1 /* vi: set sw=4 ts=4: */
2 /* Small bzip2 deflate implementation, by Rob Landley (rob@landley.net).
4 Based on bzip2 decompression code by Julian R Seward (jseward@acm.org),
5 which also acknowledges contributions by Mike Burrows, David Wheeler,
6 Peter Fenwick, Alistair Moffat, Radford Neal, Ian H. Witten,
7 Robert Sedgewick, and Jon L. Bentley.
9 This code is licensed under the LGPLv2:
10 LGPL http://www.gnu.org/copyleft/lgpl.html
14 Size and speed optimizations by Manuel Novoa III (mjn3@codepoet.org).
16 More efficient reading of Huffman codes, a streamlined read_bunzip()
17 function, and various other tweaks. In (limited) tests, approximately
18 20% faster than bzcat on x86 and about 10% faster on arm.
20 Note that about 2/3 of the time is spent in read_unzip() reversing
21 the Burrows-Wheeler transformation. Much of that time is delay
22 resulting from cache misses.
24 I would ask that anyone benefiting from this work, especially those
25 using it in commercial products, consider making a donation to my local
26 non-profit hospice organization (www.hospiceacadiana.com) in the name of
27 the woman I loved, Toni W. Hagan, who passed away Feb. 12, 2003.
41 /* Constants for Huffman coding */
43 #define GROUP_SIZE 50 /* 64 would have been more efficient */
44 #define MAX_HUFCODE_BITS 20 /* Longest Huffman code allowed */
45 #define MAX_SYMBOLS 258 /* 256 literals + RUNA + RUNB */
49 /* Status return values */
51 #define RETVAL_LAST_BLOCK (-1)
52 #define RETVAL_NOT_BZIP_DATA (-2)
53 #define RETVAL_UNEXPECTED_INPUT_EOF (-3)
54 #define RETVAL_UNEXPECTED_OUTPUT_EOF (-4)
55 #define RETVAL_DATA_ERROR (-5)
56 #define RETVAL_OUT_OF_MEMORY (-6)
57 #define RETVAL_OBSOLETE_INPUT (-7)
59 /* Other housekeeping constants */
60 #define IOBUF_SIZE 4096
62 /* This is what we know about each Huffman coding group */
64 /* We have an extra slot at the end of limit[] for a sentinal value. */
65 int limit[MAX_HUFCODE_BITS+1],base[MAX_HUFCODE_BITS],permute[MAX_SYMBOLS];
69 /* Structure holding all the housekeeping data, including IO buffers and
70 memory that persists between calls to bunzip */
73 /* State for interrupting output loop */
75 int writeCopies,writePos,writeRunCountdown,writeCount,writeCurrent;
77 /* I/O tracking data (file handles, buffers, positions, etc.) */
79 int in_fd,out_fd,inbufCount,inbufPos /*,outbufPos*/;
80 unsigned char *inbuf /*,*outbuf*/;
81 unsigned int inbufBitCount, inbufBits;
83 /* The CRC values stored in the block header and calculated from the data */
85 unsigned int crc32Table[256],headerCRC, totalCRC, writeCRC;
87 /* Intermediate buffer and its size (in bytes) */
89 unsigned int *dbuf, dbufSize;
91 /* These things are a bit too big to go on the stack */
93 unsigned char selectors[32768]; /* nSelectors=15 bits */
94 struct group_data groups[MAX_GROUPS]; /* Huffman coding tables */
96 /* For I/O error handling */
101 /* Return the next nnn bits of input. All reads from the compressed input
102 are done through this function. All reads are big endian */
104 static unsigned int get_bits(bunzip_data *bd, char bits_wanted)
108 /* If we need to get more data from the byte buffer, do so. (Loop getting
109 one byte at a time to enforce endianness and avoid unaligned access.) */
111 while (bd->inbufBitCount<bits_wanted) {
113 /* If we need to read more data from file into byte buffer, do so */
115 if(bd->inbufPos==bd->inbufCount) {
