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 */ 44struct 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 */ 95struct 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. */ 115struct 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 132struct 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 214struct 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 253struct 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 */ 286static void XZ_FUNC 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 */ 300static void XZ_FUNC 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. */ 309static __always_inline bool XZ_FUNC 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 */ 320static __always_inline uint32_t XZ_FUNC dict_get( 321 const struct dictionary *dict, uint32_t dist) 322{ 323 size_t offset = dict->pos - dist - 1; 324 325 if (dist >= dict->pos) 326 offset += dict->end; 327 328 return dict->full > 0 ? dict->buf[offset] : 0; 329} 330 331/* 332 * Put one byte into the dictionary. It is assumed that there is space for it. 333 */ 334static inline void XZ_FUNC dict_put(struct dictionary *dict, uint8_t byte) 335{ 336 dict->buf[dict->pos++] = byte; 337 338 if (dict->full < dict->pos) 339 dict->full = dict->pos; 340} 341 342/* 343 * Repeat given number of bytes from the given distance. If the distance is 344 * invalid, false is returned. On success, true is returned and *len is 345 * updated to indicate how many bytes were left to be repeated. 346 */ 347static bool XZ_FUNC dict_repeat( 348 struct dictionary *dict, uint32_t *len, uint32_t dist) 349{ 350 size_t back; 351 uint32_t left; 352 353 if (dist >= dict->full || dist >= dict->size) 354 return false; 355 356 left = min_t(size_t, dict->limit - dict->pos, *len); 357 *len -= left; 358 359 back = dict->pos - dist - 1; 360 if (dist >= dict->pos) 361 back += dict->end; 362 363 do { 364 dict->buf[dict->pos++] = dict->buf[back++]; 365 if (back == dict->end) 366 back = 0; 367 } while (--left > 0); 368 369 if (dict->full < dict->pos) 370 dict->full = dict->pos; 371 372 return true; 373} 374 375/* Copy uncompressed data as is from input to dictionary and output buffers. */ 376static void XZ_FUNC dict_uncompressed( 377 struct dictionary *dict, struct xz_buf *b, uint32_t *left) 378{ 379 size_t copy_size; 380 381 while (*left > 0 && b->in_pos < b->in_size 382 && b->out_pos < b->out_size) { 383 copy_size = min(b->in_size - b->in_pos, 384 b->out_size - b->out_pos); 385 if (copy_size > dict->end - dict->pos) 386 copy_size = dict->end - dict->pos; 387 if (copy_size > *left) 388 copy_size = *left; 389 390 *left -= copy_size; 391 392 memcpy(dict->buf + dict->pos, b->in + b->in_pos, copy_size); 393 dict->pos += copy_size; 394 395 if (dict->full < dict->pos) 396 dict->full = dict->pos; 397 398 if (DEC_IS_MULTI(dict->mode)) { 399 if (dict->pos == dict->end) 400 dict->pos = 0; 401 402 memcpy(b->out + b->out_pos, b->in + b->in_pos, 403 copy_size); 404 } 405 406 dict->start = dict->pos; 407 408 b->out_pos += copy_size; 409 b->in_pos += copy_size; 410 411 } 412} 413 414/* 415 * Flush pending data from dictionary to b->out. It is assumed that there is 416 * enough space in b->out. This is guaranteed because caller uses dict_limit() 417 * before decoding data into the dictionary. 