MFSCore.h

/*
    File:       MFSCore.h
 
    Contains:   Core MFS implementation for MFSLives.
 
    Written by: DTS
 
    Copyright:  Copyright (c) 2006 by Apple Computer, Inc., All Rights Reserved.
 
    Disclaimer: IMPORTANT:  This Apple software is supplied to you by Apple Computer, Inc.
                ("Apple") in consideration of your agreement to the following terms, and your
                use, installation, modification or redistribution of this Apple software
                constitutes acceptance of these terms.  If you do not agree with these terms,
                please do not use, install, modify or redistribute this Apple software.
 
                In consideration of your agreement to abide by the following terms, and subject
                to these terms, Apple grants you a personal, non-exclusive license, under Apple's
                copyrights in this original Apple software (the "Apple Software"), to use,
                reproduce, modify and redistribute the Apple Software, with or without
                modifications, in source and/or binary forms; provided that if you redistribute
                the Apple Software in its entirety and without modifications, you must retain
                this notice and the following text and disclaimers in all such redistributions of
                the Apple Software.  Neither the name, trademarks, service marks or logos of
                Apple Computer, Inc. may be used to endorse or promote products derived from the
                Apple Software without specific prior written permission from Apple.  Except as
                expressly stated in this notice, no other rights or licenses, express or implied,
                are granted by Apple herein, including but not limited to any patent rights that
                may be infringed by your derivative works or by other works in which the Apple
                Software may be incorporated.
 
                The Apple Software is provided by Apple on an "AS IS" basis.  APPLE MAKES NO
                WARRANTIES, EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION THE IMPLIED
                WARRANTIES OF NON-INFRINGEMENT, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
                PURPOSE, REGARDING THE APPLE SOFTWARE OR ITS USE AND OPERATION ALONE OR IN
                COMBINATION WITH YOUR PRODUCTS.
 
                IN NO EVENT SHALL APPLE BE LIABLE FOR ANY SPECIAL, INDIRECT, INCIDENTAL OR
                CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
                GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
                ARISING IN ANY WAY OUT OF THE USE, REPRODUCTION, MODIFICATION AND/OR DISTRIBUTION
                OF THE APPLE SOFTWARE, HOWEVER CAUSED AND WHETHER UNDER THEORY OF CONTRACT, TORT
                (INCLUDING NEGLIGENCE), STRICT LIABILITY OR OTHERWISE, EVEN IF APPLE HAS BEEN
                ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 
    Change History (most recent first):
 
$Log: MFSCore.h,v $
Revision 1.1  2006/07/27 15:47:47  eskimo1
First checked in.
 
 
*/
 
#ifndef _MFSCORE_H
#define _MFSCORE_H
 
/////////////////////////////////////////////////////////////////////
 
#include <stdint.h>
#include <mach/boolean.h>
#include <sys/time.h>
#include <sys/mount.h>
#include <sys/vnode.h>
#if ! KERNEL
    #include "VNodeAttr.h"
#endif
 
/////////////////////////////////////////////////////////////////////
 
/*
    Design Philosophy
    -----------------
    This module implements the core MFS core.  None of the code outside of this 
    module contains any significant understanding of the MFS volume format.  
    Also, none of the code within this module contains any significant dependency 
    on the kernel environment.
    
    The reason for this isolation is that we want to be able to build run this 
    code in user space.  There are two reasons for this:
    
      o We have a user space unit test program to exercise this code.
 
      o The MFSLives.util tool uses this code to work with MFS volumes without 
        installing kernel code, which is a handy dandy feature.
    
    The upshot of this is that the MFS core does not read or write blocks, or do 
    any locking for multi-threaded access, or allocate memory.  All of these tasks 
    are the responsibility of the caller.  In some cases I've had to bend the API 
    a little to maintain this constraint (for example, 
    MFSDirectoryBlockFindEntryByName requires that you pass it a temporary buffer), 
    but that's the price you pay.
    
    Finally, because MFS isn't really suitable as a read/write volume format 
    (on a read/write volume, what would you do if you were asked to create a 
    directory?!), I have not gone out of my way to provide read/write support in 
    the MFS core code.  For example, the MFS core API exposes the raw directory 
    offset to the caller, which isn't architecturally advantageous for a read/write 
    implementation.
 
