kernel - Remove getnewvnode() bottlenecks

[dragonfly.git] / sys / kern / vfs_mount.c

/*
 * Copyright (c) 2004 The DragonFly Project.  All rights reserved.
 * 
 * This code is derived from software contributed to The DragonFly Project
 * by Matthew Dillon <dillon@backplane.com>
 * 
 * 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.
 * 3. Neither the name of The DragonFly Project nor the names of its
 *    contributors may be used to endorse or promote products derived
 *    from this software without specific, prior written permission.
 * 
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS 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
 * COPYRIGHT HOLDERS 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.
 *
 * Copyright (c) 1989, 1993
 *      The Regents of the University of California.  All rights reserved.
 * (c) UNIX System Laboratories, Inc.
 * All or some portions of this file are derived from material licensed
 * to the University of California by American Telephone and Telegraph
 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
 * the permission of UNIX System Laboratories, Inc.
 *
 * 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.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *      This product includes software developed by the University of
 *      California, Berkeley and its contributors.
 * 4. Neither the name of the University nor the names of its contributors
 *    may be used to endorse or promote products derived from this software
 *    without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 REGENTS 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.
 */

/*
 * External virtual filesystem routines
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/malloc.h>
#include <sys/mount.h>
#include <sys/proc.h>
#include <sys/vnode.h>
#include <sys/buf.h>
#include <sys/eventhandler.h>
#include <sys/kthread.h>
#include <sys/sysctl.h>

#include <machine/limits.h>

#include <sys/buf2.h>
#include <sys/thread2.h>
#include <sys/sysref2.h>

#include <vm/vm.h>
#include <vm/vm_object.h>

struct mountscan_info {
        TAILQ_ENTRY(mountscan_info) msi_entry;
        int msi_how;
        struct mount *msi_node;
};

struct vmntvnodescan_info {
        TAILQ_ENTRY(vmntvnodescan_info) entry;
        struct vnode *vp;
};

struct vnlru_info {
        int     pass;
};

static int vnlru_nowhere = 0;
SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RD,
            &vnlru_nowhere, 0,
            "Number of times the vnlru process ran without success");


static struct lwkt_token mntid_token;
static struct mount dummymount;

/* note: mountlist exported to pstat */
struct mntlist mountlist = TAILQ_HEAD_INITIALIZER(mountlist);
static TAILQ_HEAD(,mountscan_info) mountscan_list;
static struct lwkt_token mountlist_token;

static TAILQ_HEAD(,bio_ops) bio_ops_list = TAILQ_HEAD_INITIALIZER(bio_ops_list);

/*
 * Called from vfsinit()
 */
void
vfs_mount_init(void)
{
        lwkt_token_init(&mountlist_token, "mntlist");
        lwkt_token_init(&mntid_token, "mntid");
        TAILQ_INIT(&mountscan_list);
        mount_init(&dummymount);
        dummymount.mnt_flag |= MNT_RDONLY;
        dummymount.mnt_kern_flag |= MNTK_ALL_MPSAFE;
}

/*
 * Support function called to remove a vnode from the mountlist and
 * deal with side effects for scans in progress.
 *
 * Target mnt_token is held on call.
 */
static void
vremovevnodemnt(struct vnode *vp)
{
        struct vmntvnodescan_info *info;
        struct mount *mp = vp->v_mount;

        TAILQ_FOREACH(info, &mp->mnt_vnodescan_list, entry) {
                if (info->vp == vp)
                        info->vp = TAILQ_NEXT(vp, v_nmntvnodes);
        }
        TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
}

/*
 * Allocate a new vnode and associate it with a tag, mount point, and
 * operations vector.
 *
 * A VX locked and refd vnode is returned.  The caller should setup the
 * remaining fields and vx_put() or, if he wishes to leave a vref,
 * vx_unlock() the vnode.
 */
int
getnewvnode(enum vtagtype tag, struct mount *mp,
                struct vnode **vpp, int lktimeout, int lkflags)
{
        struct vnode *vp;

        KKASSERT(mp != NULL);

        vp = allocvnode(lktimeout, lkflags);
        vp->v_tag = tag;
        vp->v_data = NULL;

        /*
         * By default the vnode is assigned the mount point's normal
         * operations vector.
         */
        vp->v_ops = &mp->mnt_vn_use_ops;

        /*
         * Placing the vnode on the mount point's queue makes it visible.
         * VNON prevents it from being messed with, however.
         */
        insmntque(vp, mp);

        /*
         * A VX locked & refd vnode is returned.
         */
        *vpp = vp;
        return (0);
}

/*
 * This function creates vnodes with special operations vectors.  The
 * mount point is optional.
 *
 * This routine is being phased out but is still used by vfs_conf to
 * create vnodes for devices prior to the root mount (with mp == NULL).
 */
int
getspecialvnode(enum vtagtype tag, struct mount *mp,
                struct vop_ops **ops,
                struct vnode **vpp, int lktimeout, int lkflags)
{
        struct vnode *vp;

        vp = allocvnode(lktimeout, lkflags);
        vp->v_tag = tag;
        vp->v_data = NULL;
        vp->v_ops = ops;

        if (mp == NULL)
                mp = &dummymount;

        /*
         * Placing the vnode on the mount point's queue makes it visible.
         * VNON prevents it from being messed with, however.
         */
        insmntque(vp, mp);

