HAMMER - Rework write pipelining

[dragonfly.git] / sys / vfs / hammer / hammer_vnops.c

/*
 * Copyright (c) 2007-2008 The DragonFly Project.  All rights reserved.
 * 
 * This code is derived from software contributed to The DragonFly Project
 * by Matthew Dillon <dillon@backplane.com>
 * 
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in
 *    the documentation and/or other materials provided with the
 *    distribution.
 * 3. Neither the name of The DragonFly Project nor the names of its
 *    contributors may be used to endorse or promote products derived
 *    from this software without specific, prior written permission.
 * 
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
 * SUCH DAMAGE.
 * 
 * $DragonFly: src/sys/vfs/hammer/hammer_vnops.c,v 1.102 2008/10/16 17:24:16 dillon Exp $
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/kernel.h>
#include <sys/fcntl.h>
#include <sys/namecache.h>
#include <sys/vnode.h>
#include <sys/lockf.h>
#include <sys/event.h>
#include <sys/stat.h>
#include <sys/dirent.h>
#include <sys/file.h>
#include <vm/vm_extern.h>
#include <vfs/fifofs/fifo.h>
#include "hammer.h"

/*
 * USERFS VNOPS
 */
/*static int hammer_vop_vnoperate(struct vop_generic_args *);*/
static int hammer_vop_fsync(struct vop_fsync_args *);
static int hammer_vop_read(struct vop_read_args *);
static int hammer_vop_write(struct vop_write_args *);
static int hammer_vop_access(struct vop_access_args *);
static int hammer_vop_advlock(struct vop_advlock_args *);
static int hammer_vop_close(struct vop_close_args *);
static int hammer_vop_ncreate(struct vop_ncreate_args *);
static int hammer_vop_getattr(struct vop_getattr_args *);
static int hammer_vop_nresolve(struct vop_nresolve_args *);
static int hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *);
static int hammer_vop_nlink(struct vop_nlink_args *);
static int hammer_vop_nmkdir(struct vop_nmkdir_args *);
static int hammer_vop_nmknod(struct vop_nmknod_args *);
static int hammer_vop_open(struct vop_open_args *);
static int hammer_vop_print(struct vop_print_args *);
static int hammer_vop_readdir(struct vop_readdir_args *);
static int hammer_vop_readlink(struct vop_readlink_args *);
static int hammer_vop_nremove(struct vop_nremove_args *);
static int hammer_vop_nrename(struct vop_nrename_args *);
static int hammer_vop_nrmdir(struct vop_nrmdir_args *);
static int hammer_vop_markatime(struct vop_markatime_args *);
static int hammer_vop_setattr(struct vop_setattr_args *);
static int hammer_vop_strategy(struct vop_strategy_args *);
static int hammer_vop_bmap(struct vop_bmap_args *ap);
static int hammer_vop_nsymlink(struct vop_nsymlink_args *);
static int hammer_vop_nwhiteout(struct vop_nwhiteout_args *);
static int hammer_vop_ioctl(struct vop_ioctl_args *);
static int hammer_vop_mountctl(struct vop_mountctl_args *);
static int hammer_vop_kqfilter (struct vop_kqfilter_args *);

static int hammer_vop_fifoclose (struct vop_close_args *);
static int hammer_vop_fiforead (struct vop_read_args *);
static int hammer_vop_fifowrite (struct vop_write_args *);
static int hammer_vop_fifokqfilter (struct vop_kqfilter_args *);

struct vop_ops hammer_vnode_vops = {
        .vop_default =          vop_defaultop,
        .vop_fsync =            hammer_vop_fsync,
        .vop_getpages =         vop_stdgetpages,
        .vop_putpages =         vop_stdputpages,
        .vop_read =             hammer_vop_read,
        .vop_write =            hammer_vop_write,
        .vop_access =           hammer_vop_access,
        .vop_advlock =          hammer_vop_advlock,
        .vop_close =            hammer_vop_close,
        .vop_ncreate =          hammer_vop_ncreate,
        .vop_getattr =          hammer_vop_getattr,
        .vop_inactive =         hammer_vop_inactive,
        .vop_reclaim =          hammer_vop_reclaim,
        .vop_nresolve =         hammer_vop_nresolve,
        .vop_nlookupdotdot =    hammer_vop_nlookupdotdot,
        .vop_nlink =            hammer_vop_nlink,
        .vop_nmkdir =           hammer_vop_nmkdir,
        .vop_nmknod =           hammer_vop_nmknod,
        .vop_open =             hammer_vop_open,
        .vop_pathconf =         vop_stdpathconf,
        .vop_print =            hammer_vop_print,
        .vop_readdir =          hammer_vop_readdir,
        .vop_readlink =         hammer_vop_readlink,
        .vop_nremove =          hammer_vop_nremove,
        .vop_nrename =          hammer_vop_nrename,
        .vop_nrmdir =           hammer_vop_nrmdir,
        .vop_markatime =        hammer_vop_markatime,
        .vop_setattr =          hammer_vop_setattr,
        .vop_bmap =             hammer_vop_bmap,
        .vop_strategy =         hammer_vop_strategy,
        .vop_nsymlink =         hammer_vop_nsymlink,
        .vop_nwhiteout =        hammer_vop_nwhiteout,
        .vop_ioctl =            hammer_vop_ioctl,
        .vop_mountctl =         hammer_vop_mountctl,
        .vop_kqfilter =         hammer_vop_kqfilter
};

struct vop_ops hammer_spec_vops = {
        .vop_default =          vop_defaultop,
        .vop_fsync =            hammer_vop_fsync,
        .vop_read =             vop_stdnoread,
        .vop_write =            vop_stdnowrite,
        .vop_access =           hammer_vop_access,
        .vop_close =            hammer_vop_close,
        .vop_markatime =        hammer_vop_markatime,
        .vop_getattr =          hammer_vop_getattr,
        .vop_inactive =         hammer_vop_inactive,
        .vop_reclaim =          hammer_vop_reclaim,
        .vop_setattr =          hammer_vop_setattr
};

struct vop_ops hammer_fifo_vops = {
        .vop_default =          fifo_vnoperate,
        .vop_fsync =            hammer_vop_fsync,
        .vop_read =             hammer_vop_fiforead,
        .vop_write =            hammer_vop_fifowrite,
        .vop_access =           hammer_vop_access,
        .vop_close =            hammer_vop_fifoclose,
        .vop_markatime =        hammer_vop_markatime,
        .vop_getattr =          hammer_vop_getattr,
        .vop_inactive =         hammer_vop_inactive,
        .vop_reclaim =          hammer_vop_reclaim,
        .vop_setattr =          hammer_vop_setattr,
        .vop_kqfilter =         hammer_vop_fifokqfilter
};

static __inline
void
hammer_knote(struct vnode *vp, int flags)
{
        if (flags)
                KNOTE(&vp->v_pollinfo.vpi_selinfo.si_note, flags);
}

#ifdef DEBUG_TRUNCATE
struct hammer_inode *HammerTruncIp;
#endif

static int hammer_dounlink(hammer_transaction_t trans, struct nchandle *nch,
                           struct vnode *dvp, struct ucred *cred,
                           int flags, int isdir);
static int hammer_vop_strategy_read(struct vop_strategy_args *ap);
static int hammer_vop_strategy_write(struct vop_strategy_args *ap);

#if 0
static
int
hammer_vop_vnoperate(struct vop_generic_args *)
{
        return (VOCALL(&hammer_vnode_vops, ap));
}
#endif

/*
 * hammer_vop_fsync { vp, waitfor }
 *
 * fsync() an inode to disk and wait for it to be completely committed
 * such that the information would not be undone if a crash occured after
 * return.
 */
static
int
hammer_vop_fsync(struct vop_fsync_args *ap)
{
        hammer_inode_t ip = VTOI(ap->a_vp);

        ++hammer_count_fsyncs;
        vfsync(ap->a_vp, ap->a_waitfor, 1, NULL, NULL);
        hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
        if (ap->a_waitfor == MNT_WAIT) {
                vn_unlock(ap->a_vp);
                hammer_wait_inode(ip);
                vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
        }
        return (ip->error);
}

/*
 * hammer_vop_read { vp, uio, ioflag, cred }
 *
 * MPALMOSTSAFE
 */
static
int
hammer_vop_read(struct vop_read_args *ap)
{
        struct hammer_transaction trans;
        hammer_inode_t ip;
        off_t offset;
        struct buf *bp;
        struct uio *uio;
        int error;
        int n;
        int seqcount;
        int ioseqcount;
        int blksize;
        int got_mplock;
        int bigread;

        if (ap->a_vp->v_type != VREG)
                return (EINVAL);
        ip = VTOI(ap->a_vp);
        error = 0;
        uio = ap->a_uio;

        /*
         * Allow the UIO's size to override the sequential heuristic.
         */
        blksize = hammer_blocksize(uio->uio_offset);
        seqcount = (uio->uio_resid + (blksize - 1)) / blksize;
        ioseqcount = ap->a_ioflag >> 16;
        if (seqcount < ioseqcount)
                seqcount = ioseqcount;

        /*
         * Temporary hack until more of HAMMER can be made MPSAFE.
         */
#ifdef SMP
        if (curthread->td_mpcount) {
                got_mplock = -1;
                hammer_start_transaction(&trans, ip->hmp);
        } else {
                got_mplock = 0;
        }
#else
        hammer_start_transaction(&trans, ip->hmp);
        got_mplock = -1;
#endif

        /*
         * If reading or writing a huge amount of data we have to break
         * atomicy and allow the operation to be interrupted by a signal
         * or it can DOS the machine.
         */
        bigread = (uio->uio_resid > 100 * 1024 * 1024);

        /*
         * Access the data typically in HAMMER_BUFSIZE blocks via the
         * buffer cache, but HAMMER may use a variable block size based
         * on the offset.
         *
         * XXX Temporary hack, delay the start transaction while we remain
         *     MPSAFE.  NOTE: ino_data.size cannot change while vnode is
         *     locked-shared.
         */
        while (uio->uio_resid > 0 && uio->uio_offset < ip->ino_data.size) {
                int64_t base_offset;
                int64_t file_limit;

                blksize = hammer_blocksize(uio->uio_offset);
                offset = (int)uio->uio_offset & (blksize - 1);
                base_offset = uio->uio_offset - offset;

                if (bigread && (error = hammer_signal_check(ip->hmp)) != 0)
                        break;

                /*
                 * MPSAFE
                 */
                bp = getcacheblk(ap->a_vp, base_offset);
                if (bp) {
                        error = 0;
                        goto skip;
                }

                /*
                 * MPUNSAFE
                 */
                if (got_mplock == 0) {
                        got_mplock = 1;
                        get_mplock();
                        hammer_start_transaction(&trans, ip->hmp);
                }

                if (hammer_cluster_enable) {
                        /*
                         * Use file_limit to prevent cluster_read() from
                         * creating buffers of the wrong block size past
                         * the demarc.
                         */
                        file_limit = ip->ino_data.size;
                        if (base_offset < HAMMER_XDEMARC &&
                            file_limit > HAMMER_XDEMARC) {
                                file_limit = HAMMER_XDEMARC;
                        }
                        error = cluster_read(ap->a_vp,
                                             file_limit, base_offset,
                                             blksize, MAXPHYS,
                                             seqcount, &bp);
                } else {
                        error = bread(ap->a_vp, base_offset, blksize, &bp);
                }
                if (error) {
                        kprintf("error %d\n", error);
                        brelse(bp);
                        break;
                }
skip:

                /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
                n = blksize - offset;
                if (n > uio->uio_resid)
                        n = uio->uio_resid;
                if (n > ip->ino_data.size - uio->uio_offset)
                        n = (int)(ip->ino_data.size - uio->uio_offset);
                error = uiomove((char *)bp->b_data + offset, n, uio);

                /* data has a lower priority then meta-data */
                bp->b_flags |= B_AGE;
                bqrelse(bp);
                if (error)
                        break;
                hammer_stats_file_read += n;
        }

        /*
         * XXX only update the atime if we had to get the MP lock.
         * XXX hack hack hack, fixme.
         */
        if (got_mplock) {
                if ((ip->flags & HAMMER_INODE_RO) == 0 &&
                    (ip->hmp->mp->mnt_flag & MNT_NOATIME) == 0) {
                        ip->ino_data.atime = trans.time;
                        hammer_modify_inode(ip, HAMMER_INODE_ATIME);
                }
                hammer_done_transaction(&trans);
                if (got_mplock > 0)
                        rel_mplock();
        }
        return (error);
}

