]>
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" |
158fe190 | 27 | #include "path-util.h" |
07630cea | 28 | #include "prioq.h" |
4a0b58c4 | 29 | #include "process-util.h" |
158fe190 | 30 | #include "psi-util.h" |
6e9feda3 | 31 | #include "set.h" |
24882e06 | 32 | #include "signal-util.h" |
158fe190 LP |
33 | #include "socket-util.h" |
34 | #include "stat-util.h" | |
55cbfaa5 | 35 | #include "string-table.h" |
07630cea | 36 | #include "string-util.h" |
442ac269 | 37 | #include "strxcpyx.h" |
07630cea | 38 | #include "time-util.h" |
fd38203a | 39 | |
c2ba3ad6 | 40 | #define DEFAULT_ACCURACY_USEC (250 * USEC_PER_MSEC) |
fd38203a | 41 | |
f8f3f926 LP |
42 | static bool EVENT_SOURCE_WATCH_PIDFD(sd_event_source *s) { |
43 | /* Returns true if this is a PID event source and can be implemented by watching EPOLLIN */ | |
44 | return s && | |
45 | s->type == SOURCE_CHILD && | |
46 | s->child.pidfd >= 0 && | |
47 | s->child.options == WEXITED; | |
48 | } | |
49 | ||
b6d5481b LP |
50 | static bool event_source_is_online(sd_event_source *s) { |
51 | assert(s); | |
52 | return s->enabled != SD_EVENT_OFF && !s->ratelimited; | |
53 | } | |
54 | ||
55 | static bool event_source_is_offline(sd_event_source *s) { | |
56 | assert(s); | |
57 | return s->enabled == SD_EVENT_OFF || s->ratelimited; | |
58 | } | |
59 | ||
55cbfaa5 | 60 | static const char* const event_source_type_table[_SOURCE_EVENT_SOURCE_TYPE_MAX] = { |
3f3548f8 ZJS |
61 | [SOURCE_IO] = "io", |
62 | [SOURCE_TIME_REALTIME] = "realtime", | |
63 | [SOURCE_TIME_BOOTTIME] = "bootime", | |
64 | [SOURCE_TIME_MONOTONIC] = "monotonic", | |
55cbfaa5 DM |
65 | [SOURCE_TIME_REALTIME_ALARM] = "realtime-alarm", |
66 | [SOURCE_TIME_BOOTTIME_ALARM] = "boottime-alarm", | |
3f3548f8 ZJS |
67 | [SOURCE_SIGNAL] = "signal", |
68 | [SOURCE_CHILD] = "child", | |
69 | [SOURCE_DEFER] = "defer", | |
70 | [SOURCE_POST] = "post", | |
71 | [SOURCE_EXIT] = "exit", | |
72 | [SOURCE_WATCHDOG] = "watchdog", | |
73 | [SOURCE_INOTIFY] = "inotify", | |
158fe190 | 74 | [SOURCE_MEMORY_PRESSURE] = "memory-pressure", |
55cbfaa5 DM |
75 | }; |
76 | ||
77 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(event_source_type, int); | |
78 | ||
b6d5481b LP |
79 | #define EVENT_SOURCE_IS_TIME(t) \ |
80 | IN_SET((t), \ | |
81 | SOURCE_TIME_REALTIME, \ | |
82 | SOURCE_TIME_BOOTTIME, \ | |
83 | SOURCE_TIME_MONOTONIC, \ | |
84 | SOURCE_TIME_REALTIME_ALARM, \ | |
85 | SOURCE_TIME_BOOTTIME_ALARM) | |
86 | ||
87 | #define EVENT_SOURCE_CAN_RATE_LIMIT(t) \ | |
88 | IN_SET((t), \ | |
89 | SOURCE_IO, \ | |
90 | SOURCE_TIME_REALTIME, \ | |
91 | SOURCE_TIME_BOOTTIME, \ | |
92 | SOURCE_TIME_MONOTONIC, \ | |
93 | SOURCE_TIME_REALTIME_ALARM, \ | |
94 | SOURCE_TIME_BOOTTIME_ALARM, \ | |
95 | SOURCE_SIGNAL, \ | |
96 | SOURCE_DEFER, \ | |
158fe190 LP |
97 | SOURCE_INOTIFY, \ |
98 | SOURCE_MEMORY_PRESSURE) | |
6a0f1f6d | 99 | |
19947509 ZJS |
100 | /* This is used to assert that we didn't pass an unexpected source type to event_source_time_prioq_put(). |
101 | * Time sources and ratelimited sources can be passed, so effectively this is the same as the | |
102 | * EVENT_SOURCE_CAN_RATE_LIMIT() macro. */ | |
103 | #define EVENT_SOURCE_USES_TIME_PRIOQ(t) EVENT_SOURCE_CAN_RATE_LIMIT(t) | |
104 | ||
fd38203a | 105 | struct sd_event { |
da7e457c | 106 | unsigned n_ref; |
fd38203a LP |
107 | |
108 | int epoll_fd; | |
cde93897 | 109 | int watchdog_fd; |
fd38203a LP |
110 | |
111 | Prioq *pending; | |
112 | Prioq *prepare; | |
c2ba3ad6 | 113 | |
a8548816 | 114 | /* timerfd_create() only supports these five clocks so far. We |
6a0f1f6d LP |
115 | * can add support for more clocks when the kernel learns to |
116 | * deal with them, too. */ | |
117 | struct clock_data realtime; | |
a8548816 | 118 | struct clock_data boottime; |
6a0f1f6d LP |
119 | struct clock_data monotonic; |
120 | struct clock_data realtime_alarm; | |
121 | struct clock_data boottime_alarm; | |
fd38203a | 122 | |
da7e457c LP |
123 | usec_t perturb; |
124 | ||
9da4cb2b LP |
125 | sd_event_source **signal_sources; /* indexed by signal number */ |
126 | Hashmap *signal_data; /* indexed by priority */ | |
fd38203a LP |
127 | |
128 | Hashmap *child_sources; | |
b6d5481b | 129 | unsigned n_online_child_sources; |
fd38203a | 130 | |
6e9feda3 LP |
131 | Set *post_sources; |
132 | ||
6203e07a | 133 | Prioq *exit; |
fd38203a | 134 | |
97ef5391 LP |
135 | Hashmap *inotify_data; /* indexed by priority */ |
136 | ||
137 | /* A list of inode structures that still have an fd open, that we need to close before the next loop iteration */ | |
ed828563 | 138 | LIST_HEAD(struct inode_data, inode_data_to_close_list); |
97ef5391 LP |
139 | |
140 | /* A list of inotify objects that already have events buffered which aren't processed yet */ | |
0601b958 | 141 | LIST_HEAD(struct inotify_data, buffered_inotify_data_list); |
97ef5391 | 142 | |
158fe190 LP |
143 | /* A list of memory pressure event sources that still need their subscription string written */ |
144 | LIST_HEAD(sd_event_source, memory_pressure_write_list); | |
145 | ||
da7e457c | 146 | pid_t original_pid; |
c2ba3ad6 | 147 | |
60a3b1e1 | 148 | uint64_t iteration; |
e475d10c | 149 | triple_timestamp timestamp; |
da7e457c | 150 | int state; |
eaa3cbef | 151 | |
6203e07a | 152 | bool exit_requested:1; |
da7e457c | 153 | bool need_process_child:1; |
cde93897 | 154 | bool watchdog:1; |
34b87517 | 155 | bool profile_delays:1; |
afc6adb5 | 156 | |
6203e07a LP |
157 | int exit_code; |
158 | ||
afc6adb5 LP |
159 | pid_t tid; |
160 | sd_event **default_event_ptr; | |
cde93897 LP |
161 | |
162 | usec_t watchdog_last, watchdog_period; | |
15b38f93 LP |
163 | |
164 | unsigned n_sources; | |
a71fe8b8 | 165 | |
5cddd924 | 166 | struct epoll_event *event_queue; |
5cddd924 | 167 | |
a71fe8b8 | 168 | LIST_HEAD(sd_event_source, sources); |
34b87517 | 169 | |
baf3fdec LP |
170 | sd_event_source *sigint_event_source, *sigterm_event_source; |
171 | ||
e6a7bee5 | 172 | usec_t last_run_usec, last_log_usec; |
34b87517 | 173 | unsigned delays[sizeof(usec_t) * 8]; |
fd38203a LP |
174 | }; |
175 | ||
b937d761 NM |
176 | static thread_local sd_event *default_event = NULL; |
177 | ||
a71fe8b8 | 178 | static void source_disconnect(sd_event_source *s); |
97ef5391 | 179 | static void event_gc_inode_data(sd_event *e, struct inode_data *d); |
a71fe8b8 | 180 | |
b937d761 NM |
181 | static sd_event *event_resolve(sd_event *e) { |
182 | return e == SD_EVENT_DEFAULT ? default_event : e; | |
183 | } | |
184 | ||
fd38203a LP |
185 | static int pending_prioq_compare(const void *a, const void *b) { |
186 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 187 | int r; |
fd38203a LP |
188 | |
189 | assert(x->pending); | |
190 | assert(y->pending); | |
191 | ||
baf76283 | 192 | /* Enabled ones first */ |
06e13147 YW |
193 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
194 | if (r != 0) | |
195 | return r; | |
fd38203a | 196 | |
b6d5481b LP |
197 | /* Non rate-limited ones first. */ |
198 | r = CMP(!!x->ratelimited, !!y->ratelimited); | |
199 | if (r != 0) | |
200 | return r; | |
201 | ||
fd38203a | 202 | /* Lower priority values first */ |
9c57a73b YW |
203 | r = CMP(x->priority, y->priority); |
204 | if (r != 0) | |
205 | return r; | |
fd38203a LP |
206 | |
207 | /* Older entries first */ | |
9c57a73b | 208 | return CMP(x->pending_iteration, y->pending_iteration); |
fd38203a LP |
209 | } |
210 | ||
211 | static int prepare_prioq_compare(const void *a, const void *b) { | |
212 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 213 | int r; |
fd38203a LP |
214 | |
215 | assert(x->prepare); | |
216 | assert(y->prepare); | |
217 | ||
8046c457 | 218 | /* Enabled ones first */ |
06e13147 YW |
219 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
220 | if (r != 0) | |
221 | return r; | |
8046c457 | 222 | |
b6d5481b LP |
223 | /* Non rate-limited ones first. */ |
224 | r = CMP(!!x->ratelimited, !!y->ratelimited); | |
225 | if (r != 0) | |
226 | return r; | |
227 | ||
fd38203a LP |
228 | /* Move most recently prepared ones last, so that we can stop |
229 | * preparing as soon as we hit one that has already been | |
230 | * prepared in the current iteration */ | |
9c57a73b YW |
231 | r = CMP(x->prepare_iteration, y->prepare_iteration); |
232 | if (r != 0) | |
233 | return r; | |
fd38203a | 234 | |
fd38203a | 235 | /* Lower priority values first */ |
9c57a73b | 236 | return CMP(x->priority, y->priority); |
fd38203a LP |
237 | } |
238 | ||
b6d5481b LP |
239 | static usec_t time_event_source_next(const sd_event_source *s) { |
240 | assert(s); | |
241 | ||
242 | /* We have two kinds of event sources that have elapsation times associated with them: the actual | |
243 | * time based ones and the ones for which a ratelimit can be in effect (where we want to be notified | |
244 | * once the ratelimit time window ends). Let's return the next elapsing time depending on what we are | |
245 | * looking at here. */ | |
246 | ||
247 | if (s->ratelimited) { /* If rate-limited the next elapsation is when the ratelimit time window ends */ | |
248 | assert(s->rate_limit.begin != 0); | |
249 | assert(s->rate_limit.interval != 0); | |
250 | return usec_add(s->rate_limit.begin, s->rate_limit.interval); | |
251 | } | |
252 | ||
253 | /* Otherwise this must be a time event source, if not ratelimited */ | |
254 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
255 | return s->time.next; | |
256 | ||
257 | return USEC_INFINITY; | |
258 | } | |
259 | ||
1bce0ffa | 260 | static usec_t time_event_source_latest(const sd_event_source *s) { |
b6d5481b LP |
261 | assert(s); |
262 | ||
263 | if (s->ratelimited) { /* For ratelimited stuff the earliest and the latest time shall actually be the | |
264 | * same, as we should avoid adding additional inaccuracy on an inaccuracy time | |
265 | * window */ | |
266 | assert(s->rate_limit.begin != 0); | |
267 | assert(s->rate_limit.interval != 0); | |
268 | return usec_add(s->rate_limit.begin, s->rate_limit.interval); | |
269 | } | |
270 | ||
271 | /* Must be a time event source, if not ratelimited */ | |
272 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
273 | return usec_add(s->time.next, s->time.accuracy); | |
274 | ||
275 | return USEC_INFINITY; | |
1bce0ffa LP |
276 | } |
277 | ||
81107b84 LP |
278 | static bool event_source_timer_candidate(const sd_event_source *s) { |
279 | assert(s); | |
280 | ||
281 | /* Returns true for event sources that either are not pending yet (i.e. where it's worth to mark them pending) | |
282 | * or which are currently ratelimited (i.e. where it's worth leaving the ratelimited state) */ | |
283 | return !s->pending || s->ratelimited; | |
284 | } | |
285 | ||
286 | static int time_prioq_compare(const void *a, const void *b, usec_t (*time_func)(const sd_event_source *s)) { | |
c2ba3ad6 | 287 | const sd_event_source *x = a, *y = b; |
06e13147 | 288 | int r; |
c2ba3ad6 | 289 | |
baf76283 | 290 | /* Enabled ones first */ |
06e13147 YW |
291 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
292 | if (r != 0) | |
293 | return r; | |
c2ba3ad6 | 294 | |
81107b84 | 295 | /* Order "non-pending OR ratelimited" before "pending AND not-ratelimited" */ |
06e13147 YW |
296 | r = CMP(!event_source_timer_candidate(x), !event_source_timer_candidate(y)); |
297 | if (r != 0) | |
298 | return r; | |
c2ba3ad6 LP |
299 | |
300 | /* Order by time */ | |
81107b84 LP |
301 | return CMP(time_func(x), time_func(y)); |
302 | } | |
303 | ||
304 | static int earliest_time_prioq_compare(const void *a, const void *b) { | |
305 | return time_prioq_compare(a, b, time_event_source_next); | |
306 | } | |
307 | ||
308 | static int latest_time_prioq_compare(const void *a, const void *b) { | |
309 | return time_prioq_compare(a, b, time_event_source_latest); | |
c2ba3ad6 LP |
310 | } |
311 | ||
6203e07a | 312 | static int exit_prioq_compare(const void *a, const void *b) { |
da7e457c | 313 | const sd_event_source *x = a, *y = b; |
06e13147 | 314 | int r; |
da7e457c | 315 | |
6203e07a LP |
316 | assert(x->type == SOURCE_EXIT); |
317 | assert(y->type == SOURCE_EXIT); | |
da7e457c | 318 | |
baf76283 | 319 | /* Enabled ones first */ |
06e13147 YW |
320 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
321 | if (r != 0) | |
322 | return r; | |
da7e457c LP |
323 | |
324 | /* Lower priority values first */ | |
6dd91b36 | 325 | return CMP(x->priority, y->priority); |
da7e457c LP |
326 | } |
327 | ||
6a0f1f6d LP |
328 | static void free_clock_data(struct clock_data *d) { |
329 | assert(d); | |
9da4cb2b | 330 | assert(d->wakeup == WAKEUP_CLOCK_DATA); |
6a0f1f6d LP |
331 | |
332 | safe_close(d->fd); | |
333 | prioq_free(d->earliest); | |
334 | prioq_free(d->latest); | |
335 | } | |
336 | ||
8301aa0b | 337 | static sd_event *event_free(sd_event *e) { |
a71fe8b8 LP |
338 | sd_event_source *s; |
339 | ||
fd38203a | 340 | assert(e); |
a71fe8b8 | 341 | |
baf3fdec LP |
342 | e->sigterm_event_source = sd_event_source_unref(e->sigterm_event_source); |
343 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
344 | ||
a71fe8b8 LP |
345 | while ((s = e->sources)) { |
346 | assert(s->floating); | |
347 | source_disconnect(s); | |
348 | sd_event_source_unref(s); | |
349 | } | |
350 | ||
15b38f93 | 351 | assert(e->n_sources == 0); |
fd38203a | 352 | |
afc6adb5 LP |
353 | if (e->default_event_ptr) |
354 | *(e->default_event_ptr) = NULL; | |
355 | ||
03e334a1 | 356 | safe_close(e->epoll_fd); |
03e334a1 | 357 | safe_close(e->watchdog_fd); |
cde93897 | 358 | |
6a0f1f6d | 359 | free_clock_data(&e->realtime); |
a8548816 | 360 | free_clock_data(&e->boottime); |
6a0f1f6d LP |
361 | free_clock_data(&e->monotonic); |
362 | free_clock_data(&e->realtime_alarm); | |
363 | free_clock_data(&e->boottime_alarm); | |
364 | ||
fd38203a LP |
365 | prioq_free(e->pending); |
366 | prioq_free(e->prepare); | |
6203e07a | 367 | prioq_free(e->exit); |
fd38203a LP |
368 | |
369 | free(e->signal_sources); | |
9da4cb2b | 370 | hashmap_free(e->signal_data); |
fd38203a | 371 | |
97ef5391 LP |
372 | hashmap_free(e->inotify_data); |
373 | ||
fd38203a | 374 | hashmap_free(e->child_sources); |
6e9feda3 | 375 | set_free(e->post_sources); |
8301aa0b | 376 | |
5cddd924 LP |
377 | free(e->event_queue); |
378 | ||
8301aa0b | 379 | return mfree(e); |
fd38203a LP |
380 | } |
381 | ||
f7262a9f | 382 | _public_ int sd_event_new(sd_event** ret) { |
fd38203a LP |
383 | sd_event *e; |
384 | int r; | |
385 | ||
305f78bf | 386 | assert_return(ret, -EINVAL); |
fd38203a | 387 | |
d08eb1fa | 388 | e = new(sd_event, 1); |
fd38203a LP |
389 | if (!e) |
390 | return -ENOMEM; | |
391 | ||
d08eb1fa LP |
392 | *e = (sd_event) { |
393 | .n_ref = 1, | |
254d1313 ZJS |
394 | .epoll_fd = -EBADF, |
395 | .watchdog_fd = -EBADF, | |
d08eb1fa | 396 | .realtime.wakeup = WAKEUP_CLOCK_DATA, |
254d1313 | 397 | .realtime.fd = -EBADF, |
d08eb1fa LP |
398 | .realtime.next = USEC_INFINITY, |
399 | .boottime.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 400 | .boottime.fd = -EBADF, |
d08eb1fa LP |
401 | .boottime.next = USEC_INFINITY, |
402 | .monotonic.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 403 | .monotonic.fd = -EBADF, |
d08eb1fa LP |
404 | .monotonic.next = USEC_INFINITY, |
405 | .realtime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 406 | .realtime_alarm.fd = -EBADF, |
d08eb1fa LP |
407 | .realtime_alarm.next = USEC_INFINITY, |
408 | .boottime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 409 | .boottime_alarm.fd = -EBADF, |
d08eb1fa LP |
410 | .boottime_alarm.next = USEC_INFINITY, |
411 | .perturb = USEC_INFINITY, | |
412 | .original_pid = getpid_cached(), | |
413 | }; | |
fd38203a | 414 | |
c983e776 EV |
415 | r = prioq_ensure_allocated(&e->pending, pending_prioq_compare); |
416 | if (r < 0) | |
fd38203a | 417 | goto fail; |
fd38203a LP |
418 | |
419 | e->epoll_fd = epoll_create1(EPOLL_CLOEXEC); | |
420 | if (e->epoll_fd < 0) { | |
421 | r = -errno; | |
422 | goto fail; | |
423 | } | |
424 | ||
7fe2903c LP |
425 | e->epoll_fd = fd_move_above_stdio(e->epoll_fd); |
426 | ||
34b87517 | 427 | if (secure_getenv("SD_EVENT_PROFILE_DELAYS")) { |
28e5e1e9 DT |
428 | 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.", |
429 | special_glyph(SPECIAL_GLYPH_ELLIPSIS)); | |
34b87517 VC |
430 | e->profile_delays = true; |
431 | } | |
432 | ||
fd38203a LP |
433 | *ret = e; |
434 | return 0; | |
435 | ||
436 | fail: | |
437 | event_free(e); | |
438 | return r; | |
439 | } | |
440 | ||
8301aa0b | 441 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event, sd_event, event_free); |
c8e9d15c YW |
442 | #define PROTECT_EVENT(e) \ |
443 | _unused_ _cleanup_(sd_event_unrefp) sd_event *_ref = sd_event_ref(e); | |
fd38203a | 444 | |
afd15bbb ZJS |
445 | _public_ sd_event_source* sd_event_source_disable_unref(sd_event_source *s) { |
446 | if (s) | |
447 | (void) sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
448 | return sd_event_source_unref(s); | |
449 | } | |
450 | ||
eaa3cbef LP |
451 | static bool event_pid_changed(sd_event *e) { |
452 | assert(e); | |
453 | ||
a2360a46 | 454 | /* We don't support people creating an event loop and keeping |
eaa3cbef LP |
455 | * it around over a fork(). Let's complain. */ |
456 | ||
df0ff127 | 457 | return e->original_pid != getpid_cached(); |
eaa3cbef LP |
458 | } |
459 | ||
366e6411 | 460 | static void source_io_unregister(sd_event_source *s) { |
fd38203a LP |
461 | assert(s); |
462 | assert(s->type == SOURCE_IO); | |
463 | ||
f6806734 | 464 | if (event_pid_changed(s->event)) |
366e6411 | 465 | return; |
f6806734 | 466 | |
fd38203a | 467 | if (!s->io.registered) |
366e6411 | 468 | return; |
fd38203a | 469 | |
d1cf2023 | 470 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->io.fd, NULL) < 0) |
f80a5d6a | 471 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll, ignoring: %m", |
55cbfaa5 | 472 | strna(s->description), event_source_type_to_string(s->type)); |
fd38203a LP |
473 | |
474 | s->io.registered = false; | |
fd38203a LP |
475 | } |
476 | ||
305f78bf LP |
477 | static int source_io_register( |
478 | sd_event_source *s, | |
479 | int enabled, | |
480 | uint32_t events) { | |
481 | ||
fd38203a LP |
482 | assert(s); |
483 | assert(s->type == SOURCE_IO); | |
baf76283 | 484 | assert(enabled != SD_EVENT_OFF); |
fd38203a | 485 | |
1eac7948 | 486 | struct epoll_event ev = { |
a82f89aa LP |
487 | .events = events | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0), |
488 | .data.ptr = s, | |
489 | }; | |
fd38203a | 490 | |
15c689d7 | 491 | if (epoll_ctl(s->event->epoll_fd, |
1eac7948 | 492 | s->io.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, |
55c540d3 | 493 | s->io.fd, &ev) < 0) |
fd38203a LP |
494 | return -errno; |
495 | ||
496 | s->io.registered = true; | |
497 | ||
498 | return 0; | |
499 | } | |
500 | ||
f8f3f926 LP |
501 | static void source_child_pidfd_unregister(sd_event_source *s) { |
502 | assert(s); | |
503 | assert(s->type == SOURCE_CHILD); | |
504 | ||
505 | if (event_pid_changed(s->event)) | |
506 | return; | |
507 | ||
508 | if (!s->child.registered) | |
509 | return; | |
510 | ||
511 | if (EVENT_SOURCE_WATCH_PIDFD(s)) | |
512 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->child.pidfd, NULL) < 0) | |
f80a5d6a | 513 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll, ignoring: %m", |
f8f3f926 LP |
514 | strna(s->description), event_source_type_to_string(s->type)); |
515 | ||
516 | s->child.registered = false; | |
517 | } | |
518 | ||
519 | static int source_child_pidfd_register(sd_event_source *s, int enabled) { | |
f8f3f926 LP |
520 | assert(s); |
521 | assert(s->type == SOURCE_CHILD); | |
522 | assert(enabled != SD_EVENT_OFF); | |
523 | ||
524 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { | |
1eac7948 | 525 | struct epoll_event ev = { |
f8f3f926 LP |
526 | .events = EPOLLIN | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0), |
527 | .data.ptr = s, | |
528 | }; | |
529 | ||
55c540d3 ZJS |
530 | if (epoll_ctl(s->event->epoll_fd, |
531 | s->child.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, | |
532 | s->child.pidfd, &ev) < 0) | |
f8f3f926 LP |
533 | return -errno; |
534 | } | |
535 | ||
536 | s->child.registered = true; | |
537 | return 0; | |
538 | } | |
539 | ||
158fe190 LP |
540 | static void source_memory_pressure_unregister(sd_event_source *s) { |
541 | assert(s); | |
542 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
543 | ||
544 | if (event_pid_changed(s->event)) | |
545 | return; | |
546 | ||
547 | if (!s->memory_pressure.registered) | |
548 | return; | |
549 | ||
550 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->memory_pressure.fd, NULL) < 0) | |
551 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll, ignoring: %m", | |
552 | strna(s->description), event_source_type_to_string(s->type)); | |
553 | ||
554 | s->memory_pressure.registered = false; | |
555 | } | |
556 | ||
557 | static int source_memory_pressure_register(sd_event_source *s, int enabled) { | |
558 | assert(s); | |
559 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
560 | assert(enabled != SD_EVENT_OFF); | |
561 | ||
562 | struct epoll_event ev = { | |
563 | .events = s->memory_pressure.write_buffer_size > 0 ? EPOLLOUT : | |
564 | (s->memory_pressure.events | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0)), | |
565 | .data.ptr = s, | |
566 | }; | |
567 | ||
568 | if (epoll_ctl(s->event->epoll_fd, | |
569 | s->memory_pressure.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, | |
570 | s->memory_pressure.fd, &ev) < 0) | |
571 | return -errno; | |
572 | ||
573 | s->memory_pressure.registered = true; | |
574 | return 0; | |
575 | } | |
576 | ||
577 | static void source_memory_pressure_add_to_write_list(sd_event_source *s) { | |
578 | assert(s); | |
579 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
580 | ||
581 | if (s->memory_pressure.in_write_list) | |
582 | return; | |
583 | ||
584 | LIST_PREPEND(memory_pressure.write_list, s->event->memory_pressure_write_list, s); | |
585 | s->memory_pressure.in_write_list = true; | |
586 | } | |
587 | ||
588 | static void source_memory_pressure_remove_from_write_list(sd_event_source *s) { | |
589 | assert(s); | |
590 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
591 | ||
592 | if (!s->memory_pressure.in_write_list) | |
593 | return; | |
594 | ||
595 | LIST_REMOVE(memory_pressure.write_list, s->event->memory_pressure_write_list, s); | |
596 | s->memory_pressure.in_write_list = false; | |
597 | } | |
598 | ||
6a0f1f6d LP |
599 | static clockid_t event_source_type_to_clock(EventSourceType t) { |
600 | ||
601 | switch (t) { | |
602 | ||
603 | case SOURCE_TIME_REALTIME: | |
604 | return CLOCK_REALTIME; | |
605 | ||
a8548816 TG |
606 | case SOURCE_TIME_BOOTTIME: |
607 | return CLOCK_BOOTTIME; | |
608 | ||
6a0f1f6d LP |
609 | case SOURCE_TIME_MONOTONIC: |
610 | return CLOCK_MONOTONIC; | |
611 | ||
612 | case SOURCE_TIME_REALTIME_ALARM: | |
613 | return CLOCK_REALTIME_ALARM; | |
614 | ||
615 | case SOURCE_TIME_BOOTTIME_ALARM: | |
616 | return CLOCK_BOOTTIME_ALARM; | |
617 | ||
618 | default: | |
619 | return (clockid_t) -1; | |
620 | } | |
621 | } | |
622 | ||
623 | static EventSourceType clock_to_event_source_type(clockid_t clock) { | |
624 | ||
625 | switch (clock) { | |
626 | ||
627 | case CLOCK_REALTIME: | |
628 | return SOURCE_TIME_REALTIME; | |
629 | ||
a8548816 TG |
630 | case CLOCK_BOOTTIME: |
631 | return SOURCE_TIME_BOOTTIME; | |
632 | ||
6a0f1f6d LP |
633 | case CLOCK_MONOTONIC: |
634 | return SOURCE_TIME_MONOTONIC; | |
635 | ||
636 | case CLOCK_REALTIME_ALARM: | |
637 | return SOURCE_TIME_REALTIME_ALARM; | |
638 | ||
639 | case CLOCK_BOOTTIME_ALARM: | |
640 | return SOURCE_TIME_BOOTTIME_ALARM; | |
641 | ||
642 | default: | |
643 | return _SOURCE_EVENT_SOURCE_TYPE_INVALID; | |
644 | } | |
645 | } | |
646 | ||
647 | static struct clock_data* event_get_clock_data(sd_event *e, EventSourceType t) { | |
648 | assert(e); | |
649 | ||
650 | switch (t) { | |
651 | ||
652 | case SOURCE_TIME_REALTIME: | |
653 | return &e->realtime; | |
654 | ||
a8548816 TG |
655 | case SOURCE_TIME_BOOTTIME: |
656 | return &e->boottime; | |
657 | ||
6a0f1f6d LP |
658 | case SOURCE_TIME_MONOTONIC: |
659 | return &e->monotonic; | |
660 | ||
661 | case SOURCE_TIME_REALTIME_ALARM: | |
662 | return &e->realtime_alarm; | |
663 | ||
664 | case SOURCE_TIME_BOOTTIME_ALARM: | |
665 | return &e->boottime_alarm; | |
666 | ||
