6 * Copyright (C) 2013 Eric Biggers
8 * This file is part of wimlib, a library for working with WIM files.
10 * wimlib is free software; you can redistribute it and/or modify it under the
11 * terms of the GNU General Public License as published by the Free
12 * Software Foundation; either version 3 of the License, or (at your option)
15 * wimlib is distributed in the hope that it will be useful, but WITHOUT ANY
16 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
17 * A PARTICULAR PURPOSE. See the GNU General Public License for more
20 * You should have received a copy of the GNU General Public License
21 * along with wimlib; if not, see http://www.gnu.org/licenses/.
25 * This is a decompressor for the LZMS compression format used by Microsoft.
26 * This format is not documented, but it is one of the formats supported by the
27 * compression API available in Windows 8, and as of Windows 8 it is one of the
28 * formats that can be used in WIM files.
30 * This decompressor only implements "raw" decompression, which decompresses a
31 * single LZMS-compressed block. This behavior is the same as that of
32 * Decompress() in the Windows 8 compression API when using a compression handle
33 * created with CreateDecompressor() with the Algorithm parameter specified as
34 * COMPRESS_ALGORITHM_LZMS | COMPRESS_RAW. Presumably, non-raw LZMS data
35 * is a container format from which the locations and sizes (both compressed and
36 * uncompressed) of the constituent blocks can be determined.
38 * An LZMS-compressed block must be read in 16-bit little endian units from both
39 * directions. One logical bitstream starts at the front of the block and
40 * proceeds forwards. Another logical bitstream starts at the end of the block
41 * and proceeds backwards. Bits read from the forwards bitstream constitute
42 * range-encoded data, whereas bits read from the backwards bitstream constitute
43 * Huffman-encoded symbols or verbatim bits. For both bitstreams, the ordering
44 * of the bits within the 16-bit coding units is such that the first bit is the
45 * high-order bit and the last bit is the low-order bit.
47 * From these two logical bitstreams, an LZMS decompressor can reconstitute the
48 * series of items that make up the LZMS data representation. Each such item
49 * may be a literal byte or a match. Matches may be either traditional LZ77
50 * matches or "delta" matches, either of which can have its offset encoded
51 * explicitly or encoded via a reference to a recently used (repeat) offset.
53 * A traditional LZ match consists of a length and offset; it asserts that the
54 * sequence of bytes beginning at the current position and extending for the
55 * length is exactly equal to the equal-length sequence of bytes at the offset
56 * back in the window. On the other hand, a delta match consists of a length,
57 * raw offset, and power. It asserts that the sequence of bytes beginning at
58 * the current position and extending for the length is equal to the bytewise
59 * sum of the two equal-length sequences of bytes (2**power) and (raw_offset *
60 * 2**power) bytes before the current position, minus bytewise the sequence of
61 * bytes beginning at (2**power + raw_offset * 2**power) bytes before the
62 * current position. Although not generally as useful as traditional LZ
63 * matches, delta matches can be helpful on some types of data. Both LZ and
64 * delta matches may overlap with the current position; in fact, the minimum
65 * offset is 1, regardless of match length.
67 * For LZ matches, up to 3 repeat offsets are allowed, similar to some other
68 * LZ-based formats such as LZX and LZMA. They must updated in an LRU fashion,
69 * except for a quirk: updates to the queue must be delayed by one LZMS item,
70 * except for the removal of a repeat match. As a result, 4 entries are
71 * actually needed in the queue, even though it is only possible to decode
72 * references to the first 3 at any given time. The queue must be initialized
73 * to the offsets {1, 2, 3, 4}.
75 * Repeat delta matches are handled similarly, but for them there are two queues
76 * updated in lock-step: one for powers and one for raw offsets. The power
77 * queue must be initialized to {0, 0, 0, 0}, and the raw offset queue must be
78 * initialized to {1, 2, 3, 4}.
80 * Bits from the range decoder must be used to disambiguate item types. The
81 * range decoder must hold two state variables: the range, which must initially
82 * be set to 0xffffffff, and the current code, which must initially be set to
83 * the first 32 bits read from the forwards bitstream. The range must be
84 * maintained above 0xffff; when it falls below 0xffff, both the range and code
85 * must be left-shifted by 16 bits and the low 16 bits of the code must be
86 * filled in with the next 16 bits from the forwards bitstream.