/*
 * hammer_vop_write { vp, uio, ioflag, cred }
 */
static
int
hammer_vop_write(struct vop_write_args *ap)
{
        struct hammer_transaction trans;
        struct hammer_inode *ip;
        hammer_mount_t hmp;
        struct uio *uio;
        int offset;
        off_t base_offset;
        struct buf *bp;
        int kflags;
        int error;
        int n;
        int flags;
        int seqcount;
        int bigwrite;

        if (ap->a_vp->v_type != VREG)
                return (EINVAL);
        ip = VTOI(ap->a_vp);
        hmp = ip->hmp;
        error = 0;
        kflags = 0;
        seqcount = ap->a_ioflag >> 16;

        if (ip->flags & HAMMER_INODE_RO)
                return (EROFS);

        /*
         * Create a transaction to cover the operations we perform.
         */
        hammer_start_transaction(&trans, hmp);
        uio = ap->a_uio;

        /*
         * Check append mode
         */
        if (ap->a_ioflag & IO_APPEND)
                uio->uio_offset = ip->ino_data.size;

        /*
         * Check for illegal write offsets.  Valid range is 0...2^63-1.
         *
         * NOTE: the base_off assignment is required to work around what
         * I consider to be a GCC-4 optimization bug.
         */
        if (uio->uio_offset < 0) {
                hammer_done_transaction(&trans);
                return (EFBIG);
        }
        base_offset = uio->uio_offset + uio->uio_resid; /* work around gcc-4 */
        if (uio->uio_resid > 0 && base_offset <= uio->uio_offset) {
                hammer_done_transaction(&trans);
                return (EFBIG);
        }

        /*
         * If reading or writing a huge amount of data we have to break
         * atomicy and allow the operation to be interrupted by a signal
         * or it can DOS the machine.
         */
        bigwrite = (uio->uio_resid > 100 * 1024 * 1024);

        /*
         * Access the data typically in HAMMER_BUFSIZE blocks via the
         * buffer cache, but HAMMER may use a variable block size based
         * on the offset.
         */
        while (uio->uio_resid > 0) {
                int fixsize = 0;
                int blksize;
                int blkmask;

                if ((error = hammer_checkspace(hmp, HAMMER_CHKSPC_WRITE)) != 0)
                        break;
                if (bigwrite && (error = hammer_signal_check(hmp)) != 0)
                        break;

                blksize = hammer_blocksize(uio->uio_offset);

                /*
                 * Do not allow HAMMER to blow out the buffer cache.  Very
                 * large UIOs can lockout other processes due to bwillwrite()
                 * mechanics.
                 *
                 * The hammer inode is not locked during these operations.
                 * The vnode is locked which can interfere with the pageout
                 * daemon for non-UIO_NOCOPY writes but should not interfere
                 * with the buffer cache.  Even so, we cannot afford to
                 * allow the pageout daemon to build up too many dirty buffer
                 * cache buffers.
                 *
                 * Only call this if we aren't being recursively called from
                 * a virtual disk device (vn), else we may deadlock.
                 */
                if ((ap->a_ioflag & IO_RECURSE) == 0)
                        bwillwrite(blksize);

                /*
                 * Control the number of pending records associated with
                 * this inode.  If too many have accumulated start a
                 * flush.  Try to maintain a pipeline with the flusher.
                 */
                if (ip->rsv_recs >= hammer_limit_inode_recs) {
                        hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
                }
                if (ip->rsv_recs >= hammer_limit_inode_recs * 2) {
                        while (ip->rsv_recs >= hammer_limit_inode_recs) {
                                tsleep(&ip->rsv_recs, 0, "hmrwww", hz);
                        }
                        hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
                }

#if 0
                /*
                 * Do not allow HAMMER to blow out system memory by
                 * accumulating too many records.   Records are so well
                 * decoupled from the buffer cache that it is possible
                 * for userland to push data out to the media via
                 * direct-write, but build up the records queued to the
                 * backend faster then the backend can flush them out.
                 * HAMMER has hit its write limit but the frontend has
                 * no pushback to slow it down.
                 */
                if (hmp->rsv_recs > hammer_limit_recs / 2) {
                        /*
                         * Get the inode on the flush list
                         */
                        if (ip->rsv_recs >= 64)
                                hammer_flush_inode(ip, HAMMER_FLUSH_SIGNAL);
                        else if (ip->rsv_recs >= 16)
                                hammer_flush_inode(ip, 0);

                        /*
                         * Keep the flusher going if the system keeps
                         * queueing records.
                         */
                        delta = hmp->count_newrecords -
                                hmp->last_newrecords;
                        if (delta < 0 || delta > hammer_limit_recs / 2) {
                                hmp->last_newrecords = hmp->count_newrecords;
                                hammer_sync_hmp(hmp, MNT_NOWAIT);
                        }

                        /*
                         * If we have gotten behind start slowing
                         * down the writers.
                         */
                        delta = (hmp->rsv_recs - hammer_limit_recs) *
                                hz / hammer_limit_recs;
                        if (delta > 0)
                                tsleep(&trans, 0, "hmrslo", delta);
                }
#endif

                /*
                 * Calculate the blocksize at the current offset and figure
                 * out how much we can actually write.
                 */
                blkmask = blksize - 1;
                offset = (int)uio->uio_offset & blkmask;
                base_offset = uio->uio_offset & ~(int64_t)blkmask;
                n = blksize - offset;
                if (n > uio->uio_resid)
                        n = uio->uio_resid;
                if (uio->uio_offset + n > ip->ino_data.size) {
                        vnode_pager_setsize(ap->a_vp, uio->uio_offset + n);
                        fixsize = 1;
                        kflags |= NOTE_EXTEND;
                }

                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(ap->a_vp, base_offset,
                                    blksize, GETBLK_BHEAVY, 0);
                        if ((bp->b_flags & B_CACHE) == 0) {
                                bqrelse(bp);
                                error = bread(ap->a_vp, base_offset,
                                              blksize, &bp);
                        }
                } else if (offset == 0 && uio->uio_resid >= blksize) {
                        /*
                         * 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(ap->a_vp, base_offset, blksize, GETBLK_BHEAVY, 0);
                        if ((bp->b_flags & B_CACHE) == 0)
                                vfs_bio_clrbuf(bp);
                } else if (base_offset >= ip->ino_data.size) {
                        /*
                         * If the base offset of the buffer is beyond the
                         * file EOF, we don't have to issue a read.
                         */
                        bp = getblk(ap->a_vp, base_offset,
                                    blksize, GETBLK_BHEAVY, 0);
                        vfs_bio_clrbuf(bp);
                } else {
                        /*
                         * Partial overwrite, read in any missing bits then
                         * replace the portion being written.
                         */
                        error = bread(ap->a_vp, base_offset, blksize, &bp);
                        if (error == 0)
                                bheavy(bp);
                }
                if (error == 0) {
                        error = uiomove((char *)bp->b_data + offset,
                                        n, uio);
                }

                /*
                 * If we screwed up we have to undo any VM size changes we
                 * made.
                 */
                if (error) {
                        brelse(bp);
                        if (fixsize) {
                                vtruncbuf(ap->a_vp, ip->ino_data.size,
                                          hammer_blocksize(ip->ino_data.size));
                        }
                        break;
                }
                kflags |= NOTE_WRITE;
                hammer_stats_file_write += n;
                /* bp->b_flags |= B_CLUSTEROK; temporarily disabled */
                if (ip->ino_data.size < uio->uio_offset) {
                        ip->ino_data.size = uio->uio_offset;
                        flags = HAMMER_INODE_DDIRTY;
                        vnode_pager_setsize(ap->a_vp, ip->ino_data.size);
                } else {
                        flags = 0;
                }
                ip->ino_data.mtime = trans.time;
                flags |= HAMMER_INODE_MTIME | HAMMER_INODE_BUFS;
                hammer_modify_inode(ip, flags);

                /*
                 * Once we dirty the buffer any cached zone-X offset
                 * becomes invalid.  HAMMER NOTE: no-history mode cannot 
                 * allow overwriting over the same data sector unless
                 * we provide UNDOs for the old data, which we don't.
                 */
                bp->b_bio2.bio_offset = NOOFFSET;

                /*
                 * Final buffer disposition.
                 *
                 * Because meta-data updates are deferred, HAMMER is
                 * especially sensitive to excessive bdwrite()s because
                 * the I/O stream is not broken up by disk reads.  So the
                 * buffer cache simply cannot keep up.
                 *
                 * WARNING!  blksize is variable.  cluster_write() is
                 * expected to not blow up if it encounters buffers that
                 * do not match the passed blksize.
                 */
                bp->b_flags |= B_AGE;
                if (ap->a_ioflag & IO_SYNC) {
                        bwrite(bp);
                } else if (ap->a_ioflag & IO_DIRECT) {
                        bawrite(bp);
                } else if (offset + n == blksize) {
                        if (hammer_cluster_enable == 0 ||
                            (ap->a_vp->v_mount->mnt_flag & MNT_NOCLUSTERW)) {
                                bawrite(bp);
                        } else {
                                cluster_write(bp, ip->ino_data.size,
                                              blksize, seqcount);
                        }
                } else {
                        bdwrite(bp);
                }
        }
        hammer_done_transaction(&trans);
        hammer_knote(ap->a_vp, kflags);
        return (error);
}

/*
 * hammer_vop_access { vp, mode, cred }
 */
static
int
hammer_vop_access(struct vop_access_args *ap)
{
        struct hammer_inode *ip = VTOI(ap->a_vp);
        uid_t uid;
        gid_t gid;
        int error;

        ++hammer_stats_file_iopsr;
        uid = hammer_to_unix_xid(&ip->ino_data.uid);
        gid = hammer_to_unix_xid(&ip->ino_data.gid);

        error = vop_helper_access(ap, uid, gid, ip->ino_data.mode,
                                  ip->ino_data.uflags);
        return (error);
}

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

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

/*
 * hammer_vop_close { vp, fflag }
 */
static
int
hammer_vop_close(struct vop_close_args *ap)
{
        /*hammer_inode_t ip = VTOI(ap->a_vp);*/
        return (vop_stdclose(ap));
}

/*
 * hammer_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
hammer_vop_ncreate(struct vop_ncreate_args *ap)
{
        struct hammer_transaction trans;
        struct hammer_inode *dip;
        struct hammer_inode *nip;
        struct nchandle *nch;
        int error;

        nch = ap->a_nch;
        dip = VTOI(ap->a_dvp);

        if (dip->flags & HAMMER_INODE_RO)
                return (EROFS);
        if ((error = hammer_checkspace(dip->hmp, HAMMER_CHKSPC_CREATE)) != 0)
                return (error);

        /*
         * Create a transaction to cover the operations we perform.
         */
        hammer_start_transaction(&trans, dip->hmp);
        ++hammer_stats_file_iopsw;

        /*
         * Create a new filesystem object of the requested type.  The
         * returned inode will be referenced and shared-locked to prevent
         * it from being moved to the flusher.
         */
        error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred,
                                    dip, nch->ncp->nc_name, nch->ncp->nc_nlen,
                                    NULL, &nip);
        if (error) {
                hkprintf("hammer_create_inode error %d\n", error);
                hammer_done_transaction(&trans);
                *ap->a_vpp = NULL;
                return (error);
        }

        /*
         * Add the new filesystem object to the directory.  This will also
         * bump the inode's link count.
         */
        error = hammer_ip_add_directory(&trans, dip,
                                        nch->ncp->nc_name, nch->ncp->nc_nlen,
                                        nip);
        if (error)
                hkprintf("hammer_ip_add_directory error %d\n", error);

        /*
         * Finish up.
         */
        if (error) {
                hammer_rel_inode(nip, 0);
                hammer_done_transaction(&trans);
                *ap->a_vpp = NULL;
        } else {
                error = hammer_get_vnode(nip, ap->a_vpp);
                hammer_done_transaction(&trans);
                hammer_rel_inode(nip, 0);
                if (error == 0) {
                        cache_setunresolved(ap->a_nch);
                        cache_setvp(ap->a_nch, *ap->a_vpp);
                }
                hammer_knote(ap->a_dvp, NOTE_WRITE);
        }
        return (error);
}

/*
 * hammer_vop_getattr { vp, vap }
 *
 * Retrieve an inode's attribute information.  When accessing inodes
 * historically we fake the atime field to ensure consistent results.
 * The atime field is stored in the B-Tree element and allowed to be
 * updated without cycling the element.
 *
 * MPSAFE
 */
static
int
hammer_vop_getattr(struct vop_getattr_args *ap)
{
        struct hammer_inode *ip = VTOI(ap->a_vp);
        struct vattr *vap = ap->a_vap;

        /*
         * We want the fsid to be different when accessing a filesystem
         * with different as-of's so programs like diff don't think
         * the files are the same.
         *
         * We also want the fsid to be the same when comparing snapshots,
         * or when comparing mirrors (which might be backed by different
         * physical devices).  HAMMER fsids are based on the PFS's
         * shared_uuid field.
         *
         * XXX there is a chance of collision here.  The va_fsid reported
         * by stat is different from the more involved fsid used in the
         * mount structure.
         */
        ++hammer_stats_file_iopsr;
        hammer_lock_sh(&ip->lock);
        vap->va_fsid = ip->pfsm->fsid_udev ^ (u_int32_t)ip->obj_asof ^
                       (u_int32_t)(ip->obj_asof >> 32);

        vap->va_fileid = ip->ino_leaf.base.obj_id;
        vap->va_mode = ip->ino_data.mode;
        vap->va_nlink = ip->ino_data.nlinks;
        vap->va_uid = hammer_to_unix_xid(&ip->ino_data.uid);
        vap->va_gid = hammer_to_unix_xid(&ip->ino_data.gid);
        vap->va_rmajor = 0;
        vap->va_rminor = 0;
        vap->va_size = ip->ino_data.size;

