kernel - Finish implementing PG_RAM / pipelined mmap operation

[dragonfly.git] / sys / vfs / union / union_vnops.c

/*
 * Copyright (c) 1992, 1993, 1994, 1995 Jan-Simon Pendry.
 * Copyright (c) 1992, 1993, 1994, 1995
 *      The Regents of the University of California.  All rights reserved.
 *
 * This code is derived from software contributed to Berkeley by
 * Jan-Simon Pendry.
 *
 * 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.
 *
 *      @(#)union_vnops.c       8.32 (Berkeley) 6/23/95
 * $FreeBSD: src/sys/miscfs/union/union_vnops.c,v 1.72 1999/12/15 23:02:14 eivind Exp $
 * $DragonFly: src/sys/vfs/union/union_vnops.c,v 1.39 2007/11/20 21:03:51 dillon Exp $
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/proc.h>
#include <sys/fcntl.h>
#include <sys/stat.h>
#include <sys/kernel.h>
#include <sys/vnode.h>
#include <sys/mount.h>
#include <sys/namei.h>
#include <sys/malloc.h>
#include <sys/buf.h>
#include <sys/lock.h>
#include <sys/sysctl.h>
#include "union.h"

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

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

int uniondebug = 0;

#if UDEBUG_ENABLED
SYSCTL_INT(_vfs, OID_AUTO, uniondebug, CTLFLAG_RW, &uniondebug, 0, "");
#else
SYSCTL_INT(_vfs, OID_AUTO, uniondebug, CTLFLAG_RD, &uniondebug, 0, "");
#endif

static int      union_access (struct vop_access_args *ap);
static int      union_advlock (struct vop_advlock_args *ap);
static int      union_bmap (struct vop_bmap_args *ap);
static int      union_close (struct vop_close_args *ap);
static int      union_create (struct vop_old_create_args *ap);
static int      union_fsync (struct vop_fsync_args *ap);
static int      union_getattr (struct vop_getattr_args *ap);
static int      union_inactive (struct vop_inactive_args *ap);
static int      union_ioctl (struct vop_ioctl_args *ap);
static int      union_link (struct vop_old_link_args *ap);
static int      union_lookup (struct vop_old_lookup_args *ap);
static int      union_lookup1 (struct vnode *udvp, struct vnode **dvp,
                                   struct vnode **vpp,
                                   struct componentname *cnp);
static int      union_mkdir (struct vop_old_mkdir_args *ap);
static int      union_mknod (struct vop_old_mknod_args *ap);
static int      union_mmap (struct vop_mmap_args *ap);
static int      union_open (struct vop_open_args *ap);
static int      union_pathconf (struct vop_pathconf_args *ap);
static int      union_print (struct vop_print_args *ap);
static int      union_read (struct vop_read_args *ap);
static int      union_readdir (struct vop_readdir_args *ap);
static int      union_readlink (struct vop_readlink_args *ap);
static int      union_reclaim (struct vop_reclaim_args *ap);
static int      union_remove (struct vop_old_remove_args *ap);
static int      union_rename (struct vop_old_rename_args *ap);
static int      union_rmdir (struct vop_old_rmdir_args *ap);
static int      union_poll (struct vop_poll_args *ap);
static int      union_setattr (struct vop_setattr_args *ap);
static int      union_strategy (struct vop_strategy_args *ap);
static int      union_getpages (struct vop_getpages_args *ap);
static int      union_putpages (struct vop_putpages_args *ap);
static int      union_symlink (struct vop_old_symlink_args *ap);
static int      union_whiteout (struct vop_old_whiteout_args *ap);
static int      union_write (struct vop_read_args *ap);

static __inline
struct vnode *
union_lock_upper(struct union_node *un, struct thread *td)
{
        struct vnode *uppervp;

        if ((uppervp = un->un_uppervp) != NULL) {
                vref(uppervp);
                vn_lock(uppervp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY);
        }
        KASSERT((uppervp == NULL || uppervp->v_sysref.refcnt > 0), ("uppervp usecount is 0"));
        return(uppervp);
}

static __inline
struct vnode *
union_ref_upper(struct union_node *un)
{
        struct vnode *uppervp;

        if ((uppervp = un->un_uppervp) != NULL) {
                vref(uppervp);
                if (uppervp->v_flag & VRECLAIMED) {
                        vrele(uppervp);
                        return (NULLVP);
                }
        }
        return (uppervp);
}

static __inline
void
union_unlock_upper(struct vnode *uppervp, struct thread *td)
{
        vput(uppervp);
}

static __inline
struct vnode *
union_lock_other(struct union_node *un, struct thread *td)
{
        struct vnode *vp;

        if (un->un_uppervp != NULL) {
                vp = union_lock_upper(un, td);
        } else if ((vp = un->un_lowervp) != NULL) {
                vref(vp);
                vn_lock(vp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY);
        }
        return(vp);
}

static __inline
void
union_unlock_other(struct vnode *vp, struct thread *td)
{
        vput(vp);
}

/*
 *      union_lookup:
 *
 *      udvp    must be exclusively locked on call and will remain 
 *              exclusively locked on return.  This is the mount point 
 *              for out filesystem.
 *
 *      dvp     Our base directory, locked and referenced.
 *              The passed dvp will be dereferenced and unlocked on return
 *              and a new dvp will be returned which is locked and 
 *              referenced in the same variable.
 *
 *      vpp     is filled in with the result if no error occured,
 *              locked and ref'd.
 *
 *              If an error is returned, *vpp is set to NULLVP.  If no
 *              error occurs, *vpp is returned with a reference and an
 *              exclusive lock.
 */

static int
union_lookup1(struct vnode *udvp, struct vnode **pdvp, struct vnode **vpp,
              struct componentname *cnp)
{
        int error;
        struct thread *td = cnp->cn_td;
        struct vnode *dvp = *pdvp;
        struct vnode *tdvp;
        struct mount *mp;

        /*
         * If stepping up the directory tree, check for going
         * back across the mount point, in which case do what
         * lookup would do by stepping back down the mount
         * hierarchy.
         */
        if (cnp->cn_flags & CNP_ISDOTDOT) {
                while ((dvp != udvp) && (dvp->v_flag & VROOT)) {
                        /*
                         * Don't do the NOCROSSMOUNT check
                         * at this level.  By definition,
                         * union fs deals with namespaces, not
                         * filesystems.
                         */
                        tdvp = dvp;
                        dvp = dvp->v_mount->mnt_vnodecovered;
                        vref(dvp);
                        vput(tdvp);
                        vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
                }
        }

        /*
         * Set return dvp to be the upperdvp 'parent directory.
         */
        *pdvp = dvp;

        /*
         * If the VOP_LOOKUP call generates an error, tdvp is invalid and no
         * changes will have been made to dvp, so we are set to return.
         */

        error = VOP_LOOKUP(dvp, &tdvp, cnp);
        if (error) {
                UDEBUG(("dvp %p error %d flags %lx\n", dvp, error, cnp->cn_flags));
                *vpp = NULL;
                return (error);
        }