667 | default: | |
668 | return NULL; | |
669 | } | |
670 | } | |
671 | ||
3e4eb8e7 YW |
672 | static void event_free_signal_data(sd_event *e, struct signal_data *d) { |
673 | assert(e); | |
674 | ||
675 | if (!d) | |
676 | return; | |
677 | ||
678 | hashmap_remove(e->signal_data, &d->priority); | |
679 | safe_close(d->fd); | |
680 | free(d); | |
681 | } | |
682 | ||
9da4cb2b LP |
683 | static int event_make_signal_data( |
684 | sd_event *e, | |
685 | int sig, | |
686 | struct signal_data **ret) { | |
4807d2d0 | 687 | |
9da4cb2b LP |
688 | struct signal_data *d; |
689 | bool added = false; | |
690 | sigset_t ss_copy; | |
691 | int64_t priority; | |
f95387cd ZJS |
692 | int r; |
693 | ||
694 | assert(e); | |
695 | ||
f6806734 | 696 | if (event_pid_changed(e)) |
9da4cb2b | 697 | return -ECHILD; |
f6806734 | 698 | |
9da4cb2b LP |
699 | if (e->signal_sources && e->signal_sources[sig]) |
700 | priority = e->signal_sources[sig]->priority; | |
701 | else | |
de05913d | 702 | priority = SD_EVENT_PRIORITY_NORMAL; |
f95387cd | 703 | |
9da4cb2b LP |
704 | d = hashmap_get(e->signal_data, &priority); |
705 | if (d) { | |
706 | if (sigismember(&d->sigset, sig) > 0) { | |
707 | if (ret) | |
708 | *ret = d; | |
709 | return 0; | |
710 | } | |
711 | } else { | |
d08eb1fa | 712 | d = new(struct signal_data, 1); |
9da4cb2b LP |
713 | if (!d) |
714 | return -ENOMEM; | |
715 | ||
d08eb1fa LP |
716 | *d = (struct signal_data) { |
717 | .wakeup = WAKEUP_SIGNAL_DATA, | |
254d1313 | 718 | .fd = -EBADF, |
d08eb1fa LP |
719 | .priority = priority, |
720 | }; | |
9da4cb2b | 721 | |
f656fdb6 | 722 | r = hashmap_ensure_put(&e->signal_data, &uint64_hash_ops, &d->priority, d); |
90f604d1 ZJS |
723 | if (r < 0) { |
724 | free(d); | |
9da4cb2b | 725 | return r; |
90f604d1 | 726 | } |
f95387cd | 727 | |
9da4cb2b LP |
728 | added = true; |
729 | } | |
730 | ||
731 | ss_copy = d->sigset; | |
732 | assert_se(sigaddset(&ss_copy, sig) >= 0); | |
733 | ||
cbff793f ZJS |
734 | r = signalfd(d->fd >= 0 ? d->fd : -1, /* the first arg must be -1 or a valid signalfd */ |
735 | &ss_copy, | |
736 | SFD_NONBLOCK|SFD_CLOEXEC); | |
9da4cb2b LP |
737 | if (r < 0) { |
738 | r = -errno; | |
739 | goto fail; | |
740 | } | |
741 | ||
742 | d->sigset = ss_copy; | |
f95387cd | 743 | |
9da4cb2b LP |
744 | if (d->fd >= 0) { |
745 | if (ret) | |
746 | *ret = d; | |
f95387cd | 747 | return 0; |
9da4cb2b LP |
748 | } |
749 | ||
7fe2903c | 750 | d->fd = fd_move_above_stdio(r); |
f95387cd | 751 | |
1eac7948 | 752 | struct epoll_event ev = { |
a82f89aa LP |
753 | .events = EPOLLIN, |
754 | .data.ptr = d, | |
755 | }; | |
f95387cd | 756 | |
15c689d7 | 757 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { |
9da4cb2b LP |
758 | r = -errno; |
759 | goto fail; | |
f95387cd ZJS |
760 | } |
761 | ||
9da4cb2b LP |
762 | if (ret) |
763 | *ret = d; | |
764 | ||
f95387cd | 765 | return 0; |
9da4cb2b LP |
766 | |
767 | fail: | |
3e4eb8e7 YW |
768 | if (added) |
769 | event_free_signal_data(e, d); | |
9da4cb2b LP |
770 | |
771 | return r; | |
772 | } | |
773 | ||
774 | static void event_unmask_signal_data(sd_event *e, struct signal_data *d, int sig) { | |
775 | assert(e); | |
776 | assert(d); | |
777 | ||
778 | /* Turns off the specified signal in the signal data | |
779 | * object. If the signal mask of the object becomes empty that | |
780 | * way removes it. */ | |
781 | ||
782 | if (sigismember(&d->sigset, sig) == 0) | |
783 | return; | |
784 | ||
785 | assert_se(sigdelset(&d->sigset, sig) >= 0); | |
786 | ||
787 | if (sigisemptyset(&d->sigset)) { | |
9da4cb2b | 788 | /* If all the mask is all-zero we can get rid of the structure */ |
3e4eb8e7 | 789 | event_free_signal_data(e, d); |
9da4cb2b LP |
790 | return; |
791 | } | |
792 | ||
01e6af73 YW |
793 | if (event_pid_changed(e)) |
794 | return; | |
795 | ||
9da4cb2b LP |
796 | assert(d->fd >= 0); |
797 | ||
798 | if (signalfd(d->fd, &d->sigset, SFD_NONBLOCK|SFD_CLOEXEC) < 0) | |
799 | log_debug_errno(errno, "Failed to unset signal bit, ignoring: %m"); | |
800 | } | |
801 | ||
802 | static void event_gc_signal_data(sd_event *e, const int64_t *priority, int sig) { | |
803 | struct signal_data *d; | |
804 | static const int64_t zero_priority = 0; | |
805 | ||
806 | assert(e); | |
807 | ||
f8f3f926 LP |
808 | /* Rechecks if the specified signal is still something we are interested in. If not, we'll unmask it, |
809 | * and possibly drop the signalfd for it. */ | |
9da4cb2b LP |
810 | |
811 | if (sig == SIGCHLD && | |
b6d5481b | 812 | e->n_online_child_sources > 0) |
9da4cb2b LP |
813 | return; |
814 | ||
815 | if (e->signal_sources && | |
816 | e->signal_sources[sig] && | |
b6d5481b | 817 | event_source_is_online(e->signal_sources[sig])) |
9da4cb2b LP |
818 | return; |
819 | ||
820 | /* | |
821 | * The specified signal might be enabled in three different queues: | |
822 | * | |
823 | * 1) the one that belongs to the priority passed (if it is non-NULL) | |
824 | * 2) the one that belongs to the priority of the event source of the signal (if there is one) | |
825 | * 3) the 0 priority (to cover the SIGCHLD case) | |
826 | * | |
827 | * Hence, let's remove it from all three here. | |
828 | */ | |
829 | ||
830 | if (priority) { | |
831 | d = hashmap_get(e->signal_data, priority); | |
832 | if (d) | |
833 | event_unmask_signal_data(e, d, sig); | |
834 | } | |
835 | ||
836 | if (e->signal_sources && e->signal_sources[sig]) { | |
837 | d = hashmap_get(e->signal_data, &e->signal_sources[sig]->priority); | |
838 | if (d) | |
839 | event_unmask_signal_data(e, d, sig); | |
840 | } | |
841 | ||
842 | d = hashmap_get(e->signal_data, &zero_priority); | |
843 | if (d) | |
844 | event_unmask_signal_data(e, d, sig); | |
f95387cd ZJS |
845 | } |
846 | ||
e1951c16 MS |
847 | static void event_source_pp_prioq_reshuffle(sd_event_source *s) { |
848 | assert(s); | |
849 | ||
850 | /* Reshuffles the pending + prepare prioqs. Called whenever the dispatch order changes, i.e. when | |
851 | * they are enabled/disabled or marked pending and such. */ | |
852 | ||
853 | if (s->pending) | |
854 | prioq_reshuffle(s->event->pending, s, &s->pending_index); | |
855 | ||
856 | if (s->prepare) | |
857 | prioq_reshuffle(s->event->prepare, s, &s->prepare_index); | |
858 | } | |
859 | ||
860 | static void event_source_time_prioq_reshuffle(sd_event_source *s) { | |
861 | struct clock_data *d; | |
862 | ||
863 | assert(s); | |
e1951c16 MS |
864 | |
865 | /* Called whenever the event source's timer ordering properties changed, i.e. time, accuracy, | |
5c08c7ab YW |
866 | * pending, enable state, and ratelimiting state. Makes sure the two prioq's are ordered |
867 | * properly again. */ | |
b6d5481b LP |
868 | |
869 | if (s->ratelimited) | |
870 | d = &s->event->monotonic; | |
5c08c7ab | 871 | else if (EVENT_SOURCE_IS_TIME(s->type)) |
b6d5481b | 872 | assert_se(d = event_get_clock_data(s->event, s->type)); |
5c08c7ab YW |
873 | else |
874 | return; /* no-op for an event source which is neither a timer nor ratelimited. */ | |
b6d5481b | 875 | |
f41315fc LP |
876 | prioq_reshuffle(d->earliest, s, &s->earliest_index); |
877 | prioq_reshuffle(d->latest, s, &s->latest_index); | |
e1951c16 MS |
878 | d->needs_rearm = true; |
879 | } | |
880 | ||
1e45e3fe LP |
881 | static void event_source_time_prioq_remove( |
882 | sd_event_source *s, | |
883 | struct clock_data *d) { | |
884 | ||
885 | assert(s); | |
886 | assert(d); | |
887 | ||
f41315fc LP |
888 | prioq_remove(d->earliest, s, &s->earliest_index); |
889 | prioq_remove(d->latest, s, &s->latest_index); | |
890 | s->earliest_index = s->latest_index = PRIOQ_IDX_NULL; | |
1e45e3fe LP |
891 | d->needs_rearm = true; |
892 | } | |
893 | ||
a71fe8b8 LP |
894 | static void source_disconnect(sd_event_source *s) { |
895 | sd_event *event; | |
897448bd | 896 | int r; |
a71fe8b8 | 897 | |
fd38203a LP |
898 | assert(s); |
899 | ||
a71fe8b8 LP |
900 | if (!s->event) |
901 | return; | |
15b38f93 | 902 | |
a71fe8b8 | 903 | assert(s->event->n_sources > 0); |
fd38203a | 904 | |
a71fe8b8 | 905 | switch (s->type) { |
fd38203a | 906 | |
a71fe8b8 LP |
907 | case SOURCE_IO: |
908 | if (s->io.fd >= 0) | |
909 | source_io_unregister(s); | |
fd38203a | 910 | |
a71fe8b8 | 911 | break; |
6a0f1f6d | 912 | |
a71fe8b8 | 913 | case SOURCE_TIME_REALTIME: |
a8548816 | 914 | case SOURCE_TIME_BOOTTIME: |
a71fe8b8 LP |
915 | case SOURCE_TIME_MONOTONIC: |
916 | case SOURCE_TIME_REALTIME_ALARM: | |
b6d5481b LP |
917 | case SOURCE_TIME_BOOTTIME_ALARM: |
918 | /* Only remove this event source from the time event source here if it is not ratelimited. If | |
919 | * it is ratelimited, we'll remove it below, separately. Why? Because the clock used might | |
920 | * differ: ratelimiting always uses CLOCK_MONOTONIC, but timer events might use any clock */ | |
921 | ||
922 | if (!s->ratelimited) { | |
923 | struct clock_data *d; | |
924 | assert_se(d = event_get_clock_data(s->event, s->type)); | |
925 | event_source_time_prioq_remove(s, d); | |
926 | } | |
927 | ||
a71fe8b8 | 928 | break; |
a71fe8b8 LP |
929 | |
930 | case SOURCE_SIGNAL: | |
931 | if (s->signal.sig > 0) { | |
9da4cb2b | 932 | |
a71fe8b8 LP |
933 | if (s->event->signal_sources) |
934 | s->event->signal_sources[s->signal.sig] = NULL; | |
4807d2d0 | 935 | |
9da4cb2b | 936 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
897448bd LP |
937 | |
938 | if (s->signal.unblock) { | |
939 | sigset_t new_ss; | |
940 | ||
941 | if (sigemptyset(&new_ss) < 0) | |
942 | log_debug_errno(errno, "Failed to reset signal set, ignoring: %m"); | |
943 | else if (sigaddset(&new_ss, s->signal.sig) < 0) | |
944 | log_debug_errno(errno, "Failed to add signal %i to signal mask, ignoring: %m", s->signal.sig); | |
945 | else { | |
946 | r = pthread_sigmask(SIG_UNBLOCK, &new_ss, NULL); | |
947 | if (r != 0) | |
948 | log_debug_errno(r, "Failed to unblock signal %i, ignoring: %m", s->signal.sig); | |
949 | } | |
950 | } | |
6a0f1f6d | 951 | } |
fd38203a | 952 | |
a71fe8b8 | 953 | break; |
fd38203a | 954 | |
a71fe8b8 | 955 | case SOURCE_CHILD: |
86587c93 YW |
956 | if (event_pid_changed(s->event)) |
957 | s->child.process_owned = false; | |
958 | ||
a71fe8b8 | 959 | if (s->child.pid > 0) { |
b6d5481b LP |
960 | if (event_source_is_online(s)) { |
961 | assert(s->event->n_online_child_sources > 0); | |
962 | s->event->n_online_child_sources--; | |
4807d2d0 | 963 | } |
fd38203a | 964 | |
4a0b58c4 | 965 | (void) hashmap_remove(s->event->child_sources, PID_TO_PTR(s->child.pid)); |
a71fe8b8 | 966 | } |
fd38203a | 967 | |
f8f3f926 LP |
968 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
969 | source_child_pidfd_unregister(s); | |
970 | else | |
971 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
972 | ||
a71fe8b8 | 973 | break; |
fd38203a | 974 | |
a71fe8b8 LP |
975 | case SOURCE_DEFER: |
976 | /* nothing */ | |
977 | break; | |
fd38203a | 978 | |
a71fe8b8 LP |
979 | case SOURCE_POST: |
980 | set_remove(s->event->post_sources, s); | |
981 | break; | |
da7e457c | 982 | |
a71fe8b8 LP |
983 | case SOURCE_EXIT: |
984 | prioq_remove(s->event->exit, s, &s->exit.prioq_index); | |
985 | break; | |
0eb2e0e3 | 986 | |
97ef5391 LP |
987 | case SOURCE_INOTIFY: { |
988 | struct inode_data *inode_data; | |
989 | ||
990 | inode_data = s->inotify.inode_data; | |
991 | if (inode_data) { | |
992 | struct inotify_data *inotify_data; | |
993 | assert_se(inotify_data = inode_data->inotify_data); | |
994 | ||
995 | /* Detach this event source from the inode object */ | |
996 | LIST_REMOVE(inotify.by_inode_data, inode_data->event_sources, s); | |
997 | s->inotify.inode_data = NULL; | |
998 | ||
999 | if (s->pending) { | |
1000 | assert(inotify_data->n_pending > 0); | |
1001 | inotify_data->n_pending--; | |
1002 | } | |
1003 | ||
1004 | /* Note that we don't reduce the inotify mask for the watch descriptor here if the inode is | |
1005 | * continued to being watched. That's because inotify doesn't really have an API for that: we | |
1006 | * can only change watch masks with access to the original inode either by fd or by path. But | |
1007 | * paths aren't stable, and keeping an O_PATH fd open all the time would mean wasting an fd | |
f21f31b2 | 1008 | * continuously and keeping the mount busy which we can't really do. We could reconstruct the |
97ef5391 LP |
1009 | * original inode from /proc/self/fdinfo/$INOTIFY_FD (as all watch descriptors are listed |
1010 | * there), but given the need for open_by_handle_at() which is privileged and not universally | |
1011 | * available this would be quite an incomplete solution. Hence we go the other way, leave the | |
1012 | * mask set, even if it is not minimized now, and ignore all events we aren't interested in | |
1013 | * anymore after reception. Yes, this sucks, but … Linux … */ | |
1014 | ||
1015 | /* Maybe release the inode data (and its inotify) */ | |
1016 | event_gc_inode_data(s->event, inode_data); | |
1017 | } | |
1018 | ||
1019 | break; | |
1020 | } | |
1021 | ||
158fe190 LP |
1022 | case SOURCE_MEMORY_PRESSURE: |
1023 | source_memory_pressure_remove_from_write_list(s); | |
1024 | source_memory_pressure_unregister(s); | |
1025 | break; | |
1026 | ||
a71fe8b8 | 1027 | default: |
04499a70 | 1028 | assert_not_reached(); |
a71fe8b8 | 1029 | } |
6e9feda3 | 1030 | |
a71fe8b8 LP |
1031 | if (s->pending) |
1032 | prioq_remove(s->event->pending, s, &s->pending_index); | |
9d3e3aa5 | 1033 | |
a71fe8b8 LP |
1034 | if (s->prepare) |
1035 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
fd38203a | 1036 | |
b6d5481b LP |
1037 | if (s->ratelimited) |
1038 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
1039 | ||
e514aa1e | 1040 | event = TAKE_PTR(s->event); |
a71fe8b8 LP |
1041 | LIST_REMOVE(sources, event->sources, s); |
1042 | event->n_sources--; | |
fd38203a | 1043 | |
f5982559 LP |
1044 | /* Note that we don't invalidate the type here, since we still need it in order to close the fd or |
1045 | * pidfd associated with this event source, which we'll do only on source_free(). */ | |
1046 | ||
a71fe8b8 LP |
1047 | if (!s->floating) |
1048 | sd_event_unref(event); | |
1049 | } | |
1050 | ||
75db809a | 1051 | static sd_event_source* source_free(sd_event_source *s) { |
a71fe8b8 | 1052 | assert(s); |
fd38203a | 1053 | |
a71fe8b8 | 1054 | source_disconnect(s); |
ab93297c NM |
1055 | |
1056 | if (s->type == SOURCE_IO && s->io.owned) | |
15723a1d LP |
1057 | s->io.fd = safe_close(s->io.fd); |
1058 | ||
f8f3f926 LP |
1059 | if (s->type == SOURCE_CHILD) { |
1060 | /* Eventually the kernel will do this automatically for us, but for now let's emulate this (unreliably) in userspace. */ | |
1061 | ||
1062 | if (s->child.process_owned) { | |
1063 | ||
1064 | if (!s->child.exited) { | |
1065 | bool sent = false; | |
1066 | ||
1067 | if (s->child.pidfd >= 0) { | |
1068 | if (pidfd_send_signal(s->child.pidfd, SIGKILL, NULL, 0) < 0) { | |
1069 | if (errno == ESRCH) /* Already dead */ | |
1070 | sent = true; | |
1071 | else if (!ERRNO_IS_NOT_SUPPORTED(errno)) | |
1072 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via pidfd_send_signal(), re-trying via kill(): %m", | |
1073 | s->child.pid); | |
1074 | } else | |
1075 | sent = true; | |
1076 | } | |
1077 | ||
1078 | if (!sent) | |
1079 | if (kill(s->child.pid, SIGKILL) < 0) | |
1080 | if (errno != ESRCH) /* Already dead */ | |
1081 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via kill(), ignoring: %m", | |
1082 | s->child.pid); | |
1083 | } | |
1084 | ||
1085 | if (!s->child.waited) { | |
1086 | siginfo_t si = {}; | |
1087 | ||
1088 | /* Reap the child if we can */ | |
1089 | (void) waitid(P_PID, s->child.pid, &si, WEXITED); | |
1090 | } | |
1091 | } | |
1092 | ||
1093 | if (s->child.pidfd_owned) | |
1094 | s->child.pidfd = safe_close(s->child.pidfd); | |
1095 | } | |
1096 | ||
158fe190 LP |
1097 | if (s->type == SOURCE_MEMORY_PRESSURE) { |
1098 | s->memory_pressure.fd = safe_close(s->memory_pressure.fd); | |
1099 | s->memory_pressure.write_buffer = mfree(s->memory_pressure.write_buffer); | |
1100 | } | |
1101 | ||
15723a1d LP |
1102 | if (s->destroy_callback) |
1103 | s->destroy_callback(s->userdata); | |
ab93297c | 1104 | |
356779df | 1105 | free(s->description); |
75db809a | 1106 | return mfree(s); |
fd38203a | 1107 | } |
8c75fe17 | 1108 | DEFINE_TRIVIAL_CLEANUP_FUNC(sd_event_source*, source_free); |
fd38203a LP |
1109 | |
1110 | static int source_set_pending(sd_event_source *s, bool b) { | |
1111 | int r; | |
1112 | ||
1113 | assert(s); | |
6203e07a | 1114 | assert(s->type != SOURCE_EXIT); |
fd38203a LP |
1115 | |
1116 | if (s->pending == b) | |
1117 | return 0; | |
1118 | ||
1119 | s->pending = b; | |
1120 | ||
1121 | if (b) { | |
1122 | s->pending_iteration = s->event->iteration; | |
1123 | ||
1124 | r = prioq_put(s->event->pending, s, &s->pending_index); | |
1125 | if (r < 0) { | |
1126 | s->pending = false; | |
1127 | return r; | |
1128 | } | |
1129 | } else | |
1130 | assert_se(prioq_remove(s->event->pending, s, &s->pending_index)); | |
1131 | ||
e1951c16 MS |
1132 | if (EVENT_SOURCE_IS_TIME(s->type)) |
1133 | event_source_time_prioq_reshuffle(s); | |
2576a19e | 1134 | |
9da4cb2b LP |
1135 | if (s->type == SOURCE_SIGNAL && !b) { |
1136 | struct signal_data *d; | |
1137 | ||
1138 | d = hashmap_get(s->event->signal_data, &s->priority); | |
1139 | if (d && d->current == s) | |
1140 | d->current = NULL; | |
1141 | } | |
1142 | ||
97ef5391 LP |
1143 | if (s->type == SOURCE_INOTIFY) { |
1144 | ||
1145 | assert(s->inotify.inode_data); | |
1146 | assert(s->inotify.inode_data->inotify_data); | |
1147 | ||
1148 | if (b) | |
1149 | s->inotify.inode_data->inotify_data->n_pending ++; | |
1150 | else { | |
1151 | assert(s->inotify.inode_data->inotify_data->n_pending > 0); | |
1152 | s->inotify.inode_data->inotify_data->n_pending --; | |
1153 | } | |
1154 | } | |
1155 | ||
efd3be9d | 1156 | return 1; |
fd38203a LP |
1157 | } |
1158 | ||
a71fe8b8 | 1159 | static sd_event_source *source_new(sd_event *e, bool floating, EventSourceType type) { |
a38cf9fb LP |
1160 | |
1161 | /* Let's allocate exactly what we need. Note that the difference of the smallest event source | |
1162 | * structure to the largest is 144 bytes on x86-64 at the time of writing, i.e. more than two cache | |
1163 | * lines. */ | |
1164 | static const size_t size_table[_SOURCE_EVENT_SOURCE_TYPE_MAX] = { | |
1165 | [SOURCE_IO] = endoffsetof_field(sd_event_source, io), | |
1166 | [SOURCE_TIME_REALTIME] = endoffsetof_field(sd_event_source, time), | |
1167 | [SOURCE_TIME_BOOTTIME] = endoffsetof_field(sd_event_source, time), | |
1168 | [SOURCE_TIME_MONOTONIC] = endoffsetof_field(sd_event_source, time), | |
1169 | [SOURCE_TIME_REALTIME_ALARM] = endoffsetof_field(sd_event_source, time), | |
1170 | [SOURCE_TIME_BOOTTIME_ALARM] = endoffsetof_field(sd_event_source, time), | |
1171 | [SOURCE_SIGNAL] = endoffsetof_field(sd_event_source, signal), | |
1172 | [SOURCE_CHILD] = endoffsetof_field(sd_event_source, child), | |
1173 | [SOURCE_DEFER] = endoffsetof_field(sd_event_source, defer), | |
1174 | [SOURCE_POST] = endoffsetof_field(sd_event_source, post), | |
1175 | [SOURCE_EXIT] = endoffsetof_field(sd_event_source, exit), | |
1176 | [SOURCE_INOTIFY] = endoffsetof_field(sd_event_source, inotify), | |
158fe190 | 1177 | [SOURCE_MEMORY_PRESSURE] = endoffsetof_field(sd_event_source, memory_pressure), |
a38cf9fb LP |
1178 | }; |
1179 | ||
fd38203a LP |
1180 | sd_event_source *s; |
1181 | ||
1182 | assert(e); | |
a38cf9fb LP |
1183 | assert(type >= 0); |
1184 | assert(type < _SOURCE_EVENT_SOURCE_TYPE_MAX); | |
1185 | assert(size_table[type] > 0); | |
fd38203a | 1186 | |
a38cf9fb LP |
1187 | /* We use expand_to_usable() here to tell gcc that it should consider this an object of the full |
1188 | * size, even if we only allocate the initial part we need. */ | |
1189 | s = expand_to_usable(malloc0(size_table[type]), sizeof(sd_event_source)); | |
fd38203a LP |
1190 | if (!s) |
1191 | return NULL; | |
1192 | ||
a38cf9fb LP |
1193 | /* Note: we cannot use compound initialization here, because sizeof(sd_event_source) is likely larger |
1194 | * than what we allocated here. */ | |
1195 | s->n_ref = 1; | |
1196 | s->event = e; | |
1197 | s->floating = floating; | |
1198 | s->type = type; | |
1199 | s->pending_index = PRIOQ_IDX_NULL; | |
1200 | s->prepare_index = PRIOQ_IDX_NULL; | |
a71fe8b8 LP |
1201 | |
1202 | if (!floating) | |
1203 | sd_event_ref(e); | |
fd38203a | 1204 | |
a71fe8b8 | 1205 | LIST_PREPEND(sources, e->sources, s); |
313cefa1 | 1206 | e->n_sources++; |
15b38f93 | 1207 | |
fd38203a LP |
1208 | return s; |
1209 | } | |
1210 | ||
b9350e70 LP |
1211 | static int io_exit_callback(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
1212 | assert(s); | |
1213 | ||
1214 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1215 | } | |
1216 | ||
f7262a9f | 1217 | _public_ int sd_event_add_io( |
fd38203a | 1218 | sd_event *e, |
151b9b96 | 1219 | sd_event_source **ret, |
fd38203a LP |
1220 | int fd, |
1221 | uint32_t events, | |
718db961 | 1222 | sd_event_io_handler_t callback, |
151b9b96 | 1223 | void *userdata) { |
fd38203a | 1224 | |
ec766a51 | 1225 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1226 | int r; |
1227 | ||
305f78bf | 1228 | assert_return(e, -EINVAL); |
b937d761 | 1229 | assert_return(e = event_resolve(e), -ENOPKG); |
8ac43fee | 1230 | assert_return(fd >= 0, -EBADF); |
2a16a986 | 1231 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 1232 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1233 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1234 | |
b9350e70 LP |
1235 | if (!callback) |
1236 | callback = io_exit_callback; | |
1237 | ||
a71fe8b8 | 1238 | s = source_new(e, !ret, SOURCE_IO); |
fd38203a LP |
1239 | if (!s) |
1240 | return -ENOMEM; | |
1241 | ||
9da4cb2b | 1242 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1243 | s->io.fd = fd; |
1244 | s->io.events = events; | |
1245 | s->io.callback = callback; | |
1246 | s->userdata = userdata; | |
baf76283 | 1247 | s->enabled = SD_EVENT_ON; |
fd38203a | 1248 | |
baf76283 | 1249 | r = source_io_register(s, s->enabled, events); |
ec766a51 | 1250 | if (r < 0) |
050f74f2 | 1251 | return r; |
fd38203a | 1252 | |
a71fe8b8 LP |
1253 | if (ret) |
1254 | *ret = s; | |
ec766a51 | 1255 | TAKE_PTR(s); |
a71fe8b8 | 1256 | |
fd38203a LP |
1257 | return 0; |
1258 | } | |
1259 | ||
52444dc4 | 1260 | static void initialize_perturb(sd_event *e) { |
6d2326e0 | 1261 | sd_id128_t id = {}; |
52444dc4 | 1262 | |
6d2326e0 YW |
1263 | /* When we sleep for longer, we try to realign the wakeup to the same time within each |
1264 | * minute/second/250ms, so that events all across the system can be coalesced into a single CPU | |
1265 | * wakeup. However, let's take some system-specific randomness for this value, so that in a network | |
1266 | * of systems with synced clocks timer events are distributed a bit. Here, we calculate a | |
1267 | * perturbation usec offset from the boot ID (or machine ID if failed, e.g. /proc is not mounted). */ | |
52444dc4 | 1268 | |
3a43da28 | 1269 | if (_likely_(e->perturb != USEC_INFINITY)) |
52444dc4 LP |
1270 | return; |
1271 | ||
1912f790 | 1272 | if (sd_id128_get_boot(&id) >= 0 || sd_id128_get_machine(&id) >= 0) |
6d2326e0 YW |
1273 | e->perturb = (id.qwords[0] ^ id.qwords[1]) % USEC_PER_MINUTE; |
1274 | else | |
1275 | e->perturb = 0; /* This is a super early process without /proc and /etc ?? */ | |
52444dc4 LP |
1276 | } |
1277 | ||
fd38203a LP |
1278 | static int event_setup_timer_fd( |
1279 | sd_event *e, | |
6a0f1f6d LP |
1280 | struct clock_data *d, |
1281 | clockid_t clock) { | |
fd38203a | 1282 | |
fd38203a | 1283 | assert(e); |
6a0f1f6d | 1284 | assert(d); |
fd38203a | 1285 | |
6a0f1f6d | 1286 | if (_likely_(d->fd >= 0)) |
fd38203a LP |
1287 | return 0; |
1288 | ||
254d1313 | 1289 | _cleanup_close_ int fd = -EBADF; |
b44d87e2 | 1290 | |
6a0f1f6d | 1291 | fd = timerfd_create(clock, TFD_NONBLOCK|TFD_CLOEXEC); |
fd38203a LP |
1292 | if (fd < 0) |
1293 | return -errno; | |
1294 | ||
7fe2903c LP |
1295 | fd = fd_move_above_stdio(fd); |
1296 | ||
1eac7948 | 1297 | struct epoll_event ev = { |
a82f89aa LP |
1298 | .events = EPOLLIN, |
1299 | .data.ptr = d, | |
1300 | }; | |
fd38203a | 1301 | |
15c689d7 | 1302 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) |
fd38203a | 1303 | return -errno; |
fd38203a | 1304 | |
b44d87e2 | 1305 | d->fd = TAKE_FD(fd); |
