]>
Commit | Line | Data |
---|---|---|
db9ecf05 | 1 | /* SPDX-License-Identifier: LGPL-2.1-or-later */ |
fd38203a LP |
2 | |
3 | #include <sys/epoll.h> | |
4 | #include <sys/timerfd.h> | |
5 | #include <sys/wait.h> | |
6 | ||
cde93897 | 7 | #include "sd-daemon.h" |
07630cea LP |
8 | #include "sd-event.h" |
9 | #include "sd-id128.h" | |
158fe190 | 10 | #include "sd-messages.h" |
07630cea | 11 | |
b5efdb8a | 12 | #include "alloc-util.h" |
f8f3f926 | 13 | #include "env-util.h" |
a137a1c3 | 14 | #include "event-source.h" |
3ffd4af2 | 15 | #include "fd-util.h" |
97ef5391 | 16 | #include "fs-util.h" |
28e5e1e9 | 17 | #include "glyph-util.h" |
fd38203a | 18 | #include "hashmap.h" |
158fe190 | 19 | #include "hexdecoct.h" |
07630cea | 20 | #include "list.h" |
3ae6b3bf | 21 | #include "logarithm.h" |
07630cea | 22 | #include "macro.h" |
158fe190 | 23 | #include "mallinfo-util.h" |
0a970718 | 24 | #include "memory-util.h" |
158fe190 | 25 | #include "missing_magic.h" |
f5947a5e | 26 | #include "missing_syscall.h" |
5545f336 | 27 | #include "missing_threads.h" |
158fe190 | 28 | #include "path-util.h" |
07630cea | 29 | #include "prioq.h" |
4a0b58c4 | 30 | #include "process-util.h" |
158fe190 | 31 | #include "psi-util.h" |
6e9feda3 | 32 | #include "set.h" |
24882e06 | 33 | #include "signal-util.h" |
158fe190 LP |
34 | #include "socket-util.h" |
35 | #include "stat-util.h" | |
55cbfaa5 | 36 | #include "string-table.h" |
07630cea | 37 | #include "string-util.h" |
442ac269 | 38 | #include "strxcpyx.h" |
07630cea | 39 | #include "time-util.h" |
fd38203a | 40 | |
c2ba3ad6 | 41 | #define DEFAULT_ACCURACY_USEC (250 * USEC_PER_MSEC) |
fd38203a | 42 | |
f8f3f926 LP |
43 | static bool EVENT_SOURCE_WATCH_PIDFD(sd_event_source *s) { |
44 | /* Returns true if this is a PID event source and can be implemented by watching EPOLLIN */ | |
45 | return s && | |
46 | s->type == SOURCE_CHILD && | |
47 | s->child.pidfd >= 0 && | |
48 | s->child.options == WEXITED; | |
49 | } | |
50 | ||
b6d5481b LP |
51 | static bool event_source_is_online(sd_event_source *s) { |
52 | assert(s); | |
53 | return s->enabled != SD_EVENT_OFF && !s->ratelimited; | |
54 | } | |
55 | ||
56 | static bool event_source_is_offline(sd_event_source *s) { | |
57 | assert(s); | |
58 | return s->enabled == SD_EVENT_OFF || s->ratelimited; | |
59 | } | |
60 | ||
55cbfaa5 | 61 | static const char* const event_source_type_table[_SOURCE_EVENT_SOURCE_TYPE_MAX] = { |
3f3548f8 ZJS |
62 | [SOURCE_IO] = "io", |
63 | [SOURCE_TIME_REALTIME] = "realtime", | |
64 | [SOURCE_TIME_BOOTTIME] = "bootime", | |
65 | [SOURCE_TIME_MONOTONIC] = "monotonic", | |
55cbfaa5 DM |
66 | [SOURCE_TIME_REALTIME_ALARM] = "realtime-alarm", |
67 | [SOURCE_TIME_BOOTTIME_ALARM] = "boottime-alarm", | |
3f3548f8 ZJS |
68 | [SOURCE_SIGNAL] = "signal", |
69 | [SOURCE_CHILD] = "child", | |
70 | [SOURCE_DEFER] = "defer", | |
71 | [SOURCE_POST] = "post", | |
72 | [SOURCE_EXIT] = "exit", | |
73 | [SOURCE_WATCHDOG] = "watchdog", | |
74 | [SOURCE_INOTIFY] = "inotify", | |
158fe190 | 75 | [SOURCE_MEMORY_PRESSURE] = "memory-pressure", |
55cbfaa5 DM |
76 | }; |
77 | ||
78 | DEFINE_PRIVATE_STRING_TABLE_LOOKUP_TO_STRING(event_source_type, int); | |
79 | ||
b6d5481b LP |
80 | #define EVENT_SOURCE_IS_TIME(t) \ |
81 | IN_SET((t), \ | |
82 | SOURCE_TIME_REALTIME, \ | |
83 | SOURCE_TIME_BOOTTIME, \ | |
84 | SOURCE_TIME_MONOTONIC, \ | |
85 | SOURCE_TIME_REALTIME_ALARM, \ | |
86 | SOURCE_TIME_BOOTTIME_ALARM) | |
87 | ||
88 | #define EVENT_SOURCE_CAN_RATE_LIMIT(t) \ | |
89 | IN_SET((t), \ | |
90 | SOURCE_IO, \ | |
91 | SOURCE_TIME_REALTIME, \ | |
92 | SOURCE_TIME_BOOTTIME, \ | |
93 | SOURCE_TIME_MONOTONIC, \ | |
94 | SOURCE_TIME_REALTIME_ALARM, \ | |
95 | SOURCE_TIME_BOOTTIME_ALARM, \ | |
96 | SOURCE_SIGNAL, \ | |
97 | SOURCE_DEFER, \ | |
158fe190 LP |
98 | SOURCE_INOTIFY, \ |
99 | SOURCE_MEMORY_PRESSURE) | |
6a0f1f6d | 100 | |
19947509 ZJS |
101 | /* This is used to assert that we didn't pass an unexpected source type to event_source_time_prioq_put(). |
102 | * Time sources and ratelimited sources can be passed, so effectively this is the same as the | |
103 | * EVENT_SOURCE_CAN_RATE_LIMIT() macro. */ | |
104 | #define EVENT_SOURCE_USES_TIME_PRIOQ(t) EVENT_SOURCE_CAN_RATE_LIMIT(t) | |
105 | ||
fd38203a | 106 | struct sd_event { |
da7e457c | 107 | unsigned n_ref; |
fd38203a LP |
108 | |
109 | int epoll_fd; | |
cde93897 | 110 | int watchdog_fd; |
fd38203a LP |
111 | |
112 | Prioq *pending; | |
113 | Prioq *prepare; | |
c2ba3ad6 | 114 | |
a8548816 | 115 | /* timerfd_create() only supports these five clocks so far. We |
6a0f1f6d LP |
116 | * can add support for more clocks when the kernel learns to |
117 | * deal with them, too. */ | |
118 | struct clock_data realtime; | |
a8548816 | 119 | struct clock_data boottime; |
6a0f1f6d LP |
120 | struct clock_data monotonic; |
121 | struct clock_data realtime_alarm; | |
122 | struct clock_data boottime_alarm; | |
fd38203a | 123 | |
da7e457c LP |
124 | usec_t perturb; |
125 | ||
9da4cb2b LP |
126 | sd_event_source **signal_sources; /* indexed by signal number */ |
127 | Hashmap *signal_data; /* indexed by priority */ | |
fd38203a LP |
128 | |
129 | Hashmap *child_sources; | |
b6d5481b | 130 | unsigned n_online_child_sources; |
fd38203a | 131 | |
6e9feda3 LP |
132 | Set *post_sources; |
133 | ||
6203e07a | 134 | Prioq *exit; |
fd38203a | 135 | |
97ef5391 LP |
136 | Hashmap *inotify_data; /* indexed by priority */ |
137 | ||
138 | /* A list of inode structures that still have an fd open, that we need to close before the next loop iteration */ | |
ed828563 | 139 | LIST_HEAD(struct inode_data, inode_data_to_close_list); |
97ef5391 LP |
140 | |
141 | /* A list of inotify objects that already have events buffered which aren't processed yet */ | |
0601b958 | 142 | LIST_HEAD(struct inotify_data, buffered_inotify_data_list); |
97ef5391 | 143 | |
158fe190 LP |
144 | /* A list of memory pressure event sources that still need their subscription string written */ |
145 | LIST_HEAD(sd_event_source, memory_pressure_write_list); | |
146 | ||
da7e457c | 147 | pid_t original_pid; |
c2ba3ad6 | 148 | |
60a3b1e1 | 149 | uint64_t iteration; |
e475d10c | 150 | triple_timestamp timestamp; |
da7e457c | 151 | int state; |
eaa3cbef | 152 | |
6203e07a | 153 | bool exit_requested:1; |
da7e457c | 154 | bool need_process_child:1; |
cde93897 | 155 | bool watchdog:1; |
34b87517 | 156 | bool profile_delays:1; |
afc6adb5 | 157 | |
6203e07a LP |
158 | int exit_code; |
159 | ||
afc6adb5 LP |
160 | pid_t tid; |
161 | sd_event **default_event_ptr; | |
cde93897 LP |
162 | |
163 | usec_t watchdog_last, watchdog_period; | |
15b38f93 LP |
164 | |
165 | unsigned n_sources; | |
a71fe8b8 | 166 | |
5cddd924 | 167 | struct epoll_event *event_queue; |
5cddd924 | 168 | |
a71fe8b8 | 169 | LIST_HEAD(sd_event_source, sources); |
34b87517 | 170 | |
baf3fdec LP |
171 | sd_event_source *sigint_event_source, *sigterm_event_source; |
172 | ||
e6a7bee5 | 173 | usec_t last_run_usec, last_log_usec; |
34b87517 | 174 | unsigned delays[sizeof(usec_t) * 8]; |
fd38203a LP |
175 | }; |
176 | ||
b937d761 NM |
177 | static thread_local sd_event *default_event = NULL; |
178 | ||
a71fe8b8 | 179 | static void source_disconnect(sd_event_source *s); |
97ef5391 | 180 | static void event_gc_inode_data(sd_event *e, struct inode_data *d); |
a71fe8b8 | 181 | |
b937d761 NM |
182 | static sd_event *event_resolve(sd_event *e) { |
183 | return e == SD_EVENT_DEFAULT ? default_event : e; | |
184 | } | |
185 | ||
fd38203a LP |
186 | static int pending_prioq_compare(const void *a, const void *b) { |
187 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 188 | int r; |
fd38203a LP |
189 | |
190 | assert(x->pending); | |
191 | assert(y->pending); | |
192 | ||
baf76283 | 193 | /* Enabled ones first */ |
06e13147 YW |
194 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
195 | if (r != 0) | |
196 | return r; | |
fd38203a | 197 | |
b6d5481b LP |
198 | /* Non rate-limited ones first. */ |
199 | r = CMP(!!x->ratelimited, !!y->ratelimited); | |
200 | if (r != 0) | |
201 | return r; | |
202 | ||
fd38203a | 203 | /* Lower priority values first */ |
9c57a73b YW |
204 | r = CMP(x->priority, y->priority); |
205 | if (r != 0) | |
206 | return r; | |
fd38203a LP |
207 | |
208 | /* Older entries first */ | |
9c57a73b | 209 | return CMP(x->pending_iteration, y->pending_iteration); |
fd38203a LP |
210 | } |
211 | ||
212 | static int prepare_prioq_compare(const void *a, const void *b) { | |
213 | const sd_event_source *x = a, *y = b; | |
9c57a73b | 214 | int r; |
fd38203a LP |
215 | |
216 | assert(x->prepare); | |
217 | assert(y->prepare); | |
218 | ||
8046c457 | 219 | /* Enabled ones first */ |
06e13147 YW |
220 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
221 | if (r != 0) | |
222 | return r; | |
8046c457 | 223 | |
b6d5481b LP |
224 | /* Non rate-limited ones first. */ |
225 | r = CMP(!!x->ratelimited, !!y->ratelimited); | |
226 | if (r != 0) | |
227 | return r; | |
228 | ||
fd38203a LP |
229 | /* Move most recently prepared ones last, so that we can stop |
230 | * preparing as soon as we hit one that has already been | |
231 | * prepared in the current iteration */ | |
9c57a73b YW |
232 | r = CMP(x->prepare_iteration, y->prepare_iteration); |
233 | if (r != 0) | |
234 | return r; | |
fd38203a | 235 | |
fd38203a | 236 | /* Lower priority values first */ |
9c57a73b | 237 | return CMP(x->priority, y->priority); |
fd38203a LP |
238 | } |
239 | ||
b6d5481b LP |
240 | static usec_t time_event_source_next(const sd_event_source *s) { |
241 | assert(s); | |
242 | ||
243 | /* We have two kinds of event sources that have elapsation times associated with them: the actual | |
244 | * time based ones and the ones for which a ratelimit can be in effect (where we want to be notified | |
245 | * once the ratelimit time window ends). Let's return the next elapsing time depending on what we are | |
246 | * looking at here. */ | |
247 | ||
248 | if (s->ratelimited) { /* If rate-limited the next elapsation is when the ratelimit time window ends */ | |
249 | assert(s->rate_limit.begin != 0); | |
250 | assert(s->rate_limit.interval != 0); | |
251 | return usec_add(s->rate_limit.begin, s->rate_limit.interval); | |
252 | } | |
253 | ||
254 | /* Otherwise this must be a time event source, if not ratelimited */ | |
255 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
256 | return s->time.next; | |
257 | ||
258 | return USEC_INFINITY; | |
259 | } | |
260 | ||
1bce0ffa | 261 | static usec_t time_event_source_latest(const sd_event_source *s) { |
b6d5481b LP |
262 | assert(s); |
263 | ||
264 | if (s->ratelimited) { /* For ratelimited stuff the earliest and the latest time shall actually be the | |
265 | * same, as we should avoid adding additional inaccuracy on an inaccuracy time | |
266 | * window */ | |
267 | assert(s->rate_limit.begin != 0); | |
268 | assert(s->rate_limit.interval != 0); | |
269 | return usec_add(s->rate_limit.begin, s->rate_limit.interval); | |
270 | } | |
271 | ||
272 | /* Must be a time event source, if not ratelimited */ | |
273 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
274 | return usec_add(s->time.next, s->time.accuracy); | |
275 | ||
276 | return USEC_INFINITY; | |
1bce0ffa LP |
277 | } |
278 | ||
81107b84 LP |
279 | static bool event_source_timer_candidate(const sd_event_source *s) { |
280 | assert(s); | |
281 | ||
282 | /* Returns true for event sources that either are not pending yet (i.e. where it's worth to mark them pending) | |
283 | * or which are currently ratelimited (i.e. where it's worth leaving the ratelimited state) */ | |
284 | return !s->pending || s->ratelimited; | |
285 | } | |
286 | ||
287 | static int time_prioq_compare(const void *a, const void *b, usec_t (*time_func)(const sd_event_source *s)) { | |
c2ba3ad6 | 288 | const sd_event_source *x = a, *y = b; |
06e13147 | 289 | int r; |
c2ba3ad6 | 290 | |
baf76283 | 291 | /* Enabled ones first */ |
06e13147 YW |
292 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
293 | if (r != 0) | |
294 | return r; | |
c2ba3ad6 | 295 | |
81107b84 | 296 | /* Order "non-pending OR ratelimited" before "pending AND not-ratelimited" */ |
06e13147 YW |
297 | r = CMP(!event_source_timer_candidate(x), !event_source_timer_candidate(y)); |
298 | if (r != 0) | |
299 | return r; | |
c2ba3ad6 LP |
300 | |
301 | /* Order by time */ | |
81107b84 LP |
302 | return CMP(time_func(x), time_func(y)); |
303 | } | |
304 | ||
305 | static int earliest_time_prioq_compare(const void *a, const void *b) { | |
306 | return time_prioq_compare(a, b, time_event_source_next); | |
307 | } | |
308 | ||
309 | static int latest_time_prioq_compare(const void *a, const void *b) { | |
310 | return time_prioq_compare(a, b, time_event_source_latest); | |
c2ba3ad6 LP |
311 | } |
312 | ||
6203e07a | 313 | static int exit_prioq_compare(const void *a, const void *b) { |
da7e457c | 314 | const sd_event_source *x = a, *y = b; |
06e13147 | 315 | int r; |
da7e457c | 316 | |
6203e07a LP |
317 | assert(x->type == SOURCE_EXIT); |
318 | assert(y->type == SOURCE_EXIT); | |
da7e457c | 319 | |
baf76283 | 320 | /* Enabled ones first */ |
06e13147 YW |
321 | r = CMP(x->enabled == SD_EVENT_OFF, y->enabled == SD_EVENT_OFF); |
322 | if (r != 0) | |
323 | return r; | |
da7e457c LP |
324 | |
325 | /* Lower priority values first */ | |
6dd91b36 | 326 | return CMP(x->priority, y->priority); |
da7e457c LP |
327 | } |
328 | ||
6a0f1f6d LP |
329 | static void free_clock_data(struct clock_data *d) { |
330 | assert(d); | |
9da4cb2b | 331 | assert(d->wakeup == WAKEUP_CLOCK_DATA); |
6a0f1f6d LP |
332 | |
333 | safe_close(d->fd); | |
334 | prioq_free(d->earliest); | |
335 | prioq_free(d->latest); | |
336 | } | |
337 | ||
8301aa0b | 338 | static sd_event *event_free(sd_event *e) { |
a71fe8b8 LP |
339 | sd_event_source *s; |
340 | ||
fd38203a | 341 | assert(e); |
a71fe8b8 | 342 | |
baf3fdec LP |
343 | e->sigterm_event_source = sd_event_source_unref(e->sigterm_event_source); |
344 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
345 | ||
a71fe8b8 LP |
346 | while ((s = e->sources)) { |
347 | assert(s->floating); | |
348 | source_disconnect(s); | |
349 | sd_event_source_unref(s); | |
350 | } | |
351 | ||
15b38f93 | 352 | assert(e->n_sources == 0); |
fd38203a | 353 | |
afc6adb5 LP |
354 | if (e->default_event_ptr) |
355 | *(e->default_event_ptr) = NULL; | |
356 | ||
03e334a1 | 357 | safe_close(e->epoll_fd); |
03e334a1 | 358 | safe_close(e->watchdog_fd); |
cde93897 | 359 | |
6a0f1f6d | 360 | free_clock_data(&e->realtime); |
a8548816 | 361 | free_clock_data(&e->boottime); |
6a0f1f6d LP |
362 | free_clock_data(&e->monotonic); |
363 | free_clock_data(&e->realtime_alarm); | |
364 | free_clock_data(&e->boottime_alarm); | |
365 | ||
fd38203a LP |
366 | prioq_free(e->pending); |
367 | prioq_free(e->prepare); | |
6203e07a | 368 | prioq_free(e->exit); |
fd38203a LP |
369 | |
370 | free(e->signal_sources); | |
9da4cb2b | 371 | hashmap_free(e->signal_data); |
fd38203a | 372 | |
97ef5391 LP |
373 | hashmap_free(e->inotify_data); |
374 | ||
fd38203a | 375 | hashmap_free(e->child_sources); |
6e9feda3 | 376 | set_free(e->post_sources); |
8301aa0b | 377 | |
5cddd924 LP |
378 | free(e->event_queue); |
379 | ||
8301aa0b | 380 | return mfree(e); |
fd38203a LP |
381 | } |
382 | ||
f7262a9f | 383 | _public_ int sd_event_new(sd_event** ret) { |
fd38203a LP |
384 | sd_event *e; |
385 | int r; | |
386 | ||
305f78bf | 387 | assert_return(ret, -EINVAL); |
fd38203a | 388 | |
d08eb1fa | 389 | e = new(sd_event, 1); |
fd38203a LP |
390 | if (!e) |
391 | return -ENOMEM; | |
392 | ||
d08eb1fa LP |
393 | *e = (sd_event) { |
394 | .n_ref = 1, | |
254d1313 ZJS |
395 | .epoll_fd = -EBADF, |
396 | .watchdog_fd = -EBADF, | |
d08eb1fa | 397 | .realtime.wakeup = WAKEUP_CLOCK_DATA, |
254d1313 | 398 | .realtime.fd = -EBADF, |
d08eb1fa LP |
399 | .realtime.next = USEC_INFINITY, |
400 | .boottime.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 401 | .boottime.fd = -EBADF, |
d08eb1fa LP |
402 | .boottime.next = USEC_INFINITY, |
403 | .monotonic.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 404 | .monotonic.fd = -EBADF, |
d08eb1fa LP |
405 | .monotonic.next = USEC_INFINITY, |
406 | .realtime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 407 | .realtime_alarm.fd = -EBADF, |
d08eb1fa LP |
408 | .realtime_alarm.next = USEC_INFINITY, |
409 | .boottime_alarm.wakeup = WAKEUP_CLOCK_DATA, | |
254d1313 | 410 | .boottime_alarm.fd = -EBADF, |
d08eb1fa LP |
411 | .boottime_alarm.next = USEC_INFINITY, |
412 | .perturb = USEC_INFINITY, | |
413 | .original_pid = getpid_cached(), | |
414 | }; | |
fd38203a | 415 | |
c983e776 EV |
416 | r = prioq_ensure_allocated(&e->pending, pending_prioq_compare); |
417 | if (r < 0) | |
fd38203a | 418 | goto fail; |
fd38203a LP |
419 | |
420 | e->epoll_fd = epoll_create1(EPOLL_CLOEXEC); | |
421 | if (e->epoll_fd < 0) { | |
422 | r = -errno; | |
423 | goto fail; | |
424 | } | |
425 | ||
7fe2903c LP |
426 | e->epoll_fd = fd_move_above_stdio(e->epoll_fd); |
427 | ||
34b87517 | 428 | if (secure_getenv("SD_EVENT_PROFILE_DELAYS")) { |
28e5e1e9 DT |
429 | 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.", |
430 | special_glyph(SPECIAL_GLYPH_ELLIPSIS)); | |
34b87517 VC |
431 | e->profile_delays = true; |
432 | } | |
433 | ||
fd38203a LP |
434 | *ret = e; |
435 | return 0; | |
436 | ||
437 | fail: | |
438 | event_free(e); | |
439 | return r; | |
440 | } | |
441 | ||
8301aa0b | 442 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event, sd_event, event_free); |
c8e9d15c YW |
443 | #define PROTECT_EVENT(e) \ |
444 | _unused_ _cleanup_(sd_event_unrefp) sd_event *_ref = sd_event_ref(e); | |
fd38203a | 445 | |
afd15bbb ZJS |
446 | _public_ sd_event_source* sd_event_source_disable_unref(sd_event_source *s) { |
447 | if (s) | |
448 | (void) sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
449 | return sd_event_source_unref(s); | |
450 | } | |
451 | ||
eaa3cbef LP |
452 | static bool event_pid_changed(sd_event *e) { |
453 | assert(e); | |
454 | ||
a2360a46 | 455 | /* We don't support people creating an event loop and keeping |
eaa3cbef LP |
456 | * it around over a fork(). Let's complain. */ |
457 | ||
df0ff127 | 458 | return e->original_pid != getpid_cached(); |
eaa3cbef LP |
459 | } |
460 | ||
366e6411 | 461 | static void source_io_unregister(sd_event_source *s) { |
fd38203a LP |
462 | assert(s); |
463 | assert(s->type == SOURCE_IO); | |
464 | ||
f6806734 | 465 | if (event_pid_changed(s->event)) |
366e6411 | 466 | return; |
f6806734 | 467 | |
fd38203a | 468 | if (!s->io.registered) |
366e6411 | 469 | return; |
fd38203a | 470 | |
d1cf2023 | 471 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->io.fd, NULL) < 0) |
f80a5d6a | 472 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll, ignoring: %m", |
55cbfaa5 | 473 | strna(s->description), event_source_type_to_string(s->type)); |
fd38203a LP |
474 | |
475 | s->io.registered = false; | |
fd38203a LP |
476 | } |
477 | ||
305f78bf LP |
478 | static int source_io_register( |
479 | sd_event_source *s, | |
480 | int enabled, | |
481 | uint32_t events) { | |
482 | ||
fd38203a LP |
483 | assert(s); |
484 | assert(s->type == SOURCE_IO); | |
baf76283 | 485 | assert(enabled != SD_EVENT_OFF); |
fd38203a | 486 | |
1eac7948 | 487 | struct epoll_event ev = { |
a82f89aa LP |
488 | .events = events | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0), |
489 | .data.ptr = s, | |
490 | }; | |
fd38203a | 491 | |
15c689d7 | 492 | if (epoll_ctl(s->event->epoll_fd, |
1eac7948 | 493 | s->io.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, |
55c540d3 | 494 | s->io.fd, &ev) < 0) |
fd38203a LP |
495 | return -errno; |
496 | ||
497 | s->io.registered = true; | |
498 | ||
499 | return 0; | |
500 | } | |
501 | ||
f8f3f926 LP |
502 | static void source_child_pidfd_unregister(sd_event_source *s) { |
503 | assert(s); | |
504 | assert(s->type == SOURCE_CHILD); | |
505 | ||
506 | if (event_pid_changed(s->event)) | |
507 | return; | |
508 | ||
509 | if (!s->child.registered) | |
510 | return; | |
511 | ||
512 | if (EVENT_SOURCE_WATCH_PIDFD(s)) | |
513 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->child.pidfd, NULL) < 0) | |
f80a5d6a | 514 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll, ignoring: %m", |
f8f3f926 LP |
515 | strna(s->description), event_source_type_to_string(s->type)); |
516 | ||
517 | s->child.registered = false; | |
518 | } | |
519 | ||
520 | static int source_child_pidfd_register(sd_event_source *s, int enabled) { | |
f8f3f926 LP |
521 | assert(s); |
522 | assert(s->type == SOURCE_CHILD); | |
523 | assert(enabled != SD_EVENT_OFF); | |
524 | ||
525 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { | |
1eac7948 | 526 | struct epoll_event ev = { |
f8f3f926 LP |
527 | .events = EPOLLIN | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0), |
528 | .data.ptr = s, | |
529 | }; | |
530 | ||
55c540d3 ZJS |
531 | if (epoll_ctl(s->event->epoll_fd, |
532 | s->child.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, | |
533 | s->child.pidfd, &ev) < 0) | |
f8f3f926 LP |
534 | return -errno; |
535 | } | |
536 | ||
537 | s->child.registered = true; | |
538 | return 0; | |
539 | } | |
540 | ||
158fe190 LP |
541 | static void source_memory_pressure_unregister(sd_event_source *s) { |
542 | assert(s); | |
543 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
544 | ||
545 | if (event_pid_changed(s->event)) | |
546 | return; | |
547 | ||
548 | if (!s->memory_pressure.registered) | |
549 | return; | |
550 | ||
551 | if (epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, s->memory_pressure.fd, NULL) < 0) | |
552 | log_debug_errno(errno, "Failed to remove source %s (type %s) from epoll, ignoring: %m", | |
553 | strna(s->description), event_source_type_to_string(s->type)); | |
554 | ||
555 | s->memory_pressure.registered = false; | |
556 | } | |
557 | ||
558 | static int source_memory_pressure_register(sd_event_source *s, int enabled) { | |
559 | assert(s); | |
560 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
561 | assert(enabled != SD_EVENT_OFF); | |
562 | ||
563 | struct epoll_event ev = { | |
564 | .events = s->memory_pressure.write_buffer_size > 0 ? EPOLLOUT : | |
565 | (s->memory_pressure.events | (enabled == SD_EVENT_ONESHOT ? EPOLLONESHOT : 0)), | |
566 | .data.ptr = s, | |
567 | }; | |
568 | ||
569 | if (epoll_ctl(s->event->epoll_fd, | |
570 | s->memory_pressure.registered ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, | |
571 | s->memory_pressure.fd, &ev) < 0) | |
572 | return -errno; | |
573 | ||
574 | s->memory_pressure.registered = true; | |
575 | return 0; | |
576 | } | |
577 | ||
578 | static void source_memory_pressure_add_to_write_list(sd_event_source *s) { | |
579 | assert(s); | |
580 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
581 | ||
582 | if (s->memory_pressure.in_write_list) | |
583 | return; | |
584 | ||
585 | LIST_PREPEND(memory_pressure.write_list, s->event->memory_pressure_write_list, s); | |
586 | s->memory_pressure.in_write_list = true; | |
587 | } | |
588 | ||
589 | static void source_memory_pressure_remove_from_write_list(sd_event_source *s) { | |
590 | assert(s); | |
591 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
592 | ||
593 | if (!s->memory_pressure.in_write_list) | |
594 | return; | |
595 | ||
596 | LIST_REMOVE(memory_pressure.write_list, s->event->memory_pressure_write_list, s); | |
597 | s->memory_pressure.in_write_list = false; | |
598 | } | |
599 | ||
6a0f1f6d LP |
600 | static clockid_t event_source_type_to_clock(EventSourceType t) { |
601 | ||
602 | switch (t) { | |
603 | ||
604 | case SOURCE_TIME_REALTIME: | |
605 | return CLOCK_REALTIME; | |
606 | ||
a8548816 TG |
607 | case SOURCE_TIME_BOOTTIME: |
608 | return CLOCK_BOOTTIME; | |
609 | ||
6a0f1f6d LP |
610 | case SOURCE_TIME_MONOTONIC: |
611 | return CLOCK_MONOTONIC; | |
612 | ||
613 | case SOURCE_TIME_REALTIME_ALARM: | |
614 | return CLOCK_REALTIME_ALARM; | |
615 | ||
616 | case SOURCE_TIME_BOOTTIME_ALARM: | |
617 | return CLOCK_BOOTTIME_ALARM; | |
618 | ||
619 | default: | |
620 | return (clockid_t) -1; | |
621 | } | |
622 | } | |
623 | ||
624 | static EventSourceType clock_to_event_source_type(clockid_t clock) { | |
625 | ||
626 | switch (clock) { | |
627 | ||
628 | case CLOCK_REALTIME: | |
629 | return SOURCE_TIME_REALTIME; | |
630 | ||
a8548816 TG |
631 | case CLOCK_BOOTTIME: |
632 | return SOURCE_TIME_BOOTTIME; | |
633 | ||
6a0f1f6d LP |
634 | case CLOCK_MONOTONIC: |
635 | return SOURCE_TIME_MONOTONIC; | |
636 | ||
637 | case CLOCK_REALTIME_ALARM: | |
638 | return SOURCE_TIME_REALTIME_ALARM; | |
639 | ||
640 | case CLOCK_BOOTTIME_ALARM: | |