        /*
         * The parent directory will have been unlocked, unless lookup
         * found the last component or if dvp == tdvp (tdvp must be locked).
         *
         * We want our dvp to remain locked and ref'd.  We also want tdvp
         * to remain locked and ref'd.
         */
        UDEBUG(("parentdir %p result %p flag %lx\n", dvp, tdvp, cnp->cn_flags));

#if 0
        if (dvp != tdvp && (cnp->cn_flags & CNP_XXXISLASTCN) == 0)
                vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
#endif

        /*
         * Lastly check if the current node is a mount point in
         * which case walk up the mount hierarchy making sure not to
         * bump into the root of the mount tree (ie. dvp != udvp).
         *
         * We use dvp as a temporary variable here, it is no longer related
         * to the dvp above.  However, we have to ensure that both *pdvp and
         * tdvp are locked on return.
         */

        dvp = tdvp;
        while (
            dvp != udvp && 
            (dvp->v_type == VDIR) &&
            (mp = dvp->v_mountedhere)
        ) {
                int relock_pdvp = 0;

                if (vfs_busy(mp, 0))
                        continue;

                if (dvp == *pdvp)
                        relock_pdvp = 1;
                vput(dvp);
                dvp = NULL;
                error = VFS_ROOT(mp, &dvp);

                vfs_unbusy(mp);

                if (relock_pdvp)
                        vn_lock(*pdvp, LK_EXCLUSIVE | LK_RETRY);

                if (error) {
                        *vpp = NULL;
                        return (error);
                }
        }
        *vpp = dvp;
        return (0);
}

/*
 * union_lookup(struct vnode *a_dvp, struct vnode **a_vpp,
 *              struct componentname *a_cnp)
 */
static int
union_lookup(struct vop_old_lookup_args *ap)
{
        int error;
        int uerror, lerror;
        struct vnode *uppervp, *lowervp;
        struct vnode *upperdvp, *lowerdvp;
        struct vnode *dvp = ap->a_dvp;          /* starting dir */
        struct union_node *dun = VTOUNION(dvp); /* associated union node */
        struct componentname *cnp = ap->a_cnp;
        struct thread *td = cnp->cn_td;
        int lockparent = cnp->cn_flags & CNP_LOCKPARENT;
        struct union_mount *um = MOUNTTOUNIONMOUNT(dvp->v_mount);
        struct ucred *saved_cred = NULL;
        int iswhiteout;
        struct vattr va;

        *ap->a_vpp = NULLVP;

        /*
         * Disallow write attemps to the filesystem mounted read-only.
         */
        if ((dvp->v_mount->mnt_flag & MNT_RDONLY) &&
            (cnp->cn_nameiop == NAMEI_DELETE || cnp->cn_nameiop == NAMEI_RENAME)) {
                return (EROFS);
        }

        /*
         * For any lookup's we do, always return with the parent locked
         */
        cnp->cn_flags |= CNP_LOCKPARENT;

        lowerdvp = dun->un_lowervp;
        uppervp = NULLVP;
        lowervp = NULLVP;
        iswhiteout = 0;

        uerror = ENOENT;
        lerror = ENOENT;

        /*
         * Get a private lock on uppervp and a reference, effectively 
         * taking it out of the union_node's control.
         *
         * We must lock upperdvp while holding our lock on dvp
         * to avoid a deadlock.
         */
        upperdvp = union_lock_upper(dun, td);

        /*
         * do the lookup in the upper level.
         * if that level comsumes additional pathnames,
         * then assume that something special is going
         * on and just return that vnode.
         */
        if (upperdvp != NULLVP) {
                /*
                 * We do not have to worry about the DOTDOT case, we've
                 * already unlocked dvp.
                 */
                UDEBUG(("A %p\n", upperdvp));

                /*
                 * Do the lookup.   We must supply a locked and referenced
                 * upperdvp to the function and will get a new locked and
                 * referenced upperdvp back with the old having been 
                 * dereferenced.
                 *
                 * If an error is returned, uppervp will be NULLVP.  If no
                 * error occurs, uppervp will be the locked and referenced
                 * return vnode or possibly NULL, depending on what is being
                 * requested.  It is possible that the returned uppervp
                 * will be the same as upperdvp.
                 */
                uerror = union_lookup1(um->um_uppervp, &upperdvp, &uppervp, cnp);
                UDEBUG((
                    "uerror %d upperdvp %p %d/%d, uppervp %p ref=%d/lck=%d\n",
                    uerror,
                    upperdvp,
                    upperdvp->v_sysref.refcnt,
                    vn_islocked(upperdvp),
                    uppervp,
                    (uppervp ? uppervp->v_sysref.refcnt : -99),
                    (uppervp ? vn_islocked(uppervp) : -99)
                ));

                /*
                 * Disallow write attemps to the filesystem mounted read-only.
                 */
                if (uerror == EJUSTRETURN && 
                    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
                    (cnp->cn_nameiop == NAMEI_CREATE || cnp->cn_nameiop == NAMEI_RENAME)) {
                        error = EROFS;
                        goto out;
                }

                /*
                 * Special case.  If cn_consume != 0 skip out.  The result
                 * of the lookup is transfered to our return variable.  If
                 * an error occured we have to throw away the results.
                 */

                if (cnp->cn_consume != 0) {
                        if ((error = uerror) == 0) {
                                *ap->a_vpp = uppervp;
                                uppervp = NULL;
                        }
                        goto out;
                }

                /*
                 * Calculate whiteout, fall through
                 */

                if (uerror == ENOENT || uerror == EJUSTRETURN) {
                        if (cnp->cn_flags & CNP_ISWHITEOUT) {
                                iswhiteout = 1;
                        } else if (lowerdvp != NULLVP) {
                                int terror;

                                terror = VOP_GETATTR(upperdvp, &va);
                                if (terror == 0 && (va.va_flags & OPAQUE))
                                        iswhiteout = 1;
                        }
                }
        }

        /*
         * in a similar way to the upper layer, do the lookup
         * in the lower layer.   this time, if there is some
         * component magic going on, then vput whatever we got
         * back from the upper layer and return the lower vnode
         * instead.
         */

        if (lowerdvp != NULLVP && !iswhiteout) {
                int nameiop;

                UDEBUG(("B %p\n", lowerdvp));

                /*
                 * Force only LOOKUPs on the lower node, since
                 * we won't be making changes to it anyway.
                 */
                nameiop = cnp->cn_nameiop;
                cnp->cn_nameiop = NAMEI_LOOKUP;
                if (um->um_op == UNMNT_BELOW) {
                        saved_cred = cnp->cn_cred;
                        cnp->cn_cred = um->um_cred;
                }

