4 * LZX decompression routines, originally based on code taken from cabextract
5 * v0.5, which was, itself, a modified version of the lzx decompression code
10 * Copyright (C) 2012, 2013 Eric Biggers
12 * This file is part of wimlib, a library for working with WIM files.
14 * wimlib is free software; you can redistribute it and/or modify it under the
15 * terms of the GNU General Public License as published by the Free
16 * Software Foundation; either version 3 of the License, or (at your option)
19 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
20 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
21 * A PARTICULAR PURPOSE. See the GNU General Public License for more
24 * You should have received a copy of the GNU General Public License
25 * along with wimlib; if not, see http://www.gnu.org/licenses/.
29 * LZX is a LZ77 and Huffman-code based compression format that has many
30 * similarities to the DEFLATE format used in zlib. The compression ratio is as
31 * good or better than DEFLATE.
33 * Some notes on the LZX compression format as used in Windows Imaging (WIM)
36 * A compressed WIM resource consists of a table of chunk offsets followed by
37 * the compressed chunks themselves. All compressed chunks except possibly the
38 * last decompress to a fixed number of bytes, by default 32768. This is quite
39 * similar to the cabinet (.cab) file format, but they are not the same.
40 * According to the cabinet format documentation, the LZX block size is
41 * independent from the CFDATA blocks, and a LZX block may span several CFDATA
42 * blocks. However, in WIMs, LZX blocks do not appear to ever span multiple WIM
43 * chunks. Note that this means any WIM chunk may be decompressed or compressed
44 * independently from any other chunk, which allows random access.
46 * A LZX compressed WIM chunk contains one or more LZX blocks of the aligned,
47 * verbatim, or uncompressed block types. For aligned and verbatim blocks, the
48 * size of the block in uncompressed bytes is specified by a bit following the 3
49 * bits that specify the block type, possibly followed by an additional 16 bits.
50 * '1' means to use the default block size (equal to 32768, the default size of
51 * a WIM chunk), while '0' means that the block size is provided by the next 16
54 * The cabinet format, as documented, allows for the possibility that a
55 * compressed CFDATA chunk is up to 6144 bytes larger than the data it
56 * uncompresses to. However, in the WIM format it appears that every chunk that
57 * would be 32768 bytes or more when compressed is actually stored fully
60 * The 'e8' preprocessing step that changes x86 call instructions to use
61 * absolute offsets instead of relative offsets relies on a filesize parameter.
62 * There is no such parameter for this in the WIM files (even though the size of
63 * the file resource could be used for this purpose), and instead a magic file
64 * size of 12000000 is used. The 'e8' preprocessing is always done, and there
65 * is no bit to indicate whether it is done or not.
69 * Some more notes about errors in Microsoft's LZX documentation:
71 * Microsoft's LZX document and their implementation of the com.ms.util.cab Java
72 * package do not concur.
74 * In the LZX document, there is a table showing the correlation between window
75 * size and the number of position slots. It states that the 1MB window = 40
76 * slots and the 2MB window = 42 slots. In the implementation, 1MB = 42 slots,
77 * 2MB = 50 slots. The actual calculation is 'find the first slot whose position
78 * base is equal to or more than the required window size'. This would explain
79 * why other tables in the document refer to 50 slots rather than 42.
81 * The constant NUM_PRIMARY_LENS used in the decompression pseudocode is not
82 * defined in the specification.
84 * The LZX document states that aligned offset blocks have their aligned offset
85 * Huffman tree AFTER the main and length trees. The implementation suggests
86 * that the aligned offset tree is BEFORE the main and length trees.
88 * The LZX document decoding algorithm states that, in an aligned offset block,
89 * if an extra_bits value is 1, 2 or 3, then that number of bits should be read
90 * and the result added to the match offset. This is correct for 1 and 2, but
91 * not 3, where just a Huffman symbol (using the aligned tree) should be read.
93 * Regarding the E8 preprocessing, the LZX document states 'No translation may
94 * be performed on the last 6 bytes of the input block'. This is correct.
95 * However, the pseudocode provided checks for the *E8 leader* up to the last 6
96 * bytes. If the leader appears between -10 and -7 bytes from the end, this
97 * would cause the next four bytes to be modified, at least one of which would
98 * be in the last 6 bytes, which is not allowed according to the spec.
