1 /* Lzma decompressor for Linux kernel. Shamelessly snarfed
4 *Linux kernel adaptation
5 *Copyright (C) 2006 Alain < alain@knaff.lu >
7 *Based on small lzma deflate implementation/Small range coder
8 *implementation for lzma.
9 *Copyright (C) 2006 Aurelien Jacobs < aurel@gnuage.org >
11 *Based on LzmaDecode.c from the LZMA SDK 4.22 (http://www.7-zip.org/)
12 *Copyright (C) 1999-2005 Igor Pavlov
14 *Copyrights of the parts, see headers below.
17 *This program is free software; you can redistribute it and/or
18 *modify it under the terms of the GNU Lesser General Public
19 *License as published by the Free Software Foundation; either
20 *version 2.1 of the License, or (at your option) any later version.
22 *This program is distributed in the hope that it will be useful,
23 *but WITHOUT ANY WARRANTY; without even the implied warranty of
24 *MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
25 *Lesser General Public License for more details.
27 *You should have received a copy of the GNU Lesser General Public
28 *License along with this library; if not, write to the Free Software
29 *Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
35 #include <linux/decompress/unlzma.h>
38 #include <linux/decompress/mm.h>
40 #define MIN(a, b) (((a) < (b)) ? (a) : (b))
42 static long long INIT read_int(unsigned char *ptr, int size)
47 for (i = 0; i < size; i++)
48 ret = (ret << 8) | ptr[size-i-1];
52 #define ENDIAN_CONVERT(x) \
53 x = (typeof(x))read_int((unsigned char *)&x, sizeof(x))
56 /* Small range coder implementation for lzma.
57 *Copyright (C) 2006 Aurelien Jacobs < aurel@gnuage.org >
59 *Based on LzmaDecode.c from the LZMA SDK 4.22 (http://www.7-zip.org/)
60 *Copyright (c) 1999-2005 Igor Pavlov
63 #include <linux/compiler.h>
65 #define LZMA_IOBUF_SIZE 0x10000
68 int (*fill)(void*, unsigned int);
76 void (*error)(char *);
80 #define RC_TOP_BITS 24
81 #define RC_MOVE_BITS 5
82 #define RC_MODEL_TOTAL_BITS 11
85 static int INIT nofill(void *buffer, unsigned int len)
90 /* Called twice: once at startup and once in rc_normalize() */
91 static void INIT rc_read(struct rc *rc)
93 rc->buffer_size = rc->fill((char *)rc->buffer, LZMA_IOBUF_SIZE);
94 if (rc->buffer_size <= 0)
95 rc->error("unexpected EOF");
97 rc->buffer_end = rc->buffer + rc->buffer_size;
101 static inline void INIT rc_init(struct rc *rc,
102 int (*fill)(void*, unsigned int),
103 char *buffer, int buffer_size)
109 rc->buffer = (uint8_t *)buffer;
110 rc->buffer_size = buffer_size;
111 rc->buffer_end = rc->buffer + rc->buffer_size;
112 rc->ptr = rc->buffer;
115 rc->range = 0xFFFFFFFF;
118 static inline void INIT rc_init_code(struct rc *rc)
122 for (i = 0; i < 5; i++) {
123 if (rc->ptr >= rc->buffer_end)
125 rc->code = (rc->code << 8) | *rc->ptr++;
130 /* Called twice, but one callsite is in inline'd rc_is_bit_0_helper() */
131 static void INIT rc_do_normalize(struct rc *rc)
133 if (rc->ptr >= rc->buffer_end)
136 rc->code = (rc->code << 8) | *rc->ptr++;
138 static inline void INIT rc_normalize(struct rc *rc)
140 if (rc->range < (1 << RC_TOP_BITS))
145 /* Why rc_is_bit_0_helper exists?
