xnu-1699.22.81.tar.gz

[apple/xnu.git] / bsd / hfs / hfs_btreeio.c

/*
 * Copyright (c) 2000-2010 Apple Inc. All rights reserved.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
 * 
 * This file contains Original Code and/or Modifications of Original Code
 * as defined in and that are subject to the Apple Public Source License
 * Version 2.0 (the 'License'). You may not use this file except in
 * compliance with the License. The rights granted to you under the License
 * may not be used to create, or enable the creation or redistribution of,
 * unlawful or unlicensed copies of an Apple operating system, or to
 * circumvent, violate, or enable the circumvention or violation of, any
 * terms of an Apple operating system software license agreement.
 * 
 * Please obtain a copy of the License at
 * http://www.opensource.apple.com/apsl/ and read it before using this file.
 * 
 * The Original Code and all software distributed under the License are
 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
 * Please see the License for the specific language governing rights and
 * limitations under the License.
 * 
 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
 */

#include <sys/param.h>
#include <sys/systm.h>
#include <sys/buf.h>
#include <sys/buf_internal.h>
#include <sys/kernel.h>
#include <sys/malloc.h>
#include <sys/mount.h>
#include <sys/vnode.h>


#include "hfs.h"
#include "hfs_cnode.h"
#include "hfs_dbg.h"
#include "hfs_endian.h"
#include "hfs_btreeio.h"

#include "hfscommon/headers/FileMgrInternal.h"
#include "hfscommon/headers/BTreesPrivate.h"

#define FORCESYNCBTREEWRITES 0

/* From bsd/vfs/vfs_bio.c */
extern int bdwrite_internal(struct buf *, int);

static int ClearBTNodes(struct vnode *vp, long blksize, off_t offset, off_t amount);
static int btree_journal_modify_block_end(struct hfsmount *hfsmp, struct buf *bp);

void btree_swap_node(struct buf *bp, __unused void *arg);

OSStatus SetBTreeBlockSize(FileReference vp, ByteCount blockSize, __unused ItemCount minBlockCount)
{
        BTreeControlBlockPtr    bTreePtr;
        
        DBG_ASSERT(vp != NULL);
        DBG_ASSERT(blockSize >= kMinNodeSize);
    if (blockSize > MAXBSIZE )
        return (fsBTBadNodeSize);

        bTreePtr = (BTreeControlBlockPtr)VTOF(vp)->fcbBTCBPtr;
        bTreePtr->nodeSize = blockSize;
        
    return (E_NONE);
}


OSStatus GetBTreeBlock(FileReference vp, u_int32_t blockNum, GetBlockOptions options, BlockDescriptor *block)
{
    OSStatus     retval = E_NONE;
    struct buf   *bp = NULL;
        u_int8_t     allow_empty_node;    

        /* If the btree block is being read using hint, it is 
         * fine for the swap code to find zeroed out nodes. 
         */
        if (options & kGetBlockHint) {
                        allow_empty_node = true;
        } else {
                        allow_empty_node = false;
        }

    if (options & kGetEmptyBlock) {
        daddr64_t blkno;
        off_t offset;

        offset = (daddr64_t)blockNum * (daddr64_t)block->blockSize;
        bp = buf_getblk(vp, (daddr64_t)blockNum, block->blockSize, 0, 0, BLK_META);
        if (bp &&
            VNOP_BLOCKMAP(vp, offset, block->blockSize, &blkno, NULL, NULL, 0, NULL) == 0) {
            buf_setblkno(bp, blkno);
        }
    } else {
        retval = buf_meta_bread(vp, (daddr64_t)blockNum, block->blockSize, NOCRED, &bp);
    }
    if (bp == NULL)
        retval = -1;    //XXX need better error

    if (retval == E_NONE) {
        block->blockHeader = bp;
        block->buffer = (char *)buf_dataptr(bp);
        block->blockNum = buf_lblkno(bp);
        block->blockReadFromDisk = (buf_fromcache(bp) == 0);    /* not found in cache ==> came from disk */

                // XXXdbg 
                block->isModified = 0;