                /*
                 * We shouldn't have to worry about locking interactions
                 * between the lower layer and our union layer (w.r.t.
                 * `..' processing) because we don't futz with lowervp
                 * locks in the union-node instantiation code path.
                 *
                 * union_lookup1() requires lowervp to be locked on entry,
                 * and it will be unlocked on return.  The ref count will
                 * not change.  On return lowervp doesn't represent anything
                 * to us so we NULL it out.
                 */
                vref(lowerdvp);
                vn_lock(lowerdvp, LK_EXCLUSIVE | LK_RETRY);
                lerror = union_lookup1(um->um_lowervp, &lowerdvp, &lowervp, cnp);
                if (lowerdvp == lowervp)
                        vrele(lowerdvp);
                else
                        vput(lowerdvp);
                lowerdvp = NULL;        /* lowerdvp invalid after vput */

                if (um->um_op == UNMNT_BELOW)
                        cnp->cn_cred = saved_cred;
                cnp->cn_nameiop = nameiop;

                if (cnp->cn_consume != 0 || lerror == EACCES) {
                        if ((error = lerror) == 0) {
                                *ap->a_vpp = lowervp;
                                lowervp = NULL;
                        }
                        goto out;
                }
        } else {
                UDEBUG(("C %p\n", lowerdvp));
                if ((cnp->cn_flags & CNP_ISDOTDOT) && dun->un_pvp != NULLVP) {
                        if ((lowervp = LOWERVP(dun->un_pvp)) != NULL) {
                                vref(lowervp);
                                vn_lock(lowervp, LK_EXCLUSIVE | LK_RETRY);
                                lerror = 0;
                        }
                }
        }

        /*
         * Ok.  Now we have uerror, uppervp, upperdvp, lerror, and lowervp.
         *
         * 1. If both layers returned an error, select the upper layer.
         *
         * 2. If the upper layer faile and the bottom layer succeeded,
         *    two subcases occur:
         *
         *      a.  The bottom vnode is not a directory, in which case
         *          just return a new union vnode referencing an
         *          empty top layer and the existing bottom layer.
         *
         *      b.  The button vnode is a directory, in which case
         *          create a new directory in the top layer and
         *          and fall through to case 3.
         *
         * 3. If the top layer succeeded then return a new union
         *    vnode referencing whatever the new top layer and
         *    whatever the bottom layer returned.
         */

        /* case 1. */
        if ((uerror != 0) && (lerror != 0)) {
                error = uerror;
                goto out;
        }

        /* case 2. */
        if (uerror != 0 /* && (lerror == 0) */ ) {
                if (lowervp->v_type == VDIR) { /* case 2b. */
                        KASSERT(uppervp == NULL, ("uppervp unexpectedly non-NULL"));
                        /*
                         * oops, uppervp has a problem, we may have to shadow.
                         */
                        uerror = union_mkshadow(um, upperdvp, cnp, &uppervp);
                        if (uerror) {
                                error = uerror;
                                goto out;
                        }
                }
        }

        /*
         * Must call union_allocvp with both the upper and lower vnodes
         * referenced and the upper vnode locked.   ap->a_vpp is returned 
         * referenced and locked.  lowervp, uppervp, and upperdvp are 
         * absorbed by union_allocvp() whether it succeeds or fails.
         *
         * upperdvp is the parent directory of uppervp which may be
         * different, depending on the path, from dvp->un_uppervp.  That's
         * why it is a separate argument.  Note that it must be unlocked.
         *
         * dvp must be locked on entry to the call and will be locked on
         * return.
         */

        if (uppervp && uppervp != upperdvp)
                vn_unlock(uppervp);
        if (lowervp)
                vn_unlock(lowervp);
        if (upperdvp)
                vn_unlock(upperdvp);

        error = union_allocvp(ap->a_vpp, dvp->v_mount, dvp, upperdvp, cnp,
                              uppervp, lowervp, 1);

        UDEBUG(("Create %p = %p %p refs=%d\n", *ap->a_vpp, uppervp, lowervp, (*ap->a_vpp) ? ((*ap->a_vpp)->v_sysref.refcnt) : -99));

        uppervp = NULL;
        upperdvp = NULL;
        lowervp = NULL;

        /* 
         *      Termination Code
         *
         *      - put away any extra junk laying around.  Note that lowervp
         *        (if not NULL) will never be the same as *ap->a_vp and 
         *        neither will uppervp, because when we set that state we 
         *        NULL-out lowervp or uppervp.  On the otherhand, upperdvp
         *        may match uppervp or *ap->a_vpp.
         *
         *      - relock/unlock dvp if appropriate.
         */

out:
        if (upperdvp) {
                if (upperdvp == uppervp || upperdvp == *ap->a_vpp)
                        vrele(upperdvp);
                else
                        vput(upperdvp);
        }

        if (uppervp)
                vput(uppervp);

        if (lowervp)
                vput(lowervp);

        /*
         * Restore LOCKPARENT state
         */

        if (!lockparent)
                cnp->cn_flags &= ~CNP_LOCKPARENT;

        UDEBUG(("Out %d vpp %p/%d lower %p upper %p\n", error, *ap->a_vpp,
                ((*ap->a_vpp) ? (*ap->a_vpp)->v_sysref.refcnt : -99),
                lowervp, uppervp));

        /*
         * dvp lock state, determine whether to relock dvp.  dvp is expected
         * to be locked on return if:
         *
         *      - there was an error (except not EJUSTRETURN), or
         *      - we hit the last component and lockparent is true
         *
         * dvp_is_locked is the current state of the dvp lock, not counting
         * the possibility that *ap->a_vpp == dvp (in which case it is locked
         * anyway).  Note that *ap->a_vpp == dvp only if no error occured.
         */

        if (*ap->a_vpp != dvp) {
                if ((error == 0 || error == EJUSTRETURN) && !lockparent) {
                        vn_unlock(dvp);
                }
        }

        /*
         * Diagnostics
         */

#ifdef DIAGNOSTIC
        if (cnp->cn_namelen == 1 &&
            cnp->cn_nameptr[0] == '.' &&
            *ap->a_vpp != dvp) {
                panic("union_lookup returning . (%p) not same as startdir (%p)", ap->a_vpp, dvp);
        }
#endif

        return (error);
}

/*
 *      union_create:
 *
 * a_dvp is locked on entry and remains locked on return.  a_vpp is returned
 * locked if no error occurs, otherwise it is garbage.
 *
 * union_create(struct vnode *a_dvp, struct vnode **a_vpp,
 *              struct componentname *a_cnp, struct vattr *a_vap)
 */
static int
union_create(struct vop_old_create_args *ap)
{
        struct union_node *dun = VTOUNION(ap->a_dvp);
        struct componentname *cnp = ap->a_cnp;
        struct thread *td = cnp->cn_td;
        struct vnode *dvp;
        int error = EROFS;

        if ((dvp = union_lock_upper(dun, td)) != NULL) {
                struct vnode *vp;
                struct mount *mp;

                error = VOP_CREATE(dvp, &vp, cnp, ap->a_vap);
                if (error == 0) {
                        mp = ap->a_dvp->v_mount;
                        vn_unlock(vp);
                        UDEBUG(("ALLOCVP-1 FROM %p REFS %d\n", vp, vp->v_sysref.refcnt));
                        error = union_allocvp(ap->a_vpp, mp, NULLVP, NULLVP,
                                cnp, vp, NULLVP, 1);
                        UDEBUG(("ALLOCVP-2B FROM %p REFS %d\n", *ap->a_vpp, vp->v_sysref.refcnt));
                }
                union_unlock_upper(dvp, td);
        }
        return (error);
}