641 | return SOURCE_TIME_BOOTTIME_ALARM; | |
642 | ||
643 | default: | |
644 | return _SOURCE_EVENT_SOURCE_TYPE_INVALID; | |
645 | } | |
646 | } | |
647 | ||
648 | static struct clock_data* event_get_clock_data(sd_event *e, EventSourceType t) { | |
649 | assert(e); | |
650 | ||
651 | switch (t) { | |
652 | ||
653 | case SOURCE_TIME_REALTIME: | |
654 | return &e->realtime; | |
655 | ||
a8548816 TG |
656 | case SOURCE_TIME_BOOTTIME: |
657 | return &e->boottime; | |
658 | ||
6a0f1f6d LP |
659 | case SOURCE_TIME_MONOTONIC: |
660 | return &e->monotonic; | |
661 | ||
662 | case SOURCE_TIME_REALTIME_ALARM: | |
663 | return &e->realtime_alarm; | |
664 | ||
665 | case SOURCE_TIME_BOOTTIME_ALARM: | |
666 | return &e->boottime_alarm; | |
667 | ||
668 | default: | |
669 | return NULL; | |
670 | } | |
671 | } | |
672 | ||
3e4eb8e7 YW |
673 | static void event_free_signal_data(sd_event *e, struct signal_data *d) { |
674 | assert(e); | |
675 | ||
676 | if (!d) | |
677 | return; | |
678 | ||
679 | hashmap_remove(e->signal_data, &d->priority); | |
680 | safe_close(d->fd); | |
681 | free(d); | |
682 | } | |
683 | ||
9da4cb2b LP |
684 | static int event_make_signal_data( |
685 | sd_event *e, | |
686 | int sig, | |
687 | struct signal_data **ret) { | |
4807d2d0 | 688 | |
9da4cb2b LP |
689 | struct signal_data *d; |
690 | bool added = false; | |
691 | sigset_t ss_copy; | |
692 | int64_t priority; | |
f95387cd ZJS |
693 | int r; |
694 | ||
695 | assert(e); | |
696 | ||
f6806734 | 697 | if (event_pid_changed(e)) |
9da4cb2b | 698 | return -ECHILD; |
f6806734 | 699 | |
9da4cb2b LP |
700 | if (e->signal_sources && e->signal_sources[sig]) |
701 | priority = e->signal_sources[sig]->priority; | |
702 | else | |
de05913d | 703 | priority = SD_EVENT_PRIORITY_NORMAL; |
f95387cd | 704 | |
9da4cb2b LP |
705 | d = hashmap_get(e->signal_data, &priority); |
706 | if (d) { | |
707 | if (sigismember(&d->sigset, sig) > 0) { | |
708 | if (ret) | |
709 | *ret = d; | |
710 | return 0; | |
711 | } | |
712 | } else { | |
d08eb1fa | 713 | d = new(struct signal_data, 1); |
9da4cb2b LP |
714 | if (!d) |
715 | return -ENOMEM; | |
716 | ||
d08eb1fa LP |
717 | *d = (struct signal_data) { |
718 | .wakeup = WAKEUP_SIGNAL_DATA, | |
254d1313 | 719 | .fd = -EBADF, |
d08eb1fa LP |
720 | .priority = priority, |
721 | }; | |
9da4cb2b | 722 | |
f656fdb6 | 723 | r = hashmap_ensure_put(&e->signal_data, &uint64_hash_ops, &d->priority, d); |
90f604d1 ZJS |
724 | if (r < 0) { |
725 | free(d); | |
9da4cb2b | 726 | return r; |
90f604d1 | 727 | } |
f95387cd | 728 | |
9da4cb2b LP |
729 | added = true; |
730 | } | |
731 | ||
732 | ss_copy = d->sigset; | |
733 | assert_se(sigaddset(&ss_copy, sig) >= 0); | |
734 | ||
cbff793f ZJS |
735 | r = signalfd(d->fd >= 0 ? d->fd : -1, /* the first arg must be -1 or a valid signalfd */ |
736 | &ss_copy, | |
737 | SFD_NONBLOCK|SFD_CLOEXEC); | |
9da4cb2b LP |
738 | if (r < 0) { |
739 | r = -errno; | |
740 | goto fail; | |
741 | } | |
742 | ||
743 | d->sigset = ss_copy; | |
f95387cd | 744 | |
9da4cb2b LP |
745 | if (d->fd >= 0) { |
746 | if (ret) | |
747 | *ret = d; | |
f95387cd | 748 | return 0; |
9da4cb2b LP |
749 | } |
750 | ||
7fe2903c | 751 | d->fd = fd_move_above_stdio(r); |
f95387cd | 752 | |
1eac7948 | 753 | struct epoll_event ev = { |
a82f89aa LP |
754 | .events = EPOLLIN, |
755 | .data.ptr = d, | |
756 | }; | |
f95387cd | 757 | |
15c689d7 | 758 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { |
9da4cb2b LP |
759 | r = -errno; |
760 | goto fail; | |
f95387cd ZJS |
761 | } |
762 | ||
9da4cb2b LP |
763 | if (ret) |
764 | *ret = d; | |
765 | ||
f95387cd | 766 | return 0; |
9da4cb2b LP |
767 | |
768 | fail: | |
3e4eb8e7 YW |
769 | if (added) |
770 | event_free_signal_data(e, d); | |
9da4cb2b LP |
771 | |
772 | return r; | |
773 | } | |
774 | ||
775 | static void event_unmask_signal_data(sd_event *e, struct signal_data *d, int sig) { | |
776 | assert(e); | |
777 | assert(d); | |
778 | ||
779 | /* Turns off the specified signal in the signal data | |
780 | * object. If the signal mask of the object becomes empty that | |
781 | * way removes it. */ | |
782 | ||
783 | if (sigismember(&d->sigset, sig) == 0) | |
784 | return; | |
785 | ||
786 | assert_se(sigdelset(&d->sigset, sig) >= 0); | |
787 | ||
788 | if (sigisemptyset(&d->sigset)) { | |
9da4cb2b | 789 | /* If all the mask is all-zero we can get rid of the structure */ |
3e4eb8e7 | 790 | event_free_signal_data(e, d); |
9da4cb2b LP |
791 | return; |
792 | } | |
793 | ||
01e6af73 YW |
794 | if (event_pid_changed(e)) |
795 | return; | |
796 | ||
9da4cb2b LP |
797 | assert(d->fd >= 0); |
798 | ||
799 | if (signalfd(d->fd, &d->sigset, SFD_NONBLOCK|SFD_CLOEXEC) < 0) | |
800 | log_debug_errno(errno, "Failed to unset signal bit, ignoring: %m"); | |
801 | } | |
802 | ||
803 | static void event_gc_signal_data(sd_event *e, const int64_t *priority, int sig) { | |
804 | struct signal_data *d; | |
805 | static const int64_t zero_priority = 0; | |
806 | ||
807 | assert(e); | |
808 | ||
f8f3f926 LP |
809 | /* Rechecks if the specified signal is still something we are interested in. If not, we'll unmask it, |
810 | * and possibly drop the signalfd for it. */ | |
9da4cb2b LP |
811 | |
812 | if (sig == SIGCHLD && | |
b6d5481b | 813 | e->n_online_child_sources > 0) |
9da4cb2b LP |
814 | return; |
815 | ||
816 | if (e->signal_sources && | |
817 | e->signal_sources[sig] && | |
b6d5481b | 818 | event_source_is_online(e->signal_sources[sig])) |
9da4cb2b LP |
819 | return; |
820 | ||
821 | /* | |
822 | * The specified signal might be enabled in three different queues: | |
823 | * | |
824 | * 1) the one that belongs to the priority passed (if it is non-NULL) | |
825 | * 2) the one that belongs to the priority of the event source of the signal (if there is one) | |
826 | * 3) the 0 priority (to cover the SIGCHLD case) | |
827 | * | |
828 | * Hence, let's remove it from all three here. | |
829 | */ | |
830 | ||
831 | if (priority) { | |
832 | d = hashmap_get(e->signal_data, priority); | |
833 | if (d) | |
834 | event_unmask_signal_data(e, d, sig); | |
835 | } | |
836 | ||
837 | if (e->signal_sources && e->signal_sources[sig]) { | |
838 | d = hashmap_get(e->signal_data, &e->signal_sources[sig]->priority); | |
839 | if (d) | |
840 | event_unmask_signal_data(e, d, sig); | |
841 | } | |
842 | ||
843 | d = hashmap_get(e->signal_data, &zero_priority); | |
844 | if (d) | |
845 | event_unmask_signal_data(e, d, sig); | |
f95387cd ZJS |
846 | } |
847 | ||
e1951c16 MS |
848 | static void event_source_pp_prioq_reshuffle(sd_event_source *s) { |
849 | assert(s); | |
850 | ||
851 | /* Reshuffles the pending + prepare prioqs. Called whenever the dispatch order changes, i.e. when | |
852 | * they are enabled/disabled or marked pending and such. */ | |
853 | ||
854 | if (s->pending) | |
855 | prioq_reshuffle(s->event->pending, s, &s->pending_index); | |
856 | ||
857 | if (s->prepare) | |
858 | prioq_reshuffle(s->event->prepare, s, &s->prepare_index); | |
859 | } | |
860 | ||
861 | static void event_source_time_prioq_reshuffle(sd_event_source *s) { | |
862 | struct clock_data *d; | |
863 | ||
864 | assert(s); | |
e1951c16 MS |
865 | |
866 | /* Called whenever the event source's timer ordering properties changed, i.e. time, accuracy, | |
5c08c7ab YW |
867 | * pending, enable state, and ratelimiting state. Makes sure the two prioq's are ordered |
868 | * properly again. */ | |
b6d5481b LP |
869 | |
870 | if (s->ratelimited) | |
871 | d = &s->event->monotonic; | |
5c08c7ab | 872 | else if (EVENT_SOURCE_IS_TIME(s->type)) |
b6d5481b | 873 | assert_se(d = event_get_clock_data(s->event, s->type)); |
5c08c7ab YW |
874 | else |
875 | return; /* no-op for an event source which is neither a timer nor ratelimited. */ | |
b6d5481b | 876 | |
f41315fc LP |
877 | prioq_reshuffle(d->earliest, s, &s->earliest_index); |
878 | prioq_reshuffle(d->latest, s, &s->latest_index); | |
e1951c16 MS |
879 | d->needs_rearm = true; |
880 | } | |
881 | ||
1e45e3fe LP |
882 | static void event_source_time_prioq_remove( |
883 | sd_event_source *s, | |
884 | struct clock_data *d) { | |
885 | ||
886 | assert(s); | |
887 | assert(d); | |
888 | ||
f41315fc LP |
889 | prioq_remove(d->earliest, s, &s->earliest_index); |
890 | prioq_remove(d->latest, s, &s->latest_index); | |
891 | s->earliest_index = s->latest_index = PRIOQ_IDX_NULL; | |
1e45e3fe LP |
892 | d->needs_rearm = true; |
893 | } | |
894 | ||
a71fe8b8 LP |
895 | static void source_disconnect(sd_event_source *s) { |
896 | sd_event *event; | |
897448bd | 897 | int r; |
a71fe8b8 | 898 | |
fd38203a LP |
899 | assert(s); |
900 | ||
a71fe8b8 LP |
901 | if (!s->event) |
902 | return; | |
15b38f93 | 903 | |
a71fe8b8 | 904 | assert(s->event->n_sources > 0); |
fd38203a | 905 | |
a71fe8b8 | 906 | switch (s->type) { |
fd38203a | 907 | |
a71fe8b8 LP |
908 | case SOURCE_IO: |
909 | if (s->io.fd >= 0) | |
910 | source_io_unregister(s); | |
fd38203a | 911 | |
a71fe8b8 | 912 | break; |
6a0f1f6d | 913 | |
a71fe8b8 | 914 | case SOURCE_TIME_REALTIME: |
a8548816 | 915 | case SOURCE_TIME_BOOTTIME: |
a71fe8b8 LP |
916 | case SOURCE_TIME_MONOTONIC: |
917 | case SOURCE_TIME_REALTIME_ALARM: | |
b6d5481b LP |
918 | case SOURCE_TIME_BOOTTIME_ALARM: |
919 | /* Only remove this event source from the time event source here if it is not ratelimited. If | |
920 | * it is ratelimited, we'll remove it below, separately. Why? Because the clock used might | |
921 | * differ: ratelimiting always uses CLOCK_MONOTONIC, but timer events might use any clock */ | |
922 | ||
923 | if (!s->ratelimited) { | |
924 | struct clock_data *d; | |
925 | assert_se(d = event_get_clock_data(s->event, s->type)); | |
926 | event_source_time_prioq_remove(s, d); | |
927 | } | |
928 | ||
a71fe8b8 | 929 | break; |
a71fe8b8 LP |
930 | |
931 | case SOURCE_SIGNAL: | |
932 | if (s->signal.sig > 0) { | |
9da4cb2b | 933 | |
a71fe8b8 LP |
934 | if (s->event->signal_sources) |
935 | s->event->signal_sources[s->signal.sig] = NULL; | |
4807d2d0 | 936 | |
9da4cb2b | 937 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
897448bd LP |
938 | |
939 | if (s->signal.unblock) { | |
940 | sigset_t new_ss; | |
941 | ||
942 | if (sigemptyset(&new_ss) < 0) | |
943 | log_debug_errno(errno, "Failed to reset signal set, ignoring: %m"); | |
944 | else if (sigaddset(&new_ss, s->signal.sig) < 0) | |
945 | log_debug_errno(errno, "Failed to add signal %i to signal mask, ignoring: %m", s->signal.sig); | |
946 | else { | |
947 | r = pthread_sigmask(SIG_UNBLOCK, &new_ss, NULL); | |
948 | if (r != 0) | |
949 | log_debug_errno(r, "Failed to unblock signal %i, ignoring: %m", s->signal.sig); | |
950 | } | |
951 | } | |
6a0f1f6d | 952 | } |
fd38203a | 953 | |
a71fe8b8 | 954 | break; |
fd38203a | 955 | |
a71fe8b8 | 956 | case SOURCE_CHILD: |
86587c93 YW |
957 | if (event_pid_changed(s->event)) |
958 | s->child.process_owned = false; | |
959 | ||
a71fe8b8 | 960 | if (s->child.pid > 0) { |
b6d5481b LP |
961 | if (event_source_is_online(s)) { |
962 | assert(s->event->n_online_child_sources > 0); | |
963 | s->event->n_online_child_sources--; | |
4807d2d0 | 964 | } |
fd38203a | 965 | |
4a0b58c4 | 966 | (void) hashmap_remove(s->event->child_sources, PID_TO_PTR(s->child.pid)); |
a71fe8b8 | 967 | } |
fd38203a | 968 | |
f8f3f926 LP |
969 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
970 | source_child_pidfd_unregister(s); | |
971 | else | |
972 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
973 | ||
a71fe8b8 | 974 | break; |
fd38203a | 975 | |
a71fe8b8 LP |
976 | case SOURCE_DEFER: |
977 | /* nothing */ | |
978 | break; | |
fd38203a | 979 | |
a71fe8b8 LP |
980 | case SOURCE_POST: |
981 | set_remove(s->event->post_sources, s); | |
982 | break; | |
da7e457c | 983 | |
a71fe8b8 LP |
984 | case SOURCE_EXIT: |
985 | prioq_remove(s->event->exit, s, &s->exit.prioq_index); | |
986 | break; | |
0eb2e0e3 | 987 | |
97ef5391 LP |
988 | case SOURCE_INOTIFY: { |
989 | struct inode_data *inode_data; | |
990 | ||
991 | inode_data = s->inotify.inode_data; | |
992 | if (inode_data) { | |
993 | struct inotify_data *inotify_data; | |
994 | assert_se(inotify_data = inode_data->inotify_data); | |
995 | ||
996 | /* Detach this event source from the inode object */ | |
997 | LIST_REMOVE(inotify.by_inode_data, inode_data->event_sources, s); | |
998 | s->inotify.inode_data = NULL; | |
999 | ||
1000 | if (s->pending) { | |
1001 | assert(inotify_data->n_pending > 0); | |
1002 | inotify_data->n_pending--; | |
1003 | } | |
1004 | ||
1005 | /* Note that we don't reduce the inotify mask for the watch descriptor here if the inode is | |
1006 | * continued to being watched. That's because inotify doesn't really have an API for that: we | |
1007 | * can only change watch masks with access to the original inode either by fd or by path. But | |
1008 | * paths aren't stable, and keeping an O_PATH fd open all the time would mean wasting an fd | |
f21f31b2 | 1009 | * continuously and keeping the mount busy which we can't really do. We could reconstruct the |
97ef5391 LP |
1010 | * original inode from /proc/self/fdinfo/$INOTIFY_FD (as all watch descriptors are listed |
1011 | * there), but given the need for open_by_handle_at() which is privileged and not universally | |
1012 | * available this would be quite an incomplete solution. Hence we go the other way, leave the | |
1013 | * mask set, even if it is not minimized now, and ignore all events we aren't interested in | |
1014 | * anymore after reception. Yes, this sucks, but … Linux … */ | |
1015 | ||
1016 | /* Maybe release the inode data (and its inotify) */ | |
1017 | event_gc_inode_data(s->event, inode_data); | |
1018 | } | |
1019 | ||
1020 | break; | |
1021 | } | |
1022 | ||
158fe190 LP |
1023 | case SOURCE_MEMORY_PRESSURE: |
1024 | source_memory_pressure_remove_from_write_list(s); | |
1025 | source_memory_pressure_unregister(s); | |
1026 | break; | |
1027 | ||
a71fe8b8 | 1028 | default: |
04499a70 | 1029 | assert_not_reached(); |
a71fe8b8 | 1030 | } |
6e9feda3 | 1031 | |
a71fe8b8 LP |
1032 | if (s->pending) |
1033 | prioq_remove(s->event->pending, s, &s->pending_index); | |
9d3e3aa5 | 1034 | |
a71fe8b8 LP |
1035 | if (s->prepare) |
1036 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
fd38203a | 1037 | |
b6d5481b LP |
1038 | if (s->ratelimited) |
1039 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
1040 | ||
e514aa1e | 1041 | event = TAKE_PTR(s->event); |
a71fe8b8 LP |
1042 | LIST_REMOVE(sources, event->sources, s); |
1043 | event->n_sources--; | |
fd38203a | 1044 | |
f5982559 LP |
1045 | /* Note that we don't invalidate the type here, since we still need it in order to close the fd or |
1046 | * pidfd associated with this event source, which we'll do only on source_free(). */ | |
1047 | ||
a71fe8b8 LP |
1048 | if (!s->floating) |
1049 | sd_event_unref(event); | |
1050 | } | |
1051 | ||
75db809a | 1052 | static sd_event_source* source_free(sd_event_source *s) { |
a71fe8b8 | 1053 | assert(s); |
fd38203a | 1054 | |
a71fe8b8 | 1055 | source_disconnect(s); |
ab93297c NM |
1056 | |
1057 | if (s->type == SOURCE_IO && s->io.owned) | |
15723a1d LP |
1058 | s->io.fd = safe_close(s->io.fd); |
1059 | ||
f8f3f926 LP |
1060 | if (s->type == SOURCE_CHILD) { |
1061 | /* Eventually the kernel will do this automatically for us, but for now let's emulate this (unreliably) in userspace. */ | |
1062 | ||
1063 | if (s->child.process_owned) { | |
1064 | ||
1065 | if (!s->child.exited) { | |
1066 | bool sent = false; | |
1067 | ||
1068 | if (s->child.pidfd >= 0) { | |
1069 | if (pidfd_send_signal(s->child.pidfd, SIGKILL, NULL, 0) < 0) { | |
1070 | if (errno == ESRCH) /* Already dead */ | |
1071 | sent = true; | |
1072 | else if (!ERRNO_IS_NOT_SUPPORTED(errno)) | |
1073 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via pidfd_send_signal(), re-trying via kill(): %m", | |
1074 | s->child.pid); | |
1075 | } else | |
1076 | sent = true; | |
1077 | } | |
1078 | ||
1079 | if (!sent) | |
1080 | if (kill(s->child.pid, SIGKILL) < 0) | |
1081 | if (errno != ESRCH) /* Already dead */ | |
1082 | log_debug_errno(errno, "Failed to kill process " PID_FMT " via kill(), ignoring: %m", | |
1083 | s->child.pid); | |
1084 | } | |
1085 | ||
1086 | if (!s->child.waited) { | |
1087 | siginfo_t si = {}; | |
1088 | ||
1089 | /* Reap the child if we can */ | |
1090 | (void) waitid(P_PID, s->child.pid, &si, WEXITED); | |
1091 | } | |
1092 | } | |
1093 | ||
1094 | if (s->child.pidfd_owned) | |
1095 | s->child.pidfd = safe_close(s->child.pidfd); | |
1096 | } | |
1097 | ||
158fe190 LP |
1098 | if (s->type == SOURCE_MEMORY_PRESSURE) { |
1099 | s->memory_pressure.fd = safe_close(s->memory_pressure.fd); | |
1100 | s->memory_pressure.write_buffer = mfree(s->memory_pressure.write_buffer); | |
1101 | } | |
1102 | ||
15723a1d LP |
1103 | if (s->destroy_callback) |
1104 | s->destroy_callback(s->userdata); | |
ab93297c | 1105 | |
356779df | 1106 | free(s->description); |
75db809a | 1107 | return mfree(s); |
fd38203a | 1108 | } |
8c75fe17 | 1109 | DEFINE_TRIVIAL_CLEANUP_FUNC(sd_event_source*, source_free); |
fd38203a LP |
1110 | |
1111 | static int source_set_pending(sd_event_source *s, bool b) { | |
1112 | int r; | |
1113 | ||
1114 | assert(s); | |
6203e07a | 1115 | assert(s->type != SOURCE_EXIT); |
fd38203a LP |
1116 | |
1117 | if (s->pending == b) | |
1118 | return 0; | |
1119 | ||
1120 | s->pending = b; | |
1121 | ||
1122 | if (b) { | |
1123 | s->pending_iteration = s->event->iteration; | |
1124 | ||
1125 | r = prioq_put(s->event->pending, s, &s->pending_index); | |
1126 | if (r < 0) { | |
1127 | s->pending = false; | |
1128 | return r; | |
1129 | } | |
1130 | } else | |
1131 | assert_se(prioq_remove(s->event->pending, s, &s->pending_index)); | |
1132 | ||
e1951c16 MS |
1133 | if (EVENT_SOURCE_IS_TIME(s->type)) |
1134 | event_source_time_prioq_reshuffle(s); | |
2576a19e | 1135 | |
9da4cb2b LP |
1136 | if (s->type == SOURCE_SIGNAL && !b) { |
1137 | struct signal_data *d; | |
1138 | ||
1139 | d = hashmap_get(s->event->signal_data, &s->priority); | |
1140 | if (d && d->current == s) | |
1141 | d->current = NULL; | |
1142 | } | |
1143 | ||
97ef5391 LP |
1144 | if (s->type == SOURCE_INOTIFY) { |
1145 | ||
1146 | assert(s->inotify.inode_data); | |
1147 | assert(s->inotify.inode_data->inotify_data); | |
1148 | ||
1149 | if (b) | |
1150 | s->inotify.inode_data->inotify_data->n_pending ++; | |
1151 | else { | |
1152 | assert(s->inotify.inode_data->inotify_data->n_pending > 0); | |
1153 | s->inotify.inode_data->inotify_data->n_pending --; | |
1154 | } | |
1155 | } | |
1156 | ||
efd3be9d | 1157 | return 1; |
fd38203a LP |
1158 | } |
1159 | ||
a71fe8b8 | 1160 | static sd_event_source *source_new(sd_event *e, bool floating, EventSourceType type) { |
a38cf9fb LP |
1161 | |
1162 | /* Let's allocate exactly what we need. Note that the difference of the smallest event source | |
1163 | * structure to the largest is 144 bytes on x86-64 at the time of writing, i.e. more than two cache | |
1164 | * lines. */ | |
1165 | static const size_t size_table[_SOURCE_EVENT_SOURCE_TYPE_MAX] = { | |
1166 | [SOURCE_IO] = endoffsetof_field(sd_event_source, io), | |
1167 | [SOURCE_TIME_REALTIME] = endoffsetof_field(sd_event_source, time), | |
1168 | [SOURCE_TIME_BOOTTIME] = endoffsetof_field(sd_event_source, time), | |
1169 | [SOURCE_TIME_MONOTONIC] = endoffsetof_field(sd_event_source, time), | |
1170 | [SOURCE_TIME_REALTIME_ALARM] = endoffsetof_field(sd_event_source, time), | |
1171 | [SOURCE_TIME_BOOTTIME_ALARM] = endoffsetof_field(sd_event_source, time), | |
1172 | [SOURCE_SIGNAL] = endoffsetof_field(sd_event_source, signal), | |
1173 | [SOURCE_CHILD] = endoffsetof_field(sd_event_source, child), | |
1174 | [SOURCE_DEFER] = endoffsetof_field(sd_event_source, defer), | |
1175 | [SOURCE_POST] = endoffsetof_field(sd_event_source, post), | |
1176 | [SOURCE_EXIT] = endoffsetof_field(sd_event_source, exit), | |
1177 | [SOURCE_INOTIFY] = endoffsetof_field(sd_event_source, inotify), | |
158fe190 | 1178 | [SOURCE_MEMORY_PRESSURE] = endoffsetof_field(sd_event_source, memory_pressure), |
a38cf9fb LP |
1179 | }; |
1180 | ||
fd38203a LP |
1181 | sd_event_source *s; |
1182 | ||
1183 | assert(e); | |
a38cf9fb LP |
1184 | assert(type >= 0); |
1185 | assert(type < _SOURCE_EVENT_SOURCE_TYPE_MAX); | |
1186 | assert(size_table[type] > 0); | |
fd38203a | 1187 | |
a38cf9fb LP |
1188 | /* We use expand_to_usable() here to tell gcc that it should consider this an object of the full |
1189 | * size, even if we only allocate the initial part we need. */ | |
1190 | s = expand_to_usable(malloc0(size_table[type]), sizeof(sd_event_source)); | |
fd38203a LP |
1191 | if (!s) |
1192 | return NULL; | |
1193 | ||
a38cf9fb LP |
1194 | /* Note: we cannot use compound initialization here, because sizeof(sd_event_source) is likely larger |
1195 | * than what we allocated here. */ | |
1196 | s->n_ref = 1; | |
1197 | s->event = e; | |
1198 | s->floating = floating; | |
1199 | s->type = type; | |
1200 | s->pending_index = PRIOQ_IDX_NULL; | |
1201 | s->prepare_index = PRIOQ_IDX_NULL; | |
a71fe8b8 LP |
1202 | |
1203 | if (!floating) | |
1204 | sd_event_ref(e); | |
fd38203a | 1205 | |
a71fe8b8 | 1206 | LIST_PREPEND(sources, e->sources, s); |
313cefa1 | 1207 | e->n_sources++; |
15b38f93 | 1208 | |
fd38203a LP |
1209 | return s; |
1210 | } | |
1211 | ||
b9350e70 LP |
1212 | static int io_exit_callback(sd_event_source *s, int fd, uint32_t revents, void *userdata) { |
1213 | assert(s); | |
1214 | ||
1215 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1216 | } | |
1217 | ||
f7262a9f | 1218 | _public_ int sd_event_add_io( |
fd38203a | 1219 | sd_event *e, |
151b9b96 | 1220 | sd_event_source **ret, |
fd38203a LP |
1221 | int fd, |
1222 | uint32_t events, | |
718db961 | 1223 | sd_event_io_handler_t callback, |
151b9b96 | 1224 | void *userdata) { |
fd38203a | 1225 | |
ec766a51 | 1226 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1227 | int r; |
1228 | ||
305f78bf | 1229 | assert_return(e, -EINVAL); |
b937d761 | 1230 | assert_return(e = event_resolve(e), -ENOPKG); |
8ac43fee | 1231 | assert_return(fd >= 0, -EBADF); |
2a16a986 | 1232 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 1233 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1234 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1235 | |
b9350e70 LP |
1236 | if (!callback) |
1237 | callback = io_exit_callback; | |
1238 | ||
a71fe8b8 | 1239 | s = source_new(e, !ret, SOURCE_IO); |
fd38203a LP |
1240 | if (!s) |
1241 | return -ENOMEM; | |
1242 | ||
9da4cb2b | 1243 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1244 | s->io.fd = fd; |
1245 | s->io.events = events; | |
1246 | s->io.callback = callback; | |
1247 | s->userdata = userdata; | |
baf76283 | 1248 | s->enabled = SD_EVENT_ON; |
fd38203a | 1249 | |
baf76283 | 1250 | r = source_io_register(s, s->enabled, events); |
ec766a51 | 1251 | if (r < 0) |
050f74f2 | 1252 | return r; |
fd38203a | 1253 | |
a71fe8b8 LP |
1254 | if (ret) |
1255 | *ret = s; | |
ec766a51 | 1256 | TAKE_PTR(s); |
a71fe8b8 | 1257 | |
fd38203a LP |
1258 | return 0; |
1259 | } | |
1260 | ||
52444dc4 | 1261 | static void initialize_perturb(sd_event *e) { |
6d2326e0 | 1262 | sd_id128_t id = {}; |
52444dc4 | 1263 | |
6d2326e0 YW |
1264 | /* When we sleep for longer, we try to realign the wakeup to the same time within each |
1265 | * minute/second/250ms, so that events all across the system can be coalesced into a single CPU | |
1266 | * wakeup. However, let's take some system-specific randomness for this value, so that in a network | |
1267 | * of systems with synced clocks timer events are distributed a bit. Here, we calculate a | |
1268 | * perturbation usec offset from the boot ID (or machine ID if failed, e.g. /proc is not mounted). */ | |
52444dc4 | 1269 | |
3a43da28 | 1270 | if (_likely_(e->perturb != USEC_INFINITY)) |
52444dc4 LP |
1271 | return; |
1272 | ||
1912f790 | 1273 | if (sd_id128_get_boot(&id) >= 0 || sd_id128_get_machine(&id) >= 0) |
6d2326e0 YW |
1274 | e->perturb = (id.qwords[0] ^ id.qwords[1]) % USEC_PER_MINUTE; |
1275 | else | |
1276 | e->perturb = 0; /* This is a super early process without /proc and /etc ?? */ | |
52444dc4 LP |
1277 | } |
1278 | ||
fd38203a LP |
1279 | static int event_setup_timer_fd( |
1280 | sd_event *e, | |
6a0f1f6d LP |
1281 | struct clock_data *d, |
1282 | clockid_t clock) { | |
fd38203a | 1283 | |
fd38203a | 1284 | assert(e); |
6a0f1f6d | 1285 | assert(d); |
fd38203a | 1286 | |
6a0f1f6d | 1287 | if (_likely_(d->fd >= 0)) |
fd38203a LP |
1288 | return 0; |
1289 | ||
254d1313 | 1290 | _cleanup_close_ int fd = -EBADF; |
b44d87e2 | 1291 | |
6a0f1f6d | 1292 | fd = timerfd_create(clock, TFD_NONBLOCK|TFD_CLOEXEC); |
