1 | /* |
2 | * Copyright (c) 2003-2017 Apple Inc. All rights reserved. |
3 | * |
4 | * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ |
5 | * |
6 | * This file contains Original Code and/or Modifications of Original Code |
7 | * as defined in and that are subject to the Apple Public Source License |
8 | * Version 2.0 (the 'License'). You may not use this file except in |
9 | * compliance with the License. The rights granted to you under the License |
10 | * may not be used to create, or enable the creation or redistribution of, |
11 | * unlawful or unlicensed copies of an Apple operating system, or to |
12 | * circumvent, violate, or enable the circumvention or violation of, any |
13 | * terms of an Apple operating system software license agreement. |
14 | * |
15 | * Please obtain a copy of the License at |
16 | * http://www.opensource.apple.com/apsl/ and read it before using this file. |
17 | * |
18 | * The Original Code and all software distributed under the License are |
19 | * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER |
20 | * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, |
21 | * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, |
22 | * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. |
23 | * Please see the License for the specific language governing rights and |
24 | * limitations under the License. |
25 | * |
26 | * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ |
27 | */ |
28 | /*- |
29 | * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org> |
30 | * All rights reserved. |
31 | * |
32 | * Redistribution and use in source and binary forms, with or without |
33 | * modification, are permitted provided that the following conditions |
34 | * are met: |
35 | * 1. Redistributions of source code must retain the above copyright |
36 | * notice, this list of conditions and the following disclaimer. |
37 | * 2. Redistributions in binary form must reproduce the above copyright |
38 | * notice, this list of conditions and the following disclaimer in the |
39 | * documentation and/or other materials provided with the distribution. |
40 | * |
41 | * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND |
42 | * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
43 | * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
44 | * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE |
45 | * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL |
46 | * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS |
47 | * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) |
48 | * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
49 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY |
50 | * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF |
51 | * SUCH DAMAGE. |
52 | * |
53 | * $FreeBSD: src/sys/sys/event.h,v 1.5.2.5 2001/12/14 19:21:22 jlemon Exp $ |
54 | */ |
55 | |
56 | #ifndef _SYS_EVENT_H_ |
57 | #define _SYS_EVENT_H_ |
58 | |
59 | #include <machine/types.h> |
60 | #include <sys/cdefs.h> |
61 | #include <stdint.h> |
62 | |
63 | /* |
64 | * Filter types |
65 | */ |
66 | #define EVFILT_READ (-1) |
67 | #define EVFILT_WRITE (-2) |
68 | #define EVFILT_AIO (-3) /* attached to aio requests */ |
69 | #define EVFILT_VNODE (-4) /* attached to vnodes */ |
70 | #define EVFILT_PROC (-5) /* attached to struct proc */ |
71 | #define EVFILT_SIGNAL (-6) /* attached to struct proc */ |
72 | #define EVFILT_TIMER (-7) /* timers */ |
73 | #define EVFILT_MACHPORT (-8) /* Mach portsets */ |
74 | #define EVFILT_FS (-9) /* Filesystem events */ |
75 | #define EVFILT_USER (-10) /* User events */ |
76 | /* (-11) unused */ |
77 | #define EVFILT_VM (-12) /* Virtual memory events */ |
78 | |
79 | #ifdef PRIVATE |
80 | #define EVFILT_SOCK (-13) /* Socket events */ |
81 | #define EVFILT_MEMORYSTATUS (-14) /* Memorystatus events */ |
82 | #endif /* PRIVATE */ |
83 | #define EVFILT_EXCEPT (-15) /* Exception events */ |
84 | #ifdef PRIVATE |
85 | #define EVFILT_WORKLOOP (-17) /* Workloop events */ |
86 | #endif /* PRIVATE */ |
87 | |
88 | #define EVFILT_SYSCOUNT 17 |
89 | #define EVFILT_THREADMARKER EVFILT_SYSCOUNT /* Internal use only */ |
90 | |
91 | #pragma pack(4) |
92 | |
93 | struct kevent { |
94 | uintptr_t ident; /* identifier for this event */ |
95 | int16_t filter; /* filter for event */ |
96 | uint16_t flags; /* general flags */ |
97 | uint32_t fflags; /* filter-specific flags */ |
98 | intptr_t data; /* filter-specific data */ |
99 | void *udata; /* opaque user data identifier */ |
100 | }; |
101 | |
102 | #ifdef KERNEL_PRIVATE |
103 | |
104 | struct user64_kevent { |
105 | uint64_t ident; /* identifier for this event */ |
106 | int16_t filter; /* filter for event */ |
107 | uint16_t flags; /* general flags */ |
108 | uint32_t fflags; /* filter-specific flags */ |
109 | int64_t data; /* filter-specific data */ |
110 | user_addr_t udata; /* opaque user data identifier */ |
111 | }; |
112 | |
113 | struct user32_kevent { |
114 | uint32_t ident; /* identifier for this event */ |
115 | int16_t filter; /* filter for event */ |
116 | uint16_t flags; /* general flags */ |
117 | uint32_t fflags; /* filter-specific flags */ |
118 | int32_t data; /* filter-specific data */ |
119 | user32_addr_t udata; /* opaque user data identifier */ |
120 | }; |
121 | |
122 | struct kevent_internal_s { |
123 | uint64_t ident; /* identifier for this event */ |
124 | int16_t filter; /* filter for event */ |
125 | uint16_t flags; /* general flags */ |
126 | int32_t qos; /* quality of service */ |
127 | uint32_t fflags; /* filter-specific flags */ |
128 | // uint32_t xflags; /* extra filter-specific flags */ |
129 | int64_t data; /* filter-specific data */ |
130 | uint64_t udata; /* opaque user data identifier */ |
131 | uint64_t ext[4]; /* filter-specific extensions */ |
132 | }; |
133 | |
134 | #endif /* KERNEL_PRIVATE */ |
135 | |
136 | #pragma pack() |
137 | |
138 | struct kevent64_s { |
139 | uint64_t ident; /* identifier for this event */ |
140 | int16_t filter; /* filter for event */ |
141 | uint16_t flags; /* general flags */ |
142 | uint32_t fflags; /* filter-specific flags */ |
143 | int64_t data; /* filter-specific data */ |
144 | uint64_t udata; /* opaque user data identifier */ |
145 | uint64_t ext[2]; /* filter-specific extensions */ |
146 | }; |
147 | |
148 | #ifdef PRIVATE |
149 | struct kevent_qos_s { |
150 | uint64_t ident; /* identifier for this event */ |
151 | int16_t filter; /* filter for event */ |
152 | uint16_t flags; /* general flags */ |
153 | int32_t qos; /* quality of service */ |
154 | uint64_t udata; /* opaque user data identifier */ |
155 | uint32_t fflags; /* filter-specific flags */ |
156 | uint32_t xflags; /* extra filter-specific flags */ |