        /*
         * A VX locked & refd vnode is returned.
         */
        *vpp = vp;
        return (0);
}

/*
 * Interlock against an unmount, return 0 on success, non-zero on failure.
 *
 * The passed flag may be 0 or LK_NOWAIT and is only used if an unmount
 * is in-progress.  
 *
 * If no unmount is in-progress LK_NOWAIT is ignored.  No other flag bits
 * are used.  A shared locked will be obtained and the filesystem will not
 * be unmountable until the lock is released.
 */
int
vfs_busy(struct mount *mp, int flags)
{
        int lkflags;

        atomic_add_int(&mp->mnt_refs, 1);
        lwkt_gettoken(&mp->mnt_token);
        if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
                if (flags & LK_NOWAIT) {
                        lwkt_reltoken(&mp->mnt_token);
                        atomic_add_int(&mp->mnt_refs, -1);
                        return (ENOENT);
                }
                /* XXX not MP safe */
                mp->mnt_kern_flag |= MNTK_MWAIT;
                /*
                 * Since all busy locks are shared except the exclusive
                 * lock granted when unmounting, the only place that a
                 * wakeup needs to be done is at the release of the
                 * exclusive lock at the end of dounmount.
                 */
                tsleep((caddr_t)mp, 0, "vfs_busy", 0);
                lwkt_reltoken(&mp->mnt_token);
                atomic_add_int(&mp->mnt_refs, -1);
                return (ENOENT);
        }
        lkflags = LK_SHARED;
        if (lockmgr(&mp->mnt_lock, lkflags))
                panic("vfs_busy: unexpected lock failure");
        lwkt_reltoken(&mp->mnt_token);
        return (0);
}

/*
 * Free a busy filesystem.
 *
 * Decrement refs before releasing the lock so e.g. a pending umount
 * doesn't give us an unexpected busy error.
 */
void
vfs_unbusy(struct mount *mp)
{
        atomic_add_int(&mp->mnt_refs, -1);
        lockmgr(&mp->mnt_lock, LK_RELEASE);
}

/*
 * Lookup a filesystem type, and if found allocate and initialize
 * a mount structure for it.
 *
 * Devname is usually updated by mount(8) after booting.
 */
int
vfs_rootmountalloc(char *fstypename, char *devname, struct mount **mpp)
{
        struct vfsconf *vfsp;
        struct mount *mp;

        if (fstypename == NULL)
                return (ENODEV);

        vfsp = vfsconf_find_by_name(fstypename);
        if (vfsp == NULL)
                return (ENODEV);
        mp = kmalloc(sizeof(struct mount), M_MOUNT, M_WAITOK | M_ZERO);
        mount_init(mp);
        lockinit(&mp->mnt_lock, "vfslock", VLKTIMEOUT, 0);

        vfs_busy(mp, 0);
        mp->mnt_vfc = vfsp;
        mp->mnt_op = vfsp->vfc_vfsops;
        vfsp->vfc_refcount++;
        mp->mnt_stat.f_type = vfsp->vfc_typenum;
        mp->mnt_flag |= MNT_RDONLY;
        mp->mnt_flag |= vfsp->vfc_flags & MNT_VISFLAGMASK;
        strncpy(mp->mnt_stat.f_fstypename, vfsp->vfc_name, MFSNAMELEN);
        copystr(devname, mp->mnt_stat.f_mntfromname, MNAMELEN - 1, 0);
        *mpp = mp;
        return (0);
}

/*
 * Basic mount structure initialization
 */
void
mount_init(struct mount *mp)
{
        lockinit(&mp->mnt_lock, "vfslock", 0, 0);
        lwkt_token_init(&mp->mnt_token, "permnt");

        TAILQ_INIT(&mp->mnt_vnodescan_list);
        TAILQ_INIT(&mp->mnt_nvnodelist);
        TAILQ_INIT(&mp->mnt_reservedvnlist);
        TAILQ_INIT(&mp->mnt_jlist);
        mp->mnt_nvnodelistsize = 0;
        mp->mnt_flag = 0;
        mp->mnt_iosize_max = MAXPHYS;
}

/*
 * Lookup a mount point by filesystem identifier.
 */
struct mount *
vfs_getvfs(fsid_t *fsid)
{
        struct mount *mp;

        lwkt_gettoken(&mountlist_token);
        TAILQ_FOREACH(mp, &mountlist, mnt_list) {
                if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
                    mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
                        break;
                }
        }
        lwkt_reltoken(&mountlist_token);
        return (mp);
}

/*
 * Get a new unique fsid.  Try to make its val[0] unique, since this value
 * will be used to create fake device numbers for stat().  Also try (but
 * not so hard) make its val[0] unique mod 2^16, since some emulators only
 * support 16-bit device numbers.  We end up with unique val[0]'s for the
 * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
 *
 * Keep in mind that several mounts may be running in parallel.  Starting
 * the search one past where the previous search terminated is both a
 * micro-optimization and a defense against returning the same fsid to
 * different mounts.
 */
void
vfs_getnewfsid(struct mount *mp)
{
        static u_int16_t mntid_base;
        fsid_t tfsid;
        int mtype;

        lwkt_gettoken(&mntid_token);
        mtype = mp->mnt_vfc->vfc_typenum;
        tfsid.val[1] = mtype;
        mtype = (mtype & 0xFF) << 24;
        for (;;) {
                tfsid.val[0] = makeudev(255,
                    mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
                mntid_base++;
                if (vfs_getvfs(&tfsid) == NULL)
                        break;
        }
        mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
        mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
        lwkt_reltoken(&mntid_token);
}

