |
2442 | assert(s); | |
2443 | ||
2444 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
2445 | } | |
2446 | ||
e67d738a | 2447 | static int event_add_inotify_fd_internal( |
97ef5391 LP |
2448 | sd_event *e, |
2449 | sd_event_source **ret, | |
e67d738a LP |
2450 | int fd, |
2451 | bool donate, | |
97ef5391 LP |
2452 | uint32_t mask, |
2453 | sd_event_inotify_handler_t callback, | |
2454 | void *userdata) { | |
2455 | ||
5bb1d7fb | 2456 | _cleanup_close_ int donated_fd = donate ? fd : -EBADF; |
e67d738a | 2457 | _cleanup_(source_freep) sd_event_source *s = NULL; |
97ef5391 LP |
2458 | struct inotify_data *inotify_data = NULL; |
2459 | struct inode_data *inode_data = NULL; | |
97ef5391 LP |
2460 | struct stat st; |
2461 | int r; | |
2462 | ||
2463 | assert_return(e, -EINVAL); | |
2464 | assert_return(e = event_resolve(e), -ENOPKG); | |
e67d738a | 2465 | assert_return(fd >= 0, -EBADF); |
97ef5391 | 2466 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 2467 | assert_return(!event_origin_changed(e), -ECHILD); |
97ef5391 | 2468 | |
b9350e70 LP |
2469 | if (!callback) |
2470 | callback = inotify_exit_callback; | |
2471 | ||
97ef5391 LP |
2472 | /* Refuse IN_MASK_ADD since we coalesce watches on the same inode, and hence really don't want to merge |
2473 | * masks. Or in other words, this whole code exists only to manage IN_MASK_ADD type operations for you, hence | |
2474 | * the user can't use them for us. */ | |
2475 | if (mask & IN_MASK_ADD) | |
2476 | return -EINVAL; | |
2477 | ||
97ef5391 LP |
2478 | if (fstat(fd, &st) < 0) |
2479 | return -errno; | |
2480 | ||
2481 | s = source_new(e, !ret, SOURCE_INOTIFY); | |
2482 | if (!s) | |
2483 | return -ENOMEM; | |
2484 | ||
2485 | s->enabled = mask & IN_ONESHOT ? SD_EVENT_ONESHOT : SD_EVENT_ON; | |
2486 | s->inotify.mask = mask; | |
2487 | s->inotify.callback = callback; | |
2488 | s->userdata = userdata; | |
2489 | ||
2490 | /* Allocate an inotify object for this priority, and an inode object within it */ | |
2491 | r = event_make_inotify_data(e, SD_EVENT_PRIORITY_NORMAL, &inotify_data); | |
2492 | if (r < 0) | |
8c75fe17 | 2493 | return r; |
97ef5391 LP |
2494 | |
2495 | r = event_make_inode_data(e, inotify_data, st.st_dev, st.st_ino, &inode_data); | |
8c75fe17 | 2496 | if (r < 0) { |
e67d738a | 2497 | event_gc_inotify_data(e, inotify_data); |
8c75fe17 ZJS |
2498 | return r; |
2499 | } | |
97ef5391 LP |
2500 | |
2501 | /* Keep the O_PATH fd around until the first iteration of the loop, so that we can still change the priority of | |
2502 | * the event source, until then, for which we need the original inode. */ | |
2503 | if (inode_data->fd < 0) { | |
e67d738a LP |
2504 | if (donated_fd >= 0) |
2505 | inode_data->fd = TAKE_FD(donated_fd); | |
2506 | else { | |
2507 | inode_data->fd = fcntl(fd, F_DUPFD_CLOEXEC, 3); | |
2508 | if (inode_data->fd < 0) { | |
2509 | r = -errno; | |
2510 | event_gc_inode_data(e, inode_data); | |
2511 | return r; | |
2512 | } | |
2513 | } | |
2514 | ||
ed828563 | 2515 | LIST_PREPEND(to_close, e->inode_data_to_close_list, inode_data); |
4a7cd0ca YW |
2516 | |
2517 | _cleanup_free_ char *path = NULL; | |
2518 | r = fd_get_path(inode_data->fd, &path); | |
2519 | if (r < 0 && r != -ENOSYS) { /* The path is optional, hence ignore -ENOSYS. */ | |
2520 | event_gc_inode_data(e, inode_data); | |
2521 | return r; | |
2522 | } | |
2523 | ||
2524 | free_and_replace(inode_data->path, path); | |
97ef5391 LP |
2525 | } |
2526 | ||
2527 | /* Link our event source to the inode data object */ | |
2528 | LIST_PREPEND(inotify.by_inode_data, inode_data->event_sources, s); | |
2529 | s->inotify.inode_data = inode_data; | |
2530 | ||
97ef5391 LP |
2531 | /* Actually realize the watch now */ |
2532 | r = inode_data_realize_watch(e, inode_data); | |
2533 | if (r < 0) | |
8c75fe17 | 2534 | return r; |
97ef5391 | 2535 | |
97ef5391 LP |
2536 | if (ret) |
2537 | *ret = s; | |
8c75fe17 | 2538 | TAKE_PTR(s); |
97ef5391 LP |
2539 | |
2540 | return 0; | |
97ef5391 LP |
2541 | } |
2542 | ||
e67d738a LP |
2543 | _public_ int sd_event_add_inotify_fd( |
2544 | sd_event *e, | |
2545 | sd_event_source **ret, | |
2546 | int fd, | |
2547 | uint32_t mask, | |
2548 | sd_event_inotify_handler_t callback, | |
2549 | void *userdata) { | |
2550 | ||
2551 | return event_add_inotify_fd_internal(e, ret, fd, /* donate= */ false, mask, callback, userdata); | |
2552 | } | |
2553 | ||
2554 | _public_ int sd_event_add_inotify( | |
2555 | sd_event *e, | |
2556 | sd_event_source **ret, | |
2557 | const char *path, | |
2558 | uint32_t mask, | |
2559 | sd_event_inotify_handler_t callback, | |
2560 | void *userdata) { | |
2561 | ||
2091c779 | 2562 | sd_event_source *s = NULL; /* avoid false maybe-uninitialized warning */ |
e67d738a LP |
2563 | int fd, r; |
2564 | ||
2565 | assert_return(path, -EINVAL); | |
2566 | ||
586c8cee ZJS |
2567 | fd = open(path, O_PATH | O_CLOEXEC | |
2568 | (mask & IN_ONLYDIR ? O_DIRECTORY : 0) | | |
2569 | (mask & IN_DONT_FOLLOW ? O_NOFOLLOW : 0)); | |
e67d738a LP |
2570 | if (fd < 0) |
2571 | return -errno; | |
2572 | ||
2573 | r = event_add_inotify_fd_internal(e, &s, fd, /* donate= */ true, mask, callback, userdata); | |
2574 | if (r < 0) | |
2575 | return r; | |
2576 | ||
2577 | (void) sd_event_source_set_description(s, path); | |
2578 | ||
2579 | if (ret) | |
2580 | *ret = s; | |
2581 | ||
2582 | return r; | |
2583 | } | |
2584 | ||
8301aa0b | 2585 | static sd_event_source* event_source_free(sd_event_source *s) { |
6680dd6b LP |
2586 | if (!s) |
2587 | return NULL; | |
da7e457c | 2588 | |
8301aa0b YW |
2589 | /* Here's a special hack: when we are called from a |
2590 | * dispatch handler we won't free the event source | |
2591 | * immediately, but we will detach the fd from the | |
2592 | * epoll. This way it is safe for the caller to unref | |
2593 | * the event source and immediately close the fd, but | |
2594 | * we still retain a valid event source object after | |
2595 | * the callback. */ | |
fd38203a | 2596 | |
76d04c3a | 2597 | if (s->dispatching) |
8301aa0b | 2598 | source_disconnect(s); |
76d04c3a | 2599 | else |
8301aa0b | 2600 | source_free(s); |
fd38203a LP |
2601 | |
2602 | return NULL; | |
2603 | } | |
2604 | ||
8301aa0b YW |
2605 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event_source, sd_event_source, event_source_free); |
2606 | ||
356779df | 2607 | _public_ int sd_event_source_set_description(sd_event_source *s, const char *description) { |
f7f53e9e | 2608 | assert_return(s, -EINVAL); |
2eeff0f4 | 2609 | assert_return(!event_origin_changed(s->event), -ECHILD); |
f7f53e9e | 2610 | |
356779df | 2611 | return free_and_strdup(&s->description, description); |
f7f53e9e TG |
2612 | } |
2613 | ||
356779df | 2614 | _public_ int sd_event_source_get_description(sd_event_source *s, const char **description) { |
f7f53e9e | 2615 | assert_return(s, -EINVAL); |
356779df | 2616 | assert_return(description, -EINVAL); |
f7f53e9e | 2617 | |
7d92a1a4 ZJS |
2618 | if (!s->description) |
2619 | return -ENXIO; | |
2620 | ||
356779df | 2621 | *description = s->description; |
f7f53e9e TG |
2622 | return 0; |
2623 | } | |
2624 | ||
adcc4ca3 | 2625 | _public_ sd_event *sd_event_source_get_event(sd_event_source *s) { |
305f78bf | 2626 | assert_return(s, NULL); |
2eeff0f4 | 2627 | assert_return(!event_origin_changed(s->event), NULL); |
eaa3cbef LP |
2628 | |
2629 | return s->event; | |
2630 | } | |
2631 | ||
f7262a9f | 2632 | _public_ int sd_event_source_get_pending(sd_event_source *s) { |
305f78bf | 2633 | assert_return(s, -EINVAL); |
6203e07a | 2634 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c | 2635 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 2636 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
2637 | |
2638 | return s->pending; | |
2639 | } | |
2640 | ||
f7262a9f | 2641 | _public_ int sd_event_source_get_io_fd(sd_event_source *s) { |
305f78bf LP |
2642 | assert_return(s, -EINVAL); |
2643 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2eeff0f4 | 2644 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
2645 | |
2646 | return s->io.fd; | |
2647 | } | |
2648 | ||
30caf8f3 | 2649 | _public_ int sd_event_source_set_io_fd(sd_event_source *s, int fd) { |
2fa48059 | 2650 | int saved_fd, r; |
30caf8f3 LP |
2651 | |
2652 | assert_return(s, -EINVAL); | |
8ac43fee | 2653 | assert_return(fd >= 0, -EBADF); |
30caf8f3 | 2654 | assert_return(s->type == SOURCE_IO, -EDOM); |
2eeff0f4 | 2655 | assert_return(!event_origin_changed(s->event), -ECHILD); |
30caf8f3 LP |
2656 | |
2657 | if (s->io.fd == fd) | |
2658 | return 0; | |
2659 | ||
2fa48059 YW |
2660 | saved_fd = s->io.fd; |
2661 | s->io.fd = fd; | |
30caf8f3 | 2662 | |
2fa48059 | 2663 | assert(event_source_is_offline(s) == !s->io.registered); |
30caf8f3 | 2664 | |
2fa48059 | 2665 | if (s->io.registered) { |
30caf8f3 LP |
2666 | s->io.registered = false; |
2667 | ||
2668 | r = source_io_register(s, s->enabled, s->io.events); | |
2669 | if (r < 0) { | |
2670 | s->io.fd = saved_fd; | |
2671 | s->io.registered = true; | |
2672 | return r; | |
2673 | } | |
2674 | ||
5a795bff | 2675 | (void) epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, saved_fd, NULL); |
30caf8f3 LP |
2676 | } |
2677 | ||
2fa48059 YW |
2678 | if (s->io.owned) |
2679 | safe_close(saved_fd); | |
2680 | ||
30caf8f3 LP |
2681 | return 0; |
2682 | } | |
2683 | ||
ab93297c NM |
2684 | _public_ int sd_event_source_get_io_fd_own(sd_event_source *s) { |
2685 | assert_return(s, -EINVAL); | |
2686 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2eeff0f4 | 2687 | assert_return(!event_origin_changed(s->event), -ECHILD); |
ab93297c NM |
2688 | |
2689 | return s->io.owned; | |
2690 | } | |
2691 | ||
2692 | _public_ int sd_event_source_set_io_fd_own(sd_event_source *s, int own) { | |
2693 | assert_return(s, -EINVAL); | |
2694 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2eeff0f4 | 2695 | assert_return(!event_origin_changed(s->event), -ECHILD); |
ab93297c NM |
2696 | |
2697 | s->io.owned = own; | |
2698 | return 0; | |
2699 | } | |
2700 | ||
f7262a9f | 2701 | _public_ int sd_event_source_get_io_events(sd_event_source *s, uint32_t* events) { |
305f78bf LP |
2702 | assert_return(s, -EINVAL); |
2703 | assert_return(events, -EINVAL); | |
2704 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2eeff0f4 | 2705 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
2706 | |
2707 | *events = s->io.events; | |
2708 | return 0; | |
2709 | } | |
2710 | ||
f7262a9f | 2711 | _public_ int sd_event_source_set_io_events(sd_event_source *s, uint32_t events) { |
fd38203a LP |
2712 | int r; |
2713 | ||
305f78bf LP |
2714 | assert_return(s, -EINVAL); |
2715 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2a16a986 | 2716 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 2717 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 2718 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a | 2719 | |
b63c8d4f DH |
2720 | /* edge-triggered updates are never skipped, so we can reset edges */ |
2721 | if (s->io.events == events && !(events & EPOLLET)) | |
fd38203a LP |
2722 | return 0; |
2723 | ||
2a0dc6cd LP |
2724 | r = source_set_pending(s, false); |
2725 | if (r < 0) | |
2726 | return r; | |
2727 | ||
b6d5481b | 2728 | if (event_source_is_online(s)) { |
e4715127 | 2729 | r = source_io_register(s, s->enabled, events); |
fd38203a LP |
2730 | if (r < 0) |
2731 | return r; | |
2732 | } | |
2733 | ||
2734 | s->io.events = events; | |
2735 | ||
2736 | return 0; | |
2737 | } | |
2738 | ||
f7262a9f | 2739 | _public_ int sd_event_source_get_io_revents(sd_event_source *s, uint32_t* revents) { |
305f78bf LP |
2740 | assert_return(s, -EINVAL); |
2741 | assert_return(revents, -EINVAL); | |
2742 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2743 | assert_return(s->pending, -ENODATA); | |
2eeff0f4 | 2744 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
2745 | |
2746 | *revents = s->io.revents; | |
2747 | return 0; | |
2748 | } | |
2749 | ||
f7262a9f | 2750 | _public_ int sd_event_source_get_signal(sd_event_source *s) { |
305f78bf LP |
2751 | assert_return(s, -EINVAL); |
2752 | assert_return(s->type == SOURCE_SIGNAL, -EDOM); | |
2eeff0f4 | 2753 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
2754 | |
2755 | return s->signal.sig; | |
2756 | } | |
2757 | ||
31927c16 | 2758 | _public_ int sd_event_source_get_priority(sd_event_source *s, int64_t *priority) { |
305f78bf | 2759 | assert_return(s, -EINVAL); |
2eeff0f4 | 2760 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a | 2761 | |
6680b8d1 ME |
2762 | *priority = s->priority; |
2763 | return 0; | |
fd38203a LP |
2764 | } |
2765 | ||
31927c16 | 2766 | _public_ int sd_event_source_set_priority(sd_event_source *s, int64_t priority) { |
97ef5391 LP |
2767 | bool rm_inotify = false, rm_inode = false; |
2768 | struct inotify_data *new_inotify_data = NULL; | |
2769 | struct inode_data *new_inode_data = NULL; | |
9da4cb2b LP |
2770 | int r; |
2771 | ||
305f78bf | 2772 | assert_return(s, -EINVAL); |
da7e457c | 2773 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 2774 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
2775 | |
2776 | if (s->priority == priority) | |
2777 | return 0; | |
2778 | ||
97ef5391 LP |
2779 | if (s->type == SOURCE_INOTIFY) { |
2780 | struct inode_data *old_inode_data; | |
2781 | ||
2782 | assert(s->inotify.inode_data); | |
2783 | old_inode_data = s->inotify.inode_data; | |
2784 | ||
2785 | /* We need the original fd to change the priority. If we don't have it we can't change the priority, | |
2786 | * anymore. Note that we close any fds when entering the next event loop iteration, i.e. for inotify | |
2787 | * events we allow priority changes only until the first following iteration. */ | |
2788 | if (old_inode_data->fd < 0) | |
2789 | return -EOPNOTSUPP; | |
2790 | ||
2791 | r = event_make_inotify_data(s->event, priority, &new_inotify_data); | |
2792 | if (r < 0) | |
2793 | return r; | |
2794 | rm_inotify = r > 0; | |
2795 | ||
