X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/2d21ac55c334faf3a56e5634905ed6987fc787d4..c18c124eaa464aaaa5549e99e5a70fc9cbb50944:/bsd/kern/kern_event.c?ds=inline diff --git a/bsd/kern/kern_event.c b/bsd/kern/kern_event.c index 5c940792d..708aef474 100644 --- a/bsd/kern/kern_event.c +++ b/bsd/kern/kern_event.c @@ -1,8 +1,8 @@ /* - * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2014 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ - * + * * This file contains Original Code and/or Modifications of Original Code * as defined in and that are subject to the Apple Public Source License * Version 2.0 (the 'License'). You may not use this file except in @@ -11,10 +11,10 @@ * unlawful or unlicensed copies of an Apple operating system, or to * circumvent, violate, or enable the circumvention or violation of, any * terms of an Apple operating system software license agreement. - * + * * Please obtain a copy of the License at * http://www.opensource.apple.com/apsl/ and read it before using this file. - * + * * The Original Code and all software distributed under the License are * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, @@ -22,7 +22,7 @@ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. * Please see the License for the specific language governing rights and * limitations under the License. - * + * * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ * */ @@ -62,7 +62,7 @@ #include #include #include -#include +#include #include #include #include @@ -78,125 +78,199 @@ #include #include #include +#include #include #include +#include -#include +#include #include #include #include +#include #include #include #include -#include "kpi_mbuf_internal.h" +#include "net/net_str_id.h" + +#include + +#if VM_PRESSURE_EVENTS +#include +#endif + +#if CONFIG_MEMORYSTATUS +#include +#endif MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system"); +#define KQ_EVENT NULL + static inline void kqlock(struct kqueue *kq); static inline void kqunlock(struct kqueue *kq); -static int kqlock2knoteuse(struct kqueue *kq, struct knote *kn); -static int kqlock2knoteusewait(struct kqueue *kq, struct knote *kn); -static int kqlock2knotedrop(struct kqueue *kq, struct knote *kn); -static int knoteuse2kqlock(struct kqueue *kq, struct knote *kn); - -static void kqueue_wakeup(struct kqueue *kq); -static int kqueue_read(struct fileproc *fp, struct uio *uio, - int flags, vfs_context_t ctx); -static int kqueue_write(struct fileproc *fp, struct uio *uio, - int flags, vfs_context_t ctx); -static int kqueue_ioctl(struct fileproc *fp, u_long com, caddr_t data, - vfs_context_t ctx); -static int kqueue_select(struct fileproc *fp, int which, void *wql, - vfs_context_t ctx); -static int kqueue_close(struct fileglob *fp, vfs_context_t ctx); -static int kqueue_kqfilter(struct fileproc *fp, struct knote *kn, vfs_context_t ctx); -extern int kqueue_stat(struct fileproc *fp, void *ub, int isstat64, vfs_context_t ctx); - -static struct fileops kqueueops = { - kqueue_read, - kqueue_write, - kqueue_ioctl, - kqueue_select, - kqueue_close, - kqueue_kqfilter, - 0 +static int kqlock2knoteuse(struct kqueue *kq, struct knote *kn); +static int kqlock2knoteusewait(struct kqueue *kq, struct knote *kn); +static int kqlock2knotedrop(struct kqueue *kq, struct knote *kn); +static int knoteuse2kqlock(struct kqueue *kq, struct knote *kn); + +static void kqueue_wakeup(struct kqueue *kq, int closed); +static int kqueue_read(struct fileproc *fp, struct uio *uio, + int flags, vfs_context_t ctx); +static int kqueue_write(struct fileproc *fp, struct uio *uio, + int flags, vfs_context_t ctx); +static int kqueue_ioctl(struct fileproc *fp, u_long com, caddr_t data, + vfs_context_t ctx); +static int kqueue_select(struct fileproc *fp, int which, void *wql, + vfs_context_t ctx); +static int kqueue_close(struct fileglob *fg, vfs_context_t ctx); +static int kqueue_kqfilter(struct fileproc *fp, struct knote *kn, + vfs_context_t ctx); +static int kqueue_drain(struct fileproc *fp, vfs_context_t ctx); + +static const struct fileops kqueueops = { + .fo_type = DTYPE_KQUEUE, + .fo_read = kqueue_read, + .fo_write = kqueue_write, + .fo_ioctl = kqueue_ioctl, + .fo_select = kqueue_select, + .fo_close = kqueue_close, + .fo_kqfilter = kqueue_kqfilter, + .fo_drain = kqueue_drain, }; -static int kevent_copyin(user_addr_t *addrp, struct kevent *kevp, struct proc *p); -static int kevent_copyout(struct kevent *kevp, user_addr_t *addrp, struct proc *p); - -static int kevent_callback(struct kqueue *kq, struct kevent *kevp, void *data); -static void kevent_continue(struct kqueue *kq, void *data, int error); -static void kevent_scan_continue(void *contp, wait_result_t wait_result); -static int kevent_process(struct kqueue *kq, kevent_callback_t callback, - void *data, int *countp, struct proc *p); -static void knote_put(struct knote *kn); -static int knote_fdpattach(struct knote *kn, struct filedesc *fdp, struct proc *p); -static void knote_drop(struct knote *kn, struct proc *p); -static void knote_activate(struct knote *kn); -static void knote_deactivate(struct knote *kn); -static void knote_enqueue(struct knote *kn); -static void knote_dequeue(struct knote *kn); -static struct knote *knote_alloc(void); -static void knote_free(struct knote *kn); - -static int filt_fileattach(struct knote *kn); -static struct filterops file_filtops = - { 1, filt_fileattach, NULL, NULL }; - -static void filt_kqdetach(struct knote *kn); -static int filt_kqueue(struct knote *kn, long hint); -static struct filterops kqread_filtops = - { 1, NULL, filt_kqdetach, filt_kqueue }; +static int kevent_internal(struct proc *p, int iskev64, user_addr_t changelist, + int nchanges, user_addr_t eventlist, int nevents, int fd, + user_addr_t utimeout, unsigned int flags, int32_t *retval); +static int kevent_copyin(user_addr_t *addrp, struct kevent64_s *kevp, + struct proc *p, int iskev64); +static int kevent_copyout(struct kevent64_s *kevp, user_addr_t *addrp, + struct proc *p, int iskev64); +char * kevent_description(struct kevent64_s *kevp, char *s, size_t n); + +static int kevent_callback(struct kqueue *kq, struct kevent64_s *kevp, + void *data); +static void kevent_continue(struct kqueue *kq, void *data, int error); +static void kqueue_scan_continue(void *contp, wait_result_t wait_result); +static int kqueue_process(struct kqueue *kq, kevent_callback_t callback, + void *data, int *countp, struct proc *p); +static int kqueue_begin_processing(struct kqueue *kq); +static void kqueue_end_processing(struct kqueue *kq); +static int knote_process(struct knote *kn, kevent_callback_t callback, + void *data, struct kqtailq *inprocessp, struct proc *p); +static void knote_put(struct knote *kn); +static int knote_fdpattach(struct knote *kn, struct filedesc *fdp, + struct proc *p); +static void knote_drop(struct knote *kn, struct proc *p); +static void knote_activate(struct knote *kn, int); +static void knote_deactivate(struct knote *kn); +static void knote_enqueue(struct knote *kn); +static void knote_dequeue(struct knote *kn); +static struct knote *knote_alloc(void); +static void knote_free(struct knote *kn); + +static int filt_fileattach(struct knote *kn); +static struct filterops file_filtops = { + .f_isfd = 1, + .f_attach = filt_fileattach, +}; -/* - * placeholder for not-yet-implemented filters - */ -static int filt_badattach(struct knote *kn); -static struct filterops bad_filtops = - { 0, filt_badattach, 0 , 0 }; +static void filt_kqdetach(struct knote *kn); +static int filt_kqueue(struct knote *kn, long hint); +static struct filterops kqread_filtops = { + .f_isfd = 1, + .f_detach = filt_kqdetach, + .f_event = filt_kqueue, +}; + +/* placeholder for not-yet-implemented filters */ +static int filt_badattach(struct knote *kn); +static struct filterops bad_filtops = { + .f_attach = filt_badattach, +}; + +static int filt_procattach(struct knote *kn); +static void filt_procdetach(struct knote *kn); +static int filt_proc(struct knote *kn, long hint); +static struct filterops proc_filtops = { + .f_attach = filt_procattach, + .f_detach = filt_procdetach, + .f_event = filt_proc, +}; -static int filt_procattach(struct knote *kn); -static void filt_procdetach(struct knote *kn); -static int filt_proc(struct knote *kn, long hint); +#if VM_PRESSURE_EVENTS +static int filt_vmattach(struct knote *kn); +static void filt_vmdetach(struct knote *kn); +static int filt_vm(struct knote *kn, long hint); +static struct filterops vm_filtops = { + .f_attach = filt_vmattach, + .f_detach = filt_vmdetach, + .f_event = filt_vm, +}; +#endif /* VM_PRESSURE_EVENTS */ -static struct filterops proc_filtops = - { 0, filt_procattach, filt_procdetach, filt_proc }; +#if CONFIG_MEMORYSTATUS +extern struct filterops memorystatus_filtops; +#endif /* CONFIG_MEMORYSTATUS */ extern struct filterops fs_filtops; extern struct filterops sig_filtops; - /* Timer filter */ -static int filt_timercompute(struct knote *kn, uint64_t *abs_time); -static void filt_timerexpire(void *knx, void *param1); -static int filt_timerattach(struct knote *kn); -static void filt_timerdetach(struct knote *kn); -static int filt_timer(struct knote *kn, long hint); +static int filt_timerattach(struct knote *kn); +static void filt_timerdetach(struct knote *kn); +static int filt_timer(struct knote *kn, long hint); +static void filt_timertouch(struct knote *kn, struct kevent64_s *kev, + long type); +static struct filterops timer_filtops = { + .f_attach = filt_timerattach, + .f_detach = filt_timerdetach, + .f_event = filt_timer, + .f_touch = filt_timertouch, +}; -static struct filterops timer_filtops = - { 0, filt_timerattach, filt_timerdetach, filt_timer }; +/* Helpers */ +static void filt_timerexpire(void *knx, void *param1); +static int filt_timervalidate(struct knote *kn); +static void filt_timerupdate(struct knote *kn); +static void filt_timercancel(struct knote *kn); -/* to avoid arming timers that fire quicker than we can handle */ -static uint64_t filt_timerfloor = 0; +#define TIMER_RUNNING 0x1 +#define TIMER_CANCELWAIT 0x2 static lck_mtx_t _filt_timerlock; -static void filt_timerlock(void); -static void filt_timerunlock(void); +static void filt_timerlock(void); +static void filt_timerunlock(void); -static zone_t knote_zone; +static zone_t knote_zone; -#define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) +#define KN_HASH(val, mask) (((val) ^ (val >> 8)) & (mask)) #if 0 extern struct filterops aio_filtops; #endif +/* Mach portset filter */ +extern struct filterops machport_filtops; + +/* User filter */ +static int filt_userattach(struct knote *kn); +static void filt_userdetach(struct knote *kn); +static int filt_user(struct knote *kn, long hint); +static void filt_usertouch(struct knote *kn, struct kevent64_s *kev, + long type); +static struct filterops user_filtops = { + .f_attach = filt_userattach, + .f_detach = filt_userdetach, + .f_event = filt_user, + .f_touch = filt_usertouch, +}; + /* - * Table for for all system-defined filters. + * Table for all system-defined filters. */ static struct filterops *sysfilt_ops[] = { &file_filtops, /* EVFILT_READ */ @@ -210,8 +284,21 @@ static struct filterops *sysfilt_ops[] = { &proc_filtops, /* EVFILT_PROC */ &sig_filtops, /* EVFILT_SIGNAL */ &timer_filtops, /* EVFILT_TIMER */ - &bad_filtops, /* EVFILT_MACHPORT */ - &fs_filtops /* EVFILT_FS */ + &machport_filtops, /* EVFILT_MACHPORT */ + &fs_filtops, /* EVFILT_FS */ + &user_filtops, /* EVFILT_USER */ + &bad_filtops, /* unused */ +#if VM_PRESSURE_EVENTS + &vm_filtops, /* EVFILT_VM */ +#else + &bad_filtops, /* EVFILT_VM */ +#endif + &file_filtops, /* EVFILT_SOCK */ +#if CONFIG_MEMORYSTATUS + &memorystatus_filtops, /* EVFILT_MEMORYSTATUS */ +#else + &bad_filtops, /* EVFILT_MEMORYSTATUS */ +#endif }; /* @@ -237,13 +324,12 @@ kqunlock(struct kqueue *kq) lck_spin_unlock(&kq->kq_lock); } -/* +/* * Convert a kq lock to a knote use referece. * * If the knote is being dropped, we can't get * a use reference, so just return with it * still locked. - * * - kq locked at entry * - unlock on exit if we get the use reference */ @@ -251,36 +337,39 @@ static int kqlock2knoteuse(struct kqueue *kq, struct knote *kn) { if (kn->kn_status & KN_DROPPING) - return 0; + return (0); kn->kn_inuse++; kqunlock(kq); - return 1; - } + return (1); +} -/* - * Convert a kq lock to a knote use referece. +/* + * Convert a kq lock to a knote use referece, + * but wait for attach and drop events to complete. * * If the knote is being