/*
 * union_whiteout(struct vnode *a_dvp, struct componentname *a_cnp,
 *                int a_flags)
 */
static int
union_whiteout(struct vop_old_whiteout_args *ap)
{
        struct union_node *un = VTOUNION(ap->a_dvp);
        struct componentname *cnp = ap->a_cnp;
        struct vnode *uppervp;
        int error = EOPNOTSUPP;

        if ((uppervp = union_lock_upper(un, cnp->cn_td)) != NULLVP) {
                error = VOP_WHITEOUT(un->un_uppervp, cnp, ap->a_flags);
                union_unlock_upper(uppervp, cnp->cn_td);
        }
        return(error);
}

/*
 *      union_mknod:
 *
 *      a_dvp is locked on entry and should remain locked on return.
 *      a_vpp is garbagre whether an error occurs or not.
 *
 * union_mknod(struct vnode *a_dvp, struct vnode **a_vpp,
 *              struct componentname *a_cnp, struct vattr *a_vap)
 */
static int
union_mknod(struct vop_old_mknod_args *ap)
{
        struct union_node *dun = VTOUNION(ap->a_dvp);
        struct componentname *cnp = ap->a_cnp;
        struct vnode *dvp;
        int error = EROFS;

        if ((dvp = union_lock_upper(dun, cnp->cn_td)) != NULL) {
                error = VOP_MKNOD(dvp, ap->a_vpp, cnp, ap->a_vap);
                union_unlock_upper(dvp, cnp->cn_td);
        }
        return (error);
}

/*
 *      union_open:
 *
 *      run open VOP.  When opening the underlying vnode we have to mimic
 *      vn_open.  What we *really* need to do to avoid screwups if the
 *      open semantics change is to call vn_open().  For example, ufs blows
 *      up if you open a file but do not vmio it prior to writing.
 *
 * union_open(struct vnode *a_vp, int a_mode,
 *            struct ucred *a_cred, struct thread *a_td)
 */
static int
union_open(struct vop_open_args *ap)
{
        struct union_node *un = VTOUNION(ap->a_vp);
        struct vnode *tvp;
        int mode = ap->a_mode;
        struct ucred *cred = ap->a_cred;
        struct thread *td = ap->a_td;
        int error = 0;
        int tvpisupper = 1;

        /*
         * If there is an existing upper vp then simply open that.
         * The upper vp takes precedence over the lower vp.  When opening
         * a lower vp for writing copy it to the uppervp and then open the
         * uppervp.
         *
         * At the end of this section tvp will be left locked.
         */
        if ((tvp = union_lock_upper(un, td)) == NULLVP) {
                /*
                 * If the lower vnode is being opened for writing, then
                 * copy the file contents to the upper vnode and open that,
                 * otherwise can simply open the lower vnode.
                 */
                tvp = un->un_lowervp;
                if ((ap->a_mode & FWRITE) && (tvp->v_type == VREG)) {
                        int docopy = !(mode & O_TRUNC);
                        error = union_copyup(un, docopy, cred, td);
                        tvp = union_lock_upper(un, td);
                } else {
                        un->un_openl++;
                        vref(tvp);
                        vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
                        tvpisupper = 0;
                }
        }

        /*
         * We are holding the correct vnode, open it.  Note
         * that in DragonFly, VOP_OPEN is responsible for associating
         * a VM object with the vnode if the vnode is mappable or the
         * underlying filesystem uses buffer cache calls on it.
         */
        if (error == 0)
                error = VOP_OPEN(tvp, mode, cred, NULL);

        /*
         * Release any locks held
         */
        if (tvpisupper) {
                if (tvp)
                        union_unlock_upper(tvp, td);
        } else {
                vput(tvp);
        }
        return (error);
}

/*
 *      union_close:
 *
 *      It is unclear whether a_vp is passed locked or unlocked.  Whatever
 *      the case we do not change it.
 *
 * union_close(struct vnode *a_vp, int a_fflag, struct ucred *a_cred,
 *              struct thread *a_td)
 */
static int
union_close(struct vop_close_args *ap)
{
        struct union_node *un = VTOUNION(ap->a_vp);
        struct vnode *vp;

        if ((vp = un->un_uppervp) == NULLVP) {
#ifdef UNION_DIAGNOSTIC
                if (un->un_openl <= 0)
                        panic("union: un_openl cnt");
#endif
                --un->un_openl;
                vp = un->un_lowervp;
        }
        ap->a_head.a_ops = *vp->v_ops;
        ap->a_vp = vp;
        return(vop_close_ap(ap));
}

/*
 * Check access permission on the union vnode.
 * The access check being enforced is to check
 * against both the underlying vnode, and any
 * copied vnode.  This ensures that no additional
 * file permissions are given away simply because
 * the user caused an implicit file copy.
 *
 * union_access(struct vnode *a_vp, int a_mode,
 *              struct ucred *a_cred, struct thread *a_td)
 */
static int
union_access(struct vop_access_args *ap)
{
        struct union_node *un = VTOUNION(ap->a_vp);
        struct thread *td = ap->a_td;
        int error = EACCES;
        struct vnode *vp;

        /*
         * Disallow write attempts on filesystems mounted read-only.
         */
        if ((ap->a_mode & VWRITE) && 
            (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)) {
                switch (ap->a_vp->v_type) {
                case VREG: 
                case VDIR:
                case VLNK:
                        return (EROFS);
                default:
                        break;
                }
        }

        if ((vp = union_lock_upper(un, td)) != NULLVP) {
                ap->a_head.a_ops = *vp->v_ops;
                ap->a_vp = vp;
                error = vop_access_ap(ap);
                union_unlock_upper(vp, td);
                return(error);
        }

        if ((vp = un->un_lowervp) != NULLVP) {
                vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
                ap->a_head.a_ops = *vp->v_ops;
                ap->a_vp = vp;

                /*
                 * Remove VWRITE from a_mode if our mount point is RW, because
                 * we want to allow writes and lowervp may be read-only.
                 */
                if ((un->un_vnode->v_mount->mnt_flag & MNT_RDONLY) == 0)
                        ap->a_mode &= ~VWRITE;

                error = vop_access_ap(ap);
                if (error == 0) {
                        struct union_mount *um;

                        um = MOUNTTOUNIONMOUNT(un->un_vnode->v_mount);

                        if (um->um_op == UNMNT_BELOW) {
                                ap->a_cred = um->um_cred;
                                error = vop_access_ap(ap);
                        }
                }
                vn_unlock(vp);
        }
        return(error);
}

/*
 * We handle getattr only to change the fsid and
 * track object sizes
 *
 * It's not clear whether VOP_GETATTR is to be
 * called with the vnode locked or not.  stat() calls
 * it with (vp) locked, and fstat calls it with
 * (vp) unlocked. 
 *
 * Because of this we cannot use our normal locking functions
 * if we do not intend to lock the main a_vp node.  At the moment
 * we are running without any specific locking at all, but beware
 * to any programmer that care must be taken if locking is added
 * to this function.
 *
 * union_getattr(struct vnode *a_vp, struct vattr *a_vap,
 *               struct ucred *a_cred, struct thread *a_td)
 */
static int
union_getattr(struct vop_getattr_args *ap)
{
        int error;
        struct union_node *un = VTOUNION(ap->a_vp);
        struct vnode *vp;
        struct vattr *vap;
        struct vattr va;