2796 | r = event_make_inode_data(s->event, new_inotify_data, old_inode_data->dev, old_inode_data->ino, &new_inode_data); | |
2797 | if (r < 0) | |
2798 | goto fail; | |
2799 | rm_inode = r > 0; | |
2800 | ||
2801 | if (new_inode_data->fd < 0) { | |
2802 | /* Duplicate the fd for the new inode object if we don't have any yet */ | |
2803 | new_inode_data->fd = fcntl(old_inode_data->fd, F_DUPFD_CLOEXEC, 3); | |
2804 | if (new_inode_data->fd < 0) { | |
2805 | r = -errno; | |
2806 | goto fail; | |
2807 | } | |
2808 | ||
ed828563 | 2809 | LIST_PREPEND(to_close, s->event->inode_data_to_close_list, new_inode_data); |
4a7cd0ca YW |
2810 | |
2811 | _cleanup_free_ char *path = NULL; | |
2812 | r = fd_get_path(new_inode_data->fd, &path); | |
2813 | if (r < 0 && r != -ENOSYS) | |
2814 | goto fail; | |
2815 | ||
2816 | free_and_replace(new_inode_data->path, path); | |
97ef5391 LP |
2817 | } |
2818 | ||
2819 | /* Move the event source to the new inode data structure */ | |
2820 | LIST_REMOVE(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2821 | LIST_PREPEND(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2822 | s->inotify.inode_data = new_inode_data; | |
2823 | ||
2824 | /* Now create the new watch */ | |
2825 | r = inode_data_realize_watch(s->event, new_inode_data); | |
2826 | if (r < 0) { | |
2827 | /* Move it back */ | |
2828 | LIST_REMOVE(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2829 | LIST_PREPEND(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2830 | s->inotify.inode_data = old_inode_data; | |
2831 | goto fail; | |
2832 | } | |
2833 | ||
2834 | s->priority = priority; | |
2835 | ||
2836 | event_gc_inode_data(s->event, old_inode_data); | |
2837 | ||
b6d5481b | 2838 | } else if (s->type == SOURCE_SIGNAL && event_source_is_online(s)) { |
9da4cb2b LP |
2839 | struct signal_data *old, *d; |
2840 | ||
2841 | /* Move us from the signalfd belonging to the old | |
2842 | * priority to the signalfd of the new priority */ | |
2843 | ||
2844 | assert_se(old = hashmap_get(s->event->signal_data, &s->priority)); | |
2845 | ||
2846 | s->priority = priority; | |
2847 | ||
2848 | r = event_make_signal_data(s->event, s->signal.sig, &d); | |
2849 | if (r < 0) { | |
2850 | s->priority = old->priority; | |
2851 | return r; | |
2852 | } | |
2853 | ||
2854 | event_unmask_signal_data(s->event, old, s->signal.sig); | |
2855 | } else | |
2856 | s->priority = priority; | |
fd38203a | 2857 | |
e1951c16 | 2858 | event_source_pp_prioq_reshuffle(s); |
fd38203a | 2859 | |
6203e07a LP |
2860 | if (s->type == SOURCE_EXIT) |
2861 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
305f78bf | 2862 | |
fd38203a | 2863 | return 0; |
97ef5391 LP |
2864 | |
2865 | fail: | |
2866 | if (rm_inode) | |
2867 | event_free_inode_data(s->event, new_inode_data); | |
2868 | ||
2869 | if (rm_inotify) | |
2870 | event_free_inotify_data(s->event, new_inotify_data); | |
2871 | ||
2872 | return r; | |
fd38203a LP |
2873 | } |
2874 | ||
cad143a8 | 2875 | _public_ int sd_event_source_get_enabled(sd_event_source *s, int *ret) { |
71193c0b ZJS |
2876 | /* Quick mode: the event source doesn't exist and we only want to query boolean enablement state. */ |
2877 | if (!s && !ret) | |
2878 | return false; | |
2879 | ||
305f78bf | 2880 | assert_return(s, -EINVAL); |
2eeff0f4 | 2881 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a | 2882 | |
cad143a8 LP |
2883 | if (ret) |
2884 | *ret = s->enabled; | |
2885 | ||
08c1eb0e | 2886 | return s->enabled != SD_EVENT_OFF; |
fd38203a LP |
2887 | } |
2888 | ||
b6d5481b LP |
2889 | static int event_source_offline( |
2890 | sd_event_source *s, | |
2891 | int enabled, | |
2892 | bool ratelimited) { | |
2893 | ||
2894 | bool was_offline; | |
fd38203a LP |
2895 | int r; |
2896 | ||
ddfde737 | 2897 | assert(s); |
b6d5481b | 2898 | assert(enabled == SD_EVENT_OFF || ratelimited); |
fd38203a | 2899 | |
ddfde737 | 2900 | /* Unset the pending flag when this event source is disabled */ |
b6d5481b LP |
2901 | if (s->enabled != SD_EVENT_OFF && |
2902 | enabled == SD_EVENT_OFF && | |
2903 | !IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
ddfde737 LP |
2904 | r = source_set_pending(s, false); |
2905 | if (r < 0) | |
2906 | return r; | |
2907 | } | |
cc567911 | 2908 | |
b6d5481b LP |
2909 | was_offline = event_source_is_offline(s); |
2910 | s->enabled = enabled; | |
2911 | s->ratelimited = ratelimited; | |
fd38203a | 2912 | |
ddfde737 | 2913 | switch (s->type) { |
fd38203a | 2914 | |
ddfde737 LP |
2915 | case SOURCE_IO: |
2916 | source_io_unregister(s); | |
2917 | break; | |
ac989a78 | 2918 | |
ddfde737 LP |
2919 | case SOURCE_SIGNAL: |
2920 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); | |
2921 | break; | |
fd38203a | 2922 | |
ddfde737 | 2923 | case SOURCE_CHILD: |
b6d5481b LP |
2924 | if (!was_offline) { |
2925 | assert(s->event->n_online_child_sources > 0); | |
2926 | s->event->n_online_child_sources--; | |
2927 | } | |
fd38203a | 2928 | |
ddfde737 LP |
2929 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
2930 | source_child_pidfd_unregister(s); | |
2931 | else | |
2932 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
2933 | break; | |
4807d2d0 | 2934 | |
ddfde737 LP |
2935 | case SOURCE_EXIT: |
2936 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
2937 | break; | |
fd38203a | 2938 | |
158fe190 LP |
2939 | case SOURCE_MEMORY_PRESSURE: |
2940 | source_memory_pressure_unregister(s); | |
2941 | break; | |
2942 | ||
2115b9b6 YW |
2943 | case SOURCE_TIME_REALTIME: |
2944 | case SOURCE_TIME_BOOTTIME: | |
2945 | case SOURCE_TIME_MONOTONIC: | |
2946 | case SOURCE_TIME_REALTIME_ALARM: | |
2947 | case SOURCE_TIME_BOOTTIME_ALARM: | |
ddfde737 LP |
2948 | case SOURCE_DEFER: |
2949 | case SOURCE_POST: | |
2950 | case SOURCE_INOTIFY: | |
2951 | break; | |
fd38203a | 2952 | |
ddfde737 | 2953 | default: |
04499a70 | 2954 | assert_not_reached(); |
ddfde737 | 2955 | } |
fd38203a | 2956 | |
2115b9b6 YW |
2957 | /* Always reshuffle time prioq, as the ratelimited flag may be changed. */ |
2958 | event_source_time_prioq_reshuffle(s); | |
2959 | ||
b6d5481b | 2960 | return 1; |
ddfde737 | 2961 | } |
f8f3f926 | 2962 | |
b6d5481b LP |
2963 | static int event_source_online( |
2964 | sd_event_source *s, | |
2965 | int enabled, | |
2966 | bool ratelimited) { | |
2967 | ||
2968 | bool was_online; | |
ddfde737 | 2969 | int r; |
fd38203a | 2970 | |
ddfde737 | 2971 | assert(s); |
b6d5481b | 2972 | assert(enabled != SD_EVENT_OFF || !ratelimited); |
305f78bf | 2973 | |
ddfde737 | 2974 | /* Unset the pending flag when this event source is enabled */ |
b6d5481b LP |
2975 | if (s->enabled == SD_EVENT_OFF && |
2976 | enabled != SD_EVENT_OFF && | |
2977 | !IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
ddfde737 LP |
2978 | r = source_set_pending(s, false); |
2979 | if (r < 0) | |
2980 | return r; | |
2981 | } | |
9d3e3aa5 | 2982 | |
b6d5481b LP |
2983 | /* Are we really ready for onlining? */ |
2984 | if (enabled == SD_EVENT_OFF || ratelimited) { | |
2985 | /* Nope, we are not ready for onlining, then just update the precise state and exit */ | |
2986 | s->enabled = enabled; | |
2987 | s->ratelimited = ratelimited; | |
2988 | return 0; | |
2989 | } | |
2990 | ||
2991 | was_online = event_source_is_online(s); | |
2992 | ||
ddfde737 | 2993 | switch (s->type) { |
ddfde737 | 2994 | case SOURCE_IO: |
b6d5481b | 2995 | r = source_io_register(s, enabled, s->io.events); |
d2eafe61 | 2996 | if (r < 0) |
ddfde737 | 2997 | return r; |
ddfde737 | 2998 | break; |
fd38203a | 2999 | |
ddfde737 LP |
3000 | case SOURCE_SIGNAL: |
3001 | r = event_make_signal_data(s->event, s->signal.sig, NULL); | |
3002 | if (r < 0) { | |
ddfde737 LP |
3003 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
3004 | return r; | |
3005 | } | |
fd38203a | 3006 | |
ddfde737 | 3007 | break; |
fd38203a | 3008 | |
ddfde737 | 3009 | case SOURCE_CHILD: |
ddfde737 LP |
3010 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
3011 | /* yes, we have pidfd */ | |
9da4cb2b | 3012 | |
b6d5481b | 3013 | r = source_child_pidfd_register(s, enabled); |
ac9f2640 | 3014 | if (r < 0) |
9da4cb2b | 3015 | return r; |
ddfde737 LP |
3016 | } else { |
3017 | /* no pidfd, or something other to watch for than WEXITED */ | |
9da4cb2b | 3018 | |
ddfde737 LP |
3019 | r = event_make_signal_data(s->event, SIGCHLD, NULL); |
3020 | if (r < 0) { | |
ddfde737 LP |
3021 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); |
3022 | return r; | |
3023 | } | |
3024 | } | |
fd38203a | 3025 | |
b6d5481b LP |
3026 | if (!was_online) |
3027 | s->event->n_online_child_sources++; | |
ddfde737 | 3028 | break; |
4807d2d0 | 3029 | |
158fe190 LP |
3030 | case SOURCE_MEMORY_PRESSURE: |
3031 | r = source_memory_pressure_register(s, enabled); | |
3032 | if (r < 0) | |
3033 | return r; | |
3034 | ||
3035 | break; | |
3036 | ||
d2eafe61 ZJS |
3037 | case SOURCE_TIME_REALTIME: |
3038 | case SOURCE_TIME_BOOTTIME: | |
3039 | case SOURCE_TIME_MONOTONIC: | |
3040 | case SOURCE_TIME_REALTIME_ALARM: | |
3041 | case SOURCE_TIME_BOOTTIME_ALARM: | |
ddfde737 | 3042 | case SOURCE_EXIT: |
ddfde737 LP |
3043 | case SOURCE_DEFER: |
3044 | case SOURCE_POST: | |
3045 | case SOURCE_INOTIFY: | |
3046 | break; | |
9da4cb2b | 3047 | |
ddfde737 | 3048 | default: |
04499a70 | 3049 | assert_not_reached(); |
ddfde737 | 3050 | } |
f8f3f926 | 3051 | |
b6d5481b LP |
3052 | s->enabled = enabled; |
3053 | s->ratelimited = ratelimited; | |
d2eafe61 ZJS |
3054 | |
3055 | /* Non-failing operations below */ | |
2115b9b6 | 3056 | if (s->type == SOURCE_EXIT) |
d2eafe61 | 3057 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); |
d2eafe61 | 3058 | |
2115b9b6 YW |
3059 | /* Always reshuffle time prioq, as the ratelimited flag may be changed. */ |
3060 | event_source_time_prioq_reshuffle(s); | |
d2eafe61 | 3061 | |
b6d5481b | 3062 | return 1; |
ddfde737 LP |
3063 | } |
3064 | ||
3065 | _public_ int sd_event_source_set_enabled(sd_event_source *s, int m) { | |
3066 | int r; | |
9da4cb2b | 3067 | |
ddfde737 | 3068 | assert_return(IN_SET(m, SD_EVENT_OFF, SD_EVENT_ON, SD_EVENT_ONESHOT), -EINVAL); |
7e922b05 ZJS |
3069 | |
3070 | /* Quick mode: if the source doesn't exist, SD_EVENT_OFF is a noop. */ | |
3071 | if (m == SD_EVENT_OFF && !s) | |
3072 | return 0; | |
3073 | ||
3074 | assert_return(s, -EINVAL); | |
2eeff0f4 | 3075 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a | 3076 | |
ddfde737 LP |
3077 | /* If we are dead anyway, we are fine with turning off sources, but everything else needs to fail. */ |
3078 | if (s->event->state == SD_EVENT_FINISHED) | |
3079 | return m == SD_EVENT_OFF ? 0 : -ESTALE; | |
305f78bf | 3080 | |
ddfde737 LP |
3081 | if (s->enabled == m) /* No change? */ |
3082 | return 0; | |
9d3e3aa5 | 3083 | |
ddfde737 | 3084 | if (m == SD_EVENT_OFF) |
b6d5481b | 3085 | r = event_source_offline(s, m, s->ratelimited); |
ddfde737 LP |
3086 | else { |
3087 | if (s->enabled != SD_EVENT_OFF) { | |
3088 | /* Switching from "on" to "oneshot" or back? If that's the case, we can take a shortcut, the | |
3089 | * event source is already enabled after all. */ | |
3090 | s->enabled = m; | |
3091 | return 0; | |
fd38203a | 3092 | } |
ddfde737 | 3093 | |
b6d5481b | 3094 | r = event_source_online(s, m, s->ratelimited); |
fd38203a | 3095 | } |
ddfde737 LP |
3096 | if (r < 0) |
3097 | return r; | |
fd38203a | 3098 | |
e1951c16 | 3099 | event_source_pp_prioq_reshuffle(s); |
fd38203a LP |
3100 | return 0; |
3101 | } | |
3102 | ||
f7262a9f | 3103 | _public_ int sd_event_source_get_time(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
3104 | assert_return(s, -EINVAL); |
3105 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 3106 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
2eeff0f4 | 3107 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
3108 | |
3109 | *usec = s->time.next; | |
3110 | return 0; | |
3111 | } | |
3112 | ||
f7262a9f | 3113 | _public_ int sd_event_source_set_time(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 3114 | int r; |
6a0f1f6d | 3115 | |
305f78bf | 3116 | assert_return(s, -EINVAL); |
6a0f1f6d | 3117 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 3118 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 3119 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a | 3120 | |
2a0dc6cd LP |
3121 | r = source_set_pending(s, false); |
3122 | if (r < 0) | |
3123 | return r; | |
2576a19e | 3124 | |
2a0dc6cd | 3125 | s->time.next = usec; |
fd38203a | 3126 | |
e1951c16 | 3127 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
3128 | return 0; |
3129 | } | |
3130 | ||
d6a83dc4 LP |
3131 | _public_ int sd_event_source_set_time_relative(sd_event_source *s, uint64_t usec) { |
3132 | usec_t t; | |
3133 | int r; | |
3134 | ||
3135 | assert_return(s, -EINVAL); | |
3136 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2eeff0f4 | 3137 | assert_return(!event_origin_changed(s->event), -ECHILD); |
d6a83dc4 | 3138 | |
ef859195 LP |
3139 | if (usec == USEC_INFINITY) |
3140 | return sd_event_source_set_time(s, USEC_INFINITY); | |
3141 | ||
d6a83dc4 LP |
3142 | r = sd_event_now(s->event, event_source_type_to_clock(s->type), &t); |
3143 | if (r < 0) | |
3144 | return r; | |
3145 | ||
496db330 YW |
3146 | usec = usec_add(t, usec); |
3147 | if (usec == USEC_INFINITY) | |
d6a83dc4 LP |
3148 | return -EOVERFLOW; |
3149 | ||
496db330 | 3150 | return sd_event_source_set_time(s, usec); |
d6a83dc4 LP |
3151 | } |
3152 | ||
f7262a9f | 3153 | _public_ int sd_event_source_get_time_accuracy(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
3154 | assert_return(s, -EINVAL); |
3155 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 3156 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
2eeff0f4 | 3157 | assert_return(!event_origin_changed(s->event), -ECHILD); |
305f78bf LP |
3158 | |
3159 | *usec = s->time.accuracy; | |
3160 | return 0; | |
3161 | } | |
3162 | ||
f7262a9f | 3163 | _public_ int sd_event_source_set_time_accuracy(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 3164 | int r; |