fd38203a LP |
1306 | return 0; |
1307 | } | |
1308 | ||
c4f1aff2 TG |
1309 | static int time_exit_callback(sd_event_source *s, uint64_t usec, void *userdata) { |
1310 | assert(s); | |
1311 | ||
1312 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1313 | } | |
1314 | ||
41c63f36 LP |
1315 | static int setup_clock_data(sd_event *e, struct clock_data *d, clockid_t clock) { |
1316 | int r; | |
1317 | ||
1318 | assert(d); | |
1319 | ||
1320 | if (d->fd < 0) { | |
1321 | r = event_setup_timer_fd(e, d, clock); | |
1322 | if (r < 0) | |
1323 | return r; | |
1324 | } | |
1325 | ||
1326 | r = prioq_ensure_allocated(&d->earliest, earliest_time_prioq_compare); | |
1327 | if (r < 0) | |
1328 | return r; | |
1329 | ||
1330 | r = prioq_ensure_allocated(&d->latest, latest_time_prioq_compare); | |
1331 | if (r < 0) | |
1332 | return r; | |
1333 | ||
1334 | return 0; | |
1335 | } | |
1336 | ||
1e45e3fe LP |
1337 | static int event_source_time_prioq_put( |
1338 | sd_event_source *s, | |
1339 | struct clock_data *d) { | |
1340 | ||
1341 | int r; | |
1342 | ||
1343 | assert(s); | |
1344 | assert(d); | |
19947509 | 1345 | assert(EVENT_SOURCE_USES_TIME_PRIOQ(s->type)); |
1e45e3fe | 1346 | |
f41315fc | 1347 | r = prioq_put(d->earliest, s, &s->earliest_index); |
1e45e3fe LP |
1348 | if (r < 0) |
1349 | return r; | |
1350 | ||
f41315fc | 1351 | r = prioq_put(d->latest, s, &s->latest_index); |
1e45e3fe | 1352 | if (r < 0) { |
f41315fc LP |
1353 | assert_se(prioq_remove(d->earliest, s, &s->earliest_index) > 0); |
1354 | s->earliest_index = PRIOQ_IDX_NULL; | |
1e45e3fe LP |
1355 | return r; |
1356 | } | |
1357 | ||
1358 | d->needs_rearm = true; | |
1359 | return 0; | |
1360 | } | |
1361 | ||
6a0f1f6d | 1362 | _public_ int sd_event_add_time( |
fd38203a | 1363 | sd_event *e, |
151b9b96 | 1364 | sd_event_source **ret, |
6a0f1f6d | 1365 | clockid_t clock, |
fd38203a | 1366 | uint64_t usec, |
c2ba3ad6 | 1367 | uint64_t accuracy, |
718db961 | 1368 | sd_event_time_handler_t callback, |
151b9b96 | 1369 | void *userdata) { |
fd38203a | 1370 | |
6a0f1f6d | 1371 | EventSourceType type; |
ec766a51 | 1372 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6a0f1f6d | 1373 | struct clock_data *d; |
fd38203a LP |
1374 | int r; |
1375 | ||
305f78bf | 1376 | assert_return(e, -EINVAL); |
b937d761 | 1377 | assert_return(e = event_resolve(e), -ENOPKG); |
f5fbe71d | 1378 | assert_return(accuracy != UINT64_MAX, -EINVAL); |
da7e457c | 1379 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1380 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1381 | |
e475d10c LP |
1382 | if (!clock_supported(clock)) /* Checks whether the kernel supports the clock */ |
1383 | return -EOPNOTSUPP; | |
1384 | ||
1385 | type = clock_to_event_source_type(clock); /* checks whether sd-event supports this clock */ | |
1386 | if (type < 0) | |
3411372e LP |
1387 | return -EOPNOTSUPP; |
1388 | ||
c4f1aff2 TG |
1389 | if (!callback) |
1390 | callback = time_exit_callback; | |
1391 | ||
1e45e3fe | 1392 | assert_se(d = event_get_clock_data(e, type)); |
c2ba3ad6 | 1393 | |
41c63f36 | 1394 | r = setup_clock_data(e, d, clock); |
c983e776 EV |
1395 | if (r < 0) |
1396 | return r; | |
fd38203a | 1397 | |
a71fe8b8 | 1398 | s = source_new(e, !ret, type); |
fd38203a LP |
1399 | if (!s) |
1400 | return -ENOMEM; | |
1401 | ||
1402 | s->time.next = usec; | |
c2ba3ad6 | 1403 | s->time.accuracy = accuracy == 0 ? DEFAULT_ACCURACY_USEC : accuracy; |
fd38203a | 1404 | s->time.callback = callback; |
f41315fc | 1405 | s->earliest_index = s->latest_index = PRIOQ_IDX_NULL; |
fd38203a | 1406 | s->userdata = userdata; |
baf76283 | 1407 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1408 | |
1e45e3fe | 1409 | r = event_source_time_prioq_put(s, d); |
c2ba3ad6 | 1410 | if (r < 0) |
ec766a51 | 1411 | return r; |
fd38203a | 1412 | |
a71fe8b8 LP |
1413 | if (ret) |
1414 | *ret = s; | |
ec766a51 | 1415 | TAKE_PTR(s); |
a71fe8b8 | 1416 | |
fd38203a LP |
1417 | return 0; |
1418 | } | |
1419 | ||
d6a83dc4 LP |
1420 | _public_ int sd_event_add_time_relative( |
1421 | sd_event *e, | |
1422 | sd_event_source **ret, | |
1423 | clockid_t clock, | |
1424 | uint64_t usec, | |
1425 | uint64_t accuracy, | |
1426 | sd_event_time_handler_t callback, | |
1427 | void *userdata) { | |
1428 | ||
1429 | usec_t t; | |
1430 | int r; | |
1431 | ||
1432 | /* Same as sd_event_add_time() but operates relative to the event loop's current point in time, and | |
1433 | * checks for overflow. */ | |
1434 | ||
1435 | r = sd_event_now(e, clock, &t); | |
1436 | if (r < 0) | |
1437 | return r; | |
1438 | ||
1439 | if (usec >= USEC_INFINITY - t) | |
1440 | return -EOVERFLOW; | |
1441 | ||
1442 | return sd_event_add_time(e, ret, clock, t + usec, accuracy, callback, userdata); | |
1443 | } | |
1444 | ||
59bc1fd7 LP |
1445 | static int signal_exit_callback(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) { |
1446 | assert(s); | |
1447 | ||
1448 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1449 | } | |
1450 | ||
f7262a9f | 1451 | _public_ int sd_event_add_signal( |
305f78bf | 1452 | sd_event *e, |
151b9b96 | 1453 | sd_event_source **ret, |
305f78bf | 1454 | int sig, |
718db961 | 1455 | sd_event_signal_handler_t callback, |
151b9b96 | 1456 | void *userdata) { |
305f78bf | 1457 | |
ec766a51 | 1458 | _cleanup_(source_freep) sd_event_source *s = NULL; |
9da4cb2b | 1459 | struct signal_data *d; |
897448bd LP |
1460 | sigset_t new_ss; |
1461 | bool block_it; | |
fd38203a LP |
1462 | int r; |
1463 | ||
305f78bf | 1464 | assert_return(e, -EINVAL); |
b937d761 | 1465 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1466 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1467 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1468 | |
897448bd LP |
1469 | /* Let's make sure our special flag stays outside of the valid signal range */ |
1470 | assert_cc(_NSIG < SD_EVENT_SIGNAL_PROCMASK); | |
1471 | ||
1472 | if (sig & SD_EVENT_SIGNAL_PROCMASK) { | |
1473 | sig &= ~SD_EVENT_SIGNAL_PROCMASK; | |
1474 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
1475 | ||
1476 | block_it = true; | |
1477 | } else { | |
1478 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
1479 | ||
1480 | r = signal_is_blocked(sig); | |
1481 | if (r < 0) | |
1482 | return r; | |
1483 | if (r == 0) | |
1484 | return -EBUSY; | |
1485 | ||
1486 | block_it = false; | |
1487 | } | |
1488 | ||
59bc1fd7 LP |
1489 | if (!callback) |
1490 | callback = signal_exit_callback; | |
1491 | ||
fd38203a LP |
1492 | if (!e->signal_sources) { |
1493 | e->signal_sources = new0(sd_event_source*, _NSIG); | |
1494 | if (!e->signal_sources) | |
1495 | return -ENOMEM; | |
1496 | } else if (e->signal_sources[sig]) | |
1497 | return -EBUSY; | |
1498 | ||
a71fe8b8 | 1499 | s = source_new(e, !ret, SOURCE_SIGNAL); |
fd38203a LP |
1500 | if (!s) |
1501 | return -ENOMEM; | |
1502 | ||
1503 | s->signal.sig = sig; | |
1504 | s->signal.callback = callback; | |
1505 | s->userdata = userdata; | |
baf76283 | 1506 | s->enabled = SD_EVENT_ON; |
fd38203a LP |
1507 | |
1508 | e->signal_sources[sig] = s; | |
fd38203a | 1509 | |
897448bd LP |
1510 | if (block_it) { |
1511 | sigset_t old_ss; | |
1512 | ||
1513 | if (sigemptyset(&new_ss) < 0) | |
1514 | return -errno; | |
1515 | ||
1516 | if (sigaddset(&new_ss, sig) < 0) | |
1517 | return -errno; | |
1518 | ||
1519 | r = pthread_sigmask(SIG_BLOCK, &new_ss, &old_ss); | |
1520 | if (r != 0) | |
1521 | return -r; | |
1522 | ||
1523 | r = sigismember(&old_ss, sig); | |
1524 | if (r < 0) | |
1525 | return -errno; | |
1526 | ||
1527 | s->signal.unblock = !r; | |
1528 | } else | |
1529 | s->signal.unblock = false; | |
1530 | ||
9da4cb2b | 1531 | r = event_make_signal_data(e, sig, &d); |
897448bd LP |
1532 | if (r < 0) { |
1533 | if (s->signal.unblock) | |
1534 | (void) pthread_sigmask(SIG_UNBLOCK, &new_ss, NULL); | |
1535 | ||
9da4cb2b | 1536 | return r; |
897448bd | 1537 | } |
fd38203a | 1538 | |
f1f00dbb LP |
1539 | /* Use the signal name as description for the event source by default */ |
1540 | (void) sd_event_source_set_description(s, signal_to_string(sig)); | |
1541 | ||
a71fe8b8 LP |
1542 | if (ret) |
1543 | *ret = s; | |
ec766a51 | 1544 | TAKE_PTR(s); |
a71fe8b8 | 1545 | |
fd38203a LP |
1546 | return 0; |
1547 | } | |
1548 | ||
b9350e70 LP |
1549 | static int child_exit_callback(sd_event_source *s, const siginfo_t *si, void *userdata) { |
1550 | assert(s); | |
1551 | ||
1552 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1553 | } | |
1554 | ||
f8f3f926 LP |
1555 | static bool shall_use_pidfd(void) { |
1556 | /* Mostly relevant for debugging, i.e. this is used in test-event.c to test the event loop once with and once without pidfd */ | |
1557 | return getenv_bool_secure("SYSTEMD_PIDFD") != 0; | |
1558 | } | |
1559 | ||
f7262a9f | 1560 | _public_ int sd_event_add_child( |
305f78bf | 1561 | sd_event *e, |
151b9b96 | 1562 | sd_event_source **ret, |
305f78bf LP |
1563 | pid_t pid, |
1564 | int options, | |
718db961 | 1565 | sd_event_child_handler_t callback, |
151b9b96 | 1566 | void *userdata) { |
305f78bf | 1567 | |
ec766a51 | 1568 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1569 | int r; |
1570 | ||
305f78bf | 1571 | assert_return(e, -EINVAL); |
b937d761 | 1572 | assert_return(e = event_resolve(e), -ENOPKG); |
305f78bf LP |
1573 | assert_return(pid > 1, -EINVAL); |
1574 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1575 | assert_return(options != 0, -EINVAL); | |
da7e457c | 1576 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1577 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1578 | |
b9350e70 LP |
1579 | if (!callback) |
1580 | callback = child_exit_callback; | |
1581 | ||
b6d5481b | 1582 | if (e->n_online_child_sources == 0) { |
ee880b37 LP |
1583 | /* Caller must block SIGCHLD before using us to watch children, even if pidfd is available, |
1584 | * for compatibility with pre-pidfd and because we don't want the reap the child processes | |
1585 | * ourselves, i.e. call waitid(), and don't want Linux' default internal logic for that to | |
1586 | * take effect. | |
1587 | * | |
1588 | * (As an optimization we only do this check on the first child event source created.) */ | |
1589 | r = signal_is_blocked(SIGCHLD); | |
1590 | if (r < 0) | |
1591 | return r; | |
1592 | if (r == 0) | |
1593 | return -EBUSY; | |
1594 | } | |
1595 | ||
d5099efc | 1596 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
fd38203a LP |
1597 | if (r < 0) |
1598 | return r; | |
1599 | ||
4a0b58c4 | 1600 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) |
fd38203a LP |
1601 | return -EBUSY; |
1602 | ||
a71fe8b8 | 1603 | s = source_new(e, !ret, SOURCE_CHILD); |
fd38203a LP |
1604 | if (!s) |
1605 | return -ENOMEM; | |
1606 | ||
f8f3f926 | 1607 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1608 | s->child.options = options; |
1609 | s->child.callback = callback; | |
1610 | s->userdata = userdata; | |
baf76283 | 1611 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1612 | |
f8f3f926 LP |
1613 | /* We always take a pidfd here if we can, even if we wait for anything else than WEXITED, so that we |
1614 | * pin the PID, and make regular waitid() handling race-free. */ | |
1615 | ||
1616 | if (shall_use_pidfd()) { | |
54988a27 | 1617 | s->child.pidfd = pidfd_open(pid, 0); |
f8f3f926 LP |
1618 | if (s->child.pidfd < 0) { |
1619 | /* Propagate errors unless the syscall is not supported or blocked */ | |
1620 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
1621 | return -errno; | |
1622 | } else | |
1623 | s->child.pidfd_owned = true; /* If we allocate the pidfd we own it by default */ | |
1624 | } else | |
254d1313 | 1625 | s->child.pidfd = -EBADF; |
f8f3f926 | 1626 | |
f8f3f926 LP |
1627 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1628 | /* We have a pidfd and we only want to watch for exit */ | |
f8f3f926 | 1629 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1630 | if (r < 0) |
f8f3f926 | 1631 | return r; |
ac9f2640 | 1632 | |
f8f3f926 LP |
1633 | } else { |
1634 | /* We have no pidfd or we shall wait for some other event than WEXITED */ | |
f8f3f926 | 1635 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1636 | if (r < 0) |
f8f3f926 | 1637 | return r; |
f8f3f926 LP |
1638 | |
1639 | e->need_process_child = true; | |
1640 | } | |
c2ba3ad6 | 1641 | |
54988a27 YW |
1642 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); |
1643 | if (r < 0) | |
1644 | return r; | |
1645 | ||
1646 | /* These must be done after everything succeeds. */ | |
1647 | s->child.pid = pid; | |
b6d5481b | 1648 | e->n_online_child_sources++; |
ac9f2640 | 1649 | |
a71fe8b8 LP |
1650 | if (ret) |
1651 | *ret = s; | |
ec766a51 | 1652 | TAKE_PTR(s); |
f8f3f926 LP |
1653 | return 0; |
1654 | } | |
1655 | ||
1656 | _public_ int sd_event_add_child_pidfd( | |
1657 | sd_event *e, | |
1658 | sd_event_source **ret, | |
1659 | int pidfd, | |
1660 | int options, | |
1661 | sd_event_child_handler_t callback, | |
1662 | void *userdata) { | |
1663 | ||
1664 | ||
1665 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
1666 | pid_t pid; | |
1667 | int r; | |
1668 | ||
1669 | assert_return(e, -EINVAL); | |
1670 | assert_return(e = event_resolve(e), -ENOPKG); | |
1671 | assert_return(pidfd >= 0, -EBADF); | |
1672 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1673 | assert_return(options != 0, -EINVAL); | |
f8f3f926 LP |
1674 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1675 | assert_return(!event_pid_changed(e), -ECHILD); | |
1676 | ||
b9350e70 LP |
1677 | if (!callback) |
1678 | callback = child_exit_callback; | |
1679 | ||
b6d5481b | 1680 | if (e->n_online_child_sources == 0) { |
ee880b37 LP |
1681 | r = signal_is_blocked(SIGCHLD); |
1682 | if (r < 0) | |
1683 | return r; | |
1684 | if (r == 0) | |
1685 | return -EBUSY; | |
1686 | } | |
1687 | ||
f8f3f926 LP |
1688 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
1689 | if (r < 0) | |
1690 | return r; | |
1691 | ||
1692 | r = pidfd_get_pid(pidfd, &pid); | |
1693 | if (r < 0) | |
1694 | return r; | |
1695 | ||
1696 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) | |
1697 | return -EBUSY; | |
1698 | ||
1699 | s = source_new(e, !ret, SOURCE_CHILD); | |
1700 | if (!s) | |
1701 | return -ENOMEM; | |
1702 | ||
1703 | s->wakeup = WAKEUP_EVENT_SOURCE; | |
1704 | s->child.pidfd = pidfd; | |
1705 | s->child.pid = pid; | |
1706 | s->child.options = options; | |
1707 | s->child.callback = callback; | |
1708 | 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) */ | |
1709 | s->userdata = userdata; | |
1710 | s->enabled = SD_EVENT_ONESHOT; | |
1711 | ||
1712 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); | |
1713 | if (r < 0) | |
1714 | return r; | |
1715 | ||
f8f3f926 LP |
1716 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1717 | /* We only want to watch for WEXITED */ | |
f8f3f926 | 1718 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1719 | if (r < 0) |
f8f3f926 | 1720 | return r; |
f8f3f926 LP |
1721 | } else { |
1722 | /* We shall wait for some other event than WEXITED */ | |
f8f3f926 | 1723 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1724 | if (r < 0) |
f8f3f926 | 1725 | return r; |
a71fe8b8 | 1726 | |
f8f3f926 LP |
1727 | e->need_process_child = true; |
1728 | } | |
1729 | ||
b6d5481b | 1730 | e->n_online_child_sources++; |
ac9f2640 | 1731 | |
f8f3f926 LP |
1732 | if (ret) |
1733 | *ret = s; | |
f8f3f926 | 1734 | TAKE_PTR(s); |
fd38203a LP |
1735 | return 0; |
1736 | } | |
1737 | ||
b9350e70 LP |
1738 | static int generic_exit_callback(sd_event_source *s, void *userdata) { |
1739 | assert(s); | |
1740 | ||
1741 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1742 | } | |
1743 | ||
f7262a9f | 1744 | _public_ int sd_event_add_defer( |
305f78bf | 1745 | sd_event *e, |
151b9b96 | 1746 | sd_event_source **ret, |
718db961 | 1747 | sd_event_handler_t callback, |
151b9b96 | 1748 | void *userdata) { |
305f78bf | 1749 | |
ec766a51 | 1750 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1751 | int r; |
1752 | ||
305f78bf | 1753 | assert_return(e, -EINVAL); |
b937d761 | 1754 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1755 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1756 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1757 | |
b9350e70 LP |
1758 | if (!callback) |
1759 | callback = generic_exit_callback; | |
1760 | ||
a71fe8b8 | 1761 | s = source_new(e, !ret, SOURCE_DEFER); |
fd38203a LP |
1762 | if (!s) |
1763 | return -ENOMEM; | |
1764 | ||
1765 | s->defer.callback = callback; | |
1766 | s->userdata = userdata; | |
baf76283 | 1767 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a LP |
1768 | |
1769 | r = source_set_pending(s, true); | |
ec766a51 | 1770 | if (r < 0) |
fd38203a | 1771 | return r; |
fd38203a | 1772 | |
a71fe8b8 LP |
1773 | if (ret) |
1774 | *ret = s; | |
ec766a51 | 1775 | TAKE_PTR(s); |
a71fe8b8 | 1776 | |
fd38203a LP |
1777 | return 0; |
1778 | } | |
1779 | ||
6e9feda3 LP |
1780 | _public_ int sd_event_add_post( |
1781 | sd_event *e, | |
1782 | sd_event_source **ret, | |
1783 | sd_event_handler_t callback, | |
1784 | void *userdata) { | |
1785 | ||
ec766a51 | 1786 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6e9feda3 LP |
1787 | int r; |
1788 | ||
1789 | assert_return(e, -EINVAL); | |
b937d761 | 1790 | assert_return(e = event_resolve(e), -ENOPKG); |
6e9feda3 LP |
1791 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1792 | assert_return(!event_pid_changed(e), -ECHILD); | |
1793 | ||
b9350e70 LP |
1794 | if (!callback) |
1795 | callback = generic_exit_callback; | |
1796 | ||
a71fe8b8 | 1797 | s = source_new(e, !ret, SOURCE_POST); |
6e9feda3 LP |
1798 | if (!s) |
1799 | return -ENOMEM; | |
1800 | ||
1801 | s->post.callback = callback; | |
1802 | s->userdata = userdata; | |
1803 | s->enabled = SD_EVENT_ON; | |
1804 | ||
de7fef4b | 1805 | r = set_ensure_put(&e->post_sources, NULL, s); |
ec766a51 | 1806 | if (r < 0) |
6e9feda3 | 1807 | return r; |
de7fef4b | 1808 | assert(r > 0); |
6e9feda3 | 1809 | |
a71fe8b8 LP |
1810 | if (ret) |
1811 | *ret = s; | |
ec766a51 | 1812 | TAKE_PTR(s); |
a71fe8b8 | 1813 | |
6e9feda3 LP |
1814 | return 0; |
1815 | } | |
1816 | ||
6203e07a | 1817 | _public_ int sd_event_add_exit( |
305f78bf | 1818 | sd_event *e, |
151b9b96 | 1819 | sd_event_source **ret, |
718db961 | 1820 | sd_event_handler_t callback, |
151b9b96 | 1821 | void *userdata) { |
305f78bf | 1822 | |
ec766a51 | 1823 | _cleanup_(source_freep) sd_event_source *s = NULL; |
da7e457c LP |
1824 | int r; |
1825 | ||
1826 | assert_return(e, -EINVAL); | |
b937d761 | 1827 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
1828 | assert_return(callback, -EINVAL); |
1829 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
1830 | assert_return(!event_pid_changed(e), -ECHILD); | |
1831 | ||
c983e776 EV |
1832 | r = prioq_ensure_allocated(&e->exit, exit_prioq_compare); |
1833 | if (r < 0) | |
1834 | return r; | |
da7e457c | 1835 | |
a71fe8b8 | 1836 | s = source_new(e, !ret, SOURCE_EXIT); |
fd38203a | 1837 | if (!s) |
da7e457c | 1838 | return -ENOMEM; |
fd38203a | 1839 | |
6203e07a | 1840 | s->exit.callback = callback; |
da7e457c | 1841 | s->userdata = userdata; |
6203e07a | 1842 | s->exit.prioq_index = PRIOQ_IDX_NULL; |
baf76283 | 1843 | s->enabled = SD_EVENT_ONESHOT; |
da7e457c | 1844 | |
6203e07a | 1845 | r = prioq_put(s->event->exit, s, &s->exit.prioq_index); |
ec766a51 | 1846 | if (r < 0) |
da7e457c | 1847 | return r; |
da7e457c | 1848 | |
a71fe8b8 LP |
1849 | if (ret) |
1850 | *ret = s; | |
ec766a51 | 1851 | TAKE_PTR(s); |
a71fe8b8 | 1852 | |
da7e457c LP |
1853 | return 0; |
1854 | } | |
1855 | ||
158fe190 LP |
1856 | int sd_event_trim_memory(void) { |
1857 | int r; | |
1858 | ||
1859 | /* A default implementation of a memory pressure callback. Simply releases our own allocation caches | |
1860 | * and glibc's. This is automatically used when people call sd_event_add_memory_pressure() with a | |
1861 | * NULL callback parameter. */ | |
1862 | ||
1863 | log_debug("Memory pressure event, trimming malloc() memory."); | |
1864 | ||
1865 | #if HAVE_GENERIC_MALLINFO | |
1866 | generic_mallinfo before_mallinfo = generic_mallinfo_get(); | |
1867 | #endif | |
1868 | ||
1869 | usec_t before_timestamp = now(CLOCK_MONOTONIC); | |
1870 | hashmap_trim_pools(); | |
1871 | r = malloc_trim(0); | |
1872 | usec_t after_timestamp = now(CLOCK_MONOTONIC); | |
1873 | ||
1874 | if (r > 0) | |
1875 | log_debug("Successfully trimmed some memory."); | |
1876 | else | |
1877 | log_debug("Couldn't trim any memory."); | |
1878 | ||
1879 | usec_t period = after_timestamp - before_timestamp; | |
1880 | ||
1881 | #if HAVE_GENERIC_MALLINFO | |
1882 | generic_mallinfo after_mallinfo = generic_mallinfo_get(); | |
1883 | size_t l = LESS_BY((size_t) before_mallinfo.hblkhd, (size_t) after_mallinfo.hblkhd) + | |
1884 | LESS_BY((size_t) before_mallinfo.arena, (size_t) after_mallinfo.arena); | |
1885 | log_struct(LOG_DEBUG, | |
1886 | LOG_MESSAGE("Memory trimming took %s, returned %s to OS.", | |
1887 | FORMAT_TIMESPAN(period, 0), | |
1888 | FORMAT_BYTES(l)), | |
1889 | "MESSAGE_ID=" SD_MESSAGE_MEMORY_TRIM_STR, | |
1890 | "TRIMMED_BYTES=%zu", l, | |
1891 | "TRIMMED_USEC=" USEC_FMT, period); | |
1892 | #else | |
1893 | log_struct(LOG_DEBUG, | |
1894 | LOG_MESSAGE("Memory trimming took %s.", | |
1895 | FORMAT_TIMESPAN(period, 0)), | |
1896 | "MESSAGE_ID=" SD_MESSAGE_MEMORY_TRIM_STR, | |
1897 | "TRIMMED_USEC=" USEC_FMT, period); | |
1898 | #endif | |
1899 | ||
1900 | return 0; | |
1901 | } | |
1902 | ||
1903 | static int memory_pressure_callback(sd_event_source *s, void *userdata) { | |
1904 | assert(s); | |
1905 | ||
1906 | sd_event_trim_memory(); | |
1907 | return 0; | |
1908 | } | |
1909 | ||
1910 | _public_ int sd_event_add_memory_pressure( | |
1911 | sd_event *e, | |
1912 | sd_event_source **ret, | |
1913 | sd_event_handler_t callback, | |
1914 | void *userdata) { | |
1915 | ||
1916 | _cleanup_free_ char *w = NULL; | |
1917 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
1918 | _cleanup_close_ int path_fd = -1, fd = -1; | |
1919 | _cleanup_free_ void *write_buffer = NULL; | |
40c5d5d2 | 1920 | const char *watch, *watch_fallback = NULL, *env; |
158fe190 LP |
1921 | size_t write_buffer_size = 0; |
1922 | struct stat st; | |
1923 | uint32_t events; | |
1924 | bool locked; | |
1925 | int r; | |
1926 | ||
1927 | assert_return(e, -EINVAL); | |
1928 | assert_return(e = event_resolve(e), -ENOPKG); | |
1929 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
1930 | assert_return(!event_pid_changed(e), -ECHILD); | |
1931 | ||
1932 | if (!callback) | |
1933 | callback = memory_pressure_callback; | |
1934 | ||
1935 | s = source_new(e, !ret, SOURCE_MEMORY_PRESSURE); | |
1936 | if (!s) | |
1937 | return -ENOMEM; | |
1938 | ||
1939 | s->wakeup = WAKEUP_EVENT_SOURCE; | |
1940 | s->memory_pressure.callback = callback; | |
1941 | s->userdata = userdata; | |
1942 | s->enabled = SD_EVENT_ON; | |
1943 | s->memory_pressure.fd = -EBADF; | |
1944 | ||
1945 | env = secure_getenv("MEMORY_PRESSURE_WATCH"); | |
1946 | if (env) { | |
1947 | if (isempty(env) || path_equal(env, "/dev/null")) | |
1948 | return log_debug_errno(SYNTHETIC_ERRNO(EHOSTDOWN), | |
1949 | "Memory pressure logic is explicitly disabled via $MEMORY_PRESSURE_WATCH."); | |
1950 | ||
1951 | if (!path_is_absolute(env) || !path_is_normalized(env)) | |
1952 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
1953 | "$MEMORY_PRESSURE_WATCH set to invalid path: %s", env); | |
1954 | ||
1955 | watch = env; | |
1956 | ||
1957 | env = secure_getenv("MEMORY_PRESSURE_WRITE"); | |
1958 | if (env) { | |
1959 | r = unbase64mem(env, SIZE_MAX, &write_buffer, &write_buffer_size); | |
1960 | if (r < 0) | |
1961 | return r; | |
1962 | } | |
1963 | ||
1964 | locked = true; | |
1965 | } else { | |
1966 | ||
1967 | r = is_pressure_supported(); | |
1968 | if (r < 0) | |
1969 | return r; | |
1970 | if (r == 0) | |
1971 | return -EOPNOTSUPP; | |
1972 | ||
1973 | /* By default we want to watch memory pressure on the local cgroup, but we'll fall back on | |
1974 | * the system wide pressure if for some reason we cannot (which could be: memory controller | |
1975 | * not delegated to us, or PSI simply not available in the kernel). On legacy cgroupv1 we'll | |
1976 | * only use the system-wide logic. */ | |
1977 | r = cg_all_unified(); | |
1978 | if (r < 0) | |
1979 | return r; | |
