1/* 2 * Copyright (c) 2000-2014 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#include <sys/codesign.h> 85 86#include <kern/locks.h> 87#include <kern/clock.h> 88#include <kern/thread_call.h> 89#include <kern/sched_prim.h> 90#include <kern/zalloc.h> 91#include <kern/assert.h> 92 93#include <libkern/libkern.h> 94#include "net/net_str_id.h" 95 96#include <mach/task.h> 97 98#if VM_PRESSURE_EVENTS 99#include <kern/vm_pressure.h> 100#endif 101 102#if CONFIG_MEMORYSTATUS 103#include <sys/kern_memorystatus.h> 104#endif 105 106MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); 107 108#define KQ_EVENT NULL 109 110static inline void kqlock(struct kqueue *kq); 111static inline void kqunlock(struct kqueue *kq); 112 113static int kqlock2knoteuse(struct kqueue *kq, struct knote *kn); 114static int kqlock2knoteusewait(struct kqueue *kq, struct knote *kn); 115static int kqlock2knotedrop(struct kqueue *kq, struct knote *kn); 116static int knoteuse2kqlock(struct kqueue *kq, struct knote *kn); 117 118static void kqueue_wakeup(struct kqueue *kq, int closed); 119static int kqueue_read(struct fileproc *fp, struct uio *uio, 120 int flags, vfs_context_t ctx); 121static int kqueue_write(struct fileproc *fp, struct uio *uio, 122 int flags, vfs_context_t ctx); 123static int kqueue_ioctl(struct fileproc *fp, u_long com, caddr_t data, 124 vfs_context_t ctx); 125static int kqueue_select(struct fileproc *fp, int which, void *wql, 126 vfs_context_t ctx); 127static int kqueue_close(struct fileglob *fg, vfs_context_t ctx); 128static int kqueue_kqfilter(struct fileproc *fp, struct knote *kn, 129 vfs_context_t ctx); 130static int kqueue_drain(struct fileproc *fp, vfs_context_t ctx); 131 132static const struct fileops kqueueops = { 133 .fo_type = DTYPE_KQUEUE, 134 .fo_read = kqueue_read, 135 .fo_write = kqueue_write, 136 .fo_ioctl = kqueue_ioctl, 137 .fo_select = kqueue_select, 138 .fo_close = kqueue_close, 139 .fo_kqfilter = kqueue_kqfilter, 140 .fo_drain = kqueue_drain, 141}; 142 143static int kevent_internal(struct proc *p, int iskev64, user_addr_t changelist, 144 int nchanges, user_addr_t eventlist, int nevents, int fd, 145 user_addr_t utimeout, unsigned int flags, int32_t *retval); 146static int kevent_copyin(user_addr_t *addrp, struct kevent64_s *kevp, 147 struct proc *p, int iskev64); 148static int kevent_copyout(struct kevent64_s *kevp, user_addr_t *addrp, 149 struct proc *p, int iskev64); 150char * kevent_description(struct kevent64_s *kevp, char *s, size_t n); 151 152static int kevent_callback(struct kqueue *kq, struct kevent64_s *kevp, 153 void *data); 154static void kevent_continue(struct kqueue *kq, void *data, int error); 155static void kqueue_scan_continue(void *contp, wait_result_t wait_result); 156static int kqueue_process(struct kqueue *kq, kevent_callback_t callback, 157 void *data, int *countp, struct proc *p); 158static int kqueue_begin_processing(struct kqueue *kq); 159static void kqueue_end_processing(struct kqueue *kq); 160static int knote_process(struct knote *kn, kevent_callback_t callback, 161 void *data, struct kqtailq *inprocessp, struct proc *p); 162static void knote_put(struct knote *kn); 163static int knote_fdpattach(struct knote *kn, struct filedesc *fdp, 164 struct proc *p); 165static void knote_drop(struct knote *kn, struct proc *p); 166static void knote_activate(struct knote *kn, int); 167static void knote_deactivate(struct knote *kn); 168static void knote_enqueue(struct knote *kn); 169static void knote_dequeue(struct knote *kn); 170static struct knote *knote_alloc(void); 171static void knote_free(struct knote *kn); 172 173static int filt_fileattach(struct knote *kn); 174static struct filterops file_filtops = { 175 .f_isfd = 1, 176 .f_attach = filt_fileattach, 177}; 178 179static void filt_kqdetach(struct knote *kn); 180static int filt_kqueue(struct knote *kn, long hint); 181static struct filterops kqread_filtops = { 182 .f_isfd = 1, 183 .f_detach = filt_kqdetach, 184 .f_event = filt_kqueue, 185}; 186 187/* placeholder for not-yet-implemented filters */ 188static int filt_badattach(struct knote *kn); 189static struct filterops bad_filtops = { 190 .f_attach = filt_badattach, 191}; 192 193static int filt_procattach(struct knote *kn); 194static void filt_procdetach(struct knote *kn); 195static int filt_proc(struct knote *kn, long hint); 196static struct filterops proc_filtops = { 197 .f_attach = filt_procattach, 198 .f_detach = filt_procdetach, 199 .f_event = filt_proc, 200}; 201 202#if VM_PRESSURE_EVENTS 203static int filt_vmattach(struct knote *kn); 204static void filt_vmdetach(struct knote *kn); 205static int filt_vm(struct knote *kn, long hint); 206static struct filterops vm_filtops = { 207 .f_attach = filt_vmattach, 208 .f_detach = filt_vmdetach, 209 .f_event = filt_vm, 210}; 211#endif /* VM_PRESSURE_EVENTS */ 212 213#if CONFIG_MEMORYSTATUS 214extern struct filterops memorystatus_filtops; 215#endif /* CONFIG_MEMORYSTATUS */ 216 217extern struct filterops fs_filtops; 218 219extern struct filterops sig_filtops; 220 221/* Timer filter */ 222static int filt_timerattach(struct knote *kn); 223static void filt_timerdetach(struct knote *kn); 224static int filt_timer(struct knote *kn, long hint); 225static void filt_timertouch(struct knote *kn, struct kevent64_s *kev, 226 long type); 227static struct filterops timer_filtops = { 228 .f_attach = filt_timerattach, 229 .f_detach = filt_timerdetach, 230 .f_event = filt_timer, 231 .f_touch = filt_timertouch, 232}; 233 234/* Helpers */ 235static void filt_timerexpire(void *knx, void *param1); 236static int filt_timervalidate(struct knote *kn); 237static void filt_timerupdate(struct knote *kn); 238static void filt_timercancel(struct knote *kn); 239 240#define TIMER_RUNNING 0x1 241#define TIMER_CANCELWAIT 0x2 242 243static lck_mtx_t _filt_timerlock; 244static void filt_timerlock(void); 245static void filt_timerunlock(void); 246 247static zone_t knote_zone; 248 249#define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) 250 251#if 0 252extern struct filterops aio_filtops; 253#endif 254 255/* Mach portset filter */ 256extern struct filterops machport_filtops; 257 258/* User filter */ 259static int filt_userattach(struct knote *kn); 260static void filt_userdetach(struct knote *kn); 261static int filt_user(struct knote *kn, long hint); 262static void filt_usertouch(struct knote *kn, struct kevent64_s *kev, 263 long type); 264static struct filterops user_filtops = { 265 .f_attach = filt_userattach, 266 .f_detach = filt_userdetach, 267 .f_event = filt_user, 268 .f_touch = filt_usertouch, 269}; 270 271/* 272 * Table for all system-defined filters. 273 */ 274static struct filterops *sysfilt_ops[] = { 275 &file_filtops, /* EVFILT_READ */ 276 &file_filtops, /* EVFILT_WRITE */ 277#if 0 278 &aio_filtops, /* EVFILT_AIO */ 279#else 280 &bad_filtops, /* EVFILT_AIO */ 281#endif 282 &file_filtops, /* EVFILT_VNODE */ 283 &proc_filtops, /* EVFILT_PROC */ 284 &sig_filtops, /* EVFILT_SIGNAL */ 285 &timer_filtops, /* EVFILT_TIMER */ 286 &machport_filtops, /* EVFILT_MACHPORT */ 287 &fs_filtops, /* EVFILT_FS */ 288 &user_filtops, /* EVFILT_USER */ 289 &bad_filtops, /* unused */ 290#if VM_PRESSURE_EVENTS 291 &vm_filtops, /* EVFILT_VM */ 292#else 293 &bad_filtops, /* EVFILT_VM */ 294#endif 295 &file_filtops, /* EVFILT_SOCK */ 296#if CONFIG_MEMORYSTATUS 297 &memorystatus_filtops, /* EVFILT_MEMORYSTATUS */ 298#else 299 &bad_filtops, /* EVFILT_MEMORYSTATUS */ 300#endif 301}; 302 303/* 304 * kqueue/note lock attributes and implementations 305 * 306 * kqueues have locks, while knotes have use counts 307 * Most of the knote state is guarded by the object lock. 308 * the knote "inuse" count and status use the kqueue lock. 309 */ 310lck_grp_attr_t * kq_lck_grp_attr; 311lck_grp_t * kq_lck_grp; 312lck_attr_t * kq_lck_attr; 313 314static inline void 315kqlock(struct kqueue *kq) 316{ 317 lck_spin_lock(&kq->kq_lock); 318} 319 320static inline void 321kqunlock(struct kqueue *kq) 322{ 323 lck_spin_unlock(&kq->kq_lock); 324} 325 326/* 327 * Convert a kq lock to a knote use referece. 328 * 329 * If the knote is being dropped, we can't get 330 * a use reference, so just return with it 331 * still locked. 332 * - kq locked at entry 333 * - unlock on exit if we get the use reference 334 */ 335static int 336kqlock2knoteuse(struct kqueue *kq, struct knote *kn) 337{ 338 if (kn->kn_status & KN_DROPPING) 339 return (0); 340 kn->kn_inuse++; 341 kqunlock(kq); 342 return (1); 343} 344 345/* 346 * Convert a kq lock to a knote use referece, 347 * but wait for attach and drop events to complete. 348 * 349 * If the knote is being dropped, we can't get 350 * a use reference, so just return with it 351 * still locked. 352 * - kq locked at entry 353 * - kq always unlocked on exit 354 */ 355static int 356kqlock2knoteusewait(struct kqueue *kq, struct knote *kn) 357{ 358 if ((kn->kn_status & (KN_DROPPING | KN_ATTACHING)) != 0) { 359 kn->kn_status |= KN_USEWAIT; 360 wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, 361 &kn->kn_status, THREAD_UNINT, 0); 362 kqunlock(kq); 363 thread_block(THREAD_CONTINUE_NULL); 364 return (0); 365 } 366 kn->kn_inuse++; 367 kqunlock(kq); 368 return (1); 369} 370 371/* 372 * Convert from a knote use reference back to kq lock. 373 * 374 * Drop a use reference and wake any waiters if 375 * this is the last one. 376 * 377 * The exit return indicates if the knote is 378 * still alive - but the kqueue lock is taken 379 * unconditionally. 380 */ 381static int 382knoteuse2kqlock(struct kqueue *kq, struct knote *kn) 383{ 384 kqlock(kq); 385 if (--kn->kn_inuse == 0) { 386 if ((kn->kn_status & KN_ATTACHING) != 0) { 387 kn->kn_status &= ~KN_ATTACHING; 388 } 389 if ((kn->kn_status & KN_USEWAIT) != 0) { 390 kn->kn_status &= ~KN_USEWAIT; 391 wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, 392 &kn->kn_status, THREAD_AWAKENED); 393 } 394 } 395 return ((kn->kn_status & KN_DROPPING) == 0); 396} 397 398/* 399 * Convert a kq lock to a knote drop reference. 400 * 401 * If the knote is in use, wait for the use count 402 * to subside. We first mark our intention to drop 403 * it - keeping other users from "piling on." 404 * If we are too late, we have to wait for the 405 * other drop to complete. 406 * 407 * - kq locked at entry 408 * - always unlocked on exit. 409 * - caller can't hold any locks that would prevent 410 * the other dropper from completing. 411 */ 412static int 413kqlock2knotedrop(struct kqueue *kq, struct knote *kn) 414{ 415 int oktodrop; 416 417 oktodrop = ((kn->kn_status & (KN_DROPPING | KN_ATTACHING)) == 0); 418 kn->kn_status |= KN_DROPPING; 419 if (oktodrop) { 420 if (kn->kn_inuse == 0) { 421 kqunlock(kq); 422 return (oktodrop); 423 } 424 } 425 kn->kn_status |= KN_USEWAIT; 426 wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, &kn->kn_status, 427 THREAD_UNINT, 0); 428 kqunlock(kq); 429 thread_block(THREAD_CONTINUE_NULL); 430 return (oktodrop); 431} 432 433/* 434 * Release a knote use count reference. 