LZX pre/post-processing improvements

[wimlib] / README

                                  INTRODUCTION

This is wimlib version 1.7.4-BETA (December 2014).  wimlib is a C library for
creating, modifying, extracting, and mounting files in the Windows Imaging
Format (WIM files).  wimlib and its command-line frontend 'wimlib-imagex'
provide a free and cross-platform alternative to Microsoft's WIMGAPI, ImageX,
and DISM.

                                  INSTALLATION

To install wimlib and wimlib-imagex on Windows, simply download and extract the
ZIP file containing the latest binaries from the SourceForge page
(http://sourceforge.net/projects/wimlib/).  You probably have already done this!

To install wimlib and wimlib-imagex on UNIX-like systems (with Linux being the
primary supported and tested platform), you must compile the source code, which
is also available at http://sourceforge.net/projects/wimlib/.  Alternatively,
check if a package has been prepared for your Linux distribution.  Example files
for Debian and RPM packaging are in the debian/ and rpm/ directories.

                                    WIM FILES

A Windows Imaging (WIM) file is an archive designed primarily for archiving
Windows filesystems.  However, it can be used on other platforms as well, with
some limitations.  Like some other archive formats such as ZIP, files in WIM
archives may be compressed.  WIM files support multiple compression formats,
including LZX, XPRESS, and LZMS.  All these formats are supported by wimlib.

A WIM file consists of one or more "images".  Each image is an independent
top-level directory structure and is logically separate from all other images in
the WIM.  Each image has a name as well as a 1-based index in the WIM file.  To
save space, WIM archives automatically combine all duplicate files across all
images.

A WIM file may be either stand-alone or split into multiple parts.  Split WIMs
are read-only and cannot be modified.

Since version 1.6.0, wimlib also supports ESD (.esd) files, except when
encrypted.  These are still WIM files but they use a newer version of the file
format.

                             IMAGEX IMPLEMENTATION

wimlib itself is a C library, and it provides a documented public API (See:
http://wimlib.sourceforge.net) for other programs to use.  However, it is also
distributed with a command-line program called "wimlib-imagex" that uses this
library to implement an imaging tool similar to Microsoft's ImageX.
wimlib-imagex supports almost all the capabilities of Microsoft's ImageX as well
as additional capabilities.  wimlib-imagex works on both UNIX-like systems and
Windows, although some features differ between the platforms.

Run `wimlib-imagex' with no arguments to see an overview of the available
commands and their syntax.  For additional documentation:

  * If you have installed wimlib-imagex on a UNIX-like system, you will find
    further documentation in the man pages; run `man wimlib-imagex' to get
    started.

  * If you have downloaded the Windows binary distribution, you will find the
    documentation for wimlib-imagex in PDF format in the "doc" directory,
    ready for viewing with any PDF viewer.  Please note that although the PDF
    files are converted from UNIX-style "man pages", they do document
    Windows-specific behavior when appropriate.

                                COMPRESSION RATIO

wimlib (and wimlib-imagex) can create XPRESS, LZX, or LZMS compressed WIM files.
wimlib includes its own compression codecs and does not use the compression API
available on some versions of Windows.

I have gradually been improving the compression codecs in wimlib.  For XPRESS
and LZX, they now usually outperform and outcompress the equivalent Microsoft
implementations.  Although results will vary depending on the data being
compressed, in the table below I present the results for a common use case:
compressing an x86 Windows PE image.  Each row displays the compression type,
the size of the resulting WIM file in bytes, and how many seconds it took to
create the file.  When applicable, the results with the equivalent Microsoft
implementation in WIMGAPI is included.

  =============================================================================
  | Compression            ||  wimlib (v1.7.2)       |  WIMGAPI (Windows 8.1) |
  =============================================================================
  | None             [1]   ||  361,314,224 in 3.4s   |  361,315,338 in 4.5s   |
  | XPRESS           [2]   ||  138,380,918 in 4.2s   |  140,457,487 in 6.3s   |
  | XPRESS (slow)    [3]   ||  135,269,627 in 11.1s  |  N/A                   |
  | LZX (quick)      [4]   ||  130,332,081 in 4.7s   |  N/A                   |
  | LZX (normal)     [5]   ||  126,714,941 in 12.9s  |  127,293,240 in 19.2s  |
  | LZX (slow)       [6]   ||  126,150,725 in 23.4s  |  N/A                   |
  | LZMS (non-solid) [7]   ||  121,909,750 in 13.3s  |  N/A                   |
  | LZMS (solid)     [8]   ||  93,650,894  in 44.4s  |  88,771,192 in 109.2   |
  | "WIMBoot"        [9]   ||  167,095,369 in 6.4s   |  169,109,650 in 10.7s  |
  | "WIMBoot" (slow) [10]  ||  165,195,668 in 9.5s   |  N/A                   |
  =============================================================================

Notes:
   [1] '--compress=none' for wimlib-imagex; '/compress:none' for DISM.

