Windows CE installation cabinet (.CAB) file format

About the format

The Windows CE installation cabinet format was designed by Microsoft make it easy to create packaged software for the Windows CE platform. After generating all his executable programs, data files, help files and so on, a developer can then write an "installation script" which lists the name of the software package, where each file should be installed on the target device's filesystem, registry entries to make on the target system's Windows Registry, and so on. A tool called "CabWiz" takes the installation script and produces a single CAB file with a special control file stored in it. On the target Windows CE device, a program called WCELOAD.EXE will read cabinet files, install them, register them as installed packages, set their registry entries, and so on.

The Cabinet format was previously developed by Microsoft for creating a simple archive format for packaged software, from entire operating systems down to ActiveX Web plugins. They also intended to displace Sun's JAR format for packaging Java software, but this did not succeed.

Overall cabinet file format

Firstly, the overall file format of a Windows CE installation cabinet is the standard Microsoft Cabinet file format. This was originally documented in the Microsoft document CABFMT.DOC as part of the Microsoft Cabinet SDK (CAB-SDK.EXE), it is now documented in the Windows Installer documentation. The format can also be gleaned from open source tools and libraries such as cabextract and libmspack. Windows CE installation cabinets typically use no compression (the NONE compression method in the cabinet specification), sometimes MSZIP compression.

Cabinet file contents

The filenames of the files in the cabinet use the MS-DOS "8.3" convention, where the first eight characters in the filename come from the "true filename", and are followed by a period and three numeric digits. The digits form a decimal number from 000 to 999. The name part before the period is currently always 8 characters (shorter names are padded on the left with "0"), but this shouldn't be relied upon. Only the numeric extension should be used to identify the file. There should not be more than one file in the cabinet with the same numeric extension. As an example:

Name in cab fileOriginal filename
BLKJAC~4.000--- (installation data)
BLACKJ~1.999Blackjack Setup.dll
SELFRE~1.006SelfRegister.dll
0Blkjack.005Blkjack.exe
00Sample.004Sample.BJL
WINGAM~1.003Win Game.wav
LOSEGA~1.002Lose Game.wav
BLACKJ~1.001Blackjack Help.htp

The order of these files within the cab file is important. The first file in the cabinet should be the ".000" file. It should then be followed by all other files in descending numerical order of their extensions.

These original filenames are not recoverable, they are not stored anywhere. The only names available are the target filenames given in the FILES section of the .000 file.

All installation data is contained in the .000 file. If a .999 file exists, it is an "application specific Setup.DLL"; a Microsoft standard Dynamic Link Library that will be loaded and executed after installation by the installation program. All other files are files for installation on the Windows CE device, and the three digits should be resolved into an integer "file number".

Format of the .000 file

All multi-byte integers are little-endian. Values given in this document are in decimal, unless otherwise specified. The overall file format of the .000 file is as follows:

Only the fixed size header has a fixed file offset, all other strings and sections may be at any file offset, in any order. They are typically in the order given above. Each file offset is given in the fixed size header.

The .000 file begins at file offset 0 with a fixed size header of exactly 100 bytes. The format of this header is as follows:

OffsetSize in bytesDescription
04An ASCII signature, "MSCE". This is 0x4543534D as a little-endian integer
44Unknown purpose, usually 0
84The overall length of this .000 header file, in bytes
124Unknown purpose, usually 0
164Unknown purpose, usually 1
204Target architecture for this cabinet: see Appendix A
244Minimal version of WinCE (major version number) required to install this cabinet, or 0 to indicate no restriction
284Minimal version of WinCE (minor version number) required to install this cabinet, or 0 to indicate no restriction
324Maximal version of WinCE (major version number) required to install this cabinet, or 0 to indicate no restriction
364Maximal version of WinCE (minor version number) required to install this cabinet, or 0 to indicate no restriction
404Minmal version of WinCE (build number) required to install this cabinet, or 0 to indicate no restriction
444Maximal version of WinCE (build number) required to install this cabinet, or 0 to indicate no restriction
482The number of entries in the STRINGS section
502The number of entries in the DIRS section
522The number of entries in the FILES section
542The number of entries in the REGHIVES section
562The number of entries in the REGKEYS section
582The number of entries in the LINKS section
604The file offset of the STRINGS section in bytes
644The file offset of the DIRS section in bytes
684The file offset of the FILES section in bytes
724The file offset of the REGHIVES section in bytes
764The file offset of the REGKEYS section in bytes
804The file offset of the LINKS section, in bytes
842The file offset of the APPNAME string in bytes
862The length of the APPNAME string in bytes, including null terminating byte
882The file offset of the PROVIDER string in bytes
902The length of the PROVIDER string in bytes, including null terminating byte
922The file offset of the UNSUPPORTED multi string in bytes
942The length of the UNSUPPORTED multi string in bytes, including null terminating byte
962Unknown purpose, usually 0
982Unknown purpose, usually 0

The APPNAME string

This is a null-terminated ASCII string that gives the name of the packaged application.

