• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * LZMA2 decoder
3  *
4  * Authors: Lasse Collin <lasse.collin@tukaani.org>
5  *          Igor Pavlov <http://7-zip.org/>
6  *
7  * This file has been put into the public domain.
8  * You can do whatever you want with this file.
9  */
10 
11 #include "xz_private.h"
12 #include "xz_lzma2.h"
13 
14 /*
15  * Range decoder initialization eats the first five bytes of each LZMA chunk.
16  */
17 #define RC_INIT_BYTES 5
18 
19 /*
20  * Minimum number of usable input buffer to safely decode one LZMA symbol.
21  * The worst case is that we decode 22 bits using probabilities and 26
22  * direct bits. This may decode at maximum of 20 bytes of input. However,
23  * lzma_main() does an extra normalization before returning, thus we
24  * need to put 21 here.
25  */
26 #define LZMA_IN_REQUIRED 21
27 
28 /*
29  * Dictionary (history buffer)
30  *
31  * These are always true:
32  *    start <= pos <= full <= end
33  *    pos <= limit <= end
34  *
35  * In multi-call mode, also these are true:
36  *    end == size
37  *    size <= size_max
38  *    allocated <= size
39  *
40  * Most of these variables are size_t to support single-call mode,
41  * in which the dictionary variables address the actual output
42  * buffer directly.
43  */
44 struct dictionary {
45 	/* Beginning of the history buffer */
46 	uint8_t *buf;
47 
48 	/* Old position in buf (before decoding more data) */
49 	size_t start;
50 
51 	/* Position in buf */
52 	size_t pos;
53 
54 	/*
55 	 * How full dictionary is. This is used to detect corrupt input that
56 	 * would read beyond the beginning of the uncompressed stream.
57 	 */
58 	size_t full;
59 
60 	/* Write limit; we don't write to buf[limit] or later bytes. */
61 	size_t limit;
62 
63 	/*
64 	 * End of the dictionary buffer. In multi-call mode, this is
65 	 * the same as the dictionary size. In single-call mode, this
66 	 * indicates the size of the output buffer.
67 	 */
68 	size_t end;
69 
70 	/*
71 	 * Size of the dictionary as specified in Block Header. This is used
72 	 * together with "full" to detect corrupt input that would make us
73 	 * read beyond the beginning of the uncompressed stream.
74 	 */
75 	uint32_t size;
76 
77 	/*
78 	 * Maximum allowed dictionary size in multi-call mode.
79 	 * This is ignored in single-call mode.
80 	 */
81 	uint32_t size_max;
82 
83 	/*
84 	 * Amount of memory currently allocated for the dictionary.
85 	 * This is used only with XZ_DYNALLOC. (With XZ_PREALLOC,
86 	 * size_max is always the same as the allocated size.)
87 	 */
88 	uint32_t allocated;
89 
90 	/* Operation mode */
91 	enum xz_mode mode;
92 };
93 
94 /* Range decoder */
95 struct rc_dec {
96 	uint32_t range;
97 	uint32_t code;
98 
99 	/*
100 	 * Number of initializing bytes remaining to be read
101 	 * by rc_read_init().
102 	 */
103 	uint32_t init_bytes_left;
104 
105 	/*
106 	 * Buffer from which we read our input. It can be either
107 	 * temp.buf or the caller-provided input buffer.
108 	 */
109 	const uint8_t *in;
110 	size_t in_pos;
111 	size_t in_limit;
112 };
113 
114 /* Probabilities for a length decoder. */
115 struct lzma_len_dec {
116 	/* Probability of match length being at least 10 */
117 	uint16_t choice;
118 
119 	/* Probability of match length being at least 18 */
120 	uint16_t choice2;
121 
122 	/* Probabilities for match lengths 2-9 */
123 	uint16_t low[POS_STATES_MAX][LEN_LOW_SYMBOLS];
124 
125 	/* Probabilities for match lengths 10-17 */
126 	uint16_t mid[POS_STATES_MAX][LEN_MID_SYMBOLS];
127 
128 	/* Probabilities for match lengths 18-273 */
129 	uint16_t high[LEN_HIGH_SYMBOLS];
130 };
131 
132 struct lzma_dec {
133 	/* Distances of latest four matches */
134 	uint32_t rep0;
135 	uint32_t rep1;
136 	uint32_t rep2;
137 	uint32_t rep3;
138 
139 	/* Types of the most recently seen LZMA symbols */
140 	enum lzma_state state;
141 
142 	/*
143 	 * Length of a match. This is updated so that dict_repeat can
144 	 * be called again to finish repeating the whole match.
145 	 */
146 	uint32_t len;
147 
148 	/*
149 	 * LZMA properties or related bit masks (number of literal
150 	 * context bits, a mask dervied from the number of literal
151 	 * position bits, and a mask dervied from the number
152 	 * position bits)
153 	 */
154 	uint32_t lc;
155 	uint32_t literal_pos_mask; /* (1 << lp) - 1 */
156 	uint32_t pos_mask;         /* (1 << pb) - 1 */
157 
158 	/* If 1, it's a match. Otherwise it's a single 8-bit literal. */
159 	uint16_t is_match[STATES][POS_STATES_MAX];
160 
161 	/* If 1, it's a repeated match. The distance is one of rep0 .. rep3. */
162 	uint16_t is_rep[STATES];
163 
164 	/*
165 	 * If 0, distance of a repeated match is rep0.
166 	 * Otherwise check is_rep1.
167 	 */
168 	uint16_t is_rep0[STATES];
169 
170 	/*
171 	 * If 0, distance of a repeated match is rep1.
172 	 * Otherwise check is_rep2.
