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/decompress.h"
112 #include "wimlib/lzx.h"
113 #include "wimlib/util.h"
117 /* Huffman decoding tables and maps from symbols to code lengths. */
120 u16 maintree_decode_table[(1 << LZX_MAINCODE_TABLEBITS) +
121 (LZX_MAINCODE_MAX_NUM_SYMBOLS * 2)]
122 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
123 u8 maintree_lens[LZX_MAINCODE_MAX_NUM_SYMBOLS];
126 u16 lentree_decode_table[(1 << LZX_LENCODE_TABLEBITS) +
127 (LZX_LENCODE_NUM_SYMBOLS * 2)]
128 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
129 u8 lentree_lens[LZX_LENCODE_NUM_SYMBOLS];
132 u16 alignedtree_decode_table[(1 << LZX_ALIGNEDCODE_TABLEBITS) +
133 (LZX_ALIGNEDCODE_NUM_SYMBOLS * 2)]
134 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
135 u8 alignedtree_lens[LZX_ALIGNEDCODE_NUM_SYMBOLS];
136 } _aligned_attribute(DECODE_TABLE_ALIGNMENT);
140 * Reads a Huffman-encoded symbol using the pre-tree.
143 read_huffsym_using_pretree(struct input_bitstream *istream,
144 const u16 pretree_decode_table[],
145 const u8 pretree_lens[], unsigned *n)
147 return read_huffsym(istream, pretree_decode_table, pretree_lens,
148 LZX_PRECODE_NUM_SYMBOLS, LZX_PRECODE_TABLEBITS, n,
149 LZX_MAX_PRE_CODEWORD_LEN);
152 /* Reads a Huffman-encoded symbol using the main tree. */
154 read_huffsym_using_maintree(struct input_bitstream *istream,
155 const struct lzx_tables *tables,
157 unsigned num_main_syms)
159 return read_huffsym(istream, tables->maintree_decode_table,
160 tables->maintree_lens, num_main_syms,
161 LZX_MAINCODE_TABLEBITS, n, LZX_MAX_MAIN_CODEWORD_LEN);
164 /* Reads a Huffman-encoded symbol using the length tree. */
166 read_huffsym_using_lentree(struct input_bitstream *istream,
167 const struct lzx_tables *tables,
170 return read_huffsym(istream, tables->lentree_decode_table,
171 tables->lentree_lens, LZX_LENCODE_NUM_SYMBOLS,
172 LZX_LENCODE_TABLEBITS, n, LZX_MAX_LEN_CODEWORD_LEN);
175 /* Reads a Huffman-encoded symbol using the aligned offset tree. */
177 read_huffsym_using_alignedtree(struct input_bitstream *istream,
178 const struct lzx_tables *tables,
181 return read_huffsym(istream, tables->alignedtree_decode_table,
182 tables->alignedtree_lens,
183 LZX_ALIGNEDCODE_NUM_SYMBOLS,
184 LZX_ALIGNEDCODE_TABLEBITS, n,
185 LZX_MAX_ALIGNED_CODEWORD_LEN);
189 * Reads the pretree from the input, then uses the pretree to decode @num_lens
190 * code length values from the input.
192 * @istream: The bit stream for the input. It is positioned on the beginning
193 * of the pretree for the code length values.
194 * @lens: An array that contains the length values from the previous time
195 * the code lengths for this Huffman tree were read, or all
196 * 0's if this is the first time.
197 * @num_lens: Number of length values to decode and return.
