xref: /linux/lib/decompress_bunzip2.c (revision a1ff5a7d78a036d6c2178ee5acd6ba4946243800)
1 /*	Small bzip2 deflate implementation, by Rob Landley (rob@landley.net).
2 
3 	Based on bzip2 decompression code by Julian R Seward (jseward@acm.org),
4 	which also acknowledges contributions by Mike Burrows, David Wheeler,
5 	Peter Fenwick, Alistair Moffat, Radford Neal, Ian H. Witten,
6 	Robert Sedgewick, and Jon L. Bentley.
7 
8 	This code is licensed under the LGPLv2:
9 		LGPL (http://www.gnu.org/copyleft/lgpl.html
10 */
11 
12 /*
13 	Size and speed optimizations by Manuel Novoa III  (mjn3@codepoet.org).
14 
15 	More efficient reading of Huffman codes, a streamlined read_bunzip()
16 	function, and various other tweaks.  In (limited) tests, approximately
17 	20% faster than bzcat on x86 and about 10% faster on arm.
18 
19 	Note that about 2/3 of the time is spent in read_unzip() reversing
20 	the Burrows-Wheeler transformation.  Much of that time is delay
21 	resulting from cache misses.
22 
23 	I would ask that anyone benefiting from this work, especially those
24 	using it in commercial products, consider making a donation to my local
25 	non-profit hospice organization in the name of the woman I loved, who
26 	passed away Feb. 12, 2003.
27 
28 		In memory of Toni W. Hagan
29 
30 		Hospice of Acadiana, Inc.
31 		2600 Johnston St., Suite 200
32 		Lafayette, LA 70503-3240
33 
34 		Phone (337) 232-1234 or 1-800-738-2226
35 		Fax   (337) 232-1297
36 
37 		https://www.hospiceacadiana.com/
38 
39 	Manuel
40  */
41 
42 /*
43 	Made it fit for running in Linux Kernel by Alain Knaff (alain@knaff.lu)
44 */
45 
46 
47 #ifdef STATIC
48 #define PREBOOT
49 #else
50 #include <linux/decompress/bunzip2.h>
51 #endif /* STATIC */
52 
53 #include <linux/decompress/mm.h>
54 #include <linux/crc32poly.h>
55 
56 #ifndef INT_MAX
57 #define INT_MAX 0x7fffffff
58 #endif
59 
60 /* Constants for Huffman coding */
61 #define MAX_GROUPS		6
62 #define GROUP_SIZE   		50	/* 64 would have been more efficient */
63 #define MAX_HUFCODE_BITS 	20	/* Longest Huffman code allowed */
64 #define MAX_SYMBOLS 		258	/* 256 literals + RUNA + RUNB */
65 #define SYMBOL_RUNA		0
66 #define SYMBOL_RUNB		1
67 
68 /* Status return values */
69 #define RETVAL_OK			0
70 #define RETVAL_LAST_BLOCK		(-1)
71 #define RETVAL_NOT_BZIP_DATA		(-2)
72 #define RETVAL_UNEXPECTED_INPUT_EOF	(-3)
73 #define RETVAL_UNEXPECTED_OUTPUT_EOF	(-4)
74 #define RETVAL_DATA_ERROR		(-5)
75 #define RETVAL_OUT_OF_MEMORY		(-6)
76 #define RETVAL_OBSOLETE_INPUT		(-7)
77 
78 /* Other housekeeping constants */
79 #define BZIP2_IOBUF_SIZE		4096
80 
81 /* This is what we know about each Huffman coding group */
82 struct group_data {
83 	/* We have an extra slot at the end of limit[] for a sentinel value. */
84 	int limit[MAX_HUFCODE_BITS+1];
85 	int base[MAX_HUFCODE_BITS];
86 	int permute[MAX_SYMBOLS];
87 	int minLen, maxLen;
88 };
89 
90 /* Structure holding all the housekeeping data, including IO buffers and
91    memory that persists between calls to bunzip */
92 struct bunzip_data {
93 	/* State for interrupting output loop */
94 	int writeCopies, writePos, writeRunCountdown, writeCount, writeCurrent;
95 	/* I/O tracking data (file handles, buffers, positions, etc.) */
96 	long (*fill)(void*, unsigned long);
97 	long inbufCount, inbufPos /*, outbufPos*/;
98 	unsigned char *inbuf /*,*outbuf*/;
99 	unsigned int inbufBitCount, inbufBits;
100 	/* The CRC values stored in the block header and calculated from the
101 	data */
102 	unsigned int crc32Table[256], headerCRC, totalCRC, writeCRC;
103 	/* Intermediate buffer and its size (in bytes) */
104 	unsigned int *dbuf, dbufSize;
105 	/* These things are a bit too big to go on the stack */
106 	unsigned char selectors[32768];		/* nSelectors = 15 bits */
107 	struct group_data groups[MAX_GROUPS];	/* Huffman coding tables */
108 	int io_error;			/* non-zero if we have IO error */
109 	int byteCount[256];
110 	unsigned char symToByte[256], mtfSymbol[256];
111 };
112 
113 
114 /* Return the next nnn bits of input.  All reads from the compressed input
115    are done through this function.  All reads are big endian */
get_bits(struct bunzip_data * bd,char bits_wanted)116 static unsigned int INIT get_bits(struct bunzip_data *bd, char bits_wanted)
117 {
118 	unsigned int bits = 0;
119 
120 	/* If we need to get more data from the byte buffer, do so.
