decompress_common: switch to subtables for Huffman decoding
[wimlib] / src / lzx_decompress.c
1 /*
2  * lzx_decompress.c
3  *
4  * A decompressor for the LZX compression format, as used in WIM files.
5  */
6
7 /*
8  * Copyright (C) 2012-2016 Eric Biggers
9  *
10  * This file is free software; you can redistribute it and/or modify it under
11  * the terms of the GNU Lesser General Public License as published by the Free
12  * Software Foundation; either version 3 of the License, or (at your option) any
13  * later version.
14  *
15  * This file is distributed in the hope that it will be useful, but WITHOUT
16  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
17  * FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more
18  * details.
19  *
20  * You should have received a copy of the GNU Lesser General Public License
21  * along with this file; if not, see http://www.gnu.org/licenses/.
22  */
23
24 /*
25  * LZX is an LZ77 and Huffman-code based compression format that has many
26  * similarities to DEFLATE (the format used by zlib/gzip).  The compression
27  * ratio is as good or better than DEFLATE.  See lzx_compress.c for a format
28  * overview, and see https://en.wikipedia.org/wiki/LZX_(algorithm) for a
29  * historical overview.  Here I make some pragmatic notes.
30  *
31  * The old specification for LZX is the document "Microsoft LZX Data Compression
32  * Format" (1997).  It defines the LZX format as used in cabinet files.  Allowed
33  * window sizes are 2^n where 15 <= n <= 21.  However, this document contains
34  * several errors, so don't read too much into it...
35  *
36  * The new specification for LZX is the document "[MS-PATCH]: LZX DELTA
37  * Compression and Decompression" (2014).  It defines the LZX format as used by
38  * Microsoft's binary patcher.  It corrects several errors in the 1997 document
39  * and extends the format in several ways --- namely, optional reference data,
40  * up to 2^25 byte windows, and longer match lengths.
41  *
42  * WIM files use a more restricted form of LZX.  No LZX DELTA extensions are
43  * present, the window is not "sliding", E8 preprocessing is done
44  * unconditionally with a fixed file size, and the maximum window size is always
45  * 2^15 bytes (equal to the size of each "chunk" in a compressed WIM resource).
46  * This code is primarily intended to implement this form of LZX.  But although
47  * not compatible with WIMGAPI, this code also supports maximum window sizes up
48  * to 2^21 bytes.
49  *
50  * TODO: Add support for window sizes up to 2^25 bytes.
51  */
52
53 #ifdef HAVE_CONFIG_H
54 #  include "config.h"
55 #endif
56
57 #include <string.h>
58
59 #include "wimlib/decompressor_ops.h"
60 #include "wimlib/decompress_common.h"
61 #include "wimlib/error.h"
62 #include "wimlib/lzx_common.h"
63 #include "wimlib/util.h"
64
65 /* These values are chosen for fast decompression.  */
66 #define LZX_MAINCODE_TABLEBITS          11
67 #define LZX_LENCODE_TABLEBITS           9
68 #define LZX_PRECODE_TABLEBITS           6
69 #define LZX_ALIGNEDCODE_TABLEBITS       7
70
71 #define LZX_READ_LENS_MAX_OVERRUN 50
72
73 struct lzx_decompressor {
74
75         DECODE_TABLE(maincode_decode_table, LZX_MAINCODE_MAX_NUM_SYMBOLS,
76                      LZX_MAINCODE_TABLEBITS, LZX_MAX_MAIN_CODEWORD_LEN);
77         u8 maincode_lens[LZX_MAINCODE_MAX_NUM_SYMBOLS + LZX_READ_LENS_MAX_OVERRUN];
78
79         DECODE_TABLE(lencode_decode_table, LZX_LENCODE_NUM_SYMBOLS,
80                      LZX_LENCODE_TABLEBITS, LZX_MAX_LEN_CODEWORD_LEN);
81         u8 lencode_lens[LZX_LENCODE_NUM_SYMBOLS + LZX_READ_LENS_MAX_OVERRUN];
82
83         union {
84                 DECODE_TABLE(alignedcode_decode_table, LZX_ALIGNEDCODE_NUM_SYMBOLS,
85                              LZX_ALIGNEDCODE_TABLEBITS, LZX_MAX_ALIGNED_CODEWORD_LEN);
86                 u8 alignedcode_lens[LZX_ALIGNEDCODE_NUM_SYMBOLS];
87         };
88
89         union {
90                 DECODE_TABLE(precode_decode_table, LZX_PRECODE_NUM_SYMBOLS,
91                              LZX_PRECODE_TABLEBITS, LZX_MAX_PRE_CODEWORD_LEN);
92                 u8 precode_lens[LZX_PRECODE_NUM_SYMBOLS];
93                 u8 extra_offset_bits[LZX_MAX_OFFSET_SLOTS];
94         };
95
96         unsigned window_order;
97         unsigned num_main_syms;
98
99         /* Like lzx_extra_offset_bits[], but does not include the entropy-coded
100          * bits of aligned offset blocks */
101         u8 extra_offset_bits_minus_aligned[LZX_MAX_OFFSET_SLOTS];
102