6a0f1f6d | 3165 | |
305f78bf | 3166 | assert_return(s, -EINVAL); |
f5fbe71d | 3167 | assert_return(usec != UINT64_MAX, -EINVAL); |
6a0f1f6d | 3168 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 3169 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 3170 | assert_return(!event_origin_changed(s->event), -ECHILD); |
eaa3cbef | 3171 | |
2a0dc6cd LP |
3172 | r = source_set_pending(s, false); |
3173 | if (r < 0) | |
3174 | return r; | |
3175 | ||
eaa3cbef LP |
3176 | if (usec == 0) |
3177 | usec = DEFAULT_ACCURACY_USEC; | |
3178 | ||
eaa3cbef LP |
3179 | s->time.accuracy = usec; |
3180 | ||
e1951c16 | 3181 | event_source_time_prioq_reshuffle(s); |
6a0f1f6d LP |
3182 | return 0; |
3183 | } | |
3184 | ||
3185 | _public_ int sd_event_source_get_time_clock(sd_event_source *s, clockid_t *clock) { | |
3186 | assert_return(s, -EINVAL); | |
3187 | assert_return(clock, -EINVAL); | |
3188 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
2eeff0f4 | 3189 | assert_return(!event_origin_changed(s->event), -ECHILD); |
eaa3cbef | 3190 | |
6a0f1f6d | 3191 | *clock = event_source_type_to_clock(s->type); |
eaa3cbef LP |
3192 | return 0; |
3193 | } | |
3194 | ||
f7262a9f | 3195 | _public_ int sd_event_source_get_child_pid(sd_event_source *s, pid_t *pid) { |
4bee8012 LP |
3196 | assert_return(s, -EINVAL); |
3197 | assert_return(pid, -EINVAL); | |
3198 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2eeff0f4 | 3199 | assert_return(!event_origin_changed(s->event), -ECHILD); |
4bee8012 LP |
3200 | |
3201 | *pid = s->child.pid; | |
3202 | return 0; | |
3203 | } | |
3204 | ||
f8f3f926 LP |
3205 | _public_ int sd_event_source_get_child_pidfd(sd_event_source *s) { |
3206 | assert_return(s, -EINVAL); | |
3207 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2eeff0f4 | 3208 | assert_return(!event_origin_changed(s->event), -ECHILD); |
f8f3f926 LP |
3209 | |
3210 | if (s->child.pidfd < 0) | |
3211 | return -EOPNOTSUPP; | |
3212 | ||
3213 | return s->child.pidfd; | |
3214 | } | |
3215 | ||
3216 | _public_ int sd_event_source_send_child_signal(sd_event_source *s, int sig, const siginfo_t *si, unsigned flags) { | |
3217 | assert_return(s, -EINVAL); | |
3218 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2eeff0f4 | 3219 | assert_return(!event_origin_changed(s->event), -ECHILD); |
f8f3f926 LP |
3220 | assert_return(SIGNAL_VALID(sig), -EINVAL); |
3221 | ||
3222 | /* If we already have seen indication the process exited refuse sending a signal early. This way we | |
3223 | * can be sure we don't accidentally kill the wrong process on PID reuse when pidfds are not | |
3224 | * available. */ | |
3225 | if (s->child.exited) | |
3226 | return -ESRCH; | |
3227 | ||
3228 | if (s->child.pidfd >= 0) { | |
3229 | siginfo_t copy; | |
3230 | ||
3231 | /* pidfd_send_signal() changes the siginfo_t argument. This is weird, let's hence copy the | |
3232 | * structure here */ | |
3233 | if (si) | |
3234 | copy = *si; | |
3235 | ||
3236 | if (pidfd_send_signal(s->child.pidfd, sig, si ? © : NULL, 0) < 0) { | |
3237 | /* Let's propagate the error only if the system call is not implemented or prohibited */ | |
3238 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
3239 | return -errno; | |
3240 | } else | |
3241 | return 0; | |
3242 | } | |
3243 | ||
3244 | /* Flags are only supported for pidfd_send_signal(), not for rt_sigqueueinfo(), hence let's refuse | |
3245 | * this here. */ | |
3246 | if (flags != 0) | |
3247 | return -EOPNOTSUPP; | |
3248 | ||
3249 | if (si) { | |
3250 | /* We use rt_sigqueueinfo() only if siginfo_t is specified. */ | |
3251 | siginfo_t copy = *si; | |
3252 | ||
3253 | if (rt_sigqueueinfo(s->child.pid, sig, ©) < 0) | |
3254 | return -errno; | |
3255 | } else if (kill(s->child.pid, sig) < 0) | |
3256 | return -errno; | |
3257 | ||
3258 | return 0; | |
3259 | } | |
3260 | ||
3261 | _public_ int sd_event_source_get_child_pidfd_own(sd_event_source *s) { | |
3262 | assert_return(s, -EINVAL); | |
3263 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2eeff0f4 | 3264 | assert_return(!event_origin_changed(s->event), -ECHILD); |
f8f3f926 LP |
3265 | |
3266 | if (s->child.pidfd < 0) | |
3267 | return -EOPNOTSUPP; | |
3268 | ||
3269 | return s->child.pidfd_owned; | |
3270 | } | |
3271 | ||
3272 | _public_ int sd_event_source_set_child_pidfd_own(sd_event_source *s, int own) { | |
3273 | assert_return(s, -EINVAL); | |
3274 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2eeff0f4 | 3275 | assert_return(!event_origin_changed(s->event), -ECHILD); |
f8f3f926 LP |
3276 | |
3277 | if (s->child.pidfd < 0) | |
3278 | return -EOPNOTSUPP; | |
3279 | ||
3280 | s->child.pidfd_owned = own; | |
3281 | return 0; | |
3282 | } | |
3283 | ||
3284 | _public_ int sd_event_source_get_child_process_own(sd_event_source *s) { | |
3285 | assert_return(s, -EINVAL); | |
3286 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2eeff0f4 | 3287 | assert_return(!event_origin_changed(s->event), -ECHILD); |
f8f3f926 LP |
3288 | |
3289 | return s->child.process_owned; | |
3290 | } | |
3291 | ||
3292 | _public_ int sd_event_source_set_child_process_own(sd_event_source *s, int own) { | |
3293 | assert_return(s, -EINVAL); | |
3294 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
2eeff0f4 | 3295 | assert_return(!event_origin_changed(s->event), -ECHILD); |
f8f3f926 LP |
3296 | |
3297 | s->child.process_owned = own; | |
3298 | return 0; | |
3299 | } | |
3300 | ||
c1ab4458 | 3301 | _public_ int sd_event_source_get_inotify_mask(sd_event_source *s, uint32_t *ret) { |
97ef5391 | 3302 | assert_return(s, -EINVAL); |
c1ab4458 | 3303 | assert_return(ret, -EINVAL); |
97ef5391 | 3304 | assert_return(s->type == SOURCE_INOTIFY, -EDOM); |
2eeff0f4 | 3305 | assert_return(!event_origin_changed(s->event), -ECHILD); |
97ef5391 | 3306 | |
c1ab4458 | 3307 | *ret = s->inotify.mask; |
97ef5391 LP |
3308 | return 0; |
3309 | } | |
3310 | ||
4a7cd0ca | 3311 | _public_ int sd_event_source_get_inotify_path(sd_event_source *s, const char **ret) { |
74c4231c YW |
3312 | assert_return(s, -EINVAL); |
3313 | assert_return(ret, -EINVAL); | |
3314 | assert_return(s->type == SOURCE_INOTIFY, -EDOM); | |
3315 | assert_return(!event_origin_changed(s->event), -ECHILD); | |
3316 | ||
4a7cd0ca YW |
3317 | if (!s->inotify.inode_data) |
3318 | return -ESTALE; /* already disconnected. */ | |
74c4231c | 3319 | |
4a7cd0ca YW |
3320 | if (!s->inotify.inode_data->path) |
3321 | return -ENOSYS; /* /proc was not mounted? */ | |
3322 | ||
3323 | *ret = s->inotify.inode_data->path; | |
3324 | return 0; | |
74c4231c YW |
3325 | } |
3326 | ||
718db961 | 3327 | _public_ int sd_event_source_set_prepare(sd_event_source *s, sd_event_handler_t callback) { |
fd38203a LP |
3328 | int r; |
3329 | ||
da7e457c | 3330 | assert_return(s, -EINVAL); |
6203e07a | 3331 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c | 3332 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 3333 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd38203a LP |
3334 | |
3335 | if (s->prepare == callback) | |
3336 | return 0; | |
3337 | ||
3338 | if (callback && s->prepare) { | |
3339 | s->prepare = callback; | |
3340 | return 0; | |
3341 | } | |
3342 | ||
3343 | r = prioq_ensure_allocated(&s->event->prepare, prepare_prioq_compare); | |
3344 | if (r < 0) | |
3345 | return r; | |
3346 | ||
3347 | s->prepare = callback; | |
3348 | ||
3349 | if (callback) { | |
3350 | r = prioq_put(s->event->prepare, s, &s->prepare_index); | |
3351 | if (r < 0) | |
3352 | return r; | |
3353 | } else | |
3354 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
3355 | ||
3356 | return 0; | |
3357 | } | |
3358 | ||
f7262a9f | 3359 | _public_ void* sd_event_source_get_userdata(sd_event_source *s) { |
da7e457c | 3360 | assert_return(s, NULL); |
2eeff0f4 | 3361 | assert_return(!event_origin_changed(s->event), NULL); |
fd38203a LP |
3362 | |
3363 | return s->userdata; | |
3364 | } | |
3365 | ||
8f726607 LP |
3366 | _public_ void *sd_event_source_set_userdata(sd_event_source *s, void *userdata) { |
3367 | void *ret; | |
3368 | ||
3369 | assert_return(s, NULL); | |
2eeff0f4 | 3370 | assert_return(!event_origin_changed(s->event), NULL); |
8f726607 LP |
3371 | |
3372 | ret = s->userdata; | |
3373 | s->userdata = userdata; | |
3374 | ||
3375 | return ret; | |
3376 | } | |
3377 | ||
b6d5481b LP |
3378 | static int event_source_enter_ratelimited(sd_event_source *s) { |
3379 | int r; | |
3380 | ||
3381 | assert(s); | |
3382 | ||
3383 | /* When an event source becomes ratelimited, we place it in the CLOCK_MONOTONIC priority queue, with | |
3384 | * the end of the rate limit time window, much as if it was a timer event source. */ | |
3385 | ||
3386 | if (s->ratelimited) | |
3387 | return 0; /* Already ratelimited, this is a NOP hence */ | |
3388 | ||
3389 | /* Make sure we can install a CLOCK_MONOTONIC event further down. */ | |
3390 | r = setup_clock_data(s->event, &s->event->monotonic, CLOCK_MONOTONIC); | |
3391 | if (r < 0) | |
3392 | return r; | |
3393 | ||
3394 | /* Timer event sources are already using the earliest/latest queues for the timer scheduling. Let's | |
3395 | * first remove them from the prioq appropriate for their own clock, so that we can use the prioq | |
3396 | * fields of the event source then for adding it to the CLOCK_MONOTONIC prioq instead. */ | |
3397 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
3398 | event_source_time_prioq_remove(s, event_get_clock_data(s->event, s->type)); | |
3399 | ||
3400 | /* Now, let's add the event source to the monotonic clock instead */ | |
3401 | r = event_source_time_prioq_put(s, &s->event->monotonic); | |
3402 | if (r < 0) | |
3403 | goto fail; | |
3404 | ||
3405 | /* And let's take the event source officially offline */ | |
3406 | r = event_source_offline(s, s->enabled, /* ratelimited= */ true); | |
3407 | if (r < 0) { | |
3408 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
3409 | goto fail; | |
3410 | } | |
3411 | ||
3412 | event_source_pp_prioq_reshuffle(s); | |
3413 | ||
3414 | log_debug("Event source %p (%s) entered rate limit state.", s, strna(s->description)); | |
3415 | return 0; | |
3416 | ||
3417 | fail: | |
3418 | /* Reinstall time event sources in the priority queue as before. This shouldn't fail, since the queue | |
3419 | * space for it should already be allocated. */ | |
3420 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
3421 | assert_se(event_source_time_prioq_put(s, event_get_clock_data(s->event, s->type)) >= 0); | |
3422 | ||
3423 | return r; | |
3424 | } | |
3425 | ||
fd69f224 | 3426 | static int event_source_leave_ratelimit(sd_event_source *s, bool run_callback) { |
b6d5481b LP |
3427 | int r; |
3428 | ||
3429 | assert(s); | |
3430 | ||
3431 | if (!s->ratelimited) | |
3432 | return 0; | |
3433 | ||
3434 | /* Let's take the event source out of the monotonic prioq first. */ | |
3435 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
3436 | ||
3437 | /* Let's then add the event source to its native clock prioq again — if this is a timer event source */ | |
3438 | if (EVENT_SOURCE_IS_TIME(s->type)) { | |
3439 | r = event_source_time_prioq_put(s, event_get_clock_data(s->event, s->type)); | |
3440 | if (r < 0) | |
3441 | goto fail; | |
3442 | } | |
3443 | ||
3444 | /* Let's try to take it online again. */ | |
3445 | r = event_source_online(s, s->enabled, /* ratelimited= */ false); | |
3446 | if (r < 0) { | |
3447 | /* Do something roughly sensible when this failed: undo the two prioq ops above */ | |
3448 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
3449 | event_source_time_prioq_remove(s, event_get_clock_data(s->event, s->type)); | |
3450 | ||
3451 | goto fail; | |
3452 | } | |
3453 | ||
3454 | event_source_pp_prioq_reshuffle(s); | |
3455 | ratelimit_reset(&s->rate_limit); | |
3456 | ||
3457 | log_debug("Event source %p (%s) left rate limit state.", s, strna(s->description)); | |
fd69f224 MS |
3458 | |
3459 | if (run_callback && s->ratelimit_expire_callback) { | |
3460 | s->dispatching = true; | |
3461 | r = s->ratelimit_expire_callback(s, s->userdata); | |
3462 | s->dispatching = false; | |
3463 | ||
3464 | if (r < 0) { | |
3465 | log_debug_errno(r, "Ratelimit expiry callback of event source %s (type %s) returned error, %s: %m", | |
3466 | strna(s->description), | |
3467 | event_source_type_to_string(s->type), | |
3468 | s->exit_on_failure ? "exiting" : "disabling"); | |
3469 | ||
3470 | if (s->exit_on_failure) | |
3471 | (void) sd_event_exit(s->event, r); | |
3472 | } | |
3473 | ||
3474 | if (s->n_ref == 0) | |
3475 | source_free(s); | |
3476 | else if (r < 0) | |
0a040e64 | 3477 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
fd69f224 MS |
3478 | |
3479 | return 1; | |
3480 | } | |
3481 | ||
b6d5481b LP |
3482 | return 0; |
3483 | ||
3484 | fail: | |
3485 | /* Do something somewhat reasonable when we cannot move an event sources out of ratelimited mode: | |
3486 | * simply put it back in it, maybe we can then process it more successfully next iteration. */ | |
3487 | assert_se(event_source_time_prioq_put(s, &s->event->monotonic) >= 0); | |
3488 | ||
3489 | return r; | |
3490 | } | |
3491 | ||
c2ba3ad6 LP |
3492 | static usec_t sleep_between(sd_event *e, usec_t a, usec_t b) { |
3493 | usec_t c; | |
3494 | assert(e); | |
3495 | assert(a <= b); | |
3496 | ||
3497 | if (a <= 0) | |
3498 | return 0; | |
393003e1 LP |
3499 | if (a >= USEC_INFINITY) |
3500 | return USEC_INFINITY; | |
c2ba3ad6 LP |
3501 | |
3502 | if (b <= a + 1) | |
3503 | return a; | |
3504 | ||
52444dc4 LP |
3505 | initialize_perturb(e); |
3506 | ||
c2ba3ad6 LP |
3507 | /* |
3508 | Find a good time to wake up again between times a and b. We | |
3509 | have two goals here: | |
3510 | ||
3511 | a) We want to wake up as seldom as possible, hence prefer | |
3512 | later times over earlier times. | |
3513 | ||
3514 | b) But if we have to wake up, then let's make sure to | |
3515 | dispatch as much as possible on the entire system. | |
3516 | ||
