1/* 2 * Copyright (c) 2000-2011 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/*- 30 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org> 31 * All rights reserved. 32 * 33 * Redistribution and use in source and binary forms, with or without 34 * modification, are permitted provided that the following conditions 35 * are met: 36 * 1. Redistributions of source code must retain the above copyright 37 * notice, this list of conditions and the following disclaimer. 38 * 2. Redistributions in binary form must reproduce the above copyright 39 * notice, this list of conditions and the following disclaimer in the 40 * documentation and/or other materials provided with the distribution. 41 * 42 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 43 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 44 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 45 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 46 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 47 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 48 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 49 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 50 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 51 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 52 * SUCH DAMAGE. 53 */ 54/* 55 * @(#)kern_event.c 1.0 (3/31/2000) 56 */ 57#include <stdint.h> 58 59#include <sys/param.h> 60#include <sys/systm.h> 61#include <sys/filedesc.h> 62#include <sys/kernel.h> 63#include <sys/proc_internal.h> 64#include <sys/kauth.h> 65#include <sys/malloc.h> 66#include <sys/unistd.h> 67#include <sys/file_internal.h> 68#include <sys/fcntl.h> 69#include <sys/select.h> 70#include <sys/queue.h> 71#include <sys/event.h> 72#include <sys/eventvar.h> 73#include <sys/protosw.h> 74#include <sys/socket.h> 75#include <sys/socketvar.h> 76#include <sys/stat.h> 77#include <sys/sysctl.h> 78#include <sys/uio.h> 79#include <sys/sysproto.h> 80#include <sys/user.h> 81#include <sys/vnode_internal.h> 82#include <string.h> 83#include <sys/proc_info.h> 84 85#include <kern/lock.h> 86#include <kern/clock.h> 87#include <kern/thread_call.h> 88#include <kern/sched_prim.h> 89#include <kern/zalloc.h> 90#include <kern/assert.h> 91 92#include <libkern/libkern.h> 93#include "net/net_str_id.h" 94 95#include <mach/task.h> 96 97#if VM_PRESSURE_EVENTS 98#include <kern/vm_pressure.h> 99#endif 100 101MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 102 103#define KQ_EVENT NULL 104 105static inline void kqlock(struct kqueue *kq); 106static inline void kqunlock(struct kqueue *kq); 107 108static int kqlock2knoteuse(struct kqueue *kq, struct knote *kn); 109static int kqlock2knoteusewait(struct kqueue *kq, struct knote *kn); 110static int kqlock2knotedrop(struct kqueue *kq, struct knote *kn); 111static int knoteuse2kqlock(struct kqueue *kq, struct knote *kn); 112 113static void kqueue_wakeup(struct kqueue *kq, int closed); 114static int kqueue_read(struct fileproc *fp, struct uio *uio, 115 int flags, vfs_context_t ctx); 116static int kqueue_write(struct fileproc *fp, struct uio *uio, 117 int flags, vfs_context_t ctx); 118static int kqueue_ioctl(struct fileproc *fp, u_long com, caddr_t data, 119 vfs_context_t ctx); 120static int kqueue_select(struct fileproc *fp, int which, void *wql, 121 vfs_context_t ctx); 122static int kqueue_close(struct fileglob *fg, vfs_context_t ctx); 123static int kqueue_kqfilter(struct fileproc *fp, struct knote *kn, vfs_context_t ctx); 124static int kqueue_drain(struct fileproc *fp, vfs_context_t ctx); 125extern int kqueue_stat(struct fileproc *fp, void *ub, int isstat64, vfs_context_t ctx); 126 127static struct fileops kqueueops = { 128 .fo_read = kqueue_read, 129 .fo_write = kqueue_write, 130 .fo_ioctl = kqueue_ioctl, 131 .fo_select = kqueue_select, 132 .fo_close = kqueue_close, 133 .fo_kqfilter = kqueue_kqfilter, 134 .fo_drain = kqueue_drain, 135}; 136 137static int kevent_internal(struct proc *p, int iskev64, user_addr_t changelist, 138 int nchanges, user_addr_t eventlist, int nevents, int fd, 139 user_addr_t utimeout, unsigned int flags, int32_t *retval); 140static int kevent_copyin(user_addr_t *addrp, struct kevent64_s *kevp, struct proc *p, int iskev64); 141static int kevent_copyout(struct kevent64_s *kevp, user_addr_t *addrp, struct proc *p, int iskev64); 142char * kevent_description(struct kevent64_s *kevp, char *s, size_t n); 143 144static int kevent_callback(struct kqueue *kq, struct kevent64_s *kevp, void *data); 145static void kevent_continue(struct kqueue *kq, void *data, int error); 146static void kqueue_scan_continue(void *contp, wait_result_t wait_result); 147static int kqueue_process(struct kqueue *kq, kevent_callback_t callback, 148 void *data, int *countp, struct proc *p); 149static int kqueue_begin_processing(struct kqueue *kq); 150static void kqueue_end_processing(struct kqueue *kq); 151static int knote_process(struct knote *kn, kevent_callback_t callback, 152 void *data, struct kqtailq *inprocessp, struct proc *p); 153static void knote_put(struct knote *kn); 154static int knote_fdpattach(struct knote *kn, struct filedesc *fdp, struct proc *p); 155static void knote_drop(struct knote *kn, struct proc *p); 156static void knote_activate(struct knote *kn, int); 157static void knote_deactivate(struct knote *kn); 158static void knote_enqueue(struct knote *kn); 159static void knote_dequeue(struct knote *kn); 160static struct knote *knote_alloc(void); 161static void knote_free(struct knote *kn); 162 163static int filt_fileattach(struct knote *kn); 164static struct filterops file_filtops = { 165 .f_isfd = 1, 166 .f_attach = filt_fileattach, 167}; 168 169static void filt_kqdetach(struct knote *kn); 170static int filt_kqueue(struct knote *kn, long hint); 171static struct filterops kqread_filtops = { 172 .f_isfd = 1, 173 .f_detach = filt_kqdetach, 174 .f_event = filt_kqueue, 175}; 176 177/* 178 * placeholder for not-yet-implemented filters 179 */ 180static int filt_badattach(struct knote *kn); 181static struct filterops bad_filtops = { 182 .f_attach = filt_badattach, 183}; 184 185static int filt_procattach(struct knote *kn); 186static void filt_procdetach(struct knote *kn); 187static int filt_proc(struct knote *kn, long hint); 188static struct filterops proc_filtops = { 189 .f_attach = filt_procattach, 190 .f_detach = filt_procdetach, 191 .f_event = filt_proc, 192}; 193 194#if VM_PRESSURE_EVENTS 195static int filt_vmattach(struct knote *kn); 196static void filt_vmdetach(struct knote *kn); 197static int filt_vm(struct knote *kn, long hint); 198static struct filterops vm_filtops = { 199 .f_attach = filt_vmattach, 200 .f_detach = filt_vmdetach, 201 .f_event = filt_vm, 202}; 203#endif /* VM_PRESSURE_EVENTS */ 204 205extern struct filterops fs_filtops; 206 207extern struct filterops sig_filtops; 208 209/* Timer filter */ 210static int filt_timerattach(struct knote *kn); 211static void filt_timerdetach(struct knote *kn); 212static int filt_timer(struct knote *kn, long hint); 213static void filt_timertouch(struct knote *kn, struct kevent64_s *kev, 214 long type); 215static struct filterops timer_filtops = { 216 .f_attach = filt_timerattach, 217 .f_detach = filt_timerdetach, 218 .f_event = filt_timer, 219 .f_touch = filt_timertouch, 220}; 221 222/* Helpers */ 223 224static void filt_timerexpire(void *knx, void *param1); 225static int filt_timervalidate(struct knote *kn); 226static void filt_timerupdate(struct knote *kn); 227static void filt_timercancel(struct knote *kn); 228 229#define TIMER_RUNNING 0x1 230#define TIMER_CANCELWAIT 0x2 231 232static lck_mtx_t _filt_timerlock; 233static void filt_timerlock(void); 234static void filt_timerunlock(void); 235 236static zone_t knote_zone; 237 238#define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 239 240#if 0 241extern struct filterops aio_filtops; 242#endif 243 244/* Mach portset filter */ 245extern struct filterops machport_filtops; 246 247/* User filter */ 248static int filt_userattach(struct knote *kn); 249static void filt_userdetach(struct knote *kn); 250static int filt_user(struct knote *kn, long hint); 251static void filt_usertouch(struct knote *kn, struct kevent64_s *kev, 252 long type); 253static struct filterops user_filtops = { 254 .f_attach = filt_userattach, 255 .f_detach = filt_userdetach, 256 .f_event = filt_user, 257 .f_touch = filt_usertouch, 258}; 259 260/* 261 * Table for for all system-defined filters. 262 */ 263static struct filterops *sysfilt_ops[] = { 264 &file_filtops, /* EVFILT_READ */ 265 &file_filtops, /* EVFILT_WRITE */ 266#if 0 267 &aio_filtops, /* EVFILT_AIO */ 268#else 269 &bad_filtops, /* EVFILT_AIO */ 270#endif 271 &file_filtops, /* EVFILT_VNODE */ 272 &proc_filtops, /* EVFILT_PROC */ 273 &sig_filtops, /* EVFILT_SIGNAL */ 274 &timer_filtops, /* EVFILT_TIMER */ 275 &machport_filtops, /* EVFILT_MACHPORT */ 276 &fs_filtops, /* EVFILT_FS */ 277 &user_filtops, /* EVFILT_USER */ 278 &bad_filtops, /* unused */ 279#if VM_PRESSURE_EVENTS 280 &vm_filtops, /* EVFILT_VM */ 281#else 282 &bad_filtops, /* EVFILT_VM */ 283#endif 284 &file_filtops, /* EVFILT_SOCK */ 285}; 286 287/* 288 * kqueue/note lock attributes and implementations 289 * 290 * kqueues have locks, while knotes have use counts 291 * Most of the knote state is guarded by the object lock. 292 * the knote "inuse" count and status use the kqueue lock. 293 */ 294lck_grp_attr_t * kq_lck_grp_attr; 295lck_grp_t * kq_lck_grp; 296lck_attr_t * kq_lck_attr; 297 298static inline void 299kqlock(struct kqueue *kq) 300{ 301 lck_spin_lock(&kq->kq_lock); 302} 303 304static inline void 305kqunlock(struct kqueue *kq) 306{ 307 lck_spin_unlock(&kq->kq_lock); 308} 309 310/* 311 * Convert a kq lock to a knote use referece. 312 * 313 * If the knote is being dropped, we can't get 314 * a use reference, so just return with it 315 * still locked. 316 * 317 * - kq locked at entry 318 * - unlock on exit if we get the use reference 319 */ 320static int 321kqlock2knoteuse(struct kqueue *kq, struct knote *kn) 322{ 323 if (kn->kn_status & KN_DROPPING) 324 return 0; 325 kn->kn_inuse++; 326 kqunlock(kq); 327 return 1; 328 } 329 330/* 331 * Convert a kq lock to a knote use referece, 332 * but wait for attach and drop events to complete. 