fd38203a LP |
1293 | if (fd < 0) |
1294 | return -errno; | |
1295 | ||
7fe2903c LP |
1296 | fd = fd_move_above_stdio(fd); |
1297 | ||
1eac7948 | 1298 | struct epoll_event ev = { |
a82f89aa LP |
1299 | .events = EPOLLIN, |
1300 | .data.ptr = d, | |
1301 | }; | |
fd38203a | 1302 | |
15c689d7 | 1303 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, fd, &ev) < 0) |
fd38203a | 1304 | return -errno; |
fd38203a | 1305 | |
b44d87e2 | 1306 | d->fd = TAKE_FD(fd); |
fd38203a LP |
1307 | return 0; |
1308 | } | |
1309 | ||
c4f1aff2 TG |
1310 | static int time_exit_callback(sd_event_source *s, uint64_t usec, void *userdata) { |
1311 | assert(s); | |
1312 | ||
1313 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1314 | } | |
1315 | ||
41c63f36 LP |
1316 | static int setup_clock_data(sd_event *e, struct clock_data *d, clockid_t clock) { |
1317 | int r; | |
1318 | ||
1319 | assert(d); | |
1320 | ||
1321 | if (d->fd < 0) { | |
1322 | r = event_setup_timer_fd(e, d, clock); | |
1323 | if (r < 0) | |
1324 | return r; | |
1325 | } | |
1326 | ||
1327 | r = prioq_ensure_allocated(&d->earliest, earliest_time_prioq_compare); | |
1328 | if (r < 0) | |
1329 | return r; | |
1330 | ||
1331 | r = prioq_ensure_allocated(&d->latest, latest_time_prioq_compare); | |
1332 | if (r < 0) | |
1333 | return r; | |
1334 | ||
1335 | return 0; | |
1336 | } | |
1337 | ||
1e45e3fe LP |
1338 | static int event_source_time_prioq_put( |
1339 | sd_event_source *s, | |
1340 | struct clock_data *d) { | |
1341 | ||
1342 | int r; | |
1343 | ||
1344 | assert(s); | |
1345 | assert(d); | |
19947509 | 1346 | assert(EVENT_SOURCE_USES_TIME_PRIOQ(s->type)); |
1e45e3fe | 1347 | |
f41315fc | 1348 | r = prioq_put(d->earliest, s, &s->earliest_index); |
1e45e3fe LP |
1349 | if (r < 0) |
1350 | return r; | |
1351 | ||
f41315fc | 1352 | r = prioq_put(d->latest, s, &s->latest_index); |
1e45e3fe | 1353 | if (r < 0) { |
f41315fc LP |
1354 | assert_se(prioq_remove(d->earliest, s, &s->earliest_index) > 0); |
1355 | s->earliest_index = PRIOQ_IDX_NULL; | |
1e45e3fe LP |
1356 | return r; |
1357 | } | |
1358 | ||
1359 | d->needs_rearm = true; | |
1360 | return 0; | |
1361 | } | |
1362 | ||
6a0f1f6d | 1363 | _public_ int sd_event_add_time( |
fd38203a | 1364 | sd_event *e, |
151b9b96 | 1365 | sd_event_source **ret, |
6a0f1f6d | 1366 | clockid_t clock, |
fd38203a | 1367 | uint64_t usec, |
c2ba3ad6 | 1368 | uint64_t accuracy, |
718db961 | 1369 | sd_event_time_handler_t callback, |
151b9b96 | 1370 | void *userdata) { |
fd38203a | 1371 | |
6a0f1f6d | 1372 | EventSourceType type; |
ec766a51 | 1373 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6a0f1f6d | 1374 | struct clock_data *d; |
fd38203a LP |
1375 | int r; |
1376 | ||
305f78bf | 1377 | assert_return(e, -EINVAL); |
b937d761 | 1378 | assert_return(e = event_resolve(e), -ENOPKG); |
f5fbe71d | 1379 | assert_return(accuracy != UINT64_MAX, -EINVAL); |
da7e457c | 1380 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1381 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1382 | |
e475d10c LP |
1383 | if (!clock_supported(clock)) /* Checks whether the kernel supports the clock */ |
1384 | return -EOPNOTSUPP; | |
1385 | ||
1386 | type = clock_to_event_source_type(clock); /* checks whether sd-event supports this clock */ | |
1387 | if (type < 0) | |
3411372e LP |
1388 | return -EOPNOTSUPP; |
1389 | ||
c4f1aff2 TG |
1390 | if (!callback) |
1391 | callback = time_exit_callback; | |
1392 | ||
1e45e3fe | 1393 | assert_se(d = event_get_clock_data(e, type)); |
c2ba3ad6 | 1394 | |
41c63f36 | 1395 | r = setup_clock_data(e, d, clock); |
c983e776 EV |
1396 | if (r < 0) |
1397 | return r; | |
fd38203a | 1398 | |
a71fe8b8 | 1399 | s = source_new(e, !ret, type); |
fd38203a LP |
1400 | if (!s) |
1401 | return -ENOMEM; | |
1402 | ||
1403 | s->time.next = usec; | |
c2ba3ad6 | 1404 | s->time.accuracy = accuracy == 0 ? DEFAULT_ACCURACY_USEC : accuracy; |
fd38203a | 1405 | s->time.callback = callback; |
f41315fc | 1406 | s->earliest_index = s->latest_index = PRIOQ_IDX_NULL; |
fd38203a | 1407 | s->userdata = userdata; |
baf76283 | 1408 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1409 | |
1e45e3fe | 1410 | r = event_source_time_prioq_put(s, d); |
c2ba3ad6 | 1411 | if (r < 0) |
ec766a51 | 1412 | return r; |
fd38203a | 1413 | |
a71fe8b8 LP |
1414 | if (ret) |
1415 | *ret = s; | |
ec766a51 | 1416 | TAKE_PTR(s); |
a71fe8b8 | 1417 | |
fd38203a LP |
1418 | return 0; |
1419 | } | |
1420 | ||
d6a83dc4 LP |
1421 | _public_ int sd_event_add_time_relative( |
1422 | sd_event *e, | |
1423 | sd_event_source **ret, | |
1424 | clockid_t clock, | |
1425 | uint64_t usec, | |
1426 | uint64_t accuracy, | |
1427 | sd_event_time_handler_t callback, | |
1428 | void *userdata) { | |
1429 | ||
1430 | usec_t t; | |
1431 | int r; | |
1432 | ||
1433 | /* Same as sd_event_add_time() but operates relative to the event loop's current point in time, and | |
1434 | * checks for overflow. */ | |
1435 | ||
1436 | r = sd_event_now(e, clock, &t); | |
1437 | if (r < 0) | |
1438 | return r; | |
1439 | ||
1440 | if (usec >= USEC_INFINITY - t) | |
1441 | return -EOVERFLOW; | |
1442 | ||
1443 | return sd_event_add_time(e, ret, clock, t + usec, accuracy, callback, userdata); | |
1444 | } | |
1445 | ||
59bc1fd7 LP |
1446 | static int signal_exit_callback(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) { |
1447 | assert(s); | |
1448 | ||
1449 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1450 | } | |
1451 | ||
f7262a9f | 1452 | _public_ int sd_event_add_signal( |
305f78bf | 1453 | sd_event *e, |
151b9b96 | 1454 | sd_event_source **ret, |
305f78bf | 1455 | int sig, |
718db961 | 1456 | sd_event_signal_handler_t callback, |
151b9b96 | 1457 | void *userdata) { |
305f78bf | 1458 | |
ec766a51 | 1459 | _cleanup_(source_freep) sd_event_source *s = NULL; |
9da4cb2b | 1460 | struct signal_data *d; |
897448bd LP |
1461 | sigset_t new_ss; |
1462 | bool block_it; | |
fd38203a LP |
1463 | int r; |
1464 | ||
305f78bf | 1465 | assert_return(e, -EINVAL); |
b937d761 | 1466 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1467 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1468 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1469 | |
897448bd LP |
1470 | /* Let's make sure our special flag stays outside of the valid signal range */ |
1471 | assert_cc(_NSIG < SD_EVENT_SIGNAL_PROCMASK); | |
1472 | ||
1473 | if (sig & SD_EVENT_SIGNAL_PROCMASK) { | |
1474 | sig &= ~SD_EVENT_SIGNAL_PROCMASK; | |
1475 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
1476 | ||
1477 | block_it = true; | |
1478 | } else { | |
1479 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
1480 | ||
1481 | r = signal_is_blocked(sig); | |
1482 | if (r < 0) | |
1483 | return r; | |
1484 | if (r == 0) | |
1485 | return -EBUSY; | |
1486 | ||
1487 | block_it = false; | |
1488 | } | |
1489 | ||
59bc1fd7 LP |
1490 | if (!callback) |
1491 | callback = signal_exit_callback; | |
1492 | ||
fd38203a LP |
1493 | if (!e->signal_sources) { |
1494 | e->signal_sources = new0(sd_event_source*, _NSIG); | |
1495 | if (!e->signal_sources) | |
1496 | return -ENOMEM; | |
1497 | } else if (e->signal_sources[sig]) | |
1498 | return -EBUSY; | |
1499 | ||
a71fe8b8 | 1500 | s = source_new(e, !ret, SOURCE_SIGNAL); |
fd38203a LP |
1501 | if (!s) |
1502 | return -ENOMEM; | |
1503 | ||
1504 | s->signal.sig = sig; | |
1505 | s->signal.callback = callback; | |
1506 | s->userdata = userdata; | |
baf76283 | 1507 | s->enabled = SD_EVENT_ON; |
fd38203a LP |
1508 | |
1509 | e->signal_sources[sig] = s; | |
fd38203a | 1510 | |
897448bd LP |
1511 | if (block_it) { |
1512 | sigset_t old_ss; | |
1513 | ||
1514 | if (sigemptyset(&new_ss) < 0) | |
1515 | return -errno; | |
1516 | ||
1517 | if (sigaddset(&new_ss, sig) < 0) | |
1518 | return -errno; | |
1519 | ||
1520 | r = pthread_sigmask(SIG_BLOCK, &new_ss, &old_ss); | |
1521 | if (r != 0) | |
1522 | return -r; | |
1523 | ||
1524 | r = sigismember(&old_ss, sig); | |
1525 | if (r < 0) | |
1526 | return -errno; | |
1527 | ||
1528 | s->signal.unblock = !r; | |
1529 | } else | |
1530 | s->signal.unblock = false; | |
1531 | ||
9da4cb2b | 1532 | r = event_make_signal_data(e, sig, &d); |
897448bd LP |
1533 | if (r < 0) { |
1534 | if (s->signal.unblock) | |
1535 | (void) pthread_sigmask(SIG_UNBLOCK, &new_ss, NULL); | |
1536 | ||
9da4cb2b | 1537 | return r; |
897448bd | 1538 | } |
fd38203a | 1539 | |
f1f00dbb LP |
1540 | /* Use the signal name as description for the event source by default */ |
1541 | (void) sd_event_source_set_description(s, signal_to_string(sig)); | |
1542 | ||
a71fe8b8 LP |
1543 | if (ret) |
1544 | *ret = s; | |
ec766a51 | 1545 | TAKE_PTR(s); |
a71fe8b8 | 1546 | |
fd38203a LP |
1547 | return 0; |
1548 | } | |
1549 | ||
b9350e70 LP |
1550 | static int child_exit_callback(sd_event_source *s, const siginfo_t *si, void *userdata) { |
1551 | assert(s); | |
1552 | ||
1553 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1554 | } | |
1555 | ||
f8f3f926 LP |
1556 | static bool shall_use_pidfd(void) { |
1557 | /* Mostly relevant for debugging, i.e. this is used in test-event.c to test the event loop once with and once without pidfd */ | |
1558 | return getenv_bool_secure("SYSTEMD_PIDFD") != 0; | |
1559 | } | |
1560 | ||
f7262a9f | 1561 | _public_ int sd_event_add_child( |
305f78bf | 1562 | sd_event *e, |
151b9b96 | 1563 | sd_event_source **ret, |
305f78bf LP |
1564 | pid_t pid, |
1565 | int options, | |
718db961 | 1566 | sd_event_child_handler_t callback, |
151b9b96 | 1567 | void *userdata) { |
305f78bf | 1568 | |
ec766a51 | 1569 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1570 | int r; |
1571 | ||
305f78bf | 1572 | assert_return(e, -EINVAL); |
b937d761 | 1573 | assert_return(e = event_resolve(e), -ENOPKG); |
305f78bf LP |
1574 | assert_return(pid > 1, -EINVAL); |
1575 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1576 | assert_return(options != 0, -EINVAL); | |
da7e457c | 1577 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1578 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1579 | |
b9350e70 LP |
1580 | if (!callback) |
1581 | callback = child_exit_callback; | |
1582 | ||
b6d5481b | 1583 | if (e->n_online_child_sources == 0) { |
ee880b37 LP |
1584 | /* Caller must block SIGCHLD before using us to watch children, even if pidfd is available, |
1585 | * for compatibility with pre-pidfd and because we don't want the reap the child processes | |
1586 | * ourselves, i.e. call waitid(), and don't want Linux' default internal logic for that to | |
1587 | * take effect. | |
1588 | * | |
1589 | * (As an optimization we only do this check on the first child event source created.) */ | |
1590 | r = signal_is_blocked(SIGCHLD); | |
1591 | if (r < 0) | |
1592 | return r; | |
1593 | if (r == 0) | |
1594 | return -EBUSY; | |
1595 | } | |
1596 | ||
d5099efc | 1597 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
fd38203a LP |
1598 | if (r < 0) |
1599 | return r; | |
1600 | ||
4a0b58c4 | 1601 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) |
fd38203a LP |
1602 | return -EBUSY; |
1603 | ||
a71fe8b8 | 1604 | s = source_new(e, !ret, SOURCE_CHILD); |
fd38203a LP |
1605 | if (!s) |
1606 | return -ENOMEM; | |
1607 | ||
f8f3f926 | 1608 | s->wakeup = WAKEUP_EVENT_SOURCE; |
fd38203a LP |
1609 | s->child.options = options; |
1610 | s->child.callback = callback; | |
1611 | s->userdata = userdata; | |
baf76283 | 1612 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a | 1613 | |
f8f3f926 LP |
1614 | /* We always take a pidfd here if we can, even if we wait for anything else than WEXITED, so that we |
1615 | * pin the PID, and make regular waitid() handling race-free. */ | |
1616 | ||
1617 | if (shall_use_pidfd()) { | |
54988a27 | 1618 | s->child.pidfd = pidfd_open(pid, 0); |
f8f3f926 LP |
1619 | if (s->child.pidfd < 0) { |
1620 | /* Propagate errors unless the syscall is not supported or blocked */ | |
1621 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
1622 | return -errno; | |
1623 | } else | |
1624 | s->child.pidfd_owned = true; /* If we allocate the pidfd we own it by default */ | |
1625 | } else | |
254d1313 | 1626 | s->child.pidfd = -EBADF; |
f8f3f926 | 1627 | |
f8f3f926 LP |
1628 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1629 | /* We have a pidfd and we only want to watch for exit */ | |
f8f3f926 | 1630 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1631 | if (r < 0) |
f8f3f926 | 1632 | return r; |
ac9f2640 | 1633 | |
f8f3f926 LP |
1634 | } else { |
1635 | /* We have no pidfd or we shall wait for some other event than WEXITED */ | |
f8f3f926 | 1636 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1637 | if (r < 0) |
f8f3f926 | 1638 | return r; |
f8f3f926 LP |
1639 | |
1640 | e->need_process_child = true; | |
1641 | } | |
c2ba3ad6 | 1642 | |
54988a27 YW |
1643 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); |
1644 | if (r < 0) | |
1645 | return r; | |
1646 | ||
1647 | /* These must be done after everything succeeds. */ | |
1648 | s->child.pid = pid; | |
b6d5481b | 1649 | e->n_online_child_sources++; |
ac9f2640 | 1650 | |
a71fe8b8 LP |
1651 | if (ret) |
1652 | *ret = s; | |
ec766a51 | 1653 | TAKE_PTR(s); |
f8f3f926 LP |
1654 | return 0; |
1655 | } | |
1656 | ||
1657 | _public_ int sd_event_add_child_pidfd( | |
1658 | sd_event *e, | |
1659 | sd_event_source **ret, | |
1660 | int pidfd, | |
1661 | int options, | |
1662 | sd_event_child_handler_t callback, | |
1663 | void *userdata) { | |
1664 | ||
1665 | ||
1666 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
1667 | pid_t pid; | |
1668 | int r; | |
1669 | ||
1670 | assert_return(e, -EINVAL); | |
1671 | assert_return(e = event_resolve(e), -ENOPKG); | |
1672 | assert_return(pidfd >= 0, -EBADF); | |
1673 | assert_return(!(options & ~(WEXITED|WSTOPPED|WCONTINUED)), -EINVAL); | |
1674 | assert_return(options != 0, -EINVAL); | |
f8f3f926 LP |
1675 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1676 | assert_return(!event_pid_changed(e), -ECHILD); | |
1677 | ||
b9350e70 LP |
1678 | if (!callback) |
1679 | callback = child_exit_callback; | |
1680 | ||
b6d5481b | 1681 | if (e->n_online_child_sources == 0) { |
ee880b37 LP |
1682 | r = signal_is_blocked(SIGCHLD); |
1683 | if (r < 0) | |
1684 | return r; | |
1685 | if (r == 0) | |
1686 | return -EBUSY; | |
1687 | } | |
1688 | ||
f8f3f926 LP |
1689 | r = hashmap_ensure_allocated(&e->child_sources, NULL); |
1690 | if (r < 0) | |
1691 | return r; | |
1692 | ||
1693 | r = pidfd_get_pid(pidfd, &pid); | |
1694 | if (r < 0) | |
1695 | return r; | |
1696 | ||
1697 | if (hashmap_contains(e->child_sources, PID_TO_PTR(pid))) | |
1698 | return -EBUSY; | |
1699 | ||
1700 | s = source_new(e, !ret, SOURCE_CHILD); | |
1701 | if (!s) | |
1702 | return -ENOMEM; | |
1703 | ||
1704 | s->wakeup = WAKEUP_EVENT_SOURCE; | |
1705 | s->child.pidfd = pidfd; | |
1706 | s->child.pid = pid; | |
1707 | s->child.options = options; | |
1708 | s->child.callback = callback; | |
1709 | 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) */ | |
1710 | s->userdata = userdata; | |
1711 | s->enabled = SD_EVENT_ONESHOT; | |
1712 | ||
1713 | r = hashmap_put(e->child_sources, PID_TO_PTR(pid), s); | |
1714 | if (r < 0) | |
1715 | return r; | |
1716 | ||
f8f3f926 LP |
1717 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
1718 | /* We only want to watch for WEXITED */ | |
f8f3f926 | 1719 | r = source_child_pidfd_register(s, s->enabled); |
ac9f2640 | 1720 | if (r < 0) |
f8f3f926 | 1721 | return r; |
f8f3f926 LP |
1722 | } else { |
1723 | /* We shall wait for some other event than WEXITED */ | |
f8f3f926 | 1724 | r = event_make_signal_data(e, SIGCHLD, NULL); |
ac9f2640 | 1725 | if (r < 0) |
f8f3f926 | 1726 | return r; |
a71fe8b8 | 1727 | |
f8f3f926 LP |
1728 | e->need_process_child = true; |
1729 | } | |
1730 | ||
b6d5481b | 1731 | e->n_online_child_sources++; |
ac9f2640 | 1732 | |
f8f3f926 LP |
1733 | if (ret) |
1734 | *ret = s; | |
f8f3f926 | 1735 | TAKE_PTR(s); |
fd38203a LP |
1736 | return 0; |
1737 | } | |
1738 | ||
b9350e70 LP |
1739 | static int generic_exit_callback(sd_event_source *s, void *userdata) { |
1740 | assert(s); | |
1741 | ||
1742 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
1743 | } | |
1744 | ||
f7262a9f | 1745 | _public_ int sd_event_add_defer( |
305f78bf | 1746 | sd_event *e, |
151b9b96 | 1747 | sd_event_source **ret, |
718db961 | 1748 | sd_event_handler_t callback, |
151b9b96 | 1749 | void *userdata) { |
305f78bf | 1750 | |
ec766a51 | 1751 | _cleanup_(source_freep) sd_event_source *s = NULL; |
fd38203a LP |
1752 | int r; |
1753 | ||
305f78bf | 1754 | assert_return(e, -EINVAL); |
b937d761 | 1755 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 1756 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 1757 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 1758 | |
b9350e70 LP |
1759 | if (!callback) |
1760 | callback = generic_exit_callback; | |
1761 | ||
a71fe8b8 | 1762 | s = source_new(e, !ret, SOURCE_DEFER); |
fd38203a LP |
1763 | if (!s) |
1764 | return -ENOMEM; | |
1765 | ||
1766 | s->defer.callback = callback; | |
1767 | s->userdata = userdata; | |
baf76283 | 1768 | s->enabled = SD_EVENT_ONESHOT; |
fd38203a LP |
1769 | |
1770 | r = source_set_pending(s, true); | |
ec766a51 | 1771 | if (r < 0) |
fd38203a | 1772 | return r; |
fd38203a | 1773 | |
a71fe8b8 LP |
1774 | if (ret) |
1775 | *ret = s; | |
ec766a51 | 1776 | TAKE_PTR(s); |
a71fe8b8 | 1777 | |
fd38203a LP |
1778 | return 0; |
1779 | } | |
1780 | ||
6e9feda3 LP |
1781 | _public_ int sd_event_add_post( |
1782 | sd_event *e, | |
1783 | sd_event_source **ret, | |
1784 | sd_event_handler_t callback, | |
1785 | void *userdata) { | |
1786 | ||
ec766a51 | 1787 | _cleanup_(source_freep) sd_event_source *s = NULL; |
6e9feda3 LP |
1788 | int r; |
1789 | ||
1790 | assert_return(e, -EINVAL); | |
b937d761 | 1791 | assert_return(e = event_resolve(e), -ENOPKG); |
6e9feda3 LP |
1792 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
1793 | assert_return(!event_pid_changed(e), -ECHILD); | |
1794 | ||
b9350e70 LP |
1795 | if (!callback) |
1796 | callback = generic_exit_callback; | |
1797 | ||
a71fe8b8 | 1798 | s = source_new(e, !ret, SOURCE_POST); |
6e9feda3 LP |
1799 | if (!s) |
1800 | return -ENOMEM; | |
1801 | ||
1802 | s->post.callback = callback; | |
1803 | s->userdata = userdata; | |
1804 | s->enabled = SD_EVENT_ON; | |
1805 | ||
de7fef4b | 1806 | r = set_ensure_put(&e->post_sources, NULL, s); |
ec766a51 | 1807 | if (r < 0) |
6e9feda3 | 1808 | return r; |
de7fef4b | 1809 | assert(r > 0); |
6e9feda3 | 1810 | |
a71fe8b8 LP |
1811 | if (ret) |
1812 | *ret = s; | |
ec766a51 | 1813 | TAKE_PTR(s); |
a71fe8b8 | 1814 | |
6e9feda3 LP |
1815 | return 0; |
1816 | } | |
1817 | ||
6203e07a | 1818 | _public_ int sd_event_add_exit( |
305f78bf | 1819 | sd_event *e, |
151b9b96 | 1820 | sd_event_source **ret, |
718db961 | 1821 | sd_event_handler_t callback, |
151b9b96 | 1822 | void *userdata) { |
305f78bf | 1823 | |
ec766a51 | 1824 | _cleanup_(source_freep) sd_event_source *s = NULL; |
da7e457c LP |
1825 | int r; |
1826 | ||
1827 | assert_return(e, -EINVAL); | |
b937d761 | 1828 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
1829 | assert_return(callback, -EINVAL); |
1830 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
1831 | assert_return(!event_pid_changed(e), -ECHILD); | |
1832 | ||
c983e776 EV |
1833 | r = prioq_ensure_allocated(&e->exit, exit_prioq_compare); |
1834 | if (r < 0) | |
1835 | return r; | |
da7e457c | 1836 | |
a71fe8b8 | 1837 | s = source_new(e, !ret, SOURCE_EXIT); |
fd38203a | 1838 | if (!s) |
da7e457c | 1839 | return -ENOMEM; |
fd38203a | 1840 | |
6203e07a | 1841 | s->exit.callback = callback; |
da7e457c | 1842 | s->userdata = userdata; |
6203e07a | 1843 | s->exit.prioq_index = PRIOQ_IDX_NULL; |
baf76283 | 1844 | s->enabled = SD_EVENT_ONESHOT; |
da7e457c | 1845 | |
6203e07a | 1846 | r = prioq_put(s->event->exit, s, &s->exit.prioq_index); |
ec766a51 | 1847 | if (r < 0) |
da7e457c | 1848 | return r; |
da7e457c | 1849 | |
a71fe8b8 LP |
1850 | if (ret) |
1851 | *ret = s; | |
ec766a51 | 1852 | TAKE_PTR(s); |
a71fe8b8 | 1853 | |
da7e457c LP |
1854 | return 0; |
1855 | } | |
1856 | ||
158fe190 LP |
1857 | int sd_event_trim_memory(void) { |
1858 | int r; | |
1859 | ||
1860 | /* A default implementation of a memory pressure callback. Simply releases our own allocation caches | |
1861 | * and glibc's. This is automatically used when people call sd_event_add_memory_pressure() with a | |
1862 | * NULL callback parameter. */ | |
1863 | ||
1864 | log_debug("Memory pressure event, trimming malloc() memory."); | |
1865 | ||
1866 | #if HAVE_GENERIC_MALLINFO | |
1867 | generic_mallinfo before_mallinfo = generic_mallinfo_get(); | |
1868 | #endif | |
1869 | ||
1870 | usec_t before_timestamp = now(CLOCK_MONOTONIC); | |
1871 | hashmap_trim_pools(); | |
1872 | r = malloc_trim(0); | |
1873 | usec_t after_timestamp = now(CLOCK_MONOTONIC); | |
1874 | ||
1875 | if (r > 0) | |
1876 | log_debug("Successfully trimmed some memory."); | |
1877 | else | |
1878 | log_debug("Couldn't trim any memory."); | |
1879 | ||
1880 | usec_t period = after_timestamp - before_timestamp; | |
1881 | ||
1882 | #if HAVE_GENERIC_MALLINFO | |
1883 | generic_mallinfo after_mallinfo = generic_mallinfo_get(); | |
1884 | size_t l = LESS_BY((size_t) before_mallinfo.hblkhd, (size_t) after_mallinfo.hblkhd) + | |
1885 | LESS_BY((size_t) before_mallinfo.arena, (size_t) after_mallinfo.arena); | |
1886 | log_struct(LOG_DEBUG, | |
1887 | LOG_MESSAGE("Memory trimming took %s, returned %s to OS.", | |
1888 | FORMAT_TIMESPAN(period, 0), | |
1889 | FORMAT_BYTES(l)), | |
1890 | "MESSAGE_ID=" SD_MESSAGE_MEMORY_TRIM_STR, | |
1891 | "TRIMMED_BYTES=%zu", l, | |
1892 | "TRIMMED_USEC=" USEC_FMT, period); | |
1893 | #else | |
1894 | log_struct(LOG_DEBUG, | |
1895 | LOG_MESSAGE("Memory trimming took %s.", | |
1896 | FORMAT_TIMESPAN(period, 0)), | |
1897 | "MESSAGE_ID=" SD_MESSAGE_MEMORY_TRIM_STR, | |
1898 | "TRIMMED_USEC=" USEC_FMT, period); | |
1899 | #endif | |
1900 | ||
1901 | return 0; | |
1902 | } | |
1903 | ||
1904 | static int memory_pressure_callback(sd_event_source *s, void *userdata) { | |
1905 | assert(s); | |
1906 | ||
1907 | sd_event_trim_memory(); | |
1908 | return 0; | |
1909 | } | |
1910 | ||
1911 | _public_ int sd_event_add_memory_pressure( | |
1912 | sd_event *e, | |
1913 | sd_event_source **ret, | |
1914 | sd_event_handler_t callback, | |
1915 | void *userdata) { | |
1916 | ||
1917 | _cleanup_free_ char *w = NULL; | |
1918 | _cleanup_(source_freep) sd_event_source *s = NULL; | |
92651a7a | 1919 | _cleanup_close_ int path_fd = -EBADF, fd = -EBADF; |
158fe190 | 1920 | _cleanup_free_ void *write_buffer = NULL; |
40c5d5d2 | 1921 | const char *watch, *watch_fallback = NULL, *env; |
158fe190 LP |
1922 | size_t write_buffer_size = 0; |
1923 | struct stat st; | |
1924 | uint32_t events; | |
1925 | bool locked; | |
1926 | int r; | |
1927 | ||
1928 | assert_return(e, -EINVAL); | |
1929 | assert_return(e = event_resolve(e), -ENOPKG); | |
1930 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
1931 | assert_return(!event_pid_changed(e), -ECHILD); | |
1932 | ||
1933 | if (!callback) | |
1934 | callback = memory_pressure_callback; | |
1935 | ||
1936 | s = source_new(e, !ret, SOURCE_MEMORY_PRESSURE); | |
1937 | if (!s) | |
1938 | return -ENOMEM; | |
1939 | ||
1940 | s->wakeup = WAKEUP_EVENT_SOURCE; | |
1941 | s->memory_pressure.callback = callback; | |
1942 | s->userdata = userdata; | |
1943 | s->enabled = SD_EVENT_ON; | |
1944 | s->memory_pressure.fd = -EBADF; | |
1945 | ||
1946 | env = secure_getenv("MEMORY_PRESSURE_WATCH"); | |
1947 | if (env) { | |
1948 | if (isempty(env) || path_equal(env, "/dev/null")) | |
1949 | return log_debug_errno(SYNTHETIC_ERRNO(EHOSTDOWN), | |
1950 | "Memory pressure logic is explicitly disabled via $MEMORY_PRESSURE_WATCH."); | |
1951 | ||
1952 | if (!path_is_absolute(env) || !path_is_normalized(env)) | |
1953 | return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG), | |
1954 | "$MEMORY_PRESSURE_WATCH set to invalid path: %s", env); | |
1955 | ||
1956 | watch = env; | |
1957 | ||
1958 | env = secure_getenv("MEMORY_PRESSURE_WRITE"); | |
1959 | if (env) { | |
1960 | r = unbase64mem(env, SIZE_MAX, &write_buffer, &write_buffer_size); | |
1961 | if (r < 0) | |
1962 | return r; | |
1963 | } | |
1964 | ||
1965 | locked = true; | |
1966 | } else { | |
1967 | ||
1968 | r = is_pressure_supported(); | |
1969 | if (r < 0) | |
1970 | return r; | |
1971 | if (r == 0) | |
1972 | return -EOPNOTSUPP; | |
1973 | ||
1974 | /* By default we want to watch memory pressure on the local cgroup, but we'll fall back on | |