435 */ 436static void 437knote_put(struct knote *kn) 438{ 439 struct kqueue *kq = kn->kn_kq; 440 441 kqlock(kq); 442 if (--kn->kn_inuse == 0) { 443 if ((kn->kn_status & KN_USEWAIT) != 0) { 444 kn->kn_status &= ~KN_USEWAIT; 445 wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, 446 &kn->kn_status, THREAD_AWAKENED); 447 } 448 } 449 kqunlock(kq); 450} 451 452static int 453filt_fileattach(struct knote *kn) 454{ 455 return (fo_kqfilter(kn->kn_fp, kn, vfs_context_current())); 456} 457 458#define f_flag f_fglob->fg_flag 459#define f_msgcount f_fglob->fg_msgcount 460#define f_cred f_fglob->fg_cred 461#define f_ops f_fglob->fg_ops 462#define f_offset f_fglob->fg_offset 463#define f_data f_fglob->fg_data 464 465static void 466filt_kqdetach(struct knote *kn) 467{ 468 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 469 470 kqlock(kq); 471 KNOTE_DETACH(&kq->kq_sel.si_note, kn); 472 kqunlock(kq); 473} 474 475/*ARGSUSED*/ 476static int 477filt_kqueue(struct knote *kn, __unused long hint) 478{ 479 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 480 481 kn->kn_data = kq->kq_count; 482 return (kn->kn_data > 0); 483} 484 485static int 486filt_procattach(struct knote *kn) 487{ 488 struct proc *p; 489 490 assert(PID_MAX < NOTE_PDATAMASK); 491 492 if ((kn->kn_sfflags & (NOTE_TRACK | NOTE_TRACKERR | NOTE_CHILD)) != 0) 493 return (ENOTSUP); 494 495 p = proc_find(kn->kn_id); 496 if (p == NULL) { 497 return (ESRCH); 498 } 499 500 const int NoteExitStatusBits = NOTE_EXIT | NOTE_EXITSTATUS; 501 502 if ((kn->kn_sfflags & NoteExitStatusBits) == NoteExitStatusBits) 503 do { 504 pid_t selfpid = proc_selfpid(); 505 506 if (p->p_ppid == selfpid) 507 break; /* parent => ok */ 508 509 if ((p->p_lflag & P_LTRACED) != 0 && 510 (p->p_oppid == selfpid)) 511 break; /* parent-in-waiting => ok */ 512 513 proc_rele(p); 514 return (EACCES); 515 } while (0); 516 517 proc_klist_lock(); 518 519 kn->kn_flags |= EV_CLEAR; /* automatically set */ 520 kn->kn_ptr.p_proc = p; /* store the proc handle */ 521 522 KNOTE_ATTACH(&p->p_klist, kn); 523 524 proc_klist_unlock(); 525 526 proc_rele(p); 527 528 return (0); 529} 530 531/* 532 * The knote may be attached to a different process, which may exit, 533 * leaving nothing for the knote to be attached to. In that case, 534 * the pointer to the process will have already been nulled out. 535 */ 536static void 537filt_procdetach(struct knote *kn) 538{ 539 struct proc *p; 540 541 proc_klist_lock(); 542 543 p = kn->kn_ptr.p_proc; 544 if (p != PROC_NULL) { 545 kn->kn_ptr.p_proc = PROC_NULL; 546 KNOTE_DETACH(&p->p_klist, kn); 547 } 548 549 proc_klist_unlock(); 550} 551 552static int 553filt_proc(struct knote *kn, long hint) 554{ 555 /* 556 * Note: a lot of bits in hint may be obtained from the knote 557 * To free some of those bits, see <rdar://problem/12592988> Freeing up 558 * bits in hint for filt_proc 559 */ 560 /* hint is 0 when called from above */ 561 if (hint != 0) { 562 u_int event; 563 564 /* ALWAYS CALLED WITH proc_klist_lock when (hint != 0) */ 565 566 /* 567 * mask off extra data 568 */ 569 event = (u_int)hint & NOTE_PCTRLMASK; 570 571 /* 572 * termination lifecycle events can happen while a debugger 573 * has reparented a process, in which case notifications 574 * should be quashed except to the tracing parent. When 575 * the debugger reaps the child (either via wait4(2) or 576 * process exit), the child will be reparented to the original 577 * parent and these knotes re-fired. 578 */ 579 if (event & NOTE_EXIT) { 580 if ((kn->kn_ptr.p_proc->p_oppid != 0) 581 && (kn->kn_kq->kq_p->p_pid != kn->kn_ptr.p_proc->p_ppid)) { 582 /* 583 * This knote is not for the current ptrace(2) parent, ignore. 584 */ 585 return 0; 586 } 587 } 588 589 /* 590 * if the user is interested in this event, record it. 591 */ 592 if (kn->kn_sfflags & event) 593 kn->kn_fflags |= event; 594 595#pragma clang diagnostic push 596#pragma clang diagnostic ignored "-Wdeprecated-declarations" 597 if ((event == NOTE_REAP) || ((event == NOTE_EXIT) && !(kn->kn_sfflags & NOTE_REAP))) { 598 kn->kn_flags |= (EV_EOF | EV_ONESHOT); 599 } 600#pragma clang diagnostic pop 601 602 603 /* 604 * The kernel has a wrapper in place that returns the same data 605 * as is collected here, in kn_data. Any changes to how 606 * NOTE_EXITSTATUS and NOTE_EXIT_DETAIL are collected 607 * should also be reflected in the proc_pidnoteexit() wrapper. 608 */ 609 if (event == NOTE_EXIT) { 610 kn->kn_data = 0; 611 if ((kn->kn_sfflags & NOTE_EXITSTATUS) != 0) { 612 kn->kn_fflags |= NOTE_EXITSTATUS; 613 kn->kn_data |= (hint & NOTE_PDATAMASK); 614 } 615 if ((kn->kn_sfflags & NOTE_EXIT_DETAIL) != 0) { 616 kn->kn_fflags |= NOTE_EXIT_DETAIL; 617 if ((kn->kn_ptr.p_proc->p_lflag & 618 P_LTERM_DECRYPTFAIL) != 0) { 619 kn->kn_data |= NOTE_EXIT_DECRYPTFAIL; 620 } 621 if ((kn->kn_ptr.p_proc->p_lflag & 622 P_LTERM_JETSAM) != 0) { 623 kn->kn_data |= NOTE_EXIT_MEMORY; 624 switch (kn->kn_ptr.p_proc->p_lflag & 625 P_JETSAM_MASK) { 626 case P_JETSAM_VMPAGESHORTAGE: 627 kn->kn_data |= NOTE_EXIT_MEMORY_VMPAGESHORTAGE; 628 break; 629 case P_JETSAM_VMTHRASHING: 630 kn->kn_data |= NOTE_EXIT_MEMORY_VMTHRASHING; 631 break; 632 case P_JETSAM_FCTHRASHING: 633 kn->kn_data |= NOTE_EXIT_MEMORY_FCTHRASHING; 634 break; 635 case P_JETSAM_VNODE: 636 kn->kn_data |= NOTE_EXIT_MEMORY_VNODE; 637 break; 638 case P_JETSAM_HIWAT: 639 kn->kn_data |= NOTE_EXIT_MEMORY_HIWAT; 640 break; 641 case P_JETSAM_PID: 642 kn->kn_data |= NOTE_EXIT_MEMORY_PID; 643 break; 644 case P_JETSAM_IDLEEXIT: 645 kn->kn_data |= NOTE_EXIT_MEMORY_IDLE; 646 break; 647 } 648 } 649 if ((kn->kn_ptr.p_proc->p_csflags & 650 CS_KILLED) != 0) { 651 kn->kn_data |= NOTE_EXIT_CSERROR; 652 } 653 } 654 } 655 } 656 657 /* atomic check, no locking need when called from above */ 658 return (kn->kn_fflags != 0); 659} 660 661#if VM_PRESSURE_EVENTS 662/* 663 * Virtual memory kevents 664 * 665 * author: Matt Jacobson [matthew_jacobson@apple.com] 666 */ 667 668static int 669filt_vmattach(struct knote *kn) 670{ 671 /* 672 * The note will be cleared once the information has been flushed to 673 * the client. If there is still pressure, we will be re-alerted. 674 */ 675 kn->kn_flags |= EV_CLEAR; 676 return (vm_knote_register(kn)); 677} 678 679static void 680filt_vmdetach(struct knote *kn) 681{ 682 vm_knote_unregister(kn); 683} 684 685static int 686filt_vm(struct knote *kn, long hint) 687{ 688 /* hint == 0 means this is just an alive? check (always true) */ 689 if (hint != 0) { 690 const pid_t pid = (pid_t)hint; 691 if ((kn->kn_sfflags & NOTE_VM_PRESSURE) && 692 (kn->kn_kq->kq_p->p_pid == pid)) { 693 kn->kn_fflags |= NOTE_VM_PRESSURE; 694 } 695 } 696 697 return (kn->kn_fflags != 0); 698} 699#endif /* VM_PRESSURE_EVENTS */ 700 701/* 702 * filt_timervalidate - process data from user 703 * 704 * Converts to either interval or deadline format. 705 * 706 * The saved-data field in the knote contains the 707 * time value. The saved filter-flags indicates 708 * the unit of measurement. 709 * 710 * After validation, either the saved-data field 711 * contains the interval in absolute time, or ext[0] 712 * contains the expected deadline. If that deadline 713 * is in the past, ext[0] is 0. 714 * 715 * Returns EINVAL for unrecognized units of time. 716 * 717 * Timer filter lock is held. 718 * 719 */ 720static int 721filt_timervalidate(struct knote *kn) 722{ 723 uint64_t multiplier; 724 uint64_t raw = 0; 725 726 switch (kn->kn_sfflags & (NOTE_SECONDS|NOTE_USECONDS|NOTE_NSECONDS)) { 727 case NOTE_SECONDS: 728 multiplier = NSEC_PER_SEC; 729 break; 730 case NOTE_USECONDS: 731 multiplier = NSEC_PER_USEC; 732 break; 733 case NOTE_NSECONDS: 734 multiplier = 1; 735 break; 736 case 0: /* milliseconds (default) */ 737 multiplier = NSEC_PER_SEC / 1000; 738 break; 739 default: 740 return (EINVAL); 741 } 742 743 /* transform the slop delta(leeway) in kn_ext[1] if passed to same time scale */ 744 if(kn->kn_sfflags & NOTE_LEEWAY){ 745 nanoseconds_to_absolutetime((uint64_t)kn->kn_ext[1] * multiplier, &raw); 746 kn->kn_ext[1] = raw; 747 } 748 749 nanoseconds_to_absolutetime((uint64_t)kn->kn_sdata * multiplier, &raw); 750 751 kn->kn_ext[0] = 0; 752 kn->kn_sdata = 0; 753 754 if (kn->kn_sfflags & NOTE_ABSOLUTE) { 755 clock_sec_t seconds; 756 clock_nsec_t nanoseconds; 757 uint64_t now; 758 759 clock_get_calendar_nanotime(&seconds, &nanoseconds); 760 nanoseconds_to_absolutetime((uint64_t)seconds * NSEC_PER_SEC + 761 nanoseconds, &now); 762 763 if (raw < now) { 764 /* time has already passed */ 765 kn->kn_ext[0] = 0; 766 } else { 767 raw -= now; 768 clock_absolutetime_interval_to_deadline(raw, 769 &kn->kn_ext[0]); 770 } 771 } else { 772 kn->kn_sdata = raw; 773 } 774 775 return (0); 776} 777 778/* 779 * filt_timerupdate - compute the next deadline 780 * 781 * Repeating timers store their interval in kn_sdata. Absolute 782 * timers have already calculated the deadline, stored in ext[0]. 783 * 784 * On return, the next deadline (or zero if no deadline is needed) 785 * is stored in kn_ext[0]. 786 * 787 * Timer filter lock is held. 788 */ 789static void 790filt_timerupdate(struct knote *kn) 791{ 792 /* if there's no interval, deadline is just in kn_ext[0] */ 793 if (kn->kn_sdata == 0) 794 return; 795 796 /* if timer hasn't fired before, fire in interval nsecs */ 797 if (kn->kn_ext[0] == 0) { 798 clock_absolutetime_interval_to_deadline(kn->kn_sdata, 799 &kn->kn_ext[0]); 800 } else { 801 /* 802 * If timer has fired before, schedule the next pop 803 * relative to the last intended deadline. 804 * 805 * We could check for whether the deadline has expired, 806 * but the thread call layer can handle that. 807 */ 808 kn->kn_ext[0] += kn->kn_sdata; 809 } 810} 811 812/* 813 * filt_timerexpire - the timer callout routine 814 * 815 * Just propagate the timer event into the knote 816 * filter routine (by going through the knote 817 * synchronization point). Pass a hint to 818 * indicate this is a real event, not just a 819 * query from above. 820 */ 821static void 822filt_timerexpire(void *knx, __unused void *spare) 823{ 824 struct klist timer_list; 825 struct knote *kn = knx; 826 827 filt_timerlock(); 828 829 kn->kn_hookid &= ~TIMER_RUNNING; 830 831 /* no "object" for timers, so fake a list */ 832 SLIST_INIT(&timer_list); 833 SLIST_INSERT_HEAD(&timer_list, kn, kn_selnext); 834 KNOTE(&timer_list, 1); 835 836 /* if someone is waiting for timer to pop */ 837 if (kn->kn_hookid & TIMER_CANCELWAIT) { 838 struct kqueue *kq = kn->kn_kq; 839 wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, &kn->kn_hook, 840 THREAD_AWAKENED); 841 } 842 843 filt_timerunlock(); 844} 845 846/* 847 * Cancel a running timer (or wait for the pop). 848 * Timer filter lock is held. 849 */ 850static void 851filt_timercancel(struct knote *kn) 852{ 853 struct kqueue *kq = kn->kn_kq; 854 thread_call_t callout = kn->kn_hook; 855 boolean_t cancelled; 856 857 if (kn->kn_hookid & TIMER_RUNNING) { 858 /* cancel the callout if we can */ 859 cancelled = thread_call_cancel(callout); 860 if (cancelled) { 861 kn->kn_hookid &= ~TIMER_RUNNING; 862 } else { 863 /* we have to wait for the expire routine. */ 864 kn->kn_hookid |= TIMER_CANCELWAIT; 865 wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, 866 &kn->kn_hook, THREAD_UNINT, 0); 867 filt_timerunlock(); 868 thread_block(THREAD_CONTINUE_NULL); 869 filt_timerlock(); 870 assert((kn->kn_hookid & TIMER_RUNNING) == 0); 871 } 872 } 873} 874 875/* 876 * Allocate a thread call for the knote's lifetime, and kick off the timer. 877 */ 878static int 879filt_timerattach(struct knote *kn) 880{ 881 thread_call_t callout; 882 int error; 883 884 callout = thread_call_allocate(filt_timerexpire, kn); 885 if (NULL == callout) 886 return (ENOMEM); 887 888 filt_timerlock(); 889 error = filt_timervalidate(kn); 890 if (error != 0) { 891 filt_timerunlock(); 892 return (error); 893 } 894 895 kn->kn_hook = (void*)callout; 896 kn->kn_hookid = 0; 897 898 /* absolute=EV_ONESHOT */ 899 if (kn->kn_sfflags & NOTE_ABSOLUTE) 900 kn->kn_flags |= EV_ONESHOT; 901 902 filt_timerupdate(kn); 903 if (kn->kn_ext[0]) { 904 kn->kn_flags |= EV_CLEAR; 905 unsigned int timer_flags = 0; 906 if (kn->kn_sfflags & NOTE_CRITICAL) 907 timer_flags |= THREAD_CALL_DELAY_USER_CRITICAL; 908 else if (kn->kn_sfflags & NOTE_BACKGROUND) 909 timer_flags |= THREAD_CALL_DELAY_USER_BACKGROUND; 910 else 911 timer_flags |= THREAD_CALL_DELAY_USER_NORMAL; 912 913 if (kn->kn_sfflags & NOTE_LEEWAY) 914 timer_flags |= THREAD_CALL_DELAY_LEEWAY; 915 916 thread_call_enter_delayed_with_leeway(callout, NULL, 917 kn->kn_ext[0], kn->kn_ext[1], timer_flags); 918 919 kn->kn_hookid |= TIMER_RUNNING; 920 } else { 921 /* fake immediate */ 922 kn->kn_data = 1; 923 } 924 925 filt_timerunlock(); 926 return (0); 927} 928 929/* 930 * Shut down the timer if it's running, and free the callout. 