X-Git-Url: https://wimlib.net/git/?p=wimlib;a=blobdiff_plain;f=src%2Flzx-decompress.c;h=1e8532741b3b9f2240db9a3a69246ed4049e0df6;hp=a66e6d8b2126384880ca943fb2979141a4182340;hb=3d8ef754a66f76c8f7121b65a4e466bce6a75f0f;hpb=a6f5add5e9811584ebd75591a6a25cb9686da9a8 diff --git a/src/lzx-decompress.c b/src/lzx-decompress.c index a66e6d8b..1e853274 100644 --- a/src/lzx-decompress.c +++ b/src/lzx-decompress.c @@ -1,13 +1,11 @@ /* * lzx-decompress.c * - * LZX decompression routines, originally based on code taken from cabextract - * v0.5, which was, itself, a modified version of the lzx decompression code - * from unlzx. + * A very fast decompressor for LZX, as used in WIM files. */ /* - * Copyright (C) 2012 Eric Biggers + * Copyright (C) 2012, 2013, 2014 Eric Biggers * * This file is part of wimlib, a library for working with WIM files. * @@ -26,261 +24,221 @@ */ /* - * LZX is a LZ77 and Huffman-code based compression format that has many - * similarities to the DEFLATE format used in zlib. The compression ratio is as - * good or better than DEFLATE. However, in WIM files only up to 32768 bytes of - * data can ever compressed be in the same LZX block, so a .tar.gz file could - * potentially be smaller than a WIM file that uses LZX compression because it - * can use a larger LZ77 window size. + * LZX is an LZ77 and Huffman-code based compression format that has many + * similarities to DEFLATE (the format used by zlib/gzip). The compression + * ratio is as good or better than DEFLATE. See lzx-compress.c for a format + * overview, and see https://en.wikipedia.org/wiki/LZX_(algorithm) for a + * historical overview. Here I make some pragmatic notes. * - * Some notes on the LZX compression format as used in Windows Imaging (WIM) - * files: + * The old specification for LZX is the document "Microsoft LZX Data Compression + * Format" (1997). It defines the LZX format as used in cabinet files. Allowed + * window sizes are 2^n where 15 <= n <= 21. However, this document contains + * several errors, so don't read too much into it... * - * A compressed WIM resource consists of a table of chunk offsets followed by - * the compressed chunks themselves. All compressed chunks except possibly the - * last decompress to WIM_CHUNK_SIZE (= 32768) bytes. This is quite similar to - * the cabinet (.cab) file format, but they are not the same. According to the - * cabinet format documentation, the LZX block size is independent from the - * CFDATA blocks, and a LZX block may span several CFDATA blocks. However, in - * WIMs, LZX blocks do not appear to ever span multiple WIM chunks. Note that - * this means any WIM chunk may be decompressed or compressed independently from - * any other chunk, which is convenient. + * The new specification for LZX is the document "[MS-PATCH]: LZX DELTA + * Compression and Decompression" (2014). It defines the LZX format as used by + * Microsoft's binary patcher. It corrects several errors in the 1997 document + * and extends the format in several ways --- namely, optional reference data, + * up to 2^25 byte windows, and longer match lengths. * - * A LZX compressed WIM chunk contains one or more LZX blocks of the aligned, - * verbatim, or uncompressed block types. For aligned and verbatim blocks, the - * size of the block in uncompressed bytes is specified by a bit following the 3 - * bits that specify the block type, possibly followed by an additional 16 bits. - * '1' means to use the default block size (equal to 32768, the size of a WIM - * chunk--- and this seems to only be valid for the first LZX block in a WIM - * chunk), while '0' means that the block size is provided by the next 16 bits. + * WIM files use a more restricted form of LZX. No LZX DELTA extensions are + * present, the window is not "sliding", E8 preprocessing is done + * unconditionally with a fixed file size, and the maximum window size is always + * 2^15 bytes (equal to the size of each "chunk" in a compressed WIM resource). + * This code is primarily intended to implement this form of LZX. But although + * not compatible with WIMGAPI, this code also supports maximum window sizes up + * to 2^21 bytes. * - * The cabinet format, as documented, allows for the possibility that a - * compressed CFDATA chunk is up to 6144 bytes larger than the data it - * uncompresses to. However, in the WIM format it appears that every chunk that - * would be 32768 bytes or more when compressed is actually stored fully - * uncompressed. - * - * The 'e8' preprocessing step that changes x86 call instructions to use - * absolute offsets instead of relative offsets relies on a filesize parameter. - * There is no such parameter for this in the WIM files (even though the size of - * the file resource could be used for this purpose), and instead a magic file - * size of 12000000 is used. The 'e8' preprocessing is always done, and there - * is no bit to indicate whether it is done or not. + * TODO: Add support for window sizes up to 2^25 bytes. */ -/* - * Some more notes about errors in Microsoft's LZX documentation: - * - * Microsoft's LZX document and their implementation of the com.ms.util.cab Java - * package do not concur. - * - * In the LZX document, there is a table showing the correlation between window - * size and the number of position slots. It states that the 1MB window = 40 - * slots and the 2MB window = 42 slots. In the implementation, 1MB = 42 slots, - * 2MB = 50 slots. The actual calculation is 'find the first