*/
/*
- * Copyright (C) 2012, 2013 Eric Biggers
+ * Copyright (C) 2012, 2013, 2014 Eric Biggers
*
* This file is part of wimlib, a library for working with WIM files.
*
*/
/*
- * LZX is a LZ77 and Huffman-code based compression format that has many
+ * LZX is an 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.
*
* last decompress to a fixed number of bytes, by default 32768. 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
+ * independent from the CFDATA blocks, and an 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 allows random access.
*
- * A LZX compressed WIM chunk contains one or more LZX blocks of the aligned,
+ * An 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.
#endif
#include "wimlib.h"
-#include "wimlib/decompress.h"
+#include "wimlib/decompressor_ops.h"
+#include "wimlib/decompress_common.h"
#include "wimlib/lzx.h"
#include "wimlib/util.h"
u8 alignedtree_lens[LZX_ALIGNEDCODE_NUM_SYMBOLS];
} _aligned_attribute(DECODE_TABLE_ALIGNMENT);
+struct lzx_decompressor {
+ u32 max_window_size;
+ unsigned num_main_syms;
+ struct lzx_tables tables;
+};
/*
* Reads a Huffman-encoded symbol using the pre-tree.
*/
-static inline int
+static inline u16
read_huffsym_using_pretree(struct input_bitstream *istream,
- const u16 pretree_decode_table[],
- const u8 pretree_lens[], unsigned *n)
+ const u16 pretree_decode_table[])
{
- return read_huffsym(istream, pretree_decode_table, pretree_lens,
- LZX_PRECODE_NUM_SYMBOLS, LZX_PRECODE_TABLEBITS, n,
- LZX_MAX_PRE_CODEWORD_LEN);
+ return read_huffsym(istream, pretree_decode_table,
+ LZX_PRECODE_TABLEBITS, LZX_MAX_PRE_CODEWORD_LEN);
}
/* Reads a Huffman-encoded symbol using the main tree. */
-static inline int
+static inline u16
read_huffsym_using_maintree(struct input_bitstream *istream,
- const struct lzx_tables *tables,
- unsigned *n,
- unsigned num_main_syms)
+ const struct lzx_tables *tables)
{
return read_huffsym(istream, tables->maintree_decode_table,
- tables->maintree_lens, num_main_syms,
- LZX_MAINCODE_TABLEBITS, n, LZX_MAX_MAIN_CODEWORD_LEN);
+ LZX_MAINCODE_TABLEBITS, LZX_MAX_MAIN_CODEWORD_LEN);
}
/* Reads a Huffman-encoded symbol using the length tree. */
-static inline int
+static inline u16
read_huffsym_using_lentree(struct input_bitstream *istream,
- const struct lzx_tables *tables,
- unsigned *n)
+ const struct lzx_tables *tables)
{
return read_huffsym(istream, tables->lentree_decode_table,
- tables->lentree_lens, LZX_LENCODE_NUM_SYMBOLS,
- LZX_LENCODE_TABLEBITS, n, LZX_MAX_LEN_CODEWORD_LEN);
+ LZX_LENCODE_TABLEBITS, LZX_MAX_LEN_CODEWORD_LEN);
}
/* Reads a Huffman-encoded symbol using the aligned offset tree. */
-static inline int
+static inline u16
read_huffsym_using_alignedtree(struct input_bitstream *istream,
- const struct lzx_tables *tables,
- unsigned *n)
+ const struct lzx_tables *tables)
{
return read_huffsym(istream, tables->alignedtree_decode_table,
- tables->alignedtree_lens,
- LZX_ALIGNEDCODE_NUM_SYMBOLS,
- LZX_ALIGNEDCODE_TABLEBITS, n,
- LZX_MAX_ALIGNED_CODEWORD_LEN);
+ LZX_ALIGNEDCODE_TABLEBITS, LZX_MAX_ALIGNED_CODEWORD_LEN);
}
/*
* code length values from the input.
*
* @istream: The bit stream for the input. It is positioned on the beginning
- * of the pretree for the code length values.
+ * 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.
+ * 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.