157 | int64_t data; /* filter-specific data */ |
158 | uint64_t ext[4]; /* filter-specific extensions */ |
159 | }; |
160 | |
161 | /* |
162 | * Type definition for names/ids of dynamically allocated kqueues. |
163 | */ |
164 | typedef uint64_t kqueue_id_t; |
165 | |
166 | #endif /* PRIVATE */ |
167 | |
168 | #define EV_SET(kevp, a, b, c, d, e, f) do { \ |
169 | struct kevent *__kevp__ = (kevp); \ |
170 | __kevp__->ident = (a); \ |
171 | __kevp__->filter = (b); \ |
172 | __kevp__->flags = (c); \ |
173 | __kevp__->fflags = (d); \ |
174 | __kevp__->data = (e); \ |
175 | __kevp__->udata = (f); \ |
176 | } while(0) |
177 | |
178 | #define EV_SET64(kevp, a, b, c, d, e, f, g, h) do { \ |
179 | struct kevent64_s *__kevp__ = (kevp); \ |
180 | __kevp__->ident = (a); \ |
181 | __kevp__->filter = (b); \ |
182 | __kevp__->flags = (c); \ |
183 | __kevp__->fflags = (d); \ |
184 | __kevp__->data = (e); \ |
185 | __kevp__->udata = (f); \ |
186 | __kevp__->ext[0] = (g); \ |
187 | __kevp__->ext[1] = (h); \ |
188 | } while(0) |
189 | |
190 | |
191 | /* kevent system call flags */ |
192 | #define KEVENT_FLAG_NONE 0x000000 /* no flag value */ |
193 | #define KEVENT_FLAG_IMMEDIATE 0x000001 /* immediate timeout */ |
194 | #define KEVENT_FLAG_ERROR_EVENTS 0x000002 /* output events only include change errors */ |
195 | |
196 | #ifdef PRIVATE |
197 | |
198 | /* |
199 | * Rather than provide an EV_SET_QOS macro for kevent_qos_t structure |
200 | * initialization, we encourage use of named field initialization support |
201 | * instead. |
202 | */ |
203 | |
204 | #define KEVENT_FLAG_STACK_EVENTS 0x000004 /* output events treated as stack (grows down) */ |
205 | #define KEVENT_FLAG_STACK_DATA 0x000008 /* output data allocated as stack (grows down) */ |
206 | // 0x000010 |
207 | #define KEVENT_FLAG_WORKQ 0x000020 /* interact with the default workq kq */ |
208 | // KEVENT_FLAG_LEGACY32 0x000040 |
209 | // KEVENT_FLAG_LEGACY64 0x000080 |
210 | // 0x000100 |
211 | #define KEVENT_FLAG_WORKQ_MANAGER 0x000200 /* obsolete */ |
212 | #define KEVENT_FLAG_WORKLOOP 0x000400 /* interact with the specified workloop kq */ |
213 | #define KEVENT_FLAG_PARKING 0x000800 /* workq thread is parking */ |
214 | // KEVENT_FLAG_KERNEL 0x001000 |
215 | // KEVENT_FLAG_DYNAMIC_KQUEUE 0x002000 |
216 | // 0x004000 |
217 | #define KEVENT_FLAG_WORKLOOP_SERVICER_ATTACH 0x008000 /* obsolete */ |
218 | #define KEVENT_FLAG_WORKLOOP_SERVICER_DETACH 0x010000 /* obsolete */ |
219 | #define KEVENT_FLAG_DYNAMIC_KQ_MUST_EXIST 0x020000 /* kq lookup by id must exist */ |
220 | #define KEVENT_FLAG_DYNAMIC_KQ_MUST_NOT_EXIST 0x040000 /* kq lookup by id must not exist */ |
221 | #define KEVENT_FLAG_WORKLOOP_NO_WQ_THREAD 0x080000 /* obsolete */ |
222 | |
223 | #ifdef XNU_KERNEL_PRIVATE |
224 | |
225 | #define KEVENT_FLAG_LEGACY32 0x0040 /* event data in legacy 32-bit format */ |
226 | #define KEVENT_FLAG_LEGACY64 0x0080 /* event data in legacy 64-bit format */ |
227 | #define KEVENT_FLAG_KERNEL 0x1000 /* caller is in-kernel */ |
228 | #define KEVENT_FLAG_DYNAMIC_KQUEUE 0x2000 /* kqueue is dynamically allocated */ |
229 | |
230 | #define KEVENT_FLAG_USER (KEVENT_FLAG_IMMEDIATE | KEVENT_FLAG_ERROR_EVENTS | \ |
231 | KEVENT_FLAG_STACK_EVENTS | KEVENT_FLAG_STACK_DATA | \ |
232 | KEVENT_FLAG_WORKQ | KEVENT_FLAG_WORKLOOP | \ |
233 | KEVENT_FLAG_DYNAMIC_KQ_MUST_EXIST | KEVENT_FLAG_DYNAMIC_KQ_MUST_NOT_EXIST) |
234 | |
235 | /* |
236 | * Since some filter ops are not part of the standard sysfilt_ops, we use |
237 | * kn_filtid starting from EVFILT_SYSCOUNT to identify these cases. This is to |
238 | * let kn_fops() get the correct fops for all cases. |
239 | */ |
240 | #define EVFILTID_KQREAD (EVFILT_SYSCOUNT) |
241 | #define EVFILTID_PIPE_R (EVFILT_SYSCOUNT + 1) |
242 | #define EVFILTID_PIPE_W (EVFILT_SYSCOUNT + 2) |
243 | #define EVFILTID_PTSD (EVFILT_SYSCOUNT + 3) |
244 | #define EVFILTID_SOREAD (EVFILT_SYSCOUNT + 4) |
245 | #define EVFILTID_SOWRITE (EVFILT_SYSCOUNT + 5) |
246 | #define EVFILTID_SCK (EVFILT_SYSCOUNT + 6) |
247 | #define EVFILTID_SOEXCEPT (EVFILT_SYSCOUNT + 7) |
248 | #define EVFILTID_SPEC (EVFILT_SYSCOUNT + 8) |
249 | #define EVFILTID_BPFREAD (EVFILT_SYSCOUNT + 9) |
250 | #define EVFILTID_NECP_FD (EVFILT_SYSCOUNT + 10) |
251 | #define EVFILTID_FSEVENT (EVFILT_SYSCOUNT + 13) |
252 | #define EVFILTID_VN (EVFILT_SYSCOUNT + 14) |
253 | #define EVFILTID_TTY (EVFILT_SYSCOUNT + 16) |
254 | #define EVFILTID_PTMX (EVFILT_SYSCOUNT + 17) |
255 | |
256 | #define EVFILTID_MAX (EVFILT_SYSCOUNT + 18) |
257 | |
258 | #endif /* defined(XNU_KERNEL_PRIVATE) */ |
259 | |
260 | #define EV_SET_QOS 0 |
261 | |
262 | #endif /* PRIVATE */ |
263 | |
264 | /* actions */ |
265 | #define EV_ADD 0x0001 /* add event to kq (implies enable) */ |
266 | #define EV_DELETE 0x0002 /* delete event from kq */ |
267 | #define EV_ENABLE 0x0004 /* enable event */ |
268 | #define EV_DISABLE 0x0008 /* disable event (not reported) */ |
269 | |
270 | /* flags */ |
271 | #define EV_ONESHOT 0x0010 /* only report one occurrence */ |
272 | #define EV_CLEAR 0x0020 /* clear event state after reporting */ |
273 | #define EV_RECEIPT 0x0040 /* force immediate event output */ |
274 | /* ... with or without EV_ERROR */ |
275 | /* ... use KEVENT_FLAG_ERROR_EVENTS */ |
276 | /* on syscalls supporting flags */ |
277 | |
278 | #define EV_DISPATCH 0x0080 /* disable event after reporting */ |
279 | #define EV_UDATA_SPECIFIC 0x0100 /* unique kevent per udata value */ |
280 | |
281 | #define EV_DISPATCH2 (EV_DISPATCH | EV_UDATA_SPECIFIC) |
282 | /* ... in combination with EV_DELETE */ |
283 | /* will defer delete until udata-specific */ |
284 | /* event enabled. EINPROGRESS will be */ |
285 | /* returned to indicate the deferral */ |
286 | |
287 | #define EV_VANISHED 0x0200 /* report that source has vanished */ |
288 | /* ... only valid with EV_DISPATCH2 */ |
289 | |
290 | #define EV_SYSFLAGS 0xF000 /* reserved by system */ |
291 | #define EV_FLAG0 0x1000 /* filter-specific flag */ |
292 | #define EV_FLAG1 0x2000 /* filter-specific flag */ |
293 | |
294 | /* returned values */ |
295 | #define EV_EOF 0x8000 /* EOF detected */ |
296 | #define EV_ERROR 0x4000 /* error, data contains errno */ |
297 | |
298 | /* |
299 | * Filter specific flags for EVFILT_READ |
300 | * |
301 | * The default behavior for EVFILT_READ is to make the "read" determination |
302 | * relative to the current file descriptor read pointer. |
303 | * |
304 | * The EV_POLL flag indicates the determination should be made via poll(2) |
305 | * semantics. These semantics dictate always returning true for regular files, |
306 | * regardless of the amount of unread data in the file. |
307 | * |
308 | * On input, EV_OOBAND specifies that filter should actively return in the |