1975 | * the system wide pressure if for some reason we cannot (which could be: memory controller | |
1976 | * not delegated to us, or PSI simply not available in the kernel). On legacy cgroupv1 we'll | |
1977 | * only use the system-wide logic. */ | |
1978 | r = cg_all_unified(); | |
1979 | if (r < 0) | |
1980 | return r; | |
1981 | if (r == 0) | |
1982 | watch = "/proc/pressure/memory"; | |
1983 | else { | |
1984 | _cleanup_free_ char *cg = NULL; | |
1985 | ||
1986 | r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &cg); | |
1987 | if (r < 0) | |
1988 | return r; | |
1989 | ||
1990 | w = path_join("/sys/fs/cgroup", cg, "memory.pressure"); | |
1991 | if (!w) | |
1992 | return -ENOMEM; | |
1993 | ||
1994 | watch = w; | |
1995 | watch_fallback = "/proc/pressure/memory"; | |
1996 | } | |
1997 | ||
1998 | /* Android uses three levels in its userspace low memory killer logic: | |
1999 | * some 70000 1000000 | |
2000 | * some 100000 1000000 | |
2001 | * full 70000 1000000 | |
2002 | * | |
2003 | * GNOME's low memory monitor uses: | |
2004 | * some 70000 1000000 | |
2005 | * some 100000 1000000 | |
2006 | * full 100000 1000000 | |
2007 | * | |
2008 | * We'll default to the middle level that both agree on */ | |
2009 | if (asprintf((char**) &write_buffer, | |
2010 | "%s " USEC_FMT " " USEC_FMT, | |
2011 | MEMORY_PRESSURE_DEFAULT_TYPE, | |
2012 | MEMORY_PRESSURE_DEFAULT_THRESHOLD_USEC, | |
2013 | MEMORY_PRESSURE_DEFAULT_WINDOW_USEC) < 0) | |
2014 | return -ENOMEM; | |
2015 | ||
2016 | write_buffer_size = strlen(write_buffer) + 1; | |
2017 | locked = false; | |
2018 | } | |
2019 | ||
2020 | path_fd = open(watch, O_PATH|O_CLOEXEC); | |
2021 | if (path_fd < 0) { | |
2022 | if (errno != ENOENT) | |
2023 | return -errno; | |
2024 | ||
2025 | /* We got ENOENT. Three options now: try the fallback if we have one, or return the error as | |
2026 | * is (if based on user/env config), or return -EOPNOTSUPP (because we picked the path, and | |
2027 | * the PSI service apparently is not supported) */ | |
2028 | if (!watch_fallback) | |
2029 | return locked ? -ENOENT : -EOPNOTSUPP; | |
2030 | ||
2031 | path_fd = open(watch_fallback, O_PATH|O_CLOEXEC); | |
63b1e67e YW |
2032 | if (path_fd < 0) { |
2033 | if (errno == ENOENT) /* PSI is not available in the kernel even under the fallback path? */ | |
2034 | return -EOPNOTSUPP; | |
158fe190 | 2035 | return -errno; |
63b1e67e | 2036 | } |
158fe190 LP |
2037 | } |
2038 | ||
2039 | if (fstat(path_fd, &st) < 0) | |
2040 | return -errno; | |
2041 | ||
2042 | if (S_ISSOCK(st.st_mode)) { | |
2043 | fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0); | |
2044 | if (fd < 0) | |
2045 | return -errno; | |
2046 | ||
2047 | r = connect_unix_path(fd, path_fd, NULL); | |
2048 | if (r < 0) | |
2049 | return r; | |
2050 | ||
2051 | events = EPOLLIN; | |
2052 | ||
2053 | } else if (S_ISREG(st.st_mode) || S_ISFIFO(st.st_mode) || S_ISCHR(st.st_mode)) { | |
2054 | fd = fd_reopen(path_fd, (write_buffer_size > 0 ? O_RDWR : O_RDONLY) |O_CLOEXEC|O_NONBLOCK|O_NOCTTY); | |
2055 | if (fd < 0) | |
2056 | return fd; | |
2057 | ||
2058 | if (S_ISREG(st.st_mode)) { | |
2059 | struct statfs sfs; | |
2060 | ||
2061 | /* If this is a regular file validate this is a procfs or cgroupfs file, where we look for EPOLLPRI */ | |
2062 | ||
2063 | if (fstatfs(fd, &sfs) < 0) | |
2064 | return -errno; | |
2065 | ||
2066 | if (!is_fs_type(&sfs, PROC_SUPER_MAGIC) && | |
2067 | !is_fs_type(&sfs, CGROUP2_SUPER_MAGIC)) | |
2068 | return -ENOTTY; | |
2069 | ||
2070 | events = EPOLLPRI; | |
2071 | } else | |
2072 | /* For fifos and char devices just watch for EPOLLIN */ | |
2073 | events = EPOLLIN; | |
2074 | ||
2075 | } else if (S_ISDIR(st.st_mode)) | |
2076 | return -EISDIR; | |
2077 | else | |
2078 | return -EBADF; | |
2079 | ||
2080 | s->memory_pressure.fd = TAKE_FD(fd); | |
2081 | s->memory_pressure.write_buffer = TAKE_PTR(write_buffer); | |
2082 | s->memory_pressure.write_buffer_size = write_buffer_size; | |
2083 | s->memory_pressure.events = events; | |
2084 | s->memory_pressure.locked = locked; | |
2085 | ||
2086 | /* So here's the thing: if we are talking to PSI we need to write the watch string before adding the | |
2087 | * fd to epoll (if we ignore this, then the watch won't work). Hence we'll not actually register the | |
2088 | * fd with the epoll right-away. Instead, we just add the event source to a list of memory pressure | |
2089 | * event sources on which writes must be executed before the first event loop iteration is | |
2090 | * executed. (We could also write the data here, right away, but we want to give the caller the | |
2091 | * freedom to call sd_event_source_set_memory_pressure_type() and | |
2092 | * sd_event_source_set_memory_pressure_rate() before we write it. */ | |
2093 | ||
2094 | if (s->memory_pressure.write_buffer_size > 0) | |
2095 | source_memory_pressure_add_to_write_list(s); | |
2096 | else { | |
2097 | r = source_memory_pressure_register(s, s->enabled); | |
2098 | if (r < 0) | |
2099 | return r; | |
2100 | } | |
2101 | ||
2102 | if (ret) | |
2103 | *ret = s; | |
2104 | TAKE_PTR(s); | |
2105 | ||
2106 | return 0; | |
2107 | } | |
2108 | ||
97ef5391 LP |
2109 | static void event_free_inotify_data(sd_event *e, struct inotify_data *d) { |
2110 | assert(e); | |
2111 | ||
2112 | if (!d) | |
2113 | return; | |
2114 | ||
2115 | assert(hashmap_isempty(d->inodes)); | |
2116 | assert(hashmap_isempty(d->wd)); | |
2117 | ||
2118 | if (d->buffer_filled > 0) | |
0601b958 | 2119 | LIST_REMOVE(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
2120 | |
2121 | hashmap_free(d->inodes); | |
2122 | hashmap_free(d->wd); | |
2123 | ||
2124 | assert_se(hashmap_remove(e->inotify_data, &d->priority) == d); | |
2125 | ||
2126 | if (d->fd >= 0) { | |
fbae5090 YW |
2127 | if (!event_pid_changed(e) && |
2128 | epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, d->fd, NULL) < 0) | |
97ef5391 LP |
2129 | log_debug_errno(errno, "Failed to remove inotify fd from epoll, ignoring: %m"); |
2130 | ||
2131 | safe_close(d->fd); | |
2132 | } | |
2133 | free(d); | |
2134 | } | |
2135 | ||
2136 | static int event_make_inotify_data( | |
2137 | sd_event *e, | |
2138 | int64_t priority, | |
2139 | struct inotify_data **ret) { | |
2140 | ||
254d1313 | 2141 | _cleanup_close_ int fd = -EBADF; |
97ef5391 | 2142 | struct inotify_data *d; |
97ef5391 LP |
2143 | int r; |
2144 | ||
2145 | assert(e); | |
2146 | ||
2147 | d = hashmap_get(e->inotify_data, &priority); | |
2148 | if (d) { | |
2149 | if (ret) | |
2150 | *ret = d; | |
2151 | return 0; | |
2152 | } | |
2153 | ||
2154 | fd = inotify_init1(IN_NONBLOCK|O_CLOEXEC); | |
2155 | if (fd < 0) | |
2156 | return -errno; | |
2157 | ||
2158 | fd = fd_move_above_stdio(fd); | |
2159 | ||
97ef5391 LP |
2160 | d = new(struct inotify_data, 1); |
2161 | if (!d) | |
2162 | return -ENOMEM; | |
2163 | ||
2164 | *d = (struct inotify_data) { | |
2165 | .wakeup = WAKEUP_INOTIFY_DATA, | |
2166 | .fd = TAKE_FD(fd), | |
2167 | .priority = priority, | |
2168 | }; | |
2169 | ||
c2484a75 | 2170 | r = hashmap_ensure_put(&e->inotify_data, &uint64_hash_ops, &d->priority, d); |
97ef5391 LP |
2171 | if (r < 0) { |
2172 | d->fd = safe_close(d->fd); | |
2173 | free(d); | |
2174 | return r; | |
2175 | } | |
2176 | ||
1eac7948 | 2177 | struct epoll_event ev = { |
97ef5391 LP |
2178 | .events = EPOLLIN, |
2179 | .data.ptr = d, | |
2180 | }; | |
2181 | ||
2182 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, d->fd, &ev) < 0) { | |
2183 | r = -errno; | |
2184 | d->fd = safe_close(d->fd); /* let's close this ourselves, as event_free_inotify_data() would otherwise | |
2185 | * remove the fd from the epoll first, which we don't want as we couldn't | |
2186 | * add it in the first place. */ | |
2187 | event_free_inotify_data(e, d); | |
2188 | return r; | |
2189 | } | |
2190 | ||
2191 | if (ret) | |
2192 | *ret = d; | |
2193 | ||
2194 | return 1; | |
2195 | } | |
2196 | ||
7a08d314 | 2197 | static int inode_data_compare(const struct inode_data *x, const struct inode_data *y) { |
90c88092 | 2198 | int r; |
97ef5391 LP |
2199 | |
2200 | assert(x); | |
2201 | assert(y); | |
2202 | ||
90c88092 YW |
2203 | r = CMP(x->dev, y->dev); |
2204 | if (r != 0) | |
2205 | return r; | |
97ef5391 | 2206 | |
6dd91b36 | 2207 | return CMP(x->ino, y->ino); |
97ef5391 LP |
2208 | } |
2209 | ||
7a08d314 YW |
2210 | static void inode_data_hash_func(const struct inode_data *d, struct siphash *state) { |
2211 | assert(d); | |
97ef5391 LP |
2212 | |
2213 | siphash24_compress(&d->dev, sizeof(d->dev), state); | |
2214 | siphash24_compress(&d->ino, sizeof(d->ino), state); | |
2215 | } | |
2216 | ||
7a08d314 | 2217 | DEFINE_PRIVATE_HASH_OPS(inode_data_hash_ops, struct inode_data, inode_data_hash_func, inode_data_compare); |
97ef5391 LP |
2218 | |
2219 | static void event_free_inode_data( | |
2220 | sd_event *e, | |
2221 | struct inode_data *d) { | |
2222 | ||
2223 | assert(e); | |
2224 | ||
2225 | if (!d) | |
2226 | return; | |
2227 | ||
64903d18 | 2228 | assert(!d->event_sources); |
97ef5391 LP |
2229 | |
2230 | if (d->fd >= 0) { | |
ed828563 | 2231 | LIST_REMOVE(to_close, e->inode_data_to_close_list, d); |
97ef5391 LP |
2232 | safe_close(d->fd); |
2233 | } | |
2234 | ||
2235 | if (d->inotify_data) { | |
2236 | ||
2237 | if (d->wd >= 0) { | |
fbae5090 | 2238 | if (d->inotify_data->fd >= 0 && !event_pid_changed(e)) { |
97ef5391 LP |
2239 | /* So here's a problem. At the time this runs the watch descriptor might already be |
2240 | * invalidated, because an IN_IGNORED event might be queued right the moment we enter | |
2241 | * the syscall. Hence, whenever we get EINVAL, ignore it entirely, since it's a very | |
2242 | * likely case to happen. */ | |
2243 | ||
2244 | if (inotify_rm_watch(d->inotify_data->fd, d->wd) < 0 && errno != EINVAL) | |
2245 | log_debug_errno(errno, "Failed to remove watch descriptor %i from inotify, ignoring: %m", d->wd); | |
2246 | } | |
2247 | ||
2248 | assert_se(hashmap_remove(d->inotify_data->wd, INT_TO_PTR(d->wd)) == d); | |
2249 | } | |
2250 | ||
2251 | assert_se(hashmap_remove(d->inotify_data->inodes, d) == d); | |
2252 | } | |
2253 | ||
2254 | free(d); | |
2255 | } | |
2256 | ||
53baf2ef LP |
2257 | static void event_gc_inotify_data( |
2258 | sd_event *e, | |
2259 | struct inotify_data *d) { | |
2260 | ||
2261 | assert(e); | |
2262 | ||
2263 | /* GCs the inotify data object if we don't need it anymore. That's the case if we don't want to watch | |
2264 | * any inode with it anymore, which in turn happens if no event source of this priority is interested | |
2265 | * in any inode any longer. That said, we maintain an extra busy counter: if non-zero we'll delay GC | |
2266 | * (under the expectation that the GC is called again once the counter is decremented). */ | |
2267 | ||
2268 | if (!d) | |
2269 | return; | |
2270 | ||
2271 | if (!hashmap_isempty(d->inodes)) | |
2272 | return; | |
2273 | ||
2274 | if (d->n_busy > 0) | |
2275 | return; | |
2276 | ||
2277 | event_free_inotify_data(e, d); | |
2278 | } | |
2279 | ||
97ef5391 LP |
2280 | static void event_gc_inode_data( |
2281 | sd_event *e, | |
2282 | struct inode_data *d) { | |
2283 | ||
2284 | struct inotify_data *inotify_data; | |
2285 | ||
2286 | assert(e); | |
2287 | ||
2288 | if (!d) | |
2289 | return; | |
2290 | ||
64903d18 | 2291 | if (d->event_sources) |
97ef5391 LP |
2292 | return; |
2293 | ||
2294 | inotify_data = d->inotify_data; | |
2295 | event_free_inode_data(e, d); | |
2296 | ||
53baf2ef | 2297 | event_gc_inotify_data(e, inotify_data); |
97ef5391 LP |
2298 | } |
2299 | ||
2300 | static int event_make_inode_data( | |
2301 | sd_event *e, | |
2302 | struct inotify_data *inotify_data, | |
2303 | dev_t dev, | |
2304 | ino_t ino, | |
2305 | struct inode_data **ret) { | |
2306 | ||
2307 | struct inode_data *d, key; | |
2308 | int r; | |
2309 | ||
2310 | assert(e); | |
2311 | assert(inotify_data); | |
2312 | ||
2313 | key = (struct inode_data) { | |
2314 | .ino = ino, | |
2315 | .dev = dev, | |
2316 | }; | |
2317 | ||
2318 | d = hashmap_get(inotify_data->inodes, &key); | |
2319 | if (d) { | |
2320 | if (ret) | |
2321 | *ret = d; | |
2322 | ||
2323 | return 0; | |
2324 | } | |
2325 | ||
2326 | r = hashmap_ensure_allocated(&inotify_data->inodes, &inode_data_hash_ops); | |
2327 | if (r < 0) | |
2328 | return r; | |
2329 | ||
2330 | d = new(struct inode_data, 1); | |
2331 | if (!d) | |
2332 | return -ENOMEM; | |
2333 | ||
2334 | *d = (struct inode_data) { | |
2335 | .dev = dev, | |
2336 | .ino = ino, | |
2337 | .wd = -1, | |
254d1313 | 2338 | .fd = -EBADF, |
97ef5391 LP |
2339 | .inotify_data = inotify_data, |
2340 | }; | |
2341 | ||
2342 | r = hashmap_put(inotify_data->inodes, d, d); | |
2343 | if (r < 0) { | |
2344 | free(d); | |
2345 | return r; | |
2346 | } | |
2347 | ||
2348 | if (ret) | |
2349 | *ret = d; | |
2350 | ||
2351 | return 1; | |
2352 | } | |
2353 | ||
2354 | static uint32_t inode_data_determine_mask(struct inode_data *d) { | |
2355 | bool excl_unlink = true; | |
2356 | uint32_t combined = 0; | |
97ef5391 LP |
2357 | |
2358 | assert(d); | |
2359 | ||
2360 | /* Combines the watch masks of all event sources watching this inode. We generally just OR them together, but | |
2361 | * the IN_EXCL_UNLINK flag is ANDed instead. | |
2362 | * | |
2363 | * Note that we add all sources to the mask here, regardless whether enabled, disabled or oneshot. That's | |
2364 | * because we cannot change the mask anymore after the event source was created once, since the kernel has no | |
f21f31b2 | 2365 | * API for that. Hence we need to subscribe to the maximum mask we ever might be interested in, and suppress |
97ef5391 LP |
2366 | * events we don't care for client-side. */ |
2367 | ||
2368 | LIST_FOREACH(inotify.by_inode_data, s, d->event_sources) { | |
2369 | ||
2370 | if ((s->inotify.mask & IN_EXCL_UNLINK) == 0) | |
2371 | excl_unlink = false; | |
2372 | ||
2373 | combined |= s->inotify.mask; | |
2374 | } | |
2375 | ||
2376 | return (combined & ~(IN_ONESHOT|IN_DONT_FOLLOW|IN_ONLYDIR|IN_EXCL_UNLINK)) | (excl_unlink ? IN_EXCL_UNLINK : 0); | |
2377 | } | |
2378 | ||
2379 | static int inode_data_realize_watch(sd_event *e, struct inode_data *d) { | |
2380 | uint32_t combined_mask; | |
2381 | int wd, r; | |
2382 | ||
2383 | assert(d); | |
2384 | assert(d->fd >= 0); | |
2385 | ||
2386 | combined_mask = inode_data_determine_mask(d); | |
2387 | ||
2388 | if (d->wd >= 0 && combined_mask == d->combined_mask) | |
2389 | return 0; | |
2390 | ||
2391 | r = hashmap_ensure_allocated(&d->inotify_data->wd, NULL); | |
2392 | if (r < 0) | |
2393 | return r; | |
2394 | ||
2395 | wd = inotify_add_watch_fd(d->inotify_data->fd, d->fd, combined_mask); | |
2396 | if (wd < 0) | |
2397 | return -errno; | |
2398 | ||
2399 | if (d->wd < 0) { | |
2400 | r = hashmap_put(d->inotify_data->wd, INT_TO_PTR(wd), d); | |
2401 | if (r < 0) { | |
2402 | (void) inotify_rm_watch(d->inotify_data->fd, wd); | |
2403 | return r; | |
2404 | } | |
2405 | ||
2406 | d->wd = wd; | |
2407 | ||
2408 | } else if (d->wd != wd) { | |
2409 | ||
2410 | log_debug("Weird, the watch descriptor we already knew for this inode changed?"); | |
2411 | (void) inotify_rm_watch(d->fd, wd); | |
2412 | return -EINVAL; | |
2413 | } | |
2414 | ||
2415 | d->combined_mask = combined_mask; | |
2416 | return 1; | |
2417 | } | |
2418 | ||
b9350e70 LP |
2419 | static int inotify_exit_callback(sd_event_source *s, const struct inotify_event *event, void *userdata) { |
2420 | assert(s); | |
2421 | ||
2422 | return sd_event_exit(sd_event_source_get_event(s), PTR_TO_INT(userdata)); | |
2423 | } | |
2424 | ||
e67d738a | 2425 | static int event_add_inotify_fd_internal( |
97ef5391 LP |
2426 | sd_event *e, |
2427 | sd_event_source **ret, | |
e67d738a LP |
2428 | int fd, |
2429 | bool donate, | |
97ef5391 LP |
2430 | uint32_t mask, |
2431 | sd_event_inotify_handler_t callback, | |
2432 | void *userdata) { | |
2433 | ||
5bb1d7fb | 2434 | _cleanup_close_ int donated_fd = donate ? fd : -EBADF; |
e67d738a | 2435 | _cleanup_(source_freep) sd_event_source *s = NULL; |
97ef5391 LP |
2436 | struct inotify_data *inotify_data = NULL; |
2437 | struct inode_data *inode_data = NULL; | |
97ef5391 LP |
2438 | struct stat st; |
2439 | int r; | |
2440 | ||
2441 | assert_return(e, -EINVAL); | |
2442 | assert_return(e = event_resolve(e), -ENOPKG); | |
e67d738a | 2443 | assert_return(fd >= 0, -EBADF); |
97ef5391 LP |
2444 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
2445 | assert_return(!event_pid_changed(e), -ECHILD); | |
2446 | ||
b9350e70 LP |
2447 | if (!callback) |
2448 | callback = inotify_exit_callback; | |
2449 | ||
97ef5391 LP |
2450 | /* Refuse IN_MASK_ADD since we coalesce watches on the same inode, and hence really don't want to merge |
2451 | * masks. Or in other words, this whole code exists only to manage IN_MASK_ADD type operations for you, hence | |
2452 | * the user can't use them for us. */ | |
2453 | if (mask & IN_MASK_ADD) | |
2454 | return -EINVAL; | |
2455 | ||
97ef5391 LP |
2456 | if (fstat(fd, &st) < 0) |
2457 | return -errno; | |
2458 | ||
2459 | s = source_new(e, !ret, SOURCE_INOTIFY); | |
2460 | if (!s) | |
2461 | return -ENOMEM; | |
2462 | ||
2463 | s->enabled = mask & IN_ONESHOT ? SD_EVENT_ONESHOT : SD_EVENT_ON; | |
2464 | s->inotify.mask = mask; | |
2465 | s->inotify.callback = callback; | |
2466 | s->userdata = userdata; | |
2467 | ||
2468 | /* Allocate an inotify object for this priority, and an inode object within it */ | |
2469 | r = event_make_inotify_data(e, SD_EVENT_PRIORITY_NORMAL, &inotify_data); | |
2470 | if (r < 0) | |
8c75fe17 | 2471 | return r; |
97ef5391 LP |
2472 | |
2473 | r = event_make_inode_data(e, inotify_data, st.st_dev, st.st_ino, &inode_data); | |
8c75fe17 | 2474 | if (r < 0) { |
e67d738a | 2475 | event_gc_inotify_data(e, inotify_data); |
8c75fe17 ZJS |
2476 | return r; |
2477 | } | |
97ef5391 LP |
2478 | |
2479 | /* Keep the O_PATH fd around until the first iteration of the loop, so that we can still change the priority of | |
2480 | * the event source, until then, for which we need the original inode. */ | |
2481 | if (inode_data->fd < 0) { | |
e67d738a LP |
2482 | if (donated_fd >= 0) |
2483 | inode_data->fd = TAKE_FD(donated_fd); | |
2484 | else { | |
2485 | inode_data->fd = fcntl(fd, F_DUPFD_CLOEXEC, 3); | |
2486 | if (inode_data->fd < 0) { | |
2487 | r = -errno; | |
2488 | event_gc_inode_data(e, inode_data); | |
2489 | return r; | |
2490 | } | |
2491 | } | |
2492 | ||
ed828563 | 2493 | LIST_PREPEND(to_close, e->inode_data_to_close_list, inode_data); |
97ef5391 LP |
2494 | } |
2495 | ||
2496 | /* Link our event source to the inode data object */ | |
2497 | LIST_PREPEND(inotify.by_inode_data, inode_data->event_sources, s); | |
2498 | s->inotify.inode_data = inode_data; | |
2499 | ||
97ef5391 LP |
2500 | /* Actually realize the watch now */ |
2501 | r = inode_data_realize_watch(e, inode_data); | |
2502 | if (r < 0) | |
8c75fe17 | 2503 | return r; |
97ef5391 | 2504 | |
97ef5391 LP |
2505 | if (ret) |
2506 | *ret = s; | |
8c75fe17 | 2507 | TAKE_PTR(s); |
97ef5391 LP |
2508 | |
2509 | return 0; | |
97ef5391 LP |
2510 | } |
2511 | ||
e67d738a LP |
2512 | _public_ int sd_event_add_inotify_fd( |
2513 | sd_event *e, | |
2514 | sd_event_source **ret, | |
2515 | int fd, | |
2516 | uint32_t mask, | |
2517 | sd_event_inotify_handler_t callback, | |
2518 | void *userdata) { | |
2519 | ||
2520 | return event_add_inotify_fd_internal(e, ret, fd, /* donate= */ false, mask, callback, userdata); | |
2521 | } | |
2522 | ||
2523 | _public_ int sd_event_add_inotify( | |
2524 | sd_event *e, | |
2525 | sd_event_source **ret, | |
2526 | const char *path, | |
2527 | uint32_t mask, | |
2528 | sd_event_inotify_handler_t callback, | |
2529 | void *userdata) { | |
2530 | ||
2091c779 | 2531 | sd_event_source *s = NULL; /* avoid false maybe-uninitialized warning */ |
e67d738a LP |
2532 | int fd, r; |
2533 | ||
2534 | assert_return(path, -EINVAL); | |
2535 | ||
586c8cee ZJS |
2536 | fd = open(path, O_PATH | O_CLOEXEC | |
2537 | (mask & IN_ONLYDIR ? O_DIRECTORY : 0) | | |
2538 | (mask & IN_DONT_FOLLOW ? O_NOFOLLOW : 0)); | |
e67d738a LP |
2539 | if (fd < 0) |
2540 | return -errno; | |
2541 | ||
2542 | r = event_add_inotify_fd_internal(e, &s, fd, /* donate= */ true, mask, callback, userdata); | |
2543 | if (r < 0) | |
2544 | return r; | |
2545 | ||
2546 | (void) sd_event_source_set_description(s, path); | |
2547 | ||
2548 | if (ret) | |
2549 | *ret = s; | |
2550 | ||
2551 | return r; | |
2552 | } | |
2553 | ||
8301aa0b | 2554 | static sd_event_source* event_source_free(sd_event_source *s) { |
6680dd6b LP |
2555 | if (!s) |
2556 | return NULL; | |
da7e457c | 2557 | |
8301aa0b YW |
2558 | /* Here's a special hack: when we are called from a |
2559 | * dispatch handler we won't free the event source | |
2560 | * immediately, but we will detach the fd from the | |
2561 | * epoll. This way it is safe for the caller to unref | |
2562 | * the event source and immediately close the fd, but | |
2563 | * we still retain a valid event source object after | |
2564 | * the callback. */ | |
fd38203a | 2565 | |
76d04c3a | 2566 | if (s->dispatching) |
8301aa0b | 2567 | source_disconnect(s); |
76d04c3a | 2568 | else |
8301aa0b | 2569 | source_free(s); |
fd38203a LP |
2570 | |
2571 | return NULL; | |
2572 | } | |
2573 | ||
8301aa0b YW |
2574 | DEFINE_PUBLIC_TRIVIAL_REF_UNREF_FUNC(sd_event_source, sd_event_source, event_source_free); |
2575 | ||
356779df | 2576 | _public_ int sd_event_source_set_description(sd_event_source *s, const char *description) { |
f7f53e9e | 2577 | assert_return(s, -EINVAL); |
f4b2933e | 2578 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 2579 | |
356779df | 2580 | return free_and_strdup(&s->description, description); |
f7f53e9e TG |
2581 | } |
2582 | ||
356779df | 2583 | _public_ int sd_event_source_get_description(sd_event_source *s, const char **description) { |
f7f53e9e | 2584 | assert_return(s, -EINVAL); |
356779df | 2585 | assert_return(description, -EINVAL); |
f4b2933e | 2586 | assert_return(!event_pid_changed(s->event), -ECHILD); |
f7f53e9e | 2587 | |
7d92a1a4 ZJS |
2588 | if (!s->description) |
2589 | return -ENXIO; | |
2590 | ||
356779df | 2591 | *description = s->description; |
f7f53e9e TG |
2592 | return 0; |
2593 | } | |
2594 | ||
adcc4ca3 | 2595 | _public_ sd_event *sd_event_source_get_event(sd_event_source *s) { |
305f78bf | 2596 | assert_return(s, NULL); |
eaa3cbef LP |
2597 | |
2598 | return s->event; | |
2599 | } | |
2600 | ||
f7262a9f | 2601 | _public_ int sd_event_source_get_pending(sd_event_source *s) { |
305f78bf | 2602 | assert_return(s, -EINVAL); |
6203e07a | 2603 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c | 2604 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2605 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2606 | |
2607 | return s->pending; | |
2608 | } | |
2609 | ||
f7262a9f | 2610 | _public_ int sd_event_source_get_io_fd(sd_event_source *s) { |
305f78bf LP |
2611 | assert_return(s, -EINVAL); |
2612 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2613 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2614 | |
2615 | return s->io.fd; | |
2616 | } | |
2617 | ||
30caf8f3 LP |
2618 | _public_ int sd_event_source_set_io_fd(sd_event_source *s, int fd) { |
2619 | int r; | |
2620 | ||
2621 | assert_return(s, -EINVAL); | |
8ac43fee | 2622 | assert_return(fd >= 0, -EBADF); |
30caf8f3 LP |
2623 | assert_return(s->type == SOURCE_IO, -EDOM); |
2624 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
2625 | ||
2626 | if (s->io.fd == fd) | |
2627 | return 0; | |
2628 | ||
b6d5481b | 2629 | if (event_source_is_offline(s)) { |
30caf8f3 LP |
2630 | s->io.fd = fd; |
2631 | s->io.registered = false; | |
2632 | } else { | |
2633 | int saved_fd; | |
2634 | ||
2635 | saved_fd = s->io.fd; | |
2636 | assert(s->io.registered); | |
2637 | ||
2638 | s->io.fd = fd; | |
2639 | s->io.registered = false; | |
2640 | ||
2641 | r = source_io_register(s, s->enabled, s->io.events); | |
2642 | if (r < 0) { | |
2643 | s->io.fd = saved_fd; | |
2644 | s->io.registered = true; | |
2645 | return r; | |
2646 | } | |
2647 | ||
5a795bff | 2648 | (void) epoll_ctl(s->event->epoll_fd, EPOLL_CTL_DEL, saved_fd, NULL); |
30caf8f3 LP |
2649 | } |
2650 | ||
2651 | return 0; | |
2652 | } | |
2653 | ||
ab93297c NM |
2654 | _public_ int sd_event_source_get_io_fd_own(sd_event_source *s) { |
2655 | assert_return(s, -EINVAL); | |
2656 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2657 | ||
2658 | return s->io.owned; | |
2659 | } | |
2660 | ||
2661 | _public_ int sd_event_source_set_io_fd_own(sd_event_source *s, int own) { | |
