diff options
| -rw-r--r-- | archival/libunarchive/decompress_bunzip2.c | 163 |
1 files changed, 145 insertions, 18 deletions
diff --git a/archival/libunarchive/decompress_bunzip2.c b/archival/libunarchive/decompress_bunzip2.c index af3731b..8694a32 100644 --- a/archival/libunarchive/decompress_bunzip2.c +++ b/archival/libunarchive/decompress_bunzip2.c @@ -7,7 +7,7 @@ Robert Sedgewick, and Jon L. Bentley. This code is licensed under the LGPLv2: - LGPL (http://www.gnu.org/copyleft/lgpl.html + LGPL http://www.gnu.org/copyleft/lgpl.html */ /* @@ -23,19 +23,8 @@ I would ask that anyone benefiting from this work, especially those using it in commercial products, consider making a donation to my local - non-profit hospice organization in the name of the woman I loved, who - passed away Feb. 12, 2003. - - In memory of Toni W. Hagan - - Hospice of Acadiana, Inc. - 2600 Johnston St., Suite 200 - Lafayette, LA 70503-3240 - - Phone (337) 232-1234 or 1-800-738-2226 - Fax (337) 232-1297 - - http://www.hospiceacadiana.com/ + non-profit hospice organization (www.hospiceacadiana.com) in the name of + the woman I loved, Toni W. Hagan, who passed away Feb. 12, 2003. Manuel */ @@ -79,51 +68,73 @@ struct group_data { /* Structure holding all the housekeeping data, including IO buffers and memory that persists between calls to bunzip */ + typedef struct { /* State for interrupting output loop */ + int writeCopies,writePos,writeRunCountdown,writeCount,writeCurrent; + /* I/O tracking data (file handles, buffers, positions, etc.) */ + int in_fd,out_fd,inbufCount,inbufPos /*,outbufPos*/; unsigned char *inbuf /*,*outbuf*/; unsigned int inbufBitCount, inbufBits; + /* The CRC values stored in the block header and calculated from the data */ + unsigned int crc32Table[256],headerCRC, totalCRC, writeCRC; + /* Intermediate buffer and its size (in bytes) */ + unsigned int *dbuf, dbufSize; + /* These things are a bit too big to go on the stack */ + unsigned char selectors[32768]; /* nSelectors=15 bits */ struct group_data groups[MAX_GROUPS]; /* Huffman coding tables */ + /* For I/O error handling */ + jmp_buf jmpbuf; } bunzip_data; /* Return the next nnn bits of input. All reads from the compressed input are done through this function. All reads are big endian */ + static unsigned int get_bits(bunzip_data *bd, char bits_wanted) { unsigned int bits=0; /* If we need to get more data from the byte buffer, do so. (Loop getting one byte at a time to enforce endianness and avoid unaligned access.) */ + while (bd->inbufBitCount<bits_wanted) { + /* If we need to read more data from file into byte buffer, do so */ + if(bd->inbufPos==bd->inbufCount) { if((bd->inbufCount = read(bd->in_fd, bd->inbuf, IOBUF_SIZE)) <= 0) longjmp(bd->jmpbuf,RETVAL_UNEXPECTED_INPUT_EOF); bd->inbufPos=0; } + /* Avoid 32-bit overflow (dump bit buffer to top of output) */ + if(bd->inbufBitCount>=24) { bits=bd->inbufBits&((1<<bd->inbufBitCount)-1); bits_wanted-=bd->inbufBitCount; bits<<=bits_wanted; bd->inbufBitCount=0; } + /* Grab next 8 bits of input from buffer. */ + bd->inbufBits=(bd->inbufBits<<8)|bd->inbuf[bd->inbufPos++]; bd->inbufBitCount+=8; } + /* Calculate result */ + bd->inbufBitCount-=bits_wanted; bits|=(bd->inbufBits>>bd->inbufBitCount)&((1<<bits_wanted)-1); @@ -143,26 +154,34 @@ static int get_next_block(bunzip_data *bd) dbuf=bd->dbuf; dbufSize=bd->dbufSize; selectors=bd->selectors; + /* Reset longjmp I/O error handling */ + i=setjmp(bd->jmpbuf); if(i) return i; + /* Read in header signature and CRC, then validate signature. (last block signature means CRC is for whole file, return now) */ + i = get_bits(bd,24); j = get_bits(bd,24); bd->headerCRC=get_bits(bd,32); if ((i == 0x177245) && (j == 0x385090)) return RETVAL_LAST_BLOCK; if ((i != 0x314159) || (j != 0x265359)) return RETVAL_NOT_BZIP_DATA; + /* We can add support for blockRandomised if anybody complains. There was some code for this in busybox 1.0.0-pre3, but nobody ever noticed that it didn't actually work. */ + if(get_bits(bd,1)) return RETVAL_OBSOLETE_INPUT; if((origPtr=get_bits(bd,24)) > dbufSize) return RETVAL_DATA_ERROR; + /* mapping table: if some byte values are never used (encoding things like ascii text), the compression code removes the gaps to have fewer symbols to deal with, and writes a sparse bitfield indicating which values were present. We make a translation table to convert the symbols back to the corresponding bytes. */ + t=get_bits(bd, 16); symTotal=0; for (i=0;i<16;i++) { @@ -172,60 +191,81 @@ static int get_next_block(bunzip_data *bd) if(k&(1<<(15-j))) symToByte[symTotal++]=(16*i)+j; } } + /* How many different Huffman coding groups does this block use? */ + groupCount=get_bits(bd,3); if (groupCount<2 || groupCount>MAX_GROUPS) return RETVAL_DATA_ERROR; + /* nSelectors: Every GROUP_SIZE many symbols we select a new Huffman coding group. Read in the group selector list, which is stored as MTF encoded bit runs. (MTF=Move To Front, as each value is used it's moved to the start of the list.) */ + if(!(nSelectors=get_bits(bd, 15))) return RETVAL_DATA_ERROR; for(i=0; i<groupCount; i++) mtfSymbol[i] = i; for(i=0; i<nSelectors; i++) { + /* Get next value */ + for(j=0;get_bits(bd,1);j++) if (j>=groupCount) return RETVAL_DATA_ERROR; + /* Decode MTF to get the next selector */ + uc = mtfSymbol[j]; for(;j;j--) mtfSymbol[j] = mtfSymbol[j-1]; mtfSymbol[0]=selectors[i]=uc; } + /* Read the Huffman coding tables for each group, which code for symTotal literal symbols, plus two run symbols (RUNA, RUNB) */ + symCount=symTotal+2; for (j=0; j<groupCount; j++) { unsigned char length[MAX_SYMBOLS],temp[MAX_HUFCODE_BITS+1]; int minLen, maxLen, pp; + /* Read Huffman code lengths for each symbol. They're stored in a way similar to mtf; record a starting value for the first symbol, and an offset from the previous value for everys symbol after that. (Subtracting 1 before the loop and then adding it back at the end is an optimization that makes the test inside the loop simpler: symbol length 0 becomes negative, so an unsigned inequality catches it.) */ + t=get_bits(bd, 5)-1; for (i = 0; i < symCount; i++) { for(;;) { if (((unsigned)t) > (MAX_HUFCODE_BITS-1)) return RETVAL_DATA_ERROR; + /* If first bit is 0, stop. Else second bit indicates whether to increment or decrement the value. Optimization: grab 2 bits and unget the second if the first was 0. */ + k = get_bits(bd,2); if (k < 2) { bd->inbufBitCount++; break; } + /* Add one if second bit 1, else subtract 1. Avoids if/else */ + t+=(((k+1)&2)-1); } + /* Correct for the initial -1, to get the final symbol length */ + length[i]=t+1; } + /* Find largest and smallest lengths in this group */ + minLen=maxLen=length[0]; for(i = 1; i < symCount; i++) { if(length[i] > maxLen) maxLen = length[i]; else if(length[i] < minLen) minLen = length[i]; } + /* Calculate permute[], base[], and limit[] tables from length[]. * * permute[] is the lookup table for converting Huffman coded symbols @@ -236,36 +276,47 @@ static int get_next_block(bunzip_data *bd) * number of bits can have. This is how the Huffman codes can vary in * length: each code with a value>limit[length] needs another bit. */ + hufGroup=bd->groups+j; hufGroup->minLen = minLen; hufGroup->maxLen = maxLen; + /* Note that minLen can't be smaller than 1, so we adjust the base and limit array pointers so we're not always wasting the first entry. We do this again when using them (during symbol decoding).