418 */ 419static uint32_t XZ_FUNC dict_flush(struct dictionary *dict, struct xz_buf *b) 420{ 421 size_t copy_size = dict->pos - dict->start; 422 423 if (DEC_IS_MULTI(dict->mode)) { 424 if (dict->pos == dict->end) 425 dict->pos = 0; 426 427 memcpy(b->out + b->out_pos, dict->buf + dict->start, 428 copy_size); 429 } 430 431 dict->start = dict->pos; 432 b->out_pos += copy_size; 433 return copy_size; 434} 435 436/***************** 437 * Range decoder * 438 *****************/ 439 440/* Reset the range decoder. */ 441static void XZ_FUNC rc_reset(struct rc_dec *rc) 442{ 443 rc->range = (uint32_t)-1; 444 rc->code = 0; 445 rc->init_bytes_left = RC_INIT_BYTES; 446} 447 448/* 449 * Read the first five initial bytes into rc->code if they haven't been 450 * read already. (Yes, the first byte gets completely ignored.) 451 */ 452static bool XZ_FUNC rc_read_init(struct rc_dec *rc, struct xz_buf *b) 453{ 454 while (rc->init_bytes_left > 0) { 455 if (b->in_pos == b->in_size) 456 return false; 457 458 rc->code = (rc->code << 8) + b->in[b->in_pos++]; 459 --rc->init_bytes_left; 460 } 461 462 return true; 463} 464 465/* Return true if there may not be enough input for the next decoding loop. */ 466static inline bool XZ_FUNC rc_limit_exceeded(const struct rc_dec *rc) 467{ 468 return rc->in_pos > rc->in_limit; 469} 470 471/* 472 * Return true if it is possible (from point of view of range decoder) that 473 * we have reached the end of the LZMA chunk. 474 */ 475static inline bool XZ_FUNC rc_is_finished(const struct rc_dec *rc) 476{ 477 return rc->code == 0; 478} 479 480/* Read the next input byte if needed. */ 481static __always_inline void XZ_FUNC rc_normalize(struct rc_dec *rc) 482{ 483 if (rc->range < RC_TOP_VALUE) { 484 rc->range <<= RC_SHIFT_BITS; 485 rc->code = (rc->code << RC_SHIFT_BITS) + rc->in[rc->in_pos++]; 486 } 487} 488 489/* 490 * Decode one bit. In some versions, this function has been splitted in three 491 * functions so that the compiler is supposed to be able to more easily avoid 492 * an extra branch. In this particular version of the LZMA decoder, this 493 * doesn't seem to be a good idea (tested with GCC 3.3.6, 3.4.6, and 4.3.3 494 * on x86). Using a non-splitted version results in nicer looking code too. 495 * 496 * NOTE: This must return an int. Do not make it return a bool or the speed 497 * of the code generated by GCC 3.x decreases 10-15 %. (GCC 4.3 doesn't care, 498 * and it generates 10-20 % faster code than GCC 3.x from this file anyway.) 499 */ 500static __always_inline int XZ_FUNC rc_bit(struct rc_dec *rc, uint16_t *prob) 501{ 502 uint32_t bound; 503 int bit; 504 505 rc_normalize(rc); 506 bound = (rc->range >> RC_BIT_MODEL_TOTAL_BITS) * *prob; 507 if (rc->code < bound) { 508 rc->range = bound; 509 *prob += (RC_BIT_MODEL_TOTAL - *prob) >> RC_MOVE_BITS; 510 bit = 0; 511 } else { 512 rc->range -= bound; 513 rc->code -= bound; 514 *prob -= *prob >> RC_MOVE_BITS; 515 bit = 1; 516 } 517 518 return bit; 519} 520 521/* Decode a bittree starting from the most significant bit. */ 522static __always_inline uint32_t