116 if((bd->inbufCount = read(bd->in_fd, bd->inbuf, IOBUF_SIZE)) <= 0)
117 longjmp(bd->jmpbuf,RETVAL_UNEXPECTED_INPUT_EOF);
121 /* Avoid 32-bit overflow (dump bit buffer to top of output) */
123 if(bd->inbufBitCount>=24) {
124 bits=bd->inbufBits&((1<<bd->inbufBitCount)-1);
125 bits_wanted-=bd->inbufBitCount;
130 /* Grab next 8 bits of input from buffer. */
132 bd->inbufBits=(bd->inbufBits<<8)|bd->inbuf[bd->inbufPos++];
133 bd->inbufBitCount+=8;
136 /* Calculate result */
138 bd->inbufBitCount-=bits_wanted;
139 bits|=(bd->inbufBits>>bd->inbufBitCount)&((1<<bits_wanted)-1);
144 /* Unpacks the next block and sets up for the inverse burrows-wheeler step. */
146 static int get_next_block(bunzip_data *bd)
148 struct group_data *hufGroup;
149 int dbufCount,nextSym,dbufSize,groupCount,*base,*limit,selector,
150 i,j,k,t,runPos,symCount,symTotal,nSelectors,byteCount[256];
151 unsigned char uc, symToByte[256], mtfSymbol[256], *selectors;
152 unsigned int *dbuf,origPtr;
155 dbufSize=bd->dbufSize;
156 selectors=bd->selectors;
158 /* Reset longjmp I/O error handling */
160 i=setjmp(bd->jmpbuf);
163 /* Read in header signature and CRC, then validate signature.
164 (last block signature means CRC is for whole file, return now) */
168 bd->headerCRC=get_bits(bd,32);
169 if ((i == 0x177245) && (j == 0x385090)) return RETVAL_LAST_BLOCK;
170 if ((i != 0x314159) || (j != 0x265359)) return RETVAL_NOT_BZIP_DATA;
172 /* We can add support for blockRandomised if anybody complains. There was
173 some code for this in busybox 1.0.0-pre3, but nobody ever noticed that
174 it didn't actually work. */
176 if(get_bits(bd,1)) return RETVAL_OBSOLETE_INPUT;
177 if((origPtr=get_bits(bd,24)) > dbufSize) return RETVAL_DATA_ERROR;
179 /* mapping table: if some byte values are never used (encoding things
180 like ascii text), the compression code removes the gaps to have fewer
181 symbols to deal with, and writes a sparse bitfield indicating which
182 values were present. We make a translation table to convert the symbols
183 back to the corresponding bytes. */
191 if(k&(1<<(15-j))) symToByte[symTotal++]=(16*i)+j;
195 /* How many different Huffman coding groups does this block use? */
197 groupCount=get_bits(bd,3);
198 if (groupCount<2 || groupCount>MAX_GROUPS) return RETVAL_DATA_ERROR;
200 /* nSelectors: Every GROUP_SIZE many symbols we select a new Huffman coding
201 group. Read in the group selector list, which is stored as MTF encoded
202 bit runs. (MTF=Move To Front, as each value is used it's moved to the
203 start of the list.) */
205 if(!(nSelectors=get_bits(bd, 15))) return RETVAL_DATA_ERROR;
206 for(i=0; i<groupCount; i++) mtfSymbol[i] = i;
207 for(i=0; i<nSelectors; i++) {
211 for(j=0;get_bits(bd,1);j++) if (j>=groupCount) return RETVAL_DATA_ERROR;
213 /* Decode MTF to get the next selector */
216 for(;j;j--) mtfSymbol[j] = mtfSymbol[j-1];
217 mtfSymbol[0]=selectors[i]=uc;
220 /* Read the Huffman coding tables for each group, which code for symTotal
221 literal symbols, plus two run symbols (RUNA, RUNB) */
224 for (j=0; j<groupCount; j++) {
225 unsigned char length[MAX_SYMBOLS],temp[MAX_HUFCODE_BITS+1];
226 int minLen, maxLen, pp;
228 /* Read Huffman code lengths for each symbol. They're stored in
229 a way similar to mtf; record a starting value for the first symbol,
230 and an offset from the previous value for everys symbol after that.
