Print this page
11204 smatch issue in zlib/deflate.c
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/contrib/zlib/deflate.c
+++ new/usr/src/contrib/zlib/deflate.c
1 1 /* deflate.c -- compress data using the deflation algorithm
2 2 * Copyright (C) 1995-2017 Jean-loup Gailly and Mark Adler
3 3 * For conditions of distribution and use, see copyright notice in zlib.h
4 4 */
5 5
6 6 /*
7 7 * ALGORITHM
8 8 *
9 9 * The "deflation" process depends on being able to identify portions
10 10 * of the input text which are identical to earlier input (within a
11 11 * sliding window trailing behind the input currently being processed).
12 12 *
13 13 * The most straightforward technique turns out to be the fastest for
14 14 * most input files: try all possible matches and select the longest.
15 15 * The key feature of this algorithm is that insertions into the string
16 16 * dictionary are very simple and thus fast, and deletions are avoided
17 17 * completely. Insertions are performed at each input character, whereas
18 18 * string matches are performed only when the previous match ends. So it
19 19 * is preferable to spend more time in matches to allow very fast string
20 20 * insertions and avoid deletions. The matching algorithm for small
21 21 * strings is inspired from that of Rabin & Karp. A brute force approach
22 22 * is used to find longer strings when a small match has been found.
23 23 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze
24 24 * (by Leonid Broukhis).
25 25 * A previous version of this file used a more sophisticated algorithm
26 26 * (by Fiala and Greene) which is guaranteed to run in linear amortized
27 27 * time, but has a larger average cost, uses more memory and is patented.
28 28 * However the F&G algorithm may be faster for some highly redundant
29 29 * files if the parameter max_chain_length (described below) is too large.
30 30 *
31 31 * ACKNOWLEDGEMENTS
32 32 *
33 33 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and
34 34 * I found it in 'freeze' written by Leonid Broukhis.
35 35 * Thanks to many people for bug reports and testing.
36 36 *
37 37 * REFERENCES
38 38 *
39 39 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification".
40 40 * Available in http://tools.ietf.org/html/rfc1951
41 41 *
42 42 * A description of the Rabin and Karp algorithm is given in the book
43 43 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252.
44 44 *
45 45 * Fiala,E.R., and Greene,D.H.
46 46 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595
47 47 *
48 48 */
49 49
50 50 #include "deflate.h"
51 51
52 52 const char deflate_copyright[] =
53 53 " deflate 1.2.11 Copyright 1995-2017 Jean-loup Gailly and Mark Adler ";
54 54 /*
55 55 If you use the zlib library in a product, an acknowledgment is welcome
56 56 in the documentation of your product. If for some reason you cannot
57 57 include such an acknowledgment, I would appreciate that you keep this
58 58 copyright string in the executable of your product.
59 59 */
60 60
61 61 /* ===========================================================================
62 62 * Function prototypes.
63 63 */
64 64 typedef enum {
65 65 need_more, /* block not completed, need more input or more output */
66 66 block_done, /* block flush performed */
67 67 finish_started, /* finish started, need only more output at next deflate */
68 68 finish_done /* finish done, accept no more input or output */
69 69 } block_state;
70 70
71 71 typedef block_state (*compress_func) OF((deflate_state *s, int flush));
72 72 /* Compression function. Returns the block state after the call. */
73 73
74 74 local int deflateStateCheck OF((z_streamp strm));
75 75 local void slide_hash OF((deflate_state *s));
76 76 local void fill_window OF((deflate_state *s));
77 77 local block_state deflate_stored OF((deflate_state *s, int flush));
78 78 local block_state deflate_fast OF((deflate_state *s, int flush));
79 79 #ifndef FASTEST
80 80 local block_state deflate_slow OF((deflate_state *s, int flush));
81 81 #endif
82 82 local block_state deflate_rle OF((deflate_state *s, int flush));
83 83 local block_state deflate_huff OF((deflate_state *s, int flush));
84 84 local void lm_init OF((deflate_state *s));
85 85 local void putShortMSB OF((deflate_state *s, uInt b));
86 86 local void flush_pending OF((z_streamp strm));
87 87 local unsigned read_buf OF((z_streamp strm, Bytef *buf, unsigned size));
88 88 #ifdef ASMV
89 89 # pragma message("Assembler code may have bugs -- use at your own risk")
90 90 void match_init OF((void)); /* asm code initialization */
91 91 uInt longest_match OF((deflate_state *s, IPos cur_match));
92 92 #else
93 93 local uInt longest_match OF((deflate_state *s, IPos cur_match));
94 94 #endif
95 95
96 96 #ifdef ZLIB_DEBUG
97 97 local void check_match OF((deflate_state *s, IPos start, IPos match,
98 98 int length));
99 99 #endif
100 100
101 101 /* ===========================================================================
102 102 * Local data
103 103 */
104 104
105 105 #define NIL 0
106 106 /* Tail of hash chains */
107 107
108 108 #ifndef TOO_FAR
109 109 # define TOO_FAR 4096
110 110 #endif
111 111 /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
112 112
113 113 /* Values for max_lazy_match, good_match and max_chain_length, depending on
114 114 * the desired pack level (0..9). The values given below have been tuned to
115 115 * exclude worst case performance for pathological files. Better values may be
116 116 * found for specific files.
117 117 */
118 118 typedef struct config_s {
119 119 ush good_length; /* reduce lazy search above this match length */
120 120 ush max_lazy; /* do not perform lazy search above this match length */
121 121 ush nice_length; /* quit search above this match length */
122 122 ush max_chain;
123 123 compress_func func;
124 124 } config;
125 125
126 126 #ifdef FASTEST
127 127 local const config configuration_table[2] = {
128 128 /* good lazy nice chain */
129 129 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
130 130 /* 1 */ {4, 4, 8, 4, deflate_fast}}; /* max speed, no lazy matches */
131 131 #else
132 132 local const config configuration_table[10] = {
133 133 /* good lazy nice chain */
134 134 /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
135 135 /* 1 */ {4, 4, 8, 4, deflate_fast}, /* max speed, no lazy matches */
136 136 /* 2 */ {4, 5, 16, 8, deflate_fast},
137 137 /* 3 */ {4, 6, 32, 32, deflate_fast},
138 138
139 139 /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
140 140 /* 5 */ {8, 16, 32, 32, deflate_slow},
141 141 /* 6 */ {8, 16, 128, 128, deflate_slow},
142 142 /* 7 */ {8, 32, 128, 256, deflate_slow},
143 143 /* 8 */ {32, 128, 258, 1024, deflate_slow},
144 144 /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* max compression */
145 145 #endif
146 146
147 147 /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
148 148 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
149 149 * meaning.
150 150 */
151 151
152 152 /* rank Z_BLOCK between Z_NO_FLUSH and Z_PARTIAL_FLUSH */
153 153 #define RANK(f) (((f) * 2) - ((f) > 4 ? 9 : 0))
154 154
155 155 /* ===========================================================================
156 156 * Update a hash value with the given input byte
157 157 * IN assertion: all calls to UPDATE_HASH are made with consecutive input
158 158 * characters, so that a running hash key can be computed from the previous
159 159 * key instead of complete recalculation each time.
160 160 */
161 161 #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
162 162
163 163
164 164 /* ===========================================================================
165 165 * Insert string str in the dictionary and set match_head to the previous head
166 166 * of the hash chain (the most recent string with same hash key). Return
167 167 * the previous length of the hash chain.
168 168 * If this file is compiled with -DFASTEST, the compression level is forced
169 169 * to 1, and no hash chains are maintained.
170 170 * IN assertion: all calls to INSERT_STRING are made with consecutive input
171 171 * characters and the first MIN_MATCH bytes of str are valid (except for
172 172 * the last MIN_MATCH-1 bytes of the input file).
173 173 */
174 174 #ifdef FASTEST
175 175 #define INSERT_STRING(s, str, match_head) \
176 176 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
177 177 match_head = s->head[s->ins_h], \
178 178 s->head[s->ins_h] = (Pos)(str))
179 179 #else
180 180 #define INSERT_STRING(s, str, match_head) \
|
↓ open down ↓ |
180 lines elided |
↑ open up ↑ |
181 181 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
182 182 match_head = s->prev[(str) & s->w_mask] = s->head[s->ins_h], \
183 183 s->head[s->ins_h] = (Pos)(str))
184 184 #endif
185 185
186 186 /* ===========================================================================
187 187 * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
188 188 * prev[] will be initialized on the fly.
189 189 */
190 190 #define CLEAR_HASH(s) \
191 - s->head[s->hash_size-1] = NIL; \
192 - zmemzero((Bytef *)s->head, (unsigned)(s->hash_size-1)*sizeof(*s->head));
191 + do { \
192 + s->head[s->hash_size-1] = NIL; \
193 + zmemzero((Bytef *)s->head, \
194 + (unsigned)(s->hash_size-1)*sizeof(*s->head)); \
195 + } while (0)
193 196
194 197 /* ===========================================================================
195 198 * Slide the hash table when sliding the window down (could be avoided with 32
196 199 * bit values at the expense of memory usage). We slide even when level == 0 to
197 200 * keep the hash table consistent if we switch back to level > 0 later.
198 201 */
199 202 local void slide_hash(s)
200 203 deflate_state *s;
201 204 {
202 205 unsigned n, m;
203 206 Posf *p;
204 207 uInt wsize = s->w_size;
205 208
206 209 n = s->hash_size;
207 210 p = &s->head[n];
208 211 do {
209 212 m = *--p;
210 213 *p = (Pos)(m >= wsize ? m - wsize : NIL);
211 214 } while (--n);
212 215 n = wsize;
213 216 #ifndef FASTEST
214 217 p = &s->prev[n];
215 218 do {
216 219 m = *--p;
217 220 *p = (Pos)(m >= wsize ? m - wsize : NIL);
218 221 /* If n is not on any hash chain, prev[n] is garbage but
219 222 * its value will never be used.
220 223 */
221 224 } while (--n);
222 225 #endif
223 226 }
224 227
225 228 /* ========================================================================= */
226 229 int ZEXPORT deflateInit_(strm, level, version, stream_size)
227 230 z_streamp strm;
228 231 int level;
229 232 const char *version;
230 233 int stream_size;
231 234 {
232 235 return deflateInit2_(strm, level, Z_DEFLATED, MAX_WBITS, DEF_MEM_LEVEL,
233 236 Z_DEFAULT_STRATEGY, version, stream_size);
234 237 /* To do: ignore strm->next_in if we use it as window */
235 238 }
236 239
237 240 /* ========================================================================= */
238 241 int ZEXPORT deflateInit2_(strm, level, method, windowBits, memLevel, strategy,
239 242 version, stream_size)
240 243 z_streamp strm;
241 244 int level;
242 245 int method;
243 246 int windowBits;
244 247 int memLevel;
245 248 int strategy;
246 249 const char *version;
247 250 int stream_size;
248 251 {
249 252 deflate_state *s;
250 253 int wrap = 1;
251 254 static const char my_version[] = ZLIB_VERSION;
252 255
253 256 ushf *overlay;
254 257 /* We overlay pending_buf and d_buf+l_buf. This works since the average
255 258 * output size for (length,distance) codes is <= 24 bits.