2662 | assert_return(s, -EINVAL); | |
2663 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2664 | ||
2665 | s->io.owned = own; | |
2666 | return 0; | |
2667 | } | |
2668 | ||
f7262a9f | 2669 | _public_ int sd_event_source_get_io_events(sd_event_source *s, uint32_t* events) { |
305f78bf LP |
2670 | assert_return(s, -EINVAL); |
2671 | assert_return(events, -EINVAL); | |
2672 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2673 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2674 | |
2675 | *events = s->io.events; | |
2676 | return 0; | |
2677 | } | |
2678 | ||
f7262a9f | 2679 | _public_ int sd_event_source_set_io_events(sd_event_source *s, uint32_t events) { |
fd38203a LP |
2680 | int r; |
2681 | ||
305f78bf LP |
2682 | assert_return(s, -EINVAL); |
2683 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2a16a986 | 2684 | assert_return(!(events & ~(EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLPRI|EPOLLERR|EPOLLHUP|EPOLLET)), -EINVAL); |
da7e457c | 2685 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2686 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2687 | |
b63c8d4f DH |
2688 | /* edge-triggered updates are never skipped, so we can reset edges */ |
2689 | if (s->io.events == events && !(events & EPOLLET)) | |
fd38203a LP |
2690 | return 0; |
2691 | ||
2a0dc6cd LP |
2692 | r = source_set_pending(s, false); |
2693 | if (r < 0) | |
2694 | return r; | |
2695 | ||
b6d5481b | 2696 | if (event_source_is_online(s)) { |
e4715127 | 2697 | r = source_io_register(s, s->enabled, events); |
fd38203a LP |
2698 | if (r < 0) |
2699 | return r; | |
2700 | } | |
2701 | ||
2702 | s->io.events = events; | |
2703 | ||
2704 | return 0; | |
2705 | } | |
2706 | ||
f7262a9f | 2707 | _public_ int sd_event_source_get_io_revents(sd_event_source *s, uint32_t* revents) { |
305f78bf LP |
2708 | assert_return(s, -EINVAL); |
2709 | assert_return(revents, -EINVAL); | |
2710 | assert_return(s->type == SOURCE_IO, -EDOM); | |
2711 | assert_return(s->pending, -ENODATA); | |
2712 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2713 | |
2714 | *revents = s->io.revents; | |
2715 | return 0; | |
2716 | } | |
2717 | ||
f7262a9f | 2718 | _public_ int sd_event_source_get_signal(sd_event_source *s) { |
305f78bf LP |
2719 | assert_return(s, -EINVAL); |
2720 | assert_return(s->type == SOURCE_SIGNAL, -EDOM); | |
2721 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
2722 | |
2723 | return s->signal.sig; | |
2724 | } | |
2725 | ||
31927c16 | 2726 | _public_ int sd_event_source_get_priority(sd_event_source *s, int64_t *priority) { |
305f78bf LP |
2727 | assert_return(s, -EINVAL); |
2728 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a | 2729 | |
6680b8d1 ME |
2730 | *priority = s->priority; |
2731 | return 0; | |
fd38203a LP |
2732 | } |
2733 | ||
31927c16 | 2734 | _public_ int sd_event_source_set_priority(sd_event_source *s, int64_t priority) { |
97ef5391 LP |
2735 | bool rm_inotify = false, rm_inode = false; |
2736 | struct inotify_data *new_inotify_data = NULL; | |
2737 | struct inode_data *new_inode_data = NULL; | |
9da4cb2b LP |
2738 | int r; |
2739 | ||
305f78bf | 2740 | assert_return(s, -EINVAL); |
da7e457c | 2741 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 2742 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
2743 | |
2744 | if (s->priority == priority) | |
2745 | return 0; | |
2746 | ||
97ef5391 LP |
2747 | if (s->type == SOURCE_INOTIFY) { |
2748 | struct inode_data *old_inode_data; | |
2749 | ||
2750 | assert(s->inotify.inode_data); | |
2751 | old_inode_data = s->inotify.inode_data; | |
2752 | ||
2753 | /* We need the original fd to change the priority. If we don't have it we can't change the priority, | |
2754 | * anymore. Note that we close any fds when entering the next event loop iteration, i.e. for inotify | |
2755 | * events we allow priority changes only until the first following iteration. */ | |
2756 | if (old_inode_data->fd < 0) | |
2757 | return -EOPNOTSUPP; | |
2758 | ||
2759 | r = event_make_inotify_data(s->event, priority, &new_inotify_data); | |
2760 | if (r < 0) | |
2761 | return r; | |
2762 | rm_inotify = r > 0; | |
2763 | ||
2764 | r = event_make_inode_data(s->event, new_inotify_data, old_inode_data->dev, old_inode_data->ino, &new_inode_data); | |
2765 | if (r < 0) | |
2766 | goto fail; | |
2767 | rm_inode = r > 0; | |
2768 | ||
2769 | if (new_inode_data->fd < 0) { | |
2770 | /* Duplicate the fd for the new inode object if we don't have any yet */ | |
2771 | new_inode_data->fd = fcntl(old_inode_data->fd, F_DUPFD_CLOEXEC, 3); | |
2772 | if (new_inode_data->fd < 0) { | |
2773 | r = -errno; | |
2774 | goto fail; | |
2775 | } | |
2776 | ||
ed828563 | 2777 | LIST_PREPEND(to_close, s->event->inode_data_to_close_list, new_inode_data); |
97ef5391 LP |
2778 | } |
2779 | ||
2780 | /* Move the event source to the new inode data structure */ | |
2781 | LIST_REMOVE(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2782 | LIST_PREPEND(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2783 | s->inotify.inode_data = new_inode_data; | |
2784 | ||
2785 | /* Now create the new watch */ | |
2786 | r = inode_data_realize_watch(s->event, new_inode_data); | |
2787 | if (r < 0) { | |
2788 | /* Move it back */ | |
2789 | LIST_REMOVE(inotify.by_inode_data, new_inode_data->event_sources, s); | |
2790 | LIST_PREPEND(inotify.by_inode_data, old_inode_data->event_sources, s); | |
2791 | s->inotify.inode_data = old_inode_data; | |
2792 | goto fail; | |
2793 | } | |
2794 | ||
2795 | s->priority = priority; | |
2796 | ||
2797 | event_gc_inode_data(s->event, old_inode_data); | |
2798 | ||
b6d5481b | 2799 | } else if (s->type == SOURCE_SIGNAL && event_source_is_online(s)) { |
9da4cb2b LP |
2800 | struct signal_data *old, *d; |
2801 | ||
2802 | /* Move us from the signalfd belonging to the old | |
2803 | * priority to the signalfd of the new priority */ | |
2804 | ||
2805 | assert_se(old = hashmap_get(s->event->signal_data, &s->priority)); | |
2806 | ||
2807 | s->priority = priority; | |
2808 | ||
2809 | r = event_make_signal_data(s->event, s->signal.sig, &d); | |
2810 | if (r < 0) { | |
2811 | s->priority = old->priority; | |
2812 | return r; | |
2813 | } | |
2814 | ||
2815 | event_unmask_signal_data(s->event, old, s->signal.sig); | |
2816 | } else | |
2817 | s->priority = priority; | |
fd38203a | 2818 | |
e1951c16 | 2819 | event_source_pp_prioq_reshuffle(s); |
fd38203a | 2820 | |
6203e07a LP |
2821 | if (s->type == SOURCE_EXIT) |
2822 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
305f78bf | 2823 | |
fd38203a | 2824 | return 0; |
97ef5391 LP |
2825 | |
2826 | fail: | |
2827 | if (rm_inode) | |
2828 | event_free_inode_data(s->event, new_inode_data); | |
2829 | ||
2830 | if (rm_inotify) | |
2831 | event_free_inotify_data(s->event, new_inotify_data); | |
2832 | ||
2833 | return r; | |
fd38203a LP |
2834 | } |
2835 | ||
cad143a8 | 2836 | _public_ int sd_event_source_get_enabled(sd_event_source *s, int *ret) { |
71193c0b ZJS |
2837 | /* Quick mode: the event source doesn't exist and we only want to query boolean enablement state. */ |
2838 | if (!s && !ret) | |
2839 | return false; | |
2840 | ||
305f78bf | 2841 | assert_return(s, -EINVAL); |
305f78bf | 2842 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 2843 | |
cad143a8 LP |
2844 | if (ret) |
2845 | *ret = s->enabled; | |
2846 | ||
08c1eb0e | 2847 | return s->enabled != SD_EVENT_OFF; |
fd38203a LP |
2848 | } |
2849 | ||
b6d5481b LP |
2850 | static int event_source_offline( |
2851 | sd_event_source *s, | |
2852 | int enabled, | |
2853 | bool ratelimited) { | |
2854 | ||
2855 | bool was_offline; | |
fd38203a LP |
2856 | int r; |
2857 | ||
ddfde737 | 2858 | assert(s); |
b6d5481b | 2859 | assert(enabled == SD_EVENT_OFF || ratelimited); |
fd38203a | 2860 | |
ddfde737 | 2861 | /* Unset the pending flag when this event source is disabled */ |
b6d5481b LP |
2862 | if (s->enabled != SD_EVENT_OFF && |
2863 | enabled == SD_EVENT_OFF && | |
2864 | !IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
ddfde737 LP |
2865 | r = source_set_pending(s, false); |
2866 | if (r < 0) | |
2867 | return r; | |
2868 | } | |
cc567911 | 2869 | |
b6d5481b LP |
2870 | was_offline = event_source_is_offline(s); |
2871 | s->enabled = enabled; | |
2872 | s->ratelimited = ratelimited; | |
fd38203a | 2873 | |
ddfde737 | 2874 | switch (s->type) { |
fd38203a | 2875 | |
ddfde737 LP |
2876 | case SOURCE_IO: |
2877 | source_io_unregister(s); | |
2878 | break; | |
ac989a78 | 2879 | |
ddfde737 LP |
2880 | case SOURCE_SIGNAL: |
2881 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); | |
2882 | break; | |
fd38203a | 2883 | |
ddfde737 | 2884 | case SOURCE_CHILD: |
b6d5481b LP |
2885 | if (!was_offline) { |
2886 | assert(s->event->n_online_child_sources > 0); | |
2887 | s->event->n_online_child_sources--; | |
2888 | } | |
fd38203a | 2889 | |
ddfde737 LP |
2890 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
2891 | source_child_pidfd_unregister(s); | |
2892 | else | |
2893 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); | |
2894 | break; | |
4807d2d0 | 2895 | |
ddfde737 LP |
2896 | case SOURCE_EXIT: |
2897 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); | |
2898 | break; | |
fd38203a | 2899 | |
158fe190 LP |
2900 | case SOURCE_MEMORY_PRESSURE: |
2901 | source_memory_pressure_unregister(s); | |
2902 | break; | |
2903 | ||
2115b9b6 YW |
2904 | case SOURCE_TIME_REALTIME: |
2905 | case SOURCE_TIME_BOOTTIME: | |
2906 | case SOURCE_TIME_MONOTONIC: | |
2907 | case SOURCE_TIME_REALTIME_ALARM: | |
2908 | case SOURCE_TIME_BOOTTIME_ALARM: | |
ddfde737 LP |
2909 | case SOURCE_DEFER: |
2910 | case SOURCE_POST: | |
2911 | case SOURCE_INOTIFY: | |
2912 | break; | |
fd38203a | 2913 | |
ddfde737 | 2914 | default: |
04499a70 | 2915 | assert_not_reached(); |
ddfde737 | 2916 | } |
fd38203a | 2917 | |
2115b9b6 YW |
2918 | /* Always reshuffle time prioq, as the ratelimited flag may be changed. */ |
2919 | event_source_time_prioq_reshuffle(s); | |
2920 | ||
b6d5481b | 2921 | return 1; |
ddfde737 | 2922 | } |
f8f3f926 | 2923 | |
b6d5481b LP |
2924 | static int event_source_online( |
2925 | sd_event_source *s, | |
2926 | int enabled, | |
2927 | bool ratelimited) { | |
2928 | ||
2929 | bool was_online; | |
ddfde737 | 2930 | int r; |
fd38203a | 2931 | |
ddfde737 | 2932 | assert(s); |
b6d5481b | 2933 | assert(enabled != SD_EVENT_OFF || !ratelimited); |
305f78bf | 2934 | |
ddfde737 | 2935 | /* Unset the pending flag when this event source is enabled */ |
b6d5481b LP |
2936 | if (s->enabled == SD_EVENT_OFF && |
2937 | enabled != SD_EVENT_OFF && | |
2938 | !IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { | |
ddfde737 LP |
2939 | r = source_set_pending(s, false); |
2940 | if (r < 0) | |
2941 | return r; | |
2942 | } | |
9d3e3aa5 | 2943 | |
b6d5481b LP |
2944 | /* Are we really ready for onlining? */ |
2945 | if (enabled == SD_EVENT_OFF || ratelimited) { | |
2946 | /* Nope, we are not ready for onlining, then just update the precise state and exit */ | |
2947 | s->enabled = enabled; | |
2948 | s->ratelimited = ratelimited; | |
2949 | return 0; | |
2950 | } | |
2951 | ||
2952 | was_online = event_source_is_online(s); | |
2953 | ||
ddfde737 | 2954 | switch (s->type) { |
ddfde737 | 2955 | case SOURCE_IO: |
b6d5481b | 2956 | r = source_io_register(s, enabled, s->io.events); |
d2eafe61 | 2957 | if (r < 0) |
ddfde737 | 2958 | return r; |
ddfde737 | 2959 | break; |
fd38203a | 2960 | |
ddfde737 LP |
2961 | case SOURCE_SIGNAL: |
2962 | r = event_make_signal_data(s->event, s->signal.sig, NULL); | |
2963 | if (r < 0) { | |
ddfde737 LP |
2964 | event_gc_signal_data(s->event, &s->priority, s->signal.sig); |
2965 | return r; | |
2966 | } | |
fd38203a | 2967 | |
ddfde737 | 2968 | break; |
fd38203a | 2969 | |
ddfde737 | 2970 | case SOURCE_CHILD: |
ddfde737 LP |
2971 | if (EVENT_SOURCE_WATCH_PIDFD(s)) { |
2972 | /* yes, we have pidfd */ | |
9da4cb2b | 2973 | |
b6d5481b | 2974 | r = source_child_pidfd_register(s, enabled); |
ac9f2640 | 2975 | if (r < 0) |
9da4cb2b | 2976 | return r; |
ddfde737 LP |
2977 | } else { |
2978 | /* no pidfd, or something other to watch for than WEXITED */ | |
9da4cb2b | 2979 | |
ddfde737 LP |
2980 | r = event_make_signal_data(s->event, SIGCHLD, NULL); |
2981 | if (r < 0) { | |
ddfde737 LP |
2982 | event_gc_signal_data(s->event, &s->priority, SIGCHLD); |
2983 | return r; | |
2984 | } | |
2985 | } | |
fd38203a | 2986 | |
b6d5481b LP |
2987 | if (!was_online) |
2988 | s->event->n_online_child_sources++; | |
ddfde737 | 2989 | break; |
4807d2d0 | 2990 | |
158fe190 LP |
2991 | case SOURCE_MEMORY_PRESSURE: |
2992 | r = source_memory_pressure_register(s, enabled); | |
2993 | if (r < 0) | |
2994 | return r; | |
2995 | ||
2996 | break; | |
2997 | ||
d2eafe61 ZJS |
2998 | case SOURCE_TIME_REALTIME: |
2999 | case SOURCE_TIME_BOOTTIME: | |
3000 | case SOURCE_TIME_MONOTONIC: | |
3001 | case SOURCE_TIME_REALTIME_ALARM: | |
3002 | case SOURCE_TIME_BOOTTIME_ALARM: | |
ddfde737 | 3003 | case SOURCE_EXIT: |
ddfde737 LP |
3004 | case SOURCE_DEFER: |
3005 | case SOURCE_POST: | |
3006 | case SOURCE_INOTIFY: | |
3007 | break; | |
9da4cb2b | 3008 | |
ddfde737 | 3009 | default: |
04499a70 | 3010 | assert_not_reached(); |
ddfde737 | 3011 | } |
f8f3f926 | 3012 | |
b6d5481b LP |
3013 | s->enabled = enabled; |
3014 | s->ratelimited = ratelimited; | |
d2eafe61 ZJS |
3015 | |
3016 | /* Non-failing operations below */ | |
2115b9b6 | 3017 | if (s->type == SOURCE_EXIT) |
d2eafe61 | 3018 | prioq_reshuffle(s->event->exit, s, &s->exit.prioq_index); |
d2eafe61 | 3019 | |
2115b9b6 YW |
3020 | /* Always reshuffle time prioq, as the ratelimited flag may be changed. */ |
3021 | event_source_time_prioq_reshuffle(s); | |
d2eafe61 | 3022 | |
b6d5481b | 3023 | return 1; |
ddfde737 LP |
3024 | } |
3025 | ||
3026 | _public_ int sd_event_source_set_enabled(sd_event_source *s, int m) { | |
3027 | int r; | |
9da4cb2b | 3028 | |
ddfde737 | 3029 | assert_return(IN_SET(m, SD_EVENT_OFF, SD_EVENT_ON, SD_EVENT_ONESHOT), -EINVAL); |
7e922b05 ZJS |
3030 | |
3031 | /* Quick mode: if the source doesn't exist, SD_EVENT_OFF is a noop. */ | |
3032 | if (m == SD_EVENT_OFF && !s) | |
3033 | return 0; | |
3034 | ||
3035 | assert_return(s, -EINVAL); | |
ddfde737 | 3036 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 3037 | |
ddfde737 LP |
3038 | /* If we are dead anyway, we are fine with turning off sources, but everything else needs to fail. */ |
3039 | if (s->event->state == SD_EVENT_FINISHED) | |
3040 | return m == SD_EVENT_OFF ? 0 : -ESTALE; | |
305f78bf | 3041 | |
ddfde737 LP |
3042 | if (s->enabled == m) /* No change? */ |
3043 | return 0; | |
9d3e3aa5 | 3044 | |
ddfde737 | 3045 | if (m == SD_EVENT_OFF) |
b6d5481b | 3046 | r = event_source_offline(s, m, s->ratelimited); |
ddfde737 LP |
3047 | else { |
3048 | if (s->enabled != SD_EVENT_OFF) { | |
3049 | /* Switching from "on" to "oneshot" or back? If that's the case, we can take a shortcut, the | |
3050 | * event source is already enabled after all. */ | |
3051 | s->enabled = m; | |
3052 | return 0; | |
fd38203a | 3053 | } |
ddfde737 | 3054 | |
b6d5481b | 3055 | r = event_source_online(s, m, s->ratelimited); |
fd38203a | 3056 | } |
ddfde737 LP |
3057 | if (r < 0) |
3058 | return r; | |
fd38203a | 3059 | |
e1951c16 | 3060 | event_source_pp_prioq_reshuffle(s); |
fd38203a LP |
3061 | return 0; |
3062 | } | |
3063 | ||
f7262a9f | 3064 | _public_ int sd_event_source_get_time(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
3065 | assert_return(s, -EINVAL); |
3066 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 3067 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf | 3068 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a LP |
3069 | |
3070 | *usec = s->time.next; | |
3071 | return 0; | |
3072 | } | |
3073 | ||
f7262a9f | 3074 | _public_ int sd_event_source_set_time(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 3075 | int r; |
6a0f1f6d | 3076 | |
305f78bf | 3077 | assert_return(s, -EINVAL); |
6a0f1f6d | 3078 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 3079 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 3080 | assert_return(!event_pid_changed(s->event), -ECHILD); |
fd38203a | 3081 | |
2a0dc6cd LP |
3082 | r = source_set_pending(s, false); |
3083 | if (r < 0) | |
3084 | return r; | |
2576a19e | 3085 | |
2a0dc6cd | 3086 | s->time.next = usec; |
fd38203a | 3087 | |
e1951c16 | 3088 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
3089 | return 0; |
3090 | } | |
3091 | ||
d6a83dc4 LP |
3092 | _public_ int sd_event_source_set_time_relative(sd_event_source *s, uint64_t usec) { |
3093 | usec_t t; | |
3094 | int r; | |
3095 | ||
3096 | assert_return(s, -EINVAL); | |
3097 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
3098 | ||
ef859195 LP |
3099 | if (usec == USEC_INFINITY) |
3100 | return sd_event_source_set_time(s, USEC_INFINITY); | |
3101 | ||
d6a83dc4 LP |
3102 | r = sd_event_now(s->event, event_source_type_to_clock(s->type), &t); |
3103 | if (r < 0) | |
3104 | return r; | |
3105 | ||
496db330 YW |
3106 | usec = usec_add(t, usec); |
3107 | if (usec == USEC_INFINITY) | |
d6a83dc4 LP |
3108 | return -EOVERFLOW; |
3109 | ||
496db330 | 3110 | return sd_event_source_set_time(s, usec); |
d6a83dc4 LP |
3111 | } |
3112 | ||
f7262a9f | 3113 | _public_ int sd_event_source_get_time_accuracy(sd_event_source *s, uint64_t *usec) { |
305f78bf LP |
3114 | assert_return(s, -EINVAL); |
3115 | assert_return(usec, -EINVAL); | |
6a0f1f6d | 3116 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
305f78bf LP |
3117 | assert_return(!event_pid_changed(s->event), -ECHILD); |
3118 | ||
3119 | *usec = s->time.accuracy; | |
3120 | return 0; | |
3121 | } | |
3122 | ||
f7262a9f | 3123 | _public_ int sd_event_source_set_time_accuracy(sd_event_source *s, uint64_t usec) { |
2a0dc6cd | 3124 | int r; |
6a0f1f6d | 3125 | |
305f78bf | 3126 | assert_return(s, -EINVAL); |
f5fbe71d | 3127 | assert_return(usec != UINT64_MAX, -EINVAL); |
6a0f1f6d | 3128 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); |
da7e457c | 3129 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
305f78bf | 3130 | assert_return(!event_pid_changed(s->event), -ECHILD); |
eaa3cbef | 3131 | |
2a0dc6cd LP |
3132 | r = source_set_pending(s, false); |
3133 | if (r < 0) | |
3134 | return r; | |
3135 | ||
eaa3cbef LP |
3136 | if (usec == 0) |
3137 | usec = DEFAULT_ACCURACY_USEC; | |
3138 | ||
eaa3cbef LP |
3139 | s->time.accuracy = usec; |
3140 | ||
e1951c16 | 3141 | event_source_time_prioq_reshuffle(s); |
6a0f1f6d LP |
3142 | return 0; |
3143 | } | |
3144 | ||
3145 | _public_ int sd_event_source_get_time_clock(sd_event_source *s, clockid_t *clock) { | |
3146 | assert_return(s, -EINVAL); | |
3147 | assert_return(clock, -EINVAL); | |
3148 | assert_return(EVENT_SOURCE_IS_TIME(s->type), -EDOM); | |
3149 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
eaa3cbef | 3150 | |
6a0f1f6d | 3151 | *clock = event_source_type_to_clock(s->type); |
eaa3cbef LP |
3152 | return 0; |
3153 | } | |
3154 | ||
f7262a9f | 3155 | _public_ int sd_event_source_get_child_pid(sd_event_source *s, pid_t *pid) { |
4bee8012 LP |
3156 | assert_return(s, -EINVAL); |
3157 | assert_return(pid, -EINVAL); | |
3158 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
3159 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
3160 | ||
3161 | *pid = s->child.pid; | |
3162 | return 0; | |
3163 | } | |
3164 | ||
f8f3f926 LP |
3165 | _public_ int sd_event_source_get_child_pidfd(sd_event_source *s) { |
3166 | assert_return(s, -EINVAL); | |
3167 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
3168 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
3169 | ||
3170 | if (s->child.pidfd < 0) | |
3171 | return -EOPNOTSUPP; | |
3172 | ||
3173 | return s->child.pidfd; | |
3174 | } | |
3175 | ||
3176 | _public_ int sd_event_source_send_child_signal(sd_event_source *s, int sig, const siginfo_t *si, unsigned flags) { | |
3177 | assert_return(s, -EINVAL); | |
3178 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
3179 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
3180 | assert_return(SIGNAL_VALID(sig), -EINVAL); | |
3181 | ||
3182 | /* If we already have seen indication the process exited refuse sending a signal early. This way we | |
3183 | * can be sure we don't accidentally kill the wrong process on PID reuse when pidfds are not | |
3184 | * available. */ | |
3185 | if (s->child.exited) | |
3186 | return -ESRCH; | |
3187 | ||
3188 | if (s->child.pidfd >= 0) { | |
3189 | siginfo_t copy; | |
3190 | ||
3191 | /* pidfd_send_signal() changes the siginfo_t argument. This is weird, let's hence copy the | |
3192 | * structure here */ | |
3193 | if (si) | |
3194 | copy = *si; | |
3195 | ||
3196 | if (pidfd_send_signal(s->child.pidfd, sig, si ? © : NULL, 0) < 0) { | |
3197 | /* Let's propagate the error only if the system call is not implemented or prohibited */ | |
3198 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) | |
3199 | return -errno; | |
3200 | } else | |
3201 | return 0; | |
3202 | } | |
3203 | ||
3204 | /* Flags are only supported for pidfd_send_signal(), not for rt_sigqueueinfo(), hence let's refuse | |
3205 | * this here. */ | |
3206 | if (flags != 0) | |
3207 | return -EOPNOTSUPP; | |
3208 | ||
3209 | if (si) { | |
3210 | /* We use rt_sigqueueinfo() only if siginfo_t is specified. */ | |
3211 | siginfo_t copy = *si; | |
3212 | ||
3213 | if (rt_sigqueueinfo(s->child.pid, sig, ©) < 0) | |
3214 | return -errno; | |
3215 | } else if (kill(s->child.pid, sig) < 0) | |
3216 | return -errno; | |
3217 | ||
3218 | return 0; | |
3219 | } | |
3220 | ||
3221 | _public_ int sd_event_source_get_child_pidfd_own(sd_event_source *s) { | |
3222 | assert_return(s, -EINVAL); | |
3223 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
3224 | ||
3225 | if (s->child.pidfd < 0) | |
3226 | return -EOPNOTSUPP; | |
3227 | ||
3228 | return s->child.pidfd_owned; | |
3229 | } | |
3230 | ||
3231 | _public_ int sd_event_source_set_child_pidfd_own(sd_event_source *s, int own) { | |
3232 | assert_return(s, -EINVAL); | |
3233 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
3234 | ||
3235 | if (s->child.pidfd < 0) | |
3236 | return -EOPNOTSUPP; | |
3237 | ||
3238 | s->child.pidfd_owned = own; | |
3239 | return 0; | |
3240 | } | |
3241 | ||
3242 | _public_ int sd_event_source_get_child_process_own(sd_event_source *s) { | |
3243 | assert_return(s, -EINVAL); | |
3244 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
3245 | ||
3246 | return s->child.process_owned; | |
3247 | } | |
3248 | ||
3249 | _public_ int sd_event_source_set_child_process_own(sd_event_source *s, int own) { | |
3250 | assert_return(s, -EINVAL); | |
3251 | assert_return(s->type == SOURCE_CHILD, -EDOM); | |
3252 | ||
3253 | s->child.process_owned = own; | |
3254 | return 0; | |
3255 | } | |
3256 | ||
97ef5391 LP |
3257 | _public_ int sd_event_source_get_inotify_mask(sd_event_source *s, uint32_t *mask) { |
3258 | assert_return(s, -EINVAL); | |
3259 | assert_return(mask, -EINVAL); | |
3260 | assert_return(s->type == SOURCE_INOTIFY, -EDOM); | |
3261 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
3262 | ||
3263 | *mask = s->inotify.mask; | |
3264 | return 0; | |
3265 | } | |
3266 | ||
718db961 | 3267 | _public_ int sd_event_source_set_prepare(sd_event_source *s, sd_event_handler_t callback) { |
fd38203a LP |
3268 | int r; |
3269 | ||
da7e457c | 3270 | assert_return(s, -EINVAL); |
6203e07a | 3271 | assert_return(s->type != SOURCE_EXIT, -EDOM); |
da7e457c LP |
3272 | assert_return(s->event->state != SD_EVENT_FINISHED, -ESTALE); |
3273 | assert_return(!event_pid_changed(s->event), -ECHILD); | |
fd38203a LP |
3274 | |
3275 | if (s->prepare == callback) | |
3276 | return 0; | |
3277 | ||
3278 | if (callback && s->prepare) { | |
3279 | s->prepare = callback; | |
3280 | return 0; | |
3281 | } | |
3282 | ||
3283 | r = prioq_ensure_allocated(&s->event->prepare, prepare_prioq_compare); | |
3284 | if (r < 0) | |
3285 | return r; | |
3286 | ||
3287 | s->prepare = callback; | |
3288 | ||
3289 | if (callback) { | |
3290 | r = prioq_put(s->event->prepare, s, &s->prepare_index); | |
3291 | if (r < 0) | |
3292 | return r; | |
3293 | } else | |
3294 | prioq_remove(s->event->prepare, s, &s->prepare_index); | |
3295 | ||
3296 | return 0; | |
3297 | } | |
3298 | ||
f7262a9f | 3299 | _public_ void* sd_event_source_get_userdata(sd_event_source *s) { |
da7e457c | 3300 | assert_return(s, NULL); |
fd38203a LP |
3301 | |
3302 | return s->userdata; | |
3303 | } | |
3304 | ||
8f726607 LP |
3305 | _public_ void *sd_event_source_set_userdata(sd_event_source *s, void *userdata) { |
3306 | void *ret; | |
3307 | ||
3308 | assert_return(s, NULL); | |