931 */ 932static void 933filt_timerdetach(struct knote *kn) 934{ 935 thread_call_t callout; 936 937 filt_timerlock(); 938 939 callout = (thread_call_t)kn->kn_hook; 940 filt_timercancel(kn); 941 942 filt_timerunlock(); 943 944 thread_call_free(callout); 945} 946 947 948 949static int 950filt_timer(struct knote *kn, long hint) 951{ 952 int result; 953 954 if (hint) { 955 /* real timer pop -- timer lock held by filt_timerexpire */ 956 kn->kn_data++; 957 958 if (((kn->kn_hookid & TIMER_CANCELWAIT) == 0) && 959 ((kn->kn_flags & EV_ONESHOT) == 0)) { 960 961 /* evaluate next time to fire */ 962 filt_timerupdate(kn); 963 964 if (kn->kn_ext[0]) { 965 unsigned int timer_flags = 0; 966 967 /* keep the callout and re-arm */ 968 if (kn->kn_sfflags & NOTE_CRITICAL) 969 timer_flags |= THREAD_CALL_DELAY_USER_CRITICAL; 970 else if (kn->kn_sfflags & NOTE_BACKGROUND) 971 timer_flags |= THREAD_CALL_DELAY_USER_BACKGROUND; 972 else 973 timer_flags |= THREAD_CALL_DELAY_USER_NORMAL; 974 975 if (kn->kn_sfflags & NOTE_LEEWAY) 976 timer_flags |= THREAD_CALL_DELAY_LEEWAY; 977 978 thread_call_enter_delayed_with_leeway(kn->kn_hook, NULL, 979 kn->kn_ext[0], kn->kn_ext[1], timer_flags); 980 981 kn->kn_hookid |= TIMER_RUNNING; 982 } 983 } 984 985 return (1); 986 } 987 988 /* user-query */ 989 filt_timerlock(); 990 991 result = (kn->kn_data != 0); 992 993 filt_timerunlock(); 994 995 return (result); 996} 997 998 999/* 1000 * filt_timertouch - update knote with new user input 1001 * 1002 * Cancel and restart the timer based on new user data. When 1003 * the user picks up a knote, clear the count of how many timer 1004 * pops have gone off (in kn_data). 1005 */ 1006static void 1007filt_timertouch(struct knote *kn, struct kevent64_s *kev, long type) 1008{ 1009 int error; 1010 filt_timerlock(); 1011 1012 switch (type) { 1013 case EVENT_REGISTER: 1014 /* cancel current call */ 1015 filt_timercancel(kn); 1016 1017 /* recalculate deadline */ 1018 kn->kn_sdata = kev->data; 1019 kn->kn_sfflags = kev->fflags; 1020 kn->kn_ext[0] = kev->ext[0]; 1021 kn->kn_ext[1] = kev->ext[1]; 1022 1023 error = filt_timervalidate(kn); 1024 if (error) { 1025 /* no way to report error, so mark it in the knote */ 1026 kn->kn_flags |= EV_ERROR; 1027 kn->kn_data = error; 1028 break; 1029 } 1030 1031 /* start timer if necessary */ 1032 filt_timerupdate(kn); 1033 1034 if (kn->kn_ext[0]) { 1035 unsigned int timer_flags = 0; 1036 if (kn->kn_sfflags & NOTE_CRITICAL) 1037 timer_flags |= THREAD_CALL_DELAY_USER_CRITICAL; 1038 else if (kn->kn_sfflags & NOTE_BACKGROUND) 1039 timer_flags |= THREAD_CALL_DELAY_USER_BACKGROUND; 1040 else 1041 timer_flags |= THREAD_CALL_DELAY_USER_NORMAL; 1042 1043 if (kn->kn_sfflags & NOTE_LEEWAY) 1044 timer_flags |= THREAD_CALL_DELAY_LEEWAY; 1045 1046 thread_call_enter_delayed_with_leeway(kn->kn_hook, NULL, 1047 kn->kn_ext[0], kn->kn_ext[1], timer_flags); 1048 1049 kn->kn_hookid |= TIMER_RUNNING; 1050 } else { 1051 /* pretend the timer has fired */ 1052 kn->kn_data = 1; 1053 } 1054 1055 break; 1056 1057 case EVENT_PROCESS: 1058 /* reset the timer pop count in kn_data */ 1059 *kev = kn->kn_kevent; 1060 kev->ext[0] = 0; 1061 kn->kn_data = 0; 1062 if (kn->kn_flags & EV_CLEAR) 1063 kn->kn_fflags = 0; 1064 break; 1065 default: 1066 panic("%s: - invalid type (%ld)", __func__, type); 1067 break; 1068 } 1069 1070 filt_timerunlock(); 1071} 1072 1073static void 1074filt_timerlock(void) 1075{ 1076 lck_mtx_lock(&_filt_timerlock); 1077} 1078 1079static void 1080filt_timerunlock(void) 1081{ 1082 lck_mtx_unlock(&_filt_timerlock); 1083} 1084 1085static int 1086filt_userattach(struct knote *kn) 1087{ 1088 /* EVFILT_USER knotes are not attached to anything in the kernel */ 1089 kn->kn_hook = NULL; 1090 if (kn->kn_fflags & NOTE_TRIGGER) { 1091 kn->kn_hookid = 1; 1092 } else { 1093 kn->kn_hookid = 0; 1094 } 1095 return (0); 1096} 1097 1098static void 1099filt_userdetach(__unused struct knote *kn) 1100{ 1101 /* EVFILT_USER knotes are not attached to anything in the kernel */ 1102} 1103 1104static int 1105filt_user(struct knote *kn, __unused long hint) 1106{ 1107 return (kn->kn_hookid); 1108} 1109 1110static void 1111filt_usertouch(struct knote *kn, struct kevent64_s *kev, long type) 1112{ 1113 uint32_t ffctrl; 1114 switch (type) { 1115 case EVENT_REGISTER: 1116 if (kev->fflags & NOTE_TRIGGER) { 1117 kn->kn_hookid = 1; 1118 } 1119 1120 ffctrl = kev->fflags & NOTE_FFCTRLMASK; 1121 kev->fflags &= NOTE_FFLAGSMASK; 1122 switch (ffctrl) { 1123 case NOTE_FFNOP: 1124 break; 1125 case NOTE_FFAND: 1126 OSBitAndAtomic(kev->fflags, &kn->kn_sfflags); 1127 break; 1128 case NOTE_FFOR: 1129 OSBitOrAtomic(kev->fflags, &kn->kn_sfflags); 1130 break; 1131 case NOTE_FFCOPY: 1132 kn->kn_sfflags = kev->fflags; 1133 break; 1134 } 1135 kn->kn_sdata = kev->data; 1136 break; 1137 case EVENT_PROCESS: 1138 *kev = kn->kn_kevent; 1139 kev->fflags = (volatile UInt32)kn->kn_sfflags; 1140 kev->data = kn->kn_sdata; 1141 if (kn->kn_flags & EV_CLEAR) { 1142 kn->kn_hookid = 0; 1143 kn->kn_data = 0; 1144 kn->kn_fflags = 0; 1145 } 1146 break; 1147 default: 1148 panic("%s: - invalid type (%ld)", __func__, type); 1149 break; 1150 } 1151} 1152 1153/* 1154 * JMM - placeholder for not-yet-implemented filters 1155 */ 1156static int 1157filt_badattach(__unused struct knote *kn) 1158{ 1159 return (ENOTSUP); 1160} 1161 1162struct kqueue * 1163kqueue_alloc(struct proc *p) 1164{ 1165 struct filedesc *fdp = p->p_fd; 1166 struct kqueue *kq; 1167 1168 MALLOC_ZONE(kq, struct kqueue *, sizeof (struct kqueue), M_KQUEUE, 1169 M_WAITOK); 1170 if (kq != NULL) { 1171 wait_queue_set_t wqs; 1172 1173 wqs = wait_queue_set_alloc(SYNC_POLICY_FIFO | 1174 SYNC_POLICY_PREPOST); 1175 if (wqs != NULL) { 1176 bzero(kq, sizeof (struct kqueue)); 1177 lck_spin_init(&kq->kq_lock, kq_lck_grp, kq_lck_attr); 1178 TAILQ_INIT(&kq->kq_head); 1179 kq->kq_wqs = wqs; 1180 kq->kq_p = p; 1181 } else { 1182 FREE_ZONE(kq, sizeof (struct kqueue), M_KQUEUE); 1183 } 1184 } 1185 1186 if (fdp->fd_knlistsize < 0) { 1187 proc_fdlock(p); 1188 if (fdp->fd_knlistsize < 0) 1189 fdp->fd_knlistsize = 0; /* this process has had a kq */ 1190 proc_fdunlock(p); 1191 } 1192 1193 return (kq); 1194} 1195 1196/* 1197 * kqueue_dealloc - detach all knotes from a kqueue and free it 1198 * 1199 * We walk each list looking for knotes referencing this 1200 * this kqueue. If we find one, we try to drop it. But 1201 * if we fail to get a drop reference, that will wait 1202 * until it is dropped. So, we can just restart again 1203 * safe in the assumption that the list will eventually 1204 * not contain any more references to this kqueue (either 1205 * we dropped them all, or someone else did). 1206 * 1207 * Assumes no new events are being added to the kqueue. 1208 * Nothing locked on entry or exit. 1209 */ 1210void 1211kqueue_dealloc(struct kqueue *kq) 1212{ 1213 struct proc *p = kq->kq_p; 1214 struct filedesc *fdp = p->p_fd; 1215 struct knote *kn; 1216 int i; 1217 1218 proc_fdlock(p); 1219 for (i = 0; i < fdp->fd_knlistsize; i++) { 1220 kn = SLIST_FIRST(&fdp->fd_knlist[i]); 1221 while (kn != NULL) { 1222 if (kq == kn->kn_kq) { 1223 kqlock(kq); 1224 proc_fdunlock(p); 1225 /* drop it ourselves or wait */ 1226 if (kqlock2knotedrop(kq, kn)) { 1227 kn->kn_fop->f_detach(kn); 1228 knote_drop(kn, p); 1229 } 1230 proc_fdlock(p); 1231 /* start over at beginning of list */ 1232 kn = SLIST_FIRST(&fdp->fd_knlist[i]); 1233 continue; 1234 } 1235 kn = SLIST_NEXT(kn, kn_link); 1236 } 1237 } 1238 if (fdp->fd_knhashmask != 0) { 1239 for (i = 0; i < (int)fdp->fd_knhashmask + 1; i++) { 1240 kn = SLIST_FIRST(&fdp->fd_knhash[i]); 1241 while (kn != NULL) { 1242 if (kq == kn->kn_kq) { 1243 kqlock(kq); 1244 proc_fdunlock(p); 1245 /* drop it ourselves or wait */ 1246 if (kqlock2knotedrop(kq, kn)) { 1247 kn->kn_fop->f_detach(kn); 1248 knote_drop(kn, p); 1249 } 1250 proc_fdlock(p); 1251 /* start over at beginning of list */ 1252 kn = SLIST_FIRST(&fdp->fd_knhash[i]); 1253 continue; 1254 } 1255 kn = SLIST_NEXT(kn, kn_link); 1256 } 1257 } 1258 } 1259 proc_fdunlock(p); 1260 1261 /* 1262 * before freeing the wait queue set for this kqueue, 1263 * make sure it is unlinked from all its containing (select) sets. 1264 */ 1265 wait_queue_unlink_all((wait_queue_t)kq->kq_wqs); 1266 wait_queue_set_free(kq->kq_wqs); 1267 lck_spin_destroy(&kq->kq_lock, kq_lck_grp); 1268 FREE_ZONE(kq, sizeof (struct kqueue), M_KQUEUE); 1269} 1270 1271int 1272kqueue_body(struct proc *p, fp_allocfn_t fp_zalloc, void *cra, int32_t *retval) 1273{ 1274 struct kqueue *kq; 1275 struct fileproc *fp; 1276 int fd, error; 1277 1278 error = falloc_withalloc(p, 1279 &fp, &fd, vfs_context_current(), fp_zalloc, cra); 1280 if (error) { 1281 return (error); 1282 } 1283 1284 kq = kqueue_alloc(p); 1285 if (kq == NULL) { 1286 fp_free(p, fd, fp); 1287 return (ENOMEM); 1288 } 1289 1290 fp->f_flag = FREAD | FWRITE; 1291 fp->f_ops = &kqueueops; 1292 fp->f_data = kq; 1293 1294 proc_fdlock(p); 1295 *fdflags(p, fd) |= UF_EXCLOSE; 1296 procfdtbl_releasefd(p, fd, NULL); 1297 fp_drop(p, fd, fp, 1); 1298 proc_fdunlock(p); 1299 1300 *retval = fd; 1301 return (error); 1302} 1303 1304int 1305kqueue(struct proc *p, __unused struct kqueue_args *uap, int32_t *retval) 1306{ 1307 return (kqueue_body(p, fileproc_alloc_init, NULL, retval)); 1308} 1309 1310static int 1311kevent_copyin(user_addr_t *addrp, struct kevent64_s *kevp, struct proc *p, 1312 int iskev64) 1313{ 1314 int advance; 1315 int error; 1316 1317 if (iskev64) { 1318 advance = sizeof (struct kevent64_s); 1319 error = copyin(*addrp, (caddr_t)kevp, advance); 1320 } else if (IS_64BIT_PROCESS(p)) { 1321 struct user64_kevent kev64; 1322 bzero(kevp, sizeof (struct kevent64_s)); 1323 1324 advance = sizeof (kev64); 1325 error = copyin(*addrp, (caddr_t)&kev64, advance); 1326 if (error) 1327 return (error); 1328 kevp->ident = kev64.ident; 1329 kevp->filter = kev64.filter; 1330 kevp->flags = kev64.flags; 1331 kevp->fflags = kev64.fflags; 1332 kevp->data = kev64.data; 1333 kevp->udata = kev64.udata; 1334 } else { 1335 struct user32_kevent kev32; 1336 bzero(kevp, sizeof (struct kevent64_s)); 1337 1338 advance = sizeof (kev32); 1339 error = copyin(*addrp, (caddr_t)&kev32, advance); 1340 if (error) 1341 return (error); 1342 kevp->ident = (uintptr_t)kev32.ident; 1343 kevp->filter = kev32.filter; 1344 kevp->flags = kev32.flags; 1345 kevp->fflags = kev32.fflags; 1346 kevp->data = (intptr_t)kev32.data; 1347 kevp->udata = CAST_USER_ADDR_T(kev32.udata); 1348 } 1349 if (!error) 1350 *addrp += advance; 1351 return (error); 1352} 1353 1354static int 1355kevent_copyout(struct kevent64_s *kevp, user_addr_t *addrp, struct proc *p, 1356 int iskev64) 1357{ 1358 int advance; 1359 int error; 1360 1361 if (iskev64) { 1362 advance = sizeof (struct kevent64_s); 1363 error = copyout((caddr_t)kevp, *addrp, advance); 1364 } else if (IS_64BIT_PROCESS(p)) { 1365 struct user64_kevent kev64; 1366 1367 /* 1368 * deal with the special case of a user-supplied 1369 * value of (uintptr_t)-1. 1370 */ 1371 kev64.ident = (kevp->ident == (uintptr_t)-1) ? 