/StormLib/stormlib/zlib/contrib/puff/puff.c

http://ghostcb.googlecode.com/ · C · 837 lines · 366 code · 57 blank · 414 comment · 112 complexity · e51db1c3e473d9de0479f24aaaa667c3 MD5 · raw file

  1. /*
  2. * puff.c
  3. * Copyright (C) 2002-2004 Mark Adler
  4. * For conditions of distribution and use, see copyright notice in puff.h
  5. * version 1.8, 9 Jan 2004
  6. *
  7. * puff.c is a simple inflate written to be an unambiguous way to specify the
  8. * deflate format. It is not written for speed but rather simplicity. As a
  9. * side benefit, this code might actually be useful when small code is more
  10. * important than speed, such as bootstrap applications. For typical deflate
  11. * data, zlib's inflate() is about four times as fast as puff(). zlib's
  12. * inflate compiles to around 20K on my machine, whereas puff.c compiles to
  13. * around 4K on my machine (a PowerPC using GNU cc). If the faster decode()
  14. * function here is used, then puff() is only twice as slow as zlib's
  15. * inflate().
  16. *
  17. * All dynamically allocated memory comes from the stack. The stack required
  18. * is less than 2K bytes. This code is compatible with 16-bit int's and
  19. * assumes that long's are at least 32 bits. puff.c uses the short data type,
  20. * assumed to be 16 bits, for arrays in order to to conserve memory. The code
  21. * works whether integers are stored big endian or little endian.
  22. *
  23. * In the comments below are "Format notes" that describe the inflate process
  24. * and document some of the less obvious aspects of the format. This source
  25. * code is meant to supplement RFC 1951, which formally describes the deflate
  26. * format:
  27. *
  28. * http://www.zlib.org/rfc-deflate.html
  29. */
  30. /*
  31. * Change history:
  32. *
  33. * 1.0 10 Feb 2002 - First version
  34. * 1.1 17 Feb 2002 - Clarifications of some comments and notes
  35. * - Update puff() dest and source pointers on negative
  36. * errors to facilitate debugging deflators
  37. * - Remove longest from struct huffman -- not needed
  38. * - Simplify offs[] index in construct()
  39. * - Add input size and checking, using longjmp() to
  40. * maintain easy readability
  41. * - Use short data type for large arrays
  42. * - Use pointers instead of long to specify source and
  43. * destination sizes to avoid arbitrary 4 GB limits
  44. * 1.2 17 Mar 2002 - Add faster version of decode(), doubles speed (!),
  45. * but leave simple version for readabilty
  46. * - Make sure invalid distances detected if pointers
  47. * are 16 bits
  48. * - Fix fixed codes table error
  49. * - Provide a scanning mode for determining size of
  50. * uncompressed data
  51. * 1.3 20 Mar 2002 - Go back to lengths for puff() parameters [Jean-loup]
  52. * - Add a puff.h file for the interface
  53. * - Add braces in puff() for else do [Jean-loup]
  54. * - Use indexes instead of pointers for readability
  55. * 1.4 31 Mar 2002 - Simplify construct() code set check
  56. * - Fix some comments
  57. * - Add FIXLCODES #define
  58. * 1.5 6 Apr 2002 - Minor comment fixes
  59. * 1.6 7 Aug 2002 - Minor format changes
  60. * 1.7 3 Mar 2003 - Added test code for distribution
  61. * - Added zlib-like license
  62. * 1.8 9 Jan 2004 - Added some comments on no distance codes case
  63. */
  64. #include <setjmp.h> /* for setjmp(), longjmp(), and jmp_buf */
  65. #include "puff.h" /* prototype for puff() */
  66. #define local static /* for local function definitions */
  67. #define NIL ((unsigned char *)0) /* for no output option */
  68. /*
  69. * Maximums for allocations and loops. It is not useful to change these --
  70. * they are fixed by the deflate format.
  71. */
  72. #define MAXBITS 15 /* maximum bits in a code */
  73. #define MAXLCODES 286 /* maximum number of literal/length codes */
  74. #define MAXDCODES 30 /* maximum number of distance codes */
  75. #define MAXCODES (MAXLCODES+MAXDCODES) /* maximum codes lengths to read */
  76. #define FIXLCODES 288 /* number of fixed literal/length codes */
  77. /* input and output state */
  78. struct state {
  79. /* output state */
  80. unsigned char *out; /* output buffer */
  81. unsigned long outlen; /* available space at out */
  82. unsigned long outcnt; /* bytes written to out so far */
  83. /* input state */
  84. unsigned char *in; /* input buffer */
  85. unsigned long inlen; /* available input at in */
  86. unsigned long incnt; /* bytes read so far */
  87. int bitbuf; /* bit buffer */
  88. int bitcnt; /* number of bits in bit buffer */
  89. /* input limit error return state for bits() and decode() */
  90. jmp_buf env;
  91. };
  92. /*
  93. * Return need bits from the input stream. This always leaves less than
  94. * eight bits in the buffer. bits() works properly for need == 0.
