2 /***************************************************************************
3 * lzx.c - LZX decompression routines *
4 * ------------------- *
6 * maintainer: Jed Wing <jedwin@ugcs.caltech.edu> *
7 * source: modified lzx.c from cabextract v0.5 *
8 * notes: This file was taken from cabextract v0.5, which was, *
9 * itself, a modified version of the lzx decompression code *
12 * platforms: In its current incarnation, this file has been tested on *
13 * two different Linux platforms (one, redhat-based, with a *
14 * 2.1.2 glibc and gcc 2.95.x, and the other, Debian, with *
15 * 2.2.4 glibc and both gcc 2.95.4 and gcc 3.0.2). Both were *
16 * Intel x86 compatible machines. *
17 ***************************************************************************/
19 /***************************************************************************
21 * Copyright(C) Stuart Caie *
23 * This library is free software; you can redistribute it and/or modify *
24 * it under the terms of the GNU Lesser General Public License as *
25 * published by the Free Software Foundation; either version 2.1 of the *
26 * License, or (at your option) any later version. *
28 ***************************************************************************/
36 typedef unsigned char UBYTE; /* 8 bits exactly */
37 typedef unsigned short UWORD; /* 16 bits (or more) */
38 typedef unsigned int ULONG; /* 32 bits (or more) */
39 typedef signed int LONG; /* 32 bits (or more) */
41 /* some constants defined by the LZX specification */
42 #define LZX_MIN_MATCH (2)
43 #define LZX_MAX_MATCH (257)
44 #define LZX_NUM_CHARS (256)
45 #define LZX_BLOCKTYPE_INVALID (0) /* also blocktypes 4-7 invalid */
46 #define LZX_BLOCKTYPE_VERBATIM (1)
47 #define LZX_BLOCKTYPE_ALIGNED (2)
48 #define LZX_BLOCKTYPE_UNCOMPRESSED (3)
49 #define LZX_PRETREE_NUM_ELEMENTS (20)
50 #define LZX_ALIGNED_NUM_ELEMENTS (8) /* aligned offset tree #elements */
51 #define LZX_NUM_PRIMARY_LENGTHS (7) /* this one missing from spec! */
52 #define LZX_NUM_SECONDARY_LENGTHS (249) /* length tree #elements */
54 /* LZX huffman defines: tweak tablebits as desired */
55 #define LZX_PRETREE_MAXSYMBOLS (LZX_PRETREE_NUM_ELEMENTS)
56 #define LZX_PRETREE_TABLEBITS (6)
57 #define LZX_MAINTREE_MAXSYMBOLS (LZX_NUM_CHARS + 50*8)
58 #define LZX_MAINTREE_TABLEBITS (12)
59 #define LZX_LENGTH_MAXSYMBOLS (LZX_NUM_SECONDARY_LENGTHS+1)
60 #define LZX_LENGTH_TABLEBITS (12)
61 #define LZX_ALIGNED_MAXSYMBOLS (LZX_ALIGNED_NUM_ELEMENTS)
62 #define LZX_ALIGNED_TABLEBITS (7)
64 #define LZX_LENTABLE_SAFETY (64) /* we allow length table decoding overruns */
66 #define LZX_DECLARE_TABLE(tbl) \
67 UWORD tbl##_table[(1<<LZX_##tbl##_TABLEBITS) + (LZX_##tbl##_MAXSYMBOLS<<1)];\
68 UBYTE tbl##_len [LZX_##tbl##_MAXSYMBOLS + LZX_LENTABLE_SAFETY]
72 UBYTE *window; /* the actual decoding window */
73 ULONG window_size; /* window size (32Kb through 2Mb) */
74 ULONG actual_size; /* window size when it was first allocated */
75 ULONG window_posn; /* current offset within the window */
76 ULONG R0, R1, R2; /* for the LRU offset system */
77 UWORD main_elements; /* number of main tree elements */
78 int header_read; /* have we started decoding at all yet? */
79 UWORD block_type; /* type of this block */
80 ULONG block_length; /* uncompressed length of this block */
81 ULONG block_remaining; /* uncompressed bytes still left to decode */
82 ULONG frames_read; /* the number of CFDATA blocks processed */
83 LONG intel_filesize; /* magic header value used for transform */
84 LONG intel_curpos; /* current offset in transform space */
85 int intel_started; /* have we seen any translatable data yet? */
87 LZX_DECLARE_TABLE(PRETREE);
88 LZX_DECLARE_TABLE(MAINTREE);
89 LZX_DECLARE_TABLE(LENGTH);
90 LZX_DECLARE_TABLE(ALIGNED);
95 /* Microsoft's LZX document and their implementation of the
96 * com.ms.util.cab Java package do not concur.