121 	   (Loop getting one byte at a time to enforce endianness and avoid
122 	   unaligned access.) */
123 	while (bd->inbufBitCount < bits_wanted) {
124 		/* If we need to read more data from file into byte buffer, do
125 		   so */
126 		if (bd->inbufPos == bd->inbufCount) {
127 			if (bd->io_error)
128 				return 0;
129 			bd->inbufCount = bd->fill(bd->inbuf, BZIP2_IOBUF_SIZE);
130 			if (bd->inbufCount <= 0) {
131 				bd->io_error = RETVAL_UNEXPECTED_INPUT_EOF;
132 				return 0;
133 			}
134 			bd->inbufPos = 0;
135 		}
136 		/* Avoid 32-bit overflow (dump bit buffer to top of output) */
137 		if (bd->inbufBitCount >= 24) {
138 			bits = bd->inbufBits&((1 << bd->inbufBitCount)-1);
139 			bits_wanted -= bd->inbufBitCount;
140 			bits <<= bits_wanted;
141 			bd->inbufBitCount = 0;
142 		}
143 		/* Grab next 8 bits of input from buffer. */
144 		bd->inbufBits = (bd->inbufBits << 8)|bd->inbuf[bd->inbufPos++];
145 		bd->inbufBitCount += 8;
146 	}
147 	/* Calculate result */
148 	bd->inbufBitCount -= bits_wanted;
149 	bits |= (bd->inbufBits >> bd->inbufBitCount)&((1 << bits_wanted)-1);
150 
151 	return bits;
152 }
153 
154 /* Unpacks the next block and sets up for the inverse burrows-wheeler step. */
155 
get_next_block(struct bunzip_data * bd)156 static int INIT get_next_block(struct bunzip_data *bd)
157 {
158 	struct group_data *hufGroup = NULL;
159 	int *base = NULL;
160 	int *limit = NULL;
161 	int dbufCount, nextSym, dbufSize, groupCount, selector,
162 		i, j, k, t, runPos, symCount, symTotal, nSelectors, *byteCount;
163 	unsigned char uc, *symToByte, *mtfSymbol, *selectors;
164 	unsigned int *dbuf, origPtr;
165 
166 	dbuf = bd->dbuf;
167 	dbufSize = bd->dbufSize;
168 	selectors = bd->selectors;
169 	byteCount = bd->byteCount;
170 	symToByte = bd->symToByte;
171 	mtfSymbol = bd->mtfSymbol;
172 
173 	/* Read in header signature and CRC, then validate signature.
174 	   (last block signature means CRC is for whole file, return now) */
175 	i = get_bits(bd, 24);
176 	j = get_bits(bd, 24);
177 	bd->headerCRC = get_bits(bd, 32);
178 	if ((i == 0x177245) && (j == 0x385090))
179 		return RETVAL_LAST_BLOCK;
180 	if ((i != 0x314159) || (j != 0x265359))
181 		return RETVAL_NOT_BZIP_DATA;
182 	/* We can add support for blockRandomised if anybody complains.
