hammer2 - Refactor frontend part 5/many

[dragonfly.git] / sys / vfs / hammer2 / hammer2_inode.c

/*
 * Copyright (c) 2011-2014 The DragonFly Project.  All rights reserved.
 *
 * This code is derived from software contributed to The DragonFly Project
 * by Matthew Dillon <dillon@dragonflybsd.org>
 * by Venkatesh Srinivas <vsrinivas@dragonflybsd.org>
 *
 * 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.
 */
#include <sys/cdefs.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/types.h>
#include <sys/lock.h>
#include <sys/uuid.h>

#include "hammer2.h"

#define INODE_DEBUG     0

static void hammer2_inode_move_to_hidden(hammer2_trans_t *trans,
                                         hammer2_cluster_t **cparentp,
                                         hammer2_cluster_t **clusterp,
                                         hammer2_tid_t inum);

RB_GENERATE2(hammer2_inode_tree, hammer2_inode, rbnode, hammer2_inode_cmp,
             hammer2_tid_t, meta.inum);

int
hammer2_inode_cmp(hammer2_inode_t *ip1, hammer2_inode_t *ip2)
{
        if (ip1->meta.inum < ip2->meta.inum)
                return(-1);
        if (ip1->meta.inum > ip2->meta.inum)
                return(1);
        return(0);
}

/*
 * HAMMER2 inode locks
 *
 * HAMMER2 offers shared and exclusive locks on inodes.  Pass a mask of
 * flags for options:
 *
 *      - pass HAMMER2_RESOLVE_SHARED if a shared lock is desired.  The
 *        inode locking function will automatically set the RDONLY flag.
 *
 *      - pass HAMMER2_RESOLVE_ALWAYS if you need the inode's meta-data.
 *        Most front-end inode locks do.
 *
 *      - pass HAMMER2_RESOLVE_NEVER if you do not want to require that
 *        the inode data be resolved.  This is used by the syncthr because
 *        it can run on an unresolved/out-of-sync cluster, and also by the
 *        vnode reclamation code to avoid unnecessary I/O (particularly when
 *        disposing of hundreds of thousands of cached vnodes).
 *
 * The inode locking function locks the inode itself, resolves any stale
 * chains in the inode's cluster, and allocates a fresh copy of the
 * cluster with 1 ref and all the underlying chains locked.
 *
 * ip->cluster will be stable while the inode is locked.
 *
 * NOTE: We don't combine the inode/chain lock because putting away an
 *       inode would otherwise confuse multiple lock holders of the inode.
 *
 * NOTE: In-memory inodes always point to hardlink targets (the actual file),
 *       and never point to a hardlink pointer.
 *
 * NOTE: If caller passes HAMMER2_RESOLVE_RDONLY the exclusive locking code
 *       will feel free to reduce the chain set in the cluster as an
 *       optimization.  It will still be validated against the quorum if
 *       appropriate, but the optimization might be able to reduce data
 *       accesses to one node.  This flag is automatically set if the inode
 *       is locked with HAMMER2_RESOLVE_SHARED.
 */
void
hammer2_inode_lock(hammer2_inode_t *ip, int how)
{
        hammer2_inode_ref(ip);

        /* 
         * Inode structure mutex
         */
        if (how & HAMMER2_RESOLVE_SHARED) {
                how |= HAMMER2_RESOLVE_RDONLY;
                hammer2_mtx_sh(&ip->lock);
        } else {
                hammer2_mtx_ex(&ip->lock);
        }
}

/*
 * Create a locked copy of ip->cluster.  Note that the copy will have a
 * ref on the cluster AND its chains and we don't want a second ref to
 * either when we lock it.
 *
 * Exclusive inode locks set the template focus chain in (ip)
 * as a hint.  Cluster locks can ALWAYS replace the focus in the
 * working copy if the hint does not work out, so beware.
 */
hammer2_cluster_t *
hammer2_inode_cluster(hammer2_inode_t *ip, int how)
{
        hammer2_cluster_t *cluster;

        cluster = hammer2_cluster_copy(&ip->cluster);
        hammer2_cluster_lock(cluster, how);
        hammer2_cluster_resolve(cluster);

        /*
         * cluster->focus will be set if resolving RESOLVE_ALWAYS, but
         * only update the cached focus in the inode structure when taking
         * out an exclusive lock.
         */
        if ((how & HAMMER2_RESOLVE_SHARED) == 0)
                ip->cluster.focus = cluster->focus;

        return cluster;
}

void
hammer2_inode_unlock(hammer2_inode_t *ip, hammer2_cluster_t *cluster)
{
        if (cluster) {
                hammer2_cluster_unlock(cluster);
                hammer2_cluster_drop(cluster);
        }
        hammer2_mtx_unlock(&ip->lock);
        hammer2_inode_drop(ip);
}

/*
 * Temporarily release a lock held shared or exclusive.  Caller must
 * hold the lock shared or exclusive on call and lock will be released
 * on return.
 *
 * Restore a lock that was temporarily released.
 */
hammer2_mtx_state_t
hammer2_inode_lock_temp_release(hammer2_inode_t *ip)
{
        return hammer2_mtx_temp_release(&ip->lock);
}

void
hammer2_inode_lock_temp_restore(hammer2_inode_t *ip, hammer2_mtx_state_t ostate)
{
        hammer2_mtx_temp_restore(&ip->lock, ostate);
}

/*
 * Upgrade a shared inode lock to exclusive and return.  If the inode lock
 * is already held exclusively this is a NOP.
 *
 * The caller MUST hold the inode lock either shared or exclusive on call
 * and will own the lock exclusively on return.
 *
 * Returns non-zero if the lock was already exclusive prior to the upgrade.
 */
int
hammer2_inode_lock_upgrade(hammer2_inode_t *ip)
{
        int wasexclusive;

        if (mtx_islocked_ex(&ip->lock)) {
                wasexclusive = 1;
        } else {
                hammer2_mtx_unlock(&ip->lock);
                hammer2_mtx_ex(&ip->lock);
                wasexclusive = 0;
        }
        return wasexclusive;
}

/*
 * Downgrade an inode lock from exclusive to shared only if the inode
 * lock was previously shared.  If the inode lock was previously exclusive,
 * this is a NOP.
 */
void
hammer2_inode_lock_downgrade(hammer2_inode_t *ip, int wasexclusive)
{
        if (wasexclusive == 0)
                mtx_downgrade(&ip->lock);
}

/*
 * Lookup an inode by inode number
 */
hammer2_inode_t *
hammer2_inode_lookup(hammer2_pfs_t *pmp, hammer2_tid_t inum)
{
        hammer2_inode_t *ip;

        KKASSERT(pmp);
        if (pmp->spmp_hmp) {
                ip = NULL;
        } else {
                hammer2_spin_ex(&pmp->inum_spin);
                ip = RB_LOOKUP(hammer2_inode_tree, &pmp->inum_tree, inum);
                if (ip)
                        hammer2_inode_ref(ip);
                hammer2_spin_unex(&pmp->inum_spin);
        }
        return(ip);
}

/*
 * Adding a ref to an inode is only legal if the inode already has at least
 * one ref.
 *
 * (can be called with spinlock held)
 */
void
hammer2_inode_ref(hammer2_inode_t *ip)
{
        atomic_add_int(&ip->refs, 1);
}

/*
 * Drop an inode reference, freeing the inode when the last reference goes
 * away.
 */
void
hammer2_inode_drop(hammer2_inode_t *ip)
{
        hammer2_pfs_t *pmp;
        hammer2_inode_t *pip;
        u_int refs;

        while (ip) {
                refs = ip->refs;
                cpu_ccfence();
                if (refs == 1) {
                        /*
                         * Transition to zero, must interlock with
                         * the inode inumber lookup tree (if applicable).
                         * It should not be possible for anyone to race
                         * the transition to 0.
                         *
                         */
                        pmp = ip->pmp;
                        KKASSERT(pmp);
                        hammer2_spin_ex(&pmp->inum_spin);

                        if (atomic_cmpset_int(&ip->refs, 1, 0)) {
                                KKASSERT(hammer2_mtx_refs(&ip->lock) == 0);
                                if (ip->flags & HAMMER2_INODE_ONRBTREE) {
                                        atomic_clear_int(&ip->flags,
                                                     HAMMER2_INODE_ONRBTREE);
                                        RB_REMOVE(hammer2_inode_tree,
                                                  &pmp->inum_tree, ip);
                                }
                                hammer2_spin_unex(&pmp->inum_spin);

                                pip = ip->pip;
                                ip->pip = NULL;
                                ip->pmp = NULL;

                                /*
                                 * Cleaning out ip->cluster isn't entirely
                                 * trivial.
                                 */
                                hammer2_inode_repoint(ip, NULL, NULL);

                                /*
                                 * We have to drop pip (if non-NULL) to
                                 * dispose of our implied reference from
                                 * ip->pip.  We can simply loop on it.
                                 */
                                kfree(ip, pmp->minode);
                                atomic_add_long(&pmp->inmem_inodes, -1);
                                ip = pip;
                                /* continue with pip (can be NULL) */
                        } else {
                                hammer2_spin_unex(&ip->pmp->inum_spin);
                        }
                } else {
                        /*
                         * Non zero transition
                         */
                        if (atomic_cmpset_int(&ip->refs, refs, refs - 1))
                                break;
                }
        }
}

