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. However, in WIM files only up to 32768 bytes of
32 * data can ever compressed be in the same LZX block, so a .tar.gz file could
33 * potentially be smaller than a WIM file that uses LZX compression because it
34 * can use a larger LZ77 window size.
36 * Some notes on the LZX compression format as used in Windows Imaging (WIM)
39 * A compressed WIM resource consists of a table of chunk offsets followed by
40 * the compressed chunks themselves. All compressed chunks except possibly the
41 * last decompress to 32768 bytes. This is quite similar to the cabinet (.cab)
42 * file format, but they are not the same. According to the cabinet format
43 * documentation, the LZX block size is independent from the CFDATA blocks, and
44 * a LZX block may span several CFDATA blocks. However, in WIMs, LZX blocks do
45 * not appear to ever span multiple WIM chunks. Note that this means any WIM
46 * chunk may be decompressed or compressed independently from any other chunk,
47 * which is convenient.
49 * A LZX compressed WIM chunk contains one or more LZX blocks of the aligned,
50 * verbatim, or uncompressed block types. For aligned and verbatim blocks, the
51 * size of the block in uncompressed bytes is specified by a bit following the 3
52 * bits that specify the block type, possibly followed by an additional 16 bits.
53 * '1' means to use the default block size (equal to 32768, the size of a WIM
54 * chunk--- and this seems to only be valid for the first LZX block in a WIM
55 * chunk), while '0' means that the block size is provided by the next 16 bits.
57 * The cabinet format, as documented, allows for the possibility that a
58 * compressed CFDATA chunk is up to 6144 bytes larger than the data it
59 * uncompresses to. However, in the WIM format it appears that every chunk that
60 * would be 32768 bytes or more when compressed is actually stored fully
63 * The 'e8' preprocessing step that changes x86 call instructions to use
64 * absolute offsets instead of relative offsets relies on a filesize parameter.
65 * There is no such parameter for this in the WIM files (even though the size of
66 * the file resource could be used for this purpose), and instead a magic file
67 * size of 12000000 is used. The 'e8' preprocessing is always done, and there
68 * is no bit to indicate whether it is done or not.
72 * Some more notes about errors in Microsoft's LZX documentation:
74 * Microsoft's LZX document and their implementation of the com.ms.util.cab Java
75 * package do not concur.
77 * In the LZX document, there is a table showing the correlation between window
78 * size and the number of position slots. It states that the 1MB window = 40
79 * slots and the 2MB window = 42 slots. In the implementation, 1MB = 42 slots,
80 * 2MB = 50 slots. The actual calculation is 'find the first slot whose position
81 * base is equal to or more than the required window size'. This would explain
82 * why other tables in the document refer to 50 slots rather than 42.
84 * The constant NUM_PRIMARY_LENS used in the decompression pseudocode is not
85 * defined in the specification.
87 * The LZX document states that aligned offset blocks have their aligned offset
88 * huffman tree AFTER the main and length trees. The implementation suggests
89 * that the aligned offset tree is BEFORE the main and length trees.
91 * The LZX document decoding algorithm states that, in an aligned offset block,
92 * if an extra_bits value is 1, 2 or 3, then that number of bits should be read
93 * and the result added to the match offset. This is correct for 1 and 2, but
94 * not 3, where just a huffman symbol (using the aligned tree) should be read.
96 * Regarding the E8 preprocessing, the LZX document states 'No translation may
97 * be performed on the last 6 bytes of the input block'. This is correct.
98 * However, the pseudocode provided checks for the *E8 leader* up to the last 6
99 * bytes. If the leader appears between -10 and -7 bytes from the end, this
100 * would cause the next four bytes to be modified, at least one of which would
101 * be in the last 6 bytes, which is not allowed according to the spec.