  95. *
  96. * Format notes:
  97. *
  98. * - Bits are stored in bytes from the least significant bit to the most
  99. * significant bit. Therefore bits are dropped from the bottom of the bit
  100. * buffer, using shift right, and new bytes are appended to the top of the
  101. * bit buffer, using shift left.
  102. */
  103. local int bits(struct state *s, int need)
  104. {
  105. long val; /* bit accumulator (can use up to 20 bits) */
  106. /* load at least need bits into val */
  107. val = s->bitbuf;
  108. while (s->bitcnt < need) {
  109. if (s->incnt == s->inlen) longjmp(s->env, 1); /* out of input */
  110. val |= (long)(s->in[s->incnt++]) << s->bitcnt; /* load eight bits */
  111. s->bitcnt += 8;
  112. }
  113. /* drop need bits and update buffer, always zero to seven bits left */
  114. s->bitbuf = (int)(val >> need);
  115. s->bitcnt -= need;
  116. /* return need bits, zeroing the bits above that */
  117. return (int)(val & ((1L << need) - 1));
  118. }
  119. /*
  120. * Process a stored block.
  121. *
  122. * Format notes:
  123. *
  124. * - After the two-bit stored block type (00), the stored block length and
  125. * stored bytes are byte-aligned for fast copying. Therefore any leftover
  126. * bits in the byte that has the last bit of the type, as many as seven, are
  127. * discarded. The value of the discarded bits are not defined and should not
  128. * be checked against any expectation.
  129. *
  130. * - The second inverted copy of the stored block length does not have to be
  131. * checked, but it's probably a good idea to do so anyway.
  132. *
  133. * - A stored block can have zero length. This is sometimes used to byte-align
  134. * subsets of the compressed data for random access or partial recovery.
  135. */
  136. local int stored(struct state *s)
  137. {
  138. unsigned len; /* length of stored block */
  139. /* discard leftover bits from current byte (assumes s->bitcnt < 8) */
  140. s->bitbuf = 0;
  141. s->bitcnt = 0;
  142. /* get length and check against its one's complement */
  143. if (s->incnt + 4 > s->inlen) return 2; /* not enough input */
  144. len = s->in[s->incnt++];
  145. len |= s->in[s->incnt++] << 8;
  146. if (s->in[s->incnt++] != (~len & 0xff) ||
  147. s->in[s->incnt++] != ((~len >> 8) & 0xff))
  148. return -2; /* didn't match complement! */
  149. /* copy len bytes from in to out */
  150. if (s->incnt + len > s->inlen) return 2; /* not enough input */
  151. if (s->out != NIL) {
  152. if (s->outcnt + len > s->outlen)
  153. return 1; /* not enough output space */
  154. while (len--)
  155. s->out[s->outcnt++] = s->in[s->incnt++];
  156. }
  157. else { /* just scanning */
  158. s->outcnt += len;
  159. s->incnt += len;
  160. }
  161. /* done with a valid stored block */
  162. return 0;
  163. }
  164. /*
  165. * Huffman code decoding tables. count[1..MAXBITS] is the number of symbols of
  166. * each length, which for a canonical code are stepped through in order.
  167. * symbol[] are the symbol values in canonical order, where the number of
  168. * entries is the sum of the counts in count[]. The decoding process can be
  169. * seen in the function decode() below.
  170. */
  171. struct huffman {
  172. short *count; /* number of symbols of each length */
  173. short *symbol; /* canonically ordered symbols */
  174. };
  175. /*
  176. * Decode a code from the stream s using huffman table h. Return the symbol or
  177. * a negative value if there is an error. If all of the lengths are zero, i.e.
  178. * an empty code, or if the code is incomplete and an invalid code is received,
  179. * then -9 is returned after reading MAXBITS bits.
  180. *
  181. * Format notes:
  182. *
  183. * - The codes as stored in the compressed data are bit-reversed relative to
  184. * a simple integer ordering of codes of the same lengths. Hence below the
  185. * bits are pulled from the compressed data one at a time and used to
  186. * build the code value reversed from what is in the stream in order to
  187. * permit simple integer comparisons for decoding. A table-based decoding
  188. * scheme (as used in zlib) does not need to do this reversal.