88 * To decode each bit, the range decoder requires a probability that is
89 * logically a real number between 0 and 1. Multiplying this probability by the
90 * current range and taking the floor gives the bound between the 0-bit region
91 * of the range and the 1-bit region of the range. However, in LZMS,
92 * probabilities are restricted to values of n/64 where n is an integer is
93 * between 1 and 63 inclusively, so the implementation may use integer
94 * operations instead. Following calculation of the bound, if the current code
95 * is in the 0-bit region, the new range becomes the current code and the
96 * decoded bit is 0; otherwise, the bound must be subtracted from both the range
97 * and the code, and the decoded bit is 1. More information about range coding
98 * can be found at https://en.wikipedia.org/wiki/Range_encoding. Furthermore,
99 * note that the LZMA format also uses range coding and has public domain code
102 * The probability used to range-decode each bit must be taken from a table, of
103 * which one instance must exist for each distinct context in which a
104 * range-decoded bit is needed. At each call of the range decoder, the
105 * appropriate probability must be obtained by indexing the appropriate
106 * probability table with the last 4 (in the context disambiguating literals
107 * from matches), 5 (in the context disambiguating LZ matches from delta
108 * matches), or 6 (in all other contexts) bits recently range-decoded in that
109 * context, ordered such that the most recently decoded bit is the low-order bit
112 * Furthermore, each probability entry itself is variable, as its value must be
113 * maintained as n/64 where n is the number of 0 bits in the most recently
114 * decoded 64 bits with that same entry. This allows the compressed
115 * representation to adapt to the input and use fewer bits to represent the most
116 * likely data; note that LZMA uses a similar scheme. Initially, the most
117 * recently 64 decoded bits for each probability entry are assumed to be
118 * 0x0000000055555555 (high order to low order); therefore, all probabilities
119 * are initially 48/64. During the course of decoding, each probability may be
120 * updated to as low as 0/64 (as a result of reading many consecutive 1 bits
121 * with that entry) or as high as 64/64 (as a result of reading many consecutive
122 * 0 bits with that entry); however, probabilities of 0/64 and 64/64 cannot be
123 * used as-is but rather must be adjusted to 1/64 and 63/64, respectively,
124 * before being used for range decoding.
126 * Representations of the LZMS items themselves must be read from the backwards
127 * bitstream. For this, there are 5 different Huffman codes used:
129 * - The literal code, used for decoding literal bytes. Each of the 256
130 * symbols represents a literal byte. This code must be rebuilt whenever
131 * 1024 symbols have been decoded with it.
133 * - The LZ offset code, used for decoding the offsets of standard LZ77
134 * matches. Each symbol represents a position slot, which corresponds to a
135 * base value and some number of extra bits which must be read and added to
136 * the base value to reconstitute the full offset. The number of symbols in
137 * this code is the number of position slots needed to represent all possible
138 * offsets in the uncompressed block. This code must be rebuilt whenever
139 * 1024 symbols have been decoded with it.
141 * - The length code, used for decoding length symbols. Each of the 54 symbols
142 * represents a length slot, which corresponds to a base value and some
143 * number of extra bits which must be read and added to the base value to
144 * reconstitute the full length. This code must be rebuilt whenever 512
145 * symbols have been decoded with it.
147 * - The delta offset code, used for decoding the offsets of delta matches.
148 * Each symbol corresponds to a position slot, which corresponds to a base
149 * value and some number of extra bits which must be read and added to the
150 * base value to reconstitute the full offset. The number of symbols in this
151 * code is equal to the number of symbols in the LZ offset code. This code
152 * must be rebuilt whenever 1024 symbols have been decoded with it.
154 * - The delta power code, used for decoding the powers of delta matches. Each
155 * of the 8 symbols corresponds to a power. This code must be rebuilt
156 * whenever 512 symbols have been decoded with it.
158 * All the LZMS Huffman codes must be built adaptively based on symbol
159 * frequencies. Initially, each code must be built assuming that all symbols
160 * have equal frequency. Following that, each code must be rebuilt whenever a
161 * certain number of symbols has been decoded with it.
163 * In general, multiple valid Huffman codes can be constructed from a set of
164 * symbol frequencies. Like other compression formats such as XPRESS, LZX, and
165 * DEFLATE, the LZMS format solves this ambiguity by requiring that all Huffman
166 * codes be constructed in canonical form. This form requires that same-length
167 * codewords be lexicographically ordered the same way as the corresponding
168 * symbols and that all shorter codewords lexicographically precede longer
171 * Codewords in all the LZMS Huffman codes are limited to 15 bits. If the
172 * canonical code for a given set of symbol frequencies has any codewords longer
173 * than 15 bits, then all frequencies must be divided by 2, rounding up, and the
174 * code construction must be attempted again.
176 * An LZMS-compressed block seemingly cannot have a compressed size greater than
177 * or equal to the uncompressed size. In such cases the block must be stored
180 * After all LZMS items have been decoded, the data must be postprocessed to
181 * translate absolute address encoded in x86 instructions into their original
182 * relative addresses.
184 * Details omitted above can be found in the code. Note that in the absence of
185 * an official specification there is no guarantee that this decompressor
186 * handles all possible cases.