146 *Because we want to always expose (rc->code < rc->bound) to optimizer
148 static inline uint32_t INIT rc_is_bit_0_helper(struct rc *rc, uint16_t *p)
151 rc->bound = *p * (rc->range >> RC_MODEL_TOTAL_BITS);
154 static inline int INIT rc_is_bit_0(struct rc *rc, uint16_t *p)
156 uint32_t t = rc_is_bit_0_helper(rc, p);
160 /* Called ~10 times, but very small, thus inlined */
161 static inline void INIT rc_update_bit_0(struct rc *rc, uint16_t *p)
163 rc->range = rc->bound;
164 *p += ((1 << RC_MODEL_TOTAL_BITS) - *p) >> RC_MOVE_BITS;
166 static inline void INIT rc_update_bit_1(struct rc *rc, uint16_t *p)
168 rc->range -= rc->bound;
169 rc->code -= rc->bound;
170 *p -= *p >> RC_MOVE_BITS;
173 /* Called 4 times in unlzma loop */
174 static int INIT rc_get_bit(struct rc *rc, uint16_t *p, int *symbol)
176 if (rc_is_bit_0(rc, p)) {
177 rc_update_bit_0(rc, p);
181 rc_update_bit_1(rc, p);
182 *symbol = *symbol * 2 + 1;
188 static inline int INIT rc_direct_bit(struct rc *rc)
192 if (rc->code >= rc->range) {
193 rc->code -= rc->range;
200 static inline void INIT
201 rc_bit_tree_decode(struct rc *rc, uint16_t *p, int num_levels, int *symbol)
207 rc_get_bit(rc, p + *symbol, symbol);
208 *symbol -= 1 << num_levels;
213 * Small lzma deflate implementation.
214 * Copyright (C) 2006 Aurelien Jacobs < aurel@gnuage.org >
216 * Based on LzmaDecode.c from the LZMA SDK 4.22 (http://www.7-zip.org/)
217 * Copyright (C) 1999-2005 Igor Pavlov
225 } __attribute__ ((packed)) ;
228 #define LZMA_BASE_SIZE 1846
229 #define LZMA_LIT_SIZE 768
231 #define LZMA_NUM_POS_BITS_MAX 4
233 #define LZMA_LEN_NUM_LOW_BITS 3
234 #define LZMA_LEN_NUM_MID_BITS 3
235 #define LZMA_LEN_NUM_HIGH_BITS 8
237 #define LZMA_LEN_CHOICE 0
238 #define LZMA_LEN_CHOICE_2 (LZMA_LEN_CHOICE + 1)
239 #define LZMA_LEN_LOW (LZMA_LEN_CHOICE_2 + 1)
240 #define LZMA_LEN_MID (LZMA_LEN_LOW \
241 + (1 << (LZMA_NUM_POS_BITS_MAX + LZMA_LEN_NUM_LOW_BITS)))
242 #define LZMA_LEN_HIGH (LZMA_LEN_MID \
243 +(1 << (LZMA_NUM_POS_BITS_MAX + LZMA_LEN_NUM_MID_BITS)))
244 #define LZMA_NUM_LEN_PROBS (LZMA_LEN_HIGH + (1 << LZMA_LEN_NUM_HIGH_BITS))
246 #define LZMA_NUM_STATES 12
247 #define LZMA_NUM_LIT_STATES 7
249 #define LZMA_START_POS_MODEL_INDEX 4
250 #define LZMA_END_POS_MODEL_INDEX 14
251 #define LZMA_NUM_FULL_DISTANCES (1 << (LZMA_END_POS_MODEL_INDEX >> 1))
253 #define LZMA_NUM_POS_SLOT_BITS 6
254 #define LZMA_NUM_LEN_TO_POS_STATES 4
256 #define LZMA_NUM_ALIGN_BITS 4
258 #define LZMA_MATCH_MIN_LEN 2
260 #define LZMA_IS_MATCH 0
261 #define LZMA_IS_REP (LZMA_IS_MATCH + (LZMA_NUM_STATES << LZMA_NUM_POS_BITS_MAX))
262 #define LZMA_IS_REP_G0 (LZMA_IS_REP + LZMA_NUM_STATES)
263 #define LZMA_IS_REP_G1 (LZMA_IS_REP_G0 + LZMA_NUM_STATES)
264 #define LZMA_IS_REP_G2 (LZMA_IS_REP_G1 + LZMA_NUM_STATES)
265 #define LZMA_IS_REP_0_LONG (LZMA_IS_REP_G2 + LZMA_NUM_STATES)
266 #define LZMA_POS_SLOT (LZMA_IS_REP_0_LONG \
267 + (LZMA_NUM_STATES << LZMA_NUM_POS_BITS_MAX))
268 #define LZMA_SPEC_POS (LZMA_POS_SLOT \
269 +(LZMA_NUM_LEN_TO_POS_STATES << LZMA_NUM_POS_SLOT_BITS))
