4 * XPRESS decompression routines.
6 * Copyright (C) 2012 Eric Biggers
8 * wimlib - Library for working with WIM files
10 * This library is free software; you can redistribute it and/or modify it under
11 * the terms of the GNU Lesser General Public License as published by the Free
12 * Software Foundation; either version 2.1 of the License, or (at your option) any
15 * This library is distributed in the hope that it will be useful, but WITHOUT ANY
16 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A
17 * PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public License along
20 * with this library; if not, write to the Free Software Foundation, Inc., 59
21 * Temple Place, Suite 330, Boston, MA 02111-1307 USA
27 * The XPRESS compression format is a LZ77-based algorithm. That means it is
28 * quite similar to LZX compression, but XPRESS is slightly simpler, so it is a
29 * little faster to compress and decompress.
31 * The XPRESS compression format is mostly documented in a file called "[MS-XCA]
32 * Xpress Compression Algorithm". In the MSDN library, it can currently be
33 * found under Open Specifications => Protocols => Windows Protocols => Windows
34 * Server Protocols => [MS-XCA] Xpress Compression Algorithm". Note that
35 * Microsoft apparently also has either a slightly different format or an
36 * entirely different format that is also called XPRESS. The other one is
37 * supposedly used in Windows' hibernation file or something, but the one used
38 * in WIM files is the one described in the above document.
40 * If you are already familiar with the LZ77 algorithm and Huffman coding, the
41 * XPRESS format is pretty simple. The compressed data begins with 256 bytes
42 * that contain 512 4-bit integers that are the lengths of the symbols in the
43 * Huffman tree used for decoding compressed literals. This is the only Huffman
44 * tree that is used for the entirety of the compressed data, and the codeword
45 * lengths are not encoded with a pretree.
47 * The rest of the compressed data is Huffman-encoded symbols. Values 0 through
48 * 255 are literal bytes. Values 256 through 511 are matches and may require
49 * extra bits or bytes to be read to get the match offset and match length.
51 * There is no notion of a "compressed block" in the XPRESS format, so in the
52 * XPRESS format, each WIM chunk (32768 bytes) will always use only one Huffman
55 * The trickiest part is probably the fact that literal bytes for match lengths
56 * are encoded "separately" from the bitstream.
58 * Also, a caveat--- according to M$'s documentation for XPRESS,
60 * "Some implementation of the decompression algorithm expect an extra
61 * symbol to mark the end of the data. Specifically, some implementations
62 * fail during decompression if the Huffman symbol 256 is not found after
65 * This is the case for WIM files--- in we must write this extra symbol "256" at
66 * the end. Otherwise Microsoft's software will fail to decompress the
67 * XPRESS-compressed data.
69 * Howeve, WIMLIB's decompressor in xpress-decomp.c currently does not care if
70 * this extra symbol is there or not.
84 /* Decodes @huffsym, a value >= XPRESS_NUM_CHARS, that is the header of a match.
86 static int xpress_decode_match(int huffsym, uint window_pos, uint window_len,
87 u8 window[], struct input_bitstream *istream)
91 u8 match_sym = (u8)huffsym;
92 u8 len_hdr = match_sym & 0xf;
93 u8 offset_bsr = match_sym >> 4;
99 ret = bitstream_read_bits(istream, offset_bsr, &match_offset);
102 match_offset |= (1 << offset_bsr);
104 if (len_hdr == 0xf) {
105 ret = bitstream_read_byte(istream);
109 if (match_len == 0xff) {
111 ret = bitstream_read_byte(istream);
116 ret = bitstream_read_byte(istream);
120 match_len |= (ret << 8);
129 match_len += XPRESS_MIN_MATCH;
132 /* Verify that the match is in the bounds of the part of the window
133 * currently in use, then copy the source of the match to the current
136 match_dest = window + window_pos;
137 match_src = match_dest - match_offset;
139 if (window_pos + match_len > window_len) {
140 ERROR("XPRESS dedecompression error: match of length %d "
141 "bytes overflows window\n", match_len);
145 if (match_src < window) {
146 ERROR("XPRESS decompression error: match of length %d bytes "
147 "references data before window (match_offset = "
148 "%d, window_pos = %d)\n", match_len,
149 match_offset, window_pos);
153 for (i = 0; i < match_len; i++)
154 match_dest[i] = match_src[i];
159 /* Decodes the Huffman-encoded matches and literal bytes in a block of
160 * XPRESS-encoded data. */
161 static int xpress_decompress_literals(struct input_bitstream *istream,
162 u8 uncompressed_data[],
163 uint uncompressed_len,
165 const u16 decode_table[])
172 while (curpos < uncompressed_len) {
173 ret = read_huffsym(istream, decode_table, lens,
174 XPRESS_NUM_SYMBOLS, XPRESS_TABLEBITS, &huffsym,
175 XPRESS_MAX_CODEWORD_LEN);
179 if (huffsym < XPRESS_NUM_CHARS) {
180 uncompressed_data[curpos++] = huffsym;
182 match_len = xpress_decode_match(huffsym, curpos,
184 uncompressed_data, istream);
194 int xpress_decompress(const void *__compressed_data, uint compressed_len,
195 void *uncompressed_data, uint uncompressed_len)
197 u8 lens[XPRESS_NUM_SYMBOLS];
198 u16 decode_table[(1 << XPRESS_TABLEBITS) + 2 * XPRESS_NUM_SYMBOLS];
199 struct input_bitstream istream;
201 const u8 *compressed_data;
205 compressed_data = __compressed_data;
208 DEBUG2("compressed_len = %d, uncompressed_len = %d\n",
209 compressed_len, uncompressed_len);
211 /* XPRESS uses only one Huffman tree. It contains 512 symbols, and the
212 * code lengths of these symbols are given literally as 4-bit integers
213 * in the first 256 bytes of the compressed data.
215 if (compressed_len < XPRESS_NUM_SYMBOLS / 2)
216 return WIMLIB_ERR_DECOMPRESSION;
218 for (i = 0; i < XPRESS_NUM_SYMBOLS / 2; i++) {
219 *lens_p++ = compressed_data[i] & 0xf;
220 *lens_p++ = compressed_data[i] >> 4;
223 ret = make_huffman_decode_table(decode_table, XPRESS_NUM_SYMBOLS,
224 XPRESS_TABLEBITS, lens,
225 XPRESS_MAX_CODEWORD_LEN);
229 init_input_bitstream(&istream, compressed_data + XPRESS_NUM_SYMBOLS / 2,
230 compressed_len - XPRESS_NUM_SYMBOLS / 2);
232 return xpress_decompress_literals(&istream, uncompressed_data,
233 uncompressed_len, lens, decode_table);