*
*/
_aligned_attribute(DECODE_TABLE_ALIGNMENT);
u8 pretree_lens[LZX_PRECODE_NUM_SYMBOLS];
unsigned i;
- u32 len;
int ret;
/* Read the code lengths of the pretree codes. There are 20 lengths of
* 4 bits each. */
for (i = 0; i < LZX_PRECODE_NUM_SYMBOLS; i++) {
- ret = bitstream_read_bits(istream, LZX_PRECODE_ELEMENT_SIZE,
- &len);
- if (ret)
- return ret;
- pretree_lens[i] = len;
+ pretree_lens[i] = bitstream_read_bits(istream,
+ LZX_PRECODE_ELEMENT_SIZE);
}
/* Make the decoding table for the pretree. */
u32 num_same;
signed char value;
- ret = read_huffsym_using_pretree(istream, pretree_decode_table,
- pretree_lens, &tree_code);
- if (ret)
- return ret;
+ tree_code = read_huffsym_using_pretree(istream,
+ pretree_decode_table);
switch (tree_code) {
case 17: /* Run of 0's */
- ret = bitstream_read_bits(istream, 4, &num_zeroes);
- if (ret)
- return ret;
+ num_zeroes = bitstream_read_bits(istream, 4);
num_zeroes += 4;
while (num_zeroes--) {
*lens = 0;
}
break;
case 18: /* Longer run of 0's */
- ret = bitstream_read_bits(istream, 5, &num_zeroes);
- if (ret)
- return ret;
+ num_zeroes = bitstream_read_bits(istream, 5);
num_zeroes += 20;
while (num_zeroes--) {
*lens = 0;
}
break;
case 19: /* Run of identical lengths */
- ret = bitstream_read_bits(istream, 1, &num_same);
- if (ret)
- return ret;
+ num_same = bitstream_read_bits(istream, 1);
num_same += 4;
- ret = read_huffsym_using_pretree(istream,
- pretree_decode_table,
- pretree_lens,
- &code);
- if (ret)
- return ret;
+ code = read_huffsym_using_pretree(istream,
+ pretree_decode_table);
value = (signed char)*lens - (signed char)code;
if (value < 0)
value += 17;
*
* @istream: The input bitstream.
* @block_size_ret: A pointer to an int into which the size of the block,
- * in bytes, will be returned.
+ * 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.
+ * (LZX_BLOCKTYPE_*) will be returned.
+ * @tables: A pointer to an 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)
+ * R0, R1, and R2 will be written (only for uncompressed
+ * blocks, which contain this information in the header)
*/
static int
lzx_read_block_header(struct input_bitstream *istream,
unsigned block_type;
unsigned block_size;
- ret = bitstream_ensure_bits(istream, 4);
- if (ret)
- return ret;
+ bitstream_ensure_bits(istream, 4);
/* 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);
+ block_type = bitstream_pop_bits(istream, 3);
/* Read the block size. This mirrors the behavior
* lzx_write_compressed_block() in lzx-compress.c; see that for more
* details. */
- if (bitstream_read_bits_nocheck(istream, 1)) {
+ if (bitstream_pop_bits(istream, 1)) {
block_size = LZX_DEFAULT_BLOCK_SIZE;
} else {
u32 tmp;
block_size = 0;
- ret = bitstream_read_bits(istream, 8, &tmp);
- if (ret)
- return ret;
+ tmp = bitstream_read_bits(istream, 8);
block_size |= tmp;
-
- ret = bitstream_read_bits(istream, 8, &tmp);
- if (ret)
- return ret;
+ tmp = bitstream_read_bits(istream, 8);
block_size <<= 8;
block_size |= tmp;
if (max_window_size >= 65536) {
- ret = bitstream_read_bits(istream, 8, &tmp);
- if (ret)
- return ret;
+ tmp = bitstream_read_bits(istream, 8);
block_size <<= 8;
block_size |= tmp;
}
* then build it. */
for (unsigned i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) {
- u32 len;
-
- ret = bitstream_read_bits(istream,
- LZX_ALIGNEDCODE_ELEMENT_SIZE,
- &len);
- if (ret)
- return ret;
- tables->alignedtree_lens[i] = len;
+ tables->alignedtree_lens[i] =
+ bitstream_read_bits(istream,
+ LZX_ALIGNEDCODE_ELEMENT_SIZE);
}
LZX_DEBUG("Building the aligned tree.");
* tree.