1980 | if (r == 0) | |
1981 | watch = "/proc/pressure/memory"; | |
1982 | else { | |
1983 | _cleanup_free_ char *cg = NULL; | |
1984 | ||
1985 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &cg); | |
1986 | if (r < 0) | |
1987 | return r; | |
1988 | ||
1989 | w = path_join("/sys/fs/cgroup", cg, "memory.pressure"); | |
1990 | if (!w) | |
1991 | return -ENOMEM; | |
1992 | ||
1993 | watch = w; | |
1994 | watch_fallback = "/proc/pressure/memory"; | |
1995 | } | |
1996 | ||
1997 | /* Android uses three levels in its userspace low memory killer logic: | |
1998 | * some 70000 1000000 | |
1999 | * some 100000 1000000 | |
2000 | * full 70000 1000000 | |
2001 | * | |
2002 | * GNOME's low memory monitor uses: | |
2003 | * some 70000 1000000 | |
2004 | * some 100000 1000000 | |
2005 | * full 100000 1000000 | |
2006 | * | |
2007 | * We'll default to the middle level that both agree on */ | |
2008 | if (asprintf((char**) &write_buffer, | |
2009 | "%s " USEC_FMT " " USEC_FMT, | |
2010 | MEMORY_PRESSURE_DEFAULT_TYPE, | |
2011 | MEMORY_PRESSURE_DEFAULT_THRESHOLD_USEC, | |
2012 | MEMORY_PRESSURE_DEFAULT_WINDOW_USEC) < 0) | |
2013 | return -ENOMEM; | |
2014 | ||
2015 | write_buffer_size = strlen(write_buffer) + 1; | |
2016 | locked = false; | |
2017 | } | |
2018 | ||
2019 | path_fd = open(watch, O_PATH|O_CLOEXEC); | |
2020 | if (path_fd < 0) { | |
2021 | if (errno != ENOENT) | |
2022 | return -errno; | |
2023 | ||
2024 | /* We got ENOENT. Three options now: try the fallback if we have one, or return the error as | |
2025 | * is (if based on user/env config), or return -EOPNOTSUPP (because we picked the path, and | |
2026 | * the PSI service apparently is not supported) */ | |
2027 | if (!watch_fallback) | |
2028 | return locked ? -ENOENT : -EOPNOTSUPP; | |
2029 | ||
2030 | path_fd = open(watch_fallback, O_PATH|O_CLOEXEC); | |
63b1e67e YW |
2031 | if (path_fd < 0) { |
2032 | if (errno == ENOENT) /* PSI is not available in the kernel even under the fallback path? */ | |
2033 | return -EOPNOTSUPP; | |
158fe190 | 2034 | return -errno; |
63b1e67e | 2035 | } |
158fe190 LP |
2036 | } |
2037 | ||
2038 | if (fstat(path_fd, &st) < 0) | |
2039 | return -errno; | |
2040 | ||
2041 | if (S_ISSOCK(st.st_mode)) { | |
2042 | fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); | |
2043 | if (fd < 0) | |
2044 | return -errno; | |
2045 | ||
2046 | r = connect_unix_path(fd, path_fd, NULL); | |
2047 | if (r < 0) | |
2048 | return r; | |
2049 | ||
2050 | events = EPOLLIN; | |
2051 | ||
2052 | } else if (S_ISREG(st.st_mode) || S_ISFIFO(st.st_mode) || S_ISCHR(st.st_mode)) { | |
2053 | fd = fd_reopen(path_fd, (write_buffer_size > 0 ? O_RDWR : O_RDONLY) |O_CLOEXEC|O_NONBLOCK|O_NOCTTY); | |
2054 | if (fd < 0) | |
2055 | return fd; | |
2056 | ||
2057 | if (S_ISREG(st.st_mode)) { | |
2058 | struct statfs sfs; | |
2059 | ||
2060 | /* If this is a regular file validate this is a procfs or cgroupfs file, where we look for EPOLLPRI */ | |
2061 | ||
2062 | if (fstatfs(fd, &sfs) < 0) | |
2063 | return -errno; | |
2064 | ||
2065 | if (!is_fs_type(&sfs, PROC_SUPER_MAGIC) && | |
2066 | !is_fs_type(&sfs, CGROUP2_SUPER_MAGIC)) | |
2067 | return -ENOTTY; | |
2068 | ||
2069 | events = EPOLLPRI; | |
2070 | } else | |
2071 | /* For fifos and char devices just watch for EPOLLIN */ | |
2072 | events = EPOLLIN; | |
2073 | ||
2074 | } else if (S_ISDIR(st.st_mode)) | |
2075 | return -EISDIR; | |
2076 | else | |
2077 | return -EBADF; | |
2078 | ||
2079 | s->memory_pressure.fd = TAKE_FD(fd); | |
2080 | s->memory_pressure.write_buffer = TAKE_PTR(write_buffer); | |
2081 | s->memory_pressure.write_buffer_size = write_buffer_size; | |
2082 | s->memory_pressure.events = events; | |
2083 | s->memory_pressure.locked = locked; | |
2084 | ||
2085 | /* So here's the thing: if we are talking to PSI we need to write the watch string before adding the | |
2086 | * fd to epoll (if we ignore this, then the watch won't work). Hence we'll not actually register the | |
2087 | * fd with the epoll right-away. Instead, we just add the event source to a list of memory pressure | |
2088 | * event sources on which writes must be executed before the first event loop iteration is | |
2089 | * executed. (We could also write the data here, right away, but we want to give the caller the | |
2090 | * freedom to call sd_event_source_set_memory_pressure_type() and | |
2091 | * sd_event_source_set_memory_pressure_rate() before we write it. */ | |
2092 | ||
2093 | if (s->memory_pressure.write_buffer_size > 0) | |
2094 | source_memory_pressure_add_to_write_list(s); | |
2095 | else { | |
2096 | r = source_memory_pressure_register(s, s->enabled); | |
2097 | if (r < 0) | |
2098 | return r; | |
2099 | } | |
2100 | ||
2101 | if (ret) | |
2102 | *ret = s; | |
2103 | TAKE_PTR(s); | |
2104 | ||
2105 | return 0; | |
2106 | } | |
2107 | ||
97ef5391 LP |
2108 | static void event_free_inotify_data(sd_event *e, struct inotify_data *d) { |
2109 | assert(e); | |
2110 | ||
2111 | if (!d) | |
2112 | return; | |
2113 | ||
2114 | assert(hashmap_isempty(d->inodes)); | |
2115 | assert(hashmap_isempty(d->wd)); | |
2116 | ||
2117 | if (d->buffer_filled > 0) | |
0601b958 | 2118 | LIST_REMOVE(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
2119 | |
2120 | hashmap_free(d->inodes); | |
2121 | hashmap_free(d->wd); | |
2122 | ||
2123 | assert_se(hashmap_remove(e->inotify_data, &d->priority) == d); | |
2124 | ||
2125 | if (d->fd >= 0) { | |
fbae5090 YW |
2126 | if (!event_pid_changed(e) && |
2127 | epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, d->fd, NULL) < 0) | |
97ef5391 LP |
2128 | log_debug_errno(errno, "Failed to remove inotify fd from epoll, ignoring: %m"); |
2129 | ||
2130 | safe_close(d->fd); | |
2131 | } | |
2132 | free(d); | |
2133 | } | |
2134 | ||
2135 | static int event_make_inotify_data( | |
2136 | sd_event *e, | |
2137 | int64_t priority, | |
2138 | struct inotify_data **ret) { | |
2139 | ||
254d1313 | 2140 | _cleanup_close_ int fd = -EBADF; |
97ef5391 | 2141 | struct inotify_data *d; |
97ef5391 LP |
2142 | int r; |
2143 | ||
2144 | assert(e); | |
2145 | ||
2146 | d = hashmap_get(e->inotify_data, &priority); | |
2147 | if (d) { | |
2148 | if (ret) | |
2149 | *ret = d; | |
2150 | return 0; | |
2151 | } | |
2152 | ||
2153 | fd = inotify_init1(IN_NONBLOCK|O_CLOEXEC); | |
2154 | if (fd < 0) | |
2155 | return -errno; | |
2156 | ||
2157 | fd = fd_move_above_stdio(fd); | |
2158 | ||
97ef5391 LP |
2159 | d = new(struct inotify_data, 1); |
2160 | if (!d) | |
2161 | return -ENOMEM; | |
2162 | ||
2163 | *d = (struct inotify_data) { | |
2164 | .wakeup = WAKEUP_INOTIFY_DATA, | |
2165 | .fd = TAKE_FD(fd), | |
2166 | .priority = priority, | |
2167 | }; | |
2168 | ||
c2484a75 | 2169 | r = hashmap_ensure_put(&e->inotify_data, &uint64_hash_ops, &d->priority, d); |
97ef5391 LP |
2170 | if (r < 0) { |
2171 | d->fd = safe_close(d->fd); | |
2172 | free(d); | |
2173 | return r; | |
2174 | } | |
2175 | ||
1eac7948 | 2176 | struct epoll_event ev = { |
97ef5391 LP |
2177 | .events = EPOLLIN, |
2178 | .data.ptr = d, | |
2179 | }; | |
2180 | ||
2181 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { | |
2182 | r = -errno; | |
2183 | d->fd = safe_close(d->fd); /* let's close this ourselves, as event_free_inotify_data() would otherwise | |
2184 | * remove the fd from the epoll first, which we don't want as we couldn't | |
2185 | * add it in the first place. */ | |
2186 | event_free_inotify_data(e, d); | |
2187 | return r; | |
2188 | } | |
2189 | ||
2190 | if (ret) | |
2191 | *ret = d; | |
2192 | ||
2193 | return 1; | |
2194 | } | |
2195 | ||
7a08d314 | 2196 | static int inode_data_compare(const struct inode_data *x, const struct inode_data *y) { |
90c88092 | 2197 | int r; |
97ef5391 LP |
2198 | |
2199 | assert(x); | |
2200 | assert(y); | |
2201 | ||
90c88092 YW |
2202 | r = CMP(x->dev, y->dev); |
2203 | if (r != 0) | |
2204 | return r; | |
97ef5391 | 2205 | |
6dd91b36 | 2206 | return CMP(x->ino, y->ino); |
97ef5391 LP |
2207 | } |
2208 | ||
7a08d314 YW |
2209 | static void inode_data_hash_func(const struct inode_data *d, struct siphash *state) { |
2210 | assert(d); | |
97ef5391 LP |
2211 | |
2212 | siphash24_compress(&d->dev, sizeof(d->dev), state); | |
2213 | siphash24_compress(&d->ino, sizeof(d->ino), state); | |
2214 | } | |
2215 | ||
7a08d314 | 2216 | DEFINE_PRIVATE_HASH_OPS(inode_data_hash_ops, struct inode_data, inode_data_hash_func, inode_data_compare); |
97ef5391 LP |
2217 | |
2218 | static void event_free_inode_data( | |
2219 | sd_event *e, | |
2220 | struct inode_data *d) { | |
2221 | ||
2222 | assert(e); | |
2223 | ||
2224 | if (!d) | |
2225 | return; | |
2226 | ||
64903d18 | 2227 | assert(!d->event_sources); |
97ef5391 LP |
2228 | |
2229 | if (d->fd >= 0) { | |
ed828563 | 2230 | LIST_REMOVE(to_close, e->inode_data_to_close_list, d); |
97ef5391 LP |
2231 | safe_close(d->fd); |
2232 | } | |
2233 | ||
2234 | if (d->inotify_data) { | |
2235 | ||
2236 | if (d->wd >= 0) { | |
fbae5090 | 2237 | if (d->inotify_data->fd >= 0 && !event_pid_changed(e)) { |
97ef5391 LP |
2238 | /* So here's a problem. At the time this runs the watch descriptor might already be |
2239 | * invalidated, because an IN_IGNORED event might be queued right the moment we enter | |
2240 | * the syscall. Hence, whenever we get EINVAL, ignore it entirely, since it's a very | |
2241 | * likely case to happen. */ | |
2242 | ||
2243 | if (inotify_rm_watch(d->inotify_data->fd, d->wd) < 0 && errno != EINVAL) | |
2244 | log_debug_errno(errno, "Failed to remove watch descriptor %i from inotify, ignoring: %m", d->wd); | |
2245 | } | |
2246 | ||
2247 | assert_se(hashmap_remove(d->inotify_data->wd, INT_TO_PTR(d->wd)) == d); | |
2248 | } | |
2249 | ||
2250 | assert_se(hashmap_remove(d->inotify_data->inodes, d) == d); | |
2251 | } | |
2252 | ||
2253 | free(d); | |
2254 | } | |
2255 | ||
53baf2ef LP |
2256 | static void event_gc_inotify_data( |
2257 | sd_event *e, | |
2258 | struct inotify_data *d) { | |
2259 | ||
2260 | assert(e); | |
2261 | ||
2262 | /* GCs the inotify data object if we don't need it anymore. That's the case if we don't want to watch | |
2263 | * any inode with it anymore, which in turn happens if no event source of this priority is interested | |
2264 | * in any inode any longer. That said, we maintain an extra busy counter: if non-zero we'll delay GC | |
2265 | * (under the expectation that the GC is called again once the counter is decremented). */ | |
2266 | ||
2267 | if (!d) | |
2268 | return; | |
2269 | ||
2270 | if (!hashmap_isempty(d->inodes)) | |
2271 | return; | |
2272 | ||
2273 | if (d->n_busy > 0) | |
2274 | return; | |
2275 | ||
2276 | event_free_inotify_data(e, d); | |
2277 | } | |
2278 | ||
97ef5391 LP |
2279 | static void event_gc_inode_data( |
2280 | sd_event *e, | |
2281 | struct inode_data *d) { | |
2282 | ||
2283 | struct inotify_data *inotify_data; | |
2284 | ||
2285 | assert(e); | |
2286 | ||
2287 | if (!d) | |
2288 | return; | |
2289 | ||
64903d18 | 2290 | if (d->event_sources) |
97ef5391 LP |
2291 | return; |
2292 | ||
2293 | inotify_data = d->inotify_data; | |
2294 | event_free_inode_data(e, d); | |
2295 | ||
53baf2ef | 2296 | event_gc_inotify_data(e, inotify_data); |
97ef5391 LP |
2297 | } |
2298 | ||
2299 | static int event_make_inode_data( | |
2300 | sd_event *e, | |
2301 | struct inotify_data *inotify_data, | |
2302 | dev_t dev, | |
2303 | ino_t ino, | |
2304 | struct inode_data **ret) { | |
2305 | ||
2306 | struct inode_data *d, key; | |
2307 | int r; | |
2308 | ||
2309 | assert(e); | |
2310 | assert(inotify_data); | |
2311 | ||
2312 | key = (struct inode_data) { | |
2313 | .ino = ino, | |
2314 | .dev = dev, | |
2315 | }; | |
2316 | ||
2317 | d = hashmap_get(inotify_data->inodes, &key); | |
2318 | if (d) { | |
2319 | if (ret) | |
2320 | *ret = d; | |
2321 | ||
2322 | return 0; | |
2323 | } | |
2324 | ||
2325 | r = hashmap_ensure_allocated(&inotify_data->inodes, &inode_data_hash_ops); | |
2326 | if (r < 0) | |
2327 | return r; | |
2328 | ||
2329 | d = new(struct inode_data, 1); | |
2330 | if (!d) | |
2331 | return -ENOMEM; | |
2332 | ||
2333 | *d = (struct inode_data) { | |
2334 | .dev = dev, | |
2335 | .ino = ino, | |
2336 | .wd = -1, | |
254d1313 | 2337 | .fd = -EBADF, |
97ef5391 LP |
2338 | .inotify_data = inotify_data, |
2339 | }; | |
2340 | ||
2341 | r = hashmap_put(inotify_data->inodes, d, d); | |
2342 | if (r < 0) { | |
2343 | free(d); | |
2344 | return r; | |
2345 | } | |
2346 | ||
2347 | if (ret) | |
2348 | *ret = d; | |
2349 | ||
2350 | return 1; | |
2351 | } | |
2352 | ||
2353 | static uint32_t inode_data_determine_mask(struct inode_data *d) { | |
2354 | bool excl_unlink = true; | |
2355 | uint32_t combined = 0; | |
97ef5391 LP |
2356 | |
2357 | assert(d); | |
2358 | ||
2359 | /* Combines the watch masks of all event sources watching this inode. We generally just OR them together, but | |
2360 | * the IN_EXCL_UNLINK flag is ANDed instead. | |
2361 | * | |
2362 | * Note that we add all sources to the mask here, regardless whether enabled, disabled or oneshot. That's | |
2363 | * because we cannot change the mask anymore after the event source was created once, since the kernel has no | |
f21f31b2 | 2364 | * API for that. Hence we need to subscribe to the maximum mask we ever might be interested in, and suppress |
97ef5391 LP |
2365 | * events we don't care for client-side. */ |
2366 | ||
2367 | LIST_FOREACH(inotify.by_inode_data, s, d->event_sources) { | |
2368 | ||
2369 | if ((s->inotify.mask & IN_EXCL_UNLINK) == 0) | |
2370 | excl_unlink = false; | |
2371 | ||
2372 | combined |= s->inotify.mask; | |
2373 | } | |
2374 | ||
2375 | return (combined & ~(IN_ONESHOT|IN_DONT_FOLLOW|IN_ONLYDIR|IN_EXCL_UNLINK)) | (excl_unlink ? IN_EXCL_UNLINK : 0); | |
2376 | } | |
2377 | ||
2378 | static int inode_data_realize_watch(sd_event *e, struct inode_data *d) { | |
2379 | uint32_t combined_mask; | |
2380 | int wd, r; | |
2381 | ||
2382 | assert(d); | |
2383 | assert(d->fd >= 0); | |
2384 | ||
2385 | combined_mask = inode_data_determine_mask(d); | |
2386 | ||
2387 | if (d->wd >= 0 && combined_mask == d->combined_mask) | |
2388 | return 0; | |
2389 | ||
2390 | r = hashmap_ensure_allocated(&d->inotify_data->wd, NULL); | |
2391 | if (r < 0) | |
2392 | return r; | |
2393 | ||
2394 | wd = inotify_add_watch_fd(d->inotify_data->fd, d->fd, combined_mask); | |
2395 | if (wd < 0) | |
2396 | return -errno; | |
2397 | ||
2398 | if (d->wd < 0) { | |
2399 | r = hashmap_put(d->inotify_data->wd, INT_TO_PTR(wd), d); | |
2400 | if (r < 0) { | |
2401 | (void) inotify_rm_watch(d->inotify_data->fd, wd); | |
2402 | return r; | |
2403 | } | |
2404 | ||
2405 | d->wd = wd; | |
2406 | ||
2407 | } else if (d->wd != wd) { | |
2408 | ||
2409 | log_debug("Weird, the watch descriptor we already knew for this inode changed?"); | |
2410 | (void) inotify_rm_watch(d->fd, wd); | |
2411 | return -EINVAL; | |
2412 | } | |
2413 | ||
2414 | d->combined_mask = combined_mask; | |
2415 | return 1; | |
2416 | } | |
2417 | ||
b9350e70 LP |
2418 | static int inotify_exit_callback(sd_event_source *s, const struct inotify_event *event, void *userdata) { |
2419 | assert(s); | |
2420 | ||
2421 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
2422 | } | |
2423 | ||
e67d738a | 2424 | static int event_add_inotify_fd_internal( |
97ef5391 LP |
2425 | sd_event *e, |
2426 | sd_event_source **ret, | |
e67d738a LP |
2427 | int fd, |
2428 | bool donate, | |
97ef5391 LP |
2429 | uint32_t mask, |
2430 | sd_event_inotify_handler_t callback, | |
2431 | void *userdata) { | |
2432 | ||
5bb1d7fb | 2433 | _cleanup_close_ int donated_fd = donate ? fd : -EBADF; |
e67d738a | 2434 | _cleanup_(source_freep) sd_event_source *s = NULL; |
97ef5391 LP |
2435 | struct inotify_data *inotify_data = NULL; |
2436 | struct inode_data *inode_data = NULL; | |
97ef5391 LP |
2437 | struct stat st; |
2438 | int r; | |
2439 | ||
2440 | assert_return(e, -EINVAL); | |
2441 | assert_return(e = event_resolve(e), -ENOPKG); | |
e67d738a | 2442 | assert_return(fd >= 0, -EBADF); |
97ef5391 LP |
2443 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2444 | assert_return(!event_pid_changed(e), -ECHILD); | |
2445 | ||
b9350e70 LP |
2446 | if (!callback) |
2447 | callback = inotify_exit_callback; | |
2448 | ||
97ef5391 LP |
2449 | /* Refuse IN_MASK_ADD since we coalesce watches on the same inode, and hence really don't want to merge |
2450 | * masks. Or in other words, this whole code exists only to manage IN_MASK_ADD type operations for you, hence | |
2451 | * the user can't use them for us. */ | |
2452 | if (mask & IN_MASK_ADD) | |
2453 | return -EINVAL; | |
2454 | ||
97ef5391 LP |
2455 | if (fstat(fd, &st) < 0) |
2456 | return -errno; | |
2457 | ||
2458 | s = source_new(e, !ret, SOURCE_INOTIFY); | |
2459 | if (!s) | |
2460 | return -ENOMEM; | |
2461 | ||
2462 | s->enabled = mask & IN_ONESHOT ? SD_EVENT_ONESHOT : SD_EVENT_ON; | |
2463 | s->inotify.mask = mask; | |
2464 | s->inotify.callback = callback; | |
2465 | s->userdata = userdata; | |
2466 | ||
2467 | /* Allocate an inotify object for this priority, and an inode object within it */ | |
2468 | r = event_make_inotify_data(e, SD_EVENT_PRIORITY_NORMAL, &inotify_data); | |
2469 | if (r < 0) | |
8c75fe17 | 2470 | return r; |
97ef5391 LP |
2471 | |
2472 | r = event_make_inode_data(e, inotify_data, st.st_dev, st.st_ino, &inode_data); | |
8c75fe17 | 2473 | if (r < 0) { |
e67d738a | 2474 | event_gc_inotify_data(e, inotify_data); |
8c75fe17 ZJS |
2475 | return r; |
2476 | } | |
97ef5391 LP |
2477 | |
2478 | /* Keep the O_PATH fd around until the first iteration of the loop, so that we can still change the priority of | |
2479 | * the event source, until then, for which we need the original inode. */ | |
2480 | if (inode_data->fd < 0) { | |
e67d738a LP |
2481 | if (donated_fd >= 0) |
2482 | inode_data->fd = TAKE_FD(donated_fd); | |
2483 | else { | |
2484 | inode_data->fd = fcntl(fd, F_DUPFD_CLOEXEC, 3); | |
2485 | if (inode_data->fd < 0) { | |
2486 | r = -errno; | |
2487 | event_gc_inode_data(e, inode_data); | |
2488 | return r; | |
2489 | } | |
2490 | } | |
2491 | ||
ed828563 | 2492 | LIST_PREPEND(to_close, e->inode_data_to_close_list, inode_data); |
97ef5391 LP |
2493 | } |
2494 | ||
2495 | /* Link our event source to the inode data object */ | |
2496 | LIST_PREPEND(inotify.by_inode_data, inode_data->event_sources, s); | |
2497 | s->inotify.inode_data = inode_data; | |
2498 | ||
97ef5391 LP |
2499 | /* Actually realize the watch now */ |
2500 | r = inode_data_realize_watch(e, inode_data); | |
2501 | if (r < 0) | |
8c75fe17 | 2502 | return r; |
97ef5391 | 2503 | |
97ef5391 LP |
2504 | if (ret) |
2505 | *ret = s; | |
8c75fe17 | 2506 | TAKE_PTR(s); |
97ef5391 LP |
2507 | |
2508 | return 0; | |
97ef5391 LP |
2509 | } |
2510 | ||
e67d738a LP |
2511 | _public_ int sd_event_add_inotify_fd( |
2512 | sd_event *e, | |
2513 | sd_event_source **ret, | |
2514 | int fd, | |
2515 | uint32_t mask, | |
2516 | sd_event_inotify_handler_t callback, | |
2517 | void *userdata) { | |
2518 | ||
2519 | return event_add_inotify_fd_internal(e, ret, fd, /* donate= */ false, mask, callback, userdata); | |
2520 | } | |
2521 | ||
2522 | _public_ int sd_event_add_inotify( | |
2523 | sd_event *e, | |
2524 | sd_event_source **ret, | |
2525 | const char *path, | |
2526 | uint32_t mask, | |
2527 | sd_event_inotify_handler_t callback, | |
2528 | void *userdata) { | |
2529 | ||
2091c779 | 2530 | sd_event_source *s = NULL; /* avoid false maybe-uninitialized warning */ |
e67d738a LP |
2531 | int fd, r; |
2532 | ||
2533 | assert_return(path, -EINVAL); | |
2534 | ||
586c8cee ZJS |
2535 | fd = open(path, O_PATH | O_CLOEXEC | |
2536 | (mask & IN_ONLYDIR ? O_DIRECTORY : 0) | | |
2537 | (mask & IN_DONT_FOLLOW ? O_NOFOLLOW : 0)); | |
e67d738a LP |
2538 | if (fd < 0) |
2539 | return -errno; | |
2540 | ||
2541 | r = event_add_inotify_fd_internal(e, &s, fd, /* donate= */ true, mask, callback, userdata); | |
2542 | if (r < 0) | |
2543 | return r; | |
2544 | ||
2545 | (void) sd_event_source_set_description(s, path); | |
2546 | ||
2547 | if (ret) | |
2548 | *ret = s; | |
2549 | ||
2550 | return r; | |
2551 | } | |
2552 | ||
8301aa0b | 2553 | static sd_event_source* event_source_free(sd_event_source *s) { |
6680dd6b LP |
2554 | if (!s) |
2555 | return NULL; | |
da7e457c | 2556 | |
8301aa0b YW |
2557 | /* Here's a special hack: when we are called from a |
2558 | * dispatch handler we won't free the event source | |
2559 | * immediately, but we will detach the fd from the | |
2560 | * epoll. This way it is safe for the caller to unref | |
2561 | * the event source and immediately close the fd, but | |
2562 | * we still retain a valid event source object after | |
2563 | * the callback. */ | |
fd38203a | 2564 | |
76d04c3a | 2565 | if (s->dispatching) |
8301aa0b | 2566 | source_disconnect(s); |
76d04c3a | 2567 | else |
8301aa0b | 2568 | source_free(s); |
fd38203a LP |
2569 | |
2570 | return NULL; | |
2571 | } | |
2572 | ||
8301aa0b YW |
2573 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event_source, sd_event_source, event_source_free); |
2574 | ||
356779df | 2575 | _public_ int sd_event_source_set_description(sd_event_source *s, const char *description) { |
f7f53e9e | 2576 | assert_return(s, -EINVAL); |
f4b2933e | 2577 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 2578 | |
356779df | 2579 | return free_and_strdup(&s->description, description); |
f7f53e9e TG |
2580 | } |
2581 | ||
356779df | 2582 | _public_ int sd_event_source_get_description(sd_event_source *s, const char **description) { |
f7f53e9e | 2583 | assert_return(s, -EINVAL); |
356779df | 2584 | assert_return(description, -EINVAL); |
f4b2933e | 2585 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 2586 | |
7d92a1a4 ZJS |
2587 | if (!s->description) |
2588 | return -ENXIO; | |
2589 | ||
356779df | 2590 | *description = s->description; |
f7f53e9e TG |
2591 | return 0; |
2592 | } | |
2593 | ||
adcc4ca3 | 2594 | _public_ sd_event *sd_event_source_get_event(sd_event_source *s) { |
305f78bf | 2595 | assert_return(s, NULL); |
eaa3cbef LP |
2596 | |
2597 | return s->event; | |
2598 | } | |
2599 | ||
f7262a9f | 2600 | _public_ int sd_event_source_get_pending(sd_event_source *s) { |
305f78bf | 2601 | assert_return(s, -EINVAL); |
6203e07a | 2602 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c | 2603 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2604 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2605 | |
2606 | return s->pending; | |
2607 | } | |
2608 | ||
f7262a9f | 2609 | _public_ int sd_event_source_get_io_fd(sd_event_source *s) { |
305f78bf LP |
2610 | assert_return(s, -EINVAL); |
2611 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2612 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2613 | |
2614 | return s->io.fd; | |
2615 | } | |
2616 | ||
30caf8f3 LP |
2617 | _public_ int sd_event_source_set_io_fd(sd_event_source *s, int fd) { |
2618 | int r; | |
2619 | ||
2620 | assert_return(s, -EINVAL); | |
8ac43fee | 2621 | assert_return(fd >= 0, -EBADF); |
30caf8f3 LP |
2622 | assert_return(s->type == SOURCE_IO, -EDOM); |
2623 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2624 | ||
2625 | if (s->io.fd == fd) | |
2626 | return 0; | |
2627 | ||
b6d5481b | 2628 | if (event_source_is_offline(s)) { |
30caf8f3 LP |
2629 | s->io.fd = fd; |
2630 | s->io.registered = false; | |
2631 | } else { | |
2632 | int saved_fd; | |
2633 | ||
2634 | saved_fd = s->io.fd; | |
2635 | assert(s->io.registered); | |
2636 | ||
2637 | s->io.fd = fd; | |
2638 | s->io.registered = false; | |
2639 | ||
2640 | r = source_io_register(s, s->enabled, s->io.events); | |
2641 | if (r < 0) { | |
2642 | s->io.fd = saved_fd; | |
2643 | s->io.registered = true; | |
2644 | return r; | |
2645 | } | |
2646 | ||
5a795bff | 2647 | (void) epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, saved_fd, NULL); |
30caf8f3 LP |
2648 | } |
2649 | ||
2650 | return 0; | |
2651 | } | |
2652 | ||
ab93297c NM |
2653 | _public_ int sd_event_source_get_io_fd_own(sd_event_source *s) { |
2654 | assert_return(s, -EINVAL); | |
2655 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2656 | ||
2657 | return s->io.owned; | |
2658 | } | |
2659 | ||
2660 | _public_ int sd_event_source_set_io_fd_own(sd_event_source *s, int own) { | |
2661 | assert_return(s, -EINVAL); | |
2662 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2663 | ||