3309 | ||
3310 | ret = s->userdata; | |
3311 | s->userdata = userdata; | |
3312 | ||
3313 | return ret; | |
3314 | } | |
3315 | ||
b6d5481b LP |
3316 | static int event_source_enter_ratelimited(sd_event_source *s) { |
3317 | int r; | |
3318 | ||
3319 | assert(s); | |
3320 | ||
3321 | /* When an event source becomes ratelimited, we place it in the CLOCK_MONOTONIC priority queue, with | |
3322 | * the end of the rate limit time window, much as if it was a timer event source. */ | |
3323 | ||
3324 | if (s->ratelimited) | |
3325 | return 0; /* Already ratelimited, this is a NOP hence */ | |
3326 | ||
3327 | /* Make sure we can install a CLOCK_MONOTONIC event further down. */ | |
3328 | r = setup_clock_data(s->event, &s->event->monotonic, CLOCK_MONOTONIC); | |
3329 | if (r < 0) | |
3330 | return r; | |
3331 | ||
3332 | /* Timer event sources are already using the earliest/latest queues for the timer scheduling. Let's | |
3333 | * first remove them from the prioq appropriate for their own clock, so that we can use the prioq | |
3334 | * fields of the event source then for adding it to the CLOCK_MONOTONIC prioq instead. */ | |
3335 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
3336 | event_source_time_prioq_remove(s, event_get_clock_data(s->event, s->type)); | |
3337 | ||
3338 | /* Now, let's add the event source to the monotonic clock instead */ | |
3339 | r = event_source_time_prioq_put(s, &s->event->monotonic); | |
3340 | if (r < 0) | |
3341 | goto fail; | |
3342 | ||
3343 | /* And let's take the event source officially offline */ | |
3344 | r = event_source_offline(s, s->enabled, /* ratelimited= */ true); | |
3345 | if (r < 0) { | |
3346 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
3347 | goto fail; | |
3348 | } | |
3349 | ||
3350 | event_source_pp_prioq_reshuffle(s); | |
3351 | ||
3352 | log_debug("Event source %p (%s) entered rate limit state.", s, strna(s->description)); | |
3353 | return 0; | |
3354 | ||
3355 | fail: | |
3356 | /* Reinstall time event sources in the priority queue as before. This shouldn't fail, since the queue | |
3357 | * space for it should already be allocated. */ | |
3358 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
3359 | assert_se(event_source_time_prioq_put(s, event_get_clock_data(s->event, s->type)) >= 0); | |
3360 | ||
3361 | return r; | |
3362 | } | |
3363 | ||
fd69f224 | 3364 | static int event_source_leave_ratelimit(sd_event_source *s, bool run_callback) { |
b6d5481b LP |
3365 | int r; |
3366 | ||
3367 | assert(s); | |
3368 | ||
3369 | if (!s->ratelimited) | |
3370 | return 0; | |
3371 | ||
3372 | /* Let's take the event source out of the monotonic prioq first. */ | |
3373 | event_source_time_prioq_remove(s, &s->event->monotonic); | |
3374 | ||
3375 | /* Let's then add the event source to its native clock prioq again — if this is a timer event source */ | |
3376 | if (EVENT_SOURCE_IS_TIME(s->type)) { | |
3377 | r = event_source_time_prioq_put(s, event_get_clock_data(s->event, s->type)); | |
3378 | if (r < 0) | |
3379 | goto fail; | |
3380 | } | |
3381 | ||
3382 | /* Let's try to take it online again. */ | |
3383 | r = event_source_online(s, s->enabled, /* ratelimited= */ false); | |
3384 | if (r < 0) { | |
3385 | /* Do something roughly sensible when this failed: undo the two prioq ops above */ | |
3386 | if (EVENT_SOURCE_IS_TIME(s->type)) | |
3387 | event_source_time_prioq_remove(s, event_get_clock_data(s->event, s->type)); | |
3388 | ||
3389 | goto fail; | |
3390 | } | |
3391 | ||
3392 | event_source_pp_prioq_reshuffle(s); | |
3393 | ratelimit_reset(&s->rate_limit); | |
3394 | ||
3395 | log_debug("Event source %p (%s) left rate limit state.", s, strna(s->description)); | |
fd69f224 MS |
3396 | |
3397 | if (run_callback && s->ratelimit_expire_callback) { | |
3398 | s->dispatching = true; | |
3399 | r = s->ratelimit_expire_callback(s, s->userdata); | |
3400 | s->dispatching = false; | |
3401 | ||
3402 | if (r < 0) { | |
3403 | log_debug_errno(r, "Ratelimit expiry callback of event source %s (type %s) returned error, %s: %m", | |
3404 | strna(s->description), | |
3405 | event_source_type_to_string(s->type), | |
3406 | s->exit_on_failure ? "exiting" : "disabling"); | |
3407 | ||
3408 | if (s->exit_on_failure) | |
3409 | (void) sd_event_exit(s->event, r); | |
3410 | } | |
3411 | ||
3412 | if (s->n_ref == 0) | |
3413 | source_free(s); | |
3414 | else if (r < 0) | |
0a040e64 | 3415 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
fd69f224 MS |
3416 | |
3417 | return 1; | |
3418 | } | |
3419 | ||
b6d5481b LP |
3420 | return 0; |
3421 | ||
3422 | fail: | |
3423 | /* Do something somewhat reasonable when we cannot move an event sources out of ratelimited mode: | |
3424 | * simply put it back in it, maybe we can then process it more successfully next iteration. */ | |
3425 | assert_se(event_source_time_prioq_put(s, &s->event->monotonic) >= 0); | |
3426 | ||
3427 | return r; | |
3428 | } | |
3429 | ||
c2ba3ad6 LP |
3430 | static usec_t sleep_between(sd_event *e, usec_t a, usec_t b) { |
3431 | usec_t c; | |
3432 | assert(e); | |
3433 | assert(a <= b); | |
3434 | ||
3435 | if (a <= 0) | |
3436 | return 0; | |
393003e1 LP |
3437 | if (a >= USEC_INFINITY) |
3438 | return USEC_INFINITY; | |
c2ba3ad6 LP |
3439 | |
3440 | if (b <= a + 1) | |
3441 | return a; | |
3442 | ||
52444dc4 LP |
3443 | initialize_perturb(e); |
3444 | ||
c2ba3ad6 LP |
3445 | /* |
3446 | Find a good time to wake up again between times a and b. We | |
3447 | have two goals here: | |
3448 | ||
3449 | a) We want to wake up as seldom as possible, hence prefer | |
3450 | later times over earlier times. | |
3451 | ||
3452 | b) But if we have to wake up, then let's make sure to | |
3453 | dispatch as much as possible on the entire system. | |
3454 | ||
3455 | We implement this by waking up everywhere at the same time | |
850516e0 | 3456 | within any given minute if we can, synchronised via the |
c2ba3ad6 | 3457 | perturbation value determined from the boot ID. If we can't, |
ba276c81 LP |
3458 | then we try to find the same spot in every 10s, then 1s and |
3459 | then 250ms step. Otherwise, we pick the last possible time | |
3460 | to wake up. | |
c2ba3ad6 LP |
3461 | */ |
3462 | ||
850516e0 LP |
3463 | c = (b / USEC_PER_MINUTE) * USEC_PER_MINUTE + e->perturb; |
3464 | if (c >= b) { | |
3465 | if (_unlikely_(c < USEC_PER_MINUTE)) | |
3466 | return b; | |
3467 | ||
3468 | c -= USEC_PER_MINUTE; | |
3469 | } | |
3470 | ||
ba276c81 LP |
3471 | if (c >= a) |
3472 | return c; | |
3473 | ||
3474 | c = (b / (USEC_PER_SEC*10)) * (USEC_PER_SEC*10) + (e->perturb % (USEC_PER_SEC*10)); | |
3475 | if (c >= b) { | |
3476 | if (_unlikely_(c < USEC_PER_SEC*10)) | |
3477 | return b; | |
3478 | ||
3479 | c -= USEC_PER_SEC*10; | |
3480 | } | |
3481 | ||
850516e0 LP |
3482 | if (c >= a) |
3483 | return c; | |
3484 | ||
3485 | c = (b / USEC_PER_SEC) * USEC_PER_SEC + (e->perturb % USEC_PER_SEC); | |
c2ba3ad6 LP |
3486 | if (c >= b) { |
3487 | if (_unlikely_(c < USEC_PER_SEC)) | |
3488 | return b; | |
3489 | ||
3490 | c -= USEC_PER_SEC; | |
3491 | } | |
3492 | ||
3493 | if (c >= a) | |
3494 | return c; | |
3495 | ||
3496 | c = (b / (USEC_PER_MSEC*250)) * (USEC_PER_MSEC*250) + (e->perturb % (USEC_PER_MSEC*250)); | |
3497 | if (c >= b) { | |
3498 | if (_unlikely_(c < USEC_PER_MSEC*250)) | |
3499 | return b; | |
3500 | ||
3501 | c -= USEC_PER_MSEC*250; | |
3502 | } | |
3503 | ||
3504 | if (c >= a) | |
3505 | return c; | |
3506 | ||
3507 | return b; | |
3508 | } | |
3509 | ||
fd38203a LP |
3510 | static int event_arm_timer( |
3511 | sd_event *e, | |
6a0f1f6d | 3512 | struct clock_data *d) { |
fd38203a LP |
3513 | |
3514 | struct itimerspec its = {}; | |
c2ba3ad6 LP |
3515 | sd_event_source *a, *b; |
3516 | usec_t t; | |
fd38203a | 3517 | |
cde93897 | 3518 | assert(e); |
6a0f1f6d | 3519 | assert(d); |
fd38203a | 3520 | |
d06441da | 3521 | if (!d->needs_rearm) |
212bbb17 | 3522 | return 0; |
7e2bf71c YW |
3523 | |
3524 | d->needs_rearm = false; | |
212bbb17 | 3525 | |
6a0f1f6d | 3526 | a = prioq_peek(d->earliest); |
19947509 | 3527 | assert(!a || EVENT_SOURCE_USES_TIME_PRIOQ(a->type)); |
b6d5481b | 3528 | if (!a || a->enabled == SD_EVENT_OFF || time_event_source_next(a) == USEC_INFINITY) { |
72aedc1e | 3529 | |
6a0f1f6d | 3530 | if (d->fd < 0) |
c57b5ca3 LP |
3531 | return 0; |
3532 | ||
3a43da28 | 3533 | if (d->next == USEC_INFINITY) |
72aedc1e LP |
3534 | return 0; |
3535 | ||
3536 | /* disarm */ | |
15c689d7 LP |
3537 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
3538 | return -errno; | |
72aedc1e | 3539 | |
3a43da28 | 3540 | d->next = USEC_INFINITY; |
fd38203a | 3541 | return 0; |
72aedc1e | 3542 | } |
fd38203a | 3543 | |
6a0f1f6d | 3544 | b = prioq_peek(d->latest); |
19947509 ZJS |
3545 | assert(!b || EVENT_SOURCE_USES_TIME_PRIOQ(b->type)); |
3546 | assert(b && b->enabled != SD_EVENT_OFF); | |
c2ba3ad6 | 3547 | |
b6d5481b | 3548 | t = sleep_between(e, time_event_source_next(a), time_event_source_latest(b)); |
6a0f1f6d | 3549 | if (d->next == t) |
fd38203a LP |
3550 | return 0; |
3551 | ||
6a0f1f6d | 3552 | assert_se(d->fd >= 0); |
fd38203a | 3553 | |
c2ba3ad6 | 3554 | if (t == 0) { |
1751bdde | 3555 | /* We don't want to disarm here, just mean some time looooong ago. */ |
fd38203a LP |
3556 | its.it_value.tv_sec = 0; |
3557 | its.it_value.tv_nsec = 1; | |
3558 | } else | |
c2ba3ad6 | 3559 | timespec_store(&its.it_value, t); |
fd38203a | 3560 | |
15c689d7 | 3561 | if (timerfd_settime(d->fd, TFD_TIMER_ABSTIME, &its, NULL) < 0) |
cde93897 | 3562 | return -errno; |
fd38203a | 3563 | |
6a0f1f6d | 3564 | d->next = t; |
fd38203a LP |
3565 | return 0; |
3566 | } | |
3567 | ||
9a800b56 | 3568 | static int process_io(sd_event *e, sd_event_source *s, uint32_t revents) { |
fd38203a LP |
3569 | assert(e); |
3570 | assert(s); | |
3571 | assert(s->type == SOURCE_IO); | |
3572 | ||
9a800b56 LP |
3573 | /* If the event source was already pending, we just OR in the |
3574 | * new revents, otherwise we reset the value. The ORing is | |
3575 | * necessary to handle EPOLLONESHOT events properly where | |
3576 | * readability might happen independently of writability, and | |
3577 | * we need to keep track of both */ | |
3578 | ||
3579 | if (s->pending) | |
3580 | s->io.revents |= revents; | |
3581 | else | |
3582 | s->io.revents = revents; | |
fd38203a | 3583 | |
fd38203a LP |
3584 | return source_set_pending(s, true); |
3585 | } | |
3586 | ||
72aedc1e | 3587 | static int flush_timer(sd_event *e, int fd, uint32_t events, usec_t *next) { |
fd38203a LP |
3588 | uint64_t x; |
3589 | ssize_t ss; | |
3590 | ||
3591 | assert(e); | |
da7e457c | 3592 | assert(fd >= 0); |
72aedc1e | 3593 | |
305f78bf | 3594 | assert_return(events == EPOLLIN, -EIO); |
fd38203a LP |
3595 | |
3596 | ss = read(fd, &x, sizeof(x)); | |
3597 | if (ss < 0) { | |
8add30a0 | 3598 | if (ERRNO_IS_TRANSIENT(errno)) |
fd38203a LP |
3599 | return 0; |
3600 | ||
3601 | return -errno; | |
3602 | } | |
3603 | ||
8d35dae7 | 3604 | if (_unlikely_(ss != sizeof(x))) |
fd38203a LP |
3605 | return -EIO; |
3606 | ||
cde93897 | 3607 | if (next) |
3a43da28 | 3608 | *next = USEC_INFINITY; |
72aedc1e | 3609 | |
fd38203a LP |
3610 | return 0; |
3611 | } | |
3612 | ||
305f78bf LP |
3613 | static int process_timer( |
3614 | sd_event *e, | |
3615 | usec_t n, | |
6a0f1f6d | 3616 | struct clock_data *d) { |
305f78bf | 3617 | |
fd38203a | 3618 | sd_event_source *s; |
fd69f224 | 3619 | bool callback_invoked = false; |
fd38203a LP |
3620 | int r; |
3621 | ||
3622 | assert(e); | |
6a0f1f6d | 3623 | assert(d); |
fd38203a LP |
3624 | |
3625 | for (;;) { | |
6a0f1f6d | 3626 | s = prioq_peek(d->earliest); |
19947509 ZJS |
3627 | assert(!s || EVENT_SOURCE_USES_TIME_PRIOQ(s->type)); |
3628 | ||
b6d5481b LP |
3629 | if (!s || time_event_source_next(s) > n) |
3630 | break; | |
3631 | ||
3632 | if (s->ratelimited) { | |
3633 | /* This is an event sources whose ratelimit window has ended. Let's turn it on | |
3634 | * again. */ | |
3635 | assert(s->ratelimited); | |
3636 | ||
fd69f224 | 3637 | r = event_source_leave_ratelimit(s, /* run_callback */ true); |
b6d5481b LP |
3638 | if (r < 0) |
3639 | return r; | |
fd69f224 MS |
3640 | else if (r == 1) |
3641 | callback_invoked = true; | |
b6d5481b LP |
3642 | |
3643 | continue; | |
3644 | } | |
3645 | ||
3646 | if (s->enabled == SD_EVENT_OFF || s->pending) | |
fd38203a LP |
3647 | break; |
3648 | ||
3649 | r = source_set_pending(s, true); | |
3650 | if (r < 0) | |
3651 | return r; | |
3652 | ||
e1951c16 | 3653 | event_source_time_prioq_reshuffle(s); |
fd38203a LP |
3654 | } |
3655 | ||
fd69f224 | 3656 | return callback_invoked; |
fd38203a LP |
3657 | } |
3658 | ||
efd3be9d YW |
3659 | static int process_child(sd_event *e, int64_t threshold, int64_t *ret_min_priority) { |
3660 | int64_t min_priority = threshold; | |
3661 | bool something_new = false; | |
fd38203a | 3662 | sd_event_source *s; |
fd38203a LP |
3663 | int r; |
3664 | ||
3665 | assert(e); | |
efd3be9d YW |
3666 | assert(ret_min_priority); |
3667 | ||
3668 | if (!e->need_process_child) { | |
3669 | *ret_min_priority = min_priority; | |
3670 | return 0; | |
3671 | } | |
fd38203a | 3672 | |
c2ba3ad6 LP |
3673 | e->need_process_child = false; |
3674 | ||
91c70071 YW |
3675 | /* So, this is ugly. We iteratively invoke waitid() with P_PID + WNOHANG for each PID we wait |
3676 | * for, instead of using P_ALL. This is because we only want to get child information of very | |
3677 | * specific child processes, and not all of them. We might not have processed the SIGCHLD event | |
3678 | * of a previous invocation and we don't want to maintain a unbounded *per-child* event queue, | |
3679 | * hence we really don't want anything flushed out of the kernel's queue that we don't care | |
3680 | * about. Since this is O(n) this means that if you have a lot of processes you probably want | |
3681 | * to handle SIGCHLD yourself. | |
3682 | * | |
3683 | * We do not reap the children here (by using WNOWAIT), this is only done after the event | |
3684 | * source is dispatched so that the callback still sees the process as a zombie. */ | |
fd38203a | 3685 | |
90e74a66 | 3686 | HASHMAP_FOREACH(s, e->child_sources) { |
fd38203a LP |
3687 | assert(s->type == SOURCE_CHILD); |
3688 | ||
efd3be9d YW |
3689 | if (s->priority > threshold) |
3690 | continue; | |
3691 | ||
fd38203a LP |
3692 | if (s->pending) |
3693 | continue; | |
3694 | ||
b6d5481b | 3695 | if (event_source_is_offline(s)) |
fd38203a LP |
3696 | continue; |
3697 | ||
f8f3f926 LP |
3698 | if (s->child.exited) |
3699 | continue; | |
3700 | ||
91c70071 YW |
3701 | if (EVENT_SOURCE_WATCH_PIDFD(s)) |
3702 | /* There's a usable pidfd known for this event source? Then don't waitid() for | |
3703 | * it here */ | |
f8f3f926 LP |
3704 | continue; |
3705 | ||
fd38203a | 3706 | zero(s->child.siginfo); |
15c689d7 LP |
3707 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, |
3708 | WNOHANG | (s->child.options & WEXITED ? WNOWAIT : 0) | s->child.options) < 0) | |
bfd9bfcc | 3709 | return negative_errno(); |
fd38203a LP |
3710 | |
3711 | if (s->child.siginfo.si_pid != 0) { | |
945c2931 | 3712 | bool zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 3713 | |
f8f3f926 LP |
3714 | if (zombie) |
3715 | s->child.exited = true; | |
3716 | ||
08cd1552 | 3717 | if (!zombie && (s->child.options & WEXITED)) { |
91c70071 YW |
3718 | /* If the child isn't dead then let's immediately remove the state |
3719 | * change from the queue, since there's no benefit in leaving it | |
3720 | * queued. */ | |
08cd1552 LP |
3721 | |
3722 | assert(s->child.options & (WSTOPPED|WCONTINUED)); | |
a5d27871 | 3723 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|(s->child.options & (WSTOPPED|WCONTINUED))); |
08cd1552 LP |
3724 | } |
3725 | ||
fd38203a LP |
3726 | r = source_set_pending(s, true); |
3727 | if (r < 0) | |
3728 | return r; | |
efd3be9d YW |
3729 | if (r > 0) { |
3730 | something_new = true; | |
3731 | min_priority = MIN(min_priority, s->priority); | |
3732 | } | |
fd38203a LP |
3733 | } |
3734 | } | |
3735 | ||
efd3be9d YW |
3736 | *ret_min_priority = min_priority; |
3737 | return something_new; | |
fd38203a LP |
3738 | } |
3739 | ||
f8f3f926 LP |
3740 | static int process_pidfd(sd_event *e, sd_event_source *s, uint32_t revents) { |
3741 | assert(e); | |
3742 | assert(s); | |
3743 | assert(s->type == SOURCE_CHILD); | |
3744 | ||
3745 | if (s->pending) | |
3746 | return 0; | |
3747 | ||
b6d5481b | 3748 | if (event_source_is_offline(s)) |
f8f3f926 LP |
3749 | return 0; |
3750 | ||
3751 | if (!EVENT_SOURCE_WATCH_PIDFD(s)) | |
3752 | return 0; | |
3753 | ||
3754 | zero(s->child.siginfo); | |
3755 | if (waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG | WNOWAIT | s->child.options) < 0) | |
3756 | return -errno; | |
3757 | ||
3758 | if (s->child.siginfo.si_pid == 0) | |
3759 | return 0; | |
3760 | ||
3761 | if (IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED)) | |
3762 | s->child.exited = true; | |
3763 | ||
3764 | return source_set_pending(s, true); | |
3765 | } | |
3766 | ||
efd3be9d | 3767 | static int process_signal(sd_event *e, struct signal_data *d, uint32_t events, int64_t *min_priority) { |
fd38203a LP |
3768 | int r; |
3769 | ||
da7e457c | 3770 | assert(e); |
97ef5391 | 3771 | assert(d); |
305f78bf | 3772 | assert_return(events == EPOLLIN, -EIO); |
efd3be9d | 3773 | assert(min_priority); |
fd38203a | 3774 | |
91c70071 YW |
3775 | /* If there's a signal queued on this priority and SIGCHLD is on this priority too, then make |
3776 | * sure to recheck the children we watch. This is because we only ever dequeue the first signal | |
3777 | * per priority, and if we dequeue one, and SIGCHLD might be enqueued later we wouldn't know, | |
3778 | * but we might have higher priority children we care about hence we need to check that | |
3779 | * explicitly. */ | |
9da4cb2b LP |
3780 | |
3781 | if (sigismember(&d->sigset, SIGCHLD)) | |
3782 | e->need_process_child = true; | |
3783 | ||
91c70071 | 3784 | /* If there's already an event source pending for this priority we don't read another */ |
9da4cb2b LP |
3785 | if (d->current) |
3786 | return 0; | |
3787 | ||
fd38203a | 3788 | for (;;) { |
0eb2e0e3 | 3789 | struct signalfd_siginfo si; |
7057bd99 | 3790 | ssize_t n; |
92daebc0 | 3791 | sd_event_source *s = NULL; |
fd38203a | 3792 | |
9da4cb2b | 3793 | n = read(d->fd, &si, sizeof(si)); |
7057bd99 | 3794 | if (n < 0) { |
8add30a0 | 3795 | if (ERRNO_IS_TRANSIENT(errno)) |
efd3be9d | 3796 | return 0; |
fd38203a LP |
3797 | |
3798 | return -errno; | |
3799 | } | |
3800 | ||
7057bd99 | 3801 | if (_unlikely_(n != sizeof(si))) |
fd38203a LP |
3802 | return -EIO; |
3803 | ||
6eb7c172 | 3804 | assert(SIGNAL_VALID(si.ssi_signo)); |
7057bd99 | 3805 | |
92daebc0 LP |
3806 | if (e->signal_sources) |
3807 | s = e->signal_sources[si.ssi_signo]; | |
92daebc0 LP |
3808 | if (!s) |
3809 | continue; | |
9da4cb2b LP |
3810 | if (s->pending) |
3811 | continue; | |
fd38203a LP |
3812 | |
3813 | s->signal.siginfo = si; | |
9da4cb2b LP |
3814 | d->current = s; |
3815 | ||
fd38203a LP |
3816 | r = source_set_pending(s, true); |
3817 | if (r < 0) | |
3818 | return r; | |
efd3be9d YW |
3819 | if (r > 0 && *min_priority >= s->priority) { |
3820 | *min_priority = s->priority; | |
3821 | return 1; /* an event source with smaller priority is queued. */ | |
3822 | } | |
9da4cb2b | 3823 | |
efd3be9d | 3824 | return 0; |
fd38203a | 3825 | } |
fd38203a LP |
3826 | } |
3827 | ||
efd3be9d | 3828 | static int event_inotify_data_read(sd_event *e, struct inotify_data *d, uint32_t revents, int64_t threshold) { |
97ef5391 LP |
3829 | ssize_t n; |
3830 | ||
3831 | assert(e); | |
3832 | assert(d); | |
3833 | ||
3834 | assert_return(revents == EPOLLIN, -EIO); | |
3835 | ||
3836 | /* If there's already an event source pending for this priority, don't read another */ | |
3837 | if (d->n_pending > 0) | |
3838 | return 0; | |
3839 | ||
3840 | /* Is the read buffer non-empty? If so, let's not read more */ | |
3841 | if (d->buffer_filled > 0) | |
3842 | return 0; | |
3843 | ||
efd3be9d YW |
3844 | if (d->priority > threshold) |
3845 | return 0; | |
3846 | ||
97ef5391 LP |
3847 | n = read(d->fd, &d->buffer, sizeof(d->buffer)); |
3848 | if (n < 0) { | |
8add30a0 | 3849 | if (ERRNO_IS_TRANSIENT(errno)) |
97ef5391 LP |
3850 | return 0; |
3851 | ||
3852 | return -errno; | |
3853 | } | |
3854 | ||
3855 | assert(n > 0); | |
3856 | d->buffer_filled = (size_t) n; | |
0601b958 | 3857 | LIST_PREPEND(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
3858 | |
3859 | return 1; | |
3860 | } | |
3861 | ||
3862 | static void event_inotify_data_drop(sd_event *e, struct inotify_data *d, size_t sz) { | |
3863 | assert(e); | |
3864 | assert(d); | |
3865 | assert(sz <= d->buffer_filled); | |
3866 | ||
3867 | if (sz == 0) | |
3868 | return; | |
3869 | ||
3870 | /* Move the rest to the buffer to the front, in order to get things properly aligned again */ | |
3871 | memmove(d->buffer.raw, d->buffer.raw + sz, d->buffer_filled - sz); | |
3872 | d->buffer_filled -= sz; | |
3873 | ||
3874 | if (d->buffer_filled == 0) | |
0601b958 | 3875 | LIST_REMOVE(buffered, e->buffered_inotify_data_list, d); |
97ef5391 LP |
3876 | } |
3877 | ||
3878 | static int event_inotify_data_process(sd_event *e, struct inotify_data *d) { | |
3879 | int r; | |
3880 | ||
3881 | assert(e); | |
3882 | assert(d); | |
3883 | ||
3884 | /* If there's already an event source pending for this priority, don't read another */ | |
3885 | if (d->n_pending > 0) | |
3886 | return 0; | |
3887 | ||
3888 | while (d->buffer_filled > 0) { | |
3889 | size_t sz; | |
3890 | ||
3891 | /* Let's validate that the event structures are complete */ | |
3892 | if (d->buffer_filled < offsetof(struct inotify_event, name)) | |
3893 | return -EIO; | |
3894 | ||
3895 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
3896 | if (d->buffer_filled < sz) | |
3897 | return -EIO; | |
3898 | ||
3899 | if (d->buffer.ev.mask & IN_Q_OVERFLOW) { | |
3900 | struct inode_data *inode_data; | |
97ef5391 LP |
3901 | |
3902 | /* The queue overran, let's pass this event to all event sources connected to this inotify | |
3903 | * object */ | |
3904 | ||
03677889 | 3905 | HASHMAP_FOREACH(inode_data, d->inodes) |
97ef5391 LP |
3906 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { |
3907 | ||
b6d5481b | 3908 | if (event_source_is_offline(s)) |
97ef5391 LP |
3909 | continue; |
3910 | ||
3911 | r = source_set_pending(s, true); | |
3912 | if (r < 0) | |
3913 | return r; | |
3914 | } | |
97ef5391 LP |
3915 | } else { |
3916 | struct inode_data *inode_data; | |
97ef5391 LP |
3917 | |
3918 | /* Find the inode object for this watch descriptor. If IN_IGNORED is set we also remove it from | |
3919 | * our watch descriptor table. */ | |
3920 | if (d->buffer.ev.mask & IN_IGNORED) { | |
3921 | ||
3922 | inode_data = hashmap_remove(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3923 | if (!inode_data) { | |
3924 | event_inotify_data_drop(e, d, sz); | |
3925 | continue; | |
3926 | } | |
3927 | ||
3928 | /* The watch descriptor was removed by the kernel, let's drop it here too */ | |
3929 | inode_data->wd = -1; | |
3930 | } else { | |
3931 | inode_data = hashmap_get(d->wd, INT_TO_PTR(d->buffer.ev.wd)); | |
3932 | if (!inode_data) { | |
3933 | event_inotify_data_drop(e, d, sz); | |
3934 | continue; | |
3935 | } | |
3936 | } | |
3937 | ||
3938 | /* Trigger all event sources that are interested in these events. Also trigger all event | |
3939 | * sources if IN_IGNORED or IN_UNMOUNT is set. */ | |
3940 | LIST_FOREACH(inotify.by_inode_data, s, inode_data->event_sources) { | |
3941 | ||
b6d5481b | 3942 | if (event_source_is_offline(s)) |
97ef5391 LP |
3943 | continue; |
3944 | ||
3945 | if ((d->buffer.ev.mask & (IN_IGNORED|IN_UNMOUNT)) == 0 && | |
3946 | (s->inotify.mask & d->buffer.ev.mask & IN_ALL_EVENTS) == 0) | |
3947 | continue; | |
3948 | ||
3949 | r = source_set_pending(s, true); | |
3950 | if (r < 0) | |
3951 | return r; | |
3952 | } | |
3953 | } | |
3954 | ||
3955 | /* Something pending now? If so, let's finish, otherwise let's read more. */ | |
3956 | if (d->n_pending > 0) | |
3957 | return 1; | |
3958 | } | |
3959 | ||
3960 | return 0; | |
3961 | } | |
3962 | ||
3963 | static int process_inotify(sd_event *e) { | |
97ef5391 LP |
3964 | int r, done = 0; |
3965 | ||
3966 | assert(e); | |
3967 | ||
0601b958 | 3968 | LIST_FOREACH(buffered, d, e->buffered_inotify_data_list) { |
97ef5391 LP |
3969 | r = event_inotify_data_process(e, d); |
3970 | if (r < 0) | |
3971 | return r; | |
3972 | if (r > 0) | |
3973 | done ++; | |
3974 | } | |
3975 | ||