173 	 */
174 	uint16_t is_rep1[STATES];
175 
176 	/* If 0, distance of a repeated match is rep2. Otherwise it is rep3. */
177 	uint16_t is_rep2[STATES];
178 
179 	/*
180 	 * If 1, the repeated match has length of one byte. Otherwise
181 	 * the length is decoded from rep_len_decoder.
182 	 */
183 	uint16_t is_rep0_long[STATES][POS_STATES_MAX];
184 
185 	/*
186 	 * Probability tree for the highest two bits of the match
187 	 * distance. There is a separate probability tree for match
188 	 * lengths of 2 (i.e. MATCH_LEN_MIN), 3, 4, and [5, 273].
189 	 */
190 	uint16_t dist_slot[DIST_STATES][DIST_SLOTS];
191 
192 	/*
193 	 * Probility trees for additional bits for match distance
194 	 * when the distance is in the range [4, 127].
195 	 */
196 	uint16_t dist_special[FULL_DISTANCES - DIST_MODEL_END];
197 
198 	/*
199 	 * Probability tree for the lowest four bits of a match
200 	 * distance that is equal to or greater than 128.
201 	 */
202 	uint16_t dist_align[ALIGN_SIZE];
203 
204 	/* Length of a normal match */
205 	struct lzma_len_dec match_len_dec;
206 
207 	/* Length of a repeated match */
208 	struct lzma_len_dec rep_len_dec;
209 
210 	/* Probabilities of literals */
211 	uint16_t literal[LITERAL_CODERS_MAX][LITERAL_CODER_SIZE];
212 };
213 
214 struct lzma2_dec {
215 	/* Position in xz_dec_lzma2_run(). */
216 	enum lzma2_seq {
217 		SEQ_CONTROL,
218 		SEQ_UNCOMPRESSED_1,
219 		SEQ_UNCOMPRESSED_2,
220 		SEQ_COMPRESSED_0,
221 		SEQ_COMPRESSED_1,
222 		SEQ_PROPERTIES,
223 		SEQ_LZMA_PREPARE,
224 		SEQ_LZMA_RUN,
225 		SEQ_COPY
226 	} sequence;
227 
228 	/* Next position after decoding the compressed size of the chunk. */
229 	enum lzma2_seq next_sequence;
230 
231 	/* Uncompressed size of LZMA chunk (2 MiB at maximum) */
232 	uint32_t uncompressed;
233 
234 	/*
235 	 * Compressed size of LZMA chunk or compressed/uncompressed
236 	 * size of uncompressed chunk (64 KiB at maximum)
237 	 */
238 	uint32_t compressed;
239 
240 	/*
241 	 * True if dictionary reset is needed. This is false before
242 	 * the first chunk (LZMA or uncompressed).
243 	 */
244 	bool need_dict_reset;
245 
246 	/*
247 	 * True if new LZMA properties are needed. This is false
248 	 * before the first LZMA chunk.
249 	 */
250 	bool need_props;
251 };
252 
253 struct xz_dec_lzma2 {
254 	/*
255 	 * The order below is important on x86 to reduce code size and
256 	 * it shouldn't hurt on other platforms. Everything up to and
257 	 * including lzma.pos_mask are in the first 128 bytes on x86-32,
258 	 * which allows using smaller instructions to access those
259 	 * variables. On x86-64, fewer variables fit into the first 128
260 	 * bytes, but this is still the best order without sacrificing
261 	 * the readability by splitting the structures.
262 	 */
263 	struct rc_dec rc;
264 	struct dictionary dict;
265 	struct lzma2_dec lzma2;
266 	struct lzma_dec lzma;
267 
268 	/*
269 	 * Temporary buffer which holds small number of input bytes between
270 	 * decoder calls. See lzma2_lzma() for details.
271 	 */
272 	struct {
273 		uint32_t size;
274 		uint8_t buf[3 * LZMA_IN_REQUIRED];
275 	} temp;
276 };
277 
278 /**************
279  * Dictionary *
280  **************/
281 
282 /*
283  * Reset the dictionary state. When in single-call mode, set up the beginning
284  * of the dictionary to point to the actual output buffer.
285  */
dict_reset(struct dictionary * dict,struct xz_buf * b)286 static void dict_reset(struct dictionary *dict, struct xz_buf *b)
287 {
288 	if (DEC_IS_SINGLE(dict->mode)) {
289 		dict->buf = b->out + b->out_pos;
290 		dict->end = b->out_size - b->out_pos;
291 	}
292 
293 	dict->start = 0;
294 	dict->pos = 0;
295 	dict->limit = 0;
296 	dict->full = 0;
297 }
298 
299 /* Set dictionary write limit */
dict_limit(struct dictionary * dict,size_t out_max)300 static void dict_limit(struct dictionary *dict, size_t out_max)
301 {
302 	if (dict->end - dict->pos <= out_max)
303 		dict->limit = dict->end;
304 	else
305 		dict->limit = dict->pos + out_max;
306 }
307 
308 /* Return true if at least one byte can be written into the dictionary. */
dict_has_space(const struct dictionary * dict)309 static inline bool dict_has_space(const struct dictionary *dict)
310 {
311 	return dict->pos < dict->limit;
312 }
313 
314 /*
315  * Get a byte from the dictionary at the given distance. The distance is
316  * assumed to valid, or as a special case, zero when the dictionary is
317  * still empty. This special case is needed for single-call decoding to
318  * avoid writing a '\0' to the end of the destination buffer.
319  */
dict_get(const struct dictionary * dict,uint32_t dist)320 static inline uint32_t dict_get(const struct dictionary *dict, uint32_t dist)
321 {
322 	size_t offset = dict->pos - dist - 1;
323 
324 	if (dist >= dict->pos)
325 		offset += dict->end;
326 
327 	return dict->full > 0 ? dict->buf[offset] : 0;
328 }
329 
330 /*
331  * Put one byte into the dictionary. It is assumed that there is space for it.