2664 | s->io.owned = own; | |
2665 | return 0; | |
2666 | } | |
2667 | ||
f7262a9f | 2668 | _public_ int sd_event_source_get_io_events(sd_event_source *s, uint32_t* events) { |
305f78bf LP |
2669 | assert_return(s, -EINVAL); |
2670 | assert_return(events, -EINVAL); | |
2671 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2672 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2673 | |
2674 | *events = s->io.events; | |
2675 | return 0; | |
2676 | } | |
2677 | ||
f7262a9f | 2678 | _public_ int sd_event_source_set_io_events(sd_event_source *s, uint32_t events) { |
fd38203a LP |
2679 | int r; |
2680 | ||
305f78bf LP |
2681 | assert_return(s, -EINVAL); |
2682 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2a16a986 | 2683 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 2684 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2685 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2686 | |
b63c8d4f DH |
2687 | /* edge-triggered updates are never skipped, so we can reset edges */ |
2688 | if (s->io.events == events && !(events & EPOLLET)) | |
fd38203a LP |
2689 | return 0; |
2690 | ||
2a0dc6cd LP |
2691 | r = source_set_pending(s, false); |
2692 | if (r < 0) | |
2693 | return r; | |
2694 | ||
b6d5481b | 2695 | if (event_source_is_online(s)) { |
e4715127 | 2696 | r = source_io_register(s, s->enabled, events); |
fd38203a LP |
2697 | if (r < 0) |
2698 | return r; | |
2699 | } | |
2700 | ||
2701 | s->io.events = events; | |
2702 | ||
2703 | return 0; | |
2704 | } | |
2705 | ||
f7262a9f | 2706 | _public_ int sd_event_source_get_io_revents(sd_event_source *s, uint32_t* revents) { |
305f78bf LP |
2707 | assert_return(s, -EINVAL); |
2708 | assert_return(revents, -EINVAL); | |
2709 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2710 | assert_return(s->pending, -ENODATA); | |
2711 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2712 | |
2713 | *revents = s->io.revents; | |
2714 | return 0; | |
2715 | } | |
2716 | ||
f7262a9f | 2717 | _public_ int sd_event_source_get_signal(sd_event_source *s) { |
305f78bf LP |
2718 | assert_return(s, -EINVAL); |
2719 | assert_return(s->type == SOURCE_SIGNAL, -EDOM); | |
2720 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2721 | |
2722 | return s->signal.sig; | |
2723 | } | |
2724 | ||
31927c16 | 2725 | _public_ int sd_event_source_get_priority(sd_event_source *s, int64_t *priority) { |
305f78bf LP |
2726 | assert_return(s, -EINVAL); |
2727 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a | 2728 | |
6680b8d1 ME |
2729 | *priority = s->priority; |
2730 | return 0; | |
fd38203a LP |
2731 | } |
2732 | ||
31927c16 | 2733 | _public_ int sd_event_source_set_priority(sd_event_source *s, int64_t priority) { |
97ef5391 LP |
2734 | bool rm_inotify = false, rm_inode = false; |
2735 | struct inotify_data *new_inotify_data = NULL; | |
2736 | struct inode_data *new_inode_data = NULL; | |
9da4cb2b LP |
2737 | int r; |
2738 | ||
305f78bf | 2739 | assert_return(s, -EINVAL); |
da7e457c | 2740 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2741 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2742 | |
2743 | if (s->priority == priority) | |
2744 | return 0; | |
2745 | ||
97ef5391 LP |
2746 | if (s->type == SOURCE_INOTIFY) { |
2747 | struct inode_data *old_inode_data; | |
2748 | ||
2749 | assert(s->inotify.inode_data); | |
2750 | old_inode_data = s->inotify.inode_data; | |
2751 | ||
2752 | /* We need the original fd to change the priority. If we don't have it we can't change the priority, | |
2753 | * anymore. Note that we close any fds when entering the next event loop iteration, i.e. for inotify | |
2754 | * events we allow priority changes only until the first following iteration. */ | |
2755 | if (old_inode_data->fd < 0) | |
2756 | return -EOPNOTSUPP; | |
2757 | ||
2758 | r = event_make_inotify_data(s->event, priority, &new_inotify_data); | |
2759 | if (r < 0) | |
2760 | return r; | |
2761 | rm_inotify = r > 0; | |
2762 | ||
2763 | r = event_make_inode_data(s->event, new_inotify_data, old_inode_data->dev, old_inode_data->ino, &new_inode_data); | |
2764 | if (r < 0) | |
2765 | goto fail; | |
2766 | rm_inode = r > 0; | |
2767 | ||
2768 | if (new_inode_data->fd < 0) { | |
2769 | /* Duplicate the fd for the new inode object if we don't have any yet */ | |
2770 | new_inode_data->fd = fcntl(old_inode_data->fd, F_DUPFD_CLOEXEC, 3); | |
2771 | if (new_inode_data->fd < 0) { | |
2772 | r = -errno; | |
2773 | goto fail; | |
2774 | } | |
2775 | ||
ed828563 | 2776 | LIST_PREPEND(to_close, s->event->inode_data_to_close_list, new_inode_data); |
97ef5391 LP |
2777 | } |
2778 | ||
2779 | /* Move the event source to the new inode data structure */ | |
2780 | LIST_REMOVE(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2781 | LIST_PREPEND(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2782 | s->inotify.inode_data = new_inode_data; | |
2783 | ||
2784 | /* Now create the new watch */ | |
2785 | r = inode_data_realize_watch(s->event, new_inode_data); | |
2786 | if (r < 0) { | |
2787 | /* Move it back */ | |
2788 | LIST_REMOVE(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2789 | LIST_PREPEND(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2790 | s->inotify.inode_data = old_inode_data; | |
2791 | goto fail; | |
2792 | } | |
2793 | ||
2794 | s->priority = priority; | |
2795 | ||
2796 | event_gc_inode_data(s->event, old_inode_data); | |
2797 | ||
b6d5481b | 2798 | } else if (s->type == SOURCE_SIGNAL && event_source_is_online(s)) { |
9da4cb2b LP |
2799 | struct signal_data *old, *d; |
2800 | ||
2801 | /* Move us from the signalfd belonging to the old | |
2802 | * priority to the signalfd of the new priority */ | |
2803 | ||
2804 | assert_se(old = hashmap_get(s->event->signal_data, &s->priority)); | |
2805 | ||
2806 | s->priority = priority; | |
2807 | ||
2808 | r = event_make_signal_data(s->event, s->signal.sig, &d); | |
2809 | if (r < 0) { | |
2810 | s->priority = old->priority; | |
2811 | return r; | |
2812 | } | |
2813 | ||
2814 | event_unmask_signal_data(s->event, old, s->signal.sig); | |
2815 | } else | |
2816 | s->priority = priority; | |
fd38203a | 2817 | |
e1951c16 | 2818 | event_source_pp_prioq_reshuffle(s); |
fd38203a | 2819 | |
6203e07a LP |
2820 | if (s->type == SOURCE_EXIT) |
2821 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
305f78bf | 2822 | |
fd38203a | 2823 | return 0; |
97ef5391 LP |
2824 | |
2825 | fail: | |
2826 | if (rm_inode) | |
2827 | event_free_inode_data(s->event, new_inode_data); | |
2828 | ||
2829 | if (rm_inotify) | |
2830 | event_free_inotify_data(s->event, new_inotify_data); | |
2831 | ||
2832 | return r; | |
fd38203a LP |
2833 | } |
2834 | ||
cad143a8 | 2835 | _public_ int sd_event_source_get_enabled(sd_event_source *s, int *ret) { |
71193c0b ZJS |
2836 | /* Quick mode: the event source doesn't exist and we only want to query boolean enablement state. */ |
2837 | if (!s && !ret) | |
2838 | return false; | |
2839 | ||
305f78bf | 2840 | assert_return(s, -EINVAL); |
305f78bf | 2841 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2842 | |
cad143a8 LP |
2843 | if (ret) |
2844 | *ret = s->enabled; | |
2845 | ||
08c1eb0e | 2846 | return s->enabled != SD_EVENT_OFF; |
fd38203a LP |
2847 | } |
2848 | ||
b6d5481b LP |
2849 | static int event_source_offline( |
2850 | sd_event_source *s, | |
2851 | int enabled, | |
2852 | bool ratelimited) { | |
2853 | ||
2854 | bool was_offline; | |
fd38203a LP |
2855 | int r; |
2856 | ||
ddfde737 | 2857 | assert(s); |
b6d5481b | 2858 | assert(enabled == SD_EVENT_OFF || ratelimited); |
fd38203a | 2859 | |
ddfde737 | 2860 | /* Unset the pending flag when this event source is disabled */ |
b6d5481b LP |
2861 | if (s->enabled != SD_EVENT_OFF && |
2862 | enabled == SD_EVENT_OFF && | |
2863 | !IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
ddfde737 LP |
2864 | r = source_set_pending(s, false); |
2865 | if (r < 0) | |
2866 | return r; | |
2867 | } | |
cc567911 | 2868 | |
b6d5481b LP |
2869 | was_offline = event_source_is_offline(s); |
2870 | s->enabled = enabled; | |
2871 | s->ratelimited = ratelimited; | |
fd38203a | 2872 | |
ddfde737 | 2873 | switch (s->type) { |
fd38203a | 2874 | |
ddfde737 LP |
2875 | case SOURCE_IO: |
2876 | source_io_unregister(s); | |
2877 | break; | |
ac989a78 | 2878 | |
ddfde737 LP |
2879 | case SOURCE_SIGNAL: |
2880 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); | |
2881 | break; | |
fd38203a | 2882 | |
ddfde737 | 2883 | case SOURCE_CHILD: |
b6d5481b LP |
2884 | if (!was_offline) { |
2885 | assert(s->event->n_online_child_sources > 0); | |
2886 | s->event->n_online_child_sources--; | |
2887 | } | |
fd38203a | 2888 | |
ddfde737 LP |
2889 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
2890 | source_child_pidfd_unregister(s); | |
2891 | else | |
2892 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
2893 | break; | |
4807d2d0 | 2894 | |
ddfde737 LP |
2895 | case SOURCE_EXIT: |
2896 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
2897 | break; | |
fd38203a | 2898 | |
158fe190 LP |
2899 | case SOURCE_MEMORY_PRESSURE: |
2900 | source_memory_pressure_unregister(s); | |
2901 | break; | |
2902 | ||
2115b9b6 YW |
2903 | case SOURCE_TIME_REALTIME: |
2904 | case SOURCE_TIME_BOOTTIME: | |
2905 | case SOURCE_TIME_MONOTONIC: | |
2906 | case SOURCE_TIME_REALTIME_ALARM: | |
2907 | case SOURCE_TIME_BOOTTIME_ALARM: | |
ddfde737 LP |
2908 | case SOURCE_DEFER: |
2909 | case SOURCE_POST: | |
2910 | case SOURCE_INOTIFY: | |
2911 | break; | |
fd38203a | 2912 | |
ddfde737 | 2913 | default: |
04499a70 | 2914 | assert_not_reached(); |
ddfde737 | 2915 | } |
fd38203a | 2916 | |
2115b9b6 YW |
2917 | /* Always reshuffle time prioq, as the ratelimited flag may be changed. */ |
2918 | event_source_time_prioq_reshuffle(s); | |
2919 | ||
b6d5481b | 2920 | return 1; |
ddfde737 | 2921 | } |
f8f3f926 | 2922 | |
b6d5481b LP |
2923 | static int event_source_online( |
2924 | sd_event_source *s, | |
2925 | int enabled, | |
2926 | bool ratelimited) { | |
2927 | ||
2928 | bool was_online; | |
ddfde737 | 2929 | int r; |
fd38203a | 2930 | |
ddfde737 | 2931 | assert(s); |
b6d5481b | 2932 | assert(enabled != SD_EVENT_OFF || !ratelimited); |
305f78bf | 2933 | |
ddfde737 | 2934 | /* Unset the pending flag when this event source is enabled */ |
b6d5481b LP |
2935 | if (s->enabled == SD_EVENT_OFF && |
2936 | enabled != SD_EVENT_OFF && | |
2937 | !IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
ddfde737 LP |
2938 | r = source_set_pending(s, false); |
2939 | if (r < 0) | |
2940 | return r; | |
2941 | } | |
9d3e3aa5 | 2942 | |
b6d5481b LP |
2943 | /* Are we really ready for onlining? */ |
2944 | if (enabled == SD_EVENT_OFF || ratelimited) { | |
2945 | /* Nope, we are not ready for onlining, then just update the precise state and exit */ | |
2946 | s->enabled = enabled; | |
2947 | s->ratelimited = ratelimited; | |
2948 | return 0; | |
2949 | } | |
2950 | ||
2951 | was_online = event_source_is_online(s); | |
2952 | ||
ddfde737 | 2953 | switch (s->type) { |
ddfde737 | 2954 | case SOURCE_IO: |
b6d5481b | 2955 | r = source_io_register(s, enabled, s->io.events); |
d2eafe61 | 2956 | if (r < 0) |
ddfde737 | 2957 | return r; |
ddfde737 | 2958 | break; |
fd38203a | 2959 | |
ddfde737 LP |
2960 | case SOURCE_SIGNAL: |
2961 | r = event_make_signal_data(s->event, s->signal.sig, NULL); | |
2962 | if (r < 0) { | |
ddfde737 LP |
2963 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
2964 | return r; | |
2965 | } | |
fd38203a | 2966 | |
ddfde737 | 2967 | break; |
fd38203a | 2968 | |
ddfde737 | 2969 | case SOURCE_CHILD: |
ddfde737 LP |
2970 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
2971 | /* yes, we have pidfd */ | |
9da4cb2b | 2972 | |
b6d5481b | 2973 | r = source_child_pidfd_register(s, enabled); |
ac9f2640 | 2974 | if (r < 0) |
9da4cb2b | 2975 | return r; |
ddfde737 LP |
2976 | } else { |
2977 | /* no pidfd, or something other to watch for than WEXITED */ | |
9da4cb2b | 2978 | |
ddfde737 LP |
2979 | r = event_make_signal_data(s->event, SIGCHLD, NULL); |
2980 | if (r < 0) { | |
ddfde737 LP |
2981 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); |
2982 | return r; | |
2983 | } | |
2984 | } | |
fd38203a | 2985 | |
b6d5481b LP |
2986 | if (!was_online) |
2987 | s->event->n_online_child_sources++; | |
ddfde737 | 2988 | break; |
4807d2d0 | 2989 | |
158fe190 LP |
2990 | case SOURCE_MEMORY_PRESSURE: |
2991 | r = source_memory_pressure_register(s, enabled); | |
2992 | if (r < 0) | |
2993 | return r; | |
2994 | ||
2995 | break; | |
2996 | ||
d2eafe61 ZJS |
2997 | case SOURCE_TIME_REALTIME: |
2998 | case SOURCE_TIME_BOOTTIME: | |
2999 | case SOURCE_TIME_MONOTONIC: | |
3000 | case SOURCE_TIME_REALTIME_ALARM: | |
3001 | case SOURCE_TIME_BOOTTIME_ALARM: | |
ddfde737 | 3002 | case SOURCE_EXIT: |
ddfde737 LP |
3003 | case SOURCE_DEFER: |
3004 | case SOURCE_POST: | |
3005 | case SOURCE_INOTIFY: | |
3006 | break; | |
9da4cb2b | 3007 | |
ddfde737 | 3008 | default: |
04499a70 | 3009 | assert_not_reached(); |
ddfde737 | 3010 | } |
f8f3f926 | 3011 | |
b6d5481b LP |
3012 | s->enabled = enabled; |
3013 | s->ratelimited = ratelimited; | |
d2eafe61 ZJS |
3014 | |
3015 | /* Non-failing operations below */ | |
2115b9b6 | 3016 | if (s->type == SOURCE_EXIT) |
d2eafe61 | 3017 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); |
d2eafe61 | 3018 | |
2115b9b6 YW |
3019 | /* Always reshuffle time prioq, as the ratelimited flag may be changed. */ |
3020 | event_source_time_prioq_reshuffle(s); | |
d2eafe61 | 3021 | |
b6d5481b | 3022 | return 1; |
ddfde737 LP |
3023 | } |
3024 | ||
3025 | _public_ int sd_event_source_set_enabled(sd_event_source *s, int m) { | |
3026 | int r; | |
9da4cb2b | 3027 | |
ddfde737 | 3028 | assert_return(IN_SET(m, SD_EVENT_OFF, SD_EVENT_ON, SD_EVENT_ONESHOT), -EINVAL); |
7e922b05 ZJS |
3029 | |
3030 | /* Quick mode: if the source doesn't exist, SD_EVENT_OFF is a noop. */ | |
3031 | if (m == SD_EVENT_OFF && !s) | |
3032 | return 0; | |
3033 | ||
3034 | assert_return(s, -EINVAL); | |
ddfde737 | 3035 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 3036 | |
ddfde737 LP |
3037 | /* If we are dead anyway, we are fine with turning off sources, but everything else needs to fail. */ |
3038 | if (s->event->state == SD_EVENT_FINISHED) | |
3039 | return m == SD_EVENT_OFF ? 0 : -ESTALE; | |
305f78bf | 3040 | |
ddfde737 LP |
3041 | if (s->enabled == m) /* No change? */ |
3042 | return 0; | |
9d3e3aa5 | 3043 | |
ddfde737 | 3044 | if (m == SD_EVENT_OFF) |
b6d5481b | 3045 | r = event_source_offline(s, m, s->ratelimited); |
ddfde737 LP |
3046 | else { |
3047 | if (s->enabled != SD_EVENT_OFF) { | |
3048 | /* Switching from "on" to "oneshot" or back? If that's the case, we can take a shortcut, the | |
3049 | * event source is already enabled after all. */ | |
3050 | s->enabled = m; | |
3051 | return 0; | |
fd38203a | 3052 | } |
ddfde737 | 3053 | |
b6d5481b | 3054 | r = event_source_online(s, m, s->ratelimited); |
fd38203a | 3055 | } |
ddfde737 LP |
3056 | if (r < 0) |
3057 | return r; | |
fd38203a | 3058 | |
e1951c16 | 3059 | event_source_pp_prioq_reshuffle(s); |
fd38203a LP |
3060 | return 0; |
3061 | } | |
3062 | ||
f7262a9f | 3063 | _public_ int sd_event_source_get_time(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
3064 | assert_return(s, -EINVAL); |
3065 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 3066 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf | 3067 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
3068 | |
3069 | *usec = s->time.next; | |
3070 | return 0; | |
3071 | } | |
3072 | ||
f7262a9f | 3073 | _public_ int sd_event_source_set_time(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 3074 | int r; |
6a0f1f6d | 3075 | |
305f78bf | 3076 | assert_return(s, -EINVAL); |
6a0f1f6d | 3077 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 3078 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 3079 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 3080 | |
2a0dc6cd LP |
3081 | r = source_set_pending(s, false); |
3082 | if (r < 0) | |
3083 | return r; | |
2576a19e | 3084 | |
2a0dc6cd | 3085 | s->time.next = usec; |
fd38203a | 3086 | |
e1951c16 | 3087 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
3088 | return 0; |
3089 | } | |
3090 | ||
d6a83dc4 LP |
3091 | _public_ int sd_event_source_set_time_relative(sd_event_source *s, uint64_t usec) { |
3092 | usec_t t; | |
3093 | int r; | |
3094 | ||
3095 | assert_return(s, -EINVAL); | |
3096 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
3097 | ||
ef859195 LP |
3098 | if (usec == USEC_INFINITY) |
3099 | return sd_event_source_set_time(s, USEC_INFINITY); | |
3100 | ||
d6a83dc4 LP |
3101 | r = sd_event_now(s->event, event_source_type_to_clock(s->type), &t); |
3102 | if (r < 0) | |
3103 | return r; | |
3104 | ||
496db330 YW |
3105 | usec = usec_add(t, usec); |
3106 | if (usec == USEC_INFINITY) | |
d6a83dc4 LP |
3107 | return -EOVERFLOW; |
3108 | ||
496db330 | 3109 | return sd_event_source_set_time(s, usec); |
d6a83dc4 LP |
3110 | } |
3111 | ||
f7262a9f | 3112 | _public_ int sd_event_source_get_time_accuracy(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
3113 | assert_return(s, -EINVAL); |
3114 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 3115 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf LP |
3116 | assert_return(!event_pid_changed(s->event), -ECHILD); |
3117 | ||
3118 | *usec = s->time.accuracy; | |
3119 | return 0; | |
3120 | } | |
3121 | ||
f7262a9f | 3122 | _public_ int sd_event_source_set_time_accuracy(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 3123 | int r; |
6a0f1f6d | 3124 | |
305f78bf | 3125 | assert_return(s, -EINVAL); |
f5fbe71d | 3126 | assert_return(usec != UINT64_MAX, -EINVAL); |
6a0f1f6d | 3127 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 3128 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 3129 | assert_return(!event_pid_changed(s->event), -ECHILD); |
eaa3cbef | 3130 | |
2a0dc6cd LP |
3131 | r = source_set_pending(s, false); |
3132 | if (r < 0) | |
3133 | return r; | |
3134 | ||
eaa3cbef LP |
3135 | if (usec == 0) |
3136 | usec = DEFAULT_ACCURACY_USEC; | |
3137 | ||
eaa3cbef LP |
3138 | s->time.accuracy = usec; |
3139 | ||
e1951c16 | 3140 | event_source_time_prioq_reshuffle(s); |
6a0f1f6d LP |
3141 | return 0; |
3142 | } | |
3143 | ||
3144 | _public_ int sd_event_source_get_time_clock(sd_event_source *s, clockid_t *clock) { | |
3145 | assert_return(s, -EINVAL); | |
3146 | assert_return(clock, -EINVAL); | |
3147 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
3148 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
eaa3cbef | 3149 | |
6a0f1f6d | 3150 | *clock = event_source_type_to_clock(s->type); |
eaa3cbef LP |
3151 | return 0; |
3152 | } | |
3153 | ||
f7262a9f | 3154 | _public_ int sd_event_source_get_child_pid(sd_event_source *s, pid_t *pid) { |
4bee8012 LP |
3155 | assert_return(s, -EINVAL); |
3156 | assert_return(pid, -EINVAL); | |
3157 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
3158 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
3159 | ||
3160 | *pid = s->child.pid; | |
3161 | return 0; | |
3162 | } | |
3163 | ||
f8f3f926 LP |
3164 | _public_ int sd_event_source_get_child_pidfd(sd_event_source *s) { |
3165 | assert_return(s, -EINVAL); | |
3166 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
3167 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
3168 | ||
3169 | if (s->child.pidfd < 0) | |
3170 | return -EOPNOTSUPP; | |
3171 | ||
3172 | return s->child.pidfd; | |
3173 | } | |
3174 | ||
3175 | _public_ int sd_event_source_send_child_signal(sd_event_source *s, int sig, const siginfo_t *si, unsigned flags) { | |
3176 | assert_return(s, -EINVAL); | |
3177 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
3178 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
3179 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
3180 | ||
3181 | /* If we already have seen indication the process exited refuse sending a signal early. This way we | |
3182 | * can be sure we don't accidentally kill the wrong process on PID reuse when pidfds are not | |
3183 | * available. */ | |
3184 | if (s->child.exited) | |
3185 | return -ESRCH; | |
3186 | ||
3187 | if (s->child.pidfd >= 0) { | |
3188 | siginfo_t copy; | |
3189 | ||
3190 | /* pidfd_send_signal() changes the siginfo_t argument. This is weird, let's hence copy the | |
3191 | * structure here */ | |
3192 | if (si) | |
3193 | copy = *si; | |
3194 | ||
3195 | if (pidfd_send_signal(s->child.pidfd, sig, si ? © : NULL, 0) < 0) { | |
3196 | /* Let's propagate the error only if the system call is not implemented or prohibited */ | |
3197 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
3198 | return -errno; | |
3199 | } else | |
3200 | return 0; | |
3201 | } | |
3202 | ||
3203 | /* Flags are only supported for pidfd_send_signal(), not for rt_sigqueueinfo(), hence let's refuse | |
3204 | * this here. */ | |
3205 | if (flags != 0) | |
3206 | return -EOPNOTSUPP; | |
3207 | ||
3208 | if (si) { | |
3209 | /* We use rt_sigqueueinfo() only if siginfo_t is specified. */ | |
3210 | siginfo_t copy = *si; | |
3211 | ||
3212 | if (rt_sigqueueinfo(s->child.pid, sig, ©) < 0) | |
3213 | return -errno; | |
3214 | } else if (kill(s->child.pid, sig) < 0) | |
3215 | return -errno; | |
3216 | ||
3217 | return 0; | |
3218 | } | |
3219 | ||
3220 | _public_ int sd_event_source_get_child_pidfd_own(sd_event_source *s) { | |
3221 | assert_return(s, -EINVAL); | |
3222 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
3223 | ||
3224 | if (s->child.pidfd < 0) | |
3225 | return -EOPNOTSUPP; | |
3226 | ||
3227 | return s->child.pidfd_owned; | |
3228 | } | |
3229 | ||
3230 | _public_ int sd_event_source_set_child_pidfd_own(sd_event_source *s, int own) { | |
3231 | assert_return(s, -EINVAL); | |
3232 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
3233 | ||
3234 | if (s->child.pidfd < 0) | |
3235 | return -EOPNOTSUPP; | |
3236 | ||
3237 | s->child.pidfd_owned = own; | |
3238 | return 0; | |
3239 | } | |
3240 | ||
3241 | _public_ int sd_event_source_get_child_process_own(sd_event_source *s) { | |
3242 | assert_return(s, -EINVAL); | |
3243 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
3244 | ||
3245 | return s->child.process_owned; | |
3246 | } | |
3247 | ||
3248 | _public_ int sd_event_source_set_child_process_own(sd_event_source *s, int own) { | |
3249 | assert_return(s, -EINVAL); | |
3250 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
3251 | ||
3252 | s->child.process_owned = own; | |
3253 | return 0; | |
3254 | } | |
3255 | ||
97ef5391 LP |
3256 | _public_ int sd_event_source_get_inotify_mask(sd_event_source *s, uint32_t *mask) { |
3257 | assert_return(s, -EINVAL); | |
3258 | assert_return(mask, -EINVAL); | |
3259 | assert_return(s->type == SOURCE_INOTIFY, -EDOM); | |
3260 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
3261 | ||
3262 | *mask = s->inotify.mask; | |
3263 | return 0; | |
3264 | } | |
3265 | ||
718db961 | 3266 | _public_ int sd_event_source_set_prepare(sd_event_source *s, sd_event_handler_t callback) { |
fd38203a LP |
3267 | int r; |
3268 | ||
da7e457c | 3269 | assert_return(s, -EINVAL); |
6203e07a | 3270 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c LP |
3271 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
3272 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
3273 | |
3274 | if (s->prepare == callback) | |
3275 | return 0; | |
3276 | ||
3277 | if (callback && s->prepare) { | |
3278 | s->prepare = callback; | |
3279 | return 0; | |
3280 | } | |
3281 | ||
3282 | r = prioq_ensure_allocated(&s->event->prepare, prepare_prioq_compare); | |
3283 | if (r < 0) | |
3284 | return r; | |
3285 | ||
3286 | s->prepare = callback; | |
3287 | ||
3288 | if (callback) { | |
3289 | r = prioq_put(s->event->prepare, s, &s->prepare_index); | |
3290 | if (r < 0) | |
3291 | return r; | |
3292 | } else | |
3293 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
3294 | ||
3295 | return 0; | |
3296 | } | |
3297 | ||
f7262a9f | 3298 | _public_ void* sd_event_source_get_userdata(sd_event_source *s) { |
da7e457c | 3299 | assert_return(s, NULL); |
fd38203a LP |
3300 | |
3301 | return s->userdata; | |
3302 | } | |
3303 | ||
8f726607 LP |
3304 | _public_ void *sd_event_source_set_userdata(sd_event_source *s, void *userdata) { |
3305 | void *ret; | |
3306 | ||
3307 | assert_return(s, NULL); | |