/*
 * Set the FSID for a new mount point to the template.  Adjust
 * the FSID to avoid collisions.
 */
int
vfs_setfsid(struct mount *mp, fsid_t *template)
{
        int didmunge = 0;

        bzero(&mp->mnt_stat.f_fsid, sizeof(mp->mnt_stat.f_fsid));
        for (;;) {
                if (vfs_getvfs(template) == NULL)
                        break;
                didmunge = 1;
                ++template->val[1];
        }
        mp->mnt_stat.f_fsid = *template;
        return(didmunge);
}

/*
 * This routine is called when we have too many vnodes.  It attempts
 * to free <count> vnodes and will potentially free vnodes that still
 * have VM backing store (VM backing store is typically the cause
 * of a vnode blowout so we want to do this).  Therefore, this operation
 * is not considered cheap.
 *
 * A number of conditions may prevent a vnode from being reclaimed.
 * the buffer cache may have references on the vnode, a directory
 * vnode may still have references due to the namei cache representing
 * underlying files, or the vnode may be in active use.   It is not
 * desireable to reuse such vnodes.  These conditions may cause the
 * number of vnodes to reach some minimum value regardless of what
 * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
 */

/*
 * This is a quick non-blocking check to determine if the vnode is a good
 * candidate for being (eventually) vgone()'d.  Returns 0 if the vnode is
 * not a good candidate, 1 if it is.
 */
static __inline int 
vmightfree(struct vnode *vp, int page_count, int pass)
{
        if (vp->v_flag & VRECLAIMED)
                return (0);
#if 0
        if ((vp->v_flag & VFREE) && TAILQ_EMPTY(&vp->v_namecache))
                return (0);
#endif
        if (sysref_isactive(&vp->v_sysref))
                return (0);
        if (vp->v_object && vp->v_object->resident_page_count >= page_count)
                return (0);

        /*
         * XXX horrible hack.  Up to four passes will be taken.  Each pass
         * makes a larger set of vnodes eligible.  For now what this really
         * means is that we try to recycle files opened only once before
         * recycling files opened multiple times.
         */
        switch(vp->v_flag & (VAGE0 | VAGE1)) {
        case 0:
                if (pass < 3)
                        return(0);
                break;
        case VAGE0:
                if (pass < 2)
                        return(0);
                break;
        case VAGE1:
                if (pass < 1)
                        return(0);
                break;
        case VAGE0 | VAGE1:
                break;
        }
        return (1);
}

/*
 * The vnode was found to be possibly vgone()able and the caller has locked it
 * (thus the usecount should be 1 now).  Determine if the vnode is actually
 * vgone()able, doing some cleanups in the process.  Returns 1 if the vnode
 * can be vgone()'d, 0 otherwise.
 *
 * Note that v_auxrefs may be non-zero because (A) this vnode is not a leaf
 * in the namecache topology and (B) this vnode has buffer cache bufs.
 * We cannot remove vnodes with non-leaf namecache associations.  We do a
 * tentitive leaf check prior to attempting to flush out any buffers but the
 * 'real' test when all is said in done is that v_auxrefs must become 0 for
 * the vnode to be freeable.
 *
 * We could theoretically just unconditionally flush when v_auxrefs != 0,
 * but flushing data associated with non-leaf nodes (which are always
 * directories), just throws it away for no benefit.  It is the buffer 
 * cache's responsibility to choose buffers to recycle from the cached
 * data point of view.
 */
static int
visleaf(struct vnode *vp)
{
        struct namecache *ncp;

        spin_lock(&vp->v_spin);
        TAILQ_FOREACH(ncp, &vp->v_namecache, nc_vnode) {
                if (!TAILQ_EMPTY(&ncp->nc_list)) {
                        spin_unlock(&vp->v_spin);
                        return(0);
                }
        }
        spin_unlock(&vp->v_spin);
        return(1);
}

/*
 * Try to clean up the vnode to the point where it can be vgone()'d, returning
 * 0 if it cannot be vgone()'d (or already has been), 1 if it can.  Unlike
 * vmightfree() this routine may flush the vnode and block.  Vnodes marked
 * VFREE are still candidates for vgone()ing because they may hold namecache
 * resources and could be blocking the namecache directory hierarchy (and
 * related vnodes) from being freed.
 */
static int
vtrytomakegoneable(struct vnode *vp, int page_count)
{
        if (vp->v_flag & VRECLAIMED)
                return (0);
        if (vp->v_sysref.refcnt > 1)
                return (0);
        if (vp->v_object && vp->v_object->resident_page_count >= page_count)
                return (0);
        if (vp->v_auxrefs && visleaf(vp)) {
                vinvalbuf(vp, V_SAVE, 0, 0);
#if 0   /* DEBUG */
                kprintf((vp->v_auxrefs ? "vrecycle: vp %p failed: %s\n" :
                        "vrecycle: vp %p succeeded: %s\n"), vp,
                        (TAILQ_FIRST(&vp->v_namecache) ? 
                            TAILQ_FIRST(&vp->v_namecache)->nc_name : "?"));
#endif
        }