slot whose position - * base is equal to or more than the required window size'. This would explain - * why other tables in the document refer to 50 slots rather than 42. - * - * The constant NUM_PRIMARY_LENS used in the decompression pseudocode is not - * defined in the specification. - * - * The LZX document states that aligned offset blocks have their aligned offset - * huffman tree AFTER the main and length trees. The implementation suggests - * that the aligned offset tree is BEFORE the main and length trees. - * - * The LZX document decoding algorithm states that, in an aligned offset block, - * if an extra_bits value is 1, 2 or 3, then that number of bits should be read - * and the result added to the match offset. This is correct for 1 and 2, but - * not 3, where just a huffman symbol (using the aligned tree) should be read. - * - * Regarding the E8 preprocessing, the LZX document states 'No translation may - * be performed on the last 6 bytes of the input block'. This is correct. - * However, the pseudocode provided checks for the *E8 leader* up to the last 6 - * bytes. If the leader appears between -10 and -7 bytes from the end, this - * would cause the next four bytes to be modified, at least one of which would - * be in the last 6 bytes, which is not allowed according to the spec. - * - * The specification states that the huffman trees must always contain at least - * one element. However, many CAB files contain blocks where the length tree is - * completely empty (because there are no matches), and this is expected to - * succeed. - */ +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include "wimlib/decompressor_ops.h" +#include "wimlib/decompress_common.h" +#include "wimlib/error.h" +#include "wimlib/lzx.h" +#include "wimlib/util.h" -#include "util.h" -#include "lzx.h" -#include "decompress.h" #include -/* Huffman decoding tables and maps from symbols to code lengths. */ +/* These values are chosen for fast decompression. */ +#define LZX_MAINCODE_TABLEBITS 11 +#define LZX_LENCODE_TABLEBITS 10 +#define LZX_PRECODE_TABLEBITS 6 +#define LZX_ALIGNEDCODE_TABLEBITS 7 + +/* Huffman decoding tables, and arrays that map symbols to codeword lengths. */ struct lzx_tables { - u16 maintree_decode_table[(1 << LZX_MAINTREE_TABLEBITS) + - (LZX_MAINTREE_NUM_SYMBOLS * 2)]; - u8 maintree_lens[LZX_MAINTREE_NUM_SYMBOLS]; + u16 maincode_decode_table[(1 << LZX_MAINCODE_TABLEBITS) + + (LZX_MAINCODE_MAX_NUM_SYMBOLS * 2)] + _aligned_attribute(DECODE_TABLE_ALIGNMENT); + u8 maincode_lens[LZX_MAINCODE_MAX_NUM_SYMBOLS]; - u16 lentree_decode_table[(1 << LZX_LENTREE_TABLEBITS) + - (LZX_LENTREE_NUM_SYMBOLS * 2)]; - u8 lentree_lens[LZX_LENTREE_NUM_SYMBOLS]; + u16 lencode_decode_table[(1 << LZX_LENCODE_TABLEBITS) + + (LZX_LENCODE_NUM_SYMBOLS * 2)] + _aligned_attribute(DECODE_TABLE_ALIGNMENT); + u8 lencode_lens[LZX_LENCODE_NUM_SYMBOLS]; - u16 alignedtree_decode_table[(1 << LZX_ALIGNEDTREE_TABLEBITS) + - (LZX_ALIGNEDTREE_NUM_SYMBOLS * 2)]; - u8 alignedtree_lens[LZX_ALIGNEDTREE_NUM_SYMBOLS]; -}; + u16 alignedcode_decode_table[(1 << LZX_ALIGNEDCODE_TABLEBITS) + + (LZX_ALIGNEDCODE_NUM_SYMBOLS * 2)] + _aligned_attribute(DECODE_TABLE_ALIGNMENT); + u8 alignedcode_lens[LZX_ALIGNEDCODE_NUM_SYMBOLS]; +} _aligned_attribute(DECODE_TABLE_ALIGNMENT); +/* The main LZX decompressor structure. + * + * Note: we keep track of most of the decompression state outside this + * structure. This structure only exists so that (1) we can store @window_order + * and @num_main_syms for multiple calls to lzx_decompress(); and (2) so that we + * don't have to allocate the large 'struct lzx_tables' on the stack. */ +struct lzx_decompressor { + unsigned window_order; + unsigned num_main_syms; + struct lzx_tables tables; +}; -/* - * Reads a Huffman-encoded symbol using the pre-tree. - */ -static inline int read_huffsym_using_pretree(struct input_bitstream *istream, - const u16 pretree_decode_table[], - const u8 pretree_lens[], unsigned *n) +/* Read a Huffman-encoded symbol using the precode. */ +static inline u16 +read_huffsym_using_precode(struct input_bitstream *istream, + const u16 precode_decode_table[]) { - return read_huffsym(istream, pretree_decode_table, pretree_lens, - LZX_PRETREE_NUM_SYMBOLS, LZX_PRETREE_TABLEBITS, n, - LZX_MAX_CODEWORD_LEN); + return read_huffsym(istream, precode_decode_table, + LZX_PRECODE_TABLEBITS, LZX_MAX_PRE_CODEWORD_LEN); } -/* Reads a Huffman-encoded symbol using the main tree. */ -static inline int read_huffsym_using_maintree(struct input_bitstream *istream, - const struct lzx_tables *tables, - unsigned *n) +/* Read a Huffman-encoded symbol using the main code. */ +static inline u16 +read_huffsym_using_maincode(struct input_bitstream *istream, + const struct lzx_tables *tables) { - return read_huffsym(istream, tables->maintree_decode_table, - tables->maintree_lens, LZX_MAINTREE_NUM_SYMBOLS, - LZX_MAINTREE_TABLEBITS, n, LZX_MAX_CODEWORD_LEN); + return read_huffsym(istream, tables->maincode_decode_table, + LZX_MAINCODE_TABLEBITS, LZX_MAX_MAIN_CODEWORD_LEN); } -/* Reads a Huffman-encoded symbol using the length tree. */ -static inline int read_huffsym_using_lentree(struct input_bitstream *istream, - const struct lzx_tables *tables, - unsigned *n) +/* Read a Huffman-encoded symbol using the length code. */ +static inline u16 +read_huffsym_using_lencode(struct input_bitstream *istream, + const struct lzx_tables *tables) { - return