hammer2 - Fix focus vs modify race

[dragonfly.git] / sys / vfs / hammer2 / hammer2_vnops.c

/*
 * Copyright (c) 2011-2018 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>
 * by Daniel Flores (GSOC 2013 - mentored by Matthew Dillon, compression)
 *
 * 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.
 */
/*
 * Kernel Filesystem interface
 *
 * NOTE! local ipdata pointers must be reloaded on any modifying operation
 *       to the inode as its underlying chain may have changed.
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/fcntl.h>
#include <sys/buf.h>
#include <sys/proc.h>
#include <sys/namei.h>
#include <sys/mount.h>
#include <sys/vnode.h>
#include <sys/mountctl.h>
#include <sys/dirent.h>
#include <sys/uio.h>
#include <sys/objcache.h>
#include <sys/event.h>
#include <sys/file.h>
#include <vfs/fifofs/fifo.h>

#include "hammer2.h"

static int hammer2_read_file(hammer2_inode_t *ip, struct uio *uio,
                                int seqcount);
static int hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
                                int ioflag, int seqcount);
static void hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize);
static void hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize);

struct objcache *cache_xops;

static __inline
void
hammer2_knote(struct vnode *vp, int flags)
{
        if (flags)
                KNOTE(&vp->v_pollinfo.vpi_kqinfo.ki_note, flags);
}

/*
 * Last reference to a vnode is going away but it is still cached.
 */
static
int
hammer2_vop_inactive(struct vop_inactive_args *ap)
{
        hammer2_inode_t *ip;
        struct vnode *vp;

        vp = ap->a_vp;
        ip = VTOI(vp);

        /*
         * Degenerate case
         */
        if (ip == NULL) {
                vrecycle(vp);
                return (0);
        }

        /*
         * Check for deleted inodes and recycle immediately on the last
         * release.  Be sure to destroy any left-over buffer cache buffers
         * so we do not waste time trying to flush them.
         *
         * Note that deleting the file block chains under the inode chain
         * would just be a waste of energy, so don't do it.
         *
         * WARNING: nvtruncbuf() can only be safely called without the inode
         *          lock held due to the way our write thread works.
         */
        if (ip->flags & HAMMER2_INODE_ISUNLINKED) {
                hammer2_key_t lbase;
                int nblksize;

                /*
                 * Detect updates to the embedded data which may be
                 * synchronized by the strategy code.  Simply mark the
                 * inode modified so it gets picked up by our normal flush.
                 */
                nblksize = hammer2_calc_logical(ip, 0, &lbase, NULL);
                nvtruncbuf(vp, 0, nblksize, 0, 0);
                vrecycle(vp);
        }
        return (0);
}

/*
 * Reclaim a vnode so that it can be reused; after the inode is
 * disassociated, the filesystem must manage it alone.
 */
static
int
hammer2_vop_reclaim(struct vop_reclaim_args *ap)
{
        hammer2_inode_t *ip;
        hammer2_pfs_t *pmp;
        struct vnode *vp;

        vp = ap->a_vp;
        ip = VTOI(vp);
        if (ip == NULL) {
                return(0);
        }
        pmp = ip->pmp;

        /*
         * The final close of a deleted file or directory marks it for
         * destruction.  The DELETED flag allows the flusher to shortcut
         * any modified blocks still unflushed (that is, just ignore them).
         *
         * HAMMER2 usually does not try to optimize the freemap by returning
         * deleted blocks to it as it does not usually know how many snapshots
         * might be referencing portions of the file/dir.
         */
        vp->v_data = NULL;
        ip->vp = NULL;

        /*
         * NOTE! We do not attempt to flush chains here, flushing is
         *       really fragile and could also deadlock.
         */
        vclrisdirty(vp);

        /*
         * A modified inode may require chain synchronization.  This
         * synchronization is usually handled by VOP_SYNC / VOP_FSYNC
         * when vfsync() is called.  However, that requires a vnode.
         *
         * When the vnode is disassociated we must keep track of any modified
         * inode via the sideq so that it is properly flushed.  We cannot
         * safely synchronize the inode from inside the reclaim due to
         * potentially deep locks held as-of when the reclaim occurs.
         * Interactions and potential deadlocks abound.
         */
        if ((ip->flags & (HAMMER2_INODE_ISUNLINKED |
                          HAMMER2_INODE_MODIFIED |
                          HAMMER2_INODE_RESIZED |
                          HAMMER2_INODE_DIRTYDATA)) &&
            (ip->flags & HAMMER2_INODE_ISDELETED) == 0) {
                hammer2_inode_sideq_t *ipul;

                ipul = kmalloc(sizeof(*ipul), pmp->minode, M_WAITOK | M_ZERO);
                ipul->ip = ip;

                hammer2_spin_ex(&pmp->list_spin);
                if ((ip->flags & HAMMER2_INODE_ONSIDEQ) == 0) {
                        /* ref -> sideq */
                        atomic_set_int(&ip->flags, HAMMER2_INODE_ONSIDEQ);
                        TAILQ_INSERT_TAIL(&pmp->sideq, ipul, entry);
                        ++pmp->sideq_count;
                        hammer2_spin_unex(&pmp->list_spin);
                } else {
                        hammer2_spin_unex(&pmp->list_spin);
                        kfree(ipul, pmp->minode);
                        hammer2_inode_drop(ip);         /* vp ref */
                }
                /* retain ref from vp for ipul */
        } else {
                hammer2_inode_drop(ip);                 /* vp ref */
        }

        /*
         * XXX handle background sync when ip dirty, kernel will no longer
         * notify us regarding this inode because there is no longer a
         * vnode attached to it.
         */

        return (0);
}

/*
 * Currently this function synchronizes the front-end inode state to the
 * backend chain topology, then flushes the inode's chain and sub-topology
 * to backend media.  This function does not flush the root topology down to
 * the inode.
 */
static
int
hammer2_vop_fsync(struct vop_fsync_args *ap)
{
        hammer2_inode_t *ip;
        struct vnode *vp;
        int error1;
        int error2;

        vp = ap->a_vp;
        ip = VTOI(vp);
        error1 = 0;

        hammer2_trans_init(ip->pmp, 0);

        /*
         * Flush dirty buffers in the file's logical buffer cache.
         * It is best to wait for the strategy code to commit the
         * buffers to the device's backing buffer cache before
         * then trying to flush the inode.
         *
         * This should be quick, but certain inode modifications cached
         * entirely in the hammer2_inode structure may not trigger a
         * buffer read until the flush so the fsync can wind up also
         * doing scattered reads.
         */
        vfsync(vp, ap->a_waitfor, 1, NULL, NULL);
        bio_track_wait(&vp->v_track_write, 0, 0);

        /*
         * Flush any inode changes
         */
        hammer2_inode_lock(ip, 0);
        if (ip->flags & (HAMMER2_INODE_RESIZED|HAMMER2_INODE_MODIFIED))
                error1 = hammer2_inode_chain_sync(ip);

        /*
         * Flush dirty chains related to the inode.
         *
         * NOTE! XXX We do not currently flush to the volume root, ultimately
         *       we will want to have a shortcut for the flushed inode stored
         *       in the volume root for recovery purposes.
         */
        error2 = hammer2_inode_chain_flush(ip);
        if (error2)
                error1 = error2;

        /*
         * We may be able to clear the vnode dirty flag.  The
         * hammer2_pfs_moderate() code depends on this usually working.
         */
        if ((ip->flags & (HAMMER2_INODE_MODIFIED |
                          HAMMER2_INODE_RESIZED |
                          HAMMER2_INODE_DIRTYDATA)) == 0 &&
            RB_EMPTY(&vp->v_rbdirty_tree) &&
            !bio_track_active(&vp->v_track_write)) {
                vclrisdirty(vp);
        }
        hammer2_inode_unlock(ip);
        hammer2_trans_done(ip->pmp, 0);

        return (error1);
}

static
int
hammer2_vop_access(struct vop_access_args *ap)
{
        hammer2_inode_t *ip = VTOI(ap->a_vp);
        uid_t uid;
        gid_t gid;
        int error;

        hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);
        uid = hammer2_to_unix_xid(&ip->meta.uid);
        gid = hammer2_to_unix_xid(&ip->meta.gid);
        error = vop_helper_access(ap, uid, gid, ip->meta.mode, ip->meta.uflags);
        hammer2_inode_unlock(ip);

        return (error);
}

static
int
hammer2_vop_getattr(struct vop_getattr_args *ap)
{
        hammer2_pfs_t *pmp;
        hammer2_inode_t *ip;
        struct vnode *vp;
        struct vattr *vap;
        hammer2_chain_t *chain;
        int i;

        vp = ap->a_vp;
        vap = ap->a_vap;

        ip = VTOI(vp);
        pmp = ip->pmp;

        hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);

        vap->va_fsid = pmp->mp->mnt_stat.f_fsid.val[0];
        vap->va_fileid = ip->meta.inum;
        vap->va_mode = ip->meta.mode;
        vap->va_nlink = ip->meta.nlinks;
        vap->va_uid = hammer2_to_unix_xid(&ip->meta.uid);
        vap->va_gid = hammer2_to_unix_xid(&ip->meta.gid);
        vap->va_rmajor = 0;
        vap->va_rminor = 0;
        vap->va_size = ip->meta.size;   /* protected by shared lock */
        vap->va_blocksize = HAMMER2_PBUFSIZE;
        vap->va_flags = ip->meta.uflags;
        hammer2_time_to_timespec(ip->meta.ctime, &vap->va_ctime);
        hammer2_time_to_timespec(ip->meta.mtime, &vap->va_mtime);
        hammer2_time_to_timespec(ip->meta.mtime, &vap->va_atime);
        vap->va_gen = 1;
        vap->va_bytes = 0;
        if (ip->meta.type == HAMMER2_OBJTYPE_DIRECTORY) {
                /*
                 * Can't really calculate directory use sans the files under
                 * it, just assume one block for now.
                 */
                vap->va_bytes += HAMMER2_INODE_BYTES;
        } else {
                for (i = 0; i < ip->cluster.nchains; ++i) {
                        if ((chain = ip->cluster.array[i].chain) != NULL) {
                                if (vap->va_bytes <
                                    chain->bref.embed.stats.data_count) {
                                        vap->va_bytes =
                                            chain->bref.embed.stats.data_count;
                                }
                        }
                }
        }
        vap->va_type = hammer2_get_vtype(ip->meta.type);
        vap->va_filerev = 0;
        vap->va_uid_uuid = ip->meta.uid;
        vap->va_gid_uuid = ip->meta.gid;
        vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
                          VA_FSID_UUID_VALID;