3517 | We implement this by waking up everywhere at the same time | |
850516e0 | 3518 | within any given minute if we can, synchronised via the |
c2ba3ad6 | 3519 | perturbation value determined from the boot ID. If we can't, |
ba276c81 LP |
3520 | then we try to find the same spot in every 10s, then 1s and |
3521 | then 250ms step. Otherwise, we pick the last possible time | |
3522 | to wake up. | |
c2ba3ad6 LP |
3523 | */ |
3524 | ||
850516e0 LP |
3525 | c = (b / USEC_PER_MINUTE) * USEC_PER_MINUTE + e->perturb; |
3526 | if (c >= b) { | |
3527 | if (_unlikely_(c < USEC_PER_MINUTE)) | |
3528 | return b; | |
3529 | ||
3530 | c -= USEC_PER_MINUTE; | |
3531 | } | |
3532 | ||
ba276c81 LP |
3533 | if (c >= a) |
3534 | return c; | |
3535 | ||
3536 | c = (b / (USEC_PER_SEC*10)) * (USEC_PER_SEC*10) + (e->perturb % (USEC_PER_SEC*10)); | |
3537 | if (c >= b) { | |
3538 | if (_unlikely_(c < USEC_PER_SEC*10)) | |
3539 | return b; | |
3540 | ||
3541 | c -= USEC_PER_SEC*10; | |
3542 | } | |
3543 | ||
850516e0 LP |
3544 | if (c >= a) |
3545 | return c; | |
3546 | ||
3547 | c = (b / USEC_PER_SEC) * USEC_PER_SEC + (e->perturb % USEC_PER_SEC); | |
c2ba3ad6 LP |
3548 | if (c >= b) { |
3549 | if (_unlikely_(c < USEC_PER_SEC)) | |
3550 | return b; | |
3551 | ||
3552 | c -= USEC_PER_SEC; | |
3553 | } | |
3554 | ||
3555 | if (c >= a) | |
3556 | return c; | |
3557 | ||
3558 | c = (b / (USEC_PER_MSEC*250)) * (USEC_PER_MSEC*250) + (e->perturb % (USEC_PER_MSEC*250)); | |
3559 | if (c >= b) { | |
3560 | if (_unlikely_(c < USEC_PER_MSEC*250)) | |
3561 | return b; | |
3562 | ||
3563 | c -= USEC_PER_MSEC*250; | |
3564 | } | |
3565 | ||
3566 | if (c >= a) | |
3567 | return c; | |
3568 | ||
3569 | return b; | |
3570 | } | |
3571 | ||
fd38203a LP |
3572 | static int event_arm_timer( |
3573 | sd_event *e, | |
6a0f1f6d | 3574 | struct clock_data *d) { |
fd38203a LP |
3575 | |
3576 | struct itimerspec its = {}; | |
c2ba3ad6 LP |
3577 | sd_event_source *a, *b; |
3578 | usec_t t; | |
fd38203a | 3579 | |
cde93897 | 3580 | assert(e); |
6a0f1f6d | 3581 | assert(d); |
fd38203a | 3582 | |
d06441da | 3583 | if (!d->needs_rearm) |
212bbb17 | 3584 | return 0; |
7e2bf71c YW |
3585 | |
3586 | d->needs_rearm = false; | |
212bbb17 | 3587 | |
6a0f1f6d | 3588 | a = prioq_peek(d->earliest); |
19947509 | 3589 | assert(!a || EVENT_SOURCE_USES_TIME_PRIOQ(a->type)); |
b6d5481b | 3590 | if (!a || a->enabled == SD_EVENT_OFF || time_event_source_next(a) == USEC_INFINITY) { |
72aedc1e | 3591 | |
6a0f1f6d | 3592 | if (d->fd < 0) |
c57b5ca3 LP |
3593 | return 0; |
3594 | ||
3a43da28 | 3595 | if (d->next == USEC_INFINITY) |
72aedc1e LP |
3596 | return 0; |
3597 | ||
3598 | /* disarm */ | |
15c689d7 LP |
3599 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
3600 | return -errno; | |
72aedc1e | 3601 | |
3a43da28 | 3602 | d->next = USEC_INFINITY; |
fd38203a | 3603 | return 0; |
72aedc1e | 3604 | } |
fd38203a | 3605 | |
6a0f1f6d | 3606 | b = prioq_peek(d->latest); |
19947509 ZJS |
3607 | assert(!b || EVENT_SOURCE_USES_TIME_PRIOQ(b->type)); |
3608 | assert(b && b->enabled != SD_EVENT_OFF); | |
c2ba3ad6 | 3609 | |
b6d5481b | 3610 | t = sleep_between(e, time_event_source_next(a), time_event_source_latest(b)); |
6a0f1f6d | 3611 | if (d->next == t) |
fd38203a LP |
3612 | return 0; |
3613 | ||
6a0f1f6d | 3614 | assert_se(d->fd >= 0); |
fd38203a | 3615 | |
c2ba3ad6 | 3616 | if (t == 0) { |
1751bdde | 3617 | /* We don't want to disarm here, just mean some time looooong ago. */ |
fd38203a LP |
3618 | its.it_value.tv_sec = 0; |
3619 | its.it_value.tv_nsec = 1; | |
3620 | } else | |
c2ba3ad6 | 3621 | timespec_store(&its.it_value, t); |
fd38203a | 3622 | |
15c689d7 | 3623 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 | 3624 | return -errno; |
fd38203a | 3625 | |
6a0f1f6d | 3626 | d->next = t; |
fd38203a LP |
3627 | return 0; |
3628 | } | |
3629 | ||
9a800b56 | 3630 | static int process_io(sd_event *e, sd_event_source *s, uint32_t revents) { |
fd38203a LP |
3631 | assert(e); |
3632 | assert(s); | |
3633 | assert(s->type == SOURCE_IO); | |
3634 | ||
9a800b56 LP |
3635 | /* If the event source was already pending, we just OR in the |
3636 | * new revents, otherwise we reset the value. The ORing is | |
3637 | * necessary to handle EPOLLONESHOT events properly where | |
3638 | * readability might happen independently of writability, and | |
3639 | * we need to keep track of both */ | |
3640 | ||
3641 | if (s->pending) | |
3642 | s->io.revents |= revents; | |
3643 | else | |
3644 | s->io.revents = revents; | |
fd38203a | 3645 | |
fd38203a LP |
3646 | return source_set_pending(s, true); |
3647 | } | |
3648 | ||
72aedc1e | 3649 | static int flush_timer(sd_event *e, int fd, uint32_t events, usec_t *next) { |
fd38203a LP |
3650 | uint64_t x; |
3651 | ssize_t ss; | |
3652 | ||
3653 | assert(e); | |
da7e457c | 3654 | assert(fd >= 0); |
72aedc1e | 3655 | |
305f78bf | 3656 | assert_return(events == EPOLLIN, -EIO); |
fd38203a LP |
3657 | |
3658 | ss = read(fd, &x, sizeof(x)); | |
3659 | if (ss < 0) { | |
8add30a0 | 3660 | if (ERRNO_IS_TRANSIENT(errno)) |
fd38203a LP |
3661 | return 0; |
3662 | ||
3663 | return -errno; | |
3664 | } | |
3665 | ||
8d35dae7 | 3666 | if (_unlikely_(ss != sizeof(x))) |
fd38203a LP |
3667 | return -EIO; |
3668 | ||
cde93897 | 3669 | if (next) |
3a43da28 | 3670 | *next = USEC_INFINITY; |
72aedc1e | 3671 | |
fd38203a LP |
3672 | return 0; |
3673 | } | |
3674 | ||
305f78bf LP |
3675 | static int process_timer( |
3676 | sd_event *e, | |
3677 | usec_t n, | |
6a0f1f6d | 3678 | struct clock_data *d) { |
305f78bf | 3679 | |
fd38203a | 3680 | sd_event_source *s; |
fd69f224 | 3681 | bool callback_invoked = false; |
fd38203a LP |
3682 | int r; |
3683 | ||
3684 | assert(e); | |
6a0f1f6d | 3685 | assert(d); |
fd38203a LP |
3686 | |
3687 | for (;;) { | |
6a0f1f6d | 3688 | s = prioq_peek(d->earliest); |
19947509 ZJS |
3689 | assert(!s || EVENT_SOURCE_USES_TIME_PRIOQ(s->type)); |
3690 | ||
b6d5481b LP |
3691 | if (!s || time_event_source_next(s) > n) |
3692 | break; | |
3693 | ||
3694 | if (s->ratelimited) { | |
3695 | /* This is an event sources whose ratelimit window has ended. Let's turn it on | |
3696 | * again. */ | |
3697 | assert(s->ratelimited); | |
3698 | ||
fd69f224 | 3699 | r = event_source_leave_ratelimit(s, /* run_callback */ true); |
b6d5481b LP |
3700 | if (r < 0) |
3701 | return r; | |
fd69f224 MS |
3702 | else if (r == 1) |
3703 | callback_invoked = true; | |
b6d5481b LP |
3704 | |
3705 | continue; | |
3706 | } | |
3707 | ||
3708 | if (s->enabled == SD_EVENT_OFF || s->pending) | |
fd38203a LP |
3709 | break; |
3710 | ||
3711 | r = source_set_pending(s, true); | |
3712 | if (r < 0) | |
3713 | return r; | |
3714 | ||
e1951c16 | 3715 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
3716 | } |
3717 | ||
fd69f224 | 3718 | return callback_invoked; |
fd38203a LP |
3719 | } |
3720 | ||
efd3be9d YW |
3721 | static int process_child(sd_event *e, int64_t threshold, int64_t *ret_min_priority) { |
3722 | int64_t min_priority = threshold; | |
3723 | bool something_new = false; | |
fd38203a | 3724 | sd_event_source *s; |
fd38203a LP |
3725 | int r; |
3726 | ||
3727 | assert(e); | |
efd3be9d YW |
3728 | assert(ret_min_priority); |
3729 | ||
3730 | if (!e->need_process_child) { | |
3731 | *ret_min_priority = min_priority; | |
3732 | return 0; | |
3733 | } | |
fd38203a | 3734 | |
c2ba3ad6 LP |
3735 | e->need_process_child = false; |
3736 | ||
91c70071 YW |
3737 | /* So, this is ugly. We iteratively invoke waitid() with P_PID + WNOHANG for each PID we wait |
3738 | * for, instead of using P_ALL. This is because we only want to get child information of very | |
3739 | * specific child processes, and not all of them. We might not have processed the SIGCHLD event | |
3740 | * of a previous invocation and we don't want to maintain a unbounded *per-child* event queue, | |
3741 | * hence we really don't want anything flushed out of the kernel's queue that we don't care | |
3742 | * about. Since this is O(n) this means that if you have a lot of processes you probably want | |
3743 | * to handle SIGCHLD yourself. | |
3744 | * | |
3745 | * We do not reap the children here (by using WNOWAIT), this is only done after the event | |
3746 | * source is dispatched so that the callback still sees the process as a zombie. */ | |
fd38203a | 3747 | |
90e74a66 | 3748 | HASHMAP_FOREACH(s, e->child_sources) { |
fd38203a LP |
3749 | assert(s->type == SOURCE_CHILD); |
3750 | ||
efd3be9d YW |
3751 | if (s->priority > threshold) |
3752 | continue; | |
3753 | ||
fd38203a LP |
3754 | if (s->pending) |
3755 | continue; | |
3756 | ||
b6d5481b | 3757 | if (event_source_is_offline(s)) |
fd38203a LP |
3758 | continue; |
3759 | ||
f8f3f926 LP |
3760 | if (s->child.exited) |
3761 | continue; | |
3762 | ||
91c70071 YW |
3763 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
3764 | /* There's a usable pidfd known for this event source? Then don't waitid() for | |
3765 | * it here */ | |
f8f3f926 LP |
3766 | continue; |
3767 | ||
fd38203a | 3768 | zero(s->child.siginfo); |
15c689d7 LP |
3769 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, |
3770 | WNOHANG | (s->child.options & WEXITED ? WNOWAIT : 0) | s->child.options) < 0) | |
bfd9bfcc | 3771 | return negative_errno(); |
fd38203a LP |
3772 | |
3773 | if (s->child.siginfo.si_pid != 0) { | |
945c2931 | 3774 | bool zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 3775 | |
f8f3f926 LP |
3776 | if (zombie) |
3777 | s->child.exited = true; | |
3778 | ||
08cd1552 | 3779 | if (!zombie && (s->child.options & WEXITED)) { |
91c70071 YW |
3780 | /* If the child isn't dead then let's immediately remove the state |
3781 | * change from the queue, since there's no benefit in leaving it | |
3782 | * queued. */ | |
08cd1552 LP |
3783 | |
3784 | assert(s->child.options & (WSTOPPED|WCONTINUED)); | |
a5d27871 | 3785 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|(s->child.options & (WSTOPPED|WCONTINUED))); |
08cd1552 LP |
3786 | } |
3787 | ||
fd38203a LP |
3788 | r = source_set_pending(s, true); |
3789 | if (r < 0) | |
3790 | return r; | |
efd3be9d YW |
3791 | if (r > 0) { |
3792 | something_new = true; | |
3793 | min_priority = MIN(min_priority, s->priority); | |
3794 | } | |
fd38203a LP |
3795 | } |
3796 | } | |
3797 | ||
efd3be9d YW |
3798 | *ret_min_priority = min_priority; |
3799 | return something_new; | |
fd38203a LP |
3800 | } |
3801 | ||
f8f3f926 LP |
3802 | static int process_pidfd(sd_event *e, sd_event_source *s, uint32_t revents) { |
3803 | assert(e); | |
3804 | assert(s); | |
3805 | assert(s->type == SOURCE_CHILD); | |
3806 | ||
3807 | if (s->pending) | |
3808 | return 0; | |
3809 | ||
b6d5481b | 3810 | if (event_source_is_offline(s)) |
f8f3f926 LP |
3811 | return 0; |
3812 | ||
3813 | if (!EVENT_SOURCE_WATCH_PIDFD(s)) | |
3814 | return 0; | |
3815 | ||
3816 | zero(s->child.siginfo); | |
3817 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG | WNOWAIT | s->child.options) < 0) | |
3818 | return -errno; | |
3819 | ||
3820 | if (s->child.siginfo.si_pid == 0) | |
3821 | return 0; | |
3822 | ||
3823 | if (IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) | |
3824 | s->child.exited = true; | |
3825 | ||
3826 | return source_set_pending(s, true); | |
3827 | } | |
3828 | ||
efd3be9d | 3829 | static int process_signal(sd_event *e, struct signal_data *d, uint32_t events, int64_t *min_priority) { |
fd38203a LP |
3830 | int r; |
3831 | ||
da7e457c | 3832 | assert(e); |
97ef5391 | 3833 | assert(d); |
305f78bf | 3834 | assert_return(events == EPOLLIN, -EIO); |
efd3be9d | 3835 | assert(min_priority); |
fd38203a | 3836 | |
91c70071 YW |
3837 | /* If there's a signal queued on this priority and SIGCHLD is on this priority too, then make |
3838 | * sure to recheck the children we watch. This is because we only ever dequeue the first signal | |
3839 | * per priority, and if we dequeue one, and SIGCHLD might be enqueued later we wouldn't know, | |
3840 | * but we might have higher priority children we care about hence we need to check that | |
3841 | * explicitly. */ | |
9da4cb2b LP |
3842 | |
3843 | if (sigismember(&d->sigset, SIGCHLD)) | |
3844 | e->need_process_child = true; | |
3845 | ||
91c70071 | 3846 | /* If there's already an event source pending for this priority we don't read another */ |
9da4cb2b LP |
3847 | if (d->current) |
3848 | return 0; | |
3849 | ||
fd38203a | 3850 | for (;;) { |
0eb2e0e3 | 3851 | struct signalfd_siginfo si; |
7057bd99 | 3852 | ssize_t n; |
92daebc0 | 3853 | sd_event_source *s = NULL; |
fd38203a | 3854 | |
9da4cb2b | 3855 | n = read(d->fd, &si, sizeof(si)); |
7057bd99 | 3856 | if (n < 0) { |
8add30a0 | 3857 | if (ERRNO_IS_TRANSIENT(errno)) |
efd3be9d | 3858 | return 0; |
fd38203a LP |
3859 | |
3860 | return -errno; | |
3861 | } | |
3862 | ||
7057bd99 | 3863 | if (_unlikely_(n != sizeof(si))) |
fd38203a LP |
3864 | return -EIO; |
3865 | ||
6eb7c172 | 3866 | assert(SIGNAL_VALID(si.ssi_signo)); |
7057bd99 | 3867 | |
92daebc0 LP |
3868 | if (e->signal_sources) |
3869 | s = e->signal_sources[si.ssi_signo]; | |
92daebc0 LP |
3870 | if (!s) |
3871 | continue; | |
9da4cb2b LP |
3872 | if (s->pending) |
3873 | continue; | |
fd38203a LP |
3874 | |
3875 | s->signal.siginfo = si; | |
9da4cb2b LP |
3876 | d->current = s; |
3877 | ||
fd38203a LP |
3878 | r = source_set_pending(s, true); |
3879 | if (r < 0) | |
3880 | return r; | |
efd3be9d YW |
3881 | if (r > 0 && *min_priority >= s->priority) { |
3882 | *min_priority = s->priority; | |
3883 | return 1; /* an event source with smaller priority is queued. */ | |
3884 | } | |
9da4cb2b | 3885 | |
efd3be9d | 3886 | return 0; |
fd38203a | 3887 | } |
fd38203a LP |
3888 | } |
3889 | ||
efd3be9d | 3890 | static int event_inotify_data_read(sd_event *e, struct inotify_data *d, uint32_t revents, int64_t threshold) { |
97ef5391 LP |
3891 | ssize_t n; |
3892 | ||
3893 | assert(e); | |
3894 | assert(d); | |
3895 | ||
3896 | assert_return(revents == EPOLLIN, -EIO); | |
3897 | ||