98 * In the LZX document, there is a table showing the correlation between
99 * window size and the number of position slots. It states that the 1MB
100 * window = 40 slots and the 2MB window = 42 slots. In the implementation,
101 * 1MB = 42 slots, 2MB = 50 slots. The actual calculation is 'find the
102 * first slot whose position base is equal to or more than the required
103 * window size'. This would explain why other tables in the document refer
104 * to 50 slots rather than 42.
106 * The constant NUM_PRIMARY_LENGTHS used in the decompression pseudocode
107 * is not defined in the specification.
109 * The LZX document does not state the uncompressed block has an
110 * uncompressed length field. Where does this length field come from, so
111 * we can know how large the block is? The implementation has it as the 24
112 * bits following after the 3 blocktype bits, before the alignment
115 * The LZX document states that aligned offset blocks have their aligned
116 * offset huffman tree AFTER the main and length trees. The implementation
117 * suggests that the aligned offset tree is BEFORE the main and length
120 * The LZX document decoding algorithm states that, in an aligned offset
121 * block, if an extra_bits value is 1, 2 or 3, then that number of bits
122 * should be read and the result added to the match offset. This is
123 * correct for 1 and 2, but not 3, where just a huffman symbol (using the
124 * aligned tree) should be read.
126 * Regarding the E8 preprocessing, the LZX document states 'No translation
127 * may be performed on the last 6 bytes of the input block'. This is
128 * correct. However, the pseudocode provided checks for the *E8 leader*
129 * up to the last 6 bytes. If the leader appears between -10 and -7 bytes
130 * from the end, this would cause the next four bytes to be modified, at
131 * least one of which would be in the last 6 bytes, which is not allowed
132 * according to the spec.
134 * The specification states that the huffman trees must always contain at
135 * least one element. However, many CAB files contain blocks where the
136 * length tree is completely empty (because there are no matches), and
137 * this is expected to succeed.
141 /* LZX uses what it calls 'position slots' to represent match offsets.
142 * What this means is that a small 'position slot' number and a small
143 * offset from that slot are encoded instead of one large offset for
145 * - position_base is an index to the position slot bases
146 * - extra_bits states how many bits of offset-from-base data is needed.
148 static const UBYTE extra_bits[51] = {
149 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6,
150 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 12, 12, 13, 13, 14, 14,
151 15, 15, 16, 16, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17, 17,
155 static const ULONG position_base[51] = {
156 0, 1, 2, 3, 4, 6, 8, 12, 16, 24, 32, 48, 64, 96, 128, 192,
157 256, 384, 512, 768, 1024, 1536, 2048, 3072, 4096, 6144, 8192, 12288, 16384, 24576, 32768, 49152,
158 65536, 98304, 131072, 196608, 262144, 393216, 524288, 655360, 786432, 917504, 1048576, 1179648, 1310720, 1441792, 1572864, 1703936,
159 1835008, 1966080, 2097152
162 struct LZXstate *LZXinit(int window)
164 struct LZXstate *pState=NULL;
165 ULONG wndsize = 1 << window;
168 /* LZX supports window sizes of 2^15 (32Kb) through 2^21 (2Mb) */
169 /* if a previously allocated window is big enough, keep it */