        hammer2_inode_unlock(ip);

        return (0);
}

static
int
hammer2_vop_setattr(struct vop_setattr_args *ap)
{
        hammer2_inode_t *ip;
        struct vnode *vp;
        struct vattr *vap;
        int error;
        int kflags = 0;
        uint64_t ctime;

        vp = ap->a_vp;
        vap = ap->a_vap;
        hammer2_update_time(&ctime);

        ip = VTOI(vp);

        if (ip->pmp->ronly)
                return (EROFS);
        if (hammer2_vfs_enospace(ip, 0, ap->a_cred) > 1)
                return (ENOSPC);

        hammer2_pfs_memory_wait(ip, 0);
        hammer2_trans_init(ip->pmp, 0);
        hammer2_inode_lock(ip, 0);
        error = 0;

        if (vap->va_flags != VNOVAL) {
                uint32_t flags;

                flags = ip->meta.uflags;
                error = vop_helper_setattr_flags(&flags, vap->va_flags,
                                     hammer2_to_unix_xid(&ip->meta.uid),
                                     ap->a_cred);
                if (error == 0) {
                        if (ip->meta.uflags != flags) {
                                hammer2_inode_modify(ip);
                                ip->meta.uflags = flags;
                                ip->meta.ctime = ctime;
                                kflags |= NOTE_ATTRIB;
                        }
                        if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
                                error = 0;
                                goto done;
                        }
                }
                goto done;
        }
        if (ip->meta.uflags & (IMMUTABLE | APPEND)) {
                error = EPERM;
                goto done;
        }
        if (vap->va_uid != (uid_t)VNOVAL || vap->va_gid != (gid_t)VNOVAL) {
                mode_t cur_mode = ip->meta.mode;
                uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
                gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);
                uuid_t uuid_uid;
                uuid_t uuid_gid;

                error = vop_helper_chown(ap->a_vp, vap->va_uid, vap->va_gid,
                                         ap->a_cred,
                                         &cur_uid, &cur_gid, &cur_mode);
                if (error == 0) {
                        hammer2_guid_to_uuid(&uuid_uid, cur_uid);
                        hammer2_guid_to_uuid(&uuid_gid, cur_gid);
                        if (bcmp(&uuid_uid, &ip->meta.uid, sizeof(uuid_uid)) ||
                            bcmp(&uuid_gid, &ip->meta.gid, sizeof(uuid_gid)) ||
                            ip->meta.mode != cur_mode
                        ) {
                                hammer2_inode_modify(ip);
                                ip->meta.uid = uuid_uid;
                                ip->meta.gid = uuid_gid;
                                ip->meta.mode = cur_mode;
                                ip->meta.ctime = ctime;
                        }
                        kflags |= NOTE_ATTRIB;
                }
        }

        /*
         * Resize the file
         */
        if (vap->va_size != VNOVAL && ip->meta.size != vap->va_size) {
                switch(vp->v_type) {
                case VREG:
                        if (vap->va_size == ip->meta.size)
                                break;
                        if (vap->va_size < ip->meta.size) {
                                hammer2_mtx_ex(&ip->truncate_lock);
                                hammer2_truncate_file(ip, vap->va_size);
                                hammer2_mtx_unlock(&ip->truncate_lock);
                                kflags |= NOTE_WRITE;
                        } else {
                                hammer2_extend_file(ip, vap->va_size);
                                kflags |= NOTE_WRITE | NOTE_EXTEND;
                        }
                        hammer2_inode_modify(ip);
                        ip->meta.mtime = ctime;
                        vclrflags(vp, VLASTWRITETS);
                        break;
                default:
                        error = EINVAL;
                        goto done;
                }
        }
#if 0
        /* atime not supported */
        if (vap->va_atime.tv_sec != VNOVAL) {
                hammer2_inode_modify(ip);
                ip->meta.atime = hammer2_timespec_to_time(&vap->va_atime);
                kflags |= NOTE_ATTRIB;
        }
#endif
        if (vap->va_mode != (mode_t)VNOVAL) {
                mode_t cur_mode = ip->meta.mode;
                uid_t cur_uid = hammer2_to_unix_xid(&ip->meta.uid);
                gid_t cur_gid = hammer2_to_unix_xid(&ip->meta.gid);

                error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
                                         cur_uid, cur_gid, &cur_mode);
                if (error == 0 && ip->meta.mode != cur_mode) {
                        hammer2_inode_modify(ip);
                        ip->meta.mode = cur_mode;
                        ip->meta.ctime = ctime;
                        kflags |= NOTE_ATTRIB;
                }
        }

        if (vap->va_mtime.tv_sec != VNOVAL) {
                hammer2_inode_modify(ip);
                ip->meta.mtime = hammer2_timespec_to_time(&vap->va_mtime);
                kflags |= NOTE_ATTRIB;
                vclrflags(vp, VLASTWRITETS);
        }

done:
        /*
         * If a truncation occurred we must call chain_sync() now in order
         * to trim the related data chains, otherwise a later expansion can
         * cause havoc.
         *
         * If an extend occured that changed the DIRECTDATA state, we must
         * call inode_fsync now in order to prepare the inode's indirect
         * block table.
         *
         * WARNING! This means we are making an adjustment to the inode's
         * chain outside of sync/fsync, and not just to inode->meta, which
         * may result in some consistency issues if a crash were to occur
         * at just the wrong time.
         */
        if (ip->flags & HAMMER2_INODE_RESIZED)
                hammer2_inode_chain_sync(ip);

        /*
         * Cleanup.
         */
        hammer2_inode_unlock(ip);
        hammer2_trans_done(ip->pmp, 1);
        hammer2_knote(ip->vp, kflags);

        return (error);
}

static
int
hammer2_vop_readdir(struct vop_readdir_args *ap)
{
        hammer2_xop_readdir_t *xop;
        hammer2_blockref_t bref;
        hammer2_inode_t *ip;
        hammer2_tid_t inum;
        hammer2_key_t lkey;
        struct uio *uio;
        off_t *cookies;
        off_t saveoff;
        int cookie_index;
        int ncookies;
        int error;
        int eofflag;
        int r;

        ip = VTOI(ap->a_vp);
        uio = ap->a_uio;
        saveoff = uio->uio_offset;
        eofflag = 0;
        error = 0;

        /*
         * Setup cookies directory entry cookies if requested
         */
        if (ap->a_ncookies) {
                ncookies = uio->uio_resid / 16 + 1;
                if (ncookies > 1024)
                        ncookies = 1024;
                cookies = kmalloc(ncookies * sizeof(off_t), M_TEMP, M_WAITOK);
        } else {
                ncookies = -1;
                cookies = NULL;
        }
        cookie_index = 0;

        hammer2_inode_lock(ip, HAMMER2_RESOLVE_SHARED);

        /*
         * Handle artificial entries.  To ensure that only positive 64 bit
         * quantities are returned to userland we always strip off bit 63.
         * The hash code is designed such that codes 0x0000-0x7FFF are not
         * used, allowing us to use these codes for articial entries.
         *
         * Entry 0 is used for '.' and entry 1 is used for '..'.  Do not
         * allow '..' to cross the mount point into (e.g.) the super-root.
         */
        if (saveoff == 0) {
                inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
                r = vop_write_dirent(&error, uio, inum, DT_DIR, 1, ".");
                if (r)
                        goto done;
                if (cookies)
                        cookies[cookie_index] = saveoff;
                ++saveoff;
                ++cookie_index;
                if (cookie_index == ncookies)
                        goto done;
        }

        if (saveoff == 1) {
                /*
                 * Be careful with lockorder when accessing ".."
                 *
                 * (ip is the current dir. xip is the parent dir).
                 */
                inum = ip->meta.inum & HAMMER2_DIRHASH_USERMSK;
                if (ip != ip->pmp->iroot)
                        inum = ip->meta.iparent & HAMMER2_DIRHASH_USERMSK;
                r = vop_write_dirent(&error, uio, inum, DT_DIR, 2, "..");
                if (r)
                        goto done;
                if (cookies)
                        cookies[cookie_index] = saveoff;
                ++saveoff;
                ++cookie_index;
                if (cookie_index == ncookies)
                        goto done;
        }

        lkey = saveoff | HAMMER2_DIRHASH_VISIBLE;
        if (hammer2_debug & 0x0020)
                kprintf("readdir: lkey %016jx\n", lkey);
        if (error)
                goto done;