100 * The specification states that the Huffman trees must always contain at least
101 * one element. However, many CAB files contain blocks where the length tree is
102 * completely empty (because there are no matches), and this is expected to
111 #include "wimlib/decompressor_ops.h"
112 #include "wimlib/decompress_common.h"
113 #include "wimlib/lzx.h"
114 #include "wimlib/util.h"
118 /* Huffman decoding tables and maps from symbols to code lengths. */
121 u16 maintree_decode_table[(1 << LZX_MAINCODE_TABLEBITS) +
122 (LZX_MAINCODE_MAX_NUM_SYMBOLS * 2)]
123 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
124 u8 maintree_lens[LZX_MAINCODE_MAX_NUM_SYMBOLS];
127 u16 lentree_decode_table[(1 << LZX_LENCODE_TABLEBITS) +
128 (LZX_LENCODE_NUM_SYMBOLS * 2)]
129 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
130 u8 lentree_lens[LZX_LENCODE_NUM_SYMBOLS];
133 u16 alignedtree_decode_table[(1 << LZX_ALIGNEDCODE_TABLEBITS) +
134 (LZX_ALIGNEDCODE_NUM_SYMBOLS * 2)]
135 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
136 u8 alignedtree_lens[LZX_ALIGNEDCODE_NUM_SYMBOLS];
137 } _aligned_attribute(DECODE_TABLE_ALIGNMENT);
139 struct lzx_decompressor {
141 unsigned num_main_syms;
142 struct lzx_tables tables;
146 * Reads a Huffman-encoded symbol using the pre-tree.
149 read_huffsym_using_pretree(struct input_bitstream *istream,
150 const u16 pretree_decode_table[],
151 const u8 pretree_lens[], unsigned *n)
153 return read_huffsym(istream, pretree_decode_table, pretree_lens,
154 LZX_PRECODE_NUM_SYMBOLS, LZX_PRECODE_TABLEBITS, n,
155 LZX_MAX_PRE_CODEWORD_LEN);
158 /* Reads a Huffman-encoded symbol using the main tree. */
160 read_huffsym_using_maintree(struct input_bitstream *istream,
161 const struct lzx_tables *tables,
163 unsigned num_main_syms)
165 return read_huffsym(istream, tables->maintree_decode_table,
166 tables->maintree_lens, num_main_syms,
167 LZX_MAINCODE_TABLEBITS, n, LZX_MAX_MAIN_CODEWORD_LEN);
170 /* Reads a Huffman-encoded symbol using the length tree. */
172 read_huffsym_using_lentree(struct input_bitstream *istream,
173 const struct lzx_tables *tables,
176 return read_huffsym(istream, tables->lentree_decode_table,
177 tables->lentree_lens, LZX_LENCODE_NUM_SYMBOLS,
178 LZX_LENCODE_TABLEBITS, n, LZX_MAX_LEN_CODEWORD_LEN);
181 /* Reads a Huffman-encoded symbol using the aligned offset tree. */
183 read_huffsym_using_alignedtree(struct input_bitstream *istream,
184 const struct lzx_tables *tables,
187 return read_huffsym(istream, tables->alignedtree_decode_table,
188 tables->alignedtree_lens,
189 LZX_ALIGNEDCODE_NUM_SYMBOLS,
190 LZX_ALIGNEDCODE_TABLEBITS, n,
191 LZX_MAX_ALIGNED_CODEWORD_LEN);
195 * Reads the pretree from the input, then uses the pretree to decode @num_lens
196 * code length values from the input.
198 * @istream: The bit stream for the input. It is positioned on the beginning
199 * of the pretree for the code length values.
200 * @lens: An array that contains the length values from the previous time
201 * the code lengths for this Huffman tree were read, or all
202 * 0's if this is the first time.
203 * @num_lens: Number of length values to decode and return.