    MFS Volume Format
    -----------------
    Inside Macintosh II contains a very rudimentary description of the MFS volume 
    format.  This description leaves a number of open issues, and I've tried to 
    cover these in the most sensible way.
 
      o Folders -- If you're old enough to remember the original Macintosh, you 
        may recall that it did support folders.  These folders were an illusion 
        maintained by the Finder.  The MFS file system is strictly flat.
    
      o Alternative Block Sizes -- All real world MFS disks had a block size of 
        512 bytes (modulo the wacky tag bytes on the 400 KB floppy drives, but 
        that doesn't affect the actual volume format).  In theory, one could create 
        an MFS volume on a disk with a different block size.  In the MFS core code 
        (and in most of the rest of MFSLives as a whole) I've avoided hard-coding 
        the block size at 512 bytes.  Rather, I've kept a clear separate between 
        quantities that are in bytes, blocks, and allocation blocks, and I don't 
        try to convert from one to the other.
        
        For example, the location the MFS directory is in blocks, and the MFS core 
        does not assume that this is a specific byte offset from the beginning 
        of the disk.  Rather, it returns the values in blocks to the caller, and 
        expects it to get that block number.
        
        I don't know if anyone ever created an MFS volume on a device with a block 
        size that's not 512, so I can't check if this approach is correct.
        
        I /do/ make the assumption that the block size is big enough to hold the 
        MFS master directory block (64 bytes).
    
      o Text Encodings -- MFS stores a name (file and volume) as a Pascal string, 
        with no associated text encoding.  For simplicity's sake, I assume that 
        this string is MacRoman.  If you have MFS disks from a Japanese system, 
        you are out of luck.
        
        If this was production code, I'd allow you to specify the encoding at mount 
        time.  However, given the paucity of text encoding conversion support within 
        the kernel, that's just too much work for this sample code.
        
        There's a big discussion of how I implement text encoding conversion below.
    
      o Dates/Time Values -- MFS date/time values are stored in local time.  
        VFS (and general common sense) requires that date/time values be returned 
        in UTC.  I originally tried to make it so that the times displayed by the 
        Finder would match the times displayed by the Finder on a Mac 128K.  I did 
        this by doing a local-to-UTC conversion in the "MFSCore.c", in the hope that 
        it would exactly match the ultimate client's UTC-to-local conversion.  
        However, it's impossible to do this correctly (see below), so I changed my 
        approach and decided that I'd treat all MFS date/time values as if they were 
        in UTC.
        
        The reasons why it's impossible to implement my previous approach are:
        
          - You might have two clients in two different time zones, and thus it's 
            impossible to local-to-UTC conversion that will satisfy both.
            
            I don't consider this to be a serious problem; it's rare for this 
            situation to arise on a typical Mac.
            
          - Local time is not linear.  In time zones that have summer time, at the 
            point where summer time changes back to winter time, a given local 
            time has two possible UTC values.  For example, if you're on British 
            Summer Time (BST is UTC+0100) and the clock goes back to GMT (UTC+0000) 
            at 0200, a local time value of 0130 corresponds to both 0030 UTC (when the 
            local time is in BST) and 0130 (an hour later, after the local time has 
            reverted to GMT).
            
            The following table illustrates this in more detail.  Notice how the 
            right column (UTC) progresses linearly, whereas the left column has 
            discontinuities.
            
            0130 local (BST) -> 0030 UTC
            0159 local (BST) -> 0059 UTC
            <<< local time goes back one hour at 0200 local >>>
            0101 local (GMT) -> 0101 UTC
            0130 local (GMT) -> 0130 UTC
            
            The upshot of this is that a given local time can have more than one 
            possible UTC value, so there's no unique transform from local time to UTC.
        
          - Even if you ignore the above (which is only really an issue during 
            local time changes), the mechanics of the client's UTC-to-local conversion 
            make it hard to do the correction.  Specifically, system routines that 
            do UTC-to-local conversion (such as <x-man-page://3/localtime_r> will 
            typically do it in the time zone that was in effect at that time.  That is, 
            if you live in Scotland (BST/GMT) and convert a UTC value from 1 Jan 1986 
            to local time, you'll get a GMT value because GMT was in effect on 1 Jan 1986.  
            OTOH, if the UTC value is 1 Jul 1986, you'll get a BST value.
            