        /* Check and endian swap B-Tree node (only if it's a valid block) */
        if (!(options & kGetEmptyBlock)) {
            /* This happens when we first open the b-tree, we might not have all the node data on hand */
            if ((((BTNodeDescriptor *)block->buffer)->kind == kBTHeaderNode) &&
                (((BTHeaderRec *)((char *)block->buffer + 14))->nodeSize != buf_count(bp)) &&
                (SWAP_BE16 (((BTHeaderRec *)((char *)block->buffer + 14))->nodeSize) != buf_count(bp))) {

                /*
                 * Don't swap the node descriptor, record offsets, or other records.
                 * This record will be invalidated and re-read with the correct node
                 * size once the B-tree control block is set up with the node size
                 * from the header record.
                 */
                retval = hfs_swap_BTNode (block, vp, kSwapBTNodeHeaderRecordOnly, allow_empty_node);

                        } else if (block->blockReadFromDisk) {
                /*
                 * The node was just read from disk, so always swap/check it.
                 * This is necessary on big endian since the test below won't trigger.
                 */
                retval = hfs_swap_BTNode (block, vp, kSwapBTNodeBigToHost, allow_empty_node);
            } else if (*((u_int16_t *)((char *)block->buffer + (block->blockSize - sizeof (u_int16_t)))) == 0x0e00) {
                                /*
                                 * The node was left in the cache in non-native order, so swap it.
                                 * This only happens on little endian, after the node is written
                                 * back to disk.
                                 */
                retval = hfs_swap_BTNode (block, vp, kSwapBTNodeBigToHost, allow_empty_node);
            }
            
                /*
                 * If we got an error, then the node is only partially swapped.
                 * We mark the buffer invalid so that the next attempt to get the
                 * node will read it and attempt to swap again, and will notice
                 * the error again.  If we didn't do this, the next attempt to get
                 * the node might use the partially swapped node as-is.
                 */
            if (retval)
                                buf_markinvalid(bp);
        }
    }
    
    if (retval) {
        if (bp)
                        buf_brelse(bp);
        block->blockHeader = NULL;
        block->buffer = NULL;
    }

    return (retval);
}


void ModifyBlockStart(FileReference vp, BlockDescPtr blockPtr)
{
        struct hfsmount *hfsmp = VTOHFS(vp);
    struct buf *bp = NULL;

        if (hfsmp->jnl == NULL) {
                return;
        }
        
    bp = (struct buf *) blockPtr->blockHeader;
    if (bp == NULL) {
                panic("hfs: ModifyBlockStart: null bp  for blockdescptr %p?!?\n", blockPtr);
                return;
    }

        journal_modify_block_start(hfsmp->jnl, bp);
        blockPtr->isModified = 1;
}

void
btree_swap_node(struct buf *bp, __unused void *arg)
{
    //  struct hfsmount *hfsmp = (struct hfsmount *)arg;
        int retval;
    struct vnode *vp = buf_vnode(bp);
    BlockDescriptor block;
                                    
    /* Prepare the block pointer */
    block.blockHeader = bp;
    block.buffer = (char *)buf_dataptr(bp);
    block.blockNum = buf_lblkno(bp);
    /* not found in cache ==> came from disk */
    block.blockReadFromDisk = (buf_fromcache(bp) == 0);
    block.blockSize = buf_count(bp);

    /* Swap the data now that this node is ready to go to disk.
     * We allow swapping of zeroed out nodes here because we might
     * be writing node whose last record just got deleted.
     */
    retval = hfs_swap_BTNode (&block, vp, kSwapBTNodeHostToBig, true);
    if (retval)
        panic("hfs: btree_swap_node: about to write corrupt node!\n");
}


static int
btree_journal_modify_block_end(struct hfsmount *hfsmp, struct buf *bp)
{
    return journal_modify_block_end(hfsmp->jnl, bp, btree_swap_node, hfsmp);
}