207 lzx_read_code_lens(struct input_bitstream *istream, u8 lens[],
210 /* Declare the decoding table and length table for the pretree. */
211 u16 pretree_decode_table[(1 << LZX_PRECODE_TABLEBITS) +
212 (LZX_PRECODE_NUM_SYMBOLS * 2)]
213 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
214 u8 pretree_lens[LZX_PRECODE_NUM_SYMBOLS];
219 /* Read the code lengths of the pretree codes. There are 20 lengths of
221 for (i = 0; i < LZX_PRECODE_NUM_SYMBOLS; i++) {
222 ret = bitstream_read_bits(istream, LZX_PRECODE_ELEMENT_SIZE,
226 pretree_lens[i] = len;
229 /* Make the decoding table for the pretree. */
230 ret = make_huffman_decode_table(pretree_decode_table,
231 LZX_PRECODE_NUM_SYMBOLS,
232 LZX_PRECODE_TABLEBITS,
234 LZX_MAX_PRE_CODEWORD_LEN);
238 /* Pointer past the last length value that needs to be filled in. */
239 u8 *lens_end = lens + num_lens;
243 /* Decode a symbol from the input. If the symbol is between 0
244 * and 16, it is the difference from the old length. If it is
245 * between 17 and 19, it is a special code that indicates that
246 * some number of the next lengths are all 0, or some number of
247 * the next lengths are all equal to the next symbol in the
255 ret = read_huffsym_using_pretree(istream, pretree_decode_table,
256 pretree_lens, &tree_code);
260 case 17: /* Run of 0's */
261 ret = bitstream_read_bits(istream, 4, &num_zeroes);
265 while (num_zeroes--) {
267 if (++lens == lens_end)
271 case 18: /* Longer run of 0's */
272 ret = bitstream_read_bits(istream, 5, &num_zeroes);
276 while (num_zeroes--) {
278 if (++lens == lens_end)
282 case 19: /* Run of identical lengths */
283 ret = bitstream_read_bits(istream, 1, &num_same);
287 ret = read_huffsym_using_pretree(istream,
288 pretree_decode_table,
293 value = (signed char)*lens - (signed char)code;
298 if (++lens == lens_end)
302 default: /* Difference from old length. */
303 value = (signed char)*lens - (signed char)tree_code;
307 if (++lens == lens_end)
315 * Reads the header for an LZX-compressed block.
317 * @istream: The input bitstream.
318 * @block_size_ret: A pointer to an int into which the size of the block,
319 * in bytes, will be returned.
320 * @block_type_ret: A pointer to an int into which the type of the block
321 * (LZX_BLOCKTYPE_*) will be returned.
322 * @tables: A pointer to a lzx_tables structure in which the
323 * main tree, the length tree, and possibly the
324 * aligned offset tree will be constructed.
325 * @queue: A pointer to the least-recently-used queue into which
326 * R0, R1, and R2 will be written (only for uncompressed
327 * blocks, which contain this information in the header)
330 lzx_read_block_header(struct input_bitstream *istream,
331 unsigned num_main_syms,
332 unsigned max_window_size,
333 unsigned *block_size_ret,
334 unsigned *block_type_ret,
335 struct lzx_tables *tables,
336 struct lzx_lru_queue *queue)
342 ret = bitstream_ensure_bits(istream, 4);
346 /* The first three bits tell us what kind of block it is, and are one
347 * of the LZX_BLOCKTYPE_* values. */
348 block_type = bitstream_read_bits_nocheck(istream, 3);
350 /* Read the block size. This mirrors the behavior
351 * lzx_write_compressed_block() in lzx-compress.c; see that for more
353 if (bitstream_read_bits_nocheck(istream, 1)) {
354 block_size = LZX_DEFAULT_BLOCK_SIZE;
359 ret = bitstream_read_bits(istream, 8, &tmp);
364 ret = bitstream_read_bits(istream, 8, &tmp);
370 if (max_window_size >= 65536) {
371 ret = bitstream_read_bits(istream, 8, &tmp);
379 switch (block_type) {
380 case LZX_BLOCKTYPE_ALIGNED:
381 /* Read the path lengths for the elements of the aligned tree,
384 for (unsigned i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) {
387 ret = bitstream_read_bits(istream,
388 LZX_ALIGNEDCODE_ELEMENT_SIZE,
392 tables->alignedtree_lens[i] = len;
395 LZX_DEBUG("Building the aligned tree.");
396 ret = make_huffman_decode_table(tables->alignedtree_decode_table,
