/* vi: set sw=4 ts=4: */
-/* Small bzip2 deflate implementation, by Rob Landley (rob@landley.net).
+/* Small bzip2 deflate implementation, by Rob Landley (rob@landley.net).
- Based on bzip2 decompression code by Julian R Seward (jseward@acm.org),
- which also acknowledges contributions by Mike Burrows, David Wheeler,
- Peter Fenwick, Alistair Moffat, Radford Neal, Ian H. Witten,
- Robert Sedgewick, and Jon L. Bentley.
+ Based on bzip2 decompression code by Julian R Seward (jseward@acm.org),
+ which also acknowledges contributions by Mike Burrows, David Wheeler,
+ Peter Fenwick, Alistair Moffat, Radford Neal, Ian H. Witten,
+ Robert Sedgewick, and Jon L. Bentley.
- This code is licensed under the LGPLv2:
- LGPL (http://www.gnu.org/copyleft/lgpl.html
+ Licensed under GPLv2 or later, see file LICENSE in this tarball for details.
*/
-#include <setjmp.h>
-#include <stdio.h>
-#include <stdlib.h>
-#include <string.h>
-#include <unistd.h>
+/*
+ Size and speed optimizations by Manuel Novoa III (mjn3@codepoet.org).
-/* Constants for huffman coding */
-#define MAX_GROUPS 6
-#define GROUP_SIZE 50 /* 64 would have been more efficient */
-#define MAX_HUFCODE_BITS 20 /* Longest huffman code allowed */
-#define MAX_SYMBOLS 258 /* 256 literals + RUNA + RUNB */
-#define SYMBOL_RUNA 0
-#define SYMBOL_RUNB 1
+ More efficient reading of Huffman codes, a streamlined read_bunzip()
+ function, and various other tweaks. In (limited) tests, approximately
+ 20% faster than bzcat on x86 and about 10% faster on arm.
+
+ Note that about 2/3 of the time is spent in read_unzip() reversing
+ the Burrows-Wheeler transformation. Much of that time is delay
+ resulting from cache misses.
+
+ 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 (www.hospiceacadiana.com) in the name of
+ the woman I loved, Toni W. Hagan, who passed away Feb. 12, 2003.
+
+ Manuel
+ */
+
+#include "libbb.h"
+#include "unarchive.h"
+
+/* Constants for Huffman coding */
+#define MAX_GROUPS 6
+#define GROUP_SIZE 50 /* 64 would have been more efficient */
+#define MAX_HUFCODE_BITS 20 /* Longest Huffman code allowed */
+#define MAX_SYMBOLS 258 /* 256 literals + RUNA + RUNB */
+#define SYMBOL_RUNA 0
+#define SYMBOL_RUNB 1
/* Status return values */
-#define RETVAL_OK 0
-#define RETVAL_LAST_BLOCK (-1)
-#define RETVAL_NOT_BZIP_DATA (-2)
-#define RETVAL_UNEXPECTED_INPUT_EOF (-3)
-#define RETVAL_UNEXPECTED_OUTPUT_EOF (-4)
-#define RETVAL_DATA_ERROR (-5)
-#define RETVAL_OUT_OF_MEMORY (-6)
-#define RETVAL_OBSOLETE_INPUT (-7)
+#define RETVAL_OK 0
+#define RETVAL_LAST_BLOCK (-1)
+#define RETVAL_NOT_BZIP_DATA (-2)
+#define RETVAL_UNEXPECTED_INPUT_EOF (-3)
+#define RETVAL_UNEXPECTED_OUTPUT_EOF (-4)
+#define RETVAL_DATA_ERROR (-5)
+#define RETVAL_OUT_OF_MEMORY (-6)
+#define RETVAL_OBSOLETE_INPUT (-7)
/* Other housekeeping constants */
-#define IOBUF_SIZE 4096
-
-char *bunzip_errors[]={NULL,"Bad file checksum","Not bzip data",
- "Unexpected input EOF","Unexpected output EOF","Data error",
- "Out of memory","Obsolete (pre 0.9.5) bzip format not supported."};
+#define IOBUF_SIZE 4096
-/* This is what we know about each huffman coding group */
+/* This is what we know about each Huffman coding group */
struct group_data {
- int limit[MAX_HUFCODE_BITS],base[MAX_HUFCODE_BITS],permute[MAX_SYMBOLS];
- char minLen, maxLen;
+ /* We have an extra slot at the end of limit[] for a sentinal value. */
+ int limit[MAX_HUFCODE_BITS+1],base[MAX_HUFCODE_BITS],permute[MAX_SYMBOLS];
+ int minLen, maxLen;
};