309 | * presence of OOB on the descriptor. It implies that filter will return |
310 | * if there is OOB data available to read OR when any other condition |
311 | * for the read are met (for example number of bytes regular data becomes >= |
312 | * low-watermark). |
313 | * If EV_OOBAND is not set on input, it implies that the filter should not actively |
314 | * return for out of band data on the descriptor. The filter will then only return |
315 | * when some other condition for read is met (ex: when number of regular data bytes |
316 | * >=low-watermark OR when socket can't receive more data (SS_CANTRCVMORE)). |
317 | * |
318 | * On output, EV_OOBAND indicates the presence of OOB data on the descriptor. |
319 | * If it was not specified as an input parameter, then the data count is the |
320 | * number of bytes before the current OOB marker, else data count is the number |
321 | * of bytes beyond OOB marker. |
322 | */ |
323 | #define EV_POLL EV_FLAG0 |
324 | #define EV_OOBAND EV_FLAG1 |
325 | |
326 | /* |
327 | * data/hint fflags for EVFILT_USER, shared with userspace |
328 | */ |
329 | |
330 | /* |
331 | * On input, NOTE_TRIGGER causes the event to be triggered for output. |
332 | */ |
333 | #define NOTE_TRIGGER 0x01000000 |
334 | |
335 | /* |
336 | * On input, the top two bits of fflags specifies how the lower twenty four |
337 | * bits should be applied to the stored value of fflags. |
338 | * |
339 | * On output, the top two bits will always be set to NOTE_FFNOP and the |
340 | * remaining twenty four bits will contain the stored fflags value. |
341 | */ |
342 | #define NOTE_FFNOP 0x00000000 /* ignore input fflags */ |
343 | #define NOTE_FFAND 0x40000000 /* and fflags */ |
344 | #define NOTE_FFOR 0x80000000 /* or fflags */ |
345 | #define NOTE_FFCOPY 0xc0000000 /* copy fflags */ |
346 | #define NOTE_FFCTRLMASK 0xc0000000 /* mask for operations */ |
347 | #define NOTE_FFLAGSMASK 0x00ffffff |
348 | |
349 | #ifdef PRIVATE |
350 | /* |
351 | * data/hint fflags for EVFILT_WORKLOOP, shared with userspace |
352 | * |
353 | * The ident for thread requests should be the dynamic ID of the workloop |
354 | * The ident for each sync waiter must be unique to that waiter [for this workloop] |
355 | * |
356 | * |
357 | * Commands: |
358 | * |
359 | * @const NOTE_WL_THREAD_REQUEST [in/out] |
360 | * The kevent represents asynchronous userspace work and its associated QoS. |
361 | * There can only be a single knote with this flag set per workloop. |
362 | * |
363 | * @const NOTE_WL_SYNC_WAIT [in/out] |
364 | * This bit is set when the caller is waiting to become the owner of a workloop. |
365 | * If the NOTE_WL_SYNC_WAKE bit is already set then the caller is not blocked, |
366 | * else it blocks until it is set. |
367 | * |
368 | * The QoS field of the knote is used to push on other owners or servicers. |
369 | * |
370 | * @const NOTE_WL_SYNC_WAKE [in/out] |
371 | * Marks the waiter knote as being eligible to become an owner |
372 | * This bit can only be set once, trying it again will fail with EALREADY. |
373 | * |
374 | * |
375 | * Flags/Modifiers: |
376 | * |
377 | * @const NOTE_WL_UPDATE_QOS [in] (only NOTE_WL_THREAD_REQUEST) |
378 | * For successful updates (EV_ADD only), learn the new userspace async QoS from |
379 | * the kevent qos field. |
380 | * |
381 | * @const NOTE_WL_END_OWNERSHIP [in] |
382 | * If the update is successful (including deletions) or returns ESTALE, and |
383 | * the caller thread or the "suspended" thread is currently owning the workloop, |
384 | * then ownership is forgotten. |
385 | * |
386 | * @const NOTE_WL_DISCOVER_OWNER [in] |
387 | * If the update is successful (including deletions), learn the owner identity |
388 | * from the loaded value during debounce. This requires an address to have been |
389 | * filled in the EV_EXTIDX_WL_ADDR ext field, but doesn't require a mask to have |
390 | * been set in the EV_EXTIDX_WL_MASK. |
391 | * |
392 | * @const NOTE_WL_IGNORE_ESTALE [in] |
393 | * If the operation would fail with ESTALE, mask the error and pretend the |
394 | * update was successful. However the operation itself didn't happen, meaning |
395 | * that: |
396 | * - attaching a new knote will not happen |
397 | * - dropping an existing knote will not happen |
398 | * - NOTE_WL_UPDATE_QOS or NOTE_WL_DISCOVER_OWNER will have no effect |
399 | * |
400 | * This modifier doesn't affect NOTE_WL_END_OWNERSHIP. |
401 | */ |
402 | #define NOTE_WL_THREAD_REQUEST 0x00000001 |
403 | #define NOTE_WL_SYNC_WAIT 0x00000004 |
404 | #define NOTE_WL_SYNC_WAKE 0x00000008 |
405 | #define NOTE_WL_COMMANDS_MASK 0x0000000f /* Mask of all the [in] commands above */ |
406 | |
407 | #define NOTE_WL_UPDATE_QOS 0x00000010 |
408 | #define NOTE_WL_END_OWNERSHIP 0x00000020 |
409 | #define NOTE_WL_UPDATE_OWNER 0 /* ... compatibility define ... */ |
410 | #define NOTE_WL_DISCOVER_OWNER 0x00000080 |
411 | #define NOTE_WL_IGNORE_ESTALE 0x00000100 |
412 | #define NOTE_WL_UPDATES_MASK 0x000001f0 /* Mask of all the [in] updates above */ |
413 | |
414 | /* |
415 | * EVFILT_WORKLOOP ext[] array indexes/meanings. |
416 | */ |
417 | #define EV_EXTIDX_WL_LANE 0 /* lane identifier [in: sync waiter] |
418 | [out: thread request] */ |
419 | #define EV_EXTIDX_WL_ADDR 1 /* debounce address [in: NULL==no debounce] */ |
420 | #define EV_EXTIDX_WL_MASK 2 /* debounce mask [in] */ |
421 | #define EV_EXTIDX_WL_VALUE 3 /* debounce value [in: not current->ESTALE] |
422 | [out: new/debounce value] */ |
423 | #endif /* PRIVATE */ |
424 | |
425 | /* |
426 | * data/hint fflags for EVFILT_{READ|WRITE}, shared with userspace |
427 | * |
428 | * The default behavior for EVFILT_READ is to make the determination |
429 | * realtive to the current file descriptor read pointer. |
430 | */ |
431 | #define NOTE_LOWAT 0x00000001 /* low water mark */ |
432 | |
433 | /* data/hint flags for EVFILT_EXCEPT, shared with userspace */ |
434 | #define NOTE_OOB 0x00000002 /* OOB data */ |
435 | |
436 | /* |
437 | * data/hint fflags for EVFILT_VNODE, shared with userspace |
438 | */ |
439 | #define NOTE_DELETE 0x00000001 /* vnode was removed */ |
440 | #define NOTE_WRITE 0x00000002 /* data contents changed */ |
441 | #define NOTE_EXTEND 0x00000004 /* size increased */ |
442 | #define NOTE_ATTRIB 0x00000008 /* attributes changed */ |
443 | #define NOTE_LINK 0x00000010 /* link count changed */ |
444 | #define NOTE_RENAME 0x00000020 /* vnode was renamed */ |
445 | #define NOTE_REVOKE 0x00000040 /* vnode access was revoked */ |
446 | #define NOTE_NONE 0x00000080 /* No specific vnode event: to test for EVFILT_READ activation*/ |
447 | #define NOTE_FUNLOCK 0x00000100 /* vnode was unlocked by flock(2) */ |
448 | |
449 | /* |