        /*
         * Use XOP for cluster scan.
         *
         * parent is the inode cluster, already locked for us.  Don't
         * double lock shared locks as this will screw up upgrades.
         */
        xop = hammer2_xop_alloc(ip, 0);
        xop->lkey = lkey;
        hammer2_xop_start(&xop->head, hammer2_xop_readdir);

        for (;;) {
                const hammer2_inode_data_t *ripdata;
                const char *dname;
                int dtype;

                error = hammer2_xop_collect(&xop->head, 0);
                error = hammer2_error_to_errno(error);
                if (error) {
                        break;
                }
                if (cookie_index == ncookies)
                        break;
                if (hammer2_debug & 0x0020)
                kprintf("cluster chain %p %p\n",
                        xop->head.cluster.focus,
                        (xop->head.cluster.focus ?
                         xop->head.cluster.focus->data : (void *)-1));
                hammer2_cluster_bref(&xop->head.cluster, &bref);

                if (bref.type == HAMMER2_BREF_TYPE_INODE) {
                        ripdata = &hammer2_xop_gdata(&xop->head)->ipdata;
                        dtype = hammer2_get_dtype(ripdata->meta.type);
                        saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
                        r = vop_write_dirent(&error, uio,
                                             ripdata->meta.inum &
                                              HAMMER2_DIRHASH_USERMSK,
                                             dtype,
                                             ripdata->meta.name_len,
                                             ripdata->filename);
                        hammer2_xop_pdata(&xop->head);
                        if (r)
                                break;
                        if (cookies)
                                cookies[cookie_index] = saveoff;
                        ++cookie_index;
                } else if (bref.type == HAMMER2_BREF_TYPE_DIRENT) {
                        uint16_t namlen;

                        dtype = hammer2_get_dtype(bref.embed.dirent.type);
                        saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
                        namlen = bref.embed.dirent.namlen;
                        if (namlen <= sizeof(bref.check.buf)) {
                                dname = bref.check.buf;
                        } else {
                                dname = hammer2_xop_gdata(&xop->head)->buf;
                        }
                        r = vop_write_dirent(&error, uio,
                                             bref.embed.dirent.inum, dtype,
                                             namlen, dname);
                        if (namlen > sizeof(bref.check.buf))
                                hammer2_xop_pdata(&xop->head);
                        if (r)
                                break;
                        if (cookies)
                                cookies[cookie_index] = saveoff;
                        ++cookie_index;
                } else {
                        /* XXX chain error */
                        kprintf("bad chain type readdir %d\n", bref.type);
                }
        }
        hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
        if (error == ENOENT) {
                error = 0;
                eofflag = 1;
                saveoff = (hammer2_key_t)-1;
        } else {
                saveoff = bref.key & HAMMER2_DIRHASH_USERMSK;
        }
done:
        hammer2_inode_unlock(ip);
        if (ap->a_eofflag)
                *ap->a_eofflag = eofflag;
        if (hammer2_debug & 0x0020)
                kprintf("readdir: done at %016jx\n", saveoff);
        uio->uio_offset = saveoff & ~HAMMER2_DIRHASH_VISIBLE;
        if (error && cookie_index == 0) {
                if (cookies) {
                        kfree(cookies, M_TEMP);
                        *ap->a_ncookies = 0;
                        *ap->a_cookies = NULL;
                }
        } else {
                if (cookies) {
                        *ap->a_ncookies = cookie_index;
                        *ap->a_cookies = cookies;
                }
        }
        return (error);
}

/*
 * hammer2_vop_readlink { vp, uio, cred }
 */
static
int
hammer2_vop_readlink(struct vop_readlink_args *ap)
{
        struct vnode *vp;
        hammer2_inode_t *ip;
        int error;

        vp = ap->a_vp;
        if (vp->v_type != VLNK)
                return (EINVAL);
        ip = VTOI(vp);

        error = hammer2_read_file(ip, ap->a_uio, 0);
        return (error);
}

static
int
hammer2_vop_read(struct vop_read_args *ap)
{
        struct vnode *vp;
        hammer2_inode_t *ip;
        struct uio *uio;
        int error;
        int seqcount;
        int bigread;

        /*
         * Read operations supported on this vnode?
         */
        vp = ap->a_vp;
        if (vp->v_type != VREG)
                return (EINVAL);

        /*
         * Misc
         */
        ip = VTOI(vp);
        uio = ap->a_uio;
        error = 0;

        seqcount = ap->a_ioflag >> 16;
        bigread = (uio->uio_resid > 100 * 1024 * 1024);

        error = hammer2_read_file(ip, uio, seqcount);
        return (error);
}

static
int
hammer2_vop_write(struct vop_write_args *ap)
{
        hammer2_inode_t *ip;
        thread_t td;
        struct vnode *vp;
        struct uio *uio;
        int error;
        int seqcount;
        int ioflag;

        /*
         * Read operations supported on this vnode?
         */
        vp = ap->a_vp;
        if (vp->v_type != VREG)
                return (EINVAL);

        /*
         * Misc
         */
        ip = VTOI(vp);
        ioflag = ap->a_ioflag;
        uio = ap->a_uio;
        error = 0;
        if (ip->pmp->ronly)
                return (EROFS);
        switch (hammer2_vfs_enospace(ip, uio->uio_resid, ap->a_cred)) {
        case 2:
                return (ENOSPC);
        case 1:
                ioflag |= IO_DIRECT;    /* semi-synchronous */
                /* fall through */
        default:
                break;
        }

        seqcount = ioflag >> 16;

        /*
         * Check resource limit
         */
        if (uio->uio_resid > 0 && (td = uio->uio_td) != NULL && td->td_proc &&
            uio->uio_offset + uio->uio_resid >
             td->td_proc->p_rlimit[RLIMIT_FSIZE].rlim_cur) {
                lwpsignal(td->td_proc, td->td_lwp, SIGXFSZ);
                return (EFBIG);
        }

        /*
         * The transaction interlocks against flush initiations
         * (note: but will run concurrently with the actual flush).
         *
         * To avoid deadlocking against the VM system, we must flag any
         * transaction related to the buffer cache or other direct
         * VM page manipulation.
         */
        if (uio->uio_segflg == UIO_NOCOPY) {
                hammer2_trans_init(ip->pmp, HAMMER2_TRANS_BUFCACHE);
        } else {
                hammer2_pfs_memory_wait(ip, 0);
                hammer2_trans_init(ip->pmp, 0);
        }
        error = hammer2_write_file(ip, uio, ioflag, seqcount);
        hammer2_trans_done(ip->pmp, 1);

        return (error);
}

/*
 * Perform read operations on a file or symlink given an UNLOCKED
 * inode and uio.
 *
 * The passed ip is not locked.
 */
static
int
hammer2_read_file(hammer2_inode_t *ip, struct uio *uio, int seqcount)
{
        hammer2_off_t size;
        struct buf *bp;
        int error;

        error = 0;

        /*
         * UIO read loop.
         *
         * WARNING! Assumes that the kernel interlocks size changes at the
         *          vnode level.
         */
        hammer2_mtx_sh(&ip->lock);
        hammer2_mtx_sh(&ip->truncate_lock);
        size = ip->meta.size;
        hammer2_mtx_unlock(&ip->lock);

        while (uio->uio_resid > 0 && uio->uio_offset < size) {
                hammer2_key_t lbase;
                hammer2_key_t leof;
                int lblksize;
                int loff;
                int n;

                lblksize = hammer2_calc_logical(ip, uio->uio_offset,
                                                &lbase, &leof);

#if 1
                bp = NULL;
                error = cluster_readx(ip->vp, leof, lbase, lblksize,
                                      B_NOTMETA | B_KVABIO,
                                      uio->uio_resid,
                                      seqcount * MAXBSIZE,
                                      &bp);
#else
                if (uio->uio_segflg == UIO_NOCOPY) {
                        bp = getblk(ip->vp, lbase, lblksize,
                                    GETBLK_BHEAVY | GETBLK_KVABIO, 0);
                        if (bp->b_flags & B_CACHE) {
                                int i;
                                int j = 0;
                                if (bp->b_xio.xio_npages != 16)
                                        kprintf("NPAGES BAD\n");
                                for (i = 0; i < bp->b_xio.xio_npages; ++i) {
                                        vm_page_t m;
                                        m = bp->b_xio.xio_pages[i];
                                        if (m == NULL || m->valid == 0) {
                                                kprintf("bp %016jx %016jx pg %d inv",
                                                        lbase, leof, i);
                                                if (m)
                                                        kprintf("m->object %p/%p", m->object, ip->vp->v_object);
                                                kprintf("\n");
                                                j = 1;
                                        }
                                }
                                if (j)
                                        kprintf("b_flags %08x, b_error %d\n", bp->b_flags, bp->b_error);
                        }
                        bqrelse(bp);
                }
                error = bread_kvabio(ip->vp, lbase, lblksize, &bp);
#endif
                if (error) {
                        brelse(bp);
                        break;
                }
                bkvasync(bp);
                loff = (int)(uio->uio_offset - lbase);
                n = lblksize - loff;
                if (n > uio->uio_resid)
                        n = uio->uio_resid;
                if (n > size - uio->uio_offset)
                        n = (int)(size - uio->uio_offset);
                bp->b_flags |= B_AGE;
                uiomovebp(bp, (char *)bp->b_data + loff, n, uio);
                bqrelse(bp);
        }
        hammer2_mtx_unlock(&ip->truncate_lock);

        return (error);
}

/*
 * Write to the file represented by the inode via the logical buffer cache.
 * The inode may represent a regular file or a symlink.
 *
 * The inode must not be locked.
 */
static
int
hammer2_write_file(hammer2_inode_t *ip, struct uio *uio,
                   int ioflag, int seqcount)
{
        hammer2_key_t old_eof;
        hammer2_key_t new_eof;
        struct buf *bp;
        int kflags;
        int error;
        int modified;

        /*
         * Setup if append
         *
         * WARNING! Assumes that the kernel interlocks size changes at the
         *          vnode level.
         */
        hammer2_mtx_ex(&ip->lock);
        hammer2_mtx_sh(&ip->truncate_lock);
        if (ioflag & IO_APPEND)
                uio->uio_offset = ip->meta.size;
        old_eof = ip->meta.size;

