xref: /freebsd/sys/kern/kern_event.c (revision c18a16ebcf5bf0bad19be10f58d9f42cbc079057)

/*-
 * SPDX-License-Identifier: BSD-2-Clause
 *
 * Copyright (c) 1999,2000,2001 Jonathan Lemon <jlemon@FreeBSD.org>
 * Copyright 2004 John-Mark Gurney <jmg@FreeBSD.org>
 * Copyright (c) 2009 Apple, Inc.
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 */

#include <sys/cdefs.h>
#include "opt_ktrace.h"
#include "opt_kqueue.h"

#ifdef COMPAT_FREEBSD11
#define	_WANT_FREEBSD11_KEVENT
#endif

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/capsicum.h>
#include <sys/kernel.h>
#include <sys/limits.h>
#include <sys/lock.h>
#include <sys/mutex.h>
#include <sys/proc.h>
#include <sys/malloc.h>
#include <sys/unistd.h>
#include <sys/file.h>
#include <sys/filedesc.h>
#include <sys/filio.h>
#include <sys/fcntl.h>
#include <sys/kthread.h>
#include <sys/selinfo.h>
#include <sys/queue.h>
#include <sys/event.h>
#include <sys/eventvar.h>
#include <sys/poll.h>
#include <sys/protosw.h>
#include <sys/resourcevar.h>
#include <sys/sbuf.h>
#include <sys/sigio.h>
#include <sys/signalvar.h>
#include <sys/socket.h>
#include <sys/socketvar.h>
#include <sys/stat.h>
#include <sys/sysctl.h>
#include <sys/sysent.h>
#include <sys/sysproto.h>
#include <sys/syscallsubr.h>
#include <sys/taskqueue.h>
#include <sys/uio.h>
#include <sys/user.h>
#ifdef KTRACE
#include <sys/ktrace.h>
#endif
#include <machine/atomic.h>
#ifdef COMPAT_FREEBSD32
#include <compat/freebsd32/freebsd32.h>
#include <compat/freebsd32/freebsd32_util.h>
#endif

#include <vm/uma.h>

static MALLOC_DEFINE(M_KQUEUE, "kqueue", "memory for kqueue system");

/*
 * This lock is used if multiple kq locks are required.  This possibly
 * should be made into a per proc lock.
 */
static struct mtx	kq_global;
MTX_SYSINIT(kq_global, &kq_global, "kqueue order", MTX_DEF);
#define KQ_GLOBAL_LOCK(lck, haslck)	do {	\
	if (!haslck)				\
		mtx_lock(lck);			\
	haslck = 1;				\
} while (0)
#define KQ_GLOBAL_UNLOCK(lck, haslck)	do {	\
	if (haslck)				\
		mtx_unlock(lck);			\
	haslck = 0;				\
} while (0)

TASKQUEUE_DEFINE_THREAD(kqueue_ctx);

static int	kevent_copyout(void *arg, struct kevent *kevp, int count);
static int	kevent_copyin(void *arg, struct kevent *kevp, int count);
static int	kqueue_register(struct kqueue *kq, struct kevent *kev,
		    struct thread *td, int mflag);
static int	kqueue_acquire(struct file *fp, struct kqueue **kqp);
static void	kqueue_release(struct kqueue *kq, int locked);
static void	kqueue_destroy(struct kqueue *kq);
static void	kqueue_drain(struct kqueue *kq, struct thread *td);
static int	kqueue_expand(struct kqueue *kq, const struct filterops *fops,
		    uintptr_t ident, int mflag);
static void	kqueue_task(void *arg, int pending);
static int	kqueue_scan(struct kqueue *kq, int maxevents,
		    struct kevent_copyops *k_ops,
		    const struct timespec *timeout,
		    struct kevent *keva, struct thread *td);
static void 	kqueue_wakeup(struct kqueue *kq);
static const struct filterops *kqueue_fo_find(int filt);
static void	kqueue_fo_release(int filt);
struct g_kevent_args;
static int	kern_kevent_generic(struct thread *td,
		    struct g_kevent_args *uap,
		    struct kevent_copyops *k_ops, const char *struct_name);

static fo_ioctl_t	kqueue_ioctl;
static fo_poll_t	kqueue_poll;
static fo_kqfilter_t	kqueue_kqfilter;
static fo_stat_t	kqueue_stat;
static fo_close_t	kqueue_close;
static fo_fill_kinfo_t	kqueue_fill_kinfo;

static const struct fileops kqueueops = {
	.fo_read = invfo_rdwr,
	.fo_write = invfo_rdwr,
	.fo_truncate = invfo_truncate,
	.fo_ioctl = kqueue_ioctl,
	.fo_poll = kqueue_poll,
	.fo_kqfilter = kqueue_kqfilter,
	.fo_stat = kqueue_stat,
	.fo_close = kqueue_close,
	.fo_chmod = invfo_chmod,
	.fo_chown = invfo_chown,
	.fo_sendfile = invfo_sendfile,
	.fo_cmp = file_kcmp_generic,
	.fo_fill_kinfo = kqueue_fill_kinfo,
};

static int 	knote_attach(struct knote *kn, struct kqueue *kq);
static void 	knote_drop(struct knote *kn, struct thread *td);
static void 	knote_drop_detached(struct knote *kn, struct thread *td);
static void 	knote_enqueue(struct knote *kn);
static void 	knote_dequeue(struct knote *kn);
static void 	knote_init(void);
static struct 	knote *knote_alloc(int mflag);
static void 	knote_free(struct knote *kn);

static void	filt_kqdetach(struct knote *kn);
static int	filt_kqueue(struct knote *kn, long hint);
static int	filt_procattach(struct knote *kn);
static void	filt_procdetach(struct knote *kn);
static int	filt_proc(struct knote *kn, long hint);
static int	filt_fileattach(struct knote *kn);
static void	filt_timerexpire(void *knx);
static void	filt_timerexpire_l(struct knote *kn, bool proc_locked);
static int	filt_timerattach(struct knote *kn);
static void	filt_timerdetach(struct knote *kn);
static void	filt_timerstart(struct knote *kn, sbintime_t to);
static void	filt_timertouch(struct knote *kn, struct kevent *kev,
		    u_long type);
static int	filt_timervalidate(struct knote *kn, sbintime_t *to);
static int	filt_timer(struct knote *kn, long hint);
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 kevent *kev,
		    u_long type);

static const struct filterops file_filtops = {
	.f_isfd = 1,
	.f_attach = filt_fileattach,
};
static const struct filterops kqread_filtops = {
	.f_isfd = 1,
	.f_detach = filt_kqdetach,
	.f_event = filt_kqueue,
};
/* XXX - move to kern_proc.c?  */
static const struct filterops proc_filtops = {
	.f_isfd = 0,
	.f_attach = filt_procattach,
	.f_detach = filt_procdetach,
	.f_event = filt_proc,
};
static const struct filterops timer_filtops = {
	.f_isfd = 0,
	.f_attach = filt_timerattach,
	.f_detach = filt_timerdetach,
	.f_event = filt_timer,
	.f_touch = filt_timertouch,
};
static const struct filterops user_filtops = {
	.f_attach = filt_userattach,
	.f_detach = filt_userdetach,
	.f_event = filt_user,
	.f_touch = filt_usertouch,
};

static uma_zone_t	knote_zone;
static unsigned int __exclusive_cache_line	kq_ncallouts;
static unsigned int 	kq_calloutmax = 4 * 1024;
SYSCTL_UINT(_kern, OID_AUTO, kq_calloutmax, CTLFLAG_RW,
    &kq_calloutmax, 0, "Maximum number of callouts allocated for kqueue");

/* XXX - ensure not influx ? */
#define KNOTE_ACTIVATE(kn, islock) do { 				\
	if ((islock))							\
		mtx_assert(&(kn)->kn_kq->kq_lock, MA_OWNED);		\
	else								\
		KQ_LOCK((kn)->kn_kq);					\
	(kn)->kn_status |= KN_ACTIVE;					\
	if (((kn)->kn_status & (KN_QUEUED | KN_DISABLED)) == 0)		\
		knote_enqueue((kn));					\
	if (!(islock))							\
		KQ_UNLOCK((kn)->kn_kq);					\
} while (0)
#define KQ_LOCK(kq) do {						\
	mtx_lock(&(kq)->kq_lock);					\
} while (0)
#define KQ_FLUX_WAKEUP(kq) do {						\
	if (((kq)->kq_state & KQ_FLUXWAIT) == KQ_FLUXWAIT) {		\
		(kq)->kq_state &= ~KQ_FLUXWAIT;				\
		wakeup((kq));						\
	}								\
} while (0)
#define KQ_UNLOCK_FLUX(kq) do {						\
	KQ_FLUX_WAKEUP(kq);						\
	mtx_unlock(&(kq)->kq_lock);					\
} while (0)
#define KQ_UNLOCK(kq) do {						\
	mtx_unlock(&(kq)->kq_lock);					\
} while (0)
#define KQ_OWNED(kq) do {						\
	mtx_assert(&(kq)->kq_lock, MA_OWNED);				\
} while (0)
#define KQ_NOTOWNED(kq) do {						\
	mtx_assert(&(kq)->kq_lock, MA_NOTOWNED);			\
} while (0)

static struct knlist *
kn_list_lock(struct knote *kn)
{
	struct knlist *knl;

	knl = kn->kn_knlist;
	if (knl != NULL)
		knl->kl_lock(knl->kl_lockarg);
	return (knl);
}

static void
kn_list_unlock(struct knlist *knl)
{
	bool do_free;

	if (knl == NULL)
		return;
	do_free = knl->kl_autodestroy && knlist_empty(knl);
	knl->kl_unlock(knl->kl_lockarg);
	if (do_free) {
		knlist_destroy(knl);
		free(knl, M_KQUEUE);
	}
}

static bool
kn_in_flux(struct knote *kn)
{

	return (kn->kn_influx > 0);
}

static void
kn_enter_flux(struct knote *kn)
{

	KQ_OWNED(kn->kn_kq);
	MPASS(kn->kn_influx < INT_MAX);
	kn->kn_influx++;
}

static bool
kn_leave_flux(struct knote *kn)
{

	KQ_OWNED(kn->kn_kq);
	MPASS(kn->kn_influx > 0);
	kn->kn_influx--;
	return (kn->kn_influx == 0);
}

#define	KNL_ASSERT_LOCK(knl, islocked) do {				\
	if (islocked)							\
		KNL_ASSERT_LOCKED(knl);				\
	else								\
		KNL_ASSERT_UNLOCKED(knl);				\
} while (0)
#ifdef INVARIANTS
#define	KNL_ASSERT_LOCKED(knl) do {					\
	knl->kl_assert_lock((knl)->kl_lockarg, LA_LOCKED);		\
} while (0)
#define	KNL_ASSERT_UNLOCKED(knl) do {					\
	knl->kl_assert_lock((knl)->kl_lockarg, LA_UNLOCKED);		\
} while (0)
#else /* !INVARIANTS */
#define	KNL_ASSERT_LOCKED(knl) do {} while (0)
#define	KNL_ASSERT_UNLOCKED(knl) do {} while (0)
#endif /* INVARIANTS */

#ifndef	KN_HASHSIZE
#define	KN_HASHSIZE		64		/* XXX should be tunable */
#endif

#define KN_HASH(val, mask)	(((val) ^ (val >> 8)) & (mask))

static int
filt_nullattach(struct knote *kn)
{

	return (ENXIO);
};

static const struct filterops null_filtops = {
	.f_isfd = 0,
	.f_attach = filt_nullattach,
};

/* XXX - make SYSINIT to add these, and move into respective modules. */
extern const struct filterops sig_filtops;
extern const struct filterops fs_filtops;