        /*
         * This sequence may seem a little strange, but we need to optimize
         * the critical path a bit.  We can't recycle vnodes with other
         * references and because we are trying to recycle an otherwise
         * perfectly fine vnode we have to invalidate the namecache in a
         * way that avoids possible deadlocks (since the vnode lock is being
         * held here).  Finally, we have to check for other references one
         * last time in case something snuck in during the inval.
         */
        if (vp->v_sysref.refcnt > 1 || vp->v_auxrefs != 0)
                return (0);
        if (cache_inval_vp_nonblock(vp))
                return (0);
        return (vp->v_sysref.refcnt <= 1 && vp->v_auxrefs == 0);
}

/*
 * Reclaim up to 1/10 of the vnodes associated with a mount point.  Try
 * to avoid vnodes which have lots of resident pages (we are trying to free
 * vnodes, not memory).  
 *
 * This routine is a callback from the mountlist scan.  The mount point
 * in question will be busied.
 *
 * NOTE: The 1/10 reclamation also ensures that the inactive data set
 *       (the vnodes being recycled by the one-time use) does not degenerate
 *       into too-small a set.  This is important because once a vnode is
 *       marked as not being one-time-use (VAGE0/VAGE1 both 0) that vnode
 *       will not be destroyed EXCEPT by this mechanism.  VM pages can still
 *       be cleaned/freed by the pageout daemon.
 */
static int
vlrureclaim(struct mount *mp, void *data)
{
        struct vnlru_info *info = data;
        struct vnode *vp;
        int done;
        int trigger;
        int usevnodes;
        int count;
        int trigger_mult = vnlru_nowhere;

        /*
         * Calculate the trigger point for the resident pages check.  The
         * minimum trigger value is approximately the number of pages in
         * the system divded by the number of vnodes.  However, due to
         * various other system memory overheads unrelated to data caching
         * it is a good idea to double the trigger (at least).  
         *
         * trigger_mult starts at 0.  If the recycler is having problems
         * finding enough freeable vnodes it will increase trigger_mult.
         * This should not happen in normal operation, even on machines with
         * low amounts of memory, but extraordinary memory use by the system
         * verses the amount of cached data can trigger it.
         *
         * (long) -> deal with 64 bit machines, intermediate overflow
         */
        usevnodes = desiredvnodes;
        if (usevnodes <= 0)
                usevnodes = 1;
        trigger = (long)vmstats.v_page_count * (trigger_mult + 2) / usevnodes;

        done = 0;
        lwkt_gettoken(&mp->mnt_token);
        count = mp->mnt_nvnodelistsize / 10 + 1;

        while (count && mp->mnt_syncer) {
                /*
                 * Next vnode.  Use the special syncer vnode to placemark
                 * the LRU.  This way the LRU code does not interfere with
                 * vmntvnodescan().
                 */
                vp = TAILQ_NEXT(mp->mnt_syncer, v_nmntvnodes);
                TAILQ_REMOVE(&mp->mnt_nvnodelist, mp->mnt_syncer, v_nmntvnodes);
                if (vp) {
                        TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp,
                                           mp->mnt_syncer, v_nmntvnodes);
                } else {
                        TAILQ_INSERT_HEAD(&mp->mnt_nvnodelist, mp->mnt_syncer,
                                          v_nmntvnodes);
                        vp = TAILQ_NEXT(mp->mnt_syncer, v_nmntvnodes);
                        if (vp == NULL)
                                break;
                }

                /*
                 * __VNODESCAN__
                 *
                 * The VP will stick around while we hold mnt_token,
                 * at least until we block, so we can safely do an initial
                 * check, and then must check again after we lock the vnode.
                 */
                if (vp->v_type == VNON ||       /* syncer or indeterminant */
                    !vmightfree(vp, trigger, info->pass) /* critical path opt */
                ) {
                        --count;
                        continue;
                }

                /*
                 * VX get the candidate vnode.  If the VX get fails the 
                 * vnode might still be on the mountlist.  Our loop depends
                 * on us at least cycling the vnode to the end of the
                 * mountlist.
                 */
                if (vx_get_nonblock(vp) != 0) {
                        --count;
                        continue;
                }

                /*
                 * Since we blocked locking the vp, make sure it is still
                 * a candidate for reclamation.  That is, it has not already
                 * been reclaimed and only has our VX reference associated
                 * with it.
                 */
                if (vp->v_type == VNON ||       /* syncer or indeterminant */
                    (vp->v_flag & VRECLAIMED) ||
                    vp->v_mount != mp ||
                    !vtrytomakegoneable(vp, trigger)    /* critical path opt */
                ) {
                        --count;
                        vx_put(vp);
                        continue;
                }

                /*
                 * All right, we are good, move the vp to the end of the
                 * mountlist and clean it out.  The vget will have returned
                 * an error if the vnode was destroyed (VRECLAIMED set), so we
                 * do not have to check again.  The vput() will move the 
                 * vnode to the free list if the vgone() was successful.
                 */
                KKASSERT(vp->v_mount == mp);
                vgone_vxlocked(vp);
                vx_put(vp);
                ++done;
                --count;
        }
        lwkt_reltoken(&mp->mnt_token);
        return (done);
}