        /*
         * Extend the file if necessary.  If the write fails at some point
         * we will truncate it back down to cover as much as we were able
         * to write.
         *
         * Doing this now makes it easier to calculate buffer sizes in
         * the loop.
         */
        kflags = 0;
        error = 0;
        modified = 0;

        if (uio->uio_offset + uio->uio_resid > old_eof) {
                new_eof = uio->uio_offset + uio->uio_resid;
                modified = 1;
                hammer2_extend_file(ip, new_eof);
                kflags |= NOTE_EXTEND;
        } else {
                new_eof = old_eof;
        }
        hammer2_mtx_unlock(&ip->lock);

        /*
         * UIO write loop
         */
        while (uio->uio_resid > 0) {
                hammer2_key_t lbase;
                int trivial;
                int endofblk;
                int lblksize;
                int loff;
                int n;

                /*
                 * Don't allow the buffer build to blow out the buffer
                 * cache.
                 */
                if ((ioflag & IO_RECURSE) == 0)
                        bwillwrite(HAMMER2_PBUFSIZE);

                /*
                 * This nominally tells us how much we can cluster and
                 * what the logical buffer size needs to be.  Currently
                 * we don't try to cluster the write and just handle one
                 * block at a time.
                 */
                lblksize = hammer2_calc_logical(ip, uio->uio_offset,
                                                &lbase, NULL);
                loff = (int)(uio->uio_offset - lbase);

                KKASSERT(lblksize <= 65536);

                /*
                 * Calculate bytes to copy this transfer and whether the
                 * copy completely covers the buffer or not.
                 */
                trivial = 0;
                n = lblksize - loff;
                if (n > uio->uio_resid) {
                        n = uio->uio_resid;
                        if (loff == lbase && uio->uio_offset + n == new_eof)
                                trivial = 1;
                        endofblk = 0;
                } else {
                        if (loff == 0)
                                trivial = 1;
                        endofblk = 1;
                }
                if (lbase >= new_eof)
                        trivial = 1;

                /*
                 * Get the buffer
                 */
                if (uio->uio_segflg == UIO_NOCOPY) {
                        /*
                         * Issuing a write with the same data backing the
                         * buffer.  Instantiate the buffer to collect the
                         * backing vm pages, then read-in any missing bits.
                         *
                         * This case is used by vop_stdputpages().
                         */
                        bp = getblk(ip->vp, lbase, lblksize,
                                    GETBLK_BHEAVY | GETBLK_KVABIO, 0);
                        if ((bp->b_flags & B_CACHE) == 0) {
                                bqrelse(bp);
                                error = bread_kvabio(ip->vp, lbase,
                                                     lblksize, &bp);
                        }
                } else if (trivial) {
                        /*
                         * Even though we are entirely overwriting the buffer
                         * we may still have to zero it out to avoid a
                         * mmap/write visibility issue.
                         */
                        bp = getblk(ip->vp, lbase, lblksize,
                                    GETBLK_BHEAVY | GETBLK_KVABIO, 0);
                        if ((bp->b_flags & B_CACHE) == 0)
                                vfs_bio_clrbuf(bp);
                } else {
                        /*
                         * Partial overwrite, read in any missing bits then
                         * replace the portion being written.
                         *
                         * (The strategy code will detect zero-fill physical
                         * blocks for this case).
                         */
                        error = bread_kvabio(ip->vp, lbase, lblksize, &bp);
                        if (error == 0)
                                bheavy(bp);
                }

                if (error) {
                        brelse(bp);
                        break;
                }

                /*
                 * Ok, copy the data in
                 */
                bkvasync(bp);
                error = uiomovebp(bp, bp->b_data + loff, n, uio);
                kflags |= NOTE_WRITE;
                modified = 1;
                if (error) {
                        brelse(bp);
                        break;
                }

                /*
                 * WARNING: Pageout daemon will issue UIO_NOCOPY writes
                 *          with IO_SYNC or IO_ASYNC set.  These writes
                 *          must be handled as the pageout daemon expects.
                 *
                 * NOTE!    H2 relies on cluster_write() here because it
                 *          cannot preallocate disk blocks at the logical
                 *          level due to not knowing what the compression
                 *          size will be at this time.
                 *
                 *          We must use cluster_write() here and we depend
                 *          on the write-behind feature to flush buffers
                 *          appropriately.  If we let the buffer daemons do
                 *          it the block allocations will be all over the
                 *          map.
                 */
                if (ioflag & IO_SYNC) {
                        bwrite(bp);
                } else if ((ioflag & IO_DIRECT) && endofblk) {
                        bawrite(bp);
                } else if (ioflag & IO_ASYNC) {
                        bawrite(bp);
                } else if (ip->vp->v_mount->mnt_flag & MNT_NOCLUSTERW) {
                        bdwrite(bp);
                } else {
#if 1
                        bp->b_flags |= B_CLUSTEROK;
                        cluster_write(bp, new_eof, lblksize, seqcount);
#else
                        bp->b_flags |= B_CLUSTEROK;
                        bdwrite(bp);
#endif
                }
        }

        /*
         * Cleanup.  If we extended the file EOF but failed to write through
         * the entire write is a failure and we have to back-up.
         */
        if (error && new_eof != old_eof) {
                hammer2_mtx_unlock(&ip->truncate_lock);
                hammer2_mtx_ex(&ip->lock);
                hammer2_mtx_ex(&ip->truncate_lock);
                hammer2_truncate_file(ip, old_eof);
                if (ip->flags & HAMMER2_INODE_MODIFIED)
                        hammer2_inode_chain_sync(ip);
                hammer2_mtx_unlock(&ip->lock);
        } else if (modified) {
                struct vnode *vp = ip->vp;

                hammer2_mtx_ex(&ip->lock);
                hammer2_inode_modify(ip);
                if (uio->uio_segflg == UIO_NOCOPY) {
                        if (vp->v_flag & VLASTWRITETS) {
                                ip->meta.mtime =
                                    (unsigned long)vp->v_lastwrite_ts.tv_sec *
                                    1000000 +
                                    vp->v_lastwrite_ts.tv_nsec / 1000;
                        }
                } else {
                        hammer2_update_time(&ip->meta.mtime);
                        vclrflags(vp, VLASTWRITETS);
                }

#if 0
                /*
                 * REMOVED - handled by hammer2_extend_file().  Do not issue
                 * a chain_sync() outside of a sync/fsync except for DIRECTDATA
                 * state changes.
                 *
                 * Under normal conditions we only issue a chain_sync if
                 * the inode's DIRECTDATA state changed.
                 */
                if (ip->flags & HAMMER2_INODE_RESIZED)
                        hammer2_inode_chain_sync(ip);
#endif
                hammer2_mtx_unlock(&ip->lock);
                hammer2_knote(ip->vp, kflags);
        }
        hammer2_trans_assert_strategy(ip->pmp);
        hammer2_mtx_unlock(&ip->truncate_lock);

        return error;
}

/*
 * Truncate the size of a file.  The inode must not be locked.
 *
 * We must unconditionally set HAMMER2_INODE_RESIZED to properly
 * ensure that any on-media data beyond the new file EOF has been destroyed.
 *
 * WARNING: nvtruncbuf() can only be safely called without the inode lock
 *          held due to the way our write thread works.  If the truncation
 *          occurs in the middle of a buffer, nvtruncbuf() is responsible
 *          for dirtying that buffer and zeroing out trailing bytes.
 *
 * WARNING! Assumes that the kernel interlocks size changes at the
 *          vnode level.
 *
 * WARNING! Caller assumes responsibility for removing dead blocks
 *          if INODE_RESIZED is set.
 */
static
void
hammer2_truncate_file(hammer2_inode_t *ip, hammer2_key_t nsize)
{
        hammer2_key_t lbase;
        int nblksize;

        hammer2_mtx_unlock(&ip->lock);
        if (ip->vp) {
                nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
                nvtruncbuf(ip->vp, nsize,
                           nblksize, (int)nsize & (nblksize - 1),
                           0);
        }
        hammer2_mtx_ex(&ip->lock);
        KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
        ip->osize = ip->meta.size;
        ip->meta.size = nsize;
        atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
        hammer2_inode_modify(ip);
}

/*
 * Extend the size of a file.  The inode must not be locked.
 *
 * Even though the file size is changing, we do not have to set the
 * INODE_RESIZED bit unless the file size crosses the EMBEDDED_BYTES
 * boundary.  When this occurs a hammer2_inode_chain_sync() is required
 * to prepare the inode cluster's indirect block table, otherwise
 * async execution of the strategy code will implode on us.
 *
 * WARNING! Assumes that the kernel interlocks size changes at the
 *          vnode level.
 *
 * WARNING! Caller assumes responsibility for transitioning out
 *          of the inode DIRECTDATA mode if INODE_RESIZED is set.
 */
static
void
hammer2_extend_file(hammer2_inode_t *ip, hammer2_key_t nsize)
{
        hammer2_key_t lbase;
        hammer2_key_t osize;
        int oblksize;
        int nblksize;

        KKASSERT((ip->flags & HAMMER2_INODE_RESIZED) == 0);
        hammer2_inode_modify(ip);
        osize = ip->meta.size;
        ip->osize = osize;
        ip->meta.size = nsize;