170 if (window < 15 || window > 21) return NULL;
172 /* allocate state and associated window */
173 pState = (struct LZXstate *)malloc(sizeof(struct LZXstate));
174 if (!(pState->window = (UBYTE *)malloc(wndsize)))
179 pState->actual_size = wndsize;
180 pState->window_size = wndsize;
182 /* calculate required position slots */
183 if (window == 20) posn_slots = 42;
184 else if (window == 21) posn_slots = 50;
185 else posn_slots = window << 1;
187 /** alternatively **/
188 /* posn_slots=i=0; while (i < wndsize) i += 1 << extra_bits[posn_slots++]; */
190 /* initialize other state */
191 pState->R0 = pState->R1 = pState->R2 = 1;
192 pState->main_elements = LZX_NUM_CHARS + (posn_slots << 3);
193 pState->header_read = 0;
194 pState->frames_read = 0;
195 pState->block_remaining = 0;
196 pState->block_type = LZX_BLOCKTYPE_INVALID;
197 pState->intel_curpos = 0;
198 pState->intel_started = 0;
199 pState->window_posn = 0;
201 /* initialise tables to 0 (because deltas will be applied to them) */
202 for (i = 0; i < LZX_MAINTREE_MAXSYMBOLS; i++) pState->MAINTREE_len[i] = 0;
203 for (i = 0; i < LZX_LENGTH_MAXSYMBOLS; i++) pState->LENGTH_len[i] = 0;
208 void LZXteardown(struct LZXstate *pState)
213 free(pState->window);
218 int LZXreset(struct LZXstate *pState)
222 pState->R0 = pState->R1 = pState->R2 = 1;
223 pState->header_read = 0;
224 pState->frames_read = 0;
225 pState->block_remaining = 0;
226 pState->block_type = LZX_BLOCKTYPE_INVALID;
227 pState->intel_curpos = 0;
228 pState->intel_started = 0;
229 pState->window_posn = 0;
231 for (i = 0; i < LZX_MAINTREE_MAXSYMBOLS + LZX_LENTABLE_SAFETY; i++) pState->MAINTREE_len[i] = 0;
232 for (i = 0; i < LZX_LENGTH_MAXSYMBOLS + LZX_LENTABLE_SAFETY; i++) pState->LENGTH_len[i] = 0;
238 /* Bitstream reading macros:
240 * INIT_BITSTREAM should be used first to set up the system
241 * READ_BITS(var,n) takes N bits from the buffer and puts them in var
243 * ENSURE_BITS(n) ensures there are at least N bits in the bit buffer
244 * PEEK_BITS(n) extracts (without removing) N bits from the bit buffer
245 * REMOVE_BITS(n) removes N bits from the bit buffer
247 * These bit access routines work by using the area beyond the MSB and the
248 * LSB as a free source of zeroes. This avoids having to mask any bits.
249 * So we have to know the bit width of the bitbuffer variable. This is
250 * sizeof(ULONG) * 8, also defined as ULONG_BITS
253 /* number of bits in ULONG. Note: This must be at multiple of 16, and at
254 * least 32 for the bitbuffer code to work (ie, it must be able to ensure
255 * up to 17 bits - that's adding 16 bits when there's one bit left, or
256 * adding 32 bits when there are no bits left. The code should work fine
257 * for machines where ULONG >= 32 bits.
259 #define ULONG_BITS (sizeof(ULONG)<<3)
261 #define INIT_BITSTREAM do { bitsleft = 0; bitbuf = 0; } while (0)
263 #define ENSURE_BITS(n) \
264 while (bitsleft < (n)) { \
265 bitbuf |= ((inpos[1]<<8)|inpos[0]) << (ULONG_BITS-16 - bitsleft); \
266 bitsleft += 16; inpos+=2; \
269 #define PEEK_BITS(n) (bitbuf >> (ULONG_BITS - (n)))
270 #define REMOVE_BITS(n) ((bitbuf <<= (n)), (bitsleft -= (n)))
272 #define READ_BITS(v,n) do { \
274 (v) = PEEK_BITS(n); \
281 #define TABLEBITS(tbl) (LZX_##tbl##_TABLEBITS)
282 #define MAXSYMBOLS(tbl) (LZX_##tbl##_MAXSYMBOLS)
283 #define SYMTABLE(tbl) (pState->tbl##_table)
284 #define LENTABLE(tbl) (pState->tbl##_len)
286 /* BUILD_TABLE(tablename) builds a huffman lookup table from code lengths.