/*
 * Attempt to recycle vnodes in a context that is always safe to block.
 * Calling vlrurecycle() from the bowels of file system code has some
 * interesting deadlock problems.
 */
static struct thread *vnlruthread;

static void 
vnlru_proc(void)
{
        struct thread *td = curthread;
        struct vnlru_info info;
        int done;

        EVENTHANDLER_REGISTER(shutdown_pre_sync, shutdown_kproc, td,
                              SHUTDOWN_PRI_FIRST);

        for (;;) {
                kproc_suspend_loop();

                /*
                 * Do some opportunistic roving.
                 */
                if (numvnodes > 100000)
                        vnode_free_rover_scan(50);
                else if (numvnodes > 10000)
                        vnode_free_rover_scan(20);
                else
                        vnode_free_rover_scan(5);

                /*
                 * Try to free some vnodes if we have too many
                 *
                 * (long) -> deal with 64 bit machines, intermediate overflow
                 */
                if (numvnodes > desiredvnodes &&
                    freevnodes > desiredvnodes * 2 / 10) {
                        int count = numvnodes - desiredvnodes;

                        if (count > freevnodes / 100)
                                count = freevnodes / 100;
                        if (count < 5)
                                count = 5;
                        freesomevnodes(count);
                }

                /*
                 * Nothing to do if most of our vnodes are already on
                 * the free list.
                 */
                if (numvnodes - freevnodes <= (long)desiredvnodes * 9 / 10) {
                        tsleep(vnlruthread, 0, "vlruwt", hz);
                        continue;
                }
                cache_hysteresis();

                /*
                 * The pass iterates through the four combinations of
                 * VAGE0/VAGE1.  We want to get rid of aged small files
                 * first.
                 */
                info.pass = 0;
                done = 0;
                while (done == 0 && info.pass < 4) {
                        done = mountlist_scan(vlrureclaim, &info,
                                              MNTSCAN_FORWARD);
                        ++info.pass;
                }

                /*
                 * The vlrureclaim() call only processes 1/10 of the vnodes
                 * on each mount.  If we couldn't find any repeat the loop
                 * at least enough times to cover all available vnodes before
                 * we start sleeping.  Complain if the failure extends past
                 * 30 second, every 30 seconds.
                 */
                if (done == 0) {
                        ++vnlru_nowhere;
                        if (vnlru_nowhere % 10 == 0)
                                tsleep(vnlruthread, 0, "vlrup", hz * 3);
                        if (vnlru_nowhere % 100 == 0)
                                kprintf("vnlru_proc: vnode recycler stopped working!\n");
                        if (vnlru_nowhere == 1000)
                                vnlru_nowhere = 900;
                } else {
                        vnlru_nowhere = 0;
                }
        }
}

/*
 * MOUNTLIST FUNCTIONS
 */

/*
 * mountlist_insert (MP SAFE)
 *
 * Add a new mount point to the mount list.
 */
void
mountlist_insert(struct mount *mp, int how)
{
        lwkt_gettoken(&mountlist_token);
        if (how == MNTINS_FIRST)
            TAILQ_INSERT_HEAD(&mountlist, mp, mnt_list);
        else
            TAILQ_INSERT_TAIL(&mountlist, mp, mnt_list);
        lwkt_reltoken(&mountlist_token);
}

/*
 * mountlist_interlock (MP SAFE)
 *
 * Execute the specified interlock function with the mountlist token
 * held.  The function will be called in a serialized fashion verses
 * other functions called through this mechanism.
 */
int
mountlist_interlock(int (*callback)(struct mount *), struct mount *mp)
{
        int error;

        lwkt_gettoken(&mountlist_token);
        error = callback(mp);
        lwkt_reltoken(&mountlist_token);
        return (error);
}

/*
 * mountlist_boot_getfirst (DURING BOOT ONLY)
 *
 * This function returns the first mount on the mountlist, which is
 * expected to be the root mount.  Since no interlocks are obtained
 * this function is only safe to use during booting.
 */

struct mount *
mountlist_boot_getfirst(void)
{
        return(TAILQ_FIRST(&mountlist));
}

/*
 * mountlist_remove (MP SAFE)
 *
 * Remove a node from the mountlist.  If this node is the next scan node
 * for any active mountlist scans, the active mountlist scan will be 
 * adjusted to skip the node, thus allowing removals during mountlist
 * scans.
 */
void
mountlist_remove(struct mount *mp)
{
        struct mountscan_info *msi;

        lwkt_gettoken(&mountlist_token);
        TAILQ_FOREACH(msi, &mountscan_list, msi_entry) {
                if (msi->msi_node == mp) {
                        if (msi->msi_how & MNTSCAN_FORWARD)
                                msi->msi_node = TAILQ_NEXT(mp, mnt_list);
                        else
                                msi->msi_node = TAILQ_PREV(mp, mntlist, mnt_list);
                }
        }
        TAILQ_REMOVE(&mountlist, mp, mnt_list);
        lwkt_reltoken(&mountlist_token);
}

/*
 * mountlist_exists (MP SAFE)
 *
 * Checks if a node exists in the mountlist.
 * This function is mainly used by VFS quota code to check if a
 * cached nullfs struct mount pointer is still valid at use time
 *
 * FIXME: there is no warranty the mp passed to that function
 * will be the same one used by VFS_ACCOUNT() later
 */
int
mountlist_exists(struct mount *mp)
{
        int node_exists = 0;
        struct mount* lmp;

        lwkt_gettoken(&mountlist_token);
        TAILQ_FOREACH(lmp, &mountlist, mnt_list) {
                if (lmp == mp) {
                        node_exists = 1;
                        break;
                }
        }
        lwkt_reltoken(&mountlist_token);
        return(node_exists);
}