450 | * data/hint fflags for EVFILT_PROC, shared with userspace |
451 | * |
452 | * Please note that EVFILT_PROC and EVFILT_SIGNAL share the same knote list |
453 | * that hangs off the proc structure. They also both play games with the hint |
454 | * passed to KNOTE(). If NOTE_SIGNAL is passed as a hint, then the lower bits |
455 | * of the hint contain the signal. IF NOTE_FORK is passed, then the lower bits |
456 | * contain the PID of the child (but the pid does not get passed through in |
457 | * the actual kevent). |
458 | */ |
459 | enum { |
460 | eNoteReapDeprecated __deprecated_enum_msg("This kqueue(2) EVFILT_PROC flag is deprecated" ) = 0x10000000 |
461 | }; |
462 | |
463 | #define NOTE_EXIT 0x80000000 /* process exited */ |
464 | #define NOTE_FORK 0x40000000 /* process forked */ |
465 | #define NOTE_EXEC 0x20000000 /* process exec'd */ |
466 | #define NOTE_REAP ((unsigned int)eNoteReapDeprecated /* 0x10000000 */) /* process reaped */ |
467 | #define NOTE_SIGNAL 0x08000000 /* shared with EVFILT_SIGNAL */ |
468 | #define NOTE_EXITSTATUS 0x04000000 /* exit status to be returned, valid for child process only */ |
469 | #define NOTE_EXIT_DETAIL 0x02000000 /* provide details on reasons for exit */ |
470 | |
471 | #define NOTE_PDATAMASK 0x000fffff /* mask for signal & exit status */ |
472 | #define NOTE_PCTRLMASK (~NOTE_PDATAMASK) |
473 | |
474 | /* |
475 | * If NOTE_EXITSTATUS is present, provide additional info about exiting process. |
476 | */ |
477 | enum { |
478 | eNoteExitReparentedDeprecated __deprecated_enum_msg("This kqueue(2) EVFILT_PROC flag is no longer sent" ) = 0x00080000 |
479 | }; |
480 | #define NOTE_EXIT_REPARENTED ((unsigned int)eNoteExitReparentedDeprecated) /* exited while reparented */ |
481 | |
482 | /* |
483 | * If NOTE_EXIT_DETAIL is present, these bits indicate specific reasons for exiting. |
484 | */ |
485 | #define NOTE_EXIT_DETAIL_MASK 0x00070000 |
486 | #define NOTE_EXIT_DECRYPTFAIL 0x00010000 |
487 | #define NOTE_EXIT_MEMORY 0x00020000 |
488 | #define NOTE_EXIT_CSERROR 0x00040000 |
489 | |
490 | #ifdef PRIVATE |
491 | |
492 | /* |
493 | * If NOTE_EXIT_MEMORY is present, these bits indicate specific jetsam condition. |
494 | */ |
495 | #define NOTE_EXIT_MEMORY_DETAIL_MASK 0xfe000000 |
496 | #define NOTE_EXIT_MEMORY_VMPAGESHORTAGE 0x80000000 /* jetsam condition: lowest jetsam priority proc killed due to vm page shortage */ |
497 | #define NOTE_EXIT_MEMORY_VMTHRASHING 0x40000000 /* jetsam condition: lowest jetsam priority proc killed due to vm thrashing */ |
498 | #define NOTE_EXIT_MEMORY_HIWAT 0x20000000 /* jetsam condition: process reached its high water mark */ |
499 | #define NOTE_EXIT_MEMORY_PID 0x10000000 /* jetsam condition: special pid kill requested */ |
500 | #define NOTE_EXIT_MEMORY_IDLE 0x08000000 /* jetsam condition: idle process cleaned up */ |
501 | #define NOTE_EXIT_MEMORY_VNODE 0X04000000 /* jetsam condition: virtual node kill */ |
502 | #define NOTE_EXIT_MEMORY_FCTHRASHING 0x02000000 /* jetsam condition: lowest jetsam priority proc killed due to filecache thrashing */ |
503 | |
504 | #endif |
505 | |
506 | /* |
507 | * data/hint fflags for EVFILT_VM, shared with userspace. |
508 | */ |
509 | #define NOTE_VM_PRESSURE 0x80000000 /* will react on memory pressure */ |
510 | #define NOTE_VM_PRESSURE_TERMINATE 0x40000000 /* will quit on memory pressure, possibly after cleaning up dirty state */ |
511 | #define NOTE_VM_PRESSURE_SUDDEN_TERMINATE 0x20000000 /* will quit immediately on memory pressure */ |
512 | #define NOTE_VM_ERROR 0x10000000 /* there was an error */ |
513 | |
514 | #ifdef PRIVATE |
515 | |
516 | /* |
517 | * data/hint fflags for EVFILT_MEMORYSTATUS, shared with userspace. |
518 | */ |
519 | #define NOTE_MEMORYSTATUS_PRESSURE_NORMAL 0x00000001 /* system memory pressure has returned to normal */ |
520 | #define NOTE_MEMORYSTATUS_PRESSURE_WARN 0x00000002 /* system memory pressure has changed to the warning state */ |
521 | #define NOTE_MEMORYSTATUS_PRESSURE_CRITICAL 0x00000004 /* system memory pressure has changed to the critical state */ |
522 | #define NOTE_MEMORYSTATUS_LOW_SWAP 0x00000008 /* system is in a low-swap state */ |
523 | #define NOTE_MEMORYSTATUS_PROC_LIMIT_WARN 0x00000010 /* process memory limit has hit a warning state */ |
524 | #define NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL 0x00000020 /* process memory limit has hit a critical state - soft limit */ |
525 | #define NOTE_MEMORYSTATUS_MSL_STATUS 0xf0000000 /* bits used to request change to process MSL status */ |
526 | |
527 | #ifdef KERNEL_PRIVATE |
528 | /* |
529 | * data/hint fflags for EVFILT_MEMORYSTATUS, but not shared with userspace. |
530 | */ |
531 | #define NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_ACTIVE 0x00000040 /* Used to restrict sending a warn event only once, per active limit, soft limits only */ |
532 | #define NOTE_MEMORYSTATUS_PROC_LIMIT_WARN_INACTIVE 0x00000080 /* Used to restrict sending a warn event only once, per inactive limit, soft limit only */ |
533 | #define NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_ACTIVE 0x00000100 /* Used to restrict sending a critical event only once per active limit, soft limit only */ |
534 | #define NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL_INACTIVE 0x00000200 /* Used to restrict sending a critical event only once per inactive limit, soft limit only */ |
535 | |
536 | /* |
537 | * Use this mask to protect the kernel private flags. |
538 | */ |
539 | #define EVFILT_MEMORYSTATUS_ALL_MASK \ |
540 | (NOTE_MEMORYSTATUS_PRESSURE_NORMAL | NOTE_MEMORYSTATUS_PRESSURE_WARN | NOTE_MEMORYSTATUS_PRESSURE_CRITICAL | NOTE_MEMORYSTATUS_LOW_SWAP | \ |
541 | NOTE_MEMORYSTATUS_PROC_LIMIT_WARN | NOTE_MEMORYSTATUS_PROC_LIMIT_CRITICAL | NOTE_MEMORYSTATUS_MSL_STATUS) |
542 | |
543 | #endif /* KERNEL_PRIVATE */ |
544 | |
545 | typedef enum vm_pressure_level { |
546 | kVMPressureNormal = 0, |
547 | kVMPressureWarning = 1, |
548 | kVMPressureUrgent = 2, |
549 | kVMPressureCritical = 3, |
550 | } vm_pressure_level_t; |
551 | |
552 | #endif /* PRIVATE */ |
553 | |
554 | /* |
555 | * data/hint fflags for EVFILT_TIMER, shared with userspace. |
556 | * The default is a (repeating) interval timer with the data |
557 | * specifying the timeout interval in milliseconds. |
558 | * |
559 | * All timeouts are implicitly EV_CLEAR events. |
560 | */ |
561 | #define NOTE_SECONDS 0x00000001 /* data is seconds */ |
562 | #define NOTE_USECONDS 0x00000002 /* data is microseconds */ |
563 | #define NOTE_NSECONDS 0x00000004 /* data is nanoseconds */ |
564 | #define NOTE_ABSOLUTE 0x00000008 /* absolute timeout */ |
565 | /* ... implicit EV_ONESHOT, timeout uses the gettimeofday epoch */ |
