subr_inflate.c revision 344379
1/* 2 * Most parts of this file are not covered by: 3 * ---------------------------------------------------------------------------- 4 * "THE BEER-WARE LICENSE" (Revision 42): 5 * <phk@FreeBSD.org> wrote this file. As long as you retain this notice you 6 * can do whatever you want with this stuff. If we meet some day, and you think 7 * this stuff is worth it, you can buy me a beer in return. Poul-Henning Kamp 8 * ---------------------------------------------------------------------------- 9 */ 10 11#include <sys/cdefs.h> 12__FBSDID("$FreeBSD: stable/11/sys/kern/subr_inflate.c 344379 2019-02-20 19:32:02Z kevans $"); 13 14#include <sys/param.h> 15#include <sys/inflate.h> 16#ifdef _KERNEL 17#include <sys/systm.h> 18#include <sys/kernel.h> 19#endif 20#include <sys/malloc.h> 21 22#ifdef _KERNEL 23static MALLOC_DEFINE(M_GZIP, "gzip_trees", "Gzip trees"); 24#endif 25 26/* needed to make inflate() work */ 27#define uch u_char 28#define ush u_short 29#define ulg u_long 30 31/* Stuff to make inflate() work */ 32#ifdef _KERNEL 33#define memzero(dest,len) bzero(dest,len) 34#endif 35#define NOMEMCPY 36#ifdef _KERNEL 37#define FPRINTF printf 38#else 39extern void putstr (char *); 40#define FPRINTF putstr 41#endif 42 43#define FLUSH(x,y) { \ 44 int foo = (*x->gz_output)(x->gz_private,x->gz_slide,y); \ 45 if (foo) \ 46 return foo; \ 47 } 48 49static const int qflag = 0; 50 51#ifndef _KERNEL /* want to use this file in kzip also */ 52extern unsigned char *kzipmalloc (int); 53extern void kzipfree (void*); 54#define malloc(x, y, z) kzipmalloc((x)) 55#define free(x, y) kzipfree((x)) 56#endif 57 58/* 59 * This came from unzip-5.12. I have changed it the flow to pass 60 * a structure pointer around, thus hopefully making it re-entrant. 61 * Poul-Henning 62 */ 63 64/* inflate.c -- put in the public domain by Mark Adler 65 version c14o, 23 August 1994 */ 66 67/* You can do whatever you like with this source file, though I would 68 prefer that if you modify it and redistribute it that you include 69 comments to that effect with your name and the date. Thank you. 70 71 History: 72 vers date who what 73 ---- --------- -------------- ------------------------------------ 74 a ~~ Feb 92 M. Adler used full (large, one-step) lookup table 75 b1 21 Mar 92 M. Adler first version with partial lookup tables 76 b2 21 Mar 92 M. Adler fixed bug in fixed-code blocks 77 b3 22 Mar 92 M. Adler sped up match copies, cleaned up some 78 b4 25 Mar 92 M. Adler added prototypes; removed window[] (now 79 is the responsibility of unzip.h--also 80 changed name to slide[]), so needs diffs 81 for unzip.c and unzip.h (this allows 82 compiling in the small model on MSDOS); 83 fixed cast of q in huft_build(); 84 b5 26 Mar 92 M. Adler got rid of unintended macro recursion. 85 b6 27 Mar 92 M. Adler got rid of nextbyte() routine. fixed 86 bug in inflate_fixed(). 87 c1 30 Mar 92 M. Adler removed lbits, dbits environment variables. 88 changed BMAX to 16 for explode. Removed 89 OUTB usage, and replaced it with flush()-- 90 this was a 20% speed improvement! Added 91 an explode.c (to replace unimplod.c) that 92 uses the huft routines here. Removed 93 register union. 94 c2 4 Apr 92 M. Adler fixed bug for file sizes a multiple of 32k. 95 c3 10 Apr 92 M. Adler reduced memory of code tables made by 96 huft_build significantly (factor of two to 97 three). 98 c4 15 Apr 92 M. Adler added NOMEMCPY do kill use of memcpy(). 99 worked around a Turbo C optimization bug. 100 c5 21 Apr 92 M. Adler added the GZ_WSIZE #define to allow reducing 101 the 32K window size for specialized 102 applications. 103 c6 31 May 92 M. Adler added some typecasts to eliminate warnings 104 c7 27 Jun 92 G. Roelofs added some more typecasts (444: MSC bug). 105 c8 5 Oct 92 J-l. Gailly added ifdef'd code to deal with PKZIP bug. 106 c9 9 Oct 92 M. Adler removed a memory error message (~line 416). 107 c10 17 Oct 92 G. Roelofs changed ULONG/UWORD/byte to ulg/ush/uch, 108 removed old inflate, renamed inflate_entry 109 to inflate, added Mark's fix to a comment. 110 c10.5 14 Dec 92 M. Adler fix up error messages for incomplete trees. 111 c11 2 Jan 93 M. Adler fixed bug in detection of incomplete 112 tables, and removed assumption that EOB is 113 the longest code (bad assumption). 114 c12 3 Jan 93 M. Adler make tables for fixed blocks only once. 115 c13 5 Jan 93 M. Adler allow all zero length codes (pkzip 2.04c 116 outputs one zero length code for an empty 117 distance tree). 118 c14 12 Mar 93 M. Adler made inflate.c standalone with the 119 introduction of inflate.h. 120 c14b 16 Jul 93 G. Roelofs added (unsigned) typecast to w at 470. 121 c14c 19 Jul 93 J. Bush changed v[N_MAX], l[288], ll[28x+3x] arrays 122 to static for Amiga. 