/* Structure holding all the housekeeping data, including IO buffers and
memory that persists between calls to bunzip */
+
typedef struct {
- /* For I/O error handling */
- jmp_buf jmpbuf;
- /* Input stream, input buffer, input bit buffer */
- int in_fd,inbufCount,inbufPos;
- unsigned char *inbuf;
+ /* 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;
- /* Output buffer */
- char outbuf[IOBUF_SIZE];
- int outbufPos;
+
/* The CRC values stored in the block header and calculated from the data */
- unsigned int crc32Table[256],headerCRC, dataCRC, totalCRC;
+
+ uint32_t headerCRC, totalCRC, writeCRC;
+ uint32_t *crc32Table;
/* Intermediate buffer and its size (in bytes) */
+
unsigned int *dbuf, dbufSize;
- /* State for interrupting output loop */
- int writePos,writeRun,writeCount,writeCurrent;
/* 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 */
+ 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)))
+ 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);
return bits;
}
-/* Decompress a block of text to into intermediate buffer */
+/* Unpacks the next block and sets up for the inverse burrows-wheeler step. */
-extern int read_bunzip_data(bunzip_data *bd)
+static int get_next_block(bunzip_data *bd)
{
struct group_data *hufGroup;
- int dbufCount,nextSym,dbufSize,origPtr,groupCount,*base,*limit,selector,
+ int dbufCount,nextSym,dbufSize,groupCount,*base,*limit,selector,
i,j,k,t,runPos,symCount,symTotal,nSelectors,byteCount[256];
unsigned char uc, symToByte[256], mtfSymbol[256], *selectors;
- unsigned int *dbuf;
-
- /* Read in header signature (borrowing mtfSymbol for temp space). */
- for(i=0;i<6;i++) mtfSymbol[i]=get_bits(bd,8);
- mtfSymbol[6]=0;
- /* Read CRC (which is stored big endian). */
- bd->headerCRC=get_bits(bd,32);
- /* Is this the last block (with CRC for file)? */
- if(!strcmp(mtfSymbol,"\x17\x72\x45\x38\x50\x90"))
- return RETVAL_LAST_BLOCK;
- /* If it's not a valid data block, barf. */
- if(strcmp(mtfSymbol,"\x31\x41\x59\x26\x53\x59"))
- return RETVAL_NOT_BZIP_DATA;
+ unsigned int *dbuf,origPtr;
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;
+
+ 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);
- memset(symToByte,0,256);
symTotal=0;
for (i=0;i<16;i++) {
if(t&(1<<(15-i))) {
k=get_bits(bd,16);
- for(j=0;j<16;j++)
+ for (j=0;j<16;j++)
if(k&(1<<(15-j))) symToByte[symTotal++]=(16*i)+j;
}
}
- /* How many different huffman coding groups does this block use? */
+
+ /* 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
+
+ /* 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. */
+ 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++) {
+ 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;
+
+ 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];
- memmove(mtfSymbol+1,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
+
+ /* 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 lengths */
- t=get_bits(bd, 5);
+
+ /* 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 (t < 1 || t > MAX_HUFCODE_BITS) return RETVAL_DATA_ERROR;
- if(!get_bits(bd, 1)) break;
- if(!get_bits(bd, 1)) t++;
- else t--;
+ 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);
}
- length[i] = t;
+
+ /* 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++) {
+ 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
+ * permute[] is the lookup table for converting Huffman coded symbols
* into decoded symbols. base[] is the amount to subtract from the
- * value of a huffman symbol of a given length when using permute[].