        /*
         * We must issue a chain_sync() when the DIRECTDATA state changes
         * to prevent confusion between the flush code and the in-memory
         * state.  This is not perfect because we are doing it outside of
         * a sync/fsync operation, so it might not be fully synchronized
         * with the meta-data topology flush.
         */
        if (osize <= HAMMER2_EMBEDDED_BYTES && nsize > HAMMER2_EMBEDDED_BYTES) {
                atomic_set_int(&ip->flags, HAMMER2_INODE_RESIZED);
                hammer2_inode_chain_sync(ip);
        }

        hammer2_mtx_unlock(&ip->lock);
        if (ip->vp) {
                oblksize = hammer2_calc_logical(ip, osize, &lbase, NULL);
                nblksize = hammer2_calc_logical(ip, nsize, &lbase, NULL);
                nvextendbuf(ip->vp,
                            osize, nsize,
                            oblksize, nblksize,
                            -1, -1, 0);
        }
        hammer2_mtx_ex(&ip->lock);
}

static
int
hammer2_vop_nresolve(struct vop_nresolve_args *ap)
{
        hammer2_xop_nresolve_t *xop;
        hammer2_inode_t *ip;
        hammer2_inode_t *dip;
        struct namecache *ncp;
        struct vnode *vp;
        int error;

        dip = VTOI(ap->a_dvp);
        xop = hammer2_xop_alloc(dip, 0);

        ncp = ap->a_nch->ncp;
        hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);

        /*
         * Note: In DragonFly the kernel handles '.' and '..'.
         */
        hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
        hammer2_xop_start(&xop->head, hammer2_xop_nresolve);

        error = hammer2_xop_collect(&xop->head, 0);
        error = hammer2_error_to_errno(error);
        if (error) {
                ip = NULL;
        } else {
                ip = hammer2_inode_get(dip->pmp, dip, &xop->head, -1);
        }
        hammer2_inode_unlock(dip);

        /*
         * Acquire the related vnode
         *
         * NOTE: For error processing, only ENOENT resolves the namecache
         *       entry to NULL, otherwise we just return the error and
         *       leave the namecache unresolved.
         *
         * NOTE: multiple hammer2_inode structures can be aliased to the
         *       same chain element, for example for hardlinks.  This
         *       use case does not 'reattach' inode associations that
         *       might already exist, but always allocates a new one.
         *
         * WARNING: inode structure is locked exclusively via inode_get
         *          but chain was locked shared.  inode_unlock()
         *          will handle it properly.
         */
        if (ip) {
                vp = hammer2_igetv(ip, &error); /* error set to UNIX error */
                if (error == 0) {
                        vn_unlock(vp);
                        cache_setvp(ap->a_nch, vp);
                } else if (error == ENOENT) {
                        cache_setvp(ap->a_nch, NULL);
                }
                hammer2_inode_unlock(ip);

                /*
                 * The vp should not be released until after we've disposed
                 * of our locks, because it might cause vop_inactive() to
                 * be called.
                 */
                if (vp)
                        vrele(vp);
        } else {
                error = ENOENT;
                cache_setvp(ap->a_nch, NULL);
        }
        hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
        KASSERT(error || ap->a_nch->ncp->nc_vp != NULL,
                ("resolve error %d/%p ap %p\n",
                 error, ap->a_nch->ncp->nc_vp, ap));

        return error;
}

static
int
hammer2_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
{
        hammer2_inode_t *dip;
        hammer2_tid_t inum;
        int error;

        dip = VTOI(ap->a_dvp);
        inum = dip->meta.iparent;
        *ap->a_vpp = NULL;

        if (inum) {
                error = hammer2_vfs_vget(ap->a_dvp->v_mount, NULL,
                                         inum, ap->a_vpp);
        } else {
                error = ENOENT;
        }
        return error;
}

static
int
hammer2_vop_nmkdir(struct vop_nmkdir_args *ap)
{
        hammer2_inode_t *dip;
        hammer2_inode_t *nip;
        struct namecache *ncp;
        const uint8_t *name;
        size_t name_len;
        hammer2_tid_t inum;
        int error;

        dip = VTOI(ap->a_dvp);
        if (dip->pmp->ronly)
                return (EROFS);
        if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
                return (ENOSPC);

        ncp = ap->a_nch->ncp;
        name = ncp->nc_name;
        name_len = ncp->nc_nlen;

        hammer2_pfs_memory_wait(dip, 1);
        hammer2_trans_init(dip->pmp, 0);

        inum = hammer2_trans_newinum(dip->pmp);

        /*
         * Create the actual inode as a hidden file in the iroot, then
         * create the directory entry.  The creation of the actual inode
         * sets its nlinks to 1 which is the value we desire.
         */
        nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
                                   NULL, 0, inum,
                                   inum, 0, 0,
                                   0, &error);
        if (error) {
                error = hammer2_error_to_errno(error);
        } else {
                error = hammer2_dirent_create(dip, name, name_len,
                                              nip->meta.inum, nip->meta.type);
                /* returns UNIX error code */
        }
        if (error) {
                if (nip) {
                        hammer2_inode_unlink_finisher(nip, 0);
                        hammer2_inode_unlock(nip);
                        nip = NULL;
                }
                *ap->a_vpp = NULL;
        } else {
                *ap->a_vpp = hammer2_igetv(nip, &error);
                hammer2_inode_unlock(nip);
        }

        /*
         * Update dip's mtime
         *
         * We can use a shared inode lock and allow the meta.mtime update
         * SMP race.  hammer2_inode_modify() is MPSAFE w/a shared lock.
         */
        if (error == 0) {
                uint64_t mtime;

                hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
                hammer2_update_time(&mtime);
                hammer2_inode_modify(dip);
                dip->meta.mtime = mtime;
                hammer2_inode_unlock(dip);
        }

        hammer2_trans_done(dip->pmp, 1);

        if (error == 0) {
                cache_setunresolved(ap->a_nch);
                cache_setvp(ap->a_nch, *ap->a_vpp);
                hammer2_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK);
        }
        return error;
}

static
int
hammer2_vop_open(struct vop_open_args *ap)
{
        return vop_stdopen(ap);
}

/*
 * hammer2_vop_advlock { vp, id, op, fl, flags }
 */
static
int
hammer2_vop_advlock(struct vop_advlock_args *ap)
{
        hammer2_inode_t *ip = VTOI(ap->a_vp);
        hammer2_off_t size;

        size = ip->meta.size;
        return (lf_advlock(ap, &ip->advlock, size));
}

static
int
hammer2_vop_close(struct vop_close_args *ap)
{
        return vop_stdclose(ap);
}

/*
 * hammer2_vop_nlink { nch, dvp, vp, cred }
 *
 * Create a hardlink from (vp) to {dvp, nch}.
 */
static
int
hammer2_vop_nlink(struct vop_nlink_args *ap)
{
        hammer2_inode_t *tdip;  /* target directory to create link in */
        hammer2_inode_t *ip;    /* inode we are hardlinking to */
        struct namecache *ncp;
        const uint8_t *name;
        size_t name_len;
        int error;

        if (ap->a_dvp->v_mount != ap->a_vp->v_mount)
                return(EXDEV);

        tdip = VTOI(ap->a_dvp);
        if (tdip->pmp->ronly)
                return (EROFS);
        if (hammer2_vfs_enospace(tdip, 0, ap->a_cred) > 1)
                return (ENOSPC);

        ncp = ap->a_nch->ncp;
        name = ncp->nc_name;
        name_len = ncp->nc_nlen;

        /*
         * ip represents the file being hardlinked.  The file could be a
         * normal file or a hardlink target if it has already been hardlinked.
         * (with the new semantics, it will almost always be a hardlink
         * target).
         *
         * Bump nlinks and potentially also create or move the hardlink
         * target in the parent directory common to (ip) and (tdip).  The
         * consolidation code can modify ip->cluster.  The returned cluster
         * is locked.
         */
        ip = VTOI(ap->a_vp);
        KASSERT(ip->pmp, ("ip->pmp is NULL %p %p", ip, ip->pmp));
        hammer2_pfs_memory_wait(ip, 0);
        hammer2_trans_init(ip->pmp, 0);

        /*
         * Target should be an indexed inode or there's no way we will ever
         * be able to find it!
         */
        KKASSERT((ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE) == 0);

        error = 0;

        /*
         * Can return NULL and error == EXDEV if the common parent
         * crosses a directory with the xlink flag set.
         */
        hammer2_inode_lock(tdip, 0);
        hammer2_inode_lock(ip, 0);

        /*
         * Create the directory entry and bump nlinks.
         */
        if (error == 0) {
                error = hammer2_dirent_create(tdip, name, name_len,
                                              ip->meta.inum, ip->meta.type);
                hammer2_inode_modify(ip);
                ++ip->meta.nlinks;
        }
        if (error == 0) {
                /*
                 * Update dip's mtime
                 */
                uint64_t mtime;

                hammer2_update_time(&mtime);
                hammer2_inode_modify(tdip);
                tdip->meta.mtime = mtime;

                cache_setunresolved(ap->a_nch);
                cache_setvp(ap->a_nch, ap->a_vp);
        }
        hammer2_inode_unlock(ip);
        hammer2_inode_unlock(tdip);

        hammer2_trans_done(ip->pmp, 1);
        hammer2_knote(ap->a_vp, NOTE_LINK);
        hammer2_knote(ap->a_dvp, NOTE_WRITE);

        return error;
}

/*
 * hammer2_vop_ncreate { nch, dvp, vpp, cred, vap }
 *
 * The operating system has already ensured that the directory entry
 * does not exist and done all appropriate namespace locking.
 */
static
int
hammer2_vop_ncreate(struct vop_ncreate_args *ap)
{
        hammer2_inode_t *dip;
        hammer2_inode_t *nip;
        struct namecache *ncp;
        const uint8_t *name;
        size_t name_len;
        hammer2_tid_t inum;
        int error;

        dip = VTOI(ap->a_dvp);
        if (dip->pmp->ronly)
                return (EROFS);
        if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
                return (ENOSPC);

        ncp = ap->a_nch->ncp;
        name = ncp->nc_name;
        name_len = ncp->nc_nlen;
        hammer2_pfs_memory_wait(dip, 1);
        hammer2_trans_init(dip->pmp, 0);

        inum = hammer2_trans_newinum(dip->pmp);