201 lzx_read_code_lens(struct input_bitstream *istream, u8 lens[],
204 /* Declare the decoding table and length table for the pretree. */
205 u16 pretree_decode_table[(1 << LZX_PRECODE_TABLEBITS) +
206 (LZX_PRECODE_NUM_SYMBOLS * 2)]
207 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
208 u8 pretree_lens[LZX_PRECODE_NUM_SYMBOLS];
213 /* Read the code lengths of the pretree codes. There are 20 lengths of
215 for (i = 0; i < LZX_PRECODE_NUM_SYMBOLS; i++) {
216 ret = bitstream_read_bits(istream, LZX_PRECODE_ELEMENT_SIZE,
220 pretree_lens[i] = len;
223 /* Make the decoding table for the pretree. */
224 ret = make_huffman_decode_table(pretree_decode_table,
225 LZX_PRECODE_NUM_SYMBOLS,
226 LZX_PRECODE_TABLEBITS,
228 LZX_MAX_PRE_CODEWORD_LEN);
232 /* Pointer past the last length value that needs to be filled in. */
233 u8 *lens_end = lens + num_lens;
237 /* Decode a symbol from the input. If the symbol is between 0
238 * and 16, it is the difference from the old length. If it is
239 * between 17 and 19, it is a special code that indicates that
240 * some number of the next lengths are all 0, or some number of
241 * the next lengths are all equal to the next symbol in the
249 ret = read_huffsym_using_pretree(istream, pretree_decode_table,
250 pretree_lens, &tree_code);
254 case 17: /* Run of 0's */
255 ret = bitstream_read_bits(istream, 4, &num_zeroes);
259 while (num_zeroes--) {
261 if (++lens == lens_end)
265 case 18: /* Longer run of 0's */
266 ret = bitstream_read_bits(istream, 5, &num_zeroes);
270 while (num_zeroes--) {
272 if (++lens == lens_end)
276 case 19: /* Run of identical lengths */
277 ret = bitstream_read_bits(istream, 1, &num_same);
281 ret = read_huffsym_using_pretree(istream,
282 pretree_decode_table,
287 value = (signed char)*lens - (signed char)code;
292 if (++lens == lens_end)
296 default: /* Difference from old length. */
297 value = (signed char)*lens - (signed char)tree_code;
301 if (++lens == lens_end)
309 * Reads the header for an LZX-compressed block.
311 * @istream: The input bitstream.
312 * @block_size_ret: A pointer to an int into which the size of the block,
313 * in bytes, will be returned.
314 * @block_type_ret: A pointer to an int into which the type of the block
315 * (LZX_BLOCKTYPE_*) will be returned.
316 * @tables: A pointer to a lzx_tables structure in which the
317 * main tree, the length tree, and possibly the
318 * aligned offset tree will be constructed.
319 * @queue: A pointer to the least-recently-used queue into which
320 * R0, R1, and R2 will be written (only for uncompressed
321 * blocks, which contain this information in the header)
324 lzx_read_block_header(struct input_bitstream *istream,
325 unsigned num_main_syms,
326 unsigned max_window_size,
327 unsigned *block_size_ret,
328 unsigned *block_type_ret,
329 struct lzx_tables *tables,
330 struct lzx_lru_queue *queue)
336 ret = bitstream_ensure_bits(istream, 4);
340 /* The first three bits tell us what kind of block it is, and are one
341 * of the LZX_BLOCKTYPE_* values. */
342 block_type = bitstream_read_bits_nocheck(istream, 3);
344 /* Read the block size. This mirrors the behavior
345 * lzx_write_compressed_block() in lzx-compress.c; see that for more
347 if (bitstream_read_bits_nocheck(istream, 1)) {
348 block_size = LZX_DEFAULT_BLOCK_SIZE;
353 ret = bitstream_read_bits(istream, 8, &tmp);
358 ret = bitstream_read_bits(istream, 8, &tmp);
364 if (max_window_size >= 65536) {
365 ret = bitstream_read_bits(istream, 8, &tmp);
373 switch (block_type) {
374 case LZX_BLOCKTYPE_ALIGNED:
375 /* Read the path lengths for the elements of the aligned tree,
378 for (unsigned i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) {
381 ret = bitstream_read_bits(istream,
382 LZX_ALIGNEDCODE_ELEMENT_SIZE,
386 tables->alignedtree_lens[i] = len;
389 LZX_DEBUG("Building the aligned tree.");