1372 (uint64_t)-1LL : (uint64_t)kevp->ident; 1373 1374 kev64.filter = kevp->filter; 1375 kev64.flags = kevp->flags; 1376 kev64.fflags = kevp->fflags; 1377 kev64.data = (int64_t) kevp->data; 1378 kev64.udata = kevp->udata; 1379 advance = sizeof (kev64); 1380 error = copyout((caddr_t)&kev64, *addrp, advance); 1381 } else { 1382 struct user32_kevent kev32; 1383 1384 kev32.ident = (uint32_t)kevp->ident; 1385 kev32.filter = kevp->filter; 1386 kev32.flags = kevp->flags; 1387 kev32.fflags = kevp->fflags; 1388 kev32.data = (int32_t)kevp->data; 1389 kev32.udata = kevp->udata; 1390 advance = sizeof (kev32); 1391 error = copyout((caddr_t)&kev32, *addrp, advance); 1392 } 1393 if (!error) 1394 *addrp += advance; 1395 return (error); 1396} 1397 1398/* 1399 * kevent_continue - continue a kevent syscall after blocking 1400 * 1401 * assume we inherit a use count on the kq fileglob. 1402 */ 1403 1404static void 1405kevent_continue(__unused struct kqueue *kq, void *data, int error) 1406{ 1407 struct _kevent *cont_args; 1408 struct fileproc *fp; 1409 int32_t *retval; 1410 int noutputs; 1411 int fd; 1412 struct proc *p = current_proc(); 1413 1414 cont_args = (struct _kevent *)data; 1415 noutputs = cont_args->eventout; 1416 retval = cont_args->retval; 1417 fd = cont_args->fd; 1418 fp = cont_args->fp; 1419 1420 fp_drop(p, fd, fp, 0); 1421 1422 /* don't restart after signals... */ 1423 if (error == ERESTART) 1424 error = EINTR; 1425 else if (error == EWOULDBLOCK) 1426 error = 0; 1427 if (error == 0) 1428 *retval = noutputs; 1429 unix_syscall_return(error); 1430} 1431 1432/* 1433 * kevent - [syscall] register and wait for kernel events 1434 * 1435 */ 1436int 1437kevent(struct proc *p, struct kevent_args *uap, int32_t *retval) 1438{ 1439 return (kevent_internal(p, 1440 0, 1441 uap->changelist, 1442 uap->nchanges, 1443 uap->eventlist, 1444 uap->nevents, 1445 uap->fd, 1446 uap->timeout, 1447 0, /* no flags from old kevent() call */ 1448 retval)); 1449} 1450 1451int 1452kevent64(struct proc *p, struct kevent64_args *uap, int32_t *retval) 1453{ 1454 return (kevent_internal(p, 1455 1, 1456 uap->changelist, 1457 uap->nchanges, 1458 uap->eventlist, 1459 uap->nevents, 1460 uap->fd, 1461 uap->timeout, 1462 uap->flags, 1463 retval)); 1464} 1465 1466static int 1467kevent_internal(struct proc *p, int iskev64, user_addr_t changelist, 1468 int nchanges, user_addr_t ueventlist, int nevents, int fd, 1469 user_addr_t utimeout, __unused unsigned int flags, 1470 int32_t *retval) 1471{ 1472 struct _kevent *cont_args; 1473 uthread_t ut; 1474 struct kqueue *kq; 1475 struct fileproc *fp; 1476 struct kevent64_s kev; 1477 int error, noutputs; 1478 struct timeval atv; 1479 1480 /* convert timeout to absolute - if we have one */ 1481 if (utimeout != USER_ADDR_NULL) { 1482 struct timeval rtv; 1483 if (IS_64BIT_PROCESS(p)) { 1484 struct user64_timespec ts; 1485 error = copyin(utimeout, &ts, sizeof(ts)); 1486 if ((ts.tv_sec & 0xFFFFFFFF00000000ull) != 0) 1487 error = EINVAL; 1488 else 1489 TIMESPEC_TO_TIMEVAL(&rtv, &ts); 1490 } else { 1491 struct user32_timespec ts; 1492 error = copyin(utimeout, &ts, sizeof(ts)); 1493 TIMESPEC_TO_TIMEVAL(&rtv, &ts); 1494 } 1495 if (error) 1496 return (error); 1497 if (itimerfix(&rtv)) 1498 return (EINVAL); 1499 getmicrouptime(&atv); 1500 timevaladd(&atv, &rtv); 1501 } else { 1502 atv.tv_sec = 0; 1503 atv.tv_usec = 0; 1504 } 1505 1506 /* get a usecount for the kq itself */ 1507 if ((error = fp_getfkq(p, fd, &fp, &kq)) != 0) 1508 return (error); 1509 1510 /* each kq should only be used for events of one type */ 1511 kqlock(kq); 1512 if (kq->kq_state & (KQ_KEV32 | KQ_KEV64)) { 1513 if (((iskev64 && (kq->kq_state & KQ_KEV32)) || 1514 (!iskev64 && (kq->kq_state & KQ_KEV64)))) { 1515 error = EINVAL; 1516 kqunlock(kq); 1517 goto errorout; 1518 } 1519 } else { 1520 kq->kq_state |= (iskev64 ? KQ_KEV64 : KQ_KEV32); 1521 } 1522 kqunlock(kq); 1523 1524 /* register all the change requests the user provided... */ 1525 noutputs = 0; 1526 while (nchanges > 0 && error == 0) { 1527 error = kevent_copyin(&changelist, &kev, p, iskev64); 1528 if (error) 1529 break; 1530 1531 kev.flags &= ~EV_SYSFLAGS; 1532 error = kevent_register(kq, &kev, p); 1533 if ((error || (kev.flags & EV_RECEIPT)) && nevents > 0) { 1534 kev.flags = EV_ERROR; 1535 kev.data = error; 1536 error = kevent_copyout(&kev, &ueventlist, p, iskev64); 1537 if (error == 0) { 1538 nevents--; 1539 noutputs++; 1540 } 1541 } 1542 nchanges--; 1543 } 1544 1545 /* store the continuation/completion data in the uthread */ 1546 ut = (uthread_t)get_bsdthread_info(current_thread()); 1547 cont_args = &ut->uu_kevent.ss_kevent; 1548 cont_args->fp = fp; 1549 cont_args->fd = fd; 1550 cont_args->retval = retval; 1551 cont_args->eventlist = ueventlist; 1552 cont_args->eventcount = nevents; 1553 cont_args->eventout = noutputs; 1554 cont_args->eventsize = iskev64; 1555 1556 if (nevents > 0 && noutputs == 0 && error == 0) 1557 error = kqueue_scan(kq, kevent_callback, 1558 kevent_continue, cont_args, 1559 &atv, p); 1560 kevent_continue(kq, cont_args, error); 1561 1562errorout: 1563 fp_drop(p, fd, fp, 0); 1564 return (error); 1565} 1566 1567 1568/* 1569 * kevent_callback - callback for each individual event 1570 * 1571 * called with nothing locked 1572 * caller holds a reference on the kqueue 1573 */ 1574static int 1575kevent_callback(__unused struct kqueue *kq, struct kevent64_s *kevp, 1576 void *data) 1577{ 1578 struct _kevent *cont_args; 1579 int error; 1580 int iskev64; 1581 1582 cont_args = (struct _kevent *)data; 1583 assert(cont_args->eventout < cont_args->eventcount); 1584 1585 iskev64 = cont_args->eventsize; 1586 1587 /* 1588 * Copy out the appropriate amount of event data for this user. 1589 */ 1590 error = kevent_copyout(kevp, &cont_args->eventlist, current_proc(), 1591 iskev64); 1592 1593 /* 1594 * If there isn't space for additional events, return 1595 * a harmless error to stop the processing here 1596 */ 1597 if (error == 0 && ++cont_args->eventout == cont_args->eventcount) 1598 error = EWOULDBLOCK; 1599 return (error); 1600} 1601 1602/* 1603 * kevent_description - format a description of a kevent for diagnostic output 1604 * 1605 * called with a 128-byte string buffer 1606 */ 1607 1608char * 1609kevent_description(struct kevent64_s *kevp, char *s, size_t n) 1610{ 1611 snprintf(s, n, 1612 "kevent=" 1613 "{.ident=%#llx, .filter=%d, .flags=%#x, .fflags=%#x, .data=%#llx, .udata=%#llx, .ext[0]=%#llx, .ext[1]=%#llx}", 1614 kevp->ident, 1615 kevp->filter, 1616 kevp->flags, 1617 kevp->fflags, 1618 kevp->data, 1619 kevp->udata, 1620 kevp->ext[0], 1621 kevp->ext[1]); 1622 1623 return (s); 1624} 1625 1626/* 1627 * kevent_register - add a new event to a kqueue 1628 * 1629 * Creates a mapping between the event source and 1630 * the kqueue via a knote data structure. 1631 * 1632 * Because many/most the event sources are file 1633 * descriptor related, the knote is linked off 1634 * the filedescriptor table for quick access. 1635 * 1636 * called with nothing locked 1637 * caller holds a reference on the kqueue 1638 */ 1639 1640int 1641kevent_register(struct kqueue *kq, struct kevent64_s *kev, 1642 __unused struct proc *ctxp) 1643{ 1644 struct proc *p = kq->kq_p; 1645 struct filedesc *fdp = p->p_fd; 1646 struct filterops *fops; 1647 struct fileproc *fp = NULL; 1648 struct knote *kn = NULL; 1649 int error = 0; 1650 1651 if (kev->filter < 0) { 1652 if (kev->filter + EVFILT_SYSCOUNT < 0) 1653 return (EINVAL); 1654 fops = sysfilt_ops[~kev->filter]; /* to 0-base index */ 1655 } else { 1656 /* 1657 * XXX 1658 * filter attach routine is responsible for insuring that 1659 * the identifier can be attached to it. 1660 */ 1661 printf("unknown filter: %d\n", kev->filter); 1662 return (EINVAL); 1663 } 1664 1665restart: 1666 /* this iocount needs to be dropped if it is not registered */ 1667 proc_fdlock(p); 1668 if (fops->f_isfd && (error = fp_lookup(p, kev->ident, &fp, 1)) != 0) { 1669 proc_fdunlock(p); 1670 return (error); 1671 } 1672 1673 if (fops->f_isfd) { 1674 /* fd-based knotes are linked off the fd table */ 1675 if (kev->ident < (u_int)fdp->fd_knlistsize) { 1676 SLIST_FOREACH(kn, &fdp->fd_knlist[kev->ident], kn_link) 1677 if (kq == kn->kn_kq && 1678 kev->filter == kn->kn_filter) 1679 break; 1680 } 1681 } else { 1682 /* hash non-fd knotes here too */ 1683 if (fdp->fd_knhashmask != 0) { 1684 struct klist *list; 1685 1686 list = &fdp->fd_knhash[ 1687 KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; 1688 SLIST_FOREACH(kn, list, kn_link) 1689 if (kev->ident == kn->kn_id && 1690 kq == kn->kn_kq && 1691 kev->filter == kn->kn_filter) 1692 break; 1693 } 1694 } 1695 1696 /* 1697 * kn now contains the matching knote, or NULL if no match 1698 */ 1699 if (kn == NULL) { 1700 if ((kev->flags & (EV_ADD|EV_DELETE)) == EV_ADD) { 1701 kn = knote_alloc(); 1702 if (kn == NULL) { 1703 proc_fdunlock(p); 1704 error = ENOMEM; 1705 goto done; 1706 } 1707 kn->kn_fp = fp; 1708 kn->kn_kq = kq; 1709 kn->kn_tq = &kq->kq_head; 1710 kn->kn_fop = fops; 1711 kn->kn_sfflags = kev->fflags; 1712 kn->kn_sdata = kev->data; 1713 kev->fflags = 0; 1714 kev->data = 0; 1715 kn->kn_kevent = *kev; 1716 kn->kn_inuse = 1; /* for f_attach() */ 1717 kn->kn_status = KN_ATTACHING; 1718 1719 /* before anyone can find it */ 1720 if (kev->flags & EV_DISABLE) 1721 kn->kn_status |= KN_DISABLED; 1722 1723 error = knote_fdpattach(kn, fdp, p); 1724 proc_fdunlock(p); 1725 1726 if (error) { 1727 knote_free(kn); 1728 goto done; 1729 } 1730 1731 /* 1732 * apply reference count to knote structure, and 1733 * do not release it at the end of this routine. 1734 */ 1735 fp = NULL; 1736 1737 error = fops->f_attach(kn); 1738 1739 kqlock(kq); 1740 1741 if (error != 0) { 1742 /* 1743 * Failed to attach correctly, so drop. 1744 * All other possible users/droppers 1745 * have deferred to us. 1746 */ 1747 kn->kn_status |= KN_DROPPING; 1748 kqunlock(kq); 1749 knote_drop(kn, p); 1750 goto done; 1751 } else if (kn->kn_status & KN_DROPPING) { 1752 /* 1753 * Attach succeeded, but someone else 1754 * deferred their drop - now we have 1755 * to do it for them (after detaching). 1756 */ 1757 kqunlock(kq); 1758 kn->kn_fop->f_detach(kn); 1759 knote_drop(kn, p); 1760 goto done; 1761 } 1762 kn->kn_status &= ~KN_ATTACHING; 1763 kqunlock(kq); 1764 } else { 1765 proc_fdunlock(p); 1766 error = ENOENT; 1767 goto done; 1768 } 1769 } else { 1770 /* existing knote - get kqueue lock */ 1771 kqlock(kq); 1772 proc_fdunlock(p); 1773 1774 if (kev->flags & EV_DELETE) { 1775 knote_dequeue(kn); 1776 kn->kn_status |= KN_DISABLED; 1777 if (kqlock2knotedrop(kq, kn)) { 1778 kn->kn_fop->f_detach(kn); 1779 knote_drop(kn, p); 1780 } 1781 goto done; 1782 } 1783 1784 /* update status flags for existing knote */ 1785 if (kev->flags & EV_DISABLE) { 1786 knote_dequeue(kn); 1787 kn->kn_status |= KN_DISABLED; 1788 } else if (kev->flags & EV_ENABLE) { 1789 kn->kn_status &= ~KN_DISABLED; 1790 if (kn->kn_status & KN_ACTIVE) 1791 knote_enqueue(kn); 1792 } 1793 1794 /* 1795 * The user may change some filter values after the 1796 * initial EV_ADD, but doing so will not reset any 1797 * filter which have already been triggered. 1798 */ 1799 kn->kn_kevent.udata = kev->udata; 1800 if (fops->f_isfd || fops->f_touch == NULL) { 1801 kn->kn_sfflags = kev->fflags; 1802 kn->kn_sdata = kev->data; 1803 } 1804 1805 /* 1806 * If somebody is in the middle of dropping this 1807 * knote - go find/insert a new one. But we have 1808 * wait for this one to go away first. Attaches 1809 * running in parallel may also drop/modify the 1810 * knote. Wait for those to complete as well and 1811 * then start over if we encounter one. 1812 */ 1813 if (!kqlock2knoteusewait(kq, kn)) { 1814 /* kqueue, proc_fdlock both unlocked */ 1815 goto restart; 1816 } 1817 1818 /* 1819 * Call touch routine to notify filter of changes 1820 * in filter values. 