/*
 * Table for all system-defined filters.
 */
static struct mtx	filterops_lock;
MTX_SYSINIT(kqueue_filterops, &filterops_lock, "protect sysfilt_ops", MTX_DEF);
static struct {
	const struct filterops *for_fop;
	int for_nolock;
	int for_refcnt;
} sysfilt_ops[EVFILT_SYSCOUNT] = {
	[~EVFILT_READ] = { &file_filtops, 1 },
	[~EVFILT_WRITE] = { &file_filtops, 1 },
	[~EVFILT_AIO] = { &null_filtops },
	[~EVFILT_VNODE] = { &file_filtops, 1 },
	[~EVFILT_PROC] = { &proc_filtops, 1 },
	[~EVFILT_SIGNAL] = { &sig_filtops, 1 },
	[~EVFILT_TIMER] = { &timer_filtops, 1 },
	[~EVFILT_PROCDESC] = { &file_filtops, 1 },
	[~EVFILT_FS] = { &fs_filtops, 1 },
	[~EVFILT_LIO] = { &null_filtops },
	[~EVFILT_USER] = { &user_filtops, 1 },
	[~EVFILT_SENDFILE] = { &null_filtops },
	[~EVFILT_EMPTY] = { &file_filtops, 1 },
};

/*
 * Simple redirection for all cdevsw style objects to call their fo_kqfilter
 * method.
 */
static int
filt_fileattach(struct knote *kn)
{

	return (fo_kqfilter(kn->kn_fp, kn));
}

/*ARGSUSED*/
static int
kqueue_kqfilter(struct file *fp, struct knote *kn)
{
	struct kqueue *kq = kn->kn_fp->f_data;

	if (kn->kn_filter != EVFILT_READ)
		return (EINVAL);

	kn->kn_status |= KN_KQUEUE;
	kn->kn_fop = &kqread_filtops;
	knlist_add(&kq->kq_sel.si_note, kn, 0);

	return (0);
}

static void
filt_kqdetach(struct knote *kn)
{
	struct kqueue *kq = kn->kn_fp->f_data;

	knlist_remove(&kq->kq_sel.si_note, kn, 0);
}

/*ARGSUSED*/
static int
filt_kqueue(struct knote *kn, long hint)
{
	struct kqueue *kq = kn->kn_fp->f_data;

	kn->kn_data = kq->kq_count;
	return (kn->kn_data > 0);
}

/* XXX - move to kern_proc.c?  */
static int
filt_procattach(struct knote *kn)
{
	struct proc *p;
	int error;
	bool exiting, immediate;

	exiting = immediate = false;
	if (kn->kn_sfflags & NOTE_EXIT)
		p = pfind_any(kn->kn_id);
	else
		p = pfind(kn->kn_id);
	if (p == NULL)
		return (ESRCH);
	if (p->p_flag & P_WEXIT)
		exiting = true;

	if ((error = p_cansee(curthread, p))) {
		PROC_UNLOCK(p);
		return (error);
	}

	kn->kn_ptr.p_proc = p;
	kn->kn_flags |= EV_CLEAR;		/* automatically set */

	/*
	 * Internal flag indicating registration done by kernel for the
	 * purposes of getting a NOTE_CHILD notification.
	 */
	if (kn->kn_flags & EV_FLAG2) {
		kn->kn_flags &= ~EV_FLAG2;
		kn->kn_data = kn->kn_sdata;		/* ppid */
		kn->kn_fflags = NOTE_CHILD;
		kn->kn_sfflags &= ~(NOTE_EXIT | NOTE_EXEC | NOTE_FORK);
		immediate = true; /* Force immediate activation of child note. */
	}
	/*
	 * Internal flag indicating registration done by kernel (for other than
	 * NOTE_CHILD).
	 */
	if (kn->kn_flags & EV_FLAG1) {
		kn->kn_flags &= ~EV_FLAG1;
	}

	knlist_add(p->p_klist, kn, 1);

	/*
	 * Immediately activate any child notes or, in the case of a zombie
	 * target process, exit notes.  The latter is necessary to handle the
	 * case where the target process, e.g. a child, dies before the kevent
	 * is registered.
	 */
	if (immediate || (exiting && filt_proc(kn, NOTE_EXIT)))
		KNOTE_ACTIVATE(kn, 0);

	PROC_UNLOCK(p);

	return (0);
}

/*
 * The knote may be attached to a different process, which may exit,
 * leaving nothing for the knote to be attached to.  So when the process
 * exits, the knote is marked as DETACHED and also flagged as ONESHOT so
 * it will be deleted when read out.  However, as part of the knote deletion,
 * this routine is called, so a check is needed to avoid actually performing
 * a detach, because the original process does not exist any more.
 */
/* XXX - move to kern_proc.c?  */
static void
filt_procdetach(struct knote *kn)
{

	knlist_remove(kn->kn_knlist, kn, 0);
	kn->kn_ptr.p_proc = NULL;
}

/* XXX - move to kern_proc.c?  */
static int
filt_proc(struct knote *kn, long hint)
{
	struct proc *p;
	u_int event;

	p = kn->kn_ptr.p_proc;
	if (p == NULL) /* already activated, from attach filter */
		return (0);

	/* Mask off extra data. */
	event = (u_int)hint & NOTE_PCTRLMASK;

	/* If the user is interested in this event, record it. */
	if (kn->kn_sfflags & event)
		kn->kn_fflags |= event;

	/* Process is gone, so flag the event as finished. */
	if (event == NOTE_EXIT) {
		kn->kn_flags |= EV_EOF | EV_ONESHOT;
		kn->kn_ptr.p_proc = NULL;
		if (kn->kn_fflags & NOTE_EXIT)
			kn->kn_data = KW_EXITCODE(p->p_xexit, p->p_xsig);
		if (kn->kn_fflags == 0)
			kn->kn_flags |= EV_DROP;
		return (1);
	}

	return (kn->kn_fflags != 0);
}

/*
 * Called when the process forked. It mostly does the same as the
 * knote(), activating all knotes registered to be activated when the
 * process forked. Additionally, for each knote attached to the
 * parent, check whether user wants to track the new process. If so
 * attach a new knote to it, and immediately report an event with the
 * child's pid.
 */
void
knote_fork(struct knlist *list, int pid)
{
	struct kqueue *kq;
	struct knote *kn;
	struct kevent kev;
	int error;

	MPASS(list != NULL);
	KNL_ASSERT_LOCKED(list);
	if (SLIST_EMPTY(&list->kl_list))
		return;

	memset(&kev, 0, sizeof(kev));
	SLIST_FOREACH(kn, &list->kl_list, kn_selnext) {
		kq = kn->kn_kq;
		KQ_LOCK(kq);
		if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) {
			KQ_UNLOCK(kq);
			continue;
		}

		/*
		 * The same as knote(), activate the event.
		 */
		if ((kn->kn_sfflags & NOTE_TRACK) == 0) {
			if (kn->kn_fop->f_event(kn, NOTE_FORK))
				KNOTE_ACTIVATE(kn, 1);
			KQ_UNLOCK(kq);
			continue;
		}

		/*
		 * The NOTE_TRACK case. In addition to the activation
		 * of the event, we need to register new events to
		 * track the child. Drop the locks in preparation for
		 * the call to kqueue_register().
		 */
		kn_enter_flux(kn);
		KQ_UNLOCK(kq);
		list->kl_unlock(list->kl_lockarg);

		/*
		 * Activate existing knote and register tracking knotes with
		 * new process.
		 *
		 * First register a knote to get just the child notice. This
		 * must be a separate note from a potential NOTE_EXIT
		 * notification since both NOTE_CHILD and NOTE_EXIT are defined
		 * to use the data field (in conflicting ways).
		 */
		kev.ident = pid;
		kev.filter = kn->kn_filter;
		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_ONESHOT |
		    EV_FLAG2;
		kev.fflags = kn->kn_sfflags;
		kev.data = kn->kn_id;		/* parent */
		kev.udata = kn->kn_kevent.udata;/* preserve udata */
		error = kqueue_register(kq, &kev, NULL, M_NOWAIT);
		if (error)
			kn->kn_fflags |= NOTE_TRACKERR;

		/*
		 * Then register another knote to track other potential events
		 * from the new process.
		 */
		kev.ident = pid;
		kev.filter = kn->kn_filter;
		kev.flags = kn->kn_flags | EV_ADD | EV_ENABLE | EV_FLAG1;
		kev.fflags = kn->kn_sfflags;
		kev.data = kn->kn_id;		/* parent */
		kev.udata = kn->kn_kevent.udata;/* preserve udata */
		error = kqueue_register(kq, &kev, NULL, M_NOWAIT);
		if (error)
			kn->kn_fflags |= NOTE_TRACKERR;
		if (kn->kn_fop->f_event(kn, NOTE_FORK))
			KNOTE_ACTIVATE(kn, 0);
		list->kl_lock(list->kl_lockarg);
		KQ_LOCK(kq);
		kn_leave_flux(kn);
		KQ_UNLOCK_FLUX(kq);
	}
}

/*
 * XXX: EVFILT_TIMER should perhaps live in kern_time.c beside the
 * interval timer support code.
 */

#define NOTE_TIMER_PRECMASK						\
    (NOTE_SECONDS | NOTE_MSECONDS | NOTE_USECONDS | NOTE_NSECONDS)

static sbintime_t
timer2sbintime(int64_t data, int flags)
{
	int64_t secs;

        /*
         * Macros for converting to the fractional second portion of an
         * sbintime_t using 64bit multiplication to improve precision.
         */
#define NS_TO_SBT(ns) (((ns) * (((uint64_t)1 << 63) / 500000000)) >> 32)
#define US_TO_SBT(us) (((us) * (((uint64_t)1 << 63) / 500000)) >> 32)
#define MS_TO_SBT(ms) (((ms) * (((uint64_t)1 << 63) / 500)) >> 32)
	switch (flags & NOTE_TIMER_PRECMASK) {
	case NOTE_SECONDS:
#ifdef __LP64__
		if (data > (SBT_MAX / SBT_1S))
			return (SBT_MAX);
#endif
		return ((sbintime_t)data << 32);
	case NOTE_MSECONDS: /* FALLTHROUGH */
	case 0:
		if (data >= 1000) {
			secs = data / 1000;
#ifdef __LP64__
			if (secs > (SBT_MAX / SBT_1S))
				return (SBT_MAX);
#endif
			return (secs << 32 | MS_TO_SBT(data % 1000));
		}
		return (MS_TO_SBT(data));
	case NOTE_USECONDS:
		if (data >= 1000000) {
			secs = data / 1000000;
#ifdef __LP64__
			if (secs > (SBT_MAX / SBT_1S))
				return (SBT_MAX);
#endif
			return (secs << 32 | US_TO_SBT(data % 1000000));
		}
		return (US_TO_SBT(data));
	case NOTE_NSECONDS:
		if (data >= 1000000000) {
			secs = data / 1000000000;
#ifdef __LP64__
			if (secs > (SBT_MAX / SBT_1S))
				return (SBT_MAX);
#endif
			return (secs << 32 | NS_TO_SBT(data % 1000000000));
		}
		return (NS_TO_SBT(data));
	default:
		break;
	}
	return (-1);
}

struct kq_timer_cb_data {
	struct callout c;
	struct proc *p;
	struct knote *kn;
	int cpuid;
	int flags;
	TAILQ_ENTRY(kq_timer_cb_data) link;
	sbintime_t next;	/* next timer event fires at */
	sbintime_t to;		/* precalculated timer period, 0 for abs */
};

#define	KQ_TIMER_CB_ENQUEUED	0x01

static void
kqtimer_sched_callout(struct kq_timer_cb_data *kc)
{
	callout_reset_sbt_on(&kc->c, kc->next, 0, filt_timerexpire, kc->kn,
	    kc->cpuid, C_ABSOLUTE);
}

void
kqtimer_proc_continue(struct proc *p)
{
	struct kq_timer_cb_data *kc, *kc1;
	struct bintime bt;
	sbintime_t now;