270 #define LZMA_ALIGN (LZMA_SPEC_POS \
271 + LZMA_NUM_FULL_DISTANCES - LZMA_END_POS_MODEL_INDEX)
272 #define LZMA_LEN_CODER (LZMA_ALIGN + (1 << LZMA_NUM_ALIGN_BITS))
273 #define LZMA_REP_LEN_CODER (LZMA_LEN_CODER + LZMA_NUM_LEN_PROBS)
274 #define LZMA_LITERAL (LZMA_REP_LEN_CODER + LZMA_NUM_LEN_PROBS)
279 uint8_t previous_byte;
283 int(*flush)(void*, unsigned int);
284 struct lzma_header *header;
289 uint32_t rep0, rep1, rep2, rep3;
292 static inline size_t INIT get_pos(struct writer *wr)
295 wr->global_pos + wr->buffer_pos;
298 static inline uint8_t INIT peek_old_byte(struct writer *wr,
303 while (offs > wr->header->dict_size)
304 offs -= wr->header->dict_size;
305 pos = wr->buffer_pos - offs;
306 return wr->buffer[pos];
308 uint32_t pos = wr->buffer_pos - offs;
309 while (pos >= wr->header->dict_size)
310 pos += wr->header->dict_size;
311 return wr->buffer[pos];
316 static inline int INIT write_byte(struct writer *wr, uint8_t byte)
318 wr->buffer[wr->buffer_pos++] = wr->previous_byte = byte;
319 if (wr->flush && wr->buffer_pos == wr->header->dict_size) {
321 wr->global_pos += wr->header->dict_size;
322 if (wr->flush((char *)wr->buffer, wr->header->dict_size)
323 != wr->header->dict_size)
330 static inline int INIT copy_byte(struct writer *wr, uint32_t offs)
332 return write_byte(wr, peek_old_byte(wr, offs));
335 static inline int INIT copy_bytes(struct writer *wr,
336 uint32_t rep0, int len)
339 if (copy_byte(wr, rep0))
342 } while (len != 0 && wr->buffer_pos < wr->header->dst_size);
347 static inline int INIT process_bit0(struct writer *wr, struct rc *rc,
348 struct cstate *cst, uint16_t *p,
349 int pos_state, uint16_t *prob,
350 int lc, uint32_t literal_pos_mask) {
352 rc_update_bit_0(rc, prob);
353 prob = (p + LZMA_LITERAL +
355 * (((get_pos(wr) & literal_pos_mask) << lc)
356 + (wr->previous_byte >> (8 - lc))))
359 if (cst->state >= LZMA_NUM_LIT_STATES) {
360 int match_byte = peek_old_byte(wr, cst->rep0);
366 bit = match_byte & 0x100;
367 prob_lit = prob + 0x100 + bit + mi;
368 if (rc_get_bit(rc, prob_lit, &mi)) {
375 } while (mi < 0x100);
378 uint16_t *prob_lit = prob + mi;
379 rc_get_bit(rc, prob_lit, &mi);
383 else if (cst->state < 10)
388 return write_byte(wr, mi);
391 static inline int INIT process_bit1(struct writer *wr, struct rc *rc,
392 struct cstate *cst, uint16_t *p,
393 int pos_state, uint16_t *prob) {
399 rc_update_bit_1(rc, prob);
400 prob = p + LZMA_IS_REP + cst->state;
401 if (rc_is_bit_0(rc, prob)) {
402 rc_update_bit_0(rc, prob);
403 cst->rep3 = cst->rep2;
404 cst->rep2 = cst->rep1;
405 cst->rep1 = cst->rep0;
406 cst->state = cst->state < LZMA_NUM_LIT_STATES ? 0 : 3;
407 prob = p + LZMA_LEN_CODER;
409 rc_update_bit_1(rc, prob);
410 prob = p + LZMA_IS_REP_G0 + cst->state;
411 if (rc_is_bit_0(rc, prob)) {
412 rc_update_bit_0(rc, prob);
413 prob = (p + LZMA_IS_REP_0_LONG
415 LZMA_NUM_POS_BITS_MAX) +
417 if (rc_is_bit_0(rc, prob)) {
418 rc_update_bit_0(rc, prob);
420 cst->state = cst->state < LZMA_NUM_LIT_STATES ?
