1/* +++ deflate.c */ 2/* deflate.c -- compress data using the deflation algorithm 3 * Copyright (C) 1995-1996 Jean-loup Gailly. 4 * For conditions of distribution and use, see copyright notice in zlib.h 5 */ 6 7/* 8 * ALGORITHM 9 * 10 * The "deflation" process depends on being able to identify portions 11 * of the input text which are identical to earlier input (within a 12 * sliding window trailing behind the input currently being processed). 13 * 14 * The most straightforward technique turns out to be the fastest for 15 * most input files: try all possible matches and select the longest. 16 * The key feature of this algorithm is that insertions into the string 17 * dictionary are very simple and thus fast, and deletions are avoided 18 * completely. Insertions are performed at each input character, whereas 19 * string matches are performed only when the previous match ends. So it 20 * is preferable to spend more time in matches to allow very fast string 21 * insertions and avoid deletions. The matching algorithm for small 22 * strings is inspired from that of Rabin & Karp. A brute force approach 23 * is used to find longer strings when a small match has been found. 24 * A similar algorithm is used in comic (by Jan-Mark Wams) and freeze 25 * (by Leonid Broukhis). 26 * A previous version of this file used a more sophisticated algorithm 27 * (by Fiala and Greene) which is guaranteed to run in linear amortized 28 * time, but has a larger average cost, uses more memory and is patented. 29 * However the F&G algorithm may be faster for some highly redundant 30 * files if the parameter max_chain_length (described below) is too large. 31 * 32 * ACKNOWLEDGEMENTS 33 * 34 * The idea of lazy evaluation of matches is due to Jan-Mark Wams, and 35 * I found it in 'freeze' written by Leonid Broukhis. 36 * Thanks to many people for bug reports and testing. 37 * 38 * REFERENCES 39 * 40 * Deutsch, L.P.,"DEFLATE Compressed Data Format Specification". 41 * Available in ftp://ds.internic.net/rfc/rfc1951.txt 42 * 43 * A description of the Rabin and Karp algorithm is given in the book 44 * "Algorithms" by R. Sedgewick, Addison-Wesley, p252. 45 * 46 * Fiala,E.R., and Greene,D.H. 47 * Data Compression with Finite Windows, Comm.ACM, 32,4 (1989) 490-595 48 * 49 */ 50 51#include <linux/module.h> 52#include <linux/zutil.h> 53#include "defutil.h" 54 55 56/* =========================================================================== 57 * Function prototypes. 58 */ 59typedef enum { 60 need_more, /* block not completed, need more input or more output */ 61 block_done, /* block flush performed */ 62 finish_started, /* finish started, need only more output at next deflate */ 63 finish_done /* finish done, accept no more input or output */ 64} block_state; 65 66typedef block_state (*compress_func) (deflate_state *s, int flush); 67/* Compression function. Returns the block state after the call. */ 68 69static void fill_window (deflate_state *s); 70static block_state deflate_stored (deflate_state *s, int flush); 71static block_state deflate_fast (deflate_state *s, int flush); 72static block_state deflate_slow (deflate_state *s, int flush); 73static void lm_init (deflate_state *s); 74static void putShortMSB (deflate_state *s, uInt b); 75static void flush_pending (z_streamp strm); 76static int read_buf (z_streamp strm, Byte *buf, unsigned size); 77static uInt longest_match (deflate_state *s, IPos cur_match); 78 79#ifdef DEBUG_ZLIB 80static void check_match (deflate_state *s, IPos start, IPos match, 81 int length); 82#endif 83 84/* =========================================================================== 85 * Local data 86 */ 87 88#define NIL 0 89/* Tail of hash chains */ 90 91#ifndef TOO_FAR 92# define TOO_FAR 4096 93#endif 94/* Matches of length 3 are discarded if their distance exceeds TOO_FAR */ 95 96#define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1) 97/* Minimum amount of lookahead, except at the end of the input file. 98 * See deflate.c for comments about the MIN_MATCH+1. 99 */ 100 101/* Values for max_lazy_match, good_match and max_chain_length, depending on 102 * the desired pack level (0..9). The values given below have been tuned to 103 * exclude worst case performance for pathological files. Better values may be 104 * found for specific files. 