            So, from the kernel's perspective, there's no simple UTC correction factor 
            that you can add to all local times to get their UTC values such that 
            clients who convert the resulting UTC values back to local time will get 
            the value you started off with.  Instead, you have to do a different 
            conversion for each potential date.  All of this makes the problem /way/ 
            too hard for this sample code.
 
        So, rather than do a poor job of local-to-UTC conversion, I decided to 
        just ignore the problem and assume (wrongly) that all MFS disks were used 
        in UTC.  This means that you get reasonably consistent results (if file 
        X was modified an hour after file Y, that difference will be reflected 
        correctly to the client), but the Finder will no show the correct local time 
        for any given file.
        
        This change also allowed me to delete a bunch of code.  Hey, if you can't 
        get it right, you might as well keep it simple.
        
        Finally, I /do/ have to do epoch conversion.  MFS date/time values start 
        at 00:00:00 1 Jan 1904.  VFS date/time values start at 00:00:00 1 Jan 1970.  
        This module does the epoch conversion internally.  MFS date/time values 
        before 00:00:00 1970 are clipped to 00:00:00 1970.
        
        Quinn says: time is hard.
 
    Text Encodings
    --------------
    For general information about text encodings in VFS, see DTS Q&A 1173 "Text 
    Encodings in VFS".
    
    <http://developer.apple.com/qa/qa2001/qa1173.html>
    
    As I mentioned above, I assume that all names on MFS volumes are in MacRoman. 
    As VFS expects to work in UTF-8, I need to be able to:
    
     1. convert from UTF-8 (decomposed or precomposed) to MacRoman
     2. convert from MacRoman to UTF-8 (decomposed)
     3. do case insensitive equality testing
 
    The kernel provides very limited support for text encoding conversion, so I have 
    to do most of this myself.  I do it using a variety of lookup tables that I 
    precompute (using the TableGenerator program, see "TableGenerator.c") and then 
    copy'n'paste into the source of the MFS core.  I use a number of tricks to make 
    these tables manageably small:
    
      o In all cases except case folding, I special case the bottom 128 code points 
        because I know that MacRoman and UTF-8 both inherit their bottom 128 
        code points from the same place (ASCII).
        
      o For UTF-8 (decomposed or precomposed) to MacRoman conversion, I first use the 
        kernel routine utf_decodestr to convert the string to precomposed UTF-16. 
        Each valid MacRoman has a unique precomposed UTF-16 value, so I have a 
        table that maps valid UTF-16 values to their MacRoman equivalents 
        (kUTF16ToMacRoman).  As the UTF-16 space is huge, this table is sparse, 
        so I generate a sorted table and binary search it.
    
        This approach means that I don't have to parse UTF-8, or do canonical 
        composition; utf_decodestr takes care of that.  It also means that I don't 
        have to worry about decomposed UTF-16, because any combining UTF-16 
        code point remaining after precomposition is necessarily not in MacRoman.
    
      o For MacRoman to UTF-8 (decomposed) conversion, I simply have a table that 
        maps the MacRoman character to its corresponding UTF-8 (decomposed) string. 
        
        One interesting case here is truncation.  If I have to truncate a string, 
        I make sure that I truncate it at a MacRoman boundary.  That is, you either 
        get all of the UTF-8 for a given MacRoman character, or none of it.
    
      o For case insensitive equality testing, I'm testing an arbitrary UTF-8 
        string against a list of MacRoman file names (to do a directory lookup 
        by name).  In this case, I convert the UTF-8 string to MacRoman and do 
        the comparison in the MacRoman space.  I do the comparison in MacRoman 
        space by simply having a table that maps every MacRoman character to 
        its upper case equivalent.
        
        If the UTF-8 string does not convert to MacRoman, it clearly isn't a match 
        for any of my MacRoman strings.
 
    The upshot of this is that, with some sneaky coding, I can handle text encodings 
    correctly while keeping my conversion tables small.
    