        /*
         * Special case for @@PFS softlinks.  The actual size of the
         * expanded softlink is "@@0x%016llx:%05d" == 26 bytes.
         * or for MAX_TID is    "@@-1:%05d" == 10 bytes.
         */
        if (ip->ino_data.obj_type == HAMMER_OBJTYPE_SOFTLINK &&
            ip->ino_data.size == 10 &&
            ip->obj_asof == HAMMER_MAX_TID &&
            ip->obj_localization == 0 &&
            strncmp(ip->ino_data.ext.symlink, "@@PFS", 5) == 0) {
                    if (ip->pfsm->pfsd.mirror_flags & HAMMER_PFSD_SLAVE)
                            vap->va_size = 26;
                    else
                            vap->va_size = 10;
        }

        /*
         * We must provide a consistent atime and mtime for snapshots
         * so people can do a 'tar cf - ... | md5' on them and get
         * consistent results.
         */
        if (ip->flags & HAMMER_INODE_RO) {
                hammer_time_to_timespec(ip->ino_data.ctime, &vap->va_atime);
                hammer_time_to_timespec(ip->ino_data.ctime, &vap->va_mtime);
        } else {
                hammer_time_to_timespec(ip->ino_data.atime, &vap->va_atime);
                hammer_time_to_timespec(ip->ino_data.mtime, &vap->va_mtime);
        }
        hammer_time_to_timespec(ip->ino_data.ctime, &vap->va_ctime);
        vap->va_flags = ip->ino_data.uflags;
        vap->va_gen = 1;        /* hammer inums are unique for all time */
        vap->va_blocksize = HAMMER_BUFSIZE;
        if (ip->ino_data.size >= HAMMER_XDEMARC) {
                vap->va_bytes = (ip->ino_data.size + HAMMER_XBUFMASK64) &
                                ~HAMMER_XBUFMASK64;
        } else if (ip->ino_data.size > HAMMER_BUFSIZE / 2) {
                vap->va_bytes = (ip->ino_data.size + HAMMER_BUFMASK64) &
                                ~HAMMER_BUFMASK64;
        } else {
                vap->va_bytes = (ip->ino_data.size + 15) & ~15;
        }

        vap->va_type = hammer_get_vnode_type(ip->ino_data.obj_type);
        vap->va_filerev = 0;    /* XXX */
        /* mtime uniquely identifies any adjustments made to the file XXX */
        vap->va_fsmid = ip->ino_data.mtime;
        vap->va_uid_uuid = ip->ino_data.uid;
        vap->va_gid_uuid = ip->ino_data.gid;
        vap->va_fsid_uuid = ip->hmp->fsid;
        vap->va_vaflags = VA_UID_UUID_VALID | VA_GID_UUID_VALID |
                          VA_FSID_UUID_VALID;

        switch (ip->ino_data.obj_type) {
        case HAMMER_OBJTYPE_CDEV:
        case HAMMER_OBJTYPE_BDEV:
                vap->va_rmajor = ip->ino_data.rmajor;
                vap->va_rminor = ip->ino_data.rminor;
                break;
        default:
                break;
        }
        hammer_unlock(&ip->lock);
        return(0);
}

/*
 * hammer_vop_nresolve { nch, dvp, cred }
 *
 * Locate the requested directory entry.
 */
static
int
hammer_vop_nresolve(struct vop_nresolve_args *ap)
{
        struct hammer_transaction trans;
        struct namecache *ncp;
        hammer_inode_t dip;
        hammer_inode_t ip;
        hammer_tid_t asof;
        struct hammer_cursor cursor;
        struct vnode *vp;
        int64_t namekey;
        int error;
        int i;
        int nlen;
        int flags;
        int ispfs;
        int64_t obj_id;
        u_int32_t localization;
        u_int32_t max_iterations;

        /*
         * Misc initialization, plus handle as-of name extensions.  Look for
         * the '@@' extension.  Note that as-of files and directories cannot
         * be modified.
         */
        dip = VTOI(ap->a_dvp);
        ncp = ap->a_nch->ncp;
        asof = dip->obj_asof;
        localization = dip->obj_localization;   /* for code consistency */
        nlen = ncp->nc_nlen;
        flags = dip->flags & HAMMER_INODE_RO;
        ispfs = 0;

        hammer_simple_transaction(&trans, dip->hmp);
        ++hammer_stats_file_iopsr;

        for (i = 0; i < nlen; ++i) {
                if (ncp->nc_name[i] == '@' && ncp->nc_name[i+1] == '@') {
                        error = hammer_str_to_tid(ncp->nc_name + i + 2,
                                                  &ispfs, &asof, &localization);
                        if (error != 0) {
                                i = nlen;
                                break;
                        }
                        if (asof != HAMMER_MAX_TID)
                                flags |= HAMMER_INODE_RO;
                        break;
                }
        }
        nlen = i;

        /*
         * If this is a PFS softlink we dive into the PFS
         */
        if (ispfs && nlen == 0) {
                ip = hammer_get_inode(&trans, dip, HAMMER_OBJID_ROOT,
                                      asof, localization,
                                      flags, &error);
                if (error == 0) {
                        error = hammer_get_vnode(ip, &vp);
                        hammer_rel_inode(ip, 0);
                } else {
                        vp = NULL;
                }
                if (error == 0) {
                        vn_unlock(vp);
                        cache_setvp(ap->a_nch, vp);
                        vrele(vp);
                }
                goto done;
        }

        /*
         * If there is no path component the time extension is relative to dip.
         * e.g. "fubar/@@<snapshot>"
         *
         * "." is handled by the kernel, but ".@@<snapshot>" is not.
         * e.g. "fubar/.@@<snapshot>"
         *
         * ".." is handled by the kernel.  We do not currently handle
         * "..@<snapshot>".
         */
        if (nlen == 0 || (nlen == 1 && ncp->nc_name[0] == '.')) {
                ip = hammer_get_inode(&trans, dip, dip->obj_id,
                                      asof, dip->obj_localization,
                                      flags, &error);
                if (error == 0) {
                        error = hammer_get_vnode(ip, &vp);
                        hammer_rel_inode(ip, 0);
                } else {
                        vp = NULL;
                }
                if (error == 0) {
                        vn_unlock(vp);
                        cache_setvp(ap->a_nch, vp);
                        vrele(vp);
                }
                goto done;
        }

        /*
         * Calculate the namekey and setup the key range for the scan.  This
         * works kinda like a chained hash table where the lower 32 bits
         * of the namekey synthesize the chain.
         *
         * The key range is inclusive of both key_beg and key_end.
         */
        namekey = hammer_directory_namekey(dip, ncp->nc_name, nlen,
                                           &max_iterations);

        error = hammer_init_cursor(&trans, &cursor, &dip->cache[1], dip);
        cursor.key_beg.localization = dip->obj_localization +
                                      hammer_dir_localization(dip);
        cursor.key_beg.obj_id = dip->obj_id;
        cursor.key_beg.key = namekey;
        cursor.key_beg.create_tid = 0;
        cursor.key_beg.delete_tid = 0;
        cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
        cursor.key_beg.obj_type = 0;

        cursor.key_end = cursor.key_beg;
        cursor.key_end.key += max_iterations;
        cursor.asof = asof;
        cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;

        /*
         * Scan all matching records (the chain), locate the one matching
         * the requested path component.
         *
         * The hammer_ip_*() functions merge in-memory records with on-disk
         * records for the purposes of the search.
         */
        obj_id = 0;
        localization = HAMMER_DEF_LOCALIZATION;

        if (error == 0) {
                error = hammer_ip_first(&cursor);
                while (error == 0) {
                        error = hammer_ip_resolve_data(&cursor);
                        if (error)
                                break;
                        if (nlen == cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF &&
                            bcmp(ncp->nc_name, cursor.data->entry.name, nlen) == 0) {
                                obj_id = cursor.data->entry.obj_id;
                                localization = cursor.data->entry.localization;
                                break;
                        }
                        error = hammer_ip_next(&cursor);
                }
        }
        hammer_done_cursor(&cursor);

        /*
         * Lookup the obj_id.  This should always succeed.  If it does not
         * the filesystem may be damaged and we return a dummy inode.
         */
        if (error == 0) {
                ip = hammer_get_inode(&trans, dip, obj_id,
                                      asof, localization,
                                      flags, &error);
                if (error == ENOENT) {
                        kprintf("HAMMER: WARNING: Missing "
                                "inode for dirent \"%s\"\n"
                                "\tobj_id = %016llx, asof=%016llx, lo=%08x\n",
                                ncp->nc_name,
                                (long long)obj_id, (long long)asof,
                                localization);
                        error = 0;
                        ip = hammer_get_dummy_inode(&trans, dip, obj_id,
                                                    asof, localization,
                                                    flags, &error);
                }
                if (error == 0) {
                        error = hammer_get_vnode(ip, &vp);
                        hammer_rel_inode(ip, 0);
                } else {
                        vp = NULL;
                }
                if (error == 0) {
                        vn_unlock(vp);
                        cache_setvp(ap->a_nch, vp);
                        vrele(vp);
                }
        } else if (error == ENOENT) {
                cache_setvp(ap->a_nch, NULL);
        }
done:
        hammer_done_transaction(&trans);
        return (error);
}

/*
 * hammer_vop_nlookupdotdot { dvp, vpp, cred }
 *
 * Locate the parent directory of a directory vnode.
 *
 * dvp is referenced but not locked.  *vpp must be returned referenced and
 * locked.  A parent_obj_id of 0 does not necessarily indicate that we are
 * at the root, instead it could indicate that the directory we were in was
 * removed.
 *
 * NOTE: as-of sequences are not linked into the directory structure.  If
 * we are at the root with a different asof then the mount point, reload
 * the same directory with the mount point's asof.   I'm not sure what this
 * will do to NFS.  We encode ASOF stamps in NFS file handles so it might not
 * get confused, but it hasn't been tested.
 */
static
int
hammer_vop_nlookupdotdot(struct vop_nlookupdotdot_args *ap)
{
        struct hammer_transaction trans;
        struct hammer_inode *dip;
        struct hammer_inode *ip;
        int64_t parent_obj_id;
        u_int32_t parent_obj_localization;
        hammer_tid_t asof;
        int error;

        dip = VTOI(ap->a_dvp);
        asof = dip->obj_asof;

        /*
         * Whos are parent?  This could be the root of a pseudo-filesystem
         * whos parent is in another localization domain.
         */
        parent_obj_id = dip->ino_data.parent_obj_id;
        if (dip->obj_id == HAMMER_OBJID_ROOT)
                parent_obj_localization = dip->ino_data.ext.obj.parent_obj_localization;
        else
                parent_obj_localization = dip->obj_localization;

        if (parent_obj_id == 0) {
                if (dip->obj_id == HAMMER_OBJID_ROOT &&
                   asof != dip->hmp->asof) {
                        parent_obj_id = dip->obj_id;
                        asof = dip->hmp->asof;
                        *ap->a_fakename = kmalloc(19, M_TEMP, M_WAITOK);
                        ksnprintf(*ap->a_fakename, 19, "0x%016llx",
                                  (long long)dip->obj_asof);
                } else {
                        *ap->a_vpp = NULL;
                        return ENOENT;
                }
        }

        hammer_simple_transaction(&trans, dip->hmp);
        ++hammer_stats_file_iopsr;

        ip = hammer_get_inode(&trans, dip, parent_obj_id,
                              asof, parent_obj_localization,
                              dip->flags, &error);
        if (ip) {
                error = hammer_get_vnode(ip, ap->a_vpp);
                hammer_rel_inode(ip, 0);
        } else {
                *ap->a_vpp = NULL;
        }
        hammer_done_transaction(&trans);
        return (error);
}

/*
 * hammer_vop_nlink { nch, dvp, vp, cred }
 */
static
int
hammer_vop_nlink(struct vop_nlink_args *ap)
{
        struct hammer_transaction trans;
        struct hammer_inode *dip;
        struct hammer_inode *ip;
        struct nchandle *nch;
        int error;

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

        nch = ap->a_nch;
        dip = VTOI(ap->a_dvp);
        ip = VTOI(ap->a_vp);

        if (dip->obj_localization != ip->obj_localization)
                return(EXDEV);

        if (dip->flags & HAMMER_INODE_RO)
                return (EROFS);
        if (ip->flags & HAMMER_INODE_RO)
                return (EROFS);
        if ((error = hammer_checkspace(dip->hmp, HAMMER_CHKSPC_CREATE)) != 0)
                return (error);