123 c14d 13 Aug 93 J-l. Gailly de-complicatified Mark's c[*p++]++ thing. 124 c14e 8 Oct 93 G. Roelofs changed memset() to memzero(). 125 c14f 22 Oct 93 G. Roelofs renamed quietflg to qflag; made Trace() 126 conditional; added inflate_free(). 127 c14g 28 Oct 93 G. Roelofs changed l/(lx+1) macro to pointer (Cray bug) 128 c14h 7 Dec 93 C. Ghisler huft_build() optimizations. 129 c14i 9 Jan 94 A. Verheijen set fixed_t{d,l} to NULL after freeing; 130 G. Roelofs check NEXTBYTE macro for GZ_EOF. 131 c14j 23 Jan 94 G. Roelofs removed Ghisler "optimizations"; ifdef'd 132 GZ_EOF check. 133 c14k 27 Feb 94 G. Roelofs added some typecasts to avoid warnings. 134 c14l 9 Apr 94 G. Roelofs fixed split comments on preprocessor lines 135 to avoid bug in Encore compiler. 136 c14m 7 Jul 94 P. Kienitz modified to allow assembler version of 137 inflate_codes() (define ASM_INFLATECODES) 138 c14n 22 Jul 94 G. Roelofs changed fprintf to FPRINTF for DLL versions 139 c14o 23 Aug 94 C. Spieler added a newline to a debug statement; 140 G. Roelofs added another typecast to avoid MSC warning 141 */ 142 143 144/* 145 Inflate deflated (PKZIP's method 8 compressed) data. The compression 146 method searches for as much of the current string of bytes (up to a 147 length of 258) in the previous 32K bytes. If it doesn't find any 148 matches (of at least length 3), it codes the next byte. Otherwise, it 149 codes the length of the matched string and its distance backwards from 150 the current position. There is a single Huffman code that codes both 151 single bytes (called "literals") and match lengths. A second Huffman 152 code codes the distance information, which follows a length code. Each 153 length or distance code actually represents a base value and a number 154 of "extra" (sometimes zero) bits to get to add to the base value. At 155 the end of each deflated block is a special end-of-block (EOB) literal/ 156 length code. The decoding process is basically: get a literal/length 157 code; if EOB then done; if a literal, emit the decoded byte; if a 158 length then get the distance and emit the referred-to bytes from the 159 sliding window of previously emitted data. 160 161 There are (currently) three kinds of inflate blocks: stored, fixed, and 162 dynamic. The compressor outputs a chunk of data at a time and decides 163 which method to use on a chunk-by-chunk basis. A chunk might typically 164 be 32K to 64K, uncompressed. If the chunk is uncompressible, then the 165 "stored" method is used. In this case, the bytes are simply stored as 166 is, eight bits per byte, with none of the above coding. The bytes are 167 preceded by a count, since there is no longer an EOB code. 168 169 If the data is compressible, then either the fixed or dynamic methods 170 are used. In the dynamic method, the compressed data is preceded by 171 an encoding of the literal/length and distance Huffman codes that are 172 to be used to decode this block. The representation is itself Huffman 173 coded, and so is preceded by a description of that code. These code 174 descriptions take up a little space, and so for small blocks, there is 175 a predefined set of codes, called the fixed codes. The fixed method is 176 used if the block ends up smaller that way (usually for quite small 177 chunks); otherwise the dynamic method is used. In the latter case, the 178 codes are customized to the probabilities in the current block and so 179 can code it much better than the pre-determined fixed codes can. 180 181 The Huffman codes themselves are decoded using a mutli-level table 182 lookup, in order to maximize the speed of decoding plus the speed of 183 building the decoding tables. See the comments below that precede the 184 lbits and dbits tuning parameters. 185 */ 186 187 188/* 189 Notes beyond the 1.93a appnote.txt: 190 191 1. Distance pointers never point before the beginning of the output 192 stream. 193 2. Distance pointers can point back across blocks, up to 32k away. 194 3. There is an implied maximum of 7 bits for the bit length table and 195 15 bits for the actual data. 196 4. If only one code exists, then it is encoded using one bit. (Zero 197 would be more efficient, but perhaps a little confusing.) If two 198 codes exist, they are coded using one bit each (0 and 1). 199 5. There is no way of sending zero distance codes--a dummy must be 200 sent if there are none. (History: a pre 2.0 version of PKZIP would 201 store blocks with no distance codes, but this was discovered to be 202 too harsh a criterion.) Valid only for 1.93a. 2.04c does allow 203 zero distance codes, which is sent as one code of zero bits in 204 length. 