OSStatus ReleaseBTreeBlock(FileReference vp, BlockDescPtr blockPtr, ReleaseBlockOptions options)
{
    struct hfsmount     *hfsmp = VTOHFS(vp);
    OSStatus    retval = E_NONE;
    struct buf *bp = NULL;

    bp = (struct buf *) blockPtr->blockHeader;

    if (bp == NULL) {
        retval = -1;
        goto exit;
    }

    if (options & kTrashBlock) {
                buf_markinvalid(bp);

                if (hfsmp->jnl && (buf_flags(bp) & B_LOCKED)) {
                        journal_kill_block(hfsmp->jnl, bp);
                } else {
                        buf_brelse(bp); /* note: B-tree code will clear blockPtr->blockHeader and blockPtr->buffer */
                }
                
                /* Don't let anyone else try to use this bp, it's been consumed */
                blockPtr->blockHeader = NULL;
                
    } else {
        if (options & kForceWriteBlock) {
                        if (hfsmp->jnl) {
                                if (blockPtr->isModified == 0) {
                                        panic("hfs: releaseblock: modified is 0 but forcewrite set! bp %p\n", bp);
                                }

                                retval = btree_journal_modify_block_end(hfsmp, bp);
                                blockPtr->isModified = 0;
                        } else {
                                retval = VNOP_BWRITE(bp);
                        }
                        
                        /* Don't let anyone else try to use this bp, it's been consumed */
                        blockPtr->blockHeader = NULL;
                        
        } else if (options & kMarkBlockDirty) {
                        struct timeval tv;
                        microuptime(&tv);
            if ((options & kLockTransaction) && hfsmp->jnl == NULL) {
                /*
                 *
                 * Set the B_LOCKED flag and unlock the buffer, causing buf_brelse to move
                 * the buffer onto the LOCKED free list.  This is necessary, otherwise
                 * getnewbuf() would try to reclaim the buffers using buf_bawrite, which
                 * isn't going to work.
                 *
                 */
                /* Don't hog all the buffers... */
                if (count_lock_queue() > kMaxLockedMetaBuffers) {
                     hfs_btsync(vp, HFS_SYNCTRANS);
                     /* Rollback sync time to cause a sync on lock release... */
                     (void) BTSetLastSync(VTOF(vp), tv.tv_sec - (kMaxSecsForFsync + 1));
                }
                buf_setflags(bp, B_LOCKED);
            }

            /* 
             * Delay-write this block.
             * If the maximum delayed buffers has been exceeded then
             * free up some buffers and fall back to an asynchronous write.
             */
                        if (hfsmp->jnl) {
                                if (blockPtr->isModified == 0) {
                                        panic("hfs: releaseblock: modified is 0 but markdirty set! bp %p\n", bp);
                                }
                                retval = btree_journal_modify_block_end(hfsmp, bp);
                                blockPtr->isModified = 0;
                        } else if (bdwrite_internal(bp, 1) != 0) {
                hfs_btsync(vp, 0);
                /* Rollback sync time to cause a sync on lock release... */
                (void) BTSetLastSync(VTOF(vp), tv.tv_sec - (kMaxSecsForFsync + 1));

                buf_clearflags(bp, B_LOCKED);
                buf_bawrite(bp);
            }
            
            /* Don't let anyone else try to use this bp, it's been consumed */
                        blockPtr->blockHeader = NULL;
                        
        } else {
                        // check if we had previously called journal_modify_block_start() 
                        // on this block and if so, abort it (which will call buf_brelse()).
                        if (hfsmp->jnl && blockPtr->isModified) {
                                // XXXdbg - I don't want to call modify_block_abort()
                                //          because I think it may be screwing up the
                                //          journal and blowing away a block that has
                                //          valid data in it.
                                //   
                                //    journal_modify_block_abort(hfsmp->jnl, bp);
                                //panic("hfs: releaseblock called for 0x%x but mod_block_start previously called.\n", bp);
                                btree_journal_modify_block_end(hfsmp, bp);
                                blockPtr->isModified = 0;
                        } else {
                                buf_brelse(bp); /* note: B-tree code will clear blockPtr->blockHeader and blockPtr->buffer */
                        }
                        
                        /* Don't let anyone else try to use this bp, it's been consumed */
                        blockPtr->blockHeader = NULL;
        }
    }

exit:
    return (retval);
}


OSStatus ExtendBTreeFile(FileReference vp, FSSize minEOF, FSSize maxEOF)
{
#pragma unused (maxEOF)