  189. *
  190. * - The first code for the shortest length is all zeros. Subsequent codes of
  191. * the same length are simply integer increments of the previous code. When
  192. * moving up a length, a zero bit is appended to the code. For a complete
  193. * code, the last code of the longest length will be all ones.
  194. *
  195. * - Incomplete codes are handled by this decoder, since they are permitted
  196. * in the deflate format. See the format notes for fixed() and dynamic().
  197. */
  198. #ifdef SLOW
  199. local int decode(struct state *s, struct huffman *h)
  200. {
  201. int len; /* current number of bits in code */
  202. int code; /* len bits being decoded */
  203. int first; /* first code of length len */
  204. int count; /* number of codes of length len */
  205. int index; /* index of first code of length len in symbol table */
  206. code = first = index = 0;
  207. for (len = 1; len <= MAXBITS; len++) {
  208. code |= bits(s, 1); /* get next bit */
  209. count = h->count[len];
  210. if (code < first + count) /* if length len, return symbol */
  211. return h->symbol[index + (code - first)];
  212. index += count; /* else update for next length */
  213. first += count;
  214. first <<= 1;
  215. code <<= 1;
  216. }
  217. return -9; /* ran out of codes */
  218. }
  219. /*
  220. * A faster version of decode() for real applications of this code. It's not
  221. * as readable, but it makes puff() twice as fast. And it only makes the code
  222. * a few percent larger.
  223. */
  224. #else /* !SLOW */
  225. local int decode(struct state *s, struct huffman *h)
  226. {
  227. int len; /* current number of bits in code */
  228. int code; /* len bits being decoded */
  229. int first; /* first code of length len */
  230. int count; /* number of codes of length len */
  231. int index; /* index of first code of length len in symbol table */
  232. int bitbuf; /* bits from stream */
  233. int left; /* bits left in next or left to process */
  234. short *next; /* next number of codes */
  235. bitbuf = s->bitbuf;
  236. left = s->bitcnt;
  237. code = first = index = 0;
  238. len = 1;
  239. next = h->count + 1;
  240. while (1) {
  241. while (left--) {
  242. code |= bitbuf & 1;
  243. bitbuf >>= 1;
  244. count = *next++;
  245. if (code < first + count) { /* if length len, return symbol */
  246. s->bitbuf = bitbuf;
  247. s->bitcnt = (s->bitcnt - len) & 7;
  248. return h->symbol[index + (code - first)];
  249. }
  250. index += count; /* else update for next length */
  251. first += count;
  252. first <<= 1;
  253. code <<= 1;
  254. len++;
  255. }
  256. left = (MAXBITS+1) - len;
  257. if (left == 0) break;
  258. if (s->incnt == s->inlen) longjmp(s->env, 1); /* out of input */
  259. bitbuf = s->in[s->incnt++];
  260. if (left > 8) left = 8;
  261. }
  262. return -9; /* ran out of codes */
  263. }
  264. #endif /* SLOW */
  265. /*
  266. * Given the list of code lengths length[0..n-1] representing a canonical
  267. * Huffman code for n symbols, construct the tables required to decode those
  268. * codes. Those tables are the number of codes of each length, and the symbols
  269. * sorted by length, retaining their original order within each length. The
  270. * return value is zero for a complete code set, negative for an over-
  271. * subscribed code set, and positive for an incomplete code set. The tables
  272. * can be used if the return value is zero or positive, but they cannot be used
  273. * if the return value is negative. If the return value is zero, it is not
  274. * possible for decode() using that table to return an error--any stream of
  275. * enough bits will resolve to a symbol. If the return value is positive, then
  276. * it is possible for decode() using that table to return an error for received
  277. * codes past the end of the incomplete lengths.
  278. *
  279. * Not used by decode(), but used for error checking, h->count[0] is the number
  280. * of the n symbols not in the code. So n - h->count[0] is the number of
  281. * codes. This is useful for checking for incomplete codes that have more than
  282. * one symbol, which is an error in a dynamic block.
  283. *
  284. * Assumption: for all i in 0..n-1, 0 <= length[i] <= MAXBITS
  285. * This is assured by the construction of the length arrays in dynamic() and
  286. * fixed() and is not verified by construct().
  287. *
  288. * Format notes:
  289. *
  290. * - Permitted and expected examples of incomplete codes are one of the fixed
  291. * codes and any code with a single symbol which in deflate is coded as one
  292. * bit instead of zero bits. See the format notes for fixed() and dynamic().