205 6. There are up to 286 literal/length codes. Code 256 represents the 206 end-of-block. Note however that the static length tree defines 207 288 codes just to fill out the Huffman codes. Codes 286 and 287 208 cannot be used though, since there is no length base or extra bits 209 defined for them. Similarly, there are up to 30 distance codes. 210 However, static trees define 32 codes (all 5 bits) to fill out the 211 Huffman codes, but the last two had better not show up in the data. 212 7. Unzip can check dynamic Huffman blocks for complete code sets. 213 The exception is that a single code would not be complete (see #4). 214 8. The five bits following the block type is really the number of 215 literal codes sent minus 257. 216 9. Length codes 8,16,16 are interpreted as 13 length codes of 8 bits 217 (1+6+6). Therefore, to output three times the length, you output 218 three codes (1+1+1), whereas to output four times the same length, 219 you only need two codes (1+3). Hmm. 220 10. In the tree reconstruction algorithm, Code = Code + Increment 221 only if BitLength(i) is not zero. (Pretty obvious.) 222 11. Correction: 4 Bits: # of Bit Length codes - 4 (4 - 19) 223 12. Note: length code 284 can represent 227-258, but length code 285 224 really is 258. The last length deserves its own, short code 225 since it gets used a lot in very redundant files. The length 226 258 is special since 258 - 3 (the min match length) is 255. 227 13. The literal/length and distance code bit lengths are read as a 228 single stream of lengths. It is possible (and advantageous) for 229 a repeat code (16, 17, or 18) to go across the boundary between 230 the two sets of lengths. 231 */ 232 233 234#define PKZIP_BUG_WORKAROUND /* PKZIP 1.93a problem--live with it */ 235 236/* 237 inflate.h must supply the uch slide[GZ_WSIZE] array and the NEXTBYTE, 238 FLUSH() and memzero macros. If the window size is not 32K, it 239 should also define GZ_WSIZE. If INFMOD is defined, it can include 240 compiled functions to support the NEXTBYTE and/or FLUSH() macros. 241 There are defaults for NEXTBYTE and FLUSH() below for use as 242 examples of what those functions need to do. Normally, you would 243 also want FLUSH() to compute a crc on the data. inflate.h also 244 needs to provide these typedefs: 245 246 typedef unsigned char uch; 247 typedef unsigned short ush; 248 typedef unsigned long ulg; 249 250 This module uses the external functions malloc() and free() (and 251 probably memset() or bzero() in the memzero() macro). Their 252 prototypes are normally found in <string.h> and <stdlib.h>. 253 */ 254#define INFMOD /* tell inflate.h to include code to be 255 * compiled */ 256 257/* Huffman code lookup table entry--this entry is four bytes for machines 258 that have 16-bit pointers (e.g. PC's in the small or medium model). 259 Valid extra bits are 0..13. e == 15 is EOB (end of block), e == 16 260 means that v is a literal, 16 < e < 32 means that v is a pointer to 261 the next table, which codes e - 16 bits, and lastly e == 99 indicates 262 an unused code. If a code with e == 99 is looked up, this implies an 263 error in the data. */ 264struct huft { 265 uch e; /* number of extra bits or operation */ 266 uch b; /* number of bits in this code or subcode */ 267 union { 268 ush n; /* literal, length base, or distance 269 * base */ 270 struct huft *t; /* pointer to next level of table */ 271 } v; 272}; 273 274 275/* Function prototypes */ 276static int huft_build(struct inflate *, unsigned *, unsigned, unsigned, const ush *, const ush *, struct huft **, int *); 277static int huft_free(struct inflate *, struct huft *); 278static int inflate_codes(struct inflate *, struct huft *, struct huft *, int, int); 279static int inflate_stored(struct inflate *); 280static int xinflate(struct inflate *); 281static int inflate_fixed(struct inflate *); 282static int inflate_dynamic(struct inflate *); 283static int inflate_block(struct inflate *, int *); 284 285/* The inflate algorithm uses a sliding 32K byte window on the uncompressed 286 stream to find repeated byte strings. This is implemented here as a 287 circular buffer. The index is updated simply by incrementing and then 288 and'ing with 0x7fff (32K-1). */ 289/* It is left to other modules to supply the 32K area. It is assumed 290 to be usable as if it were declared "uch slide[32768];" or as just 291 "uch *slide;" and then malloc'ed in the latter case. The definition 292 must be in unzip.h, included above. */ 293 294 295/* Tables for deflate from PKZIP's appnote.txt. */ 296 297/* Order of the bit length code lengths */ 298static const unsigned border[] = { 299 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15}; 300 301static const ush cplens[] = { /* Copy lengths for literal codes 257..285 */ 302 