        OSStatus        retval = 0, ret = 0;
        int64_t         actualBytesAdded, origSize;
        u_int64_t       bytesToAdd;
        u_int32_t       startAllocation;
        u_int32_t       fileblocks;
        BTreeInfoRec    btInfo;
        ExtendedVCB     *vcb;
        FCB             *filePtr;
        struct proc     *p = NULL;
        int64_t         trim = 0;
        int             lockflags = 0;

        filePtr = GetFileControlBlock(vp);

        if ( (off_t)minEOF > filePtr->fcbEOF )
        {
                bytesToAdd = minEOF - filePtr->fcbEOF;

                if (bytesToAdd < filePtr->ff_clumpsize)
                        bytesToAdd = filePtr->ff_clumpsize;             //XXX why not always be a mutiple of clump size?
        }
        else
        {
                return -1;
        }

        vcb = VTOVCB(vp);
        
        /*
         * The Extents B-tree can't have overflow extents. ExtendFileC will
         * return an error if an attempt is made to extend the Extents B-tree
         * when the resident extents are exhausted.
         */

        /* Protect allocation bitmap and extents overflow file. */
        lockflags = SFL_BITMAP;
        if (VTOC(vp)->c_fileid != kHFSExtentsFileID)
                lockflags |= SFL_EXTENTS;
        lockflags = hfs_systemfile_lock(vcb, lockflags, HFS_EXCLUSIVE_LOCK);

        (void) BTGetInformation(filePtr, 0, &btInfo);

#if 0  // XXXdbg
        /*
         * The b-tree code expects nodes to be contiguous. So when
         * the allocation block size is less than the b-tree node
         * size, we need to force disk allocations to be contiguous.
         */
        if (vcb->blockSize >= btInfo.nodeSize) {
                extendFlags = 0;
        } else {
                /* Ensure that all b-tree nodes are contiguous on disk */
                extendFlags = kEFContigMask;
        }
#endif

        origSize = filePtr->fcbEOF;
        fileblocks = filePtr->ff_blocks;
        startAllocation = vcb->nextAllocation;

        // loop trying to get a contiguous chunk that's an integer multiple
        // of the btree node size.  if we can't get a contiguous chunk that
        // is at least the node size then we break out of the loop and let
        // the error propagate back up.
        while((off_t)bytesToAdd >= btInfo.nodeSize) {
            do {
                retval = ExtendFileC(vcb, filePtr, bytesToAdd, 0,
                                     kEFContigMask | kEFMetadataMask | kEFNoClumpMask,
                                     (int64_t *)&actualBytesAdded);
                if (retval == dskFulErr && actualBytesAdded == 0) {
                    bytesToAdd >>= 1;
                    if (bytesToAdd < btInfo.nodeSize) {
                        break;
                    } else if ((bytesToAdd % btInfo.nodeSize) != 0) {
                        // make sure it's an integer multiple of the nodeSize
                        bytesToAdd -= (bytesToAdd % btInfo.nodeSize);
                    }
                }
            } while (retval == dskFulErr && actualBytesAdded == 0);

            if (retval == dskFulErr && actualBytesAdded == 0 && bytesToAdd <= btInfo.nodeSize) {
                break;
            }

            filePtr->fcbEOF = (u_int64_t)filePtr->ff_blocks * (u_int64_t)vcb->blockSize;
            bytesToAdd = minEOF - filePtr->fcbEOF;
        }

        /*
         * If a new extent was added then move the roving allocator
         * reference forward by the current b-tree file size so 
         * there's plenty of room to grow.
         */
        if ((retval == 0) &&
            ((VCBTOHFS(vcb)->hfs_flags & HFS_METADATA_ZONE) == 0) &&
            (vcb->nextAllocation > startAllocation) &&
            ((vcb->nextAllocation + fileblocks) < vcb->allocLimit)) {
                HFS_UPDATE_NEXT_ALLOCATION(vcb, vcb->nextAllocation + fileblocks); 
        }
                
        filePtr->fcbEOF = (u_int64_t)filePtr->ff_blocks * (u_int64_t)vcb->blockSize;