422 return copy_byte(wr, cst->rep0);
424 rc_update_bit_1(rc, prob);
429 rc_update_bit_1(rc, prob);
430 prob = p + LZMA_IS_REP_G1 + cst->state;
431 if (rc_is_bit_0(rc, prob)) {
432 rc_update_bit_0(rc, prob);
433 distance = cst->rep1;
435 rc_update_bit_1(rc, prob);
436 prob = p + LZMA_IS_REP_G2 + cst->state;
437 if (rc_is_bit_0(rc, prob)) {
438 rc_update_bit_0(rc, prob);
439 distance = cst->rep2;
441 rc_update_bit_1(rc, prob);
442 distance = cst->rep3;
443 cst->rep3 = cst->rep2;
445 cst->rep2 = cst->rep1;
447 cst->rep1 = cst->rep0;
448 cst->rep0 = distance;
450 cst->state = cst->state < LZMA_NUM_LIT_STATES ? 8 : 11;
451 prob = p + LZMA_REP_LEN_CODER;
454 prob_len = prob + LZMA_LEN_CHOICE;
455 if (rc_is_bit_0(rc, prob_len)) {
456 rc_update_bit_0(rc, prob_len);
457 prob_len = (prob + LZMA_LEN_LOW
459 LZMA_LEN_NUM_LOW_BITS));
461 num_bits = LZMA_LEN_NUM_LOW_BITS;
463 rc_update_bit_1(rc, prob_len);
464 prob_len = prob + LZMA_LEN_CHOICE_2;
465 if (rc_is_bit_0(rc, prob_len)) {
466 rc_update_bit_0(rc, prob_len);
467 prob_len = (prob + LZMA_LEN_MID
469 LZMA_LEN_NUM_MID_BITS));
470 offset = 1 << LZMA_LEN_NUM_LOW_BITS;
471 num_bits = LZMA_LEN_NUM_MID_BITS;
473 rc_update_bit_1(rc, prob_len);
474 prob_len = prob + LZMA_LEN_HIGH;
475 offset = ((1 << LZMA_LEN_NUM_LOW_BITS)
476 + (1 << LZMA_LEN_NUM_MID_BITS));
477 num_bits = LZMA_LEN_NUM_HIGH_BITS;
481 rc_bit_tree_decode(rc, prob_len, num_bits, &len);
484 if (cst->state < 4) {
487 cst->state += LZMA_NUM_LIT_STATES;
491 LZMA_NUM_LEN_TO_POS_STATES ? len :
492 LZMA_NUM_LEN_TO_POS_STATES - 1)
493 << LZMA_NUM_POS_SLOT_BITS);
494 rc_bit_tree_decode(rc, prob,
495 LZMA_NUM_POS_SLOT_BITS,
497 if (pos_slot >= LZMA_START_POS_MODEL_INDEX) {
499 num_bits = (pos_slot >> 1) - 1;
500 cst->rep0 = 2 | (pos_slot & 1);
501 if (pos_slot < LZMA_END_POS_MODEL_INDEX) {
502 cst->rep0 <<= num_bits;
503 prob = p + LZMA_SPEC_POS +
504 cst->rep0 - pos_slot - 1;
506 num_bits -= LZMA_NUM_ALIGN_BITS;
508 cst->rep0 = (cst->rep0 << 1) |
510 prob = p + LZMA_ALIGN;
511 cst->rep0 <<= LZMA_NUM_ALIGN_BITS;
512 num_bits = LZMA_NUM_ALIGN_BITS;
517 if (rc_get_bit(rc, prob + mi, &mi))
522 cst->rep0 = pos_slot;
523 if (++(cst->rep0) == 0)
525 if (cst->rep0 > wr->header->dict_size