183 	   There was some code for this in busybox 1.0.0-pre3, but nobody ever
184 	   noticed that it didn't actually work. */
185 	if (get_bits(bd, 1))
186 		return RETVAL_OBSOLETE_INPUT;
187 	origPtr = get_bits(bd, 24);
188 	if (origPtr >= dbufSize)
189 		return RETVAL_DATA_ERROR;
190 	/* mapping table: if some byte values are never used (encoding things
191 	   like ascii text), the compression code removes the gaps to have fewer
192 	   symbols to deal with, and writes a sparse bitfield indicating which
193 	   values were present.  We make a translation table to convert the
194 	   symbols back to the corresponding bytes. */
195 	t = get_bits(bd, 16);
196 	symTotal = 0;
197 	for (i = 0; i < 16; i++) {
198 		if (t&(1 << (15-i))) {
199 			k = get_bits(bd, 16);
200 			for (j = 0; j < 16; j++)
201 				if (k&(1 << (15-j)))
202 					symToByte[symTotal++] = (16*i)+j;
203 		}
204 	}
205 	/* How many different Huffman coding groups does this block use? */
206 	groupCount = get_bits(bd, 3);
207 	if (groupCount < 2 || groupCount > MAX_GROUPS)
208 		return RETVAL_DATA_ERROR;
209 	/* nSelectors: Every GROUP_SIZE many symbols we select a new
210 	   Huffman coding group.  Read in the group selector list,
211 	   which is stored as MTF encoded bit runs.  (MTF = Move To
212 	   Front, as each value is used it's moved to the start of the
213 	   list.) */
214 	nSelectors = get_bits(bd, 15);
215 	if (!nSelectors)
216 		return RETVAL_DATA_ERROR;
217 	for (i = 0; i < groupCount; i++)
218 		mtfSymbol[i] = i;
219 	for (i = 0; i < nSelectors; i++) {
220 		/* Get next value */
221 		for (j = 0; get_bits(bd, 1); j++)
222 			if (j >= groupCount)
223 				return RETVAL_DATA_ERROR;
224 		/* Decode MTF to get the next selector */
225 		uc = mtfSymbol[j];
226 		for (; j; j--)
227 			mtfSymbol[j] = mtfSymbol[j-1];
228 		mtfSymbol[0] = selectors[i] = uc;
229 	}
230 	/* Read the Huffman coding tables for each group, which code
231 	   for symTotal literal symbols, plus two run symbols (RUNA,
232 	   RUNB) */
233 	symCount = symTotal+2;
234 	for (j = 0; j < groupCount; j++) {
235 		unsigned char length[MAX_SYMBOLS];
236 		unsigned short temp[MAX_HUFCODE_BITS+1];
237 		int	minLen,	maxLen, pp;
238 		/* Read Huffman code lengths for each symbol.  They're
239 		   stored in a way similar to mtf; record a starting
240 		   value for the first symbol, and an offset from the
241 		   previous value for everys symbol after that.
242 		   (Subtracting 1 before the loop and then adding it
243 		   back at the end is an optimization that makes the
244 		   test inside the loop simpler: symbol length 0
245 		   becomes negative, so an unsigned inequality catches
246 		   it.) */
247 		t = get_bits(bd, 5)-1;
248 		for (i = 0; i < symCount; i++) {
249 			for (;;) {
250 				if (((unsigned)t) > (MAX_HUFCODE_BITS-1))
251 					return RETVAL_DATA_ERROR;
252 
253 				/* If first bit is 0, stop.  Else
254 				   second bit indicates whether to
255 				   increment or decrement the value.
256 				   Optimization: grab 2 bits and unget
257 				   the second if the first was 0. */
258 
259 				k = get_bits(bd, 2);
260 				if (k < 2) {
261 					bd->inbufBitCount++;
262 					break;
263 				}
264 				/* Add one if second bit 1, else
265 				 * subtract 1.  Avoids if/else */
266 				t += (((k+1)&2)-1);
267 			}
268 			/* Correct for the initial -1, to get the
269 			 * final symbol length */
270 			length[i] = t+1;
271 		}
272 		/* Find largest and smallest lengths in this group */
273 		minLen = maxLen = length[0];
274 
275 		for (i = 1; i < symCount; i++) {
276 			if (length[i] > maxLen)
277 				maxLen = length[i];
278 			else if (length[i] < minLen)
279 				minLen = length[i];
280 		}
281 
282 		/* Calculate permute[], base[], and limit[] tables from
283 		 * length[].