        /*
         * Some programs walk the filesystem hierarchy by counting
         * links to directories to avoid stat'ing all the time.
         * This means the link count on directories needs to be "correct".
         * The only way to do that is to call getattr on both layers
         * and fix up the link count.  The link count will not necessarily
         * be accurate but will be large enough to defeat the tree walkers.
         */

        vap = ap->a_vap;

        if ((vp = un->un_uppervp) != NULLVP) {
                error = VOP_GETATTR(vp, vap);
                if (error)
                        return (error);
                /* XXX isn't this dangerouso without a lock? */
                union_newsize(ap->a_vp, vap->va_size, VNOVAL);
        }

        if (vp == NULLVP) {
                vp = un->un_lowervp;
        } else if (vp->v_type == VDIR && un->un_lowervp != NULLVP) {
                vp = un->un_lowervp;
                vap = &va;
        } else {
                vp = NULLVP;
        }

        if (vp != NULLVP) {
                error = VOP_GETATTR(vp, vap);
                if (error)
                        return (error);
                /* XXX isn't this dangerous without a lock? */
                union_newsize(ap->a_vp, VNOVAL, vap->va_size);
        }

        if ((vap != ap->a_vap) && (vap->va_type == VDIR))
                ap->a_vap->va_nlink += vap->va_nlink;
        return (0);
}

/*
 * union_setattr(struct vnode *a_vp, struct vattr *a_vap,
 *               struct ucred *a_cred, struct thread *a_td)
 */
static int
union_setattr(struct vop_setattr_args *ap)
{
        struct union_node *un = VTOUNION(ap->a_vp);
        struct thread *td = ap->a_td;
        struct vattr *vap = ap->a_vap;
        struct vnode *uppervp;
        int error;

        /*
         * Disallow write attempts on filesystems mounted read-only.
         */
        if ((ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) &&
            (vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
             vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
             vap->va_mtime.tv_sec != VNOVAL || 
             vap->va_mode != (mode_t)VNOVAL)) {
                return (EROFS);
        }

        /*
         * Handle case of truncating lower object to zero size,
         * by creating a zero length upper object.  This is to
         * handle the case of open with O_TRUNC and O_CREAT.
         */
        if (un->un_uppervp == NULLVP && (un->un_lowervp->v_type == VREG)) {
                error = union_copyup(un, (ap->a_vap->va_size != 0),
                            ap->a_cred, ap->a_td);
                if (error)
                        return (error);
        }

        /*
         * Try to set attributes in upper layer,
         * otherwise return read-only filesystem error.
         */
        error = EROFS;
        if ((uppervp = union_lock_upper(un, td)) != NULLVP) {
                error = VOP_SETATTR(un->un_uppervp, ap->a_vap, ap->a_cred);
                if ((error == 0) && (ap->a_vap->va_size != VNOVAL))
                        union_newsize(ap->a_vp, ap->a_vap->va_size, VNOVAL);
                union_unlock_upper(uppervp, td);
        }
        return (error);
}

/*
 *      union_getpages:
 */

static int
union_getpages(struct vop_getpages_args *ap)
{
        int r;

        r = vnode_pager_generic_getpages(ap->a_vp, ap->a_m,
                                         ap->a_count, ap->a_reqpage,
                                         ap->a_seqaccess);
        return(r);
}

/*
 *      union_putpages:
 */

static int
union_putpages(struct vop_putpages_args *ap)
{
        int r;

        r = vnode_pager_generic_putpages(ap->a_vp, ap->a_m, ap->a_count,
                ap->a_sync, ap->a_rtvals);
        return(r);
}

/*
 * union_read(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
 *            struct ucred *a_cred)
 */
static int
union_read(struct vop_read_args *ap)
{
        struct union_node *un = VTOUNION(ap->a_vp);
        struct thread *td = ap->a_uio->uio_td;
        struct vnode *uvp;
        int error;

        uvp = union_lock_other(un, td);
        KASSERT(uvp != NULL, ("union_read: backing vnode missing!"));

        if (ap->a_vp->v_flag & VOBJBUF)
                union_vm_coherency(ap->a_vp, ap->a_uio, 0);

        error = VOP_READ(uvp, ap->a_uio, ap->a_ioflag, ap->a_cred);
        union_unlock_other(uvp, td);

        /*
         * XXX
         * perhaps the size of the underlying object has changed under
         * our feet.  take advantage of the offset information present
         * in the uio structure.
         */
        if (error == 0) {
                struct union_node *un = VTOUNION(ap->a_vp);
                off_t cur = ap->a_uio->uio_offset;

                if (uvp == un->un_uppervp) {
                        if (cur > un->un_uppersz)
                                union_newsize(ap->a_vp, cur, VNOVAL);
                } else {
                        if (cur > un->un_lowersz)
                                union_newsize(ap->a_vp, VNOVAL, cur);
                }
        }
        return (error);
}

/*
 * union_write(struct vnode *a_vp, struct uio *a_uio, int a_ioflag,
 *              struct ucred *a_cred)
 */
static int
union_write(struct vop_read_args *ap)
{
        struct union_node *un = VTOUNION(ap->a_vp);
        struct thread *td = ap->a_uio->uio_td;
        struct vnode *uppervp;
        int error;

        if ((uppervp = union_lock_upper(un, td)) == NULLVP)
                panic("union: missing upper layer in write");

        /*
         * Since our VM pages are associated with our vnode rather then
         * the real vnode, and since we do not run our reads and writes 
         * through our own VM cache, we have a VM/VFS coherency problem. 
         * We solve them by invalidating or flushing the associated VM
         * pages prior to allowing a normal read or write to occur.
         *
         * VM-backed writes (UIO_NOCOPY) have to be converted to normal
         * writes because we are not cache-coherent.  Normal writes need
         * to be made coherent with our VM-backing store, which we do by
         * first flushing any dirty VM pages associated with the write
         * range, and then destroying any clean VM pages associated with
         * the write range.
         */

        if (ap->a_uio->uio_segflg == UIO_NOCOPY) {
                ap->a_uio->uio_segflg = UIO_SYSSPACE;
        } else if (ap->a_vp->v_flag & VOBJBUF) {
                union_vm_coherency(ap->a_vp, ap->a_uio, 1);
        }

        error = VOP_WRITE(uppervp, ap->a_uio, ap->a_ioflag, ap->a_cred);

        /*
         * the size of the underlying object may be changed by the
         * write.
         */
        if (error == 0) {
                off_t cur = ap->a_uio->uio_offset;

                if (cur > un->un_uppersz)
                        union_newsize(ap->a_vp, cur, VNOVAL);
        }
        union_unlock_upper(uppervp, td);
        return (error);
}

/*
 * union_ioctl(struct vnode *a_vp, int a_command, caddr_t a_data, int a_fflag,
 *              struct ucred *a_cred, struct thread *a_td)
 */
static int
union_ioctl(struct vop_ioctl_args *ap)
{
        struct vnode *ovp = OTHERVP(ap->a_vp);

        ap->a_head.a_ops = *ovp->v_ops;
        ap->a_vp = ovp;
        return(vop_ioctl_ap(ap));
}