194 #include "wimlib/compress_common.h"
195 #include "wimlib/decompressor_ops.h"
196 #include "wimlib/decompress_common.h"
197 #include "wimlib/lzms.h"
198 #include "wimlib/util.h"
202 #define LZMS_DECODE_TABLE_BITS 10
204 /* Structure used for range decoding, reading bits forwards. This is the first
205 * logical bitstream mentioned above. */
206 struct lzms_range_decoder_raw {
207 /* The relevant part of the current range. Although the logical range
208 * for range decoding is a very large integer, only a small portion
209 * matters at any given time, and it can be normalized (shifted left)
210 * whenever it gets too small. */
213 /* The current position in the range encoded by the portion of the input
217 /* Pointer to the next little-endian 16-bit integer in the compressed
218 * input data (reading forwards). */
221 /* Number of 16-bit integers remaining in the compressed input data
222 * (reading forwards). */
223 size_t num_le16_remaining;
226 /* Structure used for reading raw bits backwards. This is the second logical
227 * bitstream mentioned above. */
228 struct lzms_input_bitstream {
229 /* Holding variable for bits that have been read from the compressed
230 * data. The bits are ordered from high-order to low-order. */
231 /* XXX: Without special-case code to handle reading more than 17 bits
232 * at a time, this needs to be 64 bits rather than 32 bits. */
235 /* Number of bits in @bitbuf that are used. */
236 unsigned num_filled_bits;
238 /* Pointer to the one past the next little-endian 16-bit integer in the
239 * compressed input data (reading backwards). */
242 /* Number of 16-bit integers remaining in the compressed input data
243 * (reading backwards). */
244 size_t num_le16_remaining;
247 /* Structure used for range decoding. This wraps around `struct
248 * lzms_range_decoder_raw' to use and maintain probability entries. */
249 struct lzms_range_decoder {
250 /* Pointer to the raw range decoder, which has no persistent knowledge
251 * of probabilities. Multiple lzms_range_decoder's share the same
252 * lzms_range_decoder_raw. */
253 struct lzms_range_decoder_raw *rd;
255 /* Bits recently decoded by this range decoder. This are used as in
256 * index into @prob_entries. */
259 /* Bitmask for @state to prevent its value from exceeding the number of
260 * probability entries. */
263 /* Probability entries being used for this range decoder. */
264 struct lzms_probability_entry prob_entries[LZMS_MAX_NUM_STATES];
267 /* Structure used for Huffman decoding, optionally using the decoded symbols as
268 * slots into a base table to determine how many extra bits need to be read to
269 * reconstitute the full value. */
270 struct lzms_huffman_decoder {
272 /* Bitstream to read Huffman-encoded symbols and verbatim bits from.
273 * Multiple lzms_huffman_decoder's share the same lzms_input_bitstream.
275 struct lzms_input_bitstream *is;
277 /* Pointer to the slot base table to use. It is indexed by the decoded
278 * Huffman symbol that specifies the slot. The entry specifies the base
279 * value to use, and the position of its high bit is the number of
280 * additional bits that must be read to reconstitute the full value.
282 * This member need not be set if only raw Huffman symbols are being
283 * read using this decoder. */
284 const u32 *slot_base_tab;
286 const u8 *extra_bits_tab;
288 /* Number of symbols that have been read using this code far. Reset to
289 * 0 whenever the code is rebuilt. */
292 /* When @num_syms_read reaches this number, the Huffman code must be
296 /* Number of symbols in the represented Huffman code. */
299 /* Running totals of symbol frequencies. These are diluted slightly
300 * whenever the code is rebuilt. */
301 u32 sym_freqs[LZMS_MAX_NUM_SYMS];
303 /* The length, in bits, of each symbol in the Huffman code. */
304 u8 lens[LZMS_MAX_NUM_SYMS];
306 /* The codeword of each symbol in the Huffman code. */
307 u16 codewords[LZMS_MAX_NUM_SYMS];
309 /* A table for quickly decoding symbols encoded using the Huffman code.