284 		 *
285 		 * permute[] is the lookup table for converting
286 		 * Huffman coded symbols into decoded symbols.  base[]
287 		 * is the amount to subtract from the value of a
288 		 * Huffman symbol of a given length when using
289 		 * permute[].
290 		 *
291 		 * limit[] indicates the largest numerical value a
292 		 * symbol with a given number of bits can have.  This
293 		 * is how the Huffman codes can vary in length: each
294 		 * code with a value > limit[length] needs another
295 		 * bit.
296 		 */
297 		hufGroup = bd->groups+j;
298 		hufGroup->minLen = minLen;
299 		hufGroup->maxLen = maxLen;
300 		/* Note that minLen can't be smaller than 1, so we
301 		   adjust the base and limit array pointers so we're
302 		   not always wasting the first entry.  We do this
303 		   again when using them (during symbol decoding).*/
304 		base = hufGroup->base-1;
305 		limit = hufGroup->limit-1;
306 		/* Calculate permute[].  Concurrently, initialize
307 		 * temp[] and limit[]. */
308 		pp = 0;
309 		for (i = minLen; i <= maxLen; i++) {
310 			temp[i] = limit[i] = 0;
311 			for (t = 0; t < symCount; t++)
312 				if (length[t] == i)
313 					hufGroup->permute[pp++] = t;
314 		}
315 		/* Count symbols coded for at each bit length */
316 		for (i = 0; i < symCount; i++)
317 			temp[length[i]]++;
318 		/* Calculate limit[] (the largest symbol-coding value
319 		 *at each bit length, which is (previous limit <<
320 		 *1)+symbols at this level), and base[] (number of
321 		 *symbols to ignore at each bit length, which is limit
322 		 *minus the cumulative count of symbols coded for
323 		 *already). */
324 		pp = t = 0;
325 		for (i = minLen; i < maxLen; i++) {
326 			pp += temp[i];
327 			/* We read the largest possible symbol size
328 			   and then unget bits after determining how
329 			   many we need, and those extra bits could be
330 			   set to anything.  (They're noise from
331 			   future symbols.)  At each level we're
332 			   really only interested in the first few
333 			   bits, so here we set all the trailing
334 			   to-be-ignored bits to 1 so they don't
335 			   affect the value > limit[length]
336 			   comparison. */
337 			limit[i] = (pp << (maxLen - i)) - 1;
338 			pp <<= 1;
339 			base[i+1] = pp-(t += temp[i]);
340 		}
341 		limit[maxLen+1] = INT_MAX; /* Sentinel value for
342 					    * reading next sym. */
343 		limit[maxLen] = pp+temp[maxLen]-1;
344 		base[minLen] = 0;
345 	}
346 	/* We've finished reading and digesting the block header.  Now
347 	   read this block's Huffman coded symbols from the file and
348 	   undo the Huffman coding and run length encoding, saving the
349 	   result into dbuf[dbufCount++] = uc */
350 
351 	/* Initialize symbol occurrence counters and symbol Move To
352 	 * Front table */
353 	for (i = 0; i < 256; i++) {
354 		byteCount[i] = 0;
355 		mtfSymbol[i] = (unsigned char)i;
356 	}
357 	/* Loop through compressed symbols. */
358 	runPos = dbufCount = symCount = selector = 0;
359 	for (;;) {
360 		/* Determine which Huffman coding group to use. */
361 		if (!(symCount--)) {
362 			symCount = GROUP_SIZE-1;
363 			if (selector >= nSelectors)
364 				return RETVAL_DATA_ERROR;
365 			hufGroup = bd->groups+selectors[selector++];
366 			base = hufGroup->base-1;
367 			limit = hufGroup->limit-1;
368 		}
369 		/* Read next Huffman-coded symbol. */
370 		/* Note: It is far cheaper to read maxLen bits and
371 		   back up than it is to read minLen bits and then an
372 		   additional bit at a time, testing as we go.