/*
 * Get the vnode associated with the given inode, allocating the vnode if
 * necessary.  The vnode will be returned exclusively locked.
 *
 * The caller must lock the inode (shared or exclusive).
 *
 * Great care must be taken to avoid deadlocks and vnode acquisition/reclaim
 * races.
 */
struct vnode *
hammer2_igetv(hammer2_inode_t *ip, hammer2_cluster_t *cparent, int *errorp)
{
        const hammer2_inode_data_t *ripdata;
        hammer2_pfs_t *pmp;
        struct vnode *vp;

        pmp = ip->pmp;
        KKASSERT(pmp != NULL);
        *errorp = 0;

        ripdata = &hammer2_cluster_rdata(cparent)->ipdata;

        for (;;) {
                /*
                 * Attempt to reuse an existing vnode assignment.  It is
                 * possible to race a reclaim so the vget() may fail.  The
                 * inode must be unlocked during the vget() to avoid a
                 * deadlock against a reclaim.
                 */
                int wasexclusive;

                vp = ip->vp;
                if (vp) {
                        /*
                         * Inode must be unlocked during the vget() to avoid
                         * possible deadlocks, but leave the ip ref intact.
                         *
                         * vnode is held to prevent destruction during the
                         * vget().  The vget() can still fail if we lost
                         * a reclaim race on the vnode.
                         */
                        hammer2_mtx_state_t ostate;

                        vhold(vp);
                        ostate = hammer2_inode_lock_temp_release(ip);
                        if (vget(vp, LK_EXCLUSIVE)) {
                                vdrop(vp);
                                hammer2_inode_lock_temp_restore(ip, ostate);
                                continue;
                        }
                        hammer2_inode_lock_temp_restore(ip, ostate);
                        vdrop(vp);
                        /* vp still locked and ref from vget */
                        if (ip->vp != vp) {
                                kprintf("hammer2: igetv race %p/%p\n",
                                        ip->vp, vp);
                                vput(vp);
                                continue;
                        }
                        *errorp = 0;
                        break;
                }

                /*
                 * No vnode exists, allocate a new vnode.  Beware of
                 * allocation races.  This function will return an
                 * exclusively locked and referenced vnode.
                 */
                *errorp = getnewvnode(VT_HAMMER2, pmp->mp, &vp, 0, 0);
                if (*errorp) {
                        kprintf("hammer2: igetv getnewvnode failed %d\n",
                                *errorp);
                        vp = NULL;
                        break;
                }

                /*
                 * Lock the inode and check for an allocation race.
                 */
                wasexclusive = hammer2_inode_lock_upgrade(ip);
                if (ip->vp != NULL) {
                        vp->v_type = VBAD;
                        vx_put(vp);
                        hammer2_inode_lock_downgrade(ip, wasexclusive);
                        continue;
                }

                switch (ripdata->meta.type) {
                case HAMMER2_OBJTYPE_DIRECTORY:
                        vp->v_type = VDIR;
                        break;
                case HAMMER2_OBJTYPE_REGFILE:
                        vp->v_type = VREG;
                        vinitvmio(vp, ripdata->meta.size,
                                  HAMMER2_LBUFSIZE,
                                  (int)ripdata->meta.size & HAMMER2_LBUFMASK);
                        break;
                case HAMMER2_OBJTYPE_SOFTLINK:
                        /*
                         * XXX for now we are using the generic file_read
                         * and file_write code so we need a buffer cache
                         * association.
                         */
                        vp->v_type = VLNK;
                        vinitvmio(vp, ripdata->meta.size,
                                  HAMMER2_LBUFSIZE,
                                  (int)ripdata->meta.size & HAMMER2_LBUFMASK);
                        break;
                case HAMMER2_OBJTYPE_CDEV:
                        vp->v_type = VCHR;
                        /* fall through */
                case HAMMER2_OBJTYPE_BDEV:
                        vp->v_ops = &pmp->mp->mnt_vn_spec_ops;
                        if (ripdata->meta.type != HAMMER2_OBJTYPE_CDEV)
                                vp->v_type = VBLK;
                        addaliasu(vp,
                                  ripdata->meta.rmajor,
                                  ripdata->meta.rminor);
                        break;
                case HAMMER2_OBJTYPE_FIFO:
                        vp->v_type = VFIFO;
                        vp->v_ops = &pmp->mp->mnt_vn_fifo_ops;
                        break;
                default:
                        panic("hammer2: unhandled objtype %d",
                              ripdata->meta.type);
                        break;
                }

                if (ip == pmp->iroot)
                        vsetflags(vp, VROOT);

                vp->v_data = ip;
                ip->vp = vp;
                hammer2_inode_ref(ip);          /* vp association */
                hammer2_inode_lock_downgrade(ip, wasexclusive);
                break;
        }

        /*
         * Return non-NULL vp and *errorp == 0, or NULL vp and *errorp != 0.
         */
        if (hammer2_debug & 0x0002) {
                kprintf("igetv vp %p refs 0x%08x aux 0x%08x\n",
                        vp, vp->v_refcnt, vp->v_auxrefs);
        }
        return (vp);
}

/*
 * Returns the inode associated with the passed-in cluster, creating the
 * inode if necessary and synchronizing it to the passed-in cluster otherwise.
 *
 * The passed-in cluster must be locked and will remain locked on return.
 * The returned inode will be locked and the caller may dispose of both
 * via hammer2_inode_unlock_ex().  However, if the caller needs to resolve
 * a hardlink it must ref/unlock/relock/drop the inode.
 *
 * The hammer2_inode structure regulates the interface between the high level
 * kernel VNOPS API and the filesystem backend (the chains).
 *
 * On return the inode is locked with the supplied cluster.
 */
hammer2_inode_t *
hammer2_inode_get(hammer2_pfs_t *pmp, hammer2_inode_t *dip,
                  hammer2_cluster_t *cluster)
{
        hammer2_inode_t *nip;
        const hammer2_inode_data_t *iptmp;
        const hammer2_inode_data_t *nipdata;

        KKASSERT(cluster == NULL ||
                 hammer2_cluster_type(cluster) == HAMMER2_BREF_TYPE_INODE);
        KKASSERT(pmp);

        /*
         * Interlocked lookup/ref of the inode.  This code is only needed
         * when looking up inodes with nlinks != 0 (TODO: optimize out
         * otherwise and test for duplicates).
         *
         * Cluster can be NULL during the initial pfs allocation.
         */
again:
        while (cluster) {
                iptmp = &hammer2_cluster_rdata(cluster)->ipdata;
                nip = hammer2_inode_lookup(pmp, iptmp->meta.inum);
                if (nip == NULL)
                        break;

                hammer2_mtx_ex(&nip->lock);

                /*
                 * Handle SMP race (not applicable to the super-root spmp
                 * which can't index inodes due to duplicative inode numbers).
                 */
                if (pmp->spmp_hmp == NULL &&
                    (nip->flags & HAMMER2_INODE_ONRBTREE) == 0) {
                        hammer2_mtx_unlock(&nip->lock);
                        hammer2_inode_drop(nip);
                        continue;
                }
                hammer2_inode_repoint(nip, NULL, cluster);

                return nip;
        }

        /*
         * We couldn't find the inode number, create a new inode.
         */
        nip = kmalloc(sizeof(*nip), pmp->minode, M_WAITOK | M_ZERO);
        spin_init(&nip->cluster_spin, "h2clspin");
        atomic_add_long(&pmp->inmem_inodes, 1);
        hammer2_pfs_memory_inc(pmp);
        hammer2_pfs_memory_wakeup(pmp);
        if (pmp->spmp_hmp)
                nip->flags = HAMMER2_INODE_SROOT;

        /*
         * Initialize nip's cluster.  A cluster is provided for normal
         * inodes but typically not for the super-root or PFS inodes.
         */
        nip->cluster.refs = 1;
        nip->cluster.pmp = pmp;
        nip->cluster.flags |= HAMMER2_CLUSTER_INODE;
        if (cluster) {
                nipdata = &hammer2_cluster_rdata(cluster)->ipdata;
                nip->meta = nipdata->meta;
                hammer2_cluster_bref(cluster, &nip->bref);
                atomic_set_int(&nip->flags, HAMMER2_INODE_METAGOOD);
                hammer2_inode_repoint(nip, NULL, cluster);
        } else {
                nip->meta.inum = 1;             /* PFS inum is always 1 XXX */
                /* mtime will be updated when a cluster is available */
                atomic_set_int(&nip->flags, HAMMER2_INODE_METAGOOD);/*XXX*/
        }

        nip->pip = dip;                         /* can be NULL */
        if (dip)
                hammer2_inode_ref(dip); /* ref dip for nip->pip */

        nip->pmp = pmp;

        /*
         * ref and lock on nip gives it state compatible to after a
         * hammer2_inode_lock() call.
         */
        nip->refs = 1;
        hammer2_mtx_init(&nip->lock, "h2inode");
        hammer2_mtx_ex(&nip->lock);
        /* combination of thread lock and chain lock == inode lock */