256 259 */
257 260
258 261 if (version == Z_NULL || version[0] != my_version[0] ||
259 262 stream_size != sizeof(z_stream)) {
260 263 return Z_VERSION_ERROR;
261 264 }
262 265 if (strm == Z_NULL) return Z_STREAM_ERROR;
263 266
264 267 strm->msg = Z_NULL;
265 268 if (strm->zalloc == (alloc_func)0) {
266 269 #ifdef Z_SOLO
267 270 return Z_STREAM_ERROR;
268 271 #else
269 272 strm->zalloc = zcalloc;
270 273 strm->opaque = (voidpf)0;
271 274 #endif
272 275 }
273 276 if (strm->zfree == (free_func)0)
274 277 #ifdef Z_SOLO
275 278 return Z_STREAM_ERROR;
276 279 #else
277 280 strm->zfree = zcfree;
278 281 #endif
279 282
280 283 #ifdef FASTEST
281 284 if (level != 0) level = 1;
282 285 #else
283 286 if (level == Z_DEFAULT_COMPRESSION) level = 6;
284 287 #endif
285 288
286 289 if (windowBits < 0) { /* suppress zlib wrapper */
287 290 wrap = 0;
288 291 windowBits = -windowBits;
289 292 }
290 293 #ifdef GZIP
291 294 else if (windowBits > 15) {
292 295 wrap = 2; /* write gzip wrapper instead */
293 296 windowBits -= 16;
294 297 }
295 298 #endif
296 299 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
297 300 windowBits < 8 || windowBits > 15 || level < 0 || level > 9 ||
298 301 strategy < 0 || strategy > Z_FIXED || (windowBits == 8 && wrap != 1)) {
299 302 return Z_STREAM_ERROR;
300 303 }
301 304 if (windowBits == 8) windowBits = 9; /* until 256-byte window bug fixed */
302 305 s = (deflate_state *) ZALLOC(strm, 1, sizeof(deflate_state));
303 306 if (s == Z_NULL) return Z_MEM_ERROR;
304 307 strm->state = (struct internal_state FAR *)s;
305 308 s->strm = strm;
306 309 s->status = INIT_STATE; /* to pass state test in deflateReset() */
307 310
308 311 s->wrap = wrap;
309 312 s->gzhead = Z_NULL;
310 313 s->w_bits = (uInt)windowBits;
311 314 s->w_size = 1 << s->w_bits;
312 315 s->w_mask = s->w_size - 1;
313 316
314 317 s->hash_bits = (uInt)memLevel + 7;
315 318 s->hash_size = 1 << s->hash_bits;
316 319 s->hash_mask = s->hash_size - 1;
317 320 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
318 321
319 322 s->window = (Bytef *) ZALLOC(strm, s->w_size, 2*sizeof(Byte));
320 323 s->prev = (Posf *) ZALLOC(strm, s->w_size, sizeof(Pos));
321 324 s->head = (Posf *) ZALLOC(strm, s->hash_size, sizeof(Pos));
322 325
323 326 s->high_water = 0; /* nothing written to s->window yet */
324 327
325 328 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
326 329
327 330 overlay = (ushf *) ZALLOC(strm, s->lit_bufsize, sizeof(ush)+2);
328 331 s->pending_buf = (uchf *) overlay;
329 332 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
330 333
331 334 if (s->window == Z_NULL || s->prev == Z_NULL || s->head == Z_NULL ||
332 335 s->pending_buf == Z_NULL) {
333 336 s->status = FINISH_STATE;
334 337 strm->msg = ERR_MSG(Z_MEM_ERROR);
335 338 deflateEnd (strm);
336 339 return Z_MEM_ERROR;
337 340 }
338 341 s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
339 342 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
340 343
341 344 s->level = level;
342 345 s->strategy = strategy;
343 346 s->method = (Byte)method;
344 347
345 348 return deflateReset(strm);
346 349 }
347 350
348 351 /* =========================================================================
349 352 * Check for a valid deflate stream state. Return 0 if ok, 1 if not.
350 353 */
351 354 local int deflateStateCheck (strm)
352 355 z_streamp strm;
353 356 {
354 357 deflate_state *s;
355 358 if (strm == Z_NULL ||
356 359 strm->zalloc == (alloc_func)0 || strm->zfree == (free_func)0)
357 360 return 1;
358 361 s = strm->state;
359 362 if (s == Z_NULL || s->strm != strm || (s->status != INIT_STATE &&
360 363 #ifdef GZIP
361 364 s->status != GZIP_STATE &&
362 365 #endif
363 366 s->status != EXTRA_STATE &&
364 367 s->status != NAME_STATE &&
365 368 s->status != COMMENT_STATE &&
366 369 s->status != HCRC_STATE &&
367 370 s->status != BUSY_STATE &&
368 371 s->status != FINISH_STATE))
369 372 return 1;
370 373 return 0;
371 374 }
372 375
373 376 /* ========================================================================= */
374 377 int ZEXPORT deflateSetDictionary (strm, dictionary, dictLength)
375 378 z_streamp strm;
376 379 const Bytef *dictionary;
377 380 uInt dictLength;
378 381 {
379 382 deflate_state *s;
380 383 uInt str, n;
381 384 int wrap;
382 385 unsigned avail;
383 386 z_const unsigned char *next;
384 387
385 388 if (deflateStateCheck(strm) || dictionary == Z_NULL)
386 389 return Z_STREAM_ERROR;
387 390 s = strm->state;
388 391 wrap = s->wrap;
389 392 if (wrap == 2 || (wrap == 1 && s->status != INIT_STATE) || s->lookahead)
390 393 return Z_STREAM_ERROR;
391 394
392 395 /* when using zlib wrappers, compute Adler-32 for provided dictionary */
393 396 if (wrap == 1)
394 397 strm->adler = adler32(strm->adler, dictionary, dictLength);
395 398 s->wrap = 0; /* avoid computing Adler-32 in read_buf */
396 399
397 400 /* if dictionary would fill window, just replace the history */
398 401 if (dictLength >= s->w_size) {
399 402 if (wrap == 0) { /* already empty otherwise */
400 403 CLEAR_HASH(s);
401 404 s->strstart = 0;
402 405 s->block_start = 0L;
403 406 s->insert = 0;
404 407 }
405 408 dictionary += dictLength - s->w_size; /* use the tail */
406 409 dictLength = s->w_size;
407 410 }
408 411
409 412 /* insert dictionary into window and hash */
410 413 avail = strm->avail_in;
411 414 next = strm->next_in;
412 415 strm->avail_in = dictLength;
413 416 strm->next_in = (z_const Bytef *)dictionary;
414 417 fill_window(s);
415 418 while (s->lookahead >= MIN_MATCH) {
416 419 str = s->strstart;
417 420 n = s->lookahead - (MIN_MATCH-1);
418 421 do {
419 422 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
420 423 #ifndef FASTEST
421 424 s->prev[str & s->w_mask] = s->head[s->ins_h];
422 425 #endif
423 426 s->head[s->ins_h] = (Pos)str;
424 427 str++;
425 428 } while (--n);
426 429 s->strstart = str;
427 430 s->lookahead = MIN_MATCH-1;
428 431 fill_window(s);
429 432 }
430 433 s->strstart += s->lookahead;
431 434 s->block_start = (long)s->strstart;
432 435 s->insert = s->lookahead;
433 436 s->lookahead = 0;
434 437 s->match_length = s->prev_length = MIN_MATCH-1;
435 438 s->match_available = 0;
436 439 strm->next_in = next;
437 440 strm->avail_in = avail;
438 441 s->wrap = wrap;
439 442 return Z_OK;
440 443 }
441 444
442 445 /* ========================================================================= */
443 446 int ZEXPORT deflateGetDictionary (strm, dictionary, dictLength)
444 447 z_streamp strm;
445 448 Bytef *dictionary;
446 449 uInt *dictLength;
447 450 {
448 451 deflate_state *s;
449 452 uInt len;
450 453
451 454 if (deflateStateCheck(strm))
452 455 return Z_STREAM_ERROR;
453 456 s = strm->state;
454 457 len = s->strstart + s->lookahead;
455 458 if (len > s->w_size)
456 459 len = s->w_size;
457 460 if (dictionary != Z_NULL && len)
458 461 zmemcpy(dictionary, s->window + s->strstart + s->lookahead - len, len);
459 462 if (dictLength != Z_NULL)
460 463 *dictLength = len;
461 464 return Z_OK;
462 465 }
463 466
464 467 /* ========================================================================= */
465 468 int ZEXPORT deflateResetKeep (strm)
466 469 z_streamp strm;
467 470 {
468 471 deflate_state *s;
469 472
470 473 if (deflateStateCheck(strm)) {
471 474 return Z_STREAM_ERROR;
472 475 }
473 476
474 477 strm->total_in = strm->total_out = 0;
475 478 strm->msg = Z_NULL; /* use zfree if we ever allocate msg dynamically */
476 479 strm->data_type = Z_UNKNOWN;
477 480
478 481 s = (deflate_state *)strm->state;
479 482 s->pending = 0;
480 483 s->pending_out = s->pending_buf;
481 484
482 485 if (s->wrap < 0) {
483 486 s->wrap = -s->wrap; /* was made negative by deflate(..., Z_FINISH); */
484 487 }
485 488 s->status =
486 489 #ifdef GZIP
487 490 s->wrap == 2 ? GZIP_STATE :
488 491 #endif
489 492 s->wrap ? INIT_STATE : BUSY_STATE;
490 493 strm->adler =
491 494 #ifdef GZIP
492 495 s->wrap == 2 ? crc32(0L, Z_NULL, 0) :
493 496 #endif
494 497 adler32(0L, Z_NULL, 0);
495 498 s->last_flush = Z_NO_FLUSH;
496 499
497 500 _tr_init(s);
498 501
499 502 return Z_OK;
500 503 }
501 504
502 505 /* ========================================================================= */
503 506 int ZEXPORT deflateReset (strm)
504 507 z_streamp strm;
505 508 {
506 509 int ret;
507 510
508 511 ret = deflateResetKeep(strm);
509 512 if (ret == Z_OK)
510 513 lm_init(strm->state);
511 514 return ret;
512 515 }
513 516
514 517 /* ========================================================================= */
515 518 int ZEXPORT deflateSetHeader (strm, head)
516 519 z_streamp strm;
517 520 gz_headerp head;
518 521 {
519 522 if (deflateStateCheck(strm) || strm->state->wrap != 2)
520 523 return Z_STREAM_ERROR;
521 524 strm->state->gzhead = head;
522 525 return Z_OK;
523 526 }
524 527
525 528 /* ========================================================================= */
526 529 int ZEXPORT deflatePending (strm, pending, bits)
527 530 unsigned *pending;
528 531 int *bits;
529 532 z_streamp strm;
530 533 {
531 534 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
532 535 if (pending != Z_NULL)
533 536 *pending = strm->state->pending;
534 537 if (bits != Z_NULL)
535 538 *bits = strm->state->bi_valid;
536 539 return Z_OK;
537 540 }
538 541
539 542 /* ========================================================================= */
540 543 int ZEXPORT deflatePrime (strm, bits, value)
541 544 z_streamp strm;
542 545 int bits;
543 546 int value;
544 547 {
545 548 deflate_state *s;
546 549 int put;
547 550
548 551 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
549 552 s = strm->state;
550 553 if ((Bytef *)(s->d_buf) < s->pending_out + ((Buf_size + 7) >> 3))
551 554 return Z_BUF_ERROR;
552 555 do {
553 556 put = Buf_size - s->bi_valid;
554 557 if (put > bits)
555 558 put = bits;
556 559 s->bi_buf |= (ush)((value & ((1 << put) - 1)) << s->bi_valid);
557 560 s->bi_valid += put;
558 561 _tr_flush_bits(s);
559 562 value >>= put;
560 563 bits -= put;
561 564 } while (bits);
562 565 return Z_OK;
563 566 }
564 567
565 568 /* ========================================================================= */
566 569 int ZEXPORT deflateParams(strm, level, strategy)
567 570 z_streamp strm;
568 571 int level;
569 572 int strategy;
570 573 {
571 574 deflate_state *s;
572 575 compress_func func;
573 576
574 577 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
575 578 s = strm->state;
576 579
577 580 #ifdef FASTEST
578 581 if (level != 0) level = 1;
579 582 #else
580 583 if (level == Z_DEFAULT_COMPRESSION) level = 6;
581 584 #endif
582 585 if (level < 0 || level > 9 || strategy < 0 || strategy > Z_FIXED) {
583 586 return Z_STREAM_ERROR;
584 587 }
585 588 func = configuration_table[s->level].func;
586 589
587 590 if ((strategy != s->strategy || func != configuration_table[level].func) &&
588 591 s->high_water) {
589 592 /* Flush the last buffer: */
590 593 int err = deflate(strm, Z_BLOCK);
591 594 if (err == Z_STREAM_ERROR)
592 595 return err;
593 596 if (strm->avail_out == 0)
594 597 return Z_BUF_ERROR;
595 598 }
596 599 if (s->level != level) {
597 600 if (s->level == 0 && s->matches != 0) {
598 601 if (s->matches == 1)
599 602 slide_hash(s);
600 603 else
601 604 CLEAR_HASH(s);
602 605 s->matches = 0;
603 606 }
604 607 s->level = level;
605 608 s->max_lazy_match = configuration_table[level].max_lazy;
606 609 s->good_match = configuration_table[level].good_length;
607 610 s->nice_match = configuration_table[level].nice_length;
608 611 s->max_chain_length = configuration_table[level].max_chain;
609 612 }
610 613 s->strategy = strategy;
611 614 return Z_OK;
612 615 }
613 616
614 617 /* ========================================================================= */
615 618 int ZEXPORT deflateTune(strm, good_length, max_lazy, nice_length, max_chain)
616 619 z_streamp strm;
617 620 int good_length;
618 621 int max_lazy;
619 622 int nice_length;
620 623 int max_chain;
621 624 {
622 625 deflate_state *s;
623 626
624 627 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
625 628 s = strm->state;
626 629 s->good_match = (uInt)good_length;
627 630 s->max_lazy_match = (uInt)max_lazy;
628 631 s->nice_match = nice_length;
629 632 s->max_chain_length = (uInt)max_chain;
630 633 return Z_OK;
631 634 }
632 635
633 636 /* =========================================================================
634 637 * For the default windowBits of 15 and memLevel of 8, this function returns
635 638 * a close to exact, as well as small, upper bound on the compressed size.