397 LZX_ALIGNEDCODE_NUM_SYMBOLS,
398 LZX_ALIGNEDCODE_TABLEBITS,
399 tables->alignedtree_lens,
400 LZX_MAX_ALIGNED_CODEWORD_LEN);
402 LZX_DEBUG("Failed to make the decode table for the "
403 "aligned offset tree");
407 /* Fall though, since the rest of the header for aligned offset
408 * blocks is the same as that for verbatim blocks */
410 case LZX_BLOCKTYPE_VERBATIM:
411 if (block_type == LZX_BLOCKTYPE_VERBATIM)
412 LZX_DEBUG("Found verbatim block.");
414 LZX_DEBUG("Reading path lengths for main tree.");
415 /* Read the path lengths for the first 256 elements of the main
417 ret = lzx_read_code_lens(istream, tables->maintree_lens,
420 LZX_DEBUG("Failed to read the code lengths for the "
421 "first 256 elements of the main tree");
425 /* Read the path lengths for the remaining elements of the main
427 LZX_DEBUG("Reading path lengths for remaining elements of "
428 "main tree (%d elements).",
429 num_main_syms - LZX_NUM_CHARS);
430 ret = lzx_read_code_lens(istream,
431 tables->maintree_lens + LZX_NUM_CHARS,
432 num_main_syms - LZX_NUM_CHARS);
434 LZX_DEBUG("Failed to read the path lengths for the "
435 "remaining elements of the main tree");
439 LZX_DEBUG("Building the Huffman decoding "
440 "table for the main tree.");
442 ret = make_huffman_decode_table(tables->maintree_decode_table,
444 LZX_MAINCODE_TABLEBITS,
445 tables->maintree_lens,
446 LZX_MAX_MAIN_CODEWORD_LEN);
448 LZX_DEBUG("Failed to make the decode "
449 "table for the main tree");
453 LZX_DEBUG("Reading path lengths for the length tree.");
454 ret = lzx_read_code_lens(istream, tables->lentree_lens,
455 LZX_LENCODE_NUM_SYMBOLS);
457 LZX_DEBUG("Failed to read the path "
458 "lengths for the length tree");
462 LZX_DEBUG("Building the length tree.");
463 ret = make_huffman_decode_table(tables->lentree_decode_table,
464 LZX_LENCODE_NUM_SYMBOLS,
465 LZX_LENCODE_TABLEBITS,
466 tables->lentree_lens,
467 LZX_MAX_LEN_CODEWORD_LEN);
469 LZX_DEBUG("Failed to build the length Huffman tree");
472 /* The bitstream of compressed literals and matches for this
473 * block directly follows and will be read in
474 * lzx_decompress_block(). */
476 case LZX_BLOCKTYPE_UNCOMPRESSED:
477 LZX_DEBUG("Found uncompressed block.");
478 /* Before reading the three LRU match offsets from the
479 * uncompressed block header, the stream needs to be aligned on
480 * a 16-bit boundary. But, unexpectedly, if the stream is
481 * *already* aligned, the correct thing to do is to throw away
482 * the next 16 bits. */
483 if (istream->bitsleft == 0) {
484 if (istream->data_bytes_left < 14) {
485 LZX_DEBUG("Insufficient length in "
486 "uncompressed block");
490 istream->data_bytes_left -= 2;
492 if (istream->data_bytes_left < 12) {
493 LZX_DEBUG("Insufficient length in "
494 "uncompressed block");
497 istream->bitsleft = 0;
500 queue->R[0] = le32_to_cpu(*(le32*)(istream->data + 0));
501 queue->R[1] = le32_to_cpu(*(le32*)(istream->data + 4));
502 queue->R[2] = le32_to_cpu(*(le32*)(istream->data + 8));
504 istream->data_bytes_left -= 12;
505 /* The uncompressed data of this block directly follows and will
506 * be read in lzx_decompress(). */
509 LZX_DEBUG("Found invalid block");
512 *block_type_ret = block_type;
513 *block_size_ret = block_size;
518 * Decodes a compressed match from a block of LZX-compressed data. A match
519 * refers to some match_offset to a point earlier in the window as well as some
520 * match_len, for which the data is to be copied to the current position in the
523 * @main_element: The start of the match data, as decoded using the main
526 * @block_type: The type of the block (LZX_BLOCKTYPE_ALIGNED or
527 * LZX_BLOCKTYPE_VERBATIM)
529 * @bytes_remaining: The amount of uncompressed data remaining to be
530 * uncompressed in this block. It is an error if the match
531 * is longer than this number.
533 * @window: A pointer to the window into which the uncompressed
534 * data is being written.