	PROC_LOCK_ASSERT(p, MA_OWNED);

	getboottimebin(&bt);
	now = bttosbt(bt);

	TAILQ_FOREACH_SAFE(kc, &p->p_kqtim_stop, link, kc1) {
		TAILQ_REMOVE(&p->p_kqtim_stop, kc, link);
		kc->flags &= ~KQ_TIMER_CB_ENQUEUED;
		if (kc->next <= now)
			filt_timerexpire_l(kc->kn, true);
		else
			kqtimer_sched_callout(kc);
	}
}

static void
filt_timerexpire_l(struct knote *kn, bool proc_locked)
{
	struct kq_timer_cb_data *kc;
	struct proc *p;
	uint64_t delta;
	sbintime_t now;

	kc = kn->kn_ptr.p_v;

	if ((kn->kn_flags & EV_ONESHOT) != 0 || kc->to == 0) {
		kn->kn_data++;
		KNOTE_ACTIVATE(kn, 0);
		return;
	}

	now = sbinuptime();
	if (now >= kc->next) {
		delta = (now - kc->next) / kc->to;
		if (delta == 0)
			delta = 1;
		kn->kn_data += delta;
		kc->next += delta * kc->to;
		if (now >= kc->next)	/* overflow */
			kc->next = now + kc->to;
		KNOTE_ACTIVATE(kn, 0);	/* XXX - handle locking */
	}

	/*
	 * Initial check for stopped kc->p is racy.  It is fine to
	 * miss the set of the stop flags, at worst we would schedule
	 * one more callout.  On the other hand, it is not fine to not
	 * schedule when we we missed clearing of the flags, we
	 * recheck them under the lock and observe consistent state.
	 */
	p = kc->p;
	if (P_SHOULDSTOP(p) || P_KILLED(p)) {
		if (!proc_locked)
			PROC_LOCK(p);
		if (P_SHOULDSTOP(p) || P_KILLED(p)) {
			if ((kc->flags & KQ_TIMER_CB_ENQUEUED) == 0) {
				kc->flags |= KQ_TIMER_CB_ENQUEUED;
				TAILQ_INSERT_TAIL(&p->p_kqtim_stop, kc, link);
			}
			if (!proc_locked)
				PROC_UNLOCK(p);
			return;
		}
		if (!proc_locked)
			PROC_UNLOCK(p);
	}
	kqtimer_sched_callout(kc);
}

static void
filt_timerexpire(void *knx)
{
	filt_timerexpire_l(knx, false);
}

/*
 * data contains amount of time to sleep
 */
static int
filt_timervalidate(struct knote *kn, sbintime_t *to)
{
	struct bintime bt;
	sbintime_t sbt;

	if (kn->kn_sdata < 0)
		return (EINVAL);
	if (kn->kn_sdata == 0 && (kn->kn_flags & EV_ONESHOT) == 0)
		kn->kn_sdata = 1;
	/*
	 * The only fflags values supported are the timer unit
	 * (precision) and the absolute time indicator.
	 */
	if ((kn->kn_sfflags & ~(NOTE_TIMER_PRECMASK | NOTE_ABSTIME)) != 0)
		return (EINVAL);

	*to = timer2sbintime(kn->kn_sdata, kn->kn_sfflags);
	if (*to < 0)
		return (EINVAL);
	if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) {
		getboottimebin(&bt);
		sbt = bttosbt(bt);
		*to = MAX(0, *to - sbt);
	}
	return (0);
}

static int
filt_timerattach(struct knote *kn)
{
	struct kq_timer_cb_data *kc;
	sbintime_t to;
	int error;

	to = -1;
	error = filt_timervalidate(kn, &to);
	if (error != 0)
		return (error);
	KASSERT(to > 0 || (kn->kn_flags & EV_ONESHOT) != 0 ||
	    (kn->kn_sfflags & NOTE_ABSTIME) != 0,
	    ("%s: periodic timer has a calculated zero timeout", __func__));
	KASSERT(to >= 0,
	    ("%s: timer has a calculated negative timeout", __func__));

	if (atomic_fetchadd_int(&kq_ncallouts, 1) + 1 > kq_calloutmax) {
		atomic_subtract_int(&kq_ncallouts, 1);
		return (ENOMEM);
	}

	if ((kn->kn_sfflags & NOTE_ABSTIME) == 0)
		kn->kn_flags |= EV_CLEAR;	/* automatically set */
	kn->kn_status &= ~KN_DETACHED;		/* knlist_add clears it */
	kn->kn_ptr.p_v = kc = malloc(sizeof(*kc), M_KQUEUE, M_WAITOK);
	kc->kn = kn;
	kc->p = curproc;
	kc->cpuid = PCPU_GET(cpuid);
	kc->flags = 0;
	callout_init(&kc->c, 1);
	filt_timerstart(kn, to);

	return (0);
}

static void
filt_timerstart(struct knote *kn, sbintime_t to)
{
	struct kq_timer_cb_data *kc;

	kc = kn->kn_ptr.p_v;
	if ((kn->kn_sfflags & NOTE_ABSTIME) != 0) {
		kc->next = to;
		kc->to = 0;
	} else {
		kc->next = to + sbinuptime();
		kc->to = to;
	}
	kqtimer_sched_callout(kc);
}

static void
filt_timerdetach(struct knote *kn)
{
	struct kq_timer_cb_data *kc;
	unsigned int old __unused;
	bool pending;

	kc = kn->kn_ptr.p_v;
	do {
		callout_drain(&kc->c);

		/*
		 * kqtimer_proc_continue() might have rescheduled this callout.
		 * Double-check, using the process mutex as an interlock.
		 */
		PROC_LOCK(kc->p);
		if ((kc->flags & KQ_TIMER_CB_ENQUEUED) != 0) {
			kc->flags &= ~KQ_TIMER_CB_ENQUEUED;
			TAILQ_REMOVE(&kc->p->p_kqtim_stop, kc, link);
		}
		pending = callout_pending(&kc->c);
		PROC_UNLOCK(kc->p);
	} while (pending);
	free(kc, M_KQUEUE);
	old = atomic_fetchadd_int(&kq_ncallouts, -1);
	KASSERT(old > 0, ("Number of callouts cannot become negative"));
	kn->kn_status |= KN_DETACHED;	/* knlist_remove sets it */
}

static void
filt_timertouch(struct knote *kn, struct kevent *kev, u_long type)
{
	struct kq_timer_cb_data *kc;
	struct kqueue *kq;
	sbintime_t to;
	int error;

	switch (type) {
	case EVENT_REGISTER:
		/* Handle re-added timers that update data/fflags */
		if (kev->flags & EV_ADD) {
			kc = kn->kn_ptr.p_v;

			/* Drain any existing callout. */
			callout_drain(&kc->c);

			/* Throw away any existing undelivered record
			 * of the timer expiration. This is done under
			 * the presumption that if a process is
			 * re-adding this timer with new parameters,
			 * it is no longer interested in what may have
			 * happened under the old parameters. If it is
			 * interested, it can wait for the expiration,
			 * delete the old timer definition, and then
			 * add the new one.
			 *
			 * This has to be done while the kq is locked:
			 *   - if enqueued, dequeue
			 *   - make it no longer active
			 *   - clear the count of expiration events
			 */
			kq = kn->kn_kq;
			KQ_LOCK(kq);
			if (kn->kn_status & KN_QUEUED)
				knote_dequeue(kn);

			kn->kn_status &= ~KN_ACTIVE;
			kn->kn_data = 0;
			KQ_UNLOCK(kq);

			/* Reschedule timer based on new data/fflags */
			kn->kn_sfflags = kev->fflags;
			kn->kn_sdata = kev->data;
			error = filt_timervalidate(kn, &to);
			if (error != 0) {
			  	kn->kn_flags |= EV_ERROR;
				kn->kn_data = error;
			} else
			  	filt_timerstart(kn, to);
		}
		break;

        case EVENT_PROCESS:
		*kev = kn->kn_kevent;
		if (kn->kn_flags & EV_CLEAR) {
			kn->kn_data = 0;
			kn->kn_fflags = 0;
		}
		break;

	default:
		panic("filt_timertouch() - invalid type (%ld)", type);
		break;
	}
}

static int
filt_timer(struct knote *kn, long hint)
{

	return (kn->kn_data != 0);
}

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 kevent *kev, u_long type)
{
	u_int 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:
			kn->kn_sfflags &= kev->fflags;
			break;

		case NOTE_FFOR:
			kn->kn_sfflags |= kev->fflags;
			break;

		case NOTE_FFCOPY:
			kn->kn_sfflags = kev->fflags;
			break;

		default:
			/* XXX Return error? */
			break;
		}
		kn->kn_sdata = kev->data;
		if (kev->flags & EV_CLEAR) {
			kn->kn_hookid = 0;
			kn->kn_data = 0;
			kn->kn_fflags = 0;
		}
		break;

        case EVENT_PROCESS:
		*kev = kn->kn_kevent;
		kev->fflags = 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("filt_usertouch() - invalid type (%ld)", type);
		break;
	}
}

int
sys_kqueue(struct thread *td, struct kqueue_args *uap)
{

	return (kern_kqueue(td, 0, NULL));
}

int
sys_kqueuex(struct thread *td, struct kqueuex_args *uap)
{
	int flags;

	if ((uap->flags & ~(KQUEUE_CLOEXEC)) != 0)
		return (EINVAL);
	flags = 0;
	if ((uap->flags & KQUEUE_CLOEXEC) != 0)
		flags |= O_CLOEXEC;
	return (kern_kqueue(td, flags, NULL));
}

static void
kqueue_init(struct kqueue *kq)
{

	mtx_init(&kq->kq_lock, "kqueue", NULL, MTX_DEF | MTX_DUPOK);
	TAILQ_INIT(&kq->kq_head);
	knlist_init_mtx(&kq->kq_sel.si_note, &kq->kq_lock);
	TASK_INIT(&kq->kq_task, 0, kqueue_task, kq);
}

int
kern_kqueue(struct thread *td, int flags, struct filecaps *fcaps)
{
	struct filedesc *fdp;
	struct kqueue *kq;
	struct file *fp;
	struct ucred *cred;
	int fd, error;

	fdp = td->td_proc->p_fd;
	cred = td->td_ucred;
	if (!chgkqcnt(cred->cr_ruidinfo, 1, lim_cur(td, RLIMIT_KQUEUES)))
		return (ENOMEM);

	error = falloc_caps(td, &fp, &fd, flags, fcaps);
	if (error != 0) {
		chgkqcnt(cred->cr_ruidinfo, -1, 0);
		return (error);
	}

	/* An extra reference on `fp' has been held for us by falloc(). */
	kq = malloc(sizeof *kq, M_KQUEUE, M_WAITOK | M_ZERO);
	kqueue_init(kq);
	kq->kq_fdp = fdp;
	kq->kq_cred = crhold(cred);

	FILEDESC_XLOCK(fdp);
	TAILQ_INSERT_HEAD(&fdp->fd_kqlist, kq, kq_list);
	FILEDESC_XUNLOCK(fdp);

	finit(fp, FREAD | FWRITE, DTYPE_KQUEUE, kq, &kqueueops);
	fdrop(fp, td);

	td->td_retval[0] = fd;
	return (0);
}

struct g_kevent_args {
	int	fd;
	const void *changelist;
	int	nchanges;
	void	*eventlist;
	int	nevents;
	const struct timespec *timeout;
};

int
sys_kevent(struct thread *td, struct kevent_args *uap)
{
	struct kevent_copyops k_ops = {
		.arg = uap,
		.k_copyout = kevent_copyout,
		.k_copyin = kevent_copyin,
		.kevent_size = sizeof(struct kevent),
	};
	struct g_kevent_args gk_args = {
		.fd = uap->fd,
		.changelist = uap->changelist,
		.nchanges = uap->nchanges,
		.eventlist = uap->eventlist,
		.nevents = uap->nevents,
		.timeout = uap->timeout,
	};

	return (kern_kevent_generic(td, &gk_args, &k_ops, "kevent"));
}

static int
kern_kevent_generic(struct thread *td, struct g_kevent_args *uap,
    struct kevent_copyops *k_ops, const char *struct_name)
{
	struct timespec ts, *tsp;
#ifdef KTRACE
	struct kevent *eventlist = uap->eventlist;
#endif
	int error;

	if (uap->timeout != NULL) {
		error = copyin(uap->timeout, &ts, sizeof(ts));
		if (error)
			return (error);
		tsp = &ts;
	} else
		tsp = NULL;