XZ_FUNC rc_bittree( 523 struct rc_dec *rc, uint16_t *probs, uint32_t limit) 524{ 525 uint32_t symbol = 1; 526 527 do { 528 if (rc_bit(rc, &probs[symbol])) 529 symbol = (symbol << 1) + 1; 530 else 531 symbol <<= 1; 532 } while (symbol < limit); 533 534 return symbol; 535} 536 537/* Decode a bittree starting from the least significant bit. */ 538static __always_inline void XZ_FUNC rc_bittree_reverse(struct rc_dec *rc, 539 uint16_t *probs, uint32_t *dest, uint32_t limit) 540{ 541 uint32_t symbol = 1; 542 uint32_t i = 0; 543 544 do { 545 if (rc_bit(rc, &probs[symbol])) { 546 symbol = (symbol << 1) + 1; 547 *dest += 1 << i; 548 } else { 549 symbol <<= 1; 550 } 551 } while (++i < limit); 552} 553 554/* Decode direct bits (fixed fifty-fifty probability) */ 555static inline void XZ_FUNC rc_direct( 556 struct rc_dec *rc, uint32_t *dest, uint32_t limit) 557{ 558 uint32_t mask; 559 560 do { 561 rc_normalize(rc); 562 rc->range >>= 1; 563 rc->code -= rc->range; 564 mask = (uint32_t)0 - (rc->code >> 31); 565 rc->code += rc->range & mask; 566 *dest = (*dest << 1) + (mask + 1); 567 } while (--limit > 0); 568} 569 570/******** 571 * LZMA * 572 ********/ 573 574/* Get pointer to literal coder probability array. */ 575static uint16_t * XZ_FUNC lzma_literal_probs(struct xz_dec_lzma2 *s) 576{ 577 uint32_t prev_byte = dict_get(&s->dict, 0); 578 uint32_t low = prev_byte >> (8 - s->lzma.lc); 579 uint32_t high = (s->dict.pos & s->lzma.literal_pos_mask) << s->lzma.lc; 580 return s->lzma.literal[low + high]; 581} 582 583/* Decode a literal (one 8-bit byte) */ 584static void XZ_FUNC lzma_literal(struct xz_dec_lzma2 *s) 585{ 586 uint16_t *probs; 587 uint32_t symbol; 588 uint32_t match_byte; 589 uint32_t match_bit; 590 uint32_t offset; 591 uint32_t i; 592 593 probs = lzma_literal_probs(s); 594 595 if (lzma_state_is_literal(s->lzma.state)) { 596 symbol = rc_bittree(&s->rc, probs, 0x100); 597 } else { 598 symbol = 1; 599 match_byte = dict_get(&s->dict, s->lzma.rep0) << 1; 600 offset = 0x100; 601 602 do { 603 match_bit = match_byte & offset; 604 match_byte <<= 1; 605 i = offset + match_bit + symbol; 606 607 if (rc_bit(&s->rc, &probs[i])) { 608 symbol = (symbol << 1) + 1; 609 offset &= match_bit; 610 } else { 611 symbol <<= 1; 612 offset &= ~match_bit; 613 } 614 } while (symbol < 0x100); 615 } 616 617 dict_put(&s->dict, (uint8_t)symbol); 618 lzma_state_literal(&s->lzma.state); 619} 620 621/* Decode the length of the match into s->lzma.len. */ 622static void XZ_FUNC lzma_len(struct xz_dec_lzma2 *s, struct lzma_len_dec *l, 623 uint32_t pos_state) 624{ 625 uint16_t *probs; 626 uint32_t limit; 627 628 if (!rc_bit(&s->rc, &l->choice)) { 629 probs = l->low[pos_state]; 630 limit = LEN_LOW_SYMBOLS; 631 s->lzma.len = MATCH_LEN_MIN; 632 } else { 633 if (!rc_bit(&s->rc, &l->choice2)) { 634 probs = l->mid[pos_state]; 635 limit = LEN_MID_SYMBOLS; 636 s->lzma.len = MATCH_LEN_MIN + LEN_LOW_SYMBOLS; 637 } else { 638 probs = l->high; 639 limit = LEN_HIGH_SYMBOLS; 640 s->lzma.len = MATCH_LEN_MIN + LEN_LOW_SYMBOLS 641 + LEN_MID_SYMBOLS; 642 } 643 } 644 645 s->lzma.len += rc_bittree(&s->rc, probs, limit) - limit; 646} 647 648/* Decode a match. The distance will be