memdisk/inflate.c

   1 #define DEBG(x)
   2 #define DEBG1(x)
   3 /* inflate.c -- Not copyrighted 1992 by Mark Adler
   4    version c10p1, 10 January 1993 */
   5
   6 /*
   7  * Adapted for booting Linux by Hannu Savolainen 1993
   8  * based on gzip-1.0.3
   9  *
  10  * Nicolas Pitre <nico@cam.org>, 1999/04/14 :
  11  *   Little mods for all variable to reside either into rodata or bss segments
  12  *   by marking constant variables with 'const' and initializing all the others
  13  *   at run-time only.  This allows for the kernel uncompressor to run
  14  *   directly from Flash or ROM memory on embedded systems.
  15  *
  16  * Adapted for MEMDISK by H. Peter Anvin, April 2003
  17  */
  18
  19 /*
  20    Inflate deflated (PKZIP's method 8 compressed) data.  The compression
  21    method searches for as much of the current string of bytes (up to a
  22    length of 258) in the previous 32 K bytes.  If it doesn't find any
  23    matches (of at least length 3), it codes the next byte.  Otherwise, it
  24    codes the length of the matched string and its distance backwards from
  25    the current position.  There is a single Huffman code that codes both
  26    single bytes (called "literals") and match lengths.  A second Huffman
  27    code codes the distance information, which follows a length code.  Each
  28    length or distance code actually represents a base value and a number
  29    of "extra" (sometimes zero) bits to get to add to the base value.  At
  30    the end of each deflated block is a special end-of-block (EOB) literal/
  31    length code.  The decoding process is basically: get a literal/length
  32    code; if EOB then done; if a literal, emit the decoded byte; if a
  33    length then get the distance and emit the referred-to bytes from the
  34    sliding window of previously emitted data.
  35
  36    There are (currently) three kinds of inflate blocks: stored, fixed, and
  37    dynamic.  The compressor deals with some chunk of data at a time, and
  38    decides which method to use on a chunk-by-chunk basis.  A chunk might
  39    typically be 32 K or 64 K.  If the chunk is incompressible, then the
  40    "stored" method is used.  In this case, the bytes are simply stored as
  41    is, eight bits per byte, with none of the above coding.  The bytes are
  42    preceded by a count, since there is no longer an EOB code.
  43
  44    If the data is compressible, then either the fixed or dynamic methods
  45    are used.  In the dynamic method, the compressed data is preceded by
  46    an encoding of the literal/length and distance Huffman codes that are
  47    to be used to decode this block.  The representation is itself Huffman
  48    coded, and so is preceded by a description of that code.  These code
  49    descriptions take up a little space, and so for small blocks, there is
  50    a predefined set of codes, called the fixed codes.  The fixed method is
  51    used if the block codes up smaller that way (usually for quite small
  52    chunks), otherwise the dynamic method is used.  In the latter case, the
  53    codes are customized to the probabilities in the current block, and so
  54    can code it much better than the pre-determined fixed codes.
  55
  56    The Huffman codes themselves are decoded using a multi-level table
  57    lookup, in order to maximize the speed of decoding plus the speed of
  58    building the decoding tables.  See the comments below that precede the
  59    lbits and dbits tuning parameters.
  60  */
  61
  62
  63 /*
  64    Notes beyond the 1.93a appnote.txt:
  65
  66    1. Distance pointers never point before the beginning of the output
  67       stream.
  68    2. Distance pointers can point back across blocks, up to 32k away.
  69    3. There is an implied maximum of 7 bits for the bit length table and
  70       15 bits for the actual data.
  71    4. If only one code exists, then it is encoded using one bit.  (Zero
  72       would be more efficient, but perhaps a little confusing.)  If two
  73       codes exist, they are coded using one bit each (0 and 1).
  74    5. There is no way of sending zero distance codes--a dummy must be
  75       sent if there are none.  (History: a pre 2.0 version of PKZIP would
  76       store blocks with no distance codes, but this was discovered to be
  77       too harsh a criterion.)  Valid only for 1.93a.  2.04c does allow
  78       zero distance codes, which is sent as one code of zero bits in
  79       length.
  80    6. There are up to 286 literal/length codes.  Code 256 represents the
  81       end-of-block.  Note however that the static length tree defines
  82       288 codes just to fill out the Huffman codes.  Codes 286 and 287
  83       cannot be used though, since there is no length base or extra bits
  84       defined for them.  Similarly, there are up to 30 distance codes.
  85       However, static trees define 32 codes (all 5 bits) to fill out the
  86       Huffman codes, but the last two had better not show up in the data.
  87    7. Unzip can check dynamic Huffman blocks for complete code sets.
  88       The exception is that a single code would not be complete (see #4).
  89    8. The five bits following the block type is really the number of
  90       literal codes sent minus 257.
  91    9. Length codes 8,16,16 are interpreted as 13 length codes of 8 bits
  92       (1+6+6).  Therefore, to output three times the length, you output
  93       three codes (1+1+1), whereas to output four times the same length,
  94       you only need two codes (1+3).  Hmm.
  95   10. In the tree reconstruction algorithm, Code = Code + Increment
  96       only if BitLength(i) is not zero.  (Pretty obvious.)