3308 | ||
3309 | ret = s->userdata; | |
3310 | s->userdata = userdata; | |
3311 | ||
3312 | return ret; | |
3313 | } | |
3314 | ||
b6d5481b LP |
3315 | static int event_source_enter_ratelimited(sd_event_source *s) { |
3316 | int r; | |
3317 | ||
3318 | assert(s); | |
3319 | ||
3320 | /* When an event source becomes ratelimited, we place it in the CLOCK_MONOTONIC priority queue, with | |
3321 | * the end of the rate limit time window, much as if it was a timer event source. */ | |
3322 | ||
3323 | if (s->ratelimited) | |
3324 | return 0; /* Already ratelimited, this is a NOP hence */ | |
3325 | ||
3326 | /* Make sure we can install a CLOCK_MONOTONIC event further down. */ | |
3327 | r = setup_clock_data(s->event, &s->event->monotonic, CLOCK_MONOTONIC); | |
3328 | if (r < 0) | |
3329 | return r; | |
3330 | ||
3331 | /* Timer event sources are already using the earliest/latest queues for the timer scheduling. Let's | |
3332 | * first remove them from the prioq appropriate for their own clock, so that we can use the prioq | |
3333 | * fields of the event source then for adding it to the CLOCK_MONOTONIC prioq instead. */ | |
3334 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
3335 | event_source_time_prioq_remove(s, event_get_clock_data(s->event, s->type)); | |
3336 | ||
3337 | /* Now, let's add the event source to the monotonic clock instead */ | |
3338 | r = event_source_time_prioq_put(s, &s->event->monotonic); | |
3339 | if (r < 0) | |
3340 | goto fail; | |
3341 | ||
3342 | /* And let's take the event source officially offline */ | |
3343 | r = event_source_offline(s, s->enabled, /* ratelimited= */ true); | |
3344 | if (r < 0) { | |
3345 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
3346 | goto fail; | |
3347 | } | |
3348 | ||
3349 | event_source_pp_prioq_reshuffle(s); | |
3350 | ||
3351 | log_debug("Event source %p (%s) entered rate limit state.", s, strna(s->description)); | |
3352 | return 0; | |
3353 | ||
3354 | fail: | |
3355 | /* Reinstall time event sources in the priority queue as before. This shouldn't fail, since the queue | |
3356 | * space for it should already be allocated. */ | |
3357 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
3358 | assert_se(event_source_time_prioq_put(s, event_get_clock_data(s->event, s->type)) >= 0); | |
3359 | ||
3360 | return r; | |
3361 | } | |
3362 | ||
fd69f224 | 3363 | static int event_source_leave_ratelimit(sd_event_source *s, bool run_callback) { |
b6d5481b LP |
3364 | int r; |
3365 | ||
3366 | assert(s); | |
3367 | ||
3368 | if (!s->ratelimited) | |
3369 | return 0; | |
3370 | ||
3371 | /* Let's take the event source out of the monotonic prioq first. */ | |
3372 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
3373 | ||
3374 | /* Let's then add the event source to its native clock prioq again — if this is a timer event source */ | |
3375 | if (EVENT_SOURCE_IS_TIME(s->type)) { | |
3376 | r = event_source_time_prioq_put(s, event_get_clock_data(s->event, s->type)); | |
3377 | if (r < 0) | |
3378 | goto fail; | |
3379 | } | |
3380 | ||
3381 | /* Let's try to take it online again. */ | |
3382 | r = event_source_online(s, s->enabled, /* ratelimited= */ false); | |
3383 | if (r < 0) { | |
3384 | /* Do something roughly sensible when this failed: undo the two prioq ops above */ | |
3385 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
3386 | event_source_time_prioq_remove(s, event_get_clock_data(s->event, s->type)); | |
3387 | ||
3388 | goto fail; | |
3389 | } | |
3390 | ||
3391 | event_source_pp_prioq_reshuffle(s); | |
3392 | ratelimit_reset(&s->rate_limit); | |
3393 | ||
3394 | log_debug("Event source %p (%s) left rate limit state.", s, strna(s->description)); | |
fd69f224 MS |
3395 | |
3396 | if (run_callback && s->ratelimit_expire_callback) { | |
3397 | s->dispatching = true; | |
3398 | r = s->ratelimit_expire_callback(s, s->userdata); | |
3399 | s->dispatching = false; | |
3400 | ||
3401 | if (r < 0) { | |
3402 | log_debug_errno(r, "Ratelimit expiry callback of event source %s (type %s) returned error, %s: %m", | |
3403 | strna(s->description), | |
3404 | event_source_type_to_string(s->type), | |
3405 | s->exit_on_failure ? "exiting" : "disabling"); | |
3406 | ||
3407 | if (s->exit_on_failure) | |
3408 | (void) sd_event_exit(s->event, r); | |
3409 | } | |
3410 | ||
3411 | if (s->n_ref == 0) | |
3412 | source_free(s); | |
3413 | else if (r < 0) | |
0a040e64 | 3414 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
fd69f224 MS |
3415 | |
3416 | return 1; | |
3417 | } | |
3418 | ||
b6d5481b LP |
3419 | return 0; |
3420 | ||
3421 | fail: | |
3422 | /* Do something somewhat reasonable when we cannot move an event sources out of ratelimited mode: | |
3423 | * simply put it back in it, maybe we can then process it more successfully next iteration. */ | |
3424 | assert_se(event_source_time_prioq_put(s, &s->event->monotonic) >= 0); | |
3425 | ||
3426 | return r; | |
3427 | } | |
3428 | ||
c2ba3ad6 LP |
3429 | static usec_t sleep_between(sd_event *e, usec_t a, usec_t b) { |
3430 | usec_t c; | |
3431 | assert(e); | |
3432 | assert(a <= b); | |
3433 | ||
3434 | if (a <= 0) | |
3435 | return 0; | |
393003e1 LP |
3436 | if (a >= USEC_INFINITY) |
3437 | return USEC_INFINITY; | |
c2ba3ad6 LP |
3438 | |
3439 | if (b <= a + 1) | |
3440 | return a; | |
3441 | ||
52444dc4 LP |
3442 | initialize_perturb(e); |
3443 | ||
c2ba3ad6 LP |
3444 | /* |
3445 | Find a good time to wake up again between times a and b. We | |
3446 | have two goals here: | |
3447 | ||
3448 | a) We want to wake up as seldom as possible, hence prefer | |
3449 | later times over earlier times. | |
3450 | ||
3451 | b) But if we have to wake up, then let's make sure to | |
3452 | dispatch as much as possible on the entire system. | |
3453 | ||
3454 | We implement this by waking up everywhere at the same time | |
850516e0 | 3455 | within any given minute if we can, synchronised via the |
c2ba3ad6 | 3456 | perturbation value determined from the boot ID. If we can't, |
ba276c81 LP |
3457 | then we try to find the same spot in every 10s, then 1s and |
3458 | then 250ms step. Otherwise, we pick the last possible time | |
3459 | to wake up. | |
c2ba3ad6 LP |
3460 | */ |
3461 | ||
850516e0 LP |
3462 | c = (b / USEC_PER_MINUTE) * USEC_PER_MINUTE + e->perturb; |
3463 | if (c >= b) { | |
3464 | if (_unlikely_(c < USEC_PER_MINUTE)) | |
3465 | return b; | |
3466 | ||
3467 | c -= USEC_PER_MINUTE; | |
3468 | } | |
3469 | ||
ba276c81 LP |
3470 | if (c >= a) |
3471 | return c; | |
3472 | ||
3473 | c = (b / (USEC_PER_SEC*10)) * (USEC_PER_SEC*10) + (e->perturb % (USEC_PER_SEC*10)); | |
3474 | if (c >= b) { | |
3475 | if (_unlikely_(c < USEC_PER_SEC*10)) | |
3476 | return b; | |
3477 | ||
3478 | c -= USEC_PER_SEC*10; | |
3479 | } | |
3480 | ||
850516e0 LP |
3481 | if (c >= a) |
3482 | return c; | |
3483 | ||
3484 | c = (b / USEC_PER_SEC) * USEC_PER_SEC + (e->perturb % USEC_PER_SEC); | |
c2ba3ad6 LP |
3485 | if (c >= b) { |
3486 | if (_unlikely_(c < USEC_PER_SEC)) | |
3487 | return b; | |
3488 | ||
3489 | c -= USEC_PER_SEC; | |
3490 | } | |
3491 | ||
3492 | if (c >= a) | |
3493 | return c; | |
3494 | ||
3495 | c = (b / (USEC_PER_MSEC*250)) * (USEC_PER_MSEC*250) + (e->perturb % (USEC_PER_MSEC*250)); | |
3496 | if (c >= b) { | |
3497 | if (_unlikely_(c < USEC_PER_MSEC*250)) | |
3498 | return b; | |
3499 | ||
3500 | c -= USEC_PER_MSEC*250; | |
3501 | } | |
3502 | ||
3503 | if (c >= a) | |
3504 | return c; | |
3505 | ||
3506 | return b; | |
3507 | } | |
3508 | ||
fd38203a LP |
3509 | static int event_arm_timer( |
3510 | sd_event *e, | |
6a0f1f6d | 3511 | struct clock_data *d) { |
fd38203a LP |
3512 | |
3513 | struct itimerspec its = {}; | |
c2ba3ad6 LP |
3514 | sd_event_source *a, *b; |
3515 | usec_t t; | |
fd38203a | 3516 | |
cde93897 | 3517 | assert(e); |
6a0f1f6d | 3518 | assert(d); |
fd38203a | 3519 | |
d06441da | 3520 | if (!d->needs_rearm) |
212bbb17 | 3521 | return 0; |
7e2bf71c YW |
3522 | |
3523 | d->needs_rearm = false; | |
212bbb17 | 3524 | |
6a0f1f6d | 3525 | a = prioq_peek(d->earliest); |
19947509 | 3526 | assert(!a || EVENT_SOURCE_USES_TIME_PRIOQ(a->type)); |
b6d5481b | 3527 | if (!a || a->enabled == SD_EVENT_OFF || time_event_source_next(a) == USEC_INFINITY) { |
72aedc1e | 3528 | |
6a0f1f6d | 3529 | if (d->fd < 0) |
c57b5ca3 LP |
3530 | return 0; |
3531 | ||
3a43da28 | 3532 | if (d->next == USEC_INFINITY) |
72aedc1e LP |
3533 | return 0; |
3534 | ||
3535 | /* disarm */ | |
15c689d7 LP |
3536 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
3537 | return -errno; | |
72aedc1e | 3538 | |
3a43da28 | 3539 | d->next = USEC_INFINITY; |
fd38203a | 3540 | return 0; |
72aedc1e | 3541 | } |
fd38203a | 3542 | |
6a0f1f6d | 3543 | b = prioq_peek(d->latest); |
19947509 ZJS |
3544 | assert(!b || EVENT_SOURCE_USES_TIME_PRIOQ(b->type)); |
3545 | assert(b && b->enabled != SD_EVENT_OFF); | |
c2ba3ad6 | 3546 | |
b6d5481b | 3547 | t = sleep_between(e, time_event_source_next(a), time_event_source_latest(b)); |
6a0f1f6d | 3548 | if (d->next == t) |
fd38203a LP |
3549 | return 0; |
3550 | ||
6a0f1f6d | 3551 | assert_se(d->fd >= 0); |
fd38203a | 3552 | |
c2ba3ad6 | 3553 | if (t == 0) { |
1751bdde | 3554 | /* We don't want to disarm here, just mean some time looooong ago. */ |
fd38203a LP |
3555 | its.it_value.tv_sec = 0; |
3556 | its.it_value.tv_nsec = 1; | |
3557 | } else | |
c2ba3ad6 | 3558 | timespec_store(&its.it_value, t); |
fd38203a | 3559 | |
15c689d7 | 3560 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 | 3561 | return -errno; |
fd38203a | 3562 | |
6a0f1f6d | 3563 | d->next = t; |
fd38203a LP |
3564 | return 0; |
3565 | } | |
3566 | ||
9a800b56 | 3567 | static int process_io(sd_event *e, sd_event_source *s, uint32_t revents) { |
fd38203a LP |
3568 | assert(e); |
3569 | assert(s); | |
3570 | assert(s->type == SOURCE_IO); | |
3571 | ||
9a800b56 LP |
3572 | /* If the event source was already pending, we just OR in the |
3573 | * new revents, otherwise we reset the value. The ORing is | |
3574 | * necessary to handle EPOLLONESHOT events properly where | |
3575 | * readability might happen independently of writability, and | |
3576 | * we need to keep track of both */ | |
3577 | ||
3578 | if (s->pending) | |
3579 | s->io.revents |= revents; | |
3580 | else | |
3581 | s->io.revents = revents; | |
fd38203a | 3582 | |
fd38203a LP |
3583 | return source_set_pending(s, true); |
3584 | } | |
3585 | ||
72aedc1e | 3586 | static int flush_timer(sd_event *e, int fd, uint32_t events, usec_t *next) { |
fd38203a LP |
3587 | uint64_t x; |
3588 | ssize_t ss; | |
3589 | ||
3590 | assert(e); | |
da7e457c | 3591 | assert(fd >= 0); |
72aedc1e | 3592 | |
305f78bf | 3593 | assert_return(events == EPOLLIN, -EIO); |
fd38203a LP |
3594 | |
3595 | ss = read(fd, &x, sizeof(x)); | |
3596 | if (ss < 0) { | |
8add30a0 | 3597 | if (ERRNO_IS_TRANSIENT(errno)) |
fd38203a LP |
3598 | return 0; |
3599 | ||
3600 | return -errno; | |
3601 | } | |
3602 | ||
8d35dae7 | 3603 | if (_unlikely_(ss != sizeof(x))) |
fd38203a LP |
3604 | return -EIO; |
3605 | ||
cde93897 | 3606 | if (next) |
3a43da28 | 3607 | *next = USEC_INFINITY; |
72aedc1e | 3608 | |
fd38203a LP |
3609 | return 0; |
3610 | } | |
3611 | ||
305f78bf LP |
3612 | static int process_timer( |
3613 | sd_event *e, | |
3614 | usec_t n, | |
6a0f1f6d | 3615 | struct clock_data *d) { |
305f78bf | 3616 | |
fd38203a | 3617 | sd_event_source *s; |
fd69f224 | 3618 | bool callback_invoked = false; |
fd38203a LP |
3619 | int r; |
3620 | ||
3621 | assert(e); | |
6a0f1f6d | 3622 | assert(d); |
fd38203a LP |
3623 | |
3624 | for (;;) { | |
6a0f1f6d | 3625 | s = prioq_peek(d->earliest); |
19947509 ZJS |
3626 | assert(!s || EVENT_SOURCE_USES_TIME_PRIOQ(s->type)); |
3627 | ||
b6d5481b LP |
3628 | if (!s || time_event_source_next(s) > n) |
3629 | break; | |
3630 | ||
3631 | if (s->ratelimited) { | |
3632 | /* This is an event sources whose ratelimit window has ended. Let's turn it on | |
3633 | * again. */ | |
3634 | assert(s->ratelimited); | |
3635 | ||
fd69f224 | 3636 | r = event_source_leave_ratelimit(s, /* run_callback */ true); |
b6d5481b LP |
3637 | if (r < 0) |
3638 | return r; | |
fd69f224 MS |
3639 | else if (r == 1) |
3640 | callback_invoked = true; | |
b6d5481b LP |
3641 | |
3642 | continue; | |
3643 | } | |
3644 | ||
3645 | if (s->enabled == SD_EVENT_OFF || s->pending) | |
fd38203a LP |
3646 | break; |
3647 | ||
3648 | r = source_set_pending(s, true); | |
3649 | if (r < 0) | |
3650 | return r; | |
3651 | ||
e1951c16 | 3652 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
3653 | } |
3654 | ||
fd69f224 | 3655 | return callback_invoked; |
fd38203a LP |
3656 | } |
3657 | ||
efd3be9d YW |
3658 | static int process_child(sd_event *e, int64_t threshold, int64_t *ret_min_priority) { |
3659 | int64_t min_priority = threshold; | |
3660 | bool something_new = false; | |
fd38203a | 3661 | sd_event_source *s; |
fd38203a LP |
3662 | int r; |
3663 | ||
3664 | assert(e); | |
efd3be9d YW |
3665 | assert(ret_min_priority); |
3666 | ||
3667 | if (!e->need_process_child) { | |
3668 | *ret_min_priority = min_priority; | |
3669 | return 0; | |
3670 | } | |
fd38203a | 3671 | |
c2ba3ad6 LP |
3672 | e->need_process_child = false; |
3673 | ||
91c70071 YW |
3674 | /* So, this is ugly. We iteratively invoke waitid() with P_PID + WNOHANG for each PID we wait |
3675 | * for, instead of using P_ALL. This is because we only want to get child information of very | |
3676 | * specific child processes, and not all of them. We might not have processed the SIGCHLD event | |
3677 | * of a previous invocation and we don't want to maintain a unbounded *per-child* event queue, | |
3678 | * hence we really don't want anything flushed out of the kernel's queue that we don't care | |
3679 | * about. Since this is O(n) this means that if you have a lot of processes you probably want | |
3680 | * to handle SIGCHLD yourself. | |
3681 | * | |
3682 | * We do not reap the children here (by using WNOWAIT), this is only done after the event | |
3683 | * source is dispatched so that the callback still sees the process as a zombie. */ | |
fd38203a | 3684 | |
90e74a66 | 3685 | HASHMAP_FOREACH(s, e->child_sources) { |
fd38203a LP |
3686 | assert(s->type == SOURCE_CHILD); |
3687 | ||
efd3be9d YW |
3688 | if (s->priority > threshold) |
3689 | continue; | |
3690 | ||
fd38203a LP |
3691 | if (s->pending) |
3692 | continue; | |
3693 | ||
b6d5481b | 3694 | if (event_source_is_offline(s)) |
fd38203a LP |
3695 | continue; |
3696 | ||
f8f3f926 LP |
3697 | if (s->child.exited) |
3698 | continue; | |
3699 | ||
91c70071 YW |
3700 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
3701 | /* There's a usable pidfd known for this event source? Then don't waitid() for | |
3702 | * it here */ | |
f8f3f926 LP |
3703 | continue; |
3704 | ||
fd38203a | 3705 | zero(s->child.siginfo); |
15c689d7 LP |
3706 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, |
3707 | WNOHANG | (s->child.options & WEXITED ? WNOWAIT : 0) | s->child.options) < 0) | |
bfd9bfcc | 3708 | return negative_errno(); |
fd38203a LP |
3709 | |
3710 | if (s->child.siginfo.si_pid != 0) { | |
945c2931 | 3711 | bool zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 3712 | |
f8f3f926 LP |
3713 | if (zombie) |
3714 | s->child.exited = true; | |
3715 | ||
08cd1552 | 3716 | if (!zombie && (s->child.options & WEXITED)) { |
91c70071 YW |
3717 | /* If the child isn't dead then let's immediately remove the state |
3718 | * change from the queue, since there's no benefit in leaving it | |
3719 | * queued. */ | |
08cd1552 LP |
3720 | |
3721 | assert(s->child.options & (WSTOPPED|WCONTINUED)); | |
a5d27871 | 3722 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|(s->child.options & (WSTOPPED|WCONTINUED))); |
08cd1552 LP |
3723 | } |
3724 | ||
fd38203a LP |
3725 | r = source_set_pending(s, true); |
3726 | if (r < 0) | |
3727 | return r; | |
efd3be9d YW |
3728 | if (r > 0) { |
3729 | something_new = true; | |
3730 | min_priority = MIN(min_priority, s->priority); | |
3731 | } | |
fd38203a LP |
3732 | } |
3733 | } | |
3734 | ||
efd3be9d YW |
3735 | *ret_min_priority = min_priority; |
3736 | return something_new; | |
fd38203a LP |
3737 | } |
3738 | ||
f8f3f926 LP |
3739 | static int process_pidfd(sd_event *e, sd_event_source *s, uint32_t revents) { |
3740 | assert(e); | |
3741 | assert(s); | |
3742 | assert(s->type == SOURCE_CHILD); | |
3743 | ||
3744 | if (s->pending) | |
3745 | return 0; | |
3746 | ||
b6d5481b | 3747 | if (event_source_is_offline(s)) |
f8f3f926 LP |
3748 | return 0; |
3749 | ||
3750 | if (!EVENT_SOURCE_WATCH_PIDFD(s)) | |
3751 | return 0; | |
3752 | ||
3753 | zero(s->child.siginfo); | |
3754 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG | WNOWAIT | s->child.options) < 0) | |
3755 | return -errno; | |
3756 | ||
3757 | if (s->child.siginfo.si_pid == 0) | |
3758 | return 0; | |
3759 | ||
3760 | if (IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) | |
3761 | s->child.exited = true; | |
3762 | ||
3763 | return source_set_pending(s, true); | |
3764 | } | |
3765 | ||
efd3be9d | 3766 | static int process_signal(sd_event *e, struct signal_data *d, uint32_t events, int64_t *min_priority) { |
fd38203a LP |
3767 | int r; |
3768 | ||
da7e457c | 3769 | assert(e); |
97ef5391 | 3770 | assert(d); |
305f78bf | 3771 | assert_return(events == EPOLLIN, -EIO); |
efd3be9d | 3772 | assert(min_priority); |
fd38203a | 3773 | |
91c70071 YW |
3774 | /* If there's a signal queued on this priority and SIGCHLD is on this priority too, then make |
3775 | * sure to recheck the children we watch. This is because we only ever dequeue the first signal | |
3776 | * per priority, and if we dequeue one, and SIGCHLD might be enqueued later we wouldn't know, | |
3777 | * but we might have higher priority children we care about hence we need to check that | |
3778 | * explicitly. */ | |
9da4cb2b LP |
3779 | |
3780 | if (sigismember(&d->sigset, SIGCHLD)) | |
3781 | e->need_process_child = true; | |
3782 | ||
91c70071 | 3783 | /* If there's already an event source pending for this priority we don't read another */ |
9da4cb2b LP |
3784 | if (d->current) |
3785 | return 0; | |
3786 | ||
fd38203a | 3787 | for (;;) { |
0eb2e0e3 | 3788 | struct signalfd_siginfo si; |
7057bd99 | 3789 | ssize_t n; |
92daebc0 | 3790 | sd_event_source *s = NULL; |
fd38203a | 3791 | |
9da4cb2b | 3792 | n = read(d->fd, &si, sizeof(si)); |
7057bd99 | 3793 | if (n < 0) { |
8add30a0 | 3794 | if (ERRNO_IS_TRANSIENT(errno)) |
efd3be9d | 3795 | return 0; |
fd38203a LP |
3796 | |
3797 | return -errno; | |
3798 | } | |
3799 | ||
7057bd99 | 3800 | if (_unlikely_(n != sizeof(si))) |
fd38203a LP |
3801 | return -EIO; |
3802 | ||
6eb7c172 | 3803 | assert(SIGNAL_VALID(si.ssi_signo)); |
7057bd99 | 3804 | |
92daebc0 LP |
3805 | if (e->signal_sources) |
3806 | s = e->signal_sources[si.ssi_signo]; | |
92daebc0 LP |
3807 | if (!s) |
3808 | continue; | |
9da4cb2b LP |
3809 | if (s->pending) |
3810 | continue; | |
fd38203a LP |
3811 | |
3812 | s->signal.siginfo = si; | |
9da4cb2b LP |
3813 | d->current = s; |
3814 | ||
fd38203a LP |
3815 | r = source_set_pending(s, true); |
3816 | if (r < 0) | |
3817 | return r; | |
efd3be9d YW |
3818 | if (r > 0 && *min_priority >= s->priority) { |
3819 | *min_priority = s->priority; | |
3820 | return 1; /* an event source with smaller priority is queued. */ | |
3821 | } | |
9da4cb2b | 3822 | |
efd3be9d | 3823 | return 0; |
fd38203a | 3824 | } |
fd38203a LP |
3825 | } |
3826 | ||
efd3be9d | 3827 | static int event_inotify_data_read(sd_event *e, struct inotify_data *d, uint32_t revents, int64_t threshold) { |
97ef5391 LP |
3828 | ssize_t n; |
3829 | ||
3830 | assert(e); | |
3831 | assert(d); | |
3832 | ||
3833 | assert_return(revents == EPOLLIN, -EIO); | |
3834 | ||
3835 | /* If there's already an event source pending for this priority, don't read another */ | |
3836 | if (d->n_pending > 0) | |
3837 | return 0; | |
3838 | ||
3839 | /* Is the read buffer non-empty? If so, let's not read more */ | |
3840 | if (d->buffer_filled > 0) | |
3841 | return 0; | |
3842 | ||
efd3be9d YW |
3843 | if (d->priority > threshold) |
3844 | return 0; | |
3845 | ||
97ef5391 LP |
3846 | n = read(d->fd, &d->buffer, sizeof(d->buffer)); |
3847 | if (n < 0) { | |
8add30a0 | 3848 | if (ERRNO_IS_TRANSIENT(errno)) |
97ef5391 LP |
3849 | return 0; |
3850 | ||
3851 | return -errno; | |
3852 | } | |
3853 | ||
3854 | assert(n > 0); | |
3855 | d->buffer_filled = (size_t) n; | |
0601b958 | 3856 | LIST_PREPEND(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
3857 | |
3858 | return 1; | |
3859 | } | |
3860 | ||
3861 | static void event_inotify_data_drop(sd_event *e, struct inotify_data *d, size_t sz) { | |
3862 | assert(e); | |
3863 | assert(d); | |
3864 | assert(sz <= d->buffer_filled); | |
3865 | ||
3866 | if (sz == 0) | |
3867 | return; | |
3868 | ||
3869 | /* Move the rest to the buffer to the front, in order to get things properly aligned again */ | |
3870 | memmove(d->buffer.raw, d->buffer.raw + sz, d->buffer_filled - sz); | |
3871 | d->buffer_filled -= sz; | |
3872 | ||
3873 | if (d->buffer_filled == 0) | |
0601b958 | 3874 | LIST_REMOVE(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
3875 | } |
3876 | ||
3877 | static int event_inotify_data_process(sd_event *e, struct inotify_data *d) { | |
3878 | int r; | |
3879 | ||
3880 | assert(e); | |
3881 | assert(d); | |
3882 | ||
3883 | /* If there's already an event source pending for this priority, don't read another */ | |
3884 | if (d->n_pending > 0) | |
3885 | return 0; | |
3886 | ||
3887 | while (d->buffer_filled > 0) { | |
3888 | size_t sz; | |
3889 | ||
3890 | /* Let's validate that the event structures are complete */ | |
3891 | if (d->buffer_filled < offsetof(struct inotify_event, name)) | |
3892 | return -EIO; | |
3893 | ||
3894 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3895 | if (d->buffer_filled < sz) | |
3896 | return -EIO; | |
3897 | ||
3898 | if (d->buffer.ev.mask & IN_Q_OVERFLOW) { | |
3899 | struct inode_data *inode_data; | |
97ef5391 LP |
3900 | |
3901 | /* The queue overran, let's pass this event to all event sources connected to this inotify | |
3902 | * object */ | |
3903 | ||
03677889 | 3904 | HASHMAP_FOREACH(inode_data, d->inodes) |
97ef5391 LP |
3905 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { |
3906 | ||
b6d5481b | 3907 | if (event_source_is_offline(s)) |
97ef5391 LP |
3908 | continue; |
3909 | ||
3910 | r = source_set_pending(s, true); | |
3911 | if (r < 0) | |
3912 | return r; | |
3913 | } | |
97ef5391 LP |
3914 | } else { |
3915 | struct inode_data *inode_data; | |
97ef5391 LP |
3916 | |
3917 | /* Find the inode object for this watch descriptor. If IN_IGNORED is set we also remove it from | |
3918 | * our watch descriptor table. */ | |
3919 | if (d->buffer.ev.mask & IN_IGNORED) { | |
3920 | ||
3921 | inode_data = hashmap_remove(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3922 | if (!inode_data) { | |
3923 | event_inotify_data_drop(e, d, sz); | |
3924 | continue; | |
3925 | } | |
3926 | ||
3927 | /* The watch descriptor was removed by the kernel, let's drop it here too */ | |
3928 | inode_data->wd = -1; | |
3929 | } else { | |
3930 | inode_data = hashmap_get(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3931 | if (!inode_data) { | |
3932 | event_inotify_data_drop(e, d, sz); | |
3933 | continue; | |
3934 | } | |
3935 | } | |
3936 | ||
3937 | /* Trigger all event sources that are interested in these events. Also trigger all event | |
3938 | * sources if IN_IGNORED or IN_UNMOUNT is set. */ | |
3939 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
3940 | ||
b6d5481b | 3941 | if (event_source_is_offline(s)) |
97ef5391 LP |
3942 | continue; |
3943 | ||
3944 | if ((d->buffer.ev.mask & (IN_IGNORED|IN_UNMOUNT)) == 0 && | |
3945 | (s->inotify.mask & d->buffer.ev.mask & IN_ALL_EVENTS) == 0) | |
3946 | continue; | |
3947 | ||
3948 | r = source_set_pending(s, true); | |
3949 | if (r < 0) | |
3950 | return r; | |
3951 | } | |
3952 | } | |
3953 | ||
3954 | /* Something pending now? If so, let's finish, otherwise let's read more. */ | |
3955 | if (d->n_pending > 0) | |
3956 | return 1; | |
3957 | } | |
3958 | ||
3959 | return 0; | |
3960 | } | |
3961 | ||
3962 | static int process_inotify(sd_event *e) { | |
97ef5391 LP |
3963 | int r, done = 0; |
3964 | ||
3965 | assert(e); | |
3966 | ||