#ifdef KTRACE
	if (KTRPOINT(td, KTR_STRUCT_ARRAY))
		ktrstructarray(struct_name, UIO_USERSPACE, uap->changelist,
		    uap->nchanges, k_ops->kevent_size);
#endif

	error = kern_kevent(td, uap->fd, uap->nchanges, uap->nevents,
	    k_ops, tsp);

#ifdef KTRACE
	if (error == 0 && KTRPOINT(td, KTR_STRUCT_ARRAY))
		ktrstructarray(struct_name, UIO_USERSPACE, eventlist,
		    td->td_retval[0], k_ops->kevent_size);
#endif

	return (error);
}

/*
 * Copy 'count' items into the destination list pointed to by uap->eventlist.
 */
static int
kevent_copyout(void *arg, struct kevent *kevp, int count)
{
	struct kevent_args *uap;
	int error;

	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
	uap = (struct kevent_args *)arg;

	error = copyout(kevp, uap->eventlist, count * sizeof *kevp);
	if (error == 0)
		uap->eventlist += count;
	return (error);
}

/*
 * Copy 'count' items from the list pointed to by uap->changelist.
 */
static int
kevent_copyin(void *arg, struct kevent *kevp, int count)
{
	struct kevent_args *uap;
	int error;

	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
	uap = (struct kevent_args *)arg;

	error = copyin(uap->changelist, kevp, count * sizeof *kevp);
	if (error == 0)
		uap->changelist += count;
	return (error);
}

#ifdef COMPAT_FREEBSD11
static int
kevent11_copyout(void *arg, struct kevent *kevp, int count)
{
	struct freebsd11_kevent_args *uap;
	struct freebsd11_kevent kev11;
	int error, i;

	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
	uap = (struct freebsd11_kevent_args *)arg;

	for (i = 0; i < count; i++) {
		kev11.ident = kevp->ident;
		kev11.filter = kevp->filter;
		kev11.flags = kevp->flags;
		kev11.fflags = kevp->fflags;
		kev11.data = kevp->data;
		kev11.udata = kevp->udata;
		error = copyout(&kev11, uap->eventlist, sizeof(kev11));
		if (error != 0)
			break;
		uap->eventlist++;
		kevp++;
	}
	return (error);
}

/*
 * Copy 'count' items from the list pointed to by uap->changelist.
 */
static int
kevent11_copyin(void *arg, struct kevent *kevp, int count)
{
	struct freebsd11_kevent_args *uap;
	struct freebsd11_kevent kev11;
	int error, i;

	KASSERT(count <= KQ_NEVENTS, ("count (%d) > KQ_NEVENTS", count));
	uap = (struct freebsd11_kevent_args *)arg;

	for (i = 0; i < count; i++) {
		error = copyin(uap->changelist, &kev11, sizeof(kev11));
		if (error != 0)
			break;
		kevp->ident = kev11.ident;
		kevp->filter = kev11.filter;
		kevp->flags = kev11.flags;
		kevp->fflags = kev11.fflags;
		kevp->data = (uintptr_t)kev11.data;
		kevp->udata = kev11.udata;
		bzero(&kevp->ext, sizeof(kevp->ext));
		uap->changelist++;
		kevp++;
	}
	return (error);
}

int
freebsd11_kevent(struct thread *td, struct freebsd11_kevent_args *uap)
{
	struct kevent_copyops k_ops = {
		.arg = uap,
		.k_copyout = kevent11_copyout,
		.k_copyin = kevent11_copyin,
		.kevent_size = sizeof(struct freebsd11_kevent),
	};
	struct g_kevent_args gk_args = {
		.fd = uap->fd,
		.changelist = uap->changelist,
		.nchanges = uap->nchanges,
		.eventlist = uap->eventlist,
		.nevents = uap->nevents,
		.timeout = uap->timeout,
	};

	return (kern_kevent_generic(td, &gk_args, &k_ops, "freebsd11_kevent"));
}
#endif

int
kern_kevent(struct thread *td, int fd, int nchanges, int nevents,
    struct kevent_copyops *k_ops, const struct timespec *timeout)
{
	cap_rights_t rights;
	struct file *fp;
	int error;

	cap_rights_init_zero(&rights);
	if (nchanges > 0)
		cap_rights_set_one(&rights, CAP_KQUEUE_CHANGE);
	if (nevents > 0)
		cap_rights_set_one(&rights, CAP_KQUEUE_EVENT);
	error = fget(td, fd, &rights, &fp);
	if (error != 0)
		return (error);

	error = kern_kevent_fp(td, fp, nchanges, nevents, k_ops, timeout);
	fdrop(fp, td);

	return (error);
}

static int
kqueue_kevent(struct kqueue *kq, struct thread *td, int nchanges, int nevents,
    struct kevent_copyops *k_ops, const struct timespec *timeout)
{
	struct kevent keva[KQ_NEVENTS];
	struct kevent *kevp, *changes;
	int i, n, nerrors, error;

	if (nchanges < 0)
		return (EINVAL);

	nerrors = 0;
	while (nchanges > 0) {
		n = nchanges > KQ_NEVENTS ? KQ_NEVENTS : nchanges;
		error = k_ops->k_copyin(k_ops->arg, keva, n);
		if (error)
			return (error);
		changes = keva;
		for (i = 0; i < n; i++) {
			kevp = &changes[i];
			if (!kevp->filter)
				continue;
			kevp->flags &= ~EV_SYSFLAGS;
			error = kqueue_register(kq, kevp, td, M_WAITOK);
			if (error || (kevp->flags & EV_RECEIPT)) {
				if (nevents == 0)
					return (error);
				kevp->flags = EV_ERROR;
				kevp->data = error;
				(void)k_ops->k_copyout(k_ops->arg, kevp, 1);
				nevents--;
				nerrors++;
			}
		}
		nchanges -= n;
	}
	if (nerrors) {
		td->td_retval[0] = nerrors;
		return (0);
	}

	return (kqueue_scan(kq, nevents, k_ops, timeout, keva, td));
}

int
kern_kevent_fp(struct thread *td, struct file *fp, int nchanges, int nevents,
    struct kevent_copyops *k_ops, const struct timespec *timeout)
{
	struct kqueue *kq;
	int error;

	error = kqueue_acquire(fp, &kq);
	if (error != 0)
		return (error);
	error = kqueue_kevent(kq, td, nchanges, nevents, k_ops, timeout);
	kqueue_release(kq, 0);
	return (error);
}

/*
 * Performs a kevent() call on a temporarily created kqueue. This can be
 * used to perform one-shot polling, similar to poll() and select().
 */
int
kern_kevent_anonymous(struct thread *td, int nevents,
    struct kevent_copyops *k_ops)
{
	struct kqueue kq = {};
	int error;

	kqueue_init(&kq);
	kq.kq_refcnt = 1;
	error = kqueue_kevent(&kq, td, nevents, nevents, k_ops, NULL);
	kqueue_drain(&kq, td);
	kqueue_destroy(&kq);
	return (error);
}

int
kqueue_add_filteropts(int filt, const struct filterops *filtops)
{
	int error;

	error = 0;
	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0) {
		printf(
"trying to add a filterop that is out of range: %d is beyond %d\n",
		    ~filt, EVFILT_SYSCOUNT);
		return EINVAL;
	}
	mtx_lock(&filterops_lock);
	if (sysfilt_ops[~filt].for_fop != &null_filtops &&
	    sysfilt_ops[~filt].for_fop != NULL)
		error = EEXIST;
	else {
		sysfilt_ops[~filt].for_fop = filtops;
		sysfilt_ops[~filt].for_refcnt = 0;
	}
	mtx_unlock(&filterops_lock);

	return (error);
}

int
kqueue_del_filteropts(int filt)
{
	int error;

	error = 0;
	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
		return EINVAL;

	mtx_lock(&filterops_lock);
	if (sysfilt_ops[~filt].for_fop == &null_filtops ||
	    sysfilt_ops[~filt].for_fop == NULL)
		error = EINVAL;
	else if (sysfilt_ops[~filt].for_refcnt != 0)
		error = EBUSY;
	else {
		sysfilt_ops[~filt].for_fop = &null_filtops;
		sysfilt_ops[~filt].for_refcnt = 0;
	}
	mtx_unlock(&filterops_lock);

	return error;
}

static const struct filterops *
kqueue_fo_find(int filt)
{

	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
		return NULL;

	if (sysfilt_ops[~filt].for_nolock)
		return sysfilt_ops[~filt].for_fop;

	mtx_lock(&filterops_lock);
	sysfilt_ops[~filt].for_refcnt++;
	if (sysfilt_ops[~filt].for_fop == NULL)
		sysfilt_ops[~filt].for_fop = &null_filtops;
	mtx_unlock(&filterops_lock);

	return sysfilt_ops[~filt].for_fop;
}

static void
kqueue_fo_release(int filt)
{

	if (filt > 0 || filt + EVFILT_SYSCOUNT < 0)
		return;

	if (sysfilt_ops[~filt].for_nolock)
		return;

	mtx_lock(&filterops_lock);
	KASSERT(sysfilt_ops[~filt].for_refcnt > 0,
	    ("filter object refcount not valid on release"));
	sysfilt_ops[~filt].for_refcnt--;
	mtx_unlock(&filterops_lock);
}

/*
 * A ref to kq (obtained via kqueue_acquire) must be held.
 */
static int
kqueue_register(struct kqueue *kq, struct kevent *kev, struct thread *td,
    int mflag)
{
	const struct filterops *fops;
	struct file *fp;
	struct knote *kn, *tkn;
	struct knlist *knl;
	int error, filt, event;
	int haskqglobal, filedesc_unlock;

	if ((kev->flags & (EV_ENABLE | EV_DISABLE)) == (EV_ENABLE | EV_DISABLE))
		return (EINVAL);

	fp = NULL;
	kn = NULL;
	knl = NULL;
	error = 0;
	haskqglobal = 0;
	filedesc_unlock = 0;

	filt = kev->filter;
	fops = kqueue_fo_find(filt);
	if (fops == NULL)
		return EINVAL;

	if (kev->flags & EV_ADD) {
		/* Reject an invalid flag pair early */
		if (kev->flags & EV_KEEPUDATA) {
			tkn = NULL;
			error = EINVAL;
			goto done;
		}

		/*
		 * Prevent waiting with locks.  Non-sleepable
		 * allocation failures are handled in the loop, only
		 * if the spare knote appears to be actually required.
		 */
		tkn = knote_alloc(mflag);
	} else {
		tkn = NULL;
	}

findkn:
	if (fops->f_isfd) {
		KASSERT(td != NULL, ("td is NULL"));
		if (kev->ident > INT_MAX)
			error = EBADF;
		else
			error = fget(td, kev->ident, &cap_event_rights, &fp);
		if (error)
			goto done;

		if ((kev->flags & EV_ADD) == EV_ADD && kqueue_expand(kq, fops,
		    kev->ident, M_NOWAIT) != 0) {
			/* try again */
			fdrop(fp, td);
			fp = NULL;
			error = kqueue_expand(kq, fops, kev->ident, mflag);
			if (error)
				goto done;
			goto findkn;
		}

		if (fp->f_type == DTYPE_KQUEUE) {
			/*
			 * If we add some intelligence about what we are doing,
			 * we should be able to support events on ourselves.
			 * We need to know when we are doing this to prevent
			 * getting both the knlist lock and the kq lock since
			 * they are the same thing.
			 */
			if (fp->f_data == kq) {
				error = EINVAL;
				goto done;
			}