103 } _aligned_attribute(DECODE_TABLE_ALIGNMENT);
104
105 /* Read a Huffman-encoded symbol using the precode. */
106 static inline unsigned
107 read_presym(const struct lzx_decompressor *d, struct input_bitstream *is)
108 {
109         return read_huffsym(is, d->precode_decode_table,
110                             LZX_PRECODE_TABLEBITS, LZX_MAX_PRE_CODEWORD_LEN);
111 }
112
113 /* Read a Huffman-encoded symbol using the main code. */
114 static inline unsigned
115 read_mainsym(const struct lzx_decompressor *d, struct input_bitstream *is)
116 {
117         return read_huffsym(is, d->maincode_decode_table,
118                             LZX_MAINCODE_TABLEBITS, LZX_MAX_MAIN_CODEWORD_LEN);
119 }
120
121 /* Read a Huffman-encoded symbol using the length code. */
122 static inline unsigned
123 read_lensym(const struct lzx_decompressor *d, struct input_bitstream *is)
124 {
125         return read_huffsym(is, d->lencode_decode_table,
126                             LZX_LENCODE_TABLEBITS, LZX_MAX_LEN_CODEWORD_LEN);
127 }
128
129 /* Read a Huffman-encoded symbol using the aligned offset code. */
130 static inline unsigned
131 read_alignedsym(const struct lzx_decompressor *d, struct input_bitstream *is)
132 {
133         return read_huffsym(is, d->alignedcode_decode_table,
134                             LZX_ALIGNEDCODE_TABLEBITS, LZX_MAX_ALIGNED_CODEWORD_LEN);
135 }
136
137 /*
138  * Read a precode from the compressed input bitstream, then use it to decode
139  * @num_lens codeword length values and write them to @lens.
140  */
141 static int
142 lzx_read_codeword_lens(struct lzx_decompressor *d, struct input_bitstream *is,
143                        u8 *lens, unsigned num_lens)
144 {
145         u8 *len_ptr = lens;
146         u8 *lens_end = lens + num_lens;
147
148         /* Read the lengths of the precode codewords.  These are stored
149          * explicitly. */
150         for (int i = 0; i < LZX_PRECODE_NUM_SYMBOLS; i++) {
151                 d->precode_lens[i] =
152                         bitstream_read_bits(is, LZX_PRECODE_ELEMENT_SIZE);
153         }
154
155         /* Build the decoding table for the precode. */
156         if (make_huffman_decode_table(d->precode_decode_table,
157                                       LZX_PRECODE_NUM_SYMBOLS,
158                                       LZX_PRECODE_TABLEBITS,
159                                       d->precode_lens,
160                                       LZX_MAX_PRE_CODEWORD_LEN))
161                 return -1;
162
163         /* Decode the codeword lengths.  */
164         do {
165                 unsigned presym;
166                 u8 len;
167
168                 /* Read the next precode symbol.  */
169                 presym = read_presym(d, is);
170                 if (presym < 17) {
171                         /* Difference from old length  */
172                         len = *len_ptr - presym;
173                         if ((s8)len < 0)
174                                 len += 17;
175                         *len_ptr++ = len;
176                 } else {
177                         /* Special RLE values  */
178
179                         unsigned run_len;
180
181                         if (presym == 17) {
182                                 /* Run of 0's  */
183                                 run_len = 4 + bitstream_read_bits(is, 4);
184                                 len = 0;
185                         } else if (presym == 18) {
186                                 /* Longer run of 0's  */
187                                 run_len = 20 + bitstream_read_bits(is, 5);
188                                 len = 0;
189                         } else {
190                                 /* Run of identical lengths  */
191                                 run_len = 4 + bitstream_read_bits(is, 1);
192                                 presym = read_presym(d, is);
193                                 if (unlikely(presym > 17))
194                                         return -1;
195                                 len = *len_ptr - presym;
196                                 if ((s8)len < 0)
197                                         len += 17;
198                         }
199
200                         do {
201                                 *len_ptr++ = len;
202                         } while (--run_len);
203                         /*
204                          * The worst case overrun is when presym == 18,
205                          * run_len == 20 + 31, and only 1 length was remaining.