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31, 303 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0}; 304 /* note: see note #13 above about the 258 in this list. */ 305 306static const ush cplext[] = { /* Extra bits for literal codes 257..285 */ 307 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2, 308 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 0, 99, 99}; /* 99==invalid */ 309 310static const ush cpdist[] = { /* Copy offsets for distance codes 0..29 */ 311 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193, 312 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145, 313 8193, 12289, 16385, 24577}; 314 315static const ush cpdext[] = { /* Extra bits for distance codes */ 316 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 317 7, 7, 8, 8, 9, 9, 10, 10, 11, 11, 318 12, 12, 13, 13}; 319 320/* And'ing with mask[n] masks the lower n bits */ 321static const ush mask[] = { 322 0x0000, 323 0x0001, 0x0003, 0x0007, 0x000f, 0x001f, 0x003f, 0x007f, 0x00ff, 324 0x01ff, 0x03ff, 0x07ff, 0x0fff, 0x1fff, 0x3fff, 0x7fff, 0xffff 325}; 326 327 328/* Macros for inflate() bit peeking and grabbing. 329 The usage is: 330 331 NEEDBITS(glbl,j) 332 x = b & mask[j]; 333 DUMPBITS(j) 334 335 where NEEDBITS makes sure that b has at least j bits in it, and 336 DUMPBITS removes the bits from b. The macros use the variable k 337 for the number of bits in b. Normally, b and k are register 338 variables for speed, and are initialized at the beginning of a 339 routine that uses these macros from a global bit buffer and count. 340 341 In order to not ask for more bits than there are in the compressed 342 stream, the Huffman tables are constructed to only ask for just 343 enough bits to make up the end-of-block code (value 256). Then no 344 bytes need to be "returned" to the buffer at the end of the last 345 block. See the huft_build() routine. 346 */ 347 348/* 349 * The following 2 were global variables. 350 * They are now fields of the inflate structure. 351 */ 352 353#define NEEDBITS(glbl,n) { \ 354 while(k<(n)) { \ 355 int c=(*glbl->gz_input)(glbl->gz_private); \ 356 if(c==GZ_EOF) \ 357 return 1; \ 358 b|=((ulg)c)<<k; \ 359 k+=8; \ 360 } \ 361 } 362 363#define DUMPBITS(n) {b>>=(n);k-=(n);} 364 365/* 366 Huffman code decoding is performed using a multi-level table lookup. 367 The fastest way to decode is to simply build a lookup table whose 368 size is determined by the longest code. However, the time it takes 369 to build this table can also be a factor if the data being decoded 370 is not very long. The most common codes are necessarily the 371 shortest codes, so those codes dominate the decoding time, and hence 372 the speed. The idea is you can have a shorter table that decodes the 373 shorter, more probable codes, and then point to subsidiary tables for 374 the longer codes. The time it costs to decode the longer codes is 375 then traded against the time it takes to make longer tables. 376 377 This results of this trade are in the variables lbits and dbits 378 below. lbits is the number of bits the first level table for literal/ 379 length codes can decode in one step, and dbits is the same thing for 380 the distance codes. Subsequent tables are also less than or equal to 381 those sizes. These values may be adjusted either when all of the 382 codes are shorter than that, in which case the longest code length in 383 bits is used, or when the shortest code is *longer* than the requested 384 table size, in which case the length of the shortest code in bits is 385 used. 386 387 There are two different values for the two tables, since they code a 388 different number of possibilities each. The literal/length table 389 codes 286 possible values, or in a flat code, a little over eight 390 bits. The distance table codes 30 possible values, or a little less 391 than five bits, flat. The optimum values for speed end up being 392 about one bit more than those, so lbits is 8+1 and dbits is 5+1. 393 The optimum values may differ though from machine to machine, and 394 possibly even between compilers. Your mileage may vary. 395 */ 396 397static const int lbits = 9; /* bits in base literal/length lookup table */ 398static const int dbits = 6; /* bits in base distance lookup table */ 399 400 401/* If BMAX needs to be larger than 16, then h and x[] should be ulg. */ 402#define BMAX 16 /* maximum bit length of any code (16 for 403 * explode) */ 404#define N_MAX 288 /* maximum number of codes in any set */ 405 406/* Given a list of code lengths and a maximum table size, make a set of 407 tables to decode that set of codes. Return zero on success, one if 408 the given code set is incomplete (the tables are still built in this 409 case), two if the input is invalid (all zero length codes or an 410 oversubscribed set of lengths), and three if not enough memory. 