  293. *
  294. * - Within a given code length, the symbols are kept in ascending order for
  295. * the code bits definition.
  296. */
  297. local int construct(struct huffman *h, short *length, int n)
  298. {
  299. int symbol; /* current symbol when stepping through length[] */
  300. int len; /* current length when stepping through h->count[] */
  301. int left; /* number of possible codes left of current length */
  302. short offs[MAXBITS+1]; /* offsets in symbol table for each length */
  303. /* count number of codes of each length */
  304. for (len = 0; len <= MAXBITS; len++)
  305. h->count[len] = 0;
  306. for (symbol = 0; symbol < n; symbol++)
  307. (h->count[length[symbol]])++; /* assumes lengths are within bounds */
  308. if (h->count[0] == n) /* no codes! */
  309. return 0; /* complete, but decode() will fail */
  310. /* check for an over-subscribed or incomplete set of lengths */
  311. left = 1; /* one possible code of zero length */
  312. for (len = 1; len <= MAXBITS; len++) {
  313. left <<= 1; /* one more bit, double codes left */
  314. left -= h->count[len]; /* deduct count from possible codes */
  315. if (left < 0) return left; /* over-subscribed--return negative */
  316. } /* left > 0 means incomplete */
  317. /* generate offsets into symbol table for each length for sorting */
  318. offs[1] = 0;
  319. for (len = 1; len < MAXBITS; len++)
  320. offs[len + 1] = offs[len] + h->count[len];
  321. /*
  322. * put symbols in table sorted by length, by symbol order within each
  323. * length
  324. */
  325. for (symbol = 0; symbol < n; symbol++)
  326. if (length[symbol] != 0)
  327. h->symbol[offs[length[symbol]]++] = symbol;
  328. /* return zero for complete set, positive for incomplete set */
  329. return left;
  330. }
  331. /*
  332. * Decode literal/length and distance codes until an end-of-block code.
  333. *
  334. * Format notes:
  335. *
  336. * - Compressed data that is after the block type if fixed or after the code
  337. * description if dynamic is a combination of literals and length/distance
  338. * pairs terminated by and end-of-block code. Literals are simply Huffman
  339. * coded bytes. A length/distance pair is a coded length followed by a
  340. * coded distance to represent a string that occurs earlier in the
  341. * uncompressed data that occurs again at the current location.
  342. *
  343. * - Literals, lengths, and the end-of-block code are combined into a single
  344. * code of up to 286 symbols. They are 256 literals (0..255), 29 length
  345. * symbols (257..285), and the end-of-block symbol (256).
  346. *
  347. * - There are 256 possible lengths (3..258), and so 29 symbols are not enough
  348. * to represent all of those. Lengths 3..10 and 258 are in fact represented
  349. * by just a length symbol. Lengths 11..257 are represented as a symbol and
  350. * some number of extra bits that are added as an integer to the base length
  351. * of the length symbol. The number of extra bits is determined by the base
  352. * length symbol. These are in the static arrays below, lens[] for the base
  353. * lengths and lext[] for the corresponding number of extra bits.
  354. *
  355. * - The reason that 258 gets its own symbol is that the longest length is used
  356. * often in highly redundant files. Note that 258 can also be coded as the
  357. * base value 227 plus the maximum extra value of 31. While a good deflate
  358. * should never do this, it is not an error, and should be decoded properly.
  359. *
  360. * - If a length is decoded, including its extra bits if any, then it is
  361. * followed a distance code. There are up to 30 distance symbols. Again
  362. * there are many more possible distances (1..32768), so extra bits are added
  363. * to a base value represented by the symbol. The distances 1..4 get their
  364. * own symbol, but the rest require extra bits. The base distances and
  365. * corresponding number of extra bits are below in the static arrays dist[]
  366. * and dext[].
  367. *
  368. * - Literal bytes are simply written to the output. A length/distance pair is
  369. * an instruction to copy previously uncompressed bytes to the output. The
  370. * copy is from distance bytes back in the output stream, copying for length
  371. * bytes.
  372. *
  373. * - Distances pointing before the beginning of the output data are not
  374. * permitted.
  375. *
  376. * - Overlapped copies, where the length is greater than the distance, are
  377. * allowed and common. For example, a distance of one and a length of 258
  378. * simply copies the last byte 258 times. A distance of four and a length of
  379. * twelve copies the last four bytes three times. A simple forward copy
  380. * ignoring whether the length is greater than the distance or not implements
  381. * this correctly. You should not use memcpy() since its behavior is not
  382. * defined for overlapped arrays. You should not use memmove() or bcopy()
  383. * since though their behavior -is- defined for overlapping arrays, it is
  384. * defined to do the wrong thing in this case.