+ * value of a Huffman symbol of a given length when using permute[].
*
* limit[] indicates the largest numerical value a symbol with a given
- * number of bits can have. It lets us know when to stop reading.
- *
- * To use these, keep reading bits until value<=limit[bitcount] or
- * you've read over 20 bits (error). Then the decoded symbol
- * equals permute[hufcode_value-base[hufcode_bitcount]].
+ * 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[] */
- pp = 0;
- for(i=minLen;i<=maxLen;i++)
- for(t=0;t<symCount;t++)
+
+ /* 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 cumulative symbols coded for at each bit length */
- for (i=minLen;i<=maxLen;i++) temp[i]=limit[i]=0;
+ }
+
+ /* 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-cumulative count of symbols coded for already). */
+ * limit minus the cumulative count of symbols coded for already). */
+
pp=t=0;
for (i=minLen; i<maxLen; i++) {
pp+=temp[i];
- limit[i]=pp-1;
+
+ /* 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]);
}
+ limit[maxLen+1] = INT_MAX; /* Sentinal value for reading next sym. */
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
+ 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 mtf table */
- memset(byteCount,0,256*sizeof(int));
- for(i=0;i<256;i++) mtfSymbol[i]=(unsigned char)i;
- /* Loop through compressed symbols */
- runPos=dbufCount=symCount=selector=0;
- for(;;) {
- /* Determine which huffman coding group to use. */
- if(!(symCount--)) {
- symCount=GROUP_SIZE-1;
- if(selector>=nSelectors) return RETVAL_DATA_ERROR;
- hufGroup=bd->groups+selectors[selector++];
- base=hufGroup->base-1;
- limit=hufGroup->limit-1;
- }
- /* Read next huffman-coded symbol */
- i = hufGroup->minLen;
- j=get_bits(bd, i);
- for(;;) {
- if (i > hufGroup->maxLen) return RETVAL_DATA_ERROR;
- if (j <= limit[i]) break;
- i++;
-
- j = (j << 1) | get_bits(bd,1);
- }
- /* Huffman decode nextSym (with bounds checking) */
- j-=base[i];
- if (j < 0 || j >= MAX_SYMBOLS) return RETVAL_DATA_ERROR;
+ /* 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),
+ there is no danger of the overread causing an unexpected EOF for a
+ valid compressed file. As a further optimization, we do the read
+ inline (falling back to a call to get_bits if the buffer runs
+ 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);
+ goto got_huff_bits;
+ }
+ bd->inbufBits=(bd->inbufBits<<8)|bd->inbuf[bd->inbufPos++];
+ bd->inbufBitCount+=8;
+ };
+ 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];
- /* If this is a repeated run, loop collecting data */
- if (nextSym == SYMBOL_RUNA || nextSym == SYMBOL_RUNB) {
+
+ /* 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.
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. */
- if (nextSym == SYMBOL_RUNA) t += runPos;
- else t += 2*runPos;
- runPos <<= 1;
- continue;
+
+ t += (runPos << nextSym); /* +runPos if RUNA; +2*runPos if RUNB */
+ if(runPos < dbufSize) 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;
uc = symToByte[mtfSymbol[0]];
byteCount[uc] += t;
- while(t--) dbuf[dbufCount++]=uc;
+ while (t--) dbuf[dbufCount++]=uc;
}
+
/* Is this the terminating symbol? */
+
if(nextSym>symTotal) break;
- /* At this point, the symbol we just decoded 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, because 0 and 1 are RUNA and RUNB.