dropped, we can't get * a use reference, so just return with it * still locked. - * * - kq locked at entry * - kq always unlocked on exit */ static int kqlock2knoteusewait(struct kqueue *kq, struct knote *kn) { - if (!kqlock2knoteuse(kq, kn)) { - kn->kn_status |= KN_DROPWAIT; - assert_wait(&kn->kn_status, THREAD_UNINT); + if ((kn->kn_status & (KN_DROPPING | KN_ATTACHING)) != 0) { + kn->kn_status |= KN_USEWAIT; + wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, + &kn->kn_status, THREAD_UNINT, 0); kqunlock(kq); thread_block(THREAD_CONTINUE_NULL); - return 0; + return (0); } - return 1; - } + kn->kn_inuse++; + kqunlock(kq); + return (1); +} -/* +/* * Convert from a knote use reference back to kq lock. * * Drop a use reference and wake any waiters if @@ -294,23 +383,28 @@ static int knoteuse2kqlock(struct kqueue *kq, struct knote *kn) { kqlock(kq); - if ((--kn->kn_inuse == 0) && - (kn->kn_status & KN_USEWAIT)) { - kn->kn_status &= ~KN_USEWAIT; - thread_wakeup(&kn->kn_inuse); + if (--kn->kn_inuse == 0) { + if ((kn->kn_status & KN_ATTACHING) != 0) { + kn->kn_status &= ~KN_ATTACHING; + } + if ((kn->kn_status & KN_USEWAIT) != 0) { + kn->kn_status &= ~KN_USEWAIT; + wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, + &kn->kn_status, THREAD_AWAKENED); + } } return ((kn->kn_status & KN_DROPPING) == 0); - } +} -/* - * Convert a kq lock to a knote drop referece. +/* + * Convert a kq lock to a knote drop reference. * * If the knote is in use, wait for the use count * to subside. We first mark our intention to drop * it - keeping other users from "piling on." * If we are too late, we have to wait for the * other drop to complete. - * + * * - kq locked at entry * - always unlocked on exit. * - caller can't hold any locks that would prevent @@ -319,27 +413,26 @@ knoteuse2kqlock(struct kqueue *kq, struct knote *kn) static int kqlock2knotedrop(struct kqueue *kq, struct knote *kn) { + int oktodrop; - if ((kn->kn_status & KN_DROPPING) == 0) { - kn->kn_status |= KN_DROPPING; - if (kn->kn_inuse > 0) { - kn->kn_status |= KN_USEWAIT; - assert_wait(&kn->kn_inuse, THREAD_UNINT); - kqunlock(kq); - thread_block(THREAD_CONTINUE_NULL); - } else + oktodrop = ((kn->kn_status & (KN_DROPPING | KN_ATTACHING)) == 0); + kn->kn_status &= ~KN_STAYQUEUED; + kn->kn_status |= KN_DROPPING; + if (oktodrop) { + if (kn->kn_inuse == 0) { kqunlock(kq); - return 1; - } else { - kn->kn_status |= KN_DROPWAIT; - assert_wait(&kn->kn_status, THREAD_UNINT); - kqunlock(kq); - thread_block(THREAD_CONTINUE_NULL); - return 0; + return (oktodrop); + } } + kn->kn_status |= KN_USEWAIT; + wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, &kn->kn_status, + THREAD_UNINT, 0); + kqunlock(kq); + thread_block(THREAD_CONTINUE_NULL); + return (oktodrop); } - -/* + +/* * Release a knote use count reference. */ static void @@ -348,30 +441,28 @@ knote_put(struct knote *kn) struct kqueue *kq = kn->kn_kq; kqlock(kq); - if ((--kn->kn_inuse == 0) && - (kn->kn_status & KN_USEWAIT)) { - kn->kn_status &= ~KN_USEWAIT; - thread_wakeup(&kn->kn_inuse); + if (--kn->kn_inuse == 0) { + if ((kn->kn_status & KN_USEWAIT) != 0) { + kn->kn_status &= ~KN_USEWAIT; + wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, + &kn->kn_status, THREAD_AWAKENED); + } } kqunlock(kq); - } - - +} static int filt_fileattach(struct knote *kn) { - return (fo_kqfilter(kn->kn_fp, kn, vfs_context_current())); } -#define f_flag f_fglob->fg_flag -#define f_type f_fglob->fg_type -#define f_msgcount f_fglob->fg_msgcount -#define f_cred f_fglob->fg_cred -#define f_ops f_fglob->fg_ops -#define f_offset f_fglob->fg_offset -#define f_data f_fglob->fg_data +#define f_flag f_fglob->fg_flag +#define f_msgcount f_fglob->fg_msgcount +#define f_cred f_fglob->fg_cred +#define f_ops f_fglob->fg_ops +#define f_offset f_fglob->fg_offset +#define f_data f_fglob->fg_data static void filt_kqdetach(struct knote *kn) @@ -399,15 +490,32 @@ filt_procattach(struct knote *kn) struct proc *p; assert(PID_MAX < NOTE_PDATAMASK); - + if ((kn->kn_sfflags & (NOTE_TRACK | NOTE_TRACKERR | NOTE_CHILD)) != 0) - return(ENOTSUP); + return (ENOTSUP); p = proc_find(kn->kn_id); if (p == NULL) { return (ESRCH); } + const int NoteExitStatusBits = NOTE_EXIT | NOTE_EXITSTATUS; + + if ((kn->kn_sfflags & NoteExitStatusBits) == NoteExitStatusBits) + do { + pid_t selfpid = proc_selfpid(); + + if (p->p_ppid == selfpid) + break; /* parent => ok */ + + if ((p->p_lflag & P_LTRACED) != 0 && + (p->p_oppid == selfpid)) + break; /* parent-in-waiting => ok */ + + proc_rele(p); + return (EACCES); + } while (0); + proc_klist_lock(); kn->kn_flags |= EV_CLEAR; /* automatically set */ @@ -433,7 +541,7 @@ filt_procdetach(struct knote *kn) struct proc *p; proc_klist_lock(); - + p = kn->kn_ptr.p_proc; if (p != PROC_NULL) { kn->kn_ptr.p_proc = PROC_NULL; @@ -446,8 +554,11 @@ filt_procdetach(struct knote *kn) static int filt_proc(struct knote *kn, long hint) { - struct proc * p; - + /* + * Note: a lot of bits in hint may be obtained from the knote + * To free some of those bits, see Freeing up + * bits in hint for filt_proc + */ /* hint is 0 when called from above */ if (hint != 0) { u_int event; @@ -459,48 +570,160 @@ filt_proc(struct knote *kn, long hint) */ event = (u_int)hint & NOTE_PCTRLMASK; + /* + * termination lifecycle events can happen while a debugger + * has reparented a process, in which case notifications + * should be quashed except to the tracing parent. When + * the debugger reaps the child (either via wait4(2) or + * process exit), the child will be reparented to the original + * parent and these knotes re-fired. + */ + if (event & NOTE_EXIT) { + if ((kn->kn_ptr.p_proc->p_oppid != 0) + && (kn->kn_kq->kq_p->p_pid != kn->kn_ptr.p_proc->p_ppid)) { + /* + * This knote is not for the current ptrace(2) parent, ignore. + */ + return 0; + } + } + /* * if the user is interested in this event, record it. */ if (kn->kn_sfflags & event) kn->kn_fflags |= event; +#pragma clang diagnostic push +#pragma clang diagnostic ignored "-Wdeprecated-declarations" + if ((event == NOTE_REAP) || ((event == NOTE_EXIT) && !(kn->kn_sfflags & NOTE_REAP))) { + kn->kn_flags |= (EV_EOF | EV_ONESHOT); + } +#pragma clang diagnostic pop + + /* - * If this is the last possible event for the - * knote, unlink this knote from the process - * before the process goes away. + * The kernel has a wrapper in place that returns the same data + * as is collected here, in kn_data. Any changes to how + * NOTE_EXITSTATUS and NOTE_EXIT_DETAIL are collected + * should also be reflected in the proc_pidnoteexit() wrapper. */ - if (event == NOTE_REAP || (event == NOTE_EXIT && !(kn->kn_sfflags & NOTE_REAP))) { - kn->kn_flags |= (EV_EOF | EV_ONESHOT); - p = kn->kn_ptr.p_proc; - if (p != PROC_NULL) { - kn->kn_ptr.p_proc = PROC_NULL; - KNOTE_DETACH(&p->p_klist, kn); + if (event == NOTE_EXIT) { + kn->kn_data = 0; + if ((kn->kn_sfflags & NOTE_EXITSTATUS) != 0) { + kn->kn_fflags |= NOTE_EXITSTATUS; + kn->kn_data |= (hint & NOTE_PDATAMASK); + } + if ((kn->kn_sfflags & NOTE_EXIT_DETAIL) != 0) { + kn->kn_fflags |= NOTE_EXIT_DETAIL; + if ((kn->kn_ptr.p_proc->p_lflag & + P_LTERM_DECRYPTFAIL) != 0) { + kn->kn_data |= NOTE_EXIT_DECRYPTFAIL; + } + if ((kn->kn_ptr.p_proc->p_lflag & + P_LTERM_JETSAM) != 0) { + kn->kn_data |= NOTE_EXIT_MEMORY; + switch (kn->kn_ptr.p_proc->p_lflag & + P_JETSAM_MASK) { + case P_JETSAM_VMPAGESHORTAGE: + kn->kn_data |= NOTE_EXIT_MEMORY_VMPAGESHORTAGE; + break; + case P_JETSAM_VMTHRASHING: + kn->kn_data |= NOTE_EXIT_MEMORY_VMTHRASHING; + break; + case P_JETSAM_FCTHRASHING: + kn->kn_data |= NOTE_EXIT_MEMORY_FCTHRASHING; + break; + case P_JETSAM_VNODE: + kn->kn_data |= NOTE_EXIT_MEMORY_VNODE; + break; + case P_JETSAM_HIWAT: + kn->kn_data |= NOTE_EXIT_MEMORY_HIWAT; + break; + case P_JETSAM_PID: + kn->kn_data |= NOTE_EXIT_MEMORY_PID; + break; + case P_JETSAM_IDLEEXIT: + kn->kn_data |= NOTE_EXIT_MEMORY_IDLE; + break; + } + } + if ((kn->kn_ptr.p_proc->p_csflags & + CS_KILLED) != 0) { + kn->kn_data |= NOTE_EXIT_CSERROR; + } } - return (1); } - } /* atomic check, no locking need when called from above */ - return (kn->kn_fflags != 0); + return (kn->kn_fflags != 0); } +#if VM_PRESSURE_EVENTS /* - * filt_timercompute - compute absolute timeout + * Virtual memory kevents + * + * author: Matt Jacobson [matthew_jacobson@apple.com] + */ + +static int +filt_vmattach(struct knote *kn) +{ + /* + * The note will be cleared once the information has been flushed to + * the client. If there is still pressure, we will be re-alerted. + */ + kn->kn_flags |= EV_CLEAR; + return (vm_knote_register(kn)); +} + +static void +filt_vmdetach(struct knote *kn) +{ + vm_knote_unregister(kn); +} + +static int +filt_vm(struct knote *kn, long hint) +{ + /* hint == 0 means this is just an alive? check (always true) */ + if (hint != 0) { + const pid_t pid = (pid_t)hint; + if ((kn->kn_sfflags & NOTE_VM_PRESSURE) && + (kn->kn_kq->kq_p->p_pid == pid)) { + kn->kn_fflags |= NOTE_VM_PRESSURE; + } + } + + return (kn->kn_fflags != 0); +} +#endif /* VM_PRESSURE_EVENTS */ + +/* + * filt_timervalidate - process data from user + * + * Converts to either interval or deadline format. * * The saved-data field in the knote contains the * time value. The saved filter-flags indicates * the unit of measurement. * - * If the timeout is not absolute, adjust it for - * the current time. + * After validation, either the saved-data field + * contains the interval in absolute time, or ext[0] + * contains the expected deadline. If that deadline + * is in the past, ext[0] is 0. + * + * Returns EINVAL for unrecognized units of time. + * + * Timer filter lock is held. + * */ static int -filt_timercompute(struct knote *kn, uint64_t *abs_time) +filt_timervalidate(struct knote *kn) { uint64_t multiplier; - uint64_t raw; + uint64_t raw = 0; switch (kn->kn_sfflags & (NOTE_SECONDS|NOTE_USECONDS|NOTE_NSECONDS)) { case NOTE_SECONDS: @@ -516,38 +739,86 @@ filt_timercompute(struct knote *kn, uint64_t *abs_time) multiplier = NSEC_PER_SEC / 1000; break; default: - return EINVAL; + return (EINVAL); } - nanoseconds_to_absolutetime((uint64_t)kn->kn_sdata * multiplier, &raw); - if (raw <= filt_timerfloor) { - *abs_time = 0; - return 0; + + /* transform the slop delta(leeway) in kn_ext[1] if passed to same time scale */ + if(kn->kn_sfflags & NOTE_LEEWAY){ + nanoseconds_to_absolutetime((uint64_t)kn->kn_ext[1] * multiplier, &raw); + kn->kn_ext[1] = raw; } - if ((kn->kn_sfflags & NOTE_ABSOLUTE) == NOTE_ABSOLUTE) { - uint32_t seconds, nanoseconds; + + nanoseconds_to_absolutetime((uint64_t)kn->kn_sdata * multiplier, &raw); + + kn->kn_ext[0] = 0; + kn->kn_sdata = 0; + + if (kn->kn_sfflags & NOTE_ABSOLUTE) { + clock_sec_t seconds; + clock_nsec_t nanoseconds; uint64_t now; clock_get_calendar_nanotime(&seconds, &nanoseconds); - nanoseconds_to_absolutetime((uint64_t)seconds * NSEC_PER_SEC + nanoseconds, - &now); - if (now >= raw + filt_timerfloor) { - *abs_time = 0; - return 0; + nanoseconds_to_absolutetime((uint64_t)seconds * NSEC_PER_SEC + + nanoseconds, &now); + + if (raw < now) { + /* time has already passed */ + kn->kn_ext[0] = 0; + } else { + raw -= now; + clock_absolutetime_interval_to_deadline(raw, + &kn->kn_ext[0]); } - raw -= now; - } - clock_absolutetime_interval_to_deadline(raw, abs_time); - return 0; + } else { + kn->kn_sdata = raw; + } + + return (0); } -/* +/* + * filt_timerupdate - compute the next deadline + * + * Repeating timers store their interval in kn_sdata. Absolute + * timers have already calculated the deadline, stored in ext[0]. + * + * On return, the next deadline (or zero if no deadline is needed) + * is stored in kn_ext[0]. + * + * Timer filter lock is held. + */ +static void +filt_timerupdate(struct knote *kn) +{ + /* if there's no interval, deadline is just in kn_ext[0] */ + if (kn->kn_sdata == 0) + return; + + /* if timer hasn't fired before, fire in interval nsecs */ + if (kn->kn_ext[0] == 0) { + clock_absolutetime_interval_to_deadline(kn->kn_sdata, + &kn->kn_ext[0]); + } else { + /* + * If timer has fired before, schedule the next pop + * relative to the last intended deadline. + * + * We could check for whether the deadline has expired, + * but the thread call layer can handle that. + */ + kn->kn_ext[0] += kn->kn_sdata; + } +} + +/* * filt_timerexpire - the timer callout routine * - * Just propagate the timer event into the knote - * filter routine (by going through the knote - * synchronization point). Pass a hint to - * indicate this is a real event, not