536 * @window_pos: The current byte offset in the window.
538 * @tables: The Huffman decoding tables for this LZX block (main
539 * code, length code, and for LZX_BLOCKTYPE_ALIGNED blocks,
540 * also the aligned offset code).
542 * @queue: The least-recently used queue for match offsets.
544 * @istream: The input bitstream.
546 * Returns the length of the match, or a negative number on error. The possible
548 * - Match would exceed the amount of data remaining to be uncompressed.
549 * - Match refers to data before the window.
550 * - The input bitstream ended unexpectedly.
553 lzx_decode_match(unsigned main_element, int block_type,
554 unsigned bytes_remaining, u8 *window,
556 const struct lzx_tables *tables,
557 struct lzx_lru_queue *queue,
558 struct input_bitstream *istream)
560 unsigned length_header;
561 unsigned position_slot;
563 unsigned match_offset;
564 unsigned additional_len;
565 unsigned num_extra_bits;
573 /* The main element is offset by 256 because values under 256 indicate a
575 main_element -= LZX_NUM_CHARS;
577 /* The length header consists of the lower 3 bits of the main element.
578 * The position slot is the rest of it. */
579 length_header = main_element & LZX_NUM_PRIMARY_LENS;
580 position_slot = main_element >> 3;
582 /* If the length_header is less than LZX_NUM_PRIMARY_LENS (= 7), it
583 * gives the match length as the offset from LZX_MIN_MATCH_LEN.
584 * Otherwise, the length is given by an additional symbol encoded using
585 * the length tree, offset by 9 (LZX_MIN_MATCH_LEN +
586 * LZX_NUM_PRIMARY_LENS) */
587 match_len = LZX_MIN_MATCH_LEN + length_header;
588 if (length_header == LZX_NUM_PRIMARY_LENS) {
589 ret = read_huffsym_using_lentree(istream, tables,
593 match_len += additional_len;
597 /* If the position_slot is 0, 1, or 2, the match offset is retrieved
598 * from the LRU queue. Otherwise, the match offset is not in the LRU
600 switch (position_slot) {
602 match_offset = queue->R[0];
605 match_offset = queue->R[1];
606 swap(queue->R[0], queue->R[1]);
609 /* The queue doesn't work quite the same as a real LRU queue,
610 * since using the R2 offset doesn't bump the R1 offset down to
612 match_offset = queue->R[2];
613 swap(queue->R[0], queue->R[2]);
616 /* Otherwise, the offset was not encoded as one the offsets in
617 * the queue. Depending on the position slot, there is a
618 * certain number of extra bits that need to be read to fully
619 * decode the match offset. */
621 /* Look up the number of extra bits that need to be read. */
622 num_extra_bits = lzx_get_num_extra_bits(position_slot);
624 /* For aligned blocks, if there are at least 3 extra bits, the
625 * actual number of extra bits is 3 less, and they encode a
626 * number of 8-byte words that are added to the offset; there
627 * is then an additional symbol read using the aligned tree that
628 * specifies the actual byte alignment. */
629 if (block_type == LZX_BLOCKTYPE_ALIGNED && num_extra_bits >= 3) {
631 /* There is an error in the LZX "specification" at this
632 * point; it indicates that a Huffman symbol is to be
633 * read only if num_extra_bits is greater than 3, but
634 * actually it is if num_extra_bits is greater than or
635 * equal to 3. (Note that in the case with
636 * num_extra_bits == 3, the assignment to verbatim_bits
637 * will just set it to 0. ) */
638 ret = bitstream_read_bits(istream, num_extra_bits - 3,
645 ret = read_huffsym_using_alignedtree(istream, tables,
650 /* For non-aligned blocks, or for aligned blocks with
651 * less than 3 extra bits, the extra bits are added
652 * directly to the match offset, and the correction for
653 * the alignment is taken to be 0. */
654 ret = bitstream_read_bits(istream, num_extra_bits,