3898 | /* If there's already an event source pending for this priority, don't read another */ | |
3899 | if (d->n_pending > 0) | |
3900 | return 0; | |
3901 | ||
3902 | /* Is the read buffer non-empty? If so, let's not read more */ | |
3903 | if (d->buffer_filled > 0) | |
3904 | return 0; | |
3905 | ||
efd3be9d YW |
3906 | if (d->priority > threshold) |
3907 | return 0; | |
3908 | ||
97ef5391 LP |
3909 | n = read(d->fd, &d->buffer, sizeof(d->buffer)); |
3910 | if (n < 0) { | |
8add30a0 | 3911 | if (ERRNO_IS_TRANSIENT(errno)) |
97ef5391 LP |
3912 | return 0; |
3913 | ||
3914 | return -errno; | |
3915 | } | |
3916 | ||
3917 | assert(n > 0); | |
3918 | d->buffer_filled = (size_t) n; | |
0601b958 | 3919 | LIST_PREPEND(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
3920 | |
3921 | return 1; | |
3922 | } | |
3923 | ||
3924 | static void event_inotify_data_drop(sd_event *e, struct inotify_data *d, size_t sz) { | |
3925 | assert(e); | |
3926 | assert(d); | |
3927 | assert(sz <= d->buffer_filled); | |
3928 | ||
3929 | if (sz == 0) | |
3930 | return; | |
3931 | ||
3932 | /* Move the rest to the buffer to the front, in order to get things properly aligned again */ | |
3933 | memmove(d->buffer.raw, d->buffer.raw + sz, d->buffer_filled - sz); | |
3934 | d->buffer_filled -= sz; | |
3935 | ||
3936 | if (d->buffer_filled == 0) | |
0601b958 | 3937 | LIST_REMOVE(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
3938 | } |
3939 | ||
3940 | static int event_inotify_data_process(sd_event *e, struct inotify_data *d) { | |
3941 | int r; | |
3942 | ||
3943 | assert(e); | |
3944 | assert(d); | |
3945 | ||
3946 | /* If there's already an event source pending for this priority, don't read another */ | |
3947 | if (d->n_pending > 0) | |
3948 | return 0; | |
3949 | ||
3950 | while (d->buffer_filled > 0) { | |
3951 | size_t sz; | |
3952 | ||
3953 | /* Let's validate that the event structures are complete */ | |
3954 | if (d->buffer_filled < offsetof(struct inotify_event, name)) | |
3955 | return -EIO; | |
3956 | ||
3957 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3958 | if (d->buffer_filled < sz) | |
3959 | return -EIO; | |
3960 | ||
3961 | if (d->buffer.ev.mask & IN_Q_OVERFLOW) { | |
3962 | struct inode_data *inode_data; | |
97ef5391 LP |
3963 | |
3964 | /* The queue overran, let's pass this event to all event sources connected to this inotify | |
3965 | * object */ | |
3966 | ||
03677889 | 3967 | HASHMAP_FOREACH(inode_data, d->inodes) |
97ef5391 LP |
3968 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { |
3969 | ||
b6d5481b | 3970 | if (event_source_is_offline(s)) |
97ef5391 LP |
3971 | continue; |
3972 | ||
3973 | r = source_set_pending(s, true); | |
3974 | if (r < 0) | |
3975 | return r; | |
3976 | } | |
97ef5391 LP |
3977 | } else { |
3978 | struct inode_data *inode_data; | |
97ef5391 LP |
3979 | |
3980 | /* Find the inode object for this watch descriptor. If IN_IGNORED is set we also remove it from | |
3981 | * our watch descriptor table. */ | |
3982 | if (d->buffer.ev.mask & IN_IGNORED) { | |
3983 | ||
3984 | inode_data = hashmap_remove(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3985 | if (!inode_data) { | |
3986 | event_inotify_data_drop(e, d, sz); | |
3987 | continue; | |
3988 | } | |
3989 | ||
3990 | /* The watch descriptor was removed by the kernel, let's drop it here too */ | |
3991 | inode_data->wd = -1; | |
3992 | } else { | |
3993 | inode_data = hashmap_get(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3994 | if (!inode_data) { | |
3995 | event_inotify_data_drop(e, d, sz); | |
3996 | continue; | |
3997 | } | |
3998 | } | |
3999 | ||
4000 | /* Trigger all event sources that are interested in these events. Also trigger all event | |
4001 | * sources if IN_IGNORED or IN_UNMOUNT is set. */ | |
4002 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
4003 | ||
b6d5481b | 4004 | if (event_source_is_offline(s)) |
97ef5391 LP |
4005 | continue; |
4006 | ||
4007 | if ((d->buffer.ev.mask & (IN_IGNORED|IN_UNMOUNT)) == 0 && | |
4008 | (s->inotify.mask & d->buffer.ev.mask & IN_ALL_EVENTS) == 0) | |
4009 | continue; | |
4010 | ||
4011 | r = source_set_pending(s, true); | |
4012 | if (r < 0) | |
4013 | return r; | |
4014 | } | |
4015 | } | |
4016 | ||
4017 | /* Something pending now? If so, let's finish, otherwise let's read more. */ | |
4018 | if (d->n_pending > 0) | |
4019 | return 1; | |
4020 | } | |
4021 | ||
4022 | return 0; | |
4023 | } | |
4024 | ||
4025 | static int process_inotify(sd_event *e) { | |
97ef5391 LP |
4026 | int r, done = 0; |
4027 | ||
4028 | assert(e); | |
4029 | ||
0601b958 | 4030 | LIST_FOREACH(buffered, d, e->buffered_inotify_data_list) { |
97ef5391 LP |
4031 | r = event_inotify_data_process(e, d); |
4032 | if (r < 0) | |
4033 | return r; | |
4034 | if (r > 0) | |
b3a9d980 | 4035 | done++; |
97ef5391 LP |
4036 | } |
4037 | ||
4038 | return done; | |
4039 | } | |
4040 | ||
158fe190 LP |
4041 | static int process_memory_pressure(sd_event_source *s, uint32_t revents) { |
4042 | assert(s); | |
4043 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
4044 | ||
4045 | if (s->pending) | |
4046 | s->memory_pressure.revents |= revents; | |
4047 | else | |
4048 | s->memory_pressure.revents = revents; | |
4049 | ||
4050 | return source_set_pending(s, true); | |
4051 | } | |
4052 | ||
4053 | static int source_memory_pressure_write(sd_event_source *s) { | |
4054 | ssize_t n; | |
4055 | int r; | |
4056 | ||
4057 | assert(s); | |
4058 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
4059 | ||
4060 | /* once we start writing, the buffer is locked, we allow no further changes. */ | |
4061 | s->memory_pressure.locked = true; | |
4062 | ||
4063 | if (s->memory_pressure.write_buffer_size > 0) { | |
4064 | n = write(s->memory_pressure.fd, s->memory_pressure.write_buffer, s->memory_pressure.write_buffer_size); | |
4065 | if (n < 0) { | |
9897f5dd LP |
4066 | if (!ERRNO_IS_TRANSIENT(errno)) { |
4067 | /* If kernel is built with CONFIG_PSI_DEFAULT_DISABLED it will expose PSI | |
4068 | * files, but then generates EOPNOSUPP on read() and write() (instead of on | |
4069 | * open()!). This sucks hard, since we can only detect this kind of failure | |
4070 | * so late. Let's make the best of it, and turn off the event source like we | |
4071 | * do for failed event source handlers. */ | |
4072 | ||
4073 | log_debug_errno(errno, "Writing memory pressure settings to kernel failed, disabling memory pressure event source: %m"); | |
4074 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); | |
4075 | return 0; | |
4076 | } | |
158fe190 LP |
4077 | |
4078 | n = 0; | |
4079 | } | |
4080 | } else | |
4081 | n = 0; | |
4082 | ||
4083 | assert(n >= 0); | |
4084 | ||
4085 | if ((size_t) n == s->memory_pressure.write_buffer_size) { | |
4086 | s->memory_pressure.write_buffer = mfree(s->memory_pressure.write_buffer); | |
4087 | ||
4088 | if (n > 0) { | |
4089 | s->memory_pressure.write_buffer_size = 0; | |
4090 | ||
4091 | /* Update epoll events mask, since we have now written everything and don't care for EPOLLOUT anymore */ | |
4092 | r = source_memory_pressure_register(s, s->enabled); | |
4093 | if (r < 0) | |
4094 | return r; | |
4095 | } | |
4096 | } else if (n > 0) { | |
4097 | _cleanup_free_ void *c = NULL; | |
4098 | ||
4099 | assert((size_t) n < s->memory_pressure.write_buffer_size); | |
4100 | ||
4101 | c = memdup((uint8_t*) s->memory_pressure.write_buffer + n, s->memory_pressure.write_buffer_size - n); | |
4102 | if (!c) | |
4103 | return -ENOMEM; | |
4104 | ||
4105 | free_and_replace(s->memory_pressure.write_buffer, c); | |
4106 | s->memory_pressure.write_buffer_size -= n; | |
4107 | return 1; | |
4108 | } | |
4109 | ||
4110 | return 0; | |
4111 | } | |
4112 | ||
4113 | static int source_memory_pressure_initiate_dispatch(sd_event_source *s) { | |
4114 | int r; | |
4115 | ||
4116 | assert(s); | |
4117 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
4118 | ||
4119 | r = source_memory_pressure_write(s); | |
4120 | if (r < 0) | |
4121 | return r; | |
4122 | if (r > 0) | |
4123 | return 1; /* if we wrote something, then don't continue with dispatching user dispatch | |
4124 | * function. Instead, shortcut it so that we wait for next EPOLLOUT immediately. */ | |
4125 | ||
4126 | /* No pending incoming IO? Then let's not continue further */ | |
4127 | if ((s->memory_pressure.revents & (EPOLLIN|EPOLLPRI)) == 0) { | |
4128 | ||
4129 | /* Treat IO errors on the notifier the same ways errors returned from a callback */ | |
4130 | if ((s->memory_pressure.revents & (EPOLLHUP|EPOLLERR|EPOLLRDHUP)) != 0) | |
4131 | return -EIO; | |
4132 | ||
4133 | return 1; /* leave dispatch, we already processed everything */ | |
4134 | } | |
4135 | ||
4136 | if (s->memory_pressure.revents & EPOLLIN) { | |
4137 | uint8_t pipe_buf[PIPE_BUF]; | |
4138 | ssize_t n; | |
4139 | ||
4140 | /* If the fd is readable, then flush out anything that might be queued */ | |
4141 | ||
4142 | n = read(s->memory_pressure.fd, pipe_buf, sizeof(pipe_buf)); | |
4143 | if (n < 0 && !ERRNO_IS_TRANSIENT(errno)) | |
4144 | return -errno; | |
4145 | } | |
4146 | ||
4147 | return 0; /* go on, dispatch to user callback */ | |
4148 | } | |
4149 | ||
fd38203a | 4150 | static int source_dispatch(sd_event_source *s) { |
8f5c235d | 4151 | EventSourceType saved_type; |
c8e9d15c | 4152 | sd_event *saved_event; |
fe8245eb | 4153 | int r = 0; |
fd38203a LP |
4154 | |
4155 | assert(s); | |
6203e07a | 4156 | assert(s->pending || s->type == SOURCE_EXIT); |
fd38203a | 4157 | |
b778cba4 LP |
4158 | /* Save the event source type, here, so that we still know it after the event callback which might |
4159 | * invalidate the event. */ | |
8f5c235d LP |
4160 | saved_type = s->type; |
4161 | ||
de02634c | 4162 | /* Similarly, store a reference to the event loop object, so that we can still access it after the |
b778cba4 | 4163 | * callback might have invalidated/disconnected the event source. */ |
c8e9d15c YW |
4164 | saved_event = s->event; |
4165 | PROTECT_EVENT(saved_event); | |
b778cba4 | 4166 | |
de02634c | 4167 | /* Check if we hit the ratelimit for this event source, and if so, let's disable it. */ |
b6d5481b LP |
4168 | assert(!s->ratelimited); |
4169 | if (!ratelimit_below(&s->rate_limit)) { | |
4170 | r = event_source_enter_ratelimited(s); | |
4171 | if (r < 0) | |
4172 | return r; | |
4173 | ||
4174 | return 1; | |
4175 | } | |
4176 | ||
945c2931 | 4177 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { |
da7e457c LP |
4178 | r = source_set_pending(s, false); |
4179 | if (r < 0) | |
4180 | return r; | |
4181 | } | |
fd38203a | 4182 | |
6e9feda3 LP |
4183 | if (s->type != SOURCE_POST) { |
4184 | sd_event_source *z; | |
6e9feda3 | 4185 | |
de02634c | 4186 | /* If we execute a non-post source, let's mark all post sources as pending. */ |
6e9feda3 | 4187 | |
90e74a66 | 4188 | SET_FOREACH(z, s->event->post_sources) { |
b6d5481b | 4189 | if (event_source_is_offline(z)) |
6e9feda3 LP |
4190 | continue; |
4191 | ||
4192 | r = source_set_pending(z, true); | |
4193 | if (r < 0) | |
4194 | return r; | |
4195 | } | |
4196 | } | |
4197 | ||
158fe190 LP |
4198 | if (s->type == SOURCE_MEMORY_PRESSURE) { |
4199 | r = source_memory_pressure_initiate_dispatch(s); | |
4200 | if (r == -EIO) /* handle EIO errors similar to callback errors */ | |
4201 | goto finish; | |
4202 | if (r < 0) | |
4203 | return r; | |
4204 | if (r > 0) /* already handled */ | |
4205 | return 1; | |
4206 | } | |
4207 | ||
baf76283 LP |
4208 | if (s->enabled == SD_EVENT_ONESHOT) { |
4209 | r = sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
4210 | if (r < 0) |
4211 | return r; | |
4212 | } | |
4213 | ||
12179984 | 4214 | s->dispatching = true; |
b7484e2a | 4215 | |
fd38203a LP |
4216 | switch (s->type) { |
4217 | ||
4218 | case SOURCE_IO: | |
4219 | r = s->io.callback(s, s->io.fd, s->io.revents, s->userdata); | |
4220 | break; | |
4221 | ||
6a0f1f6d | 4222 | case SOURCE_TIME_REALTIME: |
a8548816 | 4223 | case SOURCE_TIME_BOOTTIME: |
6a0f1f6d LP |
4224 | case SOURCE_TIME_MONOTONIC: |
4225 | case SOURCE_TIME_REALTIME_ALARM: | |
4226 | case SOURCE_TIME_BOOTTIME_ALARM: | |
fd38203a LP |
4227 | r = s->time.callback(s, s->time.next, s->userdata); |
4228 | break; | |
4229 | ||
4230 | case SOURCE_SIGNAL: | |
4231 | r = s->signal.callback(s, &s->signal.siginfo, s->userdata); | |
4232 | break; | |
4233 | ||
08cd1552 LP |
4234 | case SOURCE_CHILD: { |
4235 | bool zombie; | |
4236 | ||
945c2931 | 4237 | zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 4238 | |
fd38203a | 4239 | r = s->child.callback(s, &s->child.siginfo, s->userdata); |
08cd1552 LP |
4240 | |
4241 | /* Now, reap the PID for good. */ | |
f8f3f926 | 4242 | if (zombie) { |
cc59d290 | 4243 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|WEXITED); |
f8f3f926 LP |
4244 | s->child.waited = true; |
4245 | } | |
08cd1552 | 4246 | |
fd38203a | 4247 | break; |
08cd1552 | 4248 | } |
fd38203a LP |
4249 | |
4250 | case SOURCE_DEFER: | |
4251 | r = s->defer.callback(s, s->userdata); | |
4252 | break; | |
da7e457c | 4253 | |
6e9feda3 LP |
4254 | case SOURCE_POST: |
4255 | r = s->post.callback(s, s->userdata); | |
4256 | break; | |
4257 | ||
6203e07a LP |
4258 | case SOURCE_EXIT: |
4259 | r = s->exit.callback(s, s->userdata); | |
da7e457c | 4260 | break; |
9d3e3aa5 | 4261 | |
97ef5391 LP |
4262 | case SOURCE_INOTIFY: { |
4263 | struct sd_event *e = s->event; | |
4264 | struct inotify_data *d; | |
4265 | size_t sz; | |
4266 | ||
4267 | assert(s->inotify.inode_data); | |
4268 | assert_se(d = s->inotify.inode_data->inotify_data); | |
4269 | ||
4270 | assert(d->buffer_filled >= offsetof(struct inotify_event, name)); | |
4271 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
4272 | assert(d->buffer_filled >= sz); | |
4273 | ||
53baf2ef LP |
4274 | /* If the inotify callback destroys the event source then this likely means we don't need to |
4275 | * watch the inode anymore, and thus also won't need the inotify object anymore. But if we'd | |