0601b958 | 3967 | LIST_FOREACH(buffered, d, e->buffered_inotify_data_list) { |
97ef5391 LP |
3968 | r = event_inotify_data_process(e, d); |
3969 | if (r < 0) | |
3970 | return r; | |
3971 | if (r > 0) | |
3972 | done ++; | |
3973 | } | |
3974 | ||
3975 | return done; | |
3976 | } | |
3977 | ||
158fe190 LP |
3978 | static int process_memory_pressure(sd_event_source *s, uint32_t revents) { |
3979 | assert(s); | |
3980 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
3981 | ||
3982 | if (s->pending) | |
3983 | s->memory_pressure.revents |= revents; | |
3984 | else | |
3985 | s->memory_pressure.revents = revents; | |
3986 | ||
3987 | return source_set_pending(s, true); | |
3988 | } | |
3989 | ||
3990 | static int source_memory_pressure_write(sd_event_source *s) { | |
3991 | ssize_t n; | |
3992 | int r; | |
3993 | ||
3994 | assert(s); | |
3995 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
3996 | ||
3997 | /* once we start writing, the buffer is locked, we allow no further changes. */ | |
3998 | s->memory_pressure.locked = true; | |
3999 | ||
4000 | if (s->memory_pressure.write_buffer_size > 0) { | |
4001 | n = write(s->memory_pressure.fd, s->memory_pressure.write_buffer, s->memory_pressure.write_buffer_size); | |
4002 | if (n < 0) { | |
4003 | if (!ERRNO_IS_TRANSIENT(errno)) | |
4004 | return -errno; | |
4005 | ||
4006 | n = 0; | |
4007 | } | |
4008 | } else | |
4009 | n = 0; | |
4010 | ||
4011 | assert(n >= 0); | |
4012 | ||
4013 | if ((size_t) n == s->memory_pressure.write_buffer_size) { | |
4014 | s->memory_pressure.write_buffer = mfree(s->memory_pressure.write_buffer); | |
4015 | ||
4016 | if (n > 0) { | |
4017 | s->memory_pressure.write_buffer_size = 0; | |
4018 | ||
4019 | /* Update epoll events mask, since we have now written everything and don't care for EPOLLOUT anymore */ | |
4020 | r = source_memory_pressure_register(s, s->enabled); | |
4021 | if (r < 0) | |
4022 | return r; | |
4023 | } | |
4024 | } else if (n > 0) { | |
4025 | _cleanup_free_ void *c = NULL; | |
4026 | ||
4027 | assert((size_t) n < s->memory_pressure.write_buffer_size); | |
4028 | ||
4029 | c = memdup((uint8_t*) s->memory_pressure.write_buffer + n, s->memory_pressure.write_buffer_size - n); | |
4030 | if (!c) | |
4031 | return -ENOMEM; | |
4032 | ||
4033 | free_and_replace(s->memory_pressure.write_buffer, c); | |
4034 | s->memory_pressure.write_buffer_size -= n; | |
4035 | return 1; | |
4036 | } | |
4037 | ||
4038 | return 0; | |
4039 | } | |
4040 | ||
4041 | static int source_memory_pressure_initiate_dispatch(sd_event_source *s) { | |
4042 | int r; | |
4043 | ||
4044 | assert(s); | |
4045 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
4046 | ||
4047 | r = source_memory_pressure_write(s); | |
4048 | if (r < 0) | |
4049 | return r; | |
4050 | if (r > 0) | |
4051 | return 1; /* if we wrote something, then don't continue with dispatching user dispatch | |
4052 | * function. Instead, shortcut it so that we wait for next EPOLLOUT immediately. */ | |
4053 | ||
4054 | /* No pending incoming IO? Then let's not continue further */ | |
4055 | if ((s->memory_pressure.revents & (EPOLLIN|EPOLLPRI)) == 0) { | |
4056 | ||
4057 | /* Treat IO errors on the notifier the same ways errors returned from a callback */ | |
4058 | if ((s->memory_pressure.revents & (EPOLLHUP|EPOLLERR|EPOLLRDHUP)) != 0) | |
4059 | return -EIO; | |
4060 | ||
4061 | return 1; /* leave dispatch, we already processed everything */ | |
4062 | } | |
4063 | ||
4064 | if (s->memory_pressure.revents & EPOLLIN) { | |
4065 | uint8_t pipe_buf[PIPE_BUF]; | |
4066 | ssize_t n; | |
4067 | ||
4068 | /* If the fd is readable, then flush out anything that might be queued */ | |
4069 | ||
4070 | n = read(s->memory_pressure.fd, pipe_buf, sizeof(pipe_buf)); | |
4071 | if (n < 0 && !ERRNO_IS_TRANSIENT(errno)) | |
4072 | return -errno; | |
4073 | } | |
4074 | ||
4075 | return 0; /* go on, dispatch to user callback */ | |
4076 | } | |
4077 | ||
fd38203a | 4078 | static int source_dispatch(sd_event_source *s) { |
8f5c235d | 4079 | EventSourceType saved_type; |
c8e9d15c | 4080 | sd_event *saved_event; |
fe8245eb | 4081 | int r = 0; |
fd38203a LP |
4082 | |
4083 | assert(s); | |
6203e07a | 4084 | assert(s->pending || s->type == SOURCE_EXIT); |
fd38203a | 4085 | |
b778cba4 LP |
4086 | /* Save the event source type, here, so that we still know it after the event callback which might |
4087 | * invalidate the event. */ | |
8f5c235d LP |
4088 | saved_type = s->type; |
4089 | ||
de02634c | 4090 | /* Similarly, store a reference to the event loop object, so that we can still access it after the |
b778cba4 | 4091 | * callback might have invalidated/disconnected the event source. */ |
c8e9d15c YW |
4092 | saved_event = s->event; |
4093 | PROTECT_EVENT(saved_event); | |
b778cba4 | 4094 | |
de02634c | 4095 | /* Check if we hit the ratelimit for this event source, and if so, let's disable it. */ |
b6d5481b LP |
4096 | assert(!s->ratelimited); |
4097 | if (!ratelimit_below(&s->rate_limit)) { | |
4098 | r = event_source_enter_ratelimited(s); | |
4099 | if (r < 0) | |
4100 | return r; | |
4101 | ||
4102 | return 1; | |
4103 | } | |
4104 | ||
945c2931 | 4105 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { |
da7e457c LP |
4106 | r = source_set_pending(s, false); |
4107 | if (r < 0) | |
4108 | return r; | |
4109 | } | |
fd38203a | 4110 | |
6e9feda3 LP |
4111 | if (s->type != SOURCE_POST) { |
4112 | sd_event_source *z; | |
6e9feda3 | 4113 | |
de02634c | 4114 | /* If we execute a non-post source, let's mark all post sources as pending. */ |
6e9feda3 | 4115 | |
90e74a66 | 4116 | SET_FOREACH(z, s->event->post_sources) { |
b6d5481b | 4117 | if (event_source_is_offline(z)) |
6e9feda3 LP |
4118 | continue; |
4119 | ||
4120 | r = source_set_pending(z, true); | |
4121 | if (r < 0) | |
4122 | return r; | |
4123 | } | |
4124 | } | |
4125 | ||
158fe190 LP |
4126 | if (s->type == SOURCE_MEMORY_PRESSURE) { |
4127 | r = source_memory_pressure_initiate_dispatch(s); | |
4128 | if (r == -EIO) /* handle EIO errors similar to callback errors */ | |
4129 | goto finish; | |
4130 | if (r < 0) | |
4131 | return r; | |
4132 | if (r > 0) /* already handled */ | |
4133 | return 1; | |
4134 | } | |
4135 | ||
baf76283 LP |
4136 | if (s->enabled == SD_EVENT_ONESHOT) { |
4137 | r = sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
4138 | if (r < 0) |
4139 | return r; | |
4140 | } | |
4141 | ||
12179984 | 4142 | s->dispatching = true; |
b7484e2a | 4143 | |
fd38203a LP |
4144 | switch (s->type) { |
4145 | ||
4146 | case SOURCE_IO: | |
4147 | r = s->io.callback(s, s->io.fd, s->io.revents, s->userdata); | |
4148 | break; | |
4149 | ||
6a0f1f6d | 4150 | case SOURCE_TIME_REALTIME: |
a8548816 | 4151 | case SOURCE_TIME_BOOTTIME: |
6a0f1f6d LP |
4152 | case SOURCE_TIME_MONOTONIC: |
4153 | case SOURCE_TIME_REALTIME_ALARM: | |
4154 | case SOURCE_TIME_BOOTTIME_ALARM: | |
fd38203a LP |
4155 | r = s->time.callback(s, s->time.next, s->userdata); |
4156 | break; | |
4157 | ||
4158 | case SOURCE_SIGNAL: | |
4159 | r = s->signal.callback(s, &s->signal.siginfo, s->userdata); | |
4160 | break; | |
4161 | ||
08cd1552 LP |
4162 | case SOURCE_CHILD: { |
4163 | bool zombie; | |
4164 | ||
945c2931 | 4165 | zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 4166 | |
fd38203a | 4167 | r = s->child.callback(s, &s->child.siginfo, s->userdata); |
08cd1552 LP |
4168 | |
4169 | /* Now, reap the PID for good. */ | |
f8f3f926 | 4170 | if (zombie) { |
cc59d290 | 4171 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|WEXITED); |
f8f3f926 LP |
4172 | s->child.waited = true; |
4173 | } | |
08cd1552 | 4174 | |
fd38203a | 4175 | break; |
08cd1552 | 4176 | } |
fd38203a LP |
4177 | |
4178 | case SOURCE_DEFER: | |
4179 | r = s->defer.callback(s, s->userdata); | |
4180 | break; | |
da7e457c | 4181 | |
6e9feda3 LP |
4182 | case SOURCE_POST: |
4183 | r = s->post.callback(s, s->userdata); | |
4184 | break; | |
4185 | ||
6203e07a LP |
4186 | case SOURCE_EXIT: |
4187 | r = s->exit.callback(s, s->userdata); | |
da7e457c | 4188 | break; |
9d3e3aa5 | 4189 | |
97ef5391 LP |
4190 | case SOURCE_INOTIFY: { |
4191 | struct sd_event *e = s->event; | |
4192 | struct inotify_data *d; | |
4193 | size_t sz; | |
4194 | ||
4195 | assert(s->inotify.inode_data); | |
4196 | assert_se(d = s->inotify.inode_data->inotify_data); | |
4197 | ||
4198 | assert(d->buffer_filled >= offsetof(struct inotify_event, name)); | |
4199 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
4200 | assert(d->buffer_filled >= sz); | |
4201 | ||
53baf2ef LP |
4202 | /* If the inotify callback destroys the event source then this likely means we don't need to |
4203 | * watch the inode anymore, and thus also won't need the inotify object anymore. But if we'd | |
4204 | * free it immediately, then we couldn't drop the event from the inotify event queue without | |
4205 | * memory corruption anymore, as below. Hence, let's not free it immediately, but mark it | |
4206 | * "busy" with a counter (which will ensure it's not GC'ed away prematurely). Let's then | |
4207 | * explicitly GC it after we are done dropping the inotify event from the buffer. */ | |
4208 | d->n_busy++; | |
97ef5391 | 4209 | r = s->inotify.callback(s, &d->buffer.ev, s->userdata); |
53baf2ef | 4210 | d->n_busy--; |
97ef5391 | 4211 | |
53baf2ef LP |
4212 | /* When no event is pending anymore on this inotify object, then let's drop the event from |
4213 | * the inotify event queue buffer. */ | |
97ef5391 LP |
4214 | if (d->n_pending == 0) |
4215 | event_inotify_data_drop(e, d, sz); | |
4216 | ||
53baf2ef LP |
4217 | /* Now we don't want to access 'd' anymore, it's OK to GC now. */ |
4218 | event_gc_inotify_data(e, d); | |
97ef5391 LP |
4219 | break; |
4220 | } | |
4221 | ||
158fe190 LP |
4222 | case SOURCE_MEMORY_PRESSURE: |
4223 | r = s->memory_pressure.callback(s, s->userdata); | |
4224 | break; | |
4225 | ||
9d3e3aa5 | 4226 | case SOURCE_WATCHDOG: |
a71fe8b8 | 4227 | case _SOURCE_EVENT_SOURCE_TYPE_MAX: |
9f2a50a3 | 4228 | case _SOURCE_EVENT_SOURCE_TYPE_INVALID: |
04499a70 | 4229 | assert_not_reached(); |
fd38203a LP |
4230 | } |
4231 | ||
12179984 LP |
4232 | s->dispatching = false; |
4233 | ||
158fe190 | 4234 | finish: |
b778cba4 LP |
4235 | if (r < 0) { |
4236 | log_debug_errno(r, "Event source %s (type %s) returned error, %s: %m", | |
4237 | strna(s->description), | |
4238 | event_source_type_to_string(saved_type), | |
4239 | s->exit_on_failure ? "exiting" : "disabling"); | |
4240 | ||
4241 | if (s->exit_on_failure) | |
4242 | (void) sd_event_exit(saved_event, r); | |
4243 | } | |
12179984 LP |
4244 | |
4245 | if (s->n_ref == 0) | |
4246 | source_free(s); | |
4247 | else if (r < 0) | |
c3c50474 | 4248 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
b7484e2a | 4249 | |
6203e07a | 4250 | return 1; |
fd38203a LP |
4251 | } |
4252 | ||
4253 | static int event_prepare(sd_event *e) { | |
4254 | int r; | |
4255 | ||
4256 | assert(e); | |
4257 | ||
4258 | for (;;) { | |
4259 | sd_event_source *s; | |
4260 | ||
4261 | s = prioq_peek(e->prepare); | |
b6d5481b | 4262 | if (!s || s->prepare_iteration == e->iteration || event_source_is_offline(s)) |
fd38203a LP |
4263 | break; |
4264 | ||
4265 | s->prepare_iteration = e->iteration; | |
8656f4a6 | 4266 | prioq_reshuffle(e->prepare, s, &s->prepare_index); |
fd38203a LP |
4267 | |
4268 | assert(s->prepare); | |
12179984 | 4269 | s->dispatching = true; |
fd38203a | 4270 | r = s->prepare(s, s->userdata); |
12179984 LP |
4271 | s->dispatching = false; |
4272 | ||
b778cba4 LP |
4273 | if (r < 0) { |
4274 | log_debug_errno(r, "Prepare callback of event source %s (type %s) returned error, %s: %m", | |
4275 | strna(s->description), | |
4276 | event_source_type_to_string(s->type), | |
4277 | s->exit_on_failure ? "exiting" : "disabling"); | |
4278 | ||
4279 | if (s->exit_on_failure) | |
4280 | (void) sd_event_exit(e, r); | |
4281 | } | |
fd38203a | 4282 | |
12179984 LP |
4283 | if (s->n_ref == 0) |
4284 | source_free(s); | |
4285 | else if (r < 0) | |
c3c50474 | 4286 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
fd38203a LP |
4287 | } |
4288 | ||
4289 | return 0; | |
4290 | } | |
4291 | ||
6203e07a | 4292 | static int dispatch_exit(sd_event *e) { |
da7e457c LP |
4293 | sd_event_source *p; |
4294 | int r; | |
4295 | ||
4296 | assert(e); | |
4297 | ||
6203e07a | 4298 | p = prioq_peek(e->exit); |
19947509 ZJS |
4299 | assert(!p || p->type == SOURCE_EXIT); |
4300 | ||
b6d5481b | 4301 | if (!p || event_source_is_offline(p)) { |
da7e457c LP |
4302 | e->state = SD_EVENT_FINISHED; |
4303 | return 0; | |
4304 | } | |
4305 | ||
c8e9d15c | 4306 | PROTECT_EVENT(e); |
da7e457c | 4307 | e->iteration++; |
6203e07a | 4308 | e->state = SD_EVENT_EXITING; |
da7e457c | 4309 | r = source_dispatch(p); |
2b0c9ef7 | 4310 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
4311 | return r; |
4312 | } | |
4313 | ||
c2ba3ad6 LP |
4314 | static sd_event_source* event_next_pending(sd_event *e) { |
4315 | sd_event_source *p; | |
4316 | ||
da7e457c LP |
4317 | assert(e); |
4318 | ||
c2ba3ad6 LP |
4319 | p = prioq_peek(e->pending); |
4320 | if (!p) | |
4321 | return NULL; | |
4322 | ||
b6d5481b | 4323 | if (event_source_is_offline(p)) |
c2ba3ad6 LP |
4324 | return NULL; |
4325 | ||
4326 | return p; | |
4327 | } | |
4328 | ||
cde93897 LP |
4329 | static int arm_watchdog(sd_event *e) { |
4330 | struct itimerspec its = {}; | |
4331 | usec_t t; | |
cde93897 LP |
4332 | |
4333 | assert(e); | |
4334 | assert(e->watchdog_fd >= 0); | |
4335 | ||
4336 | t = sleep_between(e, | |
a595fb5c YW |
4337 | usec_add(e->watchdog_last, (e->watchdog_period / 2)), |
4338 | usec_add(e->watchdog_last, (e->watchdog_period * 3 / 4))); | |
cde93897 LP |
4339 | |
4340 | timespec_store(&its.it_value, t); | |
4341 | ||
75145780 LP |
4342 | /* Make sure we never set the watchdog to 0, which tells the |
4343 | * kernel to disable it. */ | |
4344 | if (its.it_value.tv_sec == 0 && its.it_value.tv_nsec == 0) | |
4345 | its.it_value.tv_nsec = 1; | |
4346 | ||
7c248223 | 4347 | return RET_NERRNO(timerfd_settime(e->watchdog_fd, TFD_TIMER_ABSTIME, &its, NULL)); |
cde93897 LP |
4348 | } |
4349 | ||
4350 | static int process_watchdog(sd_event *e) { | |
4351 | assert(e); | |
4352 | ||
4353 | if (!e->watchdog) | |
4354 | return 0; | |
4355 | ||
4356 | /* Don't notify watchdog too often */ | |
4357 | if (e->watchdog_last + e->watchdog_period / 4 > e->timestamp.monotonic) | |
4358 | return 0; | |
4359 | ||
4360 | sd_notify(false, "WATCHDOG=1"); | |
4361 | e->watchdog_last = e->timestamp.monotonic; | |
4362 | ||
4363 | return arm_watchdog(e); | |
4364 | } | |
4365 | ||
97ef5391 LP |
4366 | static void event_close_inode_data_fds(sd_event *e) { |
4367 | struct inode_data *d; | |
4368 | ||
4369 | assert(e); | |
4370 | ||
4371 | /* Close the fds pointing to the inodes to watch now. We need to close them as they might otherwise pin | |
4372 | * filesystems. But we can't close them right-away as we need them as long as the user still wants to make | |
365c2885 | 4373 | * adjustments to the event source, such as changing the priority (which requires us to remove and re-add a watch |
97ef5391 LP |
4374 | * for the inode). Hence, let's close them when entering the first iteration after they were added, as a |
4375 | * compromise. */ | |
4376 | ||
ed828563 | 4377 | while ((d = e->inode_data_to_close_list)) { |
97ef5391 LP |
4378 | assert(d->fd >= 0); |
4379 | d->fd = safe_close(d->fd); | |
4380 | ||
ed828563 | 4381 | LIST_REMOVE(to_close, e->inode_data_to_close_list, d); |
97ef5391 LP |
4382 | } |
4383 | } | |
4384 | ||
158fe190 LP |
4385 | static int event_memory_pressure_write_list(sd_event *e) { |
4386 | int r; | |
4387 | ||
4388 | assert(e); | |
4389 | ||
4390 | for (;;) { | |
4391 | sd_event_source *s; | |
4392 | ||
4393 | s = LIST_POP(memory_pressure.write_list, e->memory_pressure_write_list); | |
4394 | if (!s) | |
4395 | break; | |
4396 | ||
4397 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
4398 | assert(s->memory_pressure.write_buffer_size > 0); | |
4399 | s->memory_pressure.in_write_list = false; | |
4400 | ||
4401 | r = source_memory_pressure_write(s); | |
4402 | if (r < 0) | |
4403 | return r; | |
4404 | } | |
4405 | ||
4406 | return 0; | |
4407 | } | |
4408 | ||
c45a5a74 TG |
4409 | _public_ int sd_event_prepare(sd_event *e) { |
4410 | int r; | |
fd38203a | 4411 | |
da7e457c | 4412 | assert_return(e, -EINVAL); |
b937d761 | 4413 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
4414 | assert_return(!event_pid_changed(e), -ECHILD); |
4415 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 4416 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 4417 | |
e5446015 LP |
4418 | /* Let's check that if we are a default event loop we are executed in the correct thread. We only do |
4419 | * this check here once, since gettid() is typically not cached, and thus want to minimize | |
4420 | * syscalls */ | |
4421 | assert_return(!e->default_event_ptr || e->tid == gettid(), -EREMOTEIO); | |
4422 | ||
f814c871 | 4423 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
c8e9d15c | 4424 | PROTECT_EVENT(e); |
f814c871 | 4425 | |
6203e07a | 4426 | if (e->exit_requested) |
c45a5a74 | 4427 | goto pending; |
fd38203a LP |
4428 | |
4429 | e->iteration++; | |
4430 | ||
0be6c2f6 | 4431 | e->state = SD_EVENT_PREPARING; |
fd38203a | 4432 | r = event_prepare(e); |
0be6c2f6 | 4433 | e->state = SD_EVENT_INITIAL; |
fd38203a | 4434 | if (r < 0) |
c45a5a74 | 4435 | return r; |
fd38203a | 4436 | |
158fe190 LP |
4437 | r = event_memory_pressure_write_list(e); |
4438 | if (r < 0) | |
4439 | return r; | |
4440 | ||
6a0f1f6d LP |
4441 | r = event_arm_timer(e, &e->realtime); |
4442 | if (r < 0) | |
c45a5a74 | 4443 | return r; |
6a0f1f6d | 4444 | |
a8548816 TG |
4445 | r = event_arm_timer(e, &e->boottime); |
4446 | if (r < 0) | |
c45a5a74 | 4447 | return r; |
a8548816 | 4448 | |
6a0f1f6d LP |
4449 | r = event_arm_timer(e, &e->monotonic); |
4450 | if (r < 0) | |
c45a5a74 | 4451 | return r; |
6a0f1f6d LP |
4452 | |
4453 | r = event_arm_timer(e, &e->realtime_alarm); | |
1b5995b0 | 4454 | if (r < 0) |
c45a5a74 | 4455 | return r; |
fd38203a | 4456 | |
6a0f1f6d | 4457 | r = event_arm_timer(e, &e->boottime_alarm); |
1b5995b0 | 4458 | if (r < 0) |
c45a5a74 | 4459 | return r; |
fd38203a | 4460 | |
97ef5391 LP |
4461 | event_close_inode_data_fds(e); |
4462 | ||
0601b958 | 4463 | if (event_next_pending(e) || e->need_process_child || e->buffered_inotify_data_list) |
c45a5a74 TG |
4464 | goto pending; |
4465 | ||
2b0c9ef7 | 4466 | e->state = SD_EVENT_ARMED; |
c45a5a74 TG |
4467 | |
4468 | return 0; | |
4469 | ||
4470 | pending: | |
2b0c9ef7 | 4471 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
4472 | r = sd_event_wait(e, 0); |
4473 | if (r == 0) | |
2b0c9ef7 | 4474 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
4475 | |
4476 | return r; | |
c45a5a74 TG |
4477 | } |
4478 | ||
798445ab LP |
4479 | static int epoll_wait_usec( |
4480 | int fd, | |
4481 | struct epoll_event *events, | |
4482 | int maxevents, | |
4483 | usec_t timeout) { | |
4484 | ||
7c248223 | 4485 | int msec; |
0c14c45e LP |
4486 | /* A wrapper that uses epoll_pwait2() if available, and falls back to epoll_wait() if not. */ |
4487 | ||
4488 | #if HAVE_EPOLL_PWAIT2 | |
39f756d3 | 4489 | static bool epoll_pwait2_absent = false; |
52bb308c | 4490 | int r; |
798445ab | 4491 | |
0c14c45e LP |
4492 | /* epoll_pwait2() was added to Linux 5.11 (2021-02-14) and to glibc in 2.35 (2022-02-03). In contrast |
4493 | * to other syscalls we don't bother with our own fallback syscall wrappers on old libcs, since this | |
4494 | * is not that obvious to implement given the libc and kernel definitions differ in the last | |
4495 | * argument. Moreover, the only reason to use it is the more accurate time-outs (which is not a | |
4496 | * biggie), let's hence rely on glibc's definitions, and fallback to epoll_pwait() when that's | |
4497 | * missing. */ | |
798445ab LP |
4498 | |
4499 | if (!epoll_pwait2_absent && timeout != USEC_INFINITY) { | |
798445ab LP |
4500 | r = epoll_pwait2(fd, |
4501 | events, | |
4502 | maxevents, | |
52bb308c | 4503 | TIMESPEC_STORE(timeout), |
798445ab LP |
4504 | NULL); |
4505 | if (r >= 0) | |
4506 | return r; | |
7cb45dbf | 4507 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) |
798445ab LP |
4508 | return -errno; /* Only fallback to old epoll_wait() if the syscall is masked or not |
4509 | * supported. */ | |
4510 | ||
4511 | epoll_pwait2_absent = true; | |
4512 | } | |
39f756d3 | 4513 | #endif |
798445ab LP |
4514 | |
4515 | if (timeout == USEC_INFINITY) | |
4516 | msec = -1; | |
4517 | else { | |
4518 | usec_t k; | |
4519 | ||
4520 | k = DIV_ROUND_UP(timeout, USEC_PER_MSEC); | |
4521 | if (k >= INT_MAX) | |
4522 | msec = INT_MAX; /* Saturate */ | |
4523 | else | |
4524 | msec = (int) k; | |
4525 | } | |
4526 | ||
7c248223 | 4527 | return RET_NERRNO(epoll_wait(fd, events, maxevents, msec)); |
798445ab LP |
4528 | } |
4529 | ||
efd3be9d | 4530 | static int process_epoll(sd_event *e, usec_t timeout, int64_t threshold, int64_t *ret_min_priority) { |
319a4f4b | 4531 | size_t n_event_queue, m, n_event_max; |
efd3be9d YW |
4532 | int64_t min_priority = threshold; |
4533 | bool something_new = false; | |
798445ab | 4534 | int r; |
c45a5a74 | 4535 | |
efd3be9d YW |
4536 | assert(e); |
4537 | assert(ret_min_priority); | |
6a0f1f6d | 4538 | |
8b9708d1 | 4539 | n_event_queue = MAX(e->n_sources, 1u); |
319a4f4b | 4540 | if (!GREEDY_REALLOC(e->event_queue, n_event_queue)) |
5cddd924 | 4541 | return -ENOMEM; |
fd38203a | 4542 | |
319a4f4b LP |
4543 | n_event_max = MALLOC_ELEMENTSOF(e->event_queue); |
4544 | ||
97ef5391 | 4545 | /* If we still have inotify data buffered, then query the other fds, but don't wait on it */ |
0601b958 | 4546 | if (e->buffered_inotify_data_list) |
798445ab | 4547 | timeout = 0; |
97ef5391 | 4548 | |
8b9708d1 | 4549 | for (;;) { |
319a4f4b LP |
4550 | r = epoll_wait_usec( |
4551 | e->epoll_fd, | |
4552 | e->event_queue, | |
4553 | n_event_max, | |
4554 | timeout); | |
798445ab | 4555 | if (r < 0) |
efd3be9d | 4556 | return r; |
c45a5a74 | 4557 | |
8b9708d1 YW |
4558 | m = (size_t) r; |
4559 | ||
319a4f4b | 4560 | if (m < n_event_max) |
8b9708d1 YW |
4561 | break; |
4562 | ||
319a4f4b | 4563 | if (n_event_max >= n_event_queue * 10) |
8b9708d1 YW |
4564 | break; |
4565 | ||
319a4f4b | 4566 | if (!GREEDY_REALLOC(e->event_queue, n_event_max + n_event_queue)) |
8b9708d1 YW |
4567 | return -ENOMEM; |
4568 | ||
319a4f4b | 4569 | n_event_max = MALLOC_ELEMENTSOF(e->event_queue); |
798445ab | 4570 | timeout = 0; |
da7e457c | 4571 | } |
fd38203a | 4572 | |
efd3be9d YW |
4573 | /* Set timestamp only when this is called first time. */ |
4574 | if (threshold == INT64_MAX) | |
4575 | triple_timestamp_get(&e->timestamp); | |
fd38203a | 4576 | |
8b9708d1 | 4577 | for (size_t i = 0; i < m; i++) { |
fd38203a | 4578 | |
5cddd924 LP |
4579 | if (e->event_queue[i].data.ptr == INT_TO_PTR(SOURCE_WATCHDOG)) |
4580 | r = flush_timer(e, e->watchdog_fd, e->event_queue[i].events, NULL); | |
9da4cb2b | 4581 | else { |
5cddd924 | 4582 | WakeupType *t = e->event_queue[i].data.ptr; |
9da4cb2b LP |
4583 | |
4584 | switch (*t) { | |
4585 | ||
f8f3f926 | 4586 | case WAKEUP_EVENT_SOURCE: { |
5cddd924 | 4587 | sd_event_source *s = e->event_queue[i].data.ptr; |
f8f3f926 LP |
4588 | |
4589 | assert(s); | |
4590 | ||
efd3be9d YW |
4591 | if (s->priority > threshold) |
4592 | continue; | |
4593 | ||
4594 | min_priority = MIN(min_priority, s->priority); | |
4595 | ||
f8f3f926 LP |
4596 | switch (s->type) { |
4597 | ||
4598 | case SOURCE_IO: | |
5cddd924 | 4599 | r = process_io(e, s, e->event_queue[i].events); |
f8f3f926 LP |
4600 | break; |
4601 | ||
4602 | case SOURCE_CHILD: | |
5cddd924 | 4603 | r = process_pidfd(e, s, e->event_queue[i].events); |
f8f3f926 LP |
4604 | break; |
4605 | ||
158fe190 LP |
4606 | case SOURCE_MEMORY_PRESSURE: |
4607 | r = process_memory_pressure(s, e->event_queue[i].events); | |
4608 | break; | |
4609 | ||
f8f3f926 | 4610 | default: |
04499a70 | 4611 | assert_not_reached(); |
f8f3f926 LP |
4612 | } |
4613 | ||
9da4cb2b | 4614 | break; |
f8f3f926 | 4615 | } |
fd38203a | 4616 | |
9da4cb2b | 4617 | case WAKEUP_CLOCK_DATA: { |
5cddd924 | 4618 | struct clock_data *d = e->event_queue[i].data.ptr; |
f8f3f926 LP |
4619 | |
4620 | assert(d); | |