636 639 * They are coded as constants here for a reason--if the #define's are
637 640 * changed, then this function needs to be changed as well. The return
638 641 * value for 15 and 8 only works for those exact settings.
639 642 *
640 643 * For any setting other than those defaults for windowBits and memLevel,
641 644 * the value returned is a conservative worst case for the maximum expansion
642 645 * resulting from using fixed blocks instead of stored blocks, which deflate
643 646 * can emit on compressed data for some combinations of the parameters.
644 647 *
645 648 * This function could be more sophisticated to provide closer upper bounds for
646 649 * every combination of windowBits and memLevel. But even the conservative
647 650 * upper bound of about 14% expansion does not seem onerous for output buffer
648 651 * allocation.
649 652 */
650 653 uLong ZEXPORT deflateBound(strm, sourceLen)
651 654 z_streamp strm;
652 655 uLong sourceLen;
653 656 {
654 657 deflate_state *s;
655 658 uLong complen, wraplen;
656 659
657 660 /* conservative upper bound for compressed data */
658 661 complen = sourceLen +
659 662 ((sourceLen + 7) >> 3) + ((sourceLen + 63) >> 6) + 5;
660 663
661 664 /* if can't get parameters, return conservative bound plus zlib wrapper */
662 665 if (deflateStateCheck(strm))
663 666 return complen + 6;
664 667
665 668 /* compute wrapper length */
666 669 s = strm->state;
667 670 switch (s->wrap) {
668 671 case 0: /* raw deflate */
669 672 wraplen = 0;
670 673 break;
671 674 case 1: /* zlib wrapper */
672 675 wraplen = 6 + (s->strstart ? 4 : 0);
673 676 break;
674 677 #ifdef GZIP
675 678 case 2: /* gzip wrapper */
676 679 wraplen = 18;
677 680 if (s->gzhead != Z_NULL) { /* user-supplied gzip header */
678 681 Bytef *str;
679 682 if (s->gzhead->extra != Z_NULL)
680 683 wraplen += 2 + s->gzhead->extra_len;
681 684 str = s->gzhead->name;
682 685 if (str != Z_NULL)
683 686 do {
684 687 wraplen++;
685 688 } while (*str++);
686 689 str = s->gzhead->comment;
687 690 if (str != Z_NULL)
688 691 do {
689 692 wraplen++;
690 693 } while (*str++);
691 694 if (s->gzhead->hcrc)
692 695 wraplen += 2;
693 696 }
694 697 break;
695 698 #endif
696 699 default: /* for compiler happiness */
697 700 wraplen = 6;
698 701 }
699 702
700 703 /* if not default parameters, return conservative bound */
701 704 if (s->w_bits != 15 || s->hash_bits != 8 + 7)
702 705 return complen + wraplen;
703 706
704 707 /* default settings: return tight bound for that case */
705 708 return sourceLen + (sourceLen >> 12) + (sourceLen >> 14) +
706 709 (sourceLen >> 25) + 13 - 6 + wraplen;
707 710 }
708 711
709 712 /* =========================================================================
710 713 * Put a short in the pending buffer. The 16-bit value is put in MSB order.
711 714 * IN assertion: the stream state is correct and there is enough room in
712 715 * pending_buf.
713 716 */
714 717 local void putShortMSB (s, b)
715 718 deflate_state *s;
716 719 uInt b;
717 720 {
718 721 put_byte(s, (Byte)(b >> 8));
719 722 put_byte(s, (Byte)(b & 0xff));
720 723 }
721 724
722 725 /* =========================================================================
723 726 * Flush as much pending output as possible. All deflate() output, except for
724 727 * some deflate_stored() output, goes through this function so some
725 728 * applications may wish to modify it to avoid allocating a large
726 729 * strm->next_out buffer and copying into it. (See also read_buf()).
727 730 */
728 731 local void flush_pending(strm)
729 732 z_streamp strm;
730 733 {
731 734 unsigned len;
732 735 deflate_state *s = strm->state;
733 736
734 737 _tr_flush_bits(s);
735 738 len = s->pending;
736 739 if (len > strm->avail_out) len = strm->avail_out;
737 740 if (len == 0) return;
738 741
739 742 zmemcpy(strm->next_out, s->pending_out, len);
740 743 strm->next_out += len;
741 744 s->pending_out += len;
742 745 strm->total_out += len;
743 746 strm->avail_out -= len;
744 747 s->pending -= len;
745 748 if (s->pending == 0) {
746 749 s->pending_out = s->pending_buf;
747 750 }
748 751 }
749 752
750 753 /* ===========================================================================
751 754 * Update the header CRC with the bytes s->pending_buf[beg..s->pending - 1].
752 755 */
753 756 #define HCRC_UPDATE(beg) \
754 757 do { \
755 758 if (s->gzhead->hcrc && s->pending > (beg)) \
756 759 strm->adler = crc32(strm->adler, s->pending_buf + (beg), \
757 760 s->pending - (beg)); \
758 761 } while (0)
759 762
760 763 /* ========================================================================= */
761 764 int ZEXPORT deflate (strm, flush)
762 765 z_streamp strm;
763 766 int flush;
764 767 {
765 768 int old_flush; /* value of flush param for previous deflate call */
766 769 deflate_state *s;
767 770
768 771 if (deflateStateCheck(strm) || flush > Z_BLOCK || flush < 0) {
769 772 return Z_STREAM_ERROR;
770 773 }
771 774 s = strm->state;
772 775
773 776 if (strm->next_out == Z_NULL ||
774 777 (strm->avail_in != 0 && strm->next_in == Z_NULL) ||
775 778 (s->status == FINISH_STATE && flush != Z_FINISH)) {
776 779 ERR_RETURN(strm, Z_STREAM_ERROR);
777 780 }
778 781 if (strm->avail_out == 0) ERR_RETURN(strm, Z_BUF_ERROR);
779 782
780 783 old_flush = s->last_flush;
781 784 s->last_flush = flush;
782 785
783 786 /* Flush as much pending output as possible */
784 787 if (s->pending != 0) {
785 788 flush_pending(strm);
786 789 if (strm->avail_out == 0) {
787 790 /* Since avail_out is 0, deflate will be called again with
788 791 * more output space, but possibly with both pending and
789 792 * avail_in equal to zero. There won't be anything to do,
790 793 * but this is not an error situation so make sure we
791 794 * return OK instead of BUF_ERROR at next call of deflate:
792 795 */
793 796 s->last_flush = -1;
794 797 return Z_OK;
795 798 }
796 799
797 800 /* Make sure there is something to do and avoid duplicate consecutive
798 801 * flushes. For repeated and useless calls with Z_FINISH, we keep
799 802 * returning Z_STREAM_END instead of Z_BUF_ERROR.
800 803 */
801 804 } else if (strm->avail_in == 0 && RANK(flush) <= RANK(old_flush) &&
802 805 flush != Z_FINISH) {
803 806 ERR_RETURN(strm, Z_BUF_ERROR);
804 807 }
805 808
806 809 /* User must not provide more input after the first FINISH: */
807 810 if (s->status == FINISH_STATE && strm->avail_in != 0) {
808 811 ERR_RETURN(strm, Z_BUF_ERROR);
809 812 }
810 813
811 814 /* Write the header */
812 815 if (s->status == INIT_STATE) {
813 816 /* zlib header */
814 817 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
815 818 uInt level_flags;
816 819
817 820 if (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2)
818 821 level_flags = 0;
819 822 else if (s->level < 6)
820 823 level_flags = 1;
821 824 else if (s->level == 6)
822 825 level_flags = 2;
823 826 else
824 827 level_flags = 3;
825 828 header |= (level_flags << 6);
826 829 if (s->strstart != 0) header |= PRESET_DICT;
827 830 header += 31 - (header % 31);
828 831
829 832 putShortMSB(s, header);
830 833
831 834 /* Save the adler32 of the preset dictionary: */
832 835 if (s->strstart != 0) {
833 836 putShortMSB(s, (uInt)(strm->adler >> 16));
834 837 putShortMSB(s, (uInt)(strm->adler & 0xffff));
835 838 }
836 839 strm->adler = adler32(0L, Z_NULL, 0);
837 840 s->status = BUSY_STATE;
838 841
839 842 /* Compression must start with an empty pending buffer */
840 843 flush_pending(strm);
841 844 if (s->pending != 0) {
842 845 s->last_flush = -1;
843 846 return Z_OK;
844 847 }
845 848 }
846 849 #ifdef GZIP
847 850 if (s->status == GZIP_STATE) {
848 851 /* gzip header */
849 852 strm->adler = crc32(0L, Z_NULL, 0);
850 853 put_byte(s, 31);
851 854 put_byte(s, 139);
852 855 put_byte(s, 8);
853 856 if (s->gzhead == Z_NULL) {
854 857 put_byte(s, 0);
855 858 put_byte(s, 0);
856 859 put_byte(s, 0);
857 860 put_byte(s, 0);
858 861 put_byte(s, 0);
859 862 put_byte(s, s->level == 9 ? 2 :
860 863 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
861 864 4 : 0));
862 865 put_byte(s, OS_CODE);
863 866 s->status = BUSY_STATE;
864 867
865 868 /* Compression must start with an empty pending buffer */
866 869 flush_pending(strm);
867 870 if (s->pending != 0) {
868 871 s->last_flush = -1;
869 872 return Z_OK;
870 873 }
871 874 }
872 875 else {
873 876 put_byte(s, (s->gzhead->text ? 1 : 0) +
874 877 (s->gzhead->hcrc ? 2 : 0) +
875 878 (s->gzhead->extra == Z_NULL ? 0 : 4) +
876 879 (s->gzhead->name == Z_NULL ? 0 : 8) +
877 880 (s->gzhead->comment == Z_NULL ? 0 : 16)
878 881 );
879 882 put_byte(s, (Byte)(s->gzhead->time & 0xff));
880 883 put_byte(s, (Byte)((s->gzhead->time >> 8) & 0xff));
881 884 put_byte(s, (Byte)((s->gzhead->time >> 16) & 0xff));
882 885 put_byte(s, (Byte)((s->gzhead->time >> 24) & 0xff));
883 886 put_byte(s, s->level == 9 ? 2 :
884 887 (s->strategy >= Z_HUFFMAN_ONLY || s->level < 2 ?