4276 | * free it immediately, then we couldn't drop the event from the inotify event queue without | |
4277 | * memory corruption anymore, as below. Hence, let's not free it immediately, but mark it | |
4278 | * "busy" with a counter (which will ensure it's not GC'ed away prematurely). Let's then | |
4279 | * explicitly GC it after we are done dropping the inotify event from the buffer. */ | |
4280 | d->n_busy++; | |
97ef5391 | 4281 | r = s->inotify.callback(s, &d->buffer.ev, s->userdata); |
53baf2ef | 4282 | d->n_busy--; |
97ef5391 | 4283 | |
53baf2ef LP |
4284 | /* When no event is pending anymore on this inotify object, then let's drop the event from |
4285 | * the inotify event queue buffer. */ | |
97ef5391 LP |
4286 | if (d->n_pending == 0) |
4287 | event_inotify_data_drop(e, d, sz); | |
4288 | ||
53baf2ef LP |
4289 | /* Now we don't want to access 'd' anymore, it's OK to GC now. */ |
4290 | event_gc_inotify_data(e, d); | |
97ef5391 LP |
4291 | break; |
4292 | } | |
4293 | ||
158fe190 LP |
4294 | case SOURCE_MEMORY_PRESSURE: |
4295 | r = s->memory_pressure.callback(s, s->userdata); | |
4296 | break; | |
4297 | ||
9d3e3aa5 | 4298 | case SOURCE_WATCHDOG: |
a71fe8b8 | 4299 | case _SOURCE_EVENT_SOURCE_TYPE_MAX: |
9f2a50a3 | 4300 | case _SOURCE_EVENT_SOURCE_TYPE_INVALID: |
04499a70 | 4301 | assert_not_reached(); |
fd38203a LP |
4302 | } |
4303 | ||
12179984 LP |
4304 | s->dispatching = false; |
4305 | ||
158fe190 | 4306 | finish: |
b778cba4 LP |
4307 | if (r < 0) { |
4308 | log_debug_errno(r, "Event source %s (type %s) returned error, %s: %m", | |
4309 | strna(s->description), | |
4310 | event_source_type_to_string(saved_type), | |
4311 | s->exit_on_failure ? "exiting" : "disabling"); | |
4312 | ||
4313 | if (s->exit_on_failure) | |
4314 | (void) sd_event_exit(saved_event, r); | |
4315 | } | |
12179984 LP |
4316 | |
4317 | if (s->n_ref == 0) | |
4318 | source_free(s); | |
4319 | else if (r < 0) | |
c3c50474 | 4320 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
b7484e2a | 4321 | |
6203e07a | 4322 | return 1; |
fd38203a LP |
4323 | } |
4324 | ||
4325 | static int event_prepare(sd_event *e) { | |
4326 | int r; | |
4327 | ||
4328 | assert(e); | |
4329 | ||
4330 | for (;;) { | |
4331 | sd_event_source *s; | |
4332 | ||
4333 | s = prioq_peek(e->prepare); | |
b6d5481b | 4334 | if (!s || s->prepare_iteration == e->iteration || event_source_is_offline(s)) |
fd38203a LP |
4335 | break; |
4336 | ||
4337 | s->prepare_iteration = e->iteration; | |
8656f4a6 | 4338 | prioq_reshuffle(e->prepare, s, &s->prepare_index); |
fd38203a LP |
4339 | |
4340 | assert(s->prepare); | |
12179984 | 4341 | s->dispatching = true; |
fd38203a | 4342 | r = s->prepare(s, s->userdata); |
12179984 LP |
4343 | s->dispatching = false; |
4344 | ||
b778cba4 LP |
4345 | if (r < 0) { |
4346 | log_debug_errno(r, "Prepare callback of event source %s (type %s) returned error, %s: %m", | |
4347 | strna(s->description), | |
4348 | event_source_type_to_string(s->type), | |
4349 | s->exit_on_failure ? "exiting" : "disabling"); | |
4350 | ||
4351 | if (s->exit_on_failure) | |
4352 | (void) sd_event_exit(e, r); | |
4353 | } | |
fd38203a | 4354 | |
12179984 LP |
4355 | if (s->n_ref == 0) |
4356 | source_free(s); | |
4357 | else if (r < 0) | |
c3c50474 | 4358 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
fd38203a LP |
4359 | } |
4360 | ||
4361 | return 0; | |
4362 | } | |
4363 | ||
6203e07a | 4364 | static int dispatch_exit(sd_event *e) { |
da7e457c LP |
4365 | sd_event_source *p; |
4366 | int r; | |
4367 | ||
4368 | assert(e); | |
4369 | ||
6203e07a | 4370 | p = prioq_peek(e->exit); |
19947509 ZJS |
4371 | assert(!p || p->type == SOURCE_EXIT); |
4372 | ||
b6d5481b | 4373 | if (!p || event_source_is_offline(p)) { |
da7e457c LP |
4374 | e->state = SD_EVENT_FINISHED; |
4375 | return 0; | |
4376 | } | |
4377 | ||
c8e9d15c | 4378 | PROTECT_EVENT(e); |
da7e457c | 4379 | e->iteration++; |
6203e07a | 4380 | e->state = SD_EVENT_EXITING; |
da7e457c | 4381 | r = source_dispatch(p); |
2b0c9ef7 | 4382 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
4383 | return r; |
4384 | } | |
4385 | ||
c2ba3ad6 LP |
4386 | static sd_event_source* event_next_pending(sd_event *e) { |
4387 | sd_event_source *p; | |
4388 | ||
da7e457c LP |
4389 | assert(e); |
4390 | ||
c2ba3ad6 LP |
4391 | p = prioq_peek(e->pending); |
4392 | if (!p) | |
4393 | return NULL; | |
4394 | ||
b6d5481b | 4395 | if (event_source_is_offline(p)) |
c2ba3ad6 LP |
4396 | return NULL; |
4397 | ||
4398 | return p; | |
4399 | } | |
4400 | ||
cde93897 LP |
4401 | static int arm_watchdog(sd_event *e) { |
4402 | struct itimerspec its = {}; | |
4403 | usec_t t; | |
cde93897 LP |
4404 | |
4405 | assert(e); | |
4406 | assert(e->watchdog_fd >= 0); | |
4407 | ||
4408 | t = sleep_between(e, | |
a595fb5c YW |
4409 | usec_add(e->watchdog_last, (e->watchdog_period / 2)), |
4410 | usec_add(e->watchdog_last, (e->watchdog_period * 3 / 4))); | |
cde93897 LP |
4411 | |
4412 | timespec_store(&its.it_value, t); | |
4413 | ||
75145780 LP |
4414 | /* Make sure we never set the watchdog to 0, which tells the |
4415 | * kernel to disable it. */ | |
4416 | if (its.it_value.tv_sec == 0 && its.it_value.tv_nsec == 0) | |
4417 | its.it_value.tv_nsec = 1; | |
4418 | ||
7c248223 | 4419 | return RET_NERRNO(timerfd_settime(e->watchdog_fd, TFD_TIMER_ABSTIME, &its, NULL)); |
cde93897 LP |
4420 | } |
4421 | ||
4422 | static int process_watchdog(sd_event *e) { | |
4423 | assert(e); | |
4424 | ||
4425 | if (!e->watchdog) | |
4426 | return 0; | |
4427 | ||
4428 | /* Don't notify watchdog too often */ | |
4429 | if (e->watchdog_last + e->watchdog_period / 4 > e->timestamp.monotonic) | |
4430 | return 0; | |
4431 | ||
4432 | sd_notify(false, "WATCHDOG=1"); | |
4433 | e->watchdog_last = e->timestamp.monotonic; | |
4434 | ||
4435 | return arm_watchdog(e); | |
4436 | } | |
4437 | ||
97ef5391 LP |
4438 | static void event_close_inode_data_fds(sd_event *e) { |
4439 | struct inode_data *d; | |
4440 | ||
4441 | assert(e); | |
4442 | ||
4443 | /* Close the fds pointing to the inodes to watch now. We need to close them as they might otherwise pin | |
4444 | * filesystems. But we can't close them right-away as we need them as long as the user still wants to make | |
365c2885 | 4445 | * adjustments to the event source, such as changing the priority (which requires us to remove and re-add a watch |
97ef5391 LP |
4446 | * for the inode). Hence, let's close them when entering the first iteration after they were added, as a |
4447 | * compromise. */ | |
4448 | ||
ed828563 | 4449 | while ((d = e->inode_data_to_close_list)) { |
97ef5391 LP |
4450 | assert(d->fd >= 0); |
4451 | d->fd = safe_close(d->fd); | |
4452 | ||
ed828563 | 4453 | LIST_REMOVE(to_close, e->inode_data_to_close_list, d); |
97ef5391 LP |
4454 | } |
4455 | } | |
4456 | ||
158fe190 LP |
4457 | static int event_memory_pressure_write_list(sd_event *e) { |
4458 | int r; | |
4459 | ||
4460 | assert(e); | |
4461 | ||
4462 | for (;;) { | |
4463 | sd_event_source *s; | |
4464 | ||
4465 | s = LIST_POP(memory_pressure.write_list, e->memory_pressure_write_list); | |
4466 | if (!s) | |
4467 | break; | |
4468 | ||
4469 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
4470 | assert(s->memory_pressure.write_buffer_size > 0); | |
4471 | s->memory_pressure.in_write_list = false; | |
4472 | ||
4473 | r = source_memory_pressure_write(s); | |
4474 | if (r < 0) | |
4475 | return r; | |
4476 | } | |
4477 | ||
4478 | return 0; | |
4479 | } | |
4480 | ||
c45a5a74 TG |
4481 | _public_ int sd_event_prepare(sd_event *e) { |
4482 | int r; | |
fd38203a | 4483 | |
da7e457c | 4484 | assert_return(e, -EINVAL); |
b937d761 | 4485 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 4486 | assert_return(!event_origin_changed(e), -ECHILD); |
da7e457c | 4487 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2b0c9ef7 | 4488 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 4489 | |
e5446015 LP |
4490 | /* Let's check that if we are a default event loop we are executed in the correct thread. We only do |
4491 | * this check here once, since gettid() is typically not cached, and thus want to minimize | |
4492 | * syscalls */ | |
4493 | assert_return(!e->default_event_ptr || e->tid == gettid(), -EREMOTEIO); | |
4494 | ||
f814c871 | 4495 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
c8e9d15c | 4496 | PROTECT_EVENT(e); |
f814c871 | 4497 | |
6203e07a | 4498 | if (e->exit_requested) |
c45a5a74 | 4499 | goto pending; |
fd38203a LP |
4500 | |
4501 | e->iteration++; | |
4502 | ||
0be6c2f6 | 4503 | e->state = SD_EVENT_PREPARING; |
fd38203a | 4504 | r = event_prepare(e); |
0be6c2f6 | 4505 | e->state = SD_EVENT_INITIAL; |
fd38203a | 4506 | if (r < 0) |
c45a5a74 | 4507 | return r; |
fd38203a | 4508 | |
158fe190 LP |
4509 | r = event_memory_pressure_write_list(e); |
4510 | if (r < 0) | |
4511 | return r; | |
4512 | ||
6a0f1f6d LP |
4513 | r = event_arm_timer(e, &e->realtime); |
4514 | if (r < 0) | |
c45a5a74 | 4515 | return r; |
6a0f1f6d | 4516 | |
a8548816 TG |
4517 | r = event_arm_timer(e, &e->boottime); |
4518 | if (r < 0) | |
c45a5a74 | 4519 | return r; |
a8548816 | 4520 | |
6a0f1f6d LP |
4521 | r = event_arm_timer(e, &e->monotonic); |
4522 | if (r < 0) | |
c45a5a74 | 4523 | return r; |
6a0f1f6d LP |
4524 | |
4525 | r = event_arm_timer(e, &e->realtime_alarm); | |
1b5995b0 | 4526 | if (r < 0) |
c45a5a74 | 4527 | return r; |
fd38203a | 4528 | |
6a0f1f6d | 4529 | r = event_arm_timer(e, &e->boottime_alarm); |
1b5995b0 | 4530 | if (r < 0) |
c45a5a74 | 4531 | return r; |
fd38203a | 4532 | |
97ef5391 LP |
4533 | event_close_inode_data_fds(e); |
4534 | ||
0601b958 | 4535 | if (event_next_pending(e) || e->need_process_child || e->buffered_inotify_data_list) |
c45a5a74 TG |
4536 | goto pending; |
4537 | ||
2b0c9ef7 | 4538 | e->state = SD_EVENT_ARMED; |
c45a5a74 TG |
4539 | |
4540 | return 0; | |
4541 | ||
4542 | pending: | |
2b0c9ef7 | 4543 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
4544 | r = sd_event_wait(e, 0); |
4545 | if (r == 0) | |
2b0c9ef7 | 4546 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
4547 | |
4548 | return r; | |
c45a5a74 TG |
4549 | } |
4550 | ||
798445ab LP |
4551 | static int epoll_wait_usec( |
4552 | int fd, | |
4553 | struct epoll_event *events, | |
4554 | int maxevents, | |
4555 | usec_t timeout) { | |
4556 | ||
7c248223 | 4557 | int msec; |
0c14c45e LP |
4558 | /* A wrapper that uses epoll_pwait2() if available, and falls back to epoll_wait() if not. */ |
4559 | ||
4560 | #if HAVE_EPOLL_PWAIT2 | |
39f756d3 | 4561 | static bool epoll_pwait2_absent = false; |
52bb308c | 4562 | int r; |
798445ab | 4563 | |
0c14c45e LP |
4564 | /* epoll_pwait2() was added to Linux 5.11 (2021-02-14) and to glibc in 2.35 (2022-02-03). In contrast |
4565 | * to other syscalls we don't bother with our own fallback syscall wrappers on old libcs, since this | |
4566 | * is not that obvious to implement given the libc and kernel definitions differ in the last | |
4567 | * argument. Moreover, the only reason to use it is the more accurate time-outs (which is not a | |
4568 | * biggie), let's hence rely on glibc's definitions, and fallback to epoll_pwait() when that's | |
4569 | * missing. */ | |
798445ab LP |
4570 | |
4571 | if (!epoll_pwait2_absent && timeout != USEC_INFINITY) { | |
798445ab LP |
4572 | r = epoll_pwait2(fd, |
4573 | events, | |
4574 | maxevents, | |
52bb308c | 4575 | TIMESPEC_STORE(timeout), |
798445ab LP |
4576 | NULL); |
4577 | if (r >= 0) | |
4578 | return r; | |
7cb45dbf | 4579 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) |
798445ab LP |
4580 | return -errno; /* Only fallback to old epoll_wait() if the syscall is masked or not |
4581 | * supported. */ | |
4582 | ||
4583 | epoll_pwait2_absent = true; | |
4584 | } | |
39f756d3 | 4585 | #endif |
798445ab LP |
4586 | |
4587 | if (timeout == USEC_INFINITY) | |
4588 | msec = -1; | |
4589 | else { | |
4590 | usec_t k; | |
4591 | ||
4592 | k = DIV_ROUND_UP(timeout, USEC_PER_MSEC); | |
4593 | if (k >= INT_MAX) | |
4594 | msec = INT_MAX; /* Saturate */ | |
4595 | else | |
4596 | msec = (int) k; | |
4597 | } | |
4598 | ||
7c248223 | 4599 | return RET_NERRNO(epoll_wait(fd, events, maxevents, msec)); |
798445ab LP |
4600 | } |
4601 | ||
efd3be9d | 4602 | static int process_epoll(sd_event *e, usec_t timeout, int64_t threshold, int64_t *ret_min_priority) { |
319a4f4b | 4603 | size_t n_event_queue, m, n_event_max; |
efd3be9d YW |
4604 | int64_t min_priority = threshold; |
4605 | bool something_new = false; | |
798445ab | 4606 | int r; |
c45a5a74 | 4607 | |
efd3be9d YW |
4608 | assert(e); |
4609 | assert(ret_min_priority); | |
6a0f1f6d | 4610 | |
8b9708d1 | 4611 | n_event_queue = MAX(e->n_sources, 1u); |
319a4f4b | 4612 | if (!GREEDY_REALLOC(e->event_queue, n_event_queue)) |
5cddd924 | 4613 | return -ENOMEM; |
fd38203a | 4614 | |
319a4f4b LP |
4615 | n_event_max = MALLOC_ELEMENTSOF(e->event_queue); |
4616 | ||
97ef5391 | 4617 | /* If we still have inotify data buffered, then query the other fds, but don't wait on it */ |
0601b958 | 4618 | if (e->buffered_inotify_data_list) |
798445ab | 4619 | timeout = 0; |
97ef5391 | 4620 | |
8b9708d1 | 4621 | for (;;) { |
319a4f4b LP |
4622 | r = epoll_wait_usec( |
4623 | e->epoll_fd, | |
4624 | e->event_queue, | |
4625 | n_event_max, | |
4626 | timeout); | |
798445ab | 4627 | if (r < 0) |
efd3be9d | 4628 | return r; |
c45a5a74 | 4629 | |
8b9708d1 YW |
4630 | m = (size_t) r; |
4631 | ||
319a4f4b | 4632 | if (m < n_event_max) |
8b9708d1 YW |
4633 | break; |
4634 | ||
319a4f4b | 4635 | if (n_event_max >= n_event_queue * 10) |
8b9708d1 YW |
4636 | break; |
4637 | ||
319a4f4b | 4638 | if (!GREEDY_REALLOC(e->event_queue, n_event_max + n_event_queue)) |
8b9708d1 YW |
4639 | return -ENOMEM; |
4640 | ||
319a4f4b | 4641 | n_event_max = MALLOC_ELEMENTSOF(e->event_queue); |
798445ab | 4642 | timeout = 0; |
da7e457c | 4643 | } |
fd38203a | 4644 | |
efd3be9d YW |
4645 | /* Set timestamp only when this is called first time. */ |