    One minor quibble is that there's an asymmetry in the code: I use <sys/utfconv.h> 
    routines to got from UTF-8 to UTF-16 (and hence to MacRoman), but not the other 
    way.  I did this because the <sys/utfconv.h> routines require me to have an 
    extra buffer (for the UTF-16), which is a pain.  So I only use them if 
    they offer a real advantage.  In the case of utf8_decodestr, it handles 
    all of the ugliness of UTF-8 parsing and does pre-composition, both things 
    that I really need.  So the hassle of the extra buffer is far outweighed 
    by the benefits.  OTOH, when going the other way, it's just as easy for 
    my table to contain UTF-8 as it would be for it to contain UTF-16, so I just 
    go directly from MacRoman to UTF-8.
*/
 
/////////////////////////////////////////////////////////////////////
 
// Special inode numbers -- All of these are traditional for Macintosh-based file systems 
// starting with HFS.  MFS uses a different approach (file numbers start from 1), so 
// the MFS core does an internal translation to these numbers.
 
enum {
    kMFSRootParentInodeNumber = 1,
    kMFSRootInodeNumber       = 2,
    kMFSFirstFileInodeName    = 16
};
 
// Special block numbers -- There's only one, allowing the caller to find the 
// master directory block (MDB) to pass to MFSMDBCheck.
 
enum {
    kMFSMDBBlock = 2                // block number of the MDB (includes the start of the VABM) 
};
 
// The volume allocation block map (VABM) follows immediately after the MDB on disk 
// (that is, the start of the VABM shares the same block as the MDB).
 
extern int MFSMDBCheck(
    const void *        mdbBlockPtr, 
    uint64_t            containerBlockCount,
    size_t *            mdbAndVABMSizeInBytesPtr, 
    uint16_t *          directoryStartBlockPtr, 
    uint16_t *          directoryBlockCountPtr,
    uint16_t *          allocationBlocksStartBlockPtr, 
    uint32_t *          allocationBlockSizeInBytesPtr
);
    // Given a pointer to an MFS master directory block (MDB), this routine returns 
    // 0 if the MDB is valid and EINVAL if it isn't.
    //
    // mdbBlockPtr must be a pointer to master directory block that you've read in. 
    // This is always block kMFSMDBBlock on the disk.  The size of this block must 
    // be at least 64 bytes (which is the size of the MFS MDB).
    //
    // containerBlockCount must be the number of blocks on the disk.  The routine uses 
    // this value to range check various fields in the MDB.
    //
    // On entry, any combination of mdbAndVABMSizeInBytesPtr, directoryStartBlockPtr, 
    // directoryBlockCountPtr, allocationBlocksStartBlockPtr, and allocationBlockSizeInBytesPtr may 
    // be NULL, in which case that value is not returned.  If xxxPtr is not NULL, 
    // *xxxPtr is ignored on entry.  If xxxPtr is not NULL then, on success, 
    // *xxxPtr contains a valid value and, on error, the value of *xxxPtr will be unchanged.
    //
    // On success when mdbAndVABMSizeInBytesPtr is not NULL, *mdbAndVABMSizeInBytesPtr 
    // will be the size of the combined MDB and VABM.  The caller is expected to read in 
    // at least this many bytes when it calls MFSForkGetExtent.
    //
    // On success when directoryStartBlockPtr is not NULL, *directoryStartBlockPtr will 
    // be the block number of the first directory block.
    //
    // On success when directoryBlockCountPtr is not NULL, *directoryBlockCountPtr will 
    // be the number of directory blocks.
    //
    // To search the directory, the caller is expected to iterate over all of the directory 
    // blocks, reading in each block and presenting it to the appropriate MFS core routine 
    // (MFSDirectoryBlockIterate or MFSDirectoryBlockFindEntryByName).
    //
    // On success when allocationBlocksStartBlockPtr is not NULL, *allocationBlocksStartBlockPtr  
    // will be the block number of the block that holds the first allocation block.
    //
    // On success when allocationBlockSizeInBytesPtr is not NULL, *allocationBlockSizeInBytesPtr  
    // will be the size in bytes of each allocation block.
 