105 */ 106typedef struct config_s { 107 ush good_length; /* reduce lazy search above this match length */ 108 ush max_lazy; /* do not perform lazy search above this match length */ 109 ush nice_length; /* quit search above this match length */ 110 ush max_chain; 111 compress_func func; 112} config; 113 114static const config configuration_table[10] = { 115/* good lazy nice chain */ 116/* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */ 117/* 1 */ {4, 4, 8, 4, deflate_fast}, /* maximum speed, no lazy matches */ 118/* 2 */ {4, 5, 16, 8, deflate_fast}, 119/* 3 */ {4, 6, 32, 32, deflate_fast}, 120 121/* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */ 122/* 5 */ {8, 16, 32, 32, deflate_slow}, 123/* 6 */ {8, 16, 128, 128, deflate_slow}, 124/* 7 */ {8, 32, 128, 256, deflate_slow}, 125/* 8 */ {32, 128, 258, 1024, deflate_slow}, 126/* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* maximum compression */ 127 128/* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4 129 * For deflate_fast() (levels <= 3) good is ignored and lazy has a different 130 * meaning. 131 */ 132 133#define EQUAL 0 134/* result of memcmp for equal strings */ 135 136/* =========================================================================== 137 * Update a hash value with the given input byte 138 * IN assertion: all calls to to UPDATE_HASH are made with consecutive 139 * input characters, so that a running hash key can be computed from the 140 * previous key instead of complete recalculation each time. 141 */ 142#define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask) 143 144 145/* =========================================================================== 146 * Insert string str in the dictionary and set match_head to the previous head 147 * of the hash chain (the most recent string with same hash key). Return 148 * the previous length of the hash chain. 149 * IN assertion: all calls to to INSERT_STRING are made with consecutive 150 * input characters and the first MIN_MATCH bytes of str are valid 151 * (except for the last MIN_MATCH-1 bytes of the input file). 152 */ 153#define INSERT_STRING(s, str, match_head) \ 154 (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \ 155 s->prev[(str) & s->w_mask] = match_head = s->head[s->ins_h], \ 156 s->head[s->ins_h] = (Pos)(str)) 157 158/* =========================================================================== 159 * Initialize the hash table (avoiding 64K overflow for 16 bit systems). 160 * prev[] will be initialized on the fly. 161 */ 162#define CLEAR_HASH(s) \ 163 s->head[s->hash_size-1] = NIL; \ 164 memset((char *)s->head, 0, (unsigned)(s->hash_size-1)*sizeof(*s->head)); 165 166/* ========================================================================= */ 167int zlib_deflateInit2( 168 z_streamp strm, 169 int level, 170 int method, 171 int windowBits, 172 int memLevel, 173 int strategy 174) 175{ 176 deflate_state *s; 177 int noheader = 0; 178 deflate_workspace *mem; 179 180 ush *overlay; 181 /* We overlay pending_buf and d_buf+l_buf. This works since the average 182 * output size for (length,distance) codes is <= 24 bits. 183 */ 184 185 if (strm == NULL) return Z_STREAM_ERROR; 186 187 strm->msg = NULL; 188 189 if (level == Z_DEFAULT_COMPRESSION) level = 6; 190 191 mem = (deflate_workspace *) strm->workspace; 192 193 if (windowBits < 0) { /* undocumented feature: suppress zlib header */ 194 noheader = 1; 195 windowBits = -windowBits; 196 } 197 if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED || 198 windowBits < 9 || windowBits > 15 || level < 0 || level > 9 || 199 strategy < 0 || strategy > Z_HUFFMAN_ONLY) { 200 return Z_STREAM_ERROR; 201 } 202 s = (deflate_state *) &(mem->deflate_memory); 203 strm->state = (struct internal_state *)s; 204 s->strm = strm; 205 206 s->noheader = noheader; 207 s->w_bits = windowBits; 208 s->w_size = 1 << s->w_bits; 209 s->w_mask = s->w_size - 1; 210 211 s->hash_bits = memLevel + 7; 212 s->hash_size = 1 << s->hash_bits; 213 s->hash_mask = s->hash_size - 1; 214 s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH); 215 216 s->window = (Byte *) mem->window_memory; 217 s->prev = (Pos *) mem->prev_memory; 218 s->head = (Pos *) mem->head_memory; 219 220 s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */ 221 222 overlay = (ush *) mem->overlay_memory; 223 s->pending_buf = (uch *) overlay; 224 s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L); 225 226 s->d_buf = overlay + s->lit_bufsize/sizeof(ush); 227 s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize; 228 229 s->level = level; 230 s->strategy = strategy; 231 s->method = (Byte)method; 232 233 return zlib_deflateReset(strm); 234} 235 236/* ========================================================================= */ 237 238/* ========================================================================= */ 239int zlib_deflateReset( 240 z_streamp strm 241) 242{ 243 deflate_state *s; 244 245 if (strm == NULL || strm->state == NULL) 246 return Z_STREAM_ERROR; 247 248 strm->total_in = strm->total_out = 0; 249 strm->msg = NULL; 250 strm->data_type = Z_UNKNOWN; 251 252 s = (deflate_state *)strm->state; 253 s->pending = 0; 254 s->pending_out = s->pending_buf; 255 256 if (s->noheader < 0) { 257 s->noheader = 0; /* was set to -1 by deflate(..., Z_FINISH); */ 258 } 259 s->status = s->noheader ? BUSY_STATE : INIT_STATE; 260 strm->adler = 1; 261 s->last_flush = Z_NO_FLUSH; 262 263 zlib_tr_init(s); 264 lm_init(s); 265 266 return Z_OK; 267} 268 269/* ========================================================================= */ 270 271/* ========================================================================= 272 * Put a short in the pending buffer. The 16-bit value is put in MSB order. 