332  */
dict_put(struct dictionary * dict,uint8_t byte)333 static inline void dict_put(struct dictionary *dict, uint8_t byte)
334 {
335 	dict->buf[dict->pos++] = byte;
336 
337 	if (dict->full < dict->pos)
338 		dict->full = dict->pos;
339 }
340 
341 /*
342  * Repeat given number of bytes from the given distance. If the distance is
343  * invalid, false is returned. On success, true is returned and *len is
344  * updated to indicate how many bytes were left to be repeated.
345  */
dict_repeat(struct dictionary * dict,uint32_t * len,uint32_t dist)346 static bool dict_repeat(struct dictionary *dict, uint32_t *len, uint32_t dist)
347 {
348 	size_t back;
349 	uint32_t left;
350 
351 	if (dist >= dict->full || dist >= dict->size)
352 		return false;
353 
354 	left = min_t(size_t, dict->limit - dict->pos, *len);
355 	*len -= left;
356 
357 	back = dict->pos - dist - 1;
358 	if (dist >= dict->pos)
359 		back += dict->end;
360 
361 	do {
362 		dict->buf[dict->pos++] = dict->buf[back++];
363 		if (back == dict->end)
364 			back = 0;
365 	} while (--left > 0);
366 
367 	if (dict->full < dict->pos)
368 		dict->full = dict->pos;
369 
370 	return true;
371 }
372 
373 /* Copy uncompressed data as is from input to dictionary and output buffers. */
dict_uncompressed(struct dictionary * dict,struct xz_buf * b,uint32_t * left)374 static void dict_uncompressed(struct dictionary *dict, struct xz_buf *b,
375 			      uint32_t *left)
376 {
377 	size_t copy_size;
378 
379 	while (*left > 0 && b->in_pos < b->in_size
380 			&& b->out_pos < b->out_size) {
381 		copy_size = min(b->in_size - b->in_pos,
382 				b->out_size - b->out_pos);
383 		if (copy_size > dict->end - dict->pos)
384 			copy_size = dict->end - dict->pos;
385 		if (copy_size > *left)
386 			copy_size = *left;
387 
388 		*left -= copy_size;
389 
390 		memcpy(dict->buf + dict->pos, b->in + b->in_pos, copy_size);
391 		dict->pos += copy_size;
392 
393 		if (dict->full < dict->pos)
394 			dict->full = dict->pos;
395 
396 		if (DEC_IS_MULTI(dict->mode)) {
397 			if (dict->pos == dict->end)
398 				dict->pos = 0;
399 
400 			memcpy(b->out + b->out_pos, b->in + b->in_pos,
401 					copy_size);
402 		}
403 
404 		dict->start = dict->pos;
405 
406 		b->out_pos += copy_size;
407 		b->in_pos += copy_size;
408 	}
409 }
410 
411 /*
412  * Flush pending data from dictionary to b->out. It is assumed that there is
413  * enough space in b->out. This is guaranteed because caller uses dict_limit()
414  * before decoding data into the dictionary.
415  */
dict_flush(struct dictionary * dict,struct xz_buf * b)416 static uint32_t dict_flush(struct dictionary *dict, struct xz_buf *b)
417 {
418 	size_t copy_size = dict->pos - dict->start;
419 
420 	if (DEC_IS_MULTI(dict->mode)) {
421 		if (dict->pos == dict->end)
422 			dict->pos = 0;
423 
424 		memcpy(b->out + b->out_pos, dict->buf + dict->start,
425 				copy_size);
426 	}
427 
428 	dict->start = dict->pos;
429 	b->out_pos += copy_size;
430 	return copy_size;
431 }
432 
433 /*****************
434  * Range decoder *
435  *****************/
436 
437 /* Reset the range decoder. */
rc_reset(struct rc_dec * rc)438 static void rc_reset(struct rc_dec *rc)
439 {
440 	rc->range = (uint32_t)-1;
441 	rc->code = 0;
442 	rc->init_bytes_left = RC_INIT_BYTES;
443 }
444 
445 /*
446  * Read the first five initial bytes into rc->code if they haven't been
447  * read already. (Yes, the first byte gets completely ignored.)
448  */
rc_read_init(struct rc_dec * rc,struct xz_buf * b)449 static bool rc_read_init(struct rc_dec *rc, struct xz_buf *b)
450 {
451 	while (rc->init_bytes_left > 0) {
452 		if (b->in_pos == b->in_size)
453 			return false;
454 
455 		rc->code = (rc->code << 8) + b->in[b->in_pos++];
456 		--rc->init_bytes_left;
457 	}
458 
459 	return true;
460 }
461 
462 /* Return true if there may not be enough input for the next decoding loop. */
rc_limit_exceeded(const struct rc_dec * rc)463 static inline bool rc_limit_exceeded(const struct rc_dec *rc)
464 {
465 	return rc->in_pos > rc->in_limit;
466 }
467 
468 /*
469  * Return true if it is possible (from point of view of range decoder) that
470  * we have reached the end of the LZMA chunk.
471  */
rc_is_finished(const struct rc_dec * rc)472 static inline bool rc_is_finished(const struct rc_dec *rc)
473 {
474 	return rc->code == 0;
475 }
476 
477 /* Read the next input byte if needed. */
rc_normalize(struct rc_dec * rc)478 static __always_inline void rc_normalize(struct rc_dec *rc)
479 {
480 	if (rc->range < RC_TOP_VALUE) {
481 		rc->range <<= RC_SHIFT_BITS;
482 		rc->code = (rc->code << RC_SHIFT_BITS) + rc->in[rc->in_pos++];
483 	}
484 }
485 
486 /*
487  * Decode one bit. In some versions, this function has been splitted in three
488  * functions so that the compiler is supposed to be able to more easily avoid
489  * an extra branch. In this particular version of the LZMA decoder, this
490  * doesn't seem to be a good idea (tested with GCC 3.3.6, 3.4.6, and 4.3.3
491  * on x86). Using a non-splitted version results in nicer looking code too.