extern void MFSMDBGetError(
    const void *        mdbBlockPtr, 
    uint64_t            containerBlockCount,
    char *              errStr,
    size_t              errStrSize
);
    // If MFSMDBCheck fails with EINVAL, you can call this routine to get a 
    // non-localised description of what the MFS core does not like about the 
    // MDB.
    //
    // mdbBlockPtr and containerBlockCount are as per MFSMDBCheck.
    //
    // errStr must point to a buffer where the error is returned (as a UTF-8 
    // C string).  errStrSize is the size of that buffer.  It must be 
    // at least 1 (otherwise we couldn't returning a valid C string, 
    // with its terminating null character).  If it's too short, the 
    // routine silently truncates the string.  A value of 256 should be 
    // sufficient.
 
extern int MFSMDBGetAttr(
    const void *        mdbBlockPtr,
    struct vfs_attr *   attr
);
    // Returns information about the MFS volume.
    //
    // mdbBlockPtr is as per MFSMDBCheck.
    //
    // attr must not be NULL; it is handled as per the VFS plug-in's vfs_getattr 
    // entry point.
    //
    // Note that, with one exception, this routine always returns all of the 
    // attributes that are available for the MFS volume, because the cost of 
    // returns those values is trivial.  That is, it (typically) ignores 
    // the f_active field of attr.  The one exception is f_vol_name; because getting 
    // the volume requires MacRoman to UTF-8 conversion, which is potentially 
    // time consuming, this routine only returns f_vol_name if you request it.
 
#define kMFSDirectoryBlockIterateFromStart ((size_t)-1)
 
extern int MFSDirectoryBlockIterate(
    const void *        directoryBlockPtr, 
    size_t              directoryBlockSizeInBytes, 
    size_t *            dirOffsetPtr, 
    struct vnode_attr * attr
);
    // Allows you to iterate over every directory entry within an MFS directory block.
    // To do this, start by setting *dirOffsetPtr to kMFSDirectoryBlockIterateFromStart. 
    // Each time you call the routine, it will either fail with ENOENT 
    // (indicating that there are no more directory entries in the directory 
    // block), or update *dirOffsetPtr to be the offset of the next directory 
    // entry within the block.
    //
    // directoryBlockPtr must point to an MFS directory block.  See MFSMDBCheck 
    // for information on how to locate these.
    //
    // directoryBlockSizeInBytes must be the size of that block.
    //
    // dirOffsetPtr must not be NULL.  On entry, it must either be 
    // kMFSDirectoryBlockIterateFromStart (indicating that you're requesting the 
    // first directory entry in the block), or a value that was previously 
    // returned by this routine.  On success, it is set to the offset of the 
    // next directory entry within the directory block.  On failure, its 
    // value is unchanged.
    //
    // On entry, if attr is NULL, no attributes are returned.  OTOH, if attr is not 
    // NULL then, on success, the directory entry's attributes will be returned 
    // in *attr.  See MFSDirectoryEntryGetAttr for specific details on this.
 
extern int MFSDirectoryBlockCheckDirOffset(
    const void *        directoryBlockPtr, 
    size_t              directoryBlockSizeInBytes, 
    size_t              candidateDirOffset
);
    // Checks to see if candidateDirOffset is a valid directory offset, that is, 
    // an offset that you would get back if you iterated the directory from the 
    // start.
    //
    // directoryBlockPtr must point to an MFS directory block.  See MFSMDBCheck 
    // for information on how to locate these.
    //
    // directoryBlockSizeInBytes must be the size of that block.
    //
    // candidateDirOffset must be less than directoryBlockSizeInBytes.
    //
    // Returns EINVAL if candidateDirOffset is not valid, and 0 otherwise.
 