   [2] '--compress=XPRESS' for wimlib-imagex; '/compress:fast' for DISM.
       Compression chunk size defaults to 32768 bytes in both cases.

   [3] '--compress=XPRESS:80' for wimlib-imagex; no known equivalent for DISM.
       Compression chunk size defaults to 32768 bytes.

   [4] '--compress=LZX:20' for wimlib-imagex; no known equivalent for DISM.
       Compression chunk size defaults to 32768 bytes.

   [5] '--compress=LZX' or '--compress=LZX:50' or no option for wimlib-imagex;
       '/compress:maximum' for DISM.
       Compression chunk size defaults to 32768 bytes in both cases.

   [6] '--compress=LZX:100' for wimlib-imagex; no known equivalent for DISM.
       Compression chunk size defaults to 32768 bytes.

   [7] '--compress=LZMS' for wimlib-imagex; no known equivalent for DISM.
       Compression chunk size defaults to 131072 bytes.

   [8] '--solid' for wimlib-imagex.  Should be '/compress:recovery' for DISM,
       but only works for /Export-Image, not /Capture-Image.  Compression chunk
       size in solid blocks defaults to 33554432 for wimlib, 67108864 for DISM.

   [9] '--wimboot' for wimlib-imagex; '/wimboot' for DISM.
       This is really XPRESS compression with 4096 byte chunks, so the same as
       '--compress=XPRESS --chunk-size=4096'.

   [10] '--wimboot --compress=XPRESS:80' for wimlib-imagex;
        no known equivalent for DISM.
        Same format as [9], but trying harder to get a good compression ratio.

Note: wimlib-imagex's --compress option also accepts the "fast", "maximum", and
"recovery" aliases for XPRESS, LZX, and LZMS, respectively.

Testing environment:

    - 64 bit binaries
    - Windows 8.1 virtual machine running on Linux with VT-x
    - 4 CPUs and 4 GiB memory given to virtual machine
    - SSD-backed virtual disk
    - All tests done with page cache warmed

The compression ratio provided by wimlib is also competitive with commonly used
archive formats.  Below are file sizes that result when the Canterbury corpus is
compressed with wimlib (v1.7.2), WIMGAPI (Windows 8.1), and some other
formats/programs:

     =====================================================
     | Format                             | Size (bytes) |
     =====================================================
     | tar                                | 2,826,240    |
     | WIM (WIMGAPI, None)                | 2,814,254    |
     | WIM (wimlib, None)                 | 2,814,216    |
     | WIM (WIMGAPI, XPRESS)              | 825,536      |
     | WIM (wimlib, XPRESS)               | 790,016      |
     | tar.gz (gzip, default)             | 738,796      |
     | ZIP (Info-ZIP, default)            | 735,334      |
     | tar.gz (gzip, -9)                  | 733,971      |
     | ZIP (Info-ZIP, -9)                 | 732,297      |
     | WIM (wimlib, LZX quick)            | 704,006      |
     | WIM (WIMGAPI, LZX)                 | 651,866      |
     | WIM (wimlib, LZX normal)           | 632,614      |
     | WIM (wimlib, LZX slow)             | 625,050      |
     | WIM (wimlib, LZMS non-solid)       | 581,960      |
     | tar.bz2 (bzip, default)            | 565,008      |
     | tar.bz2 (bzip, -9)                 | 565,008      |
     | WIM (wimlib, LZX solid)            | 532,700      |
     | WIM (wimlib, LZMS solid)           | 525,990      |
     | WIM (wimlib, LZX solid, slow)      | 525,140      |
     | WIM (wimlib, LZMS solid, slow)     | 523,728      |
     | WIM (WIMGAPI, LZMS solid)          | 521,366      |
     | WIM (wimlib, LZX solid, very slow) | 520,832      |
     | tar.xz (xz, default)               | 486,916      |
     | tar.xz (xz, -9)                    | 486,904      |
     | 7z  (7-zip, default)               | 484,700      |
     | 7z  (7-zip, -9)                    | 483,239      |
     =====================================================

Note: WIM does even better on directory trees containing duplicate files, which
the Canterbury corpus doesn't have.