373 		   Because there is a trailing last block (with file
374 		   CRC), there is no danger of the overread causing an
375 		   unexpected EOF for a valid compressed file.  As a
376 		   further optimization, we do the read inline
377 		   (falling back to a call to get_bits if the buffer
378 		   runs dry).  The following (up to got_huff_bits:) is
379 		   equivalent to j = get_bits(bd, hufGroup->maxLen);
380 		 */
381 		while (bd->inbufBitCount < hufGroup->maxLen) {
382 			if (bd->inbufPos == bd->inbufCount) {
383 				j = get_bits(bd, hufGroup->maxLen);
384 				goto got_huff_bits;
385 			}
386 			bd->inbufBits =
387 				(bd->inbufBits << 8)|bd->inbuf[bd->inbufPos++];
388 			bd->inbufBitCount += 8;
389 		}
390 		bd->inbufBitCount -= hufGroup->maxLen;
391 		j = (bd->inbufBits >> bd->inbufBitCount)&
392 			((1 << hufGroup->maxLen)-1);
393 got_huff_bits:
394 		/* Figure how many bits are in next symbol and
395 		 * unget extras */
396 		i = hufGroup->minLen;
397 		while (j > limit[i])
398 			++i;
399 		bd->inbufBitCount += (hufGroup->maxLen - i);
400 		/* Huffman decode value to get nextSym (with bounds checking) */
401 		if ((i > hufGroup->maxLen)
402 			|| (((unsigned)(j = (j>>(hufGroup->maxLen-i))-base[i]))
403 				>= MAX_SYMBOLS))
404 			return RETVAL_DATA_ERROR;
405 		nextSym = hufGroup->permute[j];
406 		/* We have now decoded the symbol, which indicates
407 		   either a new literal byte, or a repeated run of the
408 		   most recent literal byte.  First, check if nextSym
409 		   indicates a repeated run, and if so loop collecting
410 		   how many times to repeat the last literal. */
411 		if (((unsigned)nextSym) <= SYMBOL_RUNB) { /* RUNA or RUNB */
412 			/* If this is the start of a new run, zero out
413 			 * counter */
414 			if (!runPos) {
415 				runPos = 1;
416 				t = 0;
417 			}
418 			/* Neat trick that saves 1 symbol: instead of
419 			   or-ing 0 or 1 at each bit position, add 1
420 			   or 2 instead.  For example, 1011 is 1 << 0
421 			   + 1 << 1 + 2 << 2.  1010 is 2 << 0 + 2 << 1
422 			   + 1 << 2.  You can make any bit pattern
423 			   that way using 1 less symbol than the basic
424 			   or 0/1 method (except all bits 0, which
425 			   would use no symbols, but a run of length 0
426 			   doesn't mean anything in this context).
427 			   Thus space is saved. */
428 			t += (runPos << nextSym);
429 			/* +runPos if RUNA; +2*runPos if RUNB */
430 
431 			runPos <<= 1;
432 			continue;
433 		}
434 		/* When we hit the first non-run symbol after a run,
435 		   we now know how many times to repeat the last
436 		   literal, so append that many copies to our buffer
437 		   of decoded symbols (dbuf) now.  (The last literal
438 		   used is the one at the head of the mtfSymbol
439 		   array.) */
440 		if (runPos) {
441 			runPos = 0;
442 			if (dbufCount+t >= dbufSize)
443 				return RETVAL_DATA_ERROR;
444 
445 			uc = symToByte[mtfSymbol[0]];
446 			byteCount[uc] += t;
447 			while (t--)
448 				dbuf[dbufCount++] = uc;
449 		}
450 		/* Is this the terminating symbol? */
451 		if (nextSym > symTotal)
452 			break;
453 		/* At this point, nextSym indicates a new literal
454 		   character.  Subtract one to get the position in the
455 		   MTF array at which this literal is currently to be
456 		   found.  (Note that the result can't be -1 or 0,
457 		   because 0 and 1 are RUNA and RUNB.  But another
458 		   instance of the first symbol in the mtf array,