566 | #define NOTE_LEEWAY 0x00000010 /* ext[1] holds leeway for power aware timers */ |
567 | #define NOTE_CRITICAL 0x00000020 /* system does minimal timer coalescing */ |
568 | #define NOTE_BACKGROUND 0x00000040 /* system does maximum timer coalescing */ |
569 | #define NOTE_MACH_CONTINUOUS_TIME 0x00000080 |
570 | /* |
571 | * NOTE_MACH_CONTINUOUS_TIME: |
572 | * with NOTE_ABSOLUTE: causes the timer to continue to tick across sleep, |
573 | * still uses gettimeofday epoch |
574 | * with NOTE_MACHTIME and NOTE_ABSOLUTE: uses mach continuous time epoch |
575 | * without NOTE_ABSOLUTE (interval timer mode): continues to tick across sleep |
576 | */ |
577 | #define NOTE_MACHTIME 0x00000100 /* data is mach absolute time units */ |
578 | /* timeout uses the mach absolute time epoch */ |
579 | |
580 | #ifdef PRIVATE |
581 | /* |
582 | * data/hint fflags for EVFILT_SOCK, shared with userspace. |
583 | * |
584 | */ |
585 | #define NOTE_CONNRESET 0x00000001 /* Received RST */ |
586 | #define NOTE_READCLOSED 0x00000002 /* Read side is shutdown */ |
587 | #define NOTE_WRITECLOSED 0x00000004 /* Write side is shutdown */ |
588 | #define NOTE_TIMEOUT 0x00000008 /* timeout: rexmt, keep-alive or persist */ |
589 | #define NOTE_NOSRCADDR 0x00000010 /* source address not available */ |
590 | #define NOTE_IFDENIED 0x00000020 /* interface denied connection */ |
591 | #define NOTE_SUSPEND 0x00000040 /* output queue suspended */ |
592 | #define NOTE_RESUME 0x00000080 /* output queue resumed */ |
593 | #define NOTE_KEEPALIVE 0x00000100 /* TCP Keepalive received */ |
594 | #define NOTE_ADAPTIVE_WTIMO 0x00000200 /* TCP adaptive write timeout */ |
595 | #define NOTE_ADAPTIVE_RTIMO 0x00000400 /* TCP adaptive read timeout */ |
596 | #define NOTE_CONNECTED 0x00000800 /* socket is connected */ |
597 | #define NOTE_DISCONNECTED 0x00001000 /* socket is disconnected */ |
598 | #define NOTE_CONNINFO_UPDATED 0x00002000 /* connection info was updated */ |
599 | #define NOTE_NOTIFY_ACK 0x00004000 /* notify acknowledgement */ |
600 | |
601 | #define EVFILT_SOCK_LEVEL_TRIGGER_MASK \ |
602 | (NOTE_READCLOSED | NOTE_WRITECLOSED | NOTE_SUSPEND | NOTE_RESUME | \ |
603 | NOTE_CONNECTED | NOTE_DISCONNECTED) |
604 | |
605 | #define EVFILT_SOCK_ALL_MASK \ |
606 | (NOTE_CONNRESET | NOTE_READCLOSED | NOTE_WRITECLOSED | NOTE_TIMEOUT | \ |
607 | NOTE_NOSRCADDR | NOTE_IFDENIED | NOTE_SUSPEND | NOTE_RESUME | \ |
608 | NOTE_KEEPALIVE | NOTE_ADAPTIVE_WTIMO | NOTE_ADAPTIVE_RTIMO | \ |
609 | NOTE_CONNECTED | NOTE_DISCONNECTED | NOTE_CONNINFO_UPDATED | \ |
610 | NOTE_NOTIFY_ACK) |
611 | |
612 | #endif /* PRIVATE */ |
613 | |
614 | /* |
615 | * data/hint fflags for EVFILT_MACHPORT, shared with userspace. |
616 | * |
617 | * Only portsets are supported at this time. |
618 | * |
619 | * The fflags field can optionally contain the MACH_RCV_MSG, MACH_RCV_LARGE, |
620 | * and related trailer receive options as defined in <mach/message.h>. |
621 | * The presence of these flags directs the kevent64() call to attempt to receive |
622 | * the message during kevent delivery, rather than just indicate that a message exists. |
623 | * On setup, The ext[0] field contains the receive buffer pointer and ext[1] contains |
624 | * the receive buffer length. Upon event delivery, the actual received message size |
625 | * is returned in ext[1]. As with mach_msg(), the buffer must be large enough to |
626 | * receive the message and the requested (or default) message trailers. In addition, |
627 | * the fflags field contains the return code normally returned by mach_msg(). |
628 | * |
629 | * If MACH_RCV_MSG is specified, and the ext[1] field specifies a zero length, the |
630 | * system call argument specifying an ouput area (kevent_qos) will be consulted. If |
631 | * the system call specified an output data area, the user-space address |
632 | * of the received message is carved from that provided output data area (if enough |
633 | * space remains there). The address and length of each received message is |
634 | * returned in the ext[0] and ext[1] fields (respectively) of the corresponding kevent. |
635 | * |
636 | * IF_MACH_RCV_VOUCHER_CONTENT is specified, the contents of the message voucher is |
637 | * extracted (as specified in the xflags field) and stored in ext[2] up to ext[3] |
638 | * length. If the input length is zero, and the system call provided a data area, |
639 | * the space for the voucher content is carved from the provided space and its |
640 | * address and length is returned in ext[2] and ext[3] respectively. |
641 | * |
642 | * If no message receipt options were provided in the fflags field on setup, no |
643 | * message is received by this call. Instead, on output, the data field simply |
644 | * contains the name of the actual port detected with a message waiting. |
645 | */ |
646 | |
647 | /* |
648 | * DEPRECATED!!!!!!!!! |
649 | * NOTE_TRACK, NOTE_TRACKERR, and NOTE_CHILD are no longer supported as of 10.5 |
650 | */ |
651 | /* additional flags for EVFILT_PROC */ |
652 | #define NOTE_TRACK 0x00000001 /* follow across forks */ |
653 | #define NOTE_TRACKERR 0x00000002 /* could not track child */ |
654 | #define NOTE_CHILD 0x00000004 /* am a child process */ |
655 | |
656 | |
657 | #ifdef PRIVATE |
658 | #endif /* PRIVATE */ |
659 | |
660 | #ifndef KERNEL |
661 | /* Temporay solution for BootX to use inode.h till kqueue moves to vfs layer */ |
662 | #include <sys/queue.h> |
663 | struct knote; |
664 | SLIST_HEAD(klist, knote); |
665 | #endif |
666 | |
667 | #ifdef KERNEL |
668 | |
669 | #ifdef XNU_KERNEL_PRIVATE |
670 | #include <sys/queue.h> |
671 | #include <kern/kern_types.h> |
672 | #include <sys/fcntl.h> /* FREAD, FWRITE */ |
673 | #include <kern/debug.h> /* panic */ |
674 | #include <pthread/priority_private.h> |
675 | |
676 | #ifdef MALLOC_DECLARE |
677 | MALLOC_DECLARE(M_KQUEUE); |
678 | #endif |
679 | |
680 | TAILQ_HEAD(kqtailq, knote); /* a list of "queued" events */ |
681 | |
682 | /* index into various kq queues */ |
683 | typedef uint8_t kq_index_t; |
684 | typedef uint16_t kn_status_t; |
685 | |
686 | #define KN_ACTIVE 0x0001 /* event has been triggered */ |
687 | #define KN_QUEUED 0x0002 /* event is on queue */ |
688 | #define KN_DISABLED 0x0004 /* event is disabled */ |
689 | #define KN_DROPPING 0x0008 /* knote is being dropped */ |
690 | #define KN_LOCKED 0x0010 /* knote is locked (kq_knlocks) */ |
691 | #define KN_ATTACHING 0x0020 /* event is pending attach */ |
692 | #define KN_STAYACTIVE 0x0040 /* force event to stay active */ |
693 | #define KN_DEFERDELETE 0x0080 /* defer delete until re-enabled */ |
694 | #define KN_ATTACHED 0x0100 /* currently attached to source */ |
695 | #define KN_DISPATCH 0x0200 /* disables as part of deliver */ |