        /*
         * Create a transaction to cover the operations we perform.
         */
        hammer_start_transaction(&trans, dip->hmp);
        ++hammer_stats_file_iopsw;

        /*
         * Add the filesystem object to the directory.  Note that neither
         * dip nor ip are referenced or locked, but their vnodes are
         * referenced.  This function will bump the inode's link count.
         */
        error = hammer_ip_add_directory(&trans, dip,
                                        nch->ncp->nc_name, nch->ncp->nc_nlen,
                                        ip);

        /*
         * Finish up.
         */
        if (error == 0) {
                cache_setunresolved(nch);
                cache_setvp(nch, ap->a_vp);
        }
        hammer_done_transaction(&trans);
        hammer_knote(ap->a_vp, NOTE_LINK);
        hammer_knote(ap->a_dvp, NOTE_WRITE);
        return (error);
}

/*
 * hammer_vop_nmkdir { 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
hammer_vop_nmkdir(struct vop_nmkdir_args *ap)
{
        struct hammer_transaction trans;
        struct hammer_inode *dip;
        struct hammer_inode *nip;
        struct nchandle *nch;
        int error;

        nch = ap->a_nch;
        dip = VTOI(ap->a_dvp);

        if (dip->flags & HAMMER_INODE_RO)
                return (EROFS);
        if ((error = hammer_checkspace(dip->hmp, HAMMER_CHKSPC_CREATE)) != 0)
                return (error);

        /*
         * Create a transaction to cover the operations we perform.
         */
        hammer_start_transaction(&trans, dip->hmp);
        ++hammer_stats_file_iopsw;

        /*
         * Create a new filesystem object of the requested type.  The
         * returned inode will be referenced but not locked.
         */
        error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred,
                                    dip, nch->ncp->nc_name, nch->ncp->nc_nlen,
                                    NULL, &nip);
        if (error) {
                hkprintf("hammer_mkdir error %d\n", error);
                hammer_done_transaction(&trans);
                *ap->a_vpp = NULL;
                return (error);
        }
        /*
         * Add the new filesystem object to the directory.  This will also
         * bump the inode's link count.
         */
        error = hammer_ip_add_directory(&trans, dip,
                                        nch->ncp->nc_name, nch->ncp->nc_nlen,
                                        nip);
        if (error)
                hkprintf("hammer_mkdir (add) error %d\n", error);

        /*
         * Finish up.
         */
        if (error) {
                hammer_rel_inode(nip, 0);
                *ap->a_vpp = NULL;
        } else {
                error = hammer_get_vnode(nip, ap->a_vpp);
                hammer_rel_inode(nip, 0);
                if (error == 0) {
                        cache_setunresolved(ap->a_nch);
                        cache_setvp(ap->a_nch, *ap->a_vpp);
                }
        }
        hammer_done_transaction(&trans);
        if (error == 0)
                hammer_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK);
        return (error);
}

/*
 * hammer_vop_nmknod { 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
hammer_vop_nmknod(struct vop_nmknod_args *ap)
{
        struct hammer_transaction trans;
        struct hammer_inode *dip;
        struct hammer_inode *nip;
        struct nchandle *nch;
        int error;

        nch = ap->a_nch;
        dip = VTOI(ap->a_dvp);

        if (dip->flags & HAMMER_INODE_RO)
                return (EROFS);
        if ((error = hammer_checkspace(dip->hmp, HAMMER_CHKSPC_CREATE)) != 0)
                return (error);

        /*
         * Create a transaction to cover the operations we perform.
         */
        hammer_start_transaction(&trans, dip->hmp);
        ++hammer_stats_file_iopsw;

        /*
         * Create a new filesystem object of the requested type.  The
         * returned inode will be referenced but not locked.
         *
         * If mknod specifies a directory a pseudo-fs is created.
         */
        error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred,
                                    dip, nch->ncp->nc_name, nch->ncp->nc_nlen,
                                    NULL, &nip);
        if (error) {
                hammer_done_transaction(&trans);
                *ap->a_vpp = NULL;
                return (error);
        }

        /*
         * Add the new filesystem object to the directory.  This will also
         * bump the inode's link count.
         */
        error = hammer_ip_add_directory(&trans, dip,
                                        nch->ncp->nc_name, nch->ncp->nc_nlen,
                                        nip);

        /*
         * Finish up.
         */
        if (error) {
                hammer_rel_inode(nip, 0);
                *ap->a_vpp = NULL;
        } else {
                error = hammer_get_vnode(nip, ap->a_vpp);
                hammer_rel_inode(nip, 0);
                if (error == 0) {
                        cache_setunresolved(ap->a_nch);
                        cache_setvp(ap->a_nch, *ap->a_vpp);
                }
        }
        hammer_done_transaction(&trans);
        if (error == 0)
                hammer_knote(ap->a_dvp, NOTE_WRITE);
        return (error);
}

/*
 * hammer_vop_open { vp, mode, cred, fp }
 */
static
int
hammer_vop_open(struct vop_open_args *ap)
{
        hammer_inode_t ip;

        ++hammer_stats_file_iopsr;
        ip = VTOI(ap->a_vp);

        if ((ap->a_mode & FWRITE) && (ip->flags & HAMMER_INODE_RO))
                return (EROFS);
        return(vop_stdopen(ap));
}

/*
 * hammer_vop_print { vp }
 */
static
int
hammer_vop_print(struct vop_print_args *ap)
{
        return EOPNOTSUPP;
}

/*
 * hammer_vop_readdir { vp, uio, cred, *eofflag, *ncookies, off_t **cookies }
 */
static
int
hammer_vop_readdir(struct vop_readdir_args *ap)
{
        struct hammer_transaction trans;
        struct hammer_cursor cursor;
        struct hammer_inode *ip;
        struct uio *uio;
        hammer_base_elm_t base;
        int error;
        int cookie_index;
        int ncookies;
        off_t *cookies;
        off_t saveoff;
        int r;
        int dtype;

        ++hammer_stats_file_iopsr;
        ip = VTOI(ap->a_vp);
        uio = ap->a_uio;
        saveoff = uio->uio_offset;

        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);
                cookie_index = 0;
        } else {
                ncookies = -1;
                cookies = NULL;
                cookie_index = 0;
        }

        hammer_simple_transaction(&trans, ip->hmp);

        /*
         * Handle artificial entries
         *
         * It should be noted that the minimum value for a directory
         * hash key on-media is 0x0000000100000000, so we can use anything
         * less then that to represent our 'special' key space.
         */
        error = 0;
        if (saveoff == 0) {
                r = vop_write_dirent(&error, uio, ip->obj_id, 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) {
                if (ip->ino_data.parent_obj_id) {
                        r = vop_write_dirent(&error, uio,
                                             ip->ino_data.parent_obj_id,
                                             DT_DIR, 2, "..");
                } else {
                        r = vop_write_dirent(&error, uio,
                                             ip->obj_id, DT_DIR, 2, "..");
                }
                if (r)
                        goto done;
                if (cookies)
                        cookies[cookie_index] = saveoff;
                ++saveoff;
                ++cookie_index;
                if (cookie_index == ncookies)
                        goto done;
        }

        /*
         * Key range (begin and end inclusive) to scan.  Directory keys
         * directly translate to a 64 bit 'seek' position.
         */
        hammer_init_cursor(&trans, &cursor, &ip->cache[1], ip);
        cursor.key_beg.localization = ip->obj_localization +
                                      hammer_dir_localization(ip);
        cursor.key_beg.obj_id = ip->obj_id;
        cursor.key_beg.create_tid = 0;
        cursor.key_beg.delete_tid = 0;
        cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
        cursor.key_beg.obj_type = 0;
        cursor.key_beg.key = saveoff;

        cursor.key_end = cursor.key_beg;
        cursor.key_end.key = HAMMER_MAX_KEY;
        cursor.asof = ip->obj_asof;
        cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;

        error = hammer_ip_first(&cursor);

        while (error == 0) {
                error = hammer_ip_resolve_data(&cursor);
                if (error)
                        break;
                base = &cursor.leaf->base;
                saveoff = base->key;
                KKASSERT(cursor.leaf->data_len > HAMMER_ENTRY_NAME_OFF);

                if (base->obj_id != ip->obj_id)
                        panic("readdir: bad record at %p", cursor.node);

                /*
                 * Convert pseudo-filesystems into softlinks
                 */
                dtype = hammer_get_dtype(cursor.leaf->base.obj_type);
                r = vop_write_dirent(
                             &error, uio, cursor.data->entry.obj_id,
                             dtype,
                             cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF ,
                             (void *)cursor.data->entry.name);
                if (r)
                        break;
                ++saveoff;
                if (cookies)
                        cookies[cookie_index] = base->key;
                ++cookie_index;
                if (cookie_index == ncookies)
                        break;
                error = hammer_ip_next(&cursor);
        }
        hammer_done_cursor(&cursor);

done:
        hammer_done_transaction(&trans);

        if (ap->a_eofflag)
                *ap->a_eofflag = (error == ENOENT);
        uio->uio_offset = saveoff;
        if (error && cookie_index == 0) {
                if (error == ENOENT)
                        error = 0;
                if (cookies) {
                        kfree(cookies, M_TEMP);
                        *ap->a_ncookies = 0;
                        *ap->a_cookies = NULL;
                }
        } else {
                if (error == ENOENT)
                        error = 0;
                if (cookies) {
                        *ap->a_ncookies = cookie_index;
                        *ap->a_cookies = cookies;
                }
        }
        return(error);
}

/*
 * hammer_vop_readlink { vp, uio, cred }
 */
static
int
hammer_vop_readlink(struct vop_readlink_args *ap)
{
        struct hammer_transaction trans;
        struct hammer_cursor cursor;
        struct hammer_inode *ip;
        char buf[32];
        u_int32_t localization;
        hammer_pseudofs_inmem_t pfsm;
        int error;

        ip = VTOI(ap->a_vp);

        /*
         * Shortcut if the symlink data was stuffed into ino_data.
         *
         * Also expand special "@@PFS%05d" softlinks (expansion only
         * occurs for non-historical (current) accesses made from the
         * primary filesystem).
         */
        if (ip->ino_data.size <= HAMMER_INODE_BASESYMLEN) {
                char *ptr;
                int bytes;

                ptr = ip->ino_data.ext.symlink;
                bytes = (int)ip->ino_data.size;
                if (bytes == 10 &&
                    ip->obj_asof == HAMMER_MAX_TID &&
                    ip->obj_localization == 0 &&
                    strncmp(ptr, "@@PFS", 5) == 0) {
                        hammer_simple_transaction(&trans, ip->hmp);
                        bcopy(ptr + 5, buf, 5);
                        buf[5] = 0;
                        localization = strtoul(buf, NULL, 10) << 16;
                        pfsm = hammer_load_pseudofs(&trans, localization,
                                                    &error);
                        if (error == 0) {
                                if (pfsm->pfsd.mirror_flags &
                                    HAMMER_PFSD_SLAVE) {
                                        /* vap->va_size == 26 */
                                        ksnprintf(buf, sizeof(buf),
                                                  "@@0x%016llx:%05d",
                                                  (long long)pfsm->pfsd.sync_end_tid,
                                                  localization >> 16);
                                } else {
                                        /* vap->va_size == 10 */
                                        ksnprintf(buf, sizeof(buf),
                                                  "@@-1:%05d",
                                                  localization >> 16);
#if 0
                                        ksnprintf(buf, sizeof(buf),
                                                  "@@0x%016llx:%05d",
                                                  (long long)HAMMER_MAX_TID,
                                                  localization >> 16);
#endif
                                }
                                ptr = buf;
                                bytes = strlen(buf);
                        }
                        if (pfsm)
                                hammer_rel_pseudofs(trans.hmp, pfsm);
                        hammer_done_transaction(&trans);
                }
                error = uiomove(ptr, bytes, ap->a_uio);
                return(error);
        }

        /*
         * Long version
         */
        hammer_simple_transaction(&trans, ip->hmp);
        ++hammer_stats_file_iopsr;
        hammer_init_cursor(&trans, &cursor, &ip->cache[1], ip);