			/*
			 * Pre-lock the filedesc before the global
			 * lock mutex, see the comment in
			 * kqueue_close().
			 */
			FILEDESC_XLOCK(td->td_proc->p_fd);
			filedesc_unlock = 1;
			KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
		}

		KQ_LOCK(kq);
		if (kev->ident < kq->kq_knlistsize) {
			SLIST_FOREACH(kn, &kq->kq_knlist[kev->ident], kn_link)
				if (kev->filter == kn->kn_filter)
					break;
		}
	} else {
		if ((kev->flags & EV_ADD) == EV_ADD) {
			error = kqueue_expand(kq, fops, kev->ident, mflag);
			if (error != 0)
				goto done;
		}

		KQ_LOCK(kq);

		/*
		 * If possible, find an existing knote to use for this kevent.
		 */
		if (kev->filter == EVFILT_PROC &&
		    (kev->flags & (EV_FLAG1 | EV_FLAG2)) != 0) {
			/* This is an internal creation of a process tracking
			 * note. Don't attempt to coalesce this with an
			 * existing note.
			 */
			;
		} else if (kq->kq_knhashmask != 0) {
			struct klist *list;

			list = &kq->kq_knhash[
			    KN_HASH((u_long)kev->ident, kq->kq_knhashmask)];
			SLIST_FOREACH(kn, list, kn_link)
				if (kev->ident == kn->kn_id &&
				    kev->filter == kn->kn_filter)
					break;
		}
	}

	/* knote is in the process of changing, wait for it to stabilize. */
	if (kn != NULL && kn_in_flux(kn)) {
		KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
		if (filedesc_unlock) {
			FILEDESC_XUNLOCK(td->td_proc->p_fd);
			filedesc_unlock = 0;
		}
		kq->kq_state |= KQ_FLUXWAIT;
		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqflxwt", 0);
		if (fp != NULL) {
			fdrop(fp, td);
			fp = NULL;
		}
		goto findkn;
	}

	/*
	 * kn now contains the matching knote, or NULL if no match
	 */
	if (kn == NULL) {
		if (kev->flags & EV_ADD) {
			kn = tkn;
			tkn = NULL;
			if (kn == NULL) {
				KQ_UNLOCK(kq);
				error = ENOMEM;
				goto done;
			}
			kn->kn_fp = fp;
			kn->kn_kq = kq;
			kn->kn_fop = fops;
			/*
			 * apply reference counts to knote structure, and
			 * do not release it at the end of this routine.
			 */
			fops = NULL;
			fp = NULL;

			kn->kn_sfflags = kev->fflags;
			kn->kn_sdata = kev->data;
			kev->fflags = 0;
			kev->data = 0;
			kn->kn_kevent = *kev;
			kn->kn_kevent.flags &= ~(EV_ADD | EV_DELETE |
			    EV_ENABLE | EV_DISABLE | EV_FORCEONESHOT);
			kn->kn_status = KN_DETACHED;
			if ((kev->flags & EV_DISABLE) != 0)
				kn->kn_status |= KN_DISABLED;
			kn_enter_flux(kn);

			error = knote_attach(kn, kq);
			KQ_UNLOCK(kq);
			if (error != 0) {
				tkn = kn;
				goto done;
			}

			if ((error = kn->kn_fop->f_attach(kn)) != 0) {
				knote_drop_detached(kn, td);
				goto done;
			}
			knl = kn_list_lock(kn);
			goto done_ev_add;
		} else {
			/* No matching knote and the EV_ADD flag is not set. */
			KQ_UNLOCK(kq);
			error = ENOENT;
			goto done;
		}
	}

	if (kev->flags & EV_DELETE) {
		kn_enter_flux(kn);
		KQ_UNLOCK(kq);
		knote_drop(kn, td);
		goto done;
	}

	if (kev->flags & EV_FORCEONESHOT) {
		kn->kn_flags |= EV_ONESHOT;
		KNOTE_ACTIVATE(kn, 1);
	}

	if ((kev->flags & EV_ENABLE) != 0)
		kn->kn_status &= ~KN_DISABLED;
	else if ((kev->flags & EV_DISABLE) != 0)
		kn->kn_status |= KN_DISABLED;

	/*
	 * The user may change some filter values after the initial EV_ADD,
	 * but doing so will not reset any filter which has already been
	 * triggered.
	 */
	kn->kn_status |= KN_SCAN;
	kn_enter_flux(kn);
	KQ_UNLOCK(kq);
	knl = kn_list_lock(kn);
	if ((kev->flags & EV_KEEPUDATA) == 0)
		kn->kn_kevent.udata = kev->udata;
	if (!fops->f_isfd && fops->f_touch != NULL) {
		fops->f_touch(kn, kev, EVENT_REGISTER);
	} else {
		kn->kn_sfflags = kev->fflags;
		kn->kn_sdata = kev->data;
	}

done_ev_add:
	/*
	 * We can get here with kn->kn_knlist == NULL.  This can happen when
	 * the initial attach event decides that the event is "completed"
	 * already, e.g., filt_procattach() is called on a zombie process.  It
	 * will call filt_proc() which will remove it from the list, and NULL
	 * kn_knlist.
	 *
	 * KN_DISABLED will be stable while the knote is in flux, so the
	 * unlocked read will not race with an update.
	 */
	if ((kn->kn_status & KN_DISABLED) == 0)
		event = kn->kn_fop->f_event(kn, 0);
	else
		event = 0;

	KQ_LOCK(kq);
	if (event)
		kn->kn_status |= KN_ACTIVE;
	if ((kn->kn_status & (KN_ACTIVE | KN_DISABLED | KN_QUEUED)) ==
	    KN_ACTIVE)
		knote_enqueue(kn);
	kn->kn_status &= ~KN_SCAN;
	kn_leave_flux(kn);
	kn_list_unlock(knl);
	KQ_UNLOCK_FLUX(kq);

done:
	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
	if (filedesc_unlock)
		FILEDESC_XUNLOCK(td->td_proc->p_fd);
	if (fp != NULL)
		fdrop(fp, td);
	knote_free(tkn);
	if (fops != NULL)
		kqueue_fo_release(filt);
	return (error);
}

static int
kqueue_acquire(struct file *fp, struct kqueue **kqp)
{
	int error;
	struct kqueue *kq;

	error = 0;

	kq = fp->f_data;
	if (fp->f_type != DTYPE_KQUEUE || kq == NULL)
		return (EBADF);
	*kqp = kq;
	KQ_LOCK(kq);
	if ((kq->kq_state & KQ_CLOSING) == KQ_CLOSING) {
		KQ_UNLOCK(kq);
		return (EBADF);
	}
	kq->kq_refcnt++;
	KQ_UNLOCK(kq);

	return error;
}

static void
kqueue_release(struct kqueue *kq, int locked)
{
	if (locked)
		KQ_OWNED(kq);
	else
		KQ_LOCK(kq);
	kq->kq_refcnt--;
	if (kq->kq_refcnt == 1)
		wakeup(&kq->kq_refcnt);
	if (!locked)
		KQ_UNLOCK(kq);
}

static void
ast_kqueue(struct thread *td, int tda __unused)
{
	taskqueue_quiesce(taskqueue_kqueue_ctx);
}

static void
kqueue_schedtask(struct kqueue *kq)
{
	KQ_OWNED(kq);
	KASSERT(((kq->kq_state & KQ_TASKDRAIN) != KQ_TASKDRAIN),
	    ("scheduling kqueue task while draining"));

	if ((kq->kq_state & KQ_TASKSCHED) != KQ_TASKSCHED) {
		taskqueue_enqueue(taskqueue_kqueue_ctx, &kq->kq_task);
		kq->kq_state |= KQ_TASKSCHED;
		ast_sched(curthread, TDA_KQUEUE);
	}
}

/*
 * Expand the kq to make sure we have storage for fops/ident pair.
 *
 * Return 0 on success (or no work necessary), return errno on failure.
 */
static int
kqueue_expand(struct kqueue *kq, const struct filterops *fops, uintptr_t ident,
    int mflag)
{
	struct klist *list, *tmp_knhash, *to_free;
	u_long tmp_knhashmask;
	int error, fd, size;

	KQ_NOTOWNED(kq);

	error = 0;
	to_free = NULL;
	if (fops->f_isfd) {
		fd = ident;
		if (kq->kq_knlistsize <= fd) {
			size = kq->kq_knlistsize;
			while (size <= fd)
				size += KQEXTENT;
			list = malloc(size * sizeof(*list), M_KQUEUE, mflag);
			if (list == NULL)
				return ENOMEM;
			KQ_LOCK(kq);
			if ((kq->kq_state & KQ_CLOSING) != 0) {
				to_free = list;
				error = EBADF;
			} else if (kq->kq_knlistsize > fd) {
				to_free = list;
			} else {
				if (kq->kq_knlist != NULL) {
					bcopy(kq->kq_knlist, list,
					    kq->kq_knlistsize * sizeof(*list));
					to_free = kq->kq_knlist;
					kq->kq_knlist = NULL;
				}
				bzero((caddr_t)list +
				    kq->kq_knlistsize * sizeof(*list),
				    (size - kq->kq_knlistsize) * sizeof(*list));
				kq->kq_knlistsize = size;
				kq->kq_knlist = list;
			}
			KQ_UNLOCK(kq);
		}
	} else {
		if (kq->kq_knhashmask == 0) {
			tmp_knhash = hashinit_flags(KN_HASHSIZE, M_KQUEUE,
			    &tmp_knhashmask, (mflag & M_WAITOK) != 0 ?
			    HASH_WAITOK : HASH_NOWAIT);
			if (tmp_knhash == NULL)
				return (ENOMEM);
			KQ_LOCK(kq);
			if ((kq->kq_state & KQ_CLOSING) != 0) {
				to_free = tmp_knhash;
				error = EBADF;
			} else if (kq->kq_knhashmask == 0) {
				kq->kq_knhash = tmp_knhash;
				kq->kq_knhashmask = tmp_knhashmask;
			} else {
				to_free = tmp_knhash;
			}
			KQ_UNLOCK(kq);
		}
	}
	free(to_free, M_KQUEUE);

	KQ_NOTOWNED(kq);
	return (error);
}

static void
kqueue_task(void *arg, int pending)
{
	struct kqueue *kq;
	int haskqglobal;

	haskqglobal = 0;
	kq = arg;

	KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
	KQ_LOCK(kq);

	KNOTE_LOCKED(&kq->kq_sel.si_note, 0);

	kq->kq_state &= ~KQ_TASKSCHED;
	if ((kq->kq_state & KQ_TASKDRAIN) == KQ_TASKDRAIN) {
		wakeup(&kq->kq_state);
	}
	KQ_UNLOCK(kq);
	KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
}