390 ret = make_huffman_decode_table(tables->alignedtree_decode_table,
391 LZX_ALIGNEDCODE_NUM_SYMBOLS,
392 LZX_ALIGNEDCODE_TABLEBITS,
393 tables->alignedtree_lens,
394 LZX_MAX_ALIGNED_CODEWORD_LEN);
396 LZX_DEBUG("Failed to make the decode table for the "
397 "aligned offset tree");
401 /* Fall though, since the rest of the header for aligned offset
402 * blocks is the same as that for verbatim blocks */
404 case LZX_BLOCKTYPE_VERBATIM:
405 if (block_type == LZX_BLOCKTYPE_VERBATIM)
406 LZX_DEBUG("Found verbatim block.");
408 LZX_DEBUG("Reading path lengths for main tree.");
409 /* Read the path lengths for the first 256 elements of the main
411 ret = lzx_read_code_lens(istream, tables->maintree_lens,
414 LZX_DEBUG("Failed to read the code lengths for the "
415 "first 256 elements of the main tree");
419 /* Read the path lengths for the remaining elements of the main
421 LZX_DEBUG("Reading path lengths for remaining elements of "
422 "main tree (%d elements).",
423 num_main_syms - LZX_NUM_CHARS);
424 ret = lzx_read_code_lens(istream,
425 tables->maintree_lens + LZX_NUM_CHARS,
426 num_main_syms - LZX_NUM_CHARS);
428 LZX_DEBUG("Failed to read the path lengths for the "
429 "remaining elements of the main tree");
433 LZX_DEBUG("Building the Huffman decoding "
434 "table for the main tree.");
436 ret = make_huffman_decode_table(tables->maintree_decode_table,
438 LZX_MAINCODE_TABLEBITS,
439 tables->maintree_lens,
440 LZX_MAX_MAIN_CODEWORD_LEN);
442 LZX_DEBUG("Failed to make the decode "
443 "table for the main tree");
447 LZX_DEBUG("Reading path lengths for the length tree.");
448 ret = lzx_read_code_lens(istream, tables->lentree_lens,
449 LZX_LENCODE_NUM_SYMBOLS);
451 LZX_DEBUG("Failed to read the path "
452 "lengths for the length tree");
456 LZX_DEBUG("Building the length tree.");
457 ret = make_huffman_decode_table(tables->lentree_decode_table,
458 LZX_LENCODE_NUM_SYMBOLS,
459 LZX_LENCODE_TABLEBITS,
460 tables->lentree_lens,
461 LZX_MAX_LEN_CODEWORD_LEN);
463 LZX_DEBUG("Failed to build the length Huffman tree");
466 /* The bitstream of compressed literals and matches for this
467 * block directly follows and will be read in
468 * lzx_decompress_block(). */
470 case LZX_BLOCKTYPE_UNCOMPRESSED:
471 LZX_DEBUG("Found uncompressed block.");
472 /* Before reading the three LRU match offsets from the
473 * uncompressed block header, the stream needs to be aligned on
474 * a 16-bit boundary. But, unexpectedly, if the stream is
475 * *already* aligned, the correct thing to do is to throw away
476 * the next 16 bits. */
477 if (istream->bitsleft == 0) {
478 if (istream->data_bytes_left < 14) {
479 LZX_DEBUG("Insufficient length in "
480 "uncompressed block");
484 istream->data_bytes_left -= 2;
486 if (istream->data_bytes_left < 12) {
487 LZX_DEBUG("Insufficient length in "
488 "uncompressed block");
491 istream->bitsleft = 0;
494 queue->R[0] = le32_to_cpu(*(le32*)(istream->data + 0));
495 queue->R[1] = le32_to_cpu(*(le32*)(istream->data + 4));
496 queue->R[2] = le32_to_cpu(*(le32*)(istream->data + 8));
498 istream->data_bytes_left -= 12;
499 /* The uncompressed data of this block directly follows and will
500 * be read in lzx_decompress(). */
503 LZX_DEBUG("Found invalid block");
506 *block_type_ret = block_type;
507 *block_size_ret = block_size;
512 * Decodes a compressed match from a block of LZX-compressed data. A match
513 * refers to some match_offset to a point earlier in the window as well as some
514 * match_len, for which the data is to be copied to the current position in the
517 * @main_element: The start of the match data, as decoded using the main
520 * @block_type: The type of the block (LZX_BLOCKTYPE_ALIGNED or
521 * LZX_BLOCKTYPE_VERBATIM)
523 * @bytes_remaining: The amount of uncompressed data remaining to be
524 * uncompressed in this block. It is an error if the match
525 * is longer than this number.