333 * 334 * If the knote is being dropped, we can't get 335 * a use reference, so just return with it 336 * still locked. 337 * 338 * - kq locked at entry 339 * - kq always unlocked on exit 340 */ 341static int 342kqlock2knoteusewait(struct kqueue *kq, struct knote *kn) 343{ 344 if ((kn->kn_status & (KN_DROPPING | KN_ATTACHING)) != 0) { 345 kn->kn_status |= KN_USEWAIT; 346 wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, &kn->kn_status, THREAD_UNINT, 0); 347 kqunlock(kq); 348 thread_block(THREAD_CONTINUE_NULL); 349 return 0; 350 } 351 kn->kn_inuse++; 352 kqunlock(kq); 353 return 1; 354 } 355 356 357/* 358 * Convert from a knote use reference back to kq lock. 359 * 360 * Drop a use reference and wake any waiters if 361 * this is the last one. 362 * 363 * The exit return indicates if the knote is 364 * still alive - but the kqueue lock is taken 365 * unconditionally. 366 */ 367static int 368knoteuse2kqlock(struct kqueue *kq, struct knote *kn) 369{ 370 kqlock(kq); 371 if (--kn->kn_inuse == 0) { 372 if ((kn->kn_status & KN_ATTACHING) != 0) { 373 kn->kn_status &= ~KN_ATTACHING; 374 } 375 if ((kn->kn_status & KN_USEWAIT) != 0) { 376 kn->kn_status &= ~KN_USEWAIT; 377 wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, &kn->kn_status, THREAD_AWAKENED); 378 } 379 } 380 return ((kn->kn_status & KN_DROPPING) == 0); 381 } 382 383/* 384 * Convert a kq lock to a knote drop referece. 385 * 386 * If the knote is in use, wait for the use count 387 * to subside. We first mark our intention to drop 388 * it - keeping other users from "piling on." 389 * If we are too late, we have to wait for the 390 * other drop to complete. 391 * 392 * - kq locked at entry 393 * - always unlocked on exit. 394 * - caller can't hold any locks that would prevent 395 * the other dropper from completing. 396 */ 397static int 398kqlock2knotedrop(struct kqueue *kq, struct knote *kn) 399{ 400 int oktodrop; 401 402 oktodrop = ((kn->kn_status & (KN_DROPPING | KN_ATTACHING)) == 0); 403 kn->kn_status |= KN_DROPPING; 404 if (oktodrop) { 405 if (kn->kn_inuse == 0) { 406 kqunlock(kq); 407 return oktodrop; 408 } 409 } 410 kn->kn_status |= KN_USEWAIT; 411 wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, &kn->kn_status, THREAD_UNINT, 0); 412 kqunlock(kq); 413 thread_block(THREAD_CONTINUE_NULL); 414 return oktodrop; 415} 416 417/* 418 * Release a knote use count reference. 419 */ 420static void 421knote_put(struct knote *kn) 422{ 423 struct kqueue *kq = kn->kn_kq; 424 425 kqlock(kq); 426 if (--kn->kn_inuse == 0) { 427 if ((kn->kn_status & KN_USEWAIT) != 0) { 428 kn->kn_status &= ~KN_USEWAIT; 429 wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, &kn->kn_status, THREAD_AWAKENED); 430 } 431 } 432 kqunlock(kq); 433 } 434 435static int 436filt_fileattach(struct knote *kn) 437{ 438 439 return (fo_kqfilter(kn->kn_fp, kn, vfs_context_current())); 440} 441 442#define f_flag f_fglob->fg_flag 443#define f_type f_fglob->fg_type 444#define f_msgcount f_fglob->fg_msgcount 445#define f_cred f_fglob->fg_cred 446#define f_ops f_fglob->fg_ops 447#define f_offset f_fglob->fg_offset 448#define f_data f_fglob->fg_data 449 450static void 451filt_kqdetach(struct knote *kn) 452{ 453 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 454 455 kqlock(kq); 456 KNOTE_DETACH(&kq->kq_sel.si_note, kn); 457 kqunlock(kq); 458} 459 460/*ARGSUSED*/ 461static int 462filt_kqueue(struct knote *kn, __unused long hint) 463{ 464 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 465 466 kn->kn_data = kq->kq_count; 467 return (kn->kn_data > 0); 468} 469 470static int 471filt_procattach(struct knote *kn) 472{ 473 struct proc *p; 474 475 assert(PID_MAX < NOTE_PDATAMASK); 476 477 if ((kn->kn_sfflags & (NOTE_TRACK | NOTE_TRACKERR | NOTE_CHILD)) != 0) 478 return(ENOTSUP); 479 480 p = proc_find(kn->kn_id); 481 if (p == NULL) { 482 return (ESRCH); 483 } 484 485 const int NoteExitStatusBits = NOTE_EXIT | NOTE_EXITSTATUS; 486 487 if ((kn->kn_sfflags & NoteExitStatusBits) == NoteExitStatusBits) 488 do { 489 pid_t selfpid = proc_selfpid(); 490 491 if (p->p_ppid == selfpid) 492 break; /* parent => ok */ 493 494 if ((p->p_lflag & P_LTRACED) != 0 && 495 (p->p_oppid == selfpid)) 496 break; /* parent-in-waiting => ok */ 497 498 proc_rele(p); 499 return (EACCES); 500 } while (0); 501 502 proc_klist_lock(); 503 504 kn->kn_flags |= EV_CLEAR; /* automatically set */ 505 kn->kn_ptr.p_proc = p; /* store the proc handle */ 506 507 KNOTE_ATTACH(&p->p_klist, kn); 508 509 proc_klist_unlock(); 510 511 proc_rele(p); 512 513 return (0); 514} 515 516/* 517 * The knote may be attached to a different process, which may exit, 518 * leaving nothing for the knote to be attached to. In that case, 519 * the pointer to the process will have already been nulled out. 520 */ 521static void 522filt_procdetach(struct knote *kn) 523{ 524 struct proc *p; 525 526 proc_klist_lock(); 527 528 p = kn->kn_ptr.p_proc; 529 if (p != PROC_NULL) { 530 kn->kn_ptr.p_proc = PROC_NULL; 531 KNOTE_DETACH(&p->p_klist, kn); 532 } 533 534 proc_klist_unlock(); 535} 536 537static int 538filt_proc(struct knote *kn, long hint) 539{ 540 /* hint is 0 when called from above */ 541 if (hint != 0) { 542 u_int event; 543 544 /* ALWAYS CALLED WITH proc_klist_lock when (hint != 0) */ 545 546 /* 547 * mask off extra data 548 */ 549 event = (u_int)hint & NOTE_PCTRLMASK; 550 551 /* 552 * termination lifecycle events can happen while a debugger 553 * has reparented a process, in which case notifications 554 * should be quashed except to the tracing parent. When 555 * the debugger reaps the child (either via wait4(2) or 556 * process exit), the child will be reparented to the original 557 * parent and these knotes re-fired. 558 */ 559 if (event & NOTE_EXIT) { 560 if ((kn->kn_ptr.p_proc->p_oppid != 0) 561 && (kn->kn_kq->kq_p->p_pid != kn->kn_ptr.p_proc->p_ppid)) { 562 /* 563 * This knote is not for the current ptrace(2) parent, ignore. 564 */ 565 return 0; 566 } 567 } 568 569 /* 570 * if the user is interested in this event, record it. 571 */ 572 if (kn->kn_sfflags & event) 573 kn->kn_fflags |= event; 574 575 if (event == NOTE_REAP || (event == NOTE_EXIT && !(kn->kn_sfflags & NOTE_REAP))) { 576 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 577 } 578 if ((event == NOTE_EXIT) && ((kn->kn_sfflags & NOTE_EXITSTATUS) != 0)) { 579 kn->kn_fflags |= NOTE_EXITSTATUS; 580 kn->kn_data = (hint & NOTE_PDATAMASK); 581 } 582 if ((event == NOTE_RESOURCEEND) && ((kn->kn_sfflags & NOTE_RESOURCEEND) != 0)) { 583 kn->kn_fflags |= NOTE_RESOURCEEND; 584 kn->kn_data = (hint & NOTE_PDATAMASK); 585 } 586#if CONFIG_EMBEDDED 587 /* If the event is one of the APPSTATE events,remove the rest */ 588 if (((event & NOTE_APPALLSTATES) != 0) && ((kn->kn_sfflags & NOTE_APPALLSTATES) != 0)) { 589 /* only one state at a time */ 590 kn->kn_fflags &= ~NOTE_APPALLSTATES; 591 kn->kn_fflags |= event; 592 } 593#endif /* CONFIG_EMBEDDED */ 594 } 595 596 /* atomic check, no locking need when called from above */ 597 return (kn->kn_fflags != 0); 598} 599 600#if VM_PRESSURE_EVENTS 601/* 602 * Virtual memory kevents 603 * 604 * author: Matt Jacobson [matthew_jacobson@apple.com] 605 */ 606 607static int 608filt_vmattach(struct knote *kn) 609{ 610 /* 611 * The note will be cleared once the information has been flushed to the client. 612 * If there is still pressure, we will be re-alerted. 613 */ 614 kn->kn_flags |= EV_CLEAR; 615 616 return vm_knote_register(kn); 617} 618 619static void 620filt_vmdetach(struct knote *kn) 621{ 622 vm_knote_unregister(kn); 623} 624 625static int 626filt_vm(struct knote *kn, long hint) 627{ 628 /* hint == 0 means this is just an alive? check (always true) */ 629 if (hint != 0) { 630 const pid_t pid = (pid_t)hint; 631 if ((kn->kn_sfflags & NOTE_VM_PRESSURE) && (kn->kn_kq->kq_p->p_pid == pid)) { 632 kn->kn_fflags |= NOTE_VM_PRESSURE; 633 } 634 } 635 636 return (kn->kn_fflags != 0); 637} 638#endif /* VM_PRESSURE_EVENTS */ 639 640/* 641 * filt_timervalidate - process data from user 642 * 643 * Converts to either interval or deadline format. 644 * 645 * The saved-data field in the knote contains the 646 * time value. The saved filter-flags indicates 647 * the unit of measurement. 648 * 649 * After validation, either the saved-data field 650 * contains the interval in absolute time, or ext[0] 651 * contains the expected deadline. If that deadline 652 * is in the past, ext[0] is 0. 653 * 654 * Returns EINVAL for unrecognized units of time. 655 * 656 * Timer filter lock is held. 657 * 658 */ 659static int 660filt_timervalidate(struct knote *kn) 661{ 662 uint64_t multiplier; 663 uint64_t raw; 664 665 switch (kn->kn_sfflags & (NOTE_SECONDS|NOTE_USECONDS|NOTE_NSECONDS)) { 666 case NOTE_SECONDS: 667 multiplier = NSEC_PER_SEC; 668 break; 669 case NOTE_USECONDS: 670 multiplier = NSEC_PER_USEC; 671 break; 672 case NOTE_NSECONDS: 673 multiplier = 1; 674 break; 675 case 0: /* milliseconds (default) */ 676 multiplier = NSEC_PER_SEC / 1000; 677 break; 678 default: 679 return EINVAL; 680 } 681 682 nanoseconds_to_absolutetime((uint64_t)kn->kn_sdata * multiplier, &raw); 683 684 kn->kn_ext[0] = 0; 685 kn->kn_sdata = 0; 686 687 if (kn->kn_sfflags & NOTE_ABSOLUTE) { 688 clock_sec_t seconds; 689 clock_nsec_t nanoseconds; 690 uint64_t now; 691 692 clock_get_calendar_nanotime(&seconds, &nanoseconds); 693 nanoseconds_to_absolutetime((uint64_t)seconds * NSEC_PER_SEC + 694 nanoseconds, &now); 695 696 if (raw < now) { 697 /* time has already passed */ 698 kn->kn_ext[0] = 0; 699 } else { 700 raw -= now; 701 clock_absolutetime_interval_to_deadline(raw, 702 &kn->kn_ext[0]); 703 } 704 } else { 705 kn->kn_sdata = raw; 706 } 707 708 return 0; 709} 710 711/* 712 * filt_timerupdate - compute the next deadline 713 * 714 * Repeating timers store their interval in kn_sdata. Absolute 715 * timers have already calculated the deadline, stored in ext[0]. 716 * 717 * On return, the next deadline (or zero if no deadline is needed) 718 * is stored in kn_ext[0]. 