459 		   position 0, would have been handled as part of a
460 		   run above.  Therefore 1 unused mtf position minus 2
461 		   non-literal nextSym values equals -1.) */
462 		if (dbufCount >= dbufSize)
463 			return RETVAL_DATA_ERROR;
464 		i = nextSym - 1;
465 		uc = mtfSymbol[i];
466 		/* Adjust the MTF array.  Since we typically expect to
467 		 *move only a small number of symbols, and are bound
468 		 *by 256 in any case, using memmove here would
469 		 *typically be bigger and slower due to function call
470 		 *overhead and other assorted setup costs. */
471 		do {
472 			mtfSymbol[i] = mtfSymbol[i-1];
473 		} while (--i);
474 		mtfSymbol[0] = uc;
475 		uc = symToByte[uc];
476 		/* We have our literal byte.  Save it into dbuf. */
477 		byteCount[uc]++;
478 		dbuf[dbufCount++] = (unsigned int)uc;
479 	}
480 	/* At this point, we've read all the Huffman-coded symbols
481 	   (and repeated runs) for this block from the input stream,
482 	   and decoded them into the intermediate buffer.  There are
483 	   dbufCount many decoded bytes in dbuf[].  Now undo the
484 	   Burrows-Wheeler transform on dbuf.  See
485 	   http://dogma.net/markn/articles/bwt/bwt.htm
486 	 */
487 	/* Turn byteCount into cumulative occurrence counts of 0 to n-1. */
488 	j = 0;
489 	for (i = 0; i < 256; i++) {
490 		k = j+byteCount[i];
491 		byteCount[i] = j;
492 		j = k;
493 	}
494 	/* Figure out what order dbuf would be in if we sorted it. */
495 	for (i = 0; i < dbufCount; i++) {
496 		uc = (unsigned char)(dbuf[i] & 0xff);
497 		dbuf[byteCount[uc]] |= (i << 8);
498 		byteCount[uc]++;
499 	}
500 	/* Decode first byte by hand to initialize "previous" byte.
501 	   Note that it doesn't get output, and if the first three
502 	   characters are identical it doesn't qualify as a run (hence
503 	   writeRunCountdown = 5). */
504 	if (dbufCount) {
505 		if (origPtr >= dbufCount)
506 			return RETVAL_DATA_ERROR;
507 		bd->writePos = dbuf[origPtr];
508 		bd->writeCurrent = (unsigned char)(bd->writePos&0xff);
509 		bd->writePos >>= 8;
510 		bd->writeRunCountdown = 5;
511 	}
512 	bd->writeCount = dbufCount;
513 
514 	return RETVAL_OK;
515 }
516 
517 /* Undo burrows-wheeler transform on intermediate buffer to produce output.
518    If start_bunzip was initialized with out_fd =-1, then up to len bytes of
519    data are written to outbuf.  Return value is number of bytes written or
520    error (all errors are negative numbers).  If out_fd!=-1, outbuf and len
521    are ignored, data is written to out_fd and return is RETVAL_OK or error.
522 */
523 
read_bunzip(struct bunzip_data * bd,char * outbuf,int len)524 static int INIT read_bunzip(struct bunzip_data *bd, char *outbuf, int len)
525 {
526 	const unsigned int *dbuf;
527 	int pos, xcurrent, previous, gotcount;
528 
529 	/* If last read was short due to end of file, return last block now */
530 	if (bd->writeCount < 0)
531 		return bd->writeCount;
532 
533 	gotcount = 0;
534 	dbuf = bd->dbuf;
535 	pos = bd->writePos;
536 	xcurrent = bd->writeCurrent;
537 
538 	/* We will always have pending decoded data to write into the output
539 	   buffer unless this is the very first call (in which case we haven't
540 	   Huffman-decoded a block into the intermediate buffer yet). */
541 
542 	if (bd->writeCopies) {
543 		/* Inside the loop, writeCopies means extra copies (beyond 1) */
544 		--bd->writeCopies;
545 		/* Loop outputting bytes */
546 		for (;;) {