        /*
         * Attempt to add the inode.  If it fails we raced another inode
         * get.  Undo all the work and try again.
         */
        if (pmp->spmp_hmp == NULL) {
                hammer2_spin_ex(&pmp->inum_spin);
                if (RB_INSERT(hammer2_inode_tree, &pmp->inum_tree, nip)) {
                        hammer2_spin_unex(&pmp->inum_spin);
                        hammer2_mtx_unlock(&nip->lock);
                        hammer2_inode_drop(nip);
                        goto again;
                }
                atomic_set_int(&nip->flags, HAMMER2_INODE_ONRBTREE);
                hammer2_spin_unex(&pmp->inum_spin);
        }

        return (nip);
}

/*
 * Create a new inode in the specified directory using the vattr to
 * figure out the type of inode.
 *
 * If no error occurs the new inode with its cluster locked is returned in
 * *nipp, otherwise an error is returned and *nipp is set to NULL.
 *
 * If vap and/or cred are NULL the related fields are not set and the
 * inode type defaults to a directory.  This is used when creating PFSs
 * under the super-root, so the inode number is set to 1 in this case.
 *
 * dip is not locked on entry.
 *
 * NOTE: When used to create a snapshot, the inode is temporarily associated
 *       with the super-root spmp. XXX should pass new pmp for snapshot.
 */
hammer2_inode_t *
hammer2_inode_create(hammer2_trans_t *trans, hammer2_inode_t *dip,
                     struct vattr *vap, struct ucred *cred,
                     const uint8_t *name, size_t name_len,
                     hammer2_cluster_t **clusterp,
                     int flags, int *errorp)
{
        const hammer2_inode_data_t *dipdata;
        hammer2_inode_data_t *nipdata;
        hammer2_cluster_t *cluster;
        hammer2_cluster_t *cparent;
        hammer2_inode_t *nip;
        hammer2_key_t key_dummy;
        hammer2_key_t lhc;
        int error;
        uid_t xuid;
        uuid_t dip_uid;
        uuid_t dip_gid;
        uint32_t dip_mode;
        uint8_t dip_comp_algo;
        uint8_t dip_check_algo;

        lhc = hammer2_dirhash(name, name_len);
        *errorp = 0;

        /*
         * Locate the inode or indirect block to create the new
         * entry in.  At the same time check for key collisions
         * and iterate until we don't get one.
         *
         * NOTE: hidden inodes do not have iterators.
         */
retry:
        hammer2_inode_lock(dip, HAMMER2_RESOLVE_ALWAYS);
        cparent = hammer2_inode_cluster(dip, HAMMER2_RESOLVE_ALWAYS);
        dipdata = &hammer2_cluster_rdata(cparent)->ipdata;
        dip_uid = dipdata->meta.uid;
        dip_gid = dipdata->meta.gid;
        dip_mode = dipdata->meta.mode;
        dip_comp_algo = dipdata->meta.comp_algo;
        dip_check_algo = dipdata->meta.check_algo;

        error = 0;
        while (error == 0) {
                cluster = hammer2_cluster_lookup(cparent, &key_dummy,
                                                 lhc, lhc, 0);
                if (cluster == NULL)
                        break;
                if ((lhc & HAMMER2_DIRHASH_VISIBLE) == 0)
                        error = ENOSPC;
                if ((lhc & HAMMER2_DIRHASH_LOMASK) == HAMMER2_DIRHASH_LOMASK)
                        error = ENOSPC;
                hammer2_cluster_unlock(cluster);
                hammer2_cluster_drop(cluster);
                cluster = NULL;
                ++lhc;
        }

        if (error == 0) {
                error = hammer2_cluster_create(trans, cparent, &cluster,
                                             lhc, 0,
                                             HAMMER2_BREF_TYPE_INODE,
                                             HAMMER2_INODE_BYTES,
                                             flags);
        }
#if INODE_DEBUG
        kprintf("CREATE INODE %*.*s chain=%p\n",
                (int)name_len, (int)name_len, name,
                (cluster ? cluster->focus : NULL));
#endif

        /*
         * Cleanup and handle retries.
         */
        if (error == EAGAIN) {
                hammer2_cluster_ref(cparent);
                hammer2_inode_unlock(dip, cparent);
                hammer2_cluster_wait(cparent);
                hammer2_cluster_drop(cparent);
                goto retry;
        }
        hammer2_inode_unlock(dip, cparent);
        cparent = NULL;

        if (error) {
                KKASSERT(cluster == NULL);
                *errorp = error;
                return (NULL);
        }

        /*
         * Set up the new inode.
         *
         * NOTE: *_get() integrates chain's lock into the inode lock.
         *
         * NOTE: Only one new inode can currently be created per
         *       transaction.  If the need arises we can adjust
         *       hammer2_trans_init() to allow more.
         *
         * NOTE: nipdata will have chain's blockset data.
         */
        KKASSERT(cluster->focus->flags & HAMMER2_CHAIN_MODIFIED);
        nipdata = &hammer2_cluster_wdata(cluster)->ipdata;
        nipdata->meta.inum = trans->inode_tid;
        hammer2_cluster_modsync(cluster);
        nip = hammer2_inode_get(dip->pmp, dip, cluster);
        nipdata = &hammer2_cluster_wdata(cluster)->ipdata;

        if (vap) {
                KKASSERT(trans->inodes_created == 0);
                nipdata->meta.type = hammer2_get_obj_type(vap->va_type);
                nipdata->meta.inum = trans->inode_tid;
                ++trans->inodes_created;

                switch (nipdata->meta.type) {
                case HAMMER2_OBJTYPE_CDEV:
                case HAMMER2_OBJTYPE_BDEV:
                        nipdata->meta.rmajor = vap->va_rmajor;
                        nipdata->meta.rminor = vap->va_rminor;
                        break;
                default:
                        break;
                }
        } else {
                nipdata->meta.type = HAMMER2_OBJTYPE_DIRECTORY;
                nipdata->meta.inum = 1;
        }
        
        /* Inherit parent's inode compression mode. */
        nip->comp_heuristic = 0;
        nipdata->meta.comp_algo = dip_comp_algo;
        nipdata->meta.check_algo = dip_check_algo;
        nipdata->meta.version = HAMMER2_INODE_VERSION_ONE;
        hammer2_update_time(&nipdata->meta.ctime);
        nipdata->meta.mtime = nipdata->meta.ctime;
        if (vap)
                nipdata->meta.mode = vap->va_mode;
        nipdata->meta.nlinks = 1;
        if (vap) {
                if (dip && dip->pmp) {
                        xuid = hammer2_to_unix_xid(&dip_uid);
                        xuid = vop_helper_create_uid(dip->pmp->mp,
                                                     dip_mode,
                                                     xuid,
                                                     cred,
                                                     &vap->va_mode);
                } else {
                        /* super-root has no dip and/or pmp */
                        xuid = 0;
                }
                if (vap->va_vaflags & VA_UID_UUID_VALID)
                        nipdata->meta.uid = vap->va_uid_uuid;
                else if (vap->va_uid != (uid_t)VNOVAL)
                        hammer2_guid_to_uuid(&nipdata->meta.uid, vap->va_uid);
                else
                        hammer2_guid_to_uuid(&nipdata->meta.uid, xuid);

                if (vap->va_vaflags & VA_GID_UUID_VALID)
                        nipdata->meta.gid = vap->va_gid_uuid;
                else if (vap->va_gid != (gid_t)VNOVAL)
                        hammer2_guid_to_uuid(&nipdata->meta.gid, vap->va_gid);
                else if (dip)
                        nipdata->meta.gid = dip_gid;
        }

        /*
         * Regular files and softlinks allow a small amount of data to be
         * directly embedded in the inode.  This flag will be cleared if
         * the size is extended past the embedded limit.
         */
        if (nipdata->meta.type == HAMMER2_OBJTYPE_REGFILE ||
            nipdata->meta.type == HAMMER2_OBJTYPE_SOFTLINK) {
                nipdata->meta.op_flags |= HAMMER2_OPFLAG_DIRECTDATA;
        }

        KKASSERT(name_len < HAMMER2_INODE_MAXNAME);
        bcopy(name, nipdata->filename, name_len);
        nipdata->meta.name_key = lhc;
        nipdata->meta.name_len = name_len;
        nip->meta = nipdata->meta;
        hammer2_cluster_modsync(cluster);
        *clusterp = cluster;

        return (nip);
}

/*
 * The cluster has been removed from the original directory and replaced
 * with a hardlink pointer.  Move the cluster to the specified parent
 * directory, change the filename to "0xINODENUMBER", and adjust the key.
 * The cluster becomes our invisible hardlink target.
 *
 * The original cluster must be deleted on entry.
 */
static
void
hammer2_hardlink_shiftup(hammer2_trans_t *trans, hammer2_cluster_t *cluster,
                        hammer2_inode_t *ip, hammer2_inode_t *dip,
                        hammer2_cluster_t *dcluster,
                        int nlinks, int *errorp)
{
        const hammer2_inode_data_t *iptmp;
        hammer2_inode_data_t *nipdata;
        hammer2_cluster_t *xcluster;
        hammer2_key_t key_dummy;
        hammer2_key_t lhc;
        hammer2_blockref_t bref;

        iptmp = &hammer2_cluster_rdata(cluster)->ipdata;
        lhc = iptmp->meta.inum;
        KKASSERT((lhc & HAMMER2_DIRHASH_VISIBLE) == 0);