719 * 720 * Timer filter lock is held. 721 */ 722static void 723filt_timerupdate(struct knote *kn) 724{ 725 /* if there's no interval, deadline is just in kn_ext[0] */ 726 if (kn->kn_sdata == 0) 727 return; 728 729 /* if timer hasn't fired before, fire in interval nsecs */ 730 if (kn->kn_ext[0] == 0) { 731 clock_absolutetime_interval_to_deadline(kn->kn_sdata, 732 &kn->kn_ext[0]); 733 } else { 734 /* 735 * If timer has fired before, schedule the next pop 736 * relative to the last intended deadline. 737 * 738 * We could check for whether the deadline has expired, 739 * but the thread call layer can handle that. 740 */ 741 kn->kn_ext[0] += kn->kn_sdata; 742 } 743} 744 745/* 746 * filt_timerexpire - the timer callout routine 747 * 748 * Just propagate the timer event into the knote 749 * filter routine (by going through the knote 750 * synchronization point). Pass a hint to 751 * indicate this is a real event, not just a 752 * query from above. 753 */ 754static void 755filt_timerexpire(void *knx, __unused void *spare) 756{ 757 struct klist timer_list; 758 struct knote *kn = knx; 759 760 filt_timerlock(); 761 762 kn->kn_hookid &= ~TIMER_RUNNING; 763 764 /* no "object" for timers, so fake a list */ 765 SLIST_INIT(&timer_list); 766 SLIST_INSERT_HEAD(&timer_list, kn, kn_selnext); 767 KNOTE(&timer_list, 1); 768 769 /* if someone is waiting for timer to pop */ 770 if (kn->kn_hookid & TIMER_CANCELWAIT) { 771 struct kqueue *kq = kn->kn_kq; 772 wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, &kn->kn_hook, 773 THREAD_AWAKENED); 774 } 775 776 filt_timerunlock(); 777} 778 779/* 780 * Cancel a running timer (or wait for the pop). 781 * Timer filter lock is held. 782 */ 783static void 784filt_timercancel(struct knote *kn) 785{ 786 struct kqueue *kq = kn->kn_kq; 787 thread_call_t callout = kn->kn_hook; 788 boolean_t cancelled; 789 790 if (kn->kn_hookid & TIMER_RUNNING) { 791 /* cancel the callout if we can */ 792 cancelled = thread_call_cancel(callout); 793 if (cancelled) { 794 kn->kn_hookid &= ~TIMER_RUNNING; 795 } else { 796 /* we have to wait for the expire routine. */ 797 kn->kn_hookid |= TIMER_CANCELWAIT; 798 wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, 799 &kn->kn_hook, THREAD_UNINT, 0); 800 filt_timerunlock(); 801 thread_block(THREAD_CONTINUE_NULL); 802 filt_timerlock(); 803 assert((kn->kn_hookid & TIMER_RUNNING) == 0); 804 } 805 } 806} 807 808/* 809 * Allocate a thread call for the knote's lifetime, and kick off the timer. 810 */ 811static int 812filt_timerattach(struct knote *kn) 813{ 814 thread_call_t callout; 815 int error; 816 817 callout = thread_call_allocate(filt_timerexpire, kn); 818 if (NULL == callout) 819 return (ENOMEM); 820 821 filt_timerlock(); 822 error = filt_timervalidate(kn); 823 if (error) { 824 filt_timerunlock(); 825 return (error); 826 } 827 828 kn->kn_hook = (void*)callout; 829 kn->kn_hookid = 0; 830 831 /* absolute=EV_ONESHOT */ 832 if (kn->kn_sfflags & NOTE_ABSOLUTE) 833 kn->kn_flags |= EV_ONESHOT; 834 835 filt_timerupdate(kn); 836 if (kn->kn_ext[0]) { 837 kn->kn_flags |= EV_CLEAR; 838 thread_call_enter_delayed(callout, kn->kn_ext[0]); 839 kn->kn_hookid |= TIMER_RUNNING; 840 } else { 841 /* fake immediate */ 842 kn->kn_data = 1; 843 } 844 845 filt_timerunlock(); 846 return (0); 847} 848 849/* 850 * Shut down the timer if it's running, and free the callout. 851 */ 852static void 853filt_timerdetach(struct knote *kn) 854{ 855 thread_call_t callout; 856 857 filt_timerlock(); 858 859 callout = (thread_call_t)kn->kn_hook; 860 filt_timercancel(kn); 861 862 filt_timerunlock(); 863 864 thread_call_free(callout); 865} 866 867 868 869static int 870filt_timer(struct knote *kn, long hint) 871{ 872 int result; 873 874 if (hint) { 875 /* real timer pop -- timer lock held by filt_timerexpire */ 876 877 kn->kn_data++; 878 879 if (((kn->kn_hookid & TIMER_CANCELWAIT) == 0) && 880 ((kn->kn_flags & EV_ONESHOT) == 0)) { 881 882 /* evaluate next time to fire */ 883 filt_timerupdate(kn); 884 885 if (kn->kn_ext[0]) { 886 /* keep the callout and re-arm */ 887 thread_call_enter_delayed(kn->kn_hook, 888 kn->kn_ext[0]); 889 kn->kn_hookid |= TIMER_RUNNING; 890 } 891 } 892 893 return 1; 894 } 895 896 /* user-query */ 897 filt_timerlock(); 898 899 result = (kn->kn_data != 0); 900 901 filt_timerunlock(); 902 return result; 903} 904 905 906/* 907 * filt_timertouch - update knote with new user input 908 * 909 * Cancel and restart the timer based on new user data. When 910 * the user picks up a knote, clear the count of how many timer 911 * pops have gone off (in kn_data). 912 */ 913static void 914filt_timertouch(struct knote *kn, struct kevent64_s *kev, long type) 915{ 916 int error; 917 filt_timerlock(); 918 919 switch (type) { 920 case EVENT_REGISTER: 921 /* cancel current call */ 922 filt_timercancel(kn); 923 924 /* recalculate deadline */ 925 kn->kn_sdata = kev->data; 926 kn->kn_sfflags = kev->fflags; 927 928 error = filt_timervalidate(kn); 929 if (error) { 930 /* no way to report error, so mark it in the knote */ 931 kn->kn_flags |= EV_ERROR; 932 kn->kn_data = error; 933 break; 934 } 935 936 /* start timer if necessary */ 937 filt_timerupdate(kn); 938 if (kn->kn_ext[0]) { 939 thread_call_enter_delayed(kn->kn_hook, kn->kn_ext[0]); 940 kn->kn_hookid |= TIMER_RUNNING; 941 } else { 942 /* pretend the timer has fired */ 943 kn->kn_data = 1; 944 } 945 946 break; 947 948 case EVENT_PROCESS: 949 /* reset the timer pop count in kn_data */ 950 *kev = kn->kn_kevent; 951 kev->ext[0] = 0; 952 kn->kn_data = 0; 953 if (kn->kn_flags & EV_CLEAR) 954 kn->kn_fflags = 0; 955 break; 956 default: 957 panic("filt_timertouch() - invalid type (%ld)", type); 958 break; 959 } 960 961 filt_timerunlock(); 962} 963 964static void 965filt_timerlock(void) 966{ 967 lck_mtx_lock(&_filt_timerlock); 968} 969 970static void 971filt_timerunlock(void) 972{ 973 lck_mtx_unlock(&_filt_timerlock); 974} 975 976static int 977filt_userattach(struct knote *kn) 978{ 979 /* EVFILT_USER knotes are not attached to anything in the kernel */ 980 kn->kn_hook = NULL; 981 if (kn->kn_fflags & NOTE_TRIGGER) { 982 kn->kn_hookid = 1; 983 } else { 984 kn->kn_hookid = 0; 985 } 986 return 0; 987} 988 989static void 990filt_userdetach(__unused struct knote *kn) 991{ 992 /* EVFILT_USER knotes are not attached to anything in the kernel */ 993} 994 995static int 996filt_user(struct knote *kn, __unused long hint) 997{ 998 return kn->kn_hookid; 999} 1000 1001static void 1002filt_usertouch(struct knote *kn, struct kevent64_s *kev, long type) 1003{ 1004 uint32_t ffctrl; 1005 switch (type) { 1006 case EVENT_REGISTER: 1007 if (kev->fflags & NOTE_TRIGGER) { 1008 kn->kn_hookid = 1; 1009 } 1010 1011 ffctrl = kev->fflags & NOTE_FFCTRLMASK; 1012 kev->fflags &= NOTE_FFLAGSMASK; 1013 switch (ffctrl) { 1014 case NOTE_FFNOP: 1015 break; 1016 case NOTE_FFAND: 1017 OSBitAndAtomic(kev->fflags, &kn->kn_sfflags); 1018 break; 1019 case NOTE_FFOR: 1020 OSBitOrAtomic(kev->fflags, &kn->kn_sfflags); 1021 break; 1022 case NOTE_FFCOPY: 1023 kn->kn_sfflags = kev->fflags; 1024 break; 1025 } 1026 kn->kn_sdata = kev->data; 1027 break; 1028 case EVENT_PROCESS: 1029 *kev = kn->kn_kevent; 1030 kev->fflags = (volatile UInt32)kn->kn_sfflags; 1031 kev->data = kn->kn_sdata; 1032 if (kn->kn_flags & EV_CLEAR) { 1033 kn->kn_hookid = 0; 1034 kn->kn_data = 0; 1035 kn->kn_fflags = 0; 1036 } 1037 break; 1038 default: 1039 panic("filt_usertouch() - invalid type (%ld)", type); 1040 break; 1041 } 1042} 1043 1044/* 1045 * JMM - placeholder for not-yet-implemented filters 1046 */ 1047static int 1048filt_badattach(__unused struct knote *kn) 1049{ 1050 return(ENOTSUP); 1051} 1052 1053 1054struct kqueue * 1055kqueue_alloc(struct proc *p) 1056{ 1057 struct filedesc *fdp = p->p_fd; 1058 struct kqueue *kq; 1059 1060 MALLOC_ZONE(kq, struct kqueue *, sizeof(struct kqueue), M_KQUEUE, M_WAITOK); 1061 if (kq != NULL) { 1062 wait_queue_set_t wqs; 1063 1064 wqs = wait_queue_set_alloc(SYNC_POLICY_FIFO | SYNC_POLICY_PREPOST); 1065 if (wqs != NULL) { 1066 bzero(kq, sizeof(struct kqueue)); 1067 lck_spin_init(&kq->kq_lock, kq_lck_grp, kq_lck_attr); 1068 TAILQ_INIT(&kq->kq_head); 1069 kq->kq_wqs = wqs; 1070 kq->kq_p = p; 1071 } else { 1072 FREE_ZONE(kq, sizeof(struct kqueue), M_KQUEUE); 1073 } 1074 } 1075 1076 if (fdp->fd_knlistsize < 0) { 1077 proc_fdlock(p); 1078 if (fdp->fd_knlistsize < 0) 1079 fdp->fd_knlistsize = 0; /* this process has had a kq */ 1080 proc_fdunlock(p); 1081 } 1082 1083 return kq; 1084} 1085 1086 1087/* 1088 * kqueue_dealloc - detach all knotes from a kqueue and free it 1089 * 1090 * We walk each list looking for knotes referencing this 1091 * this kqueue. If we find one, we try to drop it. But 1092 * if we fail to get a drop reference, that will wait 1093 * until it is dropped. So, we can just restart again 1094 * safe in the assumption that the list will eventually 1095 * not contain any more references to this kqueue (either 1096 * we dropped them all, or someone else did). 1097 * 1098 * Assumes no new events are being added to the kqueue. 