stored in s->lzma.rep0. */ 649static void XZ_FUNC lzma_match(struct xz_dec_lzma2 *s, uint32_t pos_state) 650{ 651 uint16_t *probs; 652 uint32_t dist_slot; 653 uint32_t limit; 654 655 lzma_state_match(&s->lzma.state); 656 657 s->lzma.rep3 = s->lzma.rep2; 658 s->lzma.rep2 = s->lzma.rep1; 659 s->lzma.rep1 = s->lzma.rep0; 660 661 lzma_len(s, &s->lzma.match_len_dec, pos_state); 662 663 probs = s->lzma.dist_slot[lzma_get_dist_state(s->lzma.len)]; 664 dist_slot = rc_bittree(&s->rc, probs, DIST_SLOTS) - DIST_SLOTS; 665 666 if (dist_slot < DIST_MODEL_START) { 667 s->lzma.rep0 = dist_slot; 668 } else { 669 limit = (dist_slot >> 1) - 1; 670 s->lzma.rep0 = 2 + (dist_slot & 1); 671 672 if (dist_slot < DIST_MODEL_END) { 673 s->lzma.rep0 <<= limit; 674 probs = s->lzma.dist_special + s->lzma.rep0 675 - dist_slot - 1; 676 rc_bittree_reverse(&s->rc, probs, 677 &s->lzma.rep0, limit); 678 } else { 679 rc_direct(&s->rc, &s->lzma.rep0, limit - ALIGN_BITS); 680 s->lzma.rep0 <<= ALIGN_BITS; 681 rc_bittree_reverse(&s->rc, s->lzma.dist_align, 682 &s->lzma.rep0, ALIGN_BITS); 683 } 684 } 685} 686 687/* 688 * Decode a repeated match. The distance is one of the four most recently 689 * seen matches. The distance will be stored in s->lzma.rep0. 690 */ 691static void XZ_FUNC lzma_rep_match(struct xz_dec_lzma2 *s, uint32_t pos_state) 692{ 693 uint32_t tmp; 694 695 if (!rc_bit(&s->rc, &s->lzma.is_rep0[s->lzma.state])) { 696 if (!rc_bit(&s->rc, &s->lzma.is_rep0_long[ 697 s->lzma.state][pos_state])) { 698 lzma_state_short_rep(&s->lzma.state); 699 s->lzma.len = 1; 700 return; 701 } 702 } else { 703 if (!rc_bit(&s->rc, &s->lzma.is_rep1[s->lzma.state])) { 704 tmp = s->lzma.rep1; 705 } else { 706 if (!rc_bit(&s->rc, &s->lzma.is_rep2[s->lzma.state])) { 707 tmp = s->lzma.rep2; 708 } else { 709 tmp = s->lzma.rep3; 710 s->lzma.rep3 = s->lzma.rep2; 711 } 712 713 s->lzma.rep2 = s->lzma.rep1; 714 } 715 716 s->lzma.rep1 = s->lzma.rep0; 717 s->lzma.rep0 = tmp; 718 } 719 720 lzma_state_long_rep(&s->lzma.state); 721 lzma_len(s, &s->lzma.rep_len_dec, pos_state); 722} 723 724/* LZMA decoder core */ 725static bool XZ_FUNC lzma_main(struct xz_dec_lzma2 *s) 726{ 727 uint32_t pos_state; 728 729 /* 730 * If the dictionary was reached during the previous call, try to 731 * finish the possibly pending repeat in the dictionary. 732 */ 733 if (dict_has_space(&s->dict) && s->lzma.len > 0) 734 dict_repeat(&s->dict, &s->lzma.len, s->lzma.rep0); 735 736 /* 737 * Decode more LZMA symbols. One iteration may consume up to 738 * LZMA_IN_REQUIRED - 1 bytes. 739 */ 740 while (dict_has_space(&s->dict) && !rc_limit_exceeded(&s->rc)) { 741 pos_state = s->dict.pos & s->lzma.pos_mask; 742 743 if (!rc_bit(&s->rc, &s->lzma.is_match[ 744 s->lzma.state][pos_state])) { 745 lzma_literal(s); 746 } else { 747 if (rc_bit(&s->rc, &s->lzma.is_rep[s->lzma.state])) 748 lzma_rep_match(s, pos_state); 749 else 750 lzma_match(s, pos_state); 751 752 if (!dict_repeat(&s->dict, &s->lzma.len, s->lzma.rep0)) 753 return false; 754 } 755 } 756 757 /* 758 * Having the range decoder always normalized when we are outside 759 * this function makes it easier to correctly handle end of the chunk. 