  385. */
  386. local int codes(struct state *s,
  387. struct huffman *lencode,
  388. struct huffman *distcode)
  389. {
  390. int symbol; /* decoded symbol */
  391. int len; /* length for copy */
  392. unsigned dist; /* distance for copy */
  393. static const short lens[29] = { /* Size base for length codes 257..285 */
  394. 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
  395. 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258};
  396. static const short lext[29] = { /* Extra bits for length codes 257..285 */
  397. 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2,
  398. 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 0};
  399. static const short dists[30] = { /* Offset base for distance codes 0..29 */
  400. 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
  401. 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,
  402. 8193, 12289, 16385, 24577};
  403. static const short dext[30] = { /* Extra bits for distance codes 0..29 */
  404. 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6,
  405. 7, 7, 8, 8, 9, 9, 10, 10, 11, 11,
  406. 12, 12, 13, 13};
  407. /* decode literals and length/distance pairs */
  408. do {
  409. symbol = decode(s, lencode);
  410. if (symbol < 0) return symbol; /* invalid symbol */
  411. if (symbol < 256) { /* literal: symbol is the byte */
  412. /* write out the literal */
  413. if (s->out != NIL) {
  414. if (s->outcnt == s->outlen) return 1;
  415. s->out[s->outcnt] = symbol;
  416. }
  417. s->outcnt++;
  418. }
  419. else if (symbol > 256) { /* length */
  420. /* get and compute length */
  421. symbol -= 257;
  422. if (symbol >= 29) return -9; /* invalid fixed code */
  423. len = lens[symbol] + bits(s, lext[symbol]);
  424. /* get and check distance */
  425. symbol = decode(s, distcode);
  426. if (symbol < 0) return symbol; /* invalid symbol */
  427. dist = dists[symbol] + bits(s, dext[symbol]);
  428. if (dist > s->outcnt)
  429. return -10; /* distance too far back */
  430. /* copy length bytes from distance bytes back */
  431. if (s->out != NIL) {
  432. if (s->outcnt + len > s->outlen) return 1;
  433. while (len--) {
  434. s->out[s->outcnt] = s->out[s->outcnt - dist];
  435. s->outcnt++;
  436. }
  437. }
  438. else
  439. s->outcnt += len;
  440. }
  441. } while (symbol != 256); /* end of block symbol */
  442. /* done with a valid fixed or dynamic block */
  443. return 0;
  444. }
  445. /*
  446. * Process a fixed codes block.
  447. *
  448. * Format notes:
  449. *
  450. * - This block type can be useful for compressing small amounts of data for
  451. * which the size of the code descriptions in a dynamic block exceeds the
  452. * benefit of custom codes for that block. For fixed codes, no bits are
  453. * spent on code descriptions. Instead the code lengths for literal/length
  454. * codes and distance codes are fixed. The specific lengths for each symbol
  455. * can be seen in the "for" loops below.
  456. *
  457. * - The literal/length code is complete, but has two symbols that are invalid
  458. * and should result in an error if received. This cannot be implemented
  459. * simply as an incomplete code since those two symbols are in the "middle"
  460. * of the code. They are eight bits long and the longest literal/length\
  461. * code is nine bits. Therefore the code must be constructed with those
  462. * symbols, and the invalid symbols must be detected after decoding.
  463. *
  464. * - The fixed distance codes also have two invalid symbols that should result
  465. * in an error if received. Since all of the distance codes are the same
  466. * length, this can be implemented as an incomplete code. Then the invalid
  467. * codes are detected while decoding.
  468. */
  469. local int fixed(struct state *s)
  470. {
  471. static int virgin = 1;
  472. static short lencnt[MAXBITS+1], lensym[FIXLCODES];
  473. static short distcnt[MAXBITS+1], distsym[MAXDCODES];
  474. static struct huffman lencode = {lencnt, lensym};
  475. static struct huffman distcode = {distcnt, distsym};
  476. /* build fixed huffman tables if first call (may not be thread safe) */
  477. if (virgin) {
  478. int symbol;
  479. short lengths[FIXLCODES];
  480. /* literal/length table */
  481. for (symbol = 0; symbol < 144; symbol++)
  482. lengths[symbol] = 8;
  483. for (; symbol < 256; symbol++)
  484. lengths[symbol] = 9;
  485. for (; symbol < 280; symbol++)
  486. lengths[symbol] = 7;
  487. for (; symbol < FIXLCODES; symbol++)
  488. lengths[symbol] = 8;
  489. construct(&lencode, lengths, FIXLCODES);
  490. /* distance table */
  491. for (symbol = 0; symbol < MAXDCODES; symbol++)
  492. lengths[symbol] = 5;
  493. construct(&distcode, lengths, MAXDCODES);
  494. /* do this just once */
  495. virgin = 0;
  496. }
  497. /* decode data until end-of-block code */
  498. return codes(s, &lencode, &distcode);
  499. }
  500. /*
  501. * Process a dynamic codes block.