103 * The specification states that the huffman trees must always contain at least
104 * one element. However, many CAB files contain blocks where the length tree is
105 * completely empty (because there are no matches), and this is expected to
114 #include "wimlib/decompress.h"
115 #include "wimlib/lzx.h"
116 #include "wimlib/util.h"
120 /* Huffman decoding tables and maps from symbols to code lengths. */
123 u16 maintree_decode_table[(1 << LZX_MAINCODE_TABLEBITS) +
124 (LZX_MAINCODE_MAX_NUM_SYMBOLS * 2)]
125 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
126 u8 maintree_lens[LZX_MAINCODE_MAX_NUM_SYMBOLS];
129 u16 lentree_decode_table[(1 << LZX_LENCODE_TABLEBITS) +
130 (LZX_LENCODE_NUM_SYMBOLS * 2)]
131 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
132 u8 lentree_lens[LZX_LENCODE_NUM_SYMBOLS];
135 u16 alignedtree_decode_table[(1 << LZX_ALIGNEDCODE_TABLEBITS) +
136 (LZX_ALIGNEDCODE_NUM_SYMBOLS * 2)]
137 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
138 u8 alignedtree_lens[LZX_ALIGNEDCODE_NUM_SYMBOLS];
139 } _aligned_attribute(DECODE_TABLE_ALIGNMENT);
143 * Reads a Huffman-encoded symbol using the pre-tree.
146 read_huffsym_using_pretree(struct input_bitstream *istream,
147 const u16 pretree_decode_table[],
148 const u8 pretree_lens[], unsigned *n)
150 return read_huffsym(istream, pretree_decode_table, pretree_lens,
151 LZX_PRECODE_NUM_SYMBOLS, LZX_PRECODE_TABLEBITS, n,
152 LZX_MAX_PRE_CODEWORD_LEN);
155 /* Reads a Huffman-encoded symbol using the main tree. */
157 read_huffsym_using_maintree(struct input_bitstream *istream,
158 const struct lzx_tables *tables,
160 unsigned num_main_syms)
162 return read_huffsym(istream, tables->maintree_decode_table,
163 tables->maintree_lens, num_main_syms,
164 LZX_MAINCODE_TABLEBITS, n, LZX_MAX_MAIN_CODEWORD_LEN);
167 /* Reads a Huffman-encoded symbol using the length tree. */
169 read_huffsym_using_lentree(struct input_bitstream *istream,
170 const struct lzx_tables *tables,
173 return read_huffsym(istream, tables->lentree_decode_table,
174 tables->lentree_lens, LZX_LENCODE_NUM_SYMBOLS,
175 LZX_LENCODE_TABLEBITS, n, LZX_MAX_LEN_CODEWORD_LEN);
178 /* Reads a Huffman-encoded symbol using the aligned offset tree. */
180 read_huffsym_using_alignedtree(struct input_bitstream *istream,
181 const struct lzx_tables *tables,
184 return read_huffsym(istream, tables->alignedtree_decode_table,
185 tables->alignedtree_lens,
186 LZX_ALIGNEDCODE_NUM_SYMBOLS,
187 LZX_ALIGNEDCODE_TABLEBITS, n,
188 LZX_MAX_ALIGNED_CODEWORD_LEN);
192 * Reads the pretree from the input, then uses the pretree to decode @num_lens
193 * code length values from the input.
195 * @istream: The bit stream for the input. It is positioned on the beginning
196 * of the pretree for the code length values.
197 * @lens: An array that contains the length values from the previous time
198 * the code lengths for this Huffman tree were read, or all
199 * 0's if this is the first time.
200 * @num_lens: Number of length values to decode and return.