527 * @window: A pointer to the window into which the uncompressed
528 * data is being written.
530 * @window_pos: The current byte offset in the window.
532 * @tables: The Huffman decoding tables for this LZX block (main
533 * code, length code, and for LZX_BLOCKTYPE_ALIGNED blocks,
534 * also the aligned offset code).
536 * @queue: The least-recently used queue for match offsets.
538 * @istream: The input bitstream.
540 * Returns the length of the match, or a negative number on error. The possible
542 * - Match would exceed the amount of data remaining to be uncompressed.
543 * - Match refers to data before the window.
544 * - The input bitstream ended unexpectedly.
547 lzx_decode_match(unsigned main_element, int block_type,
548 unsigned bytes_remaining, u8 *window,
550 const struct lzx_tables *tables,
551 struct lzx_lru_queue *queue,
552 struct input_bitstream *istream)
554 unsigned length_header;
555 unsigned position_slot;
557 unsigned match_offset;
558 unsigned additional_len;
559 unsigned num_extra_bits;
567 /* The main element is offset by 256 because values under 256 indicate a
569 main_element -= LZX_NUM_CHARS;
571 /* The length header consists of the lower 3 bits of the main element.
572 * The position slot is the rest of it. */
573 length_header = main_element & LZX_NUM_PRIMARY_LENS;
574 position_slot = main_element >> 3;
576 /* If the length_header is less than LZX_NUM_PRIMARY_LENS (= 7), it
577 * gives the match length as the offset from LZX_MIN_MATCH_LEN.
578 * Otherwise, the length is given by an additional symbol encoded using
579 * the length tree, offset by 9 (LZX_MIN_MATCH_LEN +
580 * LZX_NUM_PRIMARY_LENS) */
581 match_len = LZX_MIN_MATCH_LEN + length_header;
582 if (length_header == LZX_NUM_PRIMARY_LENS) {
583 ret = read_huffsym_using_lentree(istream, tables,
587 match_len += additional_len;
591 /* If the position_slot is 0, 1, or 2, the match offset is retrieved
592 * from the LRU queue. Otherwise, the match offset is not in the LRU
594 switch (position_slot) {
596 match_offset = queue->R[0];
599 match_offset = queue->R[1];
600 swap(queue->R[0], queue->R[1]);
603 /* The queue doesn't work quite the same as a real LRU queue,
604 * since using the R2 offset doesn't bump the R1 offset down to
606 match_offset = queue->R[2];
607 swap(queue->R[0], queue->R[2]);
610 /* Otherwise, the offset was not encoded as one the offsets in
611 * the queue. Depending on the position slot, there is a
612 * certain number of extra bits that need to be read to fully
613 * decode the match offset. */
615 /* Look up the number of extra bits that need to be read. */
616 num_extra_bits = lzx_get_num_extra_bits(position_slot);
618 /* For aligned blocks, if there are at least 3 extra bits, the
619 * actual number of extra bits is 3 less, and they encode a
620 * number of 8-byte words that are added to the offset; there
621 * is then an additional symbol read using the aligned tree that
622 * specifies the actual byte alignment. */
623 if (block_type == LZX_BLOCKTYPE_ALIGNED && num_extra_bits >= 3) {
625 /* There is an error in the LZX "specification" at this
626 * point; it indicates that a Huffman symbol is to be
627 * read only if num_extra_bits is greater than 3, but
628 * actually it is if num_extra_bits is greater than or
629 * equal to 3. (Note that in the case with
630 * num_extra_bits == 3, the assignment to verbatim_bits
631 * will just set it to 0. ) */
632 ret = bitstream_read_bits(istream, num_extra_bits - 3,
639 ret = read_huffsym_using_alignedtree(istream, tables,
644 /* For non-aligned blocks, or for aligned blocks with
645 * less than 3 extra bits, the extra bits are added
646 * directly to the match offset, and the correction for
647 * the alignment is taken to be 0. */