        /*
         * Create the actual inode as a hidden file in the iroot, then
         * create the directory entry.  The creation of the actual inode
         * sets its nlinks to 1 which is the value we desire.
         */
        nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
                                   NULL, 0, inum,
                                   inum, 0, 0,
                                   0, &error);

        if (error) {
                error = hammer2_error_to_errno(error);
        } else {
                error = hammer2_dirent_create(dip, name, name_len,
                                              nip->meta.inum, nip->meta.type);
        }
        if (error) {
                if (nip) {
                        hammer2_inode_unlink_finisher(nip, 0);
                        hammer2_inode_unlock(nip);
                        nip = NULL;
                }
                *ap->a_vpp = NULL;
        } else {
                *ap->a_vpp = hammer2_igetv(nip, &error);
                hammer2_inode_unlock(nip);
        }

        /*
         * Update dip's mtime
         */
        if (error == 0) {
                uint64_t mtime;

                hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
                hammer2_update_time(&mtime);
                hammer2_inode_modify(dip);
                dip->meta.mtime = mtime;
                hammer2_inode_unlock(dip);
        }

        hammer2_trans_done(dip->pmp, 1);

        if (error == 0) {
                cache_setunresolved(ap->a_nch);
                cache_setvp(ap->a_nch, *ap->a_vpp);
                hammer2_knote(ap->a_dvp, NOTE_WRITE);
        }
        return error;
}

/*
 * Make a device node (typically a fifo)
 */
static
int
hammer2_vop_nmknod(struct vop_nmknod_args *ap)
{
        hammer2_inode_t *dip;
        hammer2_inode_t *nip;
        struct namecache *ncp;
        const uint8_t *name;
        size_t name_len;
        hammer2_tid_t inum;
        int error;

        dip = VTOI(ap->a_dvp);
        if (dip->pmp->ronly)
                return (EROFS);
        if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
                return (ENOSPC);

        ncp = ap->a_nch->ncp;
        name = ncp->nc_name;
        name_len = ncp->nc_nlen;
        hammer2_pfs_memory_wait(dip, 1);
        hammer2_trans_init(dip->pmp, 0);

        /*
         * Create the device inode and then create the directory entry.
         */
        inum = hammer2_trans_newinum(dip->pmp);
        nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
                                   NULL, 0, inum,
                                   inum, 0, 0,
                                   0, &error);
        if (error == 0) {
                error = hammer2_dirent_create(dip, name, name_len,
                                              nip->meta.inum, nip->meta.type);
        }
        if (error) {
                if (nip) {
                        hammer2_inode_unlink_finisher(nip, 0);
                        hammer2_inode_unlock(nip);
                        nip = NULL;
                }
                *ap->a_vpp = NULL;
        } else {
                *ap->a_vpp = hammer2_igetv(nip, &error);
                hammer2_inode_unlock(nip);
        }

        /*
         * Update dip's mtime
         */
        if (error == 0) {
                uint64_t mtime;

                hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
                hammer2_update_time(&mtime);
                hammer2_inode_modify(dip);
                dip->meta.mtime = mtime;
                hammer2_inode_unlock(dip);
        }

        hammer2_trans_done(dip->pmp, 1);

        if (error == 0) {
                cache_setunresolved(ap->a_nch);
                cache_setvp(ap->a_nch, *ap->a_vpp);
                hammer2_knote(ap->a_dvp, NOTE_WRITE);
        }
        return error;
}

/*
 * hammer2_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
 */
static
int
hammer2_vop_nsymlink(struct vop_nsymlink_args *ap)
{
        hammer2_inode_t *dip;
        hammer2_inode_t *nip;
        struct namecache *ncp;
        const uint8_t *name;
        size_t name_len;
        hammer2_tid_t inum;
        int error;

        dip = VTOI(ap->a_dvp);
        if (dip->pmp->ronly)
                return (EROFS);
        if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
                return (ENOSPC);

        ncp = ap->a_nch->ncp;
        name = ncp->nc_name;
        name_len = ncp->nc_nlen;
        hammer2_pfs_memory_wait(dip, 1);
        hammer2_trans_init(dip->pmp, 0);

        ap->a_vap->va_type = VLNK;      /* enforce type */

        /*
         * Create the softlink as an inode and then create the directory
         * entry.
         */
        inum = hammer2_trans_newinum(dip->pmp);

        nip = hammer2_inode_create(dip->pmp->iroot, dip, ap->a_vap, ap->a_cred,
                                   NULL, 0, inum,
                                   inum, 0, 0,
                                   0, &error);
        if (error == 0) {
                error = hammer2_dirent_create(dip, name, name_len,
                                              nip->meta.inum, nip->meta.type);
        }
        if (error) {
                if (nip) {
                        hammer2_inode_unlink_finisher(nip, 0);
                        hammer2_inode_unlock(nip);
                        nip = NULL;
                }
                *ap->a_vpp = NULL;
                hammer2_trans_done(dip->pmp, 1);
                return error;
        }
        *ap->a_vpp = hammer2_igetv(nip, &error);

        /*
         * Build the softlink (~like file data) and finalize the namecache.
         */
        if (error == 0) {
                size_t bytes;
                struct uio auio;
                struct iovec aiov;

                bytes = strlen(ap->a_target);

                hammer2_inode_unlock(nip);
                bzero(&auio, sizeof(auio));
                bzero(&aiov, sizeof(aiov));
                auio.uio_iov = &aiov;
                auio.uio_segflg = UIO_SYSSPACE;
                auio.uio_rw = UIO_WRITE;
                auio.uio_resid = bytes;
                auio.uio_iovcnt = 1;
                auio.uio_td = curthread;
                aiov.iov_base = ap->a_target;
                aiov.iov_len = bytes;
                error = hammer2_write_file(nip, &auio, IO_APPEND, 0);
                /* XXX handle error */
                error = 0;
        } else {
                hammer2_inode_unlock(nip);
        }

        /*
         * Update dip's mtime
         */
        if (error == 0) {
                uint64_t mtime;

                hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
                hammer2_update_time(&mtime);
                hammer2_inode_modify(dip);
                dip->meta.mtime = mtime;
                hammer2_inode_unlock(dip);
        }

        hammer2_trans_done(dip->pmp, 1);

        /*
         * Finalize namecache
         */
        if (error == 0) {
                cache_setunresolved(ap->a_nch);
                cache_setvp(ap->a_nch, *ap->a_vpp);
                hammer2_knote(ap->a_dvp, NOTE_WRITE);
        }
        return error;
}

/*
 * hammer2_vop_nremove { nch, dvp, cred }
 */
static
int
hammer2_vop_nremove(struct vop_nremove_args *ap)
{
        hammer2_xop_unlink_t *xop;
        hammer2_inode_t *dip;
        hammer2_inode_t *ip;
        struct namecache *ncp;
        int error;
        int isopen;

        dip = VTOI(ap->a_dvp);
        if (dip->pmp->ronly)
                return (EROFS);
#if 0
        /* allow removals, except user to also bulkfree */
        if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
                return (ENOSPC);
#endif

        ncp = ap->a_nch->ncp;

        hammer2_pfs_memory_wait(dip, 1);
        hammer2_trans_init(dip->pmp, 0);
        hammer2_inode_lock(dip, 0);

        /*
         * The unlink XOP unlinks the path from the directory and
         * locates and returns the cluster associated with the real inode.
         * We have to handle nlinks here on the frontend.
         */
        xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);
        hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);

        /*
         * The namecache entry is locked so nobody can use this namespace.
         * Calculate isopen to determine if this namespace has an open vp
         * associated with it and resolve the vp only if it does.
         *
         * We try to avoid resolving the vnode if nobody has it open, but
         * note that the test is via this namespace only.
         */
        isopen = cache_isopen(ap->a_nch);
        xop->isdir = 0;
        xop->dopermanent = 0;
        hammer2_xop_start(&xop->head, hammer2_xop_unlink);

        /*
         * Collect the real inode and adjust nlinks, destroy the real
         * inode if nlinks transitions to 0 and it was the real inode
         * (else it has already been removed).
         */
        error = hammer2_xop_collect(&xop->head, 0);
        error = hammer2_error_to_errno(error);
        hammer2_inode_unlock(dip);

        if (error == 0) {
                ip = hammer2_inode_get(dip->pmp, dip, &xop->head, -1);
                hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
                if (ip) {
                        hammer2_inode_unlink_finisher(ip, isopen);
                        hammer2_inode_unlock(ip);
                }
        } else {
                hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
        }

        /*
         * Update dip's mtime
         */
        if (error == 0) {
                uint64_t mtime;

                hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
                hammer2_update_time(&mtime);
                hammer2_inode_modify(dip);
                dip->meta.mtime = mtime;
                hammer2_inode_unlock(dip);
        }

        hammer2_trans_done(dip->pmp, 1);
        if (error == 0) {
                cache_unlink(ap->a_nch);
                hammer2_knote(ap->a_dvp, NOTE_WRITE);
        }
        return (error);
}

/*
 * hammer2_vop_nrmdir { nch, dvp, cred }
 */
static
int
hammer2_vop_nrmdir(struct vop_nrmdir_args *ap)
{
        hammer2_xop_unlink_t *xop;
        hammer2_inode_t *dip;
        hammer2_inode_t *ip;
        struct namecache *ncp;
        int isopen;
        int error;

        dip = VTOI(ap->a_dvp);
        if (dip->pmp->ronly)
                return (EROFS);
#if 0
        /* allow removals, except user to also bulkfree */
        if (hammer2_vfs_enospace(dip, 0, ap->a_cred) > 1)
                return (ENOSPC);
#endif

        hammer2_pfs_memory_wait(dip, 1);
        hammer2_trans_init(dip->pmp, 0);
        hammer2_inode_lock(dip, 0);

        xop = hammer2_xop_alloc(dip, HAMMER2_XOP_MODIFYING);

        ncp = ap->a_nch->ncp;
        hammer2_xop_setname(&xop->head, ncp->nc_name, ncp->nc_nlen);
        isopen = cache_isopen(ap->a_nch);
        xop->isdir = 1;
        xop->dopermanent = 0;
        hammer2_xop_start(&xop->head, hammer2_xop_unlink);