- Another instance of the first symbol in the mtf array, position 0,
- would have been handled as part of a run.) */
+
+ /* 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,
+ because 0 and 1 are RUNA and RUNB. But another instance of the
+ 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];
- memmove(mtfSymbol+1,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. */
+
+end_of_huffman_loop:
+ if(symCount--) goto continue_this_group;
}
- /* At this point, we've finished reading huffman-coded symbols and
- compressed runs from the input stream. There are dbufCount many of
- them in dbuf[]. Now undo the Burrows-Wheeler transform on dbuf.
+
+ /* 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
*/
- /* Now we know what dbufCount is, do a better sanity check on origPtr. */
- if (origPtr<0 || origPtr>=dbufCount) return RETVAL_DATA_ERROR;
/* Turn byteCount into cumulative occurrence counts of 0 to n-1. */
+
j=0;
- for(i=0;i<256;i++) {
+ 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);
+ uc=(unsigned char)(dbuf[i] & 0xff);
dbuf[byteCount[uc]] |= (i << 8);
byteCount[uc]++;
}
- /* blockRandomised support would go here. */
- /* Using i as position, j as previous character, t as current character,
- and uc as run count */
- bd->dataCRC = 0xffffffffL;
/* 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 uc=255, which will either wrap
- to 1 or get reset). */
+ it doesn't qualify as a run (hence writeRunCountdown=5). */
+
if(dbufCount) {
+ if(origPtr>=dbufCount) return RETVAL_DATA_ERROR;
bd->writePos=dbuf[origPtr];
bd->writeCurrent=(unsigned char)(bd->writePos&0xff);
bd->writePos>>=8;
- bd->writeRun=-1;
+ bd->writeRunCountdown=5;
}
bd->writeCount=dbufCount;
return RETVAL_OK;
}
-/* Flush output buffer to disk */
-extern void flush_bunzip_outbuf(bunzip_data *bd, int out_fd)
+/* Undo burrows-wheeler transform on intermediate buffer to produce output.
+ If start_bunzip was initialized with out_fd=-1, then up to len bytes of
+ data are written to outbuf. Return value is number of bytes written or
+ error (all errors are negative numbers). If out_fd!=-1, outbuf and len
+ are ignored, data is written to out_fd and return is RETVAL_OK or error.
+*/
+
+static int read_bunzip(bunzip_data *bd, char *outbuf, int len)
{
- if(bd->outbufPos) {
- if(write(out_fd, bd->outbuf, bd->outbufPos) != bd->outbufPos)
- longjmp(bd->jmpbuf,RETVAL_UNEXPECTED_OUTPUT_EOF);
- bd->outbufPos=0;
- }
-}
+ const unsigned int *dbuf;
+ int pos,current,previous,gotcount;
+ /* If last read was short due to end of file, return last block now */
+ if(bd->writeCount<0) return bd->writeCount;
-/* Undo burrows-wheeler transform on intermediate buffer to produce output.
- If !len, write up to len bytes of data to buf. Otherwise write to out_fd.