231 (Subtracting 1 before the loop and then adding it back at the end is
232 an optimization that makes the test inside the loop simpler: symbol
233 length 0 becomes negative, so an unsigned inequality catches it.) */
236 for (i = 0; i < symCount; i++) {
238 if (((unsigned)t) > (MAX_HUFCODE_BITS-1))
239 return RETVAL_DATA_ERROR;
241 /* If first bit is 0, stop. Else second bit indicates whether
242 to increment or decrement the value. Optimization: grab 2
243 bits and unget the second if the first was 0. */
251 /* Add one if second bit 1, else subtract 1. Avoids if/else */
256 /* Correct for the initial -1, to get the final symbol length */
261 /* Find largest and smallest lengths in this group */
263 minLen=maxLen=length[0];
264 for(i = 1; i < symCount; i++) {
265 if(length[i] > maxLen) maxLen = length[i];
266 else if(length[i] < minLen) minLen = length[i];
269 /* Calculate permute[], base[], and limit[] tables from length[].
271 * permute[] is the lookup table for converting Huffman coded symbols
272 * into decoded symbols. base[] is the amount to subtract from the
273 * value of a Huffman symbol of a given length when using permute[].
275 * limit[] indicates the largest numerical value a symbol with a given
276 * number of bits can have. This is how the Huffman codes can vary in
277 * length: each code with a value>limit[length] needs another bit.
280 hufGroup=bd->groups+j;
281 hufGroup->minLen = minLen;
282 hufGroup->maxLen = maxLen;
284 /* Note that minLen can't be smaller than 1, so we adjust the base
285 and limit array pointers so we're not always wasting the first
286 entry. We do this again when using them (during symbol decoding).*/
288 base=hufGroup->base-1;
289 limit=hufGroup->limit-1;
291 /* Calculate permute[]. Concurently, initialize temp[] and limit[]. */
294 for(i=minLen;i<=maxLen;i++) {
296 for(t=0;t<symCount;t++)
297 if(length[t]==i) hufGroup->permute[pp++] = t;
300 /* Count symbols coded for at each bit length */
302 for (i=0;i<symCount;i++) temp[length[i]]++;
304 /* Calculate limit[] (the largest symbol-coding value at each bit
305 * length, which is (previous limit<<1)+symbols at this level), and
306 * base[] (number of symbols to ignore at each bit length, which is
307 * limit minus the cumulative count of symbols coded for already). */
310 for (i=minLen; i<maxLen; i++) {
313 /* We read the largest possible symbol size and then unget bits
314 after determining how many we need, and those extra bits could
315 be set to anything. (They're noise from future symbols.) At
316 each level we're really only interested in the first few bits,
317 so here we set all the trailing to-be-ignored bits to 1 so they
318 don't affect the value>limit[length] comparison. */
320 limit[i]= (pp << (maxLen - i)) - 1;
322 base[i+1]=pp-(t+=temp[i]);
324 limit[maxLen+1] = INT_MAX; /* Sentinal value for reading next sym. */
325 limit[maxLen]=pp+temp[maxLen]-1;
329 /* We've finished reading and digesting the block header. Now read this
330 block's Huffman coded symbols from the file and undo the Huffman coding
331 and run length encoding, saving the result into dbuf[dbufCount++]=uc */
333 /* Initialize symbol occurrence counters and symbol Move To Front table */
337 mtfSymbol[i]=(unsigned char)i;
340 /* Loop through compressed symbols. */
342 runPos=dbufCount=selector=0;
345 /* fetch next Huffman coding group from list. */
347 symCount=GROUP_SIZE-1;
348 if(selector>=nSelectors) return RETVAL_DATA_ERROR;
349 hufGroup=bd->groups+selectors[selector++];
350 base=hufGroup->base-1;
351 limit=hufGroup->limit-1;
354 /* Read next Huffman-coded symbol. */
356 /* Note: It is far cheaper to read maxLen bits and back up than it is
357 to read minLen bits and then an additional bit at a time, testing