read_huffsym(istream, tables->lentree_decode_table, - tables->lentree_lens, LZX_LENTREE_NUM_SYMBOLS, - LZX_LENTREE_TABLEBITS, n, LZX_MAX_CODEWORD_LEN); + return read_huffsym(istream, tables->lencode_decode_table, + LZX_LENCODE_TABLEBITS, LZX_MAX_LEN_CODEWORD_LEN); } -/* Reads a Huffman-encoded symbol using the aligned offset tree. */ -static inline int read_huffsym_using_alignedtree(struct input_bitstream *istream, - const struct lzx_tables *tables, - unsigned *n) +/* Read a Huffman-encoded symbol using the aligned offset code. */ +static inline u16 +read_huffsym_using_alignedcode(struct input_bitstream *istream, + const struct lzx_tables *tables) { - return read_huffsym(istream, tables->alignedtree_decode_table, - tables->alignedtree_lens, - LZX_ALIGNEDTREE_NUM_SYMBOLS, - LZX_ALIGNEDTREE_TABLEBITS, n, 8); + return read_huffsym(istream, tables->alignedcode_decode_table, + LZX_ALIGNEDCODE_TABLEBITS, LZX_MAX_ALIGNED_CODEWORD_LEN); } /* - * Reads the pretree from the input, then uses the pretree to decode @num_lens - * code length values from the input. + * Read the precode from the compressed input bitstream, then use it to decode + * @num_lens codeword length values. + * + * @istream: + * The input bitstream. + * + * @lens: + * An array that contains the length values from the previous time the + * codeword lengths for this Huffman code were read, or all 0's if this is + * the first time. * - * @istream: The bit stream for the input. It is positioned on the beginning - * of the pretree for the code length values. - * @lens: An array that contains the length values from the previous time - * the code lengths for this Huffman tree were read, or all - * 0's if this is the first time. - * @num_lens: Number of length values to decode and return. + * @num_lens: + * Number of length values to decode. * + * Returns 0 on success, or -1 if the data was invalid. */ -static int lzx_read_code_lens(struct input_bitstream *istream, u8 lens[], - unsigned num_lens) +static int +lzx_read_codeword_lens(struct input_bitstream *istream, u8 lens[], unsigned num_lens) { - /* Declare the decoding table and length table for the pretree. */ - u16 pretree_decode_table[(1 << LZX_PRETREE_TABLEBITS) + - (LZX_PRETREE_NUM_SYMBOLS * 2)]; - u8 pretree_lens[LZX_PRETREE_NUM_SYMBOLS]; - unsigned i; - unsigned len; + /* Declare the decoding table and length table for the precode. */ + u16 precode_decode_table[(1 << LZX_PRECODE_TABLEBITS) + + (LZX_PRECODE_NUM_SYMBOLS * 2)] + _aligned_attribute(DECODE_TABLE_ALIGNMENT); + u8 precode_lens[LZX_PRECODE_NUM_SYMBOLS]; int ret; - /* Read the code lengths of the pretree codes. There are 20 lengths of - * 4 bits each. */ - for (i = 0; i < LZX_PRETREE_NUM_SYMBOLS; i++) { - ret = bitstream_read_bits(istream, LZX_PRETREE_ELEMENT_SIZE, - &len); - if (ret != 0) - return ret; - pretree_lens[i] = len; + /* Read the lengths of the precode codewords. These are given + * explicitly. */ + for (int i = 0; i < LZX_PRECODE_NUM_SYMBOLS; i++) { + precode_lens[i] = bitstream_read_bits(istream, + LZX_PRECODE_ELEMENT_SIZE); } - /* Make the decoding table for the pretree. */ - ret = make_huffman_decode_table(pretree_decode_table, - LZX_PRETREE_NUM_SYMBOLS, - LZX_PRETREE_TABLEBITS, - pretree_lens, - LZX_MAX_CODEWORD_LEN); - if (ret != 0) + /* Make the decoding table for the precode. */ + ret = make_huffman_decode_table(precode_decode_table, + LZX_PRECODE_NUM_SYMBOLS, + LZX_PRECODE_TABLEBITS, + precode_lens, + LZX_MAX_PRE_CODEWORD_LEN); + if (ret) return ret; - /* Pointer past the last length value that needs to be filled in. */ + /* Pointer past the last length value that needs to be filled in. */ u8 *lens_end = lens + num_lens; - while (1) { - - /* Decode a symbol from the input. If the symbol is between 0 - * and 16, it is the difference from the old length. If it is - * between 17 and 19, it is a special code that indicates that - * some number of the next lengths are all 0, or some number of - * the next lengths are all equal to the next symbol in the - * input. */ - unsigned tree_code; - unsigned num_zeroes; - unsigned code; - unsigned num_same; - char value; - - ret = read_huffsym_using_pretree(istream, pretree_decode_table, - pretree_lens, &tree_code); - if (ret != 0) - return ret; - switch (tree_code) { - case 17: /* Run of 0's */ - ret = bitstream_read_bits(istream, 4, &num_zeroes); - if (ret != 0) - return ret; - num_zeroes += 4; - while (num_zeroes--) { + for (;;) { + + unsigned presym; + unsigned run_len; + signed char value; + + /* Decode a symbol from the input. + * + * If the symbol is between 0 and 16, it is the difference from + * the old length, modulo 17. + * + * If the symbol is between 17 and 19, it is a special symbol + * that indicates that some number of the next lengths are all + * 0, or that some number of the next lengths are all equal to + * the next symbol. */ + + presym = read_huffsym_using_precode(istream, + precode_decode_table); + switch (presym) { + + case 17: /* Run of 0's */ + run_len = 4 + bitstream_read_bits(istream, 4); + do { *lens = 0; if (++lens == lens_end) return 0; - } + } while (--run_len); break; - case 18: /* Longer run of 0's */ - ret = bitstream_read_bits(istream, 5, &num_zeroes); - if (ret != 0) - return ret; - num_zeroes += 20; - while (num_zeroes--) { + + case 18: /* Longer run of 0's */ + run_len = 20 + bitstream_read_bits(istream, 5); + do { *lens = 0; if (++lens == lens_end) return 0; - } + } while (--run_len); break; - case 19: /* Run of identical lengths */ - ret = bitstream_read_bits(istream, 1, &num_same); - if (ret != 0) - return ret; - num_same += 4; - ret = read_huffsym_using_pretree(istream, - pretree_decode_table, - pretree_lens, &code); - if (ret != 0) - return ret; - value = (char)*lens - (char)code; + case 19: /* Run of identical lengths */ + run_len = 4 + bitstream_read_bits(istream, 1); + presym = read_huffsym_using_precode(istream, + precode_decode_table); + value = (signed char)*lens - (signed char)presym; if (value < 0) value += 17; - while (num_same--) { + do { *lens = value; if (++lens == lens_end) return 0; - } + } while (--run_len); break; - default: /* Difference from old length. */ - value = (char)*lens - (char)tree_code; + + default: /* Difference from old length */ + value = (signed char)*lens - (signed char)presym; if (value < 0) value += 17; *lens = value; @@ -292,278 +250,226 @@ static int lzx_read_code_lens(struct input_bitstream *istream, u8 lens[], } /* - * Reads the header for an LZX-compressed block. + * Read the header of an LZX block and save the block type and size in + * *block_type_ret and *block_size_ret, respectively. * - * @istream: The input bitstream. - * @block_size_ret: A pointer to an int into which the size of the block, - * in bytes, will be returned. - * @block_type_ret: A pointer to an int into which the type of the block - * (LZX_BLOCKTYPE_*) will be returned. - * @tables: A pointer to a lzx_tables structure in which the - * main tree, the length tree, and possibly the - * aligned offset tree will be constructed. - * @queue: A pointer to the least-recently-used queue into which - * R0, R1, and R2 will be written (only for uncompressed - * blocks, which contain this information in the header) + * If the block is compressed, also update the Huffman decode @tables with the + * new Huffman codes. + * + * If the block is uncompressed, also update the match offset @queue with the + * new match offsets. + * + * Return 0 on success, or -1 if the data was invalid. */ -static int lzx_read_block_header(struct input_bitstream *istream, - unsigned *block_size_ret, - unsigned *block_type_ret, - struct lzx_tables *tables, - struct lru_queue *queue) +static int +lzx_read_block_header(struct input_bitstream *istream, + unsigned num_main_syms, + unsigned window_order, + int *block_type_ret, + u32 *block_size_ret, + struct lzx_tables *tables, + struct lzx_lru_queue *queue) { - int ret; int block_type; - unsigned block_size; - int s; - int i; - unsigned len; - int32_t R[3]; - - ret = bitstream_ensure_bits(istream, 4); - if (ret != 0) { - ERROR("LZX input stream overrun"); - return ret; - } + u32 block_size; + int ret; - /* The first three bits tell us what kind of block it is, and are one - * of the LZX_BLOCKTYPE_* values. */ - block_type = bitstream_read_bits_nocheck(istream, 3); + bitstream_ensure_bits(istream, 4); - /* The next bit indicates whether the block size is the default (32768), - * indicated by a 1 bit, or whether the block size is given by the next - * 16 bits, indicated by a 0 bit. */ - s = bitstream_read_bits_nocheck(istream, 1); + /* The first three bits tell us what kind of block it is, and should be + * one of the LZX_BLOCKTYPE_* values. */ + block_type = bitstream_pop_bits(istream, 3); - if (s == 1) { - block_size = 1 << 15; + /* Read the block size. This mirrors the behavior of + * lzx_write_compressed_block() in lzx-compress.c; see that for more + * details. */ + if (bitstream_pop_bits(istream, 1)) { + block_size = LZX_DEFAULT_BLOCK_SIZE; } else { - ret = bitstream_read_bits(istream, 16, &block_size); - if (ret != 0) - return ret; - block_size = le16_to_cpu(block_size); + u32 tmp; + block_size = 0; + + tmp = bitstream_read_bits(istream, 8); + block_size |= tmp; + tmp = bitstream_read_bits(istream, 8); + block_size <<= 8; + block_size |= tmp; + + if (window_order >= 16) { + tmp = bitstream_read_bits(istream, 8); + block_size <<= 8; + block_size |= tmp; + } } switch (block_type) { + case LZX_BLOCKTYPE_ALIGNED: - /* Read the path lengths for the elements of the aligned tree, - * then build it. */ - - for (i = 0; i < LZX_ALIGNEDTREE_NUM_SYMBOLS; i++) { - ret = bitstream_read_bits(istream, - LZX_ALIGNEDTREE_ELEMENT_SIZE, - &len); - if (ret != 0) - return ret; - tables->alignedtree_lens[i] = len; + + /* Read the aligned offset code and prepare its decode table. + */ + + for (int i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) { + tables->alignedcode_lens[i] = + bitstream_read_bits(istream, + LZX_ALIGNEDCODE_ELEMENT_SIZE); } - LZX_DEBUG("Building the aligned tree."); - ret = make_huffman_decode_table(tables->alignedtree_decode_table, - LZX_ALIGNEDTREE_NUM_SYMBOLS, - LZX_ALIGNEDTREE_TABLEBITS, - tables->alignedtree_lens, - 8); - if (ret != 0) { - ERROR("lzx_decompress(): Failed to make the decode " - "table for the aligned offset tree"); + ret = make_huffman_decode_table(tables->alignedcode_decode_table, + LZX_ALIGNEDCODE_NUM_SYMBOLS, + LZX_ALIGNEDCODE_TABLEBITS, + tables->alignedcode_lens, + LZX_MAX_ALIGNED_CODEWORD_LEN); + if (ret) return ret; - } /* Fall though, since the rest of the header for aligned offset - * blocks is the same as that for verbatim blocks */ + * blocks is the same as that for verbatim blocks. */ case LZX_BLOCKTYPE_VERBATIM: - if (block_type == LZX_BLOCKTYPE_VERBATIM) - LZX_DEBUG("Found verbatim block."); - - LZX_DEBUG("Reading path lengths for main tree."); - /* Read the path lengths for the first 256 elements of the main - * tree. */ - ret = lzx_read_code_lens(istream, tables->maintree_lens, - LZX_NUM_CHARS); - if (ret != 0) { - ERROR("lzx_decompress(): Failed to read the code " - "lengths for the first 256 elements of the " - "main tree"); + + /* Read the main code and prepare its decode table. + * + * Note that the codeword lengths in the main code are encoded + * in two parts: one part for literal symbols, and one part for + * match symbols. */ + + ret = lzx_read_codeword_lens(istream, tables->maincode_lens, + LZX_NUM_CHARS); + if (ret) return ret; - } - /* Read the path lengths for the remaining elements of the main - * tree. */ - LZX_DEBUG("Reading path lengths for remaining elements of " - "main tree (%d elements).", - LZX_MAINTREE_NUM_SYMBOLS - LZX_NUM_CHARS); - ret = lzx_read_code_lens(istream, - tables->maintree_lens + LZX_NUM_CHARS, - LZX_MAINTREE_NUM_SYMBOLS - LZX_NUM_CHARS); - if (ret != 0) { - ERROR("lzx_decompress(): Failed to read the path " - "lengths for the remaining elements of the main " - "tree"); + ret = lzx_read_codeword_lens(istream, + tables->maincode_lens + LZX_NUM_CHARS, + num_main_syms - LZX_NUM_CHARS); + if (ret) return ret; - } - LZX_DEBUG("Building the Huffman decoding " - "table for the main tree."); - - ret = make_huffman_decode_table(tables->maintree_decode_table, - LZX_MAINTREE_NUM_SYMBOLS, - LZX_MAINTREE_TABLEBITS, - tables->maintree_lens, - LZX_MAX_CODEWORD_LEN); - if (ret != 0) { - ERROR("lzx_decompress(): Failed to make the decode " - "table for the main tree"); + ret = make_huffman_decode_table(tables->maincode_decode_table, + num_main_syms, + LZX_MAINCODE_TABLEBITS, + tables->maincode_lens, + LZX_MAX_MAIN_CODEWORD_LEN); + if (ret) return ret; - } - LZX_DEBUG("Reading path lengths for the length tree."); - ret = lzx_read_code_lens(istream, tables->lentree_lens, - LZX_LENTREE_NUM_SYMBOLS); - if (ret != 0) { - ERROR("lzx_decompress(): Failed to read the path " - "lengths for the length tree"); + /* Read the length code and prepare its decode table. */ + + ret = lzx_read_codeword_lens(istream, tables->lencode_lens, + LZX_LENCODE_NUM_SYMBOLS); + if (ret) return ret; - } - LZX_DEBUG("Building the length tree."); - ret = make_huffman_decode_table(tables->lentree_decode_table, - LZX_LENTREE_NUM_SYMBOLS, - LZX_LENTREE_TABLEBITS, - tables->lentree_lens, - LZX_MAX_CODEWORD_LEN); - if (ret != 0) { - ERROR("lzx_decompress(): Failed to build the length " - "Huffman tree"); + ret = make_huffman_decode_table(tables->lencode_decode_table, + LZX_LENCODE_NUM_SYMBOLS, + LZX_LENCODE_TABLEBITS, + tables->lencode_lens, + LZX_MAX_LEN_CODEWORD_LEN); + if (ret) return ret; - } break; case LZX_BLOCKTYPE_UNCOMPRESSED: - LZX_DEBUG("Found uncompressed block."); - /* Mystery bit! */ - ret = bitstream_read_bits(istream, 1, &i); - if (ret != 0) - return ret; - align_input_bitstream(istream); - ret = bitstream_read_bytes(istream, sizeof(R), R); - if (ret != 0) - return ret; - queue->R0 = le32_to_cpu(R[0]); - queue->R1 = le32_to_cpu(R[1]); - queue->R2 = le32_to_cpu(R[2]); + /* Before reading the three LRU match offsets from the + * uncompressed block header, the stream must be aligned on a + * 16-bit boundary. But, unexpectedly, if the stream is + * *already* aligned, the correct thing to do is to throw away + * the next 16 bits. */ + + if (istream->bitsleft == 0) { + if (istream->data_bytes_left < 14) + return -1; + istream->data += 2; + istream->data_bytes_left -= 2; + } else { + if (istream->data_bytes_left < 12) + return -1; + istream->bitsleft = 0; + istream->bitbuf = 0; + } + queue->R[0] = le32_to_cpu(*(le32*)(istream->data + 0)); + queue->R[1] = le32_to_cpu(*(le32*)(istream->data + 4)); + queue->R[2] = le32_to_cpu(*(le32*)(istream->data + 8)); + istream->data += 12; + istream->data_bytes_left -= 12; break; + default: - LZX_DEBUG("Found invalid block."); - return 1; + /* Unrecognized block type. */ + return -1; } + *block_type_ret = block_type; *block_size_ret = block_size; return 0; } /* - * Decodes a compressed literal match value. It refers to some match_offset to - * a point earlier in the window, and some match_len, for which the data is to - * be copied to the current position in the window. + * Decode a match and copy its bytes into the decompression window. * - * @main_element: The start of the match data, as decoded using the main - * tree. - * @block_type: The type of the block (LZX_BLOCKTYPE_ALIGNED or - * LZX_BLOCKTYPE_VERBATIM) - * @bytes_remaining: The amount of uncompressed data remaining to be - * uncompressed. It is an error if the match - * is longer than @bytes_remaining. - * @window: A pointer to the window into which the uncompressed - * data is being written. - * @window_pos: The current position in the window. - * @tables: Contains the Huffman tables for the block (main, - * length, and also aligned offset only for - * LZX_BLOCKTYPE_ALIGNED) - * @queue: The least-recently used queue for match offsets. - * @istream: The input bitstream. - * - * Returns the length of the match, or -1 on error (match would exceed - * the amount of data needing to be uncompressed, or match refers to data before - * the window, or the input bitstream ended unexpectedly). + * Return the length of the match in bytes, or 0 if the match underflowed the + * window or overflowed the current block. */ -static int lzx_decode_match(int main_element, int block_type, - int bytes_remaining, u8 *window, int window_pos, - const struct lzx_tables *tables, - struct lru_queue *queue, - struct input_bitstream *istream) +static u32 +lzx_decode_match(unsigned main_symbol, int block_type, + u32 bytes_remaining, u8 *window, u32 window_pos, + const struct lzx_tables *tables, + struct lzx_lru_queue *queue, + struct input_bitstream *istream) { unsigned length_header; unsigned position_slot; - unsigned match_len; - unsigned match_offset; - unsigned additional_len; + u32 match_len; + u32 match_offset; unsigned num_extra_bits; - unsigned verbatim_bits; - unsigned aligned_bits; - int ret; - int i; - u8 *match_dest; - u8 *match_src; + u32 verbatim_bits; + u32 aligned_bits; - /* The main element is offset by 256 because values under 256 indicate a - * literal value. */ - main_element -= LZX_NUM_CHARS; + /* The main symbol is offset by 256 because values under 256 indicate a + * literal value. */ + main_symbol -= LZX_NUM_CHARS; /* The length header consists of the lower 3 bits of the main element. * The position slot is the rest of it. */ - length_header = main_element & LZX_NUM_PRIMARY_LENS; - position_slot = main_element >> 3; + length_header = main_symbol & LZX_NUM_PRIMARY_LENS; + position_slot = main_symbol >> 3; /* If the length_header is less than LZX_NUM_PRIMARY_LENS (= 7), it - * gives the match length as the offset from LZX_MIN_MATCH. Otherwise, - * the length is given by an additional symbol encoded using the length - * tree, offset by 9 (LZX_MIN_MATCH + LZX_NUM_PRIMARY_LENS) */ - match_len = LZX_MIN_MATCH + length_header; - if (length_header == LZX_NUM_PRIMARY_LENS) { - ret = read_huffsym_using_lentree(istream, tables, - &additional_len); - if (ret != 0) - return -1; - match_len += additional_len; - } - + * gives the match length as the offset from LZX_MIN_MATCH_LEN. + * Otherwise, the length is given by an additional symbol encoded using + * the length code, offset by 9 (LZX_MIN_MATCH_LEN + + * LZX_NUM_PRIMARY_LENS) */ + match_len = LZX_MIN_MATCH_LEN + length_header; + if (length_header == LZX_NUM_PRIMARY_LENS) + match_len += read_huffsym_using_lencode(istream, tables); /* If the position_slot is 0, 1, or 2, the match offset is retrieved * from the LRU queue. Otherwise, the match offset is not in the LRU * queue. */ - switch (position_slot) { - case 0: - match_offset = queue->R0; - break; - case 1: - match_offset = queue->R1; - swap(queue->R0, queue->R1); - break; - case 2: - /* The queue doesn't work quite the same as a real LRU queue, - * since using the R2 offset doesn't bump the R1 offset down to - * R2. */ - match_offset = queue->R2; - swap(queue->R0, queue->R2); - break; - default: + if (position_slot <= 2) { + /* Note: This isn't a real LRU queue, since using the R2 offset + * doesn't bump the R1 offset down to R2. This quirk allows all + * 3 recent offsets to be handled by the same code. (For R0, + * the swap is a no-op.) */ + match_offset = queue->R[position_slot]; + queue->R[position_slot] = queue->R[0]; + queue->R[0] = match_offset; + } else { /* Otherwise, the offset was not encoded as one the offsets in * the queue. Depending on the position slot, there is a * certain number of extra bits that need to be read to fully * decode the match offset. */ /* Look up the number of extra bits that need to be read. */ - num_extra_bits = lzx_extra_bits[position_slot]; + num_extra_bits = lzx_get_num_extra_bits(position_slot); /* For aligned blocks, if there are at least 3 extra bits, the * actual number of extra bits is 3 less, and they encode a * number of 8-byte words that are added to the offset; there - * is then an additional symbol read using the aligned tree that - * specifies the actual byte alignment. */ + * is then an additional symbol read using the aligned offset + * code that specifies the actual byte alignment. */ if (block_type == LZX_BLOCKTYPE_ALIGNED && num_extra_bits >= 3) { /* There is an error in the LZX "specification" at this @@ -573,278 +479,231 @@ static int lzx_decode_match(int main_element, int block_type, * equal to 3. (Note that in the case with * num_extra_bits == 3, the assignment to verbatim_bits * will just set it to 0. ) */ - ret = bitstream_read_bits(istream, num_extra_bits - 3, - &verbatim_bits); - if (ret != 0) - return -1; - + verbatim_bits = bitstream_read_bits(istream, + num_extra_bits - 3); verbatim_bits <<= 3; - - ret = read_huffsym_using_alignedtree(istream, tables, - &aligned_bits); - if (ret != 0) - return -1; + aligned_bits = read_huffsym_using_alignedcode(istream, + tables); } else { /* For non-aligned blocks, or for aligned blocks with * less than 3 extra bits, the extra bits are added * directly to the match offset, and the correction for * the alignment is taken to be 0. */ - ret = bitstream_read_bits(istream, num_extra_bits, - &verbatim_bits); - if (ret != 0) - return -1; - + verbatim_bits = bitstream_read_bits(istream, num_extra_bits); aligned_bits = 0; } /* Calculate the match offset. */ - match_offset = lzx_position_base[position_slot] + verbatim_bits + - aligned_bits - 2; + match_offset = lzx_position_base[position_slot] + + verbatim_bits + aligned_bits - LZX_OFFSET_OFFSET; /* Update the LRU queue. */ - queue->R2 = queue->R1; - queue->R1 = queue->R0; - queue->R0 = match_offset; - break; + queue->R[2] = queue->R[1]; + queue->R[1] = queue->R[0]; + queue->R[0] = match_offset; } - /* Verify that the match is in the bounds of the part of the window - * currently in use, then copy the source of the match to the current - * position. */ - match_dest = window + window_pos; - match_src = match_dest - match_offset; + /* Validate the match, then copy it