4646 | if (threshold == INT64_MAX) | |
fa5a0251 | 4647 | triple_timestamp_now(&e->timestamp); |
fd38203a | 4648 | |
8b9708d1 | 4649 | for (size_t i = 0; i < m; i++) { |
fd38203a | 4650 | |
5cddd924 LP |
4651 | if (e->event_queue[i].data.ptr == INT_TO_PTR(SOURCE_WATCHDOG)) |
4652 | r = flush_timer(e, e->watchdog_fd, e->event_queue[i].events, NULL); | |
9da4cb2b | 4653 | else { |
5cddd924 | 4654 | WakeupType *t = e->event_queue[i].data.ptr; |
9da4cb2b LP |
4655 | |
4656 | switch (*t) { | |
4657 | ||
f8f3f926 | 4658 | case WAKEUP_EVENT_SOURCE: { |
5cddd924 | 4659 | sd_event_source *s = e->event_queue[i].data.ptr; |
f8f3f926 LP |
4660 | |
4661 | assert(s); | |
4662 | ||
efd3be9d YW |
4663 | if (s->priority > threshold) |
4664 | continue; | |
4665 | ||
4666 | min_priority = MIN(min_priority, s->priority); | |
4667 | ||
f8f3f926 LP |
4668 | switch (s->type) { |
4669 | ||
4670 | case SOURCE_IO: | |
5cddd924 | 4671 | r = process_io(e, s, e->event_queue[i].events); |
f8f3f926 LP |
4672 | break; |
4673 | ||
4674 | case SOURCE_CHILD: | |
5cddd924 | 4675 | r = process_pidfd(e, s, e->event_queue[i].events); |
f8f3f926 LP |
4676 | break; |
4677 | ||
158fe190 LP |
4678 | case SOURCE_MEMORY_PRESSURE: |
4679 | r = process_memory_pressure(s, e->event_queue[i].events); | |
4680 | break; | |
4681 | ||
f8f3f926 | 4682 | default: |
04499a70 | 4683 | assert_not_reached(); |
f8f3f926 LP |
4684 | } |
4685 | ||
9da4cb2b | 4686 | break; |
f8f3f926 | 4687 | } |
fd38203a | 4688 | |
9da4cb2b | 4689 | case WAKEUP_CLOCK_DATA: { |
5cddd924 | 4690 | struct clock_data *d = e->event_queue[i].data.ptr; |
f8f3f926 LP |
4691 | |
4692 | assert(d); | |
4693 | ||
5cddd924 | 4694 | r = flush_timer(e, d->fd, e->event_queue[i].events, &d->next); |
9da4cb2b LP |
4695 | break; |
4696 | } | |
4697 | ||
4698 | case WAKEUP_SIGNAL_DATA: | |
efd3be9d | 4699 | r = process_signal(e, e->event_queue[i].data.ptr, e->event_queue[i].events, &min_priority); |
9da4cb2b LP |
4700 | break; |
4701 | ||
97ef5391 | 4702 | case WAKEUP_INOTIFY_DATA: |
efd3be9d | 4703 | r = event_inotify_data_read(e, e->event_queue[i].data.ptr, e->event_queue[i].events, threshold); |
97ef5391 LP |
4704 | break; |
4705 | ||
9da4cb2b | 4706 | default: |
04499a70 | 4707 | assert_not_reached(); |
9da4cb2b LP |
4708 | } |
4709 | } | |
efd3be9d YW |
4710 | if (r < 0) |
4711 | return r; | |
4712 | if (r > 0) | |
4713 | something_new = true; | |
4714 | } | |
4715 | ||
4716 | *ret_min_priority = min_priority; | |
4717 | return something_new; | |
4718 | } | |
4719 | ||
4720 | _public_ int sd_event_wait(sd_event *e, uint64_t timeout) { | |
4721 | int r; | |
4722 | ||
4723 | assert_return(e, -EINVAL); | |
4724 | assert_return(e = event_resolve(e), -ENOPKG); | |
2eeff0f4 | 4725 | assert_return(!event_origin_changed(e), -ECHILD); |
efd3be9d YW |
4726 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
4727 | assert_return(e->state == SD_EVENT_ARMED, -EBUSY); | |
4728 | ||
4729 | if (e->exit_requested) { | |
4730 | e->state = SD_EVENT_PENDING; | |
4731 | return 1; | |
4732 | } | |
4733 | ||
4734 | for (int64_t threshold = INT64_MAX; ; threshold--) { | |
4735 | int64_t epoll_min_priority, child_min_priority; | |
4736 | ||
4737 | /* There may be a possibility that new epoll (especially IO) and child events are | |
4738 | * triggered just after process_epoll() call but before process_child(), and the new IO | |
4739 | * events may have higher priority than the child events. To salvage these events, | |
4740 | * let's call epoll_wait() again, but accepts only events with higher priority than the | |
4741 | * previous. See issue https://github.com/systemd/systemd/issues/18190 and comments | |
4742 | * https://github.com/systemd/systemd/pull/18750#issuecomment-785801085 | |
4743 | * https://github.com/systemd/systemd/pull/18922#issuecomment-792825226 */ | |
4744 | ||
4745 | r = process_epoll(e, timeout, threshold, &epoll_min_priority); | |
4746 | if (r == -EINTR) { | |
4747 | e->state = SD_EVENT_PENDING; | |
4748 | return 1; | |
4749 | } | |
4750 | if (r < 0) | |
4751 | goto finish; | |
4752 | if (r == 0 && threshold < INT64_MAX) | |
4753 | /* No new epoll event. */ | |
4754 | break; | |
4755 | ||
4756 | r = process_child(e, threshold, &child_min_priority); | |
fd38203a | 4757 | if (r < 0) |
da7e457c | 4758 | goto finish; |
efd3be9d YW |
4759 | if (r == 0) |
4760 | /* No new child event. */ | |
4761 | break; | |
4762 | ||
4763 | threshold = MIN(epoll_min_priority, child_min_priority); | |
4764 | if (threshold == INT64_MIN) | |
4765 | break; | |
4766 | ||
4767 | timeout = 0; | |
fd38203a LP |
4768 | } |
4769 | ||
cde93897 LP |
4770 | r = process_watchdog(e); |
4771 | if (r < 0) | |
4772 | goto finish; | |
4773 | ||
fd69f224 | 4774 | r = process_inotify(e); |
6a0f1f6d LP |
4775 | if (r < 0) |
4776 | goto finish; | |
4777 | ||
fd69f224 | 4778 | r = process_timer(e, e->timestamp.realtime, &e->realtime); |
a8548816 TG |
4779 | if (r < 0) |
4780 | goto finish; | |
4781 | ||
fd69f224 | 4782 | r = process_timer(e, e->timestamp.boottime, &e->boottime); |
6a0f1f6d LP |
4783 | if (r < 0) |
4784 | goto finish; | |
4785 | ||
4786 | r = process_timer(e, e->timestamp.realtime, &e->realtime_alarm); | |
fd38203a | 4787 | if (r < 0) |
da7e457c | 4788 | goto finish; |
fd38203a | 4789 | |
e475d10c | 4790 | r = process_timer(e, e->timestamp.boottime, &e->boottime_alarm); |
fd38203a | 4791 | if (r < 0) |
da7e457c | 4792 | goto finish; |
fd38203a | 4793 | |
fd69f224 | 4794 | r = process_timer(e, e->timestamp.monotonic, &e->monotonic); |
97ef5391 LP |
4795 | if (r < 0) |
4796 | goto finish; | |
fd69f224 MS |
4797 | else if (r == 1) { |
4798 | /* Ratelimit expiry callback was called. Let's postpone processing pending sources and | |
4799 | * put loop in the initial state in order to evaluate (in the next iteration) also sources | |
4800 | * there were potentially re-enabled by the callback. | |
4801 | * | |
4802 | * Wondering why we treat only this invocation of process_timer() differently? Once event | |
4803 | * source is ratelimited we essentially transform it into CLOCK_MONOTONIC timer hence | |
4804 | * ratelimit expiry callback is never called for any other timer type. */ | |
4805 | r = 0; | |
4806 | goto finish; | |
4807 | } | |
97ef5391 | 4808 | |
c45a5a74 TG |
4809 | if (event_next_pending(e)) { |
4810 | e->state = SD_EVENT_PENDING; | |
c45a5a74 | 4811 | return 1; |
da7e457c LP |
4812 | } |
4813 | ||
c45a5a74 | 4814 | r = 0; |
fd38203a | 4815 | |
da7e457c | 4816 | finish: |
2b0c9ef7 | 4817 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
4818 | |
4819 | return r; | |
fd38203a LP |
4820 | } |
4821 | ||
c45a5a74 TG |
4822 | _public_ int sd_event_dispatch(sd_event *e) { |
4823 | sd_event_source *p; | |
4824 | int r; | |
4825 | ||
4826 | assert_return(e, -EINVAL); | |
b937d761 | 4827 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 4828 | assert_return(!event_origin_changed(e), -ECHILD); |
c45a5a74 TG |
4829 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
4830 | assert_return(e->state == SD_EVENT_PENDING, -EBUSY); | |
4831 | ||
4832 | if (e->exit_requested) | |
4833 | return dispatch_exit(e); | |
4834 | ||
4835 | p = event_next_pending(e); | |
4836 | if (p) { | |
c8e9d15c | 4837 | PROTECT_EVENT(e); |
c45a5a74 TG |
4838 | |
4839 | e->state = SD_EVENT_RUNNING; | |
4840 | r = source_dispatch(p); | |
2b0c9ef7 | 4841 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
4842 | return r; |
4843 | } | |
4844 | ||
2b0c9ef7 | 4845 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
4846 | |
4847 | return 1; | |
4848 | } | |
4849 | ||
34b87517 | 4850 | static void event_log_delays(sd_event *e) { |
442ac269 YW |
4851 | char b[ELEMENTSOF(e->delays) * DECIMAL_STR_MAX(unsigned) + 1], *p; |
4852 | size_t l, i; | |
34b87517 | 4853 | |
442ac269 YW |
4854 | p = b; |
4855 | l = sizeof(b); | |
4856 | for (i = 0; i < ELEMENTSOF(e->delays); i++) { | |
4857 | l = strpcpyf(&p, l, "%u ", e->delays[i]); | |
34b87517 VC |
4858 | e->delays[i] = 0; |
4859 | } | |
442ac269 | 4860 | log_debug("Event loop iterations: %s", b); |
34b87517 VC |
4861 | } |
4862 | ||
c45a5a74 TG |
4863 | _public_ int sd_event_run(sd_event *e, uint64_t timeout) { |
4864 | int r; | |
4865 | ||
4866 | assert_return(e, -EINVAL); | |
b937d761 | 4867 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 4868 | assert_return(!event_origin_changed(e), -ECHILD); |
c45a5a74 | 4869 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2b0c9ef7 | 4870 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
c45a5a74 | 4871 | |
e6a7bee5 | 4872 | if (e->profile_delays && e->last_run_usec != 0) { |
34b87517 VC |
4873 | usec_t this_run; |
4874 | unsigned l; | |
4875 | ||
4876 | this_run = now(CLOCK_MONOTONIC); | |
4877 | ||
58c34be8 | 4878 | l = log2u64(this_run - e->last_run_usec); |
cb9d621e | 4879 | assert(l < ELEMENTSOF(e->delays)); |
34b87517 VC |
4880 | e->delays[l]++; |
4881 | ||
e6a7bee5 | 4882 | if (this_run - e->last_log_usec >= 5*USEC_PER_SEC) { |
34b87517 | 4883 | event_log_delays(e); |
e6a7bee5 | 4884 | e->last_log_usec = this_run; |
34b87517 VC |
4885 | } |
4886 | } | |
4887 | ||
f814c871 | 4888 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
c8e9d15c | 4889 | PROTECT_EVENT(e); |
f814c871 | 4890 | |
c45a5a74 | 4891 | r = sd_event_prepare(e); |
53bac4e0 LP |
4892 | if (r == 0) |
4893 | /* There was nothing? Then wait... */ | |
4894 | r = sd_event_wait(e, timeout); | |
c45a5a74 | 4895 | |
34b87517 | 4896 | if (e->profile_delays) |
e6a7bee5 | 4897 | e->last_run_usec = now(CLOCK_MONOTONIC); |
34b87517 | 4898 | |
02d30981 | 4899 | if (r > 0) { |
53bac4e0 | 4900 | /* There's something now, then let's dispatch it */ |
02d30981 TG |
4901 | r = sd_event_dispatch(e); |
4902 | if (r < 0) | |
4903 | return r; | |
53bac4e0 LP |
4904 | |
4905 | return 1; | |
4906 | } | |
4907 | ||
4908 | return r; | |
c45a5a74 TG |
4909 | } |
4910 | ||
f7262a9f | 4911 | _public_ int sd_event_loop(sd_event *e) { |
fd38203a LP |
4912 | int r; |
4913 | ||
da7e457c | 4914 | assert_return(e, -EINVAL); |
b937d761 | 4915 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 4916 | assert_return(!event_origin_changed(e), -ECHILD); |
2b0c9ef7 | 4917 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 4918 | |
2eeff0f4 | 4919 | |
c8e9d15c | 4920 | PROTECT_EVENT(e); |
fd38203a | 4921 | |
da7e457c | 4922 | while (e->state != SD_EVENT_FINISHED) { |
f5fbe71d | 4923 | r = sd_event_run(e, UINT64_MAX); |
fd38203a | 4924 | if (r < 0) |
30dd293c | 4925 | return r; |
fd38203a LP |
4926 | } |
4927 | ||
30dd293c | 4928 | return e->exit_code; |
fd38203a LP |
4929 | } |
4930 | ||
9b364545 | 4931 | _public_ int sd_event_get_fd(sd_event *e) { |
9b364545 | 4932 | assert_return(e, -EINVAL); |
b937d761 | 4933 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 4934 | assert_return(!event_origin_changed(e), -ECHILD); |
9b364545 TG |
4935 | |
4936 | return e->epoll_fd; | |
4937 | } | |
4938 | ||
f7262a9f | 4939 | _public_ int sd_event_get_state(sd_event *e) { |
da7e457c | 4940 | assert_return(e, -EINVAL); |
b937d761 | 4941 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 4942 | assert_return(!event_origin_changed(e), -ECHILD); |
da7e457c LP |
4943 | |
4944 | return e->state; | |
4945 | } | |
4946 | ||
6203e07a | 4947 | _public_ int sd_event_get_exit_code(sd_event *e, int *code) { |
da7e457c | 4948 | assert_return(e, -EINVAL); |
b937d761 | 4949 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 4950 | assert_return(!event_origin_changed(e), -ECHILD); |
fd38203a | 4951 | |
6203e07a LP |
4952 | if (!e->exit_requested) |
4953 | return -ENODATA; | |
4954 | ||
7253eaa0 LP |
4955 | if (code) |
4956 | *code = e->exit_code; | |
6203e07a | 4957 | return 0; |
fd38203a LP |
4958 | } |
4959 | ||
6203e07a | 4960 | _public_ int sd_event_exit(sd_event *e, int code) { |
da7e457c | 4961 | assert_return(e, -EINVAL); |
b937d761 | 4962 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 4963 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2eeff0f4 | 4964 | assert_return(!event_origin_changed(e), -ECHILD); |
fd38203a | 4965 | |
6203e07a LP |
4966 | e->exit_requested = true; |
4967 | e->exit_code = code; | |
4968 | ||
fd38203a LP |
4969 | return 0; |
4970 | } | |
46e8c825 | 4971 | |
6a0f1f6d | 4972 | _public_ int sd_event_now(sd_event *e, clockid_t clock, uint64_t *usec) { |
46e8c825 | 4973 | assert_return(e, -EINVAL); |
b937d761 | 4974 | assert_return(e = event_resolve(e), -ENOPKG); |
46e8c825 | 4975 | assert_return(usec, -EINVAL); |
2eeff0f4 | 4976 | assert_return(!event_origin_changed(e), -ECHILD); |
46e8c825 | 4977 | |
e475d10c LP |
4978 | if (!TRIPLE_TIMESTAMP_HAS_CLOCK(clock)) |
4979 | return -EOPNOTSUPP; | |
4980 | ||
e475d10c | 4981 | if (!triple_timestamp_is_set(&e->timestamp)) { |
15c689d7 | 4982 | /* Implicitly fall back to now() if we never ran before and thus have no cached time. */ |
38a03f06 LP |
4983 | *usec = now(clock); |
4984 | return 1; | |
4985 | } | |
46e8c825 | 4986 | |
e475d10c | 4987 | *usec = triple_timestamp_by_clock(&e->timestamp, clock); |
46e8c825 LP |
4988 | return 0; |
4989 | } | |
afc6adb5 LP |
4990 | |
4991 | _public_ int sd_event_default(sd_event **ret) { | |
39883f62 | 4992 | sd_event *e = NULL; |
afc6adb5 LP |
4993 | int r; |
4994 | ||
4995 | if (!ret) | |
4996 | return !!default_event; | |
4997 | ||
4998 | if (default_event) { | |
4999 | *ret = sd_event_ref(default_event); | |
5000 | return 0; | |
5001 | } | |
5002 | ||
5003 | r = sd_event_new(&e); | |
5004 | if (r < 0) | |
5005 | return r; | |
5006 | ||
5007 | e->default_event_ptr = &default_event; | |
5008 | e->tid = gettid(); | |
5009 | default_event = e; | |
5010 | ||
5011 | *ret = e; | |
5012 | return 1; | |
5013 | } | |
5014 | ||
5015 | _public_ int sd_event_get_tid(sd_event *e, pid_t *tid) { | |
5016 | assert_return(e, -EINVAL); | |
b937d761 | 5017 | assert_return(e = event_resolve(e), -ENOPKG); |
afc6adb5 | 5018 | assert_return(tid, -EINVAL); |
2eeff0f4 | 5019 | assert_return(!event_origin_changed(e), -ECHILD); |