358 as we go. Because there is a trailing last block (with file CRC),
359 there is no danger of the overread causing an unexpected EOF for a
360 valid compressed file. As a further optimization, we do the read
361 inline (falling back to a call to get_bits if the buffer runs
362 dry). The following (up to got_huff_bits:) is equivalent to
363 j=get_bits(bd,hufGroup->maxLen);
366 while (bd->inbufBitCount<hufGroup->maxLen) {
367 if(bd->inbufPos==bd->inbufCount) {
368 j = get_bits(bd,hufGroup->maxLen);
371 bd->inbufBits=(bd->inbufBits<<8)|bd->inbuf[bd->inbufPos++];
372 bd->inbufBitCount+=8;
374 bd->inbufBitCount-=hufGroup->maxLen;
375 j = (bd->inbufBits>>bd->inbufBitCount)&((1<<hufGroup->maxLen)-1);
379 /* Figure how how many bits are in next symbol and unget extras */
382 while(j>limit[i]) ++i;
383 bd->inbufBitCount += (hufGroup->maxLen - i);
385 /* Huffman decode value to get nextSym (with bounds checking) */
387 if ((i > hufGroup->maxLen)
388 || (((unsigned)(j=(j>>(hufGroup->maxLen-i))-base[i]))
390 return RETVAL_DATA_ERROR;
391 nextSym = hufGroup->permute[j];
393 /* We have now decoded the symbol, which indicates either a new literal
394 byte, or a repeated run of the most recent literal byte. First,
395 check if nextSym indicates a repeated run, and if so loop collecting
396 how many times to repeat the last literal. */
398 if (((unsigned)nextSym) <= SYMBOL_RUNB) { /* RUNA or RUNB */
400 /* If this is the start of a new run, zero out counter */
407 /* Neat trick that saves 1 symbol: instead of or-ing 0 or 1 at
408 each bit position, add 1 or 2 instead. For example,
409 1011 is 1<<0 + 1<<1 + 2<<2. 1010 is 2<<0 + 2<<1 + 1<<2.
410 You can make any bit pattern that way using 1 less symbol than
411 the basic or 0/1 method (except all bits 0, which would use no
412 symbols, but a run of length 0 doesn't mean anything in this
413 context). Thus space is saved. */
415 t += (runPos << nextSym); /* +runPos if RUNA; +2*runPos if RUNB */
416 if(runPos < dbufSize) runPos <<= 1;
417 goto end_of_huffman_loop;
420 /* When we hit the first non-run symbol after a run, we now know
421 how many times to repeat the last literal, so append that many
422 copies to our buffer of decoded symbols (dbuf) now. (The last
423 literal used is the one at the head of the mtfSymbol array.) */
427 if(dbufCount+t>=dbufSize) return RETVAL_DATA_ERROR;
429 uc = symToByte[mtfSymbol[0]];
431 while(t--) dbuf[dbufCount++]=uc;
434 /* Is this the terminating symbol? */
436 if(nextSym>symTotal) break;
438 /* At this point, nextSym indicates a new literal character. Subtract
439 one to get the position in the MTF array at which this literal is
440 currently to be found. (Note that the result can't be -1 or 0,
441 because 0 and 1 are RUNA and RUNB. But another instance of the
442 first symbol in the mtf array, position 0, would have been handled
443 as part of a run above. Therefore 1 unused mtf position minus
444 2 non-literal nextSym values equals -1.) */
446 if(dbufCount>=dbufSize) return RETVAL_DATA_ERROR;
450 /* Adjust the MTF array. Since we typically expect to move only a
451 * small number of symbols, and are bound by 256 in any case, using
452 * memmove here would typically be bigger and slower due to function
453 * call overhead and other assorted setup costs. */
456 mtfSymbol[i] = mtfSymbol[i-1];
461 /* We have our literal byte. Save it into dbuf. */
464 dbuf[dbufCount++] = (unsigned int)uc;
466 /* Skip group initialization if we're not done with this group. Done
467 * this way to avoid compiler warning. */
470 if(symCount--) goto continue_this_group;
473 /* At this point, we've read all the Huffman-coded symbols (and repeated
474 runs) for this block from the input stream, and decoded them into the
475 intermediate buffer. There are dbufCount many decoded bytes in dbuf[].