        // XXXdbg ExtendFileC() could have returned an error even though
        // it grew the file to be big enough for our needs.  If this is
        // the case, we don't care about retval so we blow it away.
        //
        if (filePtr->fcbEOF >= (off_t)minEOF && retval != 0) {
                retval = 0;
        }

        // XXXdbg if the file grew but isn't large enough or isn't an
        // even multiple of the nodeSize then trim things back.  if
        // the file isn't large enough we trim back to the original
        // size.  otherwise we trim back to be an even multiple of the
        // btree node size.
        //
        if ((filePtr->fcbEOF < (off_t)minEOF) || ((filePtr->fcbEOF - origSize) % btInfo.nodeSize) != 0) {

                if (filePtr->fcbEOF < (off_t)minEOF) {
                        retval = dskFulErr;
                        
                        if (filePtr->fcbEOF < origSize) {
                                panic("hfs: btree file eof %lld less than orig size %lld!\n",
                                          filePtr->fcbEOF, origSize);
                        }
                        
                        trim = filePtr->fcbEOF - origSize;
                } else {
                        trim = ((filePtr->fcbEOF - origSize) % btInfo.nodeSize);
                }

                ret = TruncateFileC(vcb, filePtr, filePtr->fcbEOF - trim, 0, 0, FTOC(filePtr)->c_fileid, 0);
                filePtr->fcbEOF = (u_int64_t)filePtr->ff_blocks * (u_int64_t)vcb->blockSize;

                // XXXdbg - panic if the file didn't get trimmed back properly
                if ((filePtr->fcbEOF % btInfo.nodeSize) != 0) {
                        panic("hfs: truncate file didn't! fcbEOF %lld nsize %d fcb %p\n",
                                  filePtr->fcbEOF, btInfo.nodeSize, filePtr);
                }

                if (ret) {
                        // XXXdbg - this probably doesn't need to be a panic()
                        panic("hfs: error truncating btree files (sz 0x%llx, trim %lld, ret %ld)\n",
                              filePtr->fcbEOF, trim, (long)ret);
                        goto out;
                }
        }

        if(VTOC(vp)->c_fileid != kHFSExtentsFileID) {
                /*
                 * Get any extents overflow b-tree changes to disk ASAP!
                 */
                (void) BTFlushPath(VTOF(vcb->extentsRefNum));
                (void) hfs_fsync(vcb->extentsRefNum, MNT_WAIT, 0, p);
        }
        hfs_systemfile_unlock(vcb, lockflags);
        lockflags = 0;

        if ((filePtr->fcbEOF % btInfo.nodeSize) != 0) {
                panic("hfs: extendbtree: fcb %p has eof 0x%llx not a multiple of 0x%x (trim %llx)\n",
                          filePtr, filePtr->fcbEOF, btInfo.nodeSize, trim);
        }

        /*
         * Update the Alternate MDB or Alternate VolumeHeader
         */
        if ((VTOC(vp)->c_fileid == kHFSExtentsFileID)   ||
            (VTOC(vp)->c_fileid == kHFSCatalogFileID)   ||
            (VTOC(vp)->c_fileid == kHFSAttributesFileID)
           ) {
                VTOC(vp)->c_flag |= C_MODIFIED;
                MarkVCBDirty( vcb );
                ret = hfs_flushvolumeheader(VCBTOHFS(vcb), MNT_WAIT, HFS_ALTFLUSH);
        } else {
                VTOC(vp)->c_touch_chgtime = TRUE;
                VTOC(vp)->c_touch_modtime = TRUE;
                (void) hfs_update(vp, TRUE);
        }

        ret = ClearBTNodes(vp, btInfo.nodeSize, origSize, (filePtr->fcbEOF - origSize));
out:
        if (retval == 0)
                retval = ret;
        
        if (lockflags)
                hfs_systemfile_unlock(vcb, lockflags);
        
        return retval;
}


/*
 * Clear out (zero) new b-tree nodes on disk.
 */
static int
ClearBTNodes(struct vnode *vp, long blksize, off_t offset, off_t amount)
{
        struct hfsmount *hfsmp = VTOHFS(vp);
        struct buf *bp = NULL;
        daddr64_t blk;
        daddr64_t blkcnt;
    
        blk = offset / blksize;
        blkcnt = amount / blksize;
        
        while (blkcnt > 0) {
                bp = buf_getblk(vp, blk, blksize, 0, 0, BLK_META);
                if (bp == NULL)
                        continue;