885 888 4 : 0));
886 889 put_byte(s, s->gzhead->os & 0xff);
887 890 if (s->gzhead->extra != Z_NULL) {
888 891 put_byte(s, s->gzhead->extra_len & 0xff);
889 892 put_byte(s, (s->gzhead->extra_len >> 8) & 0xff);
890 893 }
891 894 if (s->gzhead->hcrc)
892 895 strm->adler = crc32(strm->adler, s->pending_buf,
893 896 s->pending);
894 897 s->gzindex = 0;
895 898 s->status = EXTRA_STATE;
896 899 }
897 900 }
898 901 if (s->status == EXTRA_STATE) {
899 902 if (s->gzhead->extra != Z_NULL) {
900 903 ulg beg = s->pending; /* start of bytes to update crc */
901 904 uInt left = (s->gzhead->extra_len & 0xffff) - s->gzindex;
902 905 while (s->pending + left > s->pending_buf_size) {
903 906 uInt copy = s->pending_buf_size - s->pending;
904 907 zmemcpy(s->pending_buf + s->pending,
905 908 s->gzhead->extra + s->gzindex, copy);
906 909 s->pending = s->pending_buf_size;
907 910 HCRC_UPDATE(beg);
908 911 s->gzindex += copy;
909 912 flush_pending(strm);
910 913 if (s->pending != 0) {
911 914 s->last_flush = -1;
912 915 return Z_OK;
913 916 }
914 917 beg = 0;
915 918 left -= copy;
916 919 }
917 920 zmemcpy(s->pending_buf + s->pending,
918 921 s->gzhead->extra + s->gzindex, left);
919 922 s->pending += left;
920 923 HCRC_UPDATE(beg);
921 924 s->gzindex = 0;
922 925 }
923 926 s->status = NAME_STATE;
924 927 }
925 928 if (s->status == NAME_STATE) {
926 929 if (s->gzhead->name != Z_NULL) {
927 930 ulg beg = s->pending; /* start of bytes to update crc */
928 931 int val;
929 932 do {
930 933 if (s->pending == s->pending_buf_size) {
931 934 HCRC_UPDATE(beg);
932 935 flush_pending(strm);
933 936 if (s->pending != 0) {
934 937 s->last_flush = -1;
935 938 return Z_OK;
936 939 }
937 940 beg = 0;
938 941 }
939 942 val = s->gzhead->name[s->gzindex++];
940 943 put_byte(s, val);
941 944 } while (val != 0);
942 945 HCRC_UPDATE(beg);
943 946 s->gzindex = 0;
944 947 }
945 948 s->status = COMMENT_STATE;
946 949 }
947 950 if (s->status == COMMENT_STATE) {
948 951 if (s->gzhead->comment != Z_NULL) {
949 952 ulg beg = s->pending; /* start of bytes to update crc */
950 953 int val;
951 954 do {
952 955 if (s->pending == s->pending_buf_size) {
953 956 HCRC_UPDATE(beg);
954 957 flush_pending(strm);
955 958 if (s->pending != 0) {
956 959 s->last_flush = -1;
957 960 return Z_OK;
958 961 }
959 962 beg = 0;
960 963 }
961 964 val = s->gzhead->comment[s->gzindex++];
962 965 put_byte(s, val);
963 966 } while (val != 0);
964 967 HCRC_UPDATE(beg);
965 968 }
966 969 s->status = HCRC_STATE;
967 970 }
968 971 if (s->status == HCRC_STATE) {
969 972 if (s->gzhead->hcrc) {
970 973 if (s->pending + 2 > s->pending_buf_size) {
971 974 flush_pending(strm);
972 975 if (s->pending != 0) {
973 976 s->last_flush = -1;
974 977 return Z_OK;
975 978 }
976 979 }
977 980 put_byte(s, (Byte)(strm->adler & 0xff));
978 981 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
979 982 strm->adler = crc32(0L, Z_NULL, 0);
980 983 }
981 984 s->status = BUSY_STATE;
982 985
983 986 /* Compression must start with an empty pending buffer */
984 987 flush_pending(strm);
985 988 if (s->pending != 0) {
986 989 s->last_flush = -1;
987 990 return Z_OK;
988 991 }
989 992 }
990 993 #endif
991 994
992 995 /* Start a new block or continue the current one.
993 996 */
994 997 if (strm->avail_in != 0 || s->lookahead != 0 ||
995 998 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
996 999 block_state bstate;
997 1000
998 1001 bstate = s->level == 0 ? deflate_stored(s, flush) :
999 1002 s->strategy == Z_HUFFMAN_ONLY ? deflate_huff(s, flush) :
1000 1003 s->strategy == Z_RLE ? deflate_rle(s, flush) :
1001 1004 (*(configuration_table[s->level].func))(s, flush);
1002 1005
1003 1006 if (bstate == finish_started || bstate == finish_done) {
1004 1007 s->status = FINISH_STATE;
1005 1008 }
1006 1009 if (bstate == need_more || bstate == finish_started) {
1007 1010 if (strm->avail_out == 0) {
1008 1011 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
1009 1012 }
1010 1013 return Z_OK;
1011 1014 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
1012 1015 * of deflate should use the same flush parameter to make sure
1013 1016 * that the flush is complete. So we don't have to output an
1014 1017 * empty block here, this will be done at next call. This also
1015 1018 * ensures that for a very small output buffer, we emit at most
1016 1019 * one empty block.
1017 1020 */
1018 1021 }
1019 1022 if (bstate == block_done) {
1020 1023 if (flush == Z_PARTIAL_FLUSH) {
1021 1024 _tr_align(s);
1022 1025 } else if (flush != Z_BLOCK) { /* FULL_FLUSH or SYNC_FLUSH */
1023 1026 _tr_stored_block(s, (char*)0, 0L, 0);
1024 1027 /* For a full flush, this empty block will be recognized
1025 1028 * as a special marker by inflate_sync().
1026 1029 */
1027 1030 if (flush == Z_FULL_FLUSH) {
1028 1031 CLEAR_HASH(s); /* forget history */
1029 1032 if (s->lookahead == 0) {
1030 1033 s->strstart = 0;
1031 1034 s->block_start = 0L;
1032 1035 s->insert = 0;
1033 1036 }
1034 1037 }
1035 1038 }
1036 1039 flush_pending(strm);
1037 1040 if (strm->avail_out == 0) {
1038 1041 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
1039 1042 return Z_OK;
1040 1043 }
1041 1044 }
1042 1045 }
1043 1046
1044 1047 if (flush != Z_FINISH) return Z_OK;
1045 1048 if (s->wrap <= 0) return Z_STREAM_END;
1046 1049
1047 1050 /* Write the trailer */
1048 1051 #ifdef GZIP
1049 1052 if (s->wrap == 2) {
1050 1053 put_byte(s, (Byte)(strm->adler & 0xff));
1051 1054 put_byte(s, (Byte)((strm->adler >> 8) & 0xff));
1052 1055 put_byte(s, (Byte)((strm->adler >> 16) & 0xff));
1053 1056 put_byte(s, (Byte)((strm->adler >> 24) & 0xff));
1054 1057 put_byte(s, (Byte)(strm->total_in & 0xff));
1055 1058 put_byte(s, (Byte)((strm->total_in >> 8) & 0xff));
1056 1059 put_byte(s, (Byte)((strm->total_in >> 16) & 0xff));
1057 1060 put_byte(s, (Byte)((strm->total_in >> 24) & 0xff));
1058 1061 }
1059 1062 else
1060 1063 #endif
1061 1064 {
1062 1065 putShortMSB(s, (uInt)(strm->adler >> 16));
1063 1066 putShortMSB(s, (uInt)(strm->adler & 0xffff));
1064 1067 }
1065 1068 flush_pending(strm);
1066 1069 /* If avail_out is zero, the application will call deflate again
1067 1070 * to flush the rest.
1068 1071 */
1069 1072 if (s->wrap > 0) s->wrap = -s->wrap; /* write the trailer only once! */
1070 1073 return s->pending != 0 ? Z_OK : Z_STREAM_END;
1071 1074 }
1072 1075
1073 1076 /* ========================================================================= */
1074 1077 int ZEXPORT deflateEnd (strm)
1075 1078 z_streamp strm;
1076 1079 {
1077 1080 int status;
1078 1081
1079 1082 if (deflateStateCheck(strm)) return Z_STREAM_ERROR;
1080 1083
1081 1084 status = strm->state->status;
1082 1085
1083 1086 /* Deallocate in reverse order of allocations: */
1084 1087 TRY_FREE(strm, strm->state->pending_buf);
1085 1088 TRY_FREE(strm, strm->state->head);
1086 1089 TRY_FREE(strm, strm->state->prev);
1087 1090 TRY_FREE(strm, strm->state->window);
1088 1091
1089 1092 ZFREE(strm, strm->state);
1090 1093 strm->state = Z_NULL;
1091 1094
1092 1095 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
1093 1096 }
1094 1097
1095 1098 /* =========================================================================
1096 1099 * Copy the source state to the destination state.
1097 1100 * To simplify the source, this is not supported for 16-bit MSDOS (which
1098 1101 * doesn't have enough memory anyway to duplicate compression states).
1099 1102 */
1100 1103 int ZEXPORT deflateCopy (dest, source)
1101 1104 z_streamp dest;
1102 1105 z_streamp source;
1103 1106 {
1104 1107 #ifdef MAXSEG_64K
1105 1108 return Z_STREAM_ERROR;
1106 1109 #else
1107 1110 deflate_state *ds;
1108 1111 deflate_state *ss;
1109 1112 ushf *overlay;
1110 1113
1111 1114
1112 1115 if (deflateStateCheck(source) || dest == Z_NULL) {
1113 1116 return Z_STREAM_ERROR;
1114 1117 }
1115 1118
1116 1119 ss = source->state;
1117 1120
1118 1121 zmemcpy((voidpf)dest, (voidpf)source, sizeof(z_stream));
1119 1122
1120 1123 ds = (deflate_state *) ZALLOC(dest, 1, sizeof(deflate_state));
1121 1124 if (ds == Z_NULL) return Z_MEM_ERROR;
1122 1125 dest->state = (struct internal_state FAR *) ds;
1123 1126 zmemcpy((voidpf)ds, (voidpf)ss, sizeof(deflate_state));
1124 1127 ds->strm = dest;
1125 1128
1126 1129 ds->window = (Bytef *) ZALLOC(dest, ds->w_size, 2*sizeof(Byte));
1127 1130 ds->prev = (Posf *) ZALLOC(dest, ds->w_size, sizeof(Pos));
1128 1131 ds->head = (Posf *) ZALLOC(dest, ds->hash_size, sizeof(Pos));
1129 1132 overlay = (ushf *) ZALLOC(dest, ds->lit_bufsize, sizeof(ush)+2);
1130 1133 ds->pending_buf = (uchf *) overlay;
1131 1134
1132 1135 if (ds->window == Z_NULL || ds->prev == Z_NULL || ds->head == Z_NULL ||
1133 1136 ds->pending_buf == Z_NULL) {
1134 1137 deflateEnd (dest);
1135 1138 return Z_MEM_ERROR;
1136 1139 }
1137 1140 /* following zmemcpy do not work for 16-bit MSDOS */
1138 1141 zmemcpy(ds->window, ss->window, ds->w_size * 2 * sizeof(Byte));
1139 1142 zmemcpy((voidpf)ds->prev, (voidpf)ss->prev, ds->w_size * sizeof(Pos));
1140 1143 zmemcpy((voidpf)ds->head, (voidpf)ss->head, ds->hash_size * sizeof(Pos));
1141 1144 zmemcpy(ds->pending_buf, ss->pending_buf, (uInt)ds->pending_buf_size);
1142 1145
1143 1146 ds->pending_out = ds->pending_buf + (ss->pending_out - ss->pending_buf);
1144 1147 ds->d_buf = overlay + ds->lit_bufsize/sizeof(ush);
1145 1148 ds->l_buf = ds->pending_buf + (1+sizeof(ush))*ds->lit_bufsize;
1146 1149
1147 1150 ds->l_desc.dyn_tree = ds->dyn_ltree;
1148 1151 ds->d_desc.dyn_tree = ds->dyn_dtree;
1149 1152 ds->bl_desc.dyn_tree = ds->bl_tree;
1150 1153
1151 1154 return Z_OK;
1152 1155 #endif /* MAXSEG_64K */
1153 1156 }
1154 1157
1155 1158 /* ===========================================================================
1156 1159 * Read a new buffer from the current input stream, update the adler32
1157 1160 * and total number of bytes read. All deflate() input goes through
1158 1161 * this function so some applications may wish to modify it to avoid
1159 1162 * allocating a large strm->next_in buffer and copying from it.