  97   11. Correction: 4 Bits: # of Bit Length codes - 4     (4 - 19)
  98   12. Note: length code 284 can represent 227-258, but length code 285
  99       really is 258.  The last length deserves its own, short code
 100       since it gets used a lot in very redundant files.  The length
 101       258 is special since 258 - 3 (the min match length) is 255.
 102   13. The literal/length and distance code bit lengths are read as a
 103       single stream of lengths.  It is possible (and advantageous) for
 104       a repeat code (16, 17, or 18) to go across the boundary between
 105       the two sets of lengths.
 106  */
 107
 108 #ifdef RCSID
 109 static char rcsid[] = "#Id: inflate.c,v 0.14 1993/06/10 13:27:04 jloup Exp #";
 110 #endif
 111
 112 #define slide window
 113
 114 /* Huffman code lookup table entry--this entry is four bytes for machines
 115    that have 16-bit pointers (e.g. PC's in the small or medium model).
 116    Valid extra bits are 0..13.  e == 15 is EOB (end of block), e == 16
 117    means that v is a literal, 16 < e < 32 means that v is a pointer to
 118    the next table, which codes e - 16 bits, and lastly e == 99 indicates
 119    an unused code.  If a code with e == 99 is looked up, this implies an
 120    error in the data. */
 121 struct huft {
 122   uch e;                /* number of extra bits or operation */
 123   uch b;                /* number of bits in this code or subcode */
 124   union {
 125     ush n;              /* literal, length base, or distance base */
 126     struct huft *t;     /* pointer to next level of table */
 127   } v;
 128 };
 129
 130
 131 /* Function prototypes */
 132 STATIC int huft_build OF((unsigned *, unsigned, unsigned,
 133                 const ush *, const ush *, struct huft **, int *));
 134 STATIC int huft_free OF((struct huft *));
 135 STATIC int inflate_codes OF((struct huft *, struct huft *, int, int));
 136 STATIC int inflate_stored OF((void));
 137 STATIC int inflate_fixed OF((void));
 138 STATIC int inflate_dynamic OF((void));
 139 STATIC int inflate_block OF((int *));
 140 STATIC int inflate OF((void));
 141
 142
 143 /* The inflate algorithm uses a sliding 32 K byte window on the uncompressed
 144    stream to find repeated byte strings.  This is implemented here as a
 145    circular buffer.  The index is updated simply by incrementing and then
 146    ANDing with 0x7fff (32K-1). */
 147 /* It is left to other modules to supply the 32 K area.  It is assumed
 148    to be usable as if it were declared "uch slide[32768];" or as just
 149    "uch *slide;" and then malloc'ed in the latter case.  The definition
 150    must be in unzip.h, included above. */
 151 /* unsigned wp;             current position in slide */
 152 #define wp outcnt
 153 #define flush_output(w) (wp=(w),flush_window())
 154
 155 /* Tables for deflate from PKZIP's appnote.txt. */
 156 static const unsigned border[] = {    /* Order of the bit length code lengths */
 157         16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15};
 158 static const ush cplens[] = {         /* Copy lengths for literal codes 257..285 */
 159         3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
 160         35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0};
 161         /* note: see note #13 above about the 258 in this list. */
 162 static const ush cplext[] = {         /* Extra bits for literal codes 257..285 */
 163         0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2,
 164         3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 0, 99, 99}; /* 99==invalid */
 165 static const ush cpdist[] = {         /* Copy offsets for distance codes 0..29 */
 166         1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
 167         257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,
 168         8193, 12289, 16385, 24577};
 169 static const ush cpdext[] = {         /* Extra bits for distance codes */
 170         0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6,
 171         7, 7, 8, 8, 9, 9, 10, 10, 11, 11,
 172         12, 12, 13, 13};
 173
 174
 175
 176 /* Macros for inflate() bit peeking and grabbing.
 177    The usage is:
 178
 179         NEEDBITS(j)
 180         x = b & mask_bits[j];
 181         DUMPBITS(j)
 182
 183    where NEEDBITS makes sure that b has at least j bits in it, and
 184    DUMPBITS removes the bits from b.  The macros use the variable k
 185    for the number of bits in b.  Normally, b and k are register
 186    variables for speed, and are initialized at the beginning of a
 187    routine that uses these macros from a global bit buffer and count.
 188
 189    If we assume that EOB will be the longest code, then we will never
 190    ask for bits with NEEDBITS that are beyond the end of the stream.
 191    So, NEEDBITS should not read any more bytes than are needed to
 192    meet the request.  Then no bytes need to be "returned" to the buffer
 193    at the end of the last block.
 194
 195    However, this assumption is not true for fixed blocks--the EOB code
 196    is 7 bits, but the other literal/length codes can be 8 or 9 bits.
 197    (The EOB code is shorter than other codes because fixed blocks are
 198    generally short.  So, while a block always has an EOB, many other
 199    literal/length codes have a significantly lower probability of
 200    showing up at all.)  However, by making the first table have a
 201    lookup of seven bits, the EOB code will be found in that first
 202    lookup, and so will not require that too many bits be pulled from
 203    the stream.