411 The code with value 256 is special, and the tables are constructed 412 so that no bits beyond that code are fetched when that code is 413 decoded. */ 414/* 415 * Arguments: 416 * b code lengths in bits (all assumed <= BMAX) 417 * n number of codes (assumed <= N_MAX) 418 * s number of simple-valued codes (0..s-1) 419 * d list of base values for non-simple codes 420 * e list of extra bits for non-simple codes 421 * t result: starting table 422 * m maximum lookup bits, returns actual 423 */ 424static int 425huft_build(struct inflate *glbl, unsigned *b, unsigned n, unsigned s, 426 const ush *d, const ush *e, struct huft **t, int *m) 427{ 428 unsigned a; /* counter for codes of length k */ 429 unsigned c[BMAX + 1]; /* bit length count table */ 430 unsigned el; /* length of EOB code (value 256) */ 431 unsigned f; /* i repeats in table every f entries */ 432 int g; /* maximum code length */ 433 int h; /* table level */ 434 unsigned i; /* counter, current code */ 435 unsigned j; /* counter */ 436 int k; /* number of bits in current code */ 437 int lx[BMAX + 1]; /* memory for l[-1..BMAX-1] */ 438 int *l = lx + 1; /* stack of bits per table */ 439 unsigned *p; /* pointer into c[], b[], or v[] */ 440 struct huft *q; /* points to current table */ 441 struct huft r; /* table entry for structure assignment */ 442 struct huft *u[BMAX];/* table stack */ 443 unsigned v[N_MAX]; /* values in order of bit length */ 444 int w; /* bits before this table == (l * h) */ 445 unsigned x[BMAX + 1]; /* bit offsets, then code stack */ 446 unsigned *xp; /* pointer into x */ 447 int y; /* number of dummy codes added */ 448 unsigned z; /* number of entries in current table */ 449 450 /* Generate counts for each bit length */ 451 el = n > 256 ? b[256] : BMAX; /* set length of EOB code, if any */ 452#ifdef _KERNEL 453 memzero((char *) c, sizeof(c)); 454#else 455 for (i = 0; i < BMAX+1; i++) 456 c [i] = 0; 457#endif 458 p = b; 459 i = n; 460 do { 461 c[*p]++; 462 p++; /* assume all entries <= BMAX */ 463 } while (--i); 464 if (c[0] == n) { /* null input--all zero length codes */ 465 *t = (struct huft *) NULL; 466 *m = 0; 467 return 0; 468 } 469 /* Find minimum and maximum length, bound *m by those */ 470 for (j = 1; j <= BMAX; j++) 471 if (c[j]) 472 break; 473 k = j; /* minimum code length */ 474 if ((unsigned) *m < j) 475 *m = j; 476 for (i = BMAX; i; i--) 477 if (c[i]) 478 break; 479 g = i; /* maximum code length */ 480 if ((unsigned) *m > i) 481 *m = i; 482 483 /* Adjust last length count to fill out codes, if needed */ 484 for (y = 1 << j; j < i; j++, y <<= 1) 485 if ((y -= c[j]) < 0) 486 return 2; /* bad input: more codes than bits */ 487 if ((y -= c[i]) < 0) 488 return 2; 489 c[i] += y; 490 491 /* Generate starting offsets into the value table for each length */ 492 x[1] = j = 0; 493 p = c + 1; 494 xp = x + 2; 495 while (--i) { /* note that i == g from above */ 496 *xp++ = (j += *p++); 497 } 498 499 /* Make a table of values in order of bit lengths */ 500 p = b; 501 i = 0; 502 do { 503 if ((j = *p++) != 0) 504 v[x[j]++] = i; 505 } while (++i < n); 506 507 /* Generate the Huffman codes and for each, make the table entries */ 508 x[0] = i = 0; /* first Huffman code is zero */ 509 p = v; /* grab values in bit order */ 510 h = -1; /* no tables yet--level -1 */ 511 w = l[-1] = 0; /* no bits decoded yet */ 512 u[0] = (struct huft *) NULL; /* just to keep compilers happy */ 513 q = (struct huft *) NULL; /* ditto */ 514 z = 0; /* ditto */ 515 516 /* go through the bit lengths (k already is bits in shortest code) */ 517 for (; k <= g; k++) { 518 a = c[k]; 519 while (a--) { 520 /* 521 * here i is the Huffman code of length k bits for 522 * value *p 523 */ 524 /* make tables up to required level */ 525 while (k > w + l[h]) { 526 w += l[h++]; /* add bits already decoded */ 527 528 /* 529 * compute minimum size table less than or 530 * equal to *m bits 531 */ 532 z = (z = g - w) > (unsigned) *m ? *m : z; /* upper limit */ 533 if ((f = 1 << (j = k - w)) > a + 1) { /* try a k-w bit table *//* t 534 * oo few codes for k-w 535 * bit table */ 536 f -= a + 1; /* deduct codes from 537 * patterns left */ 538 xp = c + k; 539 while (++j < z) { /* try smaller tables up 540 * to z bits */ 541 if ((f <<= 1) <= *++xp) 542 break; /* enough codes to use 543 * up j bits */ 544 f -= *xp; /* else deduct codes 545 * from patterns */ 546 } 547 } 548 if ((unsigned) w + j > el && (unsigned) w < el) 549 j = el - w; /* make EOB code end at 550 * table */ 551 z = 1 << j; /* table entries for j-bit 552 * table */ 553 l[h] = j; /* set table size in stack */ 554 555 /* allocate and link in new table */ 