273 * IN assertion: the stream state is correct and there is enough room in 274 * pending_buf. 275 */ 276static void putShortMSB( 277 deflate_state *s, 278 uInt b 279) 280{ 281 put_byte(s, (Byte)(b >> 8)); 282 put_byte(s, (Byte)(b & 0xff)); 283} 284 285/* ========================================================================= 286 * Flush as much pending output as possible. All deflate() output goes 287 * through this function so some applications may wish to modify it 288 * to avoid allocating a large strm->next_out buffer and copying into it. 289 * (See also read_buf()). 290 */ 291static void flush_pending( 292 z_streamp strm 293) 294{ 295 deflate_state *s = (deflate_state *) strm->state; 296 unsigned len = s->pending; 297 298 if (len > strm->avail_out) len = strm->avail_out; 299 if (len == 0) return; 300 301 if (strm->next_out != NULL) { 302 memcpy(strm->next_out, s->pending_out, len); 303 strm->next_out += len; 304 } 305 s->pending_out += len; 306 strm->total_out += len; 307 strm->avail_out -= len; 308 s->pending -= len; 309 if (s->pending == 0) { 310 s->pending_out = s->pending_buf; 311 } 312} 313 314/* ========================================================================= */ 315int zlib_deflate( 316 z_streamp strm, 317 int flush 318) 319{ 320 int old_flush; /* value of flush param for previous deflate call */ 321 deflate_state *s; 322 323 if (strm == NULL || strm->state == NULL || 324 flush > Z_FINISH || flush < 0) { 325 return Z_STREAM_ERROR; 326 } 327 s = (deflate_state *) strm->state; 328 329 if ((strm->next_in == NULL && strm->avail_in != 0) || 330 (s->status == FINISH_STATE && flush != Z_FINISH)) { 331 return Z_STREAM_ERROR; 332 } 333 if (strm->avail_out == 0) return Z_BUF_ERROR; 334 335 s->strm = strm; /* just in case */ 336 old_flush = s->last_flush; 337 s->last_flush = flush; 338 339 /* Write the zlib header */ 340 if (s->status == INIT_STATE) { 341 342 uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8; 343 uInt level_flags = (s->level-1) >> 1; 344 345 if (level_flags > 3) level_flags = 3; 346 header |= (level_flags << 6); 347 if (s->strstart != 0) header |= PRESET_DICT; 348 header += 31 - (header % 31); 349 350 s->status = BUSY_STATE; 351 putShortMSB(s, header); 352 353 /* Save the adler32 of the preset dictionary: */ 354 if (s->strstart != 0) { 355 putShortMSB(s, (uInt)(strm->adler >> 16)); 356 putShortMSB(s, (uInt)(strm->adler & 0xffff)); 357 } 358 strm->adler = 1L; 359 } 360 361 /* Flush as much pending output as possible */ 362 if (s->pending != 0) { 363 flush_pending(strm); 364 if (strm->avail_out == 0) { 365 /* Since avail_out is 0, deflate will be called again with 366 * more output space, but possibly with both pending and 367 * avail_in equal to zero. There won't be anything to do, 368 * but this is not an error situation so make sure we 369 * return OK instead of BUF_ERROR at next call of deflate: 370 */ 371 s->last_flush = -1; 372 return Z_OK; 373 } 374 375 /* Make sure there is something to do and avoid duplicate consecutive 376 * flushes. For repeated and useless calls with Z_FINISH, we keep 377 * returning Z_STREAM_END instead of Z_BUFF_ERROR. 378 */ 379 } else if (strm->avail_in == 0 && flush <= old_flush && 380 flush != Z_FINISH) { 381 return Z_BUF_ERROR; 382 } 383 384 /* User must not provide more input after the first FINISH: */ 385 if (s->status == FINISH_STATE && strm->avail_in != 0) { 386 return Z_BUF_ERROR; 387 } 388 389 /* Start a new block or continue the current one. 390 */ 391 if (strm->avail_in != 0 || s->lookahead != 0 || 392 (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) { 393 block_state bstate; 394 395 bstate = (*(configuration_table[s->level].func))(s, flush); 396 397 if (bstate == finish_started || bstate == finish_done) { 398 s->status = FINISH_STATE; 399 } 400 if (bstate == need_more || bstate == finish_started) { 401 if (strm->avail_out == 0) { 402 s->last_flush = -1; /* avoid BUF_ERROR next call, see above */ 403 } 404 return Z_OK; 405 /* If flush != Z_NO_FLUSH && avail_out == 0, the next call 406 * of deflate should use the same flush parameter to make sure 407 * that the flush is complete. So we don't have to output an 408 * empty block here, this will be done at next call. This also 409 * ensures that for a very small output buffer, we emit at most 410 * one empty block. 