*
* @block_type: The type of the block (LZX_BLOCKTYPE_ALIGNED or
- * LZX_BLOCKTYPE_VERBATIM)
+ * LZX_BLOCKTYPE_VERBATIM)
*
* @bytes_remaining: The amount of uncompressed data remaining to be
- * uncompressed in this block. It is an error if the match
- * is longer than this number.
+ * uncompressed in this block. It is an error if the match
+ * is longer than this number.
*
* @window: A pointer to the window into which the uncompressed
- * data is being written.
+ * data is being written.
*
* @window_pos: The current byte offset in the window.
*
*
* Returns the length of the match, or a negative number on error. The possible
* error cases are:
- * - Match would exceed the amount of data remaining to be uncompressed.
- * - Match refers to data before the window.
- * - The input bitstream ended unexpectedly.
+ * - Match would exceed the amount of data remaining to be uncompressed.
+ * - Match refers to data before the window.
+ * - The input bitstream ended unexpectedly.
*/
static int
lzx_decode_match(unsigned main_element, int block_type,
unsigned position_slot;
unsigned match_len;
unsigned match_offset;
- unsigned additional_len;
unsigned num_extra_bits;
u32 verbatim_bits;
u32 aligned_bits;
unsigned i;
- int ret;
u8 *match_dest;
u8 *match_src;
* the length tree, 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) {
- ret = read_huffsym_using_lentree(istream, tables,
- &additional_len);
- if (ret)
- return ret;
- match_len += additional_len;
- }
-
+ if (length_header == LZX_NUM_PRIMARY_LENS)
+ match_len += read_huffsym_using_lentree(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->R[0];
- break;
- case 1:
- match_offset = queue->R[1];
- swap(queue->R[0], queue->R[1]);
- 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->R[2];
- swap(queue->R[0], queue->R[2]);
- 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
* 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)
- return ret;
-
+ verbatim_bits = bitstream_read_bits(istream,
+ num_extra_bits - 3);
verbatim_bits <<= 3;
-
- ret = read_huffsym_using_alignedtree(istream, tables,
- &aligned_bits);
- if (ret)
- return ret;
+ aligned_bits = read_huffsym_using_alignedtree(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)
- return ret;
-
+ verbatim_bits = bitstream_read_bits(istream, num_extra_bits);
aligned_bits = 0;
}
queue->R[2] = queue->R[1];
queue->R[1] = queue->R[0];
queue->R[0] = match_offset;
- break;
}
/* 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. */
- if (match_len > bytes_remaining) {
+ if (unlikely(match_len > bytes_remaining)) {
LZX_DEBUG("Match of length %u bytes overflows "
"uncompressed block size", match_len);
return -1;
}
- if (match_offset > window_pos) {
+ if (unlikely(match_offset > window_pos)) {
LZX_DEBUG("Match of length %u bytes references "
"data before window (match_offset = %u, "
"window_pos = %u)",
return match_len;
}
-static void
-undo_call_insn_translation(u32 *call_insn_target, int input_pos,
- s32 file_size)
-{
- s32 abs_offset;
- s32 rel_offset;
-
- abs_offset = le32_to_cpu(*call_insn_target);
- if (abs_offset >= -input_pos && abs_offset < file_size) {
- if (abs_offset >= 0) {
- /* "good translation" */
- rel_offset = abs_offset - input_pos;
- } else {
- /* "compensating translation" */
- rel_offset = abs_offset + file_size;
- }
- *call_insn_target = cpu_to_le32(rel_offset);
- }
-}
-
-/* Undo the 'E8' preprocessing, where the targets of x86 CALL instructions were
- * changed from relative offsets to absolute offsets.
- *
- * Note that this call instruction preprocessing can and will be used on any
- * data even if it is not actually x86 machine code. In fact, this type of
- * preprocessing appears to always be used in LZX-compressed resources in WIM
- * files; there is no bit to indicate whether it is used or not, unlike in the
- * LZX compressed format as used in cabinet files, where a bit is reserved for
- * that purpose.
- *
- * Call instruction preprocessing is disabled in the last 6 bytes of the
- * uncompressed data, which really means the 5-byte call instruction cannot
- * start in the last 10 bytes of the uncompressed data. This is one of the
- * errors in the LZX documentation.