556 if ((q = (struct huft *) malloc((z + 1) * sizeof(struct huft), M_GZIP, M_WAITOK)) == 557 (struct huft *) NULL) { 558 if (h) 559 huft_free(glbl, u[0]); 560 return 3; /* not enough memory */ 561 } 562 glbl->gz_hufts += z + 1; /* track memory usage */ 563 *t = q + 1; /* link to list for 564 * huft_free() */ 565 *(t = &(q->v.t)) = (struct huft *) NULL; 566 u[h] = ++q; /* table starts after link */ 567 568 /* connect to last table, if there is one */ 569 if (h) { 570 x[h] = i; /* save pattern for 571 * backing up */ 572 r.b = (uch) l[h - 1]; /* bits to dump before 573 * this table */ 574 r.e = (uch) (16 + j); /* bits in this table */ 575 r.v.t = q; /* pointer to this table */ 576 j = (i & ((1 << w) - 1)) >> (w - l[h - 1]); 577 u[h - 1][j] = r; /* connect to last table */ 578 } 579 } 580 581 /* set up table entry in r */ 582 r.b = (uch) (k - w); 583 if (p >= v + n) 584 r.e = 99; /* out of values--invalid 585 * code */ 586 else if (*p < s) { 587 r.e = (uch) (*p < 256 ? 16 : 15); /* 256 is end-of-block 588 * code */ 589 r.v.n = *p++; /* simple code is just the 590 * value */ 591 } else { 592 r.e = (uch) e[*p - s]; /* non-simple--look up 593 * in lists */ 594 r.v.n = d[*p++ - s]; 595 } 596 597 /* fill code-like entries with r */ 598 f = 1 << (k - w); 599 for (j = i >> w; j < z; j += f) 600 q[j] = r; 601 602 /* backwards increment the k-bit code i */ 603 for (j = 1 << (k - 1); i & j; j >>= 1) 604 i ^= j; 605 i ^= j; 606 607 /* backup over finished tables */ 608 while ((i & ((1 << w) - 1)) != x[h]) 609 w -= l[--h]; /* don't need to update q */ 610 } 611 } 612 613 /* return actual size of base table */ 614 *m = l[0]; 615 616 /* Return true (1) if we were given an incomplete table */ 617 return y != 0 && g != 1; 618} 619 620/* 621 * Arguments: 622 * t table to free 623 */ 624static int 625huft_free(struct inflate *glbl, struct huft *t) 626/* Free the malloc'ed tables built by huft_build(), which makes a linked 627 list of the tables it made, with the links in a dummy first entry of 628 each table. */ 629{ 630 struct huft *p, *q; 631 632 /* Go through linked list, freeing from the malloced (t[-1]) address. */ 633 p = t; 634 while (p != (struct huft *) NULL) { 635 q = (--p)->v.t; 636 free(p, M_GZIP); 637 p = q; 638 } 639 return 0; 640} 641 642/* inflate (decompress) the codes in a deflated (compressed) block. 643 Return an error code or zero if it all goes ok. */ 644/* 645 * Arguments: 646 * tl, td literal/length and distance decoder tables 647 * bl, bd number of bits decoded by tl[] and td[] 648 */ 649static int 650inflate_codes(struct inflate *glbl, struct huft *tl, struct huft*td, int bl, 651 int bd) 652{ 653 unsigned e; /* table entry flag/number of extra bits */ 654 unsigned n, d; /* length and index for copy */ 655 unsigned w; /* current window position */ 656 struct huft *t; /* pointer to table entry */ 657 unsigned ml, md; /* masks for bl and bd bits */ 658 ulg b; /* bit buffer */ 659 unsigned k; /* number of bits in bit buffer */ 660 661 /* make local copies of globals */ 662 b = glbl->gz_bb; /* initialize bit buffer */ 663 k = glbl->gz_bk; 664 w = glbl->gz_wp; /* initialize window position */ 665 666 /* inflate the coded data */ 667 ml = mask[bl]; /* precompute masks for speed */ 668 md = mask[bd]; 669 while (1) { /* do until end of block */ 670 NEEDBITS(glbl, (unsigned) bl) 671 if ((e = (t = tl + ((unsigned) b & ml))->e) > 16) 672 do { 673 if (e == 99) 674 return 1; 675 DUMPBITS(t->b) 676 e -= 16; 677 NEEDBITS(glbl, e) 678 } while ((e = (t = t->v.t + ((unsigned) b & mask[e]))->e) > 16); 679 DUMPBITS(t->b) 680 if (e == 16) { /* then it's a literal */ 681 glbl->gz_slide[w++] = (uch) t->v.n; 682 if (w == GZ_WSIZE) { 683 FLUSH(glbl, w); 684 w = 0; 685 } 686 } else { /* it's an EOB or a length */ 687 /* exit if end of block */ 688 if (e == 15) 689 break; 690 691 /* get length of block to copy */ 692 NEEDBITS(glbl, e) 693 n = t->v.n + ((unsigned) b & mask[e]); 694 DUMPBITS(e); 695 696 /* decode distance of block to copy */ 697 NEEDBITS(glbl, (unsigned) bd) 698 if ((e = (t = td + ((unsigned) b & md))->e) > 16) 699 do { 700 if (e == 99) 701 return 1; 702 DUMPBITS(t->b) 703 e -= 16; 704 NEEDBITS(glbl, e) 705 } while ((e = (t = t->v.t + ((unsigned) b & mask[e]))->e) > 16); 706 DUMPBITS(t->b) 707 NEEDBITS(glbl, e) 708 d = w - t->v.n - ((unsigned) b & mask[e]); 709 DUMPBITS(e) 710 /* do the copy */ 711 do { 712 n -= (e = (e = GZ_WSIZE - ((d &= GZ_WSIZE - 1) > w ? d : w)) > n ? n : e); 713#ifndef NOMEMCPY 714 if (w - d >= e) { /* (this test assumes 715 * unsigned comparison) */ 716 memcpy(glbl->gz_slide + w, glbl->gz_slide + d, e); 717 w += e; 718 d += e; 719 } else /* do it slow to avoid memcpy() 720 * overlap */ 721#endif /* !NOMEMCPY */ 722 do { 723 