3976 | return done; | |
3977 | } | |
3978 | ||
158fe190 LP |
3979 | static int process_memory_pressure(sd_event_source *s, uint32_t revents) { |
3980 | assert(s); | |
3981 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
3982 | ||
3983 | if (s->pending) | |
3984 | s->memory_pressure.revents |= revents; | |
3985 | else | |
3986 | s->memory_pressure.revents = revents; | |
3987 | ||
3988 | return source_set_pending(s, true); | |
3989 | } | |
3990 | ||
3991 | static int source_memory_pressure_write(sd_event_source *s) { | |
3992 | ssize_t n; | |
3993 | int r; | |
3994 | ||
3995 | assert(s); | |
3996 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
3997 | ||
3998 | /* once we start writing, the buffer is locked, we allow no further changes. */ | |
3999 | s->memory_pressure.locked = true; | |
4000 | ||
4001 | if (s->memory_pressure.write_buffer_size > 0) { | |
4002 | n = write(s->memory_pressure.fd, s->memory_pressure.write_buffer, s->memory_pressure.write_buffer_size); | |
4003 | if (n < 0) { | |
9897f5dd LP |
4004 | if (!ERRNO_IS_TRANSIENT(errno)) { |
4005 | /* If kernel is built with CONFIG_PSI_DEFAULT_DISABLED it will expose PSI | |
4006 | * files, but then generates EOPNOSUPP on read() and write() (instead of on | |
4007 | * open()!). This sucks hard, since we can only detect this kind of failure | |
4008 | * so late. Let's make the best of it, and turn off the event source like we | |
4009 | * do for failed event source handlers. */ | |
4010 | ||
4011 | log_debug_errno(errno, "Writing memory pressure settings to kernel failed, disabling memory pressure event source: %m"); | |
4012 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); | |
4013 | return 0; | |
4014 | } | |
158fe190 LP |
4015 | |
4016 | n = 0; | |
4017 | } | |
4018 | } else | |
4019 | n = 0; | |
4020 | ||
4021 | assert(n >= 0); | |
4022 | ||
4023 | if ((size_t) n == s->memory_pressure.write_buffer_size) { | |
4024 | s->memory_pressure.write_buffer = mfree(s->memory_pressure.write_buffer); | |
4025 | ||
4026 | if (n > 0) { | |
4027 | s->memory_pressure.write_buffer_size = 0; | |
4028 | ||
4029 | /* Update epoll events mask, since we have now written everything and don't care for EPOLLOUT anymore */ | |
4030 | r = source_memory_pressure_register(s, s->enabled); | |
4031 | if (r < 0) | |
4032 | return r; | |
4033 | } | |
4034 | } else if (n > 0) { | |
4035 | _cleanup_free_ void *c = NULL; | |
4036 | ||
4037 | assert((size_t) n < s->memory_pressure.write_buffer_size); | |
4038 | ||
4039 | c = memdup((uint8_t*) s->memory_pressure.write_buffer + n, s->memory_pressure.write_buffer_size - n); | |
4040 | if (!c) | |
4041 | return -ENOMEM; | |
4042 | ||
4043 | free_and_replace(s->memory_pressure.write_buffer, c); | |
4044 | s->memory_pressure.write_buffer_size -= n; | |
4045 | return 1; | |
4046 | } | |
4047 | ||
4048 | return 0; | |
4049 | } | |
4050 | ||
4051 | static int source_memory_pressure_initiate_dispatch(sd_event_source *s) { | |
4052 | int r; | |
4053 | ||
4054 | assert(s); | |
4055 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
4056 | ||
4057 | r = source_memory_pressure_write(s); | |
4058 | if (r < 0) | |
4059 | return r; | |
4060 | if (r > 0) | |
4061 | return 1; /* if we wrote something, then don't continue with dispatching user dispatch | |
4062 | * function. Instead, shortcut it so that we wait for next EPOLLOUT immediately. */ | |
4063 | ||
4064 | /* No pending incoming IO? Then let's not continue further */ | |
4065 | if ((s->memory_pressure.revents & (EPOLLIN|EPOLLPRI)) == 0) { | |
4066 | ||
4067 | /* Treat IO errors on the notifier the same ways errors returned from a callback */ | |
4068 | if ((s->memory_pressure.revents & (EPOLLHUP|EPOLLERR|EPOLLRDHUP)) != 0) | |
4069 | return -EIO; | |
4070 | ||
4071 | return 1; /* leave dispatch, we already processed everything */ | |
4072 | } | |
4073 | ||
4074 | if (s->memory_pressure.revents & EPOLLIN) { | |
4075 | uint8_t pipe_buf[PIPE_BUF]; | |
4076 | ssize_t n; | |
4077 | ||
4078 | /* If the fd is readable, then flush out anything that might be queued */ | |
4079 | ||
4080 | n = read(s->memory_pressure.fd, pipe_buf, sizeof(pipe_buf)); | |
4081 | if (n < 0 && !ERRNO_IS_TRANSIENT(errno)) | |
4082 | return -errno; | |
4083 | } | |
4084 | ||
4085 | return 0; /* go on, dispatch to user callback */ | |
4086 | } | |
4087 | ||
fd38203a | 4088 | static int source_dispatch(sd_event_source *s) { |
8f5c235d | 4089 | EventSourceType saved_type; |
c8e9d15c | 4090 | sd_event *saved_event; |
fe8245eb | 4091 | int r = 0; |
fd38203a LP |
4092 | |
4093 | assert(s); | |
6203e07a | 4094 | assert(s->pending || s->type == SOURCE_EXIT); |
fd38203a | 4095 | |
b778cba4 LP |
4096 | /* Save the event source type, here, so that we still know it after the event callback which might |
4097 | * invalidate the event. */ | |
8f5c235d LP |
4098 | saved_type = s->type; |
4099 | ||
de02634c | 4100 | /* Similarly, store a reference to the event loop object, so that we can still access it after the |
b778cba4 | 4101 | * callback might have invalidated/disconnected the event source. */ |
c8e9d15c YW |
4102 | saved_event = s->event; |
4103 | PROTECT_EVENT(saved_event); | |
b778cba4 | 4104 | |
de02634c | 4105 | /* Check if we hit the ratelimit for this event source, and if so, let's disable it. */ |
b6d5481b LP |
4106 | assert(!s->ratelimited); |
4107 | if (!ratelimit_below(&s->rate_limit)) { | |
4108 | r = event_source_enter_ratelimited(s); | |
4109 | if (r < 0) | |
4110 | return r; | |
4111 | ||
4112 | return 1; | |
4113 | } | |
4114 | ||
945c2931 | 4115 | if (!IN_SET(s->type, SOURCE_DEFER, SOURCE_EXIT)) { |
da7e457c LP |
4116 | r = source_set_pending(s, false); |
4117 | if (r < 0) | |
4118 | return r; | |
4119 | } | |
fd38203a | 4120 | |
6e9feda3 LP |
4121 | if (s->type != SOURCE_POST) { |
4122 | sd_event_source *z; | |
6e9feda3 | 4123 | |
de02634c | 4124 | /* If we execute a non-post source, let's mark all post sources as pending. */ |
6e9feda3 | 4125 | |
90e74a66 | 4126 | SET_FOREACH(z, s->event->post_sources) { |
b6d5481b | 4127 | if (event_source_is_offline(z)) |
6e9feda3 LP |
4128 | continue; |
4129 | ||
4130 | r = source_set_pending(z, true); | |
4131 | if (r < 0) | |
4132 | return r; | |
4133 | } | |
4134 | } | |
4135 | ||
158fe190 LP |
4136 | if (s->type == SOURCE_MEMORY_PRESSURE) { |
4137 | r = source_memory_pressure_initiate_dispatch(s); | |
4138 | if (r == -EIO) /* handle EIO errors similar to callback errors */ | |
4139 | goto finish; | |
4140 | if (r < 0) | |
4141 | return r; | |
4142 | if (r > 0) /* already handled */ | |
4143 | return 1; | |
4144 | } | |
4145 | ||
baf76283 LP |
4146 | if (s->enabled == SD_EVENT_ONESHOT) { |
4147 | r = sd_event_source_set_enabled(s, SD_EVENT_OFF); | |
fd38203a LP |
4148 | if (r < 0) |
4149 | return r; | |
4150 | } | |
4151 | ||
12179984 | 4152 | s->dispatching = true; |
b7484e2a | 4153 | |
fd38203a LP |
4154 | switch (s->type) { |
4155 | ||
4156 | case SOURCE_IO: | |
4157 | r = s->io.callback(s, s->io.fd, s->io.revents, s->userdata); | |
4158 | break; | |
4159 | ||
6a0f1f6d | 4160 | case SOURCE_TIME_REALTIME: |
a8548816 | 4161 | case SOURCE_TIME_BOOTTIME: |
6a0f1f6d LP |
4162 | case SOURCE_TIME_MONOTONIC: |
4163 | case SOURCE_TIME_REALTIME_ALARM: | |
4164 | case SOURCE_TIME_BOOTTIME_ALARM: | |
fd38203a LP |
4165 | r = s->time.callback(s, s->time.next, s->userdata); |
4166 | break; | |
4167 | ||
4168 | case SOURCE_SIGNAL: | |
4169 | r = s->signal.callback(s, &s->signal.siginfo, s->userdata); | |
4170 | break; | |
4171 | ||
08cd1552 LP |
4172 | case SOURCE_CHILD: { |
4173 | bool zombie; | |
4174 | ||
945c2931 | 4175 | zombie = IN_SET(s->child.siginfo.si_code, CLD_EXITED, CLD_KILLED, CLD_DUMPED); |
08cd1552 | 4176 | |
fd38203a | 4177 | r = s->child.callback(s, &s->child.siginfo, s->userdata); |
08cd1552 LP |
4178 | |
4179 | /* Now, reap the PID for good. */ | |
f8f3f926 | 4180 | if (zombie) { |
cc59d290 | 4181 | (void) waitid(P_PID, s->child.pid, &s->child.siginfo, WNOHANG|WEXITED); |
f8f3f926 LP |
4182 | s->child.waited = true; |
4183 | } | |
08cd1552 | 4184 | |
fd38203a | 4185 | break; |
08cd1552 | 4186 | } |
fd38203a LP |
4187 | |
4188 | case SOURCE_DEFER: | |
4189 | r = s->defer.callback(s, s->userdata); | |
4190 | break; | |
da7e457c | 4191 | |
6e9feda3 LP |
4192 | case SOURCE_POST: |
4193 | r = s->post.callback(s, s->userdata); | |
4194 | break; | |
4195 | ||
6203e07a LP |
4196 | case SOURCE_EXIT: |
4197 | r = s->exit.callback(s, s->userdata); | |
da7e457c | 4198 | break; |
9d3e3aa5 | 4199 | |
97ef5391 LP |
4200 | case SOURCE_INOTIFY: { |
4201 | struct sd_event *e = s->event; | |
4202 | struct inotify_data *d; | |
4203 | size_t sz; | |
4204 | ||
4205 | assert(s->inotify.inode_data); | |
4206 | assert_se(d = s->inotify.inode_data->inotify_data); | |
4207 | ||
4208 | assert(d->buffer_filled >= offsetof(struct inotify_event, name)); | |
4209 | sz = offsetof(struct inotify_event, name) + d->buffer.ev.len; | |
4210 | assert(d->buffer_filled >= sz); | |
4211 | ||
53baf2ef LP |
4212 | /* If the inotify callback destroys the event source then this likely means we don't need to |
4213 | * watch the inode anymore, and thus also won't need the inotify object anymore. But if we'd | |
4214 | * free it immediately, then we couldn't drop the event from the inotify event queue without | |
4215 | * memory corruption anymore, as below. Hence, let's not free it immediately, but mark it | |
4216 | * "busy" with a counter (which will ensure it's not GC'ed away prematurely). Let's then | |
4217 | * explicitly GC it after we are done dropping the inotify event from the buffer. */ | |
4218 | d->n_busy++; | |
97ef5391 | 4219 | r = s->inotify.callback(s, &d->buffer.ev, s->userdata); |
53baf2ef | 4220 | d->n_busy--; |
97ef5391 | 4221 | |
53baf2ef LP |
4222 | /* When no event is pending anymore on this inotify object, then let's drop the event from |
4223 | * the inotify event queue buffer. */ | |
97ef5391 LP |
4224 | if (d->n_pending == 0) |
4225 | event_inotify_data_drop(e, d, sz); | |
4226 | ||
53baf2ef LP |
4227 | /* Now we don't want to access 'd' anymore, it's OK to GC now. */ |
4228 | event_gc_inotify_data(e, d); | |
97ef5391 LP |
4229 | break; |
4230 | } | |
4231 | ||
158fe190 LP |
4232 | case SOURCE_MEMORY_PRESSURE: |
4233 | r = s->memory_pressure.callback(s, s->userdata); | |
4234 | break; | |
4235 | ||
9d3e3aa5 | 4236 | case SOURCE_WATCHDOG: |
a71fe8b8 | 4237 | case _SOURCE_EVENT_SOURCE_TYPE_MAX: |
9f2a50a3 | 4238 | case _SOURCE_EVENT_SOURCE_TYPE_INVALID: |
04499a70 | 4239 | assert_not_reached(); |
fd38203a LP |
4240 | } |
4241 | ||
12179984 LP |
4242 | s->dispatching = false; |
4243 | ||
158fe190 | 4244 | finish: |
b778cba4 LP |
4245 | if (r < 0) { |
4246 | log_debug_errno(r, "Event source %s (type %s) returned error, %s: %m", | |
4247 | strna(s->description), | |
4248 | event_source_type_to_string(saved_type), | |
4249 | s->exit_on_failure ? "exiting" : "disabling"); | |
4250 | ||
4251 | if (s->exit_on_failure) | |
4252 | (void) sd_event_exit(saved_event, r); | |
4253 | } | |
12179984 LP |
4254 | |
4255 | if (s->n_ref == 0) | |
4256 | source_free(s); | |
4257 | else if (r < 0) | |
c3c50474 | 4258 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
b7484e2a | 4259 | |
6203e07a | 4260 | return 1; |
fd38203a LP |
4261 | } |
4262 | ||
4263 | static int event_prepare(sd_event *e) { | |
4264 | int r; | |
4265 | ||
4266 | assert(e); | |
4267 | ||
4268 | for (;;) { | |
4269 | sd_event_source *s; | |
4270 | ||
4271 | s = prioq_peek(e->prepare); | |
b6d5481b | 4272 | if (!s || s->prepare_iteration == e->iteration || event_source_is_offline(s)) |
fd38203a LP |
4273 | break; |
4274 | ||
4275 | s->prepare_iteration = e->iteration; | |
8656f4a6 | 4276 | prioq_reshuffle(e->prepare, s, &s->prepare_index); |
fd38203a LP |
4277 | |
4278 | assert(s->prepare); | |
12179984 | 4279 | s->dispatching = true; |
fd38203a | 4280 | r = s->prepare(s, s->userdata); |
12179984 LP |
4281 | s->dispatching = false; |
4282 | ||
b778cba4 LP |
4283 | if (r < 0) { |
4284 | log_debug_errno(r, "Prepare callback of event source %s (type %s) returned error, %s: %m", | |
4285 | strna(s->description), | |
4286 | event_source_type_to_string(s->type), | |
4287 | s->exit_on_failure ? "exiting" : "disabling"); | |
4288 | ||
4289 | if (s->exit_on_failure) | |
4290 | (void) sd_event_exit(e, r); | |
4291 | } | |
fd38203a | 4292 | |
12179984 LP |
4293 | if (s->n_ref == 0) |
4294 | source_free(s); | |
4295 | else if (r < 0) | |
c3c50474 | 4296 | assert_se(sd_event_source_set_enabled(s, SD_EVENT_OFF) >= 0); |
fd38203a LP |
4297 | } |
4298 | ||
4299 | return 0; | |
4300 | } | |
4301 | ||
6203e07a | 4302 | static int dispatch_exit(sd_event *e) { |
da7e457c LP |
4303 | sd_event_source *p; |
4304 | int r; | |
4305 | ||
4306 | assert(e); | |
4307 | ||
6203e07a | 4308 | p = prioq_peek(e->exit); |
19947509 ZJS |
4309 | assert(!p || p->type == SOURCE_EXIT); |
4310 | ||
b6d5481b | 4311 | if (!p || event_source_is_offline(p)) { |
da7e457c LP |
4312 | e->state = SD_EVENT_FINISHED; |
4313 | return 0; | |
4314 | } | |
4315 | ||
c8e9d15c | 4316 | PROTECT_EVENT(e); |
da7e457c | 4317 | e->iteration++; |
6203e07a | 4318 | e->state = SD_EVENT_EXITING; |
da7e457c | 4319 | r = source_dispatch(p); |
2b0c9ef7 | 4320 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
4321 | return r; |
4322 | } | |
4323 | ||
c2ba3ad6 LP |
4324 | static sd_event_source* event_next_pending(sd_event *e) { |
4325 | sd_event_source *p; | |
4326 | ||
da7e457c LP |
4327 | assert(e); |
4328 | ||
c2ba3ad6 LP |
4329 | p = prioq_peek(e->pending); |
4330 | if (!p) | |
4331 | return NULL; | |
4332 | ||
b6d5481b | 4333 | if (event_source_is_offline(p)) |
c2ba3ad6 LP |
4334 | return NULL; |
4335 | ||
4336 | return p; | |
4337 | } | |
4338 | ||
cde93897 LP |
4339 | static int arm_watchdog(sd_event *e) { |
4340 | struct itimerspec its = {}; | |
4341 | usec_t t; | |
cde93897 LP |
4342 | |
4343 | assert(e); | |
4344 | assert(e->watchdog_fd >= 0); | |
4345 | ||
4346 | t = sleep_between(e, | |
a595fb5c YW |
4347 | usec_add(e->watchdog_last, (e->watchdog_period / 2)), |
4348 | usec_add(e->watchdog_last, (e->watchdog_period * 3 / 4))); | |
cde93897 LP |
4349 | |
4350 | timespec_store(&its.it_value, t); | |
4351 | ||
75145780 LP |
4352 | /* Make sure we never set the watchdog to 0, which tells the |
4353 | * kernel to disable it. */ | |
4354 | if (its.it_value.tv_sec == 0 && its.it_value.tv_nsec == 0) | |
4355 | its.it_value.tv_nsec = 1; | |
4356 | ||
7c248223 | 4357 | return RET_NERRNO(timerfd_settime(e->watchdog_fd, TFD_TIMER_ABSTIME, &its, NULL)); |
cde93897 LP |
4358 | } |
4359 | ||
4360 | static int process_watchdog(sd_event *e) { | |
4361 | assert(e); | |
4362 | ||
4363 | if (!e->watchdog) | |
4364 | return 0; | |
4365 | ||
4366 | /* Don't notify watchdog too often */ | |
4367 | if (e->watchdog_last + e->watchdog_period / 4 > e->timestamp.monotonic) | |
4368 | return 0; | |
4369 | ||
4370 | sd_notify(false, "WATCHDOG=1"); | |
4371 | e->watchdog_last = e->timestamp.monotonic; | |
4372 | ||
4373 | return arm_watchdog(e); | |
4374 | } | |
4375 | ||
97ef5391 LP |
4376 | static void event_close_inode_data_fds(sd_event *e) { |
4377 | struct inode_data *d; | |
4378 | ||
4379 | assert(e); | |
4380 | ||
4381 | /* Close the fds pointing to the inodes to watch now. We need to close them as they might otherwise pin | |
4382 | * filesystems. But we can't close them right-away as we need them as long as the user still wants to make | |
365c2885 | 4383 | * adjustments to the event source, such as changing the priority (which requires us to remove and re-add a watch |
97ef5391 LP |
4384 | * for the inode). Hence, let's close them when entering the first iteration after they were added, as a |
4385 | * compromise. */ | |
4386 | ||
ed828563 | 4387 | while ((d = e->inode_data_to_close_list)) { |
97ef5391 LP |
4388 | assert(d->fd >= 0); |
4389 | d->fd = safe_close(d->fd); | |
4390 | ||
ed828563 | 4391 | LIST_REMOVE(to_close, e->inode_data_to_close_list, d); |
97ef5391 LP |
4392 | } |
4393 | } | |
4394 | ||
158fe190 LP |
4395 | static int event_memory_pressure_write_list(sd_event *e) { |
4396 | int r; | |
4397 | ||
4398 | assert(e); | |
4399 | ||
4400 | for (;;) { | |
4401 | sd_event_source *s; | |
4402 | ||
4403 | s = LIST_POP(memory_pressure.write_list, e->memory_pressure_write_list); | |
4404 | if (!s) | |
4405 | break; | |
4406 | ||
4407 | assert(s->type == SOURCE_MEMORY_PRESSURE); | |
4408 | assert(s->memory_pressure.write_buffer_size > 0); | |
4409 | s->memory_pressure.in_write_list = false; | |
4410 | ||
4411 | r = source_memory_pressure_write(s); | |
4412 | if (r < 0) | |
4413 | return r; | |
4414 | } | |
4415 | ||
4416 | return 0; | |
4417 | } | |
4418 | ||
c45a5a74 TG |
4419 | _public_ int sd_event_prepare(sd_event *e) { |
4420 | int r; | |
fd38203a | 4421 | |
da7e457c | 4422 | assert_return(e, -EINVAL); |
b937d761 | 4423 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
4424 | assert_return(!event_pid_changed(e), -ECHILD); |
4425 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 4426 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 4427 | |
e5446015 LP |
4428 | /* Let's check that if we are a default event loop we are executed in the correct thread. We only do |
4429 | * this check here once, since gettid() is typically not cached, and thus want to minimize | |
4430 | * syscalls */ | |
4431 | assert_return(!e->default_event_ptr || e->tid == gettid(), -EREMOTEIO); | |
4432 | ||
f814c871 | 4433 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
c8e9d15c | 4434 | PROTECT_EVENT(e); |
f814c871 | 4435 | |
6203e07a | 4436 | if (e->exit_requested) |
c45a5a74 | 4437 | goto pending; |
fd38203a LP |
4438 | |
4439 | e->iteration++; | |
4440 | ||
0be6c2f6 | 4441 | e->state = SD_EVENT_PREPARING; |
fd38203a | 4442 | r = event_prepare(e); |
0be6c2f6 | 4443 | e->state = SD_EVENT_INITIAL; |
fd38203a | 4444 | if (r < 0) |
c45a5a74 | 4445 | return r; |
fd38203a | 4446 | |
158fe190 LP |
4447 | r = event_memory_pressure_write_list(e); |
4448 | if (r < 0) | |
4449 | return r; | |
4450 | ||
6a0f1f6d LP |
4451 | r = event_arm_timer(e, &e->realtime); |
4452 | if (r < 0) | |
c45a5a74 | 4453 | return r; |
6a0f1f6d | 4454 | |
a8548816 TG |
4455 | r = event_arm_timer(e, &e->boottime); |
4456 | if (r < 0) | |
c45a5a74 | 4457 | return r; |
a8548816 | 4458 | |
6a0f1f6d LP |
4459 | r = event_arm_timer(e, &e->monotonic); |
4460 | if (r < 0) | |
c45a5a74 | 4461 | return r; |
6a0f1f6d LP |
4462 | |
4463 | r = event_arm_timer(e, &e->realtime_alarm); | |
1b5995b0 | 4464 | if (r < 0) |
c45a5a74 | 4465 | return r; |
fd38203a | 4466 | |
6a0f1f6d | 4467 | r = event_arm_timer(e, &e->boottime_alarm); |
1b5995b0 | 4468 | if (r < 0) |
c45a5a74 | 4469 | return r; |
fd38203a | 4470 | |
97ef5391 LP |
4471 | event_close_inode_data_fds(e); |
4472 | ||
0601b958 | 4473 | if (event_next_pending(e) || e->need_process_child || e->buffered_inotify_data_list) |
c45a5a74 TG |
4474 | goto pending; |
4475 | ||
2b0c9ef7 | 4476 | e->state = SD_EVENT_ARMED; |
c45a5a74 TG |
4477 | |
4478 | return 0; | |
4479 | ||
4480 | pending: | |
2b0c9ef7 | 4481 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
4482 | r = sd_event_wait(e, 0); |
4483 | if (r == 0) | |
2b0c9ef7 | 4484 | e->state = SD_EVENT_ARMED; |
6d148a84 TG |
4485 | |
4486 | return r; | |
c45a5a74 TG |
4487 | } |
4488 | ||
798445ab LP |
4489 | static int epoll_wait_usec( |
4490 | int fd, | |
4491 | struct epoll_event *events, | |
4492 | int maxevents, | |
4493 | usec_t timeout) { | |
4494 | ||
7c248223 | 4495 | int msec; |
0c14c45e LP |
4496 | /* A wrapper that uses epoll_pwait2() if available, and falls back to epoll_wait() if not. */ |
4497 | ||
4498 | #if HAVE_EPOLL_PWAIT2 | |
39f756d3 | 4499 | static bool epoll_pwait2_absent = false; |
52bb308c | 4500 | int r; |
798445ab | 4501 | |
0c14c45e LP |
4502 | /* epoll_pwait2() was added to Linux 5.11 (2021-02-14) and to glibc in 2.35 (2022-02-03). In contrast |
4503 | * to other syscalls we don't bother with our own fallback syscall wrappers on old libcs, since this | |
4504 | * is not that obvious to implement given the libc and kernel definitions differ in the last | |
4505 | * argument. Moreover, the only reason to use it is the more accurate time-outs (which is not a | |
4506 | * biggie), let's hence rely on glibc's definitions, and fallback to epoll_pwait() when that's | |
4507 | * missing. */ | |
798445ab LP |
4508 | |
4509 | if (!epoll_pwait2_absent && timeout != USEC_INFINITY) { | |
798445ab LP |
4510 | r = epoll_pwait2(fd, |
4511 | events, | |
4512 | maxevents, | |
52bb308c | 4513 | TIMESPEC_STORE(timeout), |
798445ab LP |
4514 | NULL); |
4515 | if (r >= 0) | |
4516 | return r; | |
7cb45dbf | 4517 | if (!ERRNO_IS_NOT_SUPPORTED(errno) && !ERRNO_IS_PRIVILEGE(errno)) |
798445ab LP |
4518 | return -errno; /* Only fallback to old epoll_wait() if the syscall is masked or not |
4519 | * supported. */ | |
4520 | ||
4521 | epoll_pwait2_absent = true; | |
4522 | } | |
39f756d3 | 4523 | #endif |
798445ab LP |
4524 | |
4525 | if (timeout == USEC_INFINITY) | |
4526 | msec = -1; | |
4527 | else { | |
4528 | usec_t k; | |
4529 | ||
4530 | k = DIV_ROUND_UP(timeout, USEC_PER_MSEC); | |
4531 | if (k >= INT_MAX) | |
4532 | msec = INT_MAX; /* Saturate */ | |
4533 | else | |
4534 | msec = (int) k; | |
4535 | } | |
4536 | ||
7c248223 | 4537 | return RET_NERRNO(epoll_wait(fd, events, maxevents, msec)); |
798445ab LP |
4538 | } |
4539 | ||
efd3be9d | 4540 | static int process_epoll(sd_event *e, usec_t timeout, int64_t threshold, int64_t *ret_min_priority) { |
319a4f4b | 4541 | size_t n_event_queue, m, n_event_max; |
efd3be9d YW |
4542 | int64_t min_priority = threshold; |
4543 | bool something_new = false; | |
798445ab | 4544 | int r; |
c45a5a74 | 4545 | |
efd3be9d YW |
4546 | assert(e); |
4547 | assert(ret_min_priority); | |
6a0f1f6d | 4548 | |
8b9708d1 | 4549 | n_event_queue = MAX(e->n_sources, 1u); |
319a4f4b | 4550 | if (!GREEDY_REALLOC(e->event_queue, n_event_queue)) |
5cddd924 | 4551 | return -ENOMEM; |
fd38203a | 4552 | |
319a4f4b LP |
4553 | n_event_max = MALLOC_ELEMENTSOF(e->event_queue); |
4554 | ||
97ef5391 | 4555 | /* If we still have inotify data buffered, then query the other fds, but don't wait on it */ |
0601b958 | 4556 | if (e->buffered_inotify_data_list) |
798445ab | 4557 | timeout = 0; |
97ef5391 | 4558 | |
8b9708d1 | 4559 | for (;;) { |
319a4f4b LP |
4560 | r = epoll_wait_usec( |
4561 | e->epoll_fd, | |
4562 | e->event_queue, | |
4563 | n_event_max, | |
4564 | timeout); | |
798445ab | 4565 | if (r < 0) |
efd3be9d | 4566 | return r; |
c45a5a74 | 4567 | |
8b9708d1 YW |
4568 | m = (size_t) r; |
4569 | ||
319a4f4b | 4570 | if (m < n_event_max) |
8b9708d1 YW |
4571 | break; |
4572 | ||
319a4f4b | 4573 | if (n_event_max >= n_event_queue * 10) |
8b9708d1 YW |
4574 | break; |
4575 | ||
319a4f4b | 4576 | if (!GREEDY_REALLOC(e->event_queue, n_event_max + n_event_queue)) |
8b9708d1 YW |
4577 | return -ENOMEM; |
4578 | ||
319a4f4b | 4579 | n_event_max = MALLOC_ELEMENTSOF(e->event_queue); |
798445ab | 4580 | timeout = 0; |
da7e457c | 4581 | } |
fd38203a | 4582 | |
efd3be9d YW |
4583 | /* Set timestamp only when this is called first time. */ |
4584 | if (threshold == INT64_MAX) | |
4585 | triple_timestamp_get(&e->timestamp); | |
fd38203a | 4586 | |
8b9708d1 | 4587 | for (size_t i = 0; i < m; i++) { |
fd38203a | 4588 | |
5cddd924 LP |
4589 | if (e->event_queue[i].data.ptr == INT_TO_PTR(SOURCE_WATCHDOG)) |
4590 | r = flush_timer(e, e->watchdog_fd, e->event_queue[i].events, NULL); | |
9da4cb2b | 4591 | else { |
5cddd924 | 4592 | WakeupType *t = e->event_queue[i].data.ptr; |
9da4cb2b LP |
4593 | |
4594 | switch (*t) { | |
4595 | ||
f8f3f926 | 4596 | case WAKEUP_EVENT_SOURCE: { |
5cddd924 | 4597 | sd_event_source *s = e->event_queue[i].data.ptr; |
f8f3f926 LP |
4598 | |
4599 | assert(s); | |
4600 | ||
efd3be9d YW |
4601 | if (s->priority > threshold) |
4602 | continue; | |
4603 | ||
4604 | min_priority = MIN(min_priority, s->priority); | |
4605 | ||
f8f3f926 LP |
4606 | switch (s->type) { |
4607 | ||
4608 | case SOURCE_IO: | |
5cddd924 | 4609 | r = process_io(e, s, e->event_queue[i].events); |
f8f3f926 LP |
4610 | break; |
4611 | ||
4612 | case SOURCE_CHILD: | |
5cddd924 | 4613 | r = process_pidfd(e, s, e->event_queue[i].events); |
f8f3f926 LP |
4614 | break; |
4615 | ||
158fe190 LP |
4616 | case SOURCE_MEMORY_PRESSURE: |
4617 | r = process_memory_pressure(s, e->event_queue[i].events); | |
4618 | break; | |
4619 | ||
f8f3f926 | 4620 | default: |
04499a70 | 4621 | assert_not_reached(); |
f8f3f926 LP |
4622 | } |
4623 | ||
9da4cb2b | 4624 | break; |
f8f3f926 | 4625 | } |
fd38203a | 4626 | |