492  *
493  * NOTE: This must return an int. Do not make it return a bool or the speed
494  * of the code generated by GCC 3.x decreases 10-15 %. (GCC 4.3 doesn't care,
495  * and it generates 10-20 % faster code than GCC 3.x from this file anyway.)
496  */
rc_bit(struct rc_dec * rc,uint16_t * prob)497 static __always_inline int rc_bit(struct rc_dec *rc, uint16_t *prob)
498 {
499 	uint32_t bound;
500 	int bit;
501 
502 	rc_normalize(rc);
503 	bound = (rc->range >> RC_BIT_MODEL_TOTAL_BITS) * *prob;
504 	if (rc->code < bound) {
505 		rc->range = bound;
506 		*prob += (RC_BIT_MODEL_TOTAL - *prob) >> RC_MOVE_BITS;
507 		bit = 0;
508 	} else {
509 		rc->range -= bound;
510 		rc->code -= bound;
511 		*prob -= *prob >> RC_MOVE_BITS;
512 		bit = 1;
513 	}
514 
515 	return bit;
516 }
517 
518 /* Decode a bittree starting from the most significant bit. */
rc_bittree(struct rc_dec * rc,uint16_t * probs,uint32_t limit)519 static __always_inline uint32_t rc_bittree(struct rc_dec *rc,
520 					   uint16_t *probs, uint32_t limit)
521 {
522 	uint32_t symbol = 1;
523 
524 	do {
525 		if (rc_bit(rc, &probs[symbol]))
526 			symbol = (symbol << 1) + 1;
527 		else
528 			symbol <<= 1;
529 	} while (symbol < limit);
530 
531 	return symbol;
532 }
533 
534 /* Decode a bittree starting from the least significant bit. */
rc_bittree_reverse(struct rc_dec * rc,uint16_t * probs,uint32_t * dest,uint32_t limit)535 static __always_inline void rc_bittree_reverse(struct rc_dec *rc,
536 					       uint16_t *probs,
537 					       uint32_t *dest, uint32_t limit)
538 {
539 	uint32_t symbol = 1;
540 	uint32_t i = 0;
541 
542 	do {
543 		if (rc_bit(rc, &probs[symbol])) {
544 			symbol = (symbol << 1) + 1;
545 			*dest += 1 << i;
546 		} else {
547 			symbol <<= 1;
548 		}
549 	} while (++i < limit);
550 }
551 
552 /* Decode direct bits (fixed fifty-fifty probability) */
rc_direct(struct rc_dec * rc,uint32_t * dest,uint32_t limit)553 static inline void rc_direct(struct rc_dec *rc, uint32_t *dest, uint32_t limit)
554 {
555 	uint32_t mask;
556 
557 	do {
558 		rc_normalize(rc);
559 		rc->range >>= 1;
560 		rc->code -= rc->range;
561 		mask = (uint32_t)0 - (rc->code >> 31);
562 		rc->code += rc->range & mask;
563 		*dest = (*dest << 1) + (mask + 1);
564 	} while (--limit > 0);
565 }
566 
567 /********
568  * LZMA *
569  ********/
570 
571 /* Get pointer to literal coder probability array. */
lzma_literal_probs(struct xz_dec_lzma2 * s)572 static uint16_t *lzma_literal_probs(struct xz_dec_lzma2 *s)
573 {
574 	uint32_t prev_byte = dict_get(&s->dict, 0);
575 	uint32_t low = prev_byte >> (8 - s->lzma.lc);
576 	uint32_t high = (s->dict.pos & s->lzma.literal_pos_mask) << s->lzma.lc;
577 	return s->lzma.literal[low + high];
578 }
579 
580 /* Decode a literal (one 8-bit byte) */
lzma_literal(struct xz_dec_lzma2 * s)581 static void lzma_literal(struct xz_dec_lzma2 *s)
582 {
583 	uint16_t *probs;
584 	uint32_t symbol;
585 	uint32_t match_byte;
586 	uint32_t match_bit;
587 	uint32_t offset;
588 	uint32_t i;
589 
590 	probs = lzma_literal_probs(s);
591 
592 	if (lzma_state_is_literal(s->lzma.state)) {
593 		symbol = rc_bittree(&s->rc, probs, 0x100);
594 	} else {
595 		symbol = 1;
596 		match_byte = dict_get(&s->dict, s->lzma.rep0) << 1;
597 		offset = 0x100;
598 
599 		do {
600 			match_bit = match_byte & offset;
601 			match_byte <<= 1;
602 			i = offset + match_bit + symbol;
603 
604 			if (rc_bit(&s->rc, &probs[i])) {
605 				symbol = (symbol << 1) + 1;
606 				offset &= match_bit;
607 			} else {
608 				symbol <<= 1;
609 				offset &= ~match_bit;
610 			}
611 		} while (symbol < 0x100);
612 	}
613 
614 	dict_put(&s->dict, (uint8_t)symbol);
615 	lzma_state_literal(&s->lzma.state);
616 }
617 
618 /* Decode the length of the match into s->lzma.len. */
lzma_len(struct xz_dec_lzma2 * s,struct lzma_len_dec * l,uint32_t pos_state)619 static void lzma_len(struct xz_dec_lzma2 *s, struct lzma_len_dec *l,
620 		     uint32_t pos_state)
621 {
622 	uint16_t *probs;
623 	uint32_t limit;
624 
625 	if (!rc_bit(&s->rc, &l->choice)) {
