cannam@89: /* blast.c
cannam@89:  * Copyright (C) 2003 Mark Adler
cannam@89:  * For conditions of distribution and use, see copyright notice in blast.h
cannam@89:  * version 1.1, 16 Feb 2003
cannam@89:  *
cannam@89:  * blast.c decompresses data compressed by the PKWare Compression Library.
cannam@89:  * This function provides functionality similar to the explode() function of
cannam@89:  * the PKWare library, hence the name "blast".
cannam@89:  *
cannam@89:  * This decompressor is based on the excellent format description provided by
cannam@89:  * Ben Rudiak-Gould in comp.compression on August 13, 2001.  Interestingly, the
cannam@89:  * example Ben provided in the post is incorrect.  The distance 110001 should
cannam@89:  * instead be 111000.  When corrected, the example byte stream becomes:
cannam@89:  *
cannam@89:  *    00 04 82 24 25 8f 80 7f
cannam@89:  *
cannam@89:  * which decompresses to "AIAIAIAIAIAIA" (without the quotes).
cannam@89:  */
cannam@89: 
cannam@89: /*
cannam@89:  * Change history:
cannam@89:  *
cannam@89:  * 1.0  12 Feb 2003     - First version
cannam@89:  * 1.1  16 Feb 2003     - Fixed distance check for > 4 GB uncompressed data
cannam@89:  */
cannam@89: 
cannam@89: #include <setjmp.h>             /* for setjmp(), longjmp(), and jmp_buf */
cannam@89: #include "blast.h"              /* prototype for blast() */
cannam@89: 
cannam@89: #define local static            /* for local function definitions */
cannam@89: #define MAXBITS 13              /* maximum code length */
cannam@89: #define MAXWIN 4096             /* maximum window size */
cannam@89: 
cannam@89: /* input and output state */
cannam@89: struct state {
cannam@89:     /* input state */
cannam@89:     blast_in infun;             /* input function provided by user */
cannam@89:     void *inhow;                /* opaque information passed to infun() */
cannam@89:     unsigned char *in;          /* next input location */
cannam@89:     unsigned left;              /* available input at in */
cannam@89:     int bitbuf;                 /* bit buffer */
cannam@89:     int bitcnt;                 /* number of bits in bit buffer */
cannam@89: 
cannam@89:     /* input limit error return state for bits() and decode() */
cannam@89:     jmp_buf env;
cannam@89: 
cannam@89:     /* output state */
cannam@89:     blast_out outfun;           /* output function provided by user */
cannam@89:     void *outhow;               /* opaque information passed to outfun() */
cannam@89:     unsigned next;              /* index of next write location in out[] */
cannam@89:     int first;                  /* true to check distances (for first 4K) */
cannam@89:     unsigned char out[MAXWIN];  /* output buffer and sliding window */
cannam@89: };
cannam@89: 
cannam@89: /*
cannam@89:  * Return need bits from the input stream.  This always leaves less than
cannam@89:  * eight bits in the buffer.  bits() works properly for need == 0.
cannam@89:  *
cannam@89:  * Format notes:
cannam@89:  *
cannam@89:  * - Bits are stored in bytes from the least significant bit to the most
cannam@89:  *   significant bit.  Therefore bits are dropped from the bottom of the bit
cannam@89:  *   buffer, using shift right, and new bytes are appended to the top of the
cannam@89:  *   bit buffer, using shift left.