        /*
         * Key range (begin and end inclusive) to scan.  Directory keys
         * directly translate to a 64 bit 'seek' position.
         */
        cursor.key_beg.localization = ip->obj_localization +
                                      HAMMER_LOCALIZE_MISC;
        cursor.key_beg.obj_id = ip->obj_id;
        cursor.key_beg.create_tid = 0;
        cursor.key_beg.delete_tid = 0;
        cursor.key_beg.rec_type = HAMMER_RECTYPE_FIX;
        cursor.key_beg.obj_type = 0;
        cursor.key_beg.key = HAMMER_FIXKEY_SYMLINK;
        cursor.asof = ip->obj_asof;
        cursor.flags |= HAMMER_CURSOR_ASOF;

        error = hammer_ip_lookup(&cursor);
        if (error == 0) {
                error = hammer_ip_resolve_data(&cursor);
                if (error == 0) {
                        KKASSERT(cursor.leaf->data_len >=
                                 HAMMER_SYMLINK_NAME_OFF);
                        error = uiomove(cursor.data->symlink.name,
                                        cursor.leaf->data_len -
                                                HAMMER_SYMLINK_NAME_OFF,
                                        ap->a_uio);
                }
        }
        hammer_done_cursor(&cursor);
        hammer_done_transaction(&trans);
        return(error);
}

/*
 * hammer_vop_nremove { nch, dvp, cred }
 */
static
int
hammer_vop_nremove(struct vop_nremove_args *ap)
{
        struct hammer_transaction trans;
        struct hammer_inode *dip;
        int error;

        dip = VTOI(ap->a_dvp);

        if (hammer_nohistory(dip) == 0 &&
            (error = hammer_checkspace(dip->hmp, HAMMER_CHKSPC_REMOVE)) != 0) {
                return (error);
        }

        hammer_start_transaction(&trans, dip->hmp);
        ++hammer_stats_file_iopsw;
        error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0, 0);
        hammer_done_transaction(&trans);
        if (error == 0)
                hammer_knote(ap->a_dvp, NOTE_WRITE);
        return (error);
}

/*
 * hammer_vop_nrename { fnch, tnch, fdvp, tdvp, cred }
 */
static
int
hammer_vop_nrename(struct vop_nrename_args *ap)
{
        struct hammer_transaction trans;
        struct namecache *fncp;
        struct namecache *tncp;
        struct hammer_inode *fdip;
        struct hammer_inode *tdip;
        struct hammer_inode *ip;
        struct hammer_cursor cursor;
        int64_t namekey;
        u_int32_t max_iterations;
        int nlen, error;

        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);
        tdip = VTOI(ap->a_tdvp);
        fncp = ap->a_fnch->ncp;
        tncp = ap->a_tnch->ncp;
        ip = VTOI(fncp->nc_vp);
        KKASSERT(ip != NULL);

        if (fdip->obj_localization != tdip->obj_localization)
                return(EXDEV);
        if (fdip->obj_localization != ip->obj_localization)
                return(EXDEV);

        if (fdip->flags & HAMMER_INODE_RO)
                return (EROFS);
        if (tdip->flags & HAMMER_INODE_RO)
                return (EROFS);
        if (ip->flags & HAMMER_INODE_RO)
                return (EROFS);
        if ((error = hammer_checkspace(fdip->hmp, HAMMER_CHKSPC_CREATE)) != 0)
                return (error);

        hammer_start_transaction(&trans, fdip->hmp);
        ++hammer_stats_file_iopsw;

        /*
         * Remove tncp from the target directory and then link ip as
         * tncp. XXX pass trans to dounlink
         *
         * Force the inode sync-time to match the transaction so it is
         * in-sync with the creation of the target directory entry.
         */
        error = hammer_dounlink(&trans, ap->a_tnch, ap->a_tdvp,
                                ap->a_cred, 0, -1);
        if (error == 0 || error == ENOENT) {
                error = hammer_ip_add_directory(&trans, tdip,
                                                tncp->nc_name, tncp->nc_nlen,
                                                ip);
                if (error == 0) {
                        ip->ino_data.parent_obj_id = tdip->obj_id;
                        ip->ino_data.ctime = trans.time;
                        hammer_modify_inode(ip, HAMMER_INODE_DDIRTY);
                }
        }
        if (error)
                goto failed; /* XXX */

        /*
         * Locate the record in the originating directory and remove it.
         *
         * Calculate the namekey and setup the key range for the scan.  This
         * works kinda like a chained hash table where the lower 32 bits
         * of the namekey synthesize the chain.
         *
         * The key range is inclusive of both key_beg and key_end.
         */
        namekey = hammer_directory_namekey(fdip, fncp->nc_name, fncp->nc_nlen,
                                           &max_iterations);
retry:
        hammer_init_cursor(&trans, &cursor, &fdip->cache[1], fdip);
        cursor.key_beg.localization = fdip->obj_localization +
                                      hammer_dir_localization(fdip);
        cursor.key_beg.obj_id = fdip->obj_id;
        cursor.key_beg.key = namekey;
        cursor.key_beg.create_tid = 0;
        cursor.key_beg.delete_tid = 0;
        cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
        cursor.key_beg.obj_type = 0;

        cursor.key_end = cursor.key_beg;
        cursor.key_end.key += max_iterations;
        cursor.asof = fdip->obj_asof;
        cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;

        /*
         * Scan all matching records (the chain), locate the one matching
         * the requested path component.
         *
         * The hammer_ip_*() functions merge in-memory records with on-disk
         * records for the purposes of the search.
         */
        error = hammer_ip_first(&cursor);
        while (error == 0) {
                if (hammer_ip_resolve_data(&cursor) != 0)
                        break;
                nlen = cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF;
                KKASSERT(nlen > 0);
                if (fncp->nc_nlen == nlen &&
                    bcmp(fncp->nc_name, cursor.data->entry.name, nlen) == 0) {
                        break;
                }
                error = hammer_ip_next(&cursor);
        }

        /*
         * If all is ok we have to get the inode so we can adjust nlinks.
         *
         * WARNING: hammer_ip_del_directory() may have to terminate the
         * cursor to avoid a recursion.  It's ok to call hammer_done_cursor()
         * twice.
         */
        if (error == 0)
                error = hammer_ip_del_directory(&trans, &cursor, fdip, ip);

        /*
         * XXX A deadlock here will break rename's atomicy for the purposes
         * of crash recovery.
         */
        if (error == EDEADLK) {
                hammer_done_cursor(&cursor);
                goto retry;
        }

        /*
         * Cleanup and tell the kernel that the rename succeeded.
         */
        hammer_done_cursor(&cursor);
        if (error == 0) {
                cache_rename(ap->a_fnch, ap->a_tnch);
                hammer_knote(ap->a_fdvp, NOTE_WRITE);
                hammer_knote(ap->a_tdvp, NOTE_WRITE);
                if (ip->vp)
                        hammer_knote(ip->vp, NOTE_RENAME);
        }

failed:
        hammer_done_transaction(&trans);
        return (error);
}

/*
 * hammer_vop_nrmdir { nch, dvp, cred }
 */
static
int
hammer_vop_nrmdir(struct vop_nrmdir_args *ap)
{
        struct hammer_transaction trans;
        struct hammer_inode *dip;
        int error;

        dip = VTOI(ap->a_dvp);

        if (hammer_nohistory(dip) == 0 &&
            (error = hammer_checkspace(dip->hmp, HAMMER_CHKSPC_REMOVE)) != 0) {
                return (error);
        }

        hammer_start_transaction(&trans, dip->hmp);
        ++hammer_stats_file_iopsw;
        error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp, ap->a_cred, 0, 1);
        hammer_done_transaction(&trans);
        if (error == 0)
                hammer_knote(ap->a_dvp, NOTE_WRITE | NOTE_LINK);
        return (error);
}

/*
 * hammer_vop_markatime { vp, cred }
 */
static
int
hammer_vop_markatime(struct vop_markatime_args *ap)
{
        struct hammer_transaction trans;
        struct hammer_inode *ip;

        ip = VTOI(ap->a_vp);
        if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
                return (EROFS);
        if (ip->flags & HAMMER_INODE_RO)
                return (EROFS);
        if (ip->hmp->mp->mnt_flag & MNT_NOATIME)
                return (0);
        hammer_start_transaction(&trans, ip->hmp);
        ++hammer_stats_file_iopsw;

        ip->ino_data.atime = trans.time;
        hammer_modify_inode(ip, HAMMER_INODE_ATIME);
        hammer_done_transaction(&trans);
        hammer_knote(ap->a_vp, NOTE_ATTRIB);
        return (0);
}

/*
 * hammer_vop_setattr { vp, vap, cred }
 */
static
int
hammer_vop_setattr(struct vop_setattr_args *ap)
{
        struct hammer_transaction trans;
        struct vattr *vap;
        struct hammer_inode *ip;
        int modflags;
        int error;
        int truncating;
        int blksize;
        int kflags;
        int64_t aligned_size;
        u_int32_t flags;

        vap = ap->a_vap;
        ip = ap->a_vp->v_data;
        modflags = 0;
        kflags = 0;

        if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
                return(EROFS);
        if (ip->flags & HAMMER_INODE_RO)
                return (EROFS);
        if (hammer_nohistory(ip) == 0 &&
            (error = hammer_checkspace(ip->hmp, HAMMER_CHKSPC_REMOVE)) != 0) {
                return (error);
        }

        hammer_start_transaction(&trans, ip->hmp);
        ++hammer_stats_file_iopsw;
        error = 0;

        if (vap->va_flags != VNOVAL) {
                flags = ip->ino_data.uflags;
                error = vop_helper_setattr_flags(&flags, vap->va_flags,
                                         hammer_to_unix_xid(&ip->ino_data.uid),
                                         ap->a_cred);
                if (error == 0) {
                        if (ip->ino_data.uflags != flags) {
                                ip->ino_data.uflags = flags;
                                ip->ino_data.ctime = trans.time;
                                modflags |= HAMMER_INODE_DDIRTY;
                                kflags |= NOTE_ATTRIB;
                        }
                        if (ip->ino_data.uflags & (IMMUTABLE | APPEND)) {
                                error = 0;
                                goto done;
                        }
                }
                goto done;
        }
        if (ip->ino_data.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->ino_data.mode;
                uid_t cur_uid = hammer_to_unix_xid(&ip->ino_data.uid);
                gid_t cur_gid = hammer_to_unix_xid(&ip->ino_data.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) {
                        hammer_guid_to_uuid(&uuid_uid, cur_uid);
                        hammer_guid_to_uuid(&uuid_gid, cur_gid);
                        if (bcmp(&uuid_uid, &ip->ino_data.uid,
                                 sizeof(uuid_uid)) ||
                            bcmp(&uuid_gid, &ip->ino_data.gid,
                                 sizeof(uuid_gid)) ||
                            ip->ino_data.mode != cur_mode
                        ) {
                                ip->ino_data.uid = uuid_uid;
                                ip->ino_data.gid = uuid_gid;
                                ip->ino_data.mode = cur_mode;
                                ip->ino_data.ctime = trans.time;
                                modflags |= HAMMER_INODE_DDIRTY;
                        }
                        kflags |= NOTE_ATTRIB;
                }
        }
        while (vap->va_size != VNOVAL && ip->ino_data.size != vap->va_size) {
                switch(ap->a_vp->v_type) {
                case VREG:
                        if (vap->va_size == ip->ino_data.size)
                                break;
                        /*
                         * XXX break atomicy, we can deadlock the backend
                         * if we do not release the lock.  Probably not a
                         * big deal here.
                         */
                        blksize = hammer_blocksize(vap->va_size);
                        if (vap->va_size < ip->ino_data.size) {
                                vtruncbuf(ap->a_vp, vap->va_size, blksize);
                                truncating = 1;
                                kflags |= NOTE_WRITE;
                        } else {
                                vnode_pager_setsize(ap->a_vp, vap->va_size);
                                truncating = 0;
                                kflags |= NOTE_WRITE | NOTE_EXTEND;
                        }
                        ip->ino_data.size = vap->va_size;
                        ip->ino_data.mtime = trans.time;
                        modflags |= HAMMER_INODE_MTIME | HAMMER_INODE_DDIRTY;

                        /*
                         * on-media truncation is cached in the inode until
                         * the inode is synchronized.
                         */
                        if (truncating) {
                                hammer_ip_frontend_trunc(ip, vap->va_size);
#ifdef DEBUG_TRUNCATE
                                if (HammerTruncIp == NULL)
                                        HammerTruncIp = ip;
#endif
                                if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
                                        ip->flags |= HAMMER_INODE_TRUNCATED;
                                        ip->trunc_off = vap->va_size;
#ifdef DEBUG_TRUNCATE
                                        if (ip == HammerTruncIp)
                                        kprintf("truncate1 %016llx\n",
                                                (long long)ip->trunc_off);
#endif
                                } else if (ip->trunc_off > vap->va_size) {
                                        ip->trunc_off = vap->va_size;
#ifdef DEBUG_TRUNCATE
                                        if (ip == HammerTruncIp)
                                        kprintf("truncate2 %016llx\n",
                                                (long long)ip->trunc_off);
#endif
                                } else {
#ifdef DEBUG_TRUNCATE
                                        if (ip == HammerTruncIp)
                                        kprintf("truncate3 %016llx (ignored)\n",
                                                (long long)vap->va_size);
#endif
                                }
                        }

