1 /* vi: set sw=4 ts=4: */
3 * Small lzma deflate implementation.
4 * Copyright (C) 2006 Aurelien Jacobs <aurel@gnuage.org>
6 * Based on LzmaDecode.c from the LZMA SDK 4.22 (http://www.7-zip.org/)
7 * Copyright (C) 1999-2005 Igor Pavlov
9 * Licensed under GPLv2 or later, see file LICENSE in this source tree.
12 #include "bb_archive.h"
14 #if ENABLE_FEATURE_LZMA_FAST
15 # define speed_inline ALWAYS_INLINE
19 # define size_inline ALWAYS_INLINE
27 /* Was keeping rc on stack in unlzma and separately allocating buffer,
28 * but with "buffer 'attached to' allocated rc" code is smaller: */
29 /* uint8_t *buffer; */
30 #define RC_BUFFER ((uint8_t*)(rc+1))
34 /* Had provisions for variable buffer, but we don't need it here */
35 /* int buffer_size; */
36 #define RC_BUFFER_SIZE 0x10000
43 #define RC_TOP_BITS 24
44 #define RC_MOVE_BITS 5
45 #define RC_MODEL_TOTAL_BITS 11
48 /* Called twice: once at startup (LZMA_FAST only) and once in rc_normalize() */
49 static size_inline void rc_read(rc_t *rc)
51 int buffer_size = safe_read(rc->fd, RC_BUFFER, RC_BUFFER_SIZE);
52 //TODO: return -1 instead
53 //This will make unlzma delete broken unpacked file on unpack errors
55 bb_error_msg_and_die("unexpected EOF");
57 rc->buffer_end = RC_BUFFER + buffer_size;
60 /* Called twice, but one callsite is in speed_inline'd rc_is_bit_1() */
61 static void rc_do_normalize(rc_t *rc)
63 if (rc->ptr >= rc->buffer_end)
66 rc->code = (rc->code << 8) | *rc->ptr++;
70 static ALWAYS_INLINE rc_t* rc_init(int fd) /*, int buffer_size) */
75 rc = xzalloc(sizeof(*rc) + RC_BUFFER_SIZE);
78 /* rc->ptr = rc->buffer_end; */
80 for (i = 0; i < 5; i++) {
81 #if ENABLE_FEATURE_LZMA_FAST
82 if (rc->ptr >= rc->buffer_end)
84 rc->code = (rc->code << 8) | *rc->ptr++;
89 rc->range = 0xFFFFFFFF;
94 static ALWAYS_INLINE void rc_free(rc_t *rc)
99 static ALWAYS_INLINE void rc_normalize(rc_t *rc)
101 if (rc->range < (1 << RC_TOP_BITS)) {
106 /* rc_is_bit_1 is called 9 times */
107 static speed_inline int rc_is_bit_1(rc_t *rc, uint16_t *p)
110 rc->bound = *p * (rc->range >> RC_MODEL_TOTAL_BITS);
111 if (rc->code < rc->bound) {
112 rc->range = rc->bound;
113 *p += ((1 << RC_MODEL_TOTAL_BITS) - *p) >> RC_MOVE_BITS;
116 rc->range -= rc->bound;
117 rc->code -= rc->bound;
118 *p -= *p >> RC_MOVE_BITS;
122 /* Called 4 times in unlzma loop */
123 static speed_inline int rc_get_bit(rc_t *rc, uint16_t *p, int *symbol)
125 int ret = rc_is_bit_1(rc, p);
126 *symbol = *symbol * 2 + ret;
131 static ALWAYS_INLINE int rc_direct_bit(rc_t *rc)
135 if (rc->code >= rc->range) {
136 rc->code -= rc->range;
143 static speed_inline void
144 rc_bit_tree_decode(rc_t *rc, uint16_t *p, int num_levels, int *symbol)
150 rc_get_bit(rc, p + *symbol, symbol);
151 *symbol -= 1 << num_levels;
159 } PACKED lzma_header_t;
162 /* #defines will force compiler to compute/optimize each one with each usage.