411 */ 412 } 413 if (bstate == block_done) { 414 if (flush == Z_PARTIAL_FLUSH) { 415 zlib_tr_align(s); 416 } else if (flush == Z_PACKET_FLUSH) { 417 /* Output just the 3-bit `stored' block type value, 418 but not a zero length. */ 419 zlib_tr_stored_type_only(s); 420 } else { /* FULL_FLUSH or SYNC_FLUSH */ 421 zlib_tr_stored_block(s, (char*)0, 0L, 0); 422 /* For a full flush, this empty block will be recognized 423 * as a special marker by inflate_sync(). 424 */ 425 if (flush == Z_FULL_FLUSH) { 426 CLEAR_HASH(s); /* forget history */ 427 } 428 } 429 flush_pending(strm); 430 if (strm->avail_out == 0) { 431 s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */ 432 return Z_OK; 433 } 434 } 435 } 436 Assert(strm->avail_out > 0, "bug2"); 437 438 if (flush != Z_FINISH) return Z_OK; 439 if (s->noheader) return Z_STREAM_END; 440 441 /* Write the zlib trailer (adler32) */ 442 putShortMSB(s, (uInt)(strm->adler >> 16)); 443 putShortMSB(s, (uInt)(strm->adler & 0xffff)); 444 flush_pending(strm); 445 /* If avail_out is zero, the application will call deflate again 446 * to flush the rest. 447 */ 448 s->noheader = -1; /* write the trailer only once! */ 449 return s->pending != 0 ? Z_OK : Z_STREAM_END; 450} 451 452/* ========================================================================= */ 453int zlib_deflateEnd( 454 z_streamp strm 455) 456{ 457 int status; 458 deflate_state *s; 459 460 if (strm == NULL || strm->state == NULL) return Z_STREAM_ERROR; 461 s = (deflate_state *) strm->state; 462 463 status = s->status; 464 if (status != INIT_STATE && status != BUSY_STATE && 465 status != FINISH_STATE) { 466 return Z_STREAM_ERROR; 467 } 468 469 strm->state = NULL; 470 471 return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK; 472} 473 474/* ========================================================================= 475 * Copy the source state to the destination state. 476 */ 477 478/* =========================================================================== 479 * Read a new buffer from the current input stream, update the adler32 480 * and total number of bytes read. All deflate() input goes through 481 * this function so some applications may wish to modify it to avoid 482 * allocating a large strm->next_in buffer and copying from it. 483 * (See also flush_pending()). 484 */ 485static int read_buf( 486 z_streamp strm, 487 Byte *buf, 488 unsigned size 489) 490{ 491 unsigned len = strm->avail_in; 492 493 if (len > size) len = size; 494 if (len == 0) return 0; 495 496 strm->avail_in -= len; 497 498 if (!((deflate_state *)(strm->state))->noheader) { 499 strm->adler = zlib_adler32(strm->adler, strm->next_in, len); 500 } 501 memcpy(buf, strm->next_in, len); 502 strm->next_in += len; 503 strm->total_in += len; 504 505 return (int)len; 506} 507 508/* =========================================================================== 509 * Initialize the "longest match" routines for a new zlib stream 510 */ 511static void lm_init( 512 deflate_state *s 513) 514{ 515 s->window_size = (ulg)2L*s->w_size; 516 517 CLEAR_HASH(s); 518 519 /* Set the default configuration parameters: 520 */ 521 s->max_lazy_match = configuration_table[s->level].max_lazy; 522 s->good_match = configuration_table[s->level].good_length; 523 s->nice_match = configuration_table[s->level].nice_length; 524 s->max_chain_length = configuration_table[s->level].max_chain; 525 526 s->strstart = 0; 527 s->block_start = 0L; 528 s->lookahead = 0; 529 s->match_length = s->prev_length = MIN_MATCH-1; 530 s->match_available = 0; 531 s->ins_h = 0; 532} 533 534/* =========================================================================== 535 * Set match_start to the longest match starting at the given string and 536 * return its length. Matches shorter or equal to prev_length are discarded, 537 * in which case the result is equal to prev_length and match_start is 538 * garbage. 539 * IN assertions: cur_match is the head of the hash chain for the current 540 * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1 541 * OUT assertion: the match length is not greater than s->lookahead. 542 */ 543/* For 80x86 and 680x0, an optimized version will be provided in match.asm or 544 * match.S. The code will be functionally equivalent. 545 */ 546static uInt longest_match( 547 deflate_state *s, 548 IPos cur_match /* current match */ 549) 550{ 551 unsigned chain_length = s->max_chain_length;/* max hash chain length */ 552 register Byte *scan = s->window + s->strstart; /* current string */ 553 register Byte *match; /* matched string */ 554 register int len; /* length of current match */ 555 int best_len = s->prev_length; /* best match length so far */ 556 int nice_match = s->nice_match; /* stop if match long enough */ 557 IPos limit = s->strstart > (IPos)MAX_DIST(s) ? 558 s->strstart - (IPos)MAX_DIST(s) : NIL; 559 /* Stop when cur_match becomes <= limit. To simplify the code, 560 * we prevent matches with the string of window index 0. 