 204  */
 205
 206 STATIC ulg bb;                         /* bit buffer */
 207 STATIC unsigned bk;                    /* bits in bit buffer */
 208
 209 STATIC const ush mask_bits[] = {
 210     0x0000,
 211     0x0001, 0x0003, 0x0007, 0x000f, 0x001f, 0x003f, 0x007f, 0x00ff,
 212     0x01ff, 0x03ff, 0x07ff, 0x0fff, 0x1fff, 0x3fff, 0x7fff, 0xffff
 213 };
 214
 215 #define NEXTBYTE()  (uch)get_byte()
 216 #define NEEDBITS(n) {while(k<(n)){b|=((ulg)NEXTBYTE())<<k;k+=8;}}
 217 #define DUMPBITS(n) {b>>=(n);k-=(n);}
 218
 219
 220 /*
 221    Huffman code decoding is performed using a multi-level table lookup.
 222    The fastest way to decode is to simply build a lookup table whose
 223    size is determined by the longest code.  However, the time it takes
 224    to build this table can also be a factor if the data being decoded
 225    is not very long.  The most common codes are necessarily the
 226    shortest codes, so those codes dominate the decoding time, and hence
 227    the speed.  The idea is you can have a shorter table that decodes the
 228    shorter, more probable codes, and then point to subsidiary tables for
 229    the longer codes.  The time it costs to decode the longer codes is
 230    then traded against the time it takes to make longer tables.
 231
 232    This results of this trade are in the variables lbits and dbits
 233    below.  lbits is the number of bits the first level table for literal/
 234    length codes can decode in one step, and dbits is the same thing for
 235    the distance codes.  Subsequent tables are also less than or equal to
 236    those sizes.  These values may be adjusted either when all of the
 237    codes are shorter than that, in which case the longest code length in
 238    bits is used, or when the shortest code is *longer* than the requested
 239    table size, in which case the length of the shortest code in bits is
 240    used.
 241
 242    There are two different values for the two tables, since they code a
 243    different number of possibilities each.  The literal/length table
 244    codes 286 possible values, or in a flat code, a little over eight
 245    bits.  The distance table codes 30 possible values, or a little less
 246    than five bits, flat.  The optimum values for speed end up being
 247    about one bit more than those, so lbits is 8+1 and dbits is 5+1.
 248    The optimum values may differ though from machine to machine, and
 249    possibly even between compilers.  Your mileage may vary.
 250  */
 251
 252
 253 STATIC const int lbits = 9;          /* bits in base literal/length lookup table */
 254 STATIC const int dbits = 6;          /* bits in base distance lookup table */
 255
 256
 257 /* If BMAX needs to be larger than 16, then h and x[] should be ulg. */
 258 #define BMAX 16         /* maximum bit length of any code (16 for explode) */
 259 #define N_MAX 288       /* maximum number of codes in any set */
 260
 261
 262 STATIC unsigned hufts;         /* track memory usage */
 263
 264
 265 STATIC int huft_build(b, n, s, d, e, t, m)
 266 unsigned *b;            /* code lengths in bits (all assumed <= BMAX) */
 267 unsigned n;             /* number of codes (assumed <= N_MAX) */
 268 unsigned s;             /* number of simple-valued codes (0..s-1) */
 269 const ush *d;                 /* list of base values for non-simple codes */
 270 const ush *e;                 /* list of extra bits for non-simple codes */
 271 struct huft **t;        /* result: starting table */
 272 int *m;                 /* maximum lookup bits, returns actual */
 273 /* Given a list of code lengths and a maximum table size, make a set of
 274    tables to decode that set of codes.  Return zero on success, one if
 275    the given code set is incomplete (the tables are still built in this
 276    case), two if the input is invalid (all zero length codes or an
 277    oversubscribed set of lengths), and three if not enough memory. */
 278 {
 279   unsigned a;                   /* counter for codes of length k */
 280   unsigned c[BMAX+1];           /* bit length count table */
 281   unsigned f;                   /* i repeats in table every f entries */
 282   int g;                        /* maximum code length */
 283   int h;                        /* table level */
 284   register unsigned i;          /* counter, current code */
 285   register unsigned j;          /* counter */
 286   register int k;               /* number of bits in current code */
 287   int l;                        /* bits per table (returned in m) */
 288   register unsigned *p;         /* pointer into c[], b[], or v[] */
 289   register struct huft *q;      /* points to current table */
 290   struct huft r;                /* table entry for structure assignment */
 291   struct huft *u[BMAX];         /* table stack */