  502. *
  503. * Format notes:
  504. *
  505. * - A dynamic block starts with a description of the literal/length and
  506. * distance codes for that block. New dynamic blocks allow the compressor to
  507. * rapidly adapt to changing data with new codes optimized for that data.
  508. *
  509. * - The codes used by the deflate format are "canonical", which means that
  510. * the actual bits of the codes are generated in an unambiguous way simply
  511. * from the number of bits in each code. Therefore the code descriptions
  512. * are simply a list of code lengths for each symbol.
  513. *
  514. * - The code lengths are stored in order for the symbols, so lengths are
  515. * provided for each of the literal/length symbols, and for each of the
  516. * distance symbols.
  517. *
  518. * - If a symbol is not used in the block, this is represented by a zero as
  519. * as the code length. This does not mean a zero-length code, but rather
  520. * that no code should be created for this symbol. There is no way in the
  521. * deflate format to represent a zero-length code.
  522. *
  523. * - The maximum number of bits in a code is 15, so the possible lengths for
  524. * any code are 1..15.
  525. *
  526. * - The fact that a length of zero is not permitted for a code has an
  527. * interesting consequence. Normally if only one symbol is used for a given
  528. * code, then in fact that code could be represented with zero bits. However
  529. * in deflate, that code has to be at least one bit. So for example, if
  530. * only a single distance base symbol appears in a block, then it will be
  531. * represented by a single code of length one, in particular one 0 bit. This
  532. * is an incomplete code, since if a 1 bit is received, it has no meaning,
  533. * and should result in an error. So incomplete distance codes of one symbol
  534. * should be permitted, and the receipt of invalid codes should be handled.
  535. *
  536. * - It is also possible to have a single literal/length code, but that code
  537. * must be the end-of-block code, since every dynamic block has one. This
  538. * is not the most efficient way to create an empty block (an empty fixed
  539. * block is fewer bits), but it is allowed by the format. So incomplete
  540. * literal/length codes of one symbol should also be permitted.
  541. *
  542. * - If there are only literal codes and no lengths, then there are no distance
  543. * codes. This is represented by one distance code with zero bits.
  544. *
  545. * - The list of up to 286 length/literal lengths and up to 30 distance lengths
  546. * are themselves compressed using Huffman codes and run-length encoding. In
  547. * the list of code lengths, a 0 symbol means no code, a 1..15 symbol means
  548. * that length, and the symbols 16, 17, and 18 are run-length instructions.
  549. * Each of 16, 17, and 18 are follwed by extra bits to define the length of
  550. * the run. 16 copies the last length 3 to 6 times. 17 represents 3 to 10
  551. * zero lengths, and 18 represents 11 to 138 zero lengths. Unused symbols
  552. * are common, hence the special coding for zero lengths.
  553. *
  554. * - The symbols for 0..18 are Huffman coded, and so that code must be
  555. * described first. This is simply a sequence of up to 19 three-bit values
  556. * representing no code (0) or the code length for that symbol (1..7).
  557. *
  558. * - A dynamic block starts with three fixed-size counts from which is computed
  559. * the number of literal/length code lengths, the number of distance code
  560. * lengths, and the number of code length code lengths (ok, you come up with
  561. * a better name!) in the code descriptions. For the literal/length and
  562. * distance codes, lengths after those provided are considered zero, i.e. no
  563. * code. The code length code lengths are received in a permuted order (see
  564. * the order[] array below) to make a short code length code length list more
  565. * likely. As it turns out, very short and very long codes are less likely
  566. * to be seen in a dynamic code description, hence what may appear initially
  567. * to be a peculiar ordering.
  568. *
  569. * - Given the number of literal/length code lengths (nlen) and distance code
  570. * lengths (ndist), then they are treated as one long list of nlen + ndist
  571. * code lengths. Therefore run-length coding can and often does cross the
  572. * boundary between the two sets of lengths.