1099 * Nothing locked on entry or exit. 1100 */ 1101void 1102kqueue_dealloc(struct kqueue *kq) 1103{ 1104 struct proc *p = kq->kq_p; 1105 struct filedesc *fdp = p->p_fd; 1106 struct knote *kn; 1107 int i; 1108 1109 proc_fdlock(p); 1110 for (i = 0; i < fdp->fd_knlistsize; i++) { 1111 kn = SLIST_FIRST(&fdp->fd_knlist[i]); 1112 while (kn != NULL) { 1113 if (kq == kn->kn_kq) { 1114 kqlock(kq); 1115 proc_fdunlock(p); 1116 /* drop it ourselves or wait */ 1117 if (kqlock2knotedrop(kq, kn)) { 1118 kn->kn_fop->f_detach(kn); 1119 knote_drop(kn, p); 1120 } 1121 proc_fdlock(p); 1122 /* start over at beginning of list */ 1123 kn = SLIST_FIRST(&fdp->fd_knlist[i]); 1124 continue; 1125 } 1126 kn = SLIST_NEXT(kn, kn_link); 1127 } 1128 } 1129 if (fdp->fd_knhashmask != 0) { 1130 for (i = 0; i < (int)fdp->fd_knhashmask + 1; i++) { 1131 kn = SLIST_FIRST(&fdp->fd_knhash[i]); 1132 while (kn != NULL) { 1133 if (kq == kn->kn_kq) { 1134 kqlock(kq); 1135 proc_fdunlock(p); 1136 /* drop it ourselves or wait */ 1137 if (kqlock2knotedrop(kq, kn)) { 1138 kn->kn_fop->f_detach(kn); 1139 knote_drop(kn, p); 1140 } 1141 proc_fdlock(p); 1142 /* start over at beginning of list */ 1143 kn = SLIST_FIRST(&fdp->fd_knhash[i]); 1144 continue; 1145 } 1146 kn = SLIST_NEXT(kn, kn_link); 1147 } 1148 } 1149 } 1150 proc_fdunlock(p); 1151 1152 /* 1153 * before freeing the wait queue set for this kqueue, 1154 * make sure it is unlinked from all its containing (select) sets. 1155 */ 1156 wait_queue_unlink_all((wait_queue_t)kq->kq_wqs); 1157 wait_queue_set_free(kq->kq_wqs); 1158 lck_spin_destroy(&kq->kq_lock, kq_lck_grp); 1159 FREE_ZONE(kq, sizeof(struct kqueue), M_KQUEUE); 1160} 1161 1162int 1163kqueue(struct proc *p, __unused struct kqueue_args *uap, int32_t *retval) 1164{ 1165 struct kqueue *kq; 1166 struct fileproc *fp; 1167 int fd, error; 1168 1169 error = falloc(p, &fp, &fd, vfs_context_current()); 1170 if (error) { 1171 return (error); 1172 } 1173 1174 kq = kqueue_alloc(p); 1175 if (kq == NULL) { 1176 fp_free(p, fd, fp); 1177 return (ENOMEM); 1178 } 1179 1180 fp->f_flag = FREAD | FWRITE; 1181 fp->f_type = DTYPE_KQUEUE; 1182 fp->f_ops = &kqueueops; 1183 fp->f_data = (caddr_t)kq; 1184 1185 proc_fdlock(p); 1186 procfdtbl_releasefd(p, fd, NULL); 1187 fp_drop(p, fd, fp, 1); 1188 proc_fdunlock(p); 1189 1190 *retval = fd; 1191 return (error); 1192} 1193 1194static int 1195kevent_copyin(user_addr_t *addrp, struct kevent64_s *kevp, struct proc *p, int iskev64) 1196{ 1197 int advance; 1198 int error; 1199 1200 if (iskev64) { 1201 advance = sizeof(struct kevent64_s); 1202 error = copyin(*addrp, (caddr_t)kevp, advance); 1203 } else if (IS_64BIT_PROCESS(p)) { 1204 struct user64_kevent kev64; 1205 bzero(kevp, sizeof(struct kevent64_s)); 1206 1207 advance = sizeof(kev64); 1208 error = copyin(*addrp, (caddr_t)&kev64, advance); 1209 if (error) 1210 return error; 1211 kevp->ident = kev64.ident; 1212 kevp->filter = kev64.filter; 1213 kevp->flags = kev64.flags; 1214 kevp->fflags = kev64.fflags; 1215 kevp->data = kev64.data; 1216 kevp->udata = kev64.udata; 1217 } else { 1218 struct user32_kevent kev32; 1219 bzero(kevp, sizeof(struct kevent64_s)); 1220 1221 advance = sizeof(kev32); 1222 error = copyin(*addrp, (caddr_t)&kev32, advance); 1223 if (error) 1224 return error; 1225 kevp->ident = (uintptr_t)kev32.ident; 1226 kevp->filter = kev32.filter; 1227 kevp->flags = kev32.flags; 1228 kevp->fflags = kev32.fflags; 1229 kevp->data = (intptr_t)kev32.data; 1230 kevp->udata = CAST_USER_ADDR_T(kev32.udata); 1231 } 1232 if (!error) 1233 *addrp += advance; 1234 return error; 1235} 1236 1237static int 1238kevent_copyout(struct kevent64_s *kevp, user_addr_t *addrp, struct proc *p, int iskev64) 1239{ 1240 int advance; 1241 int error; 1242 1243 if (iskev64) { 1244 advance = sizeof(struct kevent64_s); 1245 error = copyout((caddr_t)kevp, *addrp, advance); 1246 } else if (IS_64BIT_PROCESS(p)) { 1247 struct user64_kevent kev64; 1248 1249 /* 1250 * deal with the special case of a user-supplied 1251 * value of (uintptr_t)-1. 1252 */ 1253 kev64.ident = (kevp->ident == (uintptr_t)-1) ? 1254 (uint64_t)-1LL : (uint64_t)kevp->ident; 1255 1256 kev64.filter = kevp->filter; 1257 kev64.flags = kevp->flags; 1258 kev64.fflags = kevp->fflags; 1259 kev64.data = (int64_t) kevp->data; 1260 kev64.udata = kevp->udata; 1261 advance = sizeof(kev64); 1262 error = copyout((caddr_t)&kev64, *addrp, advance); 1263 } else { 1264 struct user32_kevent kev32; 1265 1266 kev32.ident = (uint32_t)kevp->ident; 1267 kev32.filter = kevp->filter; 1268 kev32.flags = kevp->flags; 1269 kev32.fflags = kevp->fflags; 1270 kev32.data = (int32_t)kevp->data; 1271 kev32.udata = kevp->udata; 1272 advance = sizeof(kev32); 1273 error = copyout((caddr_t)&kev32, *addrp, advance); 1274 } 1275 if (!error) 1276 *addrp += advance; 1277 return error; 1278} 1279 1280/* 1281 * kevent_continue - continue a kevent syscall after blocking 1282 * 1283 * assume we inherit a use count on the kq fileglob. 1284 */ 1285 1286static void 1287kevent_continue(__unused struct kqueue *kq, void *data, int error) 1288{ 1289 struct _kevent *cont_args; 1290 struct fileproc *fp; 1291 int32_t *retval; 1292 int noutputs; 1293 int fd; 1294 struct proc *p = current_proc(); 1295 1296 cont_args = (struct _kevent *)data; 1297 noutputs = cont_args->eventout; 1298 retval = cont_args->retval; 1299 fd = cont_args->fd; 1300 fp = cont_args->fp; 1301 1302 fp_drop(p, fd, fp, 0); 1303 1304 /* don't restart after signals... */ 1305 if (error == ERESTART) 1306 error = EINTR; 1307 else if (error == EWOULDBLOCK) 1308 error = 0; 1309 if (error == 0) 1310 *retval = noutputs; 1311 unix_syscall_return(error); 1312} 1313 1314/* 1315 * kevent - [syscall] register and wait for kernel events 1316 * 1317 */ 1318int 1319kevent(struct proc *p, struct kevent_args *uap, int32_t *retval) 1320{ 1321 return kevent_internal(p, 1322 0, 1323 uap->changelist, 1324 uap->nchanges, 1325 uap->eventlist, 1326 uap->nevents, 1327 uap->fd, 1328 uap->timeout, 1329 0, /* no flags from old kevent() call */ 1330 retval); 1331} 1332 1333int 1334kevent64(struct proc *p, struct kevent64_args *uap, int32_t *retval) 1335{ 1336 return kevent_internal(p, 1337 1, 1338 uap->changelist, 1339 uap->nchanges, 1340 uap->eventlist, 1341 uap->nevents, 1342 uap->fd, 1343 uap->timeout, 1344 uap->flags, 1345 retval); 1346} 1347 1348static int 1349kevent_internal(struct proc *p, int iskev64, user_addr_t changelist, 1350 int nchanges, user_addr_t ueventlist, int nevents, int fd, 1351 user_addr_t utimeout, __unused unsigned int flags, 1352 int32_t *retval) 1353{ 1354 struct _kevent *cont_args; 1355 uthread_t ut; 1356 struct kqueue *kq; 1357 struct fileproc *fp; 1358 struct kevent64_s kev; 1359 int error, noutputs; 1360 struct timeval atv; 1361 1362 /* convert timeout to absolute - if we have one */ 1363 if (utimeout != USER_ADDR_NULL) { 1364 struct timeval rtv; 1365 if (IS_64BIT_PROCESS(p)) { 1366 struct user64_timespec ts; 1367 error = copyin(utimeout, &ts, sizeof(ts)); 1368 if ((ts.tv_sec & 0xFFFFFFFF00000000ull) != 0) 1369 error = EINVAL; 1370 else 1371 TIMESPEC_TO_TIMEVAL(&rtv, &ts); 1372 } else { 1373 struct user32_timespec ts; 1374 error = copyin(utimeout, &ts, sizeof(ts)); 1375 TIMESPEC_TO_TIMEVAL(&rtv, &ts); 1376 } 1377 if (error) 1378 return error; 1379 if (itimerfix(&rtv)) 1380 return EINVAL; 1381 getmicrouptime(&atv); 1382 timevaladd(&atv, &rtv); 1383 } else { 1384 atv.tv_sec = 0; 1385 atv.tv_usec = 0; 1386 } 1387 1388 /* get a usecount for the kq itself */ 1389 if ((error = fp_getfkq(p, fd, &fp, &kq)) != 0) 1390 return(error); 1391 1392 /* each kq should only be used for events of one type */ 1393 kqlock(kq); 1394 if (kq->kq_state & (KQ_KEV32 | KQ_KEV64)) { 1395 if (((iskev64 && (kq->kq_state & KQ_KEV32)) || 1396 (!iskev64 && (kq->kq_state & KQ_KEV64)))) { 1397 error = EINVAL; 1398 kqunlock(kq); 1399 goto errorout; 1400 } 1401 } else { 1402 kq->kq_state |= (iskev64 ? KQ_KEV64 : KQ_KEV32); 1403 } 1404 kqunlock(kq); 1405 1406 /* register all the change requests the user provided... */ 1407 noutputs = 0; 1408 while (nchanges > 0 && error == 0) { 1409 error = kevent_copyin(&changelist, &kev, p, iskev64); 1410 if (error) 1411 break; 1412 1413 kev.flags &= ~EV_SYSFLAGS; 1414 error = kevent_register(kq, &kev, p); 1415 if ((error || (kev.flags & EV_RECEIPT)) && nevents > 0) { 1416 kev.flags = EV_ERROR; 1417 kev.data = error; 1418 error = kevent_copyout(&kev, &ueventlist, p, iskev64); 1419 if (error == 0) { 1420 nevents--; 1421 noutputs++; 1422 } 1423 } 1424 nchanges--; 1425 } 1426 1427 /* store the continuation/completion data in the uthread */ 1428 ut = (uthread_t)get_bsdthread_info(current_thread()); 1429 cont_args = &ut->uu_kevent.ss_kevent; 1430 cont_args->fp = fp; 1431 cont_args->fd = fd; 1432 cont_args->retval = retval; 1433 cont_args->eventlist = ueventlist; 1434 cont_args->eventcount = nevents; 1435 cont_args->eventout = noutputs; 1436 cont_args->eventsize = iskev64; 1437 1438 if (nevents > 0 && noutputs == 0 && error == 0) 1439 error = kqueue_scan(kq, kevent_callback, 1440 kevent_continue, cont_args, 1441 &atv, p); 1442 kevent_continue(kq, cont_args, error); 1443 1444errorout: 1445 fp_drop(p, fd, fp, 0); 1446 return error; 1447} 1448 1449 1450/* 1451 * kevent_callback - callback for each individual event 1452 * 1453 * called with nothing locked 1454 * caller holds a reference on the kqueue 1455 */ 1456 1457static int 1458kevent_callback(__unused struct kqueue *kq, struct kevent64_s *kevp, 1459 void *data) 1460{ 1461 struct _kevent *cont_args; 1462 int error; 1463 int iskev64; 1464 1465 cont_args = (struct _kevent *)data; 1466 assert(cont_args->eventout < cont_args->eventcount); 1467 1468 iskev64 = cont_args->eventsize; 1469 1470 /* 1471 * Copy out the appropriate amount of event data for this user. 1472 */ 1473 error = kevent_copyout(kevp, &cont_args->eventlist, current_proc(), iskev64); 1474 1475 /* 1476 * If there isn't space for additional events, return 1477 * a harmless error to stop the processing here 1478 */ 1479 if (error == 0 && ++cont_args->eventout == cont_args->eventcount) 1480 error = EWOULDBLOCK; 1481 return error; 1482} 1483 1484/* 1485 * kevent_description - format a description of a kevent for diagnostic output 1486 * 1487 * called with a 128-byte string buffer 1488 */ 1489 1490char * 1491kevent_description(struct kevent64_s *kevp, char *s, size_t n) 1492{ 1493 snprintf(s, n, 1494 "kevent=" 1495 "{.ident=%#llx, .filter=%d, .flags=%#x, .fflags=%#x, .data=%#llx, .udata=%#llx, .ext[0]=%#llx, .ext[1]=%#llx}", 1496 kevp->ident, 1497 kevp->filter, 1498 kevp->flags, 1499 kevp->fflags, 1500 kevp->data, 1501 kevp->udata, 1502 kevp->ext[0], 1503 kevp->ext[1]); 1504 return s; 1505} 1506 1507/* 1508 * kevent_register - add a new event to a kqueue 1509 * 1510 * Creates a mapping between the event source and 1511 * the kqueue via a knote data structure. 1512 * 1513 * Because many/most the event sources are file 1514 * descriptor related, the knote is linked off 1515 * the filedescriptor table for quick access. 