/*
 * union_poll(struct vnode *a_vp, int a_events, struct ucred *a_cred,
 *            struct thread *a_td)
 */
static int
union_poll(struct vop_poll_args *ap)
{
        struct vnode *ovp = OTHERVP(ap->a_vp);

        ap->a_head.a_ops = *ovp->v_ops;
        ap->a_vp = ovp;
        return(vop_poll_ap(ap));
}

/*
 * union_mmap(struct vnode *a_vp, int a_fflags, struct ucred *a_cred,
 *            struct thread *a_td)
 */
static int
union_mmap(struct vop_mmap_args *ap)
{
        struct vnode *ovp = OTHERVP(ap->a_vp);

        ap->a_head.a_ops = *ovp->v_ops;
        ap->a_vp = ovp;
        return (vop_mmap_ap(ap));
}

/*
 * union_fsync(struct vnode *a_vp, struct ucred *a_cred, int a_waitfor,
 *              struct thread *a_td)
 */
static int
union_fsync(struct vop_fsync_args *ap)
{
        int error = 0;
        struct thread *td = ap->a_td;
        struct vnode *targetvp;
        struct union_node *un = VTOUNION(ap->a_vp);

        if ((targetvp = union_lock_other(un, td)) != NULLVP) {
                error = VOP_FSYNC(targetvp, ap->a_waitfor, 0);
                union_unlock_other(targetvp, td);
        }

        return (error);
}

/*
 *      union_remove:
 *
 *      Remove the specified cnp.  The dvp and vp are passed to us locked
 *      and must remain locked on return.
 *
 * union_remove(struct vnode *a_dvp, struct vnode *a_vp,
 *              struct componentname *a_cnp)
 */
static int
union_remove(struct vop_old_remove_args *ap)
{
        struct union_node *dun = VTOUNION(ap->a_dvp);
        struct union_node *un = VTOUNION(ap->a_vp);
        struct componentname *cnp = ap->a_cnp;
        struct thread *td = cnp->cn_td;
        struct vnode *uppervp;
        struct vnode *upperdvp;
        int error;

        if ((upperdvp = union_lock_upper(dun, td)) == NULLVP)
                panic("union remove: null upper vnode");

        if ((uppervp = union_lock_upper(un, td)) != NULLVP) {
                if (union_dowhiteout(un, cnp->cn_cred, td))
                        cnp->cn_flags |= CNP_DOWHITEOUT;
                error = VOP_REMOVE(upperdvp, uppervp, cnp);
#if 0
                /* XXX */
                if (!error)
                        union_removed_upper(un);
#endif
                union_unlock_upper(uppervp, td);
        } else {
                error = union_mkwhiteout(
                            MOUNTTOUNIONMOUNT(ap->a_dvp->v_mount),
                            upperdvp, ap->a_cnp, un->un_path);
        }
        union_unlock_upper(upperdvp, td);
        return (error);
}

/*
 *      union_link:
 *
 *      tdvp will be locked on entry, vp will not be locked on entry.
 *      tdvp should remain locked on return and vp should remain unlocked
 *      on return.
 *
 * union_link(struct vnode *a_tdvp, struct vnode *a_vp,
 *            struct componentname *a_cnp)
 */
static int
union_link(struct vop_old_link_args *ap)
{
        struct componentname *cnp = ap->a_cnp;
        struct thread *td = cnp->cn_td;
        struct union_node *dun = VTOUNION(ap->a_tdvp);
        struct vnode *vp;
        struct vnode *tdvp;
        int error = 0;

        if (ap->a_tdvp->v_ops != ap->a_vp->v_ops) {
                vp = ap->a_vp;
        } else {
                struct union_node *tun = VTOUNION(ap->a_vp);

                if (tun->un_uppervp == NULLVP) {
                        vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
#if 0
                        if (dun->un_uppervp == tun->un_dirvp) {
                                if (dun->un_flags & UN_ULOCK) {
                                        dun->un_flags &= ~UN_ULOCK;
                                        vn_unlock(dun->un_uppervp);
                                }
                        }
#endif
                        error = union_copyup(tun, 1, cnp->cn_cred, td);
#if 0
                        if (dun->un_uppervp == tun->un_dirvp) {
                                vn_lock(dun->un_uppervp, 
                                        LK_EXCLUSIVE | LK_RETRY);
                                dun->un_flags |= UN_ULOCK;
                        }
#endif
                        vn_unlock(ap->a_vp);
                }
                vp = tun->un_uppervp;
        }

        if (error)
                return (error);

        /*
         * Make sure upper is locked, then unlock the union directory we were 
         * called with to avoid a deadlock while we are calling VOP_LINK on 
         * the upper (with tdvp locked and vp not locked).  Our ap->a_tdvp
         * is expected to be locked on return.
         */

        if ((tdvp = union_lock_upper(dun, td)) == NULLVP)
                return (EROFS);

        vn_unlock(ap->a_tdvp);  /* unlock calling node */
        error = VOP_LINK(tdvp, vp, cnp); /* call link on upper */

        /*
         * We have to unlock tdvp prior to relocking our calling node in
         * order to avoid a deadlock.
         */
        union_unlock_upper(tdvp, td);
        vn_lock(ap->a_tdvp, LK_EXCLUSIVE | LK_RETRY);
        return (error);
}

/*
 * union_rename(struct vnode *a_fdvp, struct vnode *a_fvp,
 *              struct componentname *a_fcnp, struct vnode *a_tdvp,
 *              struct vnode *a_tvp, struct componentname *a_tcnp)
 */
static int
union_rename(struct vop_old_rename_args *ap)
{
        int error;
        struct vnode *fdvp = ap->a_fdvp;
        struct vnode *fvp = ap->a_fvp;
        struct vnode *tdvp = ap->a_tdvp;
        struct vnode *tvp = ap->a_tvp;

        /*
         * Figure out what fdvp to pass to our upper or lower vnode.  If we
         * replace the fdvp, release the original one and ref the new one.
         */

        if (fdvp->v_tag == VT_UNION) {  /* always true */
                struct union_node *un = VTOUNION(fdvp);
                if (un->un_uppervp == NULLVP) {
                        /*
                         * this should never happen in normal
                         * operation but might if there was
                         * a problem creating the top-level shadow
                         * directory.
                         */
                        error = EXDEV;
                        goto bad;
                }
                fdvp = un->un_uppervp;
                vref(fdvp);
                vrele(ap->a_fdvp);
        }