561 */ 562 Pos *prev = s->prev; 563 uInt wmask = s->w_mask; 564 565#ifdef UNALIGNED_OK 566 /* Compare two bytes at a time. Note: this is not always beneficial. 567 * Try with and without -DUNALIGNED_OK to check. 568 */ 569 register Byte *strend = s->window + s->strstart + MAX_MATCH - 1; 570 register ush scan_start = *(ush*)scan; 571 register ush scan_end = *(ush*)(scan+best_len-1); 572#else 573 register Byte *strend = s->window + s->strstart + MAX_MATCH; 574 register Byte scan_end1 = scan[best_len-1]; 575 register Byte scan_end = scan[best_len]; 576#endif 577 578 /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16. 579 * It is easy to get rid of this optimization if necessary. 580 */ 581 Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever"); 582 583 /* Do not waste too much time if we already have a good match: */ 584 if (s->prev_length >= s->good_match) { 585 chain_length >>= 2; 586 } 587 /* Do not look for matches beyond the end of the input. This is necessary 588 * to make deflate deterministic. 589 */ 590 if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead; 591 592 Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead"); 593 594 do { 595 Assert(cur_match < s->strstart, "no future"); 596 match = s->window + cur_match; 597 598 /* Skip to next match if the match length cannot increase 599 * or if the match length is less than 2: 600 */ 601#if (defined(UNALIGNED_OK) && MAX_MATCH == 258) 602 /* This code assumes sizeof(unsigned short) == 2. Do not use 603 * UNALIGNED_OK if your compiler uses a different size. 604 */ 605 if (*(ush*)(match+best_len-1) != scan_end || 606 *(ush*)match != scan_start) continue; 607 608 /* It is not necessary to compare scan[2] and match[2] since they are 609 * always equal when the other bytes match, given that the hash keys 610 * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at 611 * strstart+3, +5, ... up to strstart+257. We check for insufficient 612 * lookahead only every 4th comparison; the 128th check will be made 613 * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is 614 * necessary to put more guard bytes at the end of the window, or 615 * to check more often for insufficient lookahead. 616 */ 617 Assert(scan[2] == match[2], "scan[2]?"); 618 scan++, match++; 619 do { 620 } while (*(ush*)(scan+=2) == *(ush*)(match+=2) && 621 *(ush*)(scan+=2) == *(ush*)(match+=2) && 622 *(ush*)(scan+=2) == *(ush*)(match+=2) && 623 *(ush*)(scan+=2) == *(ush*)(match+=2) && 624 scan < strend); 625 /* The funny "do {}" generates better code on most compilers */ 626 627 /* Here, scan <= window+strstart+257 */ 628 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); 629 if (*scan == *match) scan++; 630 631 len = (MAX_MATCH - 1) - (int)(strend-scan); 632 scan = strend - (MAX_MATCH-1); 633 634#else /* UNALIGNED_OK */ 635 636 if (match[best_len] != scan_end || 637 match[best_len-1] != scan_end1 || 638 *match != *scan || 639 *++match != scan[1]) continue; 640 641 /* The check at best_len-1 can be removed because it will be made 642 * again later. (This heuristic is not always a win.) 643 * It is not necessary to compare scan[2] and match[2] since they 644 * are always equal when the other bytes match, given that 645 * the hash keys are equal and that HASH_BITS >= 8. 646 */ 647 scan += 2, match++; 648 Assert(*scan == *match, "match[2]?"); 649 650 /* We check for insufficient lookahead only every 8th comparison; 651 * the 256th check will be made at strstart+258. 652 */ 653 do { 654 } while (*++scan == *++match && *++scan == *++match && 655 *++scan == *++match && *++scan == *++match && 656 *++scan == *++match && *++scan == *++match && 657 *++scan == *++match && *++scan == *++match && 658 scan < strend); 659 660 Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan"); 661 662 len = MAX_MATCH - (int)(strend - scan); 663 scan = strend - MAX_MATCH; 664 665#endif /* UNALIGNED_OK */ 666 667 if (len > best_len) { 668 s->match_start = cur_match; 669 best_len = len; 670 if (len >= nice_match) break; 671#ifdef UNALIGNED_OK 672 scan_end = *(ush*)(scan+best_len-1); 673#else 674 scan_end1 = scan[best_len-1]; 675 scan_end = scan[best_len]; 676#endif 677 } 678 } while ((cur_match = prev[cur_match & wmask]) > limit 679 && --chain_length != 0); 680 681 if ((uInt)best_len <= s->lookahead) return best_len; 682 return s->lookahead; 683} 684 685#ifdef DEBUG_ZLIB 686/* =========================================================================== 687 * Check that the match at match_start is indeed a match. 