1516 * 1517 * called with nothing locked 1518 * caller holds a reference on the kqueue 1519 */ 1520 1521int 1522kevent_register(struct kqueue *kq, struct kevent64_s *kev, __unused struct proc *ctxp) 1523{ 1524 struct proc *p = kq->kq_p; 1525 struct filedesc *fdp = p->p_fd; 1526 struct filterops *fops; 1527 struct fileproc *fp = NULL; 1528 struct knote *kn = NULL; 1529 int error = 0; 1530 1531 if (kev->filter < 0) { 1532 if (kev->filter + EVFILT_SYSCOUNT < 0) 1533 return (EINVAL); 1534 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 1535 } else { 1536 /* 1537 * XXX 1538 * filter attach routine is responsible for insuring that 1539 * the identifier can be attached to it. 1540 */ 1541 printf("unknown filter: %d\n", kev->filter); 1542 return (EINVAL); 1543 } 1544 1545 restart: 1546 /* this iocount needs to be dropped if it is not registered */ 1547 proc_fdlock(p); 1548 if (fops->f_isfd && (error = fp_lookup(p, kev->ident, &fp, 1)) != 0) { 1549 proc_fdunlock(p); 1550 return(error); 1551 } 1552 1553 if (fops->f_isfd) { 1554 /* fd-based knotes are linked off the fd table */ 1555 if (kev->ident < (u_int)fdp->fd_knlistsize) { 1556 SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link) 1557 if (kq == kn->kn_kq && 1558 kev->filter == kn->kn_filter) 1559 break; 1560 } 1561 } else { 1562 /* hash non-fd knotes here too */ 1563 if (fdp->fd_knhashmask != 0) { 1564 struct klist *list; 1565 1566 list = &fdp->fd_knhash[ 1567 KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; 1568 SLIST_FOREACH(kn, list, kn_link) 1569 if (kev->ident == kn->kn_id && 1570 kq == kn->kn_kq && 1571 kev->filter == kn->kn_filter) 1572 break; 1573 } 1574 } 1575 1576 /* 1577 * kn now contains the matching knote, or NULL if no match 1578 */ 1579 if (kn == NULL) { 1580 if ((kev->flags & (EV_ADD|EV_DELETE)) == EV_ADD) { 1581 kn = knote_alloc(); 1582 if (kn == NULL) { 1583 proc_fdunlock(p); 1584 error = ENOMEM; 1585 goto done; 1586 } 1587 kn->kn_fp = fp; 1588 kn->kn_kq = kq; 1589 kn->kn_tq = &kq->kq_head; 1590 kn->kn_fop = fops; 1591 kn->kn_sfflags = kev->fflags; 1592 kn->kn_sdata = kev->data; 1593 kev->fflags = 0; 1594 kev->data = 0; 1595 kn->kn_kevent = *kev; 1596 kn->kn_inuse = 1; /* for f_attach() */ 1597 kn->kn_status = KN_ATTACHING; 1598 1599 /* before anyone can find it */ 1600 if (kev->flags & EV_DISABLE) 1601 kn->kn_status |= KN_DISABLED; 1602 1603 error = knote_fdpattach(kn, fdp, p); 1604 proc_fdunlock(p); 1605 1606 if (error) { 1607 knote_free(kn); 1608 goto done; 1609 } 1610 1611 /* 1612 * apply reference count to knote structure, and 1613 * do not release it at the end of this routine. 1614 */ 1615 fp = NULL; 1616 1617 error = fops->f_attach(kn); 1618 1619 kqlock(kq); 1620 1621 if (error != 0) { 1622 /* 1623 * Failed to attach correctly, so drop. 1624 * All other possible users/droppers 1625 * have deferred to us. 1626 */ 1627 kn->kn_status |= KN_DROPPING; 1628 kqunlock(kq); 1629 knote_drop(kn, p); 1630 goto done; 1631 } else if (kn->kn_status & KN_DROPPING) { 1632 /* 1633 * Attach succeeded, but someone else 1634 * deferred their drop - now we have 1635 * to do it for them (after detaching). 1636 */ 1637 kqunlock(kq); 1638 kn->kn_fop->f_detach(kn); 1639 knote_drop(kn, p); 1640 goto done; 1641 } 1642 kn->kn_status &= ~KN_ATTACHING; 1643 kqunlock(kq); 1644 } else { 1645 proc_fdunlock(p); 1646 error = ENOENT; 1647 goto done; 1648 } 1649 } else { 1650 /* existing knote - get kqueue lock */ 1651 kqlock(kq); 1652 proc_fdunlock(p); 1653 1654 if (kev->flags & EV_DELETE) { 1655 knote_dequeue(kn); 1656 kn->kn_status |= KN_DISABLED; 1657 if (kqlock2knotedrop(kq, kn)) { 1658 kn->kn_fop->f_detach(kn); 1659 knote_drop(kn, p); 1660 } 1661 goto done; 1662 } 1663 1664 /* update status flags for existing knote */ 1665 if (kev->flags & EV_DISABLE) { 1666 knote_dequeue(kn); 1667 kn->kn_status |= KN_DISABLED; 1668 } else if (kev->flags & EV_ENABLE) { 1669 kn->kn_status &= ~KN_DISABLED; 1670 if (kn->kn_status & KN_ACTIVE) 1671 knote_enqueue(kn); 1672 } 1673 1674 /* 1675 * The user may change some filter values after the 1676 * initial EV_ADD, but doing so will not reset any 1677 * filter which have already been triggered. 1678 */ 1679 kn->kn_kevent.udata = kev->udata; 1680 if (fops->f_isfd || fops->f_touch == NULL) { 1681 kn->kn_sfflags = kev->fflags; 1682 kn->kn_sdata = kev->data; 1683 } 1684 1685 /* 1686 * If somebody is in the middle of dropping this 1687 * knote - go find/insert a new one. But we have 1688 * wait for this one to go away first. Attaches 1689 * running in parallel may also drop/modify the 1690 * knote. Wait for those to complete as well and 1691 * then start over if we encounter one. 1692 */ 1693 if (!kqlock2knoteusewait(kq, kn)) { 1694 /* kqueue, proc_fdlock both unlocked */ 1695 goto restart; 1696 } 1697 1698 /* 1699 * Call touch routine to notify filter of changes 1700 * in filter values. 1701 */ 1702 if (!fops->f_isfd && fops->f_touch != NULL) 1703 fops->f_touch(kn, kev, EVENT_REGISTER); 1704 } 1705 /* still have use ref on knote */ 1706 1707 /* 1708 * If the knote is not marked to always stay enqueued, 1709 * invoke the filter routine to see if it should be 1710 * enqueued now. 1711 */ 1712 if ((kn->kn_status & KN_STAYQUEUED) == 0 && kn->kn_fop->f_event(kn, 0)) { 1713 if (knoteuse2kqlock(kq, kn)) 1714 knote_activate(kn, 1); 1715 kqunlock(kq); 1716 } else { 1717 knote_put(kn); 1718 } 1719 1720done: 1721 if (fp != NULL) 1722 fp_drop(p, kev->ident, fp, 0); 1723 return (error); 1724} 1725 1726 1727/* 1728 * knote_process - process a triggered event 1729 * 1730 * Validate that it is really still a triggered event 1731 * by calling the filter routines (if necessary). Hold 1732 * a use reference on the knote to avoid it being detached. 1733 * If it is still considered triggered, invoke the callback 1734 * routine provided and move it to the provided inprocess 1735 * queue. 1736 * 1737 * caller holds a reference on the kqueue. 1738 * kqueue locked on entry and exit - but may be dropped 1739 */ 1740static int 1741knote_process(struct knote *kn, 1742 kevent_callback_t callback, 1743 void *data, 1744 struct kqtailq *inprocessp, 1745 struct proc *p) 1746{ 1747 struct kqueue *kq = kn->kn_kq; 1748 struct kevent64_s kev; 1749 int touch; 1750 int result; 1751 int error; 1752 1753 /* 1754 * Determine the kevent state we want to return. 1755 * 1756 * Some event states need to be revalidated before returning 1757 * them, others we take the snapshot at the time the event 1758 * was enqueued. 1759 * 1760 * Events with non-NULL f_touch operations must be touched. 1761 * Triggered events must fill in kev for the callback. 1762 * 1763 * Convert our lock to a use-count and call the event's 1764 * filter routine(s) to update. 1765 */ 1766 if ((kn->kn_status & KN_DISABLED) != 0) { 1767 result = 0; 1768 touch = 0; 1769 } else { 1770 int revalidate; 1771 1772 result = 1; 1773 revalidate = ((kn->kn_status & KN_STAYQUEUED) != 0 || 1774 (kn->kn_flags & EV_ONESHOT) == 0); 1775 touch = (!kn->kn_fop->f_isfd && kn->kn_fop->f_touch != NULL); 1776 1777 if (revalidate || touch) { 1778 if (revalidate) 1779 knote_deactivate(kn); 1780 1781 /* call the filter/touch routines with just a ref */ 1782 if (kqlock2knoteuse(kq, kn)) { 1783 1784 /* if we have to revalidate, call the filter */ 1785 if (revalidate) { 1786 result = kn->kn_fop->f_event(kn, 0); 1787 } 1788 1789 /* capture the kevent data - using touch if specified */ 1790 if (result && touch) { 1791 kn->kn_fop->f_touch(kn, &kev, EVENT_PROCESS); 1792 } 1793 1794 /* convert back to a kqlock - bail if the knote went away */ 1795 if (!knoteuse2kqlock(kq, kn)) { 1796 return EJUSTRETURN; 1797 } else if (result) { 1798 /* if revalidated as alive, make sure it's active */ 1799 if (!(kn->kn_status & KN_ACTIVE)) { 1800 knote_activate(kn, 0); 1801 } 1802 1803 /* capture all events that occurred during filter */ 1804 if (!touch) { 1805 kev = kn->kn_kevent; 1806 } 1807 1808 } else if ((kn->kn_status & KN_STAYQUEUED) == 0) { 1809 /* was already dequeued, so just bail on this one */ 1810 return EJUSTRETURN; 1811 } 1812 } else { 1813 return EJUSTRETURN; 1814 } 1815 } else { 1816 kev = kn->kn_kevent; 1817 } 1818 } 1819 1820 /* move knote onto inprocess queue */ 1821 assert(kn->kn_tq == &kq->kq_head); 1822 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1823 kn->kn_tq = inprocessp; 1824 TAILQ_INSERT_TAIL(inprocessp, kn, kn_tqe); 1825 1826 /* 1827 * Determine how to dispatch the knote for future event handling. 1828 * not-fired: just return (do not callout). 1829 * One-shot: deactivate it. 1830 * Clear: deactivate and clear the state. 1831 * Dispatch: don't clear state, just deactivate it and mark it disabled. 1832 * All others: just leave where they are. 1833 */ 1834 1835 if (result == 0) { 1836 return EJUSTRETURN; 1837 } else if ((kn->kn_flags & EV_ONESHOT) != 0) { 1838 knote_deactivate(kn); 1839 if (kqlock2knotedrop(kq, kn)) { 1840 kn->kn_fop->f_detach(kn); 1841 knote_drop(kn, p); 1842 } 1843 } else if ((kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) != 0) { 1844 if ((kn->kn_flags & EV_DISPATCH) != 0) { 1845 /* deactivate and disable all dispatch knotes */ 1846 knote_deactivate(kn); 1847 kn->kn_status |= KN_DISABLED; 1848 } else if (!touch || kn->kn_fflags == 0) { 1849 /* only deactivate if nothing since the touch */ 1850 knote_deactivate(kn); 1851 } 1852 if (!touch && (kn->kn_flags & EV_CLEAR) != 0) { 1853 /* manually clear non-touch knotes */ 1854 kn->kn_data = 0; 1855 kn->kn_fflags = 0; 1856 } 1857 kqunlock(kq); 1858 } else { 1859 /* 1860 * leave on inprocess queue. We'll 1861 * move all the remaining ones back 1862 * the kq queue and wakeup any 1863 * waiters when we are done. 1864 */ 1865 kqunlock(kq); 1866 } 1867 1868 /* callback to handle each event as we find it */ 1869 error = (callback)(kq, &kev, data); 1870 1871 kqlock(kq); 1872 return error; 1873} 1874 1875/* 1876 * Return 0 to indicate that processing should proceed, 1877 * -1 if there is nothing to process. 1878 * 1879 * Called with kqueue locked and returns the same way, 1880 * but may drop lock temporarily. 