                        /*
                         * If truncating we have to clean out a portion of
                         * the last block on-disk.  We do this in the
                         * front-end buffer cache.
                         */
                        aligned_size = (vap->va_size + (blksize - 1)) &
                                       ~(int64_t)(blksize - 1);
                        if (truncating && vap->va_size < aligned_size) {
                                struct buf *bp;
                                int offset;

                                aligned_size -= blksize;

                                offset = (int)vap->va_size & (blksize - 1);
                                error = bread(ap->a_vp, aligned_size,
                                              blksize, &bp);
                                hammer_ip_frontend_trunc(ip, aligned_size);
                                if (error == 0) {
                                        bzero(bp->b_data + offset,
                                              blksize - offset);
                                        /* must de-cache direct-io offset */
                                        bp->b_bio2.bio_offset = NOOFFSET;
                                        bdwrite(bp);
                                } else {
                                        kprintf("ERROR %d\n", error);
                                        brelse(bp);
                                }
                        }
                        break;
                case VDATABASE:
                        if ((ip->flags & HAMMER_INODE_TRUNCATED) == 0) {
                                ip->flags |= HAMMER_INODE_TRUNCATED;
                                ip->trunc_off = vap->va_size;
                        } else if (ip->trunc_off > vap->va_size) {
                                ip->trunc_off = vap->va_size;
                        }
                        hammer_ip_frontend_trunc(ip, vap->va_size);
                        ip->ino_data.size = vap->va_size;
                        ip->ino_data.mtime = trans.time;
                        modflags |= HAMMER_INODE_MTIME | HAMMER_INODE_DDIRTY;
                        kflags |= NOTE_ATTRIB;
                        break;
                default:
                        error = EINVAL;
                        goto done;
                }
                break;
        }
        if (vap->va_atime.tv_sec != VNOVAL) {
                ip->ino_data.atime = hammer_timespec_to_time(&vap->va_atime);
                modflags |= HAMMER_INODE_ATIME;
                kflags |= NOTE_ATTRIB;
        }
        if (vap->va_mtime.tv_sec != VNOVAL) {
                ip->ino_data.mtime = hammer_timespec_to_time(&vap->va_mtime);
                modflags |= HAMMER_INODE_MTIME;
                kflags |= NOTE_ATTRIB;
        }
        if (vap->va_mode != (mode_t)VNOVAL) {
                mode_t   cur_mode = ip->ino_data.mode;
                uid_t cur_uid = hammer_to_unix_xid(&ip->ino_data.uid);
                gid_t cur_gid = hammer_to_unix_xid(&ip->ino_data.gid);

                error = vop_helper_chmod(ap->a_vp, vap->va_mode, ap->a_cred,
                                         cur_uid, cur_gid, &cur_mode);
                if (error == 0 && ip->ino_data.mode != cur_mode) {
                        ip->ino_data.mode = cur_mode;
                        ip->ino_data.ctime = trans.time;
                        modflags |= HAMMER_INODE_DDIRTY;
                        kflags |= NOTE_ATTRIB;
                }
        }
done:
        if (error == 0)
                hammer_modify_inode(ip, modflags);
        hammer_done_transaction(&trans);
        hammer_knote(ap->a_vp, kflags);
        return (error);
}

/*
 * hammer_vop_nsymlink { nch, dvp, vpp, cred, vap, target }
 */
static
int
hammer_vop_nsymlink(struct vop_nsymlink_args *ap)
{
        struct hammer_transaction trans;
        struct hammer_inode *dip;
        struct hammer_inode *nip;
        struct nchandle *nch;
        hammer_record_t record;
        int error;
        int bytes;

        ap->a_vap->va_type = VLNK;

        nch = ap->a_nch;
        dip = VTOI(ap->a_dvp);

        if (dip->flags & HAMMER_INODE_RO)
                return (EROFS);
        if ((error = hammer_checkspace(dip->hmp, HAMMER_CHKSPC_CREATE)) != 0)
                return (error);

        /*
         * Create a transaction to cover the operations we perform.
         */
        hammer_start_transaction(&trans, dip->hmp);
        ++hammer_stats_file_iopsw;

        /*
         * Create a new filesystem object of the requested type.  The
         * returned inode will be referenced but not locked.
         */

        error = hammer_create_inode(&trans, ap->a_vap, ap->a_cred,
                                    dip, nch->ncp->nc_name, nch->ncp->nc_nlen,
                                    NULL, &nip);
        if (error) {
                hammer_done_transaction(&trans);
                *ap->a_vpp = NULL;
                return (error);
        }

        /*
         * Add a record representing the symlink.  symlink stores the link
         * as pure data, not a string, and is no \0 terminated.
         */
        if (error == 0) {
                bytes = strlen(ap->a_target);

                if (bytes <= HAMMER_INODE_BASESYMLEN) {
                        bcopy(ap->a_target, nip->ino_data.ext.symlink, bytes);
                } else {
                        record = hammer_alloc_mem_record(nip, bytes);
                        record->type = HAMMER_MEM_RECORD_GENERAL;

                        record->leaf.base.localization = nip->obj_localization +
                                                         HAMMER_LOCALIZE_MISC;
                        record->leaf.base.key = HAMMER_FIXKEY_SYMLINK;
                        record->leaf.base.rec_type = HAMMER_RECTYPE_FIX;
                        record->leaf.data_len = bytes;
                        KKASSERT(HAMMER_SYMLINK_NAME_OFF == 0);
                        bcopy(ap->a_target, record->data->symlink.name, bytes);
                        error = hammer_ip_add_record(&trans, record);
                }

                /*
                 * Set the file size to the length of the link.
                 */
                if (error == 0) {
                        nip->ino_data.size = bytes;
                        hammer_modify_inode(nip, HAMMER_INODE_DDIRTY);
                }
        }
        if (error == 0)
                error = hammer_ip_add_directory(&trans, dip, nch->ncp->nc_name,
                                                nch->ncp->nc_nlen, nip);

        /*
         * Finish up.
         */
        if (error) {
                hammer_rel_inode(nip, 0);
                *ap->a_vpp = NULL;
        } else {
                error = hammer_get_vnode(nip, ap->a_vpp);
                hammer_rel_inode(nip, 0);
                if (error == 0) {
                        cache_setunresolved(ap->a_nch);
                        cache_setvp(ap->a_nch, *ap->a_vpp);
                        hammer_knote(ap->a_dvp, NOTE_WRITE);
                }
        }
        hammer_done_transaction(&trans);
        return (error);
}

/*
 * hammer_vop_nwhiteout { nch, dvp, cred, flags }
 */
static
int
hammer_vop_nwhiteout(struct vop_nwhiteout_args *ap)
{
        struct hammer_transaction trans;
        struct hammer_inode *dip;
        int error;

        dip = VTOI(ap->a_dvp);

        if (hammer_nohistory(dip) == 0 &&
            (error = hammer_checkspace(dip->hmp, HAMMER_CHKSPC_CREATE)) != 0) {
                return (error);
        }

        hammer_start_transaction(&trans, dip->hmp);
        ++hammer_stats_file_iopsw;
        error = hammer_dounlink(&trans, ap->a_nch, ap->a_dvp,
                                ap->a_cred, ap->a_flags, -1);
        hammer_done_transaction(&trans);

        return (error);
}

/*
 * hammer_vop_ioctl { vp, command, data, fflag, cred }
 */
static
int
hammer_vop_ioctl(struct vop_ioctl_args *ap)
{
        struct hammer_inode *ip = ap->a_vp->v_data;

        ++hammer_stats_file_iopsr;
        return(hammer_ioctl(ip, ap->a_command, ap->a_data,
                            ap->a_fflag, ap->a_cred));
}

static
int
hammer_vop_mountctl(struct vop_mountctl_args *ap)
{
        static const struct mountctl_opt extraopt[] = {
                { HMNT_NOHISTORY,       "nohistory" },
                { HMNT_MASTERID,        "master" },
                { 0, NULL}

        };
        struct hammer_mount *hmp;
        struct mount *mp;
        int usedbytes;
        int error;

        error = 0;
        usedbytes = 0;
        mp = ap->a_head.a_ops->head.vv_mount;
        KKASSERT(mp->mnt_data != NULL);
        hmp = (struct hammer_mount *)mp->mnt_data;

        switch(ap->a_op) {

        case MOUNTCTL_SET_EXPORT:
                if (ap->a_ctllen != sizeof(struct export_args))
                        error = EINVAL;
                else
                        error = hammer_vfs_export(mp, ap->a_op,
                                      (const struct export_args *)ap->a_ctl);
                break;
        case MOUNTCTL_MOUNTFLAGS:
        {
                /*
                 * Call standard mountctl VOP function
                 * so we get user mount flags.
                 */
                error = vop_stdmountctl(ap);
                if (error)
                        break;

                usedbytes = *ap->a_res;

                if (usedbytes > 0 && usedbytes < ap->a_buflen) {
                        usedbytes += vfs_flagstostr(hmp->hflags, extraopt, ap->a_buf,
                                                    ap->a_buflen - usedbytes,
                                                    &error);
                }

                *ap->a_res += usedbytes;
                break;
        }
        default:
                error = vop_stdmountctl(ap);
                break;
        }
        return(error);
}

/*
 * hammer_vop_strategy { vp, bio }
 *
 * Strategy call, used for regular file read & write only.  Note that the
 * bp may represent a cluster.
 *
 * To simplify operation and allow better optimizations in the future,
 * this code does not make any assumptions with regards to buffer alignment
 * or size.
 */
static
int
hammer_vop_strategy(struct vop_strategy_args *ap)
{
        struct buf *bp;
        int error;

        bp = ap->a_bio->bio_buf;

        switch(bp->b_cmd) {
        case BUF_CMD_READ:
                error = hammer_vop_strategy_read(ap);
                break;
        case BUF_CMD_WRITE:
                error = hammer_vop_strategy_write(ap);
                break;
        default:
                bp->b_error = error = EINVAL;
                bp->b_flags |= B_ERROR;
                biodone(ap->a_bio);
                break;
        }
        return (error);
}

/*
 * Read from a regular file.  Iterate the related records and fill in the
 * BIO/BUF.  Gaps are zero-filled.
 *
 * The support code in hammer_object.c should be used to deal with mixed
 * in-memory and on-disk records.
 *
 * NOTE: Can be called from the cluster code with an oversized buf.
 *
 * XXX atime update
 */
static
int
hammer_vop_strategy_read(struct vop_strategy_args *ap)
{
        struct hammer_transaction trans;
        struct hammer_inode *ip;
        struct hammer_inode *dip;
        struct hammer_cursor cursor;
        hammer_base_elm_t base;
        hammer_off_t disk_offset;
        struct bio *bio;
        struct bio *nbio;
        struct buf *bp;
        int64_t rec_offset;
        int64_t ran_end;
        int64_t tmp64;
        int error;
        int boff;
        int roff;
        int n;

        bio = ap->a_bio;
        bp = bio->bio_buf;
        ip = ap->a_vp->v_data;

        /*
         * The zone-2 disk offset may have been set by the cluster code via
         * a BMAP operation, or else should be NOOFFSET.
         *
         * Checking the high bits for a match against zone-2 should suffice.
         */
        nbio = push_bio(bio);
        if ((nbio->bio_offset & HAMMER_OFF_ZONE_MASK) ==
            HAMMER_ZONE_LARGE_DATA) {
                error = hammer_io_direct_read(ip->hmp, nbio, NULL);
                return (error);
        }

        /*
         * Well, that sucked.  Do it the hard way.  If all the stars are
         * aligned we may still be able to issue a direct-read.
         */
        hammer_simple_transaction(&trans, ip->hmp);
        hammer_init_cursor(&trans, &cursor, &ip->cache[1], ip);

        /*
         * Key range (begin and end inclusive) to scan.  Note that the key's
         * stored in the actual records represent BASE+LEN, not BASE.  The
         * first record containing bio_offset will have a key > bio_offset.
         */
        cursor.key_beg.localization = ip->obj_localization +
                                      HAMMER_LOCALIZE_MISC;
        cursor.key_beg.obj_id = ip->obj_id;
        cursor.key_beg.create_tid = 0;
        cursor.key_beg.delete_tid = 0;
        cursor.key_beg.obj_type = 0;
        cursor.key_beg.key = bio->bio_offset + 1;
        cursor.asof = ip->obj_asof;
        cursor.flags |= HAMMER_CURSOR_ASOF;

        cursor.key_end = cursor.key_beg;
        KKASSERT(ip->ino_data.obj_type == HAMMER_OBJTYPE_REGFILE);
#if 0
        if (ip->ino_data.obj_type == HAMMER_OBJTYPE_DBFILE) {
                cursor.key_beg.rec_type = HAMMER_RECTYPE_DB;
                cursor.key_end.rec_type = HAMMER_RECTYPE_DB;
                cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
        } else
#endif
        {
                ran_end = bio->bio_offset + bp->b_bufsize;
                cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
                cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
                tmp64 = ran_end + MAXPHYS + 1;  /* work-around GCC-4 bug */
                if (tmp64 < ran_end)
                        cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
                else
                        cursor.key_end.key = ran_end + MAXPHYS + 1;
        }
        cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;

        error = hammer_ip_first(&cursor);
        boff = 0;

        while (error == 0) {
                /*
                 * Get the base file offset of the record.  The key for
                 * data records is (base + bytes) rather then (base).
                 */
                base = &cursor.leaf->base;
                rec_offset = base->key - cursor.leaf->data_len;