 292   unsigned v[N_MAX];            /* values in order of bit length */
 293   register int w;               /* bits before this table == (l * h) */
 294   unsigned x[BMAX+1];           /* bit offsets, then code stack */
 295   unsigned *xp;                 /* pointer into x */
 296   int y;                        /* number of dummy codes added */
 297   unsigned z;                   /* number of entries in current table */
 298
 299 DEBG("huft1 ");
 300
 301   /* Generate counts for each bit length */
 302   memzero(c, sizeof(c));
 303   p = b;  i = n;
 304   do {
 305     Tracecv(*p, (stderr, (n-i >= ' ' && n-i <= '~' ? "%c %d\n" : "0x%x %d\n"),
 306             n-i, *p));
 307     c[*p]++;                    /* assume all entries <= BMAX */
 308     p++;                      /* Can't combine with above line (Solaris bug) */
 309   } while (--i);
 310   if (c[0] == n)                /* null input--all zero length codes */
 311   {
 312     *t = (struct huft *)NULL;
 313     *m = 0;
 314     return 0;
 315   }
 316
 317 DEBG("huft2 ");
 318
 319   /* Find minimum and maximum length, bound *m by those */
 320   l = *m;
 321   for (j = 1; j <= BMAX; j++)
 322     if (c[j])
 323       break;
 324   k = j;                        /* minimum code length */
 325   if ((unsigned)l < j)
 326     l = j;
 327   for (i = BMAX; i; i--)
 328     if (c[i])
 329       break;
 330   g = i;                        /* maximum code length */
 331   if ((unsigned)l > i)
 332     l = i;
 333   *m = l;
 334
 335 DEBG("huft3 ");
 336
 337   /* Adjust last length count to fill out codes, if needed */
 338   for (y = 1 << j; j < i; j++, y <<= 1)
 339     if ((y -= c[j]) < 0)
 340       return 2;                 /* bad input: more codes than bits */
 341   if ((y -= c[i]) < 0)
 342     return 2;
 343   c[i] += y;
 344
 345 DEBG("huft4 ");
 346
 347   /* Generate starting offsets into the value table for each length */
 348   x[1] = j = 0;
 349   p = c + 1;  xp = x + 2;
 350   while (--i) {                 /* note that i == g from above */
 351     *xp++ = (j += *p++);
 352   }
 353
 354 DEBG("huft5 ");
 355
 356   /* Make a table of values in order of bit lengths */
 357   p = b;  i = 0;
 358   do {
 359     if ((j = *p++) != 0)
 360       v[x[j]++] = i;
 361   } while (++i < n);
 362
 363 DEBG("h6 ");
 364
 365   /* Generate the Huffman codes and for each, make the table entries */
 366   x[0] = i = 0;                 /* first Huffman code is zero */
 367   p = v;                        /* grab values in bit order */
 368   h = -1;                       /* no tables yet--level -1 */
 369   w = -l;                       /* bits decoded == (l * h) */
 370   u[0] = (struct huft *)NULL;   /* just to keep compilers happy */
 371   q = (struct huft *)NULL;      /* ditto */
 372   z = 0;                        /* ditto */
 373 DEBG("h6a ");
 374
 375   /* go through the bit lengths (k already is bits in shortest code) */
 376   for (; k <= g; k++)
 377   {
 378 DEBG("h6b ");
 379     a = c[k];
 380     while (a--)
 381     {
 382 DEBG("h6b1 ");
 383       /* here i is the Huffman code of length k bits for value *p */
 384       /* make tables up to required level */
 385       while (k > w + l)
 386       {
 387 DEBG1("1 ");
 388         h++;
 389         w += l;                 /* previous table always l bits */
 390
 391         /* compute minimum size table less than or equal to l bits */
 392         z = (z = g - w) > (unsigned)l ? l : z;  /* upper limit on table size */
 393         if ((f = 1 << (j = k - w)) > a + 1)     /* try a k-w bit table */
 394         {                       /* too few codes for k-w bit table */
 395 DEBG1("2 ");
 396           f -= a + 1;           /* deduct codes from patterns left */
 397           xp = c + k;
 398           while (++j < z)       /* try smaller tables up to z bits */
 399           {
 400             if ((f <<= 1) <= *++xp)
 401               break;            /* enough codes to use up j bits */
 402             f -= *xp;           /* else deduct codes from patterns */
 403           }
 404         }
 405 DEBG1("3 ");
 406         z = 1 << j;             /* table entries for j-bit table */
 407
 408         /* allocate and link in new table */
 409         if ((q = (struct huft *)malloc((z + 1)*sizeof(struct huft))) ==
 410             (struct huft *)NULL)
 411         {
 412           if (h)
 413             huft_free(u[0]);
 414           return 3;             /* not enough memory */
 415         }
 416 DEBG1("4 ");
 417         hufts += z + 1;         /* track memory usage */
 418         *t = q + 1;             /* link to list for huft_free() */
 419         *(t = &(q->v.t)) = (struct huft *)NULL;
 420         u[h] = ++q;             /* table starts after link */
 421
 422 DEBG1("5 ");