1881 */ 1882static int 1883kqueue_begin_processing(struct kqueue *kq) 1884{ 1885 for (;;) { 1886 if (kq->kq_count == 0) { 1887 return -1; 1888 } 1889 1890 /* if someone else is processing the queue, wait */ 1891 if (kq->kq_nprocess != 0) { 1892 wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, &kq->kq_nprocess, THREAD_UNINT, 0); 1893 kq->kq_state |= KQ_PROCWAIT; 1894 kqunlock(kq); 1895 thread_block(THREAD_CONTINUE_NULL); 1896 kqlock(kq); 1897 } else { 1898 kq->kq_nprocess = 1; 1899 return 0; 1900 } 1901 } 1902} 1903 1904/* 1905 * Called with kqueue lock held. 1906 */ 1907static void 1908kqueue_end_processing(struct kqueue *kq) 1909{ 1910 kq->kq_nprocess = 0; 1911 if (kq->kq_state & KQ_PROCWAIT) { 1912 kq->kq_state &= ~KQ_PROCWAIT; 1913 wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, &kq->kq_nprocess, THREAD_AWAKENED); 1914 } 1915} 1916 1917/* 1918 * kqueue_process - process the triggered events in a kqueue 1919 * 1920 * Walk the queued knotes and validate that they are 1921 * really still triggered events by calling the filter 1922 * routines (if necessary). Hold a use reference on 1923 * the knote to avoid it being detached. For each event 1924 * that is still considered triggered, invoke the 1925 * callback routine provided. 1926 * 1927 * caller holds a reference on the kqueue. 1928 * kqueue locked on entry and exit - but may be dropped 1929 * kqueue list locked (held for duration of call) 1930 */ 1931 1932static int 1933kqueue_process(struct kqueue *kq, 1934 kevent_callback_t callback, 1935 void *data, 1936 int *countp, 1937 struct proc *p) 1938{ 1939 struct kqtailq inprocess; 1940 struct knote *kn; 1941 int nevents; 1942 int error; 1943 1944 TAILQ_INIT(&inprocess); 1945 1946 if (kqueue_begin_processing(kq) == -1) { 1947 *countp = 0; 1948 /* Nothing to process */ 1949 return 0; 1950 } 1951 1952 /* 1953 * Clear any pre-posted status from previous runs, so we only 1954 * detect events that occur during this run. 1955 */ 1956 wait_queue_sub_clearrefs(kq->kq_wqs); 1957 1958 /* 1959 * loop through the enqueued knotes, processing each one and 1960 * revalidating those that need it. As they are processed, 1961 * they get moved to the inprocess queue (so the loop can end). 1962 */ 1963 error = 0; 1964 nevents = 0; 1965 1966 while (error == 0 && 1967 (kn = TAILQ_FIRST(&kq->kq_head)) != NULL) { 1968 error = knote_process(kn, callback, data, &inprocess, p); 1969 if (error == EJUSTRETURN) 1970 error = 0; 1971 else 1972 nevents++; 1973 } 1974 1975 /* 1976 * With the kqueue still locked, move any knotes 1977 * remaining on the inprocess queue back to the 1978 * kq's queue and wake up any waiters. 1979 */ 1980 while ((kn = TAILQ_FIRST(&inprocess)) != NULL) { 1981 assert(kn->kn_tq == &inprocess); 1982 TAILQ_REMOVE(&inprocess, kn, kn_tqe); 1983 kn->kn_tq = &kq->kq_head; 1984 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 1985 } 1986 1987 kqueue_end_processing(kq); 1988 1989 *countp = nevents; 1990 return error; 1991} 1992 1993 1994static void 1995kqueue_scan_continue(void *data, wait_result_t wait_result) 1996{ 1997 thread_t self = current_thread(); 1998 uthread_t ut = (uthread_t)get_bsdthread_info(self); 1999 struct _kqueue_scan * cont_args = &ut->uu_kevent.ss_kqueue_scan; 2000 struct kqueue *kq = (struct kqueue *)data; 2001 int error; 2002 int count; 2003 2004 /* convert the (previous) wait_result to a proper error */ 2005 switch (wait_result) { 2006 case THREAD_AWAKENED: 2007 kqlock(kq); 2008 error = kqueue_process(kq, cont_args->call, cont_args, &count, current_proc()); 2009 if (error == 0 && count == 0) { 2010 wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, KQ_EVENT, 2011 THREAD_ABORTSAFE, cont_args->deadline); 2012 kq->kq_state |= KQ_SLEEP; 2013 kqunlock(kq); 2014 thread_block_parameter(kqueue_scan_continue, kq); 2015 /* NOTREACHED */ 2016 } 2017 kqunlock(kq); 2018 break; 2019 case THREAD_TIMED_OUT: 2020 error = EWOULDBLOCK; 2021 break; 2022 case THREAD_INTERRUPTED: 2023 error = EINTR; 2024 break; 2025 default: 2026 panic("kevent_scan_cont() - invalid wait_result (%d)", wait_result); 2027 error = 0; 2028 } 2029 2030 /* call the continuation with the results */ 2031 assert(cont_args->cont != NULL); 2032 (cont_args->cont)(kq, cont_args->data, error); 2033} 2034 2035 2036/* 2037 * kqueue_scan - scan and wait for events in a kqueue 2038 * 2039 * Process the triggered events in a kqueue. 2040 * 2041 * If there are no events triggered arrange to 2042 * wait for them. If the caller provided a 2043 * continuation routine, then kevent_scan will 2044 * also. 2045 * 2046 * The callback routine must be valid. 2047 * The caller must hold a use-count reference on the kq. 2048 */ 2049 2050int 2051kqueue_scan(struct kqueue *kq, 2052 kevent_callback_t callback, 2053 kqueue_continue_t continuation, 2054 void *data, 2055 struct timeval *atvp, 2056 struct proc *p) 2057{ 2058 thread_continue_t cont = THREAD_CONTINUE_NULL; 2059 uint64_t deadline; 2060 int error; 2061 int first; 2062 2063 assert(callback != NULL); 2064 2065 first = 1; 2066 for (;;) { 2067 wait_result_t wait_result; 2068 int count; 2069 2070 /* 2071 * Make a pass through the kq to find events already 2072 * triggered. 2073 */ 2074 kqlock(kq); 2075 error = kqueue_process(kq, callback, data, &count, p); 2076 if (error || count) 2077 break; /* lock still held */ 2078 2079 /* looks like we have to consider blocking */ 2080 if (first) { 2081 first = 0; 2082 /* convert the timeout to a deadline once */ 2083 if (atvp->tv_sec || atvp->tv_usec) { 2084 uint64_t now; 2085 2086 clock_get_uptime(&now); 2087 nanoseconds_to_absolutetime((uint64_t)atvp->tv_sec * NSEC_PER_SEC + 2088 atvp->tv_usec * NSEC_PER_USEC, 2089 &deadline); 2090 if (now >= deadline) { 2091 /* non-blocking call */ 2092 error = EWOULDBLOCK; 2093 break; /* lock still held */ 2094 } 2095 deadline -= now; 2096 clock_absolutetime_interval_to_deadline(deadline, &deadline); 2097 } else { 2098 deadline = 0; /* block forever */ 2099 } 2100 2101 if (continuation) { 2102 uthread_t ut = (uthread_t)get_bsdthread_info(current_thread()); 2103 struct _kqueue_scan *cont_args = &ut->uu_kevent.ss_kqueue_scan; 2104 2105 cont_args->call = callback; 2106 cont_args->cont = continuation; 2107 cont_args->deadline = deadline; 2108 cont_args->data = data; 2109 cont = kqueue_scan_continue; 2110 } 2111 } 2112 2113 /* go ahead and wait */ 2114 wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, KQ_EVENT, THREAD_ABORTSAFE, deadline); 2115 kq->kq_state |= KQ_SLEEP; 2116 kqunlock(kq); 2117 wait_result = thread_block_parameter(cont, kq); 2118 /* NOTREACHED if (continuation != NULL) */ 2119 2120 switch (wait_result) { 2121 case THREAD_AWAKENED: 2122 continue; 2123 case THREAD_TIMED_OUT: 2124 return EWOULDBLOCK; 2125 case THREAD_INTERRUPTED: 2126 return EINTR; 2127 default: 2128 panic("kevent_scan - bad wait_result (%d)", 2129 wait_result); 2130 error = 0; 2131 } 2132 } 2133 kqunlock(kq); 2134 return error; 2135} 2136 2137 2138/* 2139 * XXX 2140 * This could be expanded to call kqueue_scan, if desired. 2141 */ 2142/*ARGSUSED*/ 2143static int 2144kqueue_read(__unused struct fileproc *fp, 2145 __unused struct uio *uio, 2146 __unused int flags, 2147 __unused vfs_context_t ctx) 2148{ 2149 return (ENXIO); 2150} 2151 2152/*ARGSUSED*/ 2153static int 2154kqueue_write(__unused struct fileproc *fp, 2155 __unused struct uio *uio, 2156 __unused int flags, 2157 __unused vfs_context_t ctx) 2158{ 2159 return (ENXIO); 2160} 2161 2162/*ARGSUSED*/ 2163static int 2164kqueue_ioctl(__unused struct fileproc *fp, 2165 __unused u_long com, 2166 __unused caddr_t data, 2167 __unused vfs_context_t ctx) 2168{ 2169 return (ENOTTY); 2170} 2171 2172/*ARGSUSED*/ 2173static int 2174kqueue_select(struct fileproc *fp, int which, void *wql, __unused vfs_context_t ctx) 2175{ 2176 struct kqueue *kq = (struct kqueue *)fp->f_data; 2177 struct knote *kn; 2178 struct kqtailq inprocessq; 2179 int retnum = 0; 2180 2181 if (which != FREAD) 2182 return 0; 2183 2184 TAILQ_INIT(&inprocessq); 2185 2186 kqlock(kq); 2187 /* 2188 * If this is the first pass, link the wait queue associated with the 2189 * the kqueue onto the wait queue set for the select(). Normally we 2190 * use selrecord() for this, but it uses the wait queue within the 2191 * selinfo structure and we need to use the main one for the kqueue to 2192 * catch events from KN_STAYQUEUED sources. So we do the linkage manually. 2193 * (The select() call will unlink them when it ends). 2194 */ 2195 if (wql != NULL) { 2196 thread_t cur_act = current_thread(); 2197 struct uthread * ut = get_bsdthread_info(cur_act); 2198 2199 kq->kq_state |= KQ_SEL; 2200 wait_queue_link_noalloc((wait_queue_t)kq->kq_wqs, ut->uu_wqset, 2201 (wait_queue_link_t)wql); 2202 } 2203 2204 if (kqueue_begin_processing(kq) == -1) { 2205 kqunlock(kq); 2206 return 0; 2207 } 2208 2209 if (kq->kq_count != 0) { 2210 /* 2211 * there is something queued - but it might be a 2212 * KN_STAYQUEUED knote, which may or may not have 2213 * any events pending. So, we have to walk the 2214 * list of knotes to see, and peek at the stay- 2215 * queued ones to be really sure. 2216 */ 2217 while ((kn = (struct knote*)TAILQ_FIRST(&kq->kq_head)) != NULL) { 2218 if ((kn->kn_status & KN_STAYQUEUED) == 0) { 2219 retnum = 1; 2220 goto out; 2221 } 2222 2223 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 2224 TAILQ_INSERT_TAIL(&inprocessq, kn, kn_tqe); 2225 2226 if (kqlock2knoteuse(kq, kn)) { 2227 unsigned peek; 2228 2229 peek = kn->kn_fop->f_peek(kn); 2230 if (knoteuse2kqlock(kq, kn)) { 2231 if (peek > 0) { 2232 retnum = 1; 2233 goto out; 2234 } 2235 } else { 2236 retnum = 0; 2237 } 2238 } 2239 } 2240 } 2241 2242out: 2243 /* Return knotes to active queue */ 2244 while ((kn = TAILQ_FIRST(&inprocessq)) != NULL) { 2245 TAILQ_REMOVE(&inprocessq, kn, kn_tqe); 2246 kn->kn_tq = &kq->kq_head; 2247 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 2248 } 2249 2250 kqueue_end_processing(kq); 2251 kqunlock(kq); 2252 return retnum; 2253} 2254 2255/* 2256 * kqueue_close - 2257 */ 2258/*ARGSUSED*/ 2259static int 2260kqueue_close(struct fileglob *fg, __unused vfs_context_t ctx) 2261{ 2262 struct kqueue *kq = (struct kqueue *)fg->fg_data; 2263 2264 kqueue_dealloc(kq); 2265 fg->fg_data = NULL; 2266 return (0); 2267} 2268 2269/*ARGSUSED*/ 2270/* 2271 * The callers has taken a use-count reference on this kqueue and will donate it 2272 * to the kqueue we are being added to. This keeps the kqueue from closing until 2273 * that relationship is torn down. 2274 */ 2275static int 2276kqueue_kqfilter(__unused struct fileproc *fp, struct knote *kn, __unused vfs_context_t ctx) 2277{ 2278 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 2279 struct kqueue *parentkq = kn->kn_kq; 2280 2281 if (parentkq == kq || 2282 kn->kn_filter != EVFILT_READ) 2283 return (1); 2284 2285 /* 2286 * We have to avoid creating a cycle when nesting kqueues 2287 * inside another. Rather than trying to walk the whole 2288 * potential DAG of nested kqueues, we just use a simple 2289 * ceiling protocol. When a kqueue is inserted into another, 2290 * we check that the (future) parent is not already nested 2291 * into another kqueue at a lower level than the potenial 2292 * child (because it could indicate a cycle). If that test 2293 * passes, we just mark the nesting levels accordingly. 