                /*
                 * Calculate the gap, if any, and zero-fill it.
                 *
                 * n is the offset of the start of the record verses our
                 * current seek offset in the bio.
                 */
                n = (int)(rec_offset - (bio->bio_offset + boff));
                if (n > 0) {
                        if (n > bp->b_bufsize - boff)
                                n = bp->b_bufsize - boff;
                        bzero((char *)bp->b_data + boff, n);
                        boff += n;
                        n = 0;
                }

                /*
                 * Calculate the data offset in the record and the number
                 * of bytes we can copy.
                 *
                 * There are two degenerate cases.  First, boff may already
                 * be at bp->b_bufsize.  Secondly, the data offset within
                 * the record may exceed the record's size.
                 */
                roff = -n;
                rec_offset += roff;
                n = cursor.leaf->data_len - roff;
                if (n <= 0) {
                        kprintf("strategy_read: bad n=%d roff=%d\n", n, roff);
                        n = 0;
                } else if (n > bp->b_bufsize - boff) {
                        n = bp->b_bufsize - boff;
                }

                /*
                 * Deal with cached truncations.  This cool bit of code
                 * allows truncate()/ftruncate() to avoid having to sync
                 * the file.
                 *
                 * If the frontend is truncated then all backend records are
                 * subject to the frontend's truncation.
                 *
                 * If the backend is truncated then backend records on-disk
                 * (but not in-memory) are subject to the backend's
                 * truncation.  In-memory records owned by the backend
                 * represent data written after the truncation point on the
                 * backend and must not be truncated.
                 *
                 * Truncate operations deal with frontend buffer cache
                 * buffers and frontend-owned in-memory records synchronously.
                 */
                if (ip->flags & HAMMER_INODE_TRUNCATED) {
                        if (hammer_cursor_ondisk(&cursor) ||
                            cursor.iprec->flush_state == HAMMER_FST_FLUSH) {
                                if (ip->trunc_off <= rec_offset)
                                        n = 0;
                                else if (ip->trunc_off < rec_offset + n)
                                        n = (int)(ip->trunc_off - rec_offset);
                        }
                }
                if (ip->sync_flags & HAMMER_INODE_TRUNCATED) {
                        if (hammer_cursor_ondisk(&cursor)) {
                                if (ip->sync_trunc_off <= rec_offset)
                                        n = 0;
                                else if (ip->sync_trunc_off < rec_offset + n)
                                        n = (int)(ip->sync_trunc_off - rec_offset);
                        }
                }

                /*
                 * Try to issue a direct read into our bio if possible,
                 * otherwise resolve the element data into a hammer_buffer
                 * and copy.
                 *
                 * The buffer on-disk should be zerod past any real
                 * truncation point, but may not be for any synthesized
                 * truncation point from above.
                 */
                disk_offset = cursor.leaf->data_offset + roff;
                if (boff == 0 && n == bp->b_bufsize &&
                    hammer_cursor_ondisk(&cursor) &&
                    (disk_offset & HAMMER_BUFMASK) == 0) {
                        KKASSERT((disk_offset & HAMMER_OFF_ZONE_MASK) ==
                                 HAMMER_ZONE_LARGE_DATA);
                        nbio->bio_offset = disk_offset;
                        error = hammer_io_direct_read(trans.hmp, nbio,
                                                      cursor.leaf);
                        goto done;
                } else if (n) {
                        error = hammer_ip_resolve_data(&cursor);
                        if (error == 0) {
                                bcopy((char *)cursor.data + roff,
                                      (char *)bp->b_data + boff, n);
                        }
                }
                if (error)
                        break;

                /*
                 * Iterate until we have filled the request.
                 */
                boff += n;
                if (boff == bp->b_bufsize)
                        break;
                error = hammer_ip_next(&cursor);
        }

        /*
         * There may have been a gap after the last record
         */
        if (error == ENOENT)
                error = 0;
        if (error == 0 && boff != bp->b_bufsize) {
                KKASSERT(boff < bp->b_bufsize);
                bzero((char *)bp->b_data + boff, bp->b_bufsize - boff);
                /* boff = bp->b_bufsize; */
        }
        bp->b_resid = 0;
        bp->b_error = error;
        if (error)
                bp->b_flags |= B_ERROR;
        biodone(ap->a_bio);

done:
        /*
         * Cache the b-tree node for the last data read in cache[1].
         *
         * If we hit the file EOF then also cache the node in the
         * governing director's cache[3], it will be used to initialize
         * the inode's cache[1] for any inodes looked up via the directory.
         *
         * This doesn't reduce disk accesses since the B-Tree chain is
         * likely cached, but it does reduce cpu overhead when looking
         * up file offsets for cpdup/tar/cpio style iterations.
         */
        if (cursor.node)
                hammer_cache_node(&ip->cache[1], cursor.node);
        if (ran_end >= ip->ino_data.size) {
                dip = hammer_find_inode(&trans, ip->ino_data.parent_obj_id,
                                        ip->obj_asof, ip->obj_localization);
                if (dip) {
                        hammer_cache_node(&dip->cache[3], cursor.node);
                        hammer_rel_inode(dip, 0);
                }
        }
        hammer_done_cursor(&cursor);
        hammer_done_transaction(&trans);
        return(error);
}

/*
 * BMAP operation - used to support cluster_read() only.
 *
 * (struct vnode *vp, off_t loffset, off_t *doffsetp, int *runp, int *runb)
 *
 * This routine may return EOPNOTSUPP if the opration is not supported for
 * the specified offset.  The contents of the pointer arguments do not
 * need to be initialized in that case. 
 *
 * If a disk address is available and properly aligned return 0 with 
 * *doffsetp set to the zone-2 address, and *runp / *runb set appropriately
 * to the run-length relative to that offset.  Callers may assume that
 * *doffsetp is valid if 0 is returned, even if *runp is not sufficiently
 * large, so return EOPNOTSUPP if it is not sufficiently large.
 */
static
int
hammer_vop_bmap(struct vop_bmap_args *ap)
{
        struct hammer_transaction trans;
        struct hammer_inode *ip;
        struct hammer_cursor cursor;
        hammer_base_elm_t base;
        int64_t rec_offset;
        int64_t ran_end;
        int64_t tmp64;
        int64_t base_offset;
        int64_t base_disk_offset;
        int64_t last_offset;
        hammer_off_t last_disk_offset;
        hammer_off_t disk_offset;
        int     rec_len;
        int     error;
        int     blksize;

        ++hammer_stats_file_iopsr;
        ip = ap->a_vp->v_data;

        /*
         * We can only BMAP regular files.  We can't BMAP database files,
         * directories, etc.
         */
        if (ip->ino_data.obj_type != HAMMER_OBJTYPE_REGFILE)
                return(EOPNOTSUPP);

        /*
         * bmap is typically called with runp/runb both NULL when used
         * for writing.  We do not support BMAP for writing atm.
         */
        if (ap->a_cmd != BUF_CMD_READ)
                return(EOPNOTSUPP);

        /*
         * Scan the B-Tree to acquire blockmap addresses, then translate
         * to raw addresses.
         */
        hammer_simple_transaction(&trans, ip->hmp);
#if 0
        kprintf("bmap_beg %016llx ip->cache %p\n",
                (long long)ap->a_loffset, ip->cache[1]);
#endif
        hammer_init_cursor(&trans, &cursor, &ip->cache[1], ip);

        /*
         * Key range (begin and end inclusive) to scan.  Note that the key's
         * stored in the actual records represent BASE+LEN, not BASE.  The
         * first record containing bio_offset will have a key > bio_offset.
         */
        cursor.key_beg.localization = ip->obj_localization +
                                      HAMMER_LOCALIZE_MISC;
        cursor.key_beg.obj_id = ip->obj_id;
        cursor.key_beg.create_tid = 0;
        cursor.key_beg.delete_tid = 0;
        cursor.key_beg.obj_type = 0;
        if (ap->a_runb)
                cursor.key_beg.key = ap->a_loffset - MAXPHYS + 1;
        else
                cursor.key_beg.key = ap->a_loffset + 1;
        if (cursor.key_beg.key < 0)
                cursor.key_beg.key = 0;
        cursor.asof = ip->obj_asof;
        cursor.flags |= HAMMER_CURSOR_ASOF;

        cursor.key_end = cursor.key_beg;
        KKASSERT(ip->ino_data.obj_type == HAMMER_OBJTYPE_REGFILE);

        ran_end = ap->a_loffset + MAXPHYS;
        cursor.key_beg.rec_type = HAMMER_RECTYPE_DATA;
        cursor.key_end.rec_type = HAMMER_RECTYPE_DATA;
        tmp64 = ran_end + MAXPHYS + 1;  /* work-around GCC-4 bug */
        if (tmp64 < ran_end)
                cursor.key_end.key = 0x7FFFFFFFFFFFFFFFLL;
        else
                cursor.key_end.key = ran_end + MAXPHYS + 1;

        cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE;

        error = hammer_ip_first(&cursor);
        base_offset = last_offset = 0;
        base_disk_offset = last_disk_offset = 0;

        while (error == 0) {
                /*
                 * Get the base file offset of the record.  The key for
                 * data records is (base + bytes) rather then (base).
                 *
                 * NOTE: rec_offset + rec_len may exceed the end-of-file.
                 * The extra bytes should be zero on-disk and the BMAP op
                 * should still be ok.
                 */
                base = &cursor.leaf->base;
                rec_offset = base->key - cursor.leaf->data_len;
                rec_len    = cursor.leaf->data_len;

                /*
                 * Incorporate any cached truncation.
                 *
                 * NOTE: Modifications to rec_len based on synthesized
                 * truncation points remove the guarantee that any extended
                 * data on disk is zero (since the truncations may not have
                 * taken place on-media yet).
                 */
                if (ip->flags & HAMMER_INODE_TRUNCATED) {
                        if (hammer_cursor_ondisk(&cursor) ||
                            cursor.iprec->flush_state == HAMMER_FST_FLUSH) {
                                if (ip->trunc_off <= rec_offset)
                                        rec_len = 0;
                                else if (ip->trunc_off < rec_offset + rec_len)
                                        rec_len = (int)(ip->trunc_off - rec_offset);
                        }
                }
                if (ip->sync_flags & HAMMER_INODE_TRUNCATED) {
                        if (hammer_cursor_ondisk(&cursor)) {
                                if (ip->sync_trunc_off <= rec_offset)
                                        rec_len = 0;
                                else if (ip->sync_trunc_off < rec_offset + rec_len)
                                        rec_len = (int)(ip->sync_trunc_off - rec_offset);
                        }
                }

                /*
                 * Accumulate information.  If we have hit a discontiguous
                 * block reset base_offset unless we are already beyond the
                 * requested offset.  If we are, that's it, we stop.
                 */
                if (error)
                        break;
                if (hammer_cursor_ondisk(&cursor)) {
                        disk_offset = cursor.leaf->data_offset;
                        if (rec_offset != last_offset ||
                            disk_offset != last_disk_offset) {
                                if (rec_offset > ap->a_loffset)
                                        break;
                                base_offset = rec_offset;
                                base_disk_offset = disk_offset;
                        }
                        last_offset = rec_offset + rec_len;
                        last_disk_offset = disk_offset + rec_len;
                }
                error = hammer_ip_next(&cursor);
        }

#if 0
        kprintf("BMAP %016llx:  %016llx - %016llx\n",
                (long long)ap->a_loffset,
                (long long)base_offset,
                (long long)last_offset);
        kprintf("BMAP %16s:  %016llx - %016llx\n", "",
                (long long)base_disk_offset,
                (long long)last_disk_offset);
#endif

        if (cursor.node) {
                hammer_cache_node(&ip->cache[1], cursor.node);
#if 0
                kprintf("bmap_end2 %016llx ip->cache %p\n",
                        (long long)ap->a_loffset, ip->cache[1]);
#endif
        }
        hammer_done_cursor(&cursor);
        hammer_done_transaction(&trans);

        /*
         * If we couldn't find any records or the records we did find were
         * all behind the requested offset, return failure.  A forward
         * truncation can leave a hole w/ no on-disk records.
         */
        if (last_offset == 0 || last_offset < ap->a_loffset)
                return (EOPNOTSUPP);