696 | #define KN_UDATA_SPECIFIC 0x0400 /* udata is part of matching */ |
697 | #define KN_SUPPRESSED 0x0800 /* event is suppressed during delivery */ |
698 | #define KN_MERGE_QOS 0x1000 /* f_event() / f_* ran concurrently and |
699 | overrides must merge */ |
700 | #define KN_REQVANISH 0x2000 /* requested EV_VANISH */ |
701 | #define KN_VANISHED 0x4000 /* has vanished */ |
702 | // 0x8000 |
703 | |
704 | /* combination defines deferred-delete mode enabled */ |
705 | #define KN_DISPATCH2 (KN_DISPATCH | KN_UDATA_SPECIFIC) |
706 | |
707 | #define KNOTE_KQ_BITSIZE 42 |
708 | _Static_assert(KNOTE_KQ_BITSIZE >= VM_KERNEL_POINTER_SIGNIFICANT_BITS, |
709 | "Make sure sign extending kn_kq_packed is legit" ); |
710 | |
711 | struct kqueue; |
712 | struct knote { |
713 | TAILQ_ENTRY(knote) kn_tqe; /* linkage for tail queue */ |
714 | SLIST_ENTRY(knote) kn_link; /* linkage for search list */ |
715 | SLIST_ENTRY(knote) kn_selnext; /* klist element chain */ |
716 | uintptr_t kn_filtid:8, /* filter id to index filter ops */ |
717 | kn_req_index:4, /* requested qos index */ |
718 | kn_qos_index:4, /* in-use qos index */ |
719 | kn_qos_override:4, /* qos override index */ |
720 | kn_vnode_kqok:1, |
721 | kn_vnode_use_ofst:1; |
722 | #if __LP64__ |
723 | intptr_t kn_kq_packed : KNOTE_KQ_BITSIZE; |
724 | #else |
725 | intptr_t kn_kq_packed; |
726 | #endif |
727 | union { |
728 | void *kn_hook; |
729 | uint64_t kn_hook_data; |
730 | }; |
731 | int64_t kn_sdata; /* saved data field */ |
732 | union { |
733 | struct fileproc *p_fp; /* file data pointer */ |
734 | struct proc *p_proc; /* proc pointer */ |
735 | struct ipc_mqueue *p_mqueue; /* pset pointer */ |
736 | } kn_ptr; |
737 | struct kevent_internal_s kn_kevent; |
738 | int kn_sfflags; /* saved filter flags */ |
739 | int kn_hookid; |
740 | uint16_t kn_inuse; /* inuse count */ |
741 | kn_status_t kn_status; /* status bits */ |
742 | |
743 | #define kn_id kn_kevent.ident |
744 | #define kn_filter kn_kevent.filter |
745 | #define kn_flags kn_kevent.flags |
746 | #define kn_qos kn_kevent.qos |
747 | #define kn_udata kn_kevent.udata |
748 | #define kn_fflags kn_kevent.fflags |
749 | #define kn_xflags kn_kevent.xflags |
750 | #define kn_data kn_kevent.data |
751 | #define kn_ext kn_kevent.ext |
752 | #define kn_fp kn_ptr.p_fp |
753 | }; |
754 | |
755 | static inline struct kqueue * |
756 | knote_get_kq(struct knote *kn) |
757 | { |
758 | return (struct kqueue *)kn->kn_kq_packed; |
759 | } |
760 | |
761 | static inline int knote_get_seltype(struct knote *kn) |
762 | { |
763 | switch (kn->kn_filter) { |
764 | case EVFILT_READ: |
765 | return FREAD; |
766 | case EVFILT_WRITE: |
767 | return FWRITE; |
768 | default: |
769 | panic("%s(%p): invalid filter %d\n" , |
770 | __func__, kn, kn->kn_filter); |
771 | return 0; |
772 | } |
773 | } |
774 | |
775 | static inline void knote_set_error(struct knote *kn, int error) |
776 | { |
777 | kn->kn_flags |= EV_ERROR; |
778 | kn->kn_data = error; |
779 | } |
780 | |
781 | struct filt_process_s { |
782 | int fp_fd; |
783 | unsigned int fp_flags; |
784 | user_addr_t fp_data_out; |
785 | user_size_t fp_data_size; |
786 | user_size_t fp_data_resid; |
787 | }; |
788 | typedef struct filt_process_s *filt_process_data_t; |
789 | |
790 | /* |
791 | * Filter operators |
792 | * |
793 | * These routines, provided by each filter, are called to attach, detach, deliver events, |
794 | * change/update filter registration and process/deliver events: |
795 | * |
796 | * - the f_attach, f_touch, f_process, f_peek and f_detach callbacks are always |
797 | * serialized with respect to each other for the same knote. |
798 | * |
799 | * - the f_event routine is called with a use-count taken on the knote to |
800 | * prolongate its lifetime and protect against drop, but is not otherwise |
801 | * serialized with other routine calls. |
802 | * |
803 | * - the f_detach routine is always called last, and is serialized with all |
804 | * other callbacks, including f_event calls. |
805 | * |
806 | * |
807 | * Here are more details: |
808 | * |
809 | * f_isfd - |
810 | * identifies if the "ident" field in the kevent structure is a file-descriptor. |
811 | * |
812 | * If so, the knote is associated with the file descriptor prior to attach and |
813 | * auto-removed when the file descriptor is closed (this latter behavior may change |
814 | * for EV_DISPATCH2 kevent types to allow delivery of events identifying unintended |
815 | * closes). |
816 | * |
817 | * Otherwise the knote is hashed by the ident and has no auto-close behavior. |
818 | * |
819 | * f_adjusts_qos - |
820 | * identifies if the filter can adjust its QoS during its lifetime. |
821 | * |
822 | * Filters using this facility should request the new overrides they want |
823 | * using the appropriate FILTER_{RESET,ADJUST}_EVENT_QOS extended codes. |
824 | * |
825 | * Currently, EVFILT_MACHPORT is the only filter using this facility. |
826 | * |
827 | * f_extended_codes - |
828 | * identifies if the filter returns extended codes from its routines |
829 | * (see FILTER_ACTIVE, ...) or 0 / 1 values. |
830 | * |
831 | * f_attach - |
832 | * called to attach the knote to the underlying object that will be delivering events |
833 | * through it when EV_ADD is supplied and no existing matching event is found |
834 | * |
835 | * provided a knote that is pre-attached to the fd or hashed (see above) but is |
836 | * specially marked to avoid concurrent access until the attach is complete. The |
837 | * kevent structure embedded in this knote has been filled in with a sanitized |
838 | * version of the user-supplied kevent data. However, the user-supplied filter-specific |
839 | * flags (fflags) and data fields have been moved into the knote's kn_sfflags and kn_sdata |
840 | * fields respectively. These are usually interpretted as a set of "interest" flags and |
841 | * data by each filter - to be matched against delivered events. |
842 | * |
843 | * The attach operator indicated errors by setting the EV_ERROR flog in the flags field |
844 | * embedded in the knote's kevent structure - with the specific error indicated in the |
845 | * corresponding data field. |
846 | * |
847 | * The return value indicates if the knote should already be considered "activated" at |
848 | * the time of attach (one or more of the interest events has already occured). |
849 | * |
850 | * f_detach - |
851 | * called to disassociate the knote from the underlying object delivering events |
852 | * the filter should not attempt to deliver events through this knote after this |
853 | * operation returns control to the kq system. |
854 | * |
855 | * f_event - |
856 | * if the knote() function (or KNOTE() macro) is called against a list of knotes, |