        // XXXdbg
                if (hfsmp->jnl) {
                        // XXXdbg -- skipping this for now since it makes a transaction
                        //           become *way* too large
                    //journal_modify_block_start(hfsmp->jnl, bp);
                }
                bzero((char *)buf_dataptr(bp), blksize);

                buf_markaged(bp);

        // XXXdbg
                if (hfsmp->jnl) {
                        // XXXdbg -- skipping this for now since it makes a transaction
                        //           become *way* too large
                        //journal_modify_block_end(hfsmp->jnl, bp);

                        // XXXdbg - remove this once we decide what to do with the
                        //          writes to the journal
                        if ((blk % 32) == 0)
                            VNOP_BWRITE(bp);
                        else
                            buf_bawrite(bp);
                } else {
                        /* wait/yield every 32 blocks so we don't hog all the buffers */
                        if ((blk % 32) == 0)
                                VNOP_BWRITE(bp);
                        else
                                buf_bawrite(bp);
                }
                --blkcnt;
                ++blk;
        }

        return (0);
}


extern char  hfs_attrname[];

/*
 * Create an HFS+ Attribute B-tree File.
 *
 * No global resources should be held.
 */
int
hfs_create_attr_btree(struct hfsmount *hfsmp, u_int32_t nodesize, u_int32_t nodecnt)
{
        struct vnode* vp = NULLVP;
        struct cat_desc cndesc;
        struct cat_attr cnattr;
        struct cat_fork cfork;
        BlockDescriptor blkdesc;
        BTNodeDescriptor  *ndp;
        BTHeaderRec  *bthp;
        BTreeControlBlockPtr btcb = NULL;
        struct buf *bp = NULL;
        void * buffer;
        u_int8_t *bitmap;
        u_int16_t *index;
        u_int32_t node_num, num_map_nodes;
        u_int32_t bytes_per_map_record;
        u_int32_t temp;
        u_int16_t  offset;
        int intrans = 0;
        int result;
        int newvnode_flags = 0;
        
again:
        /*
         * Serialize creation using HFS_CREATING_BTREE flag.
         */
        lck_mtx_lock(&hfsmp->hfs_mutex);
        if (hfsmp->hfs_flags & HFS_CREATING_BTREE) {
                        /* Someone else beat us, wait for them to finish. */
                        (void) msleep(hfsmp->hfs_attribute_cp, &hfsmp->hfs_mutex,
                                      PDROP | PINOD, "hfs_create_attr_btree", 0);
                        if (hfsmp->hfs_attribute_vp) {
                                return (0);
                        }
                        goto again;
        }
        hfsmp->hfs_flags |= HFS_CREATING_BTREE;
        lck_mtx_unlock(&hfsmp->hfs_mutex);

        /* Check if were out of usable disk space. */
        if ((hfs_freeblks(hfsmp, 1) == 0)) {
                result = ENOSPC;
                goto exit;
        }

        /*
         * Set up Attribute B-tree vnode
         * (this must be done before we start a transaction
         *  or take any system file locks)
         */
        bzero(&cndesc, sizeof(cndesc));
        cndesc.cd_parentcnid = kHFSRootParentID;
        cndesc.cd_flags |= CD_ISMETA;
        cndesc.cd_nameptr = (const u_int8_t *)hfs_attrname;
        cndesc.cd_namelen = strlen(hfs_attrname);
        cndesc.cd_cnid = kHFSAttributesFileID;

        bzero(&cnattr, sizeof(cnattr));
        cnattr.ca_linkcount = 1;
        cnattr.ca_mode = S_IFREG;
        cnattr.ca_fileid = cndesc.cd_cnid;

        bzero(&cfork, sizeof(cfork));
        cfork.cf_clump = nodesize * nodecnt;