626 		probs = l->low[pos_state];
627 		limit = LEN_LOW_SYMBOLS;
628 		s->lzma.len = MATCH_LEN_MIN;
629 	} else {
630 		if (!rc_bit(&s->rc, &l->choice2)) {
631 			probs = l->mid[pos_state];
632 			limit = LEN_MID_SYMBOLS;
633 			s->lzma.len = MATCH_LEN_MIN + LEN_LOW_SYMBOLS;
634 		} else {
635 			probs = l->high;
636 			limit = LEN_HIGH_SYMBOLS;
637 			s->lzma.len = MATCH_LEN_MIN + LEN_LOW_SYMBOLS
638 					+ LEN_MID_SYMBOLS;
639 		}
640 	}
641 
642 	s->lzma.len += rc_bittree(&s->rc, probs, limit) - limit;
643 }
644 
645 /* Decode a match. The distance will be stored in s->lzma.rep0. */
lzma_match(struct xz_dec_lzma2 * s,uint32_t pos_state)646 static void lzma_match(struct xz_dec_lzma2 *s, uint32_t pos_state)
647 {
648 	uint16_t *probs;
649 	uint32_t dist_slot;
650 	uint32_t limit;
651 
652 	lzma_state_match(&s->lzma.state);
653 
654 	s->lzma.rep3 = s->lzma.rep2;
655 	s->lzma.rep2 = s->lzma.rep1;
656 	s->lzma.rep1 = s->lzma.rep0;
657 
658 	lzma_len(s, &s->lzma.match_len_dec, pos_state);
659 
660 	probs = s->lzma.dist_slot[lzma_get_dist_state(s->lzma.len)];
661 	dist_slot = rc_bittree(&s->rc, probs, DIST_SLOTS) - DIST_SLOTS;
662 
663 	if (dist_slot < DIST_MODEL_START) {
664 		s->lzma.rep0 = dist_slot;
665 	} else {
666 		limit = (dist_slot >> 1) - 1;
667 		s->lzma.rep0 = 2 + (dist_slot & 1);
668 
669 		if (dist_slot < DIST_MODEL_END) {
670 			s->lzma.rep0 <<= limit;
671 			probs = s->lzma.dist_special + s->lzma.rep0
672 					- dist_slot - 1;
673 			rc_bittree_reverse(&s->rc, probs,
674 					&s->lzma.rep0, limit);
675 		} else {
676 			rc_direct(&s->rc, &s->lzma.rep0, limit - ALIGN_BITS);
677 			s->lzma.rep0 <<= ALIGN_BITS;
678 			rc_bittree_reverse(&s->rc, s->lzma.dist_align,
679 					&s->lzma.rep0, ALIGN_BITS);
680 		}
681 	}
682 }
683 
684 /*
685  * Decode a repeated match. The distance is one of the four most recently
686  * seen matches. The distance will be stored in s->lzma.rep0.
687  */
lzma_rep_match(struct xz_dec_lzma2 * s,uint32_t pos_state)688 static void lzma_rep_match(struct xz_dec_lzma2 *s, uint32_t pos_state)
689 {
690 	uint32_t tmp;
691 
692 	if (!rc_bit(&s->rc, &s->lzma.is_rep0[s->lzma.state])) {
693 		if (!rc_bit(&s->rc, &s->lzma.is_rep0_long[
694 				s->lzma.state][pos_state])) {
695 			lzma_state_short_rep(&s->lzma.state);
696 			s->lzma.len = 1;
697 			return;
698 		}
699 	} else {
700 		if (!rc_bit(&s->rc, &s->lzma.is_rep1[s->lzma.state])) {
701 			tmp = s->lzma.rep1;
702 		} else {
703 			if (!rc_bit(&s->rc, &s->lzma.is_rep2[s->lzma.state])) {
704 				tmp = s->lzma.rep2;
705 			} else {
706 				tmp = s->lzma.rep3;
707 				s->lzma.rep3 = s->lzma.rep2;
708 			}
709 
710 			s->lzma.rep2 = s->lzma.rep1;
711 		}
712 
713 		s->lzma.rep1 = s->lzma.rep0;
714 		s->lzma.rep0 = tmp;
715 	}
716 
717 	lzma_state_long_rep(&s->lzma.state);
718 	lzma_len(s, &s->lzma.rep_len_dec, pos_state);
719 }
720 
721 /* LZMA decoder core */
lzma_main(struct xz_dec_lzma2 * s)722 static bool lzma_main(struct xz_dec_lzma2 *s)
723 {
724 	uint32_t pos_state;
725 
726 	/*
727 	 * If the dictionary was reached during the previous call, try to
728 	 * finish the possibly pending repeat in the dictionary.
729 	 */
730 	if (dict_has_space(&s->dict) && s->lzma.len > 0)
731 		dict_repeat(&s->dict, &s->lzma.len, s->lzma.rep0);
732 
733 	/*
734 	 * Decode more LZMA symbols. One iteration may consume up to
735 	 * LZMA_IN_REQUIRED - 1 bytes.
736 	 */
737 	while (dict_has_space(&s->dict) && !rc_limit_exceeded(&s->rc)) {
738 		pos_state = s->dict.pos & s->lzma.pos_mask;
739 
740 		if (!rc_bit(&s->rc, &s->lzma.is_match[
741 				s->lzma.state][pos_state])) {
742 			lzma_literal(s);
743 		} else {
744 			if (rc_bit(&s->rc, &s->lzma.is_rep[s->lzma.state]))
745 				lzma_rep_match(s, pos_state);
746 			else
747 				lzma_match(s, pos_state);
748 
749 			if (!dict_repeat(&s->dict, &s->lzma.len, s->lzma.rep0))
750 				return false;
751 		}
752 	}
753 
754 	/*
755 	 * Having the range decoder always normalized when we are outside
756 	 * this function makes it easier to correctly handle end of the chunk.