204 lzx_read_code_lens(struct input_bitstream *istream, u8 lens[],
207 /* Declare the decoding table and length table for the pretree. */
208 u16 pretree_decode_table[(1 << LZX_PRECODE_TABLEBITS) +
209 (LZX_PRECODE_NUM_SYMBOLS * 2)]
210 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
211 u8 pretree_lens[LZX_PRECODE_NUM_SYMBOLS];
216 /* Read the code lengths of the pretree codes. There are 20 lengths of
218 for (i = 0; i < LZX_PRECODE_NUM_SYMBOLS; i++) {
219 ret = bitstream_read_bits(istream, LZX_PRECODE_ELEMENT_SIZE,
223 pretree_lens[i] = len;
226 /* Make the decoding table for the pretree. */
227 ret = make_huffman_decode_table(pretree_decode_table,
228 LZX_PRECODE_NUM_SYMBOLS,
229 LZX_PRECODE_TABLEBITS,
231 LZX_MAX_PRE_CODEWORD_LEN);
235 /* Pointer past the last length value that needs to be filled in. */
236 u8 *lens_end = lens + num_lens;
240 /* Decode a symbol from the input. If the symbol is between 0
241 * and 16, it is the difference from the old length. If it is
242 * between 17 and 19, it is a special code that indicates that
243 * some number of the next lengths are all 0, or some number of
244 * the next lengths are all equal to the next symbol in the
252 ret = read_huffsym_using_pretree(istream, pretree_decode_table,
253 pretree_lens, &tree_code);
257 case 17: /* Run of 0's */
258 ret = bitstream_read_bits(istream, 4, &num_zeroes);
262 while (num_zeroes--) {
264 if (++lens == lens_end)
268 case 18: /* Longer run of 0's */
269 ret = bitstream_read_bits(istream, 5, &num_zeroes);
273 while (num_zeroes--) {
275 if (++lens == lens_end)
279 case 19: /* Run of identical lengths */
280 ret = bitstream_read_bits(istream, 1, &num_same);
284 ret = read_huffsym_using_pretree(istream,
285 pretree_decode_table,
290 value = (signed char)*lens - (signed char)code;
295 if (++lens == lens_end)
299 default: /* Difference from old length. */
300 value = (signed char)*lens - (signed char)tree_code;
304 if (++lens == lens_end)
312 * Reads the header for an LZX-compressed block.
314 * @istream: The input bitstream.
315 * @block_size_ret: A pointer to an int into which the size of the block,
316 * in bytes, will be returned.
317 * @block_type_ret: A pointer to an int into which the type of the block
318 * (LZX_BLOCKTYPE_*) will be returned.
319 * @tables: A pointer to a lzx_tables structure in which the
320 * main tree, the length tree, and possibly the
321 * aligned offset tree will be constructed.
322 * @queue: A pointer to the least-recently-used queue into which
323 * R0, R1, and R2 will be written (only for uncompressed
324 * blocks, which contain this information in the header)
327 lzx_read_block_header(struct input_bitstream *istream,
328 unsigned num_main_syms,
329 unsigned max_window_size,
330 unsigned *block_size_ret,
331 unsigned *block_type_ret,
332 struct lzx_tables *tables,
333 struct lzx_lru_queue *queue)
342 ret = bitstream_ensure_bits(istream, 4);
344 LZX_DEBUG("LZX input stream overrun");
348 /* The first three bits tell us what kind of block it is, and are one
349 * of the LZX_BLOCKTYPE_* values. */
350 block_type = bitstream_read_bits_nocheck(istream, 3);
352 /* Read the block size. This mirrors the behavior
353 * lzx_write_compressed_block() in lzx-compress.c; see that for more
355 s = bitstream_read_bits_nocheck(istream, 1);
357 block_size = LZX_DEFAULT_BLOCK_SIZE;
362 ret = bitstream_read_bits(istream, 8, &tmp);
367 ret = bitstream_read_bits(istream, 8, &tmp);
373 if (max_window_size >= 65536) {
374 ret = bitstream_read_bits(istream, 8, &tmp);
382 switch (block_type) {
383 case LZX_BLOCKTYPE_ALIGNED:
384 /* Read the path lengths for the elements of the aligned tree,
387 for (i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) {
388 ret = bitstream_read_bits(istream,
389 LZX_ALIGNEDCODE_ELEMENT_SIZE,
393 tables->alignedtree_lens[i] = len;
396 LZX_DEBUG("Building the aligned tree.");
397 ret = make_huffman_decode_table(tables->alignedtree_decode_table,