        result = hfs_getnewvnode(hfsmp, NULL, NULL, &cndesc, 0, &cnattr, 
                                                         &cfork, &vp, &newvnode_flags);
        if (result) {
                goto exit;
        }
        /*
         * Set up Attribute B-tree control block
         */
        MALLOC(btcb, BTreeControlBlock *, sizeof(BTreeControlBlock), M_TEMP, M_WAITOK);
        bzero(btcb, sizeof(BTreeControlBlock));

        btcb->nodeSize          = nodesize;
        btcb->maxKeyLength      = kHFSPlusAttrKeyMaximumLength;
        btcb->btreeType         = 0xFF;
        btcb->attributes        = kBTVariableIndexKeysMask | kBTBigKeysMask;
        btcb->version           = kBTreeVersion;
        btcb->writeCount        = 1;
        btcb->flags             = 0;  /* kBTHeaderDirty */
        btcb->fileRefNum        = vp;
        btcb->getBlockProc      = GetBTreeBlock;
        btcb->releaseBlockProc  = ReleaseBTreeBlock;
        btcb->setEndOfForkProc  = ExtendBTreeFile;
        btcb->keyCompareProc    = (KeyCompareProcPtr)hfs_attrkeycompare;
        VTOF(vp)->fcbBTCBPtr    = btcb;

        /*
         * Allocate some space
         */
        if (hfs_start_transaction(hfsmp) != 0) {
                result = EINVAL;
                goto exit;
        }
        intrans = 1;

        /* Note ExtendBTreeFile will acquire the necessary system file locks. */
        result = ExtendBTreeFile(vp, nodesize, cfork.cf_clump);
        if (result)
                goto exit;

        btcb->totalNodes = VTOF(vp)->ff_size / nodesize;

        /*
         * Figure out how many map nodes we'll need.
         *
         * bytes_per_map_record = the number of bytes in the map record of a
         * map node.  Since that is the only record in the node, it is the size
         * of the node minus the node descriptor at the start, and two record
         * offsets at the end of the node.  The "- 2" is to round the size down
         * to a multiple of 4 bytes (since sizeof(BTNodeDescriptor) is not a
         * multiple of 4).
         *
         * The value "temp" here is the number of *bits* in the map record of
         * the header node.
         */
        bytes_per_map_record = nodesize - sizeof(BTNodeDescriptor) - 2*sizeof(u_int16_t) - 2;
        temp = 8 * (nodesize - sizeof(BTNodeDescriptor) 
                        - sizeof(BTHeaderRec)
                        - kBTreeHeaderUserBytes
                        - 4 * sizeof(u_int16_t));
        if (btcb->totalNodes > temp) {
                num_map_nodes = howmany(btcb->totalNodes - temp, bytes_per_map_record * 8);
        }
        else {
                num_map_nodes = 0;
        }
        
        btcb->freeNodes = btcb->totalNodes - 1 - num_map_nodes;
        
        /*
         * Initialize the b-tree header on disk
         */
        bp = buf_getblk(vp, 0, nodesize, 0, 0, BLK_META);
        if (bp == NULL) {
                result = EIO;
                goto exit;
        }

        buffer = (void *)buf_dataptr(bp);
        blkdesc.buffer = buffer;
        blkdesc.blockHeader = (void *)bp;
        blkdesc.blockReadFromDisk = 0;
        blkdesc.isModified = 0;

        ModifyBlockStart(vp, &blkdesc);

        if (buf_size(bp) != nodesize)
                panic("hfs_create_attr_btree: bad buffer size (%d)\n", buf_size(bp));

        bzero(buffer, nodesize);
        index = (u_int16_t *)buffer;

        /* FILL IN THE NODE DESCRIPTOR:  */
        ndp = (BTNodeDescriptor *)buffer;
        if (num_map_nodes != 0)
                ndp->fLink = 1;
        ndp->kind = kBTHeaderNode;
        ndp->numRecords = 3;
        offset = sizeof(BTNodeDescriptor);
        index[(nodesize / 2) - 1] = offset;