        /*
         * Locate the inode or indirect block to create the new
         * entry in.  lhc represents the inode number so there is
         * no collision iteration.
         *
         * There should be no key collisions with invisible inode keys.
         *
         * WARNING! Must use inode_lock_ex() on dip to handle a stale
         *          dip->cluster cache.
         */
        *errorp = 0;
        xcluster = hammer2_cluster_lookup(dcluster, &key_dummy,
                                      lhc, lhc, 0);
        if (xcluster) {
                kprintf("X3 chain %p dip %p dchain %p dip->chain %p\n",
                        xcluster->focus, dip, dcluster->focus,
                        dip->cluster.focus);
                hammer2_cluster_unlock(xcluster);
                hammer2_cluster_drop(xcluster);
                xcluster = NULL;
                *errorp = ENOSPC;
#if 0
                Debugger("X3");
#endif
        }

        /*
         * Handle the error case
         */
        if (*errorp) {
                panic("error2");
                KKASSERT(xcluster == NULL);
                return;
        }

        /*
         * Use xcluster as a placeholder for (lhc).  Duplicate cluster to the
         * same target bref as xcluster and then delete xcluster.  The
         * duplication occurs after xcluster in flush order even though
         * xcluster is deleted after the duplication. XXX
         *
         * WARNING! Duplications (to a different parent) can cause indirect
         *          blocks to be inserted, refactor xcluster.
         *
         * WARNING! Only key and keybits is extracted from a passed-in bref.
         */
        hammer2_cluster_bref(cluster, &bref);
        bref.key = lhc;                 /* invisible dir entry key */
        bref.keybits = 0;
        hammer2_cluster_rename(trans, &bref, dcluster, cluster, 0);

        /*
         * cluster is now 'live' again.. adjust the filename.
         *
         * Directory entries are inodes but this is a hidden hardlink
         * target.  The name isn't used but to ease debugging give it
         * a name after its inode number.
         */
        hammer2_cluster_modify(trans, cluster, 0);
        nipdata = &hammer2_cluster_wdata(cluster)->ipdata;
        ksnprintf(nipdata->filename, sizeof(nipdata->filename),
                  "0x%016jx", (intmax_t)nipdata->meta.inum);
        nipdata->meta.name_len = strlen(nipdata->filename);
        nipdata->meta.name_key = lhc;
        nipdata->meta.nlinks += nlinks;

        /*
         * Resync ip->meta.  Some fields have to be retained.
         */
        nipdata->meta.size = ip->meta.size;
        nipdata->meta.mtime = ip->meta.mtime;
        ip->meta = nipdata->meta;

        hammer2_cluster_modsync(cluster);
}

/*
 * Connect the target inode represented by (cluster) to the media topology
 * at (dip, name, len).  The caller can pass a rough *chainp, this function
 * will issue lookup()s to position the parent chain properly for the
 * chain insertion.
 *
 * If hlink is TRUE this function creates an OBJTYPE_HARDLINK directory
 * entry instead of connecting (cluster).
 *
 * If hlink is FALSE this function expects (cluster) to be unparented.
 */
int
hammer2_inode_connect(hammer2_trans_t *trans,
                      hammer2_inode_t *ip, hammer2_cluster_t **clusterp,
                      int hlink,
                      hammer2_inode_t *dip, hammer2_cluster_t *dcluster,
                      const uint8_t *name, size_t name_len,
                      hammer2_key_t lhc)
{
        hammer2_inode_data_t *wipdata;
        hammer2_cluster_t *ocluster;
        hammer2_cluster_t *ncluster;
        hammer2_key_t key_dummy;
        int error;

        /*
         * Since ocluster is either disconnected from the topology or
         * represents a hardlink terminus which is always a parent of or
         * equal to dip, we should be able to safely lock dip->chain for
         * our setup.
         *
         * WARNING! Must use inode_lock_ex() on dip to handle a stale
         *          dip->cluster.
         *
         * If name is non-NULL we calculate lhc, else we use the passed-in
         * lhc.
         */
        ocluster = *clusterp;

        if (name) {
                lhc = hammer2_dirhash(name, name_len);

                /*
                 * Locate the inode or indirect block to create the new
                 * entry in.  At the same time check for key collisions
                 * and iterate until we don't get one.
                 */
                error = 0;
                while (error == 0) {
                        ncluster = hammer2_cluster_lookup(dcluster, &key_dummy,
                                                      lhc, lhc, 0);
                        if (ncluster == NULL)
                                break;
                        if ((lhc & HAMMER2_DIRHASH_LOMASK) ==
                            HAMMER2_DIRHASH_LOMASK) {
                                error = ENOSPC;
                        }
                        hammer2_cluster_unlock(ncluster);
                        hammer2_cluster_drop(ncluster);
                        ncluster = NULL;
                        ++lhc;
                }
        } else {
                /*
                 * Reconnect to specific key (used when moving
                 * unlinked-but-open files into the hidden directory).
                 */
                ncluster = hammer2_cluster_lookup(dcluster, &key_dummy,
                                                  lhc, lhc, 0);
                KKASSERT(ncluster == NULL);
                error = 0;
        }

        if (error == 0) {
                if (hlink) {
                        /*
                         * Hardlink pointer needed, create totally fresh
                         * directory entry.
                         *
                         * We must refactor ocluster because it might have
                         * been shifted into an indirect cluster by the
                         * create.
                         */
                        KKASSERT(ncluster == NULL);
                        error = hammer2_cluster_create(trans,
                                                       dcluster, &ncluster,
                                                       lhc, 0,
                                                       HAMMER2_BREF_TYPE_INODE,
                                                       HAMMER2_INODE_BYTES,
                                                       0);
                } else {
                        /*
                         * Reconnect the original cluster under the new name.
                         * Original cluster must have already been deleted by
                         * teh caller.
                         *
                         * WARNING! Can cause held-over clusters to require a
                         *          refactor.  Fortunately we have none (our
                         *          locked clusters are passed into and
                         *          modified by the call).
                         */
                        ncluster = ocluster;
                        ocluster = NULL;
                        error = hammer2_cluster_create(trans,
                                                       dcluster, &ncluster,
                                                       lhc, 0,
                                                       HAMMER2_BREF_TYPE_INODE,
                                                       HAMMER2_INODE_BYTES,
                                                       0);
                }
        }

        /*
         * Unlock stuff.
         */
        KKASSERT(error != EAGAIN);

        /*
         * ncluster should be NULL on error, leave ocluster
         * (ocluster == *clusterp) alone.
         */
        if (error) {
                KKASSERT(ncluster == NULL);
                return (error);
        }

        /*
         * Directory entries are inodes so if the name has changed we have
         * to update the inode.
         *
         * When creating an OBJTYPE_HARDLINK entry remember to unlock the
         * cluster, the caller will access the hardlink via the actual hardlink
         * target file and not the hardlink pointer entry, so we must still
         * return ocluster.
         */
        if (hlink && hammer2_hardlink_enable >= 0) {
                /*
                 * Create the HARDLINK pointer.  oip represents the hardlink
                 * target in this situation.
                 *
                 * We will return ocluster (the hardlink target).
                 */
                hammer2_cluster_modify(trans, ncluster, 0);
                KKASSERT(name_len < HAMMER2_INODE_MAXNAME);
                wipdata = &hammer2_cluster_wdata(ncluster)->ipdata;
                bcopy(name, wipdata->filename, name_len);
                wipdata->meta.name_key = lhc;
                wipdata->meta.name_len = name_len;
                wipdata->meta.target_type =
                            hammer2_cluster_rdata(ocluster)->ipdata.meta.type;
                wipdata->meta.type = HAMMER2_OBJTYPE_HARDLINK;
                wipdata->meta.inum =
                            hammer2_cluster_rdata(ocluster)->ipdata.meta.inum;
                wipdata->meta.version = HAMMER2_INODE_VERSION_ONE;
                wipdata->meta.nlinks = 1;
                wipdata->meta.op_flags = HAMMER2_OPFLAG_DIRECTDATA;
                hammer2_cluster_modsync(ncluster);
                hammer2_cluster_unlock(ncluster);
                hammer2_cluster_drop(ncluster);
                ncluster = ocluster;
                ocluster = NULL;
        } else {
                /*
                 * ncluster is a duplicate of ocluster at the new location.
                 * We must fixup the name stored in the inode data.
                 * The bref key has already been adjusted by inode_connect().
                 */
                hammer2_cluster_modify(trans, ncluster, 0);
                wipdata = &hammer2_cluster_wdata(ncluster)->ipdata;

                KKASSERT(name_len < HAMMER2_INODE_MAXNAME);
                bcopy(name, wipdata->filename, name_len);
                wipdata->meta.name_key = lhc;
                wipdata->meta.name_len = name_len;
                wipdata->meta.nlinks = 1;
                hammer2_cluster_modsync(ncluster);

                /*
                 * Resync the in-memory inode, some fields must be retained.
                 */
                if (ip) {       /* XXX move_to_hidden passes NULL */
                        wipdata->meta.size = ip->meta.size;
                        wipdata->meta.mtime = ip->meta.mtime;
                        ip->meta = wipdata->meta;
                }
        }