662 /* Calculate the match offset. */
663 match_offset = lzx_position_base[position_slot] +
664 verbatim_bits + aligned_bits - LZX_OFFSET_OFFSET;
666 /* Update the LRU queue. */
667 queue->R[2] = queue->R[1];
668 queue->R[1] = queue->R[0];
669 queue->R[0] = match_offset;
673 /* Verify that the match is in the bounds of the part of the window
674 * currently in use, then copy the source of the match to the current
677 if (match_len > bytes_remaining) {
678 LZX_DEBUG("Match of length %u bytes overflows "
679 "uncompressed block size", match_len);
683 if (match_offset > window_pos) {
684 LZX_DEBUG("Match of length %u bytes references "
685 "data before window (match_offset = %u, "
687 match_len, match_offset, window_pos);
691 match_dest = window + window_pos;
692 match_src = match_dest - match_offset;
695 printf("Match: src %u, dst %u, len %u\n", match_src - window,
699 for (i = 0; i < match_len; i++) {
700 match_dest[i] = match_src[i];
701 putchar(match_src[i]);
706 for (i = 0; i < match_len; i++)
707 match_dest[i] = match_src[i];
714 undo_call_insn_translation(u32 *call_insn_target, s32 input_pos,
720 abs_offset = le32_to_cpu(*call_insn_target);
721 if (abs_offset >= -input_pos && abs_offset < file_size) {
722 if (abs_offset >= 0) {
723 /* "good translation" */
724 rel_offset = abs_offset - input_pos;
726 /* "compensating translation" */
727 rel_offset = abs_offset + file_size;
729 *call_insn_target = cpu_to_le32(rel_offset);
733 /* Undo the 'E8' preprocessing, where the targets of x86 CALL instructions were
734 * changed from relative offsets to absolute offsets.
736 * Note that this call instruction preprocessing can and will be used on any
737 * data even if it is not actually x86 machine code. In fact, this type of
738 * preprocessing appears to always be used in LZX-compressed resources in WIM
739 * files; there is no bit to indicate whether it is used or not, unlike in the
740 * LZX compressed format as used in cabinet files, where a bit is reserved for
743 * Call instruction preprocessing is disabled in the last 6 bytes of the
744 * uncompressed data, which really means the 5-byte call instruction cannot
745 * start in the last 10 bytes of the uncompressed data. This is one of the
746 * errors in the LZX documentation.
748 * Call instruction preprocessing does not appear to be disabled after the
749 * 32768th chunk of a WIM stream, which is apparently is yet another difference
750 * from the LZX compression used in cabinet files.
752 * Call instruction processing is supposed to take the file size as a parameter,
753 * as it is used in calculating the translated jump targets. But in WIM files,
754 * this file size is always the same (LZX_WIM_MAGIC_FILESIZE == 12000000).*/
756 undo_call_insn_preprocessing(u8 *uncompressed_data, s32 uncompressed_size)
758 for (s32 i = 0; i < uncompressed_size - 10; i++) {
759 if (uncompressed_data[i] == 0xe8) {
760 undo_call_insn_translation((u32*)&uncompressed_data[i + 1],
762 LZX_WIM_MAGIC_FILESIZE);
769 * Decompresses a LZX-compressed block of data from which the header has already
772 * @block_type: The type of the block (LZX_BLOCKTYPE_VERBATIM or
773 * LZX_BLOCKTYPE_ALIGNED)
774 * @block_size: The size of the block, in bytes.
775 * @num_main_syms: Number of symbols in the main alphabet.
776 * @window: Pointer to the decompression window.
777 * @window_pos: The current position in the window. Will be 0 for the first
779 * @tables: The Huffman decoding tables for the block (main, length, and
780 * aligned offset, the latter only for LZX_BLOCKTYPE_ALIGNED)
781 * @queue: The least-recently-used queue for match offsets.
782 * @istream: The input bitstream for the compressed literals.