398 LZX_ALIGNEDCODE_NUM_SYMBOLS,
399 LZX_ALIGNEDCODE_TABLEBITS,
400 tables->alignedtree_lens,
401 LZX_MAX_ALIGNED_CODEWORD_LEN);
403 LZX_DEBUG("Failed to make the decode table for the "
404 "aligned offset tree");
408 /* Fall though, since the rest of the header for aligned offset
409 * blocks is the same as that for verbatim blocks */
411 case LZX_BLOCKTYPE_VERBATIM:
412 if (block_type == LZX_BLOCKTYPE_VERBATIM)
413 LZX_DEBUG("Found verbatim block.");
415 LZX_DEBUG("Reading path lengths for main tree.");
416 /* Read the path lengths for the first 256 elements of the main
418 ret = lzx_read_code_lens(istream, tables->maintree_lens,
421 LZX_DEBUG("Failed to read the code lengths for the "
422 "first 256 elements of the main tree");
426 /* Read the path lengths for the remaining elements of the main
428 LZX_DEBUG("Reading path lengths for remaining elements of "
429 "main tree (%d elements).",
430 num_main_syms - LZX_NUM_CHARS);
431 ret = lzx_read_code_lens(istream,
432 tables->maintree_lens + LZX_NUM_CHARS,
433 num_main_syms - LZX_NUM_CHARS);
435 LZX_DEBUG("Failed to read the path lengths for the "
436 "remaining elements of the main tree");
440 LZX_DEBUG("Building the Huffman decoding "
441 "table for the main tree.");
443 ret = make_huffman_decode_table(tables->maintree_decode_table,
445 LZX_MAINCODE_TABLEBITS,
446 tables->maintree_lens,
447 LZX_MAX_MAIN_CODEWORD_LEN);
449 LZX_DEBUG("Failed to make the decode "
450 "table for the main tree");
454 LZX_DEBUG("Reading path lengths for the length tree.");
455 ret = lzx_read_code_lens(istream, tables->lentree_lens,
456 LZX_LENCODE_NUM_SYMBOLS);
458 LZX_DEBUG("Failed to read the path "
459 "lengths for the length tree");
463 LZX_DEBUG("Building the length tree.");
464 ret = make_huffman_decode_table(tables->lentree_decode_table,
465 LZX_LENCODE_NUM_SYMBOLS,
466 LZX_LENCODE_TABLEBITS,
467 tables->lentree_lens,
468 LZX_MAX_LEN_CODEWORD_LEN);
470 LZX_DEBUG("Failed to build the length Huffman tree");
473 /* The bitstream of compressed literals and matches for this
474 * block directly follows and will be read in
475 * lzx_decompress_block(). */
477 case LZX_BLOCKTYPE_UNCOMPRESSED:
478 LZX_DEBUG("Found uncompressed block.");
479 /* Before reading the three LRU match offsets from the
480 * uncompressed block header, the stream needs to be aligned on
481 * a 16-bit boundary. But, unexpectedly, if the stream is
482 * *already* aligned, the correct thing to do is to throw away
483 * the next 16 bits. */
484 if (istream->bitsleft == 0) {
485 if (istream->data_bytes_left < 14) {
486 LZX_DEBUG("Insufficient length in "
487 "uncompressed block");
491 istream->data_bytes_left -= 2;
493 if (istream->data_bytes_left < 12) {
494 LZX_DEBUG("Insufficient length in "
495 "uncompressed block");
498 istream->bitsleft = 0;
501 queue->R[0] = le32_to_cpu(*(u32*)(istream->data + 0));
502 queue->R[1] = le32_to_cpu(*(u32*)(istream->data + 4));
503 queue->R[2] = le32_to_cpu(*(u32*)(istream->data + 8));
505 istream->data_bytes_left -= 12;
506 /* The uncompressed data of this block directly follows and will
507 * be read in lzx_decompress(). */
510 LZX_DEBUG("Found invalid block");
513 *block_type_ret = block_type;
514 *block_size_ret = block_size;
519 * Decodes a compressed match from a block of LZX-compressed data. A match
520 * refers to some match_offset to a point earlier in the window as well as some
521 * match_len, for which the data is to be copied to the current position in the
524 * @main_element: The start of the match data, as decoded using the main
527 * @block_type: The type of the block (LZX_BLOCKTYPE_ALIGNED or
528 * LZX_BLOCKTYPE_VERBATIM)
530 * @bytes_remaining: The amount of uncompressed data remaining to be
531 * uncompressed in this block. It is an error if the match
532 * is longer than this number.
534 * @window: A pointer to the window into which the uncompressed
535 * data is being written.
537 * @window_pos: The current byte offset in the window.