287 * In reality, it just calls make_decode_table() with the appropriate
288 * values - they're all fixed by some #defines anyway, so there's no point
289 * writing each call out in full by hand.
291 #define BUILD_TABLE(tbl) \
292 if (make_decode_table( \
293 MAXSYMBOLS(tbl), TABLEBITS(tbl), LENTABLE(tbl), SYMTABLE(tbl) \
294 )) { return DECR_ILLEGALDATA; }
297 /* READ_HUFFSYM(tablename, var) decodes one huffman symbol from the
298 * bitstream using the stated table and puts it in var.
300 #define READ_HUFFSYM(tbl,var) do { \
302 hufftbl = SYMTABLE(tbl); \
303 if ((i = hufftbl[PEEK_BITS(TABLEBITS(tbl))]) >= MAXSYMBOLS(tbl)) { \
304 j = 1 << (ULONG_BITS - TABLEBITS(tbl)); \
306 j >>= 1; i <<= 1; i |= (bitbuf & j) ? 1 : 0; \
307 if (!j) { return DECR_ILLEGALDATA; } \
308 } while ((i = hufftbl[i]) >= MAXSYMBOLS(tbl)); \
310 j = LENTABLE(tbl)[(var) = i]; \
315 /* READ_LENGTHS(tablename, first, last) reads in code lengths for symbols
316 * first to last in the given table. The code lengths are stored in their
317 * own special LZX way.
319 #define READ_LENGTHS(tbl,first,last) do { \
320 lb.bb = bitbuf; lb.bl = bitsleft; lb.ip = inpos; \
321 if (lzx_read_lens(pState, LENTABLE(tbl),(first),(last),&lb)) { \
322 return DECR_ILLEGALDATA; \
324 bitbuf = lb.bb; bitsleft = lb.bl; inpos = lb.ip; \
328 /* make_decode_table(nsyms, nbits, length[], table[])
330 * This function was coded by David Tritscher. It builds a fast huffman
331 * decoding table out of just a canonical huffman code lengths table.
333 * nsyms = total number of symbols in this huffman tree.
334 * nbits = any symbols with a code length of nbits or less can be decoded
335 * in one lookup of the table.
336 * length = A table to get code lengths from [0 to syms-1]
337 * table = The table to fill up with decoded symbols and pointers.
339 * Returns 0 for OK or 1 for error
342 static int make_decode_table(ULONG nsyms, ULONG nbits, UBYTE *length, UWORD *table) {
345 register UBYTE bit_num = 1;
347 ULONG pos = 0; /* the current position in the decode table */
348 ULONG table_mask = 1 << nbits;
349 ULONG bit_mask = table_mask >> 1; /* don't do 0 length codes */
350 ULONG next_symbol = bit_mask; /* base of allocation for long codes */
352 /* fill entries for codes short enough for a direct mapping */
353 while (bit_num <= nbits) {
354 for (sym = 0; sym < nsyms; sym++) {
355 if (length[sym] == bit_num) {
358 if((pos += bit_mask) > table_mask) return 1; /* table overrun */
360 /* fill all possible lookups of this symbol with the symbol itself */
362 while (fill-- > 0) table[leaf++] = sym;
369 /* if there are any codes longer than nbits */
370 if (pos != table_mask) {
371 /* clear the remainder of the table */
372 for (sym = pos; sym < table_mask; sym++) table[sym] = 0;
374 /* give ourselves room for codes to grow by up to 16 more bits */
379 while (bit_num <= 16) {
380 for (sym = 0; sym < nsyms; sym++) {
381 if (length[sym] == bit_num) {
383 for (fill = 0; fill < bit_num - nbits; fill++) {
384 /* if this path hasn't been taken yet, 'allocate' two entries */
385 if (table[leaf] == 0) {
386 table[(next_symbol << 1)] = 0;
387 table[(next_symbol << 1) + 1] = 0;
388 table[leaf] = next_symbol++;
390 /* follow the path and select either left or right for next bit */