- Returns len ? bytes written : RETVAL_OK. Notice all errors negative #'s. */
-extern int write_bunzip_data(bunzip_data *bd, int out_fd, char *outbuf, int len)
-{
- unsigned int *dbuf=bd->dbuf;
- int count,pos,current, run,copies,outbyte,previous,gotcount=0;
-
- for(;;) {
- /* If last read was short due to end of file, return last block now */
- if(bd->writeCount<0) return bd->writeCount;
- /* If we need to refill dbuf, do it. */
- if(!bd->writeCount) {
- int i=read_bunzip_data(bd);
- if(i) {
- if(i==RETVAL_LAST_BLOCK) {
- bd->writeCount=i;
- return gotcount;
- } else return i;
+ gotcount = 0;
+ dbuf=bd->dbuf;
+ pos=bd->writePos;
+ current=bd->writeCurrent;
+
+ /* We will always have pending decoded data to write into the output
+ buffer unless this is the very first call (in which case we haven't
+ 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;
}
- }
- /* Loop generating output */
- count=bd->writeCount;
- pos=bd->writePos;
- current=bd->writeCurrent;
- run=bd->writeRun;
- while(count) {
- /* If somebody (like busybox tar) wants a certain number of bytes of
- data from memory instead of written to a file, humor them */
- if(len && bd->outbufPos>=len) goto dataus_interruptus;
- count--;
+
+ /* 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;
+ }
+decode_next_byte:
+ if (!bd->writeCount--) break;
/* Follow sequence vector to undo Burrows-Wheeler transform */
previous=current;
pos=dbuf[pos];
current=pos&0xff;
pos>>=8;
- /* Whenever we see 3 consecutive copies of the same byte,
- the 4th is a repeat count */
- if(run++==3) {
- copies=current;
- outbyte=previous;
- current=-1;
+
+ /* 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 {
- copies=1;
- outbyte=current;
- }
- /* Output bytes to buffer, flushing to file if necessary */
- while(copies--) {
- if(bd->outbufPos == IOBUF_SIZE) flush_bunzip_outbuf(bd,out_fd);
- bd->outbuf[bd->outbufPos++] = outbyte;
- bd->dataCRC = (bd->dataCRC << 8)
- ^ bd->crc32Table[(bd->dataCRC >> 24) ^ outbyte];
+
+ /* 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;
}
- if(current!=previous) run=0;
}
+
/* Decompression of this block completed successfully */
- bd->dataCRC=~(bd->dataCRC);
- bd->totalCRC=((bd->totalCRC << 1) | (bd->totalCRC >> 31)) ^ bd->dataCRC;
+
+ 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->dataCRC!=bd->headerCRC) {
+
+ if(bd->writeCRC!=bd->headerCRC) {
bd->totalCRC=bd->headerCRC+1;
return RETVAL_LAST_BLOCK;
}
-dataus_interruptus:
- bd->writeCount=count;
- if(len) {
- gotcount+=bd->outbufPos;
- memcpy(outbuf,bd->outbuf,len);
- /* If we got enough data, checkpoint loop state and return */
- if((len-=bd->outbufPos)<1) {
- bd->outbufPos-=len;
- if(bd->outbufPos)
- memmove(bd->outbuf,bd->outbuf+len,bd->outbufPos);
- bd->writePos=pos;
- bd->writeCurrent=current;
- bd->writeRun=run;
- return gotcount;
- }
- }
}
+
+ /* 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;
+ return (previous!=RETVAL_LAST_BLOCK) ? previous : gotcount;
+ }
+ bd->writeCRC=~0;
+ pos=bd->writePos;
+ current=bd->writeCurrent;
+ goto decode_next_byte;
}
-/* Allocate the structure, read file header. If !len, src_fd contains
- filehandle to read from. Else inbuf contains data. */
-extern int start_bunzip(bunzip_data **bdp, int src_fd, char *inbuf, int len)
+/* 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)
{
bunzip_data *bd;
- unsigned int i,j,c;
+ unsigned int i;