                                  NTFS SUPPORT

WIM images may contain data, such as alternate data streams and
compression/encryption flags, that are best represented on the NTFS filesystem
used on Windows.  Also, WIM images may contain security descriptors which are
specific to Windows and cannot be represented on other operating systems.
wimlib handles this NTFS-specific or Windows-specific data in a
platform-dependent way:

  * In the Windows version of wimlib and wimlib-imagex, NTFS-specific and
    Windows-specific data are supported natively.

  * In the UNIX version of wimlib and wimlib-imagex, NTFS-specific and
    Windows-specific data are ordinarily ignored; however, there is also special
    support for capturing and extracting images directly to/from unmounted NTFS
    volumes.  This was made possible with the help of libntfs-3g from the
    NTFS-3g project.

For both platforms the code for NTFS capture and extraction is complete enough
that it is possible to apply an image from the "install.wim" contained in recent
Windows installation media (Vista, Windows 7, or Windows 8) directly to an NTFS
filesystem, and then boot Windows from it after preparing the Boot Configuration
Data.  In addition, a Windows installation can be captured (or backed up) into a
WIM file, and then re-applied later.

                                   WINDOWS PE

A major use for wimlib and wimlib-imagex is to create customized images of
Windows PE, the Windows Preinstallation Environment, on either UNIX-like systems
or Windows without having to rely on Microsoft's software and its restrictions
and limitations.

Windows PE is a lightweight version of Windows that can run entirely from memory
and can be used to install Windows from local media or a network drive or
perform maintenance.  It is the operating system that runs when you boot from
the Windows installation media.

You can find Windows PE on the installation DVD for Windows Vista, Windows 7, or
Windows 8, in the file `sources/boot.wim'.  Windows PE can also be found in the
Windows Automated Installation Kit (WAIK), which is free to download from
Microsoft, inside the `WinPE.cab' file, which you can extract natively on
Windows, or on UNIX-like systems if you install either the `cabextract' or
`p7zip' programs.

In addition, Windows installations and recovery partitions frequently contain a
WIM containing an image of the Windows Recovery Environment, which is similar to
Windows PE.

A shell script `mkwinpeimg' is distributed with wimlib on UNIX-like systems to
ease the process of creating and customizing a bootable Windows PE image.

                                  DEPENDENCIES

This section documents the dependencies of wimlib and the programs distributed
with it, when building for a UNIX-like system from source.  If you have
downloaded the Windows binary distribution of wimlib and wimlib-imagex then all
dependencies were already included and this section is irrelevant.

* libxml2 (required)
        This is a commonly used free library to read and write XML documents.
        Almost all Linux distributions should include this; however, you may
        need to install the header files, which might be in a package named
        "libxml2-dev" or similar.  For more information see http://xmlsoft.org/.

* libfuse (optional but recommended)
        Unless configured --without-fuse, wimlib requires a non-ancient version
        of libfuse.  Most Linux distributions already include this, but make
        sure you have the libfuse package installed, and also libfuse-dev if
        your distribution distributes header files separately.  FUSE also
        requires a kernel module.  If the kernel module is available it should
        automatically be loaded if you try to mount a WIM image.  For more
        information see http://fuse.sourceforge.net/.

* libattr (optional but recommended)
        Unless configured --without-fuse, wimlib also requires libattr.  Almost
        all Linux distributions should include this; however, you may need to
        install the header files, which might be in a package named "attr-dev",
        "libattr1-dev", or similar.

* libntfs-3g (optional but recommended)
        Unless configured --without-ntfs-3g, wimlib requires the library and
        headers for libntfs-3g version 2011-4-12 or later to be installed.

* OpenSSL / libcrypto (optional)
        wimlib can use the SHA-1 message digest implementation from libcrypto
        (usually provided by OpenSSL) instead of compiling in yet another SHA-1
        implementation.

* cdrkit (optional)
* mtools (optional)
* syslinux (optional)
* cabextract (optional)
        The `mkwinpeimg' shell script will look for several other programs
        depending on what options are given to it.  Depending on your Linux
        distribution, you may already have these programs installed, or they may
        be in the software repository.  Making an ISO filesystem requires
        `mkisofs' from `cdrkit' (http://www.cdrkit.org).  Making a disk image
        requires `mtools' (http://www.gnu.org/software/mtools) and `syslinux'
        (http://www.syslinux.org).  Retrieving files from the Windows Automated
        Installation Kit requires `cabextract' (http://www.cabextract.org.uk).