 423         /* connect to last table, if there is one */
 424         if (h)
 425         {
 426           x[h] = i;             /* save pattern for backing up */
 427           r.b = (uch)l;         /* bits to dump before this table */
 428           r.e = (uch)(16 + j);  /* bits in this table */
 429           r.v.t = q;            /* pointer to this table */
 430           j = i >> (w - l);     /* (get around Turbo C bug) */
 431           u[h-1][j] = r;        /* connect to last table */
 432         }
 433 DEBG1("6 ");
 434       }
 435 DEBG("h6c ");
 436
 437       /* set up table entry in r */
 438       r.b = (uch)(k - w);
 439       if (p >= v + n)
 440         r.e = 99;               /* out of values--invalid code */
 441       else if (*p < s)
 442       {
 443         r.e = (uch)(*p < 256 ? 16 : 15);    /* 256 is end-of-block code */
 444         r.v.n = (ush)(*p);             /* simple code is just the value */
 445         p++;                           /* one compiler does not like *p++ */
 446       }
 447       else
 448       {
 449         r.e = (uch)e[*p - s];   /* non-simple--look up in lists */
 450         r.v.n = d[*p++ - s];
 451       }
 452 DEBG("h6d ");
 453
 454       /* fill code-like entries with r */
 455       f = 1 << (k - w);
 456       for (j = i >> w; j < z; j += f)
 457         q[j] = r;
 458
 459       /* backwards increment the k-bit code i */
 460       for (j = 1 << (k - 1); i & j; j >>= 1)
 461         i ^= j;
 462       i ^= j;
 463
 464       /* backup over finished tables */
 465       while ((i & ((1 << w) - 1)) != x[h])
 466       {
 467         h--;                    /* don't need to update q */
 468         w -= l;
 469       }
 470 DEBG("h6e ");
 471     }
 472 DEBG("h6f ");
 473   }
 474
 475 DEBG("huft7 ");
 476
 477   /* Return true (1) if we were given an incomplete table */
 478   return y != 0 && g != 1;
 479 }
 480
 481
 482
 483 STATIC int huft_free(t)
 484 struct huft *t;         /* table to free */
 485 /* Free the malloc'ed tables built by huft_build(), which makes a linked
 486    list of the tables it made, with the links in a dummy first entry of
 487    each table. */
 488 {
 489   register struct huft *p, *q;
 490
 491
 492   /* Go through linked list, freeing from the malloced (t[-1]) address. */
 493   p = t;
 494   while (p != (struct huft *)NULL)
 495   {
 496     q = (--p)->v.t;
 497     free((char*)p);
 498     p = q;
 499   }
 500   return 0;
 501 }
 502
 503
 504 STATIC int inflate_codes(tl, td, bl, bd)
 505 struct huft *tl, *td;   /* literal/length and distance decoder tables */
 506 int bl, bd;             /* number of bits decoded by tl[] and td[] */
 507 /* inflate (decompress) the codes in a deflated (compressed) block.
 508    Return an error code or zero if it all goes ok. */
 509 {
 510   register unsigned e;  /* table entry flag/number of extra bits */
 511   unsigned n, d;        /* length and index for copy */
 512   unsigned w;           /* current window position */
 513   struct huft *t;       /* pointer to table entry */
 514   unsigned ml, md;      /* masks for bl and bd bits */
 515   register ulg b;       /* bit buffer */
 516   register unsigned k;  /* number of bits in bit buffer */
 517
 518
 519   /* make local copies of globals */
 520   b = bb;                       /* initialize bit buffer */
 521   k = bk;
 522   w = wp;                       /* initialize window position */
 523
 524   /* inflate the coded data */
 525   ml = mask_bits[bl];           /* precompute masks for speed */
 526   md = mask_bits[bd];
 527   for (;;)                      /* do until end of block */
 528   {
 529     NEEDBITS((unsigned)bl)
 530     if ((e = (t = tl + ((unsigned)b & ml))->e) > 16)
 531       do {
 532         if (e == 99)
 533           return 1;
 534         DUMPBITS(t->b)
 535         e -= 16;
 536         NEEDBITS(e)
 537       } while ((e = (t = t->v.t + ((unsigned)b & mask_bits[e]))->e) > 16);
 538     DUMPBITS(t->b)
 539     if (e == 16)                /* then it's a literal */
 540     {
 541       slide[w++] = (uch)t->v.n;
 542       Tracevv((stderr, "%c", slide[w-1]));
 543       if (w == WSIZE)
 544       {
 545         flush_output(w);
 546         w = 0;
 547       }
 548     }
 549     else                        /* it's an EOB or a length */
 550     {
 551       /* exit if end of block */
 552       if (e == 15)
 553         break;
 554
 555       /* get length of block to copy */
 556       NEEDBITS(e)
 557       n = t->v.n + ((unsigned)b & mask_bits[e]);
 558       DUMPBITS(e);
 559
 560       /* decode distance of block to copy */
 561       NEEDBITS((unsigned)bd)
 562       if ((e = (t = td + ((unsigned)b & md))->e) > 16)
 563         do {
 564           if (e == 99)
 565             return 1;
 566           DUMPBITS(t->b)