648 ret = bitstream_read_bits(istream, num_extra_bits,
656 /* Calculate the match offset. */
657 match_offset = lzx_position_base[position_slot] +
658 verbatim_bits + aligned_bits - LZX_OFFSET_OFFSET;
660 /* Update the LRU queue. */
661 queue->R[2] = queue->R[1];
662 queue->R[1] = queue->R[0];
663 queue->R[0] = match_offset;
667 /* Verify that the match is in the bounds of the part of the window
668 * currently in use, then copy the source of the match to the current
671 if (match_len > bytes_remaining) {
672 LZX_DEBUG("Match of length %u bytes overflows "
673 "uncompressed block size", match_len);
677 if (match_offset > window_pos) {
678 LZX_DEBUG("Match of length %u bytes references "
679 "data before window (match_offset = %u, "
681 match_len, match_offset, window_pos);
685 match_dest = window + window_pos;
686 match_src = match_dest - match_offset;
689 printf("Match: src %u, dst %u, len %u\n", match_src - window,
693 for (i = 0; i < match_len; i++) {
694 match_dest[i] = match_src[i];
695 putchar(match_src[i]);
700 for (i = 0; i < match_len; i++)
701 match_dest[i] = match_src[i];
708 undo_call_insn_translation(u32 *call_insn_target, int input_pos,
714 abs_offset = le32_to_cpu(*call_insn_target);
715 if (abs_offset >= -input_pos && abs_offset < file_size) {
716 if (abs_offset >= 0) {
717 /* "good translation" */
718 rel_offset = abs_offset - input_pos;
720 /* "compensating translation" */
721 rel_offset = abs_offset + file_size;
723 *call_insn_target = cpu_to_le32(rel_offset);
727 /* Undo the 'E8' preprocessing, where the targets of x86 CALL instructions were
728 * changed from relative offsets to absolute offsets.
730 * Note that this call instruction preprocessing can and will be used on any
731 * data even if it is not actually x86 machine code. In fact, this type of
732 * preprocessing appears to always be used in LZX-compressed resources in WIM
733 * files; there is no bit to indicate whether it is used or not, unlike in the
734 * LZX compressed format as used in cabinet files, where a bit is reserved for
737 * Call instruction preprocessing is disabled in the last 6 bytes of the
738 * uncompressed data, which really means the 5-byte call instruction cannot
739 * start in the last 10 bytes of the uncompressed data. This is one of the
740 * errors in the LZX documentation.
742 * Call instruction preprocessing does not appear to be disabled after the
743 * 32768th chunk of a WIM stream, which is apparently is yet another difference
744 * from the LZX compression used in cabinet files.
746 * Call instruction processing is supposed to take the file size as a parameter,
747 * as it is used in calculating the translated jump targets. But in WIM files,
748 * this file size is always the same (LZX_WIM_MAGIC_FILESIZE == 12000000).*/
750 undo_call_insn_preprocessing(u8 uncompressed_data[], int uncompressed_data_len)
752 for (int i = 0; i < uncompressed_data_len - 10; i++) {
753 if (uncompressed_data[i] == 0xe8) {
754 undo_call_insn_translation((u32*)&uncompressed_data[i + 1],
756 LZX_WIM_MAGIC_FILESIZE);
763 * Decompresses a LZX-compressed block of data from which the header has already
766 * @block_type: The type of the block (LZX_BLOCKTYPE_VERBATIM or
767 * LZX_BLOCKTYPE_ALIGNED)
768 * @block_size: The size of the block, in bytes.
769 * @num_main_syms: Number of symbols in the main alphabet.
770 * @window: Pointer to the decompression window.
771 * @window_pos: The current position in the window. Will be 0 for the first
773 * @tables: The Huffman decoding tables for the block (main, length, and
774 * aligned offset, the latter only for LZX_BLOCKTYPE_ALIGNED)
775 * @queue: The least-recently-used queue for match offsets.
776 * @istream: The input bitstream for the compressed literals.