757 	 */
758 	rc_normalize(&s->rc);
759 
760 	return true;
761 }
762 
763 /*
764  * Reset the LZMA decoder and range decoder state. Dictionary is nore reset
765  * here, because LZMA state may be reset without resetting the dictionary.
766  */
lzma_reset(struct xz_dec_lzma2 * s)767 static void lzma_reset(struct xz_dec_lzma2 *s)
768 {
769 	uint16_t *probs;
770 	size_t i;
771 
772 	s->lzma.state = STATE_LIT_LIT;
773 	s->lzma.rep0 = 0;
774 	s->lzma.rep1 = 0;
775 	s->lzma.rep2 = 0;
776 	s->lzma.rep3 = 0;
777 
778 	/*
779 	 * All probabilities are initialized to the same value. This hack
780 	 * makes the code smaller by avoiding a separate loop for each
781 	 * probability array.
782 	 *
783 	 * This could be optimized so that only that part of literal
784 	 * probabilities that are actually required. In the common case
785 	 * we would write 12 KiB less.
786 	 */
787 	probs = s->lzma.is_match[0];
788 	for (i = 0; i < PROBS_TOTAL; ++i)
789 		probs[i] = RC_BIT_MODEL_TOTAL / 2;
790 
791 	rc_reset(&s->rc);
792 }
793 
794 /*
795  * Decode and validate LZMA properties (lc/lp/pb) and calculate the bit masks
796  * from the decoded lp and pb values. On success, the LZMA decoder state is
797  * reset and true is returned.
798  */
lzma_props(struct xz_dec_lzma2 * s,uint8_t props)799 static bool lzma_props(struct xz_dec_lzma2 *s, uint8_t props)
800 {
801 	if (props > (4 * 5 + 4) * 9 + 8)
802 		return false;
803 
804 	s->lzma.pos_mask = 0;
805 	while (props >= 9 * 5) {
806 		props -= 9 * 5;
807 		++s->lzma.pos_mask;
808 	}
809 
810 	s->lzma.pos_mask = (1 << s->lzma.pos_mask) - 1;
811 
812 	s->lzma.literal_pos_mask = 0;
813 	while (props >= 9) {
814 		props -= 9;
815 		++s->lzma.literal_pos_mask;
816 	}
817 
818 	s->lzma.lc = props;
819 
820 	if (s->lzma.lc + s->lzma.literal_pos_mask > 4)
821 		return false;
822 
823 	s->lzma.literal_pos_mask = (1 << s->lzma.literal_pos_mask) - 1;
824 
825 	lzma_reset(s);
826 
827 	return true;
828 }
829 
830 /*********
831  * LZMA2 *
832  *********/
833 
834 /*
835  * The LZMA decoder assumes that if the input limit (s->rc.in_limit) hasn't
836  * been exceeded, it is safe to read up to LZMA_IN_REQUIRED bytes. This
837  * wrapper function takes care of making the LZMA decoder's assumption safe.
838  *
839  * As long as there is plenty of input left to be decoded in the current LZMA
840  * chunk, we decode directly from the caller-supplied input buffer until
841  * there's LZMA_IN_REQUIRED bytes left. Those remaining bytes are copied into
842  * s->temp.buf, which (hopefully) gets filled on the next call to this
843  * function. We decode a few bytes from the temporary buffer so that we can
844  * continue decoding from the caller-supplied input buffer again.
845  */
lzma2_lzma(struct xz_dec_lzma2 * s,struct xz_buf * b)846 static bool lzma2_lzma(struct xz_dec_lzma2 *s, struct xz_buf *b)
847 {
848 	size_t in_avail;
849 	uint32_t tmp;
850 
851 	in_avail = b->in_size - b->in_pos;
852 	if (s->temp.size > 0 || s->lzma2.compressed == 0) {
853 		tmp = 2 * LZMA_IN_REQUIRED - s->temp.size;
854 		if (tmp > s->lzma2.compressed - s->temp.size)
855 			tmp = s->lzma2.compressed - s->temp.size;
856 		if (tmp > in_avail)
857 			tmp = in_avail;
858 
859 		memcpy(s->temp.buf + s->temp.size, b->in + b->in_pos, tmp);
860 
861 		if (s->temp.size + tmp == s->lzma2.compressed) {
862 			memzero(s->temp.buf + s->temp.size + tmp,
863 					sizeof(s->temp.buf)
864 						- s->temp.size - tmp);
865 			s->rc.in_limit = s->temp.size + tmp;
866 		} else if (s->temp.size + tmp < LZMA_IN_REQUIRED) {
867 			s->temp.size += tmp;
868 			b->in_pos += tmp;
869 			return true;
870 		} else {
871 			s->rc.in_limit = s->temp.size + tmp - LZMA_IN_REQUIRED;
872 		}
873 
874 		s->rc.in = s->temp.buf;
875 		s->rc.in_pos = 0;
876 
877 		if (!lzma_main(s) || s->rc.in_pos > s->temp.size + tmp)
878 			return false;
879 
880 		s->lzma2.compressed -= s->rc.in_pos;
881 
882 		if (s->rc.in_pos < s->temp.size) {
883 			s->temp.size -= s->rc.in_pos;
884 			memmove(s->temp.buf, s->temp.buf + s->rc.in_pos,
885 					s->temp.size);
886 			return true;
887 		}
888 
889 		b->in_pos += s->rc.in_pos - s->temp.size;
890 		s->temp.size = 0;
891 	}
892 
893 	in_avail = b->in_size - b->in_pos;
894 	if (in_avail >= LZMA_IN_REQUIRED) {
895 		s->rc.in = b->in;
896 		s->rc.in_pos = b->in_pos;
897 
898 		if (in_avail >= s->lzma2.compressed + LZMA_IN_REQUIRED)
899 			s->rc.in_limit = b->in_pos + s->lzma2.compressed;
900 		else
901 			s->rc.in_limit = b->in_size - LZMA_IN_REQUIRED;
902 
903 		if (!lzma_main(s))
904 			return false;
905 
906 		in_avail = s->rc.in_pos - b->in_pos;
907 		if (in_avail > s->lzma2.compressed)
908 			return false;
909 
910 		s->lzma2.compressed -= in_avail;
911 		b->in_pos = s->rc.in_pos;
912 	}
913 
914 	in_avail = b->in_size - b->in_pos;
915 	if (in_avail < LZMA_IN_REQUIRED) {
916 		if (in_avail > s->lzma2.compressed)
917 			in_avail = s->lzma2.compressed;
918 
919 		memcpy(s->temp.buf, b->in + b->in_pos, in_avail);
920 		s->temp.size = in_avail;
921 		b->in_pos += in_avail;
922 	}
923 
924 	return true;
925 }
926 
927 /*
928  * Take care of the LZMA2 control layer, and forward the job of actual LZMA
929  * decoding or copying of uncompressed chunks to other functions.