/*
 * mountlist_scan (MP SAFE)
 *
 * Safely scan the mount points on the mount list.  Unless otherwise 
 * specified each mount point will be busied prior to the callback and
 * unbusied afterwords.  The callback may safely remove any mount point
 * without interfering with the scan.  If the current callback
 * mount is removed the scanner will not attempt to unbusy it.
 *
 * If a mount node cannot be busied it is silently skipped.
 *
 * The callback return value is aggregated and a total is returned.  A return
 * value of < 0 is not aggregated and will terminate the scan.
 *
 * MNTSCAN_FORWARD      - the mountlist is scanned in the forward direction
 * MNTSCAN_REVERSE      - the mountlist is scanned in reverse
 * MNTSCAN_NOBUSY       - the scanner will make the callback without busying
 *                        the mount node.
 */
int
mountlist_scan(int (*callback)(struct mount *, void *), void *data, int how)
{
        struct mountscan_info info;
        struct mount *mp;
        int count;
        int res;

        lwkt_gettoken(&mountlist_token);

        info.msi_how = how;
        info.msi_node = NULL;   /* paranoia */
        TAILQ_INSERT_TAIL(&mountscan_list, &info, msi_entry);

        res = 0;

        if (how & MNTSCAN_FORWARD) {
                info.msi_node = TAILQ_FIRST(&mountlist);
                while ((mp = info.msi_node) != NULL) {
                        if (how & MNTSCAN_NOBUSY) {
                                count = callback(mp, data);
                        } else if (vfs_busy(mp, LK_NOWAIT) == 0) {
                                count = callback(mp, data);
                                if (mp == info.msi_node)
                                        vfs_unbusy(mp);
                        } else {
                                count = 0;
                        }
                        if (count < 0)
                                break;
                        res += count;
                        if (mp == info.msi_node)
                                info.msi_node = TAILQ_NEXT(mp, mnt_list);
                }
        } else if (how & MNTSCAN_REVERSE) {
                info.msi_node = TAILQ_LAST(&mountlist, mntlist);
                while ((mp = info.msi_node) != NULL) {
                        if (how & MNTSCAN_NOBUSY) {
                                count = callback(mp, data);
                        } else if (vfs_busy(mp, LK_NOWAIT) == 0) {
                                count = callback(mp, data);
                                if (mp == info.msi_node)
                                        vfs_unbusy(mp);
                        } else {
                                count = 0;
                        }
                        if (count < 0)
                                break;
                        res += count;
                        if (mp == info.msi_node)
                                info.msi_node = TAILQ_PREV(mp, mntlist, mnt_list);
                }
        }
        TAILQ_REMOVE(&mountscan_list, &info, msi_entry);
        lwkt_reltoken(&mountlist_token);
        return(res);
}

/*
 * MOUNT RELATED VNODE FUNCTIONS
 */

static struct kproc_desc vnlru_kp = {
        "vnlru",
        vnlru_proc,
        &vnlruthread
};
SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)

/*
 * Move a vnode from one mount queue to another.
 *
 * MPSAFE
 */
void
insmntque(struct vnode *vp, struct mount *mp)
{
        struct mount *omp;

        /*
         * Delete from old mount point vnode list, if on one.
         */
        if ((omp = vp->v_mount) != NULL) {
                lwkt_gettoken(&omp->mnt_token);
                KKASSERT(omp == vp->v_mount);
                KASSERT(omp->mnt_nvnodelistsize > 0,
                        ("bad mount point vnode list size"));
                vremovevnodemnt(vp);
                omp->mnt_nvnodelistsize--;
                lwkt_reltoken(&omp->mnt_token);
        }

        /*
         * Insert into list of vnodes for the new mount point, if available.
         * The 'end' of the LRU list is the vnode prior to mp->mnt_syncer.
         */
        if (mp == NULL) {
                vp->v_mount = NULL;
                return;
        }
        lwkt_gettoken(&mp->mnt_token);
        vp->v_mount = mp;
        if (mp->mnt_syncer) {
                TAILQ_INSERT_BEFORE(mp->mnt_syncer, vp, v_nmntvnodes);
        } else {
                TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
        }
        mp->mnt_nvnodelistsize++;
        lwkt_reltoken(&mp->mnt_token);
}


/*
 * Scan the vnodes under a mount point and issue appropriate callbacks.
 *
 * The fastfunc() callback is called with just the mountlist token held
 * (no vnode lock).  It may not block and the vnode may be undergoing
 * modifications while the caller is processing it.  The vnode will
 * not be entirely destroyed, however, due to the fact that the mountlist
 * token is held.  A return value < 0 skips to the next vnode without calling
 * the slowfunc(), a return value > 0 terminates the loop.
 *
 * The slowfunc() callback is called after the vnode has been successfully
 * locked based on passed flags.  The vnode is skipped if it gets rearranged
 * or destroyed while blocking on the lock.  A non-zero return value from
 * the slow function terminates the loop.  The slow function is allowed to
 * arbitrarily block.  The scanning code guarentees consistency of operation
 * even if the slow function deletes or moves the node, or blocks and some
 * other thread deletes or moves the node.
 *
 * NOTE: We hold vmobj_token to prevent a VM object from being destroyed
 *       out from under the fastfunc()'s vnode test.  It will not prevent
 *       v_object from getting NULL'd out but it will ensure that the
 *       pointer (if we race) will remain stable.
 */
int
vmntvnodescan(
    struct mount *mp, 
    int flags,
    int (*fastfunc)(struct mount *mp, struct vnode *vp, void *data),
    int (*slowfunc)(struct mount *mp, struct vnode *vp, void *data),
    void *data
) {
        struct vmntvnodescan_info info;
        struct vnode *vp;
        int r = 0;
        int maxcount = mp->mnt_nvnodelistsize * 2;
        int stopcount = 0;
        int count = 0;