just a - * query from above. + * Just propagate the timer event into the knote + * filter routine (by going through the knote + * synchronization point). Pass a hint to + * indicate this is a real event, not just a + * query from above. */ static void filt_timerexpire(void *knx, __unused void *spare) @@ -555,167 +826,250 @@ filt_timerexpire(void *knx, __unused void *spare) struct klist timer_list; struct knote *kn = knx; + filt_timerlock(); + + kn->kn_hookid &= ~TIMER_RUNNING; + /* no "object" for timers, so fake a list */ SLIST_INIT(&timer_list); - SLIST_INSERT_HEAD(&timer_list, kn, kn_selnext); + SLIST_INSERT_HEAD(&timer_list, kn, kn_selnext); KNOTE(&timer_list, 1); + + /* if someone is waiting for timer to pop */ + if (kn->kn_hookid & TIMER_CANCELWAIT) { + struct kqueue *kq = kn->kn_kq; + wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, &kn->kn_hook, + THREAD_AWAKENED); + } + + filt_timerunlock(); +} + +/* + * Cancel a running timer (or wait for the pop). + * Timer filter lock is held. + */ +static void +filt_timercancel(struct knote *kn) +{ + struct kqueue *kq = kn->kn_kq; + thread_call_t callout = kn->kn_hook; + boolean_t cancelled; + + if (kn->kn_hookid & TIMER_RUNNING) { + /* cancel the callout if we can */ + cancelled = thread_call_cancel(callout); + if (cancelled) { + kn->kn_hookid &= ~TIMER_RUNNING; + } else { + /* we have to wait for the expire routine. */ + kn->kn_hookid |= TIMER_CANCELWAIT; + wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, + &kn->kn_hook, THREAD_UNINT, 0); + filt_timerunlock(); + thread_block(THREAD_CONTINUE_NULL); + filt_timerlock(); + assert((kn->kn_hookid & TIMER_RUNNING) == 0); + } + } } /* - * data contains amount of time to sleep, in milliseconds, - * or a pointer to a timespec structure. - */ + * Allocate a thread call for the knote's lifetime, and kick off the timer. + */ static int filt_timerattach(struct knote *kn) { thread_call_t callout; - uint64_t deadline; int error; - error = filt_timercompute(kn, &deadline); - if (error) - return (error); + callout = thread_call_allocate(filt_timerexpire, kn); + if (NULL == callout) + return (ENOMEM); - if (deadline) { - callout = thread_call_allocate(filt_timerexpire, kn); - if (NULL == callout) - return (ENOMEM); - } else { - /* handle as immediate */ - kn->kn_sdata = 0; - callout = NULL; + filt_timerlock(); + error = filt_timervalidate(kn); + if (error != 0) { + filt_timerunlock(); + return (error); } - filt_timerlock(); - kn->kn_hook = (caddr_t)callout; + kn->kn_hook = (void*)callout; + kn->kn_hookid = 0; /* absolute=EV_ONESHOT */ if (kn->kn_sfflags & NOTE_ABSOLUTE) - kn->kn_flags |= EV_ONESHOT; + kn->kn_flags |= EV_ONESHOT; - if (deadline) { - /* all others - if not faking immediate */ + filt_timerupdate(kn); + if (kn->kn_ext[0]) { kn->kn_flags |= EV_CLEAR; - thread_call_enter_delayed(callout, deadline); - kn->kn_hookid = 0; + unsigned int timer_flags = 0; + if (kn->kn_sfflags & NOTE_CRITICAL) + timer_flags |= THREAD_CALL_DELAY_USER_CRITICAL; + else if (kn->kn_sfflags & NOTE_BACKGROUND) + timer_flags |= THREAD_CALL_DELAY_USER_BACKGROUND; + else + timer_flags |= THREAD_CALL_DELAY_USER_NORMAL; + + if (kn->kn_sfflags & NOTE_LEEWAY) + timer_flags |= THREAD_CALL_DELAY_LEEWAY; + + thread_call_enter_delayed_with_leeway(callout, NULL, + kn->kn_ext[0], kn->kn_ext[1], timer_flags); + + kn->kn_hookid |= TIMER_RUNNING; } else { /* fake immediate */ - kn->kn_hookid = 1; + kn->kn_data = 1; } + filt_timerunlock(); return (0); } +/* + * Shut down the timer if it's running, and free the callout. + */ static void filt_timerdetach(struct knote *kn) { thread_call_t callout; filt_timerlock(); + callout = (thread_call_t)kn->kn_hook; - if (callout != NULL) { - boolean_t cancelled; + filt_timercancel(kn); - /* cancel the callout if we can */ - cancelled = thread_call_cancel(callout); - if (cancelled) { - /* got it, just free it */ - kn->kn_hook = NULL; - filt_timerunlock(); - thread_call_free(callout); - return; - } - /* we have to wait for the expire routine. */ - kn->kn_hookid = -1; /* we are detaching */ - assert_wait(&kn->kn_hook, THREAD_UNINT); - filt_timerunlock(); - thread_block(THREAD_CONTINUE_NULL); - assert(kn->kn_hook == NULL); - return; - } - /* nothing to do */ - filt_timerunlock(); + filt_timerunlock(); + + thread_call_free(callout); } static int -filt_timer(struct knote *kn, __unused long hint) +filt_timer(struct knote *kn, long hint) { int result; - - if (hint) { - /* real timer pop */ - thread_call_t callout; - boolean_t detaching; - filt_timerlock(); - + if (hint) { + /* real timer pop -- timer lock held by filt_timerexpire */ kn->kn_data++; - detaching = (kn->kn_hookid < 0); - callout = (thread_call_t)kn->kn_hook; + if (((kn->kn_hookid & TIMER_CANCELWAIT) == 0) && + ((kn->kn_flags & EV_ONESHOT) == 0)) { - if (!detaching && (kn->kn_flags & EV_ONESHOT) == 0) { - uint64_t deadline; - int error; + /* evaluate next time to fire */ + filt_timerupdate(kn); + + if (kn->kn_ext[0]) { + unsigned int timer_flags = 0; - /* user input data may have changed - deal */ - error = filt_timercompute(kn, &deadline); - if (error) { - kn->kn_flags |= EV_ERROR; - kn->kn_data = error; - } else if (deadline == 0) { - /* revert to fake immediate */ - kn->kn_flags &= ~EV_CLEAR; - kn->kn_sdata = 0; - kn->kn_hookid = 1; - } else { /* keep the callout and re-arm */ - thread_call_enter_delayed(callout, deadline); - filt_timerunlock(); - return 1; + if (kn->kn_sfflags & NOTE_CRITICAL) + timer_flags |= THREAD_CALL_DELAY_USER_CRITICAL; + else if (kn->kn_sfflags & NOTE_BACKGROUND) + timer_flags |= THREAD_CALL_DELAY_USER_BACKGROUND; + else + timer_flags |= THREAD_CALL_DELAY_USER_NORMAL; + + if (kn->kn_sfflags & NOTE_LEEWAY) + timer_flags |= THREAD_CALL_DELAY_LEEWAY; + + thread_call_enter_delayed_with_leeway(kn->kn_hook, NULL, + kn->kn_ext[0], kn->kn_ext[1], timer_flags); + + kn->kn_hookid |= TIMER_RUNNING; } } - kn->kn_hook = NULL; - filt_timerunlock(); - thread_call_free(callout); - - /* if someone is waiting for timer to pop */ - if (detaching) - thread_wakeup(&kn->kn_hook); - return 1; - } + return (1); + } /* user-query */ filt_timerlock(); - /* change fake timer to real if needed */ - while (kn->kn_hookid > 0 && kn->kn_sdata > 0) { - int error; + result = (kn->kn_data != 0); - /* update the fake timer (make real) */ - kn->kn_hookid = 0; - kn->kn_data = 0; - filt_timerunlock(); - error = filt_timerattach(kn); - filt_timerlock(); + filt_timerunlock(); + + return (result); +} + + +/* + * filt_timertouch - update knote with new user input + * + * Cancel and restart the timer based on new user data. When + * the user picks up a knote, clear the count of how many timer + * pops have gone off (in kn_data). + */ +static void +filt_timertouch(struct knote *kn, struct kevent64_s *kev, long type) +{ + int error; + filt_timerlock(); + + switch (type) { + case EVENT_REGISTER: + /* cancel current call */ + filt_timercancel(kn); + + /* recalculate deadline */ + kn->kn_sdata = kev->data; + kn->kn_sfflags = kev->fflags; + kn->kn_ext[0] = kev->ext[0]; + kn->kn_ext[1] = kev->ext[1]; + + error = filt_timervalidate(kn); if (error) { + /* no way to report error, so mark it in the knote */ kn->kn_flags |= EV_ERROR; kn->kn_data = error; - filt_timerunlock(); - return 1; + break; } - } - /* if still fake, pretend it fired */ - if (kn->kn_hookid > 0) - kn->kn_data = 1; + /* start timer if necessary */ + filt_timerupdate(kn); + + if (kn->kn_ext[0]) { + unsigned int timer_flags = 0; + if (kn->kn_sfflags & NOTE_CRITICAL) + timer_flags |= THREAD_CALL_DELAY_USER_CRITICAL; + else if (kn->kn_sfflags & NOTE_BACKGROUND) + timer_flags |= THREAD_CALL_DELAY_USER_BACKGROUND; + else + timer_flags |= THREAD_CALL_DELAY_USER_NORMAL; + + if (kn->kn_sfflags & NOTE_LEEWAY) + timer_flags |= THREAD_CALL_DELAY_LEEWAY; + + thread_call_enter_delayed_with_leeway(kn->kn_hook, NULL, + kn->kn_ext[0], kn->kn_ext[1], timer_flags); + + kn->kn_hookid |= TIMER_RUNNING; + } else { + /* pretend the timer has fired */ + kn->kn_data = 1; + } + + break; + + case EVENT_PROCESS: + /* reset the timer pop count in kn_data */ + *kev = kn->kn_kevent; + kev->ext[0] = 0; + kn->kn_data = 0; + if (kn->kn_flags & EV_CLEAR) + kn->kn_fflags = 0; + break; + default: + panic("%s: - invalid type (%ld)", __func__, type); + break; + } - result = (kn->kn_data != 0); filt_timerunlock(); - return result; } static void @@ -730,42 +1084,118 @@ filt_timerunlock(void) lck_mtx_unlock(&_filt_timerlock); } +static int +filt_userattach(struct knote *kn) +{ + /* EVFILT_USER knotes are not attached to anything in the kernel */ + kn->kn_hook = NULL; + if (kn->kn_fflags & NOTE_TRIGGER) { + kn->kn_hookid = 1; + } else { + kn->kn_hookid = 0; + } + return (0); +} + +static void +filt_userdetach(__unused struct knote *kn) +{ + /* EVFILT_USER knotes are not attached to anything in the kernel */ +} + +static int +filt_user(struct knote *kn, __unused long hint) +{ + return (kn->kn_hookid); +} + +static void +filt_usertouch(struct knote *kn, struct kevent64_s *kev, long type) +{ + uint32_t ffctrl; + switch (type) { + case EVENT_REGISTER: + if (kev->fflags & NOTE_TRIGGER) { + kn->kn_hookid = 1; + } + + ffctrl = kev->fflags & NOTE_FFCTRLMASK; + kev->fflags &= NOTE_FFLAGSMASK; + switch (ffctrl) { + case NOTE_FFNOP: + break; + case NOTE_FFAND: + OSBitAndAtomic(kev->fflags, &kn->kn_sfflags); + break; + case NOTE_FFOR: + OSBitOrAtomic(kev->fflags, &kn->kn_sfflags); + break; + case NOTE_FFCOPY: + kn->kn_sfflags = kev->fflags; + break; + } + kn->kn_sdata = kev->data; + break; + case EVENT_PROCESS: + *kev = kn->kn_kevent; + kev->fflags = (volatile UInt32)kn->kn_sfflags; + kev->data = kn->kn_sdata; + if (kn->kn_flags & EV_CLEAR) { + kn->kn_hookid = 0; + kn->kn_data = 0; + kn->kn_fflags = 0; + } + break; + default: + panic("%s: - invalid type (%ld)", __func__, type); + break; + } +} + /* * JMM - placeholder for not-yet-implemented filters - */ + */ static int filt_badattach(__unused struct knote *kn) { - return(ENOTSUP); + return (ENOTSUP); } - struct kqueue * kqueue_alloc(struct proc *p) { struct filedesc *fdp = p->p_fd; struct kqueue *kq; - MALLOC_ZONE(kq, struct kqueue *, sizeof(struct kqueue), M_KQUEUE, M_WAITOK); + MALLOC_ZONE(kq, struct kqueue *, sizeof (struct kqueue), M_KQUEUE, + M_WAITOK); if (kq != NULL) { - bzero(kq, sizeof(struct kqueue)); - lck_spin_init(&kq->kq_lock, kq_lck_grp, kq_lck_attr); - TAILQ_INIT(&kq->kq_head); - TAILQ_INIT(&kq->kq_inprocess); - kq->kq_p = p; + wait_queue_set_t wqs; + + wqs = wait_queue_set_alloc(SYNC_POLICY_FIFO | + SYNC_POLICY_PREPOST); + if (wqs != NULL) { + bzero(kq, sizeof (struct kqueue)); + lck_spin_init(&kq->kq_lock, kq_lck_grp, kq_lck_attr); + TAILQ_INIT(&kq->kq_head); + kq->kq_wqs = wqs; + kq->kq_p = p; + } else { + FREE_ZONE(kq, sizeof (struct kqueue), M_KQUEUE); + kq = NULL; + } } if (fdp->fd_knlistsize < 0) { proc_fdlock(p); if (fdp->fd_knlistsize < 0) - fdp->fd_knlistsize = 0; /* this process has had a kq */ + fdp->fd_knlistsize = 0; /* this process has had a kq */ proc_fdunlock(p); } - return kq; + return (kq); } - /* * kqueue_dealloc - detach all knotes from a kqueue and free it * @@ -830,18 +1260,26 @@ kqueue_dealloc(struct kqueue *kq) } } proc_fdunlock(p); + + /* + * before freeing the wait queue set for this kqueue, + * make sure it is unlinked from all its containing (select) sets. + */ + wait_queue_unlink_all((wait_queue_t)kq->kq_wqs); + wait_queue_set_free(kq->kq_wqs); lck_spin_destroy(&kq->kq_lock, kq_lck_grp); - FREE_ZONE(kq, sizeof(struct kqueue), M_KQUEUE); + FREE_ZONE(kq, sizeof (struct kqueue), M_KQUEUE); } int -kqueue(struct proc *p, __unused struct kqueue_args *uap, register_t *retval) +kqueue_body(struct proc *p, fp_allocfn_t fp_zalloc, void *cra, int32_t *retval) { struct kqueue *kq; struct fileproc *fp; int fd, error; - error = falloc(p, &fp, &fd, vfs_context_current()); + error = falloc_withalloc(p, + &fp, &fd, vfs_context_current(), fp_zalloc, cra); if (error) { return (error); } @@ -853,11 +1291,11 @@ kqueue(struct proc *p, __unused struct kqueue_args *uap, register_t *retval) } fp->f_flag = FREAD | FWRITE; - fp->f_type = DTYPE_KQUEUE; fp->f_ops = &kqueueops; - fp->f_data = (caddr_t)kq; + fp->f_data = kq; proc_fdlock(p); + *fdflags(p, fd) |= UF_EXCLOSE; procfdtbl_releasefd(p, fd, NULL); fp_drop(p, fd, fp, 1); proc_fdunlock(p); @@ -867,89 +1305,97 @@ kqueue(struct proc *p, __unused struct kqueue_args *uap, register_t *retval) } int -kqueue_portset_np(__unused struct proc *p, - __unused