476 Now undo the Burrows-Wheeler transform on dbuf.
477 See http://dogma.net/markn/articles/bwt/bwt.htm
480 /* Turn byteCount into cumulative occurrence counts of 0 to n-1. */
489 /* Figure out what order dbuf would be in if we sorted it. */
491 for (i=0;i<dbufCount;i++) {
492 uc=(unsigned char)(dbuf[i] & 0xff);
493 dbuf[byteCount[uc]] |= (i << 8);
497 /* Decode first byte by hand to initialize "previous" byte. Note that it
498 doesn't get output, and if the first three characters are identical
499 it doesn't qualify as a run (hence writeRunCountdown=5). */
502 if(origPtr>=dbufCount) return RETVAL_DATA_ERROR;
503 bd->writePos=dbuf[origPtr];
504 bd->writeCurrent=(unsigned char)(bd->writePos&0xff);
506 bd->writeRunCountdown=5;
508 bd->writeCount=dbufCount;
513 /* Undo burrows-wheeler transform on intermediate buffer to produce output.
514 If start_bunzip was initialized with out_fd=-1, then up to len bytes of
515 data are written to outbuf. Return value is number of bytes written or
516 error (all errors are negative numbers). If out_fd!=-1, outbuf and len
517 are ignored, data is written to out_fd and return is RETVAL_OK or error.
520 static int read_bunzip(bunzip_data *bd, char *outbuf, int len)
522 const unsigned int *dbuf;
523 int pos,current,previous,gotcount;
525 /* If last read was short due to end of file, return last block now */
526 if(bd->writeCount<0) return bd->writeCount;
531 current=bd->writeCurrent;
533 /* We will always have pending decoded data to write into the output
534 buffer unless this is the very first call (in which case we haven't
535 Huffman-decoded a block into the intermediate buffer yet). */
537 if (bd->writeCopies) {
539 /* Inside the loop, writeCopies means extra copies (beyond 1) */
543 /* Loop outputting bytes */
547 /* If the output buffer is full, snapshot state and return */
549 if(gotcount >= len) {
551 bd->writeCurrent=current;
556 /* Write next byte into output buffer, updating CRC */
558 outbuf[gotcount++] = current;
559 bd->writeCRC=(((bd->writeCRC)<<8)
560 ^bd->crc32Table[((bd->writeCRC)>>24)^current]);
562 /* Loop now if we're outputting multiple copies of this byte */
564 if (bd->writeCopies) {
569 if (!bd->writeCount--) break;
570 /* Follow sequence vector to undo Burrows-Wheeler transform */
576 /* After 3 consecutive copies of the same byte, the 4th is a repeat
577 count. We count down from 4 instead
578 * of counting up because testing for non-zero is faster */
580 if(--bd->writeRunCountdown) {
581 if(current!=previous) bd->writeRunCountdown=4;
584 /* We have a repeated run, this byte indicates the count */
586 bd->writeCopies=current;
588 bd->writeRunCountdown=5;
590 /* Sometimes there are just 3 bytes (run length 0) */
592 if(!bd->writeCopies) goto decode_next_byte;
594 /* Subtract the 1 copy we'd output anyway to get extras */
600 /* Decompression of this block completed successfully */