2294 */ 2295 2296 kqlock(parentkq); 2297 if (parentkq->kq_level > 0 && 2298 parentkq->kq_level < kq->kq_level) 2299 { 2300 kqunlock(parentkq); 2301 return (1); 2302 } else { 2303 /* set parent level appropriately */ 2304 if (parentkq->kq_level == 0) 2305 parentkq->kq_level = 2; 2306 if (parentkq->kq_level < kq->kq_level + 1) 2307 parentkq->kq_level = kq->kq_level + 1; 2308 kqunlock(parentkq); 2309 2310 kn->kn_fop = &kqread_filtops; 2311 kqlock(kq); 2312 KNOTE_ATTACH(&kq->kq_sel.si_note, kn); 2313 /* indicate nesting in child, if needed */ 2314 if (kq->kq_level == 0) 2315 kq->kq_level = 1; 2316 kqunlock(kq); 2317 return (0); 2318 } 2319} 2320 2321/* 2322 * kqueue_drain - called when kq is closed 2323 */ 2324/*ARGSUSED*/ 2325static int 2326kqueue_drain(struct fileproc *fp, __unused vfs_context_t ctx) 2327{ 2328 struct kqueue *kq = (struct kqueue *)fp->f_fglob->fg_data; 2329 kqlock(kq); 2330 kqueue_wakeup(kq, 1); 2331 kqunlock(kq); 2332 return 0; 2333} 2334 2335/*ARGSUSED*/ 2336int 2337kqueue_stat(struct fileproc *fp, void *ub, int isstat64, __unused vfs_context_t ctx) 2338{ 2339 2340 struct kqueue *kq = (struct kqueue *)fp->f_data; 2341 if (isstat64 != 0) { 2342 struct stat64 *sb64 = (struct stat64 *)ub; 2343 2344 bzero((void *)sb64, sizeof(*sb64)); 2345 sb64->st_size = kq->kq_count; 2346 if (kq->kq_state & KQ_KEV64) 2347 sb64->st_blksize = sizeof(struct kevent64_s); 2348 else 2349 sb64->st_blksize = sizeof(struct kevent); 2350 sb64->st_mode = S_IFIFO; 2351 } else { 2352 struct stat *sb = (struct stat *)ub; 2353 2354 bzero((void *)sb, sizeof(*sb)); 2355 sb->st_size = kq->kq_count; 2356 if (kq->kq_state & KQ_KEV64) 2357 sb->st_blksize = sizeof(struct kevent64_s); 2358 else 2359 sb->st_blksize = sizeof(struct kevent); 2360 sb->st_mode = S_IFIFO; 2361 } 2362 2363 return (0); 2364} 2365 2366/* 2367 * Called with the kqueue locked 2368 */ 2369static void 2370kqueue_wakeup(struct kqueue *kq, int closed) 2371{ 2372 if ((kq->kq_state & (KQ_SLEEP | KQ_SEL)) != 0 || kq->kq_nprocess > 0) { 2373 kq->kq_state &= ~(KQ_SLEEP | KQ_SEL); 2374 wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, KQ_EVENT, 2375 (closed) ? THREAD_INTERRUPTED : THREAD_AWAKENED); 2376 } 2377} 2378 2379void 2380klist_init(struct klist *list) 2381{ 2382 SLIST_INIT(list); 2383} 2384 2385 2386/* 2387 * Query/Post each knote in the object's list 2388 * 2389 * The object lock protects the list. It is assumed 2390 * that the filter/event routine for the object can 2391 * determine that the object is already locked (via 2392 * the hint) and not deadlock itself. 2393 * 2394 * The object lock should also hold off pending 2395 * detach/drop operations. But we'll prevent it here 2396 * too - just in case. 2397 */ 2398void 2399knote(struct klist *list, long hint) 2400{ 2401 struct knote *kn; 2402 2403 SLIST_FOREACH(kn, list, kn_selnext) { 2404 struct kqueue *kq = kn->kn_kq; 2405 2406 kqlock(kq); 2407 if (kqlock2knoteuse(kq, kn)) { 2408 int result; 2409 2410 /* call the event with only a use count */ 2411 result = kn->kn_fop->f_event(kn, hint); 2412 2413 /* if its not going away and triggered */ 2414 if (knoteuse2kqlock(kq, kn) && result) 2415 knote_activate(kn, 1); 2416 /* lock held again */ 2417 } 2418 kqunlock(kq); 2419 } 2420} 2421 2422/* 2423 * attach a knote to the specified list. Return true if this is the first entry. 2424 * The list is protected by whatever lock the object it is associated with uses. 2425 */ 2426int 2427knote_attach(struct klist *list, struct knote *kn) 2428{ 2429 int ret = SLIST_EMPTY(list); 2430 SLIST_INSERT_HEAD(list, kn, kn_selnext); 2431 return ret; 2432} 2433 2434/* 2435 * detach a knote from the specified list. Return true if that was the last entry. 2436 * The list is protected by whatever lock the object it is associated with uses. 2437 */ 2438int 2439knote_detach(struct klist *list, struct knote *kn) 2440{ 2441 SLIST_REMOVE(list, kn, knote, kn_selnext); 2442 return SLIST_EMPTY(list); 2443} 2444 2445/* 2446 * For a given knote, link a provided wait queue directly with the kqueue. 2447 * Wakeups will happen via recursive wait queue support. But nothing will move 2448 * the knote to the active list at wakeup (nothing calls knote()). Instead, 2449 * we permanently enqueue them here. 2450 * 2451 * kqueue and knote references are held by caller. 2452 * 2453 * caller provides the wait queue link structure. 2454 */ 2455int 2456knote_link_wait_queue(struct knote *kn, struct wait_queue *wq, wait_queue_link_t wql) 2457{ 2458 struct kqueue *kq = kn->kn_kq; 2459 kern_return_t kr; 2460 2461 kr = wait_queue_link_noalloc(wq, kq->kq_wqs, wql); 2462 if (kr == KERN_SUCCESS) { 2463 knote_markstayqueued(kn); 2464 return 0; 2465 } else { 2466 return EINVAL; 2467 } 2468} 2469 2470/* 2471 * Unlink the provided wait queue from the kqueue associated with a knote. 2472 * Also remove it from the magic list of directly attached knotes. 2473 * 2474 * Note that the unlink may have already happened from the other side, so 2475 * ignore any failures to unlink and just remove it from the kqueue list. 2476 * 2477 * On success, caller is responsible for the link structure 2478 */ 2479int 2480knote_unlink_wait_queue(struct knote *kn, struct wait_queue *wq, wait_queue_link_t *wqlp) 2481{ 2482 struct kqueue *kq = kn->kn_kq; 2483 kern_return_t kr; 2484 2485 kr = wait_queue_unlink_nofree(wq, kq->kq_wqs, wqlp); 2486 kqlock(kq); 2487 kn->kn_status &= ~KN_STAYQUEUED; 2488 knote_dequeue(kn); 2489 kqunlock(kq); 2490 return (kr != KERN_SUCCESS) ? EINVAL : 0; 2491} 2492 2493/* 2494 * remove all knotes referencing a specified fd 2495 * 2496 * Essentially an inlined knote_remove & knote_drop 2497 * when we know for sure that the thing is a file 2498 * 2499 * Entered with the proc_fd lock already held. 2500 * It returns the same way, but may drop it temporarily. 2501 */ 2502void 2503knote_fdclose(struct proc *p, int fd) 2504{ 2505 struct filedesc *fdp = p->p_fd; 2506 struct klist *list; 2507 struct knote *kn; 2508 2509 list = &fdp->fd_knlist[fd]; 2510 while ((kn = SLIST_FIRST(list)) != NULL) { 2511 struct kqueue *kq = kn->kn_kq; 2512 2513 if (kq->kq_p != p) 2514 panic("knote_fdclose: proc mismatch (kq->kq_p=%p != p=%p)", kq->kq_p, p); 2515 2516 kqlock(kq); 2517 proc_fdunlock(p); 2518 2519 /* 2520 * Convert the lock to a drop ref. 2521 * If we get it, go ahead and drop it. 2522 * Otherwise, we waited for it to 2523 * be dropped by the other guy, so 2524 * it is safe to move on in the list. 2525 */ 2526 if (kqlock2knotedrop(kq, kn)) { 2527 kn->kn_fop->f_detach(kn); 2528 knote_drop(kn, p); 2529 } 2530 2531 proc_fdlock(p); 2532 2533 /* the fd tables may have changed - start over */ 2534 list = &fdp->fd_knlist[fd]; 2535 } 2536} 2537 2538/* proc_fdlock held on entry (and exit) */ 2539static int 2540knote_fdpattach(struct knote *kn, struct filedesc *fdp, struct proc *p) 2541{ 2542 struct klist *list = NULL; 2543 2544 if (! kn->kn_fop->f_isfd) { 2545 if (fdp->fd_knhashmask == 0) 2546 fdp->fd_knhash = hashinit(CONFIG_KN_HASHSIZE, M_KQUEUE, 2547 &fdp->fd_knhashmask); 2548 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 2549 } else { 2550 if ((u_int)fdp->fd_knlistsize <= kn->kn_id) { 2551 u_int size = 0; 2552 2553 if (kn->kn_id >= (uint64_t)p->p_rlimit[RLIMIT_NOFILE].rlim_cur 2554 || kn->kn_id >= (uint64_t)maxfiles) 2555 return (EINVAL); 2556 2557 /* have to grow the fd_knlist */ 2558 size = fdp->fd_knlistsize; 2559 while (size <= kn->kn_id) 2560 size += KQEXTENT; 2561 2562 if (size >= (UINT_MAX/sizeof(struct klist *))) 2563 return (EINVAL); 2564 2565 MALLOC(list, struct klist *, 2566 size * sizeof(struct klist *), M_KQUEUE, M_WAITOK); 2567 if (list == NULL) 2568 return (ENOMEM); 2569 2570 bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list, 2571 fdp->fd_knlistsize * sizeof(struct klist *)); 2572 bzero((caddr_t)list + 2573 fdp->fd_knlistsize * sizeof(struct klist *), 2574 (size - fdp->fd_knlistsize) * sizeof(struct klist *)); 2575 FREE(fdp->fd_knlist, M_KQUEUE); 2576 fdp->fd_knlist = list; 2577 fdp->fd_knlistsize = size; 2578 } 2579 list = &fdp->fd_knlist[kn->kn_id]; 2580 } 2581 SLIST_INSERT_HEAD(list, kn, kn_link); 2582 return (0); 2583} 2584 2585 2586 2587/* 2588 * should be called at spl == 0, since we don't want to hold spl 2589 * while calling fdrop and free. 2590 */ 2591static void 2592knote_drop(struct knote *kn, __unused struct proc *ctxp) 2593{ 2594 struct kqueue *kq = kn->kn_kq; 2595 struct proc *p = kq->kq_p; 2596 struct filedesc *fdp = p->p_fd; 2597 struct klist *list; 2598 int needswakeup; 2599 2600 proc_fdlock(p); 2601 if (kn->kn_fop->f_isfd) 2602 list = &fdp->fd_knlist[kn->kn_id]; 2603 else 2604 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 2605 2606 SLIST_REMOVE(list, kn, knote, kn_link); 2607 kqlock(kq); 2608 knote_dequeue(kn); 2609 needswakeup = (kn->kn_status & KN_USEWAIT); 2610 kqunlock(kq); 2611 proc_fdunlock(p); 2612 2613 if (needswakeup) 2614 wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, &kn->kn_status, THREAD_AWAKENED); 2615 2616 if (kn->kn_fop->f_isfd) 2617 fp_drop(p, kn->kn_id, kn->kn_fp, 0); 2618 2619 knote_free(kn); 2620} 2621 2622/* called with kqueue lock held */ 2623static void 2624knote_activate(struct knote *kn, int propagate) 2625{ 2626 struct kqueue *kq = kn->kn_kq; 2627 2628 kn->kn_status |= KN_ACTIVE; 2629 knote_enqueue(kn); 2630 kqueue_wakeup(kq, 0); 2631 2632 /* this is