 567           e -= 16;
 568           NEEDBITS(e)
 569         } while ((e = (t = t->v.t + ((unsigned)b & mask_bits[e]))->e) > 16);
 570       DUMPBITS(t->b)
 571       NEEDBITS(e)
 572       d = w - t->v.n - ((unsigned)b & mask_bits[e]);
 573       DUMPBITS(e)
 574       Tracevv((stderr,"\\[%d,%d]", w-d, n));
 575
 576       /* do the copy */
 577       do {
 578         n -= (e = (e = WSIZE - ((d &= WSIZE-1) > w ? d : w)) > n ? n : e);
 579 #if !defined(NOMEMCPY) && !defined(DEBUG)
 580         if (w - d >= e)         /* (this test assumes unsigned comparison) */
 581         {
 582           memcpy(slide + w, slide + d, e);
 583           w += e;
 584           d += e;
 585         }
 586         else                      /* do it slow to avoid memcpy() overlap */
 587 #endif /* !NOMEMCPY */
 588           do {
 589             slide[w++] = slide[d++];
 590             Tracevv((stderr, "%c", slide[w-1]));
 591           } while (--e);
 592         if (w == WSIZE)
 593         {
 594           flush_output(w);
 595           w = 0;
 596         }
 597       } while (n);
 598     }
 599   }
 600
 601
 602   /* restore the globals from the locals */
 603   wp = w;                       /* restore global window pointer */
 604   bb = b;                       /* restore global bit buffer */
 605   bk = k;
 606
 607   /* done */
 608   return 0;
 609 }
 610
 611
 612
 613 STATIC int inflate_stored()
 614 /* "decompress" an inflated type 0 (stored) block. */
 615 {
 616   unsigned n;           /* number of bytes in block */
 617   unsigned w;           /* current window position */
 618   register ulg b;       /* bit buffer */
 619   register unsigned k;  /* number of bits in bit buffer */
 620
 621 DEBG("<stor");
 622
 623   /* make local copies of globals */
 624   b = bb;                       /* initialize bit buffer */
 625   k = bk;
 626   w = wp;                       /* initialize window position */
 627
 628
 629   /* go to byte boundary */
 630   n = k & 7;
 631   DUMPBITS(n);
 632
 633
 634   /* get the length and its complement */
 635   NEEDBITS(16)
 636   n = ((unsigned)b & 0xffff);
 637   DUMPBITS(16)
 638   NEEDBITS(16)
 639   if (n != (unsigned)((~b) & 0xffff))
 640     return 1;                   /* error in compressed data */
 641   DUMPBITS(16)
 642
 643
 644   /* read and output the compressed data */
 645   while (n--)
 646   {
 647     NEEDBITS(8)
 648     slide[w++] = (uch)b;
 649     if (w == WSIZE)
 650     {
 651       flush_output(w);
 652       w = 0;
 653     }
 654     DUMPBITS(8)
 655   }
 656
 657
 658   /* restore the globals from the locals */
 659   wp = w;                       /* restore global window pointer */
 660   bb = b;                       /* restore global bit buffer */
 661   bk = k;
 662
 663   DEBG(">");
 664   return 0;
 665 }
 666
 667
 668
 669 STATIC int inflate_fixed()
 670 /* decompress an inflated type 1 (fixed Huffman codes) block.  We should
 671    either replace this with a custom decoder, or at least precompute the
 672    Huffman tables. */
 673 {
 674   int i;                /* temporary variable */
 675   struct huft *tl;      /* literal/length code table */
 676   struct huft *td;      /* distance code table */
 677   int bl;               /* lookup bits for tl */
 678   int bd;               /* lookup bits for td */
 679   unsigned l[288];      /* length list for huft_build */
 680
 681 DEBG("<fix");
 682
 683   /* set up literal table */
 684   for (i = 0; i < 144; i++)
 685     l[i] = 8;
 686   for (; i < 256; i++)
 687     l[i] = 9;
 688   for (; i < 280; i++)
 689     l[i] = 7;
 690   for (; i < 288; i++)          /* make a complete, but wrong code set */
 691     l[i] = 8;
 692   bl = 7;
 693   if ((i = huft_build(l, 288, 257, cplens, cplext, &tl, &bl)) != 0)
 694     return i;
 695
 696
 697   /* set up distance table */
 698   for (i = 0; i < 30; i++)      /* make an incomplete code set */
 699     l[i] = 5;
 700   bd = 5;
 701   if ((i = huft_build(l, 30, 0, cpdist, cpdext, &td, &bd)) > 1)
 702   {
 703     huft_free(tl);
 704
 705     DEBG(">");
 706     return i;
 707   }
 708
 709
 710   /* decompress until an end-of-block code */
 711   if (inflate_codes(tl, td, bl, bd))
 712     return 1;
 713
 714
 715   /* free the decoding tables, return */
 716   huft_free(tl);
 717   huft_free(td);
 718   return 0;
 719 }
 720
 721
 722
 723 STATIC int inflate_dynamic()
 724 /* decompress an inflated type 2 (dynamic Huffman codes) block. */
 725 {
 726   int i;                /* temporary variables */
 727   unsigned j;
 728   unsigned l;           /* last length */
 729   unsigned m;           /* mask for bit lengths table */
 730   unsigned n;           /* number of lengths to get */
 731   struct huft *tl;      /* literal/length code table */