1160 1163 * (See also flush_pending()).
1161 1164 */
1162 1165 local unsigned read_buf(strm, buf, size)
1163 1166 z_streamp strm;
1164 1167 Bytef *buf;
1165 1168 unsigned size;
1166 1169 {
1167 1170 unsigned len = strm->avail_in;
1168 1171
1169 1172 if (len > size) len = size;
1170 1173 if (len == 0) return 0;
1171 1174
1172 1175 strm->avail_in -= len;
1173 1176
1174 1177 zmemcpy(buf, strm->next_in, len);
1175 1178 if (strm->state->wrap == 1) {
1176 1179 strm->adler = adler32(strm->adler, buf, len);
1177 1180 }
1178 1181 #ifdef GZIP
1179 1182 else if (strm->state->wrap == 2) {
1180 1183 strm->adler = crc32(strm->adler, buf, len);
1181 1184 }
1182 1185 #endif
1183 1186 strm->next_in += len;
1184 1187 strm->total_in += len;
1185 1188
1186 1189 return len;
1187 1190 }
1188 1191
1189 1192 /* ===========================================================================
1190 1193 * Initialize the "longest match" routines for a new zlib stream
1191 1194 */
1192 1195 local void lm_init (s)
1193 1196 deflate_state *s;
1194 1197 {
1195 1198 s->window_size = (ulg)2L*s->w_size;
1196 1199
1197 1200 CLEAR_HASH(s);
1198 1201
1199 1202 /* Set the default configuration parameters:
1200 1203 */
1201 1204 s->max_lazy_match = configuration_table[s->level].max_lazy;
1202 1205 s->good_match = configuration_table[s->level].good_length;
1203 1206 s->nice_match = configuration_table[s->level].nice_length;
1204 1207 s->max_chain_length = configuration_table[s->level].max_chain;
1205 1208
1206 1209 s->strstart = 0;
1207 1210 s->block_start = 0L;
1208 1211 s->lookahead = 0;
1209 1212 s->insert = 0;
1210 1213 s->match_length = s->prev_length = MIN_MATCH-1;
1211 1214 s->match_available = 0;
1212 1215 s->ins_h = 0;
1213 1216 #ifndef FASTEST
1214 1217 #ifdef ASMV
1215 1218 match_init(); /* initialize the asm code */
1216 1219 #endif
1217 1220 #endif
1218 1221 }
1219 1222
1220 1223 #ifndef FASTEST
1221 1224 /* ===========================================================================
1222 1225 * Set match_start to the longest match starting at the given string and
1223 1226 * return its length. Matches shorter or equal to prev_length are discarded,
1224 1227 * in which case the result is equal to prev_length and match_start is
1225 1228 * garbage.
1226 1229 * IN assertions: cur_match is the head of the hash chain for the current
1227 1230 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
1228 1231 * OUT assertion: the match length is not greater than s->lookahead.
1229 1232 */
1230 1233 #ifndef ASMV
1231 1234 /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
1232 1235 * match.S. The code will be functionally equivalent.
1233 1236 */
1234 1237 local uInt longest_match(s, cur_match)
1235 1238 deflate_state *s;
1236 1239 IPos cur_match; /* current match */
1237 1240 {
1238 1241 unsigned chain_length = s->max_chain_length;/* max hash chain length */
1239 1242 register Bytef *scan = s->window + s->strstart; /* current string */
1240 1243 register Bytef *match; /* matched string */
1241 1244 register int len; /* length of current match */
1242 1245 int best_len = (int)s->prev_length; /* best match length so far */
1243 1246 int nice_match = s->nice_match; /* stop if match long enough */
1244 1247 IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
1245 1248 s->strstart - (IPos)MAX_DIST(s) : NIL;
1246 1249 /* Stop when cur_match becomes <= limit. To simplify the code,
1247 1250 * we prevent matches with the string of window index 0.
1248 1251 */
1249 1252 Posf *prev = s->prev;
1250 1253 uInt wmask = s->w_mask;
1251 1254
1252 1255 #ifdef UNALIGNED_OK
1253 1256 /* Compare two bytes at a time. Note: this is not always beneficial.
1254 1257 * Try with and without -DUNALIGNED_OK to check.
1255 1258 */
1256 1259 register Bytef *strend = s->window + s->strstart + MAX_MATCH - 1;
1257 1260 register ush scan_start = *(ushf*)scan;
1258 1261 register ush scan_end = *(ushf*)(scan+best_len-1);
1259 1262 #else
1260 1263 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1261 1264 register Byte scan_end1 = scan[best_len-1];
1262 1265 register Byte scan_end = scan[best_len];
1263 1266 #endif
1264 1267
1265 1268 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1266 1269 * It is easy to get rid of this optimization if necessary.
1267 1270 */
1268 1271 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1269 1272
1270 1273 /* Do not waste too much time if we already have a good match: */
1271 1274 if (s->prev_length >= s->good_match) {
1272 1275 chain_length >>= 2;
1273 1276 }
1274 1277 /* Do not look for matches beyond the end of the input. This is necessary
1275 1278 * to make deflate deterministic.
1276 1279 */
1277 1280 if ((uInt)nice_match > s->lookahead) nice_match = (int)s->lookahead;
1278 1281
1279 1282 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1280 1283
1281 1284 do {
1282 1285 Assert(cur_match < s->strstart, "no future");
1283 1286 match = s->window + cur_match;
1284 1287
1285 1288 /* Skip to next match if the match length cannot increase
1286 1289 * or if the match length is less than 2. Note that the checks below
1287 1290 * for insufficient lookahead only occur occasionally for performance
1288 1291 * reasons. Therefore uninitialized memory will be accessed, and
1289 1292 * conditional jumps will be made that depend on those values.
1290 1293 * However the length of the match is limited to the lookahead, so
1291 1294 * the output of deflate is not affected by the uninitialized values.
1292 1295 */
1293 1296 #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
1294 1297 /* This code assumes sizeof(unsigned short) == 2. Do not use
1295 1298 * UNALIGNED_OK if your compiler uses a different size.
1296 1299 */
1297 1300 if (*(ushf*)(match+best_len-1) != scan_end ||
1298 1301 *(ushf*)match != scan_start) continue;
1299 1302
1300 1303 /* It is not necessary to compare scan[2] and match[2] since they are
1301 1304 * always equal when the other bytes match, given that the hash keys
1302 1305 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
1303 1306 * strstart+3, +5, ... up to strstart+257. We check for insufficient
1304 1307 * lookahead only every 4th comparison; the 128th check will be made
1305 1308 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
1306 1309 * necessary to put more guard bytes at the end of the window, or
1307 1310 * to check more often for insufficient lookahead.
1308 1311 */
1309 1312 Assert(scan[2] == match[2], "scan[2]?");
1310 1313 scan++, match++;
1311 1314 do {
1312 1315 } while (*(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1313 1316 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1314 1317 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1315 1318 *(ushf*)(scan+=2) == *(ushf*)(match+=2) &&
1316 1319 scan < strend);
1317 1320 /* The funny "do {}" generates better code on most compilers */
1318 1321
1319 1322 /* Here, scan <= window+strstart+257 */
1320 1323 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1321 1324 if (*scan == *match) scan++;
1322 1325
1323 1326 len = (MAX_MATCH - 1) - (int)(strend-scan);
1324 1327 scan = strend - (MAX_MATCH-1);
1325 1328
1326 1329 #else /* UNALIGNED_OK */
1327 1330
1328 1331 if (match[best_len] != scan_end ||
1329 1332 match[best_len-1] != scan_end1 ||
1330 1333 *match != *scan ||
1331 1334 *++match != scan[1]) continue;
1332 1335
1333 1336 /* The check at best_len-1 can be removed because it will be made
1334 1337 * again later. (This heuristic is not always a win.)
1335 1338 * It is not necessary to compare scan[2] and match[2] since they
1336 1339 * are always equal when the other bytes match, given that
1337 1340 * the hash keys are equal and that HASH_BITS >= 8.
1338 1341 */
1339 1342 scan += 2, match++;
1340 1343 Assert(*scan == *match, "match[2]?");
1341 1344
1342 1345 /* We check for insufficient lookahead only every 8th comparison;
1343 1346 * the 256th check will be made at strstart+258.
1344 1347 */
1345 1348 do {
1346 1349 } while (*++scan == *++match && *++scan == *++match &&
1347 1350 *++scan == *++match && *++scan == *++match &&
1348 1351 *++scan == *++match && *++scan == *++match &&
1349 1352 *++scan == *++match && *++scan == *++match &&
1350 1353 scan < strend);
1351 1354
1352 1355 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1353 1356
1354 1357 len = MAX_MATCH - (int)(strend - scan);
1355 1358 scan = strend - MAX_MATCH;
1356 1359
1357 1360 #endif /* UNALIGNED_OK */
1358 1361
1359 1362 if (len > best_len) {
1360 1363 s->match_start = cur_match;
1361 1364 best_len = len;
1362 1365 if (len >= nice_match) break;
1363 1366 #ifdef UNALIGNED_OK
1364 1367 scan_end = *(ushf*)(scan+best_len-1);
1365 1368 #else
1366 1369 scan_end1 = scan[best_len-1];
1367 1370 scan_end = scan[best_len];
1368 1371 #endif
1369 1372 }
1370 1373 } while ((cur_match = prev[cur_match & wmask]) > limit
1371 1374 && --chain_length != 0);
1372 1375
1373 1376 if ((uInt)best_len <= s->lookahead) return (uInt)best_len;
1374 1377 return s->lookahead;
1375 1378 }
1376 1379 #endif /* ASMV */
1377 1380
1378 1381 #else /* FASTEST */
1379 1382
1380 1383 /* ---------------------------------------------------------------------------
1381 1384 * Optimized version for FASTEST only
1382 1385 */
1383 1386 local uInt longest_match(s, cur_match)
1384 1387 deflate_state *s;
1385 1388 IPos cur_match; /* current match */
1386 1389 {
1387 1390 register Bytef *scan = s->window + s->strstart; /* current string */
1388 1391 register Bytef *match; /* matched string */
1389 1392 register int len; /* length of current match */
1390 1393 register Bytef *strend = s->window + s->strstart + MAX_MATCH;
1391 1394
1392 1395 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
1393 1396 * It is easy to get rid of this optimization if necessary.