857 | * this operator will be called on each knote in the list. |
858 | * |
859 | * The "hint" parameter is completely filter-specific, but usually indicates an |
860 | * event or set of events that have occured against the source object associated |
861 | * with the list. |
862 | * |
863 | * The return value indicates if the knote should already be considered "activated" at |
864 | * the time of attach (one or more of the interest events has already occured). |
865 | * |
866 | * f_process - |
867 | * called when attempting to deliver triggered events to user-space. |
868 | * |
869 | * If the knote was previously activated, this operator will be called when a |
870 | * thread is trying to deliver events to user-space. The filter gets one last |
871 | * chance to determine if the event/events are still interesting for this knote |
872 | * (are the conditions still right to deliver an event). If so, the filter |
873 | * fills in the output kevent structure with the information to be delivered. |
874 | * |
875 | * The input context/data parameter is used during event delivery. Some |
876 | * filters allow additional data delivery as part of event delivery. This |
877 | * context field indicates if space was made available for these additional |
878 | * items and how that space is to be allocated/carved-out. |
879 | * |
880 | * The filter may set EV_CLEAR or EV_ONESHOT in the output flags field to indicate |
881 | * special post-delivery dispositions for the knote. |
882 | * |
883 | * EV_CLEAR - indicates that all matching events have been delivered. Even |
884 | * though there were events to deliver now, there will not be any |
885 | * more until some additional events are delivered to the knote |
886 | * via the f_event operator, or the interest set is changed via |
887 | * the f_touch operator. The knote can remain deactivated after |
888 | * processing this event delivery. |
889 | * |
890 | * EV_ONESHOT - indicates that this is the last event to be delivered via |
891 | * this knote. It will automatically be deleted upon delivery |
892 | * (or if in dispatch-mode, upon re-enablement after this delivery). |
893 | * |
894 | * The return value indicates if the knote has delivered an output event. |
895 | * Unless one of the special output flags was set in the output kevent, a non- |
896 | * zero return value ALSO indicates that the knote should be re-activated |
897 | * for future event processing (in case it delivers level-based or a multi-edge |
898 | * type events like message queues that already exist). |
899 | * |
900 | * NOTE: In the future, the boolean may change to an enum that allows more |
901 | * explicit indication of just delivering a current event vs delivering |
902 | * an event with more events still pending. |
903 | * |
904 | * f_touch - |
905 | * called to update the knote with new state from the user during |
906 | * EVFILT_ADD/ENABLE/DISABLE on an already-attached knote. |
907 | * |
908 | * f_touch should copy relevant new data from the kevent into the knote. |
909 | * |
910 | * operator must lock against concurrent f_event operations. |
911 | * |
912 | * A return value of 1 indicates that the knote should now be considered |
913 | * 'activated'. |
914 | * |
915 | * f_touch can set EV_ERROR with specific error in the data field to |
916 | * return an error to the client. You should return 1 to indicate that |
917 | * the kevent needs to be activated and processed. |
918 | * |
919 | * f_peek - |
920 | * For knotes marked KN_STAYACTIVE, indicate if the knote is truly active |
921 | * at the moment (not used for event delivery, but for status checks). |
922 | * |
923 | * f_allow_drop - |
924 | * |
925 | * [OPTIONAL] If this function is non-null, then it indicates that the |
926 | * filter wants to validate EV_DELETE events. This is necessary if |
927 | * a particular filter needs to synchronize knote deletion with its own |
928 | * filter lock. |
929 | * |
930 | * When true is returned, the the EV_DELETE is allowed and can proceed. |
931 | * |
932 | * If false is returned, the EV_DELETE doesn't proceed, and the passed in |
933 | * kevent is used for the copyout to userspace. |
934 | * |
935 | * Currently, EVFILT_WORKLOOP is the only filter using this facility. |
936 | * |
937 | * f_post_register_wait - |
938 | * [OPTIONAL] called when attach or touch return the FILTER_REGISTER_WAIT |
939 | * extended code bit. It is possible to use this facility when the last |
940 | * register command wants to wait. |
941 | * |
942 | * Currently, EVFILT_WORKLOOP is the only filter using this facility. |
943 | */ |
944 | |
945 | struct _kevent_register; |
946 | struct knote_lock_ctx; |
947 | struct proc; |
948 | struct uthread; |
949 | struct waitq; |
950 | |
951 | struct filterops { |
952 | bool f_isfd; /* true if ident == filedescriptor */ |
953 | bool f_adjusts_qos; /* true if the filter can override the knote */ |
954 | bool f_extended_codes; /* hooks return extended codes */ |
955 | |
956 | int (*f_attach)(struct knote *kn, struct kevent_internal_s *kev); |
957 | void (*f_detach)(struct knote *kn); |
958 | int (*f_event)(struct knote *kn, long hint); |
959 | int (*f_touch)(struct knote *kn, struct kevent_internal_s *kev); |
960 | int (*f_process)(struct knote *kn, struct filt_process_s *data, struct kevent_internal_s *kev); |
961 | int (*f_peek)(struct knote *kn); |
962 | |
963 | /* optional & advanced */ |
964 | bool (*f_allow_drop)(struct knote *kn, struct kevent_internal_s *kev); |
965 | void (*f_post_register_wait)(struct uthread *uth, struct knote_lock_ctx *ctx, |
966 | struct _kevent_register *ss_kr); |
967 | }; |
968 | |
969 | /* |
970 | * Extended codes returned by filter routines when f_extended_codes is set. |
971 | * |
972 | * FILTER_ACTIVE |
973 | * The filter is active and a call to f_process() may return an event. |
974 | * |
975 | * For f_process() the meaning is slightly different: the knote will be |
976 | * activated again as long as f_process returns FILTER_ACTIVE, unless |
977 | * EV_CLEAR is set, which require a new f_event to reactivate the knote. |
978 | * |
979 | * Valid: f_attach, f_event, f_touch, f_process, f_peek |
980 | * Implicit: - |
981 | * Ignored: - |
982 | * |
983 | * FILTER_REGISTER_WAIT |
984 | * The filter wants its f_post_register_wait() to be called. |
985 | * |
986 | * Note: It is only valid to ask for this behavior for a workloop kqueue, |
987 | * and is really only meant to be used by EVFILT_WORKLOOP. |
988 | * |
989 | * Valid: f_attach, f_touch |
990 | * Implicit: - |
991 | * Ignored: f_event, f_process, f_peek |
992 | * |
993 | * FILTER_UPDATE_REQ_QOS |
994 | * The filter wants the passed in QoS to be updated as the new intrinsic qos |