688 */ 689static void check_match( 690 deflate_state *s, 691 IPos start, 692 IPos match, 693 int length 694) 695{ 696 /* check that the match is indeed a match */ 697 if (memcmp((char *)s->window + match, 698 (char *)s->window + start, length) != EQUAL) { 699 fprintf(stderr, " start %u, match %u, length %d\n", 700 start, match, length); 701 do { 702 fprintf(stderr, "%c%c", s->window[match++], s->window[start++]); 703 } while (--length != 0); 704 z_error("invalid match"); 705 } 706 if (z_verbose > 1) { 707 fprintf(stderr,"\\[%d,%d]", start-match, length); 708 do { putc(s->window[start++], stderr); } while (--length != 0); 709 } 710} 711#else 712# define check_match(s, start, match, length) 713#endif 714 715/* =========================================================================== 716 * Fill the window when the lookahead becomes insufficient. 717 * Updates strstart and lookahead. 718 * 719 * IN assertion: lookahead < MIN_LOOKAHEAD 720 * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD 721 * At least one byte has been read, or avail_in == 0; reads are 722 * performed for at least two bytes (required for the zip translate_eol 723 * option -- not supported here). 724 */ 725static void fill_window( 726 deflate_state *s 727) 728{ 729 register unsigned n, m; 730 register Pos *p; 731 unsigned more; /* Amount of free space at the end of the window. */ 732 uInt wsize = s->w_size; 733 734 do { 735 more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart); 736 737 /* Deal with !@#$% 64K limit: */ 738 if (more == 0 && s->strstart == 0 && s->lookahead == 0) { 739 more = wsize; 740 741 } else if (more == (unsigned)(-1)) { 742 /* Very unlikely, but possible on 16 bit machine if strstart == 0 743 * and lookahead == 1 (input done one byte at time) 744 */ 745 more--; 746 747 /* If the window is almost full and there is insufficient lookahead, 748 * move the upper half to the lower one to make room in the upper half. 749 */ 750 } else if (s->strstart >= wsize+MAX_DIST(s)) { 751 752 memcpy((char *)s->window, (char *)s->window+wsize, 753 (unsigned)wsize); 754 s->match_start -= wsize; 755 s->strstart -= wsize; /* we now have strstart >= MAX_DIST */ 756 s->block_start -= (long) wsize; 757 758 /* Slide the hash table (could be avoided with 32 bit values 759 at the expense of memory usage). We slide even when level == 0 760 to keep the hash table consistent if we switch back to level > 0 761 later. (Using level 0 permanently is not an optimal usage of 762 zlib, so we don't care about this pathological case.) 763 */ 764 n = s->hash_size; 765 p = &s->head[n]; 766 do { 767 m = *--p; 768 *p = (Pos)(m >= wsize ? m-wsize : NIL); 769 } while (--n); 770 771 n = wsize; 772 p = &s->prev[n]; 773 do { 774 m = *--p; 775 *p = (Pos)(m >= wsize ? m-wsize : NIL); 776 /* If n is not on any hash chain, prev[n] is garbage but 777 * its value will never be used. 778 */ 779 } while (--n); 780 more += wsize; 781 } 782 if (s->strm->avail_in == 0) return; 783 784 /* If there was no sliding: 785 * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 && 786 * more == window_size - lookahead - strstart 787 * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1) 788 * => more >= window_size - 2*WSIZE + 2 789 * In the BIG_MEM or MMAP case (not yet supported), 790 * window_size == input_size + MIN_LOOKAHEAD && 791 * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD. 792 * Otherwise, window_size == 2*WSIZE so more >= 2. 793 * If there was sliding, more >= WSIZE. So in all cases, more >= 2. 794 */ 795 Assert(more >= 2, "more < 2"); 796 797 n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more); 798 s->lookahead += n; 799 800 /* Initialize the hash value now that we have some input: */ 801 if (s->lookahead >= MIN_MATCH) { 802 s->ins_h = s->window[s->strstart]; 803 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]); 804#if MIN_MATCH != 3 805 Call UPDATE_HASH() MIN_MATCH-3 more times 806#endif 807 } 808 /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage, 809 * but this is not important since only literal bytes will be emitted. 810 */ 811 812 } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0); 813} 814 815/* =========================================================================== 816 * Flush the current block, with given end-of-file flag. 817 * IN assertion: strstart is set to the end of the current match. 818 */ 819#define FLUSH_BLOCK_ONLY(s, eof) { \ 820 zlib_tr_flush_block(s, (s->block_start >= 0L ? \ 821 (char *)&s->window[(unsigned)s->block_start] : \ 822 NULL), \ 823 (ulg)((long)s->strstart - s->block_start), \ 824 (eof)); \ 825 s->block_start = s->strstart; \ 826 flush_pending(s->strm); \ 827 Tracev((stderr,"[FLUSH]")); \ 828} 829 830/* Same but force premature exit if necessary. */ 831#define FLUSH_BLOCK(s, eof) { \ 832 FLUSH_BLOCK_ONLY(s, eof); \ 833 if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \ 834} 835 836/* =========================================================================== 837 * Copy without compression as much as possible from the input stream, return 838 * the current block state. 