1821 */ 1822 if (!fops->f_isfd && fops->f_touch != NULL) 1823 fops->f_touch(kn, kev, EVENT_REGISTER); 1824 } 1825 /* still have use ref on knote */ 1826 1827 /* 1828 * If the knote is not marked to always stay enqueued, 1829 * invoke the filter routine to see if it should be 1830 * enqueued now. 1831 */ 1832 if ((kn->kn_status & KN_STAYQUEUED) == 0 && kn->kn_fop->f_event(kn, 0)) { 1833 if (knoteuse2kqlock(kq, kn)) 1834 knote_activate(kn, 1); 1835 kqunlock(kq); 1836 } else { 1837 knote_put(kn); 1838 } 1839 1840done: 1841 if (fp != NULL) 1842 fp_drop(p, kev->ident, fp, 0); 1843 return (error); 1844} 1845 1846 1847/* 1848 * knote_process - process a triggered event 1849 * 1850 * Validate that it is really still a triggered event 1851 * by calling the filter routines (if necessary). Hold 1852 * a use reference on the knote to avoid it being detached. 1853 * If it is still considered triggered, invoke the callback 1854 * routine provided and move it to the provided inprocess 1855 * queue. 1856 * 1857 * caller holds a reference on the kqueue. 1858 * kqueue locked on entry and exit - but may be dropped 1859 */ 1860static int 1861knote_process(struct knote *kn, 1862 kevent_callback_t callback, 1863 void *data, 1864 struct kqtailq *inprocessp, 1865 struct proc *p) 1866{ 1867 struct kqueue *kq = kn->kn_kq; 1868 struct kevent64_s kev; 1869 int touch; 1870 int result; 1871 int error; 1872 1873 /* 1874 * Determine the kevent state we want to return. 1875 * 1876 * Some event states need to be revalidated before returning 1877 * them, others we take the snapshot at the time the event 1878 * was enqueued. 1879 * 1880 * Events with non-NULL f_touch operations must be touched. 1881 * Triggered events must fill in kev for the callback. 1882 * 1883 * Convert our lock to a use-count and call the event's 1884 * filter routine(s) to update. 1885 */ 1886 if ((kn->kn_status & KN_DISABLED) != 0) { 1887 result = 0; 1888 touch = 0; 1889 } else { 1890 int revalidate; 1891 1892 result = 1; 1893 revalidate = ((kn->kn_status & KN_STAYQUEUED) != 0 || 1894 (kn->kn_flags & EV_ONESHOT) == 0); 1895 touch = (!kn->kn_fop->f_isfd && kn->kn_fop->f_touch != NULL); 1896 1897 if (revalidate || touch) { 1898 if (revalidate) 1899 knote_deactivate(kn); 1900 1901 /* call the filter/touch routines with just a ref */ 1902 if (kqlock2knoteuse(kq, kn)) { 1903 /* if we have to revalidate, call the filter */ 1904 if (revalidate) { 1905 result = kn->kn_fop->f_event(kn, 0); 1906 } 1907 1908 /* 1909 * capture the kevent data - using touch if 1910 * specified 1911 */ 1912 if (result && touch) { 1913 kn->kn_fop->f_touch(kn, &kev, 1914 EVENT_PROCESS); 1915 } 1916 1917 /* 1918 * convert back to a kqlock - bail if the knote 1919 * went away 1920 */ 1921 if (!knoteuse2kqlock(kq, kn)) { 1922 return (EJUSTRETURN); 1923 } else if (result) { 1924 /* 1925 * if revalidated as alive, make sure 1926 * it's active 1927 */ 1928 if (!(kn->kn_status & KN_ACTIVE)) { 1929 knote_activate(kn, 0); 1930 } 1931 1932 /* 1933 * capture all events that occurred 1934 * during filter 1935 */ 1936 if (!touch) { 1937 kev = kn->kn_kevent; 1938 } 1939 1940 } else if ((kn->kn_status & KN_STAYQUEUED) == 0) { 1941 /* 1942 * was already dequeued, so just bail on 1943 * this one 1944 */ 1945 return (EJUSTRETURN); 1946 } 1947 } else { 1948 return (EJUSTRETURN); 1949 } 1950 } else { 1951 kev = kn->kn_kevent; 1952 } 1953 } 1954 1955 /* move knote onto inprocess queue */ 1956 assert(kn->kn_tq == &kq->kq_head); 1957 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 1958 kn->kn_tq = inprocessp; 1959 TAILQ_INSERT_TAIL(inprocessp, kn, kn_tqe); 1960 1961 /* 1962 * Determine how to dispatch the knote for future event handling. 1963 * not-fired: just return (do not callout). 1964 * One-shot: deactivate it. 1965 * Clear: deactivate and clear the state. 1966 * Dispatch: don't clear state, just deactivate it and mark it disabled. 1967 * All others: just leave where they are. 1968 */ 1969 1970 if (result == 0) { 1971 return (EJUSTRETURN); 1972 } else if ((kn->kn_flags & EV_ONESHOT) != 0) { 1973 knote_deactivate(kn); 1974 if (kqlock2knotedrop(kq, kn)) { 1975 kn->kn_fop->f_detach(kn); 1976 knote_drop(kn, p); 1977 } 1978 } else if ((kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) != 0) { 1979 if ((kn->kn_flags & EV_DISPATCH) != 0) { 1980 /* deactivate and disable all dispatch knotes */ 1981 knote_deactivate(kn); 1982 kn->kn_status |= KN_DISABLED; 1983 } else if (!touch || kn->kn_fflags == 0) { 1984 /* only deactivate if nothing since the touch */ 1985 knote_deactivate(kn); 1986 } 1987 if (!touch && (kn->kn_flags & EV_CLEAR) != 0) { 1988 /* manually clear non-touch knotes */ 1989 kn->kn_data = 0; 1990 kn->kn_fflags = 0; 1991 } 1992 kqunlock(kq); 1993 } else { 1994 /* 1995 * leave on inprocess queue. We'll 1996 * move all the remaining ones back 1997 * the kq queue and wakeup any 1998 * waiters when we are done. 1999 */ 2000 kqunlock(kq); 2001 } 2002 2003 /* callback to handle each event as we find it */ 2004 error = (callback)(kq, &kev, data); 2005 2006 kqlock(kq); 2007 return (error); 2008} 2009 2010/* 2011 * Return 0 to indicate that processing should proceed, 2012 * -1 if there is nothing to process. 2013 * 2014 * Called with kqueue locked and returns the same way, 2015 * but may drop lock temporarily. 2016 */ 2017static int 2018kqueue_begin_processing(struct kqueue *kq) 2019{ 2020 for (;;) { 2021 if (kq->kq_count == 0) { 2022 return (-1); 2023 } 2024 2025 /* if someone else is processing the queue, wait */ 2026 if (kq->kq_nprocess != 0) { 2027 wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, 2028 &kq->kq_nprocess, THREAD_UNINT, 0); 2029 kq->kq_state |= KQ_PROCWAIT; 2030 kqunlock(kq); 2031 thread_block(THREAD_CONTINUE_NULL); 2032 kqlock(kq); 2033 } else { 2034 kq->kq_nprocess = 1; 2035 return (0); 2036 } 2037 } 2038} 2039 2040/* 2041 * Called with kqueue lock held. 2042 */ 2043static void 2044kqueue_end_processing(struct kqueue *kq) 2045{ 2046 kq->kq_nprocess = 0; 2047 if (kq->kq_state & KQ_PROCWAIT) { 2048 kq->kq_state &= ~KQ_PROCWAIT; 2049 wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, 2050 &kq->kq_nprocess, THREAD_AWAKENED); 2051 } 2052} 2053 2054/* 2055 * kqueue_process - process the triggered events in a kqueue 2056 * 2057 * Walk the queued knotes and validate that they are 2058 * really still triggered events by calling the filter 2059 * routines (if necessary). Hold a use reference on 2060 * the knote to avoid it being detached. For each event 2061 * that is still considered triggered, invoke the 2062 * callback routine provided. 2063 * 2064 * caller holds a reference on the kqueue. 2065 * kqueue locked on entry and exit - but may be dropped 2066 * kqueue list locked (held for duration of call) 2067 */ 2068 2069static int 2070kqueue_process(struct kqueue *kq, 2071 kevent_callback_t callback, 2072 void *data, 2073 int *countp, 2074 struct proc *p) 2075{ 2076 struct kqtailq inprocess; 2077 struct knote *kn; 2078 int nevents; 2079 int error; 2080 2081 TAILQ_INIT(&inprocess); 2082 2083 if (kqueue_begin_processing(kq) == -1) { 2084 *countp = 0; 2085 /* Nothing to process */ 2086 return (0); 2087 } 2088 2089 /* 2090 * Clear any pre-posted status from previous runs, so we 2091 * only detect events that occur during this run. 2092 */ 2093 wait_queue_sub_clearrefs(kq->kq_wqs); 2094 2095 /* 2096 * loop through the enqueued knotes, processing each one and 2097 * revalidating those that need it. As they are processed, 2098 * they get moved to the inprocess queue (so the loop can end). 2099 */ 2100 error = 0; 2101 nevents = 0; 2102 2103 while (error == 0 && 2104 (kn = TAILQ_FIRST(&kq->kq_head)) != NULL) { 2105 error = knote_process(kn, callback, data, &inprocess, p); 2106 if (error == EJUSTRETURN) 2107 error = 0; 2108 else 2109 nevents++; 2110 } 2111 2112 /* 2113 * With the kqueue still locked, move any knotes 2114 * remaining on the inprocess queue back to the 2115 * kq's queue and wake up any waiters. 2116 */ 2117 while ((kn = TAILQ_FIRST(&inprocess)) != NULL) { 2118 assert(kn->kn_tq == &inprocess); 2119 TAILQ_REMOVE(&inprocess, kn, kn_tqe); 2120 kn->kn_tq = &kq->kq_head; 2121 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 2122 } 2123 2124 kqueue_end_processing(kq); 2125 2126 *countp = nevents; 2127 return (error); 2128} 2129 2130 2131static void 2132kqueue_scan_continue(void *data, wait_result_t wait_result) 2133{ 2134 thread_t self = current_thread(); 2135 uthread_t ut = (uthread_t)get_bsdthread_info(self); 2136 struct _kqueue_scan * cont_args = &ut->uu_kevent.ss_kqueue_scan; 2137 struct kqueue *kq = (struct kqueue *)data; 2138 int error; 2139 int count; 2140 2141 /* convert the (previous) wait_result to a proper error */ 2142 switch (wait_result) { 2143 case THREAD_AWAKENED: 2144 kqlock(kq); 2145 error = kqueue_process(kq, cont_args->call, cont_args, &count, 2146 current_proc()); 2147 if (error == 0 && count == 0) { 2148 wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, 2149 KQ_EVENT, THREAD_ABORTSAFE, cont_args->deadline); 2150 kq->kq_state |= KQ_SLEEP; 2151 kqunlock(kq); 2152 thread_block_parameter(kqueue_scan_continue, kq); 2153 /* NOTREACHED */ 2154 } 2155 kqunlock(kq); 2156 break; 2157 case THREAD_TIMED_OUT: 2158 error = EWOULDBLOCK; 2159 break; 2160 case THREAD_INTERRUPTED: 2161 error = EINTR; 2162 break; 2163 default: 2164 panic("%s: - invalid wait_result (%d)", __func__, 2165 wait_result); 2166 error = 0; 2167 } 2168 2169 /* call the continuation with the results */ 2170 assert(cont_args->cont != NULL); 2171 (cont_args->cont)(kq, cont_args->data, error); 2172} 2173 2174 2175/* 2176 * kqueue_scan - scan and wait for events in a kqueue 2177 * 2178 * Process the triggered events in a kqueue. 2179 * 2180 * If there are no events triggered arrange to 2181 * wait for them. If the caller provided a 2182 * continuation routine, then kevent_scan will 2183 * also. 2184 * 2185 * The callback routine must be valid. 2186 * The caller must hold a use-count reference on the kq. 2187 */ 2188 2189int 2190kqueue_scan(struct kqueue *kq, 2191 kevent_callback_t callback, 2192 kqueue_continue_t continuation, 2193 void *data, 2194 struct timeval *atvp, 2195 struct proc *p) 2196{ 2197 thread_continue_t cont = THREAD_CONTINUE_NULL; 2198 uint64_t deadline; 2199 int error; 2200 int first; 2201 2202 assert(callback != NULL); 2203 2204 first = 1; 2205 for (;;) { 2206 wait_result_t wait_result; 2207 int count; 2208 2209 /* 2210 * Make a pass through the kq to find events already 2211 * triggered. 2212 */ 2213 kqlock(kq); 2214 error = kqueue_process(kq, callback, data, &count, p); 2215 if (error || count) 2216 break; /* lock still held */ 2217 2218 /* looks like we have to consider blocking */ 2219 if (first) { 2220 first = 0; 2221 /* convert the timeout to a deadline once */ 2222 if (atvp->tv_sec || atvp->tv_usec) { 2223 uint64_t now; 2224 2225 clock_get_uptime(&now); 2226 nanoseconds_to_absolutetime((uint64_t)atvp->tv_sec * NSEC_PER_SEC + 2227 atvp->tv_usec * (long)NSEC_PER_USEC, 2228 &deadline); 2229 if (now >= deadline) { 2230 /* non-blocking call */ 2231 error = EWOULDBLOCK; 2232 break; /* lock still held */ 2233 } 2234 deadline -= now; 2235 clock_absolutetime_interval_to_deadline(deadline, &deadline); 2236 } else { 2237 deadline = 0; /* block forever */ 2238 } 2239 2240 if (continuation) { 2241 uthread_t ut = (uthread_t)get_bsdthread_info(current_thread()); 2242 struct _kqueue_scan *cont_args = &ut->uu_kevent.ss_kqueue_scan; 2243 2244 cont_args->call = callback; 2245 cont_args->cont = continuation; 2246 cont_args->deadline = deadline; 2247 cont_args->data = data; 2248 cont = kqueue_scan_continue; 2249 } 2250 } 2251 2252 /* go ahead and wait */ 2253 wait_queue_assert_wait_with_leeway((wait_queue_t)kq->kq_wqs, 2254 KQ_EVENT, THREAD_ABORTSAFE, TIMEOUT_URGENCY_USER_NORMAL, 2255 deadline, 0); 2256 kq->kq_state |= KQ_SLEEP; 2257 kqunlock(kq); 2258 wait_result = thread_block_parameter(cont, kq); 2259 /* NOTREACHED if (continuation != NULL) */ 2260 2261 switch (wait_result) { 2262 case THREAD_AWAKENED: 2263 continue; 2264 case THREAD_TIMED_OUT: 2265 return (EWOULDBLOCK); 2266 case THREAD_INTERRUPTED: 2267 return (EINTR); 2268 default: 2269 panic("%s: - bad wait_result (%d)", __func__, 2270 wait_result); 2271 error = 0; 2272 } 2273 } 2274 kqunlock(kq); 2275 return (error); 2276} 2277 2278 2279/* 2280 * XXX 2281 * This could be expanded to call kqueue_scan, if desired. 