539 * @tables: The Huffman decoding tables for this LZX block (main
540 * code, length code, and for LZX_BLOCKTYPE_ALIGNED blocks,
541 * also the aligned offset code).
543 * @queue: The least-recently used queue for match offsets.
545 * @istream: The input bitstream.
547 * Returns the length of the match, or a negative number on error. The possible
549 * - Match would exceed the amount of data remaining to be uncompressed.
550 * - Match refers to data before the window.
551 * - The input bitstream ended unexpectedly.
554 lzx_decode_match(unsigned main_element, int block_type,
555 unsigned bytes_remaining, u8 *window,
557 const struct lzx_tables *tables,
558 struct lzx_lru_queue *queue,
559 struct input_bitstream *istream)
561 unsigned length_header;
562 unsigned position_slot;
564 unsigned match_offset;
565 unsigned additional_len;
566 unsigned num_extra_bits;
567 unsigned verbatim_bits;
568 unsigned aligned_bits;
574 /* The main element is offset by 256 because values under 256 indicate a
576 main_element -= LZX_NUM_CHARS;
578 /* The length header consists of the lower 3 bits of the main element.
579 * The position slot is the rest of it. */
580 length_header = main_element & LZX_NUM_PRIMARY_LENS;
581 position_slot = main_element >> 3;
583 /* If the length_header is less than LZX_NUM_PRIMARY_LENS (= 7), it
584 * gives the match length as the offset from LZX_MIN_MATCH_LEN.
585 * Otherwise, the length is given by an additional symbol encoded using
586 * the length tree, offset by 9 (LZX_MIN_MATCH_LEN +
587 * LZX_NUM_PRIMARY_LENS) */
588 match_len = LZX_MIN_MATCH_LEN + length_header;
589 if (length_header == LZX_NUM_PRIMARY_LENS) {
590 ret = read_huffsym_using_lentree(istream, tables,
594 match_len += additional_len;
598 /* If the position_slot is 0, 1, or 2, the match offset is retrieved
599 * from the LRU queue. Otherwise, the match offset is not in the LRU
601 switch (position_slot) {
603 match_offset = queue->R[0];
606 match_offset = queue->R[1];
607 swap(queue->R[0], queue->R[1]);
610 /* The queue doesn't work quite the same as a real LRU queue,
611 * since using the R2 offset doesn't bump the R1 offset down to
613 match_offset = queue->R[2];
614 swap(queue->R[0], queue->R[2]);
617 /* Otherwise, the offset was not encoded as one the offsets in
618 * the queue. Depending on the position slot, there is a
619 * certain number of extra bits that need to be read to fully
620 * decode the match offset. */
622 /* Look up the number of extra bits that need to be read. */
623 num_extra_bits = lzx_get_num_extra_bits(position_slot);
625 /* For aligned blocks, if there are at least 3 extra bits, the
626 * actual number of extra bits is 3 less, and they encode a
627 * number of 8-byte words that are added to the offset; there
628 * is then an additional symbol read using the aligned tree that
629 * specifies the actual byte alignment. */
630 if (block_type == LZX_BLOCKTYPE_ALIGNED && num_extra_bits >= 3) {
632 /* There is an error in the LZX "specification" at this
633 * point; it indicates that a Huffman symbol is to be
634 * read only if num_extra_bits is greater than 3, but
635 * actually it is if num_extra_bits is greater than or
636 * equal to 3. (Note that in the case with
637 * num_extra_bits == 3, the assignment to verbatim_bits
638 * will just set it to 0. ) */
639 ret = bitstream_read_bits(istream, num_extra_bits - 3,
646 ret = read_huffsym_using_alignedtree(istream, tables,
651 /* For non-aligned blocks, or for aligned blocks with
652 * less than 3 extra bits, the extra bits are added
653 * directly to the match offset, and the correction for
654 * the alignment is taken to be 0. */
655 ret = bitstream_read_bits(istream, num_extra_bits,