163 * Have heart and use enum instead. */
165 LZMA_BASE_SIZE = 1846,
168 LZMA_NUM_POS_BITS_MAX = 4,
170 LZMA_LEN_NUM_LOW_BITS = 3,
171 LZMA_LEN_NUM_MID_BITS = 3,
172 LZMA_LEN_NUM_HIGH_BITS = 8,
175 LZMA_LEN_CHOICE_2 = (LZMA_LEN_CHOICE + 1),
176 LZMA_LEN_LOW = (LZMA_LEN_CHOICE_2 + 1),
177 LZMA_LEN_MID = (LZMA_LEN_LOW \
178 + (1 << (LZMA_NUM_POS_BITS_MAX + LZMA_LEN_NUM_LOW_BITS))),
179 LZMA_LEN_HIGH = (LZMA_LEN_MID \
180 + (1 << (LZMA_NUM_POS_BITS_MAX + LZMA_LEN_NUM_MID_BITS))),
181 LZMA_NUM_LEN_PROBS = (LZMA_LEN_HIGH + (1 << LZMA_LEN_NUM_HIGH_BITS)),
183 LZMA_NUM_STATES = 12,
184 LZMA_NUM_LIT_STATES = 7,
186 LZMA_START_POS_MODEL_INDEX = 4,
187 LZMA_END_POS_MODEL_INDEX = 14,
188 LZMA_NUM_FULL_DISTANCES = (1 << (LZMA_END_POS_MODEL_INDEX >> 1)),
190 LZMA_NUM_POS_SLOT_BITS = 6,
191 LZMA_NUM_LEN_TO_POS_STATES = 4,
193 LZMA_NUM_ALIGN_BITS = 4,
195 LZMA_MATCH_MIN_LEN = 2,
198 LZMA_IS_REP = (LZMA_IS_MATCH + (LZMA_NUM_STATES << LZMA_NUM_POS_BITS_MAX)),
199 LZMA_IS_REP_G0 = (LZMA_IS_REP + LZMA_NUM_STATES),
200 LZMA_IS_REP_G1 = (LZMA_IS_REP_G0 + LZMA_NUM_STATES),
201 LZMA_IS_REP_G2 = (LZMA_IS_REP_G1 + LZMA_NUM_STATES),
202 LZMA_IS_REP_0_LONG = (LZMA_IS_REP_G2 + LZMA_NUM_STATES),
203 LZMA_POS_SLOT = (LZMA_IS_REP_0_LONG \
204 + (LZMA_NUM_STATES << LZMA_NUM_POS_BITS_MAX)),
205 LZMA_SPEC_POS = (LZMA_POS_SLOT \
206 + (LZMA_NUM_LEN_TO_POS_STATES << LZMA_NUM_POS_SLOT_BITS)),
207 LZMA_ALIGN = (LZMA_SPEC_POS \
208 + LZMA_NUM_FULL_DISTANCES - LZMA_END_POS_MODEL_INDEX),
209 LZMA_LEN_CODER = (LZMA_ALIGN + (1 << LZMA_NUM_ALIGN_BITS)),
210 LZMA_REP_LEN_CODER = (LZMA_LEN_CODER + LZMA_NUM_LEN_PROBS),
211 LZMA_LITERAL = (LZMA_REP_LEN_CODER + LZMA_NUM_LEN_PROBS),
215 IF_DESKTOP(long long) int FAST_FUNC
216 unpack_lzma_stream(transformer_aux_data_t *aux UNUSED_PARAM, int src_fd, int dst_fd)
218 IF_DESKTOP(long long total_written = 0;)
219 lzma_header_t header;
221 uint32_t pos_state_mask;
222 uint32_t literal_pos_mask;
229 uint8_t previous_byte = 0;
230 size_t buffer_pos = 0, global_pos = 0;
233 uint32_t rep0 = 1, rep1 = 1, rep2 = 1, rep3 = 1;
235 if (full_read(src_fd, &header, sizeof(header)) != sizeof(header)
236 || header.pos >= (9 * 5 * 5)
238 bb_error_msg("bad lzma header");
246 pos_state_mask = (1 << pb) - 1;
247 literal_pos_mask = (1 << lp) - 1;
249 header.dict_size = SWAP_LE32(header.dict_size);
250 header.dst_size = SWAP_LE64(header.dst_size);