391 leaf = table[leaf] << 1;
392 if ((pos >> (15-fill)) & 1) leaf++;
396 if ((pos += bit_mask) > table_mask) return 1; /* table overflow */
405 if (pos == table_mask) return 0;
407 /* either erroneous table, or all elements are 0 - let's find out. */
408 for (sym = 0; sym < nsyms; sym++) if (length[sym]) return 1;
418 static int lzx_read_lens(struct LZXstate *pState, UBYTE *lens, ULONG first, ULONG last, struct lzx_bits *lb) {
422 register ULONG bitbuf = lb->bb;
423 register int bitsleft = lb->bl;
424 UBYTE *inpos = lb->ip;
427 for (x = 0; x < 20; x++) {
429 LENTABLE(PRETREE)[x] = y;
431 BUILD_TABLE(PRETREE);
433 for (x = first; x < last; ) {
434 READ_HUFFSYM(PRETREE, z);
436 READ_BITS(y, 4); y += 4;
437 while (y--) lens[x++] = 0;
440 READ_BITS(y, 5); y += 20;
441 while (y--) lens[x++] = 0;
444 READ_BITS(y, 1); y += 4;
445 READ_HUFFSYM(PRETREE, z);
446 z = lens[x] - z; if (z < 0) z += 17;
447 while (y--) lens[x++] = z;
450 z = lens[x] - z; if (z < 0) z += 17;
461 int LZXdecompress(struct LZXstate *pState, unsigned char *inpos, unsigned char *outpos, int inlen, int outlen) {
462 UBYTE *endinp = inpos + inlen;
463 UBYTE *window = pState->window;
464 UBYTE *runsrc, *rundest;
465 UWORD *hufftbl; /* used in READ_HUFFSYM macro as chosen decoding table */
467 ULONG window_posn = pState->window_posn;
468 ULONG window_size = pState->window_size;
469 ULONG R0 = pState->R0;
470 ULONG R1 = pState->R1;
471 ULONG R2 = pState->R2;
473 register ULONG bitbuf;
474 register int bitsleft;
475 ULONG match_offset, i,j,k; /* ijk used in READ_HUFFSYM macro */
476 struct lzx_bits lb; /* used in READ_LENGTHS macro */
478 int togo = outlen, this_run, main_element, aligned_bits;
479 int match_length, length_footer, extra, verbatim_bits;
484 /* read header if necessary */
485 if (!pState->header_read) {
487 READ_BITS(k, 1); if (k) { READ_BITS(i,16); READ_BITS(j,16); }
488 pState->intel_filesize = (i << 16) | j; /* or 0 if not encoded */
489 pState->header_read = 1;
492 /* main decoding loop */
494 /* last block finished, new block expected */
495 if (pState->block_remaining == 0) {
496 if (pState->block_type == LZX_BLOCKTYPE_UNCOMPRESSED) {
497 if (pState->block_length & 1) inpos++; /* realign bitstream to word */
501 READ_BITS(pState->block_type, 3);
504 pState->block_remaining = pState->block_length = (i << 8) | j;
506 switch (pState->block_type) {
507 case LZX_BLOCKTYPE_ALIGNED:
508 for (i = 0; i < 8; i++) { READ_BITS(j, 3); LENTABLE(ALIGNED)[i] = j; }
509 BUILD_TABLE(ALIGNED);
510 /* rest of aligned header is same as verbatim */
512 case LZX_BLOCKTYPE_VERBATIM:
513 READ_LENGTHS(MAINTREE, 0, 256);
514 READ_LENGTHS(MAINTREE, 256, pState->main_elements);
515 BUILD_TABLE(MAINTREE);
516 if (LENTABLE(MAINTREE)[0xE8] != 0) pState->intel_started = 1;
518 READ_LENGTHS(LENGTH, 0, LZX_NUM_SECONDARY_LENGTHS);
522 case LZX_BLOCKTYPE_UNCOMPRESSED:
523 pState->intel_started = 1; /* because we can't assume otherwise */
524 ENSURE_BITS(16); /* get up to 16 pad bits into the buffer */
525 if (bitsleft > 16) inpos -= 2; /* and align the bitstream! */
526 R0 = inpos[0]|(inpos[1]<<8)|(inpos[2]<<16)|(inpos[3]<<24);inpos+=4;
527 R1 = inpos[0]|(inpos[1]<<8)|(inpos[2]<<16)|(inpos[3]<<24);inpos+=4;
528 R2 = inpos[0]|(inpos[1]<<8)|(inpos[2]<<16)|(inpos[3]<<24);inpos+=4;
532 return DECR_ILLEGALDATA;
536 /* buffer exhaustion check */
537 if (inpos > endinp) {
538 /* it's possible to have a file where the next run is less than
539 * 16 bits in size. In this case, the READ_HUFFSYM() macro used
540 * in building the tables will exhaust the buffer, so we should
541 * allow for this, but not allow those accidentally read bits to
542 * be used (so we check that there are at least 16 bits
543 * remaining - in this boundary case they aren't really part of
544 * the compressed data)
546 if (inpos > (endinp+2) || bitsleft < 16) return DECR_ILLEGALDATA;
549 while ((this_run = pState->block_remaining) > 0 && togo > 0) {
550 if (this_run > togo) this_run = togo;
552 pState->block_remaining -= this_run;
554 /* apply 2^x-1 mask */
555 window_posn &= window_size - 1;
556 /* runs can't straddle the window wraparound */
557 if ((window_posn + this_run) > window_size)
558 return DECR_DATAFORMAT;
560 switch (pState->block_type) {
562 case LZX_BLOCKTYPE_VERBATIM:
563 while (this_run > 0) {
564 READ_HUFFSYM(MAINTREE, main_element);
566 if (main_element < LZX_NUM_CHARS) {
567 /* literal: 0 to LZX_NUM_CHARS-1 */
568 window[window_posn++] = main_element;
572 /* match: LZX_NUM_CHARS + ((slot<<3) | length_header (3 bits)) */
573 main_element -= LZX_NUM_CHARS;
575 match_length = main_element & LZX_NUM_PRIMARY_LENGTHS;
576 if (match_length == LZX_NUM_PRIMARY_LENGTHS) {
577 READ_HUFFSYM(LENGTH, length_footer);
578 match_length += length_footer;
580 match_length += LZX_MIN_MATCH;
582 match_offset = main_element >> 3;
584 if (match_offset > 2) {
585 /* not repeated offset */
586 if (match_offset != 3) {
587 extra = extra_bits[match_offset];
588 READ_BITS(verbatim_bits, extra);
589 match_offset = position_base[match_offset] - 2 + verbatim_bits;
595 /* update repeated offset LRU queue */
596 R2 = R1; R1 = R0; R0 = match_offset;
598 else if (match_offset == 0) {
601 else if (match_offset == 1) {
603 R1 = R0; R0 = match_offset;
605 else /* match_offset == 2 */ {
607 R2 = R0; R0 = match_offset;
610 rundest = window + window_posn;
611 this_run -= match_length;
613 /* copy any wrapped around source data */
614 if (window_posn >= match_offset) {
616 runsrc = rundest - match_offset;
618 runsrc = rundest + (window_size - match_offset);
619 copy_length = match_offset - window_posn;
620 if (copy_length < match_length) {
621 match_length -= copy_length;
622 window_posn += copy_length;
623 while (copy_length-- > 0) *rundest++ = *runsrc++;
627 window_posn += match_length;
629 /* copy match data - no worries about destination wraps */
630 while (match_length-- > 0) *rundest++ = *runsrc++;
636 case LZX_BLOCKTYPE_ALIGNED:
637 while (this_run > 0) {
638 READ_HUFFSYM(MAINTREE, main_element);
640 if (main_element < LZX_NUM_CHARS) {
641 /* literal: 0 to LZX_NUM_CHARS-1 */
642 window[window_posn++] = main_element;
646 /* match: LZX_NUM_CHARS + ((slot<<3) | length_header (3 bits)) */
647 main_element -= LZX_NUM_CHARS;
649 match_length = main_element & LZX_NUM_PRIMARY_LENGTHS;
650 if (match_length == LZX_NUM_PRIMARY_LENGTHS) {
651 READ_HUFFSYM(LENGTH, length_footer);
652 match_length += length_footer;
654 match_length += LZX_MIN_MATCH;
656 match_offset = main_element >> 3;
658 if (match_offset > 2) {
659 /* not repeated offset */
660 extra = extra_bits[match_offset];
661 match_offset = position_base[match_offset] - 2;
663 /* verbatim and aligned bits */
665 READ_BITS(verbatim_bits, extra);
666 match_offset += (verbatim_bits << 3);
667 READ_HUFFSYM(ALIGNED, aligned_bits);
668 match_offset += aligned_bits;
670 else if (extra == 3) {
671 /* aligned bits only */
672 READ_HUFFSYM(ALIGNED, aligned_bits);
673 match_offset += aligned_bits;
675 else if (extra > 0) { /* extra==1, extra==2 */
676 /* verbatim bits only */
677 READ_BITS(verbatim_bits, extra);
678 match_offset += verbatim_bits;
680 else /* extra == 0 */ {
685 /* update repeated offset LRU queue */
686 R2 = R1; R1 = R0; R0 = match_offset;
688 else if (match_offset == 0) {
691 else if (match_offset == 1) {
693 R1 = R0; R0 = match_offset;
695 else /* match_offset == 2 */ {
697 R2 = R0; R0 = match_offset;
700 rundest = window + window_posn;
701 this_run -= match_length;
703 /* copy any wrapped around source data */
704 if (window_posn >= match_offset) {
706 runsrc = rundest - match_offset;
708 runsrc = rundest + (window_size - match_offset);
709 copy_length = match_offset - window_posn;
710 if (copy_length < match_length) {
711 match_length -= copy_length;
712 window_posn += copy_length;
713 while (copy_length-- > 0) *rundest++ = *runsrc++;
717 window_posn += match_length;
719 /* copy match data - no worries about destination wraps */
720 while (match_length-- > 0) *rundest++ = *runsrc++;
726 case LZX_BLOCKTYPE_UNCOMPRESSED:
727 if ((inpos + this_run) > endinp) return DECR_ILLEGALDATA;
728 memcpy(window + window_posn, inpos, (size_t) this_run);
729 inpos += this_run; window_posn += this_run;
733 return DECR_ILLEGALDATA; /* might as well */
739 if (togo != 0) return DECR_ILLEGALDATA;
740 memcpy(outpos, window + ((!window_posn) ? window_size : window_posn) - outlen, (size_t) outlen);
742 pState->window_posn = window_posn;
747 /* intel E8 decoding */
748 if ((pState->frames_read++ < 32768) && pState->intel_filesize != 0) {
749 if (outlen <= 6 || !pState->intel_started) {
750 pState->intel_curpos += outlen;
753 UBYTE *data = outpos;
754 UBYTE *dataend = data + outlen - 10;
755 LONG curpos = pState->intel_curpos;
756 LONG filesize = pState->intel_filesize;
757 LONG abs_off, rel_off;
759 pState->intel_curpos = curpos + outlen;
761 while (data < dataend) {
762 if (*data++ != 0xE8) { curpos++; continue; }
763 abs_off = data[0] | (data[1]<<8) | (data[2]<<16) | (data[3]<<24);
764 if ((abs_off >= -curpos) && (abs_off < filesize)) {
765 rel_off = (abs_off >= 0) ? abs_off - curpos : abs_off + filesize;
766 data[0] = (UBYTE) rel_off;
767 data[1] = (UBYTE) (rel_off >> 8);
768 data[2] = (UBYTE) (rel_off >> 16);
769 data[3] = (UBYTE) (rel_off >> 24);
779 #ifdef LZX_CHM_TESTDRIVER
780 int main(int c, char **v)
783 struct LZXstate state;
792 fout = fopen(v[2], "wb");
795 fin = fopen(v[i], "rb");
796 ilen = fread(ibuf, 1, 16384, fin);
797 status = LZXdecompress(&state, ibuf, obuf, ilen, 32768);
802 fwrite(obuf, 1, 32768, fout);
804 case DECR_DATAFORMAT:
805 printf("bad format\n");
807 case DECR_ILLEGALDATA:
808 printf("illegal data\n");
811 printf("no memory\n");