*/ + base=hufGroup->base-1; limit=hufGroup->limit-1; + /* Calculate permute[]. Concurently, initialize temp[] and limit[]. */ + pp=0; for(i=minLen;i<=maxLen;i++) { temp[i]=limit[i]=0; for(t=0;t<symCount;t++) if(length[t]==i) hufGroup->permute[pp++] = t; } + /* Count symbols coded for at each bit length */ + for (i=0;i<symCount;i++) temp[length[i]]++; + /* Calculate limit[] (the largest symbol-coding value at each bit * length, which is (previous limit<<1)+symbols at this level), and * base[] (number of symbols to ignore at each bit length, which is * limit minus the cumulative count of symbols coded for already). */ + pp=t=0; for (i=minLen; i<maxLen; i++) { pp+=temp[i]; + /* We read the largest possible symbol size and then unget bits after determining how many we need, and those extra bits could be set to anything. (They're noise from future symbols.) At each level we're really only interested in the first few bits, so here we set all the trailing to-be-ignored bits to 1 so they don't affect the value>limit[length] comparison. */ + limit[i]= (pp << (maxLen - i)) - 1; pp<<=1; base[i+1]=pp-(t+=temp[i]); @@ -274,26 +325,34 @@ static int get_next_block(bunzip_data *bd) limit[maxLen]=pp+temp[maxLen]-1; base[minLen]=0; } + /* We've finished reading and digesting the block header. Now read this block's Huffman coded symbols from the file and undo the Huffman coding and run length encoding, saving the result into dbuf[dbufCount++]=uc */ /* Initialize symbol occurrence counters and symbol Move To Front table */ + for(i=0;i<256;i++) { byteCount[i] = 0; mtfSymbol[i]=(unsigned char)i; } + /* Loop through compressed symbols. */ + runPos=dbufCount=selector=0; for(;;) { + /* fetch next Huffman coding group from list. */ + symCount=GROUP_SIZE-1; if(selector>=nSelectors) return RETVAL_DATA_ERROR; hufGroup=bd->groups+selectors[selector++]; base=hufGroup->base-1; limit=hufGroup->limit-1; continue_this_group: + /* Read next Huffman-coded symbol. */ + /* Note: It is far cheaper to read maxLen bits and back up than it is to read minLen bits and then an additional bit at a time, testing as we go. Because there is a trailing last block (with file CRC), @@ -303,6 +362,7 @@ continue_this_group: dry). The following (up to got_huff_bits:) is equivalent to j=get_bits(bd,hufGroup->maxLen); */ + while (bd->inbufBitCount<hufGroup->maxLen) { if(bd->inbufPos==bd->inbufCount) { j = get_bits(bd,hufGroup->maxLen); @@ -313,27 +373,37 @@ continue_this_group: }; bd->inbufBitCount-=hufGroup->maxLen; j = (bd->inbufBits>>bd->inbufBitCount)&((1<<hufGroup->maxLen)-1); + got_huff_bits: + /* Figure how how many bits are in next symbol and unget extras */ + i=hufGroup->minLen; while(j>limit[i]) ++i; bd->inbufBitCount += (hufGroup->maxLen - i); + /* Huffman decode value to get nextSym (with bounds checking) */ + if ((i > hufGroup->maxLen) || (((unsigned)(j=(j>>(hufGroup->maxLen-i))-base[i])) >= MAX_SYMBOLS)) return RETVAL_DATA_ERROR; nextSym = hufGroup->permute[j]; + /* We have now decoded the symbol, which indicates either a new literal byte, or a repeated run of the most recent literal byte. First, check if nextSym indicates a repeated run, and if so loop collecting how many times to repeat the last literal. */ + if (((unsigned)nextSym) <= SYMBOL_RUNB) { /* RUNA or RUNB */ + /* If this is the start of a new run, zero out counter */ + if(!runPos) { runPos = 1; t = 0; } + /* Neat trick that saves 1 symbol: instead of or-ing 0 or 1 at each bit position, add 1 or 2 instead. For example, 1011 is 1<<0 + 1<<1 + 2<<2. 