        /* FILL IN THE HEADER RECORD:  */
        bthp = (BTHeaderRec *)((u_int8_t *)buffer + offset);
        bthp->nodeSize     = nodesize;
        bthp->totalNodes   = btcb->totalNodes;
        bthp->freeNodes    = btcb->freeNodes;
        bthp->clumpSize    = cfork.cf_clump;
        bthp->btreeType    = 0xFF;
        bthp->attributes   = kBTVariableIndexKeysMask | kBTBigKeysMask;
        bthp->maxKeyLength = kHFSPlusAttrKeyMaximumLength;
        bthp->keyCompareType = kHFSBinaryCompare;
        offset += sizeof(BTHeaderRec);
        index[(nodesize / 2) - 2] = offset;

        /* FILL IN THE USER RECORD:  */
        offset += kBTreeHeaderUserBytes;
        index[(nodesize / 2) - 3] = offset;

        /* Mark the header node and map nodes in use in the map record.
         *
         * NOTE: Assumes that the header node's map record has at least
         * (num_map_nodes + 1) bits.
         */
        bitmap = (u_int8_t *) buffer + offset;
        temp = num_map_nodes + 1;       /* +1 for the header node */
        while (temp >= 8) {
                *(bitmap++) = 0xFF;
                temp -= 8;
        }
        *bitmap = ~(0xFF >> temp);
        
        offset += nodesize - sizeof(BTNodeDescriptor) - sizeof(BTHeaderRec)
                           - kBTreeHeaderUserBytes - (4 * sizeof(int16_t));
        index[(nodesize / 2) - 4] = offset;

        if (hfsmp->jnl) {
                result = btree_journal_modify_block_end(hfsmp, bp);
        } else {
                result = VNOP_BWRITE(bp);
        }
        if (result)
                goto exit;

        /* Create the map nodes: node numbers 1 .. num_map_nodes */
        for (node_num=1; node_num <= num_map_nodes; ++node_num) {
                bp = buf_getblk(vp, node_num, nodesize, 0, 0, BLK_META);
                if (bp == NULL) {
                        result = EIO;
                        goto exit;
                }
                buffer = (void *)buf_dataptr(bp);
                blkdesc.buffer = buffer;
                blkdesc.blockHeader = (void *)bp;
                blkdesc.blockReadFromDisk = 0;
                blkdesc.isModified = 0;
        
                ModifyBlockStart(vp, &blkdesc);
                
                bzero(buffer, nodesize);
                index = (u_int16_t *)buffer;
        
                /* Fill in the node descriptor */
                ndp = (BTNodeDescriptor *)buffer;
                if (node_num != num_map_nodes)
                        ndp->fLink = node_num + 1;
                ndp->kind = kBTMapNode;
                ndp->numRecords = 1;
                offset = sizeof(BTNodeDescriptor);
                index[(nodesize / 2) - 1] = offset;
        
        
                /* Fill in the map record's offset */
                /* Note: We assume that the map record is all zeroes */
                offset = sizeof(BTNodeDescriptor) + bytes_per_map_record;
                index[(nodesize / 2) - 2] = offset;
        
                if (hfsmp->jnl) {
                        result = btree_journal_modify_block_end(hfsmp, bp);
                } else {
                        result = VNOP_BWRITE(bp);
                }
                if (result)
                        goto exit;
        }
        
        /* Update vp/cp for attribute btree */
        lck_mtx_lock(&hfsmp->hfs_mutex);
        hfsmp->hfs_attribute_cp = VTOC(vp);
        hfsmp->hfs_attribute_vp = vp;
        lck_mtx_unlock(&hfsmp->hfs_mutex);

        (void) hfs_flushvolumeheader(hfsmp, MNT_WAIT, HFS_ALTFLUSH);
exit:
        if (vp) {
                hfs_unlock(VTOC(vp));
        }
        if (result) {
                if (btcb) {
                        FREE (btcb, M_TEMP);
                }
                if (vp) {
                        vnode_put(vp);
                }
                /* XXX need to give back blocks ? */
        }
        if (intrans) {
                hfs_end_transaction(hfsmp);
        }

        /*
         * All done, clear HFS_CREATING_BTREE, and wake up any sleepers.
         */
        lck_mtx_lock(&hfsmp->hfs_mutex);
        hfsmp->hfs_flags &= ~HFS_CREATING_BTREE;
        wakeup((caddr_t)hfsmp->hfs_attribute_cp);
        lck_mtx_unlock(&hfsmp->hfs_mutex);

        return (result);
}