+ const unsigned int BZh0=(((unsigned int)'B')<<24)+(((unsigned int)'Z')<<16)
+ +(((unsigned int)'h')<<8)+(unsigned int)'0';
/* Figure out how much data to allocate */
+
i=sizeof(bunzip_data);
- if(!len) i+=IOBUF_SIZE;
+ if(in_fd!=-1) i+=IOBUF_SIZE;
+
/* Allocate bunzip_data. Most fields initialize to zero. */
- if(!(bd=*bdp=malloc(i))) return RETVAL_OUT_OF_MEMORY;
- memset(bd,0,sizeof(bunzip_data));
- if(len) {
+
+ bd=*bdp=xzalloc(i);
+
+ /* Setup input buffer */
+
+ if(-1==(bd->in_fd=in_fd)) {
bd->inbuf=inbuf;
bd->inbufCount=len;
- bd->in_fd=-1;
- } else {
- bd->inbuf=(char *)(bd+1);
- bd->in_fd=src_fd;
- }
+ } 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;
- }
+
+ bd->crc32Table = crc32_filltable(1);
+
/* Setup for I/O error handling via longjmp */
+
i=setjmp(bd->jmpbuf);
if(i) return i;
- /* Ensure that file starts with "BZh" */
- for(i=0;i<3;i++) if(get_bits(bd,8)!="BZh"[i]) return RETVAL_NOT_BZIP_DATA;
- /* Next byte ascii '1'-'9', indicates block size in units of 100k of
+
+ /* 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. */
- i=get_bits(bd,8);
- if (i<'1' || i>'9') return RETVAL_NOT_BZIP_DATA;
- bd->dbufSize=100000*(i-'0');
- if(!(bd->dbuf=malloc(bd->dbufSize * sizeof(int))))
- return RETVAL_OUT_OF_MEMORY;
+
+ bd->dbufSize=100000*(i-BZh0);
+
+ bd->dbuf=xmalloc(bd->dbufSize * sizeof(int));
return RETVAL_OK;
}
-extern char *uncompressStream(int src_fd, int dst_fd)
+/* Example usage: decompress src_fd to dst_fd. (Stops at end of bzip data,
+ not end of file.) */
+
+USE_DESKTOP(long long) int
+uncompressStream(int src_fd, int dst_fd)
{
+ USE_DESKTOP(long long total_written = 0;)
+ char *outbuf;
bunzip_data *bd;
int i;
- if(!(i=start_bunzip(&bd,src_fd,0,0))) {
- i=write_bunzip_data(bd,dst_fd,0,0);
- if(i==RETVAL_LAST_BLOCK && bd->headerCRC==bd->totalCRC) i=RETVAL_OK;
+ outbuf=xmalloc(IOBUF_SIZE);
+ i=start_bunzip(&bd,src_fd,0,0);
+ if(!i) {
+ for (;;) {
+ if((i=read_bunzip(bd,outbuf,IOBUF_SIZE)) <= 0) break;
+ if(i!=write(dst_fd,outbuf,i)) {
+ i=RETVAL_UNEXPECTED_OUTPUT_EOF;
+ break;
+ }
+ USE_DESKTOP(total_written += i;)
+ }
}
- flush_bunzip_outbuf(bd,dst_fd);
- if(bd->dbuf) free(bd->dbuf);
+
+ /* 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) {
+ bb_error_msg("compressed file ends unexpectedly");
+ } else {
+ bb_error_msg("decompression failed");
+ }
+ free(bd->dbuf);
free(bd);
- return bunzip_errors[-i];
-}
+ free(outbuf);
-/* This new version is not yet properly integrated with tar */
-extern ssize_t read_bz2(int fd, void *buf, size_t count)
-{
-#warning FIXME
- return(0);
+ return i ? i : USE_DESKTOP(total_written) + 0;
}
-extern void BZ2_bzReadOpen(int fd, void *unused, int nUnused)
-{
-#warning FIXME
- return;
-}
-extern void BZ2_bzReadClose(void)
-{
-#warning FIXME
-}
+#ifdef TESTING
+
+static char * const bunzip_errors[]={NULL,"Bad file checksum","Not bzip data",
+ "Unexpected input EOF","Unexpected output EOF","Data error",
+ "Out of memory","Obsolete (pre 0.9.5) bzip format not supported."};
-#if 0
/* Dumb little test thing, decompress stdin to stdout */
int main(int argc, char *argv[])
{
- char *c=uncompressStream(0,1);
- fprintf(stderr,"\n%s\n", c ? c : "Completed OK");
+ int i=uncompressStream(0,1);
+ char c;
+
+ if(i<0) fprintf(stderr,"%s\n", bunzip_errors[-i]);
+ else if(read(0,&c,1)) fprintf(stderr,"Trailing garbage ignored\n");
+ return -i;
}
#endif
projects / oweals / busybox.git / blobdiff