1010 is 2<<0 + 2<<1 + 1<<2. @@ -341,14 +411,17 @@ got_huff_bits: the basic or 0/1 method (except all bits 0, which would use no symbols, but a run of length 0 doesn't mean anything in this context). Thus space is saved. */ + t += (runPos << nextSym); /* +runPos if RUNA; +2*runPos if RUNB */ runPos <<= 1; goto end_of_huffman_loop; } + /* When we hit the first non-run symbol after a run, we now know how many times to repeat the last literal, so append that many copies to our buffer of decoded symbols (dbuf) now. (The last literal used is the one at the head of the mtfSymbol array.) */ + if(runPos) { runPos=0; if(dbufCount+t>=dbufSize) return RETVAL_DATA_ERROR; @@ -357,8 +430,11 @@ got_huff_bits: byteCount[uc] += t; while(t--) dbuf[dbufCount++]=uc; } + /* Is this the terminating symbol? */ + if(nextSym>symTotal) break; + /* At this point, nextSym indicates a new literal character. Subtract one to get the position in the MTF array at which this literal is currently to be found. (Note that the result can't be -1 or 0, @@ -366,48 +442,62 @@ got_huff_bits: first symbol in the mtf array, position 0, would have been handled as part of a run above. Therefore 1 unused mtf position minus 2 non-literal nextSym values equals -1.) */ + if(dbufCount>=dbufSize) return RETVAL_DATA_ERROR; i = nextSym - 1; uc = mtfSymbol[i]; + /* Adjust the MTF array. Since we typically expect to move only a * small number of symbols, and are bound by 256 in any case, using * memmove here would typically be bigger and slower due to function * call overhead and other assorted setup costs. */ + do { mtfSymbol[i] = mtfSymbol[i-1]; } while (--i); mtfSymbol[0] = uc; uc=symToByte[uc]; + /* We have our literal byte. Save it into dbuf. */ + byteCount[uc]++; dbuf[dbufCount++] = (unsigned int)uc; - /* Skip group initialization if we're not done with this group. Done this - * way to avoid compiler warning. */ + + /* Skip group initialization if we're not done with this group. Done + * this way to avoid compiler warning. */ + end_of_huffman_loop: if(symCount--) goto continue_this_group; } + /* At this point, we've read all the Huffman-coded symbols (and repeated runs) for this block from the input stream, and decoded them into the intermediate buffer. There are dbufCount many decoded bytes in dbuf[]. Now undo the Burrows-Wheeler transform on dbuf. See http://dogma.net/markn/articles/bwt/bwt.htm */ + /* Turn byteCount into cumulative occurrence counts of 0 to n-1. */ + j=0; for(i=0;i<256;i++) { k=j+byteCount[i]; byteCount[i] = j; j=k; } + /* Figure out what order dbuf would be in if we sorted it. */ + for (i=0;i<dbufCount;i++) { uc=(unsigned char)(dbuf[i] & 0xff); dbuf[byteCount[uc]] |= (i << 8); byteCount[uc]++; } + /* Decode first byte by hand to initialize "previous" byte. Note that it doesn't get output, and if the first three characters are identical it doesn't qualify as a run (hence writeRunCountdown=5). */ + if(dbufCount) { if(origPtr>=dbufCount) return RETVAL_DATA_ERROR; bd->writePos=dbuf[origPtr]; @@ -445,22 +535,32 @@ static int read_bunzip(bunzip_data *bd, char *outbuf, int len) Huffman-decoded a block into the intermediate buffer yet). */ if (bd->writeCopies) { + /* Inside the loop, writeCopies means extra copies (beyond 1) */ + --bd->writeCopies; + /* Loop outputting bytes */ + for(;;) { + /* If the output buffer is full, snapshot state and return */ + if(gotcount >= len) { bd->writePos=pos; bd->writeCurrent=current; bd->writeCopies++; return len; } + /* Write next byte into output buffer, updating CRC */ + outbuf[gotcount++] = current; bd->writeCRC=(((bd->writeCRC)<<8) ^bd->crc32Table[((bd->writeCRC)>>24)^current]); + /* Loop now if we're outputting multiple copies of this byte */ + if (bd->writeCopies) { --bd->writeCopies; continue; @@ -472,26 +572,38 @@ decode_next_byte: pos=dbuf[pos]; current=pos&0xff; pos>>=8; + /* After 3 consecutive copies of the same byte, the 4th is a repeat count. We count down from 4 instead * of counting up because testing for non-zero is faster */ + if(--bd->writeRunCountdown) { if(current!