252 if (header.dict_size == 0)
255 buffer = xmalloc(MIN(header.dst_size, header.dict_size));
257 num_probs = LZMA_BASE_SIZE + (LZMA_LIT_SIZE << (lc + lp));
258 p = xmalloc(num_probs * sizeof(*p));
259 num_probs += LZMA_LITERAL - LZMA_BASE_SIZE;
260 for (i = 0; i < num_probs; i++)
261 p[i] = (1 << RC_MODEL_TOTAL_BITS) >> 1;
263 rc = rc_init(src_fd); /*, RC_BUFFER_SIZE); */
265 while (global_pos + buffer_pos < header.dst_size) {
266 int pos_state = (buffer_pos + global_pos) & pos_state_mask;
267 uint16_t *prob = p + LZMA_IS_MATCH + (state << LZMA_NUM_POS_BITS_MAX) + pos_state;
269 if (!rc_is_bit_1(rc, prob)) {
270 static const char next_state[LZMA_NUM_STATES] =
271 { 0, 0, 0, 0, 1, 2, 3, 4, 5, 6, 4, 5 };
274 prob = (p + LZMA_LITERAL
275 + (LZMA_LIT_SIZE * ((((buffer_pos + global_pos) & literal_pos_mask) << lc)
276 + (previous_byte >> (8 - lc))
281 if (state >= LZMA_NUM_LIT_STATES) {
283 uint32_t pos = buffer_pos - rep0;
285 while (pos >= header.dict_size)
286 pos += header.dict_size;
287 match_byte = buffer[pos];
292 bit = match_byte & 0x100;
293 bit ^= (rc_get_bit(rc, prob + 0x100 + bit + mi, &mi) << 8); /* 0x100 or 0 */
296 } while (mi < 0x100);
299 rc_get_bit(rc, prob + mi, &mi);
302 state = next_state[state];
304 previous_byte = (uint8_t) mi;
305 #if ENABLE_FEATURE_LZMA_FAST
307 buffer[buffer_pos++] = previous_byte;
308 if (buffer_pos == header.dict_size) {
310 global_pos += header.dict_size;
311 if (full_write(dst_fd, buffer, header.dict_size) != (ssize_t)header.dict_size)
313 IF_DESKTOP(total_written += header.dict_size;)
322 #define prob_len prob2
324 prob2 = p + LZMA_IS_REP + state;
325 if (!rc_is_bit_1(rc, prob2)) {
329 state = state < LZMA_NUM_LIT_STATES ? 0 : 3;
330 prob2 = p + LZMA_LEN_CODER;
332 prob2 += LZMA_IS_REP_G0 - LZMA_IS_REP;
333 if (!rc_is_bit_1(rc, prob2)) {
334 prob2 = (p + LZMA_IS_REP_0_LONG
335 + (state << LZMA_NUM_POS_BITS_MAX)
338 if (!rc_is_bit_1(rc, prob2)) {
339 #if ENABLE_FEATURE_LZMA_FAST
340 uint32_t pos = buffer_pos - rep0;
341 state = state < LZMA_NUM_LIT_STATES ? 9 : 11;
342 while (pos >= header.dict_size)
343 pos += header.dict_size;
344 previous_byte = buffer[pos];
347 state = state < LZMA_NUM_LIT_STATES ? 9 : 11;
355 prob2 += LZMA_IS_REP_G1 - LZMA_IS_REP_G0;
357 if (rc_is_bit_1(rc, prob2)) {
358 prob2 += LZMA_IS_REP_G2 - LZMA_IS_REP_G1;
360 if (rc_is_bit_1(rc, prob2)) {