glbl->gz_slide[w++] = glbl->gz_slide[d++]; 724 } while (--e); 725 if (w == GZ_WSIZE) { 726 FLUSH(glbl, w); 727 w = 0; 728 } 729 } while (n); 730 } 731 } 732 733 /* restore the globals from the locals */ 734 glbl->gz_wp = w; /* restore global window pointer */ 735 glbl->gz_bb = b; /* restore global bit buffer */ 736 glbl->gz_bk = k; 737 738 /* done */ 739 return 0; 740} 741 742/* "decompress" an inflated type 0 (stored) block. */ 743static int 744inflate_stored(struct inflate *glbl) 745{ 746 unsigned n; /* number of bytes in block */ 747 unsigned w; /* current window position */ 748 ulg b; /* bit buffer */ 749 unsigned k; /* number of bits in bit buffer */ 750 751 /* make local copies of globals */ 752 b = glbl->gz_bb; /* initialize bit buffer */ 753 k = glbl->gz_bk; 754 w = glbl->gz_wp; /* initialize window position */ 755 756 /* go to byte boundary */ 757 n = k & 7; 758 DUMPBITS(n); 759 760 /* get the length and its complement */ 761 NEEDBITS(glbl, 16) 762 n = ((unsigned) b & 0xffff); 763 DUMPBITS(16) 764 NEEDBITS(glbl, 16) 765 if (n != (unsigned) ((~b) & 0xffff)) 766 return 1; /* error in compressed data */ 767 DUMPBITS(16) 768 /* read and output the compressed data */ 769 while (n--) { 770 NEEDBITS(glbl, 8) 771 glbl->gz_slide[w++] = (uch) b; 772 if (w == GZ_WSIZE) { 773 FLUSH(glbl, w); 774 w = 0; 775 } 776 DUMPBITS(8) 777 } 778 779 /* restore the globals from the locals */ 780 glbl->gz_wp = w; /* restore global window pointer */ 781 glbl->gz_bb = b; /* restore global bit buffer */ 782 glbl->gz_bk = k; 783 return 0; 784} 785 786/* decompress an inflated type 1 (fixed Huffman codes) block. We should 787 either replace this with a custom decoder, or at least precompute the 788 Huffman tables. */ 789static int 790inflate_fixed(struct inflate *glbl) 791{ 792 /* if first time, set up tables for fixed blocks */ 793 if (glbl->gz_fixed_tl == (struct huft *) NULL) { 794 int i; /* temporary variable */ 795 static unsigned l[288]; /* length list for huft_build */ 796 797 /* literal table */ 798 for (i = 0; i < 144; i++) 799 l[i] = 8; 800 for (; i < 256; i++) 801 l[i] = 9; 802 for (; i < 280; i++) 803 l[i] = 7; 804 for (; i < 288; i++) /* make a complete, but wrong code 805 * set */ 806 l[i] = 8; 807 glbl->gz_fixed_bl = 7; 808 if ((i = huft_build(glbl, l, 288, 257, cplens, cplext, 809 &glbl->gz_fixed_tl, &glbl->gz_fixed_bl)) != 0) { 810 glbl->gz_fixed_tl = (struct huft *) NULL; 811 return i; 812 } 813 /* distance table */ 814 for (i = 0; i < 30; i++) /* make an incomplete code 815 * set */ 816 l[i] = 5; 817 glbl->gz_fixed_bd = 5; 818 if ((i = huft_build(glbl, l, 30, 0, cpdist, cpdext, 819 &glbl->gz_fixed_td, &glbl->gz_fixed_bd)) > 1) { 820 huft_free(glbl, glbl->gz_fixed_tl); 821 glbl->gz_fixed_tl = (struct huft *) NULL; 822 return i; 823 } 824 } 825 /* decompress until an end-of-block code */ 826 return inflate_codes(glbl, glbl->gz_fixed_tl, glbl->gz_fixed_td, glbl->gz_fixed_bl, glbl->gz_fixed_bd) != 0; 827} 828 829/* decompress an inflated type 2 (dynamic Huffman codes) block. */ 830static int 831inflate_dynamic(struct inflate *glbl) 832{ 833 int i; /* temporary variables */ 834 unsigned j; 835 unsigned l; /* last length */ 836 unsigned m; /* mask for bit lengths table */ 837 unsigned n; /* number of lengths to get */ 838 struct huft *tl; /* literal/length code table */ 839 struct huft *td; /* distance code table */ 840 int bl; /* lookup bits for tl */ 841 int bd; /* lookup bits for td */ 842 unsigned nb; /* number of bit length codes */ 843 unsigned nl; /* number of literal/length codes */ 844 unsigned nd; /* number of distance codes */ 845#ifdef PKZIP_BUG_WORKAROUND 846 unsigned ll[288 + 32]; /* literal/length and distance code 847 * lengths */ 848#else 849 unsigned ll[286 + 30]; /* literal/length and distance code 850 * lengths */ 851#endif 852 ulg b; /* bit buffer */ 853 unsigned k; /* number of bits in bit buffer */ 854 855 /* make local bit buffer */ 856 b = glbl->gz_bb; 857 k = glbl->gz_bk; 858 859 /* read in table lengths */ 860 NEEDBITS(glbl, 5) 861 nl = 257 + ((unsigned) b & 0x1f); /* number of 862 * literal/length codes */ 863 DUMPBITS(5) 864 NEEDBITS(glbl, 5) 865 nd = 1 + ((unsigned) b & 0x1f); /* number of distance codes */ 866 DUMPBITS(5) 867 NEEDBITS(glbl, 4) 868 nb = 4 + ((unsigned) b & 0xf); /* number of bit length codes */ 869 DUMPBITS(4) 870#ifdef PKZIP_BUG_WORKAROUND 871 if (nl > 288 || nd > 32) 872#else 873 if (nl > 286 || nd > 30) 874#endif 875 return 1; /* bad lengths */ 876 /* read in bit-length-code lengths */ 877 for (j = 0; j < nb; j++) { 878 NEEDBITS(glbl, 3) 879 ll[border[j]] = (unsigned) b & 7; 880 DUMPBITS(3) 881 } 882 for (; j < 19; j++) 883 ll[border[j]] = 0; 884 885 /* build decoding table for trees--single level, 7 bit lookup */ 886 bl = 7; 887 if ((i = huft_build(glbl, ll, 19, 19, NULL, NULL, &tl, &bl)) != 0) { 888 if (i == 1) 889 huft_free(glbl, tl); 890 return i; /* incomplete code set */ 891 } 892 /* read in literal and distance code lengths */ 893 n = nl + nd; 894 m = mask[bl]; 895 i = l = 0; 896 while ((unsigned) i < n) { 897 NEEDBITS(glbl, (unsigned) bl) 898 j = (td = tl + ((unsigned) b & m))->b; 899 DUMPBITS(j) 900 j = td->v.n; 901 if (j < 16) /* length of code in bits (0..15) */ 902 ll[i++] = l = j; /* save last length in l */ 903 else if (j == 16) { /* repeat last length 3 to 6 times */ 904 NEEDBITS(glbl, 2) 905 j = 3 + ((unsigned) b & 3); 906 DUMPBITS(2) 907 if ((unsigned) i + j > n) 908 return 1; 909 while (j--) 910 ll[i++] = l; 911 } else if (j == 17) { /* 3 to 10 zero length codes */ 912 NEEDBITS(glbl, 3) 913 j = 3 + ((unsigned) b & 7); 914 DUMPBITS(3) 915 if ((unsigned) i + j > n) 916 return 1; 917 while (j--) 918 ll[i++] = 0; 919 l = 0; 920 } else { /* j == 18: 11 to 138 zero length codes */ 921 NEEDBITS(glbl, 7) 922 j = 11 + ((unsigned) b & 0x7f); 923 DUMPBITS(7) 924 if ((unsigned) i + j > n) 925 return 1; 926 while (j--) 927 ll[i++] = 0; 928 l = 0; 929 } 930 } 931 932 /* free decoding table for trees */ 933 huft_free(glbl, tl); 934 935 /* restore the global bit buffer */ 936 glbl->gz_bb = b; 937 glbl->gz_bk = k; 938 939 /* build the decoding tables for literal/length and distance codes */ 940 bl = lbits; 941 i = huft_build(glbl, ll, nl, 257, cplens, cplext, &tl, &bl); 942 if (i != 0) { 943 if (i == 1 && !qflag) { 944 FPRINTF("(incomplete l-tree) "); 945 huft_free(glbl, tl); 946 } 947 return i; /* incomplete code set */ 948 } 949 bd = dbits; 950 i = huft_build(glbl, ll + nl, nd, 0, cpdist, cpdext, &td, &bd); 951 if (i != 0) { 952 if (i == 1 && !qflag) { 953 FPRINTF("(incomplete d-tree) "); 954#ifdef PKZIP_BUG_WORKAROUND 955 i = 0; 956 } 957#else 958 huft_free(glbl, td); 959 } 960 huft_free(glbl, tl); 961 return i; /* incomplete code set */ 962#endif 963 } 964 /* decompress until an end-of-block code */ 965 if (inflate_codes(glbl, tl, td, bl, bd)) 966 return 1; 967 968 /* free the decoding tables, return */ 969 huft_free(glbl, tl); 970 huft_free(glbl, td); 971 return 0; 972} 973 974/* decompress an inflated block */ 975/* 976 * Arguments: 977 * e last block flag 978 */ 979static int 980inflate_block(struct inflate *glbl, int *e) 981{ 982 unsigned t; /* block type */ 983 ulg b; /* bit buffer */ 984 unsigned k; /* number of bits in bit buffer */ 985 986 /* make local bit buffer */ 987 b = glbl->gz_bb; 988 k = glbl->gz_bk; 989 990 /* read in last block bit */ 991 NEEDBITS(glbl, 1) 992 * e = (int) b & 1; 993 DUMPBITS(1) 994 /* read in block type */ 995 NEEDBITS(glbl, 2) 996 t = (unsigned) b & 3; 997 DUMPBITS(2) 998 /* restore the global bit buffer */ 999 glbl->gz_bb = b; 1000 glbl->gz_bk = k; 1001 1002 /* inflate that block type */ 1003 if (t == 2) 1004 return inflate_dynamic(glbl); 1005 if (t == 0) 1006 return inflate_stored(glbl); 1007 if (t == 1) 1008 return inflate_fixed(glbl); 1009 /* bad block type */ 1010 return 2; 1011} 1012 1013 1014 1015/* decompress an inflated entry */ 1016static int 1017xinflate(struct inflate *glbl) 1018{ 1019 int e; /* last block flag */ 1020 int r; /* result code */ 1021 unsigned h; /* maximum struct huft's malloc'ed */ 1022 1023 glbl->gz_fixed_tl = (struct huft *) NULL; 1024 1025 /* initialize window, bit buffer */ 1026 glbl->gz_wp = 0; 1027 glbl->gz_bk = 0; 1028 glbl->gz_bb = 0; 1029 1030 /* decompress until the last block */ 1031 h = 0; 1032 do { 1033 glbl->gz_hufts = 0; 1034 if ((r = inflate_block(glbl, &e)) != 0) 1035 return r; 1036 if (glbl->gz_hufts > h) 1037 h = glbl->gz_hufts; 1038 } while (!e); 1039 1040 /* flush out slide */ 1041 FLUSH(glbl, glbl->gz_wp); 1042 1043 /* return success */ 1044 return 0; 1045} 1046 1047/* Nobody uses this - why not? */ 1048int 1049inflate(struct inflate *glbl) 1050{ 1051 int i; 1052#ifdef _KERNEL 1053 u_char *p = NULL; 1054 1055 if (!glbl->gz_slide) 1056 p = glbl->gz_slide = malloc(GZ_WSIZE, M_GZIP, M_WAITOK); 1057#endif 1058 if (!glbl->gz_slide) 1059#ifdef _KERNEL 1060 return(ENOMEM); 1061#else 1062 return 3; /* kzip expects 3 */ 1063#endif 1064 i = xinflate(glbl); 1065 1066 if (glbl->gz_fixed_td != (struct huft *) NULL) { 1067 huft_free(glbl, glbl->gz_fixed_td); 1068 glbl->gz_fixed_td = (struct huft *) NULL; 1069 } 1070 if (glbl->gz_fixed_tl != (struct huft *) NULL) { 1071 huft_free(glbl, glbl->gz_fixed_tl); 1072 glbl->gz_fixed_tl = (struct huft *) NULL; 1073 } 1074#ifdef _KERNEL 1075 if (p == glbl->gz_slide) { 1076 free(glbl->gz_slide, M_GZIP); 1077 glbl->gz_slide = NULL; 1078 } 1079#endif 1080 return i; 1081} 1082/* ----------------------- END INFLATE.C */ 1083