9da4cb2b | 4627 | case WAKEUP_CLOCK_DATA: { |
5cddd924 | 4628 | struct clock_data *d = e->event_queue[i].data.ptr; |
f8f3f926 LP |
4629 | |
4630 | assert(d); | |
4631 | ||
5cddd924 | 4632 | r = flush_timer(e, d->fd, e->event_queue[i].events, &d->next); |
9da4cb2b LP |
4633 | break; |
4634 | } | |
4635 | ||
4636 | case WAKEUP_SIGNAL_DATA: | |
efd3be9d | 4637 | r = process_signal(e, e->event_queue[i].data.ptr, e->event_queue[i].events, &min_priority); |
9da4cb2b LP |
4638 | break; |
4639 | ||
97ef5391 | 4640 | case WAKEUP_INOTIFY_DATA: |
efd3be9d | 4641 | r = event_inotify_data_read(e, e->event_queue[i].data.ptr, e->event_queue[i].events, threshold); |
97ef5391 LP |
4642 | break; |
4643 | ||
9da4cb2b | 4644 | default: |
04499a70 | 4645 | assert_not_reached(); |
9da4cb2b LP |
4646 | } |
4647 | } | |
efd3be9d YW |
4648 | if (r < 0) |
4649 | return r; | |
4650 | if (r > 0) | |
4651 | something_new = true; | |
4652 | } | |
4653 | ||
4654 | *ret_min_priority = min_priority; | |
4655 | return something_new; | |
4656 | } | |
4657 | ||
4658 | _public_ int sd_event_wait(sd_event *e, uint64_t timeout) { | |
4659 | int r; | |
4660 | ||
4661 | assert_return(e, -EINVAL); | |
4662 | assert_return(e = event_resolve(e), -ENOPKG); | |
4663 | assert_return(!event_pid_changed(e), -ECHILD); | |
4664 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
4665 | assert_return(e->state == SD_EVENT_ARMED, -EBUSY); | |
4666 | ||
4667 | if (e->exit_requested) { | |
4668 | e->state = SD_EVENT_PENDING; | |
4669 | return 1; | |
4670 | } | |
4671 | ||
4672 | for (int64_t threshold = INT64_MAX; ; threshold--) { | |
4673 | int64_t epoll_min_priority, child_min_priority; | |
4674 | ||
4675 | /* There may be a possibility that new epoll (especially IO) and child events are | |
4676 | * triggered just after process_epoll() call but before process_child(), and the new IO | |
4677 | * events may have higher priority than the child events. To salvage these events, | |
4678 | * let's call epoll_wait() again, but accepts only events with higher priority than the | |
4679 | * previous. See issue https://github.com/systemd/systemd/issues/18190 and comments | |
4680 | * https://github.com/systemd/systemd/pull/18750#issuecomment-785801085 | |
4681 | * https://github.com/systemd/systemd/pull/18922#issuecomment-792825226 */ | |
4682 | ||
4683 | r = process_epoll(e, timeout, threshold, &epoll_min_priority); | |
4684 | if (r == -EINTR) { | |
4685 | e->state = SD_EVENT_PENDING; | |
4686 | return 1; | |
4687 | } | |
4688 | if (r < 0) | |
4689 | goto finish; | |
4690 | if (r == 0 && threshold < INT64_MAX) | |
4691 | /* No new epoll event. */ | |
4692 | break; | |
4693 | ||
4694 | r = process_child(e, threshold, &child_min_priority); | |
fd38203a | 4695 | if (r < 0) |
da7e457c | 4696 | goto finish; |
efd3be9d YW |
4697 | if (r == 0) |
4698 | /* No new child event. */ | |
4699 | break; | |
4700 | ||
4701 | threshold = MIN(epoll_min_priority, child_min_priority); | |
4702 | if (threshold == INT64_MIN) | |
4703 | break; | |
4704 | ||
4705 | timeout = 0; | |
fd38203a LP |
4706 | } |
4707 | ||
cde93897 LP |
4708 | r = process_watchdog(e); |
4709 | if (r < 0) | |
4710 | goto finish; | |
4711 | ||
fd69f224 | 4712 | r = process_inotify(e); |
6a0f1f6d LP |
4713 | if (r < 0) |
4714 | goto finish; | |
4715 | ||
fd69f224 | 4716 | r = process_timer(e, e->timestamp.realtime, &e->realtime); |
a8548816 TG |
4717 | if (r < 0) |
4718 | goto finish; | |
4719 | ||
fd69f224 | 4720 | r = process_timer(e, e->timestamp.boottime, &e->boottime); |
6a0f1f6d LP |
4721 | if (r < 0) |
4722 | goto finish; | |
4723 | ||
4724 | r = process_timer(e, e->timestamp.realtime, &e->realtime_alarm); | |
fd38203a | 4725 | if (r < 0) |
da7e457c | 4726 | goto finish; |
fd38203a | 4727 | |
e475d10c | 4728 | r = process_timer(e, e->timestamp.boottime, &e->boottime_alarm); |
fd38203a | 4729 | if (r < 0) |
da7e457c | 4730 | goto finish; |
fd38203a | 4731 | |
fd69f224 | 4732 | r = process_timer(e, e->timestamp.monotonic, &e->monotonic); |
97ef5391 LP |
4733 | if (r < 0) |
4734 | goto finish; | |
fd69f224 MS |
4735 | else if (r == 1) { |
4736 | /* Ratelimit expiry callback was called. Let's postpone processing pending sources and | |
4737 | * put loop in the initial state in order to evaluate (in the next iteration) also sources | |
4738 | * there were potentially re-enabled by the callback. | |
4739 | * | |
4740 | * Wondering why we treat only this invocation of process_timer() differently? Once event | |
4741 | * source is ratelimited we essentially transform it into CLOCK_MONOTONIC timer hence | |
4742 | * ratelimit expiry callback is never called for any other timer type. */ | |
4743 | r = 0; | |
4744 | goto finish; | |
4745 | } | |
97ef5391 | 4746 | |
c45a5a74 TG |
4747 | if (event_next_pending(e)) { |
4748 | e->state = SD_EVENT_PENDING; | |
c45a5a74 | 4749 | return 1; |
da7e457c LP |
4750 | } |
4751 | ||
c45a5a74 | 4752 | r = 0; |
fd38203a | 4753 | |
da7e457c | 4754 | finish: |
2b0c9ef7 | 4755 | e->state = SD_EVENT_INITIAL; |
da7e457c LP |
4756 | |
4757 | return r; | |
fd38203a LP |
4758 | } |
4759 | ||
c45a5a74 TG |
4760 | _public_ int sd_event_dispatch(sd_event *e) { |
4761 | sd_event_source *p; | |
4762 | int r; | |
4763 | ||
4764 | assert_return(e, -EINVAL); | |
b937d761 | 4765 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
4766 | assert_return(!event_pid_changed(e), -ECHILD); |
4767 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
4768 | assert_return(e->state == SD_EVENT_PENDING, -EBUSY); | |
4769 | ||
4770 | if (e->exit_requested) | |
4771 | return dispatch_exit(e); | |
4772 | ||
4773 | p = event_next_pending(e); | |
4774 | if (p) { | |
c8e9d15c | 4775 | PROTECT_EVENT(e); |
c45a5a74 TG |
4776 | |
4777 | e->state = SD_EVENT_RUNNING; | |
4778 | r = source_dispatch(p); | |
2b0c9ef7 | 4779 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
4780 | return r; |
4781 | } | |
4782 | ||
2b0c9ef7 | 4783 | e->state = SD_EVENT_INITIAL; |
c45a5a74 TG |
4784 | |
4785 | return 1; | |
4786 | } | |
4787 | ||
34b87517 | 4788 | static void event_log_delays(sd_event *e) { |
442ac269 YW |
4789 | char b[ELEMENTSOF(e->delays) * DECIMAL_STR_MAX(unsigned) + 1], *p; |
4790 | size_t l, i; | |
34b87517 | 4791 | |
442ac269 YW |
4792 | p = b; |
4793 | l = sizeof(b); | |
4794 | for (i = 0; i < ELEMENTSOF(e->delays); i++) { | |
4795 | l = strpcpyf(&p, l, "%u ", e->delays[i]); | |
34b87517 VC |
4796 | e->delays[i] = 0; |
4797 | } | |
442ac269 | 4798 | log_debug("Event loop iterations: %s", b); |
34b87517 VC |
4799 | } |
4800 | ||
c45a5a74 TG |
4801 | _public_ int sd_event_run(sd_event *e, uint64_t timeout) { |
4802 | int r; | |
4803 | ||
4804 | assert_return(e, -EINVAL); | |
b937d761 | 4805 | assert_return(e = event_resolve(e), -ENOPKG); |
c45a5a74 TG |
4806 | assert_return(!event_pid_changed(e), -ECHILD); |
4807 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); | |
2b0c9ef7 | 4808 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
c45a5a74 | 4809 | |
e6a7bee5 | 4810 | if (e->profile_delays && e->last_run_usec != 0) { |
34b87517 VC |
4811 | usec_t this_run; |
4812 | unsigned l; | |
4813 | ||
4814 | this_run = now(CLOCK_MONOTONIC); | |
4815 | ||
58c34be8 | 4816 | l = log2u64(this_run - e->last_run_usec); |
cb9d621e | 4817 | assert(l < ELEMENTSOF(e->delays)); |
34b87517 VC |
4818 | e->delays[l]++; |
4819 | ||
e6a7bee5 | 4820 | if (this_run - e->last_log_usec >= 5*USEC_PER_SEC) { |
34b87517 | 4821 | event_log_delays(e); |
e6a7bee5 | 4822 | e->last_log_usec = this_run; |
34b87517 VC |
4823 | } |
4824 | } | |
4825 | ||
f814c871 | 4826 | /* Make sure that none of the preparation callbacks ends up freeing the event source under our feet */ |
c8e9d15c | 4827 | PROTECT_EVENT(e); |
f814c871 | 4828 | |
c45a5a74 | 4829 | r = sd_event_prepare(e); |
53bac4e0 LP |
4830 | if (r == 0) |
4831 | /* There was nothing? Then wait... */ | |
4832 | r = sd_event_wait(e, timeout); | |
c45a5a74 | 4833 | |
34b87517 | 4834 | if (e->profile_delays) |
e6a7bee5 | 4835 | e->last_run_usec = now(CLOCK_MONOTONIC); |
34b87517 | 4836 | |
02d30981 | 4837 | if (r > 0) { |
53bac4e0 | 4838 | /* There's something now, then let's dispatch it */ |
02d30981 TG |
4839 | r = sd_event_dispatch(e); |
4840 | if (r < 0) | |
4841 | return r; | |
53bac4e0 LP |
4842 | |
4843 | return 1; | |
4844 | } | |
4845 | ||
4846 | return r; | |
c45a5a74 TG |
4847 | } |
4848 | ||
f7262a9f | 4849 | _public_ int sd_event_loop(sd_event *e) { |
fd38203a LP |
4850 | int r; |
4851 | ||
da7e457c | 4852 | assert_return(e, -EINVAL); |
b937d761 | 4853 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c | 4854 | assert_return(!event_pid_changed(e), -ECHILD); |
2b0c9ef7 | 4855 | assert_return(e->state == SD_EVENT_INITIAL, -EBUSY); |
da7e457c | 4856 | |
c8e9d15c | 4857 | PROTECT_EVENT(e); |
fd38203a | 4858 | |
da7e457c | 4859 | while (e->state != SD_EVENT_FINISHED) { |
f5fbe71d | 4860 | r = sd_event_run(e, UINT64_MAX); |
fd38203a | 4861 | if (r < 0) |
30dd293c | 4862 | return r; |
fd38203a LP |
4863 | } |
4864 | ||
30dd293c | 4865 | return e->exit_code; |
fd38203a LP |
4866 | } |
4867 | ||
9b364545 | 4868 | _public_ int sd_event_get_fd(sd_event *e) { |
9b364545 | 4869 | assert_return(e, -EINVAL); |
b937d761 | 4870 | assert_return(e = event_resolve(e), -ENOPKG); |
9b364545 TG |
4871 | assert_return(!event_pid_changed(e), -ECHILD); |
4872 | ||
4873 | return e->epoll_fd; | |
4874 | } | |
4875 | ||
f7262a9f | 4876 | _public_ int sd_event_get_state(sd_event *e) { |
da7e457c | 4877 | assert_return(e, -EINVAL); |
b937d761 | 4878 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
4879 | assert_return(!event_pid_changed(e), -ECHILD); |
4880 | ||
4881 | return e->state; | |
4882 | } | |
4883 | ||
6203e07a | 4884 | _public_ int sd_event_get_exit_code(sd_event *e, int *code) { |
da7e457c | 4885 | assert_return(e, -EINVAL); |
b937d761 | 4886 | assert_return(e = event_resolve(e), -ENOPKG); |
6203e07a | 4887 | assert_return(code, -EINVAL); |
da7e457c | 4888 | assert_return(!event_pid_changed(e), -ECHILD); |
fd38203a | 4889 | |
6203e07a LP |
4890 | if (!e->exit_requested) |
4891 | return -ENODATA; | |
4892 | ||
4893 | *code = e->exit_code; | |
4894 | return 0; | |
fd38203a LP |
4895 | } |
4896 | ||
6203e07a | 4897 | _public_ int sd_event_exit(sd_event *e, int code) { |
da7e457c | 4898 | assert_return(e, -EINVAL); |
b937d761 | 4899 | assert_return(e = event_resolve(e), -ENOPKG); |
da7e457c LP |
4900 | assert_return(e->state != SD_EVENT_FINISHED, -ESTALE); |
4901 | assert_return(!event_pid_changed(e), -ECHILD); | |
fd38203a | 4902 | |
6203e07a LP |
4903 | e->exit_requested = true; |
4904 | e->exit_code = code; | |
4905 | ||
fd38203a LP |
4906 | return 0; |
4907 | } | |
46e8c825 | 4908 | |
6a0f1f6d | 4909 | _public_ int sd_event_now(sd_event *e, clockid_t clock, uint64_t *usec) { |
46e8c825 | 4910 | assert_return(e, -EINVAL); |
b937d761 | 4911 | assert_return(e = event_resolve(e), -ENOPKG); |
46e8c825 | 4912 | assert_return(usec, -EINVAL); |
46e8c825 LP |
4913 | assert_return(!event_pid_changed(e), -ECHILD); |
4914 | ||
e475d10c LP |
4915 | if (!TRIPLE_TIMESTAMP_HAS_CLOCK(clock)) |
4916 | return -EOPNOTSUPP; | |
4917 | ||
e475d10c | 4918 | if (!triple_timestamp_is_set(&e->timestamp)) { |
15c689d7 | 4919 | /* Implicitly fall back to now() if we never ran before and thus have no cached time. */ |
38a03f06 LP |
4920 | *usec = now(clock); |
4921 | return 1; | |
4922 | } | |
46e8c825 | 4923 | |
e475d10c | 4924 | *usec = triple_timestamp_by_clock(&e->timestamp, clock); |
46e8c825 LP |
4925 | return 0; |
4926 | } | |
afc6adb5 LP |
4927 | |
4928 | _public_ int sd_event_default(sd_event **ret) { | |
39883f62 | 4929 | sd_event *e = NULL; |
afc6adb5 LP |
4930 | int r; |
4931 | ||
4932 | if (!ret) | |
4933 | return !!default_event; | |
4934 | ||
4935 | if (default_event) { | |
4936 | *ret = sd_event_ref(default_event); | |
4937 | return 0; | |
4938 | } | |
4939 | ||
4940 | r = sd_event_new(&e); | |
4941 | if (r < 0) | |
4942 | return r; | |
4943 | ||
4944 | e->default_event_ptr = &default_event; | |
4945 | e->tid = gettid(); | |
4946 | default_event = e; | |
4947 | ||
4948 | *ret = e; | |
4949 | return 1; | |
4950 | } | |
4951 | ||
4952 | _public_ int sd_event_get_tid(sd_event *e, pid_t *tid) { | |
4953 | assert_return(e, -EINVAL); | |
b937d761 | 4954 | assert_return(e = event_resolve(e), -ENOPKG); |
afc6adb5 | 4955 | assert_return(tid, -EINVAL); |
76b54375 | 4956 | assert_return(!event_pid_changed(e), -ECHILD); |
afc6adb5 | 4957 | |
76b54375 LP |
4958 | if (e->tid != 0) { |
4959 | *tid = e->tid; | |
4960 | return 0; | |
4961 | } | |
4962 | ||
4963 | return -ENXIO; | |
afc6adb5 | 4964 | } |
cde93897 LP |
4965 | |
4966 | _public_ int sd_event_set_watchdog(sd_event *e, int b) { | |
4967 | int r; | |
4968 | ||
4969 | assert_return(e, -EINVAL); | |
b937d761 | 4970 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 | 4971 | assert_return(!event_pid_changed(e), -ECHILD); |
cde93897 LP |
4972 | |
4973 | if (e->watchdog == !!b) | |
4974 | return e->watchdog; | |
4975 | ||
4976 | if (b) { | |
09812eb7 LP |
4977 | r = sd_watchdog_enabled(false, &e->watchdog_period); |
4978 | if (r <= 0) | |
cde93897 | 4979 | return r; |
cde93897 LP |
4980 | |
4981 | /* Issue first ping immediately */ | |
4982 | sd_notify(false, "WATCHDOG=1"); | |
4983 | e->watchdog_last = now(CLOCK_MONOTONIC); | |
4984 | ||
4985 | e->watchdog_fd = timerfd_create(CLOCK_MONOTONIC, TFD_NONBLOCK|TFD_CLOEXEC); | |
4986 | if (e->watchdog_fd < 0) | |
4987 | return -errno; | |
4988 | ||
4989 | r = arm_watchdog(e); | |
4990 | if (r < 0) | |
4991 | goto fail; | |
4992 | ||
1eac7948 | 4993 | struct epoll_event ev = { |
a82f89aa LP |
4994 | .events = EPOLLIN, |
4995 | .data.ptr = INT_TO_PTR(SOURCE_WATCHDOG), | |
4996 | }; | |
cde93897 | 4997 | |
15c689d7 | 4998 | if (epoll_ctl(e->epoll_fd, EPOLL_CTL_ADD, e->watchdog_fd, &ev) < 0) { |
cde93897 LP |
4999 | r = -errno; |
5000 | goto fail; | |
5001 | } | |
5002 | ||
5003 | } else { | |
5004 | if (e->watchdog_fd >= 0) { | |
5a795bff | 5005 | (void) epoll_ctl(e->epoll_fd, EPOLL_CTL_DEL, e->watchdog_fd, NULL); |
03e334a1 | 5006 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
5007 | } |
5008 | } | |
5009 | ||
5010 | e->watchdog = !!b; | |
5011 | return e->watchdog; | |
5012 | ||
5013 | fail: | |
03e334a1 | 5014 | e->watchdog_fd = safe_close(e->watchdog_fd); |
cde93897 LP |
5015 | return r; |
5016 | } | |
8f726607 LP |
5017 | |
5018 | _public_ int sd_event_get_watchdog(sd_event *e) { | |
5019 | assert_return(e, -EINVAL); | |
b937d761 | 5020 | assert_return(e = event_resolve(e), -ENOPKG); |
8f726607 LP |
5021 | assert_return(!event_pid_changed(e), -ECHILD); |
5022 | ||
5023 | return e->watchdog; | |
5024 | } | |
60a3b1e1 LP |
5025 | |
5026 | _public_ int sd_event_get_iteration(sd_event *e, uint64_t *ret) { | |
5027 | assert_return(e, -EINVAL); | |
b937d761 | 5028 | assert_return(e = event_resolve(e), -ENOPKG); |
60a3b1e1 LP |
5029 | assert_return(!event_pid_changed(e), -ECHILD); |
5030 | ||
5031 | *ret = e->iteration; | |
5032 | return 0; | |
5033 | } | |
15723a1d LP |
5034 | |
5035 | _public_ int sd_event_source_set_destroy_callback(sd_event_source *s, sd_event_destroy_t callback) { | |
5036 | assert_return(s, -EINVAL); | |
5037 | ||
5038 | s->destroy_callback = callback; | |
5039 | return 0; | |
5040 | } | |
5041 | ||
5042 | _public_ int sd_event_source_get_destroy_callback(sd_event_source *s, sd_event_destroy_t *ret) { | |
5043 | assert_return(s, -EINVAL); | |
5044 | ||
5045 | if (ret) | |
5046 | *ret = s->destroy_callback; | |
5047 | ||
5048 | return !!s->destroy_callback; | |
5049 | } | |
2382c936 YW |
5050 | |
5051 | _public_ int sd_event_source_get_floating(sd_event_source *s) { | |
5052 | assert_return(s, -EINVAL); | |
5053 | ||
5054 | return s->floating; | |
5055 | } | |
5056 | ||
5057 | _public_ int sd_event_source_set_floating(sd_event_source *s, int b) { | |
5058 | assert_return(s, -EINVAL); | |
5059 | ||
5060 | if (s->floating == !!b) | |
5061 | return 0; | |
5062 | ||
5063 | if (!s->event) /* Already disconnected */ | |
5064 | return -ESTALE; | |
5065 | ||
5066 | s->floating = b; | |
5067 | ||
5068 | if (b) { | |
5069 | sd_event_source_ref(s); | |
5070 | sd_event_unref(s->event); | |
5071 | } else { | |
5072 | sd_event_ref(s->event); | |
5073 | sd_event_source_unref(s); | |
5074 | } | |
5075 | ||
5076 | return 1; | |
5077 | } | |
b778cba4 LP |
5078 | |
5079 | _public_ int sd_event_source_get_exit_on_failure(sd_event_source *s) { | |
5080 | assert_return(s, -EINVAL); | |
5081 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
5082 | ||
5083 | return s->exit_on_failure; | |
5084 | } | |
5085 | ||
5086 | _public_ int sd_event_source_set_exit_on_failure(sd_event_source *s, int b) { | |
5087 | assert_return(s, -EINVAL); | |
5088 | assert_return(s->type != SOURCE_EXIT, -EDOM); | |
5089 | ||
5090 | if (s->exit_on_failure == !!b) | |
5091 | return 0; | |
5092 | ||
5093 | s->exit_on_failure = b; | |
5094 | return 1; | |
5095 | } | |
b6d5481b LP |
5096 | |
5097 | _public_ int sd_event_source_set_ratelimit(sd_event_source *s, uint64_t interval, unsigned burst) { | |
5098 | int r; | |
5099 | ||
5100 | assert_return(s, -EINVAL); | |
5101 | ||
5102 | /* Turning on ratelimiting on event source types that don't support it, is a loggable offense. Doing | |
5103 | * so is a programming error. */ | |
5104 | assert_return(EVENT_SOURCE_CAN_RATE_LIMIT(s->type), -EDOM); | |
5105 | ||
5106 | /* When ratelimiting is configured we'll always reset the rate limit state first and start fresh, | |
5107 | * non-ratelimited. */ | |
fd69f224 | 5108 | r = event_source_leave_ratelimit(s, /* run_callback */ false); |
b6d5481b LP |
5109 | if (r < 0) |
5110 | return r; | |
5111 | ||
5112 | s->rate_limit = (RateLimit) { interval, burst }; | |
5113 | return 0; | |
fd69f224 MS |
5114 | } |
5115 | ||
5116 | _public_ int sd_event_source_set_ratelimit_expire_callback(sd_event_source *s, sd_event_handler_t callback) { | |
5117 | assert_return(s, -EINVAL); | |
5118 | ||
5119 | s->ratelimit_expire_callback = callback; | |
5120 | return 0; | |
b6d5481b LP |
5121 | } |
5122 | ||
5123 | _public_ int sd_event_source_get_ratelimit(sd_event_source *s, uint64_t *ret_interval, unsigned *ret_burst) { | |
5124 | assert_return(s, -EINVAL); | |
5125 | ||
6dd3b818 YW |
5126 | /* Querying whether an event source has ratelimiting configured is not a loggable offense, hence |
5127 | * don't use assert_return(). Unlike turning on ratelimiting it's not really a programming error. */ | |
b6d5481b LP |
5128 | if (!EVENT_SOURCE_CAN_RATE_LIMIT(s->type)) |
5129 | return -EDOM; | |
5130 | ||
5131 | if (!ratelimit_configured(&s->rate_limit)) | |
5132 | return -ENOEXEC; | |
5133 | ||
5134 | if (ret_interval) | |
5135 | *ret_interval = s->rate_limit.interval; | |
5136 | if (ret_burst) | |
5137 | *ret_burst = s->rate_limit.burst; | |
5138 | ||
5139 | return 0; | |
5140 | } | |
5141 | ||
5142 | _public_ int sd_event_source_is_ratelimited(sd_event_source *s) { | |
5143 | assert_return(s, -EINVAL); | |
5144 | ||
5145 | if (!EVENT_SOURCE_CAN_RATE_LIMIT(s->type)) | |
5146 | return false; | |
5147 | ||
5148 | if (!ratelimit_configured(&s->rate_limit)) | |
5149 | return false; | |
5150 | ||
5151 | return s->ratelimited; | |
5152 | } | |
baf3fdec LP |
5153 | |
5154 | _public_ int sd_event_set_signal_exit(sd_event *e, int b) { | |
5155 | bool change = false; | |
5156 | int r; | |
5157 | ||
5158 | assert_return(e, -EINVAL); | |
5159 | ||
5160 | if (b) { | |
5161 | /* We want to maintain pointers to these event sources, so that we can destroy them when told | |
5162 | * so. But we also don't want them to pin the event loop itself. Hence we mark them as | |
5163 | * floating after creation (and undo this before deleting them again). */ | |
5164 | ||
5165 | if (!e->sigint_event_source) { | |
5166 | r = sd_event_add_signal(e, &e->sigint_event_source, SIGINT | SD_EVENT_SIGNAL_PROCMASK, NULL, NULL); | |
5167 | if (r < 0) | |
5168 | return r; | |
5169 | ||
5170 | assert(sd_event_source_set_floating(e->sigint_event_source, true) >= 0); | |
5171 | change = true; | |
5172 | } | |
5173 | ||
5174 | if (!e->sigterm_event_source) { | |
5175 | r = sd_event_add_signal(e, &e->sigterm_event_source, SIGTERM | SD_EVENT_SIGNAL_PROCMASK, NULL, NULL); | |
5176 | if (r < 0) { | |
5177 | if (change) { | |
5178 | assert(sd_event_source_set_floating(e->sigint_event_source, false) >= 0); | |
5179 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
5180 | } | |
5181 | ||
5182 | return r; | |
5183 | } | |
5184 | ||
5185 | assert(sd_event_source_set_floating(e->sigterm_event_source, true) >= 0); | |
5186 | change = true; | |
5187 | } | |
5188 | ||
5189 | } else { | |
5190 | if (e->sigint_event_source) { | |
5191 | assert(sd_event_source_set_floating(e->sigint_event_source, false) >= 0); | |
5192 | e->sigint_event_source = sd_event_source_unref(e->sigint_event_source); | |
5193 | change = true; | |
5194 | } | |
5195 | ||
5196 | if (e->sigterm_event_source) { | |
5197 | assert(sd_event_source_set_floating(e->sigterm_event_source, false) >= 0); | |
5198 | e->sigterm_event_source = sd_event_source_unref(e->sigterm_event_source); | |
5199 | change = true; | |
5200 | } | |
5201 | } | |
5202 | ||
5203 | return change; | |
5204 | } | |
158fe190 LP |
5205 | |
5206 | _public_ int sd_event_source_set_memory_pressure_type(sd_event_source *s, const char *ty) { | |
5207 | _cleanup_free_ char *b = NULL; | |
5208 | _cleanup_free_ void *w = NULL; | |
5209 | ||
5210 | assert_return(s, -EINVAL); | |
5211 | assert_return(s->type == SOURCE_MEMORY_PRESSURE, -EDOM); | |
5212 | assert_return(ty, -EINVAL); | |
5213 | ||
5214 | if (!STR_IN_SET(ty, "some", "full")) | |
5215 | return -EINVAL; | |
5216 | ||
5217 | if (s->memory_pressure.locked) /* Refuse adjusting parameters, if caller told us how to watch for events */ | |
5218 | return -EBUSY; | |
5219 | ||
5220 | char* space = memchr(s->memory_pressure.write_buffer, ' ', s->memory_pressure.write_buffer_size); | |
5221 | if (!space) | |
5222 | return -EINVAL; | |
5223 | ||
5224 | size_t l = (char*) space - (char*) s->memory_pressure.write_buffer; | |
5225 | b = memdup_suffix0(s->memory_pressure.write_buffer, l); | |
5226 | if (!b) | |
5227 | return -ENOMEM; | |
5228 | if (!STR_IN_SET(b, "some", "full")) | |
5229 | return -EINVAL; | |
5230 | ||
5231 | if (streq(b, ty)) | |
5232 | return 0; | |
5233 | ||
5234 | size_t nl = strlen(ty) + (s->memory_pressure.write_buffer_size - l); | |
5235 | w = new(char, nl); | |
5236 | if (!w) | |
5237 | return -ENOMEM; | |
5238 | ||
5239 | memcpy(stpcpy(w, ty), space, (s->memory_pressure.write_buffer_size - l)); | |
5240 | ||
5241 | free_and_replace(s->memory_pressure.write_buffer, w); | |
5242 | s->memory_pressure.write_buffer_size = nl; | |
5243 | s->memory_pressure.locked = false; | |
5244 | ||
5245 | return 1; | |
5246 | } | |
5247 | ||
5248 | _public_ int sd_event_source_set_memory_pressure_period(sd_event_source *s, uint64_t threshold_usec, uint64_t window_usec) { | |
5249 | _cleanup_free_ char *b = NULL; | |
5250 | _cleanup_free_ void *w = NULL; | |
5251 | ||
5252 | assert_return(s, -EINVAL); | |
5253 | assert_return(s->type == SOURCE_MEMORY_PRESSURE, -EDOM); | |
5254 | ||
5255 | if (threshold_usec <= 0 || threshold_usec >= UINT64_MAX) | |
5256 | return -ERANGE; | |
5257 | if (window_usec <= 0 || window_usec >= UINT64_MAX) | |
5258 | return -ERANGE; | |
5259 | if (threshold_usec > window_usec) | |
5260 | return -EINVAL; | |
5261 | ||
5262 | if (s->memory_pressure.locked) /* Refuse adjusting parameters, if caller told us how to watch for events */ | |
5263 | return -EBUSY; | |
5264 | ||
5265 | char* space = memchr(s->memory_pressure.write_buffer, ' ', s->memory_pressure.write_buffer_size); | |
5266 | if (!space) | |
5267 | return -EINVAL; | |
5268 | ||
5269 | size_t l = (char*) space - (char*) s->memory_pressure.write_buffer; | |
5270 | b = memdup_suffix0(s->memory_pressure.write_buffer, l); | |
5271 | if (!b) | |
5272 | return -ENOMEM; | |
5273 | if (!STR_IN_SET(b, "some", "full")) | |
5274 | return -EINVAL; | |
5275 | ||
5276 | if (asprintf((char**) &w, | |
5277 | "%s " USEC_FMT " " USEC_FMT "", | |
5278 | b, | |
5279 | threshold_usec, | |
5280 | window_usec) < 0) | |
5281 | return -EINVAL; | |
5282 | ||
5283 | l = strlen(w) + 1; | |
5284 | if (memcmp_nn(s->memory_pressure.write_buffer, s->memory_pressure.write_buffer_size, w, l) == 0) | |
5285 | return 0; | |
5286 | ||
5287 | free_and_replace(s->memory_pressure.write_buffer, w); | |
5288 | s->memory_pressure.write_buffer_size = l; | |
5289 | s->memory_pressure.locked = false; | |
5290 | ||
5291 | return 1; | |
5292 | } |