  573. *
  574. * - So to summarize, the code description at the start of a dynamic block is
  575. * three counts for the number of code lengths for the literal/length codes,
  576. * the distance codes, and the code length codes. This is followed by the
  577. * code length code lengths, three bits each. This is used to construct the
  578. * code length code which is used to read the remainder of the lengths. Then
  579. * the literal/length code lengths and distance lengths are read as a single
  580. * set of lengths using the code length codes. Codes are constructed from
  581. * the resulting two sets of lengths, and then finally you can start
  582. * decoding actual compressed data in the block.
  583. *
  584. * - For reference, a "typical" size for the code description in a dynamic
  585. * block is around 80 bytes.
  586. */
  587. local int dynamic(struct state *s)
  588. {
  589. int nlen, ndist, ncode; /* number of lengths in descriptor */
  590. int index; /* index of lengths[] */
  591. int err; /* construct() return value */
  592. short lengths[MAXCODES]; /* descriptor code lengths */
  593. short lencnt[MAXBITS+1], lensym[MAXLCODES]; /* lencode memory */
  594. short distcnt[MAXBITS+1], distsym[MAXDCODES]; /* distcode memory */
  595. struct huffman lencode = {lencnt, lensym}; /* length code */
  596. struct huffman distcode = {distcnt, distsym}; /* distance code */
  597. static const short order[19] = /* permutation of code length codes */
  598. {16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15};
  599. /* get number of lengths in each table, check lengths */
  600. nlen = bits(s, 5) + 257;
  601. ndist = bits(s, 5) + 1;
  602. ncode = bits(s, 4) + 4;
  603. if (nlen > MAXLCODES || ndist > MAXDCODES)
  604. return -3; /* bad counts */
  605. /* read code length code lengths (really), missing lengths are zero */
  606. for (index = 0; index < ncode; index++)
  607. lengths[order[index]] = bits(s, 3);
  608. for (; index < 19; index++)
  609. lengths[order[index]] = 0;
  610. /* build huffman table for code lengths codes (use lencode temporarily) */
  611. err = construct(&lencode, lengths, 19);
  612. if (err != 0) return -4; /* require complete code set here */
  613. /* read length/literal and distance code length tables */
  614. index = 0;
  615. while (index < nlen + ndist) {
  616. int symbol; /* decoded value */
  617. int len; /* last length to repeat */
  618. symbol = decode(s, &lencode);
  619. if (symbol < 16) /* length in 0..15 */
  620. lengths[index++] = symbol;
  621. else { /* repeat instruction */
  622. len = 0; /* assume repeating zeros */
  623. if (symbol == 16) { /* repeat last length 3..6 times */
  624. if (index == 0) return -5; /* no last length! */
  625. len = lengths[index - 1]; /* last length */
  626. symbol = 3 + bits(s, 2);
  627. }
  628. else if (symbol == 17) /* repeat zero 3..10 times */
  629. symbol = 3 + bits(s, 3);
  630. else /* == 18, repeat zero 11..138 times */
  631. symbol = 11 + bits(s, 7);
  632. if (index + symbol > nlen + ndist)
  633. return -6; /* too many lengths! */
  634. while (symbol--) /* repeat last or zero symbol times */
  635. lengths[index++] = len;
  636. }
  637. }
  638. /* build huffman table for literal/length codes */
  639. err = construct(&lencode, lengths, nlen);
  640. if (err < 0 || (err > 0 && nlen - lencode.count[0] != 1))
  641. return -7; /* only allow incomplete codes if just one code */
  642. /* build huffman table for distance codes */
  643. err = construct(&distcode, lengths + nlen, ndist);
  644. if (err < 0 || (err > 0 && ndist - distcode.count[0] != 1))
  645. return -8; /* only allow incomplete codes if just one code */
  646. /* decode data until end-of-block code */
  647. return codes(s, &lencode, &distcode);
  648. }
  649. /*
  650. * Inflate source to dest. On return, destlen and sourcelen are updated to the
  651. * size of the uncompressed data and the size of the deflate data respectively.
  652. * On success, the return value of puff() is zero. If there is an error in the
  653. * source data, i.e. it is not in the deflate format, then a negative value is
  654. * returned. If there is not enough input available or there is not enough
  655. * output space, then a positive error is returned. In that case, destlen and
  656. * sourcelen are not updated to facilitate retrying from the beginning with the
  657. * provision of more input data or more output space. In the case of invalid
  658. * inflate data (a negative error), the dest and source pointers are updated to
  659. * facilitate the debugging of deflators.
  660. *
  661. * puff() also has a mode to determine the size of the uncompressed output with
  662. * no output written. For this dest must be (unsigned char *)0. In this case,
  663. * the input value of *destlen is ignored, and on return *destlen is set to the
  664. * size of the uncompressed output.