/*
 * Scan, update kn_data (if not ONESHOT), and copyout triggered events.
 * We treat KN_MARKER knotes as if they are in flux.
 */
static int
kqueue_scan(struct kqueue *kq, int maxevents, struct kevent_copyops *k_ops,
    const struct timespec *tsp, struct kevent *keva, struct thread *td)
{
	struct kevent *kevp;
	struct knote *kn, *marker;
	struct knlist *knl;
	sbintime_t asbt, rsbt;
	int count, error, haskqglobal, influx, nkev, touch;

	count = maxevents;
	nkev = 0;
	error = 0;
	haskqglobal = 0;

	if (maxevents == 0)
		goto done_nl;
	if (maxevents < 0) {
		error = EINVAL;
		goto done_nl;
	}

	rsbt = 0;
	if (tsp != NULL) {
		if (!timespecvalid_interval(tsp)) {
			error = EINVAL;
			goto done_nl;
		}
		if (timespecisset(tsp)) {
			if (tsp->tv_sec <= INT32_MAX) {
				rsbt = tstosbt(*tsp);
				if (TIMESEL(&asbt, rsbt))
					asbt += tc_tick_sbt;
				if (asbt <= SBT_MAX - rsbt)
					asbt += rsbt;
				else
					asbt = 0;
				rsbt >>= tc_precexp;
			} else
				asbt = 0;
		} else
			asbt = -1;
	} else
		asbt = 0;
	marker = knote_alloc(M_WAITOK);
	marker->kn_status = KN_MARKER;
	KQ_LOCK(kq);

retry:
	kevp = keva;
	if (kq->kq_count == 0) {
		if (asbt == -1) {
			error = EWOULDBLOCK;
		} else {
			kq->kq_state |= KQ_SLEEP;
			error = msleep_sbt(kq, &kq->kq_lock, PSOCK | PCATCH,
			    "kqread", asbt, rsbt, C_ABSOLUTE);
		}
		if (error == 0)
			goto retry;
		/* don't restart after signals... */
		if (error == ERESTART)
			error = EINTR;
		else if (error == EWOULDBLOCK)
			error = 0;
		goto done;
	}

	TAILQ_INSERT_TAIL(&kq->kq_head, marker, kn_tqe);
	influx = 0;
	while (count) {
		KQ_OWNED(kq);
		kn = TAILQ_FIRST(&kq->kq_head);

		if ((kn->kn_status == KN_MARKER && kn != marker) ||
		    kn_in_flux(kn)) {
			if (influx) {
				influx = 0;
				KQ_FLUX_WAKEUP(kq);
			}
			kq->kq_state |= KQ_FLUXWAIT;
			error = msleep(kq, &kq->kq_lock, PSOCK,
			    "kqflxwt", 0);
			continue;
		}

		TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
		if ((kn->kn_status & KN_DISABLED) == KN_DISABLED) {
			kn->kn_status &= ~KN_QUEUED;
			kq->kq_count--;
			continue;
		}
		if (kn == marker) {
			KQ_FLUX_WAKEUP(kq);
			if (count == maxevents)
				goto retry;
			goto done;
		}
		KASSERT(!kn_in_flux(kn),
		    ("knote %p is unexpectedly in flux", kn));

		if ((kn->kn_flags & EV_DROP) == EV_DROP) {
			kn->kn_status &= ~KN_QUEUED;
			kn_enter_flux(kn);
			kq->kq_count--;
			KQ_UNLOCK(kq);
			/*
			 * We don't need to lock the list since we've
			 * marked it as in flux.
			 */
			knote_drop(kn, td);
			KQ_LOCK(kq);
			continue;
		} else if ((kn->kn_flags & EV_ONESHOT) == EV_ONESHOT) {
			kn->kn_status &= ~KN_QUEUED;
			kn_enter_flux(kn);
			kq->kq_count--;
			KQ_UNLOCK(kq);
			/*
			 * We don't need to lock the list since we've
			 * marked the knote as being in flux.
			 */
			*kevp = kn->kn_kevent;
			knote_drop(kn, td);
			KQ_LOCK(kq);
			kn = NULL;
		} else {
			kn->kn_status |= KN_SCAN;
			kn_enter_flux(kn);
			KQ_UNLOCK(kq);
			if ((kn->kn_status & KN_KQUEUE) == KN_KQUEUE)
				KQ_GLOBAL_LOCK(&kq_global, haskqglobal);
			knl = kn_list_lock(kn);
			if (kn->kn_fop->f_event(kn, 0) == 0) {
				KQ_LOCK(kq);
				KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE |
				    KN_SCAN);
				kn_leave_flux(kn);
				kq->kq_count--;
				kn_list_unlock(knl);
				influx = 1;
				continue;
			}
			touch = (!kn->kn_fop->f_isfd &&
			    kn->kn_fop->f_touch != NULL);
			if (touch)
				kn->kn_fop->f_touch(kn, kevp, EVENT_PROCESS);
			else
				*kevp = kn->kn_kevent;
			KQ_LOCK(kq);
			KQ_GLOBAL_UNLOCK(&kq_global, haskqglobal);
			if (kn->kn_flags & (EV_CLEAR | EV_DISPATCH)) {
				/*
				 * Manually clear knotes who weren't
				 * 'touch'ed.
				 */
				if (touch == 0 && kn->kn_flags & EV_CLEAR) {
					kn->kn_data = 0;
					kn->kn_fflags = 0;
				}
				if (kn->kn_flags & EV_DISPATCH)
					kn->kn_status |= KN_DISABLED;
				kn->kn_status &= ~(KN_QUEUED | KN_ACTIVE);
				kq->kq_count--;
			} else
				TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);

			kn->kn_status &= ~KN_SCAN;
			kn_leave_flux(kn);
			kn_list_unlock(knl);
			influx = 1;
		}

		/* we are returning a copy to the user */
		kevp++;
		nkev++;
		count--;

		if (nkev == KQ_NEVENTS) {
			influx = 0;
			KQ_UNLOCK_FLUX(kq);
			error = k_ops->k_copyout(k_ops->arg, keva, nkev);
			nkev = 0;
			kevp = keva;
			KQ_LOCK(kq);
			if (error)
				break;
		}
	}
	TAILQ_REMOVE(&kq->kq_head, marker, kn_tqe);
done:
	KQ_OWNED(kq);
	KQ_UNLOCK_FLUX(kq);
	knote_free(marker);
done_nl:
	KQ_NOTOWNED(kq);
	if (nkev != 0)
		error = k_ops->k_copyout(k_ops->arg, keva, nkev);
	td->td_retval[0] = maxevents - count;
	return (error);
}

/*ARGSUSED*/
static int
kqueue_ioctl(struct file *fp, u_long cmd, void *data,
	struct ucred *active_cred, struct thread *td)
{
	/*
	 * Enabling sigio causes two major problems:
	 * 1) infinite recursion:
	 * Synopsys: kevent is being used to track signals and have FIOASYNC
	 * set.  On receipt of a signal this will cause a kqueue to recurse
	 * into itself over and over.  Sending the sigio causes the kqueue
	 * to become ready, which in turn posts sigio again, forever.
	 * Solution: this can be solved by setting a flag in the kqueue that
	 * we have a SIGIO in progress.
	 * 2) locking problems:
	 * Synopsys: Kqueue is a leaf subsystem, but adding signalling puts
	 * us above the proc and pgrp locks.
	 * Solution: Post a signal using an async mechanism, being sure to
	 * record a generation count in the delivery so that we do not deliver
	 * a signal to the wrong process.
	 *
	 * Note, these two mechanisms are somewhat mutually exclusive!
	 */
#if 0
	struct kqueue *kq;

	kq = fp->f_data;
	switch (cmd) {
	case FIOASYNC:
		if (*(int *)data) {
			kq->kq_state |= KQ_ASYNC;
		} else {
			kq->kq_state &= ~KQ_ASYNC;
		}
		return (0);

	case FIOSETOWN:
		return (fsetown(*(int *)data, &kq->kq_sigio));

	case FIOGETOWN:
		*(int *)data = fgetown(&kq->kq_sigio);
		return (0);
	}
#endif

	return (ENOTTY);
}

/*ARGSUSED*/
static int
kqueue_poll(struct file *fp, int events, struct ucred *active_cred,
	struct thread *td)
{
	struct kqueue *kq;
	int revents = 0;
	int error;

	if ((error = kqueue_acquire(fp, &kq)))
		return POLLERR;

	KQ_LOCK(kq);
	if (events & (POLLIN | POLLRDNORM)) {
		if (kq->kq_count) {
			revents |= events & (POLLIN | POLLRDNORM);
		} else {
			selrecord(td, &kq->kq_sel);
			if (SEL_WAITING(&kq->kq_sel))
				kq->kq_state |= KQ_SEL;
		}
	}
	kqueue_release(kq, 1);
	KQ_UNLOCK(kq);
	return (revents);
}

/*ARGSUSED*/
static int
kqueue_stat(struct file *fp, struct stat *st, struct ucred *active_cred)
{

	bzero((void *)st, sizeof *st);
	/*
	 * We no longer return kq_count because the unlocked value is useless.
	 * If you spent all this time getting the count, why not spend your
	 * syscall better by calling kevent?
	 *
	 * XXX - This is needed for libc_r.
	 */
	st->st_mode = S_IFIFO;
	return (0);
}

static void
kqueue_drain(struct kqueue *kq, struct thread *td)
{
	struct knote *kn;
	int i;

	KQ_LOCK(kq);

	KASSERT((kq->kq_state & KQ_CLOSING) != KQ_CLOSING,
	    ("kqueue already closing"));
	kq->kq_state |= KQ_CLOSING;
	if (kq->kq_refcnt > 1)
		msleep(&kq->kq_refcnt, &kq->kq_lock, PSOCK, "kqclose", 0);

	KASSERT(kq->kq_refcnt == 1, ("other refs are out there!"));

	KASSERT(knlist_empty(&kq->kq_sel.si_note),
	    ("kqueue's knlist not empty"));

	for (i = 0; i < kq->kq_knlistsize; i++) {
		while ((kn = SLIST_FIRST(&kq->kq_knlist[i])) != NULL) {
			if (kn_in_flux(kn)) {
				kq->kq_state |= KQ_FLUXWAIT;
				msleep(kq, &kq->kq_lock, PSOCK, "kqclo1", 0);
				continue;
			}
			kn_enter_flux(kn);
			KQ_UNLOCK(kq);
			knote_drop(kn, td);
			KQ_LOCK(kq);
		}
	}
	if (kq->kq_knhashmask != 0) {
		for (i = 0; i <= kq->kq_knhashmask; i++) {
			while ((kn = SLIST_FIRST(&kq->kq_knhash[i])) != NULL) {
				if (kn_in_flux(kn)) {
					kq->kq_state |= KQ_FLUXWAIT;
					msleep(kq, &kq->kq_lock, PSOCK,
					       "kqclo2", 0);
					continue;
				}
				kn_enter_flux(kn);
				KQ_UNLOCK(kq);
				knote_drop(kn, td);
				KQ_LOCK(kq);
			}
		}
	}

	if ((kq->kq_state & KQ_TASKSCHED) == KQ_TASKSCHED) {
		kq->kq_state |= KQ_TASKDRAIN;
		msleep(&kq->kq_state, &kq->kq_lock, PSOCK, "kqtqdr", 0);
	}

	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
		selwakeuppri(&kq->kq_sel, PSOCK);
		if (!SEL_WAITING(&kq->kq_sel))
			kq->kq_state &= ~KQ_SEL;
	}

	KQ_UNLOCK(kq);
}

static void
kqueue_destroy(struct kqueue *kq)
{

	KASSERT(kq->kq_fdp == NULL,
	    ("kqueue still attached to a file descriptor"));
	seldrain(&kq->kq_sel);
	knlist_destroy(&kq->kq_sel.si_note);
	mtx_destroy(&kq->kq_lock);

	if (kq->kq_knhash != NULL)
		free(kq->kq_knhash, M_KQUEUE);
	if (kq->kq_knlist != NULL)
		free(kq->kq_knlist, M_KQUEUE);

	funsetown(&kq->kq_sigio);
}

/*ARGSUSED*/
static int
kqueue_close(struct file *fp, struct thread *td)
{
	struct kqueue *kq = fp->f_data;
	struct filedesc *fdp;
	int error;
	int filedesc_unlock;

	if ((error = kqueue_acquire(fp, &kq)))
		return error;
	kqueue_drain(kq, td);