        /*
         * We are replacing ocluster with ncluster, unlock ocluster.  In the
         * case where ocluster is left unchanged the code above sets
         * ncluster to ocluster and ocluster to NULL, resulting in a NOP here.
         */
        if (ocluster) {
                hammer2_cluster_unlock(ocluster);
                hammer2_cluster_drop(ocluster);
        }
        *clusterp = ncluster;

        return (0);
}

/*
 * Repoint ip->cluster's chains to cluster's chains and fixup the default
 * focus.  Only valid elements are repointed.  Invalid elements have to be
 * adjusted by the appropriate slave sync threads.
 *
 * Caller must hold the inode and cluster exclusive locked, if not NULL,
 * must also be locked.
 *
 * Cluster may be NULL to clean out any chains in ip->cluster.
 */
void
hammer2_inode_repoint(hammer2_inode_t *ip, hammer2_inode_t *pip,
                      hammer2_cluster_t *cluster)
{
        hammer2_chain_t *dropch[HAMMER2_MAXCLUSTER];
        hammer2_chain_t *ochain;
        hammer2_chain_t *nchain;
        hammer2_inode_t *opip;
        int i;

        bzero(dropch, sizeof(dropch));

        /*
         * Replace chains in ip->cluster with chains from cluster and
         * adjust the focus if necessary.
         *
         * NOTE: nchain and/or ochain can be NULL due to gaps
         *       in the cluster arrays.
         */
        hammer2_spin_ex(&ip->cluster_spin);
        for (i = 0; cluster && i < cluster->nchains; ++i) {
                /*
                 * Do not replace invalid elements as this might race
                 * syncthr replacements.
                 */
                if (cluster->array[i].flags & HAMMER2_CITEM_INVALID)
                        continue;

                /*
                 * Do not replace elements which are the same.  Also handle
                 * element count discrepancies.
                 */
                nchain = cluster->array[i].chain;
                if (i < ip->cluster.nchains) {
                        ochain = ip->cluster.array[i].chain;
                        if (ochain == nchain)
                                continue;
                } else {
                        ochain = NULL;
                }

                /*
                 * Make adjustments
                 */
                ip->cluster.array[i].chain = nchain;
                ip->cluster.array[i].flags &= ~HAMMER2_CITEM_INVALID;
                ip->cluster.array[i].flags |= cluster->array[i].flags &
                                              HAMMER2_CITEM_INVALID;
                if (nchain)
                        hammer2_chain_ref(nchain);
                dropch[i] = ochain;
        }

        /*
         * Release any left-over chains in ip->cluster.
         */
        while (i < ip->cluster.nchains) {
                nchain = ip->cluster.array[i].chain;
                if (nchain) {
                        ip->cluster.array[i].chain = NULL;
                        ip->cluster.array[i].flags |= HAMMER2_CITEM_INVALID;
                }
                dropch[i] = nchain;
                ++i;
        }

        /*
         * Fixup fields.  Note that the inode-embedded cluster is never
         * directly locked.
         */
        if (cluster) {
                ip->cluster.nchains = cluster->nchains;
                ip->cluster.focus = cluster->focus;
                ip->cluster.flags = cluster->flags & ~HAMMER2_CLUSTER_LOCKED;
        } else {
                ip->cluster.nchains = 0;
                ip->cluster.focus = NULL;
                ip->cluster.flags &= ~HAMMER2_CLUSTER_ZFLAGS;
        }

        /*
         * Repoint ip->pip if requested (non-NULL pip).
         */
        if (pip && ip->pip != pip) {
                opip = ip->pip;
                hammer2_inode_ref(pip);
                ip->pip = pip;
        } else {
                opip = NULL;
        }
        hammer2_spin_unex(&ip->cluster_spin);

        /*
         * Cleanup outside of spinlock
         */
        while (--i >= 0) {
                if (dropch[i])
                        hammer2_chain_drop(dropch[i]);
        }
        if (opip)
                hammer2_inode_drop(opip);
}

/*
 * Repoint a single element from the cluster to the ip.  Used by the
 * synchronization threads to piecemeal update inodes.  Does not change
 * focus and requires inode to be re-locked to clean-up flags (XXX).
 */
void
hammer2_inode_repoint_one(hammer2_inode_t *ip, hammer2_cluster_t *cluster,
                          int idx)
{
        hammer2_chain_t *ochain;
        hammer2_chain_t *nchain;
        int i;

        hammer2_spin_ex(&ip->cluster_spin);
        KKASSERT(idx < cluster->nchains);
        if (idx < ip->cluster.nchains) {
                ochain = ip->cluster.array[idx].chain;
                nchain = cluster->array[idx].chain;
        } else {
                ochain = NULL;
                nchain = cluster->array[idx].chain;
                ip->cluster.nchains = idx + 1;
                for (i = ip->cluster.nchains; i <= idx; ++i) {
                        bzero(&ip->cluster.array[i],
                              sizeof(ip->cluster.array[i]));
                        ip->cluster.array[i].flags |= HAMMER2_CITEM_INVALID;
                }
        }
        if (ochain != nchain) {
                /*
                 * Make adjustments.
                 */
                ip->cluster.array[idx].chain = nchain;
                ip->cluster.array[idx].flags &= ~HAMMER2_CITEM_INVALID;
                ip->cluster.array[idx].flags |= cluster->array[idx].flags &
                                                HAMMER2_CITEM_INVALID;
        }
        hammer2_spin_unex(&ip->cluster_spin);
        if (ochain != nchain) {
                if (nchain)
                        hammer2_chain_ref(nchain);
                if (ochain)
                        hammer2_chain_drop(ochain);
        }
}

/*
 * Unlink the file from the specified directory inode.  The directory inode
 * does not need to be locked.
 *
 * isdir determines whether a directory/non-directory check should be made.
 * No check is made if isdir is set to -1.
 *
 * isopen specifies whether special unlink-with-open-descriptor handling
 * must be performed.  If set to -1 the caller is deleting a PFS and we
 * check whether the chain is mounted or not (chain->pmp != NULL).  1 is
 * implied if it is mounted.
 *
 * If isopen is 1 and nlinks drops to 0 this function must move the chain
 * to a special hidden directory until last-close occurs on the file.
 *
 * NOTE!  The underlying file can still be active with open descriptors
 *        or if the chain is being manually held (e.g. for rename).
 *
 *        The caller is responsible for fixing up ip->chain if e.g. a
 *        rename occurs (see chain_duplicate()).
 *
 * NOTE!  The chain is not deleted if it is moved to the hidden directory,
 *        but otherwise will be deleted.
 */
int
hammer2_unlink_file(hammer2_trans_t *trans, hammer2_inode_t *dip,
                    const uint8_t *name, size_t name_len,
                    int isdir, int *hlinkp, struct nchandle *nch,
                    int nlinks)
{
        const hammer2_inode_data_t *ripdata;
        hammer2_inode_data_t *wipdata;
        hammer2_cluster_t *cparent;
        hammer2_cluster_t *hcluster;
        hammer2_cluster_t *hparent;
        hammer2_cluster_t *cluster;
        hammer2_cluster_t *dparent;
        hammer2_cluster_t *dcluster;
        hammer2_key_t key_dummy;
        hammer2_key_t key_next;
        hammer2_key_t lhc;
        int last_link;
        int error;
        int hlink;
        uint8_t type;

        error = 0;
        hlink = 0;
        hcluster = NULL;
        hparent = NULL;
        lhc = hammer2_dirhash(name, name_len);

again:
        /*
         * Search for the filename in the directory
         */
        hammer2_inode_lock(dip, HAMMER2_RESOLVE_ALWAYS);
        cparent = hammer2_inode_cluster(dip, HAMMER2_RESOLVE_ALWAYS);
        cluster = hammer2_cluster_lookup(cparent, &key_next,
                                     lhc, lhc + HAMMER2_DIRHASH_LOMASK, 0);
        while (cluster) {
                if (hammer2_cluster_type(cluster) == HAMMER2_BREF_TYPE_INODE) {
                        ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
                        if (ripdata->meta.name_len == name_len &&
                            bcmp(ripdata->filename, name, name_len) == 0) {
                                break;
                        }
                }
                cluster = hammer2_cluster_next(cparent, cluster, &key_next,
                                               key_next,
                                               lhc + HAMMER2_DIRHASH_LOMASK,
                                               0);
        }
        hammer2_inode_unlock(dip, NULL);        /* retain cparent */

        /*
         * Not found or wrong type (isdir < 0 disables the type check).
         * If a hardlink pointer, type checks use the hardlink target.
         */
        if (cluster == NULL) {
                error = ENOENT;
                goto done;
        }
        ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
        type = ripdata->meta.type;
        if (type == HAMMER2_OBJTYPE_HARDLINK) {
                hlink = 1;
                type = ripdata->meta.target_type;
        }

        if (type == HAMMER2_OBJTYPE_DIRECTORY && isdir == 0) {
                error = ENOTDIR;
                goto done;
        }
        if (type != HAMMER2_OBJTYPE_DIRECTORY && isdir >= 1) {
                error = EISDIR;
                goto done;
        }

        /*
         * Hardlink must be resolved.  We can't hold the parent locked
         * while we do this or we could deadlock.  The physical file will
         * be located at or above the current directory.
         *
         * We loop to reacquire the hardlink origination.
         *
         * NOTE: hammer2_hardlink_find() will locate the hardlink target,
         *       returning a modified hparent and hcluster.
         */
        if (ripdata->meta.type == HAMMER2_OBJTYPE_HARDLINK) {
                if (hcluster == NULL) {
                        hcluster = cluster;
                        cluster = NULL; /* safety */
                        hammer2_cluster_unlock(cparent);
                        hammer2_cluster_drop(cparent);
                        cparent = NULL; /* safety */
                        ripdata = NULL; /* safety (associated w/cparent) */
                        error = hammer2_hardlink_find(dip, &hparent, &hcluster);

                        /*
                         * If we couldn't find the hardlink target then some
                         * parent directory containing the hardlink pointer
                         * probably got renamed to above the original target,
                         * a case not yet handled by H2.
                         */
                        if (error) {
                                kprintf("H2 unlink_file: hardlink target for "
                                        "\"%s\" not found\n",
                                        name);
                                kprintf("(likely due to known directory "
                                        "rename bug)\n");
                                goto done;
                        }
                        goto again;
                }
        }

        /*
         * If this is a directory the directory must be empty.  However, if
         * isdir < 0 we are doing a rename and the directory does not have
         * to be empty, and if isdir > 1 we are deleting a PFS/snapshot
         * and the directory does not have to be empty.
         *
         * NOTE: We check the full key range here which covers both visible
         *       and invisible entries.  Theoretically there should be no
         *       invisible (hardlink target) entries if there are no visible
         *       entries.
         */
        if (type == HAMMER2_OBJTYPE_DIRECTORY && isdir == 1) {
                dparent = hammer2_cluster_lookup_init(cluster, 0);
                dcluster = hammer2_cluster_lookup(dparent, &key_dummy,
                                                  0, (hammer2_key_t)-1,
                                                  HAMMER2_LOOKUP_NODATA);
                if (dcluster) {
                        hammer2_cluster_unlock(dcluster);
                        hammer2_cluster_drop(dcluster);
                        hammer2_cluster_lookup_done(dparent);
                        error = ENOTEMPTY;
                        goto done;
                }
                hammer2_cluster_lookup_done(dparent);
                dparent = NULL;
                /* dcluster NULL */
        }

        /*
         * If this was a hardlink then (cparent, cluster) is the hardlink
         * pointer, which we can simply destroy outright.  Discard the
         * clusters and replace with the hardlink target.
         */
        if (hcluster) {
                hammer2_cluster_delete(trans, cparent, cluster,
                                       HAMMER2_DELETE_PERMANENT);
                hammer2_cluster_unlock(cparent);
                hammer2_cluster_drop(cparent);
                hammer2_cluster_unlock(cluster);
                hammer2_cluster_drop(cluster);
                cparent = hparent;
                cluster = hcluster;
                hparent = NULL;
                hcluster = NULL;
        }

        /*
         * This leaves us with the hardlink target or non-hardlinked file
         * or directory in (cparent, cluster).
         *
         * Delete the target when nlinks reaches 0 with special handling
         * to avoid I/O (to avoid actually updating the inode) for the 1->0
         * transition, if possible.  This optimization makes rm -rf very
         * fast.
         *
         * NOTE! In DragonFly the vnops function calls cache_unlink() after
         *       calling us here to clean out the namecache association,
         *       (which does not represent a ref for the open-test), and to
         *       force finalization of the vnode if/when the last ref gets
         *       dropped.
         *
         * NOTE! Files are unlinked by rename and then relinked.  nch will be
         *       passed as NULL in this situation.  hammer2_inode_connect()
         *       will bump nlinks.
         */
        KKASSERT(cluster != NULL);

        /*
         * Note: nlinks is negative when decrementing, positive when
         *       incrementing.
         */
        ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
        last_link = (ripdata->meta.nlinks + nlinks == 0);

        if (last_link) {
                /*
                 * Target nlinks has reached 0, file now unlinked (but may
                 * still be open).
                 *
                 * nlinks will be -1 for a normal remove().  If this is the
                 * last link we must flag the inode on deactivation. XXX race ?
                 */
                hammer2_inode_t *ip;

                if (nlinks == -1) {
                        ip = hammer2_inode_lookup(trans->pmp,
                                                  ripdata->meta.inum);
                        if (ip) {
                                atomic_set_int(&ip->flags,
                                               HAMMER2_INODE_ISUNLINKED);
                                hammer2_inode_drop(ip);
                        }
                }

                if (nch && cache_isopen(nch)) {
                        /*
                         * If an unlinked file is still open we must update
                         * the inodes link count.
                         */
                        hammer2_cluster_modify(trans, cluster, 0);
                        wipdata = &hammer2_cluster_wdata(cluster)->ipdata;
                        ripdata = wipdata;
                        wipdata->meta.nlinks += nlinks;
                        /* XXX race */
                        /* XXX debugging */
                        if ((int64_t)wipdata->meta.nlinks < 0) {
                                wipdata->meta.nlinks = 0;
                        }
                        hammer2_cluster_modsync(cluster);
                        hammer2_inode_move_to_hidden(trans, &cparent, &cluster,
                                                     wipdata->meta.inum);
                } else {
                        /*
                         * This won't get everything if a vnode is still
                         * present, but the cache_unlink() call the caller
                         * makes will.
                         */
                        hammer2_cluster_delete(trans, cparent, cluster,
                                               HAMMER2_DELETE_PERMANENT);
                }
        } else if (hlink == 0) {
                /*
                 * In this situation a normal non-hardlinked file (which can
                 * only have nlinks == 1) still has a non-zero nlinks, the
                 * caller must be doing a RENAME operation and so is passing
                 * a nlinks adjustment of 0, and only wishes to remove file
                 * in order to be able to reconnect it under a different name.
                 *
                 * In this situation we do a temporary deletion of the
                 * chain in order to allow the file to be reconnected in
                 * a different location.
                 */
                KKASSERT(nlinks == 0);
                hammer2_cluster_delete(trans, cparent, cluster, 0);
        } else {
                /*
                 * Links remain, must update the inode link count.
                 */
                hammer2_cluster_modify(trans, cluster, 0);
                wipdata = &hammer2_cluster_wdata(cluster)->ipdata;
                ripdata = wipdata;
                wipdata->meta.nlinks += nlinks;
                /* XXX debugging */
                if ((int64_t)wipdata->meta.nlinks < 0) {
                        wipdata->meta.nlinks = 0;
                }
                hammer2_cluster_modsync(cluster);
        }

        error = 0;
done:
        if (cparent) {
                hammer2_cluster_unlock(cparent);
                hammer2_cluster_drop(cparent);
        }
        if (cluster) {
                hammer2_cluster_unlock(cluster);
                hammer2_cluster_drop(cluster);
        }
        if (hparent) {
                hammer2_cluster_unlock(hparent);
                hammer2_cluster_drop(hparent);
        }
        if (hcluster) {
                hammer2_cluster_unlock(hcluster);
                hammer2_cluster_drop(hcluster);
        }
        if (hlinkp)
                *hlinkp = hlink;

        return error;
}

/*
 * This is called from the mount code to initialize pmp->ihidden
 */
void
hammer2_inode_install_hidden(hammer2_pfs_t *pmp)
{
        hammer2_trans_t trans;
        hammer2_cluster_t *cparent;
        hammer2_cluster_t *cluster;
        hammer2_cluster_t *scan;
        const hammer2_inode_data_t *ripdata;
        hammer2_inode_data_t *wipdata;
        hammer2_key_t key_dummy;
        hammer2_key_t key_next;
        int error;
        int count;
        int dip_check_algo;
        int dip_comp_algo;

        if (pmp->ihidden)
                return;

        /*
         * Find the hidden directory
         */
        bzero(&key_dummy, sizeof(key_dummy));
        hammer2_trans_init(&trans, pmp, 0);

        /*
         * Setup for lookup, retrieve iroot's check and compression
         * algorithm request which was likely generated by newfs_hammer2.
         *
         * The check/comp fields will probably never be used since inodes
         * are renamed into the hidden directory and not created relative to
         * the hidden directory, chain creation inherits from bref.methods,
         * and data chains inherit from their respective file inode *_algo
         * fields.
         */
        hammer2_inode_lock(pmp->iroot, HAMMER2_RESOLVE_ALWAYS);
        cparent = hammer2_inode_cluster(pmp->iroot, HAMMER2_RESOLVE_ALWAYS);
        ripdata = &hammer2_cluster_rdata(cparent)->ipdata;
        dip_check_algo = ripdata->meta.check_algo;
        dip_comp_algo = ripdata->meta.comp_algo;
        ripdata = NULL;

        cluster = hammer2_cluster_lookup(cparent, &key_dummy,
                                         HAMMER2_INODE_HIDDENDIR,
                                         HAMMER2_INODE_HIDDENDIR,
                                         0);
        if (cluster) {
                pmp->ihidden = hammer2_inode_get(pmp, pmp->iroot, cluster);
                hammer2_inode_ref(pmp->ihidden);

                /*
                 * Remove any unlinked files which were left open as-of
                 * any system crash.
                 *
                 * Don't pass NODATA, we need the inode data so the delete
                 * can do proper statistics updates.
                 */
                count = 0;
                scan = hammer2_cluster_lookup(cluster, &key_next,
                                              0, HAMMER2_TID_MAX, 0);
                while (scan) {
                        if (hammer2_cluster_type(scan) ==
                            HAMMER2_BREF_TYPE_INODE) {
                                hammer2_cluster_delete(&trans, cluster, scan,
                                                   HAMMER2_DELETE_PERMANENT);
                                ++count;
                        }
                        scan = hammer2_cluster_next(cluster, scan, &key_next,
                                                    0, HAMMER2_TID_MAX, 0);
                }

                hammer2_inode_unlock(pmp->ihidden, cluster);
                hammer2_inode_unlock(pmp->iroot, cparent);
                hammer2_trans_done(&trans);
                kprintf("hammer2: PFS loaded hidden dir, "
                        "removed %d dead entries\n", count);
                return;
        }

        /*
         * Create the hidden directory
         */
        error = hammer2_cluster_create(&trans, cparent, &cluster,
                                       HAMMER2_INODE_HIDDENDIR, 0,
                                       HAMMER2_BREF_TYPE_INODE,
                                       HAMMER2_INODE_BYTES,
                                       0);
        hammer2_inode_unlock(pmp->iroot, cparent);

        hammer2_cluster_modify(&trans, cluster, 0);
        wipdata = &hammer2_cluster_wdata(cluster)->ipdata;
        wipdata->meta.type = HAMMER2_OBJTYPE_DIRECTORY;
        wipdata->meta.inum = HAMMER2_INODE_HIDDENDIR;
        wipdata->meta.nlinks = 1;
        wipdata->meta.comp_algo = dip_comp_algo;
        wipdata->meta.check_algo = dip_check_algo;
        hammer2_cluster_modsync(cluster);
        kprintf("hammer2: PFS root missing hidden directory, creating\n");

        pmp->ihidden = hammer2_inode_get(pmp, pmp->iroot, cluster);
        hammer2_inode_ref(pmp->ihidden);
        hammer2_inode_unlock(pmp->ihidden, cluster);
        hammer2_trans_done(&trans);
}

/*
 * If an open file is unlinked H2 needs to retain the file in the topology
 * to ensure that its backing store is not recovered by the bulk free scan.
 * This also allows us to avoid having to special-case the CHAIN_DELETED flag.
 *
 * To do this the file is moved to a hidden directory in the PFS root and
 * renamed.  The hidden directory must be created if it does not exist.
 */
static
void
hammer2_inode_move_to_hidden(hammer2_trans_t *trans,
                             hammer2_cluster_t **cparentp,
                             hammer2_cluster_t **clusterp,
                             hammer2_tid_t inum)
{
        hammer2_cluster_t *dcluster;
        hammer2_pfs_t *pmp;
        int error;

        pmp = (*clusterp)->pmp;
        KKASSERT(pmp != NULL);
        KKASSERT(pmp->ihidden != NULL);

        hammer2_cluster_delete(trans, *cparentp, *clusterp, 0);
        hammer2_inode_lock(pmp->ihidden, HAMMER2_RESOLVE_ALWAYS);
        dcluster = hammer2_inode_cluster(pmp->ihidden, HAMMER2_RESOLVE_ALWAYS);
        error = hammer2_inode_connect(trans,
                                      NULL/*XXX*/, clusterp, 0,
                                      pmp->ihidden, dcluster,
                                      NULL, 0, inum);
        hammer2_inode_unlock(pmp->ihidden, dcluster);
        KKASSERT(error == 0);
}

/*
 * Given an exclusively locked inode and cluster we consolidate the cluster
 * for hardlink creation, adding (nlinks) to the file's link count and
 * potentially relocating the inode to (cdip) which is a parent directory
 * common to both the current location of the inode and the intended new
 * hardlink.
 *
 * Replaces (*clusterp) if consolidation occurred, unlocking the old cluster
 * and returning a new locked cluster.
 *
 * NOTE!  This function will also replace ip->cluster.
 */
int
hammer2_hardlink_consolidate(hammer2_trans_t *trans,
                             hammer2_inode_t *ip,
                             hammer2_cluster_t **clusterp,
                             hammer2_inode_t *cdip,
                             hammer2_cluster_t *cdcluster,
                             int nlinks)
{
        const hammer2_inode_data_t *ripdata;
        hammer2_inode_data_t *wipdata;
        hammer2_cluster_t *cluster;
        hammer2_cluster_t *cparent;
        int error;

        cluster = *clusterp;
        ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
        if (nlinks == 0 &&                      /* no hardlink needed */
            (ripdata->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
                return (0);
        }

        if (hammer2_hardlink_enable == 0) {     /* disallow hardlinks */
                hammer2_cluster_unlock(cluster);
                hammer2_cluster_drop(cluster);
                *clusterp = NULL;
                return (ENOTSUP);
        }

        cparent = NULL;

        /*
         * If no change in the hardlink's target directory is required and
         * this is already a hardlink target, all we need to do is adjust
         * the link count.
         */
        ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
        if (cdip == ip->pip &&
            (ripdata->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0) {
                if (nlinks) {
                        hammer2_cluster_modify(trans, cluster, 0);
                        wipdata = &hammer2_cluster_wdata(cluster)->ipdata;
                        wipdata->meta.nlinks += nlinks;
                        hammer2_cluster_modsync(cluster);
                        ripdata = wipdata;
                }
                error = 0;
                goto done;
        }

        /*
         * Cluster is the real inode.  The originating directory is locked
         * by the caller so we can manipulate it without worrying about races
         * against other lookups.
         *
         * If cluster is visible we need to delete it from the current
         * location and create a hardlink pointer in its place.  If it is
         * not visible we need only delete it.  Then later cluster will be
         * renamed to a parent directory and converted (if necessary) to
         * a hidden inode (via shiftup).
         *
         * NOTE! We must hold cparent locked through the delete/create/rename
         *       operation to ensure that other threads block resolving to
         *       the same hardlink, otherwise the other threads may not see
         *       the hardlink.
         */
        KKASSERT((cluster->focus->flags & HAMMER2_CHAIN_DELETED) == 0);
        cparent = hammer2_cluster_parent(cluster);

        hammer2_cluster_delete(trans, cparent, cluster, 0);

        ripdata = &hammer2_cluster_rdata(cluster)->ipdata;
        KKASSERT(ripdata->meta.type != HAMMER2_OBJTYPE_HARDLINK);
        if (ripdata->meta.name_key & HAMMER2_DIRHASH_VISIBLE) {
                hammer2_cluster_t *ncluster;
                hammer2_key_t lhc;

                ncluster = NULL;
                lhc = cluster->focus->bref.key;
                error = hammer2_cluster_create(trans, cparent, &ncluster,
                                             lhc, 0,
                                             HAMMER2_BREF_TYPE_INODE,
                                             HAMMER2_INODE_BYTES,
                                             0);
                hammer2_cluster_modify(trans, ncluster, 0);
                wipdata = &hammer2_cluster_wdata(ncluster)->ipdata;

                /* wipdata->comp_algo = ripdata->comp_algo; */
                wipdata->meta.comp_algo = 0;
                wipdata->meta.check_algo = 0;
                wipdata->meta.version = HAMMER2_INODE_VERSION_ONE;
                wipdata->meta.inum = ripdata->meta.inum;
                wipdata->meta.target_type = ripdata->meta.type;
                wipdata->meta.type = HAMMER2_OBJTYPE_HARDLINK;
                wipdata->meta.uflags = 0;
                wipdata->meta.rmajor = 0;
                wipdata->meta.rminor = 0;
                wipdata->meta.ctime = 0;
                wipdata->meta.mtime = 0;
                wipdata->meta.atime = 0;
                wipdata->meta.btime = 0;
                bzero(&wipdata->meta.uid, sizeof(wipdata->meta.uid));
                bzero(&wipdata->meta.gid, sizeof(wipdata->meta.gid));
                wipdata->meta.op_flags = HAMMER2_OPFLAG_DIRECTDATA;
                wipdata->meta.cap_flags = 0;
                wipdata->meta.mode = 0;
                wipdata->meta.size = 0;
                wipdata->meta.nlinks = 1;
                wipdata->meta.iparent = 0;      /* XXX */
                wipdata->meta.pfs_type = 0;
                wipdata->meta.pfs_inum = 0;
                bzero(&wipdata->meta.pfs_clid, sizeof(wipdata->meta.pfs_clid));
                bzero(&wipdata->meta.pfs_fsid, sizeof(wipdata->meta.pfs_fsid));
                wipdata->meta.data_quota = 0;
                /* wipdata->data_count = 0; */
                wipdata->meta.inode_quota = 0;
                /* wipdata->inode_count = 0; */
                wipdata->meta.attr_tid = 0;
                wipdata->meta.dirent_tid = 0;
                bzero(&wipdata->u, sizeof(wipdata->u));
                bcopy(ripdata->filename, wipdata->filename,
                      ripdata->meta.name_len);
                wipdata->meta.name_key = ncluster->focus->bref.key;
                wipdata->meta.name_len = ripdata->meta.name_len;
                /* XXX transaction ids */
                hammer2_cluster_modsync(ncluster);
                hammer2_cluster_unlock(ncluster);
                hammer2_cluster_drop(ncluster);
        }
        ripdata = wipdata;

        /*
         * cluster represents the hardlink target and is now flagged deleted.
         * duplicate it to the parent directory and adjust nlinks.
         *
         * WARNING! The shiftup() call can cause ncluster to be moved into
         *          an indirect block, and our ncluster will wind up pointing
         *          to the older/original version.
         */
        KKASSERT(cluster->focus->flags & HAMMER2_CHAIN_DELETED);
        hammer2_hardlink_shiftup(trans, cluster, ip, cdip, cdcluster,
                                 nlinks, &error);

        if (error == 0)
                hammer2_inode_repoint(ip, cdip, cluster);

done:
        /*
         * Cleanup, cluster/ncluster already dealt with.
         *
         * Return the shifted cluster in *clusterp.
         */
        if (cparent) {
                hammer2_cluster_unlock(cparent);
                hammer2_cluster_drop(cparent);
        }
        *clusterp = cluster;

        return (error);
}

/*
 * If (*ochainp) is non-NULL it points to the forward OBJTYPE_HARDLINK
 * inode while (*chainp) points to the resolved (hidden hardlink
 * target) inode.  In this situation when nlinks is 1 we wish to
 * deconsolidate the hardlink, moving it back to the directory that now
 * represents the only remaining link.
 */
int
hammer2_hardlink_deconsolidate(hammer2_trans_t *trans,
                               hammer2_inode_t *dip,
                               hammer2_chain_t **chainp,
                               hammer2_chain_t **ochainp)
{
        if (*ochainp == NULL)
                return (0);
        /* XXX */
        return (0);
}

/*
 * The caller presents a locked cluster with an obj_type of
 * HAMMER2_OBJTYPE_HARDLINK in (*clusterp).  This routine will locate
 * the inode and replace (*clusterp) with a new locked cluster containing
 * the target hardlink, also locked.  The original cluster will be
 * unlocked and released.
 *
 * If cparentp is not NULL a locked cluster representing the hardlink's
 * parent is also returned.
 *
 * If we are unable to locate the hardlink target EIO is returned,
 * (*cparentp) is set to NULL, the original passed-in (*clusterp)
 * will be unlocked and released and (*clusterp) will be set to NULL
 * as well.
 */
int
hammer2_hardlink_find(hammer2_inode_t *dip,
                      hammer2_cluster_t **cparentp,
                      hammer2_cluster_t **clusterp)
{
        const hammer2_inode_data_t *ipdata;
        hammer2_cluster_t *cluster;
        hammer2_cluster_t *cparent;
        hammer2_cluster_t *rcluster;
        hammer2_inode_t *ip;
        hammer2_inode_t *pip;
        hammer2_key_t key_dummy;
        hammer2_key_t lhc;

        cluster = *clusterp;
        pip = dip;
        hammer2_inode_ref(pip);         /* for loop */

        /*
         * Locate the hardlink.  pip is referenced and not locked.
         * Unlock and release (*clusterp) after extracting the needed
         * data.
         */
        ipdata = &hammer2_cluster_rdata(cluster)->ipdata;
        lhc = ipdata->meta.inum;
        ipdata = NULL;                  /* safety */
        hammer2_cluster_unlock(cluster);
        hammer2_cluster_drop(cluster);
        *clusterp = NULL;               /* safety */

        rcluster = NULL;
        cparent = NULL;

        while ((ip = pip) != NULL) {
                hammer2_inode_lock(ip, HAMMER2_RESOLVE_ALWAYS);
                cparent = hammer2_inode_cluster(ip, HAMMER2_RESOLVE_ALWAYS);
                hammer2_inode_drop(ip);                 /* loop */
                KKASSERT(hammer2_cluster_type(cparent) ==
                         HAMMER2_BREF_TYPE_INODE);
                rcluster = hammer2_cluster_lookup(cparent, &key_dummy,
                                             lhc, lhc, 0);
                if (rcluster)
                        break;
                hammer2_cluster_lookup_done(cparent);   /* discard parent */
                cparent = NULL;                         /* safety */
                pip = ip->pip;          /* safe, ip held locked */
                if (pip)
                        hammer2_inode_ref(pip);         /* loop */
                hammer2_inode_unlock(ip, NULL);
        }

        /*
         * chain is locked, ip is locked.  Unlock ip, return the locked
         * chain.  *ipp is already set w/a ref count and not locked.
         *
         * (cparent is already unlocked).
         */
        *clusterp = rcluster;
        if (rcluster) {
                if (cparentp) {
                        *cparentp = cparent;
                        hammer2_inode_unlock(ip, NULL);
                } else {
                        hammer2_inode_unlock(ip, cparent);
                }
                return (0);
        } else {
                if (cparentp)
                        *cparentp = NULL;
                if (ip)
                        hammer2_inode_unlock(ip, cparent);
                return (EIO);
        }
}

/*
 * Find the directory common to both fdip and tdip.
 *
 * Returns a held but not locked inode.  Caller typically locks the inode,
 * and when through unlocks AND drops it.
 */
hammer2_inode_t *
hammer2_inode_common_parent(hammer2_inode_t *fdip, hammer2_inode_t *tdip)
{
        hammer2_inode_t *scan1;
        hammer2_inode_t *scan2;

        /*
         * We used to have a depth field but it complicated matters too
         * much for directory renames.  So now its ugly.  Check for
         * simple cases before giving up and doing it the expensive way.
         *
         * XXX need a bottom-up topology stability lock
         */
        if (fdip == tdip || fdip == tdip->pip) {
                hammer2_inode_ref(fdip);
                return(fdip);
        }
        if (fdip->pip == tdip) {
                hammer2_inode_ref(tdip);
                return(tdip);
        }

        /*
         * XXX not MPSAFE
         */
        for (scan1 = fdip; scan1->pmp == fdip->pmp; scan1 = scan1->pip) {
                scan2 = tdip;
                while (scan2->pmp == tdip->pmp) {
                        if (scan1 == scan2) {
                                hammer2_inode_ref(scan1);
                                return(scan1);
                        }
                        scan2 = scan2->pip;
                        if (scan2 == NULL)
                                break;
                }
        }
        panic("hammer2_inode_common_parent: no common parent %p %p\n",
              fdip, tdip);
        /* NOT REACHED */
        return(NULL);
}

/*
 * Synchronize the inode's frontend state with the chain state prior
 * to any explicit flush of the inode or any strategy write call.
 *
 * Called with a locked inode.
 */
void
hammer2_inode_fsync(hammer2_trans_t *trans, hammer2_inode_t *ip, 
                    hammer2_cluster_t *cparent)
{
        const hammer2_inode_data_t *ripdata;
        hammer2_inode_data_t *wipdata;
        hammer2_cluster_t *dparent;
        hammer2_cluster_t *cluster;
        hammer2_key_t lbase;
        hammer2_key_t key_next;
        int dosync = 0;

        ripdata = &hammer2_cluster_rdata(cparent)->ipdata;    /* target file */

        if (ip->flags & HAMMER2_INODE_MTIME) {
                wipdata = hammer2_cluster_modify_ip(trans, ip, cparent, 0);
                atomic_clear_int(&ip->flags, HAMMER2_INODE_MTIME);
                wipdata->meta.mtime = ip->meta.mtime;
                dosync = 1;
                ripdata = wipdata;
        }
        if ((ip->flags & HAMMER2_INODE_RESIZED) &&
            ip->meta.size < ripdata->meta.size) {
                wipdata = hammer2_cluster_modify_ip(trans, ip, cparent, 0);
                wipdata->meta.size = ip->meta.size;
                dosync = 1;
                ripdata = wipdata;
                atomic_clear_int(&ip->flags, HAMMER2_INODE_RESIZED);

                /*
                 * We must delete any chains beyond the EOF.  The chain
                 * straddling the EOF will be pending in the bioq.
                 */
                lbase = (ripdata->meta.size + HAMMER2_PBUFMASK64) &
                        ~HAMMER2_PBUFMASK64;
                dparent = hammer2_cluster_lookup_init(&ip->cluster, 0);
                cluster = hammer2_cluster_lookup(dparent, &key_next,
                                                 lbase, (hammer2_key_t)-1,
                                                 HAMMER2_LOOKUP_NODATA);
                while (cluster) {
                        /*
                         * Degenerate embedded case, nothing to loop on
                         */
                        switch (hammer2_cluster_type(cluster)) {
                        case HAMMER2_BREF_TYPE_INODE:
                                hammer2_cluster_unlock(cluster);
                                hammer2_cluster_drop(cluster);
                                cluster = NULL;
                                break;
                        case HAMMER2_BREF_TYPE_DATA:
                                hammer2_cluster_delete(trans, dparent, cluster,
                                                   HAMMER2_DELETE_PERMANENT);
                                /* fall through */
                        default:
                                cluster = hammer2_cluster_next(dparent, cluster,
                                                   &key_next,
                                                   key_next, (hammer2_key_t)-1,
                                                   HAMMER2_LOOKUP_NODATA);
                                break;
                        }
                }
                hammer2_cluster_lookup_done(dparent);
        } else
        if ((ip->flags & HAMMER2_INODE_RESIZED) &&
            ip->meta.size > ripdata->meta.size) {
                wipdata = hammer2_cluster_modify_ip(trans, ip, cparent, 0);
                wipdata->meta.size = ip->meta.size;
                atomic_clear_int(&ip->flags, HAMMER2_INODE_RESIZED);

                /*
                 * When resizing larger we may not have any direct-data
                 * available.
                 */
                if ((wipdata->meta.op_flags & HAMMER2_OPFLAG_DIRECTDATA) &&
                    ip->meta.size > HAMMER2_EMBEDDED_BYTES) {
                        wipdata->meta.op_flags &= ~HAMMER2_OPFLAG_DIRECTDATA;
                        bzero(&wipdata->u.blockset,
                              sizeof(wipdata->u.blockset));
                }
                dosync = 1;
                ripdata = wipdata;
        }
        if (dosync)
                hammer2_cluster_modsync(cparent);
}