  665. *
  666. * The return codes are:
  667. *
  668. * 2: available inflate data did not terminate
  669. * 1: output space exhausted before completing inflate
  670. * 0: successful inflate
  671. * -1: invalid block type (type == 3)
  672. * -2: stored block length did not match one's complement
  673. * -3: dynamic block code description: too many length or distance codes
  674. * -4: dynamic block code description: code lengths codes incomplete
  675. * -5: dynamic block code description: repeat lengths with no first length
  676. * -6: dynamic block code description: repeat more than specified lengths
  677. * -7: dynamic block code description: invalid literal/length code lengths
  678. * -8: dynamic block code description: invalid distance code lengths
  679. * -9: invalid literal/length or distance code in fixed or dynamic block
  680. * -10: distance is too far back in fixed or dynamic block
  681. *
  682. * Format notes:
  683. *
  684. * - Three bits are read for each block to determine the kind of block and
  685. * whether or not it is the last block. Then the block is decoded and the
  686. * process repeated if it was not the last block.
  687. *
  688. * - The leftover bits in the last byte of the deflate data after the last
  689. * block (if it was a fixed or dynamic block) are undefined and have no
  690. * expected values to check.
  691. */
  692. int puff(unsigned char *dest, /* pointer to destination pointer */
  693. unsigned long *destlen, /* amount of output space */
  694. unsigned char *source, /* pointer to source data pointer */
  695. unsigned long *sourcelen) /* amount of input available */
  696. {
  697. struct state s; /* input/output state */
  698. int last, type; /* block information */
  699. int err; /* return value */
  700. /* initialize output state */
  701. s.out = dest;
  702. s.outlen = *destlen; /* ignored if dest is NIL */
  703. s.outcnt = 0;
  704. /* initialize input state */
  705. s.in = source;
  706. s.inlen = *sourcelen;
  707. s.incnt = 0;
  708. s.bitbuf = 0;
  709. s.bitcnt = 0;
  710. /* return if bits() or decode() tries to read past available input */
  711. if (setjmp(s.env) != 0) /* if came back here via longjmp() */
  712. err = 2; /* then skip do-loop, return error */
  713. else {
  714. /* process blocks until last block or error */
  715. do {
  716. last = bits(&s, 1); /* one if last block */
  717. type = bits(&s, 2); /* block type 0..3 */
  718. err = type == 0 ? stored(&s) :
  719. (type == 1 ? fixed(&s) :
  720. (type == 2 ? dynamic(&s) :
  721. -1)); /* type == 3, invalid */
  722. if (err != 0) break; /* return with error */
  723. } while (!last);
  724. }
  725. /* update the lengths and return */
  726. if (err <= 0) {
  727. *destlen = s.outcnt;
  728. *sourcelen = s.incnt;
  729. }
  730. return err;
  731. }
  732. #ifdef TEST
  733. /* Example of how to use puff() */
  734. #include <stdio.h>
  735. #include <stdlib.h>
  736. #include <sys/types.h>
  737. #include <sys/stat.h>
  738. local unsigned char *yank(char *name, unsigned long *len)
  739. {
  740. unsigned long size;
  741. unsigned char *buf;
  742. FILE *in;
  743. struct stat s;
  744. *len = 0;
  745. if (stat(name, &s)) return NULL;
  746. if ((s.st_mode & S_IFMT) != S_IFREG) return NULL;
  747. size = (unsigned long)(s.st_size);
  748. if (size == 0 || (off_t)size != s.st_size) return NULL;
  749. in = fopen(name, "r");
  750. if (in == NULL) return NULL;
  751. buf = malloc(size);
  752. if (buf != NULL && fread(buf, 1, size, in) != size) {
  753. free(buf);
  754. buf = NULL;
  755. }
  756. fclose(in);
  757. *len = size;
  758. return buf;
  759. }
  760. int main(int argc, char **argv)
  761. {
  762. int ret;
  763. unsigned char *source;
  764. unsigned long len, sourcelen, destlen;
  765. if (argc < 2) return 2;
  766. source = yank(argv[1], &len);
  767. if (source == NULL) return 2;
  768. sourcelen = len;
  769. ret = puff(NIL, &destlen, source, &sourcelen);
  770. if (ret)
  771. printf("puff() failed with return code %d\n", ret);
  772. else {
  773. printf("puff() succeeded uncompressing %lu bytes\n", destlen);
  774. if (sourcelen < len) printf("%lu compressed bytes unused\n",
  775. len - sourcelen);
  776. }
  777. free(source);
  778. return ret;
  779. }
  780. #endif