struct kqueue_portset_np_args *uap, - __unused register_t *retval) -{ - /* JMM - Placeholder for now */ - return (ENOTSUP); -} - -int -kqueue_from_portset_np(__unused struct proc *p, - __unused struct kqueue_from_portset_np_args *uap, - __unused register_t *retval) +kqueue(struct proc *p, __unused struct kqueue_args *uap, int32_t *retval) { - /* JMM - Placeholder for now */ - return (ENOTSUP); + return (kqueue_body(p, fileproc_alloc_init, NULL, retval)); } static int -kevent_copyin(user_addr_t *addrp, struct kevent *kevp, struct proc *p) +kevent_copyin(user_addr_t *addrp, struct kevent64_s *kevp, struct proc *p, + int iskev64) { int advance; int error; - if (IS_64BIT_PROCESS(p)) { - struct user_kevent kev64; + if (iskev64) { + advance = sizeof (struct kevent64_s); + error = copyin(*addrp, (caddr_t)kevp, advance); + } else if (IS_64BIT_PROCESS(p)) { + struct user64_kevent kev64; + bzero(kevp, sizeof (struct kevent64_s)); - advance = sizeof(kev64); + advance = sizeof (kev64); error = copyin(*addrp, (caddr_t)&kev64, advance); if (error) - return error; - kevp->ident = CAST_DOWN(uintptr_t, kev64.ident); + return (error); + kevp->ident = kev64.ident; kevp->filter = kev64.filter; kevp->flags = kev64.flags; kevp->fflags = kev64.fflags; - kevp->data = CAST_DOWN(intptr_t, kev64.data); + kevp->data = kev64.data; kevp->udata = kev64.udata; } else { - /* - * compensate for legacy in-kernel kevent layout - * where the udata field is alredy 64-bit. - */ - advance = sizeof(*kevp) + sizeof(void *) - sizeof(user_addr_t); - error = copyin(*addrp, (caddr_t)kevp, advance); + struct user32_kevent kev32; + bzero(kevp, sizeof (struct kevent64_s)); + + advance = sizeof (kev32); + error = copyin(*addrp, (caddr_t)&kev32, advance); + if (error) + return (error); + kevp->ident = (uintptr_t)kev32.ident; + kevp->filter = kev32.filter; + kevp->flags = kev32.flags; + kevp->fflags = kev32.fflags; + kevp->data = (intptr_t)kev32.data; + kevp->udata = CAST_USER_ADDR_T(kev32.udata); } if (!error) *addrp += advance; - return error; + return (error); } static int -kevent_copyout(struct kevent *kevp, user_addr_t *addrp, struct proc *p) +kevent_copyout(struct kevent64_s *kevp, user_addr_t *addrp, struct proc *p, + int iskev64) { int advance; int error; - if (IS_64BIT_PROCESS(p)) { - struct user_kevent kev64; + if (iskev64) { + advance = sizeof (struct kevent64_s); + error = copyout((caddr_t)kevp, *addrp, advance); + } else if (IS_64BIT_PROCESS(p)) { + struct user64_kevent kev64; /* * deal with the special case of a user-supplied * value of (uintptr_t)-1. */ kev64.ident = (kevp->ident == (uintptr_t)-1) ? - (uint64_t)-1LL : (uint64_t)kevp->ident; + (uint64_t)-1LL : (uint64_t)kevp->ident; kev64.filter = kevp->filter; kev64.flags = kevp->flags; kev64.fflags = kevp->fflags; kev64.data = (int64_t) kevp->data; kev64.udata = kevp->udata; - advance = sizeof(kev64); + advance = sizeof (kev64); error = copyout((caddr_t)&kev64, *addrp, advance); } else { - /* - * compensate for legacy in-kernel kevent layout - * where the udata field is alredy 64-bit. - */ - advance = sizeof(*kevp) + sizeof(void *) - sizeof(user_addr_t); - error = copyout((caddr_t)kevp, *addrp, advance); + struct user32_kevent kev32; + + kev32.ident = (uint32_t)kevp->ident; + kev32.filter = kevp->filter; + kev32.flags = kevp->flags; + kev32.fflags = kevp->fflags; + kev32.data = (int32_t)kevp->data; + kev32.udata = kevp->udata; + advance = sizeof (kev32); + error = copyout((caddr_t)&kev32, *addrp, advance); } if (!error) *addrp += advance; - return error; + return (error); } /* @@ -963,7 +1409,7 @@ kevent_continue(__unused struct kqueue *kq, void *data, int error) { struct _kevent *cont_args; struct fileproc *fp; - register_t *retval; + int32_t *retval; int noutputs; int fd; struct proc *p = current_proc(); @@ -990,43 +1436,69 @@ kevent_continue(__unused struct kqueue *kq, void *data, int error) * kevent - [syscall] register and wait for kernel events * */ +int +kevent(struct proc *p, struct kevent_args *uap, int32_t *retval) +{ + return (kevent_internal(p, + 0, + uap->changelist, + uap->nchanges, + uap->eventlist, + uap->nevents, + uap->fd, + uap->timeout, + 0, /* no flags from old kevent() call */ + retval)); +} int -kevent(struct proc *p, struct kevent_args *uap, register_t *retval) -{ - user_addr_t changelist = uap->changelist; - user_addr_t ueventlist = uap->eventlist; - int nchanges = uap->nchanges; - int nevents = uap->nevents; - int fd = uap->fd; +kevent64(struct proc *p, struct kevent64_args *uap, int32_t *retval) +{ + return (kevent_internal(p, + 1, + uap->changelist, + uap->nchanges, + uap->eventlist, + uap->nevents, + uap->fd, + uap->timeout, + uap->flags, + retval)); +} +static int +kevent_internal(struct proc *p, int iskev64, user_addr_t changelist, + int nchanges, user_addr_t ueventlist, int nevents, int fd, + user_addr_t utimeout, __unused unsigned int flags, + int32_t *retval) +{ struct _kevent *cont_args; uthread_t ut; struct kqueue *kq; struct fileproc *fp; - struct kevent kev; + struct kevent64_s kev; int error, noutputs; struct timeval atv; /* convert timeout to absolute - if we have one */ - if (uap->timeout != USER_ADDR_NULL) { + if (utimeout != USER_ADDR_NULL) { struct timeval rtv; - if ( IS_64BIT_PROCESS(p) ) { - struct user_timespec ts; - error = copyin( uap->timeout, &ts, sizeof(ts) ); + if (IS_64BIT_PROCESS(p)) { + struct user64_timespec ts; + error = copyin(utimeout, &ts, sizeof(ts)); if ((ts.tv_sec & 0xFFFFFFFF00000000ull) != 0) error = EINVAL; else TIMESPEC_TO_TIMEVAL(&rtv, &ts); } else { - struct timespec ts; - error = copyin( uap->timeout, &ts, sizeof(ts) ); + struct user32_timespec ts; + error = copyin(utimeout, &ts, sizeof(ts)); TIMESPEC_TO_TIMEVAL(&rtv, &ts); } if (error) - return error; + return (error); if (itimerfix(&rtv)) - return EINVAL; + return (EINVAL); getmicrouptime(&atv); timevaladd(&atv, &rtv); } else { @@ -1036,21 +1508,35 @@ kevent(struct proc *p, struct kevent_args *uap, register_t *retval) /* get a usecount for the kq itself */ if ((error = fp_getfkq(p, fd, &fp, &kq)) != 0) - return(error); + return (error); + + /* each kq should only be used for events of one type */ + kqlock(kq); + if (kq->kq_state & (KQ_KEV32 | KQ_KEV64)) { + if (((iskev64 && (kq->kq_state & KQ_KEV32)) || + (!iskev64 && (kq->kq_state & KQ_KEV64)))) { + error = EINVAL; + kqunlock(kq); + goto errorout; + } + } else { + kq->kq_state |= (iskev64 ? KQ_KEV64 : KQ_KEV32); + } + kqunlock(kq); /* register all the change requests the user provided... */ noutputs = 0; while (nchanges > 0 && error == 0) { - error = kevent_copyin(&changelist, &kev, p); + error = kevent_copyin(&changelist, &kev, p, iskev64); if (error) break; - + kev.flags &= ~EV_SYSFLAGS; error = kevent_register(kq, &kev, p); if ((error || (kev.flags & EV_RECEIPT)) && nevents > 0) { kev.flags = EV_ERROR; kev.data = error; - error = kevent_copyout(&kev, &ueventlist, p); + error = kevent_copyout(&kev, &ueventlist, p, iskev64); if (error == 0) { nevents--; noutputs++; @@ -1061,52 +1547,83 @@ kevent(struct proc *p, struct kevent_args *uap, register_t *retval) /* store the continuation/completion data in the uthread */ ut = (uthread_t)get_bsdthread_info(current_thread()); - cont_args = (struct _kevent *)&ut->uu_kevent.ss_kevent; + cont_args = &ut->uu_kevent.ss_kevent; cont_args->fp = fp; cont_args->fd = fd; cont_args->retval = retval; cont_args->eventlist = ueventlist; cont_args->eventcount = nevents; cont_args->eventout = noutputs; + cont_args->eventsize = iskev64; if (nevents > 0 && noutputs == 0 && error == 0) - error = kevent_scan(kq, kevent_callback, - kevent_continue, cont_args, - &atv, p); + error = kqueue_scan(kq, kevent_callback, + kevent_continue, cont_args, + &atv, p); kevent_continue(kq, cont_args, error); - /* NOTREACHED */ - return error; + +errorout: + fp_drop(p, fd, fp, 0); + return (error); } /* * kevent_callback - callback for each individual event * - * called with nothing locked - * caller holds a reference on the kqueue + * called with nothing locked + * caller holds a reference on the kqueue */ - static int -kevent_callback(__unused struct kqueue *kq, struct kevent *kevp, void *data) +kevent_callback(__unused struct kqueue *kq, struct kevent64_s *kevp, + void *data) { struct _kevent *cont_args; int error; + int iskev64; cont_args = (struct _kevent *)data; assert(cont_args->eventout < cont_args->eventcount); + iskev64 = cont_args->eventsize; + /* * Copy out the appropriate amount of event data for this user. */ - error = kevent_copyout(kevp, &cont_args->eventlist, current_proc()); + error = kevent_copyout(kevp, &cont_args->eventlist, current_proc(), + iskev64); /* * If there isn't space for additional events, return * a harmless error to stop the processing here */ if (error == 0 && ++cont_args->eventout == cont_args->eventcount) - error = EWOULDBLOCK; - return error; + error = EWOULDBLOCK; + return (error); +} + +/* + * kevent_description - format a description of a kevent for diagnostic output + * + * called with a 128-byte string buffer + */ + +char * +kevent_description(struct kevent64_s *kevp, char *s, size_t n) +{ + snprintf(s, n, + "kevent=" + "{.ident=%#llx, .filter=%d, .flags=%#x, .fflags=%#x, .data=%#llx, .udata=%#llx, .ext[0]=%#llx, .ext[1]=%#llx}", + kevp->ident, + kevp->filter, + kevp->flags, + kevp->fflags, + kevp->data, + kevp->udata, + kevp->ext[0], + kevp->ext[1]); + + return (s); } /* @@ -1124,7 +1641,8 @@ kevent_callback(__unused struct kqueue *kq, struct kevent *kevp, void *data) */ int -kevent_register(struct kqueue *kq, struct kevent *kev, __unused struct proc *ctxp) +kevent_register(struct kqueue *kq, struct kevent64_s *kev, + __unused struct proc *ctxp) { struct proc *p = kq->kq_p; struct filedesc *fdp = p->p_fd; @@ -1147,12 +1665,14 @@ kevent_register(struct kqueue *kq, struct kevent *kev, __unused struct proc *ctx return (EINVAL); } +restart: /* this iocount needs to be dropped if it is not registered */ - if (fops->f_isfd && (error = fp_lookup(p, kev->ident, &fp, 0)) != 0) - return(error); - - restart: proc_fdlock(p); + if (fops->f_isfd && (error = fp_lookup(p, kev->ident, &fp, 1)) != 0) { + proc_fdunlock(p); + return (error); + } + if (fops->f_isfd) { /* fd-based knotes are linked off the fd table */ if (kev->ident < (u_int)fdp->fd_knlistsize) { @@ -1165,7 +1685,7 @@ kevent_register(struct kqueue *kq, struct kevent *kev, __unused struct proc *ctx /* hash non-fd knotes here too */ if (fdp->fd_knhashmask != 0) { struct klist *list; - + list = &fdp->fd_knhash[ KN_HASH((u_long)kev->ident, fdp->fd_knhashmask)]; SLIST_FOREACH(kn, list, kn_link) @@ -1197,7 +1717,7 @@ kevent_register(struct kqueue *kq, struct kevent *kev, __unused struct proc *ctx kev->data = 0; kn->kn_kevent = *kev; kn->kn_inuse = 1; /* for f_attach() */ - kn->kn_status = 0; + kn->kn_status = KN_ATTACHING; /* before anyone can find it */ if (kev->flags & EV_DISABLE) @@ -1217,14 +1737,33 @@ kevent_register(struct kqueue *kq, struct kevent *kev, __unused struct proc *ctx */ fp = NULL; - /* - * If the attach fails here, we can drop it knowing - * that nobody else has a reference to the knote. - */ - if ((error = fops->f_attach(kn)) != 0) { + error = fops->f_attach(kn); + + kqlock(kq); + + if (error != 0) { + /* + * Failed to attach correctly, so drop. + * All other possible users/droppers + * have deferred to us. + */ + kn->kn_status |= KN_DROPPING; + kqunlock(kq); + knote_drop(kn, p); + goto done; + } else if (kn->kn_status & KN_DROPPING) { + /* + * Attach succeeded, but someone else + * deferred their drop - now we have + * to do it for them (after detaching). + */ + kqunlock(kq); + kn->kn_fop->f_detach(kn); knote_drop(kn, p); goto done; } + kn->kn_status &= ~KN_ATTACHING; + kqunlock(kq); } else { proc_fdunlock(p); error = ENOENT; @@ -1234,7 +1773,7 @@ kevent_register(struct kqueue *kq, struct kevent *kev, __unused struct proc *ctx /* existing knote - get kqueue lock */ kqlock(kq); proc_fdunlock(p); - + if (kev->flags & EV_DELETE) { knote_dequeue(kn); kn->kn_status |= KN_DISABLED; @@ -1244,7 +1783,7 @@ kevent_register(struct kqueue *kq, struct kevent *kev, __unused struct proc *ctx } goto done; } - + /* update status flags for existing knote */ if (kev->flags & EV_DISABLE) { knote_dequeue(kn); @@ -1255,29 +1794,47 @@ kevent_register(struct kqueue *kq, struct kevent *kev, __unused struct proc *ctx knote_enqueue(kn); } + /* + * The user may change some filter values after the + * initial EV_ADD, but doing so will not reset