=previous) bd->writeRunCountdown=4; } else { + /* We have a repeated run, this byte indicates the count */ + bd->writeCopies=current; current=previous; bd->writeRunCountdown=5; + /* Sometimes there are just 3 bytes (run length 0) */ + if(!bd->writeCopies) goto decode_next_byte; + /* Subtract the 1 copy we'd output anyway to get extras */ + --bd->writeCopies; } } + /* Decompression of this block completed successfully */ + bd->writeCRC=~bd->writeCRC; bd->totalCRC=((bd->totalCRC<<1) | (bd->totalCRC>>31)) ^ bd->writeCRC; + /* If this block had a CRC error, force file level CRC error. */ + if(bd->writeCRC!=bd->headerCRC) { bd->totalCRC=bd->headerCRC+1; return RETVAL_LAST_BLOCK; @@ -500,6 +612,7 @@ decode_next_byte: /* Refill the intermediate buffer by Huffman-decoding next block of input */ /* (previous is just a convenient unused temp variable here) */ + previous=get_next_block(bd); if(previous) { bd->writeCount=previous; @@ -514,6 +627,7 @@ decode_next_byte: /* Allocate the structure, read file header. If in_fd==-1, inbuf must contain a complete bunzip file (len bytes long). If in_fd!=-1, inbuf and len are ignored, and data is read from file handle into temporary buffer. */ + static int start_bunzip(bunzip_data **bdp, int in_fd, unsigned char *inbuf, int len) { @@ -523,33 +637,44 @@ static int start_bunzip(bunzip_data **bdp, int in_fd, unsigned char *inbuf, +(((unsigned int)'h')<<8)+(unsigned int)'0'; /* Figure out how much data to allocate */ + i=sizeof(bunzip_data); if(in_fd!=-1) i+=IOBUF_SIZE; + /* Allocate bunzip_data. Most fields initialize to zero. */ + bd=*bdp=xmalloc(i); memset(bd,0,sizeof(bunzip_data)); + /* Setup input buffer */ + if(-1==(bd->in_fd=in_fd)) { bd->inbuf=inbuf; bd->inbufCount=len; } else bd->inbuf=(unsigned char *)(bd+1); + /* Init the CRC32 table (big endian) */ + for(i=0;i<256;i++) { c=i<<24; for(j=8;j;j--) c=c&0x80000000 ? (c<<1)^0x04c11db7 : (c<<1); bd->crc32Table[i]=c; } + /* Setup for I/O error handling via longjmp */ + i=setjmp(bd->jmpbuf); if(i) return i; /* Ensure that file starts with "BZh['1'-'9']." */ + i = get_bits(bd,32); if (((unsigned int)(i-BZh0-1)) >= 9) return RETVAL_NOT_BZIP_DATA; /* Fourth byte (ascii '1'-'9'), indicates block size in units of 100k of uncompressed data. Allocate intermediate buffer for block. */ + bd->dbufSize=100000*(i-BZh0); bd->dbuf=xmalloc(bd->dbufSize * sizeof(int)); @@ -558,6 +683,7 @@ static int start_bunzip(bunzip_data **bdp, int in_fd, unsigned char *inbuf, /* Example usage: decompress src_fd to dst_fd. (Stops at end of bzip data, not end of file.) */ + extern int uncompressStream(int src_fd, int dst_fd) { char *outbuf; @@ -574,15 +700,16 @@ extern int uncompressStream(int src_fd, int dst_fd) } } } + /* Check CRC and release memory */ + if(i==RETVAL_LAST_BLOCK) { if (bd->headerCRC!=bd->totalCRC) { bb_error_msg("Data integrity error when decompressing."); } else { i=RETVAL_OK; } - } - else if (i==RETVAL_UNEXPECTED_OUTPUT_EOF) { + } else if (i==RETVAL_UNEXPECTED_OUTPUT_EOF) { bb_error_msg("Compressed file ends unexpectedly"); } else { bb_error_msg("Decompression failed"); |