995 | * for this knote. If the kevent `qos` field is 0, no update is performed. |
996 | * |
997 | * This also will reset the event QoS, so FILTER_ADJUST_EVENT_QOS() must |
998 | * also be used if an override should be maintained. |
999 | * |
1000 | * Valid: f_touch |
1001 | * Implicit: f_attach |
1002 | * Ignored: f_event, f_process, f_peek |
1003 | * |
1004 | * FILTER_RESET_EVENT_QOS |
1005 | * FILTER_ADJUST_EVENT_QOS(qos) |
1006 | * The filter wants the QoS of the next event delivery to be overridden |
1007 | * at the specified QoS. This allows for the next event QoS to be elevated |
1008 | * from the knote requested qos (See FILTER_UPDATE_REQ_QOS). |
1009 | * |
1010 | * Event QoS Overrides are reset when a particular knote is no longer |
1011 | * active. Hence this is ignored if FILTER_ACTIVE isn't also returned. |
1012 | * |
1013 | * Races between an f_event() and any other f_* routine asking for |
1014 | * a specific QoS override are handled generically and the filters do not |
1015 | * have to worry about them. |
1016 | * |
1017 | * To use this facility, filters MUST set their f_adjusts_qos bit to true. |
1018 | * |
1019 | * It is expected that filters will return the new QoS they expect to be |
1020 | * applied from any f_* callback except for f_process() where no specific |
1021 | * information should be provided. Filters should not try to hide no-ops, |
1022 | * kevent will already optimize these away. |
1023 | * |
1024 | * Valid: f_touch, f_attach, f_event, f_process |
1025 | * Implicit: - |
1026 | * Ignored: f_peek |
1027 | */ |
1028 | #define FILTER_ACTIVE 0x00000001 |
1029 | #define FILTER_REGISTER_WAIT 0x00000002 |
1030 | #define FILTER_UPDATE_REQ_QOS 0x00000004 |
1031 | #define FILTER_ADJUST_EVENT_QOS_BIT 0x00000008 |
1032 | #define FILTER_ADJUST_EVENT_QOS_MASK 0x00000070 |
1033 | #define FILTER_ADJUST_EVENT_QOS_SHIFT 4 |
1034 | #define FILTER_ADJUST_EVENT_QOS(qos) \ |
1035 | (((qos) << FILTER_ADJUST_EVENT_QOS_SHIFT) | FILTER_ADJUST_EVENT_QOS_BIT) |
1036 | #define FILTER_RESET_EVENT_QOS FILTER_ADJUST_EVENT_QOS_BIT |
1037 | |
1038 | #define filter_call(_ops, call) \ |
1039 | ((_ops)->f_extended_codes ? (_ops)->call : !!((_ops)->call)) |
1040 | |
1041 | SLIST_HEAD(klist, knote); |
1042 | extern void knote_init(void); |
1043 | extern void klist_init(struct klist *list); |
1044 | |
1045 | #define KNOTE(list, hint) knote(list, hint) |
1046 | #define KNOTE_ATTACH(list, kn) knote_attach(list, kn) |
1047 | #define KNOTE_DETACH(list, kn) knote_detach(list, kn) |
1048 | |
1049 | extern void knote(struct klist *list, long hint); |
1050 | extern int knote_attach(struct klist *list, struct knote *kn); |
1051 | extern int knote_detach(struct klist *list, struct knote *kn); |
1052 | extern void knote_vanish(struct klist *list); |
1053 | extern void knote_link_waitqset_lazy_alloc(struct knote *kn); |
1054 | extern boolean_t knote_link_waitqset_should_lazy_alloc(struct knote *kn); |
1055 | extern int knote_link_waitq(struct knote *kn, struct waitq *wq, uint64_t *reserved_link); |
1056 | extern int knote_unlink_waitq(struct knote *kn, struct waitq *wq); |
1057 | extern void knote_fdclose(struct proc *p, int fd); |
1058 | extern void knote_markstayactive(struct knote *kn); |
1059 | extern void knote_clearstayactive(struct knote *kn); |
1060 | extern const struct filterops *knote_fops(struct knote *kn); |
1061 | extern void knote_set_error(struct knote *kn, int error); |
1062 | |
1063 | extern struct turnstile *kqueue_turnstile(struct kqueue *); |
1064 | extern struct turnstile *kqueue_alloc_turnstile(struct kqueue *); |
1065 | |
1066 | int kevent_exit_on_workloop_ownership_leak(thread_t thread); |
1067 | int kevent_proc_copy_uptrs(void *proc, uint64_t *buf, int bufsize); |
1068 | int kevent_copyout_proc_dynkqids(void *proc, user_addr_t ubuf, |
1069 | uint32_t ubufsize, int32_t *nkqueues_out); |
1070 | int kevent_copyout_dynkqinfo(void *proc, kqueue_id_t kq_id, user_addr_t ubuf, |
1071 | uint32_t ubufsize, int32_t *size_out); |
1072 | int kevent_copyout_dynkqextinfo(void *proc, kqueue_id_t kq_id, user_addr_t ubuf, |
1073 | uint32_t ubufsize, int32_t *nknotes_out); |
1074 | |
1075 | #elif defined(KERNEL_PRIVATE) /* !XNU_KERNEL_PRIVATE: kexts still need a klist structure definition */ |
1076 | |
1077 | #include <sys/queue.h> |
1078 | struct proc; |
1079 | struct knote; |
1080 | SLIST_HEAD(klist, knote); |
1081 | |
1082 | #endif /* !XNU_KERNEL_PRIVATE && KERNEL_PRIVATE */ |
1083 | |
1084 | #ifdef KERNEL_PRIVATE |
1085 | #ifdef PRIVATE |
1086 | |
1087 | /* make these private functions available to the pthread kext */ |
1088 | extern int kevent_qos_internal(struct proc *p, int fd, |
1089 | user_addr_t changelist, int nchanges, |
1090 | user_addr_t eventlist, int nevents, |
1091 | user_addr_t data_out, user_size_t *data_available, |
1092 | unsigned int flags, int32_t *retval); |
1093 | |
1094 | extern int kevent_id_internal(struct proc *p, kqueue_id_t *id, |
1095 | user_addr_t changelist, int nchanges, |
1096 | user_addr_t eventlist, int nevents, |
1097 | user_addr_t data_out, user_size_t *data_available, |
1098 | unsigned int flags, int32_t *retval); |
1099 | |
1100 | #endif /* PRIVATE */ |
1101 | #endif /* KERNEL_PRIVATE */ |
1102 | |
1103 | #else /* KERNEL */ |
1104 | |
1105 | #include <sys/types.h> |
1106 | |
1107 | struct timespec; |
1108 | |
1109 | __BEGIN_DECLS |
1110 | int kqueue(void); |
1111 | int kevent(int kq, |
1112 | const struct kevent *changelist, int nchanges, |
1113 | struct kevent *eventlist, int nevents, |
1114 | const struct timespec *timeout); |
1115 | int kevent64(int kq, |
1116 | const struct kevent64_s *changelist, int nchanges, |
1117 | struct kevent64_s *eventlist, int nevents, |
1118 | unsigned int flags, |
1119 | const struct timespec *timeout); |
1120 | |
1121 | #ifdef PRIVATE |
1122 | int kevent_qos(int kq, |
1123 | const struct kevent_qos_s *changelist, int nchanges, |
1124 | struct kevent_qos_s *eventlist, int nevents, |
1125 | void *data_out, size_t *data_available, |
1126 | unsigned int flags); |
1127 | |
1128 | int kevent_id(kqueue_id_t id, |
1129 | const struct kevent_qos_s *changelist, int nchanges, |
1130 | struct kevent_qos_s *eventlist, int nevents, |
1131 | void *data_out, size_t *data_available, |
1132 | unsigned int flags); |
1133 | #endif /* PRIVATE */ |
1134 | |
1135 | __END_DECLS |
1136 | |
1137 | |
1138 | #endif /* KERNEL */ |
1139 | |
1140 | #ifdef PRIVATE |
1141 | |
1142 | /* Flags for pending events notified by kernel via return-to-kernel ast */ |
1143 | #define R2K_WORKLOOP_PENDING_EVENTS 0x1 |
1144 | #define R2K_WORKQ_PENDING_EVENTS 0x2 |
1145 | |
1146 | #endif /* PRIVATE */ |
1147 | |
1148 | #endif /* !_SYS_EVENT_H_ */ |
1149 | |