839 * This function does not insert new strings in the dictionary since 840 * uncompressible data is probably not useful. This function is used 841 * only for the level=0 compression option. 842 * NOTE: this function should be optimized to avoid extra copying from 843 * window to pending_buf. 844 */ 845static block_state deflate_stored( 846 deflate_state *s, 847 int flush 848) 849{ 850 /* Stored blocks are limited to 0xffff bytes, pending_buf is limited 851 * to pending_buf_size, and each stored block has a 5 byte header: 852 */ 853 ulg max_block_size = 0xffff; 854 ulg max_start; 855 856 if (max_block_size > s->pending_buf_size - 5) { 857 max_block_size = s->pending_buf_size - 5; 858 } 859 860 /* Copy as much as possible from input to output: */ 861 for (;;) { 862 /* Fill the window as much as possible: */ 863 if (s->lookahead <= 1) { 864 865 Assert(s->strstart < s->w_size+MAX_DIST(s) || 866 s->block_start >= (long)s->w_size, "slide too late"); 867 868 fill_window(s); 869 if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more; 870 871 if (s->lookahead == 0) break; /* flush the current block */ 872 } 873 Assert(s->block_start >= 0L, "block gone"); 874 875 s->strstart += s->lookahead; 876 s->lookahead = 0; 877 878 /* Emit a stored block if pending_buf will be full: */ 879 max_start = s->block_start + max_block_size; 880 if (s->strstart == 0 || (ulg)s->strstart >= max_start) { 881 /* strstart == 0 is possible when wraparound on 16-bit machine */ 882 s->lookahead = (uInt)(s->strstart - max_start); 883 s->strstart = (uInt)max_start; 884 FLUSH_BLOCK(s, 0); 885 } 886 /* Flush if we may have to slide, otherwise block_start may become 887 * negative and the data will be gone: 888 */ 889 if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) { 890 FLUSH_BLOCK(s, 0); 891 } 892 } 893 FLUSH_BLOCK(s, flush == Z_FINISH); 894 return flush == Z_FINISH ? finish_done : block_done; 895} 896 897/* =========================================================================== 898 * Compress as much as possible from the input stream, return the current 899 * block state. 900 * This function does not perform lazy evaluation of matches and inserts 901 * new strings in the dictionary only for unmatched strings or for short 902 * matches. It is used only for the fast compression options. 903 */ 904static block_state deflate_fast( 905 deflate_state *s, 906 int flush 907) 908{ 909 IPos hash_head = NIL; /* head of the hash chain */ 910 int bflush; /* set if current block must be flushed */ 911 912 for (;;) { 913 /* Make sure that we always have enough lookahead, except 914 * at the end of the input file. We need MAX_MATCH bytes 915 * for the next match, plus MIN_MATCH bytes to insert the 916 * string following the next match. 917 */ 918 if (s->lookahead < MIN_LOOKAHEAD) { 919 fill_window(s); 920 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { 921 return need_more; 922 } 923 if (s->lookahead == 0) break; /* flush the current block */ 924 } 925 926 /* Insert the string window[strstart .. strstart+2] in the 927 * dictionary, and set hash_head to the head of the hash chain: 928 */ 929 if (s->lookahead >= MIN_MATCH) { 930 INSERT_STRING(s, s->strstart, hash_head); 931 } 932 933 /* Find the longest match, discarding those <= prev_length. 934 * At this point we have always match_length < MIN_MATCH 935 */ 936 if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) { 937 /* To simplify the code, we prevent matches with the string 938 * of window index 0 (in particular we have to avoid a match 939 * of the string with itself at the start of the input file). 940 */ 941 if (s->strategy != Z_HUFFMAN_ONLY) { 942 s->match_length = longest_match (s, hash_head); 943 } 944 /* longest_match() sets match_start */ 945 } 946 if (s->match_length >= MIN_MATCH) { 947 check_match(s, s->strstart, s->match_start, s->match_length); 948 949 bflush = zlib_tr_tally(s, s->strstart - s->match_start, 950 s->match_length - MIN_MATCH); 951 952 s->lookahead -= s->match_length; 953 954 /* Insert new strings in the hash table only if the match length 955 * is not too large. This saves time but degrades compression. 956 */ 957 if (s->match_length <= s->max_insert_length && 958 s->lookahead >= MIN_MATCH) { 959 s->match_length--; /* string at strstart already in hash table */ 960 do { 961 s->strstart++; 962 INSERT_STRING(s, s->strstart, hash_head); 963 /* strstart never exceeds WSIZE-MAX_MATCH, so there are 964 * always MIN_MATCH bytes ahead. 