to the current position. */ - if (match_len > bytes_remaining) { - ERROR("lzx_decode_match(): Match of length %d bytes overflows " - "uncompressed block size", match_len); - return -1; - } + if (unlikely(match_len > bytes_remaining)) + return 0; - if (match_src < window) { - ERROR("lzx_decode_match(): Match of length %d bytes references " - "data before window (match_offset = %d, window_pos = %d)", - match_len, match_offset, window_pos); - return -1; - } + if (unlikely(match_offset > window_pos)) + return 0; -#if 0 - printf("Match: src %u, dst %u, len %u\n", match_src - window, - match_dest - window, - match_len); - putchar('|'); - for (i = 0; i < match_len; i++) { - match_dest[i] = match_src[i]; - putchar(match_src[i]); - } - putchar('|'); - putchar('\n'); -#else - for (i = 0; i < match_len; i++) - match_dest[i] = match_src[i]; -#endif + lz_copy(&window[window_pos], match_len, match_offset, + &window[window_pos + bytes_remaining]); return match_len; } - - -/* Undo the 'E8' preprocessing, where the targets of x86 CALL instructions were - * changed from relative offsets to absolute offsets. This type of - * preprocessing can be used on any binary data even if it is not actually - * machine code. It seems to always be used in WIM files, even though there is - * no bit to indicate that it actually is used, unlike in the LZX compressed - * format as used in other file formats, where a bit is reserved for that - * purpose. */ -static void undo_call_insn_preprocessing(u8 uncompressed_data[], - unsigned uncompressed_data_len) -{ - int i = 0; - int file_size = LZX_MAGIC_FILESIZE; - int32_t abs_offset; - int32_t rel_offset; - - /* Not enabled in the last 6 bytes, which means the 5-byte call - * instruction cannot start in the last *10* bytes. */ - while (i < uncompressed_data_len - 10) { - if (uncompressed_data[i] != 0xe8) { - i++; - continue; - } - abs_offset = le32_to_cpu(*(int32_t*)(uncompressed_data + i + 1)); - - if (abs_offset >= -i && abs_offset < file_size) { - if (abs_offset >= 0) { - /* "good translation" */ - rel_offset = abs_offset - i; - } else { - /* "compensating translation" */ - rel_offset = abs_offset + file_size; - } - *(int32_t*)(uncompressed_data + i + 1) = - cpu_to_le32(rel_offset); - } - i += 5; - } -} - /* - * Decompresses a compressed block of data from which the header has already - * been read. + * Decompress an LZX-compressed block of data. + * + * @block_type: + * The type of the block (LZX_BLOCKTYPE_VERBATIM or LZX_BLOCKTYPE_ALIGNED). + * + * @block_size: + * The size of the block, in bytes. + * + * @window: + * Pointer to the beginning of the decompression window. + * + * @window_pos: + * The position in the window at which the block starts. * - * @block_type: The type of the block (LZX_BLOCKTYPE_VERBATIM or - * LZX_BLOCKTYPE_ALIGNED) - * @block_size: The size of the block, in bytes. - * @window: Pointer to the decompression window. - * @window_pos: The current position in the window. Will be 0 for the first - * block. - * @tables: The Huffman decoding tables for the block (main, length, and - * aligned offset, the latter only for LZX_BLOCKTYPE_ALIGNED) - * @queue: The least-recently-used queue for match offsets. - * @istream: The input bitstream for the compressed literals. + * @tables: + * The Huffman decoding tables for the block. + * + * @queue: + * The least-recently-used queue for match offsets. + * + * @istream: + * The input bitstream, positioned at the start of the block data. + * + * Returns 0 on success, or -1 if the data was invalid. */ -static int lzx_decompress_block(int block_type, int block_size, u8 *window, - int window_pos, - const struct lzx_tables *tables, - struct lru_queue *queue, - struct input_bitstream *istream) +static int +lzx_decompress_block(int block_type, u32 block_size, + u8 *window, u32 window_pos, + const struct lzx_tables *tables, + struct lzx_lru_queue *queue, + struct input_bitstream *istream) { - unsigned bytes_remaining; - unsigned main_element; - int match_len; - int ret; - - bytes_remaining = block_size; - while (bytes_remaining > 0) { - - ret = read_huffsym_using_maintree(istream, tables, - &main_element); - if (ret != 0) - return ret; - - if (main_element < LZX_NUM_CHARS) { - /* literal: 0 to LZX_NUM_CHARS - 1 */ - window[window_pos + block_size - bytes_remaining] = - main_element; - bytes_remaining--; + u32 block_end; + unsigned main_symbol; + u32 match_len; + + block_end = window_pos + block_size; + while (window_pos < block_end) { + main_symbol = read_huffsym_using_maincode(istream, tables); + if (main_symbol < LZX_NUM_CHARS) { + /* Literal */ + window[window_pos++] = main_symbol; } else { - /* match: LZX_NUM_CHARS to LZX_MAINTREE_NUM_SYMBOLS - 1 */ - match_len = lzx_decode_match(main_element, - block_type, bytes_remaining, window, - block_size + window_pos - - bytes_remaining, - tables, queue, istream); - if (match_len == -1) - return 1; - - bytes_remaining -= match_len; + /* Match */ + match_len = lzx_decode_match(main_symbol, + block_type, + block_end - window_pos, + window, + window_pos, + tables, + queue, + istream); + if (unlikely(match_len == 0)) + return -1; + window_pos += match_len; } } return 0; } -/* - * Decompresses a block of LZX-compressed data using a window size of 32768. - * - * @compressed_data: A pointer to the compressed data. - * @compressed_len: The length of the compressed data, in bytes. - * @uncompressed_data: A pointer to the buffer into which to write the - * uncompressed