cannam@89:  */
cannam@89: local int bits(struct state *s, int need)
cannam@89: {
cannam@89:     int val;            /* bit accumulator */
cannam@89: 
cannam@89:     /* load at least need bits into val */
cannam@89:     val = s->bitbuf;
cannam@89:     while (s->bitcnt < need) {
cannam@89:         if (s->left == 0) {
cannam@89:             s->left = s->infun(s->inhow, &(s->in));
cannam@89:             if (s->left == 0) longjmp(s->env, 1);       /* out of input */
cannam@89:         }
cannam@89:         val |= (int)(*(s->in)++) << s->bitcnt;          /* load eight bits */
cannam@89:         s->left--;
cannam@89:         s->bitcnt += 8;
cannam@89:     }
cannam@89: 
cannam@89:     /* drop need bits and update buffer, always zero to seven bits left */
cannam@89:     s->bitbuf = val >> need;
cannam@89:     s->bitcnt -= need;
cannam@89: 
cannam@89:     /* return need bits, zeroing the bits above that */
cannam@89:     return val & ((1 << need) - 1);
cannam@89: }
cannam@89: 
cannam@89: /*
cannam@89:  * Huffman code decoding tables.  count[1..MAXBITS] is the number of symbols of
cannam@89:  * each length, which for a canonical code are stepped through in order.
cannam@89:  * symbol[] are the symbol values in canonical order, where the number of
cannam@89:  * entries is the sum of the counts in count[].  The decoding process can be
cannam@89:  * seen in the function decode() below.
cannam@89:  */
cannam@89: struct huffman {
cannam@89:     short *count;       /* number of symbols of each length */
cannam@89:     short *symbol;      /* canonically ordered symbols */
cannam@89: };
cannam@89: 
cannam@89: /*
cannam@89:  * Decode a code from the stream s using huffman table h.  Return the symbol or
cannam@89:  * a negative value if there is an error.  If all of the lengths are zero, i.e.
cannam@89:  * an empty code, or if the code is incomplete and an invalid code is received,
cannam@89:  * then -9 is returned after reading MAXBITS bits.
cannam@89:  *
cannam@89:  * Format notes:
cannam@89:  *
cannam@89:  * - The codes as stored in the compressed data are bit-reversed relative to
cannam@89:  *   a simple integer ordering of codes of the same lengths.  Hence below the
cannam@89:  *   bits are pulled from the compressed data one at a time and used to
cannam@89:  *   build the code value reversed from what is in the stream in order to
cannam@89:  *   permit simple integer comparisons for decoding.
cannam@89:  *
cannam@89:  * - The first code for the shortest length is all ones.  Subsequent codes of
cannam@89:  *   the same length are simply integer decrements of the previous code.  When
cannam@89:  *   moving up a length, a one bit is appended to the code.  For a complete
cannam@89:  *   code, the last code of the longest length will be all zeros.  To support
cannam@89:  *   this ordering, the bits pulled during decoding are inverted to apply the
cannam@89:  *   more "natural" ordering starting with all zeros and incrementing.
cannam@89:  */
cannam@89: local int decode(struct state *s, struct huffman *h)
cannam@89: {
cannam@89:     int len;            /* current number of bits in code */
cannam@89:     int code;           /* len bits being decoded */
cannam@89:     int first;          /* first code of length len */
cannam@89:     int count;          /* number of codes of length len */
cannam@89:     int index;          /* index of first code of length len in symbol table */
cannam@89:     int bitbuf;         /* bits from stream */
cannam@89:     int left;           /* bits left in next or left to process */
cannam@89:     short *next;        /* next number of codes */
cannam@89: 
cannam@89:     bitbuf = s->bitbuf;
cannam@89:     left = s->bitcnt;
cannam@89:     code = first = index = 0;
cannam@89:     len = 1;
cannam@89:     next = h->count + 1;
cannam@89:     while (1) {
cannam@89:         while (left--) {
cannam@89:             code |= (bitbuf & 1) ^ 1;   /* invert code */
cannam@89:             bitbuf >>= 1;
cannam@89:             count = *next++;
cannam@89:             if (code < first + count) { /* if length len, return symbol */
cannam@89:                 s->bitbuf = bitbuf;
cannam@89:                 s->bitcnt = (s->bitcnt - len) & 7;
cannam@89:                 return h->symbol[index + (code - first)];
cannam@89:             }
cannam@89:             index += count;             /* else update for next length */
cannam@89:             first += count;
cannam@89:             first <<= 1;
cannam@89:             code <<= 1;
cannam@89:             len++;
cannam@89:         }
cannam@89:         left = (MAXBITS+1) - len;
cannam@89:         if (left == 0) break;
cannam@89:         if (s->left == 0) {
cannam@89:             s->left = s->infun(s->inhow, &(s->in));
cannam@89:             if (s->left == 0) longjmp(s->env, 1);       /* out of input */
cannam@89:         }
cannam@89:         bitbuf = *(s->in)++;
cannam@89:         s->left--;
cannam@89:         if (left > 8) left = 8;
cannam@89:     }
cannam@89:     return -9;                          /* ran out of codes */
cannam@89: }
cannam@89: 
cannam@89: /*
cannam@89:  * Given a list of repeated code lengths rep[0..n-1], where each byte is a
cannam@89:  * count (high four bits + 1) and a code length (low four bits), generate the
cannam@89:  * list of code lengths.  This compaction reduces the size of the object code.