- *
- * Call instruction preprocessing does not appear to be disabled after the
- * 32768th chunk of a WIM stream, which is apparently is yet another difference
- * from the LZX compression used in cabinet files.
- *
- * Call instruction processing is supposed to take the file size as a parameter,
- * as it is used in calculating the translated jump targets. But in WIM files,
- * this file size is always the same (LZX_WIM_MAGIC_FILESIZE == 12000000).*/
-static void
-undo_call_insn_preprocessing(u8 uncompressed_data[], int uncompressed_data_len)
-{
- for (int i = 0; i < uncompressed_data_len - 10; i++) {
- if (uncompressed_data[i] == 0xe8) {
- undo_call_insn_translation((u32*)&uncompressed_data[i + 1],
- i,
- LZX_WIM_MAGIC_FILESIZE);
- i += 4;
- }
- }
-}
-
/*
- * Decompresses a LZX-compressed block of data from which the header has already
+ * Decompresses an LZX-compressed block of data from which the header has already
* been read.
*
* @block_type: The type of the block (LZX_BLOCKTYPE_VERBATIM or
- * LZX_BLOCKTYPE_ALIGNED)
+ * LZX_BLOCKTYPE_ALIGNED)
* @block_size: The size of the block, in bytes.
- * @num_main_syms: Number of symbols in the main alphabet.
* @window: Pointer to the decompression window.
* @window_pos: The current position in the window. Will be 0 for the first
- * block.
+ * block.
* @tables: The Huffman decoding tables for the block (main, length, and
- * aligned offset, the latter only for LZX_BLOCKTYPE_ALIGNED)
+ * 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.
*/
static int
lzx_decompress_block(int block_type, unsigned block_size,
- unsigned num_main_syms,
u8 *window,
unsigned window_pos,
const struct lzx_tables *tables,
{
unsigned main_element;
unsigned end;
- int ret;
int match_len;
end = window_pos + block_size;
while (window_pos < end) {
- ret = read_huffsym_using_maintree(istream, tables,
- &main_element,
- num_main_syms);
- if (ret)
- return ret;
-
+ main_element = read_huffsym_using_maintree(istream, tables);
if (main_element < LZX_NUM_CHARS) {
/* literal: 0 to LZX_NUM_CHARS - 1 */
window[window_pos++] = main_element;
tables,
queue,
istream);
- if (match_len < 0)
+ if (unlikely(match_len < 0))
return match_len;
window_pos += match_len;
}
return 0;
}
-/* API function documented in wimlib.h */
-WIMLIBAPI int
-wimlib_lzx_decompress2(const void *compressed_data, unsigned compressed_len,
- void *uncompressed_data, unsigned uncompressed_len,
- u32 max_window_size)
+static int
+lzx_decompress(const void *compressed_data, size_t compressed_size,
+ void *uncompressed_data, size_t uncompressed_size,
+ void *_ctx)
{
- struct lzx_tables tables;
+ struct lzx_decompressor *ctx = _ctx;
struct input_bitstream istream;
struct lzx_lru_queue queue;
unsigned window_pos;
unsigned block_size;
unsigned block_type;
- unsigned num_main_syms;
int ret;
bool e8_preprocessing_done;
- LZX_DEBUG("compressed_data = %p, compressed_len = %u, "
- "uncompressed_data = %p, uncompressed_len = %u, "
+ LZX_DEBUG("compressed_data = %p, compressed_size = %zu, "
+ "uncompressed_data = %p, uncompressed_size = %zu, "
"max_window_size=%u).",
- compressed_data, compressed_len,
- uncompressed_data, uncompressed_len, max_window_size);
+ compressed_data, compressed_size,
+ uncompressed_data, uncompressed_size,
+ ctx->max_window_size);
- if (!lzx_window_size_valid(max_window_size)) {
- LZX_DEBUG("Window size of %u is invalid!",
- max_window_size);
- return -1;
- }
-