369 state = state < LZMA_NUM_LIT_STATES ? 8 : 11;
370 prob2 = p + LZMA_REP_LEN_CODER;
373 prob_len = prob2 + LZMA_LEN_CHOICE;
374 num_bits = LZMA_LEN_NUM_LOW_BITS;
375 if (!rc_is_bit_1(rc, prob_len)) {
376 prob_len += LZMA_LEN_LOW - LZMA_LEN_CHOICE
377 + (pos_state << LZMA_LEN_NUM_LOW_BITS);
380 prob_len += LZMA_LEN_CHOICE_2 - LZMA_LEN_CHOICE;
381 if (!rc_is_bit_1(rc, prob_len)) {
382 prob_len += LZMA_LEN_MID - LZMA_LEN_CHOICE_2
383 + (pos_state << LZMA_LEN_NUM_MID_BITS);
384 offset = 1 << LZMA_LEN_NUM_LOW_BITS;
385 num_bits += LZMA_LEN_NUM_MID_BITS - LZMA_LEN_NUM_LOW_BITS;
387 prob_len += LZMA_LEN_HIGH - LZMA_LEN_CHOICE_2;
388 offset = ((1 << LZMA_LEN_NUM_LOW_BITS)
389 + (1 << LZMA_LEN_NUM_MID_BITS));
390 num_bits += LZMA_LEN_NUM_HIGH_BITS - LZMA_LEN_NUM_LOW_BITS;
393 rc_bit_tree_decode(rc, prob_len, num_bits, &len);
400 state += LZMA_NUM_LIT_STATES;
401 prob3 = p + LZMA_POS_SLOT +
402 ((len < LZMA_NUM_LEN_TO_POS_STATES ? len :
403 LZMA_NUM_LEN_TO_POS_STATES - 1)
404 << LZMA_NUM_POS_SLOT_BITS);
405 rc_bit_tree_decode(rc, prob3,
406 LZMA_NUM_POS_SLOT_BITS, &pos_slot);
408 if (pos_slot >= LZMA_START_POS_MODEL_INDEX) {
409 int i2, mi2, num_bits2 = (pos_slot >> 1) - 1;
410 rep0 = 2 | (pos_slot & 1);
411 if (pos_slot < LZMA_END_POS_MODEL_INDEX) {
413 prob3 = p + LZMA_SPEC_POS + rep0 - pos_slot - 1;
415 for (; num_bits2 != LZMA_NUM_ALIGN_BITS; num_bits2--)
416 rep0 = (rep0 << 1) | rc_direct_bit(rc);
417 rep0 <<= LZMA_NUM_ALIGN_BITS;
418 prob3 = p + LZMA_ALIGN;
422 while (num_bits2--) {
423 if (rc_get_bit(rc, prob3 + mi2, &mi2))
432 len += LZMA_MATCH_MIN_LEN;
433 IF_NOT_FEATURE_LZMA_FAST(string:)
435 uint32_t pos = buffer_pos - rep0;
436 while (pos >= header.dict_size)
437 pos += header.dict_size;
438 previous_byte = buffer[pos];
439 IF_NOT_FEATURE_LZMA_FAST(one_byte2:)
440 buffer[buffer_pos++] = previous_byte;
441 if (buffer_pos == header.dict_size) {
443 global_pos += header.dict_size;
444 if (full_write(dst_fd, buffer, header.dict_size) != (ssize_t)header.dict_size)
446 IF_DESKTOP(total_written += header.dict_size;)
449 } while (len != 0 && buffer_pos < header.dst_size);
454 IF_NOT_DESKTOP(int total_written = 0; /* success */)
455 IF_DESKTOP(total_written += buffer_pos;)
456 if (full_write(dst_fd, buffer, buffer_pos) != (ssize_t)buffer_pos) {
458 total_written = -1; /* failure */
463 return total_written;