547 			/* If the output buffer is full, snapshot
548 			 * state and return */
549 			if (gotcount >= len) {
550 				bd->writePos = pos;
551 				bd->writeCurrent = xcurrent;
552 				bd->writeCopies++;
553 				return len;
554 			}
555 			/* Write next byte into output buffer, updating CRC */
556 			outbuf[gotcount++] = xcurrent;
557 			bd->writeCRC = (((bd->writeCRC) << 8)
558 				^bd->crc32Table[((bd->writeCRC) >> 24)
559 				^xcurrent]);
560 			/* Loop now if we're outputting multiple
561 			 * copies of this byte */
562 			if (bd->writeCopies) {
563 				--bd->writeCopies;
564 				continue;
565 			}
566 decode_next_byte:
567 			if (!bd->writeCount--)
568 				break;
569 			/* Follow sequence vector to undo
570 			 * Burrows-Wheeler transform */
571 			previous = xcurrent;
572 			pos = dbuf[pos];
573 			xcurrent = pos&0xff;
574 			pos >>= 8;
575 			/* After 3 consecutive copies of the same
576 			   byte, the 4th is a repeat count.  We count
577 			   down from 4 instead *of counting up because
578 			   testing for non-zero is faster */
579 			if (--bd->writeRunCountdown) {
580 				if (xcurrent != previous)
581 					bd->writeRunCountdown = 4;
582 			} else {
583 				/* We have a repeated run, this byte
584 				 * indicates the count */
585 				bd->writeCopies = xcurrent;
586 				xcurrent = previous;
587 				bd->writeRunCountdown = 5;
588 				/* Sometimes there are just 3 bytes
589 				 * (run length 0) */
590 				if (!bd->writeCopies)
591 					goto decode_next_byte;
592 				/* Subtract the 1 copy we'd output
593 				 * anyway to get extras */
594 				--bd->writeCopies;
595 			}
596 		}
597 		/* Decompression of this block completed successfully */
598 		bd->writeCRC = ~bd->writeCRC;
599 		bd->totalCRC = ((bd->totalCRC << 1) |
600 				(bd->totalCRC >> 31)) ^ bd->writeCRC;
601 		/* If this block had a CRC error, force file level CRC error. */
602 		if (bd->writeCRC != bd->headerCRC) {
603 			bd->totalCRC = bd->headerCRC+1;
604 			return RETVAL_LAST_BLOCK;
605 		}
606 	}
607 
608 	/* Refill the intermediate buffer by Huffman-decoding next
609 	 * block of input */
610 	/* (previous is just a convenient unused temp variable here) */
611 	previous = get_next_block(bd);
612 	if (previous) {
613 		bd->writeCount = previous;
614 		return (previous != RETVAL_LAST_BLOCK) ? previous : gotcount;
615 	}
616 	bd->writeCRC = 0xffffffffUL;
617 	pos = bd->writePos;
618 	xcurrent = bd->writeCurrent;
619 	goto decode_next_byte;
620 }
621 
nofill(void * buf,unsigned long len)622 static long INIT nofill(void *buf, unsigned long len)
623 {
624 	return -1;
625 }
626 
627 /* Allocate the structure, read file header.  If in_fd ==-1, inbuf must contain
628    a complete bunzip file (len bytes long).  If in_fd!=-1, inbuf and len are
629    ignored, and data is read from file handle into temporary buffer. */
start_bunzip(struct bunzip_data ** bdp,void * inbuf,long len,long (* fill)(void *,unsigned long))630 static int INIT start_bunzip(struct bunzip_data **bdp, void *inbuf, long len,
631 			     long (*fill)(void*, unsigned long))
632 {
633 	struct bunzip_data *bd;
634 	unsigned int i, j, c;
635 	const unsigned int BZh0 =
636 		(((unsigned int)'B') << 24)+(((unsigned int)'Z') << 16)
637 		+(((unsigned int)'h') << 8)+(unsigned int)'0';
638 
639 	/* Figure out how much data to allocate */
640 	i = sizeof(struct bunzip_data);
641 
642 	/* Allocate bunzip_data.  Most fields initialize to zero. */