 732   struct huft *td;      /* distance code table */
 733   int bl;               /* lookup bits for tl */
 734   int bd;               /* lookup bits for td */
 735   unsigned nb;          /* number of bit length codes */
 736   unsigned nl;          /* number of literal/length codes */
 737   unsigned nd;          /* number of distance codes */
 738 #ifdef PKZIP_BUG_WORKAROUND
 739   unsigned ll[288+32];  /* literal/length and distance code lengths */
 740 #else
 741   unsigned ll[286+30];  /* literal/length and distance code lengths */
 742 #endif
 743   register ulg b;       /* bit buffer */
 744   register unsigned k;  /* number of bits in bit buffer */
 745
 746 DEBG("<dyn");
 747
 748   /* make local bit buffer */
 749   b = bb;
 750   k = bk;
 751
 752
 753   /* read in table lengths */
 754   NEEDBITS(5)
 755   nl = 257 + ((unsigned)b & 0x1f);      /* number of literal/length codes */
 756   DUMPBITS(5)
 757   NEEDBITS(5)
 758   nd = 1 + ((unsigned)b & 0x1f);        /* number of distance codes */
 759   DUMPBITS(5)
 760   NEEDBITS(4)
 761   nb = 4 + ((unsigned)b & 0xf);         /* number of bit length codes */
 762   DUMPBITS(4)
 763 #ifdef PKZIP_BUG_WORKAROUND
 764   if (nl > 288 || nd > 32)
 765 #else
 766   if (nl > 286 || nd > 30)
 767 #endif
 768     return 1;                   /* bad lengths */
 769
 770 DEBG("dyn1 ");
 771
 772   /* read in bit-length-code lengths */
 773   for (j = 0; j < nb; j++)
 774   {
 775     NEEDBITS(3)
 776     ll[border[j]] = (unsigned)b & 7;
 777     DUMPBITS(3)
 778   }
 779   for (; j < 19; j++)
 780     ll[border[j]] = 0;
 781
 782 DEBG("dyn2 ");
 783
 784   /* build decoding table for trees--single level, 7 bit lookup */
 785   bl = 7;
 786   if ((i = huft_build(ll, 19, 19, NULL, NULL, &tl, &bl)) != 0)
 787   {
 788     if (i == 1)
 789       huft_free(tl);
 790     return i;                   /* incomplete code set */
 791   }
 792
 793 DEBG("dyn3 ");
 794
 795   /* read in literal and distance code lengths */
 796   n = nl + nd;
 797   m = mask_bits[bl];
 798   i = l = 0;
 799   while ((unsigned)i < n)
 800   {
 801     NEEDBITS((unsigned)bl)
 802     j = (td = tl + ((unsigned)b & m))->b;
 803     DUMPBITS(j)
 804     j = td->v.n;
 805     if (j < 16)                 /* length of code in bits (0..15) */
 806       ll[i++] = l = j;          /* save last length in l */
 807     else if (j == 16)           /* repeat last length 3 to 6 times */
 808     {
 809       NEEDBITS(2)
 810       j = 3 + ((unsigned)b & 3);
 811       DUMPBITS(2)
 812       if ((unsigned)i + j > n)
 813         return 1;
 814       while (j--)
 815         ll[i++] = l;
 816     }
 817     else if (j == 17)           /* 3 to 10 zero length codes */
 818     {
 819       NEEDBITS(3)
 820       j = 3 + ((unsigned)b & 7);
 821       DUMPBITS(3)
 822       if ((unsigned)i + j > n)
 823         return 1;
 824       while (j--)
 825         ll[i++] = 0;
 826       l = 0;
 827     }
 828     else                        /* j == 18: 11 to 138 zero length codes */
 829     {
 830       NEEDBITS(7)
 831       j = 11 + ((unsigned)b & 0x7f);
 832       DUMPBITS(7)
 833       if ((unsigned)i + j > n)
 834         return 1;
 835       while (j--)
 836         ll[i++] = 0;
 837       l = 0;
 838     }
 839   }
 840
 841 DEBG("dyn4 ");
 842
 843   /* free decoding table for trees */
 844   huft_free(tl);
 845
 846 DEBG("dyn5 ");
 847
 848   /* restore the global bit buffer */
 849   bb = b;
 850   bk = k;
 851
 852 DEBG("dyn5a ");
 853
 854   /* build the decoding tables for literal/length and distance codes */
 855   bl = lbits;
 856   if ((i = huft_build(ll, nl, 257, cplens, cplext, &tl, &bl)) != 0)
 857   {
 858 DEBG("dyn5b ");
 859     if (i == 1) {
 860       error(" incomplete literal tree");
 861       huft_free(tl);
 862     }
 863     return i;                   /* incomplete code set */
 864   }
 865 DEBG("dyn5c ");
 866   bd = dbits;
 867   if ((i = huft_build(ll + nl, nd, 0, cpdist, cpdext, &td, &bd)) != 0)
 868   {
 869 DEBG("dyn5d ");
 870     if (i == 1) {
 871       error(" incomplete distance tree");
 872 #ifdef PKZIP_BUG_WORKAROUND
 873       i = 0;
 874     }
 875 #else
 876       huft_free(td);
 877     }
 878     huft_free(tl);
 879     return i;                   /* incomplete code set */
 880 #endif
 881   }
 882
 883 DEBG("dyn6 ");
 884
 885   /* decompress until an end-of-block code */
 886   if (inflate_codes(tl, td, bl, bd))
 887     return 1;
 888
 889 DEBG("dyn7 ");
 890
 891   /* free the decoding tables, return */
 892   huft_free(tl);
 893   huft_free(td);
 894
 895   DEBG(">");
 896   return 0;
 897 }
 898
 899
 900
 901 STATIC int inflate_block(e)
 902 int *e;                 /* last block flag */
 903 /* decompress an inflated block */
 904 {
 905   unsigned t;           /* block type */