	/*
	 * We could be called due to the knote_drop() doing fdrop(),
	 * called from kqueue_register().  In this case the global
	 * lock is owned, and filedesc sx is locked before, to not
	 * take the sleepable lock after non-sleepable.
	 */
	fdp = kq->kq_fdp;
	kq->kq_fdp = NULL;
	if (!sx_xlocked(FILEDESC_LOCK(fdp))) {
		FILEDESC_XLOCK(fdp);
		filedesc_unlock = 1;
	} else
		filedesc_unlock = 0;
	TAILQ_REMOVE(&fdp->fd_kqlist, kq, kq_list);
	if (filedesc_unlock)
		FILEDESC_XUNLOCK(fdp);

	kqueue_destroy(kq);
	chgkqcnt(kq->kq_cred->cr_ruidinfo, -1, 0);
	crfree(kq->kq_cred);
	free(kq, M_KQUEUE);
	fp->f_data = NULL;

	return (0);
}

static int
kqueue_fill_kinfo(struct file *fp, struct kinfo_file *kif, struct filedesc *fdp)
{
	struct kqueue *kq = fp->f_data;

	kif->kf_type = KF_TYPE_KQUEUE;
	kif->kf_un.kf_kqueue.kf_kqueue_addr = (uintptr_t)kq;
	kif->kf_un.kf_kqueue.kf_kqueue_count = kq->kq_count;
	kif->kf_un.kf_kqueue.kf_kqueue_state = kq->kq_state;
	return (0);
}

static void
kqueue_wakeup(struct kqueue *kq)
{
	KQ_OWNED(kq);

	if ((kq->kq_state & KQ_SLEEP) == KQ_SLEEP) {
		kq->kq_state &= ~KQ_SLEEP;
		wakeup(kq);
	}
	if ((kq->kq_state & KQ_SEL) == KQ_SEL) {
		selwakeuppri(&kq->kq_sel, PSOCK);
		if (!SEL_WAITING(&kq->kq_sel))
			kq->kq_state &= ~KQ_SEL;
	}
	if (!knlist_empty(&kq->kq_sel.si_note))
		kqueue_schedtask(kq);
	if ((kq->kq_state & KQ_ASYNC) == KQ_ASYNC) {
		pgsigio(&kq->kq_sigio, SIGIO, 0);
	}
}

/*
 * Walk down a list of knotes, activating them if their event has triggered.
 *
 * There is a possibility to optimize in the case of one kq watching another.
 * Instead of scheduling a task to wake it up, you could pass enough state
 * down the chain to make up the parent kqueue.  Make this code functional
 * first.
 */
void
knote(struct knlist *list, long hint, int lockflags)
{
	struct kqueue *kq;
	struct knote *kn, *tkn;
	int error;

	if (list == NULL)
		return;

	KNL_ASSERT_LOCK(list, lockflags & KNF_LISTLOCKED);

	if ((lockflags & KNF_LISTLOCKED) == 0)
		list->kl_lock(list->kl_lockarg);

	/*
	 * If we unlock the list lock (and enter influx), we can
	 * eliminate the kqueue scheduling, but this will introduce
	 * four lock/unlock's for each knote to test.  Also, marker
	 * would be needed to keep iteration position, since filters
	 * or other threads could remove events.
	 */
	SLIST_FOREACH_SAFE(kn, &list->kl_list, kn_selnext, tkn) {
		kq = kn->kn_kq;
		KQ_LOCK(kq);
		if (kn_in_flux(kn) && (kn->kn_status & KN_SCAN) == 0) {
			/*
			 * Do not process the influx notes, except for
			 * the influx coming from the kq unlock in the
			 * kqueue_scan().  In the later case, we do
			 * not interfere with the scan, since the code
			 * fragment in kqueue_scan() locks the knlist,
			 * and cannot proceed until we finished.
			 */
			KQ_UNLOCK(kq);
		} else if ((lockflags & KNF_NOKQLOCK) != 0) {
			kn_enter_flux(kn);
			KQ_UNLOCK(kq);
			error = kn->kn_fop->f_event(kn, hint);
			KQ_LOCK(kq);
			kn_leave_flux(kn);
			if (error)
				KNOTE_ACTIVATE(kn, 1);
			KQ_UNLOCK_FLUX(kq);
		} else {
			if (kn->kn_fop->f_event(kn, hint))
				KNOTE_ACTIVATE(kn, 1);
			KQ_UNLOCK(kq);
		}
	}
	if ((lockflags & KNF_LISTLOCKED) == 0)
		list->kl_unlock(list->kl_lockarg);
}

/*
 * add a knote to a knlist
 */
void
knlist_add(struct knlist *knl, struct knote *kn, int islocked)
{

	KNL_ASSERT_LOCK(knl, islocked);
	KQ_NOTOWNED(kn->kn_kq);
	KASSERT(kn_in_flux(kn), ("knote %p not in flux", kn));
	KASSERT((kn->kn_status & KN_DETACHED) != 0,
	    ("knote %p was not detached", kn));
	if (!islocked)
		knl->kl_lock(knl->kl_lockarg);
	SLIST_INSERT_HEAD(&knl->kl_list, kn, kn_selnext);
	if (!islocked)
		knl->kl_unlock(knl->kl_lockarg);
	KQ_LOCK(kn->kn_kq);
	kn->kn_knlist = knl;
	kn->kn_status &= ~KN_DETACHED;
	KQ_UNLOCK(kn->kn_kq);
}

static void
knlist_remove_kq(struct knlist *knl, struct knote *kn, int knlislocked,
    int kqislocked)
{

	KASSERT(!kqislocked || knlislocked, ("kq locked w/o knl locked"));
	KNL_ASSERT_LOCK(knl, knlislocked);
	mtx_assert(&kn->kn_kq->kq_lock, kqislocked ? MA_OWNED : MA_NOTOWNED);
	KASSERT(kqislocked || kn_in_flux(kn), ("knote %p not in flux", kn));
	KASSERT((kn->kn_status & KN_DETACHED) == 0,
	    ("knote %p was already detached", kn));
	if (!knlislocked)
		knl->kl_lock(knl->kl_lockarg);
	SLIST_REMOVE(&knl->kl_list, kn, knote, kn_selnext);
	kn->kn_knlist = NULL;
	if (!knlislocked)
		kn_list_unlock(knl);
	if (!kqislocked)
		KQ_LOCK(kn->kn_kq);
	kn->kn_status |= KN_DETACHED;
	if (!kqislocked)
		KQ_UNLOCK(kn->kn_kq);
}

/*
 * remove knote from the specified knlist
 */
void
knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
{

	knlist_remove_kq(knl, kn, islocked, 0);
}

int
knlist_empty(struct knlist *knl)
{

	KNL_ASSERT_LOCKED(knl);
	return (SLIST_EMPTY(&knl->kl_list));
}

static struct mtx knlist_lock;
MTX_SYSINIT(knlist_lock, &knlist_lock, "knlist lock for lockless objects",
    MTX_DEF);
static void knlist_mtx_lock(void *arg);
static void knlist_mtx_unlock(void *arg);

static void
knlist_mtx_lock(void *arg)
{

	mtx_lock((struct mtx *)arg);
}

static void
knlist_mtx_unlock(void *arg)
{

	mtx_unlock((struct mtx *)arg);
}

static void
knlist_mtx_assert_lock(void *arg, int what)
{

	if (what == LA_LOCKED)
		mtx_assert((struct mtx *)arg, MA_OWNED);
	else
		mtx_assert((struct mtx *)arg, MA_NOTOWNED);
}

void
knlist_init(struct knlist *knl, void *lock, void (*kl_lock)(void *),
    void (*kl_unlock)(void *),
    void (*kl_assert_lock)(void *, int))
{

	if (lock == NULL)
		knl->kl_lockarg = &knlist_lock;
	else
		knl->kl_lockarg = lock;

	if (kl_lock == NULL)
		knl->kl_lock = knlist_mtx_lock;
	else
		knl->kl_lock = kl_lock;
	if (kl_unlock == NULL)
		knl->kl_unlock = knlist_mtx_unlock;
	else
		knl->kl_unlock = kl_unlock;
	if (kl_assert_lock == NULL)
		knl->kl_assert_lock = knlist_mtx_assert_lock;
	else
		knl->kl_assert_lock = kl_assert_lock;

	knl->kl_autodestroy = 0;
	SLIST_INIT(&knl->kl_list);
}

void
knlist_init_mtx(struct knlist *knl, struct mtx *lock)
{

	knlist_init(knl, lock, NULL, NULL, NULL);
}

struct knlist *
knlist_alloc(struct mtx *lock)
{
	struct knlist *knl;

	knl = malloc(sizeof(struct knlist), M_KQUEUE, M_WAITOK);
	knlist_init_mtx(knl, lock);
	return (knl);
}

void
knlist_destroy(struct knlist *knl)
{

	KASSERT(KNLIST_EMPTY(knl),
	    ("destroying knlist %p with knotes on it", knl));
}

void
knlist_detach(struct knlist *knl)
{

	KNL_ASSERT_LOCKED(knl);
	knl->kl_autodestroy = 1;
	if (knlist_empty(knl)) {
		knlist_destroy(knl);
		free(knl, M_KQUEUE);
	}
}

/*
 * Even if we are locked, we may need to drop the lock to allow any influx
 * knotes time to "settle".
 */
void
knlist_cleardel(struct knlist *knl, struct thread *td, int islocked, int killkn)
{
	struct knote *kn, *kn2;
	struct kqueue *kq;

	KASSERT(!knl->kl_autodestroy, ("cleardel for autodestroy %p", knl));
	if (islocked)
		KNL_ASSERT_LOCKED(knl);
	else {
		KNL_ASSERT_UNLOCKED(knl);
again:		/* need to reacquire lock since we have dropped it */
		knl->kl_lock(knl->kl_lockarg);
	}

	SLIST_FOREACH_SAFE(kn, &knl->kl_list, kn_selnext, kn2) {
		kq = kn->kn_kq;
		KQ_LOCK(kq);
		if (kn_in_flux(kn)) {
			KQ_UNLOCK(kq);
			continue;
		}
		knlist_remove_kq(knl, kn, 1, 1);
		if (killkn) {
			kn_enter_flux(kn);
			KQ_UNLOCK(kq);
			knote_drop_detached(kn, td);
		} else {
			/* Make sure cleared knotes disappear soon */
			kn->kn_flags |= EV_EOF | EV_ONESHOT;
			KQ_UNLOCK(kq);
		}
		kq = NULL;
	}

	if (!SLIST_EMPTY(&knl->kl_list)) {
		/* there are still in flux knotes remaining */
		kn = SLIST_FIRST(&knl->kl_list);
		kq = kn->kn_kq;
		KQ_LOCK(kq);
		KASSERT(kn_in_flux(kn), ("knote removed w/o list lock"));
		knl->kl_unlock(knl->kl_lockarg);
		kq->kq_state |= KQ_FLUXWAIT;
		msleep(kq, &kq->kq_lock, PSOCK | PDROP, "kqkclr", 0);
		kq = NULL;
		goto again;
	}

	if (islocked)
		KNL_ASSERT_LOCKED(knl);
	else {
		knl->kl_unlock(knl->kl_lockarg);
		KNL_ASSERT_UNLOCKED(knl);
	}
}

/*
 * Remove all knotes referencing a specified fd must be called with FILEDESC
 * lock.  This prevents a race where a new fd comes along and occupies the
 * entry and we attach a knote to the fd.
 */
void
knote_fdclose(struct thread *td, int fd)
{
	struct filedesc *fdp = td->td_proc->p_fd;
	struct kqueue *kq;
	struct knote *kn;
	int influx;

	FILEDESC_XLOCK_ASSERT(fdp);