930  */
xz_dec_lzma2_run(struct xz_dec_lzma2 * s,struct xz_buf * b)931 XZ_EXTERN enum xz_ret xz_dec_lzma2_run(struct xz_dec_lzma2 *s,
932 				       struct xz_buf *b)
933 {
934 	uint32_t tmp;
935 
936 	while (b->in_pos < b->in_size || s->lzma2.sequence == SEQ_LZMA_RUN) {
937 		switch (s->lzma2.sequence) {
938 		case SEQ_CONTROL:
939 			/*
940 			 * LZMA2 control byte
941 			 *
942 			 * Exact values:
943 			 *   0x00   End marker
944 			 *   0x01   Dictionary reset followed by
945 			 *          an uncompressed chunk
946 			 *   0x02   Uncompressed chunk (no dictionary reset)
947 			 *
948 			 * Highest three bits (s->control & 0xE0):
949 			 *   0xE0   Dictionary reset, new properties and state
950 			 *          reset, followed by LZMA compressed chunk
951 			 *   0xC0   New properties and state reset, followed
952 			 *          by LZMA compressed chunk (no dictionary
953 			 *          reset)
954 			 *   0xA0   State reset using old properties,
955 			 *          followed by LZMA compressed chunk (no
956 			 *          dictionary reset)
957 			 *   0x80   LZMA chunk (no dictionary or state reset)
958 			 *
959 			 * For LZMA compressed chunks, the lowest five bits
960 			 * (s->control & 1F) are the highest bits of the
961 			 * uncompressed size (bits 16-20).
962 			 *
963 			 * A new LZMA2 stream must begin with a dictionary
964 			 * reset. The first LZMA chunk must set new
965 			 * properties and reset the LZMA state.
966 			 *
967 			 * Values that don't match anything described above
968 			 * are invalid and we return XZ_DATA_ERROR.
969 			 */
970 			tmp = b->in[b->in_pos++];
971 
972 			if (tmp == 0x00)
973 				return XZ_STREAM_END;
974 
975 			if (tmp >= 0xE0 || tmp == 0x01) {
976 				s->lzma2.need_props = true;
977 				s->lzma2.need_dict_reset = false;
978 				dict_reset(&s->dict, b);
979 			} else if (s->lzma2.need_dict_reset) {
980 				return XZ_DATA_ERROR;
981 			}
982 
983 			if (tmp >= 0x80) {
984 				s->lzma2.uncompressed = (tmp & 0x1F) << 16;
985 				s->lzma2.sequence = SEQ_UNCOMPRESSED_1;
986 
987 				if (tmp >= 0xC0) {
988 					/*
989 					 * When there are new properties,
990 					 * state reset is done at
991 					 * SEQ_PROPERTIES.
992 					 */
993 					s->lzma2.need_props = false;
994 					s->lzma2.next_sequence
995 							= SEQ_PROPERTIES;
996 
997 				} else if (s->lzma2.need_props) {
998 					return XZ_DATA_ERROR;
999 
1000 				} else {
1001 					s->lzma2.next_sequence
1002 							= SEQ_LZMA_PREPARE;
1003 					if (tmp >= 0xA0)
1004 						lzma_reset(s);
1005 				}
1006 			} else {
1007 				if (tmp > 0x02)
1008 					return XZ_DATA_ERROR;
1009 
1010 				s->lzma2.sequence = SEQ_COMPRESSED_0;
1011 				s->lzma2.next_sequence = SEQ_COPY;
1012 			}
1013 
1014 			break;
1015 
1016 		case SEQ_UNCOMPRESSED_1:
1017 			s->lzma2.uncompressed
1018 					+= (uint32_t)b->in[b->in_pos++] << 8;
1019 			s->lzma2.sequence = SEQ_UNCOMPRESSED_2;
1020 			break;
1021 
1022 		case SEQ_UNCOMPRESSED_2:
1023 			s->lzma2.uncompressed
1024 					+= (uint32_t)b->in[b->in_pos++] + 1;
1025 			s->lzma2.sequence = SEQ_COMPRESSED_0;
1026 			break;
1027 
1028 		case SEQ_COMPRESSED_0:
1029 			s->lzma2.compressed
1030 					= (uint32_t)b->in[b->in_pos++] << 8;
1031 			s->lzma2.sequence = SEQ_COMPRESSED_1;
1032 			break;
1033 
1034 		case SEQ_COMPRESSED_1:
1035 			s->lzma2.compressed
1036 					+= (uint32_t)b->in[b->in_pos++] + 1;
1037 			s->lzma2.sequence = s->lzma2.next_sequence;
1038 			break;
1039 
1040 		case SEQ_PROPERTIES:
1041 			if (!lzma_props(s, b->in[b->in_pos++]))
1042 				return XZ_DATA_ERROR;
1043 
1044 			s->lzma2.sequence = SEQ_LZMA_PREPARE;
1045 
1046 		/* Fall through */
1047 
1048 		case SEQ_LZMA_PREPARE:
1049 			if (s->lzma2.compressed < RC_INIT_BYTES)
1050 				return XZ_DATA_ERROR;
1051 
1052 			if (!rc_read_init(&s->rc, b))
1053 				return XZ_OK;
1054 
1055 			s->lzma2.compressed -= RC_INIT_BYTES;
1056 			s->lzma2.sequence = SEQ_LZMA_RUN;
1057 
1058 		/* Fall through */
1059 
1060 		case SEQ_LZMA_RUN:
1061 			/*
1062 			 * Set dictionary limit to indicate how much we want
1063 			 * to be encoded at maximum. Decode new data into the
1064 			 * dictionary. Flush the new data from dictionary to
1065 			 * b->out. Check if we finished decoding this chunk.