any + * filter which have already been triggered. + */ + kn->kn_kevent.udata = kev->udata; + if (fops->f_isfd || fops->f_touch == NULL) { + kn->kn_sfflags = kev->fflags; + kn->kn_sdata = kev->data; + } + /* * If somebody is in the middle of dropping this * knote - go find/insert a new one. But we have - * wait for this one to go away first. + * wait for this one to go away first. Attaches + * running in parallel may also drop/modify the + * knote. Wait for those to complete as well and + * then start over if we encounter one. */ - if (!kqlock2knoteusewait(kq, kn)) - /* kqueue unlocked */ + if (!kqlock2knoteusewait(kq, kn)) { + /* kqueue, proc_fdlock both unlocked */ goto restart; + } /* - * The user may change some filter values after the - * initial EV_ADD, but doing so will not reset any - * filter which have already been triggered. + * Call touch routine to notify filter of changes + * in filter values. */ - kn->kn_sfflags = kev->fflags; - kn->kn_sdata = kev->data; - kn->kn_kevent.udata = kev->udata; + if (!fops->f_isfd && fops->f_touch != NULL) + fops->f_touch(kn, kev, EVENT_REGISTER); } - /* still have use ref on knote */ - if (kn->kn_fop->f_event(kn, 0)) { + + /* + * If the knote is not marked to always stay enqueued, + * invoke the filter routine to see if it should be + * enqueued now. + */ + if ((kn->kn_status & KN_STAYQUEUED) == 0 && kn->kn_fop->f_event(kn, 0)) { if (knoteuse2kqlock(kq, kn)) - knote_activate(kn); + knote_activate(kn, 1); kqunlock(kq); } else { knote_put(kn); @@ -1289,156 +1846,297 @@ done: return (error); } + /* - * kevent_process - process the triggered events in a kqueue + * knote_process - process a triggered event * - * Walk the queued knotes and validate that they are - * really still triggered events by calling the filter - * routines (if necessary). Hold a use reference on - * the knote to avoid it being detached. For each event - * that is still considered triggered, invoke the - * callback routine provided. + * Validate that it is really still a triggered event + * by calling the filter routines (if necessary). Hold + * a use reference on the knote to avoid it being detached. + * If it is still considered triggered, invoke the callback + * routine provided and move it to the provided inprocess + * queue. * * caller holds a reference on the kqueue. * kqueue locked on entry and exit - but may be dropped */ - static int -kevent_process(struct kqueue *kq, - kevent_callback_t callback, - void *data, - int *countp, - struct proc *p) +knote_process(struct knote *kn, + kevent_callback_t callback, + void *data, + struct kqtailq *inprocessp, + struct proc *p) { - struct knote *kn; - struct kevent kev; - int nevents; + struct kqueue *kq = kn->kn_kq; + struct kevent64_s kev; + int touch; + int result; int error; - restart: - if (kq->kq_count == 0) { - *countp = 0; - return 0; - } - - /* if someone else is processing the queue, wait */ - if (!TAILQ_EMPTY(&kq->kq_inprocess)) { - assert_wait(&kq->kq_inprocess, THREAD_UNINT); - kq->kq_state |= KQ_PROCWAIT; - kqunlock(kq); - thread_block(THREAD_CONTINUE_NULL); - kqlock(kq); - goto restart; - } - - error = 0; - nevents = 0; - while (error == 0 && - (kn = TAILQ_FIRST(&kq->kq_head)) != NULL) { + /* + * Determine the kevent state we want to return. + * + * Some event states need to be revalidated before returning + * them, others we take the snapshot at the time the event + * was enqueued. + * + * Events with non-NULL f_touch operations must be touched. + * Triggered events must fill in kev for the callback. + * + * Convert our lock to a use-count and call the event's + * filter routine(s) to update. + */ + if ((kn->kn_status & KN_DISABLED) != 0) { + result = 0; + touch = 0; + } else { + int revalidate; - /* - * Take note off the active queue. - * - * Non-EV_ONESHOT events must be re-validated. - * - * Convert our lock to a use-count and call the event's - * filter routine to update. - * - * If the event is valid, or triggered while the kq - * is unlocked, move to the inprocess queue for processing. - */ + result = 1; + revalidate = ((kn->kn_status & KN_STAYQUEUED) != 0 || + (kn->kn_flags & EV_ONESHOT) == 0); + touch = (!kn->kn_fop->f_isfd && kn->kn_fop->f_touch != NULL); - if ((kn->kn_flags & EV_ONESHOT) == 0) { - int result; - knote_deactivate(kn); + if (revalidate || touch) { + if (revalidate) + knote_deactivate(kn); + /* call the filter/touch routines with just a ref */ if (kqlock2knoteuse(kq, kn)) { - - /* call the filter with just a ref */ - result = kn->kn_fop->f_event(kn, 0); + /* if we have to revalidate, call the filter */ + if (revalidate) { + result = kn->kn_fop->f_event(kn, 0); + } + + /* + * capture the kevent data - using touch if + * specified + */ + if (result && touch) { + kn->kn_fop->f_touch(kn, &kev, + EVENT_PROCESS); + } - /* if it's still alive, make sure it's active */ - if (knoteuse2kqlock(kq, kn) && result) { - /* may have been reactivated in filter*/ + /* + * convert back to a kqlock - bail if the knote + * went away + */ + if (!knoteuse2kqlock(kq, kn)) { + return (EJUSTRETURN); + } else if (result) { + /* + * if revalidated as alive, make sure + * it's active + */ if (!(kn->kn_status & KN_ACTIVE)) { - knote_activate(kn); + knote_activate(kn, 0); } - } else { - continue; + + /* + * capture all events that occurred + * during filter + */ + if (!touch) { + kev = kn->kn_kevent; + } + + } else if ((kn->kn_status & KN_STAYQUEUED) == 0) { + /* + * was already dequeued, so just bail on + * this one + */ + return (EJUSTRETURN); } } else { - continue; + return (EJUSTRETURN); } + } else { + kev = kn->kn_kevent; } + } - /* knote is active: move onto inprocess queue */ - assert(kn->kn_tq == &kq->kq_head); - TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); - kn->kn_tq = &kq->kq_inprocess; - TAILQ_INSERT_TAIL(&kq->kq_inprocess, kn, kn_tqe); + /* move knote onto inprocess queue */ + assert(kn->kn_tq == &kq->kq_head); + TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); + kn->kn_tq = inprocessp; + TAILQ_INSERT_TAIL(inprocessp, kn, kn_tqe); - /* - * Got a valid triggered knote with the kqueue - * still locked. Snapshot the data, and determine - * how to dispatch the knote for future events. - */ - kev = kn->kn_kevent; + /* + * Determine how to dispatch the knote for future event handling. + * not-fired: just return (do not callout). + * One-shot: deactivate it. + * Clear: deactivate and clear the state. + * Dispatch: don't clear state, just deactivate it and mark it disabled. + * All others: just leave where they are. + */ - /* now what happens to it? */ - if (kn->kn_flags & EV_ONESHOT) { + if (result == 0) { + return (EJUSTRETURN); + } else if ((kn->kn_flags & EV_ONESHOT) != 0) { + knote_deactivate(kn); + if (kqlock2knotedrop(kq, kn)) { + kn->kn_fop->f_detach(kn); + knote_drop(kn, p); + } + } else if ((kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) != 0) { + if ((kn->kn_flags & EV_DISPATCH) != 0) { + /* deactivate and disable all dispatch knotes */ knote_deactivate(kn); - if (kqlock2knotedrop(kq, kn)) { - kn->kn_fop->f_detach(kn); - knote_drop(kn, p); - } - } else if (kn->kn_flags & EV_CLEAR) { + kn->kn_status |= KN_DISABLED; + } else if (!touch || kn->kn_fflags == 0) { + /* only deactivate if nothing since the touch */ knote_deactivate(kn); + } + if (!touch && (kn->kn_flags & EV_CLEAR) != 0) { + /* manually clear non-touch knotes */ kn->kn_data = 0; kn->kn_fflags = 0; + } + kqunlock(kq); + } else { + /* + * leave on inprocess queue. We'll + * move all the remaining ones back + * the kq queue and wakeup any + * waiters when we are done. + */ + kqunlock(kq); + } + + /* callback to handle each event as we find it */ + error = (callback)(kq, &kev, data); + + kqlock(kq); + return (error); +} + +/* + * Return 0 to indicate that processing should proceed, + * -1 if there is nothing to process. + * + * Called with kqueue locked and returns the same way, + * but may drop lock temporarily. + */ +static int +kqueue_begin_processing(struct kqueue *kq) +{ + for (;;) { + if (kq->kq_count == 0) { + return (-1); + } + + /* if someone else is processing the queue, wait */ + if (kq->kq_nprocess != 0) { + wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, + &kq->kq_nprocess, THREAD_UNINT, 0); + kq->kq_state |= KQ_PROCWAIT; kqunlock(kq); + thread_block(THREAD_CONTINUE_NULL); + kqlock(kq); } else { - /* - * leave on in-process queue. We'll - * move all the remaining ones back - * the kq queue and wakeup any - * waiters when we are done. - */ - kqunlock(kq); + kq->kq_nprocess = 1; + return (0); } + } +} - /* callback to handle each event as we find it */ - error = (callback)(kq, &kev, data); - nevents++; +/* + * Called with kqueue lock held. + */ +static void +kqueue_end_processing(struct kqueue *kq) +{ + kq->kq_nprocess = 0; + if (kq->kq_state & KQ_PROCWAIT) { + kq->kq_state &= ~KQ_PROCWAIT; + wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, + &kq->kq_nprocess, THREAD_AWAKENED); + } +} - kqlock(kq); +/* + * kqueue_process - process the triggered events in a kqueue + * + * Walk the queued knotes and validate that they are + * really still triggered events by calling the filter + * routines (if necessary). Hold a use reference on + * the knote to avoid it being detached. For each event + * that is still considered triggered, invoke the + * callback routine provided. + * + * caller holds a reference on the kqueue. + * kqueue locked on entry and exit - but may be dropped + * kqueue list locked (held for duration of call) + */ + +static int +kqueue_process(struct kqueue *kq, + kevent_callback_t callback, + void *data, + int *countp, + struct proc *p) +{ + struct kqtailq inprocess; + struct knote *kn; + int nevents; + int error; + + TAILQ_INIT(&inprocess); + + if (kqueue_begin_processing(kq) == -1) { + *countp = 0; + /* Nothing to process */ + return (0); + } + + /* + * Clear any pre-posted status from previous runs, so we + * only detect events that occur during this run. + */ + wait_queue_sub_clearrefs(kq->kq_wqs); + + /* + * loop through the enqueued knotes, processing each one and + * revalidating those that need it. As they are processed, + * they get moved to the inprocess queue (so the loop can end). + */ + error = 0; + nevents = 0; + + while (error == 0 && + (kn = TAILQ_FIRST(&kq->kq_head)) != NULL) { + error = knote_process(kn, callback, data, &inprocess, p); + if (error == EJUSTRETURN) + error = 0; + else + nevents++; } /* * With the kqueue still locked, move any knotes - * remaining on the in-process queue back to the + * remaining on the inprocess queue back to the * kq's queue and wake up any waiters. */ - while ((kn = TAILQ_FIRST(&kq->kq_inprocess)) != NULL) { - assert(kn->kn_tq == &kq->kq_inprocess); - TAILQ_REMOVE(&kq->kq_inprocess, kn, kn_tqe); + while ((kn = TAILQ_FIRST(&inprocess)) != NULL) { + assert(kn->kn_tq == &inprocess); + TAILQ_REMOVE(&inprocess, kn, kn_tqe); kn->kn_tq = &kq->kq_head; TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); } - if (kq->kq_state & KQ_PROCWAIT) { - kq->kq_state &= ~KQ_PROCWAIT; - thread_wakeup(&kq->kq_inprocess); - } + + kqueue_end_processing(kq); *countp = nevents; - return error; + return (error); } static void -kevent_scan_continue(void *data, wait_result_t wait_result) +kqueue_scan_continue(void *data, wait_result_t wait_result) { - uthread_t ut = (uthread_t)get_bsdthread_info(current_thread()); - struct _kevent_scan * cont_args = &ut->uu_kevent.ss_kevent_scan; + thread_t self = current_thread(); + uthread_t ut = (uthread_t)get_bsdthread_info(self); + struct _kqueue_scan * cont_args = &ut->uu_kevent.ss_kqueue_scan; struct kqueue *kq = (struct kqueue *)data; int error; int count; @@ -1447,27 +2145,30 @@ kevent_scan_continue(void *data, wait_result_t wait_result) switch (wait_result) { case THREAD_AWAKENED: kqlock(kq); - error = kevent_process(kq, cont_args->call, cont_args, &count, current_proc()); + error = kqueue_process(kq, cont_args->call, cont_args, &count, + current_proc()); if (error == 0 && count == 0) { - assert_wait_deadline(kq, THREAD_ABORTSAFE, cont_args->deadline); + wait_queue_assert_wait((wait_queue_t)kq->kq_wqs, + KQ_EVENT, THREAD_ABORTSAFE, cont_args->deadline); kq->kq_state |= KQ_SLEEP; kqunlock(kq); - thread_block_parameter(kevent_scan_continue, kq); + thread_block_parameter(kqueue_scan_continue, kq); /* NOTREACHED */ } kqunlock(kq); break; case THREAD_TIMED_OUT: - error = EWOULDBLOCK; + error = EWOULDBLOCK; break; case THREAD_INTERRUPTED: error = EINTR; break; default: - panic("kevent_scan_cont() - invalid wait_result (%d)", wait_result); + panic("%s: - invalid wait_result (%d)", __func__, + wait_result); error = 0; } - + /* call the continuation with the results */ assert(cont_args->cont != NULL); (cont_args->cont)(kq, cont_args->data, error); @@ -1475,7 +2176,7 @@ kevent_scan_continue(void *data, wait_result_t wait_result) /* - * kevent_scan - scan and wait for events in a kqueue + * kqueue_scan - scan and wait for events in a kqueue * * Process the triggered events in a kqueue. * @@ -1489,9 +2190,9 @@ kevent_scan_continue(void *data, wait_result_t wait_result) */ int -kevent_scan(struct kqueue *kq, +kqueue_scan(struct kqueue *kq, kevent_callback_t callback, - kevent_continue_t continuation, + kqueue_continue_t continuation, void *data, struct timeval *atvp, struct proc *p) @@ -1513,7 +2214,7 @@ kevent_scan(struct kqueue *kq, * triggered. */ kqlock(kq); - error = kevent_process(kq, callback, data, &count, p); + error = kqueue_process(kq, callback, data, &count, p); if (error || count) break; /* lock still held */ @@ -1523,10 +2224,10 @@ kevent_scan(struct kqueue *kq, /* convert the timeout to a deadline once */ if (atvp->tv_sec || atvp->tv_usec) { uint64_t now; - + clock_get_uptime(&now); nanoseconds_to_absolutetime((uint64_t)atvp->tv_sec * NSEC_PER_SEC + - atvp->tv_usec * NSEC_PER_USEC, + atvp->tv_usec * (long)NSEC_PER_USEC, &deadline); if (now >= deadline) { /* non-blocking call */ @@ -1541,18 +2242,20 @@ kevent_scan(struct kqueue *kq, if (continuation) { uthread_t ut = (uthread_t)get_bsdthread_info(current_thread()); - struct _kevent_scan *cont_args = &ut->uu_kevent.ss_kevent_scan; - + struct _kqueue_scan *cont_args = &ut->uu_kevent.ss_kqueue_scan; + cont_args->call = callback; cont_args->cont = continuation; cont_args->deadline = deadline; cont_args->data = data; - cont = kevent_scan_continue; + cont = kqueue_scan_continue; } } /* go ahead and wait */ - assert_wait_deadline(kq, THREAD_ABORTSAFE, deadline); + wait_queue_assert_wait_with_leeway((wait_queue_t)kq->kq_wqs, + KQ_EVENT, THREAD_ABORTSAFE, TIMEOUT_URGENCY_USER_NORMAL, + deadline, 0); kq->kq_state |= KQ_SLEEP; kqunlock(kq); wait_result = thread_block_parameter(cont, kq); @@ -1562,17 +2265,17 @@ kevent_scan(struct kqueue *kq, case THREAD_AWAKENED: continue; case THREAD_TIMED_OUT: - return EWOULDBLOCK; + return (EWOULDBLOCK); case THREAD_INTERRUPTED: - return EINTR; + return (EINTR); default: - panic("kevent_scan - bad wait_result (%d)", - wait_result); + panic("%s: - bad wait_result (%d)", __func__, + wait_result); error = 0; } } kqunlock(kq); - return error; + return (error); } @@ -1582,51 +2285,115 @@ kevent_scan(struct kqueue *kq, */ /*ARGSUSED*/ static int -kqueue_read(__unused struct fileproc *fp, - __unused struct uio *uio, - __unused int flags, - __unused vfs_context_t ctx) +kqueue_read(__unused struct fileproc *fp, + __unused struct uio *uio, + __unused int flags, + __unused vfs_context_t ctx) { return (ENXIO); } /*ARGSUSED*/ static int -kqueue_write(__unused struct fileproc *fp, - __unused struct uio *uio, - __unused int flags, - __unused vfs_context_t ctx) +kqueue_write(__unused struct fileproc *fp, + __unused struct uio *uio, + __unused int flags, + __unused vfs_context_t ctx) { return (ENXIO); } /*ARGSUSED*/ static int -kqueue_ioctl(__unused struct fileproc *fp, - __unused u_long com, - __unused caddr_t data, - __unused vfs_context_t ctx) +kqueue_ioctl(__unused struct fileproc *fp, + __unused u_long com, + __unused caddr_t data, + __unused vfs_context_t ctx) { return (ENOTTY); } /*ARGSUSED*/ static int -kqueue_select(struct fileproc *fp, int which, void *wql, vfs_context_t ctx) +kqueue_select(struct fileproc *fp, int which, void *wql, + __unused vfs_context_t ctx) { struct kqueue *kq = (struct kqueue *)fp->f_data; + struct knote *kn; + struct kqtailq inprocessq; int retnum = 0; - if (which == FREAD) { - kqlock(kq); - if (kq->kq_count) { - retnum = 1; - } else { - selrecord(vfs_context_proc(ctx), &kq->kq_sel, wql); - kq->kq_state |= KQ_SEL; + if (which != FREAD) + return (0); + + TAILQ_INIT(&inprocessq); + + kqlock(kq); + /* + * If this is the first pass, link the wait queue associated with the + * the kqueue onto the wait queue set for the select(). Normally we + * use selrecord() for this, but it uses the wait queue within the + * selinfo structure and we need to use the main one for the kqueue to + * catch events from KN_STAYQUEUED sources. So we do the linkage manually. + * (The select() call will unlink them when it ends). + */ + if (wql != NULL) { + thread_t cur_act = current_thread(); + struct uthread * ut = get_bsdthread_info(cur_act); + + kq->kq_state |= KQ_SEL; + wait_queue_link_noalloc((wait_queue_t)kq->kq_wqs, ut->uu_wqset, + (wait_queue_link_t)wql); + } + + if (kqueue_begin_processing(kq) == -1) { + kqunlock(kq); + return (0); + } + + if (kq->kq_count != 0) { + /* + * there is something queued - but it might be a + * KN_STAYQUEUED knote, which may or may not have + * any events pending. So, we have to walk the + * list of knotes to see, and peek at the stay- + * queued ones to be really sure. + */ + while ((kn = (struct knote *)TAILQ_FIRST(&kq->kq_head)) != NULL) { + if ((kn->kn_status & KN_STAYQUEUED) == 0) { + retnum = 1; + goto out; + } + + TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe); + TAILQ_INSERT_TAIL(&inprocessq, kn, kn_tqe); + + if (kqlock2knoteuse(kq, kn)) { + unsigned peek; + + peek = kn->kn_fop->f_peek(kn); + if (knoteuse2kqlock(kq, kn)) { + if (peek > 0) { + retnum = 1; + goto out; + } + } else { + retnum = 0; + } + } } - kqunlock(kq); } + +out: + /* Return knotes to active queue */ + while ((kn = TAILQ_FIRST(&inprocessq)) != NULL) { + TAILQ_REMOVE(&inprocessq, kn, kn_tqe); + kn->kn_tq = &kq->kq_head; + TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe); + } + + kqueue_end_processing(kq); + kqunlock(kq); return (retnum); } @@ -1672,7 +2439,7 @@ kqueue_kqfilter(__unused struct fileproc *fp, struct knote *kn, __unused vfs_con */ kqlock(parentkq); - if (parentkq->kq_level > 0 && + if (parentkq->kq_level > 0 && parentkq->kq_level < kq->kq_level) { kqunlock(parentkq); @@ -1696,28 +2463,47 @@ kqueue_kqfilter(__unused struct fileproc *fp, struct knote *kn, __unused vfs_con } } +/* + * kqueue_drain - called when kq is closed + */ /*ARGSUSED*/ -int -kqueue_stat(struct fileproc *fp, void *ub, int isstat64, __unused vfs_context_t ctx) +static int +kqueue_drain(struct fileproc *fp, __unused vfs_context_t ctx) { - struct stat *sb = (struct stat *)0; /* warning avoidance ; protected by isstat64 */ - struct stat64 * sb64 = (struct stat64 *)0; /* warning avoidance ; protected by isstat64 */ + struct kqueue *kq = (struct kqueue *)fp->f_fglob->fg_data; + kqlock(kq); + kqueue_wakeup(kq, 1); + kqunlock(kq); + return (0); +} - struct kqueue *kq = (struct kqueue *)fp->f_data; +/*ARGSUSED*/ +int +kqueue_stat(struct kqueue *kq, void *ub, int isstat64, proc_t p) +{ + kqlock(kq); if (isstat64 != 0) { - sb64 = (struct stat64 *)ub; + struct stat64 *sb64 = (struct stat64 *)ub; + bzero((void *)sb64, sizeof(*sb64)); sb64->st_size = kq->kq_count; - sb64->st_blksize = sizeof(struct kevent); + if (kq->kq_state & KQ_KEV64) + sb64->st_blksize = sizeof(struct kevent64_s); + else + sb64->st_blksize = IS_64BIT_PROCESS(p) ? sizeof(struct user64_kevent) : sizeof(struct user32_kevent); sb64->st_mode = S_IFIFO; } else { - sb = (struct stat *)ub; + struct stat *sb = (struct stat *)ub; + bzero((void *)sb, sizeof(*sb)); sb->st_size = kq->kq_count; - sb->st_blksize = sizeof(struct kevent); + if (kq->kq_state & KQ_KEV64) + sb->st_blksize = sizeof(struct kevent64_s); + else + sb->st_blksize = IS_64BIT_PROCESS(p) ? sizeof(struct user64_kevent) : sizeof(struct user32_kevent); sb->st_mode = S_IFIFO; } - + kqunlock(kq); return (0); } @@ -1725,18 +2511,13 @@ kqueue_stat(struct fileproc *fp, void *ub, int isstat64, __unused vfs_context_t * Called with the kqueue locked */ static void -kqueue_wakeup(struct kqueue *kq) +kqueue_wakeup(struct kqueue *kq, int closed) { - - if (kq->kq_state & KQ_SLEEP) { - kq->kq_state &= ~KQ_SLEEP; - thread_wakeup(kq); - } - if (kq->kq_state & KQ_SEL) { - kq->kq_state &= ~KQ_SEL; - selwakeup(&kq->kq_sel); + if ((kq->kq_state & (KQ_SLEEP | KQ_SEL)) != 0 || kq->kq_nprocess > 0) { + kq->kq_state &= ~(KQ_SLEEP | KQ_SEL); + wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, KQ_EVENT, + (closed) ? THREAD_INTERRUPTED : THREAD_AWAKENED); } - KNOTE(&kq->kq_sel.si_note, 0); } void @@ -1752,7 +2533,7 @@ klist_init(struct klist *list) * The object lock protects the list. It is assumed * that the filter/event routine for the object can * determine that the object is already locked (via - * the hind) and not deadlock itself. + * the hint) and not deadlock itself. * * The object lock should also hold off pending * detach/drop operations. But we'll prevent it here @@ -1775,7 +2556,7 @@ knote(struct klist *list, long hint) /* if its not going away and triggered */ if (knoteuse2kqlock(kq, kn) && result) - knote_activate(kn); + knote_activate(kn, 1); /* lock held again */ } kqunlock(kq); @@ -1791,7 +2572,7 @@ knote_attach(struct klist *list, struct knote *kn) { int ret = SLIST_EMPTY(list); SLIST_INSERT_HEAD(list, kn, kn_selnext); - return ret; + return (ret); } /* @@ -1802,7 +2583,52 @@ int knote_detach(struct klist *list, struct knote *kn) { SLIST_REMOVE(list, kn, knote, kn_selnext); - return SLIST_EMPTY(list); + return (SLIST_EMPTY(list)); +} + +/* + * For a given knote, link a provided wait queue directly with the kqueue. + * Wakeups will happen via recursive wait queue support. But nothing will move + * the knote to the active list at wakeup (nothing calls knote()). Instead, + * we permanently enqueue them here. + * + * kqueue and knote references are held by caller. + * + * caller provides the wait queue link structure. + */ +int +knote_link_wait_queue(struct knote *kn, struct wait_queue *wq, wait_queue_link_t wql) +{ + struct kqueue *kq = kn->kn_kq; + kern_return_t kr; + + kr = wait_queue_link_noalloc(wq, kq->kq_wqs, wql); + if (kr == KERN_SUCCESS) { + knote_markstayqueued(kn); + return (0); + } else { + return (EINVAL); + } +} + +/* + * Unlink the provided wait queue from the kqueue associated with a knote. + * Also remove it from the magic list of directly attached knotes. + * + * Note that the unlink may have already happened from the other side, so + * ignore any failures to unlink and just remove it from the kqueue list. + * + * On success, caller is responsible for the link structure + */ +int +knote_unlink_wait_queue(struct knote *kn, struct wait_queue *wq, wait_queue_link_t *wqlp) +{ + struct kqueue *kq = kn->kn_kq; + kern_return_t kr; + + kr = wait_queue_unlink_nofree(wq, kq->kq_wqs, wqlp); + knote_clearstayqueued(kn); + return ((kr != KERN_SUCCESS) ? EINVAL : 0); } /* @@ -1810,7 +2636,7 @@ knote_detach(struct klist *list, struct knote *kn) * * Essentially an inlined knote_remove & knote_drop * when we know for sure that the thing is a file - * + * * Entered with the proc_fd lock already held. * It returns the same way, but may drop it temporarily. */ @@ -1826,7 +2652,8 @@ knote_fdclose(struct proc *p, int fd) struct kqueue *kq = kn->kn_kq; if (kq->kq_p != p) - panic("knote_fdclose: proc mismatch (kq->kq_p=%p != p=%p)", kq->kq_p, p); + panic("%s: proc mismatch (kq->kq_p=%p != p=%p)", + __func__, kq->kq_p, p); kqlock(kq); proc_fdunlock(p); @@ -1842,7 +2669,7 @@ knote_fdclose(struct proc *p, int fd) kn->kn_fop->f_detach(kn); knote_drop(kn, p); } - + proc_fdlock(p); /* the fd tables may have changed - start over */ @@ -1852,7 +2679,7 @@ knote_fdclose(struct proc *p, int fd) /* proc_fdlock held on entry (and exit) */ static int -knote_fdpattach(struct knote *kn, struct filedesc *fdp, __unused struct proc *p) +knote_fdpattach(struct knote *kn, struct filedesc *fdp, struct proc *p) { struct klist *list = NULL; @@ -1865,20 +2692,28 @@ knote_fdpattach(struct knote *kn, struct filedesc *fdp, __unused struct proc *p) if ((u_int)fdp->fd_knlistsize <= kn->kn_id) { u_int size = 0; + if (kn->kn_id >= (uint64_t)p->p_rlimit[RLIMIT_NOFILE].rlim_cur + || kn->kn_id >= (uint64_t)maxfiles) + return (EINVAL); + /* have to grow the fd_knlist */ size = fdp->fd_knlistsize; while (size <= kn->kn_id) size += KQEXTENT; + + if (size >= (UINT_MAX/sizeof(struct klist *))) + return (EINVAL); + MALLOC(list, struct klist *, - size * sizeof(struct klist *), M_KQUEUE, M_WAITOK); + size * sizeof(struct