cannam@89:  * Then given the list of code lengths length[0..n-1] representing a canonical
cannam@89:  * Huffman code for n symbols, construct the tables required to decode those
cannam@89:  * codes.  Those tables are the number of codes of each length, and the symbols
cannam@89:  * sorted by length, retaining their original order within each length.  The
cannam@89:  * return value is zero for a complete code set, negative for an over-
cannam@89:  * subscribed code set, and positive for an incomplete code set.  The tables
cannam@89:  * can be used if the return value is zero or positive, but they cannot be used
cannam@89:  * if the return value is negative.  If the return value is zero, it is not
cannam@89:  * possible for decode() using that table to return an error--any stream of
cannam@89:  * enough bits will resolve to a symbol.  If the return value is positive, then
cannam@89:  * it is possible for decode() using that table to return an error for received
cannam@89:  * codes past the end of the incomplete lengths.
cannam@89:  */
cannam@89: local int construct(struct huffman *h, const unsigned char *rep, int n)
cannam@89: {
cannam@89:     int symbol;         /* current symbol when stepping through length[] */
cannam@89:     int len;            /* current length when stepping through h->count[] */
cannam@89:     int left;           /* number of possible codes left of current length */
cannam@89:     short offs[MAXBITS+1];      /* offsets in symbol table for each length */
cannam@89:     short length[256];  /* code lengths */
cannam@89: 
cannam@89:     /* convert compact repeat counts into symbol bit length list */
cannam@89:     symbol = 0;
cannam@89:     do {
cannam@89:         len = *rep++;
cannam@89:         left = (len >> 4) + 1;
cannam@89:         len &= 15;
cannam@89:         do {
cannam@89:             length[symbol++] = len;
cannam@89:         } while (--left);
cannam@89:     } while (--n);
cannam@89:     n = symbol;
cannam@89: 
cannam@89:     /* count number of codes of each length */
cannam@89:     for (len = 0; len <= MAXBITS; len++)
cannam@89:         h->count[len] = 0;
cannam@89:     for (symbol = 0; symbol < n; symbol++)
cannam@89:         (h->count[length[symbol]])++;   /* assumes lengths are within bounds */
cannam@89:     if (h->count[0] == n)               /* no codes! */
cannam@89:         return 0;                       /* complete, but decode() will fail */
cannam@89: 
cannam@89:     /* check for an over-subscribed or incomplete set of lengths */
cannam@89:     left = 1;                           /* one possible code of zero length */
cannam@89:     for (len = 1; len <= MAXBITS; len++) {
cannam@89:         left <<= 1;                     /* one more bit, double codes left */
cannam@89:         left -= h->count[len];          /* deduct count from possible codes */
cannam@89:         if (left < 0) return left;      /* over-subscribed--return negative */
cannam@89:     }                                   /* left > 0 means incomplete */
cannam@89: 
cannam@89:     /* generate offsets into symbol table for each length for sorting */
cannam@89:     offs[1] = 0;
cannam@89:     for (len = 1; len < MAXBITS; len++)
cannam@89:         offs[len + 1] = offs[len] + h->count[len];
cannam@89: 
cannam@89:     /*
cannam@89:      * put symbols in table sorted by length, by symbol order within each
cannam@89:      * length
cannam@89:      */
cannam@89:     for (symbol = 0; symbol < n; symbol++)
cannam@89:         if (length[symbol] != 0)
cannam@89:             h->symbol[offs[length[symbol]]++] = symbol;
cannam@89: 
cannam@89:     /* return zero for complete set, positive for incomplete set */
cannam@89:     return left;
cannam@89: }
cannam@89: 
cannam@89: /*
cannam@89:  * Decode PKWare Compression Library stream.