The PROVIDER string

This is a null-terminated ASCII string that gives the name of the author or publisher of the packaged application.

The UNSUPPORTED multi string

If this string has more than zero bytes length according to the fixed size header, then this is a sequence of consecutive null terminated ASCII strings. The sequence always terminates with an empty string. Each string is the ASCII name of a Windows CE platform where this software is not supported and will not run. Platforms seen in this sequence include "PALM PC", "HPC", "JORDAN", "PALM PC2" and "POCKETPC".

The STRINGS section

The STRINGS section is a number of common ASCII strings that may be used in the installation data. In other sections, strings from the STRINGS section are referred to by their string ID. The total number of entries in this section is given in the fixed size header. There is no padding between string entries. The format of each STRING entry is as follows:

  • 2 bytes: An integer string ID
  • 2 bytes: The length of the string, including the null terminator byte
  • X bytes: the string itself, for the length given above

As an example, the hexadecimal byte stream 04 00 09 00 53 6F 66 74 77 61 72 65 00 decodes to "ID=4 (04 00), length=9 (09 00), string=Software". The hexadecimal byte stream 05 00 0A 00 4D 69 63 72 6F 73 6F 66 74 00 decodes to "ID=5 (05 00), length=10 (0A 00), string=Microsoft".

The DIRS section

The DIRS section is a list of directory paths used in the installation. In other sections, paths from the DIRS section are referred to by their directory ID. The total number of entries in this section is given in the fixed size header. There is no padding between directory entries. The format of each directory entry is as follows:

  • 2 bytes: An integer directory ID
  • 2 bytes: The length of the following directory specification, in bytes
  • X bytes: A directory specification. This is simply an array of 16-bit (2 byte) numbers, where each number is a string ID. The final number is always 0, which terminates the list

For example, the hexidecimal byte stream 01 00 0A 00 01 00 02 00 03 00 04 00 00 translates "dir ID=1 (01 00), length=10 (0A 00), specification={1, 2, 3, 4}". Assuming string ID 1 is "foo", string ID 2 is "bar", string ID 3 is "baz" and string ID 4 is "quux", the specification translates from {1, 2, 3, 4} to "foo\bar\baz\quux".

The directory specification may begin with a path member of the form "%CEn%", where n is a decimal integer from 1 to 17. This is an abbreviation for one of the CE standard directories, as listed in Appendix B.

The FILES section

The FILES section is a list of files to be installed on the target device. All files in the list should be installed. In other sections, files from the FILES section are referred to by their file ID. The total number of entries in this section is given in the fixed size header. There is no padding between file entries. The format of each file entry is as follows:

  • 2 bytes: An integer file ID. The file for installation is the file in the cabinet which has this integer as a three digit file extension
  • 2 bytes: The directory ID for where this file should be installed
  • 2 bytes: Unknown purpose, usually the same as the file ID
  • 4 bytes: File flags. A 32-bit integer. The following bits are known:
    FlagDescription
    Bit 31 (0x80000000) If bit is set, this file is a reference-counting shared file. It is not deleted at uninstall time unless its reference count is 0
    Bit 30 (0x40000000)If bit is set, ignore file date (stored in the cabinet file) and always overwrite target (on CE device). Mutually exclusive with bit 29
    Bit 29 (0x20000000)If bit is set, do not overwrite target if target is newer. Mutually exclusive with bit 30
    Bit 28 (0x10000000)If bit is set, self-register this DLL
    Bit 10 (0x00000400)If bit is set, do not copy this file to the target unless the target already exists. Mutually exclusive with bit 4
    Bit 4 (0x00000010)If bit is set, do not overwrite target if it already exists. Mutually exclusive with bit 10
    Bit 1 (0x00000002)If bit is set, do not skip this file
    Bit 0 (0x00000001)If bit is set, warn the user if this file is skipped
    All other bits are unknown.
  • 2 bytes: the length of the installed file's filename, including the null terminator
  • X bytes: the installed file's filename, with the length given above. A null terminated ASCII string

The REGHIVES section

The REGHIVES section is a list of hives in the Windows Registry. These hives are then used in the REGKEYS section for actual registry entries. In other sections, hives from the REGHIVES section are referred to by their hive ID. The total number of entries in this section is given in the fixed size header. There is no padding between hive entries. The format of each hive entry is as follows:

  • 2 bytes: An integer hive ID
  • 2 bytes: The hive root, an integer from 1 to 4:
    1. HKCR: HKEY_CLASSES_ROOT
    2. HKCU: HKEY_CURRENT_USER
    3. HKLM: HKEY_LOCAL_MACHINE
    4. HKU: HKEY_USERS
  • 2 bytes: Unknown purpose, usually 0
  • 2 bytes: Length of the registry hive specification to follow, in bytes
  • X bytes: A registry hive specification. This is simply an array of 16-bit (2 byte) numbers, where each number is a string ID. The final number is always 0, which terminates the list

The REGKEYS section

The REGKEYS is a list of registry entries -- key / data pairs -- to be added to the Windows Registry upon installation. All entries in the list should be added. The total number of entries in this section is given in the fixed size header. There is no padding between entries. The format of each entry is as follows:

  • 2 bytes: An integer key ID
  • 2 bytes: The hive ID of the hive to store this entry in
  • 2 bytes: A Boolean value, either 1 if variable substitution should be performed or 0 if not. The entry data should have substrings such as %InstallDir% or one the %CEn% values
  • 4 bytes: The type of entry data, and flags.
    FlagDescription
    Bit 16 set, bit 0 set (0x00010001)This key is of type TYPE_DWORD, a 32-bit integer
    Bit 16 clear, bit 0 clear (0x00000000)This key is of type TYPE_SZ, a null terminated ASCII string
    Bit 16 set, bit 0 clear (0x00010000)This key is of type TYPE_MULTI_SZ, a list of null terminated ASCII strings
    Bit 16 clear, bit 0 set (0x00000001)This key is of type TYPE_BINARY, raw binary data in no particular format
    Bit 1 (0x00000002)This is the only known flag, NOCLOBBER. If this bit is set, the registry entry should not be overwritten if it already exists in the registry
    All other bits are undefined.
  • 2 bytes: The length of entry data to follow, in bytes
  • X bytes: Registry entry data. This begins with a null terminated ASCII string, giving the name of the registry key. If the empty string is used here, this is the default key for the given hive. Immediately following the null byte of the string is the data for the registry entry, whose format depends on the entry type given in the flags:
    • TYPE_DWORD: four bytes, giving a single 32-bit integer
    • TYPE_SZ: a single null terminated ASCII string
    • TYPE_MULTI_SZ: a sequence of consecutive null terminated ASCII strings. The sequence always terminates with an empty string
    • TYPE_BINARY: no specific format, just raw data

The LINKS section is a list of symbolic links to be made in the standard installation directories, to link to installed directories or files. The total number of links in this section is given in the fixed size header. There is no padding between links. Each link has the following format:

  • 2 bytes: An integer link ID
  • 2 bytes: Unknown purpose
  • 2 bytes: The base directory where the link will be stored, either 0 to indicate "%InstallDir%", the default directory where the application is installed, or a number from 1 to 17 to indicate "%CEn%", one of the standard installation directories as defined in Appendix B. The link specification, given below, should be added to this base directory, to get the full name of the link
  • 2 bytes: Integer ID of the file or directory that the link should point to. It is either a file ID, or a directory ID, depending on the link type. If it is a directory ID, the ID 0 can be used to refer to %InstallDir%
  • 2 bytes: The link type. This is 0 if the link target is a directory, or 1 if the link target is a file
  • 2 bytes: The length of the link specification in bytes
  • X bytes: The link specification. This is simply an array of 16-bit (2 byte) numbers, where each number is a string ID. The final number is always 0, which terminates the list

Appendix A: a list of processor architectures

ValueDescription
0No specific architecture
103SHx SH3
104SHx SH4
386Intel 386
486Intel 486
586Intel Pentium
601PowerPC 601
603PowerPC 603
604PowerPC 604
620PowerPC 620
821Motorola 821
1824ARM 720
2080ARM 820
2336ARM 920
2577StrongARM
4000MIPS R4000
10003Hitachi SH3
10004Hitachi SH3E
10005Hitachi SH4
21064Alpha 21064
70001ARM 7TDMI

Appendix B: a list of %CEn% shortcuts

First, for a Handheld PC:

ValueStandard directory
%CE1%\Program Files
%CE2%\Windows
%CE3%\Windows\Desktop
%CE4%\Windows\StartUp
%CE5%\My Documents
%CE6%\Program Files\Accessories
%CE7%\Program Files\Communications
%CE8%\Program Files\Games
%CE9%\Program Files\Pocket Outlook
%CE10%\Program Files\Office
%CE11%\Windows\Programs
%CE12%\Windows\Programs\Accessories
%CE13%\Windows\Programs\Communications
%CE14%\Windows\Programs\Games
%CE15%\Windows\Fonts
%CE16%\Windows\Recent
%CE17%\Windows\Favorites

For a Pocket PC version 1.0 or 2.0, as per Handheld PCs, except

ValueStandard directory
%CE3%undefined
%CE9%undefined
%CE10%undefined
%CE11%\Windows\Start Menu\Programs
%CE12%\Windows\Start Menu\Accessories
%CE13%\Windows\Start Menu\Communications
%CE14%\Windows\Start Menu\Games
%CE16%undefined
%CE17%\Windows\Start Menu

For a Pocket PC version 3.0, as for 1.0 or 2.0, except

ValueStandard directory
%CE6%undefined
%CE7%undefined
%CE8%undefined
%CE12%undefined
%CE13%undefined
%CE16%undefined