663 /* Calculate the match offset. */
664 match_offset = lzx_position_base[position_slot] +
665 verbatim_bits + aligned_bits - LZX_OFFSET_OFFSET;
667 /* Update the LRU queue. */
668 queue->R[2] = queue->R[1];
669 queue->R[1] = queue->R[0];
670 queue->R[0] = match_offset;
674 /* Verify that the match is in the bounds of the part of the window
675 * currently in use, then copy the source of the match to the current
678 if (match_len > bytes_remaining) {
679 LZX_DEBUG("Match of length %u bytes overflows "
680 "uncompressed block size", match_len);
684 if (match_offset > window_pos) {
685 LZX_DEBUG("Match of length %u bytes references "
686 "data before window (match_offset = %u, "
688 match_len, match_offset, window_pos);
692 match_dest = window + window_pos;
693 match_src = match_dest - match_offset;
696 printf("Match: src %u, dst %u, len %u\n", match_src - window,
700 for (i = 0; i < match_len; i++) {
701 match_dest[i] = match_src[i];
702 putchar(match_src[i]);
707 for (i = 0; i < match_len; i++)
708 match_dest[i] = match_src[i];
715 undo_call_insn_translation(u32 *call_insn_target, int input_pos,
721 abs_offset = le32_to_cpu(*call_insn_target);
722 if (abs_offset >= -input_pos && abs_offset < file_size) {
723 if (abs_offset >= 0) {
724 /* "good translation" */
725 rel_offset = abs_offset - input_pos;
727 /* "compensating translation" */
728 rel_offset = abs_offset + file_size;
730 *call_insn_target = cpu_to_le32(rel_offset);
734 /* Undo the 'E8' preprocessing, where the targets of x86 CALL instructions were
735 * changed from relative offsets to absolute offsets.
737 * Note that this call instruction preprocessing can and will be used on any
738 * data even if it is not actually x86 machine code. In fact, this type of
739 * preprocessing appears to always be used in LZX-compressed resources in WIM
740 * files; there is no bit to indicate whether it is used or not, unlike in the
741 * LZX compressed format as used in cabinet files, where a bit is reserved for
744 * Call instruction preprocessing is disabled in the last 6 bytes of the
745 * uncompressed data, which really means the 5-byte call instruction cannot
746 * start in the last 10 bytes of the uncompressed data. This is one of the
747 * errors in the LZX documentation.
749 * Call instruction preprocessing does not appear to be disabled after the
750 * 32768th chunk of a WIM stream, which is apparently is yet another difference
751 * from the LZX compression used in cabinet files.
753 * Call instruction processing is supposed to take the file size as a parameter,
754 * as it is used in calculating the translated jump targets. But in WIM files,
755 * this file size is always the same (LZX_WIM_MAGIC_FILESIZE == 12000000).*/
757 undo_call_insn_preprocessing(u8 uncompressed_data[], int uncompressed_data_len)
759 for (int i = 0; i < uncompressed_data_len - 10; i++) {
760 if (uncompressed_data[i] == 0xe8) {
761 undo_call_insn_translation((u32*)&uncompressed_data[i + 1],
763 LZX_WIM_MAGIC_FILESIZE);
770 * Decompresses a LZX-compressed block of data from which the header has already
773 * @block_type: The type of the block (LZX_BLOCKTYPE_VERBATIM or
774 * LZX_BLOCKTYPE_ALIGNED)
775 * @block_size: The size of the block, in bytes.
776 * @num_main_syms: Number of symbols in the main alphabet.
777 * @window: Pointer to the decompression window.
778 * @window_pos: The current position in the window. Will be 0 for the first
780 * @tables: The Huffman decoding tables for the block (main, length, and
781 * aligned offset, the latter only for LZX_BLOCKTYPE_ALIGNED)
782 * @queue: The least-recently-used queue for match offsets.
783 * @istream: The input bitstream for the compressed literals.