206                          * So LZX_READ_LENS_MAX_OVERRUN == 50.
207                          *
208                          * Overrun while reading the first half of maincode_lens
209                          * can corrupt the previous values in the second half.
210                          * This doesn't really matter because the resulting
211                          * lengths will still be in range, and data that
212                          * generates overruns is invalid anyway.
213                          */
214                 }
215         } while (len_ptr < lens_end);
216
217         return 0;
218 }
219
220 /*
221  * Read the header of an LZX block.  For all block types, the block type and
222  * size is saved in *block_type_ret and *block_size_ret, respectively.  For
223  * compressed blocks, the codeword lengths are also saved.  For uncompressed
224  * blocks, the recent offsets queue is also updated.
225  */
226 static int
227 lzx_read_block_header(struct lzx_decompressor *d, struct input_bitstream *is,
228                       u32 recent_offsets[], int *block_type_ret,
229                       u32 *block_size_ret)
230 {
231         int block_type;
232         u32 block_size;
233
234         bitstream_ensure_bits(is, 4);
235
236         /* Read the block type. */
237         block_type = bitstream_pop_bits(is, 3);
238
239         /* Read the block size. */
240         if (bitstream_pop_bits(is, 1)) {
241                 block_size = LZX_DEFAULT_BLOCK_SIZE;
242         } else {
243                 block_size = bitstream_read_bits(is, 16);
244                 if (d->window_order >= 16) {
245                         block_size <<= 8;
246                         block_size |= bitstream_read_bits(is, 8);
247                 }
248         }
249
250         switch (block_type) {
251
252         case LZX_BLOCKTYPE_ALIGNED:
253
254                 /* Read the aligned offset codeword lengths. */
255
256                 for (int i = 0; i < LZX_ALIGNEDCODE_NUM_SYMBOLS; i++) {
257                         d->alignedcode_lens[i] =
258                                 bitstream_read_bits(is,
259                                                     LZX_ALIGNEDCODE_ELEMENT_SIZE);
260                 }
261
262                 /* Fall though, since the rest of the header for aligned offset
263                  * blocks is the same as that for verbatim blocks.  */
264
265         case LZX_BLOCKTYPE_VERBATIM:
266
267                 /* Read the main codeword lengths, which are divided into two
268                  * parts: literal symbols and match headers. */
269
270                 if (lzx_read_codeword_lens(d, is, d->maincode_lens,
271                                            LZX_NUM_CHARS))
272                         return -1;
273
274                 if (lzx_read_codeword_lens(d, is, d->maincode_lens + LZX_NUM_CHARS,
275                                            d->num_main_syms - LZX_NUM_CHARS))
276                         return -1;
277
278
279                 /* Read the length codeword lengths. */
280
281                 if (lzx_read_codeword_lens(d, is, d->lencode_lens,
282                                            LZX_LENCODE_NUM_SYMBOLS))
283                         return -1;
284
285                 break;
286
287         case LZX_BLOCKTYPE_UNCOMPRESSED:
288                 /*
289                  * The header of an uncompressed block contains new values for
290                  * the recent offsets queue, starting on the next 16-bit
291                  * boundary in the bitstream.  Careful: if the stream is
292                  * *already* aligned, the correct thing to do is to throw away
293                  * the next 16 bits (this is probably a mistake in the format).
294                  */
295                 bitstream_ensure_bits(is, 1);
296                 bitstream_align(is);
297                 recent_offsets[0] = bitstream_read_u32(is);
298                 recent_offsets[1] = bitstream_read_u32(is);
299                 recent_offsets[2] = bitstream_read_u32(is);
300
301                 /* Offsets of 0 are invalid.  */
302                 if (recent_offsets[0] == 0 || recent_offsets[1] == 0 ||
303                     recent_offsets[2] == 0)
304                         return -1;
305                 break;
306
307         default:
308                 /* Unrecognized block type.  */
309                 return -1;
310         }
311
312         *block_type_ret = block_type;
313         *block_size_ret = block_size;
314         return 0;
315 }
316
317 /* Decompress a block of LZX-compressed data. */
318 static int
319 lzx_decompress_block(struct lzx_decompressor *d, struct input_bitstream *is,
320                      int block_type, u32 block_size,
321                      u8 * const out_begin, u8 *out_next, u32 recent_offsets[])
322 {
323         u8 * const block_end = out_next + block_size;
324         unsigned min_aligned_offset_slot;
325
326         /*
327          * Build the Huffman decode tables.  We always need to build the main
328          * and length decode tables.  For aligned blocks we additionally need to
329          * build the aligned offset decode table.