afc6adb5 | 5020 | |
76b54375 LP |
5021 | if (e->tid != 0) { |
5022 | *tid = e->tid; | |
5023 | return 0; | |
5024 | } | |
5025 | ||
5026 | return -ENXIO; | |
afc6adb5 | 5027 | } |
cde93897 LP |
5028 | |
5029 | _public_ int sd_event_set_watchdog(sd_event *e, int b) { | |
5030 | int r; | |
5031 | ||
5032 | assert_return(e, -EINVAL); | |
b937d761 | 5033 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 5034 | assert_return(!event_origin_changed(e), -ECHILD); |
cde93897 LP |
5035 | |
5036 | if (e->watchdog == !!b) | |
5037 | return e->watchdog; | |
5038 | ||
5039 | if (b) { | |
09812eb7 LP |
5040 | r = sd_watchdog_enabled(false, &e->watchdog_period); |
5041 | if (r <= 0) | |
cde93897 | 5042 | return r; |
cde93897 LP |
5043 | |
5044 | /* Issue first ping immediately */ | |
5045 | sd_notify(false, "WATCHDOG=1"); | |
5046 | e->watchdog_last = now(CLOCK_MONOTONIC); | |
5047 | ||
5048 | e->watchdog_fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC); | |
5049 | if (e->watchdog_fd < 0) | |
5050 | return -errno; | |
5051 | ||
5052 | r = arm_watchdog(e); | |
5053 | if (r < 0) | |
5054 | goto fail; | |
5055 | ||
1eac7948 | 5056 | struct epoll_event ev = { |
a82f89aa LP |
5057 | .events = EPOLLIN, |
5058 | .data.ptr = INT_TO_PTR(SOURCE_WATCHDOG), | |
5059 | }; | |
cde93897 | 5060 | |
15c689d7 | 5061 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, e->watchdog_fd, &ev) < 0) { |
cde93897 LP |
5062 | r = -errno; |
5063 | goto fail; | |
5064 | } | |
5065 | ||
5066 | } else { | |
5067 | if (e->watchdog_fd >= 0) { | |
5a795bff | 5068 | (void) epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, e->watchdog_fd, NULL); |
03e334a1 | 5069 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
5070 | } |
5071 | } | |
5072 | ||
0a6a5965 | 5073 | e->watchdog = b; |
cde93897 LP |
5074 | return e->watchdog; |
5075 | ||
5076 | fail: | |
03e334a1 | 5077 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
5078 | return r; |
5079 | } | |
8f726607 LP |
5080 | |
5081 | _public_ int sd_event_get_watchdog(sd_event *e) { | |
5082 | assert_return(e, -EINVAL); | |
b937d761 | 5083 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 5084 | assert_return(!event_origin_changed(e), -ECHILD); |
8f726607 LP |
5085 | |
5086 | return e->watchdog; | |
5087 | } | |
60a3b1e1 LP |
5088 | |
5089 | _public_ int sd_event_get_iteration(sd_event *e, uint64_t *ret) { | |
5090 | assert_return(e, -EINVAL); | |
b937d761 | 5091 | assert_return(e = event_resolve(e), -ENOPKG); |
2eeff0f4 | 5092 | assert_return(!event_origin_changed(e), -ECHILD); |
60a3b1e1 LP |
5093 | |
5094 | *ret = e->iteration; | |
5095 | return 0; | |
5096 | } | |
15723a1d LP |
5097 | |
5098 | _public_ int sd_event_source_set_destroy_callback(sd_event_source *s, sd_event_destroy_t callback) { | |
5099 | assert_return(s, -EINVAL); | |
2eeff0f4 LB |
5100 | assert_return(s->event, -EINVAL); |
5101 | assert_return(!event_origin_changed(s->event), -ECHILD); | |
15723a1d LP |
5102 | |
5103 | s->destroy_callback = callback; | |
5104 | return 0; | |
5105 | } | |
5106 | ||
5107 | _public_ int sd_event_source_get_destroy_callback(sd_event_source *s, sd_event_destroy_t *ret) { | |
5108 | assert_return(s, -EINVAL); | |
2eeff0f4 | 5109 | assert_return(!event_origin_changed(s->event), -ECHILD); |
15723a1d LP |
5110 | |
5111 | if (ret) | |
5112 | *ret = s->destroy_callback; | |
5113 | ||
5114 | return !!s->destroy_callback; | |
5115 | } | |
2382c936 YW |
5116 | |
5117 | _public_ int sd_event_source_get_floating(sd_event_source *s) { | |
5118 | assert_return(s, -EINVAL); | |
2eeff0f4 | 5119 | assert_return(!event_origin_changed(s->event), -ECHILD); |
2382c936 YW |
5120 | |
5121 | return s->floating; | |
5122 | } | |
5123 | ||
5124 | _public_ int sd_event_source_set_floating(sd_event_source *s, int b) { | |
5125 | assert_return(s, -EINVAL); | |
2eeff0f4 | 5126 | assert_return(!event_origin_changed(s->event), -ECHILD); |
2382c936 YW |
5127 | |
5128 | if (s->floating == !!b) | |
5129 | return 0; | |
5130 | ||
5131 | if (!s->event) /* Already disconnected */ | |
5132 | return -ESTALE; | |
5133 | ||
5134 | s->floating = b; | |
5135 | ||
5136 | if (b) { | |
5137 | sd_event_source_ref(s); | |
5138 | sd_event_unref(s->event); | |
5139 | } else { | |
5140 | sd_event_ref(s->event); | |
5141 | sd_event_source_unref(s); | |
5142 | } | |
5143 | ||
5144 | return 1; | |
5145 | } | |
b778cba4 LP |
5146 | |
5147 | _public_ int sd_event_source_get_exit_on_failure(sd_event_source *s) { | |
5148 | assert_return(s, -EINVAL); | |
5149 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
2eeff0f4 | 5150 | assert_return(!event_origin_changed(s->event), -ECHILD); |
b778cba4 LP |
5151 | |
5152 | return s->exit_on_failure; | |
5153 | } | |
5154 | ||
5155 | _public_ int sd_event_source_set_exit_on_failure(sd_event_source *s, int b) { | |
5156 | assert_return(s, -EINVAL); | |
5157 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
2eeff0f4 | 5158 | assert_return(!event_origin_changed(s->event), -ECHILD); |
b778cba4 LP |
5159 | |
5160 | if (s->exit_on_failure == !!b) | |
5161 | return 0; | |
5162 | ||
5163 | s->exit_on_failure = b; | |
5164 | return 1; | |
5165 | } | |
b6d5481b LP |
5166 | |
5167 | _public_ int sd_event_source_set_ratelimit(sd_event_source *s, uint64_t interval, unsigned burst) { | |
5168 | int r; | |
5169 | ||
5170 | assert_return(s, -EINVAL); | |
2eeff0f4 | 5171 | assert_return(!event_origin_changed(s->event), -ECHILD); |
b6d5481b LP |
5172 | |
5173 | /* Turning on ratelimiting on event source types that don't support it, is a loggable offense. Doing | |
5174 | * so is a programming error. */ | |
5175 | assert_return(EVENT_SOURCE_CAN_RATE_LIMIT(s->type), -EDOM); | |
5176 | ||
5177 | /* When ratelimiting is configured we'll always reset the rate limit state first and start fresh, | |
5178 | * non-ratelimited. */ | |
fd69f224 | 5179 | r = event_source_leave_ratelimit(s, /* run_callback */ false); |
b6d5481b LP |
5180 | if (r < 0) |
5181 | return r; | |
5182 | ||
5183 | s->rate_limit = (RateLimit) { interval, burst }; | |
5184 | return 0; | |
fd69f224 MS |
5185 | } |
5186 | ||
5187 | _public_ int sd_event_source_set_ratelimit_expire_callback(sd_event_source *s, sd_event_handler_t callback) { | |
5188 | assert_return(s, -EINVAL); | |
2eeff0f4 | 5189 | assert_return(!event_origin_changed(s->event), -ECHILD); |
fd69f224 MS |
5190 | |
5191 | s->ratelimit_expire_callback = callback; | |
5192 | return 0; | |
b6d5481b LP |
5193 | } |
5194 | ||
5195 | _public_ int sd_event_source_get_ratelimit(sd_event_source *s, uint64_t *ret_interval, unsigned *ret_burst) { | |
5196 | assert_return(s, -EINVAL); | |
2eeff0f4 | 5197 | assert_return(!event_origin_changed(s->event), -ECHILD); |
b6d5481b | 5198 | |
6dd3b818 YW |
5199 | /* Querying whether an event source has ratelimiting configured is not a loggable offense, hence |
5200 | * don't use assert_return(). Unlike turning on ratelimiting it's not really a programming error. */ | |
b6d5481b LP |
5201 | if (!EVENT_SOURCE_CAN_RATE_LIMIT(s->type)) |
5202 | return -EDOM; | |
5203 | ||
5204 | if (!ratelimit_configured(&s->rate_limit)) | |
5205 | return -ENOEXEC; | |
5206 | ||
5207 | if (ret_interval) | |
5208 | *ret_interval = s->rate_limit.interval; | |
5209 | if (ret_burst) | |
5210 | *ret_burst = s->rate_limit.burst; | |
5211 | ||
5212 | return 0; | |
5213 | } | |
5214 | ||
5215 | _public_ int sd_event_source_is_ratelimited(sd_event_source *s) { | |
5216 | assert_return(s, -EINVAL); | |
2eeff0f4 | 5217 | assert_return(!event_origin_changed(s->event), -ECHILD); |
b6d5481b LP |
5218 | |
5219 | if (!EVENT_SOURCE_CAN_RATE_LIMIT(s->type)) | |
5220 | return false; | |
5221 | ||
5222 | if (!ratelimit_configured(&s->rate_limit)) | |
5223 | return false; | |
5224 | ||
5225 | return s->ratelimited; | |
5226 | } | |
baf3fdec | 5227 | |
2fdc274c LP |
5228 | _public_ int sd_event_source_leave_ratelimit(sd_event_source *s) { |
5229 | int r; | |
5230 | ||
5231 | assert_return(s, -EINVAL); | |
5232 | ||
5233 | if (!EVENT_SOURCE_CAN_RATE_LIMIT(s->type)) | |
5234 | return 0; | |
5235 | ||
5236 | if (!ratelimit_configured(&s->rate_limit)) | |
5237 | return 0; | |
5238 | ||
5239 | if (!s->ratelimited) | |
5240 | return 0; | |
5241 | ||
5242 | r = event_source_leave_ratelimit(s, /* run_callback */ false); | |
5243 | if (r < 0) | |
5244 | return r; | |
5245 | ||
5246 | return 1; /* tell caller that we indeed just left the ratelimit state */ | |
5247 | } | |
5248 | ||
baf3fdec LP |
5249 | _public_ int sd_event_set_signal_exit(sd_event *e, int b) { |
5250 | bool change = false; | |
5251 | int r; | |
5252 | ||
5253 | assert_return(e, -EINVAL); | |
5254 | ||
5255 | if (b) { | |
5256 | /* We want to maintain pointers to these event sources, so that we can destroy them when told | |
5257 | * so. But we also don't want them to pin the event loop itself. Hence we mark them as | |
5258 | * floating after creation (and undo this before deleting them again). */ | |
5259 | ||
5260 | if (!e->sigint_event_source) { | |
5261 | r = sd_event_add_signal(e, &e->sigint_event_source, SIGINT | SD_EVENT_SIGNAL_PROCMASK, NULL, NULL); | |
5262 | if (r < 0) | |
5263 | return r; | |
5264 | ||
5265 | assert(sd_event_source_set_floating(e->sigint_event_source, true) >= 0); | |
5266 | change = true; | |
5267 | } | |
5268 | ||
5269 | if (!e->sigterm_event_source) { | |
5270 | r = sd_event_add_signal(e, &e->sigterm_event_source, SIGTERM | SD_EVENT_SIGNAL_PROCMASK, NULL, NULL); | |
5271 | if (r < 0) { | |
5272 | if (change) { | |
5273 | assert(sd_event_source_set_floating(e->sigint_event_source, false) >= 0); | |
5274 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
5275 | } | |
5276 | ||
5277 | return r; | |
5278 | } | |
5279 | ||
5280 | assert(sd_event_source_set_floating(e->sigterm_event_source, true) >= 0); | |
5281 | change = true; | |
5282 | } | |
5283 | ||
5284 | } else { | |
5285 | if (e->sigint_event_source) { | |
5286 | assert(sd_event_source_set_floating(e->sigint_event_source, false) >= 0); | |
5287 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
5288 | change = true; | |
5289 | } | |
5290 | ||
5291 | if (e->sigterm_event_source) { | |
5292 | assert(sd_event_source_set_floating(e->sigterm_event_source, false) >= 0); | |
5293 | e->sigterm_event_source = sd_event_source_unref(e->sigterm_event_source); | |
5294 | change = true; | |
5295 | } | |
5296 | } | |
5297 | ||
5298 | return change; | |
5299 | } | |
158fe190 LP |
5300 | |
5301 | _public_ int sd_event_source_set_memory_pressure_type(sd_event_source *s, const char *ty) { | |
5302 | _cleanup_free_ char *b = NULL; | |
5303 | _cleanup_free_ void *w = NULL; | |
5304 | ||
5305 | assert_return(s, -EINVAL); | |
5306 | assert_return(s->type == SOURCE_MEMORY_PRESSURE, -EDOM); | |
5307 | assert_return(ty, -EINVAL); | |
2eeff0f4 | 5308 | assert_return(!event_origin_changed(s->event), -ECHILD); |
158fe190 LP |
5309 | |
5310 | if (!STR_IN_SET(ty, "some", "full")) | |
5311 | return -EINVAL; | |
5312 | ||
5313 | if (s->memory_pressure.locked) /* Refuse adjusting parameters, if caller told us how to watch for events */ | |
5314 | return -EBUSY; | |
5315 | ||
5316 | char* space = memchr(s->memory_pressure.write_buffer, ' ', s->memory_pressure.write_buffer_size); | |
5317 | if (!space) | |
5318 | return -EINVAL; | |
5319 | ||
5320 | size_t l = (char*) space - (char*) s->memory_pressure.write_buffer; | |
5321 | b = memdup_suffix0(s->memory_pressure.write_buffer, l); | |
5322 | if (!b) | |
5323 | return -ENOMEM; | |
5324 | if (!STR_IN_SET(b, "some", "full")) | |
5325 | return -EINVAL; | |
5326 | ||
5327 | if (streq(b, ty)) | |
5328 | return 0; | |
5329 | ||
5330 | size_t nl = strlen(ty) + (s->memory_pressure.write_buffer_size - l); | |
5331 | w = new(char, nl); | |
5332 | if (!w) | |
5333 | return -ENOMEM; | |
5334 | ||
5335 | memcpy(stpcpy(w, ty), space, (s->memory_pressure.write_buffer_size - l)); | |
5336 | ||
5337 | free_and_replace(s->memory_pressure.write_buffer, w); | |
5338 | s->memory_pressure.write_buffer_size = nl; | |
5339 | s->memory_pressure.locked = false; | |
5340 | ||
5341 | return 1; | |
5342 | } | |
5343 | ||
5344 | _public_ int sd_event_source_set_memory_pressure_period(sd_event_source *s, uint64_t threshold_usec, uint64_t window_usec) { | |
5345 | _cleanup_free_ char *b = NULL; | |
5346 | _cleanup_free_ void *w = NULL; | |
5347 | ||
5348 | assert_return(s, -EINVAL); | |
5349 | assert_return(s->type == SOURCE_MEMORY_PRESSURE, -EDOM); | |
2eeff0f4 | 5350 | assert_return(!event_origin_changed(s->event), -ECHILD); |
158fe190 LP |
5351 | |
5352 | if (threshold_usec <= 0 || threshold_usec >= UINT64_MAX) | |
5353 | return -ERANGE; | |
5354 | if (window_usec <= 0 || window_usec >= UINT64_MAX) | |
5355 | return -ERANGE; | |
5356 | if (threshold_usec > window_usec) | |
5357 | return -EINVAL; | |
5358 | ||
5359 | if (s->memory_pressure.locked) /* Refuse adjusting parameters, if caller told us how to watch for events */ | |
5360 | return -EBUSY; | |
5361 | ||
5362 | char* space = memchr(s->memory_pressure.write_buffer, ' ', s->memory_pressure.write_buffer_size); | |
5363 | if (!space) | |
5364 | return -EINVAL; | |
5365 | ||
5366 | size_t l = (char*) space - (char*) s->memory_pressure.write_buffer; | |
5367 | b = memdup_suffix0(s->memory_pressure.write_buffer, l); | |
5368 | if (!b) | |
5369 | return -ENOMEM; | |
5370 | if (!STR_IN_SET(b, "some", "full")) | |
5371 | return -EINVAL; | |
5372 | ||
5373 | if (asprintf((char**) &w, | |
5374 | "%s " USEC_FMT " " USEC_FMT "", | |
5375 | b, | |
5376 | threshold_usec, | |
5377 | window_usec) < 0) | |
5378 | return -EINVAL; | |
5379 | ||
5380 | l = strlen(w) + 1; | |
5381 | if (memcmp_nn(s->memory_pressure.write_buffer, s->memory_pressure.write_buffer_size, w, l) == 0) | |
5382 | return 0; | |
5383 | ||
5384 | free_and_replace(s->memory_pressure.write_buffer, w); | |
5385 | s->memory_pressure.write_buffer_size = l; | |
5386 | s->memory_pressure.locked = false; | |
5387 | ||
5388 | return 1; | |
5389 | } |