602 bd->writeCRC=~bd->writeCRC;
603 bd->totalCRC=((bd->totalCRC<<1) | (bd->totalCRC>>31)) ^ bd->writeCRC;
605 /* If this block had a CRC error, force file level CRC error. */
607 if(bd->writeCRC!=bd->headerCRC) {
608 bd->totalCRC=bd->headerCRC+1;
609 return RETVAL_LAST_BLOCK;
613 /* Refill the intermediate buffer by Huffman-decoding next block of input */
614 /* (previous is just a convenient unused temp variable here) */
616 previous=get_next_block(bd);
618 bd->writeCount=previous;
619 return (previous!=RETVAL_LAST_BLOCK) ? previous : gotcount;
621 bd->writeCRC=0xffffffffUL;
623 current=bd->writeCurrent;
624 goto decode_next_byte;
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. */
631 static int start_bunzip(bunzip_data **bdp, int in_fd, unsigned char *inbuf,
636 const unsigned int BZh0=(((unsigned int)'B')<<24)+(((unsigned int)'Z')<<16)
637 +(((unsigned int)'h')<<8)+(unsigned int)'0';
639 /* Figure out how much data to allocate */
641 i=sizeof(bunzip_data);
642 if(in_fd!=-1) i+=IOBUF_SIZE;
644 /* Allocate bunzip_data. Most fields initialize to zero. */
647 memset(bd,0,sizeof(bunzip_data));
649 /* Setup input buffer */
651 if(-1==(bd->in_fd=in_fd)) {
654 } else bd->inbuf=(unsigned char *)(bd+1);
656 /* Init the CRC32 table (big endian) */
661 c=c&0x80000000 ? (c<<1)^0x04c11db7 : (c<<1);
665 /* Setup for I/O error handling via longjmp */
667 i=setjmp(bd->jmpbuf);
670 /* Ensure that file starts with "BZh['1'-'9']." */
673 if (((unsigned int)(i-BZh0-1)) >= 9) return RETVAL_NOT_BZIP_DATA;
675 /* Fourth byte (ascii '1'-'9'), indicates block size in units of 100k of
676 uncompressed data. Allocate intermediate buffer for block. */
678 bd->dbufSize=100000*(i-BZh0);
680 bd->dbuf=xmalloc(bd->dbufSize * sizeof(int));
684 /* Example usage: decompress src_fd to dst_fd. (Stops at end of bzip data,
687 extern int uncompressStream(int src_fd, int dst_fd)
693 outbuf=xmalloc(IOBUF_SIZE);
694 if(!(i=start_bunzip(&bd,src_fd,0,0))) {
696 if((i=read_bunzip(bd,outbuf,IOBUF_SIZE)) <= 0) break;
697 if(i!=write(dst_fd,outbuf,i)) {
698 i=RETVAL_UNEXPECTED_OUTPUT_EOF;
704 /* Check CRC and release memory */
706 if(i==RETVAL_LAST_BLOCK) {
707 if (bd->headerCRC!=bd->totalCRC) {
708 bb_error_msg("Data integrity error when decompressing.");
712 } else if (i==RETVAL_UNEXPECTED_OUTPUT_EOF) {
713 bb_error_msg("Compressed file ends unexpectedly");
715 bb_error_msg("Decompression failed");
717 if(bd->dbuf) free(bd->dbuf);
726 static char * const bunzip_errors[]={NULL,"Bad file checksum","Not bzip data",
727 "Unexpected input EOF","Unexpected output EOF","Data error",
728 "Out of memory","Obsolete (pre 0.9.5) bzip format not supported."};
730 /* Dumb little test thing, decompress stdin to stdout */
731 int main(int argc, char *argv[])
733 int i=uncompressStream(0,1);
736 if(i) fprintf(stderr,"%s\n", bunzip_errors[-i]);
737 else if(read(0,&c,1)) fprintf(stderr,"Trailing garbage ignored\n");