klist *), M_KQUEUE, M_WAITOK); if (list == NULL) return (ENOMEM); - + bcopy((caddr_t)fdp->fd_knlist, (caddr_t)list, - fdp->fd_knlistsize * sizeof(struct klist *)); + fdp->fd_knlistsize * sizeof(struct klist *)); bzero((caddr_t)list + - fdp->fd_knlistsize * sizeof(struct klist *), - (size - fdp->fd_knlistsize) * sizeof(struct klist *)); + fdp->fd_knlistsize * sizeof(struct klist *), + (size - fdp->fd_knlistsize) * sizeof(struct klist *)); FREE(fdp->fd_knlist, M_KQUEUE); fdp->fd_knlist = list; fdp->fd_knlistsize = size; @@ -1900,8 +2735,9 @@ knote_drop(struct knote *kn, __unused struct proc *ctxp) { struct kqueue *kq = kn->kn_kq; struct proc *p = kq->kq_p; - struct filedesc *fdp = p->p_fd; + struct filedesc *fdp = p->p_fd; struct klist *list; + int needswakeup; proc_fdlock(p); if (kn->kn_fop->f_isfd) @@ -1912,11 +2748,14 @@ knote_drop(struct knote *kn, __unused struct proc *ctxp) SLIST_REMOVE(list, kn, knote, kn_link); kqlock(kq); knote_dequeue(kn); - if (kn->kn_status & KN_DROPWAIT) - thread_wakeup(&kn->kn_status); + needswakeup = (kn->kn_status & KN_USEWAIT); kqunlock(kq); proc_fdunlock(p); + if (needswakeup) + wait_queue_wakeup_all((wait_queue_t)kq->kq_wqs, &kn->kn_status, + THREAD_AWAKENED); + if (kn->kn_fop->f_isfd) fp_drop(p, kn->kn_id, kn->kn_fp, 0); @@ -1925,19 +2764,23 @@ knote_drop(struct knote *kn, __unused struct proc *ctxp) /* called with kqueue lock held */ static void -knote_activate(struct knote *kn) +knote_activate(struct knote *kn, int propagate) { struct kqueue *kq = kn->kn_kq; kn->kn_status |= KN_ACTIVE; knote_enqueue(kn); - kqueue_wakeup(kq); - } + kqueue_wakeup(kq, 0); + + /* this is a real event: wake up the parent kq, too */ + if (propagate) + KNOTE(&kq->kq_sel.si_note, 0); +} /* called with kqueue lock held */ static void knote_deactivate(struct knote *kn) -{ +{ kn->kn_status &= ~KN_ACTIVE; knote_dequeue(kn); } @@ -1946,12 +2789,12 @@ knote_deactivate(struct knote *kn) static void knote_enqueue(struct knote *kn) { - struct kqueue *kq = kn->kn_kq; - - if ((kn->kn_status & (KN_QUEUED | KN_DISABLED)) == 0) { + if ((kn->kn_status & (KN_QUEUED | KN_STAYQUEUED)) == KN_STAYQUEUED || + (kn->kn_status & (KN_QUEUED | KN_STAYQUEUED | KN_DISABLED)) == 0) { struct kqtailq *tq = kn->kn_tq; + struct kqueue *kq = kn->kn_kq; - TAILQ_INSERT_TAIL(tq, kn, kn_tqe); + TAILQ_INSERT_TAIL(tq, kn, kn_tqe); kn->kn_status |= KN_QUEUED; kq->kq_count++; } @@ -1963,11 +2806,10 @@ knote_dequeue(struct knote *kn) { struct kqueue *kq = kn->kn_kq; - //assert((kn->kn_status & KN_DISABLED) == 0); - if ((kn->kn_status & KN_QUEUED) == KN_QUEUED) { + if ((kn->kn_status & (KN_QUEUED | KN_STAYQUEUED)) == KN_QUEUED) { struct kqtailq *tq = kn->kn_tq; - TAILQ_REMOVE(tq, kn, kn_tqe); + TAILQ_REMOVE(tq, kn, kn_tqe); kn->kn_tq = &kq->kq_head; kn->kn_status &= ~KN_QUEUED; kq->kq_count--; @@ -1977,10 +2819,11 @@ knote_dequeue(struct knote *kn) void knote_init(void) { - knote_zone = zinit(sizeof(struct knote), 8192*sizeof(struct knote), 8192, "knote zone"); + knote_zone = zinit(sizeof(struct knote), 8192*sizeof(struct knote), + 8192, "knote zone"); /* allocate kq lock group attribute and group */ - kq_lck_grp_attr= lck_grp_attr_alloc_init(); + kq_lck_grp_attr = lck_grp_attr_alloc_init(); kq_lck_grp = lck_grp_alloc_init("kqueue", kq_lck_grp_attr); @@ -1989,6 +2832,16 @@ knote_init(void) /* Initialize the timer filter lock */ lck_mtx_init(&_filt_timerlock, kq_lck_grp, kq_lck_attr); + +#if VM_PRESSURE_EVENTS + /* Initialize the vm pressure list lock */ + vm_pressure_init(kq_lck_grp, kq_lck_attr); +#endif + +#if CONFIG_MEMORYSTATUS + /* Initialize the memorystatus list lock */ + memorystatus_kevent_init(kq_lck_grp, kq_lck_attr); +#endif } SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL) @@ -2015,271 +2868,620 @@ knote_free(struct knote *kn) #include #include +#ifndef ROUNDUP64 +#define ROUNDUP64(x) P2ROUNDUP((x), sizeof (u_int64_t)) +#endif + +#ifndef ADVANCE64 +#define ADVANCE64(p, n) (void*)((char *)(p) + ROUNDUP64(n)) +#endif + +static lck_grp_attr_t *kev_lck_grp_attr; +static lck_attr_t *kev_lck_attr; +static lck_grp_t *kev_lck_grp; +static decl_lck_rw_data(,kev_lck_data); +static lck_rw_t *kev_rwlock = &kev_lck_data; static int kev_attach(struct socket *so, int proto, struct proc *p); static int kev_detach(struct socket *so); -static int kev_control(struct socket *so, u_long cmd, caddr_t data, struct ifnet *ifp, struct proc *p); - -struct pr_usrreqs event_usrreqs = { - pru_abort_notsupp, pru_accept_notsupp, kev_attach, pru_bind_notsupp, pru_connect_notsupp, - pru_connect2_notsupp, kev_control, kev_detach, pru_disconnect_notsupp, - pru_listen_notsupp, pru_peeraddr_notsupp, pru_rcvd_notsupp, pru_rcvoob_notsupp, - pru_send_notsupp, pru_sense_null, pru_shutdown_notsupp, pru_sockaddr_notsupp, - pru_sosend_notsupp, soreceive, pru_sopoll_notsupp +static int kev_control(struct socket *so, u_long cmd, caddr_t data, + struct ifnet *ifp, struct proc *p); +static lck_mtx_t * event_getlock(struct socket *, int); +static int event_lock(struct socket *, int, void *); +static int event_unlock(struct socket *, int, void *); + +static int event_sofreelastref(struct socket *); +static void kev_delete(struct kern_event_pcb *); + +static struct pr_usrreqs event_usrreqs = { + .pru_attach = kev_attach, + .pru_control = kev_control, + .pru_detach = kev_detach, + .pru_soreceive = soreceive, }; -struct protosw eventsw[] = { - { - SOCK_RAW, &systemdomain, SYSPROTO_EVENT, PR_ATOMIC, - 0, 0, 0, 0, - 0, - 0, 0, 0, 0, -#if __APPLE__ - 0, -#endif - &event_usrreqs, - 0, 0, 0, -#if __APPLE__ - {0, 0}, 0, {0} -#endif - } +static struct protosw eventsw[] = { +{ + .pr_type = SOCK_RAW, + .pr_protocol = SYSPROTO_EVENT, + .pr_flags = PR_ATOMIC, + .pr_usrreqs = &event_usrreqs, + .pr_lock = event_lock, + .pr_unlock = event_unlock, + .pr_getlock = event_getlock, +} }; +__private_extern__ int kevt_getstat SYSCTL_HANDLER_ARGS; +__private_extern__ int kevt_pcblist SYSCTL_HANDLER_ARGS; + +SYSCTL_NODE(_net_systm, OID_AUTO, kevt, + CTLFLAG_RW|CTLFLAG_LOCKED, 0, "Kernel event family"); + +struct kevtstat kevtstat; +SYSCTL_PROC(_net_systm_kevt, OID_AUTO, stats, + CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, + kevt_getstat, "S,kevtstat", ""); + +SYSCTL_PROC(_net_systm_kevt, OID_AUTO, pcblist, + CTLTYPE_STRUCT | CTLFLAG_RD | CTLFLAG_LOCKED, 0, 0, + kevt_pcblist, "S,xkevtpcb", ""); + +static lck_mtx_t * +event_getlock(struct socket *so, int locktype) +{ +#pragma unused(locktype) + struct kern_event_pcb *ev_pcb = (struct kern_event_pcb *)so->so_pcb; + + if (so->so_pcb != NULL) { + if (so->so_usecount < 0) + panic("%s: so=%p usecount=%d lrh= %s\n", __func__, + so, so->so_usecount, solockhistory_nr(so)); + /* NOTREACHED */ + } else { + panic("%s: so=%p NULL NO so_pcb %s\n", __func__, + so, solockhistory_nr(so)); + /* NOTREACHED */ + } + return (&ev_pcb->evp_mtx); +} + +static int +event_lock(struct socket *so, int refcount, void *lr) +{ + void *lr_saved; + + if (lr == NULL) + lr_saved = __builtin_return_address(0); + else + lr_saved = lr; + + if (so->so_pcb != NULL) { + lck_mtx_lock(&((struct kern_event_pcb *)so->so_pcb)->evp_mtx); + } else { + panic("%s: so=%p NO PCB! lr=%p lrh= %s\n", __func__, + so, lr_saved, solockhistory_nr(so)); + /* NOTREACHED */ + } + + if (so->so_usecount < 0) { + panic("%s: so=%p so_pcb=%p lr=%p ref=%d lrh= %s\n", __func__, + so, so->so_pcb, lr_saved, so->so_usecount, + solockhistory_nr(so)); + /* NOTREACHED */ + } + + if (refcount) + so->so_usecount++; + + so->lock_lr[so->next_lock_lr] = lr_saved; + so->next_lock_lr = (so->next_lock_lr+1) % SO_LCKDBG_MAX; + return (0); +} + +static int +event_unlock(struct socket *so, int refcount, void *lr) +{ + void *lr_saved; + lck_mtx_t *mutex_held; + + if (lr == NULL) + lr_saved = __builtin_return_address(0); + else + lr_saved = lr; + + if (refcount) + so->so_usecount--; + + if (so->so_usecount < 0) { + panic("%s: so=%p usecount=%d lrh= %s\n", __func__, + so, so->so_usecount, solockhistory_nr(so)); + /* NOTREACHED */ + } + if (so->so_pcb == NULL) { + panic("%s: so=%p NO PCB usecount=%d lr=%p lrh= %s\n", __func__, + so, so->so_usecount, (void *)lr_saved, + solockhistory_nr(so)); + /* NOTREACHED */ + } + mutex_held = (&((struct kern_event_pcb *)so->so_pcb)->evp_mtx); + + lck_mtx_assert(mutex_held, LCK_MTX_ASSERT_OWNED); + so->unlock_lr[so->next_unlock_lr] = lr_saved; + so->next_unlock_lr = (so->next_unlock_lr+1) % SO_LCKDBG_MAX; + + if (so->so_usecount == 0) { + VERIFY(so->so_flags & SOF_PCBCLEARING); + event_sofreelastref(so); + } else { + lck_mtx_unlock(mutex_held); + } + + return (0); +} + +static int +event_sofreelastref(struct socket *so) +{ + struct kern_event_pcb *ev_pcb = (struct kern_event_pcb *)so->so_pcb; + + lck_mtx_assert(&(ev_pcb->evp_mtx), LCK_MTX_ASSERT_OWNED); + + so->so_pcb = NULL; + + /* + * Disable upcall in the event another thread is in kev_post_msg() + * appending record to the receive socket buffer, since sbwakeup() + * may release the socket lock otherwise. + */ + so->so_rcv.sb_flags &= ~SB_UPCALL; + so->so_snd.sb_flags &= ~SB_UPCALL; + so->so_event = sonullevent; + lck_mtx_unlock(&(ev_pcb->evp_mtx)); + + lck_mtx_assert(&(ev_pcb->evp_mtx), LCK_MTX_ASSERT_NOTOWNED); + lck_rw_lock_exclusive(kev_rwlock); + LIST_REMOVE(ev_pcb, evp_link); + kevtstat.kes_pcbcount--; + kevtstat.kes_gencnt++; + lck_rw_done(kev_rwlock); + kev_delete(ev_pcb); + + sofreelastref(so, 1); + return (0); +} + +static int event_proto_count = (sizeof (eventsw) / sizeof (struct protosw)); + static struct kern_event_head kern_event_head; -static u_long static_event_id = 0; -struct domain *sysdom = &systemdomain; -static lck_mtx_t *sys_mtx; +static u_int32_t static_event_id = 0; + +#define EVPCB_ZONE_MAX 65536 +#define EVPCB_ZONE_NAME "kerneventpcb" +static struct zone *ev_pcb_zone; /* - * Install the protosw's for the NKE manager. Invoked at - * extension load time + * Install the protosw's for the NKE manager. Invoked at extension load time */ -int -kern_event_init(void) +void +kern_event_init(struct domain *dp) { - int retval; + struct protosw *pr; + int i; + + VERIFY(!(dp->dom_flags & DOM_INITIALIZED)); + VERIFY(dp == systemdomain); + + kev_lck_grp_attr = lck_grp_attr_alloc_init(); + if (kev_lck_grp_attr == NULL) { + panic("%s: lck_grp_attr_alloc_init failed\n", __func__); + /* NOTREACHED */ + } + + kev_lck_grp = lck_grp_alloc_init("Kernel Event Protocol", + kev_lck_grp_attr); + if (kev_lck_grp == NULL) { + panic("%s: lck_grp_alloc_init failed\n", __func__); + /* NOTREACHED */ + } + + kev_lck_attr = lck_attr_alloc_init(); + if (kev_lck_attr == NULL) { + panic("%s: lck_attr_alloc_init failed\n", __func__); + /* NOTREACHED */ + } + + lck_rw_init(kev_rwlock, kev_lck_grp, kev_lck_attr); + if (kev_rwlock == NULL) { + panic("%s: lck_mtx_alloc_init failed\n", __func__); + /* NOTREACHED */ + } + + for (i = 0, pr = &eventsw[0]; i < event_proto_count; i++, pr++) + net_add_proto(pr, dp, 1); - if ((retval = net_add_proto(eventsw, &systemdomain)) != 0) { - log(LOG_WARNING, "Can't install kernel events protocol (%d)\n", retval); - return(retval); + ev_pcb_zone = zinit(sizeof(struct kern_event_pcb), + EVPCB_ZONE_MAX * sizeof(struct kern_event_pcb), 0, EVPCB_ZONE_NAME); + if (ev_pcb_zone == NULL) { + panic("%s: failed allocating ev_pcb_zone", __func__); + /* NOTREACHED */ } - - /* - * Use the domain mutex for all system event sockets - */ - sys_mtx = sysdom->dom_mtx; - - return(KERN_SUCCESS); + zone_change(ev_pcb_zone, Z_EXPAND, TRUE); + zone_change(ev_pcb_zone, Z_CALLERACCT, TRUE); } static int kev_attach(struct socket *so, __unused int proto, __unused struct proc *p) { - int error; - struct kern_event_pcb *ev_pcb; + int error = 0; + struct kern_event_pcb *ev_pcb; - error = soreserve(so, KEV_SNDSPACE, KEV_RECVSPACE); - if (error) - return error; + error = soreserve(so, KEV_SNDSPACE, KEV_RECVSPACE); + if (error != 0) + return (error); - MALLOC(ev_pcb, struct kern_event_pcb *, sizeof(struct kern_event_pcb), M_PCB, M_WAITOK); - if (ev_pcb == 0) - return ENOBUFS; + if ((ev_pcb = (struct kern_event_pcb *)zalloc(ev_pcb_zone)) == NULL) { + return (ENOBUFS); + } + bzero(ev_pcb, sizeof(struct kern_event_pcb)); + lck_mtx_init(&ev_pcb->evp_mtx, kev_lck_grp, kev_lck_attr); - ev_pcb->ev_socket = so; - ev_pcb->vendor_code_filter = 0xffffffff; + ev_pcb->evp_socket = so; + ev_pcb->evp_vendor_code_filter = 0xffffffff; - so->so_pcb = (caddr_t) ev_pcb; - lck_mtx_lock(sys_mtx); - LIST_INSERT_HEAD(&kern_event_head, ev_pcb, ev_link); - lck_mtx_unlock(sys_mtx); + so->so_pcb = (caddr_t) ev_pcb; + lck_rw_lock_exclusive(kev_rwlock); + LIST_INSERT_HEAD(&kern_event_head, ev_pcb, evp_link); + kevtstat.kes_pcbcount++; + kevtstat.kes_gencnt++; + lck_rw_done(kev_rwlock); - return 0; + return (error); } +static void +kev_delete(struct kern_event_pcb *ev_pcb) +{ + VERIFY(ev_pcb != NULL); + lck_mtx_destroy(&ev_pcb->evp_mtx, kev_lck_grp); + zfree(ev_pcb_zone, ev_pcb); +} static int kev_detach(struct socket *so) { - struct kern_event_pcb *ev_pcb = (struct kern_event_pcb *) so->so_pcb; + struct kern_event_pcb *ev_pcb = (struct kern_event_pcb *) so->so_pcb; - if (ev_pcb != 0) { - LIST_REMOVE(ev_pcb, ev_link); - FREE(ev_pcb, M_PCB); - so->so_pcb = 0; + if (ev_pcb != NULL) { + soisdisconnected(so); so->so_flags |= SOF_PCBCLEARING; - } + } - return 0; + return (0); } /* * For now, kev_vendor_code and mbuf_tags use the same * mechanism. */ - errno_t kev_vendor_code_find( const char *string, u_int32_t *out_vendor_code) { if (strlen(string) >= KEV_VENDOR_CODE_MAX_STR_LEN) { - return EINVAL; + return (EINVAL); } - return mbuf_tag_id_find_internal(string, out_vendor_code, 1); + return (net_str_id_find_internal(string, out_vendor_code, + NSI_VENDOR_CODE, 1)); } -errno_t kev_msg_post(struct kev_msg *event_msg) +errno_t +kev_msg_post(struct kev_msg *event_msg) { - mbuf_tag_id_t min_vendor, max_vendor; - - mbuf_tag_id_first_last(&min_vendor, &max_vendor); - + mbuf_tag_id_t min_vendor, max_vendor; + + net_str_id_first_last(&min_vendor, &max_vendor, NSI_VENDOR_CODE); + if (event_msg == NULL) - return EINVAL; - - /* Limit third parties to posting events for registered vendor codes only */ + return (EINVAL); + + /* + * Limit third parties to posting events for registered vendor codes + * only + */ if (event_msg->vendor_code < min_vendor || - event_msg->vendor_code > max_vendor) - { - return EINVAL; + event_msg->vendor_code > max_vendor) { + OSIncrementAtomic64((SInt64 *)&kevtstat.kes_badvendor); + return (EINVAL); } - - return kev_post_msg(event_msg); + return (kev_post_msg(event_msg)); } - -int kev_post_msg(struct kev_msg *event_msg) +int +kev_post_msg(struct kev_msg *event_msg) { - struct mbuf *m, *m2; - struct kern_event_pcb *ev_pcb; - struct kern_event_msg *ev; - char *tmp; - unsigned long total_size; - int i; + struct mbuf *m, *m2; + struct kern_event_pcb *ev_pcb; + struct kern_event_msg *ev; + char *tmp; + u_int32_t total_size; + int i; /* Verify the message is small enough to fit in one mbuf w/o cluster */ total_size = KEV_MSG_HEADER_SIZE; - + for (i = 0; i < 5; i++) { if (event_msg->dv[i].data_length == 0) break; total_size += event_msg->dv[i].data_length; } - + if (total_size > MLEN) { - return EMSGSIZE; - } - - m = m_get(M_DONTWAIT, MT_DATA); - if (m == 0) - return ENOBUFS; - - ev = mtod(m, struct kern_event_msg *); - total_size = KEV_MSG_HEADER_SIZE; - - tmp = (char *) &ev->event_data[0]; - for (i = 0; i < 5; i++) { - if (event_msg->dv[i].data_length == 0) - break; - - total_size += event_msg->dv[i].data_length; - bcopy(event_msg->dv[i].data_ptr, tmp, - event_msg->dv[i].data_length); - tmp += event_msg->dv[i].data_length; - } - - ev->id = ++static_event_id; - ev->total_size = total_size; - ev->vendor_code = event_msg->vendor_code; - ev->kev_class = event_msg->kev_class; - ev->kev_subclass = event_msg->kev_subclass; - ev->event_code = event_msg->event_code; - - m->m_len = total_size; - lck_mtx_lock(sys_mtx); - for (ev_pcb = LIST_FIRST(&kern_event_head); - ev_pcb; - ev_pcb = LIST_NEXT(ev_pcb, ev_link)) { - - if (ev_pcb->vendor_code_filter != KEV_ANY_VENDOR) { - if (ev_pcb->vendor_code_filter != ev->vendor_code) - continue; - - if (ev_pcb->class_filter != KEV_ANY_CLASS) { - if (ev_pcb->class_filter != ev->kev_class) - continue; - - if ((ev_pcb->subclass_filter != KEV_ANY_SUBCLASS) && - (ev_pcb->subclass_filter != ev->kev_subclass)) - continue; - } - } - - m2 = m_copym(m, 0, m->m_len, M_NOWAIT); - if (m2 == 0) { - m_free(m); - lck_mtx_unlock(sys_mtx); - return ENOBUFS; - } - /* the socket is already locked because we hold the sys_mtx here */ - if (sbappendrecord(&ev_pcb->ev_socket->so_rcv, m2)) - sorwakeup(ev_pcb->ev_socket); - } - - m_free(m); - lck_mtx_unlock(sys_mtx); - return 0; + OSIncrementAtomic64((SInt64 *)&kevtstat.kes_toobig); + return (EMSGSIZE); + } + + m = m_get(M_DONTWAIT, MT_DATA); + if (m == 0) { + OSIncrementAtomic64((SInt64 *)&kevtstat.kes_nomem); + return (ENOMEM); + } + ev = mtod(m, struct kern_event_msg *); + total_size = KEV_MSG_HEADER_SIZE; + + tmp = (char *) &ev->event_data[0]; + for (i = 0; i < 5; i++) { + if (event_msg->dv[i].data_length == 0) + break; + + total_size += event_msg->dv[i].data_length; + bcopy(event_msg->dv[i].data_ptr, tmp, + event_msg->dv[i].data_length); + tmp += event_msg->dv[i].data_length; + } + + ev->id = ++static_event_id; + ev->total_size = total_size; + ev->vendor_code = event_msg->vendor_code; + ev->kev_class = event_msg->kev_class; + ev->kev_subclass = event_msg->kev_subclass; + ev->event_code = event_msg->event_code; + + m->m_len = total_size; + lck_rw_lock_shared(kev_rwlock); + for (ev_pcb = LIST_FIRST(&kern_event_head); + ev_pcb; + ev_pcb = LIST_NEXT(ev_pcb, evp_link)) { + lck_mtx_lock(&ev_pcb->evp_mtx); + if (ev_pcb->evp_socket->so_pcb == NULL) { + lck_mtx_unlock(&ev_pcb->evp_mtx); + continue; + } + if (ev_pcb->evp_vendor_code_filter != KEV_ANY_VENDOR) { + if (ev_pcb->evp_vendor_code_filter != ev->vendor_code) { + lck_mtx_unlock(&ev_pcb->evp_mtx); + continue; + } + + if (ev_pcb->evp_class_filter != KEV_ANY_CLASS) { + if (ev_pcb->evp_class_filter != ev->kev_class) { + lck_mtx_unlock(&ev_pcb->evp_mtx); + continue; + } + + if ((ev_pcb->evp_subclass_filter != + KEV_ANY_SUBCLASS) && + (ev_pcb->evp_subclass_filter != + ev->kev_subclass)) { + lck_mtx_unlock(&ev_pcb->evp_mtx); + continue; + } + } + } + + m2 = m_copym(m, 0, m->m_len, M_NOWAIT); + if (m2 == 0) { + OSIncrementAtomic64((SInt64 *)&kevtstat.kes_nomem); + m_free(m); + lck_mtx_unlock(&ev_pcb->evp_mtx); + lck_rw_done(kev_rwlock); + return (ENOMEM); + } + if (sbappendrecord(&ev_pcb->evp_socket->so_rcv, m2)) { + /* + * We use "m" for the socket stats as it would be + * unsafe to use "m2" + */ + so_inc_recv_data_stat(ev_pcb->evp_socket, + 1, m->m_len, SO_TC_BE); + + sorwakeup(ev_pcb->evp_socket); + OSIncrementAtomic64((SInt64 *)&kevtstat.kes_posted); + } else { + OSIncrementAtomic64((SInt64 *)&kevtstat.kes_fullsock); + } + lck_mtx_unlock(&ev_pcb->evp_mtx); + } + m_free(m); + lck_rw_done(kev_rwlock); + + return (0); } static int -kev_control(struct socket *so, - u_long cmd, - caddr_t data, - __unused struct ifnet *ifp, - __unused struct proc *p) +kev_control(struct socket *so, + u_long cmd, + caddr_t data, + __unused struct ifnet *ifp, + __unused struct proc *p) { struct kev_request *kev_req = (struct kev_request *) data; struct kern_event_pcb *ev_pcb; struct kev_vendor_code *kev_vendor; - u_long *id_value = (u_long *) data; - - + u_int32_t *id_value = (u_int32_t *) data; + switch (cmd) { - case SIOCGKEVID: *id_value = static_event_id; break; - case SIOCSKEVFILT: ev_pcb = (struct kern_event_pcb *) so->so_pcb; - ev_pcb->vendor_code_filter = kev_req->vendor_code; - ev_pcb->class_filter = kev_req->kev_class; - ev_pcb->subclass_filter = kev_req->kev_subclass; + ev_pcb->evp_vendor_code_filter = kev_req->vendor_code; + ev_pcb->evp_class_filter = kev_req->kev_class; + ev_pcb->evp_subclass_filter = kev_req->kev_subclass; break; - case SIOCGKEVFILT: ev_pcb = (struct kern_event_pcb *) so->so_pcb; - kev_req->vendor_code = ev_pcb->vendor_code_filter; - kev_req->kev_class = ev_pcb->class_filter; - kev_req->kev_subclass = ev_pcb->subclass_filter; + kev_req->vendor_code = ev_pcb->evp_vendor_code_filter; + kev_req->kev_class = ev_pcb->evp_class_filter; + kev_req->kev_subclass = ev_pcb->evp_subclass_filter; break; - case SIOCGKEVVENDOR: - kev_vendor = (struct kev_vendor_code*)data; - + kev_vendor = (struct kev_vendor_code *)data; /* Make sure string is NULL terminated */ kev_vendor->vendor_string[KEV_VENDOR_CODE_MAX_STR_LEN-1] = 0; - - return mbuf_tag_id_find_internal(kev_vendor->vendor_string, - &kev_vendor->vendor_code, 0); - + return (net_str_id_find_internal(kev_vendor->vendor_string, + &kev_vendor->vendor_code, NSI_VENDOR_CODE, 0)); default: - return ENOTSUP; + return (ENOTSUP); + } + + return (0); +} + +int +kevt_getstat SYSCTL_HANDLER_ARGS +{ +#pragma unused(oidp, arg1, arg2) + int error = 0; + + lck_rw_lock_shared(kev_rwlock); + + if (req->newptr != USER_ADDR_NULL) { + error = EPERM; + goto done; + } + if (req->oldptr == USER_ADDR_NULL) { + req->oldidx = sizeof(struct kevtstat); + goto done; + } + + error = SYSCTL_OUT(req, &kevtstat, + MIN(sizeof(struct kevtstat), req->oldlen)); +done: + lck_rw_done(kev_rwlock); + + return (error); +} + +__private_extern__ int +kevt_pcblist SYSCTL_HANDLER_ARGS +{ +#pragma unused(oidp, arg1, arg2) + int error = 0; + int n, i; + struct xsystmgen xsg; + void *buf = NULL; + size_t item_size = ROUNDUP64(sizeof (struct xkevtpcb)) + + ROUNDUP64(sizeof (struct xsocket_n)) + + 2 * ROUNDUP64(sizeof (struct xsockbuf_n)) + + ROUNDUP64(sizeof (struct xsockstat_n)); + struct kern_event_pcb *ev_pcb; + + buf = _MALLOC(item_size, M_TEMP, M_WAITOK | M_ZERO); + if (buf == NULL) + return (ENOMEM); + + lck_rw_lock_shared(kev_rwlock); + + n = kevtstat.kes_pcbcount; + + if (req->oldptr == USER_ADDR_NULL) { + req->oldidx = (n + n/8) * item_size; + goto done; + } + if (req->newptr != USER_ADDR_NULL) { + error = EPERM; + goto done; + } + bzero(&xsg, sizeof (xsg)); + xsg.xg_len = sizeof (xsg); + xsg.xg_count = n; + xsg.xg_gen = kevtstat.kes_gencnt; + xsg.xg_sogen = so_gencnt; + error = SYSCTL_OUT(req, &xsg, sizeof (xsg)); + if (error) { + goto done; + } + /* + * We are done if there is no pcb + */ + if (n == 0) { + goto done; + } + + i = 0; + for (i = 0, ev_pcb = LIST_FIRST(&kern_event_head); + i < n && ev_pcb != NULL; + i++, ev_pcb = LIST_NEXT(ev_pcb, evp_link)) { + struct xkevtpcb *xk = (struct xkevtpcb *)buf; + struct xsocket_n *xso = (struct xsocket_n *) + ADVANCE64(xk, sizeof (*xk)); + struct xsockbuf_n *xsbrcv = (struct xsockbuf_n *) + ADVANCE64(xso, sizeof (*xso)); + struct xsockbuf_n *xsbsnd = (struct xsockbuf_n *) + ADVANCE64(xsbrcv, sizeof (*xsbrcv)); + struct xsockstat_n *xsostats = (struct xsockstat_n *) + ADVANCE64(xsbsnd, sizeof (*xsbsnd)); + + bzero(buf, item_size); + + lck_mtx_lock(&ev_pcb->evp_mtx); + + xk->kep_len = sizeof(struct xkevtpcb); + xk->kep_kind = XSO_EVT; + xk->kep_evtpcb = (uint64_t)VM_KERNEL_ADDRPERM(ev_pcb); + xk->kep_vendor_code_filter = ev_pcb->evp_vendor_code_filter; + xk->kep_class_filter = ev_pcb->evp_class_filter; + xk->kep_subclass_filter = ev_pcb->evp_subclass_filter; + + sotoxsocket_n(ev_pcb->evp_socket, xso); + sbtoxsockbuf_n(ev_pcb->evp_socket ? + &ev_pcb->evp_socket->so_rcv : NULL, xsbrcv); + sbtoxsockbuf_n(ev_pcb->evp_socket ? + &ev_pcb->evp_socket->so_snd : NULL, xsbsnd); + sbtoxsockstat_n(ev_pcb->evp_socket, xsostats); + + lck_mtx_unlock(&ev_pcb->evp_mtx); + + error = SYSCTL_OUT(req, buf, item_size); + } + + if (error == 0) { + /* + * Give the user an updated idea of our state. + * If the generation differs from what we told + * her before, she knows that something happened + * while we were processing this request, and it + * might be necessary to retry. + */ + bzero(&xsg, sizeof (xsg)); + xsg.xg_len = sizeof (xsg); + xsg.xg_count = n; + xsg.xg_gen = kevtstat.kes_gencnt; + xsg.xg_sogen = so_gencnt; + error = SYSCTL_OUT(req, &xsg, sizeof (xsg)); + if (error) { + goto done; + } } - - return 0; + +done: + lck_rw_done(kev_rwlock); + + return (error); } #endif /* SOCKETS */ @@ -2290,18 +3492,37 @@ fill_kqueueinfo(struct kqueue *kq, struct kqueue_info * kinfo) { struct vinfo_stat * st; - /* No need for the funnel as fd is kept alive */ - st = &kinfo->kq_stat; st->vst_size = kq->kq_count; - st->vst_blksize = sizeof(struct kevent); + if (kq->kq_state & KQ_KEV64) + st->vst_blksize = sizeof(struct kevent64_s); + else + st->vst_blksize = sizeof(struct kevent); st->vst_mode = S_IFIFO; if (kq->kq_state & KQ_SEL) kinfo->kq_state |= PROC_KQUEUE_SELECT; if (kq->kq_state & KQ_SLEEP) kinfo->kq_state |= PROC_KQUEUE_SLEEP; - return(0); + return (0); +} + + +void +knote_markstayqueued(struct knote *kn) +{ + kqlock(kn->kn_kq); + kn->kn_status |= KN_STAYQUEUED; + knote_enqueue(kn); + kqunlock(kn->kn_kq); } +void +knote_clearstayqueued(struct knote *kn) +{ + kqlock(kn->kn_kq); + kn->kn_status &= ~KN_STAYQUEUED; + knote_dequeue(kn); + kqunlock(kn->kn_kq); +}