        /*
         * Figure out what fvp to pass to our upper or lower vnode.  If we
         * replace the fvp, release the original one and ref the new one.
         */

        if (fvp->v_tag == VT_UNION) {   /* always true */
                struct union_node *un = VTOUNION(fvp);
#if 0
                struct union_mount *um = MOUNTTOUNIONMOUNT(fvp->v_mount);
#endif

                if (un->un_uppervp == NULLVP) {
                        switch(fvp->v_type) {
                        case VREG:
                                vn_lock(un->un_vnode, LK_EXCLUSIVE | LK_RETRY);
                                error = union_copyup(un, 1, ap->a_fcnp->cn_cred, ap->a_fcnp->cn_td);
                                vn_unlock(un->un_vnode);
                                if (error)
                                        goto bad;
                                break;
                        case VDIR:
                                /*
                                 * XXX not yet.
                                 *
                                 * There is only one way to rename a directory
                                 * based in the lowervp, and that is to copy
                                 * the entire directory hierarchy.  Otherwise
                                 * it would not last across a reboot.
                                 */
#if 0
                                vrele(fvp);
                                fvp = NULL;
                                vn_lock(fdvp, LK_EXCLUSIVE | LK_RETRY);
                                error = union_mkshadow(um, fdvp, 
                                            ap->a_fcnp, &un->un_uppervp);
                                vn_unlock(fdvp);
                                if (un->un_uppervp)
                                        vn_unlock(un->un_uppervp);
                                if (error)
                                        goto bad;
                                break;
#endif
                        default:
                                error = EXDEV;
                                goto bad;
                        }
                }

                if (un->un_lowervp != NULLVP)
                        ap->a_fcnp->cn_flags |= CNP_DOWHITEOUT;
                fvp = un->un_uppervp;
                vref(fvp);
                vrele(ap->a_fvp);
        }

        /*
         * Figure out what tdvp (destination directory) to pass to the
         * lower level.  If we replace it with uppervp, we need to vput the 
         * old one.  The exclusive lock is transfered to what we will pass
         * down in the VOP_RENAME and we replace uppervp with a simple
         * reference.
         */

        if (tdvp->v_tag == VT_UNION) {
                struct union_node *un = VTOUNION(tdvp);

                if (un->un_uppervp == NULLVP) {
                        /*
                         * this should never happen in normal
                         * operation but might if there was
                         * a problem creating the top-level shadow
                         * directory.
                         */
                        error = EXDEV;
                        goto bad;
                }

                /*
                 * new tdvp is a lock and reference on uppervp, put away
                 * the old tdvp.
                 */
                tdvp = union_lock_upper(un, ap->a_tcnp->cn_td);
                vput(ap->a_tdvp);
        }

        /*
         * Figure out what tvp (destination file) to pass to the
         * lower level.
         *
         * If the uppervp file does not exist put away the (wrong)
         * file and change tvp to NULL.
         */

        if (tvp != NULLVP && tvp->v_tag == VT_UNION) {
                struct union_node *un = VTOUNION(tvp);

                tvp = union_lock_upper(un, ap->a_tcnp->cn_td);
                vput(ap->a_tvp);
                /* note: tvp may be NULL */
        }

        /*
         * VOP_RENAME releases/vputs prior to returning, so we have no
         * cleanup to do.
         */

        return (VOP_RENAME(fdvp, fvp, ap->a_fcnp, tdvp, tvp, ap->a_tcnp));

        /*
         * Error.  We still have to release / vput the various elements.
         */

bad:
        vrele(fdvp);
        if (fvp)
                vrele(fvp);
        vput(tdvp);
        if (tvp != NULLVP) {
                if (tvp != tdvp)
                        vput(tvp);
                else
                        vrele(tvp);
        }
        return (error);
}

/*
 * union_mkdir(struct vnode *a_dvp, struct vnode **a_vpp,
 *              struct componentname *a_cnp, struct vattr *a_vap)
 */
static int
union_mkdir(struct vop_old_mkdir_args *ap)
{
        struct union_node *dun = VTOUNION(ap->a_dvp);
        struct componentname *cnp = ap->a_cnp;
        struct thread *td = cnp->cn_td;
        struct vnode *upperdvp;
        int error = EROFS;

        if ((upperdvp = union_lock_upper(dun, td)) != NULLVP) {
                struct vnode *vp;

                error = VOP_MKDIR(upperdvp, &vp, cnp, ap->a_vap);
                union_unlock_upper(upperdvp, td);

                if (error == 0) {
                        vn_unlock(vp);
                        UDEBUG(("ALLOCVP-2 FROM %p REFS %d\n", vp, vp->v_sysref.refcnt));
                        error = union_allocvp(ap->a_vpp, ap->a_dvp->v_mount,
                                ap->a_dvp, NULLVP, cnp, vp, NULLVP, 1);
                        UDEBUG(("ALLOCVP-2B FROM %p REFS %d\n", *ap->a_vpp, vp->v_sysref.refcnt));
                }
        }
        return (error);
}

/*
 * union_rmdir(struct vnode *a_dvp, struct vnode *a_vp,
 *              struct componentname *a_cnp)
 */
static int
union_rmdir(struct vop_old_rmdir_args *ap)
{
        struct union_node *dun = VTOUNION(ap->a_dvp);
        struct union_node *un = VTOUNION(ap->a_vp);
        struct componentname *cnp = ap->a_cnp;
        struct thread *td = cnp->cn_td;
        struct vnode *upperdvp;
        struct vnode *uppervp;
        int error;

        if ((upperdvp = union_lock_upper(dun, td)) == NULLVP)
                panic("union rmdir: null upper vnode");

        if ((uppervp = union_lock_upper(un, td)) != NULLVP) {
                if (union_dowhiteout(un, cnp->cn_cred, td))
                        cnp->cn_flags |= CNP_DOWHITEOUT;
                error = VOP_RMDIR(upperdvp, uppervp, ap->a_cnp);
                union_unlock_upper(uppervp, td);
        } else {
                error = union_mkwhiteout(
                            MOUNTTOUNIONMOUNT(ap->a_dvp->v_mount),
                            dun->un_uppervp, ap->a_cnp, un->un_path);
        }
        union_unlock_upper(upperdvp, td);
        return (error);
}

/*
 *      union_symlink:
 *
 *      dvp is locked on entry and remains locked on return.  a_vpp is garbage
 *      (unused).
 *
 * union_symlink(struct vnode *a_dvp, struct vnode **a_vpp,
 *              struct componentname *a_cnp, struct vattr *a_vap,
 *              char *a_target)
 */
static int
union_symlink(struct vop_old_symlink_args *ap)
{
        struct union_node *dun = VTOUNION(ap->a_dvp);
        struct componentname *cnp = ap->a_cnp;
        struct thread *td = cnp->cn_td;
        struct vnode *dvp;
        int error = EROFS;

        if ((dvp = union_lock_upper(dun, td)) != NULLVP) {
                error = VOP_SYMLINK(dvp, ap->a_vpp, cnp, ap->a_vap,
                            ap->a_target);
                union_unlock_upper(dvp, td);
        }
        return (error);
}

/*
 * union_readdir works in concert with getdirentries and
 * readdir(3) to provide a list of entries in the unioned
 * directories.  getdirentries is responsible for walking
 * down the union stack.  readdir(3) is responsible for
 * eliminating duplicate names from the returned data stream.
 *
 * union_readdir(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred,
 *               int *a_eofflag, off_t *a_cookies, int a_ncookies)
 */
static int
union_readdir(struct vop_readdir_args *ap)
{
        struct union_node *un = VTOUNION(ap->a_vp);
        struct thread *td = ap->a_uio->uio_td;
        struct vnode *uvp;
        int error = 0;

        if ((uvp = union_ref_upper(un)) != NULLVP) {
                ap->a_head.a_ops = *uvp->v_ops;
                ap->a_vp = uvp;
                error = vop_readdir_ap(ap);
                vrele(uvp);
        }
        return(error);
}