779 lzx_decompress_block(int block_type, unsigned block_size,
780 unsigned num_main_syms,
783 const struct lzx_tables *tables,
784 struct lzx_lru_queue *queue,
785 struct input_bitstream *istream)
787 unsigned main_element;
792 end = window_pos + block_size;
793 while (window_pos < end) {
794 ret = read_huffsym_using_maintree(istream, tables,
800 if (main_element < LZX_NUM_CHARS) {
801 /* literal: 0 to LZX_NUM_CHARS - 1 */
802 window[window_pos++] = main_element;
804 /* match: LZX_NUM_CHARS to num_main_syms - 1 */
805 match_len = lzx_decode_match(main_element,
815 window_pos += match_len;
821 /* API function documented in wimlib.h */
823 wimlib_lzx_decompress2(const void *compressed_data, unsigned compressed_len,
824 void *uncompressed_data, unsigned uncompressed_len,
827 struct lzx_tables tables;
828 struct input_bitstream istream;
829 struct lzx_lru_queue queue;
833 unsigned num_main_syms;
835 bool e8_preprocessing_done;
837 LZX_DEBUG("compressed_data = %p, compressed_len = %u, "
838 "uncompressed_data = %p, uncompressed_len = %u, "
839 "max_window_size=%u).",
840 compressed_data, compressed_len,
841 uncompressed_data, uncompressed_len, max_window_size);
843 if (!lzx_window_size_valid(max_window_size)) {
844 LZX_DEBUG("Window size of %u is invalid!",
849 num_main_syms = lzx_get_num_main_syms(max_window_size);
851 if (uncompressed_len > max_window_size) {
852 LZX_DEBUG("Uncompressed chunk size of %u exceeds "
853 "window size of %u!",
854 uncompressed_len, max_window_size);
858 memset(tables.maintree_lens, 0, sizeof(tables.maintree_lens));
859 memset(tables.lentree_lens, 0, sizeof(tables.lentree_lens));
860 lzx_lru_queue_init(&queue);
861 init_input_bitstream(&istream, compressed_data, compressed_len);
863 e8_preprocessing_done = false; /* Set to true if there may be 0xe8 bytes
864 in the uncompressed data. */
866 /* The compressed data will consist of one or more blocks. The
867 * following loop decompresses one block, and it runs until there all
868 * the compressed data has been decompressed, so there are no more
872 window_pos < uncompressed_len;
873 window_pos += block_size)
875 LZX_DEBUG("Reading block header.");
876 ret = lzx_read_block_header(&istream, num_main_syms,
877 max_window_size, &block_size,
878 &block_type, &tables, &queue);
882 LZX_DEBUG("block_size = %u, window_pos = %u",
883 block_size, window_pos);
885 if (block_size > uncompressed_len - window_pos) {
886 LZX_DEBUG("Expected a block size of at "
887 "most %u bytes (found %u bytes)",
888 uncompressed_len - window_pos, block_size);
892 switch (block_type) {
893 case LZX_BLOCKTYPE_VERBATIM:
894 case LZX_BLOCKTYPE_ALIGNED:
895 if (block_type == LZX_BLOCKTYPE_VERBATIM)
896 LZX_DEBUG("LZX_BLOCKTYPE_VERBATIM");
898 LZX_DEBUG("LZX_BLOCKTYPE_ALIGNED");
899 ret = lzx_decompress_block(block_type,
910 if (tables.maintree_lens[0xe8] != 0)
911 e8_preprocessing_done = true;
913 case LZX_BLOCKTYPE_UNCOMPRESSED:
914 LZX_DEBUG("LZX_BLOCKTYPE_UNCOMPRESSED");
915 if (istream.data_bytes_left < block_size) {
916 LZX_DEBUG("Unexpected end of input when "
917 "reading %u bytes from LZX bitstream "
918 "(only have %u bytes left)",
919 block_size, istream.data_bytes_left);
922 memcpy(&((u8*)uncompressed_data)[window_pos], istream.data,
924 istream.data += block_size;
925 istream.data_bytes_left -= block_size;
926 /* Re-align bitstream if an odd number of bytes were
928 if (istream.data_bytes_left && (block_size & 1)) {
929 istream.data_bytes_left--;
932 e8_preprocessing_done = true;
936 if (e8_preprocessing_done)
937 undo_call_insn_preprocessing(uncompressed_data, uncompressed_len);
941 /* API function documented in wimlib.h */
943 wimlib_lzx_decompress(const void *compressed_data, unsigned compressed_len,
944 void *uncompressed_data, unsigned uncompressed_len)
946 return wimlib_lzx_decompress2(compressed_data, compressed_len,
947 uncompressed_data, uncompressed_len,