1394 1397 */
1395 1398 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
1396 1399
1397 1400 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
1398 1401
1399 1402 Assert(cur_match < s->strstart, "no future");
1400 1403
1401 1404 match = s->window + cur_match;
1402 1405
1403 1406 /* Return failure if the match length is less than 2:
1404 1407 */
1405 1408 if (match[0] != scan[0] || match[1] != scan[1]) return MIN_MATCH-1;
1406 1409
1407 1410 /* The check at best_len-1 can be removed because it will be made
1408 1411 * again later. (This heuristic is not always a win.)
1409 1412 * It is not necessary to compare scan[2] and match[2] since they
1410 1413 * are always equal when the other bytes match, given that
1411 1414 * the hash keys are equal and that HASH_BITS >= 8.
1412 1415 */
1413 1416 scan += 2, match += 2;
1414 1417 Assert(*scan == *match, "match[2]?");
1415 1418
1416 1419 /* We check for insufficient lookahead only every 8th comparison;
1417 1420 * the 256th check will be made at strstart+258.
1418 1421 */
1419 1422 do {
1420 1423 } while (*++scan == *++match && *++scan == *++match &&
1421 1424 *++scan == *++match && *++scan == *++match &&
1422 1425 *++scan == *++match && *++scan == *++match &&
1423 1426 *++scan == *++match && *++scan == *++match &&
1424 1427 scan < strend);
1425 1428
1426 1429 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
1427 1430
1428 1431 len = MAX_MATCH - (int)(strend - scan);
1429 1432
1430 1433 if (len < MIN_MATCH) return MIN_MATCH - 1;
1431 1434
1432 1435 s->match_start = cur_match;
1433 1436 return (uInt)len <= s->lookahead ? (uInt)len : s->lookahead;
1434 1437 }
1435 1438
1436 1439 #endif /* FASTEST */
1437 1440
1438 1441 #ifdef ZLIB_DEBUG
1439 1442
1440 1443 #define EQUAL 0
1441 1444 /* result of memcmp for equal strings */
1442 1445
1443 1446 /* ===========================================================================
1444 1447 * Check that the match at match_start is indeed a match.
1445 1448 */
1446 1449 local void check_match(s, start, match, length)
1447 1450 deflate_state *s;
1448 1451 IPos start, match;
1449 1452 int length;
1450 1453 {
1451 1454 /* check that the match is indeed a match */
1452 1455 if (zmemcmp(s->window + match,
1453 1456 s->window + start, length) != EQUAL) {
1454 1457 fprintf(stderr, " start %u, match %u, length %d\n",
1455 1458 start, match, length);
1456 1459 do {
1457 1460 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
1458 1461 } while (--length != 0);
1459 1462 z_error("invalid match");
1460 1463 }
1461 1464 if (z_verbose > 1) {
1462 1465 fprintf(stderr,"\\[%d,%d]", start-match, length);
1463 1466 do { putc(s->window[start++], stderr); } while (--length != 0);
1464 1467 }
1465 1468 }
1466 1469 #else
1467 1470 # define check_match(s, start, match, length)
1468 1471 #endif /* ZLIB_DEBUG */
1469 1472
1470 1473 /* ===========================================================================
1471 1474 * Fill the window when the lookahead becomes insufficient.
1472 1475 * Updates strstart and lookahead.
1473 1476 *
1474 1477 * IN assertion: lookahead < MIN_LOOKAHEAD
1475 1478 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
1476 1479 * At least one byte has been read, or avail_in == 0; reads are
1477 1480 * performed for at least two bytes (required for the zip translate_eol
1478 1481 * option -- not supported here).
1479 1482 */
1480 1483 local void fill_window(s)
1481 1484 deflate_state *s;
1482 1485 {
1483 1486 unsigned n;
1484 1487 unsigned more; /* Amount of free space at the end of the window. */
1485 1488 uInt wsize = s->w_size;
1486 1489
1487 1490 Assert(s->lookahead < MIN_LOOKAHEAD, "already enough lookahead");
1488 1491
1489 1492 do {
1490 1493 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
1491 1494
1492 1495 /* Deal with !@#$% 64K limit: */
1493 1496 if (sizeof(int) <= 2) {
1494 1497 if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
1495 1498 more = wsize;
1496 1499
1497 1500 } else if (more == (unsigned)(-1)) {
1498 1501 /* Very unlikely, but possible on 16 bit machine if
1499 1502 * strstart == 0 && lookahead == 1 (input done a byte at time)
1500 1503 */
1501 1504 more--;
1502 1505 }
1503 1506 }
1504 1507
1505 1508 /* If the window is almost full and there is insufficient lookahead,
1506 1509 * move the upper half to the lower one to make room in the upper half.
1507 1510 */
1508 1511 if (s->strstart >= wsize+MAX_DIST(s)) {
1509 1512
1510 1513 zmemcpy(s->window, s->window+wsize, (unsigned)wsize - more);
1511 1514 s->match_start -= wsize;
1512 1515 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
1513 1516 s->block_start -= (long) wsize;
1514 1517 slide_hash(s);
1515 1518 more += wsize;
1516 1519 }
1517 1520 if (s->strm->avail_in == 0) break;
1518 1521
1519 1522 /* If there was no sliding:
1520 1523 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
1521 1524 * more == window_size - lookahead - strstart
1522 1525 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
1523 1526 * => more >= window_size - 2*WSIZE + 2
1524 1527 * In the BIG_MEM or MMAP case (not yet supported),
1525 1528 * window_size == input_size + MIN_LOOKAHEAD &&
1526 1529 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
1527 1530 * Otherwise, window_size == 2*WSIZE so more >= 2.
1528 1531 * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
1529 1532 */
1530 1533 Assert(more >= 2, "more < 2");
1531 1534
1532 1535 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
1533 1536 s->lookahead += n;
1534 1537
1535 1538 /* Initialize the hash value now that we have some input: */
1536 1539 if (s->lookahead + s->insert >= MIN_MATCH) {
1537 1540 uInt str = s->strstart - s->insert;
1538 1541 s->ins_h = s->window[str];
1539 1542 UPDATE_HASH(s, s->ins_h, s->window[str + 1]);
1540 1543 #if MIN_MATCH != 3
1541 1544 Call UPDATE_HASH() MIN_MATCH-3 more times
1542 1545 #endif
1543 1546 while (s->insert) {
1544 1547 UPDATE_HASH(s, s->ins_h, s->window[str + MIN_MATCH-1]);
1545 1548 #ifndef FASTEST
1546 1549 s->prev[str & s->w_mask] = s->head[s->ins_h];
1547 1550 #endif
1548 1551 s->head[s->ins_h] = (Pos)str;
1549 1552 str++;
1550 1553 s->insert--;
1551 1554 if (s->lookahead + s->insert < MIN_MATCH)
1552 1555 break;
1553 1556 }
1554 1557 }
1555 1558 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
1556 1559 * but this is not important since only literal bytes will be emitted.
1557 1560 */
1558 1561
1559 1562 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
1560 1563
1561 1564 /* If the WIN_INIT bytes after the end of the current data have never been
1562 1565 * written, then zero those bytes in order to avoid memory check reports of
1563 1566 * the use of uninitialized (or uninitialised as Julian writes) bytes by
1564 1567 * the longest match routines. Update the high water mark for the next
1565 1568 * time through here. WIN_INIT is set to MAX_MATCH since the longest match
1566 1569 * routines allow scanning to strstart + MAX_MATCH, ignoring lookahead.
1567 1570 */
1568 1571 if (s->high_water < s->window_size) {
1569 1572 ulg curr = s->strstart + (ulg)(s->lookahead);
1570 1573 ulg init;
1571 1574
1572 1575 if (s->high_water < curr) {
1573 1576 /* Previous high water mark below current data -- zero WIN_INIT
1574 1577 * bytes or up to end of window, whichever is less.
1575 1578 */
1576 1579 init = s->window_size - curr;
1577 1580 if (init > WIN_INIT)
1578 1581 init = WIN_INIT;
1579 1582 zmemzero(s->window + curr, (unsigned)init);
1580 1583 s->high_water = curr + init;
1581 1584 }
1582 1585 else if (s->high_water < (ulg)curr + WIN_INIT) {
1583 1586 /* High water mark at or above current data, but below current data
1584 1587 * plus WIN_INIT -- zero out to current data plus WIN_INIT, or up
1585 1588 * to end of window, whichever is less.
1586 1589 */
1587 1590 init = (ulg)curr + WIN_INIT - s->high_water;
1588 1591 if (init > s->window_size - s->high_water)
1589 1592 init = s->window_size - s->high_water;
1590 1593 zmemzero(s->window + s->high_water, (unsigned)init);
1591 1594 s->high_water += init;
1592 1595 }
1593 1596 }
1594 1597
1595 1598 Assert((ulg)s->strstart <= s->window_size - MIN_LOOKAHEAD,
1596 1599 "not enough room for search");
1597 1600 }
1598 1601
1599 1602 /* ===========================================================================
1600 1603 * Flush the current block, with given end-of-file flag.
1601 1604 * IN assertion: strstart is set to the end of the current match.
1602 1605 */
1603 1606 #define FLUSH_BLOCK_ONLY(s, last) { \
1604 1607 _tr_flush_block(s, (s->block_start >= 0L ? \
1605 1608 (charf *)&s->window[(unsigned)s->block_start] : \
1606 1609 (charf *)Z_NULL), \
1607 1610 (ulg)((long)s->strstart - s->block_start), \
1608 1611 (last)); \
1609 1612 s->block_start = s->strstart; \
1610 1613 flush_pending(s->strm); \
1611 1614 Tracev((stderr,"[FLUSH]")); \
1612 1615 }
1613 1616
1614 1617 /* Same but force premature exit if necessary. */
1615 1618 #define FLUSH_BLOCK(s, last) { \
1616 1619 FLUSH_BLOCK_ONLY(s, last); \
1617 1620 if (s->strm->avail_out == 0) return (last) ? finish_started : need_more; \
1618 1621 }
1619 1622
1620 1623 /* Maximum stored block length in deflate format (not including header). */
1621 1624 #define MAX_STORED 65535
1622 1625
1623 1626 /* Minimum of a and b. */
1624 1627 #define MIN(a, b) ((a) > (b) ? (b) : (a))
1625 1628
1626 1629 /* ===========================================================================
1627 1630 * Copy without compression as much as possible from the input stream, return
1628 1631 * the current block state.
1629 1632 *
1630 1633 * In case deflateParams() is used to later switch to a non-zero compression
1631 1634 * level, s->matches (otherwise unused when storing) keeps track of the number
1632 1635 * of hash table slides to perform. If s->matches is 1, then one hash table
1633 1636 * slide will be done when switching. If s->matches is 2, the maximum value
1634 1637 * allowed here, then the hash table will be cleared, since two or more slides
1635 1638 * is the same as a clear.
1636 1639 *
1637 1640 * deflate_stored() is written to minimize the number of times an input byte is
1638 1641 * copied. It is most efficient with large input and output buffers, which
1639 1642 * maximizes the opportunites to have a single copy from next_in to next_out.
1640 1643 */
1641 1644 local block_state deflate_stored(s, flush)
1642 1645 deflate_state *s;
1643 1646 int flush;
1644 1647 {
1645 1648 /* Smallest worthy block size when not flushing or finishing. By default
1646 1649 * this is 32K. This can be as small as 507 bytes for memLevel == 1. For
1647 1650 * large input and output buffers, the stored block size will be larger.
1648 1651 */
1649 1652 unsigned min_block = MIN(s->pending_buf_size - 5, s->w_size);
1650 1653
1651 1654 /* Copy as many min_block or larger stored blocks directly to next_out as
1652 1655 * possible. If flushing, copy the remaining available input to next_out as
1653 1656 * stored blocks, if there is enough space.