330          */
331
332         if (make_huffman_decode_table(d->maincode_decode_table,
333                                       d->num_main_syms,
334                                       LZX_MAINCODE_TABLEBITS,
335                                       d->maincode_lens,
336                                       LZX_MAX_MAIN_CODEWORD_LEN))
337                 return -1;
338
339         if (make_huffman_decode_table(d->lencode_decode_table,
340                                       LZX_LENCODE_NUM_SYMBOLS,
341                                       LZX_LENCODE_TABLEBITS,
342                                       d->lencode_lens,
343                                       LZX_MAX_LEN_CODEWORD_LEN))
344                 return -1;
345
346         if (block_type == LZX_BLOCKTYPE_ALIGNED) {
347                 if (make_huffman_decode_table(d->alignedcode_decode_table,
348                                               LZX_ALIGNEDCODE_NUM_SYMBOLS,
349                                               LZX_ALIGNEDCODE_TABLEBITS,
350                                               d->alignedcode_lens,
351                                               LZX_MAX_ALIGNED_CODEWORD_LEN))
352                         return -1;
353                 min_aligned_offset_slot = LZX_MIN_ALIGNED_OFFSET_SLOT;
354                 memcpy(d->extra_offset_bits, d->extra_offset_bits_minus_aligned,
355                        sizeof(lzx_extra_offset_bits));
356         } else {
357                 min_aligned_offset_slot = LZX_MAX_OFFSET_SLOTS;
358                 memcpy(d->extra_offset_bits, lzx_extra_offset_bits,
359                        sizeof(lzx_extra_offset_bits));
360         }
361
362         /* Decode the literals and matches. */
363
364         do {
365                 unsigned mainsym;
366                 unsigned length;
367                 u32 offset;
368                 unsigned offset_slot;
369
370                 mainsym = read_mainsym(d, is);
371                 if (mainsym < LZX_NUM_CHARS) {
372                         /* Literal */
373                         *out_next++ = mainsym;
374                         continue;
375                 }
376
377                 /* Match */
378
379                 /* Decode the length header and offset slot.  */
380                 STATIC_ASSERT(LZX_NUM_CHARS % LZX_NUM_LEN_HEADERS == 0);
381                 length = mainsym % LZX_NUM_LEN_HEADERS;
382                 offset_slot = (mainsym - LZX_NUM_CHARS) / LZX_NUM_LEN_HEADERS;
383
384                 /* If needed, read a length symbol to decode the full length. */
385                 if (length == LZX_NUM_PRIMARY_LENS)
386                         length += read_lensym(d, is);
387                 length += LZX_MIN_MATCH_LEN;
388
389                 if (offset_slot < LZX_NUM_RECENT_OFFSETS) {
390                         /* Repeat offset  */
391
392                         /* Note: This isn't a real LRU queue, since using the R2
393                          * offset doesn't bump the R1 offset down to R2. */
394                         offset = recent_offsets[offset_slot];
395                         recent_offsets[offset_slot] = recent_offsets[0];
396                 } else {
397                         /* Explicit offset  */
398                         offset = bitstream_read_bits(is, d->extra_offset_bits[offset_slot]);
399                         if (offset_slot >= min_aligned_offset_slot) {
400                                 offset = (offset << LZX_NUM_ALIGNED_OFFSET_BITS) |
401                                          read_alignedsym(d, is);
402                         }
403                         offset += lzx_offset_slot_base[offset_slot];
404
405                         /* Update the match offset LRU queue.  */
406                         STATIC_ASSERT(LZX_NUM_RECENT_OFFSETS == 3);
407                         recent_offsets[2] = recent_offsets[1];
408                         recent_offsets[1] = recent_offsets[0];
409                 }
410                 recent_offsets[0] = offset;
411
412                 /* Validate the match, then copy it to the current position.  */
413
414                 if (unlikely(length > block_end - out_next))
415                         return -1;
416
417                 if (unlikely(offset > out_next - out_begin))
418                         return -1;
419
420                 lz_copy(out_next, length, offset, block_end, LZX_MIN_MATCH_LEN);
421
422                 out_next += length;
423
424         } while (out_next != block_end);
425
426         return 0;
427 }
428
429 static int