        /*
         * Collect the real inode and adjust nlinks, destroy the real
         * inode if nlinks transitions to 0 and it was the real inode
         * (else it has already been removed).
         */
        error = hammer2_xop_collect(&xop->head, 0);
        error = hammer2_error_to_errno(error);
        hammer2_inode_unlock(dip);

        if (error == 0) {
                ip = hammer2_inode_get(dip->pmp, dip, &xop->head, -1);
                hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
                if (ip) {
                        hammer2_inode_unlink_finisher(ip, isopen);
                        hammer2_inode_unlock(ip);
                }
        } else {
                hammer2_xop_retire(&xop->head, HAMMER2_XOPMASK_VOP);
        }

        /*
         * Update dip's mtime
         */
        if (error == 0) {
                uint64_t mtime;

                hammer2_inode_lock(dip, HAMMER2_RESOLVE_SHARED);
                hammer2_update_time(&mtime);
                hammer2_inode_modify(dip);
                dip->meta.mtime = mtime;
                hammer2_inode_unlock(dip);
        }

        hammer2_trans_done(dip->pmp, 1);
        if (error == 0) {
                cache_unlink(ap->a_nch);
                hammer2_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK);
        }
        return (error);
}

/*
 * hammer2_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
 */
static
int
hammer2_vop_nrename(struct vop_nrename_args *ap)
{
        struct namecache *fncp;
        struct namecache *tncp;
        hammer2_inode_t *fdip;  /* source directory */
        hammer2_inode_t *tdip;  /* target directory */
        hammer2_inode_t *ip;    /* file being renamed */
        hammer2_inode_t *tip;   /* replaced target during rename or NULL */
        const uint8_t *fname;
        size_t fname_len;
        const uint8_t *tname;
        size_t tname_len;
        int error;
        int update_tdip;
        int update_fdip;
        hammer2_key_t tlhc;

        if (ap->a_fdvp->v_mount != ap->a_tdvp->v_mount)
                return(EXDEV);
        if (ap->a_fdvp->v_mount != ap->a_fnch->ncp->nc_vp->v_mount)
                return(EXDEV);

        fdip = VTOI(ap->a_fdvp);        /* source directory */
        tdip = VTOI(ap->a_tdvp);        /* target directory */

        if (fdip->pmp->ronly)
                return (EROFS);
        if (hammer2_vfs_enospace(fdip, 0, ap->a_cred) > 1)
                return (ENOSPC);

        fncp = ap->a_fnch->ncp;         /* entry name in source */
        fname = fncp->nc_name;
        fname_len = fncp->nc_nlen;

        tncp = ap->a_tnch->ncp;         /* entry name in target */
        tname = tncp->nc_name;
        tname_len = tncp->nc_nlen;

        hammer2_pfs_memory_wait(tdip, 0);
        hammer2_trans_init(tdip->pmp, 0);

        update_tdip = 0;
        update_fdip = 0;

        ip = VTOI(fncp->nc_vp);
        hammer2_inode_ref(ip);          /* extra ref */

        /*
         * Lookup the target name to determine if a directory entry
         * is being overwritten.  We only hold related inode locks
         * temporarily, the operating system is expected to protect
         * against rename races.
         */
        tip = tncp->nc_vp ? VTOI(tncp->nc_vp) : NULL;
        if (tip)
                hammer2_inode_ref(tip); /* extra ref */

        /*
         * Can return NULL and error == EXDEV if the common parent
         * crosses a directory with the xlink flag set.
         *
         * For now try to avoid deadlocks with a simple pointer address
         * test.  (tip) can be NULL.
         */
        error = 0;
        if (fdip <= tdip) {
                hammer2_inode_lock(fdip, 0);
                hammer2_inode_lock(tdip, 0);
        } else {
                hammer2_inode_lock(tdip, 0);
                hammer2_inode_lock(fdip, 0);
        }
        if (tip) {
                if (ip <= tip) {
                        hammer2_inode_lock(ip, 0);
                        hammer2_inode_lock(tip, 0);
                } else {
                        hammer2_inode_lock(tip, 0);
                        hammer2_inode_lock(ip, 0);
                }
        } else {
                hammer2_inode_lock(ip, 0);
        }

#if 0
        /*
         * Delete the target namespace.
         *
         * REMOVED - NOW FOLDED INTO XOP_NRENAME OPERATION
         */
        {
                hammer2_xop_unlink_t *xop2;
                hammer2_inode_t *tip;
                int isopen;

                /*
                 * The unlink XOP unlinks the path from the directory and
                 * locates and returns the cluster associated with the real
                 * inode.  We have to handle nlinks here on the frontend.
                 */
                xop2 = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
                hammer2_xop_setname(&xop2->head, tname, tname_len);
                isopen = cache_isopen(ap->a_tnch);
                xop2->isdir = -1;
                xop2->dopermanent = 0;
                hammer2_xop_start(&xop2->head, hammer2_xop_unlink);

                /*
                 * Collect the real inode and adjust nlinks, destroy the real
                 * inode if nlinks transitions to 0 and it was the real inode
                 * (else it has already been removed).
                 */
                tnch_error = hammer2_xop_collect(&xop2->head, 0);
                tnch_error = hammer2_error_to_errno(tnch_error);
                /* hammer2_inode_unlock(tdip); */

                if (tnch_error == 0) {
                        tip = hammer2_inode_get(tdip->pmp, NULL,
                                                &xop2->head, -1);
                        hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
                        if (tip) {
                                hammer2_inode_unlink_finisher(tip, isopen);
                                hammer2_inode_unlock(tip);
                        }
                } else {
                        hammer2_xop_retire(&xop2->head, HAMMER2_XOPMASK_VOP);
                }
                /* hammer2_inode_lock(tdip, 0); */

                if (tnch_error && tnch_error != ENOENT) {
                        error = tnch_error;
                        goto done2;
                }
                update_tdip = 1;
        }
#endif

        /*
         * Resolve the collision space for (tdip, tname, tname_len)
         *
         * tdip must be held exclusively locked to prevent races since
         * multiple filenames can end up in the same collision space.
         */
        {
                hammer2_xop_scanlhc_t *sxop;
                hammer2_tid_t lhcbase;

                tlhc = hammer2_dirhash(tname, tname_len);
                lhcbase = tlhc;
                sxop = hammer2_xop_alloc(tdip, HAMMER2_XOP_MODIFYING);
                sxop->lhc = tlhc;
                hammer2_xop_start(&sxop->head, hammer2_xop_scanlhc);
                while ((error = hammer2_xop_collect(&sxop->head, 0)) == 0) {
                        if (tlhc != sxop->head.cluster.focus->bref.key)
                                break;
                        ++tlhc;
                }
                error = hammer2_error_to_errno(error);
                hammer2_xop_retire(&sxop->head, HAMMER2_XOPMASK_VOP);

                if (error) {
                        if (error != ENOENT)
                                goto done2;
                        ++tlhc;
                        error = 0;
                }
                if ((lhcbase ^ tlhc) & ~HAMMER2_DIRHASH_LOMASK) {
                        error = ENOSPC;
                        goto done2;
                }
        }

        /*
         * Ready to go, issue the rename to the backend.  Note that meta-data
         * updates to the related inodes occur separately from the rename
         * operation.
         *
         * NOTE: While it is not necessary to update ip->meta.name*, doing
         *       so aids catastrophic recovery and debugging.
         */
        if (error == 0) {
                hammer2_xop_nrename_t *xop4;

                xop4 = hammer2_xop_alloc(fdip, HAMMER2_XOP_MODIFYING);
                xop4->lhc = tlhc;
                xop4->ip_key = ip->meta.name_key;
                hammer2_xop_setip2(&xop4->head, ip);
                hammer2_xop_setip3(&xop4->head, tdip);
                hammer2_xop_setname(&xop4->head, fname, fname_len);
                hammer2_xop_setname2(&xop4->head, tname, tname_len);
                hammer2_xop_start(&xop4->head, hammer2_xop_nrename);

                error = hammer2_xop_collect(&xop4->head, 0);
                error = hammer2_error_to_errno(error);
                hammer2_xop_retire(&xop4->head, HAMMER2_XOPMASK_VOP);

                if (error == ENOENT)
                        error = 0;

                /*
                 * Update inode meta-data.
                 *
                 * WARNING!  The in-memory inode (ip) structure does not
                 *           maintain a copy of the inode's filename buffer.
                 */
                if (error == 0 &&
                    (ip->meta.name_key & HAMMER2_DIRHASH_VISIBLE)) {
                        hammer2_inode_modify(ip);
                        ip->meta.name_len = tname_len;
                        ip->meta.name_key = tlhc;
                }
                if (error == 0) {
                        hammer2_inode_modify(ip);
                        ip->meta.iparent = tdip->meta.inum;
                }
                update_fdip = 1;
                update_tdip = 1;
        }

done2:
        /*
         * If no error, the backend has replaced the target directory entry.
         * We must adjust nlinks on the original replace target if it exists.
         */
        if (error == 0 && tip) {
                int isopen;

                isopen = cache_isopen(ap->a_tnch);
                hammer2_inode_unlink_finisher(tip, isopen);
        }

        /*
         * Update directory mtimes to represent the something changed.
         */
        if (update_fdip || update_tdip) {
                uint64_t mtime;

                hammer2_update_time(&mtime);
                if (update_fdip) {
                        hammer2_inode_modify(fdip);
                        fdip->meta.mtime = mtime;
                }
                if (update_tdip) {
                        hammer2_inode_modify(tdip);
                        tdip->meta.mtime = mtime;
                }
        }
        if (tip) {
                hammer2_inode_unlock(tip);
                hammer2_inode_drop(tip);
        }
        hammer2_inode_unlock(ip);
        hammer2_inode_unlock(tdip);
        hammer2_inode_unlock(fdip);
        hammer2_inode_drop(ip);
        hammer2_trans_done(tdip->pmp, 1);

        /*
         * Issue the namecache update after unlocking all the internal
         * hammer2 structures, otherwise we might deadlock.
         *
         * WARNING! The target namespace must be updated atomically,
         *          and we depend on cache_rename() to handle that for
         *          us.  Do not do a separate cache_unlink() because
         *          that leaves a small window of opportunity for other
         *          threads to allocate the target namespace before we
         *          manage to complete our rename.
         *
         * WARNING! cache_rename() (and cache_unlink()) will properly
         *          set VREF_FINALIZE on any attached vnode.  Do not
         *          call cache_setunresolved() manually before-hand as
         *          this will prevent the flag from being set later via
         *          cache_rename().  If VREF_FINALIZE is not properly set
         *          and the inode is no longer in the topology, related
         *          chains can remain dirty indefinitely.
         */
        if (error == 0 && tip) {
                /*cache_unlink(ap->a_tnch); see above */
                /*cache_setunresolved(ap->a_tnch); see above */
        }
        if (error == 0) {
                cache_rename(ap->a_fnch, ap->a_tnch);
                hammer2_knote(ap->a_fdvp, NOTE_WRITE);
                hammer2_knote(ap->a_tdvp, NOTE_WRITE);
                hammer2_knote(fncp->nc_vp, NOTE_RENAME);
        }

        return (error);
}

/*
 * hammer2_vop_ioctl { vp, command, data, fflag, cred }
 */
static
int
hammer2_vop_ioctl(struct vop_ioctl_args *ap)
{
        hammer2_inode_t *ip;
        int error;

        ip = VTOI(ap->a_vp);

        error = hammer2_ioctl(ip, ap->a_command, (void *)ap->a_data,
                              ap->a_fflag, ap->a_cred);
        return (error);
}

static
int
hammer2_vop_mountctl(struct vop_mountctl_args *ap)
{
        struct mount *mp;
        hammer2_pfs_t *pmp;
        int rc;

        switch (ap->a_op) {
        case (MOUNTCTL_SET_EXPORT):
                mp = ap->a_head.a_ops->head.vv_mount;
                pmp = MPTOPMP(mp);

                if (ap->a_ctllen != sizeof(struct export_args))
                        rc = (EINVAL);
                else
                        rc = vfs_export(mp, &pmp->export,
                                        (const struct export_args *)ap->a_ctl);
                break;
        default:
                rc = vop_stdmountctl(ap);
                break;
        }
        return (rc);
}

/*
 * KQFILTER
 */
static void filt_hammer2detach(struct knote *kn);
static int filt_hammer2read(struct knote *kn, long hint);
static int filt_hammer2write(struct knote *kn, long hint);
static int filt_hammer2vnode(struct knote *kn, long hint);

static struct filterops hammer2read_filtops =
        { FILTEROP_ISFD | FILTEROP_MPSAFE,
          NULL, filt_hammer2detach, filt_hammer2read };
static struct filterops hammer2write_filtops =
        { FILTEROP_ISFD | FILTEROP_MPSAFE,
          NULL, filt_hammer2detach, filt_hammer2write };
static struct filterops hammer2vnode_filtops =
        { FILTEROP_ISFD | FILTEROP_MPSAFE,
          NULL, filt_hammer2detach, filt_hammer2vnode };

static
int
hammer2_vop_kqfilter(struct vop_kqfilter_args *ap)
{
        struct vnode *vp = ap->a_vp;
        struct knote *kn = ap->a_kn;

        switch (kn->kn_filter) {
        case EVFILT_READ:
                kn->kn_fop = &hammer2read_filtops;
                break;
        case EVFILT_WRITE:
                kn->kn_fop = &hammer2write_filtops;
                break;
        case EVFILT_VNODE:
                kn->kn_fop = &hammer2vnode_filtops;
                break;
        default:
                return (EOPNOTSUPP);
        }

        kn->kn_hook = (caddr_t)vp;

        knote_insert(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);

        return(0);
}

static void
filt_hammer2detach(struct knote *kn)
{
        struct vnode *vp = (void *)kn->kn_hook;

        knote_remove(&vp->v_pollinfo.vpi_kqinfo.ki_note, kn);
}

static int
filt_hammer2read(struct knote *kn, long hint)
{
        struct vnode *vp = (void *)kn->kn_hook;
        hammer2_inode_t *ip = VTOI(vp);
        off_t off;

        if (hint == NOTE_REVOKE) {
                kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
                return(1);
        }
        off = ip->meta.size - kn->kn_fp->f_offset;
        kn->kn_data = (off < INTPTR_MAX) ? off : INTPTR_MAX;
        if (kn->kn_sfflags & NOTE_OLDAPI)
                return(1);
        return (kn->kn_data != 0);
}


static int
filt_hammer2write(struct knote *kn, long hint)
{
        if (hint == NOTE_REVOKE)
                kn->kn_flags |= (EV_EOF | EV_NODATA | EV_ONESHOT);
        kn->kn_data = 0;
        return (1);
}

static int
filt_hammer2vnode(struct knote *kn, long hint)
{
        if (kn->kn_sfflags & hint)
                kn->kn_fflags |= hint;
        if (hint == NOTE_REVOKE) {
                kn->kn_flags |= (EV_EOF | EV_NODATA);
                return (1);
        }
        return (kn->kn_fflags != 0);
}

/*
 * FIFO VOPS
 */
static
int
hammer2_vop_markatime(struct vop_markatime_args *ap)
{
        hammer2_inode_t *ip;
        struct vnode *vp;

        vp = ap->a_vp;
        ip = VTOI(vp);

        if (ip->pmp->ronly)
                return (EROFS);
        return(0);
}

static
int
hammer2_vop_fifokqfilter(struct vop_kqfilter_args *ap)
{
        int error;

        error = VOCALL(&fifo_vnode_vops, &ap->a_head);
        if (error)
                error = hammer2_vop_kqfilter(ap);
        return(error);
}

/*
 * VOPS vector
 */
struct vop_ops hammer2_vnode_vops = {
        .vop_default    = vop_defaultop,
        .vop_fsync      = hammer2_vop_fsync,
        .vop_getpages   = vop_stdgetpages,
        .vop_putpages   = vop_stdputpages,
        .vop_access     = hammer2_vop_access,
        .vop_advlock    = hammer2_vop_advlock,
        .vop_close      = hammer2_vop_close,
        .vop_nlink      = hammer2_vop_nlink,
        .vop_ncreate    = hammer2_vop_ncreate,
        .vop_nsymlink   = hammer2_vop_nsymlink,
        .vop_nremove    = hammer2_vop_nremove,
        .vop_nrmdir     = hammer2_vop_nrmdir,
        .vop_nrename    = hammer2_vop_nrename,
        .vop_getattr    = hammer2_vop_getattr,
        .vop_setattr    = hammer2_vop_setattr,
        .vop_readdir    = hammer2_vop_readdir,
        .vop_readlink   = hammer2_vop_readlink,
        .vop_read       = hammer2_vop_read,
        .vop_write      = hammer2_vop_write,
        .vop_open       = hammer2_vop_open,
        .vop_inactive   = hammer2_vop_inactive,
        .vop_reclaim    = hammer2_vop_reclaim,
        .vop_nresolve   = hammer2_vop_nresolve,
        .vop_nlookupdotdot = hammer2_vop_nlookupdotdot,
        .vop_nmkdir     = hammer2_vop_nmkdir,
        .vop_nmknod     = hammer2_vop_nmknod,
        .vop_ioctl      = hammer2_vop_ioctl,
        .vop_mountctl   = hammer2_vop_mountctl,
        .vop_bmap       = hammer2_vop_bmap,
        .vop_strategy   = hammer2_vop_strategy,
        .vop_kqfilter   = hammer2_vop_kqfilter
};

struct vop_ops hammer2_spec_vops = {
        .vop_default =          vop_defaultop,
        .vop_fsync =            hammer2_vop_fsync,
        .vop_read =             vop_stdnoread,
        .vop_write =            vop_stdnowrite,
        .vop_access =           hammer2_vop_access,
        .vop_close =            hammer2_vop_close,
        .vop_markatime =        hammer2_vop_markatime,
        .vop_getattr =          hammer2_vop_getattr,
        .vop_inactive =         hammer2_vop_inactive,
        .vop_reclaim =          hammer2_vop_reclaim,
        .vop_setattr =          hammer2_vop_setattr
};

struct vop_ops hammer2_fifo_vops = {
        .vop_default =          fifo_vnoperate,
        .vop_fsync =            hammer2_vop_fsync,
#if 0
        .vop_read =             hammer2_vop_fiforead,
        .vop_write =            hammer2_vop_fifowrite,
#endif
        .vop_access =           hammer2_vop_access,
#if 0
        .vop_close =            hammer2_vop_fifoclose,
#endif
        .vop_markatime =        hammer2_vop_markatime,
        .vop_getattr =          hammer2_vop_getattr,
        .vop_inactive =         hammer2_vop_inactive,
        .vop_reclaim =          hammer2_vop_reclaim,
        .vop_setattr =          hammer2_vop_setattr,
        .vop_kqfilter =         hammer2_vop_fifokqfilter
};