1654 1657 */
1655 1658 unsigned len, left, have, last = 0;
1656 1659 unsigned used = s->strm->avail_in;
1657 1660 do {
1658 1661 /* Set len to the maximum size block that we can copy directly with the
1659 1662 * available input data and output space. Set left to how much of that
1660 1663 * would be copied from what's left in the window.
1661 1664 */
1662 1665 len = MAX_STORED; /* maximum deflate stored block length */
1663 1666 have = (s->bi_valid + 42) >> 3; /* number of header bytes */
1664 1667 if (s->strm->avail_out < have) /* need room for header */
1665 1668 break;
1666 1669 /* maximum stored block length that will fit in avail_out: */
1667 1670 have = s->strm->avail_out - have;
1668 1671 left = s->strstart - s->block_start; /* bytes left in window */
1669 1672 if (len > (ulg)left + s->strm->avail_in)
1670 1673 len = left + s->strm->avail_in; /* limit len to the input */
1671 1674 if (len > have)
1672 1675 len = have; /* limit len to the output */
1673 1676
1674 1677 /* If the stored block would be less than min_block in length, or if
1675 1678 * unable to copy all of the available input when flushing, then try
1676 1679 * copying to the window and the pending buffer instead. Also don't
1677 1680 * write an empty block when flushing -- deflate() does that.
1678 1681 */
1679 1682 if (len < min_block && ((len == 0 && flush != Z_FINISH) ||
1680 1683 flush == Z_NO_FLUSH ||
1681 1684 len != left + s->strm->avail_in))
1682 1685 break;
1683 1686
1684 1687 /* Make a dummy stored block in pending to get the header bytes,
1685 1688 * including any pending bits. This also updates the debugging counts.
1686 1689 */
1687 1690 last = flush == Z_FINISH && len == left + s->strm->avail_in ? 1 : 0;
1688 1691 _tr_stored_block(s, (char *)0, 0L, last);
1689 1692
1690 1693 /* Replace the lengths in the dummy stored block with len. */
1691 1694 s->pending_buf[s->pending - 4] = len;
1692 1695 s->pending_buf[s->pending - 3] = len >> 8;
1693 1696 s->pending_buf[s->pending - 2] = ~len;
1694 1697 s->pending_buf[s->pending - 1] = ~len >> 8;
1695 1698
1696 1699 /* Write the stored block header bytes. */
1697 1700 flush_pending(s->strm);
1698 1701
1699 1702 #ifdef ZLIB_DEBUG
1700 1703 /* Update debugging counts for the data about to be copied. */
1701 1704 s->compressed_len += len << 3;
1702 1705 s->bits_sent += len << 3;
1703 1706 #endif
1704 1707
1705 1708 /* Copy uncompressed bytes from the window to next_out. */
1706 1709 if (left) {
1707 1710 if (left > len)
1708 1711 left = len;
1709 1712 zmemcpy(s->strm->next_out, s->window + s->block_start, left);
1710 1713 s->strm->next_out += left;
1711 1714 s->strm->avail_out -= left;
1712 1715 s->strm->total_out += left;
1713 1716 s->block_start += left;
1714 1717 len -= left;
1715 1718 }
1716 1719
1717 1720 /* Copy uncompressed bytes directly from next_in to next_out, updating
1718 1721 * the check value.
1719 1722 */
1720 1723 if (len) {
1721 1724 read_buf(s->strm, s->strm->next_out, len);
1722 1725 s->strm->next_out += len;
1723 1726 s->strm->avail_out -= len;
1724 1727 s->strm->total_out += len;
1725 1728 }
1726 1729 } while (last == 0);
1727 1730
1728 1731 /* Update the sliding window with the last s->w_size bytes of the copied
1729 1732 * data, or append all of the copied data to the existing window if less
1730 1733 * than s->w_size bytes were copied. Also update the number of bytes to
1731 1734 * insert in the hash tables, in the event that deflateParams() switches to
1732 1735 * a non-zero compression level.
1733 1736 */
1734 1737 used -= s->strm->avail_in; /* number of input bytes directly copied */
1735 1738 if (used) {
1736 1739 /* If any input was used, then no unused input remains in the window,
1737 1740 * therefore s->block_start == s->strstart.
1738 1741 */
1739 1742 if (used >= s->w_size) { /* supplant the previous history */
1740 1743 s->matches = 2; /* clear hash */
1741 1744 zmemcpy(s->window, s->strm->next_in - s->w_size, s->w_size);
1742 1745 s->strstart = s->w_size;
1743 1746 }
1744 1747 else {
1745 1748 if (s->window_size - s->strstart <= used) {
1746 1749 /* Slide the window down. */
1747 1750 s->strstart -= s->w_size;
1748 1751 zmemcpy(s->window, s->window + s->w_size, s->strstart);
1749 1752 if (s->matches < 2)
1750 1753 s->matches++; /* add a pending slide_hash() */
1751 1754 }
1752 1755 zmemcpy(s->window + s->strstart, s->strm->next_in - used, used);
1753 1756 s->strstart += used;
1754 1757 }
1755 1758 s->block_start = s->strstart;
1756 1759 s->insert += MIN(used, s->w_size - s->insert);
1757 1760 }
1758 1761 if (s->high_water < s->strstart)
1759 1762 s->high_water = s->strstart;
1760 1763
1761 1764 /* If the last block was written to next_out, then done. */
1762 1765 if (last)
1763 1766 return finish_done;
1764 1767
1765 1768 /* If flushing and all input has been consumed, then done. */
1766 1769 if (flush != Z_NO_FLUSH && flush != Z_FINISH &&
1767 1770 s->strm->avail_in == 0 && (long)s->strstart == s->block_start)
1768 1771 return block_done;
1769 1772
1770 1773 /* Fill the window with any remaining input. */
1771 1774 have = s->window_size - s->strstart - 1;
1772 1775 if (s->strm->avail_in > have && s->block_start >= (long)s->w_size) {
1773 1776 /* Slide the window down. */
1774 1777 s->block_start -= s->w_size;
1775 1778 s->strstart -= s->w_size;
1776 1779 zmemcpy(s->window, s->window + s->w_size, s->strstart);
1777 1780 if (s->matches < 2)
1778 1781 s->matches++; /* add a pending slide_hash() */
1779 1782 have += s->w_size; /* more space now */
1780 1783 }
1781 1784 if (have > s->strm->avail_in)
1782 1785 have = s->strm->avail_in;
1783 1786 if (have) {
1784 1787 read_buf(s->strm, s->window + s->strstart, have);
1785 1788 s->strstart += have;
1786 1789 }
1787 1790 if (s->high_water < s->strstart)
1788 1791 s->high_water = s->strstart;
1789 1792
1790 1793 /* There was not enough avail_out to write a complete worthy or flushed
1791 1794 * stored block to next_out. Write a stored block to pending instead, if we
1792 1795 * have enough input for a worthy block, or if flushing and there is enough
1793 1796 * room for the remaining input as a stored block in the pending buffer.
1794 1797 */
1795 1798 have = (s->bi_valid + 42) >> 3; /* number of header bytes */
1796 1799 /* maximum stored block length that will fit in pending: */
1797 1800 have = MIN(s->pending_buf_size - have, MAX_STORED);
1798 1801 min_block = MIN(have, s->w_size);
1799 1802 left = s->strstart - s->block_start;
1800 1803 if (left >= min_block ||
1801 1804 ((left || flush == Z_FINISH) && flush != Z_NO_FLUSH &&
1802 1805 s->strm->avail_in == 0 && left <= have)) {
1803 1806 len = MIN(left, have);
1804 1807 last = flush == Z_FINISH && s->strm->avail_in == 0 &&
1805 1808 len == left ? 1 : 0;
1806 1809 _tr_stored_block(s, (charf *)s->window + s->block_start, len, last);
1807 1810 s->block_start += len;
1808 1811 flush_pending(s->strm);
1809 1812 }
1810 1813
1811 1814 /* We've done all we can with the available input and output. */
1812 1815 return last ? finish_started : need_more;
1813 1816 }
1814 1817
1815 1818 /* ===========================================================================
1816 1819 * Compress as much as possible from the input stream, return the current
1817 1820 * block state.
1818 1821 * This function does not perform lazy evaluation of matches and inserts
1819 1822 * new strings in the dictionary only for unmatched strings or for short
1820 1823 * matches. It is used only for the fast compression options.
1821 1824 */
1822 1825 local block_state deflate_fast(s, flush)
1823 1826 deflate_state *s;
1824 1827 int flush;
1825 1828 {
1826 1829 IPos hash_head; /* head of the hash chain */
1827 1830 int bflush; /* set if current block must be flushed */
1828 1831
1829 1832 for (;;) {
1830 1833 /* Make sure that we always have enough lookahead, except
1831 1834 * at the end of the input file. We need MAX_MATCH bytes
1832 1835 * for the next match, plus MIN_MATCH bytes to insert the
1833 1836 * string following the next match.
1834 1837 */
1835 1838 if (s->lookahead < MIN_LOOKAHEAD) {
1836 1839 fill_window(s);
1837 1840 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1838 1841 return need_more;
1839 1842 }
1840 1843 if (s->lookahead == 0) break; /* flush the current block */
1841 1844 }
1842 1845
1843 1846 /* Insert the string window[strstart .. strstart+2] in the
1844 1847 * dictionary, and set hash_head to the head of the hash chain:
1845 1848 */
1846 1849 hash_head = NIL;
1847 1850 if (s->lookahead >= MIN_MATCH) {
1848 1851 INSERT_STRING(s, s->strstart, hash_head);
1849 1852 }
1850 1853
1851 1854 /* Find the longest match, discarding those <= prev_length.
1852 1855 * At this point we have always match_length < MIN_MATCH
1853 1856 */
1854 1857 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
1855 1858 /* To simplify the code, we prevent matches with the string
1856 1859 * of window index 0 (in particular we have to avoid a match
1857 1860 * of the string with itself at the start of the input file).
1858 1861 */
1859 1862 s->match_length = longest_match (s, hash_head);
1860 1863 /* longest_match() sets match_start */
1861 1864 }
1862 1865 if (s->match_length >= MIN_MATCH) {
1863 1866 check_match(s, s->strstart, s->match_start, s->match_length);
1864 1867
1865 1868 _tr_tally_dist(s, s->strstart - s->match_start,
1866 1869 s->match_length - MIN_MATCH, bflush);
1867 1870
1868 1871 s->lookahead -= s->match_length;
1869 1872
1870 1873 /* Insert new strings in the hash table only if the match length
1871 1874 * is not too large. This saves time but degrades compression.
1872 1875 */
1873 1876 #ifndef FASTEST
1874 1877 if (s->match_length <= s->max_insert_length &&
1875 1878 s->lookahead >= MIN_MATCH) {
1876 1879 s->match_length--; /* string at strstart already in table */
1877 1880 do {
1878 1881 s->strstart++;
1879 1882 INSERT_STRING(s, s->strstart, hash_head);
1880 1883 /* strstart never exceeds WSIZE-MAX_MATCH, so there are
1881 1884 * always MIN_MATCH bytes ahead.
1882 1885 */
1883 1886 } while (--s->match_length != 0);
1884 1887 s->strstart++;
1885 1888 } else
1886 1889 #endif
1887 1890 {
1888 1891 s->strstart += s->match_length;
1889 1892 s->match_length = 0;
1890 1893 s->ins_h = s->window[s->strstart];
1891 1894 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
1892 1895 #if MIN_MATCH != 3
1893 1896 Call UPDATE_HASH() MIN_MATCH-3 more times
1894 1897 #endif
1895 1898 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
1896 1899 * matter since it will be recomputed at next deflate call.