786 lzx_decompress_block(int block_type, unsigned block_size,
787 unsigned num_main_syms,
790 const struct lzx_tables *tables,
791 struct lzx_lru_queue *queue,
792 struct input_bitstream *istream)
794 unsigned main_element;
799 end = window_pos + block_size;
800 while (window_pos < end) {
801 ret = read_huffsym_using_maintree(istream, tables,
807 if (main_element < LZX_NUM_CHARS) {
808 /* literal: 0 to LZX_NUM_CHARS - 1 */
809 window[window_pos++] = main_element;
811 /* match: LZX_NUM_CHARS to num_main_syms - 1 */
812 match_len = lzx_decode_match(main_element,
822 window_pos += match_len;
829 wimlib_lzx_decompress2(const void *compressed_data, unsigned compressed_len,
830 void *uncompressed_data, unsigned uncompressed_len,
833 struct lzx_tables tables;
834 struct input_bitstream istream;
835 struct lzx_lru_queue queue;
839 unsigned num_main_syms;
841 bool e8_preprocessing_done;
843 LZX_DEBUG("compressed_data = %p, compressed_len = %u, "
844 "uncompressed_data = %p, uncompressed_len = %u, "
845 "max_window_size=%u).",
846 compressed_data, compressed_len,
847 uncompressed_data, uncompressed_len, max_window_size);
849 if (!lzx_window_size_valid(max_window_size)) {
850 LZX_DEBUG("Window size of %u is invalid!",
855 num_main_syms = lzx_get_num_main_syms(max_window_size);
857 if (uncompressed_len > max_window_size) {
858 LZX_DEBUG("Uncompressed chunk size of %u exceeds "
859 "window size of %u!",
860 uncompressed_len, max_window_size);
864 memset(tables.maintree_lens, 0, sizeof(tables.maintree_lens));
865 memset(tables.lentree_lens, 0, sizeof(tables.lentree_lens));
866 lzx_lru_queue_init(&queue);
867 init_input_bitstream(&istream, compressed_data, compressed_len);
869 e8_preprocessing_done = false; /* Set to true if there may be 0xe8 bytes
870 in the uncompressed data. */
872 /* The compressed data will consist of one or more blocks. The
873 * following loop decompresses one block, and it runs until there all
874 * the compressed data has been decompressed, so there are no more
878 window_pos < uncompressed_len;
879 window_pos += block_size)
881 LZX_DEBUG("Reading block header.");
882 ret = lzx_read_block_header(&istream, num_main_syms,
883 max_window_size, &block_size,
884 &block_type, &tables, &queue);
888 LZX_DEBUG("block_size = %u, window_pos = %u",
889 block_size, window_pos);
891 if (block_size > uncompressed_len - window_pos) {
892 LZX_DEBUG("Expected a block size of at "
893 "most %u bytes (found %u bytes)",
894 uncompressed_len - window_pos, block_size);
898 switch (block_type) {
899 case LZX_BLOCKTYPE_VERBATIM:
900 case LZX_BLOCKTYPE_ALIGNED:
901 if (block_type == LZX_BLOCKTYPE_VERBATIM)
902 LZX_DEBUG("LZX_BLOCKTYPE_VERBATIM");
904 LZX_DEBUG("LZX_BLOCKTYPE_ALIGNED");
905 ret = lzx_decompress_block(block_type,
916 if (tables.maintree_lens[0xe8] != 0)
917 e8_preprocessing_done = true;
919 case LZX_BLOCKTYPE_UNCOMPRESSED:
920 LZX_DEBUG("LZX_BLOCKTYPE_UNCOMPRESSED");
921 if (istream.data_bytes_left < block_size) {
922 LZX_DEBUG("Unexpected end of input when "
923 "reading %u bytes from LZX bitstream "
924 "(only have %u bytes left)",
925 block_size, istream.data_bytes_left);
928 memcpy(&((u8*)uncompressed_data)[window_pos], istream.data,
930 istream.data += block_size;
931 istream.data_bytes_left -= block_size;
932 /* Re-align bitstream if an odd number of bytes were
934 if (istream.data_bytes_left && (block_size & 1)) {
935 istream.data_bytes_left--;
938 e8_preprocessing_done = true;
942 if (e8_preprocessing_done)
943 undo_call_insn_preprocessing(uncompressed_data, uncompressed_len);
947 /* API function documented in wimlib.h */
949 wimlib_lzx_decompress(const void *compressed_data, unsigned compressed_len,
950 void *uncompressed_data, unsigned uncompressed_len)
952 return wimlib_lzx_decompress2(compressed_data, compressed_len,
953 uncompressed_data, uncompressed_len,