4621 | ||
5cddd924 | 4622 | r = flush_timer(e, d->fd, e->event_queue[i].events, &d->next); |
9da4cb2b LP |
4623 | break; |
4624 | } | |
4625 | ||
4626 | case WAKEUP_SIGNAL_DATA: | |
efd3be9d | 4627 | r = process_signal(e, e->event_queue[i].data.ptr, e->event_queue[i].events, &min_priority); |
9da4cb2b LP |
4628 | break; |
4629 | ||
97ef5391 | 4630 | case WAKEUP_INOTIFY_DATA: |
efd3be9d | 4631 | r = event_inotify_data_read(e, e->event_queue[i].data.ptr, e->event_queue[i].events, threshold); |
97ef5391 LP |
4632 | break; |
4633 | ||
9da4cb2b | 4634 | default: |
04499a70 | 4635 | assert_not_reached(); |
9da4cb2b LP |
4636 | } |
4637 | } | |
efd3be9d YW |
4638 | if (r < 0) |
4639 | return r; | |
4640 | if (r > 0) | |
4641 | something_new = true; | |
4642 | } | |
4643 | ||
4644 | *ret_min_priority = min_priority; | |
4645 | return something_new; | |
4646 | } | |
4647 | ||
4648 | _public_ int sd_event_wait(sd_event *e, uint64_t timeout) { | |
4649 | int r; | |
4650 | ||
4651 | assert_return(e, -EINVAL); | |
4652 | assert_return(e = event_resolve(e), -ENOPKG); | |
4653 | assert_return(!event_pid_changed(e), -ECHILD); | |
4654 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
4655 | assert_return(e->state == SD_EVENT_ARMED, -EBUSY); | |
4656 | ||
4657 | if (e->exit_requested) { | |
4658 | e->state = SD_EVENT_PENDING; | |
4659 | return 1; | |
4660 | } | |
4661 | ||
4662 | for (int64_t threshold = INT64_MAX; ; threshold--) { | |
4663 | int64_t epoll_min_priority, child_min_priority; | |
4664 | ||
4665 | /* There may be a possibility that new epoll (especially IO) and child events are | |
4666 | * triggered just after process_epoll() call but before process_child(), and the new IO | |
4667 | * events may have higher priority than the child events. To salvage these events, | |
4668 | * let's call epoll_wait() again, but accepts only events with higher priority than the | |
4669 | * previous. See issue https://github.com/systemd/systemd/issues/18190 and comments | |
4670 | * https://github.com/systemd/systemd/pull/18750#issuecomment-785801085 | |
4671 | * https://github.com/systemd/systemd/pull/18922#issuecomment-792825226 */ | |
4672 | ||
4673 | r = process_epoll(e, timeout, threshold, &epoll_min_priority); | |
4674 | if (r == -EINTR) { | |
4675 | e->state = SD_EVENT_PENDING; | |
4676 | return 1; | |
4677 | } | |
4678 | if (r < 0) | |
4679 | goto finish; | |
4680 | if (r == 0 && threshold < INT64_MAX) | |
4681 | /* No new epoll event. */ | |
4682 | break; | |
4683 | ||
4684 | r = process_child(e, threshold, &child_min_priority); | |
fd38203a | 4685 | if (r < 0) |
da7e457c | 4686 | goto finish; |
efd3be9d YW |
4687 | if (r == 0) |
4688 | /* No new child event. */ | |
4689 | break; | |
4690 | ||
4691 | threshold = MIN(epoll_min_priority, child_min_priority); | |
4692 | if (threshold == INT64_MIN) | |
4693 | break; | |
4694 | ||
4695 | timeout = 0; | |
fd38203a LP |
4696 | } |
4697 | ||
cde93897 LP |
4698 | r = process_watchdog(e); |
4699 | if (r < 0) | |
4700 | goto finish; | |
4701 | ||
fd69f224 | 4702 | r = process_inotify(e); |
6a0f1f6d LP |
4703 | if (r < 0) |
4704 | goto finish; | |
4705 | ||
fd69f224 | 4706 | r = process_timer(e, e->timestamp.realtime, &e->realtime); |
a8548816 TG |
4707 | if (r < 0) |
4708 | goto finish; | |
4709 | ||
fd69f224 | 4710 | r = process_timer(e, e->timestamp.boottime, &e->boottime); |
6a0f1f6d LP |
4711 | if (r < 0) |
4712 | goto finish; | |
4713 | ||
4714 | r = process_timer(e, e->timestamp.realtime, &e->realtime_alarm); | |
fd38203a | 4715 | if (r < 0) |
da7e457c | 4716 | goto finish; |
fd38203a | 4717 | |
e475d10c | 4718 | r = process_timer(e, e->timestamp.boottime, &e->boottime_alarm); |
fd38203a | 4719 | if (r < 0) |
da7e457c | 4720 | goto finish; |
fd38203a | 4721 | |
fd69f224 | 4722 | r = process_timer(e, e->timestamp.monotonic, &e->monotonic); |
97ef5391 LP |
4723 | if (r < 0) |
4724 | goto finish; | |
fd69f224 MS |
4725 | else if (r == 1) { |
4726 | /* Ratelimit expiry callback was called. Let's postpone processing pending sources and | |
4727 | * put loop in the initial state in order to evaluate (in the next iteration) also sources | |
4728 | * there were potentially re-enabled by the callback. | |
4729 | * | |
4730 | * Wondering why we treat only this invocation of process_timer() differently? Once event | |
4731 | * source is ratelimited we essentially transform it into CLOCK_MONOTONIC timer hence | |
4732 | * ratelimit expiry callback is never called for any other timer type. */ | |
4733 | r = 0; | |
4734 | goto finish; | |
4735 | } | |
97ef5391 | 4736 | |
c45a5a74 TG |
4737 | if (event_next_pending(e)) { |
4738 | e->state = SD_EVENT_PENDING; | |
c45a5a74 | 4739 | return 1; |
da7e457c LP |
4740 | } |
4741 | ||
c45a5a74 | 4742 | r = 0; |
fd38203a | 4743 | |
da7e457c | 4744 | finish: |
2b0c9ef7 | 4745 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
4746 | |
4747 | return r; | |
fd38203a LP |
4748 | } |
4749 | ||
c45a5a74 TG |
4750 | _public_ int sd_event_dispatch(sd_event *e) { |
4751 | sd_event_source *p; | |
4752 | int r; | |
4753 | ||
4754 | assert_return(e, -EINVAL); | |
b937d761 | 4755 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
4756 | assert_return(!event_pid_changed(e), -ECHILD); |
4757 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
4758 | assert_return(e->state == SD_EVENT_PENDING, -EBUSY); | |
4759 | ||
4760 | if (e->exit_requested) | |
4761 | return dispatch_exit(e); | |
4762 | ||
4763 | p = event_next_pending(e); | |
4764 | if (p) { | |
c8e9d15c | 4765 | PROTECT_EVENT(e); |
c45a5a74 TG |
4766 | |
4767 | e->state = SD_EVENT_RUNNING; | |
4768 | r = source_dispatch(p); | |
2b0c9ef7 | 4769 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
4770 | return r; |
4771 | } | |
4772 | ||
2b0c9ef7 | 4773 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
4774 | |
4775 | return 1; | |
4776 | } | |
4777 | ||
34b87517 | 4778 | static void event_log_delays(sd_event *e) { |
442ac269 YW |
4779 | char b[ELEMENTSOF(e->delays) * DECIMAL_STR_MAX(unsigned) + 1], *p; |
4780 | size_t l, i; | |
34b87517 | 4781 | |
442ac269 YW |
4782 | p = b; |
4783 | l = sizeof(b); | |
4784 | for (i = 0; i < ELEMENTSOF(e->delays); i++) { | |
4785 | l = strpcpyf(&p, l, "%u ", e->delays[i]); | |
34b87517 VC |
4786 | e->delays[i] = 0; |
4787 | } | |
442ac269 | 4788 | log_debug("Event loop iterations: %s", b); |
34b87517 VC |
4789 | } |
4790 | ||
c45a5a74 TG |
4791 | _public_ int sd_event_run(sd_event *e, uint64_t timeout) { |
4792 | int r; | |
4793 | ||
4794 | assert_return(e, -EINVAL); | |
b937d761 | 4795 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
4796 | assert_return(!event_pid_changed(e), -ECHILD); |
4797 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 4798 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
c45a5a74 | 4799 | |
e6a7bee5 | 4800 | if (e->profile_delays && e->last_run_usec != 0) { |
34b87517 VC |
4801 | usec_t this_run; |
4802 | unsigned l; | |
4803 | ||
4804 | this_run = now(CLOCK_MONOTONIC); | |
4805 | ||
58c34be8 | 4806 | l = log2u64(this_run - e->last_run_usec); |
cb9d621e | 4807 | assert(l < ELEMENTSOF(e->delays)); |
34b87517 VC |
4808 | e->delays[l]++; |
4809 | ||
e6a7bee5 | 4810 | if (this_run - e->last_log_usec >= 5*USEC_PER_SEC) { |
34b87517 | 4811 | event_log_delays(e); |
e6a7bee5 | 4812 | e->last_log_usec = this_run; |
34b87517 VC |
4813 | } |
4814 | } | |
4815 | ||
f814c871 | 4816 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
c8e9d15c | 4817 | PROTECT_EVENT(e); |
f814c871 | 4818 | |
c45a5a74 | 4819 | r = sd_event_prepare(e); |
53bac4e0 LP |
4820 | if (r == 0) |
4821 | /* There was nothing? Then wait... */ | |
4822 | r = sd_event_wait(e, timeout); | |
c45a5a74 | 4823 | |
34b87517 | 4824 | if (e->profile_delays) |
e6a7bee5 | 4825 | e->last_run_usec = now(CLOCK_MONOTONIC); |
34b87517 | 4826 | |
02d30981 | 4827 | if (r > 0) { |
53bac4e0 | 4828 | /* There's something now, then let's dispatch it */ |
02d30981 TG |
4829 | r = sd_event_dispatch(e); |
4830 | if (r < 0) | |
4831 | return r; | |
53bac4e0 LP |
4832 | |
4833 | return 1; | |
4834 | } | |
4835 | ||
4836 | return r; | |
c45a5a74 TG |
4837 | } |
4838 | ||
f7262a9f | 4839 | _public_ int sd_event_loop(sd_event *e) { |
fd38203a LP |
4840 | int r; |
4841 | ||
da7e457c | 4842 | assert_return(e, -EINVAL); |
b937d761 | 4843 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 4844 | assert_return(!event_pid_changed(e), -ECHILD); |
2b0c9ef7 | 4845 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 4846 | |
c8e9d15c | 4847 | PROTECT_EVENT(e); |
fd38203a | 4848 | |
da7e457c | 4849 | while (e->state != SD_EVENT_FINISHED) { |
f5fbe71d | 4850 | r = sd_event_run(e, UINT64_MAX); |
fd38203a | 4851 | if (r < 0) |
30dd293c | 4852 | return r; |
fd38203a LP |
4853 | } |
4854 | ||
30dd293c | 4855 | return e->exit_code; |
fd38203a LP |
4856 | } |
4857 | ||
9b364545 | 4858 | _public_ int sd_event_get_fd(sd_event *e) { |
9b364545 | 4859 | assert_return(e, -EINVAL); |
b937d761 | 4860 | assert_return(e = event_resolve(e), -ENOPKG); |
9b364545 TG |
4861 | assert_return(!event_pid_changed(e), -ECHILD); |
4862 | ||
4863 | return e->epoll_fd; | |
4864 | } | |
4865 | ||
f7262a9f | 4866 | _public_ int sd_event_get_state(sd_event *e) { |
da7e457c | 4867 | assert_return(e, -EINVAL); |
b937d761 | 4868 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
4869 | assert_return(!event_pid_changed(e), -ECHILD); |
4870 | ||
4871 | return e->state; | |
4872 | } | |
4873 | ||
6203e07a | 4874 | _public_ int sd_event_get_exit_code(sd_event *e, int *code) { |
da7e457c | 4875 | assert_return(e, -EINVAL); |
b937d761 | 4876 | assert_return(e = event_resolve(e), -ENOPKG); |
6203e07a | 4877 | assert_return(code, -EINVAL); |
da7e457c | 4878 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 4879 | |
6203e07a LP |
4880 | if (!e->exit_requested) |
4881 | return -ENODATA; | |
4882 | ||
4883 | *code = e->exit_code; | |
4884 | return 0; | |
fd38203a LP |
4885 | } |
4886 | ||
6203e07a | 4887 | _public_ int sd_event_exit(sd_event *e, int code) { |
da7e457c | 4888 | assert_return(e, -EINVAL); |
b937d761 | 4889 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
4890 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
4891 | assert_return(!event_pid_changed(e), -ECHILD); | |
fd38203a | 4892 | |
6203e07a LP |
4893 | e->exit_requested = true; |
4894 | e->exit_code = code; | |
4895 | ||
fd38203a LP |
4896 | return 0; |
4897 | } | |
46e8c825 | 4898 | |
6a0f1f6d | 4899 | _public_ int sd_event_now(sd_event *e, clockid_t clock, uint64_t *usec) { |
46e8c825 | 4900 | assert_return(e, -EINVAL); |
b937d761 | 4901 | assert_return(e = event_resolve(e), -ENOPKG); |
46e8c825 | 4902 | assert_return(usec, -EINVAL); |
46e8c825 LP |
4903 | assert_return(!event_pid_changed(e), -ECHILD); |
4904 | ||
e475d10c LP |
4905 | if (!TRIPLE_TIMESTAMP_HAS_CLOCK(clock)) |
4906 | return -EOPNOTSUPP; | |
4907 | ||
e475d10c | 4908 | if (!triple_timestamp_is_set(&e->timestamp)) { |
15c689d7 | 4909 | /* Implicitly fall back to now() if we never ran before and thus have no cached time. */ |
38a03f06 LP |
4910 | *usec = now(clock); |
4911 | return 1; | |
4912 | } | |
46e8c825 | 4913 | |
e475d10c | 4914 | *usec = triple_timestamp_by_clock(&e->timestamp, clock); |
46e8c825 LP |
4915 | return 0; |
4916 | } | |
afc6adb5 LP |
4917 | |
4918 | _public_ int sd_event_default(sd_event **ret) { | |
39883f62 | 4919 | sd_event *e = NULL; |
afc6adb5 LP |
4920 | int r; |
4921 | ||
4922 | if (!ret) | |
4923 | return !!default_event; | |
4924 | ||
4925 | if (default_event) { | |
4926 | *ret = sd_event_ref(default_event); | |
4927 | return 0; | |
4928 | } | |
4929 | ||
4930 | r = sd_event_new(&e); | |
4931 | if (r < 0) | |
4932 | return r; | |
4933 | ||
4934 | e->default_event_ptr = &default_event; | |
4935 | e->tid = gettid(); | |
4936 | default_event = e; | |
4937 | ||
4938 | *ret = e; | |
4939 | return 1; | |
4940 | } | |
4941 | ||
4942 | _public_ int sd_event_get_tid(sd_event *e, pid_t *tid) { | |
4943 | assert_return(e, -EINVAL); | |
b937d761 | 4944 | assert_return(e = event_resolve(e), -ENOPKG); |
afc6adb5 | 4945 | assert_return(tid, -EINVAL); |
76b54375 | 4946 | assert_return(!event_pid_changed(e), -ECHILD); |
afc6adb5 | 4947 | |
76b54375 LP |
4948 | if (e->tid != 0) { |
4949 | *tid = e->tid; | |
4950 | return 0; | |
4951 | } | |
4952 | ||
4953 | return -ENXIO; | |
afc6adb5 | 4954 | } |
cde93897 LP |
4955 | |
4956 | _public_ int sd_event_set_watchdog(sd_event *e, int b) { | |
4957 | int r; | |
4958 | ||
4959 | assert_return(e, -EINVAL); | |
b937d761 | 4960 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 | 4961 | assert_return(!event_pid_changed(e), -ECHILD); |
cde93897 LP |
4962 | |
4963 | if (e->watchdog == !!b) | |
4964 | return e->watchdog; | |
4965 | ||
4966 | if (b) { | |
09812eb7 LP |
4967 | r = sd_watchdog_enabled(false, &e->watchdog_period); |
4968 | if (r <= 0) | |
cde93897 | 4969 | return r; |
cde93897 LP |
4970 | |
4971 | /* Issue first ping immediately */ | |
4972 | sd_notify(false, "WATCHDOG=1"); | |
4973 | e->watchdog_last = now(CLOCK_MONOTONIC); | |
4974 | ||
4975 | e->watchdog_fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC); | |
4976 | if (e->watchdog_fd < 0) | |
4977 | return -errno; | |
4978 | ||
4979 | r = arm_watchdog(e); | |
4980 | if (r < 0) | |
4981 | goto fail; | |
4982 | ||
1eac7948 | 4983 | struct epoll_event ev = { |
a82f89aa LP |
4984 | .events = EPOLLIN, |
4985 | .data.ptr = INT_TO_PTR(SOURCE_WATCHDOG), | |
4986 | }; | |
cde93897 | 4987 | |
15c689d7 | 4988 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, e->watchdog_fd, &ev) < 0) { |
cde93897 LP |
4989 | r = -errno; |
4990 | goto fail; | |
4991 | } | |
4992 | ||
4993 | } else { | |
4994 | if (e->watchdog_fd >= 0) { | |
5a795bff | 4995 | (void) epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, e->watchdog_fd, NULL); |
03e334a1 | 4996 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
4997 | } |
4998 | } | |
4999 | ||
5000 | e->watchdog = !!b; | |
5001 | return e->watchdog; | |
5002 | ||
5003 | fail: | |
03e334a1 | 5004 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
5005 | return r; |
5006 | } | |
8f726607 LP |
5007 | |
5008 | _public_ int sd_event_get_watchdog(sd_event *e) { | |
5009 | assert_return(e, -EINVAL); | |
b937d761 | 5010 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 LP |
5011 | assert_return(!event_pid_changed(e), -ECHILD); |
5012 | ||
5013 | return e->watchdog; | |
5014 | } | |
60a3b1e1 LP |
5015 | |
5016 | _public_ int sd_event_get_iteration(sd_event *e, uint64_t *ret) { | |
5017 | assert_return(e, -EINVAL); | |
b937d761 | 5018 | assert_return(e = event_resolve(e), -ENOPKG); |
60a3b1e1 LP |
5019 | assert_return(!event_pid_changed(e), -ECHILD); |
5020 | ||
5021 | *ret = e->iteration; | |
5022 | return 0; | |
5023 | } | |
15723a1d LP |
5024 | |
5025 | _public_ int sd_event_source_set_destroy_callback(sd_event_source *s, sd_event_destroy_t callback) { | |
5026 | assert_return(s, -EINVAL); | |
5027 | ||
5028 | s->destroy_callback = callback; | |
5029 | return 0; | |
5030 | } | |
5031 | ||
5032 | _public_ int sd_event_source_get_destroy_callback(sd_event_source *s, sd_event_destroy_t *ret) { | |
5033 | assert_return(s, -EINVAL); | |
5034 | ||
5035 | if (ret) | |
5036 | *ret = s->destroy_callback; | |
5037 | ||
5038 | return !!s->destroy_callback; | |
5039 | } | |
2382c936 YW |
5040 | |
5041 | _public_ int sd_event_source_get_floating(sd_event_source *s) { | |
5042 | assert_return(s, -EINVAL); | |
5043 | ||
5044 | return s->floating; | |
5045 | } | |
5046 | ||
5047 | _public_ int sd_event_source_set_floating(sd_event_source *s, int b) { | |
5048 | assert_return(s, -EINVAL); | |
5049 | ||
5050 | if (s->floating == !!b) | |
5051 | return 0; | |
5052 | ||
5053 | if (!s->event) /* Already disconnected */ | |
5054 | return -ESTALE; | |
5055 | ||
5056 | s->floating = b; | |
5057 | ||
5058 | if (b) { | |
5059 | sd_event_source_ref(s); | |
5060 | sd_event_unref(s->event); | |
5061 | } else { | |
5062 | sd_event_ref(s->event); | |
5063 | sd_event_source_unref(s); | |
5064 | } | |
5065 | ||
5066 | return 1; | |
5067 | } | |
b778cba4 LP |
5068 | |
5069 | _public_ int sd_event_source_get_exit_on_failure(sd_event_source *s) { | |
5070 | assert_return(s, -EINVAL); | |
5071 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
5072 | ||
5073 | return s->exit_on_failure; | |
5074 | } | |
5075 | ||
5076 | _public_ int sd_event_source_set_exit_on_failure(sd_event_source *s, int b) { | |
5077 | assert_return(s, -EINVAL); | |
5078 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
5079 | ||
5080 | if (s->exit_on_failure == !!b) | |
5081 | return 0; | |
5082 | ||
5083 | s->exit_on_failure = b; | |
5084 | return 1; | |
5085 | } | |
b6d5481b LP |
5086 | |
5087 | _public_ int sd_event_source_set_ratelimit(sd_event_source *s, uint64_t interval, unsigned burst) { | |
5088 | int r; | |
5089 | ||
5090 | assert_return(s, -EINVAL); | |
5091 | ||
5092 | /* Turning on ratelimiting on event source types that don't support it, is a loggable offense. Doing | |
5093 | * so is a programming error. */ | |
5094 | assert_return(EVENT_SOURCE_CAN_RATE_LIMIT(s->type), -EDOM); | |
5095 | ||
5096 | /* When ratelimiting is configured we'll always reset the rate limit state first and start fresh, | |
5097 | * non-ratelimited. */ | |
fd69f224 | 5098 | r = event_source_leave_ratelimit(s, /* run_callback */ false); |
b6d5481b LP |
5099 | if (r < 0) |
5100 | return r; | |
5101 | ||
5102 | s->rate_limit = (RateLimit) { interval, burst }; | |
5103 | return 0; | |
fd69f224 MS |
5104 | } |
5105 | ||
5106 | _public_ int sd_event_source_set_ratelimit_expire_callback(sd_event_source *s, sd_event_handler_t callback) { | |
5107 | assert_return(s, -EINVAL); | |
5108 | ||
5109 | s->ratelimit_expire_callback = callback; | |
5110 | return 0; | |
b6d5481b LP |
5111 | } |
5112 | ||
5113 | _public_ int sd_event_source_get_ratelimit(sd_event_source *s, uint64_t *ret_interval, unsigned *ret_burst) { | |
5114 | assert_return(s, -EINVAL); | |
5115 | ||
6dd3b818 YW |
5116 | /* Querying whether an event source has ratelimiting configured is not a loggable offense, hence |
5117 | * don't use assert_return(). Unlike turning on ratelimiting it's not really a programming error. */ | |
b6d5481b LP |
5118 | if (!EVENT_SOURCE_CAN_RATE_LIMIT(s->type)) |
5119 | return -EDOM; | |
5120 | ||
5121 | if (!ratelimit_configured(&s->rate_limit)) | |
5122 | return -ENOEXEC; | |
5123 | ||
5124 | if (ret_interval) | |
5125 | *ret_interval = s->rate_limit.interval; | |
5126 | if (ret_burst) | |
5127 | *ret_burst = s->rate_limit.burst; | |
5128 | ||
5129 | return 0; | |
5130 | } | |
5131 | ||
5132 | _public_ int sd_event_source_is_ratelimited(sd_event_source *s) { | |
5133 | assert_return(s, -EINVAL); | |
5134 | ||
5135 | if (!EVENT_SOURCE_CAN_RATE_LIMIT(s->type)) | |
5136 | return false; | |
5137 | ||
5138 | if (!ratelimit_configured(&s->rate_limit)) | |
5139 | return false; | |
5140 | ||
5141 | return s->ratelimited; | |
5142 | } | |
baf3fdec LP |
5143 | |
5144 | _public_ int sd_event_set_signal_exit(sd_event *e, int b) { | |
5145 | bool change = false; | |
5146 | int r; | |
5147 | ||
5148 | assert_return(e, -EINVAL); | |
5149 | ||
5150 | if (b) { | |
5151 | /* We want to maintain pointers to these event sources, so that we can destroy them when told | |
5152 | * so. But we also don't want them to pin the event loop itself. Hence we mark them as | |
5153 | * floating after creation (and undo this before deleting them again). */ | |
5154 | ||
5155 | if (!e->sigint_event_source) { | |
5156 | r = sd_event_add_signal(e, &e->sigint_event_source, SIGINT | SD_EVENT_SIGNAL_PROCMASK, NULL, NULL); | |
5157 | if (r < 0) | |
5158 | return r; | |
5159 | ||
5160 | assert(sd_event_source_set_floating(e->sigint_event_source, true) >= 0); | |
5161 | change = true; | |
5162 | } | |
5163 | ||
5164 | if (!e->sigterm_event_source) { | |
5165 | r = sd_event_add_signal(e, &e->sigterm_event_source, SIGTERM | SD_EVENT_SIGNAL_PROCMASK, NULL, NULL); | |
5166 | if (r < 0) { | |
5167 | if (change) { | |
5168 | assert(sd_event_source_set_floating(e->sigint_event_source, false) >= 0); | |
5169 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
5170 | } | |
5171 | ||
5172 | return r; | |
5173 | } | |
5174 | ||
5175 | assert(sd_event_source_set_floating(e->sigterm_event_source, true) >= 0); | |
5176 | change = true; | |
5177 | } | |
5178 | ||
5179 | } else { | |
5180 | if (e->sigint_event_source) { | |
5181 | assert(sd_event_source_set_floating(e->sigint_event_source, false) >= 0); | |
5182 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
5183 | change = true; | |
5184 | } | |
5185 | ||
5186 | if (e->sigterm_event_source) { | |
5187 | assert(sd_event_source_set_floating(e->sigterm_event_source, false) >= 0); | |
5188 | e->sigterm_event_source = sd_event_source_unref(e->sigterm_event_source); | |
5189 | change = true; | |
5190 | } | |
5191 | } | |
5192 | ||
5193 | return change; | |
5194 | } | |
158fe190 LP |
5195 | |
5196 | _public_ int sd_event_source_set_memory_pressure_type(sd_event_source *s, const char *ty) { | |
5197 | _cleanup_free_ char *b = NULL; | |
5198 | _cleanup_free_ void *w = NULL; | |
5199 | ||
5200 | assert_return(s, -EINVAL); | |
5201 | assert_return(s->type == SOURCE_MEMORY_PRESSURE, -EDOM); | |
5202 | assert_return(ty, -EINVAL); | |
5203 | ||
5204 | if (!STR_IN_SET(ty, "some", "full")) | |
5205 | return -EINVAL; | |
5206 | ||
5207 | if (s->memory_pressure.locked) /* Refuse adjusting parameters, if caller told us how to watch for events */ | |
5208 | return -EBUSY; | |
5209 | ||
5210 | char* space = memchr(s->memory_pressure.write_buffer, ' ', s->memory_pressure.write_buffer_size); | |
5211 | if (!space) | |
5212 | return -EINVAL; | |
5213 | ||
5214 | size_t l = (char*) space - (char*) s->memory_pressure.write_buffer; | |
5215 | b = memdup_suffix0(s->memory_pressure.write_buffer, l); | |
5216 | if (!b) | |
5217 | return -ENOMEM; | |
5218 | if (!STR_IN_SET(b, "some", "full")) | |
5219 | return -EINVAL; | |
5220 | ||
5221 | if (streq(b, ty)) | |
5222 | return 0; | |
5223 | ||
5224 | size_t nl = strlen(ty) + (s->memory_pressure.write_buffer_size - l); | |
5225 | w = new(char, nl); | |
5226 | if (!w) | |
5227 | return -ENOMEM; | |
5228 | ||
5229 | memcpy(stpcpy(w, ty), space, (s->memory_pressure.write_buffer_size - l)); | |
5230 | ||
5231 | free_and_replace(s->memory_pressure.write_buffer, w); | |
5232 | s->memory_pressure.write_buffer_size = nl; | |
5233 | s->memory_pressure.locked = false; | |
5234 | ||
5235 | return 1; | |
5236 | } | |
5237 | ||
5238 | _public_ int sd_event_source_set_memory_pressure_period(sd_event_source *s, uint64_t threshold_usec, uint64_t window_usec) { | |
5239 | _cleanup_free_ char *b = NULL; | |
5240 | _cleanup_free_ void *w = NULL; | |
5241 | ||
5242 | assert_return(s, -EINVAL); | |
5243 | assert_return(s->type == SOURCE_MEMORY_PRESSURE, -EDOM); | |
5244 | ||
5245 | if (threshold_usec <= 0 || threshold_usec >= UINT64_MAX) | |
5246 | return -ERANGE; | |
5247 | if (window_usec <= 0 || window_usec >= UINT64_MAX) | |
5248 | return -ERANGE; | |
5249 | if (threshold_usec > window_usec) | |
5250 | return -EINVAL; | |
5251 | ||
5252 | if (s->memory_pressure.locked) /* Refuse adjusting parameters, if caller told us how to watch for events */ | |
5253 | return -EBUSY; | |
5254 | ||
5255 | char* space = memchr(s->memory_pressure.write_buffer, ' ', s->memory_pressure.write_buffer_size); | |
5256 | if (!space) | |
5257 | return -EINVAL; | |
5258 | ||
5259 | size_t l = (char*) space - (char*) s->memory_pressure.write_buffer; | |
5260 | b = memdup_suffix0(s->memory_pressure.write_buffer, l); | |
5261 | if (!b) | |
5262 | return -ENOMEM; | |
5263 | if (!STR_IN_SET(b, "some", "full")) | |
5264 | return -EINVAL; | |
5265 | ||
5266 | if (asprintf((char**) &w, | |
5267 | "%s " USEC_FMT " " USEC_FMT "", | |
5268 | b, | |
5269 | threshold_usec, | |
5270 | window_usec) < 0) | |
5271 | return -EINVAL; | |
5272 | ||
5273 | l = strlen(w) + 1; | |
5274 | if (memcmp_nn(s->memory_pressure.write_buffer, s->memory_pressure.write_buffer_size, w, l) == 0) | |
5275 | return 0; | |
5276 | ||
5277 | free_and_replace(s->memory_pressure.write_buffer, w); | |
5278 | s->memory_pressure.write_buffer_size = l; | |
5279 | s->memory_pressure.locked = false; | |
5280 | ||
5281 | return 1; | |
5282 | } |