311 u16 decode_table[(1U << LZMS_DECODE_TABLE_BITS) + 2 * LZMS_MAX_NUM_SYMS]
312 _aligned_attribute(DECODE_TABLE_ALIGNMENT);
315 /* State of the LZMS decompressor. */
316 struct lzms_decompressor {
318 /* Pointer to the beginning of the uncompressed data buffer. */
321 /* Pointer to the next position in the uncompressed data buffer. */
324 /* Pointer to one past the end of the uncompressed data buffer. */
327 /* Range decoder, which reads bits from the beginning of the compressed
328 * block, going forwards. */
329 struct lzms_range_decoder_raw rd;
331 /* Input bitstream, which reads from the end of the compressed block,
332 * going backwards. */
333 struct lzms_input_bitstream is;
335 /* Range decoders. */
336 struct lzms_range_decoder main_range_decoder;
337 struct lzms_range_decoder match_range_decoder;
338 struct lzms_range_decoder lz_match_range_decoder;
339 struct lzms_range_decoder lz_repeat_match_range_decoders[LZMS_NUM_RECENT_OFFSETS - 1];
340 struct lzms_range_decoder delta_match_range_decoder;
341 struct lzms_range_decoder delta_repeat_match_range_decoders[LZMS_NUM_RECENT_OFFSETS - 1];
343 /* Huffman decoders. */
344 struct lzms_huffman_decoder literal_decoder;
345 struct lzms_huffman_decoder lz_offset_decoder;
346 struct lzms_huffman_decoder length_decoder;
347 struct lzms_huffman_decoder delta_power_decoder;
348 struct lzms_huffman_decoder delta_offset_decoder;
350 /* LRU (least-recently-used) queues for match information. */
351 struct lzms_lru_queues lru;
353 /* Used for postprocessing. */
354 s32 last_target_usages[65536];
357 /* Initialize the input bitstream @is to read forwards from the specified
358 * compressed data buffer @in that is @in_limit 16-bit integers long. */
360 lzms_input_bitstream_init(struct lzms_input_bitstream *is,
361 const le16 *in, size_t in_limit)
364 is->num_filled_bits = 0;
365 is->in = in + in_limit;
366 is->num_le16_remaining = in_limit;
369 /* Ensures that @num_bits bits are buffered in the input bitstream. */
371 lzms_input_bitstream_ensure_bits(struct lzms_input_bitstream *is,
374 while (is->num_filled_bits < num_bits) {
377 LZMS_ASSERT(is->num_filled_bits + 16 <= sizeof(is->bitbuf) * 8);
379 if (unlikely(is->num_le16_remaining == 0))
382 next = le16_to_cpu(*--is->in);
383 is->num_le16_remaining--;
385 is->bitbuf |= next << (sizeof(is->bitbuf) * 8 - is->num_filled_bits - 16);
386 is->num_filled_bits += 16;
392 /* Returns the next @num_bits bits that are buffered in the input bitstream. */
394 lzms_input_bitstream_peek_bits(struct lzms_input_bitstream *is,
397 LZMS_ASSERT(is->num_filled_bits >= num_bits);
398 return is->bitbuf >> (sizeof(is->bitbuf) * 8 - num_bits);
401 /* Removes the next @num_bits bits that are buffered in the input bitstream. */
403 lzms_input_bitstream_remove_bits(struct lzms_input_bitstream *is,
406 LZMS_ASSERT(is->num_filled_bits >= num_bits);
407 is->bitbuf <<= num_bits;
408 is->num_filled_bits -= num_bits;
411 /* Removes and returns the next @num_bits bits that are buffered in the input
414 lzms_input_bitstream_pop_bits(struct lzms_input_bitstream *is,
417 u32 bits = lzms_input_bitstream_peek_bits(is, num_bits);
418 lzms_input_bitstream_remove_bits(is, num_bits);
422 /* Reads the next @num_bits from the input bitstream. */
424 lzms_input_bitstream_read_bits(struct lzms_input_bitstream *is,
427 if (unlikely(lzms_input_bitstream_ensure_bits(is, num_bits)))
429 return lzms_input_bitstream_pop_bits(is, num_bits);
432 /* Initialize the range decoder @rd to read forwards from the specified
433 * compressed data buffer @in that is @in_limit 16-bit integers long. */
435 lzms_range_decoder_raw_init(struct lzms_range_decoder_raw *rd,
436 const le16 *in, size_t in_limit)
438 rd->range = 0xffffffff;
439 rd->code = ((u32)le16_to_cpu(in[0]) << 16) |
440 ((u32)le16_to_cpu(in[1]) << 0);
442 rd->num_le16_remaining = in_limit - 2;
445 /* Ensures the current range of the range decoder has at least 16 bits of
448 lzms_range_decoder_raw_normalize(struct lzms_range_decoder_raw *rd)
450 if (rd->range <= 0xffff) {
452 if (unlikely(rd->num_le16_remaining == 0))
454 rd->code = (rd->code << 16) | le16_to_cpu(*rd->in++);
455 rd->num_le16_remaining--;
460 /* Decode and return the next bit from the range decoder (raw version).
462 * @prob is the chance out of LZMS_PROBABILITY_MAX that the next bit is 0.
465 lzms_range_decoder_raw_decode_bit(struct lzms_range_decoder_raw *rd, u32 prob)
469 /* Ensure the range has at least 16 bits of precision. */
470 lzms_range_decoder_raw_normalize(rd);
472 /* Based on the probability, calculate the bound between the 0-bit
473 * region and the 1-bit region of the range. */
474 bound = (rd->range >> LZMS_PROBABILITY_BITS) * prob;
476 if (rd->code < bound) {
477 /* Current code is in the 0-bit region of the range. */
481 /* Current code is in the 1-bit region of the range. */
488 /* Decode and return the next bit from the range decoder. This wraps around
489 * lzms_range_decoder_raw_decode_bit() to handle using and updating the
490 * appropriate probability table. */
492 lzms_range_decode_bit(struct lzms_range_decoder *dec)
494 struct lzms_probability_entry *prob_entry;
498 /* Load the probability entry corresponding to the current state. */
499 prob_entry = &dec->prob_entries[dec->state];
501 /* Treat the number of zero bits in the most recently decoded
502 * LZMS_PROBABILITY_MAX bits with this probability entry as the chance,
503 * out of LZMS_PROBABILITY_MAX, that the next bit will be a 0. However,
504 * don't allow 0% or 100% probabilities. */
505 prob = prob_entry->num_recent_zero_bits;
506 if (prob == LZMS_PROBABILITY_MAX)
507 prob = LZMS_PROBABILITY_MAX - 1;
511 /* Decode the next bit. */
512 bit = lzms_range_decoder_raw_decode_bit(dec->rd, prob);
514 /* Update the state based on the newly decoded bit. */
515 dec->state = (((dec->state << 1) | bit) & dec->mask);
517 /* Update the recent bits, including the cached count of 0's. */
518 BUILD_BUG_ON(LZMS_PROBABILITY_MAX > sizeof(prob_entry->recent_bits) * 8);
520 if (prob_entry->recent_bits & (1ULL << (LZMS_PROBABILITY_MAX - 1))) {
521 /* Replacing 1 bit with 0 bit; increment the zero count.