- num_main_syms = lzx_get_num_main_syms(max_window_size);
-
- if (uncompressed_len > max_window_size) {
- LZX_DEBUG("Uncompressed chunk size of %u exceeds "
+ if (uncompressed_size > ctx->max_window_size) {
+ LZX_DEBUG("Uncompressed size of %zu exceeds "
"window size of %u!",
- uncompressed_len, max_window_size);
+ uncompressed_size, ctx->max_window_size);
return -1;
}
- memset(tables.maintree_lens, 0, sizeof(tables.maintree_lens));
- memset(tables.lentree_lens, 0, sizeof(tables.lentree_lens));
+ memset(ctx->tables.maintree_lens, 0, sizeof(ctx->tables.maintree_lens));
+ memset(ctx->tables.lentree_lens, 0, sizeof(ctx->tables.lentree_lens));
lzx_lru_queue_init(&queue);
- init_input_bitstream(&istream, compressed_data, compressed_len);
+ init_input_bitstream(&istream, compressed_data, compressed_size);
e8_preprocessing_done = false; /* Set to true if there may be 0xe8 bytes
in the uncompressed data. */
* blocks. */
for (window_pos = 0;
- window_pos < uncompressed_len;
+ window_pos < uncompressed_size;
window_pos += block_size)
{
LZX_DEBUG("Reading block header.");
- ret = lzx_read_block_header(&istream, num_main_syms,
- max_window_size, &block_size,
- &block_type, &tables, &queue);
+ ret = lzx_read_block_header(&istream, ctx->num_main_syms,
+ ctx->max_window_size, &block_size,
+ &block_type, &ctx->tables, &queue);
if (ret)
return ret;
LZX_DEBUG("block_size = %u, window_pos = %u",
block_size, window_pos);
- if (block_size > uncompressed_len - window_pos) {
+ if (block_size > uncompressed_size - window_pos) {
LZX_DEBUG("Expected a block size of at "
- "most %u bytes (found %u bytes)",
- uncompressed_len - window_pos, block_size);
+ "most %zu bytes (found %u bytes)",
+ uncompressed_size - window_pos, block_size);
return -1;
}
LZX_DEBUG("LZX_BLOCKTYPE_ALIGNED");
ret = lzx_decompress_block(block_type,
block_size,
- num_main_syms,
uncompressed_data,
window_pos,
- &tables,
+ &ctx->tables,
&queue,
&istream);
if (ret)
return ret;
- if (tables.maintree_lens[0xe8] != 0)
+ if (ctx->tables.maintree_lens[0xe8] != 0)
e8_preprocessing_done = true;
break;
case LZX_BLOCKTYPE_UNCOMPRESSED:
}
}
if (e8_preprocessing_done)
- undo_call_insn_preprocessing(uncompressed_data, uncompressed_len);
+ lzx_undo_e8_preprocessing(uncompressed_data, uncompressed_size);
return 0;
}
-/* API function documented in wimlib.h */
-WIMLIBAPI int
-wimlib_lzx_decompress(const void *compressed_data, unsigned compressed_len,
- void *uncompressed_data, unsigned uncompressed_len)
+static void
+lzx_free_decompressor(void *_ctx)
{
- return wimlib_lzx_decompress2(compressed_data, compressed_len,
- uncompressed_data, uncompressed_len,
- 32768);
+ struct lzx_decompressor *ctx = _ctx;
+
+ ALIGNED_FREE(ctx);
}
+
+static int
+lzx_create_decompressor(size_t max_window_size,
+ const struct wimlib_decompressor_params_header *params,
+ void **ctx_ret)
+{
+ struct lzx_decompressor *ctx;
+
+ if (!lzx_window_size_valid(max_window_size))
+ return WIMLIB_ERR_INVALID_PARAM;
+
+ ctx = ALIGNED_MALLOC(sizeof(struct lzx_decompressor),
+ DECODE_TABLE_ALIGNMENT);
+ if (ctx == NULL)
+ return WIMLIB_ERR_NOMEM;
+
+ ctx->max_window_size = max_window_size;
+ ctx->num_main_syms = lzx_get_num_main_syms(max_window_size);
+
+ *ctx_ret = ctx;
+ return 0;
+}
+
+const struct decompressor_ops lzx_decompressor_ops = {
+ .create_decompressor = lzx_create_decompressor,
+ .decompress = lzx_decompress,
+ .free_decompressor = lzx_free_decompressor,
+};
git://wimlib.net / wimlib / blobdiff