        /*
         * Figure out the block size at the requested offset and adjust
         * our limits so the cluster_read() does not create inappropriately
         * sized buffer cache buffers.
         */
        blksize = hammer_blocksize(ap->a_loffset);
        if (hammer_blocksize(base_offset) != blksize) {
                base_offset = hammer_blockdemarc(base_offset, ap->a_loffset);
        }
        if (last_offset != ap->a_loffset &&
            hammer_blocksize(last_offset - 1) != blksize) {
                last_offset = hammer_blockdemarc(ap->a_loffset,
                                                 last_offset - 1);
        }

        /*
         * Returning EOPNOTSUPP simply prevents the direct-IO optimization
         * from occuring.
         */
        disk_offset = base_disk_offset + (ap->a_loffset - base_offset);

        if ((disk_offset & HAMMER_OFF_ZONE_MASK) != HAMMER_ZONE_LARGE_DATA) {
                /*
                 * Only large-data zones can be direct-IOd
                 */
                error = EOPNOTSUPP;
        } else if ((disk_offset & HAMMER_BUFMASK) ||
                   (last_offset - ap->a_loffset) < blksize) {
                /*
                 * doffsetp is not aligned or the forward run size does
                 * not cover a whole buffer, disallow the direct I/O.
                 */
                error = EOPNOTSUPP;
        } else {
                /*
                 * We're good.
                 */
                *ap->a_doffsetp = disk_offset;
                if (ap->a_runb) {
                        *ap->a_runb = ap->a_loffset - base_offset;
                        KKASSERT(*ap->a_runb >= 0);
                }
                if (ap->a_runp) {
                        *ap->a_runp = last_offset - ap->a_loffset;
                        KKASSERT(*ap->a_runp >= 0);
                }
                error = 0;
        }
        return(error);
}

/*
 * Write to a regular file.   Because this is a strategy call the OS is
 * trying to actually get data onto the media.
 */
static
int
hammer_vop_strategy_write(struct vop_strategy_args *ap)
{
        hammer_record_t record;
        hammer_mount_t hmp;
        hammer_inode_t ip;
        struct bio *bio;
        struct buf *bp;
        int blksize;
        int bytes;
        int error;

        bio = ap->a_bio;
        bp = bio->bio_buf;
        ip = ap->a_vp->v_data;
        hmp = ip->hmp;

        blksize = hammer_blocksize(bio->bio_offset);
        KKASSERT(bp->b_bufsize == blksize);

        if (ip->flags & HAMMER_INODE_RO) {
                bp->b_error = EROFS;
                bp->b_flags |= B_ERROR;
                biodone(ap->a_bio);
                return(EROFS);
        }

        /*
         * Interlock with inode destruction (no in-kernel or directory
         * topology visibility).  If we queue new IO while trying to
         * destroy the inode we can deadlock the vtrunc call in
         * hammer_inode_unloadable_check().
         *
         * Besides, there's no point flushing a bp associated with an
         * inode that is being destroyed on-media and has no kernel
         * references.
         */
        if ((ip->flags | ip->sync_flags) &
            (HAMMER_INODE_DELETING|HAMMER_INODE_DELETED)) {
                bp->b_resid = 0;
                biodone(ap->a_bio);
                return(0);
        }

        /*
         * Reserve space and issue a direct-write from the front-end. 
         * NOTE: The direct_io code will hammer_bread/bcopy smaller
         * allocations.
         *
         * An in-memory record will be installed to reference the storage
         * until the flusher can get to it.
         *
         * Since we own the high level bio the front-end will not try to
         * do a direct-read until the write completes.
         *
         * NOTE: The only time we do not reserve a full-sized buffers
         * worth of data is if the file is small.  We do not try to
         * allocate a fragment (from the small-data zone) at the end of
         * an otherwise large file as this can lead to wildly separated
         * data.
         */
        KKASSERT((bio->bio_offset & HAMMER_BUFMASK) == 0);
        KKASSERT(bio->bio_offset < ip->ino_data.size);
        if (bio->bio_offset || ip->ino_data.size > HAMMER_BUFSIZE / 2)
                bytes = bp->b_bufsize;
        else
                bytes = ((int)ip->ino_data.size + 15) & ~15;

        record = hammer_ip_add_bulk(ip, bio->bio_offset, bp->b_data,
                                    bytes, &error);
        if (record) {
                hammer_io_direct_write(hmp, record, bio);
                if (ip->rsv_recs > 1 && hmp->rsv_recs > hammer_limit_recs)
                        hammer_flush_inode(ip, 0);
        } else {
                bp->b_bio2.bio_offset = NOOFFSET;
                bp->b_error = error;
                bp->b_flags |= B_ERROR;
                biodone(ap->a_bio);
        }
        return(error);
}

/*
 * dounlink - disconnect a directory entry
 *
 * XXX whiteout support not really in yet
 */
static int
hammer_dounlink(hammer_transaction_t trans, struct nchandle *nch,
                struct vnode *dvp, struct ucred *cred, 
                int flags, int isdir)
{
        struct namecache *ncp;
        hammer_inode_t dip;
        hammer_inode_t ip;
        struct hammer_cursor cursor;
        int64_t namekey;
        u_int32_t max_iterations;
        int nlen, error;

        /*
         * Calculate the namekey and setup the key range for the scan.  This
         * works kinda like a chained hash table where the lower 32 bits
         * of the namekey synthesize the chain.
         *
         * The key range is inclusive of both key_beg and key_end.
         */
        dip = VTOI(dvp);
        ncp = nch->ncp;

        if (dip->flags & HAMMER_INODE_RO)
                return (EROFS);

        namekey = hammer_directory_namekey(dip, ncp->nc_name, ncp->nc_nlen,
                                           &max_iterations);
retry:
        hammer_init_cursor(trans, &cursor, &dip->cache[1], dip);
        cursor.key_beg.localization = dip->obj_localization +
                                      hammer_dir_localization(dip);
        cursor.key_beg.obj_id = dip->obj_id;
        cursor.key_beg.key = namekey;
        cursor.key_beg.create_tid = 0;
        cursor.key_beg.delete_tid = 0;
        cursor.key_beg.rec_type = HAMMER_RECTYPE_DIRENTRY;
        cursor.key_beg.obj_type = 0;

        cursor.key_end = cursor.key_beg;
        cursor.key_end.key += max_iterations;
        cursor.asof = dip->obj_asof;
        cursor.flags |= HAMMER_CURSOR_END_INCLUSIVE | HAMMER_CURSOR_ASOF;

        /*
         * Scan all matching records (the chain), locate the one matching
         * the requested path component.  info->last_error contains the
         * error code on search termination and could be 0, ENOENT, or
         * something else.
         *
         * The hammer_ip_*() functions merge in-memory records with on-disk
         * records for the purposes of the search.
         */
        error = hammer_ip_first(&cursor);

        while (error == 0) {
                error = hammer_ip_resolve_data(&cursor);
                if (error)
                        break;
                nlen = cursor.leaf->data_len - HAMMER_ENTRY_NAME_OFF;
                KKASSERT(nlen > 0);
                if (ncp->nc_nlen == nlen &&
                    bcmp(ncp->nc_name, cursor.data->entry.name, nlen) == 0) {
                        break;
                }
                error = hammer_ip_next(&cursor);
        }

        /*
         * If all is ok we have to get the inode so we can adjust nlinks.
         * To avoid a deadlock with the flusher we must release the inode
         * lock on the directory when acquiring the inode for the entry.
         *
         * If the target is a directory, it must be empty.
         */
        if (error == 0) {
                hammer_unlock(&cursor.ip->lock);
                ip = hammer_get_inode(trans, dip, cursor.data->entry.obj_id,
                                      dip->hmp->asof,
                                      cursor.data->entry.localization,
                                      0, &error);
                hammer_lock_sh(&cursor.ip->lock);
                if (error == ENOENT) {
                        kprintf("HAMMER: WARNING: Removing "
                                "dirent w/missing inode \"%s\"\n"
                                "\tobj_id = %016llx\n",
                                ncp->nc_name,
                                (long long)cursor.data->entry.obj_id);
                        error = 0;
                }

                /*
                 * If isdir >= 0 we validate that the entry is or is not a
                 * directory.  If isdir < 0 we don't care.
                 */
                if (error == 0 && isdir >= 0 && ip) {
                        if (isdir &&
                            ip->ino_data.obj_type != HAMMER_OBJTYPE_DIRECTORY) {
                                error = ENOTDIR;
                        } else if (isdir == 0 &&
                            ip->ino_data.obj_type == HAMMER_OBJTYPE_DIRECTORY) {
                                error = EISDIR;
                        }
                }

                /*
                 * If we are trying to remove a directory the directory must
                 * be empty.
                 *
                 * The check directory code can loop and deadlock/retry.  Our
                 * own cursor's node locks must be released to avoid a 3-way
                 * deadlock with the flusher if the check directory code
                 * blocks.
                 *
                 * If any changes whatsoever have been made to the cursor
                 * set EDEADLK and retry.
                 */
                if (error == 0 && ip && ip->ino_data.obj_type ==
                                        HAMMER_OBJTYPE_DIRECTORY) {
                        hammer_unlock_cursor(&cursor);
                        error = hammer_ip_check_directory_empty(trans, ip);
                        hammer_lock_cursor(&cursor);
                        if (cursor.flags & HAMMER_CURSOR_RETEST) {
                                kprintf("HAMMER: Warning: avoided deadlock "
                                        "on rmdir '%s'\n",
                                        ncp->nc_name);
                                error = EDEADLK;
                        }
                }

                /*
                 * Delete the directory entry.
                 *
                 * WARNING: hammer_ip_del_directory() may have to terminate
                 * the cursor to avoid a deadlock.  It is ok to call
                 * hammer_done_cursor() twice.
                 */
                if (error == 0) {
                        error = hammer_ip_del_directory(trans, &cursor,
                                                        dip, ip);
                }
                hammer_done_cursor(&cursor);
                if (error == 0) {
                        cache_setunresolved(nch);
                        cache_setvp(nch, NULL);
                        /* XXX locking */
                        if (ip && ip->vp) {
                                hammer_knote(ip->vp, NOTE_DELETE);
                                cache_inval_vp(ip->vp, CINV_DESTROY);
                        }
                }
                if (ip)
                        hammer_rel_inode(ip, 0);
        } else {
                hammer_done_cursor(&cursor);
        }
        if (error == EDEADLK)
                goto retry;

        return (error);
}

/************************************************************************
 *                          FIFO AND SPECFS OPS                         *
 ************************************************************************
 *
 */

static int
hammer_vop_fifoclose (struct vop_close_args *ap)
{
        /* XXX update itimes */
        return (VOCALL(&fifo_vnode_vops, &ap->a_head));
}

static int
hammer_vop_fiforead (struct vop_read_args *ap)
{
        int error;

        error = VOCALL(&fifo_vnode_vops, &ap->a_head);
        /* XXX update access time */
        return (error);
}

static int
hammer_vop_fifowrite (struct vop_write_args *ap)
{
        int error;

        error = VOCALL(&fifo_vnode_vops, &ap->a_head);
        /* XXX update access time */
        return (error);
}

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

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

/************************************************************************
 *                          KQFILTER OPS                                *
 ************************************************************************
 *
 */
static void filt_hammerdetach(struct knote *kn);
static int filt_hammerread(struct knote *kn, long hint);
static int filt_hammerwrite(struct knote *kn, long hint);
static int filt_hammervnode(struct knote *kn, long hint);

static struct filterops hammerread_filtops =
        { 1, NULL, filt_hammerdetach, filt_hammerread };
static struct filterops hammerwrite_filtops =
        { 1, NULL, filt_hammerdetach, filt_hammerwrite };
static struct filterops hammervnode_filtops =
        { 1, NULL, filt_hammerdetach, filt_hammervnode };

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

        switch (kn->kn_filter) {
        case EVFILT_READ:
                kn->kn_fop = &hammerread_filtops;
                break;
        case EVFILT_WRITE:
                kn->kn_fop = &hammerwrite_filtops;
                break;
        case EVFILT_VNODE:
                kn->kn_fop = &hammervnode_filtops;
                break;
        default:
                return (1);
        }

        kn->kn_hook = (caddr_t)vp;

        lwkt_gettoken(&vlock, &vp->v_token);
        SLIST_INSERT_HEAD(&vp->v_pollinfo.vpi_selinfo.si_note, kn, kn_selnext);
        lwkt_reltoken(&vlock);

        return(0);
}

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

        lwkt_gettoken(&vlock, &vp->v_token);
        SLIST_REMOVE(&vp->v_pollinfo.vpi_selinfo.si_note,
                     kn, knote, kn_selnext);
        lwkt_reltoken(&vlock);
}

static int
filt_hammerread(struct knote *kn, long hint)
{
        struct vnode *vp = (void *)kn->kn_hook;
        hammer_inode_t ip = VTOI(vp);

        if (hint == NOTE_REVOKE) {
                kn->kn_flags |= (EV_EOF | EV_ONESHOT);
                return(1);
        }
        kn->kn_data = ip->ino_data.size - kn->kn_fp->f_offset;
        return (kn->kn_data != 0);
}

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

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