                                 CONFIGURATION

This section documents the most important options that may be passed to the
"configure" script when building from source:

--without-ntfs-3g
        If libntfs-3g is not available or is not version 2011-4-12 or later,
        wimlib can be built without it, in which case it will not be possible to
        capture or apply WIM images directly from/to NTFS volumes.

        The default is --with-ntfs-3g when building for any UNIX-like system,
        and --without-ntfs-3g when building for Windows.

--without-fuse
        The --without-fuse option completely disables support for mounting WIM
        images.  This removes dependencies on libfuse, librt, and libattr.  The
        wimmount, wimmountrw, and wimunmount commands will not work.

        The default is --with-fuse when building for Linux, and --without-fuse
        otherwise.

--without-libcrypto
        Build in functions for SHA-1 rather than using external SHA-1 functions
        from libcrypto (usually provided by OpenSSL).

        The default is to use libcrypto if it is found on your system.

                                  PORTABILITY

wimlib works on both UNIX-like systems (Linux, Mac OS X, FreeBSD, etc.) and
Windows (XP and later).

As much code as possible is shared among all supported platforms, but there
necessarily are some differences in what features are supported on each platform
and how they are implemented.  Most notable is that file tree scanning and
extraction are implemented separately for Windows, UNIX, and UNIX (NTFS-3g
mode), to ensure a fast and feature-rich implementation of each platform/mode.

wimlib is mainly used on x86 and x86_64 CPUs, but it should also work on a
number of other GCC-supported 32-bit or 64-bit architectures.  It has been
tested on the ARM architecture.

Currently, gcc and clang are the only supported compilers.  A few nonstandard
extensions are used in the code.

                                   REFERENCES

The WIM file format is partially specified in a document that can be found in
the Microsoft Download Center.  However, this document really only provides an
overview of the format and is not a formal specification.  It also does not
cover later extensions of the format, such as solid blocks.

With regards to the supported compression formats:

- Microsoft has official documentation for XPRESS that is of reasonable quality.
- Microsoft has official documentation for LZX, but in two different documents,
  neither of which is completely applicable to its use in the WIM format, and
  the first of which contains multiple errors.
- There does not seem to be any official documentation for LZMS, so my comments
  and code in src/lzms-decompress.c may in fact be the best documentation
  available for this particular compression format.

The algorithms used by wimlib's compression and decompression codecs are
inspired by a variety of sources, including open source projects and computer
science papers.

The code in ntfs-3g_apply.c and ntfs-3g_capture.c uses the NTFS-3g library,
which is a library for reading and writing to NTFS filesystems (the filesystem
used by recent versions of Windows).  See
http://www.tuxera.com/community/ntfs-3g-download/ for more information.

A limited number of other free programs can handle some parts of the WIM
file format:

  * 7-zip is able to extract and create WIMs (as well as files in many
    other archive formats).  However, wimlib is designed specifically to handle
    WIM files and provides features previously only available in Microsoft's
    implementation, such as the ability to mount WIMs read-write as well as
    read-only, the ability to create compressed WIMs, the correct handling of
    security descriptors and hard links, support for LZMS compression, and
    support for solid archives.
  * ImagePyX (https://github.com/maxpat78/ImagePyX) is a Python program that
    provides similar capabilities to wimlib-imagex.  One thing to note, though,
    is that it does not support compression and decompression by itself, but
    instead relies on external native code, such as the codecs from wimlib.

If you are looking for an archive format that provides features similar to WIM
but was designed primarily for UNIX, you may want to consider SquashFS
(http://squashfs.sourceforge.net/).  However, you may find that wimlib works
surprisingly well on UNIX.  It will store hard links and symbolic links, and it
has optional support for storing UNIX owners, groups, modes, and special files
such as device nodes and FIFOs.  Actually, I use it to back up my own files on
Linux!

                             LICENSE AND DISCLAIMER

See COPYING for information about the license.

wimlib is independently developed and does not contain any code, data, or files
copyrighted by Microsoft.  It is not known to be affected by any patents.

On UNIX-like systems, if you do not want wimlib to be dynamically linked with
libcrypto (OpenSSL), configure with --without-libcrypto.  This replaces the SHA1
implementation with built-in code and there will be no difference in
functionality.

wimlib comes with no warranty whatsoever.  Please submit a bug report (to
ebiggers3@gmail.com) if you find a bug in wimlib and/or wimlib-imagex.