430 lzx_decompress(const void *restrict compressed_data, size_t compressed_size,
431                void *restrict uncompressed_data, size_t uncompressed_size,
432                void *restrict _d)
433 {
434         struct lzx_decompressor *d = _d;
435         u8 * const out_begin = uncompressed_data;
436         u8 *out_next = out_begin;
437         u8 * const out_end = out_begin + uncompressed_size;
438         struct input_bitstream is;
439         STATIC_ASSERT(LZX_NUM_RECENT_OFFSETS == 3);
440         u32 recent_offsets[LZX_NUM_RECENT_OFFSETS] = {1, 1, 1};
441         unsigned may_have_e8_byte = 0;
442
443         init_input_bitstream(&is, compressed_data, compressed_size);
444
445         /* Codeword lengths begin as all 0's for delta encoding purposes. */
446         memset(d->maincode_lens, 0, d->num_main_syms);
447         memset(d->lencode_lens, 0, LZX_LENCODE_NUM_SYMBOLS);
448
449         /* Decompress blocks until we have all the uncompressed data. */
450
451         while (out_next != out_end) {
452                 int block_type;
453                 u32 block_size;
454
455                 if (lzx_read_block_header(d, &is, recent_offsets,
456                                           &block_type, &block_size))
457                         return -1;
458
459                 if (block_size < 1 || block_size > out_end - out_next)
460                         return -1;
461
462                 if (likely(block_type != LZX_BLOCKTYPE_UNCOMPRESSED)) {
463
464                         /* Compressed block */
465                         if (lzx_decompress_block(d, &is, block_type, block_size,
466                                                  out_begin, out_next,
467                                                  recent_offsets))
468                                 return -1;
469
470                         /* If the first E8 byte was in this block, then it must
471                          * have been encoded as a literal using mainsym E8. */
472                         may_have_e8_byte |= d->maincode_lens[0xE8];
473                 } else {
474
475                         /* Uncompressed block */
476                         if (bitstream_read_bytes(&is, out_next, block_size))
477                                 return -1;
478
479                         /* Re-align the bitstream if needed. */
480                         if (block_size & 1)
481                                 bitstream_read_byte(&is);
482
483                         /* There may have been an E8 byte in the block. */
484                         may_have_e8_byte = 1;
485                 }
486                 out_next += block_size;
487         }
488
489         /* Postprocess the data unless it cannot possibly contain E8 bytes. */
490         if (may_have_e8_byte)
491                 lzx_postprocess(uncompressed_data, uncompressed_size);
492
493         return 0;
494 }
495
496 static int
497 lzx_create_decompressor(size_t max_block_size, void **d_ret)
498 {
499         unsigned window_order;
500         struct lzx_decompressor *d;
501
502         window_order = lzx_get_window_order(max_block_size);
503         if (window_order == 0)
504                 return WIMLIB_ERR_INVALID_PARAM;
505
506         d = ALIGNED_MALLOC(sizeof(*d), DECODE_TABLE_ALIGNMENT);
507         if (!d)
508                 return WIMLIB_ERR_NOMEM;
509
510         d->window_order = window_order;
511         d->num_main_syms = lzx_get_num_main_syms(window_order);
512
513         /* Initialize 'd->extra_offset_bits_minus_aligned'. */
514         STATIC_ASSERT(sizeof(d->extra_offset_bits_minus_aligned) ==
515                       sizeof(lzx_extra_offset_bits));
516         STATIC_ASSERT(sizeof(d->extra_offset_bits) ==
517                       sizeof(lzx_extra_offset_bits));
518         memcpy(d->extra_offset_bits_minus_aligned, lzx_extra_offset_bits,
519                sizeof(lzx_extra_offset_bits));
520         for (unsigned offset_slot = LZX_MIN_ALIGNED_OFFSET_SLOT;
521              offset_slot < LZX_MAX_OFFSET_SLOTS; offset_slot++)
522         {
523                 d->extra_offset_bits_minus_aligned[offset_slot] -=
524                                 LZX_NUM_ALIGNED_OFFSET_BITS;
525         }
526
527         *d_ret = d;
528         return 0;
529 }
530
531 static void
532 lzx_free_decompressor(void *_d)
533 {
534         ALIGNED_FREE(_d);
535 }
536
537 const struct decompressor_ops lzx_decompressor_ops = {
538         .create_decompressor = lzx_create_decompressor,
539         .decompress          = lzx_decompress,
540         .free_decompressor   = lzx_free_decompressor,
541 };