1066 			 * In case the dictionary got full but we didn't fill
1067 			 * the output buffer yet, we may run this loop
1068 			 * multiple times without changing s->lzma2.sequence.
1069 			 */
1070 			dict_limit(&s->dict, min_t(size_t,
1071 					b->out_size - b->out_pos,
1072 					s->lzma2.uncompressed));
1073 			if (!lzma2_lzma(s, b))
1074 				return XZ_DATA_ERROR;
1075 
1076 			s->lzma2.uncompressed -= dict_flush(&s->dict, b);
1077 
1078 			if (s->lzma2.uncompressed == 0) {
1079 				if (s->lzma2.compressed > 0 || s->lzma.len > 0
1080 						|| !rc_is_finished(&s->rc))
1081 					return XZ_DATA_ERROR;
1082 
1083 				rc_reset(&s->rc);
1084 				s->lzma2.sequence = SEQ_CONTROL;
1085 
1086 			} else if (b->out_pos == b->out_size
1087 					|| (b->in_pos == b->in_size
1088 						&& s->temp.size
1089 						< s->lzma2.compressed)) {
1090 				return XZ_OK;
1091 			}
1092 
1093 			break;
1094 
1095 		case SEQ_COPY:
1096 			dict_uncompressed(&s->dict, b, &s->lzma2.compressed);
1097 			if (s->lzma2.compressed > 0)
1098 				return XZ_OK;
1099 
1100 			s->lzma2.sequence = SEQ_CONTROL;
1101 			break;
1102 		}
1103 	}
1104 
1105 	return XZ_OK;
1106 }
1107 
xz_dec_lzma2_create(enum xz_mode mode,uint32_t dict_max)1108 XZ_EXTERN struct xz_dec_lzma2 *xz_dec_lzma2_create(enum xz_mode mode,
1109 						   uint32_t dict_max)
1110 {
1111 	struct xz_dec_lzma2 *s = kmalloc(sizeof(*s), GFP_KERNEL);
1112 	if (s == NULL)
1113 		return NULL;
1114 
1115 	s->dict.mode = mode;
1116 	s->dict.size_max = dict_max;
1117 
1118 	if (DEC_IS_PREALLOC(mode)) {
1119 		s->dict.buf = vmalloc(dict_max);
1120 		if (s->dict.buf == NULL) {
1121 			kfree(s);
1122 			return NULL;
1123 		}
1124 	} else if (DEC_IS_DYNALLOC(mode)) {
1125 		s->dict.buf = NULL;
1126 		s->dict.allocated = 0;
1127 	}
1128 
1129 	return s;
1130 }
1131 
xz_dec_lzma2_reset(struct xz_dec_lzma2 * s,uint8_t props)1132 XZ_EXTERN enum xz_ret xz_dec_lzma2_reset(struct xz_dec_lzma2 *s, uint8_t props)
1133 {
1134 	/* This limits dictionary size to 3 GiB to keep parsing simpler. */
1135 	if (props > 39)
1136 		return XZ_OPTIONS_ERROR;
1137 
1138 	s->dict.size = 2 + (props & 1);
1139 	s->dict.size <<= (props >> 1) + 11;
1140 
1141 	if (DEC_IS_MULTI(s->dict.mode)) {
1142 		if (s->dict.size > s->dict.size_max)
1143 			return XZ_MEMLIMIT_ERROR;
1144 
1145 		s->dict.end = s->dict.size;
1146 
1147 		if (DEC_IS_DYNALLOC(s->dict.mode)) {
1148 			if (s->dict.allocated < s->dict.size) {
1149 				vfree(s->dict.buf);
1150 				s->dict.buf = vmalloc(s->dict.size);
1151 				if (s->dict.buf == NULL) {
1152 					s->dict.allocated = 0;
1153 					return XZ_MEM_ERROR;
1154 				}
1155 			}
1156 		}
1157 	}
1158 
1159 	s->lzma.len = 0;
1160 
1161 	s->lzma2.sequence = SEQ_CONTROL;
1162 	s->lzma2.need_dict_reset = true;
1163 
1164 	s->temp.size = 0;
1165 
1166 	return XZ_OK;
1167 }
1168 
xz_dec_lzma2_end(struct xz_dec_lzma2 * s)1169 XZ_EXTERN void xz_dec_lzma2_end(struct xz_dec_lzma2 *s)
1170 {
1171 	if (DEC_IS_MULTI(s->dict.mode))
1172 		vfree(s->dict.buf);
1173 
1174 	kfree(s);
1175 }
1176