enum {
    kMFSDirectoryBlockFindEntryByNameTempBufferSize = MAXPATHLEN
};
 
extern int MFSDirectoryBlockFindEntryByName(
    const void *        directoryBlockPtr, 
    size_t              directoryBlockSizeInBytes, 
    const char *        utf8Name,
    size_t              utf8NameLen,
    void *              tempBuffer,
    size_t *            dirOffsetPtr, 
    struct vnode_attr * attr
);
    // Searches, by name, for a directory entry within an MFS directory block. 
    // To do this, first allocate a temporary buffer of size 
    // kMFSDirectoryBlockFindEntryByNameTempBufferSize, and set its first byte 
    // to 0.  Then call this routine.  If it finds the named directory entry, 
    // it returns 0.  If it fails to find the entry, it returns ENOENT. 
    // Other errors can occur if the input name is not valid.
    //
    // If you want to search multiple directory blocks for the same name, you 
    // can speed things up by preserving the contents of the temporary buffer 
    // across multilpe calls to this routine.  This routine uses the temporary 
    // buffer to cache the MacRoman conversion of the UTF-8 name.
    //
    // directoryBlockPtr must point to an MFS directory block.  See MFSMDBCheck 
    // for information on how to locate these.
    //
    // directoryBlockSizeInBytes must be the size of that block.
    //
    // utf8Name must be a pointer to a UTF-8 string (either precomposed or decomposed) 
    // of the name that you're looking for.
    //
    // utf8NameLen is the length of that name in bytes.
    //
    // tempBuffer must point to a buffer of at least 
    // kMFSDirectoryBlockFindEntryByNameTempBufferSize bytes.  See the discussion 
    // above for information about how to set up this buffer.
    //
    // dirOffsetPtr must not be NULL.  On entry, *dirOffsetPtr is ignored.  Or success, 
    // *dirOffsetPtr will be the offset of the found directory entry within the 
    // directory block.
    //
    // On entry, if attr is NULL, no attributes are returned.  OTOH, if attr is not 
    // NULL then, on success, the directory entry's attributes will be returned 
    // in *attr.  See MFSDirectoryEntryGetAttr for specific details on this.
 
extern int MFSDirectoryEntryGetAttr(
    const void *        directoryBlockPtr, 
    size_t              dirOffset, 
    struct vnode_attr * attr
);
    // Gets attributes for an MFS directory entry.
    //
    // directoryBlockPtr must point to an MFS directory block.  See MFSMDBCheck 
    // for information on how to locate these.
    //
    // dirOffset must be the offset of a directory entry within that block. 
    // You typically get this by calling (MFSDirectoryBlockIterate or 
    // MFSDirectoryBlockFindEntryByName).
    // 
    // attr must not be NULL; it is handled as per the VFS plug-in's VNOPGetattr 
    // entry point.
    //
    // Note that, with one exception, this routine always returns all of the 
    // attributes that are available for the directory entry, because the cost of 
    // returns those values is trivial.  That is, it (typically) ignores 
    // the va_active field of attr.  The one exception is va_name; because getting 
    // the directory entry's name requires MacRoman to UTF-8 conversion, which is 
    // potentially time consuming, this routine only returns va_name if you request it.
 
extern int MFSDirectoryEntryGetFinderInfo(
    const void *        directoryBlockPtr, 
    size_t              dirOffset, 
    void *              finderInfoPtr
);
    // Gets the Finder information for an MFS directory entry.
    //
    // directoryBlockPtr must point to an MFS directory block.  See MFSMDBCheck 
    // for information on how to locate these.
    //
    // dirOffset must be the offset of a directory entry within that block. 
    // You typically get this by calling (MFSDirectoryBlockIterate or 
    // MFSDirectoryBlockFindEntryByName).
    //
    // finderInfoPtr must point to a buffer of 16 bytes.  On success, this 
    // buffer will contain the directory entry's Finder information.
    //
    // IMPORTANT
    // MFS only supports 16 bytes of Finder information (the 16 bytes of extended 
    // Finder information were added with HFS).  This corresponds to the FileInfo 
    // structure in "Finder.h"
 
// The MFSForkInfo structure is used to return information about a file's fork.
// 
// firstAllocationBlock is the allocation block number of the first allocation 
// block of the fork.  The caller is not expected to interpret this.  Rather, 
// if it needs to get the extents of the fork, it should call MFSForkGetExtent.
//
// lengthInBytes is the logical length of the fork.
//
// physicalLengthInBytes is the logical length of the fork, that is, the number 
// of bytes that it's consuming on the disk.
 
struct MFSForkInfo {
    uint16_t            firstAllocationBlock;
    uint32_t            lengthInBytes;
    uint32_t            physicalLengthInBytes;
};
typedef struct MFSForkInfo MFSForkInfo;
 
extern int MFSDirectoryEntryGetForkInfo(
    const void *        directoryBlockPtr, 
    size_t              dirOffset,
    size_t              forkIndex,
    MFSForkInfo *       forkInfo
);
    // Returns information about a file's fork.
    //
    // directoryBlockPtr must point to an MFS directory block.  See MFSMDBCheck 
    // for information on how to locate these.
    //
    // dirOffset must be the offset of a directory entry within that block. 
    // You typically get this by calling (MFSDirectoryBlockIterate or 
    // MFSDirectoryBlockFindEntryByName).
    //
    // forkIndex must either be 0 for the data fork or 1 for the resource fork.
    //
    // forkInfo must not be NULL.  On entry, *forkInfo is ignored.  On success, 
    // *forkInfo contains the fork information.
 
extern int MFSForkGetExtent(
    const void *        mdbAndVABMPtr,
    const MFSForkInfo * forkInfo,
    uint32_t            forkOffsetInBytes,
    uint32_t *          offsetFromFirstAllocationBlockInBytesPtr,
    uint32_t *          contiguousPhysicalBytesPtr
);
    // Returns information about the location of a fork on disk.
    // 
    // mdbAndVABMPtr must be a pointer to the combined MDB and VABM. 
    // You must get this by reading X bytes starting at the MDB 
    // block (kMFSMDBBlock), where X is the number returned by a call to 
    // MFSMDBCheck (in the *mdbAndVABMSizeInBytesPtr parameter).
    //
    // forkInfo must be a pointer to the fork's information, as returned 
    // by MFSDirectoryEntryGetForkInfo.
    //
    // forkOffsetInBytes is the offset into the fork whose location you 
    // wish to obtain.  This must be an even multiple of the allocation 
    // block size (as returned by MFSMDBCheck).
    //
    // offsetFromFirstAllocationBlockInBytesPtr must not be NULL.  
    // On success, *offsetFromFirstAllocationBlockInBytesPtr will be the 
    // offset, in bytes, from the start of the first allocation block 
    // to forkOffset'th byte of the fork.  You can map this to a disk 
    // block number by dividing it by the block size and then adding it 
    // to the disk block number of the first allocation block (returned 
    // by allocationBlocksStartBlockPtr).
    //
    // contiguousPhysicalBytesPtr must not be NULL.  
    // On success, *contiguousPhysicalBytesPtr will be the number of 
    // physically contiguous bytes of fork data to be found at 
    // *offsetFromFirstAllocationBlockInBytesPtr.
    //
    // Returns 0 on success, EINVAL if you try to get an extent for a non-
    // existant fork (that is, one whose forkInfo->lengthInBytes is 0), or 
    // EPIPE if there are bytes at forkOffset within the file.
    //
    // You can use this routine to find all of a fork's data by setting 
    // forkOffset 0, calling this routine to get the first extent, then 
    // adding contiguousPhysicalBytes to forkOffset, and repeating the call. 
    // When you get EPIPE, you have all of the extents.
 
enum {
    kUTF8ToMFSNameTempBufferSize = 255 * sizeof(uint16_t)
};
 
extern errno_t UTF8ToMFSName(const char *utf8Name, size_t utf8NameLen, void *tempBuffer, uint8_t *mfsName);
extern size_t  MFSNameToUTF8(const uint8_t *name, char *utf8Name, size_t utf8NameSize);
extern void MFSNameToUpper(uint8_t *mfsName);
extern boolean_t MFSNameEqualToUpper(const uint8_t *mfsName, const uint8_t *mfsNameUpper);
extern struct timespec MFSDateTimeToTimeSpec(uint32_t mfsDateTime);
    // These are exported purely for the benefit of the test engine.  Thus, the comments 
    // are attached to the implementation.
    // 
    // Well, not quite.  MFSNameToUTF8 is actually used by the MFSLives.util tool to 
    // convert OSTypes to UTF-8.  That's kinda lame, but the alternative was to write 
    // a bunch more code that doesn't really add any value.
 
#endif