        lwkt_gettoken(&mp->mnt_token);
        lwkt_gettoken(&vmobj_token);

        /*
         * If asked to do one pass stop after iterating available vnodes.
         * Under heavy loads new vnodes can be added while we are scanning,
         * so this isn't perfect.  Create a slop factor of 2x.
         */
        if (flags & VMSC_ONEPASS)
                stopcount = mp->mnt_nvnodelistsize;

        info.vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
        TAILQ_INSERT_TAIL(&mp->mnt_vnodescan_list, &info, entry);

        while ((vp = info.vp) != NULL) {
                if (--maxcount == 0) {
                        kprintf("Warning: excessive fssync iteration\n");
                        maxcount = mp->mnt_nvnodelistsize * 2;
                }

                /*
                 * Skip if visible but not ready, or special (e.g.
                 * mp->mnt_syncer) 
                 */
                if (vp->v_type == VNON)
                        goto next;
                KKASSERT(vp->v_mount == mp);

                /*
                 * Quick test.  A negative return continues the loop without
                 * calling the slow test.  0 continues onto the slow test.
                 * A positive number aborts the loop.
                 */
                if (fastfunc) {
                        if ((r = fastfunc(mp, vp, data)) < 0) {
                                r = 0;
                                goto next;
                        }
                        if (r)
                                break;
                }

                /*
                 * Get a vxlock on the vnode, retry if it has moved or isn't
                 * in the mountlist where we expect it.
                 */
                if (slowfunc) {
                        int error;

                        switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) {
                        case VMSC_GETVP:
                                error = vget(vp, LK_EXCLUSIVE);
                                break;
                        case VMSC_GETVP|VMSC_NOWAIT:
                                error = vget(vp, LK_EXCLUSIVE|LK_NOWAIT);
                                break;
                        case VMSC_GETVX:
                                vx_get(vp);
                                error = 0;
                                break;
                        default:
                                error = 0;
                                break;
                        }
                        if (error)
                                goto next;
                        /*
                         * Do not call the slow function if the vnode is
                         * invalid or if it was ripped out from under us
                         * while we (potentially) blocked.
                         */
                        if (info.vp == vp && vp->v_type != VNON)
                                r = slowfunc(mp, vp, data);

                        /*
                         * Cleanup
                         */
                        switch(flags & (VMSC_GETVP|VMSC_GETVX|VMSC_NOWAIT)) {
                        case VMSC_GETVP:
                        case VMSC_GETVP|VMSC_NOWAIT:
                                vput(vp);
                                break;
                        case VMSC_GETVX:
                                vx_put(vp);
                                break;
                        default:
                                break;
                        }
                        if (r != 0)
                                break;
                }

next:
                /*
                 * Yield after some processing.  Depending on the number
                 * of vnodes, we might wind up running for a long time.
                 * Because threads are not preemptable, time critical
                 * userland processes might starve.  Give them a chance
                 * now and then.
                 */
                if (++count == 10000) {
                        /* We really want to yield a bit, so we simply sleep a tick */
                        tsleep(mp, 0, "vnodescn", 1);
                        count = 0;
                }

                /*
                 * If doing one pass this decrements to zero.  If it starts
                 * at zero it is effectively unlimited for the purposes of
                 * this loop.
                 */
                if (--stopcount == 0)
                        break;

                /*
                 * Iterate.  If the vnode was ripped out from under us
                 * info.vp will already point to the next vnode, otherwise
                 * we have to obtain the next valid vnode ourselves.
                 */
                if (info.vp == vp)
                        info.vp = TAILQ_NEXT(vp, v_nmntvnodes);
        }

        TAILQ_REMOVE(&mp->mnt_vnodescan_list, &info, entry);
        lwkt_reltoken(&vmobj_token);
        lwkt_reltoken(&mp->mnt_token);
        return(r);
}