523 prob_entry->num_recent_zero_bits++;
526 if (!(prob_entry->recent_bits & (1ULL << (LZMS_PROBABILITY_MAX - 1)))) {
527 /* Replacing 0 bit with 1 bit; decrement the zero count.
529 prob_entry->num_recent_zero_bits--;
532 prob_entry->recent_bits = (prob_entry->recent_bits << 1) | bit;
534 /* Return the decoded bit. */
539 /* Build the decoding table for a new adaptive Huffman code using the alphabet
540 * used in the specified Huffman decoder, with the symbol frequencies
543 lzms_rebuild_adaptive_huffman_code(struct lzms_huffman_decoder *dec)
546 /* XXX: This implementation makes use of code already implemented for
547 * the XPRESS and LZX compression formats. However, since for the
548 * adaptive codes used in LZMS we don't actually need the explicit codes
549 * themselves, only the decode tables, it may be possible to optimize
550 * this by somehow directly building or updating the Huffman decode
551 * table. This may be a worthwhile optimization because the adaptive
552 * codes change many times throughout a decompression run. */
553 LZMS_DEBUG("Rebuilding adaptive Huffman code (num_syms=%u)",
555 make_canonical_huffman_code(dec->num_syms, LZMS_MAX_CODEWORD_LEN,
556 dec->sym_freqs, dec->lens, dec->codewords);
557 #if defined(ENABLE_LZMS_DEBUG)
560 make_huffman_decode_table(dec->decode_table, dec->num_syms,
561 LZMS_DECODE_TABLE_BITS, dec->lens,
562 LZMS_MAX_CODEWORD_LEN);
563 LZMS_ASSERT(ret == 0);
566 /* Decode and return the next Huffman-encoded symbol from the LZMS-compressed
567 * block using the specified Huffman decoder. */
569 lzms_huffman_decode_symbol(struct lzms_huffman_decoder *dec)
571 const u16 *decode_table = dec->decode_table;
572 struct lzms_input_bitstream *is = dec->is;
577 /* The Huffman codes used in LZMS are adaptive and must be rebuilt
578 * whenever a certain number of symbols have been read. Each such
579 * rebuild uses the current symbol frequencies, but the format also
580 * requires that the symbol frequencies be halved after each code
581 * rebuild. This diminishes the effect of old symbols on the current
582 * Huffman codes, thereby causing the Huffman codes to be more locally
584 if (dec->num_syms_read == dec->rebuild_freq) {
585 lzms_rebuild_adaptive_huffman_code(dec);
586 for (unsigned i = 0; i < dec->num_syms; i++) {
587 dec->sym_freqs[i] >>= 1;
588 dec->sym_freqs[i] += 1;
590 dec->num_syms_read = 0;
593 /* XXX: Copied from read_huffsym() (decompress_common.h), since this
594 * uses a different input bitstream type. Should unify the
595 * implementations. */
596 lzms_input_bitstream_ensure_bits(is, LZMS_MAX_CODEWORD_LEN);
598 /* Index the decode table by the next table_bits bits of the input. */
599 key_bits = lzms_input_bitstream_peek_bits(is, LZMS_DECODE_TABLE_BITS);
600 entry = decode_table[key_bits];
601 if (likely(entry < 0xC000)) {
602 /* Fast case: The decode table directly provided the symbol and
603 * codeword length. The low 11 bits are the symbol, and the
604 * high 5 bits are the codeword length. */
605 lzms_input_bitstream_remove_bits(is, entry >> 11);
608 /* Slow case: The codeword for the symbol is longer than
609 * table_bits, so the symbol does not have an entry directly in
610 * the first (1 << table_bits) entries of the decode table.
611 * Traverse the appropriate binary tree bit-by-bit in order to
612 * decode the symbol. */
613 lzms_input_bitstream_remove_bits(is, LZMS_DECODE_TABLE_BITS);
615 key_bits = (entry & 0x3FFF) + lzms_input_bitstream_pop_bits(is, 1);
616 } while ((entry = decode_table[key_bits]) >= 0xC000);
620 /* Tally and return the decoded symbol. */
621 ++dec->sym_freqs[sym];
622 ++dec->num_syms_read;
626 /* Decode a number from the LZMS bitstream, encoded as a Huffman-encoded symbol
627 * specifying a "slot" (whose corresponding value is looked up in a static
628 * table) plus the number specified by a number of extra bits depending on the
631 lzms_decode_value(struct lzms_huffman_decoder *dec)
634 unsigned num_extra_bits;
637 LZMS_ASSERT(dec->slot_base_tab != NULL);
638 LZMS_ASSERT(dec->extra_bits_tab != NULL);
640 /* Read the slot (position slot, length slot, etc.), which is encoded as
641 * a Huffman symbol. */
642 slot = lzms_huffman_decode_symbol(dec);
644 /* Get the number of extra bits needed to represent the range of values
645 * that share the slot. */
646 num_extra_bits = dec->extra_bits_tab[slot];
648 /* Read the number of extra bits and add them to the slot base to form
649 * the final decoded value. */
650 extra_bits = lzms_input_bitstream_read_bits(dec->is, num_extra_bits);
651 return dec->slot_base_tab[slot] + extra_bits;
654 /* Copy a literal to the output buffer. */
656 lzms_copy_literal(struct lzms_decompressor *ctx, u8 literal)
658 *ctx->out_next++ = literal;
662 /* Validate an LZ match and copy it to the output buffer. */
664 lzms_copy_lz_match(struct lzms_decompressor *ctx, u32 length, u32 offset)
669 if (length > ctx->out_end - ctx->out_next) {
670 LZMS_DEBUG("Match overrun!");
673 if (offset > ctx->out_next - ctx->out_begin) {
674 LZMS_DEBUG("Match underrun!");
678 out_next = ctx->out_next;
679 matchptr = out_next - offset;
681 *out_next++ = *matchptr++;
683 ctx->out_next = out_next;
687 /* Validate a delta match and copy it to the output buffer. */
689 lzms_copy_delta_match(struct lzms_decompressor *ctx, u32 length,
690 u32 power, u32 raw_offset)
692 u32 offset1 = 1U << power;
693 u32 offset2 = raw_offset << power;
694 u32 offset = offset1 + offset2;
700 if (length > ctx->out_end - ctx->out_next) {
701 LZMS_DEBUG("Match overrun!");
704 if (offset > ctx->out_next - ctx->out_begin) {
705 LZMS_DEBUG("Match underrun!");
709 out_next = ctx->out_next;
710 matchptr1 = out_next - offset1;
711 matchptr2 = out_next - offset2;
712 matchptr = out_next - offset;
715 *out_next++ = *matchptr1++ + *matchptr2++ - *matchptr++;
717 ctx->out_next = out_next;
721 /* Decode a (length, offset) pair from the input. */
723 lzms_decode_lz_match(struct lzms_decompressor *ctx)
728 /* Decode the match offset. The next range-encoded bit indicates
729 * whether it's a repeat offset or an explicit offset. */
731 bit = lzms_range_decode_bit(&ctx->lz_match_range_decoder);
733 /* Explicit offset. */
734 offset = lzms_decode_value(&ctx->lz_offset_decoder);
739 for (i = 0; i < LZMS_NUM_RECENT_OFFSETS - 1; i++)
740 if (!lzms_range_decode_bit(&ctx->lz_repeat_match_range_decoders[i]))
743 offset = ctx->lru.lz.recent_offsets[i];
745 for (; i < LZMS_NUM_RECENT_OFFSETS; i++)
746 ctx->lru.lz.recent_offsets[i] = ctx->lru.lz.recent_offsets[i + 1];
749 /* Decode match length, which is always given explicitly (there is no
750 * LRU queue for repeat lengths). */
751 length = lzms_decode_value(&ctx->length_decoder);
753 ctx->lru.lz.upcoming_offset = offset;
755 LZMS_DEBUG("Decoded %s LZ match: length=%u, offset=%u",
756 (bit ? "repeat" : "explicit"), length, offset);
758 /* Validate the match and copy it to the output. */
759 return lzms_copy_lz_match(ctx, length, offset);
762 /* Decodes a "delta" match from the input. */
764 lzms_decode_delta_match(struct lzms_decompressor *ctx)
767 u32 length, power, raw_offset;
769 /* Decode the match power and raw offset. The next range-encoded bit
770 * indicates whether these data are a repeat, or given explicitly. */
772 bit = lzms_range_decode_bit(&ctx->delta_match_range_decoder);
774 power = lzms_huffman_decode_symbol(&ctx->delta_power_decoder);
775 raw_offset = lzms_decode_value(&ctx->delta_offset_decoder);
779 for (i = 0; i < LZMS_NUM_RECENT_OFFSETS - 1; i++)
780 if (!lzms_range_decode_bit(&ctx->delta_repeat_match_range_decoders[i]))
783 power = ctx->lru.delta.recent_powers[i];
784 raw_offset = ctx->lru.delta.recent_offsets[i];
786 for (; i < LZMS_NUM_RECENT_OFFSETS; i++) {
787 ctx->lru.delta.recent_powers[i] = ctx->lru.delta.recent_powers[i + 1];
788 ctx->lru.delta.recent_offsets[i] = ctx->lru.delta.recent_offsets[i + 1];
792 length = lzms_decode_value(&ctx->length_decoder);
794 ctx->lru.delta.upcoming_power = power;
795 ctx->lru.delta.upcoming_offset = raw_offset;
797 LZMS_DEBUG("Decoded %s delta match: length=%u, power=%u, raw_offset=%u",
798 (bit ? "repeat" : "explicit"), length, power, raw_offset);
800 /* Validate the match and copy it to the output. */
801 return lzms_copy_delta_match(ctx, length, power, raw_offset);
804 /* Decode an LZ or delta match. */
806 lzms_decode_match(struct lzms_decompressor *ctx)
808 if (!lzms_range_decode_bit(&ctx->match_range_decoder))
809 return lzms_decode_lz_match(ctx);
811 return lzms_decode_delta_match(ctx);
814 /* Decode a literal byte encoded using the literal Huffman code. */
816 lzms_decode_literal(struct lzms_decompressor *ctx)
818 u8 literal = lzms_huffman_decode_symbol(&ctx->literal_decoder);
819 LZMS_DEBUG("Decoded literal: 0x%02x", literal);
820 return lzms_copy_literal(ctx, literal);
823 /* Decode the next LZMS match or literal. */
825 lzms_decode_item(struct lzms_decompressor *ctx)
829 ctx->lru.lz.upcoming_offset = 0;
830 ctx->lru.delta.upcoming_power = 0;
831 ctx->lru.delta.upcoming_offset = 0;
833 if (lzms_range_decode_bit(&ctx->main_range_decoder))
834 ret = lzms_decode_match(ctx);
836 ret = lzms_decode_literal(ctx);
841 lzms_update_lru_queues(&ctx->lru);
846 lzms_init_range_decoder(struct lzms_range_decoder *dec,
847 struct lzms_range_decoder_raw *rd, u32 num_states)
851 dec->mask = num_states - 1;
852 for (u32 i = 0; i < num_states; i++) {
853 dec->prob_entries[i].num_recent_zero_bits = LZMS_INITIAL_PROBABILITY;
854 dec->prob_entries[i].recent_bits = LZMS_INITIAL_RECENT_BITS;
859 lzms_init_huffman_decoder(struct lzms_huffman_decoder *dec,
860 struct lzms_input_bitstream *is,
861 const u32 *slot_base_tab,
862 const u8 *extra_bits_tab,
864 unsigned rebuild_freq)
867 dec->slot_base_tab = slot_base_tab;
868 dec->extra_bits_tab = extra_bits_tab;
869 dec->num_syms = num_syms;
870 dec->num_syms_read = rebuild_freq;
871 dec->rebuild_freq = rebuild_freq;
872 for (unsigned i = 0; i < num_syms; i++)
873 dec->sym_freqs[i] = 1;
876 /* Prepare to decode items from an LZMS-compressed block. */
878 lzms_init_decompressor(struct lzms_decompressor *ctx,
879 const void *cdata, unsigned clen,
880 void *ubuf, unsigned ulen)
882 unsigned num_position_slots;
884 LZMS_DEBUG("Initializing decompressor (clen=%u, ulen=%u)", clen, ulen);
886 /* Initialize output pointers. */
887 ctx->out_begin = ubuf;
888 ctx->out_next = ubuf;
889 ctx->out_end = (u8*)ubuf + ulen;
891 /* Initialize the raw range decoder (reading forwards). */
892 lzms_range_decoder_raw_init(&ctx->rd, cdata, clen / 2);
894 /* Initialize the input bitstream for Huffman symbols (reading
896 lzms_input_bitstream_init(&ctx->is, cdata, clen / 2);
898 /* Calculate the number of position slots needed for this compressed
900 num_position_slots = lzms_get_position_slot(ulen - 1) + 1;
902 LZMS_DEBUG("Using %u position slots", num_position_slots);
904 /* Initialize Huffman decoders for each alphabet used in the compressed
906 lzms_init_huffman_decoder(&ctx->literal_decoder, &ctx->is,
907 NULL, NULL, LZMS_NUM_LITERAL_SYMS,
908 LZMS_LITERAL_CODE_REBUILD_FREQ);
910 lzms_init_huffman_decoder(&ctx->lz_offset_decoder, &ctx->is,
911 lzms_position_slot_base,
912 lzms_extra_position_bits,
914 LZMS_LZ_OFFSET_CODE_REBUILD_FREQ);
916 lzms_init_huffman_decoder(&ctx->length_decoder, &ctx->is,
917 lzms_length_slot_base,
918 lzms_extra_length_bits,
920 LZMS_LENGTH_CODE_REBUILD_FREQ);
922 lzms_init_huffman_decoder(&ctx->delta_offset_decoder, &ctx->is,
923 lzms_position_slot_base,
924 lzms_extra_position_bits,
926 LZMS_DELTA_OFFSET_CODE_REBUILD_FREQ);
928 lzms_init_huffman_decoder(&ctx->delta_power_decoder, &ctx->is,
929 NULL, NULL, LZMS_NUM_DELTA_POWER_SYMS,
930 LZMS_DELTA_POWER_CODE_REBUILD_FREQ);
933 /* Initialize range decoders, all of which wrap around the same
934 * lzms_range_decoder_raw. */
935 lzms_init_range_decoder(&ctx->main_range_decoder,
936 &ctx->rd, LZMS_NUM_MAIN_STATES);
938 lzms_init_range_decoder(&ctx->match_range_decoder,
939 &ctx->rd, LZMS_NUM_MATCH_STATES);
941 lzms_init_range_decoder(&ctx->lz_match_range_decoder,
942 &ctx->rd, LZMS_NUM_LZ_MATCH_STATES);
944 for (size_t i = 0; i < ARRAY_LEN(ctx->lz_repeat_match_range_decoders); i++)
945 lzms_init_range_decoder(&ctx->lz_repeat_match_range_decoders[i],
946 &ctx->rd, LZMS_NUM_LZ_REPEAT_MATCH_STATES);
948 lzms_init_range_decoder(&ctx->delta_match_range_decoder,
949 &ctx->rd, LZMS_NUM_DELTA_MATCH_STATES);
951 for (size_t i = 0; i < ARRAY_LEN(ctx->delta_repeat_match_range_decoders); i++)
952 lzms_init_range_decoder(&ctx->delta_repeat_match_range_decoders[i],
953 &ctx->rd, LZMS_NUM_DELTA_REPEAT_MATCH_STATES);
955 /* Initialize LRU match information. */
956 lzms_init_lru_queues(&ctx->lru);
958 LZMS_DEBUG("Decompressor successfully initialized");
961 /* Decode the series of literals and matches from the LZMS-compressed data.
962 * Returns 0 on success; nonzero if the compressed data is invalid. */
964 lzms_decode_items(const u8 *cdata, size_t clen, u8 *ubuf, size_t ulen,
965 struct lzms_decompressor *ctx)
967 /* Initialize the LZMS decompressor. */
968 lzms_init_decompressor(ctx, cdata, clen, ubuf, ulen);
970 /* Decode the sequence of items. */
971 while (ctx->out_next != ctx->out_end) {
972 LZMS_DEBUG("Position %u", ctx->out_next - ctx->out_begin);
973 if (lzms_decode_item(ctx))
980 lzms_decompress(const void *compressed_data, size_t compressed_size,
981 void *uncompressed_data, size_t uncompressed_size, void *_ctx)
983 struct lzms_decompressor *ctx = _ctx;
985 /* The range decoder requires that a minimum of 4 bytes of compressed
986 * data be initially available. */
987 if (compressed_size < 4) {
988 LZMS_DEBUG("Compressed size too small (got %zu, expected >= 4)",
993 /* An LZMS-compressed data block should be evenly divisible into 16-bit
995 if (compressed_size % 2 != 0) {
996 LZMS_DEBUG("Compressed size not divisible by 2 (got %zu)",
1001 /* Handle the trivial case where nothing needs to be decompressed.
1002 * (Necessary because a window of size 0 does not have a valid position
1004 if (uncompressed_size == 0)
1007 /* The x86 post-processor requires that the uncompressed length fit into
1008 * a signed 32-bit integer. Also, the position slot table cannot be
1009 * searched for a position of INT32_MAX or greater. */
1010 if (uncompressed_size >= INT32_MAX) {
1011 LZMS_DEBUG("Uncompressed length too large "
1012 "(got %zu, expected < INT32_MAX)",
1017 /* Decode the literals and matches. */
1018 if (lzms_decode_items(compressed_data, compressed_size,
1019 uncompressed_data, uncompressed_size, ctx))
1022 /* Postprocess the data. */
1023 lzms_x86_filter(uncompressed_data, uncompressed_size,
1024 ctx->last_target_usages, true);
1026 LZMS_DEBUG("Decompression successful.");
1031 lzms_free_decompressor(void *_ctx)
1033 struct lzms_decompressor *ctx = _ctx;
1039 lzms_create_decompressor(size_t max_block_size,
1040 const struct wimlib_decompressor_params_header *params,
1043 struct lzms_decompressor *ctx;
1045 ctx = MALLOC(sizeof(struct lzms_decompressor));
1047 return WIMLIB_ERR_NOMEM;
1049 /* Initialize position and length slot data if not done already. */
1056 const struct decompressor_ops lzms_decompressor_ops = {
1057 .create_decompressor = lzms_create_decompressor,
1058 .decompress = lzms_decompress,
1059 .free_decompressor = lzms_free_decompressor,