a real event: wake up the parent kq, too */ 2633 if (propagate) 2634 KNOTE(&kq->kq_sel.si_note, 0); 2635} 2636 2637/* called with kqueue lock held */ 2638static void 2639knote_deactivate(struct knote *kn) 2640{ 2641 kn->kn_status &= ~KN_ACTIVE; 2642 knote_dequeue(kn); 2643} 2644 2645/* called with kqueue lock held */ 2646static void 2647knote_enqueue(struct knote *kn) 2648{ 2649 if ((kn->kn_status & (KN_QUEUED | KN_STAYQUEUED)) == KN_STAYQUEUED || 2650 (kn->kn_status & (KN_QUEUED | KN_STAYQUEUED | KN_DISABLED)) == 0) { 2651 struct kqtailq *tq = kn->kn_tq; 2652 struct kqueue *kq = kn->kn_kq; 2653 2654 TAILQ_INSERT_TAIL(tq, kn, kn_tqe); 2655 kn->kn_status |= KN_QUEUED; 2656 kq->kq_count++; 2657 } 2658} 2659 2660/* called with kqueue lock held */ 2661static void 2662knote_dequeue(struct knote *kn) 2663{ 2664 struct kqueue *kq = kn->kn_kq; 2665 2666 if ((kn->kn_status & (KN_QUEUED | KN_STAYQUEUED)) == KN_QUEUED) { 2667 struct kqtailq *tq = kn->kn_tq; 2668 2669 TAILQ_REMOVE(tq, kn, kn_tqe); 2670 kn->kn_tq = &kq->kq_head; 2671 kn->kn_status &= ~KN_QUEUED; 2672 kq->kq_count--; 2673 } 2674} 2675 2676void 2677knote_init(void) 2678{ 2679 knote_zone = zinit(sizeof(struct knote), 8192*sizeof(struct knote), 8192, "knote zone"); 2680 2681 /* allocate kq lock group attribute and group */ 2682 kq_lck_grp_attr= lck_grp_attr_alloc_init(); 2683 2684 kq_lck_grp = lck_grp_alloc_init("kqueue", kq_lck_grp_attr); 2685 2686 /* Allocate kq lock attribute */ 2687 kq_lck_attr = lck_attr_alloc_init(); 2688 2689 /* Initialize the timer filter lock */ 2690 lck_mtx_init(&_filt_timerlock, kq_lck_grp, kq_lck_attr); 2691 2692#if VM_PRESSURE_EVENTS 2693 /* Initialize the vm pressure list lock */ 2694 vm_pressure_init(kq_lck_grp, kq_lck_attr); 2695#endif 2696} 2697SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL) 2698 2699static struct knote * 2700knote_alloc(void) 2701{ 2702 return ((struct knote *)zalloc(knote_zone)); 2703} 2704 2705static void 2706knote_free(struct knote *kn) 2707{ 2708 zfree(knote_zone, kn); 2709} 2710 2711#if SOCKETS 2712#include <sys/param.h> 2713#include <sys/socket.h> 2714#include <sys/protosw.h> 2715#include <sys/domain.h> 2716#include <sys/mbuf.h> 2717#include <sys/kern_event.h> 2718#include <sys/malloc.h> 2719#include <sys/sys_domain.h> 2720#include <sys/syslog.h> 2721 2722 2723static int kev_attach(struct socket *so, int proto, struct proc *p); 2724static int kev_detach(struct socket *so); 2725static int kev_control(struct socket *so, u_long cmd, caddr_t data, struct ifnet *ifp, struct proc *p); 2726 2727struct pr_usrreqs event_usrreqs = { 2728 pru_abort_notsupp, pru_accept_notsupp, kev_attach, pru_bind_notsupp, pru_connect_notsupp, 2729 pru_connect2_notsupp, kev_control, kev_detach, pru_disconnect_notsupp, 2730 pru_listen_notsupp, pru_peeraddr_notsupp, pru_rcvd_notsupp, pru_rcvoob_notsupp, 2731 pru_send_notsupp, pru_sense_null, pru_shutdown_notsupp, pru_sockaddr_notsupp, 2732 pru_sosend_notsupp, soreceive, pru_sopoll_notsupp 2733}; 2734 2735struct protosw eventsw[] = { 2736 { 2737 .pr_type = SOCK_RAW, 2738 .pr_domain = &systemdomain, 2739 .pr_protocol = SYSPROTO_EVENT, 2740 .pr_flags = PR_ATOMIC, 2741 .pr_usrreqs = &event_usrreqs, 2742 } 2743}; 2744 2745static 2746struct kern_event_head kern_event_head; 2747 2748static u_int32_t static_event_id = 0; 2749struct domain *sysdom = &systemdomain; 2750static lck_mtx_t *sys_mtx; 2751 2752/* 2753 * Install the protosw's for the NKE manager. Invoked at 2754 * extension load time 2755 */ 2756int 2757kern_event_init(void) 2758{ 2759 int retval; 2760 2761 if ((retval = net_add_proto(eventsw, &systemdomain)) != 0) { 2762 log(LOG_WARNING, "Can't install kernel events protocol (%d)\n", retval); 2763 return(retval); 2764 } 2765 2766 /* 2767 * Use the domain mutex for all system event sockets 2768 */ 2769 sys_mtx = sysdom->dom_mtx; 2770 2771 return(KERN_SUCCESS); 2772} 2773 2774static int 2775kev_attach(struct socket *so, __unused int proto, __unused struct proc *p) 2776{ 2777 int error; 2778 struct kern_event_pcb *ev_pcb; 2779 2780 error = soreserve(so, KEV_SNDSPACE, KEV_RECVSPACE); 2781 if (error) 2782 return error; 2783 2784 MALLOC(ev_pcb, struct kern_event_pcb *, sizeof(struct kern_event_pcb), M_PCB, M_WAITOK); 2785 if (ev_pcb == 0) 2786 return ENOBUFS; 2787 2788 ev_pcb->ev_socket = so; 2789 ev_pcb->vendor_code_filter = 0xffffffff; 2790 2791 so->so_pcb = (caddr_t) ev_pcb; 2792 lck_mtx_lock(sys_mtx); 2793 LIST_INSERT_HEAD(&kern_event_head, ev_pcb, ev_link); 2794 lck_mtx_unlock(sys_mtx); 2795 2796 return 0; 2797} 2798 2799 2800static int 2801kev_detach(struct socket *so) 2802{ 2803 struct kern_event_pcb *ev_pcb = (struct kern_event_pcb *) so->so_pcb; 2804 2805 if (ev_pcb != 0) { 2806 LIST_REMOVE(ev_pcb, ev_link); 2807 FREE(ev_pcb, M_PCB); 2808 so->so_pcb = 0; 2809 so->so_flags |= SOF_PCBCLEARING; 2810 } 2811 2812 return 0; 2813} 2814 2815/* 2816 * For now, kev_vendor_code and mbuf_tags use the same 2817 * mechanism. 2818 */ 2819 2820errno_t kev_vendor_code_find( 2821 const char *string, 2822 u_int32_t *out_vendor_code) 2823{ 2824 if (strlen(string) >= KEV_VENDOR_CODE_MAX_STR_LEN) { 2825 return EINVAL; 2826 } 2827 return net_str_id_find_internal(string, out_vendor_code, NSI_VENDOR_CODE, 1); 2828} 2829 2830errno_t kev_msg_post(struct kev_msg *event_msg) 2831{ 2832 mbuf_tag_id_t min_vendor, max_vendor; 2833 2834 net_str_id_first_last(&min_vendor, &max_vendor, NSI_VENDOR_CODE); 2835 2836 if (event_msg == NULL) 2837 return EINVAL; 2838 2839 /* Limit third parties to posting events for registered vendor codes only */ 2840 if (event_msg->vendor_code < min_vendor || 2841 event_msg->vendor_code > max_vendor) 2842 { 2843 return EINVAL; 2844 } 2845 2846 return kev_post_msg(event_msg); 2847} 2848 2849 2850int kev_post_msg(struct kev_msg *event_msg) 2851{ 2852 struct mbuf *m, *m2; 2853 struct kern_event_pcb *ev_pcb; 2854 struct kern_event_msg *ev; 2855 char *tmp; 2856 u_int32_t total_size; 2857 int i; 2858 2859 /* Verify the message is small enough to fit in one mbuf w/o cluster */ 2860 total_size = KEV_MSG_HEADER_SIZE; 2861 2862 for (i = 0; i < 5; i++) { 2863 if (event_msg->dv[i].data_length == 0) 2864 break; 2865 total_size += event_msg->dv[i].data_length; 2866 } 2867 2868 if (total_size > MLEN) { 2869 return EMSGSIZE; 2870 } 2871 2872 m = m_get(M_DONTWAIT, MT_DATA); 2873 if (m == 0) 2874 return ENOBUFS; 2875 2876 ev = mtod(m, struct kern_event_msg *); 2877 total_size = KEV_MSG_HEADER_SIZE; 2878 2879 tmp = (char *) &ev->event_data[0]; 2880 for (i = 0; i < 5; i++) { 2881 if (event_msg->dv[i].data_length == 0) 2882 break; 2883 2884 total_size += event_msg->dv[i].data_length; 2885 bcopy(event_msg->dv[i].data_ptr, tmp, 2886 event_msg->dv[i].data_length); 2887 tmp += event_msg->dv[i].data_length; 2888 } 2889 2890 ev->id = ++static_event_id; 2891 ev->total_size = total_size; 2892 ev->vendor_code = event_msg->vendor_code; 2893 ev->kev_class = event_msg->kev_class; 2894 ev->kev_subclass = event_msg->kev_subclass; 2895 ev->event_code = event_msg->event_code; 2896 2897 m->m_len = total_size; 2898 lck_mtx_lock(sys_mtx); 2899 for (ev_pcb = LIST_FIRST(&kern_event_head); 2900 ev_pcb; 2901 ev_pcb = LIST_NEXT(ev_pcb, ev_link)) { 2902 2903 if (ev_pcb->vendor_code_filter != KEV_ANY_VENDOR) { 2904 if (ev_pcb->vendor_code_filter != ev->vendor_code) 2905 continue; 2906 2907 if (ev_pcb->class_filter != KEV_ANY_CLASS) { 2908 if (ev_pcb->class_filter != ev->kev_class) 2909 continue; 2910 2911 if ((ev_pcb->subclass_filter != KEV_ANY_SUBCLASS) && 2912 (ev_pcb->subclass_filter != ev->kev_subclass)) 2913 continue; 2914 } 2915 } 2916 2917 m2 = m_copym(m, 0, m->m_len, M_NOWAIT); 2918 if (m2 == 0) { 2919 m_free(m); 2920 lck_mtx_unlock(sys_mtx); 2921 return ENOBUFS; 2922 } 2923 /* the socket is already locked because we hold the sys_mtx here */ 2924 if (sbappendrecord(&ev_pcb->ev_socket->so_rcv, m2)) 2925 sorwakeup(ev_pcb->ev_socket); 2926 } 2927 2928 m_free(m); 2929 lck_mtx_unlock(sys_mtx); 2930 return 0; 2931} 2932 2933static int 2934kev_control(struct socket *so, 2935 u_long cmd, 2936 caddr_t data, 2937 __unused struct ifnet *ifp, 2938 __unused struct proc *p) 2939{ 2940 struct kev_request *kev_req = (struct kev_request *) data; 2941 struct kern_event_pcb *ev_pcb; 2942 struct kev_vendor_code *kev_vendor; 2943 u_int32_t *id_value = (u_int32_t *) data; 2944 2945 2946 switch (cmd) { 2947 2948 case SIOCGKEVID: 2949 *id_value = static_event_id; 2950 break; 2951 2952 case SIOCSKEVFILT: 2953 ev_pcb = (struct kern_event_pcb *) so->so_pcb; 2954 ev_pcb->vendor_code_filter = kev_req->vendor_code; 2955 ev_pcb->class_filter = kev_req->kev_class; 2956 ev_pcb->subclass_filter = kev_req->kev_subclass; 2957 break; 2958 2959 case SIOCGKEVFILT: 2960 ev_pcb = (struct kern_event_pcb *) so->so_pcb; 2961 kev_req->vendor_code = ev_pcb->vendor_code_filter; 2962 kev_req->kev_class = ev_pcb->class_filter; 2963 kev_req->kev_subclass = ev_pcb->subclass_filter; 2964 break; 2965 2966 case SIOCGKEVVENDOR: 2967 kev_vendor = (struct kev_vendor_code*)data; 2968 2969 /* Make sure string is NULL terminated */ 2970 kev_vendor->vendor_string[KEV_VENDOR_CODE_MAX_STR_LEN-1] = 0; 2971 2972 return net_str_id_find_internal(kev_vendor->vendor_string, 2973 &kev_vendor->vendor_code, NSI_VENDOR_CODE, 0); 2974 2975 default: 2976 return ENOTSUP; 2977 } 2978 2979 return 0; 2980} 2981 2982#endif /* SOCKETS */ 2983 2984 2985int 2986fill_kqueueinfo(struct kqueue *kq, struct kqueue_info * kinfo) 2987{ 2988 struct vinfo_stat * st; 2989 2990 /* No need for the funnel as fd is kept alive */ 2991 2992 st = &kinfo->kq_stat; 2993 2994 st->vst_size = kq->kq_count; 2995 if (kq->kq_state & KQ_KEV64) 2996 st->vst_blksize = sizeof(struct kevent64_s); 2997 else 2998 st->vst_blksize = sizeof(struct kevent); 2999 st->vst_mode = S_IFIFO; 3000 if (kq->kq_state & KQ_SEL) 3001 kinfo->kq_state |= PROC_KQUEUE_SELECT; 3002 if (kq->kq_state & KQ_SLEEP) 3003 kinfo->kq_state |= PROC_KQUEUE_SLEEP; 3004 3005 return(0); 3006} 3007 3008 3009void 3010knote_markstayqueued(struct knote *kn) 3011{ 3012 kqlock(kn->kn_kq); 3013 kn->kn_status |= KN_STAYQUEUED; 3014 knote_enqueue(kn); 3015 kqunlock(kn->kn_kq); 3016} 3017