526 || cst->rep0 > get_pos(wr))
530 len += LZMA_MATCH_MIN_LEN;
532 return copy_bytes(wr, cst->rep0, len);
537 STATIC inline int INIT unlzma(unsigned char *buf, int in_len,
538 int(*fill)(void*, unsigned int),
539 int(*flush)(void*, unsigned int),
540 unsigned char *output,
542 void(*error)(char *x)
545 struct lzma_header header;
547 uint32_t pos_state_mask;
548 uint32_t literal_pos_mask;
555 unsigned char *inbuf;
563 inbuf = malloc(LZMA_IOBUF_SIZE);
565 error("Could not allocate input bufer");
570 cst.rep0 = cst.rep1 = cst.rep2 = cst.rep3 = 1;
575 wr.previous_byte = 0;
578 rc_init(&rc, fill, inbuf, in_len);
580 for (i = 0; i < sizeof(header); i++) {
581 if (rc.ptr >= rc.buffer_end)
583 ((unsigned char *)&header)[i] = *rc.ptr++;
586 if (header.pos >= (9 * 5 * 5)) {
603 pos_state_mask = (1 << pb) - 1;
604 literal_pos_mask = (1 << lp) - 1;
606 ENDIAN_CONVERT(header.dict_size);
607 ENDIAN_CONVERT(header.dst_size);
609 if (header.dict_size == 0)
610 header.dict_size = 1;
615 wr.bufsize = MIN(header.dst_size, header.dict_size);
616 wr.buffer = large_malloc(wr.bufsize);
618 if (wr.buffer == NULL)
621 num_probs = LZMA_BASE_SIZE + (LZMA_LIT_SIZE << (lc + lp));
622 p = (uint16_t *) large_malloc(num_probs * sizeof(*p));
625 num_probs = LZMA_LITERAL + (LZMA_LIT_SIZE << (lc + lp));
626 for (i = 0; i < num_probs; i++)
627 p[i] = (1 << RC_MODEL_TOTAL_BITS) >> 1;
631 while (get_pos(&wr) < header.dst_size) {
632 int pos_state = get_pos(&wr) & pos_state_mask;
633 uint16_t *prob = p + LZMA_IS_MATCH +
634 (cst.state << LZMA_NUM_POS_BITS_MAX) + pos_state;
635 if (rc_is_bit_0(&rc, prob)) {
636 if (process_bit0(&wr, &rc, &cst, p, pos_state, prob,
637 lc, literal_pos_mask)) {
638 error("LZMA data is corrupt");
642 if (process_bit1(&wr, &rc, &cst, p, pos_state, prob)) {
643 error("LZMA data is corrupt");
649 if (rc.buffer_size <= 0)
654 *posp = rc.ptr-rc.buffer;
655 if (!wr.flush || wr.flush(wr.buffer, wr.buffer_pos) == wr.buffer_pos)
661 large_free(wr.buffer);
670 STATIC int INIT decompress(unsigned char *buf, int in_len,
671 int(*fill)(void*, unsigned int),
672 int(*flush)(void*, unsigned int),
673 unsigned char *output,
675 void(*error)(char *x)
678 return unlzma(buf, in_len - 4, fill, flush, output, posp, error);