/*
 * Remove any vnodes in the vnode table belonging to mount point mp.
 *
 * If FORCECLOSE is not specified, there should not be any active ones,
 * return error if any are found (nb: this is a user error, not a
 * system error). If FORCECLOSE is specified, detach any active vnodes
 * that are found.
 *
 * If WRITECLOSE is set, only flush out regular file vnodes open for
 * writing.
 *
 * SKIPSYSTEM causes any vnodes marked VSYSTEM to be skipped.
 *
 * `rootrefs' specifies the base reference count for the root vnode
 * of this filesystem. The root vnode is considered busy if its
 * v_sysref.refcnt exceeds this value. On a successful return, vflush()
 * will call vrele() on the root vnode exactly rootrefs times.
 * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
 * be zero.
 */
#ifdef DIAGNOSTIC
static int busyprt = 0;         /* print out busy vnodes */
SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
#endif

static int vflush_scan(struct mount *mp, struct vnode *vp, void *data);

struct vflush_info {
        int flags;
        int busy;
        thread_t td;
};

int
vflush(struct mount *mp, int rootrefs, int flags)
{
        struct thread *td = curthread;  /* XXX */
        struct vnode *rootvp = NULL;
        int error;
        struct vflush_info vflush_info;

        if (rootrefs > 0) {
                KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
                    ("vflush: bad args"));
                /*
                 * Get the filesystem root vnode. We can vput() it
                 * immediately, since with rootrefs > 0, it won't go away.
                 */
                if ((error = VFS_ROOT(mp, &rootvp)) != 0) {
                        if ((flags & FORCECLOSE) == 0)
                                return (error);
                        rootrefs = 0;
                        /* continue anyway */
                }
                if (rootrefs)
                        vput(rootvp);
        }

        vflush_info.busy = 0;
        vflush_info.flags = flags;
        vflush_info.td = td;
        vmntvnodescan(mp, VMSC_GETVX, NULL, vflush_scan, &vflush_info);

        if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
                /*
                 * If just the root vnode is busy, and if its refcount
                 * is equal to `rootrefs', then go ahead and kill it.
                 */
                KASSERT(vflush_info.busy > 0, ("vflush: not busy"));
                KASSERT(rootvp->v_sysref.refcnt >= rootrefs, ("vflush: rootrefs"));
                if (vflush_info.busy == 1 && rootvp->v_sysref.refcnt == rootrefs) {
                        vx_lock(rootvp);
                        vgone_vxlocked(rootvp);
                        vx_unlock(rootvp);
                        vflush_info.busy = 0;
                }
        }
        if (vflush_info.busy)
                return (EBUSY);
        for (; rootrefs > 0; rootrefs--)
                vrele(rootvp);
        return (0);
}

/*
 * The scan callback is made with an VX locked vnode.
 */
static int
vflush_scan(struct mount *mp, struct vnode *vp, void *data)
{
        struct vflush_info *info = data;
        struct vattr vattr;
        int flags = info->flags;

        /*
         * Skip over a vnodes marked VSYSTEM.
         */
        if ((flags & SKIPSYSTEM) && (vp->v_flag & VSYSTEM)) {
                return(0);
        }

        /*
         * Do not force-close VCHR or VBLK vnodes
         */
        if (vp->v_type == VCHR || vp->v_type == VBLK)
                flags &= ~(WRITECLOSE|FORCECLOSE);

        /*
         * If WRITECLOSE is set, flush out unlinked but still open
         * files (even if open only for reading) and regular file
         * vnodes open for writing. 
         */
        if ((flags & WRITECLOSE) &&
            (vp->v_type == VNON ||
            (VOP_GETATTR(vp, &vattr) == 0 &&
            vattr.va_nlink > 0)) &&
            (vp->v_writecount == 0 || vp->v_type != VREG)) {
                return(0);
        }

        /*
         * If we are the only holder (refcnt of 1) or the vnode is in
         * termination (refcnt < 0), we can vgone the vnode.
         */
        if (vp->v_sysref.refcnt <= 1) {
                vgone_vxlocked(vp);
                return(0);
        }

        /*
         * If FORCECLOSE is set, forcibly destroy the vnode and then move
         * it to a dummymount structure so vop_*() functions don't deref
         * a NULL pointer.
         */
        if (flags & FORCECLOSE) {
                vhold(vp);
                vgone_vxlocked(vp);
                if (vp->v_mount == NULL)
                        insmntque(vp, &dummymount);
                vdrop(vp);
                return(0);
        }
        if (vp->v_type == VCHR || vp->v_type == VBLK)
                kprintf("vflush: Warning, cannot destroy busy device vnode\n");
#ifdef DIAGNOSTIC
        if (busyprt)
                vprint("vflush: busy vnode", vp);
#endif
        ++info->busy;
        return(0);
}

void
add_bio_ops(struct bio_ops *ops)
{
        TAILQ_INSERT_TAIL(&bio_ops_list, ops, entry);
}

void
rem_bio_ops(struct bio_ops *ops)
{
        TAILQ_REMOVE(&bio_ops_list, ops, entry);
}

/*
 * This calls the bio_ops io_sync function either for a mount point
 * or generally.
 *
 * WARNING: softdeps is weirdly coded and just isn't happy unless
 * io_sync is called with a NULL mount from the general syncing code.
 */
void
bio_ops_sync(struct mount *mp)
{
        struct bio_ops *ops;

        if (mp) {
                if ((ops = mp->mnt_bioops) != NULL)
                        ops->io_sync(mp);
        } else {
                TAILQ_FOREACH(ops, &bio_ops_list, entry) {
                        ops->io_sync(NULL);
                }
        }
}

/*
 * Lookup a mount point by nch
 */
struct mount *
mount_get_by_nc(struct namecache *ncp)
{
        struct mount *mp = NULL;

        lwkt_gettoken(&mountlist_token);
        TAILQ_FOREACH(mp, &mountlist, mnt_list) {
                if (ncp == mp->mnt_ncmountpt.ncp)
                        break;
        }
        lwkt_reltoken(&mountlist_token);
        return (mp);
}