643 	bd = *bdp = malloc(i);
644 	if (!bd)
645 		return RETVAL_OUT_OF_MEMORY;
646 	memset(bd, 0, sizeof(struct bunzip_data));
647 	/* Setup input buffer */
648 	bd->inbuf = inbuf;
649 	bd->inbufCount = len;
650 	if (fill != NULL)
651 		bd->fill = fill;
652 	else
653 		bd->fill = nofill;
654 
655 	/* Init the CRC32 table (big endian) */
656 	for (i = 0; i < 256; i++) {
657 		c = i << 24;
658 		for (j = 8; j; j--)
659 			c = c&0x80000000 ? (c << 1)^(CRC32_POLY_BE) : (c << 1);
660 		bd->crc32Table[i] = c;
661 	}
662 
663 	/* Ensure that file starts with "BZh['1'-'9']." */
664 	i = get_bits(bd, 32);
665 	if (((unsigned int)(i-BZh0-1)) >= 9)
666 		return RETVAL_NOT_BZIP_DATA;
667 
668 	/* Fourth byte (ascii '1'-'9'), indicates block size in units of 100k of
669 	   uncompressed data.  Allocate intermediate buffer for block. */
670 	bd->dbufSize = 100000*(i-BZh0);
671 
672 	bd->dbuf = large_malloc(bd->dbufSize * sizeof(int));
673 	if (!bd->dbuf)
674 		return RETVAL_OUT_OF_MEMORY;
675 	return RETVAL_OK;
676 }
677 
678 /* Example usage: decompress src_fd to dst_fd.  (Stops at end of bzip2 data,
679    not end of file.) */
bunzip2(unsigned char * buf,long len,long (* fill)(void *,unsigned long),long (* flush)(void *,unsigned long),unsigned char * outbuf,long * pos,void (* error)(char * x))680 STATIC int INIT bunzip2(unsigned char *buf, long len,
681 			long (*fill)(void*, unsigned long),
682 			long (*flush)(void*, unsigned long),
683 			unsigned char *outbuf,
684 			long *pos,
685 			void(*error)(char *x))
686 {
687 	struct bunzip_data *bd;
688 	int i = -1;
689 	unsigned char *inbuf;
690 
691 	if (flush)
692 		outbuf = malloc(BZIP2_IOBUF_SIZE);
693 
694 	if (!outbuf) {
695 		error("Could not allocate output buffer");
696 		return RETVAL_OUT_OF_MEMORY;
697 	}
698 	if (buf)
699 		inbuf = buf;
700 	else
701 		inbuf = malloc(BZIP2_IOBUF_SIZE);
702 	if (!inbuf) {
703 		error("Could not allocate input buffer");
704 		i = RETVAL_OUT_OF_MEMORY;
705 		goto exit_0;
706 	}
707 	i = start_bunzip(&bd, inbuf, len, fill);
708 	if (!i) {
709 		for (;;) {
710 			i = read_bunzip(bd, outbuf, BZIP2_IOBUF_SIZE);
711 			if (i <= 0)
712 				break;
713 			if (!flush)
714 				outbuf += i;
715 			else
716 				if (i != flush(outbuf, i)) {
717 					i = RETVAL_UNEXPECTED_OUTPUT_EOF;
718 					break;
719 				}
720 		}
721 	}
722 	/* Check CRC and release memory */
723 	if (i == RETVAL_LAST_BLOCK) {
724 		if (bd->headerCRC != bd->totalCRC)
725 			error("Data integrity error when decompressing.");
726 		else
727 			i = RETVAL_OK;
728 	} else if (i == RETVAL_UNEXPECTED_OUTPUT_EOF) {
729 		error("Compressed file ends unexpectedly");
730 	}
731 	if (!bd)
732 		goto exit_1;
733 	if (bd->dbuf)
734 		large_free(bd->dbuf);
735 	if (pos)
736 		*pos = bd->inbufPos;
737 	free(bd);
738 exit_1:
739 	if (!buf)
740 		free(inbuf);
741 exit_0:
742 	if (flush)
743 		free(outbuf);
744 	return i;
745 }
746 
747 #ifdef PREBOOT
__decompress(unsigned char * buf,long len,long (* fill)(void *,unsigned long),long (* flush)(void *,unsigned long),unsigned char * outbuf,long olen,long * pos,void (* error)(char * x))748 STATIC int INIT __decompress(unsigned char *buf, long len,
749 			long (*fill)(void*, unsigned long),
750 			long (*flush)(void*, unsigned long),
751 			unsigned char *outbuf, long olen,
752 			long *pos,
753 			void (*error)(char *x))
754 {
755 	return bunzip2(buf, len - 4, fill, flush, outbuf, pos, error);
756 }
757 #endif
758