cannam@89:  *
cannam@89:  * Format notes:
cannam@89:  *
cannam@89:  * - First byte is 0 if literals are uncoded or 1 if they are coded.  Second
cannam@89:  *   byte is 4, 5, or 6 for the number of extra bits in the distance code.
cannam@89:  *   This is the base-2 logarithm of the dictionary size minus six.
cannam@89:  *
cannam@89:  * - Compressed data is a combination of literals and length/distance pairs
cannam@89:  *   terminated by an end code.  Literals are either Huffman coded or
cannam@89:  *   uncoded bytes.  A length/distance pair is a coded length followed by a
cannam@89:  *   coded distance to represent a string that occurs earlier in the
cannam@89:  *   uncompressed data that occurs again at the current location.
cannam@89:  *
cannam@89:  * - A bit preceding a literal or length/distance pair indicates which comes
cannam@89:  *   next, 0 for literals, 1 for length/distance.
cannam@89:  *
cannam@89:  * - If literals are uncoded, then the next eight bits are the literal, in the
cannam@89:  *   normal bit order in th stream, i.e. no bit-reversal is needed. Similarly,
cannam@89:  *   no bit reversal is needed for either the length extra bits or the distance
cannam@89:  *   extra bits.
cannam@89:  *
cannam@89:  * - Literal bytes are simply written to the output.  A length/distance pair is
cannam@89:  *   an instruction to copy previously uncompressed bytes to the output.  The
cannam@89:  *   copy is from distance bytes back in the output stream, copying for length
cannam@89:  *   bytes.
cannam@89:  *
cannam@89:  * - Distances pointing before the beginning of the output data are not
cannam@89:  *   permitted.
cannam@89:  *
cannam@89:  * - Overlapped copies, where the length is greater than the distance, are
cannam@89:  *   allowed and common.  For example, a distance of one and a length of 518
cannam@89:  *   simply copies the last byte 518 times.  A distance of four and a length of
cannam@89:  *   twelve copies the last four bytes three times.  A simple forward copy
cannam@89:  *   ignoring whether the length is greater than the distance or not implements
cannam@89:  *   this correctly.