/*
 * union_readlink(struct vnode *a_vp, struct uio *a_uio, struct ucred *a_cred)
 */
static int
union_readlink(struct vop_readlink_args *ap)
{
        int error;
        struct union_node *un = VTOUNION(ap->a_vp);
        struct uio *uio = ap->a_uio;
        struct thread *td = uio->uio_td;
        struct vnode *vp;

        vp = union_lock_other(un, td);
        KASSERT(vp != NULL, ("union_readlink: backing vnode missing!"));

        ap->a_head.a_ops = *vp->v_ops;
        ap->a_vp = vp;
        error = vop_readlink_ap(ap);
        union_unlock_other(vp, td);

        return (error);
}

/*
 *      union_inactive:
 *
 *      Called with the vnode locked.  We are expected to unlock the vnode.
 *
 * union_inactive(struct vnode *a_vp, struct thread *a_td)
 */
static int
union_inactive(struct vop_inactive_args *ap)
{
        struct vnode *vp = ap->a_vp;
        /*struct thread *td = ap->a_td;*/
        struct union_node *un = VTOUNION(vp);
        struct vnode **vpp;

        /*
         * Do nothing (and _don't_ bypass).
         * Wait to vrele lowervp until reclaim,
         * so that until then our union_node is in the
         * cache and reusable.
         *
         * NEEDSWORK: Someday, consider inactive'ing
         * the lowervp and then trying to reactivate it
         * with capabilities (v_id)
         * like they do in the name lookup cache code.
         * That's too much work for now.
         */

        if (un->un_dircache != 0) {
                for (vpp = un->un_dircache; *vpp != NULLVP; vpp++)
                        vrele(*vpp);
                kfree (un->un_dircache, M_TEMP);
                un->un_dircache = 0;
        }

#if 0
        if ((un->un_flags & UN_ULOCK) && un->un_uppervp) {
                un->un_flags &= ~UN_ULOCK;
                vn_unlock(un->un_uppervp);
        }
#endif

        if ((un->un_flags & UN_CACHED) == 0)
                vgone_vxlocked(vp);

        return (0);
}

/*
 * union_reclaim(struct vnode *a_vp)
 */
static int
union_reclaim(struct vop_reclaim_args *ap)
{
        union_freevp(ap->a_vp);

        return (0);
}

/*
 *      union_bmap:
 *
 *      There isn't much we can do.  We cannot push through to the real vnode
 *      to get to the underlying device because this will bypass data
 *      cached by the real vnode.
 *
 *      For some reason we cannot return the 'real' vnode either, it seems
 *      to blow up memory maps.
 *
 * union_bmap(struct vnode *a_vp, off_t a_loffset,
 *            off_t *a_doffsetp, int *a_runp, int *a_runb)
 */
static int
union_bmap(struct vop_bmap_args *ap)
{
        return(EOPNOTSUPP);
}

/*
 * union_print(struct vnode *a_vp)
 */
static int
union_print(struct vop_print_args *ap)
{
        struct vnode *vp = ap->a_vp;

        kprintf("\ttag VT_UNION, vp=%p, uppervp=%p, lowervp=%p\n",
                        vp, UPPERVP(vp), LOWERVP(vp));
        if (UPPERVP(vp) != NULLVP)
                vprint("union: upper", UPPERVP(vp));
        if (LOWERVP(vp) != NULLVP)
                vprint("union: lower", LOWERVP(vp));

        return (0);
}

/*
 * union_pathconf(struct vnode *a_vp, int a_name, int *a_retval)
 */
static int
union_pathconf(struct vop_pathconf_args *ap)
{
        int error;
        struct thread *td = curthread;          /* XXX */
        struct union_node *un = VTOUNION(ap->a_vp);
        struct vnode *vp;

        vp = union_lock_other(un, td);
        KASSERT(vp != NULL, ("union_pathconf: backing vnode missing!"));

        ap->a_head.a_ops = *vp->v_ops;
        ap->a_vp = vp;
        error = vop_pathconf_ap(ap);
        union_unlock_other(vp, td);

        return (error);
}

/*
 * union_advlock(struct vnode *a_vp, caddr_t a_id, int a_op,
 *               struct flock *a_fl, int a_flags)
 */
static int
union_advlock(struct vop_advlock_args *ap)
{
        struct vnode *ovp = OTHERVP(ap->a_vp);

        ap->a_head.a_ops = *ovp->v_ops;
        ap->a_vp = ovp;
        return (vop_advlock_ap(ap));
}


/*
 * XXX - vop_strategy must be hand coded because it has no
 * YYY - and it is not coherent with anything
 *
 * vnode in its arguments.
 * This goes away with a merged VM/buffer cache.
 *
 * union_strategy(struct vnode *a_vp, struct bio *a_bio)
 */
static int
union_strategy(struct vop_strategy_args *ap)
{
        struct bio *bio = ap->a_bio;
        struct buf *bp = bio->bio_buf;
        struct vnode *othervp = OTHERVP(ap->a_vp);

#ifdef DIAGNOSTIC
        if (othervp == NULLVP)
                panic("union_strategy: nil vp");
        if (bp->b_cmd != BUF_CMD_READ && (othervp == LOWERVP(ap->a_vp)))
                panic("union_strategy: writing to lowervp");
#endif
        return (vn_strategy(othervp, bio));
}

/*
 * Global vfs data structures
 */
struct vop_ops union_vnode_vops = {
        .vop_default =          vop_defaultop,
        .vop_access =           union_access,
        .vop_advlock =          union_advlock,
        .vop_bmap =             union_bmap,
        .vop_close =            union_close,
        .vop_old_create =       union_create,
        .vop_fsync =            union_fsync,
        .vop_getpages =         union_getpages,
        .vop_putpages =         union_putpages,
        .vop_getattr =          union_getattr,
        .vop_inactive =         union_inactive,
        .vop_ioctl =            union_ioctl,
        .vop_old_link =         union_link,
        .vop_old_lookup =       union_lookup,
        .vop_old_mkdir =        union_mkdir,
        .vop_old_mknod =        union_mknod,
        .vop_mmap =             union_mmap,
        .vop_open =             union_open,
        .vop_pathconf =         union_pathconf,
        .vop_poll =             union_poll,
        .vop_print =            union_print,
        .vop_read =             union_read,
        .vop_readdir =          union_readdir,
        .vop_readlink =         union_readlink,
        .vop_reclaim =          union_reclaim,
        .vop_old_remove =       union_remove,
        .vop_old_rename =       union_rename,
        .vop_old_rmdir =        union_rmdir,
        .vop_setattr =          union_setattr,
        .vop_strategy =         union_strategy,
        .vop_old_symlink =      union_symlink,
        .vop_old_whiteout =     union_whiteout,
        .vop_write =            union_write
};