data. - * @uncompressed_len: The length of the uncompressed data. - * - * Return non-zero on failure. - */ -int lzx_decompress(const void *compressed_data, unsigned compressed_len, - void *uncompressed_data, unsigned uncompressed_len) +static int +lzx_decompress(const void *compressed_data, size_t compressed_size, + void *uncompressed_data, size_t uncompressed_size, + void *_dec) { - struct lzx_tables tables; + struct lzx_decompressor *dec = _dec; struct input_bitstream istream; - struct lru_queue queue; - unsigned bytes_remaining; - unsigned block_size; - unsigned block_type; + struct lzx_lru_queue queue; + u32 window_pos; + int block_type; + u32 block_size; + bool may_have_e8_byte; int ret; - LZX_DEBUG("lzx_decompress (compressed_data = %p, compressed_len = %d, " - "uncompressed_data = %p, uncompressed_len = %d).", - compressed_data, compressed_len, - uncompressed_data, uncompressed_len); + init_input_bitstream(&istream, compressed_data, compressed_size); - wimlib_assert(uncompressed_len <= 32768); + /* Initialize the recent offsets queue. */ + lzx_lru_queue_init(&queue); - memset(tables.maintree_lens, 0, sizeof(tables.maintree_lens)); - memset(tables.lentree_lens, 0, sizeof(tables.lentree_lens)); - queue.R0 = 1; - queue.R1 = 1; - queue.R2 = 1; - bytes_remaining = uncompressed_len; + /* Codeword lengths begin as all 0's for delta encoding purposes. */ + memset(dec->tables.maincode_lens, 0, sizeof(dec->tables.maincode_lens)); + memset(dec->tables.lencode_lens, 0, sizeof(dec->tables.lencode_lens)); - init_input_bitstream(&istream, compressed_data, compressed_len); + /* Set this to true if there may be 0xe8 bytes in the uncompressed data. + */ + may_have_e8_byte = false; /* The compressed data will consist of one or more blocks. The * following loop decompresses one block, and it runs until there all * the compressed data has been decompressed, so there are no more * blocks. */ - while (bytes_remaining != 0) { - - LZX_DEBUG("Reading block header."); - ret = lzx_read_block_header(&istream, &block_size, - &block_type, &tables, &queue); - if (ret != 0) + for (window_pos = 0; + window_pos < uncompressed_size; + window_pos += block_size) + { + ret = lzx_read_block_header(&istream, dec->num_main_syms, + dec->window_order, &block_type, + &block_size, &dec->tables, &queue); + if (ret) return ret; - LZX_DEBUG("block_size = %u, bytes_remaining = %u", - block_size, bytes_remaining); + if (block_size > uncompressed_size - window_pos) + return -1; - if (block_size > bytes_remaining) { - ERROR("lzx_decompress(): Expected a block size of at " - "most %u bytes (found %u bytes)", - bytes_remaining, block_size); - return 1; - } + if (block_type != LZX_BLOCKTYPE_UNCOMPRESSED) { - switch (block_type) { - case LZX_BLOCKTYPE_VERBATIM: - case LZX_BLOCKTYPE_ALIGNED: - if (block_type == LZX_BLOCKTYPE_VERBATIM) - LZX_DEBUG("LZX_BLOCKTYPE_VERBATIM"); - else - LZX_DEBUG("LZX_BLOCKTYPE_ALIGNED"); + /* Compressed block. */ ret = lzx_decompress_block(block_type, block_size, uncompressed_data, - uncompressed_len - - bytes_remaining, - &tables, &queue, &istream); - if (ret != 0) + window_pos, + &dec->tables, + &queue, + &istream); + if (ret) return ret; - break; - case LZX_BLOCKTYPE_UNCOMPRESSED: - LZX_DEBUG("LZX_BLOCKTYPE_UNCOMPRESSED"); - ret = bitstream_read_bytes(&istream, block_size, - uncompressed_data + - uncompressed_len - - bytes_remaining); - if (ret != 0) - return ret; - if (block_size & 1) - align_input_bitstream(&istream); - break; - default: - wimlib_assert(0); - break; - } - bytes_remaining -= block_size; + /* If the first 0xe8 byte was in this block, it must + * have been encoded as a literal using mainsym 0xe8. */ + if (dec->tables.maincode_lens[0xe8] != 0) + may_have_e8_byte = true; + } else { + + /* Uncompressed block. */ + + if (istream.data_bytes_left < block_size) + return -1; - if (bytes_remaining != 0) - LZX_DEBUG("%d bytes remaining.", bytes_remaining); + memcpy(&((u8*)uncompressed_data)[window_pos], istream.data, + block_size); + istream.data += block_size; + istream.data_bytes_left -= block_size; + + /* Re-align the bitstream if an odd number of bytes was + * read. */ + if (istream.data_bytes_left && (block_size & 1)) { + istream.data_bytes_left--; + istream.data++; + } + may_have_e8_byte = true; + } } - if (uncompressed_len >= 10) - undo_call_insn_preprocessing(uncompressed_data, - uncompressed_len); + /* Postprocess the data unless it cannot possibly contain 0xe8 bytes */ + if (may_have_e8_byte) + lzx_undo_e8_preprocessing(uncompressed_data, uncompressed_size); return 0; } + +static void +lzx_free_decompressor(void *_dec) +{ + struct lzx_decompressor *dec = _dec; + + ALIGNED_FREE(dec); +} + +static int +lzx_create_decompressor(size_t max_block_size, void **dec_ret) +{ + struct lzx_decompressor *dec; + unsigned window_order; + + window_order = lzx_get_window_order(max_block_size); + if (window_order == 0) + return WIMLIB_ERR_INVALID_PARAM; + + /* The aligned allocation is needed to ensure that the lzx_tables are + * aligned properly. */ + dec = ALIGNED_MALLOC(sizeof(struct lzx_decompressor), + DECODE_TABLE_ALIGNMENT); + if (!dec) + return WIMLIB_ERR_NOMEM; + + dec->window_order = window_order; + dec->num_main_syms = lzx_get_num_main_syms(window_order); + + *dec_ret = dec; + return 0; +} + +const struct decompressor_ops lzx_decompressor_ops = { + .create_decompressor = lzx_create_decompressor, + .decompress = lzx_decompress, + .free_decompressor = lzx_free_decompressor, +};