760 */ 761 rc_normalize(&s->rc); 762 763 return true; 764} 765 766/* 767 * Reset the LZMA decoder and range decoder state. Dictionary is nore reset 768 * here, because LZMA state may be reset without resetting the dictionary. 769 */ 770static void XZ_FUNC lzma_reset(struct xz_dec_lzma2 *s) 771{ 772 uint16_t *probs; 773 size_t i; 774 775 s->lzma.state = STATE_LIT_LIT; 776 s->lzma.rep0 = 0; 777 s->lzma.rep1 = 0; 778 s->lzma.rep2 = 0; 779 s->lzma.rep3 = 0; 780 781 /* 782 * All probabilities are initialized to the same value. This hack 783 * makes the code smaller by avoiding a separate loop for each 784 * probability array. 785 * 786 * This could be optimized so that only that part of literal 787 * probabilities that are actually required. In the common case 788 * we would write 12 KiB less. 789 */ 790 probs = s->lzma.is_match[0]; 791 for (i = 0; i < PROBS_TOTAL; ++i) 792 probs[i] = RC_BIT_MODEL_TOTAL / 2; 793 794 rc_reset(&s->rc); 795} 796 797/* 798 * Decode and validate LZMA properties (lc/lp/pb) and calculate the bit masks 799 * from the decoded lp and pb values. On success, the LZMA decoder state is 800 * reset and true is returned. 801 */ 802static bool XZ_FUNC lzma_props(struct xz_dec_lzma2 *s, uint8_t props) 803{ 804 if (props > (4 * 5 + 4) * 9 + 8) 805 return false; 806 807 s->lzma.pos_mask = 0; 808 while (props >= 9 * 5) { 809 props -= 9 * 5; 810 ++s->lzma.pos_mask; 811 } 812 813 s->lzma.pos_mask = (1 << s->lzma.pos_mask) - 1; 814 815 s->lzma.literal_pos_mask = 0; 816 while (props >= 9) { 817 props -= 9; 818 ++s->lzma.literal_pos_mask; 819 } 820 821 s->lzma.lc = props; 822 823 if (s->lzma.lc + s->lzma.literal_pos_mask > 4) 824 return false; 825 826 s->lzma.literal_pos_mask = (1 << s->lzma.literal_pos_mask) - 1; 827 828 lzma_reset(s); 829 830 return true; 831} 832 833/********* 834 * LZMA2 * 835 *********/ 836 837/* 838 * The LZMA decoder assumes that if the input limit (s->rc.in_limit) hasn't 839 * been exceeded, it is safe to read up to LZMA_IN_REQUIRED bytes. This 840 * wrapper function takes care of making the LZMA decoder's assumption safe. 841 * 842 * As long as there is plenty of input left to be decoded in the current LZMA 843 * chunk, we decode directly from the caller-supplied input buffer until 844 * there's LZMA_IN_REQUIRED bytes left. Those remaining bytes are copied into 845 * s->temp.buf, which (hopefully) gets filled on the next call to this 846 * function. We decode a few bytes from the temporary buffer so that we can 847 * continue decoding from the caller-supplied input buffer again. 848 */ 849static bool XZ_FUNC lzma2_lzma(struct xz_dec_lzma2 *s, struct xz_buf *b) 850{ 851 size_t in_avail; 852 uint32_t tmp; 853 854 in_avail = b->in_size - b->in_pos; 855 if (s->temp.size > 0 || s->lzma2.compressed == 0) { 856 tmp = 2 * LZMA_IN_REQUIRED - s->temp.size; 857 if (tmp > s->lzma2.compressed - s->temp.size) 858 tmp = s->lzma2.compressed - s->temp.size; 859 if (tmp > in_avail) 860 tmp = in_avail; 861 862 memcpy(s->temp.buf + s->temp.size, b->in + b->in_pos, tmp); 863 864 if (s->temp.size + tmp == s->lzma2.compressed) { 865 memzero(s->temp.buf + s->temp.size + tmp, 866 sizeof(s->temp.buf) 867 - s->temp.size - tmp); 868 s->rc.in_limit = s->temp.size + tmp; 869 } else if (s->temp.size + tmp < LZMA_IN_REQUIRED) { 870 s->temp.size += tmp; 871 b->in_pos += tmp; 872 return true; 873 } else { 874 s->rc.in_limit = s->temp.size + tmp - LZMA_IN_REQUIRED; 875 } 876 877 s->rc.in = s->temp.buf; 878 s->rc.in_pos = 0; 879 880 if (!lzma_main(s) || s->rc.in_pos > s->temp.size + tmp) 881 return false; 882 883 s->lzma2.compressed -= s->rc.in_pos; 884 885 if (s->rc.in_pos < s->temp.size) { 886 s->temp.size -= s->rc.in_pos; 887 memmove(s->temp.buf, s->temp.buf + s->rc.in_pos, 888 s->temp.size); 889 return true; 890 } 891 892 b->in_pos += s->rc.in_pos - s->temp.size; 893 s->temp.size = 0; 894 } 895 896 in_avail = b->in_size - b->in_pos; 897 if (in_avail >= LZMA_IN_REQUIRED) { 898 s->rc.in = b->in; 899 s->rc.in_pos = b->in_pos; 900 901 if (in_avail >= s->lzma2.compressed + LZMA_IN_REQUIRED) 902 s->rc.in_limit = b->in_pos + s->lzma2.compressed; 903 else 904 s->rc.in_limit = b->in_size - LZMA_IN_REQUIRED; 905 906 if (!lzma_main(s)) 907 return false; 908 909 in_avail = s->rc.in_pos - b->in_pos; 910 if (in_avail > s->lzma2.compressed) 911 return false; 912 913 s->lzma2.compressed -= in_avail; 914 b->in_pos = s->rc.in_pos; 915 } 916 917 in_avail = b->in_size - b->in_pos; 918 if (in_avail < LZMA_IN_REQUIRED) { 919 if (in_avail > s->lzma2.compressed) 920 in_avail = s->lzma2.compressed; 921 922 memcpy(s->temp.buf, b->in + b->in_pos, in_avail); 923 s->temp.size = in_avail; 924 b->in_pos += in_avail; 925 } 926 927 return true; 928} 929 930/* 931 * Take care of the LZMA2 control layer, and forward the job of actual LZMA 932 * decoding or copying of uncompressed chunks to other functions. 933 */ 934XZ_EXTERN NOINLINE enum xz_ret XZ_FUNC xz_dec_lzma2_run( 935 struct xz_dec_lzma2 *s, struct xz_buf *b) 936{ 937 uint32_t tmp; 938 939 while (b->in_pos < b->in_size || s->lzma2.sequence == SEQ_LZMA_RUN) { 940 switch (s->lzma2.sequence) { 941 case SEQ_CONTROL: 942 /* 943 * LZMA2 control byte 944 * 945 * Exact values: 946 * 0x00 End marker 947 * 0x01 Dictionary reset followed by 948 * an uncompressed chunk 949 * 0x02 Uncompressed chunk (no dictionary reset) 950 * 951 * Highest three bits (s->control & 0xE0): 952 * 0xE0 Dictionary reset, new properties and state 953 * reset, followed by LZMA compressed chunk 954 * 0xC0 New properties and state reset, followed 955 * by LZMA compressed chunk (no dictionary 956 * reset) 957 * 0xA0 State reset using old properties, 958 * followed by LZMA compressed chunk (no 959 * dictionary reset) 960 * 0x80 LZMA chunk (no dictionary or state reset) 961 * 962 * For LZMA compressed chunks, the lowest five bits 963 * (s->control & 1F) are the highest bits of the 964 * uncompressed size (bits 16-20). 965 * 966 * A new LZMA2 stream must begin with a dictionary 967 * reset. The first LZMA chunk must set new 968 * properties and reset the LZMA state. 969 * 970 * Values that don't match anything described above 971 * are invalid and we return XZ_DATA_ERROR. 