	/*
	 * We shouldn't have to worry about new kevents appearing on fd
	 * since filedesc is locked.
	 */
	TAILQ_FOREACH(kq, &fdp->fd_kqlist, kq_list) {
		KQ_LOCK(kq);

again:
		influx = 0;
		while (kq->kq_knlistsize > fd &&
		    (kn = SLIST_FIRST(&kq->kq_knlist[fd])) != NULL) {
			if (kn_in_flux(kn)) {
				/* someone else might be waiting on our knote */
				if (influx)
					wakeup(kq);
				kq->kq_state |= KQ_FLUXWAIT;
				msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
				goto again;
			}
			kn_enter_flux(kn);
			KQ_UNLOCK(kq);
			influx = 1;
			knote_drop(kn, td);
			KQ_LOCK(kq);
		}
		KQ_UNLOCK_FLUX(kq);
	}
}

static int
knote_attach(struct knote *kn, struct kqueue *kq)
{
	struct klist *list;

	KASSERT(kn_in_flux(kn), ("knote %p not marked influx", kn));
	KQ_OWNED(kq);

	if ((kq->kq_state & KQ_CLOSING) != 0)
		return (EBADF);
	if (kn->kn_fop->f_isfd) {
		if (kn->kn_id >= kq->kq_knlistsize)
			return (ENOMEM);
		list = &kq->kq_knlist[kn->kn_id];
	} else {
		if (kq->kq_knhash == NULL)
			return (ENOMEM);
		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];
	}
	SLIST_INSERT_HEAD(list, kn, kn_link);
	return (0);
}

static void
knote_drop(struct knote *kn, struct thread *td)
{

	if ((kn->kn_status & KN_DETACHED) == 0)
		kn->kn_fop->f_detach(kn);
	knote_drop_detached(kn, td);
}

static void
knote_drop_detached(struct knote *kn, struct thread *td)
{
	struct kqueue *kq;
	struct klist *list;

	kq = kn->kn_kq;

	KASSERT((kn->kn_status & KN_DETACHED) != 0,
	    ("knote %p still attached", kn));
	KQ_NOTOWNED(kq);

	KQ_LOCK(kq);
	for (;;) {
		KASSERT(kn->kn_influx >= 1,
		    ("knote_drop called on %p with influx %d",
		    kn, kn->kn_influx));
		if (kn->kn_influx == 1)
			break;
		kq->kq_state |= KQ_FLUXWAIT;
		msleep(kq, &kq->kq_lock, PSOCK, "kqflxwt", 0);
	}

	if (kn->kn_fop->f_isfd)
		list = &kq->kq_knlist[kn->kn_id];
	else
		list = &kq->kq_knhash[KN_HASH(kn->kn_id, kq->kq_knhashmask)];

	if (!SLIST_EMPTY(list))
		SLIST_REMOVE(list, kn, knote, kn_link);
	if (kn->kn_status & KN_QUEUED)
		knote_dequeue(kn);
	KQ_UNLOCK_FLUX(kq);

	if (kn->kn_fop->f_isfd) {
		fdrop(kn->kn_fp, td);
		kn->kn_fp = NULL;
	}
	kqueue_fo_release(kn->kn_kevent.filter);
	kn->kn_fop = NULL;
	knote_free(kn);
}

static void
knote_enqueue(struct knote *kn)
{
	struct kqueue *kq = kn->kn_kq;

	KQ_OWNED(kn->kn_kq);
	KASSERT((kn->kn_status & KN_QUEUED) == 0, ("knote already queued"));

	TAILQ_INSERT_TAIL(&kq->kq_head, kn, kn_tqe);
	kn->kn_status |= KN_QUEUED;
	kq->kq_count++;
	kqueue_wakeup(kq);
}

static void
knote_dequeue(struct knote *kn)
{
	struct kqueue *kq = kn->kn_kq;

	KQ_OWNED(kn->kn_kq);
	KASSERT(kn->kn_status & KN_QUEUED, ("knote not queued"));

	TAILQ_REMOVE(&kq->kq_head, kn, kn_tqe);
	kn->kn_status &= ~KN_QUEUED;
	kq->kq_count--;
}

static void
knote_init(void)
{

	knote_zone = uma_zcreate("KNOTE", sizeof(struct knote), NULL, NULL,
	    NULL, NULL, UMA_ALIGN_PTR, 0);
	ast_register(TDA_KQUEUE, ASTR_ASTF_REQUIRED, 0, ast_kqueue);
}
SYSINIT(knote, SI_SUB_PSEUDO, SI_ORDER_ANY, knote_init, NULL);

static struct knote *
knote_alloc(int mflag)
{

	return (uma_zalloc(knote_zone, mflag | M_ZERO));
}

static void
knote_free(struct knote *kn)
{

	uma_zfree(knote_zone, kn);
}

/*
 * Register the kev w/ the kq specified by fd.
 */
int
kqfd_register(int fd, struct kevent *kev, struct thread *td, int mflag)
{
	struct kqueue *kq;
	struct file *fp;
	cap_rights_t rights;
	int error;

	error = fget(td, fd, cap_rights_init_one(&rights, CAP_KQUEUE_CHANGE),
	    &fp);
	if (error != 0)
		return (error);
	if ((error = kqueue_acquire(fp, &kq)) != 0)
		goto noacquire;

	error = kqueue_register(kq, kev, td, mflag);
	kqueue_release(kq, 0);

noacquire:
	fdrop(fp, td);
	return (error);
}

struct knote_status_export_bit {
	int kn_status_bit;
	int knt_status_bit;
};

#define	ST(name) \
    { .kn_status_bit = KN_##name, .knt_status_bit = KNOTE_STATUS_##name }
static const struct knote_status_export_bit knote_status_export_bits[] = {
	ST(ACTIVE),
	ST(QUEUED),
	ST(DISABLED),
	ST(DETACHED),
	ST(KQUEUE),
};
#undef ST

static int
knote_status_export(int kn_status)
{
	const struct knote_status_export_bit *b;
	unsigned i;
	int res;

	res = 0;
	for (i = 0; i < nitems(knote_status_export_bits); i++) {
		b = &knote_status_export_bits[i];
		if ((kn_status & b->kn_status_bit) != 0)
			res |= b->knt_status_bit;
	}
	return (res);
}

static int
kern_proc_kqueue_report_one(struct sbuf *s, struct proc *p,
    int kq_fd, struct kqueue *kq, struct knote *kn, bool compat32 __unused)
{
	struct kinfo_knote kin;
#ifdef COMPAT_FREEBSD32
	struct kinfo_knote32 kin32;
#endif
	int error;

	if (kn->kn_status == KN_MARKER)
		return (0);

	memset(&kin, 0, sizeof(kin));
	kin.knt_kq_fd = kq_fd;
	memcpy(&kin.knt_event, &kn->kn_kevent, sizeof(struct kevent));
	kin.knt_status = knote_status_export(kn->kn_status);
	kn_enter_flux(kn);
	KQ_UNLOCK_FLUX(kq);
	if (kn->kn_fop->f_userdump != NULL)
		(void)kn->kn_fop->f_userdump(p, kn, &kin);
#ifdef COMPAT_FREEBSD32
	if (compat32) {
		freebsd32_kinfo_knote_to_32(&kin, &kin32);
		error = sbuf_bcat(s, &kin32, sizeof(kin32));
	} else
#endif
		error = sbuf_bcat(s, &kin, sizeof(kin));
	KQ_LOCK(kq);
	kn_leave_flux(kn);
	return (error);
}

static int
kern_proc_kqueue_report(struct sbuf *s, struct proc *p, int kq_fd,
    struct kqueue *kq, bool compat32)
{
	struct knote *kn;
	int error, i;

	error = 0;
	KQ_LOCK(kq);
	for (i = 0; i < kq->kq_knlistsize; i++) {
		SLIST_FOREACH(kn, &kq->kq_knlist[i], kn_link) {
			error = kern_proc_kqueue_report_one(s, p, kq_fd,
			    kq, kn, compat32);
			if (error != 0)
				goto out;
		}
	}
	if (kq->kq_knhashmask == 0)
		goto out;
	for (i = 0; i <= kq->kq_knhashmask; i++) {
		SLIST_FOREACH(kn, &kq->kq_knhash[i], kn_link) {
			error = kern_proc_kqueue_report_one(s, p, kq_fd,
			    kq, kn, compat32);
			if (error != 0)
				goto out;
		}
	}
out:
	KQ_UNLOCK_FLUX(kq);
	return (error);
}

struct kern_proc_kqueues_out1_cb_args {
	struct sbuf *s;
	bool compat32;
};

static int
kern_proc_kqueues_out1_cb(struct proc *p, int fd, struct file *fp, void *arg)
{
	struct kqueue *kq;
	struct kern_proc_kqueues_out1_cb_args *a;

	if (fp->f_type != DTYPE_KQUEUE)
		return (0);
	a = arg;
	kq = fp->f_data;
	return (kern_proc_kqueue_report(a->s, p, fd, kq, a->compat32));
}

static int
kern_proc_kqueues_out1(struct thread *td, struct proc *p, struct sbuf *s,
    bool compat32)
{
	struct kern_proc_kqueues_out1_cb_args a;

	a.s = s;
	a.compat32 = compat32;
	return (fget_remote_foreach(td, p, kern_proc_kqueues_out1_cb, &a));
}

int
kern_proc_kqueues_out(struct proc *p, struct sbuf *sb, size_t maxlen,
    bool compat32)
{
	struct sbuf *s, sm;
	size_t sb_len;
	int error;

	if (maxlen == -1 || maxlen == 0)
		sb_len = 128;
	else
		sb_len = maxlen;
	s = sbuf_new(&sm, NULL, sb_len, maxlen == -1 ? SBUF_AUTOEXTEND :
	    SBUF_FIXEDLEN);
	error = kern_proc_kqueues_out1(curthread, p, s, compat32);
	sbuf_finish(s);
	if (error == 0) {
		sbuf_bcat(sb, sbuf_data(s), MIN(sbuf_len(s), maxlen == -1 ?
		    SIZE_T_MAX : maxlen));
	}
	sbuf_delete(s);
	return (error);
}

static int
sysctl_kern_proc_kqueue_one(struct thread *td, struct sbuf *s, struct proc *p,
    int kq_fd, bool compat32)
{
	struct file *fp;
	struct kqueue *kq;
	int error;

	error = fget_remote(td, p, kq_fd, &fp);
	if (error == 0) {
		if (fp->f_type != DTYPE_KQUEUE) {
			error = EINVAL;
		} else {
			kq = fp->f_data;
			error = kern_proc_kqueue_report(s, p, kq_fd, kq,
			    compat32);
		}
		fdrop(fp, td);
	}
	return (error);
}

static int
sysctl_kern_proc_kqueue(SYSCTL_HANDLER_ARGS)
{
	struct thread *td;
	struct proc *p;
	struct sbuf *s, sm;
	int error, error1, *name;
	bool compat32;

	name = (int *)arg1;
	if ((u_int)arg2 > 2 || (u_int)arg2 == 0)
		return (EINVAL);

	error = pget((pid_t)name[0], PGET_HOLD | PGET_CANDEBUG, &p);
	if (error != 0)
		return (error);

	td = curthread;
#ifdef FREEBSD_COMPAT32
	compat32 = SV_CURPROC_FLAG(SV_ILP32);
#else
	compat32 = false;
#endif

	s = sbuf_new_for_sysctl(&sm, NULL, 0, req);
	if (s == NULL) {
		error = ENOMEM;
		goto out;
	}
	sbuf_clear_flags(s, SBUF_INCLUDENUL);

	if ((u_int)arg2 == 1) {
		error = kern_proc_kqueues_out1(td, p, s, compat32);
	} else {
		error = sysctl_kern_proc_kqueue_one(td, s, p,
		    name[1] /* kq_fd */, compat32);
	}

	error1 = sbuf_finish(s);
	if (error == 0)
		error = error1;
	sbuf_delete(s);

out:
	PRELE(p);
	return (error);
}

static SYSCTL_NODE(_kern_proc, KERN_PROC_KQUEUE, kq,
    CTLFLAG_RD | CTLFLAG_MPSAFE,
    sysctl_kern_proc_kqueue, "KQueue events");