2282 */ 2283/*ARGSUSED*/ 2284static int 2285kqueue_read(__unused struct fileproc *fp, 2286 __unused struct uio *uio, 2287 __unused int flags, 2288 __unused vfs_context_t ctx) 2289{ 2290 return (ENXIO); 2291} 2292 2293/*ARGSUSED*/ 2294static int 2295kqueue_write(__unused struct fileproc *fp, 2296 __unused struct uio *uio, 2297 __unused int flags, 2298 __unused vfs_context_t ctx) 2299{ 2300 return (ENXIO); 2301} 2302 2303/*ARGSUSED*/ 2304static int 2305kqueue_ioctl(__unused struct fileproc *fp, 2306 __unused u_long com, 2307 __unused caddr_t data, 2308 __unused vfs_context_t ctx) 2309{ 2310 return (ENOTTY); 2311} 2312 2313/*ARGSUSED*/ 2314static int 2315kqueue_select(struct fileproc *fp, int which, void *wql, 2316 __unused vfs_context_t ctx) 2317{ 2318 struct kqueue *kq = (struct kqueue *)fp->f_data; 2319 struct knote *kn; 2320 struct kqtailq inprocessq; 2321 int retnum = 0; 2322 2323 if (which != FREAD) 2324 return (0); 2325 2326 TAILQ_INIT(&inprocessq); 2327 2328 kqlock(kq); 2329 /* 2330 * If this is the first pass, link the wait queue associated with the 2331 * the kqueue onto the wait queue set for the select(). Normally we 2332 * use selrecord() for this, but it uses the wait queue within the 2333 * selinfo structure and we need to use the main one for the kqueue to 2334 * catch events from KN_STAYQUEUED sources. So we do the linkage manually. 2335 * (The select() call will unlink them when it ends). 2336 */ 2337 if (wql != NULL) { 2338 thread_t cur_act = current_thread(); 2339 struct uthread * ut = get_bsdthread_info(cur_act); 2340 2341 kq->kq_state |= KQ_SEL; 2342 wait_queue_link_noalloc((wait_queue_t)kq->kq_wqs, ut->uu_wqset, 2343 (wait_queue_link_t)wql); 2344 } 2345 2346 if (kqueue_begin_processing(kq) == -1) { 2347 kqunlock(kq); 2348 return (0); 2349 } 2350 2351 if (kq->kq_count != 0) { 2352 /* 2353 * there is something queued - but it might be a 2354 * KN_STAYQUEUED knote, which may or may not have 2355 * any events pending. So, we have to walk the 2356 * list of knotes to see, and peek at the stay- 2357 * queued ones to be really sure. 2358 */ 2359 while ((kn = (struct knote *)TAILQ_FIRST(&kq->kq_head)) != NULL) { 2360 if ((kn->kn_status & KN_STAYQUEUED) == 0) { 2361 retnum = 1; 2362 goto out; 2363 } 2364 2365 TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); 2366 TAILQ_INSERT_TAIL(&inprocessq, kn, kn_tqe); 2367 2368 if (kqlock2knoteuse(kq, kn)) { 2369 unsigned peek; 2370 2371 peek = kn->kn_fop->f_peek(kn); 2372 if (knoteuse2kqlock(kq, kn)) { 2373 if (peek > 0) { 2374 retnum = 1; 2375 goto out; 2376 } 2377 } else { 2378 retnum = 0; 2379 } 2380 } 2381 } 2382 } 2383 2384out: 2385 /* Return knotes to active queue */ 2386 while ((kn = TAILQ_FIRST(&inprocessq)) != NULL) { 2387 TAILQ_REMOVE(&inprocessq, kn, kn_tqe); 2388 kn->kn_tq = &kq->kq_head; 2389 TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); 2390 } 2391 2392 kqueue_end_processing(kq); 2393 kqunlock(kq); 2394 return (retnum); 2395} 2396 2397/* 2398 * kqueue_close - 2399 */ 2400/*ARGSUSED*/ 2401static int 2402kqueue_close(struct fileglob *fg, __unused vfs_context_t ctx) 2403{ 2404 struct kqueue *kq = (struct kqueue *)fg->fg_data; 2405 2406 kqueue_dealloc(kq); 2407 fg->fg_data = NULL; 2408 return (0); 2409} 2410 2411/*ARGSUSED*/ 2412/* 2413 * The callers has taken a use-count reference on this kqueue and will donate it 2414 * to the kqueue we are being added to. This keeps the kqueue from closing until 2415 * that relationship is torn down. 2416 */ 2417static int 2418kqueue_kqfilter(__unused struct fileproc *fp, struct knote *kn, __unused vfs_context_t ctx) 2419{ 2420 struct kqueue *kq = (struct kqueue *)kn->kn_fp->f_data; 2421 struct kqueue *parentkq = kn->kn_kq; 2422 2423 if (parentkq == kq || 2424 kn->kn_filter != EVFILT_READ) 2425 return (1); 2426 2427 /* 2428 * We have to avoid creating a cycle when nesting kqueues 2429 * inside another. Rather than trying to walk the whole 2430 * potential DAG of nested kqueues, we just use a simple 2431 * ceiling protocol. When a kqueue is inserted into another, 2432 * we check that the (future) parent is not already nested 2433 * into another kqueue at a lower level than the potenial 2434 * child (because it could indicate a cycle). If that test 2435 * passes, we just mark the nesting levels accordingly. 2436 */ 2437 2438 kqlock(parentkq); 2439 if (parentkq->kq_level > 0 && 2440 parentkq->kq_level < kq->kq_level) 2441 { 2442 kqunlock(parentkq); 2443 return (1); 2444 } else { 2445 /* set parent level appropriately */ 2446 if (parentkq->kq_level == 0) 2447 parentkq->kq_level = 2; 2448 if (parentkq->kq_level < kq->kq_level + 1) 2449 parentkq->kq_level = kq->kq_level + 1; 2450 kqunlock(parentkq); 2451 2452 kn->kn_fop = &kqread_filtops; 2453 kqlock(kq); 2454 KNOTE_ATTACH(&kq->kq_sel.si_note, kn); 2455 /* indicate nesting in child, if needed */ 2456 if (kq->kq_level == 0) 2457 kq->kq_level = 1; 2458 kqunlock(kq); 2459 return (0); 2460 } 2461} 2462 2463/* 2464 * kqueue_drain - called when kq is closed 2465 */ 2466/*ARGSUSED*/ 2467static int 2468kqueue_drain(struct fileproc *fp, __unused vfs_context_t ctx) 2469{ 2470 struct kqueue *kq = (struct kqueue *)fp->f_fglob->fg_data; 2471 kqlock(kq); 2472 kqueue_wakeup(kq, 1); 2473 kqunlock(kq); 2474 return (0); 2475} 2476 2477/*ARGSUSED*/ 2478int 2479kqueue_stat(struct kqueue *kq, void *ub, int isstat64, proc_t p) 2480{ 2481 kqlock(kq); 2482 if (isstat64 != 0) { 2483 struct stat64 *sb64 = (struct stat64 *)ub; 2484 2485 bzero((void *)sb64, sizeof(*sb64)); 2486 sb64->st_size = kq->kq_count; 2487 if (kq->kq_state & KQ_KEV64) 2488 sb64->st_blksize = sizeof(struct kevent64_s); 2489 else 2490 sb64->st_blksize = IS_64BIT_PROCESS(p) ? sizeof(struct user64_kevent) : sizeof(struct user32_kevent); 2491 sb64->st_mode = S_IFIFO; 2492 } else { 2493 struct stat *sb = (struct stat *)ub; 2494 2495 bzero((void *)sb, sizeof(*sb)); 2496 sb->st_size = kq->kq_count; 2497 if (kq->kq_state & KQ_KEV64) 2498 sb->st_blksize = sizeof(struct kevent64_s); 2499 else 2500 sb->st_blksize = IS_64BIT_PROCESS(p) ? sizeof(struct user64_kevent) : sizeof(struct user32_kevent); 2501 sb->st_mode = S_IFIFO; 2502 } 2503 kqunlock(kq); 2504 return (0); 2505} 2506 2507/* 2508 * Called with the kqueue locked 2509 */ 2510static void 2511kqueue_wakeup(struct kqueue *kq, int closed) 2512{ 2513 if ((kq->kq_state & (KQ_SLEEP | KQ_SEL)) != 0 || kq->kq_nprocess > 0) { 2514 kq->kq_state &= ~(KQ_SLEEP | KQ_SEL); 2515 wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, KQ_EVENT, 2516 (closed) ? THREAD_INTERRUPTED : THREAD_AWAKENED); 2517 } 2518} 2519 2520void 2521klist_init(struct klist *list) 2522{ 2523 SLIST_INIT(list); 2524} 2525 2526 2527/* 2528 * Query/Post each knote in the object's list 2529 * 2530 * The object lock protects the list. It is assumed 2531 * that the filter/event routine for the object can 2532 * determine that the object is already locked (via 2533 * the hint) and not deadlock itself. 2534 * 2535 * The object lock should also hold off pending 2536 * detach/drop operations. But we'll prevent it here 2537 * too - just in case. 2538 */ 2539void 2540knote(struct klist *list, long hint) 2541{ 2542 struct knote *kn; 2543 2544 SLIST_FOREACH(kn, list, kn_selnext) { 2545 struct kqueue *kq = kn->kn_kq; 2546 2547 kqlock(kq); 2548 if (kqlock2knoteuse(kq, kn)) { 2549 int result; 2550 2551 /* call the event with only a use count */ 2552 result = kn->kn_fop->f_event(kn, hint); 2553 2554 /* if its not going away and triggered */ 2555 if (knoteuse2kqlock(kq, kn) && result) 2556 knote_activate(kn, 1); 2557 /* lock held again */ 2558 } 2559 kqunlock(kq); 2560 } 2561} 2562 2563/* 2564 * attach a knote to the specified list. Return true if this is the first entry. 2565 * The list is protected by whatever lock the object it is associated with uses. 2566 */ 2567int 2568knote_attach(struct klist *list, struct knote *kn) 2569{ 2570 int ret = SLIST_EMPTY(list); 2571 SLIST_INSERT_HEAD(list, kn, kn_selnext); 2572 return (ret); 2573} 2574 2575/* 2576 * detach a knote from the specified list. Return true if that was the last entry. 2577 * The list is protected by whatever lock the object it is associated with uses. 2578 */ 2579int 2580knote_detach(struct klist *list, struct knote *kn) 2581{ 2582 SLIST_REMOVE(list, kn, knote, kn_selnext); 2583 return (SLIST_EMPTY(list)); 2584} 2585 2586/* 2587 * For a given knote, link a provided wait queue directly with the kqueue. 2588 * Wakeups will happen via recursive wait queue support. But nothing will move 2589 * the knote to the active list at wakeup (nothing calls knote()). Instead, 2590 * we permanently enqueue them here. 2591 * 2592 * kqueue and knote references are held by caller. 2593 * 2594 * caller provides the wait queue link structure. 2595 */ 2596int 2597knote_link_wait_queue(struct knote *kn, struct wait_queue *wq, wait_queue_link_t wql) 2598{ 2599 struct kqueue *kq = kn->kn_kq; 2600 kern_return_t kr; 2601 2602 kr = wait_queue_link_noalloc(wq, kq->kq_wqs, wql); 2603 if (kr == KERN_SUCCESS) { 2604 knote_markstayqueued(kn); 2605 return (0); 2606 } else { 2607 return (EINVAL); 2608 } 2609} 2610 2611/* 2612 * Unlink the provided wait queue from the kqueue associated with a knote. 2613 * Also remove it from the magic list of directly attached knotes. 2614 * 2615 * Note that the unlink may have already happened from the other side, so 2616 * ignore any failures to unlink and just remove it from the kqueue list. 2617 * 2618 * On success, caller is responsible for the link structure 2619 */ 2620int 2621knote_unlink_wait_queue(struct knote *kn, struct wait_queue *wq, wait_queue_link_t *wqlp) 2622{ 2623 struct kqueue *kq = kn->kn_kq; 2624 kern_return_t kr; 2625 2626 kr = wait_queue_unlink_nofree(wq, kq->kq_wqs, wqlp); 2627 kqlock(kq); 2628 kn->kn_status &= ~KN_STAYQUEUED; 2629 knote_dequeue(kn); 2630 kqunlock(kq); 2631 return ((kr != KERN_SUCCESS) ? EINVAL : 0); 2632} 2633 2634/* 2635 * remove all knotes referencing a specified fd 2636 * 2637 * Essentially an inlined knote_remove & knote_drop 2638 * when we know for sure that the thing is a file 2639 * 2640 * Entered with the proc_fd lock already held. 2641 * It returns the same way, but may drop it temporarily. 