972 */ 973 tmp = b->in[b->in_pos++]; 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 == 0x00) 1008 return XZ_STREAM_END; 1009 1010 if (tmp > 0x02) 1011 return XZ_DATA_ERROR; 1012 1013 s->lzma2.sequence = SEQ_COMPRESSED_0; 1014 s->lzma2.next_sequence = SEQ_COPY; 1015 } 1016 1017 break; 1018 1019 case SEQ_UNCOMPRESSED_1: 1020 s->lzma2.uncompressed 1021 += (uint32_t)b->in[b->in_pos++] << 8; 1022 s->lzma2.sequence = SEQ_UNCOMPRESSED_2; 1023 break; 1024 1025 case SEQ_UNCOMPRESSED_2: 1026 s->lzma2.uncompressed 1027 += (uint32_t)b->in[b->in_pos++] + 1; 1028 s->lzma2.sequence = SEQ_COMPRESSED_0; 1029 break; 1030 1031 case SEQ_COMPRESSED_0: 1032 s->lzma2.compressed 1033 = (uint32_t)b->in[b->in_pos++] << 8; 1034 s->lzma2.sequence = SEQ_COMPRESSED_1; 1035 break; 1036 1037 case SEQ_COMPRESSED_1: 1038 s->lzma2.compressed 1039 += (uint32_t)b->in[b->in_pos++] + 1; 1040 s->lzma2.sequence = s->lzma2.next_sequence; 1041 break; 1042 1043 case SEQ_PROPERTIES: 1044 if (!lzma_props(s, b->in[b->in_pos++])) 1045 return XZ_DATA_ERROR; 1046 1047 s->lzma2.sequence = SEQ_LZMA_PREPARE; 1048 1049 case SEQ_LZMA_PREPARE: 1050 if (s->lzma2.compressed < RC_INIT_BYTES) 1051 return XZ_DATA_ERROR; 1052 1053 if (!rc_read_init(&s->rc, b)) 1054 return XZ_OK; 1055 1056 s->lzma2.compressed -= RC_INIT_BYTES; 1057 s->lzma2.sequence = SEQ_LZMA_RUN; 1058 1059 case SEQ_LZMA_RUN: 1060 /* 1061 * Set dictionary limit to indicate how much we want 1062 * to be encoded at maximum. Decode new data into the 1063 * dictionary. Flush the new data from dictionary to 1064 * b->out. Check if we finished decoding this chunk. 1065 * In case the dictionary got full but we didn't fill 1066 * the output buffer yet, we may run this loop 1067 * multiple times without changing s->lzma2.sequence. 1068 */ 1069 dict_limit(&s->dict, min_t(size_t, 1070 b->out_size - b->out_pos, 1071 s->lzma2.uncompressed)); 1072 if (!lzma2_lzma(s, b)) 1073 return XZ_DATA_ERROR; 1074 1075 s->lzma2.uncompressed -= dict_flush(&s->dict, b); 1076 1077 if (s->lzma2.uncompressed == 0) { 1078 if (s->lzma2.compressed > 0 || s->lzma.len > 0 1079 || !rc_is_finished(&s->rc)) 1080 return XZ_DATA_ERROR; 1081 1082 rc_reset(&s->rc); 1083 s->lzma2.sequence = SEQ_CONTROL; 1084 1085 } else if (b->out_pos == b->out_size 1086 || (b->in_pos == b->in_size 1087 && s->temp.size 1088 < s->lzma2.compressed)) { 1089 return XZ_OK; 1090 } 1091 1092 break; 1093 1094 case SEQ_COPY: 1095 dict_uncompressed(&s->dict, b, &s->lzma2.compressed); 1096 if (s->lzma2.compressed > 0) 1097 return XZ_OK; 1098 1099 s->lzma2.sequence = SEQ_CONTROL; 1100 break; 1101 } 1102 } 1103 1104 return XZ_OK; 1105} 1106 1107XZ_EXTERN struct xz_dec_lzma2 * XZ_FUNC xz_dec_lzma2_create( 1108 enum xz_mode mode, uint32_t dict_max) 1109{ 1110 struct xz_dec_lzma2 *s = kmalloc(sizeof(*s), GFP_KERNEL); 1111 if (s == NULL) 1112 return NULL; 1113 1114 s->dict.mode = mode; 1115 s->dict.size_max = dict_max; 1116 1117 if (DEC_IS_PREALLOC(mode)) { 1118 s->dict.buf = vmalloc(dict_max); 1119 if (s->dict.buf == NULL) { 1120 kfree(s); 1121 return NULL; 1122 } 1123 } else if (DEC_IS_DYNALLOC(mode)) { 1124 s->dict.buf = NULL; 1125 s->dict.allocated = 0; 1126 } 1127 1128 return s; 1129} 1130 1131XZ_EXTERN enum xz_ret XZ_FUNC xz_dec_lzma2_reset( 1132 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 1169XZ_EXTERN void XZ_FUNC 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