965 */ 966 } while (--s->match_length != 0); 967 s->strstart++; 968 } else { 969 s->strstart += s->match_length; 970 s->match_length = 0; 971 s->ins_h = s->window[s->strstart]; 972 UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]); 973#if MIN_MATCH != 3 974 Call UPDATE_HASH() MIN_MATCH-3 more times 975#endif 976 /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not 977 * matter since it will be recomputed at next deflate call. 978 */ 979 } 980 } else { 981 /* No match, output a literal byte */ 982 Tracevv((stderr,"%c", s->window[s->strstart])); 983 bflush = zlib_tr_tally (s, 0, s->window[s->strstart]); 984 s->lookahead--; 985 s->strstart++; 986 } 987 if (bflush) FLUSH_BLOCK(s, 0); 988 } 989 FLUSH_BLOCK(s, flush == Z_FINISH); 990 return flush == Z_FINISH ? finish_done : block_done; 991} 992 993/* =========================================================================== 994 * Same as above, but achieves better compression. We use a lazy 995 * evaluation for matches: a match is finally adopted only if there is 996 * no better match at the next window position. 997 */ 998static block_state deflate_slow( 999 deflate_state *s, 1000 int flush 1001) 1002{ 1003 IPos hash_head = NIL; /* head of hash chain */ 1004 int bflush; /* set if current block must be flushed */ 1005 1006 /* Process the input block. */ 1007 for (;;) { 1008 /* Make sure that we always have enough lookahead, except 1009 * at the end of the input file. We need MAX_MATCH bytes 1010 * for the next match, plus MIN_MATCH bytes to insert the 1011 * string following the next match. 1012 */ 1013 if (s->lookahead < MIN_LOOKAHEAD) { 1014 fill_window(s); 1015 if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) { 1016 return need_more; 1017 } 1018 if (s->lookahead == 0) break; /* flush the current block */ 1019 } 1020 1021 /* Insert the string window[strstart .. strstart+2] in the 1022 * dictionary, and set hash_head to the head of the hash chain: 1023 */ 1024 if (s->lookahead >= MIN_MATCH) { 1025 INSERT_STRING(s, s->strstart, hash_head); 1026 } 1027 1028 /* Find the longest match, discarding those <= prev_length. 1029 */ 1030 s->prev_length = s->match_length, s->prev_match = s->match_start; 1031 s->match_length = MIN_MATCH-1; 1032 1033 if (hash_head != NIL && s->prev_length < s->max_lazy_match && 1034 s->strstart - hash_head <= MAX_DIST(s)) { 1035 /* To simplify the code, we prevent matches with the string 1036 * of window index 0 (in particular we have to avoid a match 1037 * of the string with itself at the start of the input file). 1038 */ 1039 if (s->strategy != Z_HUFFMAN_ONLY) { 1040 s->match_length = longest_match (s, hash_head); 1041 } 1042 /* longest_match() sets match_start */ 1043 1044 if (s->match_length <= 5 && (s->strategy == Z_FILTERED || 1045 (s->match_length == MIN_MATCH && 1046 s->strstart - s->match_start > TOO_FAR))) { 1047 1048 /* If prev_match is also MIN_MATCH, match_start is garbage 1049 * but we will ignore the current match anyway. 1050 */ 1051 s->match_length = MIN_MATCH-1; 1052 } 1053 } 1054 /* If there was a match at the previous step and the current 1055 * match is not better, output the previous match: 1056 */ 1057 if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) { 1058 uInt max_insert = s->strstart + s->lookahead - MIN_MATCH; 1059 /* Do not insert strings in hash table beyond this. */ 1060 1061 check_match(s, s->strstart-1, s->prev_match, s->prev_length); 1062 1063 bflush = zlib_tr_tally(s, s->strstart -1 - s->prev_match, 1064 s->prev_length - MIN_MATCH); 1065 1066 /* Insert in hash table all strings up to the end of the match. 1067 * strstart-1 and strstart are already inserted. If there is not 1068 * enough lookahead, the last two strings are not inserted in 1069 * the hash table. 1070 */ 1071 s->lookahead -= s->prev_length-1; 1072 s->prev_length -= 2; 1073 do { 1074 if (++s->strstart <= max_insert) { 1075 INSERT_STRING(s, s->strstart, hash_head); 1076 } 1077 } while (--s->prev_length != 0); 1078 s->match_available = 0; 1079 s->match_length = MIN_MATCH-1; 1080 s->strstart++; 1081 1082 if (bflush) FLUSH_BLOCK(s, 0); 1083 1084 } else if (s->match_available) { 1085 /* If there was no match at the previous position, output a 1086 * single literal. If there was a match but the current match 1087 * is longer, truncate the previous match to a single literal. 1088 */ 1089 Tracevv((stderr,"%c", s->window[s->strstart-1])); 1090 if (zlib_tr_tally (s, 0, s->window[s->strstart-1])) { 1091 FLUSH_BLOCK_ONLY(s, 0); 1092 } 1093 s->strstart++; 1094 s->lookahead--; 1095 if (s->strm->avail_out == 0) return need_more; 1096 } else { 1097 /* There is no previous match to compare with, wait for 1098 * the next step to decide. 1099 */ 1100 s->match_available = 1; 1101 s->strstart++; 1102 s->lookahead--; 1103 } 1104 } 1105 Assert (flush != Z_NO_FLUSH, "no flush?"); 1106 if (s->match_available) { 1107 Tracevv((stderr,"%c", s->window[s->strstart-1])); 1108 zlib_tr_tally (s, 0, s->window[s->strstart-1]); 1109 s->match_available = 0; 1110 } 1111 FLUSH_BLOCK(s, flush == Z_FINISH); 1112 return flush == Z_FINISH ? finish_done : block_done; 1113} 1114 1115int zlib_deflate_workspacesize(void) 1116{ 1117 return sizeof(deflate_workspace); 1118} 1119