cannam@89:  */
cannam@89: local int decomp(struct state *s)
cannam@89: {
cannam@89:     int lit;            /* true if literals are coded */
cannam@89:     int dict;           /* log2(dictionary size) - 6 */
cannam@89:     int symbol;         /* decoded symbol, extra bits for distance */
cannam@89:     int len;            /* length for copy */
cannam@89:     int dist;           /* distance for copy */
cannam@89:     int copy;           /* copy counter */
cannam@89:     unsigned char *from, *to;   /* copy pointers */
cannam@89:     static int virgin = 1;                              /* build tables once */
cannam@89:     static short litcnt[MAXBITS+1], litsym[256];        /* litcode memory */
cannam@89:     static short lencnt[MAXBITS+1], lensym[16];         /* lencode memory */
cannam@89:     static short distcnt[MAXBITS+1], distsym[64];       /* distcode memory */
cannam@89:     static struct huffman litcode = {litcnt, litsym};   /* length code */
cannam@89:     static struct huffman lencode = {lencnt, lensym};   /* length code */
cannam@89:     static struct huffman distcode = {distcnt, distsym};/* distance code */
cannam@89:         /* bit lengths of literal codes */
cannam@89:     static const unsigned char litlen[] = {
cannam@89:         11, 124, 8, 7, 28, 7, 188, 13, 76, 4, 10, 8, 12, 10, 12, 10, 8, 23, 8,
cannam@89:         9, 7, 6, 7, 8, 7, 6, 55, 8, 23, 24, 12, 11, 7, 9, 11, 12, 6, 7, 22, 5,
cannam@89:         7, 24, 6, 11, 9, 6, 7, 22, 7, 11, 38, 7, 9, 8, 25, 11, 8, 11, 9, 12,
cannam@89:         8, 12, 5, 38, 5, 38, 5, 11, 7, 5, 6, 21, 6, 10, 53, 8, 7, 24, 10, 27,
cannam@89:         44, 253, 253, 253, 252, 252, 252, 13, 12, 45, 12, 45, 12, 61, 12, 45,
cannam@89:         44, 173};
cannam@89:         /* bit lengths of length codes 0..15 */
cannam@89:     static const unsigned char lenlen[] = {2, 35, 36, 53, 38, 23};
cannam@89:         /* bit lengths of distance codes 0..63 */
cannam@89:     static const unsigned char distlen[] = {2, 20, 53, 230, 247, 151, 248};
cannam@89:     static const short base[16] = {     /* base for length codes */
cannam@89:         3, 2, 4, 5, 6, 7, 8, 9, 10, 12, 16, 24, 40, 72, 136, 264};
cannam@89:     static const char extra[16] = {     /* extra bits for length codes */
cannam@89:         0, 0, 0, 0, 0, 0, 0, 0, 1, 2, 3, 4, 5, 6, 7, 8};
cannam@89: 
cannam@89:     /* set up decoding tables (once--might not be thread-safe) */
cannam@89:     if (virgin) {
cannam@89:         construct(&litcode, litlen, sizeof(litlen));
cannam@89:         construct(&lencode, lenlen, sizeof(lenlen));
cannam@89:         construct(&distcode, distlen, sizeof(distlen));
cannam@89:         virgin = 0;
cannam@89:     }
cannam@89: 
cannam@89:     /* read header */
cannam@89:     lit = bits(s, 8);
cannam@89:     if (lit > 1) return -1;
cannam@89:     dict = bits(s, 8);
cannam@89:     if (dict < 4 || dict > 6) return -2;
cannam@89: 
cannam@89:     /* decode literals and length/distance pairs */
cannam@89:     do {
cannam@89:         if (bits(s, 1)) {
cannam@89:             /* get length */
cannam@89:             symbol = decode(s, &lencode);
cannam@89:             len = base[symbol] + bits(s, extra[symbol]);
cannam@89:             if (len == 519) break;              /* end code */
cannam@89: 
cannam@89:             /* get distance */
cannam@89:             symbol = len == 2 ? 2 : dict;
cannam@89:             dist = decode(s, &distcode) << symbol;
cannam@89:             dist += bits(s, symbol);
cannam@89:             dist++;
cannam@89:             if (s->first && dist > s->next)
cannam@89:                 return -3;              /* distance too far back */
cannam@89: 
cannam@89:             /* copy length bytes from distance bytes back */
cannam@89:             do {
cannam@89:                 to = s->out + s->next;
cannam@89:                 from = to - dist;
cannam@89:                 copy = MAXWIN;
cannam@89:                 if (s->next < dist) {
cannam@89:                     from += copy;
cannam@89:                     copy = dist;
cannam@89:                 }
cannam@89:                 copy -= s->next;
cannam@89:                 if (copy > len) copy = len;
cannam@89:                 len -= copy;
cannam@89:                 s->next += copy;
cannam@89:                 do {
cannam@89:                     *to++ = *from++;
cannam@89:                 } while (--copy);
cannam@89:                 if (s->next == MAXWIN) {
cannam@89:                     if (s->outfun(s->outhow, s->out, s->next)) return 1;
cannam@89:                     s->next = 0;
cannam@89:                     s->first = 0;
cannam@89:                 }
cannam@89:             } while (len != 0);
cannam@89:         }
cannam@89:         else {
cannam@89:             /* get literal and write it */
cannam@89:             symbol = lit ? decode(s, &litcode) : bits(s, 8);
cannam@89:             s->out[s->next++] = symbol;
cannam@89:             if (s->next == MAXWIN) {
cannam@89:                 if (s->outfun(s->outhow, s->out, s->next)) return 1;
cannam@89:                 s->next = 0;
cannam@89:                 s->first = 0;
cannam@89:             }
cannam@89:         }
cannam@89:     } while (1);
cannam@89:     return 0;
cannam@89: }
cannam@89: 
cannam@89: /* See comments in blast.h */
cannam@89: int blast(blast_in infun, void *inhow, blast_out outfun, void *outhow)
cannam@89: {
cannam@89:     struct state s;             /* input/output state */
cannam@89:     int err;                    /* return value */
cannam@89: 
cannam@89:     /* initialize input state */
cannam@89:     s.infun = infun;
cannam@89:     s.inhow = inhow;
cannam@89:     s.left = 0;
cannam@89:     s.bitbuf = 0;
cannam@89:     s.bitcnt = 0;
cannam@89: 
cannam@89:     /* initialize output state */
cannam@89:     s.outfun = outfun;
cannam@89:     s.outhow = outhow;
cannam@89:     s.next = 0;
cannam@89:     s.first = 1;
cannam@89: 
cannam@89:     /* return if bits() or decode() tries to read past available input */
cannam@89:     if (setjmp(s.env) != 0)             /* if came back here via longjmp(), */
cannam@89:         err = 2;                        /*  then skip decomp(), return error */
cannam@89:     else
cannam@89:         err = decomp(&s);               /* decompress */
cannam@89: 
cannam@89:     /* write any leftover output and update the error code if needed */
cannam@89:     if (err != 1 && s.next && s.outfun(s.outhow, s.out, s.next) && err == 0)
cannam@89:         err = 1;
cannam@89:     return err;
cannam@89: }
cannam@89: 
cannam@89: #ifdef TEST
cannam@89: /* Example of how to use blast() */
cannam@89: #include <stdio.h>
cannam@89: #include <stdlib.h>
cannam@89: 
cannam@89: #define CHUNK 16384
cannam@89: 
cannam@89: local unsigned inf(void *how, unsigned char **buf)
cannam@89: {
cannam@89:     static unsigned char hold[CHUNK];
cannam@89: 
cannam@89:     *buf = hold;
cannam@89:     return fread(hold, 1, CHUNK, (FILE *)how);
cannam@89: }
cannam@89: 
cannam@89: local int outf(void *how, unsigned char *buf, unsigned len)
cannam@89: {
cannam@89:     return fwrite(buf, 1, len, (FILE *)how) != len;
cannam@89: }
cannam@89: 
cannam@89: /* Decompress a PKWare Compression Library stream from stdin to stdout */
cannam@89: int main(void)
cannam@89: {
cannam@89:     int ret, n;
cannam@89: 
cannam@89:     /* decompress to stdout */
cannam@89:     ret = blast(inf, stdin, outf, stdout);
cannam@89:     if (ret != 0) fprintf(stderr, "blast error: %d\n", ret);
cannam@89: 
cannam@89:     /* see if there are any leftover bytes */
cannam@89:     n = 0;
cannam@89:     while (getchar() != EOF) n++;
cannam@89:     if (n) fprintf(stderr, "blast warning: %d unused bytes of input\n", n);
cannam@89: 
cannam@89:     /* return blast() error code */
cannam@89:     return ret;
cannam@89: }
cannam@89: #endif