2642 */ 2643void 2644knote_fdclose(struct proc *p, int fd) 2645{ 2646 struct filedesc *fdp = p->p_fd; 2647 struct klist *list; 2648 struct knote *kn; 2649 2650 list = &fdp->fd_knlist[fd]; 2651 while ((kn = SLIST_FIRST(list)) != NULL) { 2652 struct kqueue *kq = kn->kn_kq; 2653 2654 if (kq->kq_p != p) 2655 panic("%s: proc mismatch (kq->kq_p=%p != p=%p)", 2656 __func__, kq->kq_p, p); 2657 2658 kqlock(kq); 2659 proc_fdunlock(p); 2660 2661 /* 2662 * Convert the lock to a drop ref. 2663 * If we get it, go ahead and drop it. 2664 * Otherwise, we waited for it to 2665 * be dropped by the other guy, so 2666 * it is safe to move on in the list. 2667 */ 2668 if (kqlock2knotedrop(kq, kn)) { 2669 kn->kn_fop->f_detach(kn); 2670 knote_drop(kn, p); 2671 } 2672 2673 proc_fdlock(p); 2674 2675 /* the fd tables may have changed - start over */ 2676 list = &fdp->fd_knlist[fd]; 2677 } 2678} 2679 2680/* proc_fdlock held on entry (and exit) */ 2681static int 2682knote_fdpattach(struct knote *kn, struct filedesc *fdp, struct proc *p) 2683{ 2684 struct klist *list = NULL; 2685 2686 if (! kn->kn_fop->f_isfd) { 2687 if (fdp->fd_knhashmask == 0) 2688 fdp->fd_knhash = hashinit(CONFIG_KN_HASHSIZE, M_KQUEUE, 2689 &fdp->fd_knhashmask); 2690 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 2691 } else { 2692 if ((u_int)fdp->fd_knlistsize <= kn->kn_id) { 2693 u_int size = 0; 2694 2695 if (kn->kn_id >= (uint64_t)p->p_rlimit[RLIMIT_NOFILE].rlim_cur 2696 || kn->kn_id >= (uint64_t)maxfiles) 2697 return (EINVAL); 2698 2699 /* have to grow the fd_knlist */ 2700 size = fdp->fd_knlistsize; 2701 while (size <= kn->kn_id) 2702 size += KQEXTENT; 2703 2704 if (size >= (UINT_MAX/sizeof(struct klist *))) 2705 return (EINVAL); 2706 2707 MALLOC(list, struct klist *, 2708 size * sizeof(struct klist *), M_KQUEUE, M_WAITOK); 2709 if (list == NULL) 2710 return (ENOMEM); 2711 2712 bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list, 2713 fdp->fd_knlistsize * sizeof(struct klist *)); 2714 bzero((caddr_t)list + 2715 fdp->fd_knlistsize * sizeof(struct klist *), 2716 (size - fdp->fd_knlistsize) * sizeof(struct klist *)); 2717 FREE(fdp->fd_knlist, M_KQUEUE); 2718 fdp->fd_knlist = list; 2719 fdp->fd_knlistsize = size; 2720 } 2721 list = &fdp->fd_knlist[kn->kn_id]; 2722 } 2723 SLIST_INSERT_HEAD(list, kn, kn_link); 2724 return (0); 2725} 2726 2727 2728 2729/* 2730 * should be called at spl == 0, since we don't want to hold spl 2731 * while calling fdrop and free. 2732 */ 2733static void 2734knote_drop(struct knote *kn, __unused struct proc *ctxp) 2735{ 2736 struct kqueue *kq = kn->kn_kq; 2737 struct proc *p = kq->kq_p; 2738 struct filedesc *fdp = p->p_fd; 2739 struct klist *list; 2740 int needswakeup; 2741 2742 proc_fdlock(p); 2743 if (kn->kn_fop->f_isfd) 2744 list = &fdp->fd_knlist[kn->kn_id]; 2745 else 2746 list = &fdp->fd_knhash[KN_HASH(kn->kn_id, fdp->fd_knhashmask)]; 2747 2748 SLIST_REMOVE(list, kn, knote, kn_link); 2749 kqlock(kq); 2750 knote_dequeue(kn); 2751 needswakeup = (kn->kn_status & KN_USEWAIT); 2752 kqunlock(kq); 2753 proc_fdunlock(p); 2754 2755 if (needswakeup) 2756 wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, &kn->kn_status, 2757 THREAD_AWAKENED); 2758 2759 if (kn->kn_fop->f_isfd) 2760 fp_drop(p, kn->kn_id, kn->kn_fp, 0); 2761 2762 knote_free(kn); 2763} 2764 2765/* called with kqueue lock held */ 2766static void 2767knote_activate(struct knote *kn, int propagate) 2768{ 2769 struct kqueue *kq = kn->kn_kq; 2770 2771 kn->kn_status |= KN_ACTIVE; 2772 knote_enqueue(kn); 2773 kqueue_wakeup(kq, 0); 2774 2775 /* this is a real event: wake up the parent kq, too */ 2776 if (propagate) 2777 KNOTE(&kq->kq_sel.si_note, 0); 2778} 2779 2780/* called with kqueue lock held */ 2781static void 2782knote_deactivate(struct knote *kn) 2783{ 2784 kn->kn_status &= ~KN_ACTIVE; 2785 knote_dequeue(kn); 2786} 2787 2788/* called with kqueue lock held */ 2789static void 2790knote_enqueue(struct knote *kn) 2791{ 2792 if ((kn->kn_status & (KN_QUEUED | KN_STAYQUEUED)) == KN_STAYQUEUED || 2793 (kn->kn_status & (KN_QUEUED | KN_STAYQUEUED | KN_DISABLED)) == 0) { 2794 struct kqtailq *tq = kn->kn_tq; 2795 struct kqueue *kq = kn->kn_kq; 2796 2797 TAILQ_INSERT_TAIL(tq, kn, kn_tqe); 2798 kn->kn_status |= KN_QUEUED; 2799 kq->kq_count++; 2800 } 2801} 2802 2803/* called with kqueue lock held */ 2804static void 2805knote_dequeue(struct knote *kn) 2806{ 2807 struct kqueue *kq = kn->kn_kq; 2808 2809 if ((kn->kn_status & (KN_QUEUED | KN_STAYQUEUED)) == KN_QUEUED) { 2810 struct kqtailq *tq = kn->kn_tq; 2811 2812 TAILQ_REMOVE(tq, kn, kn_tqe); 2813 kn->kn_tq = &kq->kq_head; 2814 kn->kn_status &= ~KN_QUEUED; 2815 kq->kq_count--; 2816 } 2817} 2818 2819void 2820knote_init(void) 2821{ 2822 knote_zone = zinit(sizeof(struct knote), 8192*sizeof(struct knote), 2823 8192, "knote zone"); 2824 2825 /* allocate kq lock group attribute and group */ 2826 kq_lck_grp_attr = lck_grp_attr_alloc_init(); 2827 2828 kq_lck_grp = lck_grp_alloc_init("kqueue", kq_lck_grp_attr); 2829 2830 /* Allocate kq lock attribute */ 2831 kq_lck_attr = lck_attr_alloc_init(); 2832 2833 /* Initialize the timer filter lock */ 2834 lck_mtx_init(&_filt_timerlock, kq_lck_grp, kq_lck_attr); 2835 2836#if VM_PRESSURE_EVENTS 2837 /* Initialize the vm pressure list lock */ 2838 vm_pressure_init(kq_lck_grp, kq_lck_attr); 2839#endif 2840 2841#if CONFIG_MEMORYSTATUS 2842 /* Initialize the memorystatus list lock */ 2843 memorystatus_kevent_init(kq_lck_grp, kq_lck_attr); 2844#endif 2845} 2846SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL) 2847 2848static struct knote * 2849knote_alloc(void) 2850{ 2851 return ((struct knote *)zalloc(knote_zone)); 2852} 2853 2854static void 2855knote_free(struct knote *kn) 2856{ 2857 zfree(knote_zone, kn); 2858} 2859 2860#if SOCKETS 2861#include <sys/param.h> 2862#include <sys/socket.h> 2863#include <sys/protosw.h> 2864#include <sys/domain.h> 2865#include <sys/mbuf.h> 2866#include <sys/kern_event.h> 2867#include <sys/malloc.h> 2868#include <sys/sys_domain.h> 2869#include <sys/syslog.h> 2870 2871#ifndef ROUNDUP64 2872#define ROUNDUP64(x) P2ROUNDUP((x), sizeof (u_int64_t)) 2873#endif 2874 2875#ifndef ADVANCE64 2876#define ADVANCE64(p, n) (void*)((char *)(p) + ROUNDUP64(n)) 2877#endif 2878 2879static lck_grp_attr_t *kev_lck_grp_attr; 2880static lck_attr_t *kev_lck_attr; 2881static lck_grp_t *kev_lck_grp; 2882static decl_lck_rw_data(,kev_lck_data); 2883static lck_rw_t *kev_rwlock = &kev_lck_data; 2884 2885static int kev_attach(struct socket *so, int proto, struct proc *p); 2886static int kev_detach(struct socket *so); 2887static int kev_control(struct socket *so, u_long cmd, caddr_t data, 2888 struct ifnet *ifp, struct proc *p); 2889static lck_mtx_t * event_getlock(struct socket *, int); 2890static int event_lock(struct socket *, int, void *); 2891static int event_unlock(struct socket *, int, void *); 2892 2893static int event_sofreelastref(struct socket *); 2894static void kev_delete(struct kern_event_pcb *); 2895 2896static struct pr_usrreqs event_usrreqs = { 2897 .pru_attach = kev_attach, 2898 .pru_control = kev_control, 2899 .pru_detach = kev_detach, 2900 .pru_soreceive = soreceive, 2901}; 2902 2903static struct protosw eventsw[] = { 2904{ 2905 .pr_type = SOCK_RAW, 2906 .pr_protocol = SYSPROTO_EVENT, 2907 .pr_flags = PR_ATOMIC, 2908 .pr_usrreqs = &event_usrreqs, 2909 .pr_lock = event_lock, 2910 .pr_unlock = event_unlock, 2911 .pr_getlock = event_getlock, 2912} 2913}; 2914 2915__private_extern__ int kevt_getstat SYSCTL_HANDLER_ARGS; 2916__private_extern__ int kevt_pcblist SYSCTL_HANDLER_ARGS; 2917 2918SYSCTL_NODE(_net_systm, OID_AUTO, kevt, 2919 CTLFLAG_RW|CTLFLAG_LOCKED, 0, "Kernel event family"); 2920 2921struct kevtstat kevtstat; 2922SYSCTL_PROC(_net_systm_kevt, OID_AUTO, stats, 2923 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, 2924 kevt_getstat, "S,kevtstat", ""); 2925 2926SYSCTL_PROC(_net_systm_kevt, OID_AUTO, pcblist, 2927 CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, 2928 kevt_pcblist, "S,xkevtpcb", ""); 2929 2930static lck_mtx_t * 2931event_getlock(struct socket *so, int locktype) 2932{ 2933#pragma unused(locktype) 2934 struct kern_event_pcb *ev_pcb = (struct kern_event_pcb *)so->so_pcb; 2935 2936 if (so->so_pcb != NULL) { 2937 if (so->so_usecount < 0) 2938 panic("%s: so=%p usecount=%d lrh= %s\n", __func__, 2939 so, so->so_usecount, solockhistory_nr(so)); 2940 /* NOTREACHED */ 2941 } else { 2942 panic("%s: so=%p NULL NO so_pcb %s\n", __func__, 2943 so, solockhistory_nr(so)); 2944 /* NOTREACHED */ 2945 } 2946 return (&ev_pcb->evp_mtx); 2947} 2948 2949static int 2950event_lock(struct socket *so, int refcount, void *lr) 2951{ 2952 void *lr_saved; 2953 2954 if (lr == NULL) 2955 lr_saved = __builtin_return_address(0); 2956 else 2957 lr_saved = lr; 2958 2959 if (so->so_pcb != NULL) { 2960 lck_mtx_lock(&((struct kern_event_pcb *)so->so_pcb)->evp_mtx); 2961 } else { 2962 panic("%s: so=%p NO PCB! lr=%p lrh= %s\n", __func__, 2963 so, lr_saved, solockhistory_nr(so)); 2964 /* NOTREACHED */ 2965 } 2966 2967 if (so->so_usecount < 0) { 2968 panic("%s: so=%p so_pcb=%p lr=%p ref=%d lrh= %s\n", __func__, 2969 so, so->so_pcb, lr_saved, so->so_usecount, 2970 solockhistory_nr(so)); 2971 /* NOTREACHED */ 2972 } 2973 2974 if (refcount) 2975 so->so_usecount++; 2976 2977 so->lock_lr[so->next_lock_lr] = lr_saved; 2978 so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX; 2979 return (0); 2980} 2981 2982static int 2983event_unlock(struct socket *so, int refcount, void *lr) 2984{ 2985 void *lr_saved; 2986 lck_mtx_t *mutex_held; 2987 2988 if (lr == NULL) 2989 lr_saved = __builtin_return_address(0); 2990 else 2991 lr_saved = lr; 2992 2993 if (refcount) 2994 so->so_usecount--; 2995 2996 if (so->so_usecount < 0) { 2997 panic("%s: so=%p usecount=%d lrh= %s\n", __func__, 2998 so, so->so_usecount, solockhistory_nr(so)); 2999 /* NOTREACHED */ 3000 } 3001 if (so->so_pcb == NULL) { 3002 panic("%s: so=%p NO PCB usecount=%d lr=%p lrh= %s\n", __func__, 3003 so, so->so_usecount, (void *)lr_saved, 3004 solockhistory_nr(so)); 3005 /* NOTREACHED */ 3006 } 3007 mutex_held = (&((struct kern_event_pcb *)so->so_pcb)->evp_mtx); 3008 3009 lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); 3010 so->unlock_lr[so->next_unlock_lr] = lr_saved; 3011 so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX; 3012 3013 if (so->so_usecount == 0) { 3014 VERIFY(so->so_flags & SOF_PCBCLEARING); 3015 event_sofreelastref(so); 3016 } else { 3017 lck_mtx_unlock(mutex_held); 3018 } 3019 3020 return (0); 3021} 3022 3023static int 3024event_sofreelastref(struct socket *so) 3025{ 3026 struct kern_event_pcb *ev_pcb = (struct kern_event_pcb *)so->so_pcb; 3027 3028 lck_mtx_assert(&(ev_pcb->evp_mtx), LCK_MTX_ASSERT_OWNED); 3029 3030 so->so_pcb = NULL; 3031 3032 /* 3033 * Disable upcall in the event another thread is in kev_post_msg() 3034 * appending record to the receive socket buffer, since sbwakeup() 3035 * may release the socket lock otherwise. 3036 */ 3037 so->so_rcv.sb_flags &= ~SB_UPCALL; 3038 so->so_snd.sb_flags &= ~SB_UPCALL; 3039 so->so_event = sonullevent; 3040 lck_mtx_unlock(&(ev_pcb->evp_mtx)); 3041 3042 lck_mtx_assert(&(ev_pcb->evp_mtx), LCK_MTX_ASSERT_NOTOWNED); 3043 lck_rw_lock_exclusive(kev_rwlock); 3044 LIST_REMOVE(ev_pcb, evp_link); 3045 kevtstat.kes_pcbcount--; 3046 kevtstat.kes_gencnt++; 3047 lck_rw_done(kev_rwlock); 3048 kev_delete(ev_pcb); 3049 3050 sofreelastref(so, 1); 3051 return (0); 3052} 3053 3054static int event_proto_count = (sizeof (eventsw) / sizeof (struct protosw)); 3055 3056static 3057struct kern_event_head kern_event_head; 3058 3059static u_int32_t static_event_id = 0; 3060 3061#define EVPCB_ZONE_MAX 65536 3062#define EVPCB_ZONE_NAME "kerneventpcb" 3063static struct zone *ev_pcb_zone; 3064 3065/* 3066 * Install the protosw's for the NKE manager. Invoked at extension load time 3067 */ 3068void 3069kern_event_init(struct domain *dp) 3070{ 3071 struct protosw *pr; 3072 int i; 3073 3074 VERIFY(!