 906   register ulg b;       /* bit buffer */
 907   register unsigned k;  /* number of bits in bit buffer */
 908
 909   DEBG("<blk");
 910
 911   /* make local bit buffer */
 912   b = bb;
 913   k = bk;
 914
 915
 916   /* read in last block bit */
 917   NEEDBITS(1)
 918   *e = (int)b & 1;
 919   DUMPBITS(1)
 920
 921
 922   /* read in block type */
 923   NEEDBITS(2)
 924   t = (unsigned)b & 3;
 925   DUMPBITS(2)
 926
 927
 928   /* restore the global bit buffer */
 929   bb = b;
 930   bk = k;
 931
 932   /* inflate that block type */
 933   if (t == 2)
 934     return inflate_dynamic();
 935   if (t == 0)
 936     return inflate_stored();
 937   if (t == 1)
 938     return inflate_fixed();
 939
 940   DEBG(">");
 941
 942   /* bad block type */
 943   return 2;
 944 }
 945
 946
 947
 948 STATIC int inflate()
 949 /* decompress an inflated entry */
 950 {
 951   int e;                /* last block flag */
 952   int r;                /* result code */
 953   unsigned h;           /* maximum struct huft's malloc'ed */
 954   void *ptr;
 955
 956   /* initialize window, bit buffer */
 957   wp = 0;
 958   bk = 0;
 959   bb = 0;
 960
 961
 962   /* decompress until the last block */
 963   h = 0;
 964   do {
 965     hufts = 0;
 966     gzip_mark(&ptr);
 967     if ((r = inflate_block(&e)) != 0) {
 968       gzip_release(&ptr);
 969       return r;
 970     }
 971     gzip_release(&ptr);
 972     if (hufts > h)
 973       h = hufts;
 974   } while (!e);
 975
 976   /* Undo too much lookahead. The next read will be byte aligned so we
 977    * can discard unused bits in the last meaningful byte.
 978    */
 979   while (bk >= 8) {
 980     bk -= 8;
 981     unget_byte();
 982   }
 983
 984   /* flush out slide */
 985   flush_output(wp);
 986
 987
 988   /* return success */
 989 #ifdef DEBUG
 990   fprintf(stderr, "<%u> ", h);
 991 #endif /* DEBUG */
 992   return 0;
 993 }
 994
 995 /**********************************************************************
 996  *
 997  * The following are support routines for inflate.c
 998  *
 999  **********************************************************************/
1000
1001 static ulg crc_32_tab[256];
1002 static ulg crc;         /* initialized in makecrc() so it'll reside in bss */
1003 #define CRC_VALUE (crc ^ 0xffffffffL)
1004
1005 /*
1006  * Code to compute the CRC-32 table. Borrowed from
1007  * gzip-1.0.3/makecrc.c.
1008  */
1009
1010 static void
1011 makecrc(void)
1012 {
1013 /* Not copyrighted 1990 Mark Adler      */
1014
1015   unsigned long c;      /* crc shift register */
1016   unsigned long e;      /* polynomial exclusive-or pattern */
1017   int i;                /* counter for all possible eight bit values */
1018   int k;                /* byte being shifted into crc apparatus */
1019
1020   /* terms of polynomial defining this crc (except x^32): */
1021   static const int p[] = {0,1,2,4,5,7,8,10,11,12,16,22,23,26};
1022
1023   /* Make exclusive-or pattern from polynomial */
1024   e = 0;
1025   for (i = 0; i < sizeof(p)/sizeof(int); i++)
1026     e |= 1L << (31 - p[i]);
1027
1028   crc_32_tab[0] = 0;
1029
1030   for (i = 1; i < 256; i++)
1031   {
1032     c = 0;
1033     for (k = i | 256; k != 1; k >>= 1)
1034     {
1035       c = c & 1 ? (c >> 1) ^ e : c >> 1;
1036       if (k & 1)
1037         c ^= e;
1038     }
1039     crc_32_tab[i] = c;
1040   }
1041
1042   /* this is initialized here so this code could reside in ROM */
1043   crc = (ulg)0xffffffffL; /* shift register contents */
1044 }
1045
1046 /* gzip flag byte */
1047 #define ASCII_FLAG   0x01 /* bit 0 set: file probably ASCII text */
1048 #define CONTINUATION 0x02 /* bit 1 set: continuation of multi-part gzip file */
1049 #define EXTRA_FIELD  0x04 /* bit 2 set: extra field present */
1050 #define ORIG_NAME    0x08 /* bit 3 set: original file name present */
1051 #define COMMENT      0x10 /* bit 4 set: file comment present */
1052 #define ENCRYPTED    0x20 /* bit 5 set: file is encrypted */
1053 #define RESERVED     0xC0 /* bit 6,7:   reserved */
1054
1055 /*
1056  * Do the uncompression!
1057  */
1058 int gunzip()
1059 {
1060     int res;
1061
1062     /* Decompress */
1063     if ((res = inflate())) {
1064             switch (res) {
1065             case 0:
1066                     break;
1067             case 1:
1068                     error("invalid compressed format (err=1)");
1069                     break;
1070             case 2:
1071                     error("invalid compressed format (err=2)");
1072                     break;
1073             case 3:
1074                     error("out of memory");
1075                     break;
1076             default:
1077                     error("invalid compressed format (other)");
1078             }
1079             return -1;
1080     }
1081
1082     return 0;
1083 }