785 lzx_decompress_block(int block_type, unsigned block_size,
786 unsigned num_main_syms,
789 const struct lzx_tables *tables,
790 struct lzx_lru_queue *queue,
791 struct input_bitstream *istream)
793 unsigned main_element;
798 end = window_pos + block_size;
799 while (window_pos < end) {
800 ret = read_huffsym_using_maintree(istream, tables,
806 if (main_element < LZX_NUM_CHARS) {
807 /* literal: 0 to LZX_NUM_CHARS - 1 */
808 window[window_pos++] = main_element;
810 /* match: LZX_NUM_CHARS to num_main_syms - 1 */
811 match_len = lzx_decode_match(main_element,
821 window_pos += match_len;
828 lzx_decompress(const void *compressed_data, size_t compressed_size,
829 void *uncompressed_data, size_t uncompressed_size,
832 struct lzx_decompressor *ctx = _ctx;
833 struct input_bitstream istream;
834 struct lzx_lru_queue queue;
839 bool e8_preprocessing_done;
841 LZX_DEBUG("compressed_data = %p, compressed_size = %zu, "
842 "uncompressed_data = %p, uncompressed_size = %zu, "
843 "max_window_size=%u).",
844 compressed_data, compressed_size,
845 uncompressed_data, uncompressed_size,
846 ctx->max_window_size);
848 if (uncompressed_size > ctx->max_window_size) {
849 LZX_DEBUG("Uncompressed size of %zu exceeds "
850 "window size of %u!",
851 uncompressed_size, ctx->max_window_size);
855 memset(ctx->tables.maintree_lens, 0, sizeof(ctx->tables.maintree_lens));
856 memset(ctx->tables.lentree_lens, 0, sizeof(ctx->tables.lentree_lens));
857 lzx_lru_queue_init(&queue);
858 init_input_bitstream(&istream, compressed_data, compressed_size);
860 e8_preprocessing_done = false; /* Set to true if there may be 0xe8 bytes
861 in the uncompressed data. */
863 /* The compressed data will consist of one or more blocks. The
864 * following loop decompresses one block, and it runs until there all
865 * the compressed data has been decompressed, so there are no more
869 window_pos < uncompressed_size;
870 window_pos += block_size)
872 LZX_DEBUG("Reading block header.");
873 ret = lzx_read_block_header(&istream, ctx->num_main_syms,
874 ctx->max_window_size, &block_size,
875 &block_type, &ctx->tables, &queue);
879 LZX_DEBUG("block_size = %u, window_pos = %u",
880 block_size, window_pos);
882 if (block_size > uncompressed_size - window_pos) {
883 LZX_DEBUG("Expected a block size of at "
884 "most %zu bytes (found %u bytes)",
885 uncompressed_size - window_pos, block_size);
889 switch (block_type) {
890 case LZX_BLOCKTYPE_VERBATIM:
891 case LZX_BLOCKTYPE_ALIGNED:
892 if (block_type == LZX_BLOCKTYPE_VERBATIM)
893 LZX_DEBUG("LZX_BLOCKTYPE_VERBATIM");
895 LZX_DEBUG("LZX_BLOCKTYPE_ALIGNED");
896 ret = lzx_decompress_block(block_type,
907 if (ctx->tables.maintree_lens[0xe8] != 0)
908 e8_preprocessing_done = true;
910 case LZX_BLOCKTYPE_UNCOMPRESSED:
911 LZX_DEBUG("LZX_BLOCKTYPE_UNCOMPRESSED");
912 if (istream.data_bytes_left < block_size) {
913 LZX_DEBUG("Unexpected end of input when "
914 "reading %u bytes from LZX bitstream "
915 "(only have %u bytes left)",
916 block_size, istream.data_bytes_left);
919 memcpy(&((u8*)uncompressed_data)[window_pos], istream.data,
921 istream.data += block_size;
922 istream.data_bytes_left -= block_size;
923 /* Re-align bitstream if an odd number of bytes were
925 if (istream.data_bytes_left && (block_size & 1)) {
926 istream.data_bytes_left--;
929 e8_preprocessing_done = true;
933 if (e8_preprocessing_done)
934 undo_call_insn_preprocessing(uncompressed_data, uncompressed_size);
939 lzx_free_decompressor(void *_ctx)
941 struct lzx_decompressor *ctx = _ctx;
947 lzx_create_decompressor(size_t max_window_size,
948 const struct wimlib_decompressor_params_header *params,
951 struct lzx_decompressor *ctx;
953 if (!lzx_window_size_valid(max_window_size))
954 return WIMLIB_ERR_INVALID_PARAM;
956 ctx = MALLOC(sizeof(struct lzx_decompressor));
958 return WIMLIB_ERR_NOMEM;
960 ctx->max_window_size = max_window_size;
961 ctx->num_main_syms = lzx_get_num_main_syms(max_window_size);
967 const struct decompressor_ops lzx_decompressor_ops = {
968 .create_decompressor = lzx_create_decompressor,
969 .decompress = lzx_decompress,
970 .free_decompressor = lzx_free_decompressor,