(dp->dom_flags & DOM_INITIALIZED)); 3075 VERIFY(dp == systemdomain); 3076 3077 kev_lck_grp_attr = lck_grp_attr_alloc_init(); 3078 if (kev_lck_grp_attr == NULL) { 3079 panic("%s: lck_grp_attr_alloc_init failed\n", __func__); 3080 /* NOTREACHED */ 3081 } 3082 3083 kev_lck_grp = lck_grp_alloc_init("Kernel Event Protocol", 3084 kev_lck_grp_attr); 3085 if (kev_lck_grp == NULL) { 3086 panic("%s: lck_grp_alloc_init failed\n", __func__); 3087 /* NOTREACHED */ 3088 } 3089 3090 kev_lck_attr = lck_attr_alloc_init(); 3091 if (kev_lck_attr == NULL) { 3092 panic("%s: lck_attr_alloc_init failed\n", __func__); 3093 /* NOTREACHED */ 3094 } 3095 3096 lck_rw_init(kev_rwlock, kev_lck_grp, kev_lck_attr); 3097 if (kev_rwlock == NULL) { 3098 panic("%s: lck_mtx_alloc_init failed\n", __func__); 3099 /* NOTREACHED */ 3100 } 3101 3102 for (i = 0, pr = &eventsw[0]; i < event_proto_count; i++, pr++) 3103 net_add_proto(pr, dp, 1); 3104 3105 ev_pcb_zone = zinit(sizeof(struct kern_event_pcb), 3106 EVPCB_ZONE_MAX * sizeof(struct kern_event_pcb), 0, EVPCB_ZONE_NAME); 3107 if (ev_pcb_zone == NULL) { 3108 panic("%s: failed allocating ev_pcb_zone", __func__); 3109 /* NOTREACHED */ 3110 } 3111 zone_change(ev_pcb_zone, Z_EXPAND, TRUE); 3112 zone_change(ev_pcb_zone, Z_CALLERACCT, TRUE); 3113} 3114 3115static int 3116kev_attach(struct socket *so, __unused int proto, __unused struct proc *p) 3117{ 3118 int error = 0; 3119 struct kern_event_pcb *ev_pcb; 3120 3121 error = soreserve(so, KEV_SNDSPACE, KEV_RECVSPACE); 3122 if (error != 0) 3123 return (error); 3124 3125 if ((ev_pcb = (struct kern_event_pcb *)zalloc(ev_pcb_zone)) == NULL) { 3126 return (ENOBUFS); 3127 } 3128 bzero(ev_pcb, sizeof(struct kern_event_pcb)); 3129 lck_mtx_init(&ev_pcb->evp_mtx, kev_lck_grp, kev_lck_attr); 3130 3131 ev_pcb->evp_socket = so; 3132 ev_pcb->evp_vendor_code_filter = 0xffffffff; 3133 3134 so->so_pcb = (caddr_t) ev_pcb; 3135 lck_rw_lock_exclusive(kev_rwlock); 3136 LIST_INSERT_HEAD(&kern_event_head, ev_pcb, evp_link); 3137 kevtstat.kes_pcbcount++; 3138 kevtstat.kes_gencnt++; 3139 lck_rw_done(kev_rwlock); 3140 3141 return (error); 3142} 3143 3144static void 3145kev_delete(struct kern_event_pcb *ev_pcb) 3146{ 3147 VERIFY(ev_pcb != NULL); 3148 lck_mtx_destroy(&ev_pcb->evp_mtx, kev_lck_grp); 3149 zfree(ev_pcb_zone, ev_pcb); 3150} 3151 3152static int 3153kev_detach(struct socket *so) 3154{ 3155 struct kern_event_pcb *ev_pcb = (struct kern_event_pcb *) so->so_pcb; 3156 3157 if (ev_pcb != NULL) { 3158 soisdisconnected(so); 3159 so->so_flags |= SOF_PCBCLEARING; 3160 } 3161 3162 return (0); 3163} 3164 3165/* 3166 * For now, kev_vendor_code and mbuf_tags use the same 3167 * mechanism. 3168 */ 3169errno_t kev_vendor_code_find( 3170 const char *string, 3171 u_int32_t *out_vendor_code) 3172{ 3173 if (strlen(string) >= KEV_VENDOR_CODE_MAX_STR_LEN) { 3174 return (EINVAL); 3175 } 3176 return (net_str_id_find_internal(string, out_vendor_code, 3177 NSI_VENDOR_CODE, 1)); 3178} 3179 3180errno_t 3181kev_msg_post(struct kev_msg *event_msg) 3182{ 3183 mbuf_tag_id_t min_vendor, max_vendor; 3184 3185 net_str_id_first_last(&min_vendor, &max_vendor, NSI_VENDOR_CODE); 3186 3187 if (event_msg == NULL) 3188 return (EINVAL); 3189 3190 /* 3191 * Limit third parties to posting events for registered vendor codes 3192 * only 3193 */ 3194 if (event_msg->vendor_code < min_vendor || 3195 event_msg->vendor_code > max_vendor) { 3196 OSIncrementAtomic64((SInt64 *)&kevtstat.kes_badvendor); 3197 return (EINVAL); 3198 } 3199 return (kev_post_msg(event_msg)); 3200} 3201 3202int 3203kev_post_msg(struct kev_msg *event_msg) 3204{ 3205 struct mbuf *m, *m2; 3206 struct kern_event_pcb *ev_pcb; 3207 struct kern_event_msg *ev; 3208 char *tmp; 3209 u_int32_t total_size; 3210 int i; 3211 3212 /* Verify the message is small enough to fit in one mbuf w/o cluster */ 3213 total_size = KEV_MSG_HEADER_SIZE; 3214 3215 for (i = 0; i < 5; i++) { 3216 if (event_msg->dv[i].data_length == 0) 3217 break; 3218 total_size += event_msg->dv[i].data_length; 3219 } 3220 3221 if (total_size > MLEN) { 3222 OSIncrementAtomic64((SInt64 *)&kevtstat.kes_toobig); 3223 return (EMSGSIZE); 3224 } 3225 3226 m = m_get(M_DONTWAIT, MT_DATA); 3227 if (m == 0) { 3228 OSIncrementAtomic64((SInt64 *)&kevtstat.kes_nomem); 3229 return (ENOMEM); 3230 } 3231 ev = mtod(m, struct kern_event_msg *); 3232 total_size = KEV_MSG_HEADER_SIZE; 3233 3234 tmp = (char *) &ev->event_data[0]; 3235 for (i = 0; i < 5; i++) { 3236 if (event_msg->dv[i].data_length == 0) 3237 break; 3238 3239 total_size += event_msg->dv[i].data_length; 3240 bcopy(event_msg->dv[i].data_ptr, tmp, 3241 event_msg->dv[i].data_length); 3242 tmp += event_msg->dv[i].data_length; 3243 } 3244 3245 ev->id = ++static_event_id; 3246 ev->total_size = total_size; 3247 ev->vendor_code = event_msg->vendor_code; 3248 ev->kev_class = event_msg->kev_class; 3249 ev->kev_subclass = event_msg->kev_subclass; 3250 ev->event_code = event_msg->event_code; 3251 3252 m->m_len = total_size; 3253 lck_rw_lock_shared(kev_rwlock); 3254 for (ev_pcb = LIST_FIRST(&kern_event_head); 3255 ev_pcb; 3256 ev_pcb = LIST_NEXT(ev_pcb, evp_link)) { 3257 lck_mtx_lock(&ev_pcb->evp_mtx); 3258 if (ev_pcb->evp_socket->so_pcb == NULL) { 3259 lck_mtx_unlock(&ev_pcb->evp_mtx); 3260 continue; 3261 } 3262 if (ev_pcb->evp_vendor_code_filter != KEV_ANY_VENDOR) { 3263 if (ev_pcb->evp_vendor_code_filter != ev->vendor_code) { 3264 lck_mtx_unlock(&ev_pcb->evp_mtx); 3265 continue; 3266 } 3267 3268 if (ev_pcb->evp_class_filter != KEV_ANY_CLASS) { 3269 if (ev_pcb->evp_class_filter != ev->kev_class) { 3270 lck_mtx_unlock(&ev_pcb->evp_mtx); 3271 continue; 3272 } 3273 3274 if ((ev_pcb->evp_subclass_filter != 3275 KEV_ANY_SUBCLASS) && 3276 (ev_pcb->evp_subclass_filter != 3277 ev->kev_subclass)) { 3278 lck_mtx_unlock(&ev_pcb->evp_mtx); 3279 continue; 3280 } 3281 } 3282 } 3283 3284 m2 = m_copym(m, 0, m->m_len, M_NOWAIT); 3285 if (m2 == 0) { 3286 OSIncrementAtomic64((SInt64 *)&kevtstat.kes_nomem); 3287 m_free(m); 3288 lck_mtx_unlock(&ev_pcb->evp_mtx); 3289 lck_rw_done(kev_rwlock); 3290 return (ENOMEM); 3291 } 3292 if (sbappendrecord(&ev_pcb->evp_socket->so_rcv, m2)) { 3293 /* 3294 * We use "m" for the socket stats as it would be 3295 * unsafe to use "m2" 3296 */ 3297 so_inc_recv_data_stat(ev_pcb->evp_socket, 3298 1, m->m_len, SO_TC_BE); 3299 3300 sorwakeup(ev_pcb->evp_socket); 3301 OSIncrementAtomic64((SInt64 *)&kevtstat.kes_posted); 3302 } else { 3303 OSIncrementAtomic64((SInt64 *)&kevtstat.kes_fullsock); 3304 } 3305 lck_mtx_unlock(&ev_pcb->evp_mtx); 3306 } 3307 m_free(m); 3308 lck_rw_done(kev_rwlock); 3309 3310 return (0); 3311} 3312 3313static int 3314kev_control(struct socket *so, 3315 u_long cmd, 3316 caddr_t data, 3317 __unused struct ifnet *ifp, 3318 __unused struct proc *p) 3319{ 3320 struct kev_request *kev_req = (struct kev_request *) data; 3321 struct kern_event_pcb *ev_pcb; 3322 struct kev_vendor_code *kev_vendor; 3323 u_int32_t *id_value = (u_int32_t *) data; 3324 3325 switch (cmd) { 3326 case SIOCGKEVID: 3327 *id_value = static_event_id; 3328 break; 3329 case SIOCSKEVFILT: 3330 ev_pcb = (struct kern_event_pcb *) so->so_pcb; 3331 ev_pcb->evp_vendor_code_filter = kev_req->vendor_code; 3332 ev_pcb->evp_class_filter = kev_req->kev_class; 3333 ev_pcb->evp_subclass_filter = kev_req->kev_subclass; 3334 break; 3335 case SIOCGKEVFILT: 3336 ev_pcb = (struct kern_event_pcb *) so->so_pcb; 3337 kev_req->vendor_code = ev_pcb->evp_vendor_code_filter; 3338 kev_req->kev_class = ev_pcb->evp_class_filter; 3339 kev_req->kev_subclass = ev_pcb->evp_subclass_filter; 3340 break; 3341 case SIOCGKEVVENDOR: 3342 kev_vendor = (struct kev_vendor_code *)data; 3343 /* Make sure string is NULL terminated */ 3344 kev_vendor->vendor_string[KEV_VENDOR_CODE_MAX_STR_LEN-1] = 0; 3345 return (net_str_id_find_internal(kev_vendor->vendor_string, 3346 &kev_vendor->vendor_code, NSI_VENDOR_CODE, 0)); 3347 default: 3348 return (ENOTSUP); 3349 } 3350 3351 return (0); 3352} 3353 3354int 3355kevt_getstat SYSCTL_HANDLER_ARGS 3356{ 3357#pragma unused(oidp, arg1, arg2) 3358 int error = 0; 3359 3360 lck_rw_lock_shared(kev_rwlock); 3361 3362 if (req->newptr != USER_ADDR_NULL) { 3363 error = EPERM; 3364 goto done; 3365 } 3366 if (req->oldptr == USER_ADDR_NULL) { 3367 req->oldidx = sizeof(struct kevtstat); 3368 goto done; 3369 } 3370 3371 error = SYSCTL_OUT(req, &kevtstat, 3372 MIN(sizeof(struct kevtstat), req->oldlen)); 3373done: 3374 lck_rw_done(kev_rwlock); 3375 3376 return (error); 3377} 3378 3379__private_extern__ int 3380kevt_pcblist SYSCTL_HANDLER_ARGS 3381{ 3382#pragma unused(oidp, arg1, arg2) 3383 int error = 0; 3384 int n, i; 3385 struct xsystmgen xsg; 3386 void *buf = NULL; 3387 size_t item_size = ROUNDUP64(sizeof (struct xkevtpcb)) + 3388 ROUNDUP64(sizeof (struct xsocket_n)) + 3389 2 * ROUNDUP64(sizeof (struct xsockbuf_n)) + 3390 ROUNDUP64(sizeof (struct xsockstat_n)); 3391 struct kern_event_pcb *ev_pcb; 3392 3393 buf = _MALLOC(item_size, M_TEMP, M_WAITOK | M_ZERO); 3394 if (buf == NULL) 3395 return (ENOMEM); 3396 3397 lck_rw_lock_shared(kev_rwlock); 3398 3399 n = kevtstat.kes_pcbcount; 3400 3401 if (req->oldptr == USER_ADDR_NULL) { 3402 req->oldidx = (n + n/8) * item_size; 3403 goto done; 3404 } 3405 if (req->newptr != USER_ADDR_NULL) { 3406 error = EPERM; 3407 goto done; 3408 } 3409 bzero(&xsg, sizeof (xsg)); 3410 xsg.xg_len = sizeof (xsg); 3411 xsg.xg_count = n; 3412 xsg.xg_gen = kevtstat.kes_gencnt; 3413 xsg.xg_sogen = so_gencnt; 3414 error = SYSCTL_OUT(req, &xsg, sizeof (xsg)); 3415 if (error) { 3416 goto done; 3417 } 3418 /* 3419 * We are done if there is no pcb 3420 */ 3421 if (n == 0) { 3422 goto done; 3423 } 3424 3425 i = 0; 3426 for (i = 0, ev_pcb = LIST_FIRST(&kern_event_head); 3427 i < n && ev_pcb != NULL; 3428 i++, ev_pcb = LIST_NEXT(ev_pcb, evp_link)) { 3429 struct xkevtpcb *xk = (struct xkevtpcb *)buf; 3430 struct xsocket_n *xso = (struct xsocket_n *) 3431 ADVANCE64(xk, sizeof (*xk)); 3432 struct xsockbuf_n *xsbrcv = (struct xsockbuf_n *) 3433 ADVANCE64(xso, sizeof (*xso)); 3434 struct xsockbuf_n *xsbsnd = (struct xsockbuf_n *) 3435 ADVANCE64(xsbrcv, sizeof (*xsbrcv)); 3436 struct xsockstat_n *xsostats = (struct xsockstat_n *) 3437 ADVANCE64(xsbsnd, sizeof (*xsbsnd)); 3438 3439 bzero(buf, item_size); 3440 3441 lck_mtx_lock(&ev_pcb->evp_mtx); 3442 3443 xk->kep_len = sizeof(struct xkevtpcb); 3444 xk->kep_kind = XSO_EVT; 3445 xk->kep_evtpcb = (uint64_t)VM_KERNEL_ADDRPERM(ev_pcb); 3446 xk->kep_vendor_code_filter = ev_pcb->evp_vendor_code_filter; 3447 xk->kep_class_filter = ev_pcb->evp_class_filter; 3448 xk->kep_subclass_filter = ev_pcb->evp_subclass_filter; 3449 3450 sotoxsocket_n(ev_pcb->evp_socket, xso); 3451 sbtoxsockbuf_n(ev_pcb->evp_socket ? 3452 &ev_pcb->evp_socket->so_rcv : NULL, xsbrcv); 3453 sbtoxsockbuf_n(ev_pcb->evp_socket ? 3454 &ev_pcb->evp_socket->so_snd : NULL, xsbsnd); 3455 sbtoxsockstat_n(ev_pcb->evp_socket, xsostats); 3456 3457 lck_mtx_unlock(&ev_pcb->evp_mtx); 3458 3459 error = SYSCTL_OUT(req, buf, item_size); 3460 } 3461 3462 if (error == 0) { 3463 /* 3464 * Give the user an updated idea of our state. 3465 * If the generation differs from what we told 3466 * her before, she knows that something happened 3467 * while we were processing this request, and it 3468 * might be necessary to retry. 3469 */ 3470 bzero(&xsg, sizeof (xsg)); 3471 xsg.xg_len = sizeof (xsg); 3472 xsg.xg_count = n; 3473 xsg.xg_gen = kevtstat.kes_gencnt; 3474 xsg.xg_sogen = so_gencnt; 3475 error = SYSCTL_OUT(req, &xsg, sizeof (xsg)); 3476 if (error) { 3477 goto done; 3478 } 3479 } 3480 3481done: 3482 lck_rw_done(kev_rwlock); 3483 3484 return (error); 3485} 3486 3487#endif /* SOCKETS */ 3488 3489 3490int 3491fill_kqueueinfo(struct kqueue *kq, struct kqueue_info * kinfo) 3492{ 3493 struct vinfo_stat * st; 3494 3495 st = &kinfo->kq_stat; 3496 3497 st->vst_size = kq->kq_count; 3498 if (kq->kq_state & KQ_KEV64) 3499 st->vst_blksize = sizeof(struct kevent64_s); 3500 else 3501 st->vst_blksize = sizeof(struct kevent); 3502 st->vst_mode = S_IFIFO; 3503 if (kq->kq_state & KQ_SEL) 3504 kinfo->kq_state |= PROC_KQUEUE_SELECT; 3505 if (kq->kq_state & KQ_SLEEP) 3506 kinfo->kq_state |= PROC_KQUEUE_SLEEP; 3507 3508 return (0); 3509} 3510 3511 3512void 3513knote_markstayqueued(struct knote *kn) 3514{ 3515 kqlock(kn->kn_kq); 3516 kn->kn_status |= KN_STAYQUEUED; 3517 knote_enqueue(kn); 3518 kqunlock(kn->kn_kq); 3519} 3520