1897 1900 */
1898 1901 }
1899 1902 } else {
1900 1903 /* No match, output a literal byte */
1901 1904 Tracevv((stderr,"%c", s->window[s->strstart]));
1902 1905 _tr_tally_lit (s, s->window[s->strstart], bflush);
1903 1906 s->lookahead--;
1904 1907 s->strstart++;
1905 1908 }
1906 1909 if (bflush) FLUSH_BLOCK(s, 0);
1907 1910 }
1908 1911 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
1909 1912 if (flush == Z_FINISH) {
1910 1913 FLUSH_BLOCK(s, 1);
1911 1914 return finish_done;
1912 1915 }
1913 1916 if (s->last_lit)
1914 1917 FLUSH_BLOCK(s, 0);
1915 1918 return block_done;
1916 1919 }
1917 1920
1918 1921 #ifndef FASTEST
1919 1922 /* ===========================================================================
1920 1923 * Same as above, but achieves better compression. We use a lazy
1921 1924 * evaluation for matches: a match is finally adopted only if there is
1922 1925 * no better match at the next window position.
1923 1926 */
1924 1927 local block_state deflate_slow(s, flush)
1925 1928 deflate_state *s;
1926 1929 int flush;
1927 1930 {
1928 1931 IPos hash_head; /* head of hash chain */
1929 1932 int bflush; /* set if current block must be flushed */
1930 1933
1931 1934 /* Process the input block. */
1932 1935 for (;;) {
1933 1936 /* Make sure that we always have enough lookahead, except
1934 1937 * at the end of the input file. We need MAX_MATCH bytes
1935 1938 * for the next match, plus MIN_MATCH bytes to insert the
1936 1939 * string following the next match.
1937 1940 */
1938 1941 if (s->lookahead < MIN_LOOKAHEAD) {
1939 1942 fill_window(s);
1940 1943 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1941 1944 return need_more;
1942 1945 }
1943 1946 if (s->lookahead == 0) break; /* flush the current block */
1944 1947 }
1945 1948
1946 1949 /* Insert the string window[strstart .. strstart+2] in the
1947 1950 * dictionary, and set hash_head to the head of the hash chain:
1948 1951 */
1949 1952 hash_head = NIL;
1950 1953 if (s->lookahead >= MIN_MATCH) {
1951 1954 INSERT_STRING(s, s->strstart, hash_head);
1952 1955 }
1953 1956
1954 1957 /* Find the longest match, discarding those <= prev_length.
1955 1958 */
1956 1959 s->prev_length = s->match_length, s->prev_match = s->match_start;
1957 1960 s->match_length = MIN_MATCH-1;
1958 1961
1959 1962 if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1960 1963 s->strstart - hash_head <= MAX_DIST(s)) {
1961 1964 /* To simplify the code, we prevent matches with the string
1962 1965 * of window index 0 (in particular we have to avoid a match
1963 1966 * of the string with itself at the start of the input file).
1964 1967 */
1965 1968 s->match_length = longest_match (s, hash_head);
1966 1969 /* longest_match() sets match_start */
1967 1970
1968 1971 if (s->match_length <= 5 && (s->strategy == Z_FILTERED
1969 1972 #if TOO_FAR <= 32767
1970 1973 || (s->match_length == MIN_MATCH &&
1971 1974 s->strstart - s->match_start > TOO_FAR)
1972 1975 #endif
1973 1976 )) {
1974 1977
1975 1978 /* If prev_match is also MIN_MATCH, match_start is garbage
1976 1979 * but we will ignore the current match anyway.
1977 1980 */
1978 1981 s->match_length = MIN_MATCH-1;
1979 1982 }
1980 1983 }
1981 1984 /* If there was a match at the previous step and the current
1982 1985 * match is not better, output the previous match:
1983 1986 */
1984 1987 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1985 1988 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1986 1989 /* Do not insert strings in hash table beyond this. */
1987 1990
1988 1991 check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1989 1992
1990 1993 _tr_tally_dist(s, s->strstart -1 - s->prev_match,
1991 1994 s->prev_length - MIN_MATCH, bflush);
1992 1995
1993 1996 /* Insert in hash table all strings up to the end of the match.
1994 1997 * strstart-1 and strstart are already inserted. If there is not
1995 1998 * enough lookahead, the last two strings are not inserted in
1996 1999 * the hash table.
1997 2000 */
1998 2001 s->lookahead -= s->prev_length-1;
1999 2002 s->prev_length -= 2;
2000 2003 do {
2001 2004 if (++s->strstart <= max_insert) {
2002 2005 INSERT_STRING(s, s->strstart, hash_head);
2003 2006 }
2004 2007 } while (--s->prev_length != 0);
2005 2008 s->match_available = 0;
2006 2009 s->match_length = MIN_MATCH-1;
2007 2010 s->strstart++;
2008 2011
2009 2012 if (bflush) FLUSH_BLOCK(s, 0);
2010 2013
2011 2014 } else if (s->match_available) {
2012 2015 /* If there was no match at the previous position, output a
2013 2016 * single literal. If there was a match but the current match
2014 2017 * is longer, truncate the previous match to a single literal.
2015 2018 */
2016 2019 Tracevv((stderr,"%c", s->window[s->strstart-1]));
2017 2020 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2018 2021 if (bflush) {
2019 2022 FLUSH_BLOCK_ONLY(s, 0);
2020 2023 }
2021 2024 s->strstart++;
2022 2025 s->lookahead--;
2023 2026 if (s->strm->avail_out == 0) return need_more;
2024 2027 } else {
2025 2028 /* There is no previous match to compare with, wait for
2026 2029 * the next step to decide.
2027 2030 */
2028 2031 s->match_available = 1;
2029 2032 s->strstart++;
2030 2033 s->lookahead--;
2031 2034 }
2032 2035 }
2033 2036 Assert (flush != Z_NO_FLUSH, "no flush?");
2034 2037 if (s->match_available) {
2035 2038 Tracevv((stderr,"%c", s->window[s->strstart-1]));
2036 2039 _tr_tally_lit(s, s->window[s->strstart-1], bflush);
2037 2040 s->match_available = 0;
2038 2041 }
2039 2042 s->insert = s->strstart < MIN_MATCH-1 ? s->strstart : MIN_MATCH-1;
2040 2043 if (flush == Z_FINISH) {
2041 2044 FLUSH_BLOCK(s, 1);
2042 2045 return finish_done;
2043 2046 }
2044 2047 if (s->last_lit)
2045 2048 FLUSH_BLOCK(s, 0);
2046 2049 return block_done;
2047 2050 }
2048 2051 #endif /* FASTEST */
2049 2052
2050 2053 /* ===========================================================================
2051 2054 * For Z_RLE, simply look for runs of bytes, generate matches only of distance
2052 2055 * one. Do not maintain a hash table. (It will be regenerated if this run of
2053 2056 * deflate switches away from Z_RLE.)
2054 2057 */
2055 2058 local block_state deflate_rle(s, flush)
2056 2059 deflate_state *s;
2057 2060 int flush;
2058 2061 {
2059 2062 int bflush; /* set if current block must be flushed */
2060 2063 uInt prev; /* byte at distance one to match */
2061 2064 Bytef *scan, *strend; /* scan goes up to strend for length of run */
2062 2065
2063 2066 for (;;) {
2064 2067 /* Make sure that we always have enough lookahead, except
2065 2068 * at the end of the input file. We need MAX_MATCH bytes
2066 2069 * for the longest run, plus one for the unrolled loop.
2067 2070 */
2068 2071 if (s->lookahead <= MAX_MATCH) {
2069 2072 fill_window(s);
2070 2073 if (s->lookahead <= MAX_MATCH && flush == Z_NO_FLUSH) {
2071 2074 return need_more;
2072 2075 }
2073 2076 if (s->lookahead == 0) break; /* flush the current block */
2074 2077 }
2075 2078
2076 2079 /* See how many times the previous byte repeats */
2077 2080 s->match_length = 0;
2078 2081 if (s->lookahead >= MIN_MATCH && s->strstart > 0) {
2079 2082 scan = s->window + s->strstart - 1;
2080 2083 prev = *scan;
2081 2084 if (prev == *++scan && prev == *++scan && prev == *++scan) {
2082 2085 strend = s->window + s->strstart + MAX_MATCH;
2083 2086 do {
2084 2087 } while (prev == *++scan && prev == *++scan &&
2085 2088 prev == *++scan && prev == *++scan &&
2086 2089 prev == *++scan && prev == *++scan &&
2087 2090 prev == *++scan && prev == *++scan &&
2088 2091 scan < strend);
2089 2092 s->match_length = MAX_MATCH - (uInt)(strend - scan);
2090 2093 if (s->match_length > s->lookahead)
2091 2094 s->match_length = s->lookahead;
2092 2095 }
2093 2096 Assert(scan <= s->window+(uInt)(s->window_size-1), "wild scan");
2094 2097 }
2095 2098
2096 2099 /* Emit match if have run of MIN_MATCH or longer, else emit literal */
2097 2100 if (s->match_length >= MIN_MATCH) {
2098 2101 check_match(s, s->strstart, s->strstart - 1, s->match_length);
2099 2102
2100 2103 _tr_tally_dist(s, 1, s->match_length - MIN_MATCH, bflush);
2101 2104
2102 2105 s->lookahead -= s->match_length;
2103 2106 s->strstart += s->match_length;
2104 2107 s->match_length = 0;
2105 2108 } else {
2106 2109 /* No match, output a literal byte */
2107 2110 Tracevv((stderr,"%c", s->window[s->strstart]));
2108 2111 _tr_tally_lit (s, s->window[s->strstart], bflush);
2109 2112 s->lookahead--;
2110 2113 s->strstart++;
2111 2114 }
2112 2115 if (bflush) FLUSH_BLOCK(s, 0);
2113 2116 }
2114 2117 s->insert = 0;
2115 2118 if (flush == Z_FINISH) {
2116 2119 FLUSH_BLOCK(s, 1);
2117 2120 return finish_done;
2118 2121 }
2119 2122 if (s->last_lit)
2120 2123 FLUSH_BLOCK(s, 0);
2121 2124 return block_done;
2122 2125 }
2123 2126
2124 2127 /* ===========================================================================
2125 2128 * For Z_HUFFMAN_ONLY, do not look for matches. Do not maintain a hash table.
2126 2129 * (It will be regenerated if this run of deflate switches away from Huffman.)
2127 2130 */
2128 2131 local block_state deflate_huff(s, flush)
2129 2132 deflate_state *s;
2130 2133 int flush;
2131 2134 {
2132 2135 int bflush; /* set if current block must be flushed */
2133 2136
2134 2137 for (;;) {
2135 2138 /* Make sure that we have a literal to write. */
2136 2139 if (s->lookahead == 0) {
2137 2140 fill_window(s);
2138 2141 if (s->lookahead == 0) {
2139 2142 if (flush == Z_NO_FLUSH)
2140 2143 return need_more;
2141 2144 break; /* flush the current block */
2142 2145 }
2143 2146 }
2144 2147
2145 2148 /* Output a literal byte */
2146 2149 s->match_length = 0;
2147 2150 Tracevv((stderr,"%c", s->window[s->strstart]));
2148 2151 _tr_tally_lit (s, s->window[s->strstart], bflush);
2149 2152 s->lookahead--;
2150 2153 s->strstart++;
2151 2154 if (bflush) FLUSH_BLOCK(s, 0);
2152 2155 }
2153 2156 s->insert = 0;
2154 2157 if (flush == Z_FINISH) {
2155 2158 FLUSH_BLOCK(s, 1);
2156 2159 return finish_done;
2157 2160 }
2158 2161 if (s->last_lit)
2159 2162 FLUSH_BLOCK(s, 0);
2160 2163 return block_done;
2161 2164 }
|
↓ open down ↓ |
1959 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX