Line data Source code
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 : /* architecture-specific bits */
56 : #ifdef CONFIG_ZLIB_DFLTCC
57 : # include "../zlib_dfltcc/dfltcc.h"
58 : #else
59 : #define DEFLATE_RESET_HOOK(strm) do {} while (0)
60 : #define DEFLATE_HOOK(strm, flush, bstate) 0
61 : #define DEFLATE_NEED_CHECKSUM(strm) 1
62 : #define DEFLATE_DFLTCC_ENABLED() 0
63 : #endif
64 :
65 : /* ===========================================================================
66 : * Function prototypes.
67 : */
68 :
69 : typedef block_state (*compress_func) (deflate_state *s, int flush);
70 : /* Compression function. Returns the block state after the call. */
71 :
72 : static void fill_window (deflate_state *s);
73 : static block_state deflate_stored (deflate_state *s, int flush);
74 : static block_state deflate_fast (deflate_state *s, int flush);
75 : static block_state deflate_slow (deflate_state *s, int flush);
76 : static void lm_init (deflate_state *s);
77 : static void putShortMSB (deflate_state *s, uInt b);
78 : static int read_buf (z_streamp strm, Byte *buf, unsigned size);
79 : static uInt longest_match (deflate_state *s, IPos cur_match);
80 :
81 : #ifdef DEBUG_ZLIB
82 : static void check_match (deflate_state *s, IPos start, IPos match,
83 : int length);
84 : #endif
85 :
86 : /* ===========================================================================
87 : * Local data
88 : */
89 :
90 : #define NIL 0
91 : /* Tail of hash chains */
92 :
93 : #ifndef TOO_FAR
94 : # define TOO_FAR 4096
95 : #endif
96 : /* Matches of length 3 are discarded if their distance exceeds TOO_FAR */
97 :
98 : #define MIN_LOOKAHEAD (MAX_MATCH+MIN_MATCH+1)
99 : /* Minimum amount of lookahead, except at the end of the input file.
100 : * See deflate.c for comments about the MIN_MATCH+1.
101 : */
102 :
103 : /* Workspace to be allocated for deflate processing */
104 : typedef struct deflate_workspace {
105 : /* State memory for the deflator */
106 : deflate_state deflate_memory;
107 : #ifdef CONFIG_ZLIB_DFLTCC
108 : /* State memory for s390 hardware deflate */
109 : struct dfltcc_state dfltcc_memory;
110 : #endif
111 : Byte *window_memory;
112 : Pos *prev_memory;
113 : Pos *head_memory;
114 : char *overlay_memory;
115 : } deflate_workspace;
116 :
117 : #ifdef CONFIG_ZLIB_DFLTCC
118 : /* dfltcc_state must be doubleword aligned for DFLTCC call */
119 : static_assert(offsetof(struct deflate_workspace, dfltcc_memory) % 8 == 0);
120 : #endif
121 :
122 : /* Values for max_lazy_match, good_match and max_chain_length, depending on
123 : * the desired pack level (0..9). The values given below have been tuned to
124 : * exclude worst case performance for pathological files. Better values may be
125 : * found for specific files.
126 : */
127 : typedef struct config_s {
128 : ush good_length; /* reduce lazy search above this match length */
129 : ush max_lazy; /* do not perform lazy search above this match length */
130 : ush nice_length; /* quit search above this match length */
131 : ush max_chain;
132 : compress_func func;
133 : } config;
134 :
135 : static const config configuration_table[10] = {
136 : /* good lazy nice chain */
137 : /* 0 */ {0, 0, 0, 0, deflate_stored}, /* store only */
138 : /* 1 */ {4, 4, 8, 4, deflate_fast}, /* maximum speed, no lazy matches */
139 : /* 2 */ {4, 5, 16, 8, deflate_fast},
140 : /* 3 */ {4, 6, 32, 32, deflate_fast},
141 :
142 : /* 4 */ {4, 4, 16, 16, deflate_slow}, /* lazy matches */
143 : /* 5 */ {8, 16, 32, 32, deflate_slow},
144 : /* 6 */ {8, 16, 128, 128, deflate_slow},
145 : /* 7 */ {8, 32, 128, 256, deflate_slow},
146 : /* 8 */ {32, 128, 258, 1024, deflate_slow},
147 : /* 9 */ {32, 258, 258, 4096, deflate_slow}}; /* maximum compression */
148 :
149 : /* Note: the deflate() code requires max_lazy >= MIN_MATCH and max_chain >= 4
150 : * For deflate_fast() (levels <= 3) good is ignored and lazy has a different
151 : * meaning.
152 : */
153 :
154 : #define EQUAL 0
155 : /* result of memcmp for equal strings */
156 :
157 : /* ===========================================================================
158 : * Update a hash value with the given input byte
159 : * IN assertion: all calls to UPDATE_HASH are made with consecutive
160 : * input characters, so that a running hash key can be computed from the
161 : * previous key instead of complete recalculation each time.
162 : */
163 : #define UPDATE_HASH(s,h,c) (h = (((h)<<s->hash_shift) ^ (c)) & s->hash_mask)
164 :
165 :
166 : /* ===========================================================================
167 : * Insert string str in the dictionary and set match_head to the previous head
168 : * of the hash chain (the most recent string with same hash key). Return
169 : * the previous length of the hash chain.
170 : * IN assertion: all calls to INSERT_STRING are made with consecutive
171 : * input characters and the first MIN_MATCH bytes of str are valid
172 : * (except for the last MIN_MATCH-1 bytes of the input file).
173 : */
174 : #define INSERT_STRING(s, str, match_head) \
175 : (UPDATE_HASH(s, s->ins_h, s->window[(str) + (MIN_MATCH-1)]), \
176 : s->prev[(str) & s->w_mask] = match_head = s->head[s->ins_h], \
177 : s->head[s->ins_h] = (Pos)(str))
178 :
179 : /* ===========================================================================
180 : * Initialize the hash table (avoiding 64K overflow for 16 bit systems).
181 : * prev[] will be initialized on the fly.
182 : */
183 : #define CLEAR_HASH(s) \
184 : s->head[s->hash_size-1] = NIL; \
185 : memset((char *)s->head, 0, (unsigned)(s->hash_size-1)*sizeof(*s->head));
186 :
187 : /* ========================================================================= */
188 0 : int zlib_deflateInit2(
189 : z_streamp strm,
190 : int level,
191 : int method,
192 : int windowBits,
193 : int memLevel,
194 : int strategy
195 : )
196 : {
197 0 : deflate_state *s;
198 0 : int noheader = 0;
199 0 : deflate_workspace *mem;
200 0 : char *next;
201 :
202 0 : ush *overlay;
203 : /* We overlay pending_buf and d_buf+l_buf. This works since the average
204 : * output size for (length,distance) codes is <= 24 bits.
205 : */
206 :
207 0 : if (strm == NULL) return Z_STREAM_ERROR;
208 :
209 0 : strm->msg = NULL;
210 :
211 0 : if (level == Z_DEFAULT_COMPRESSION) level = 6;
212 :
213 0 : mem = (deflate_workspace *) strm->workspace;
214 :
215 0 : if (windowBits < 0) { /* undocumented feature: suppress zlib header */
216 0 : noheader = 1;
217 0 : windowBits = -windowBits;
218 : }
219 0 : if (memLevel < 1 || memLevel > MAX_MEM_LEVEL || method != Z_DEFLATED ||
220 0 : windowBits < 9 || windowBits > 15 || level < 0 || level > 9 ||
221 0 : strategy < 0 || strategy > Z_HUFFMAN_ONLY) {
222 : return Z_STREAM_ERROR;
223 : }
224 :
225 : /*
226 : * Direct the workspace's pointers to the chunks that were allocated
227 : * along with the deflate_workspace struct.
228 : */
229 0 : next = (char *) mem;
230 0 : next += sizeof(*mem);
231 : #ifdef CONFIG_ZLIB_DFLTCC
232 : /*
233 : * DFLTCC requires the window to be page aligned.
234 : * Thus, we overallocate and take the aligned portion of the buffer.
235 : */
236 : mem->window_memory = (Byte *) PTR_ALIGN(next, PAGE_SIZE);
237 : #else
238 0 : mem->window_memory = (Byte *) next;
239 : #endif
240 0 : next += zlib_deflate_window_memsize(windowBits);
241 0 : mem->prev_memory = (Pos *) next;
242 0 : next += zlib_deflate_prev_memsize(windowBits);
243 0 : mem->head_memory = (Pos *) next;
244 0 : next += zlib_deflate_head_memsize(memLevel);
245 0 : mem->overlay_memory = next;
246 :
247 0 : s = (deflate_state *) &(mem->deflate_memory);
248 0 : strm->state = (struct internal_state *)s;
249 0 : s->strm = strm;
250 :
251 0 : s->noheader = noheader;
252 0 : s->w_bits = windowBits;
253 0 : s->w_size = 1 << s->w_bits;
254 0 : s->w_mask = s->w_size - 1;
255 :
256 0 : s->hash_bits = memLevel + 7;
257 0 : s->hash_size = 1 << s->hash_bits;
258 0 : s->hash_mask = s->hash_size - 1;
259 0 : s->hash_shift = ((s->hash_bits+MIN_MATCH-1)/MIN_MATCH);
260 :
261 0 : s->window = (Byte *) mem->window_memory;
262 0 : s->prev = (Pos *) mem->prev_memory;
263 0 : s->head = (Pos *) mem->head_memory;
264 :
265 0 : s->lit_bufsize = 1 << (memLevel + 6); /* 16K elements by default */
266 :
267 0 : overlay = (ush *) mem->overlay_memory;
268 0 : s->pending_buf = (uch *) overlay;
269 0 : s->pending_buf_size = (ulg)s->lit_bufsize * (sizeof(ush)+2L);
270 :
271 0 : s->d_buf = overlay + s->lit_bufsize/sizeof(ush);
272 0 : s->l_buf = s->pending_buf + (1+sizeof(ush))*s->lit_bufsize;
273 :
274 0 : s->level = level;
275 0 : s->strategy = strategy;
276 0 : s->method = (Byte)method;
277 :
278 0 : return zlib_deflateReset(strm);
279 : }
280 :
281 : /* ========================================================================= */
282 0 : int zlib_deflateReset(
283 : z_streamp strm
284 : )
285 : {
286 0 : deflate_state *s;
287 :
288 0 : if (strm == NULL || strm->state == NULL)
289 : return Z_STREAM_ERROR;
290 :
291 0 : strm->total_in = strm->total_out = 0;
292 0 : strm->msg = NULL;
293 0 : strm->data_type = Z_UNKNOWN;
294 :
295 0 : s = (deflate_state *)strm->state;
296 0 : s->pending = 0;
297 0 : s->pending_out = s->pending_buf;
298 :
299 0 : if (s->noheader < 0) {
300 0 : s->noheader = 0; /* was set to -1 by deflate(..., Z_FINISH); */
301 : }
302 0 : s->status = s->noheader ? BUSY_STATE : INIT_STATE;
303 0 : strm->adler = 1;
304 0 : s->last_flush = Z_NO_FLUSH;
305 :
306 0 : zlib_tr_init(s);
307 0 : lm_init(s);
308 :
309 0 : DEFLATE_RESET_HOOK(strm);
310 :
311 0 : return Z_OK;
312 : }
313 :
314 : /* =========================================================================
315 : * Put a short in the pending buffer. The 16-bit value is put in MSB order.
316 : * IN assertion: the stream state is correct and there is enough room in
317 : * pending_buf.
318 : */
319 0 : static void putShortMSB(
320 : deflate_state *s,
321 : uInt b
322 : )
323 : {
324 0 : put_byte(s, (Byte)(b >> 8));
325 0 : put_byte(s, (Byte)(b & 0xff));
326 0 : }
327 :
328 : /* ========================================================================= */
329 0 : int zlib_deflate(
330 : z_streamp strm,
331 : int flush
332 : )
333 : {
334 0 : int old_flush; /* value of flush param for previous deflate call */
335 0 : deflate_state *s;
336 :
337 0 : if (strm == NULL || strm->state == NULL ||
338 0 : flush > Z_FINISH || flush < 0) {
339 : return Z_STREAM_ERROR;
340 : }
341 0 : s = (deflate_state *) strm->state;
342 :
343 0 : if ((strm->next_in == NULL && strm->avail_in != 0) ||
344 0 : (s->status == FINISH_STATE && flush != Z_FINISH)) {
345 : return Z_STREAM_ERROR;
346 : }
347 0 : if (strm->avail_out == 0) return Z_BUF_ERROR;
348 :
349 0 : s->strm = strm; /* just in case */
350 0 : old_flush = s->last_flush;
351 0 : s->last_flush = flush;
352 :
353 : /* Write the zlib header */
354 0 : if (s->status == INIT_STATE) {
355 :
356 0 : uInt header = (Z_DEFLATED + ((s->w_bits-8)<<4)) << 8;
357 0 : uInt level_flags = (s->level-1) >> 1;
358 :
359 0 : if (level_flags > 3) level_flags = 3;
360 0 : header |= (level_flags << 6);
361 0 : if (s->strstart != 0) header |= PRESET_DICT;
362 0 : header += 31 - (header % 31);
363 :
364 0 : s->status = BUSY_STATE;
365 0 : putShortMSB(s, header);
366 :
367 : /* Save the adler32 of the preset dictionary: */
368 0 : if (s->strstart != 0) {
369 0 : putShortMSB(s, (uInt)(strm->adler >> 16));
370 0 : putShortMSB(s, (uInt)(strm->adler & 0xffff));
371 : }
372 0 : strm->adler = 1L;
373 : }
374 :
375 : /* Flush as much pending output as possible */
376 0 : if (s->pending != 0) {
377 0 : flush_pending(strm);
378 0 : if (strm->avail_out == 0) {
379 : /* Since avail_out is 0, deflate will be called again with
380 : * more output space, but possibly with both pending and
381 : * avail_in equal to zero. There won't be anything to do,
382 : * but this is not an error situation so make sure we
383 : * return OK instead of BUF_ERROR at next call of deflate:
384 : */
385 0 : s->last_flush = -1;
386 0 : return Z_OK;
387 : }
388 :
389 : /* Make sure there is something to do and avoid duplicate consecutive
390 : * flushes. For repeated and useless calls with Z_FINISH, we keep
391 : * returning Z_STREAM_END instead of Z_BUFF_ERROR.
392 : */
393 0 : } else if (strm->avail_in == 0 && flush <= old_flush &&
394 0 : flush != Z_FINISH) {
395 : return Z_BUF_ERROR;
396 : }
397 :
398 : /* User must not provide more input after the first FINISH: */
399 0 : if (s->status == FINISH_STATE && strm->avail_in != 0) {
400 : return Z_BUF_ERROR;
401 : }
402 :
403 : /* Start a new block or continue the current one.
404 : */
405 0 : if (strm->avail_in != 0 || s->lookahead != 0 ||
406 0 : (flush != Z_NO_FLUSH && s->status != FINISH_STATE)) {
407 0 : block_state bstate;
408 :
409 0 : bstate = DEFLATE_HOOK(strm, flush, &bstate) ? bstate :
410 0 : (*(configuration_table[s->level].func))(s, flush);
411 :
412 0 : if (bstate == finish_started || bstate == finish_done) {
413 0 : s->status = FINISH_STATE;
414 : }
415 0 : if (bstate == need_more || bstate == finish_started) {
416 0 : if (strm->avail_out == 0) {
417 0 : s->last_flush = -1; /* avoid BUF_ERROR next call, see above */
418 : }
419 0 : return Z_OK;
420 : /* If flush != Z_NO_FLUSH && avail_out == 0, the next call
421 : * of deflate should use the same flush parameter to make sure
422 : * that the flush is complete. So we don't have to output an
423 : * empty block here, this will be done at next call. This also
424 : * ensures that for a very small output buffer, we emit at most
425 : * one empty block.
426 : */
427 : }
428 0 : if (bstate == block_done) {
429 0 : if (flush == Z_PARTIAL_FLUSH) {
430 0 : zlib_tr_align(s);
431 0 : } else if (flush == Z_PACKET_FLUSH) {
432 : /* Output just the 3-bit `stored' block type value,
433 : but not a zero length. */
434 0 : zlib_tr_stored_type_only(s);
435 : } else { /* FULL_FLUSH or SYNC_FLUSH */
436 0 : zlib_tr_stored_block(s, (char*)0, 0L, 0);
437 : /* For a full flush, this empty block will be recognized
438 : * as a special marker by inflate_sync().
439 : */
440 0 : if (flush == Z_FULL_FLUSH) {
441 0 : CLEAR_HASH(s); /* forget history */
442 : }
443 : }
444 0 : flush_pending(strm);
445 0 : if (strm->avail_out == 0) {
446 0 : s->last_flush = -1; /* avoid BUF_ERROR at next call, see above */
447 0 : return Z_OK;
448 : }
449 : }
450 : }
451 0 : Assert(strm->avail_out > 0, "bug2");
452 :
453 0 : if (flush != Z_FINISH) return Z_OK;
454 0 : if (s->noheader) return Z_STREAM_END;
455 :
456 : /* Write the zlib trailer (adler32) */
457 0 : putShortMSB(s, (uInt)(strm->adler >> 16));
458 0 : putShortMSB(s, (uInt)(strm->adler & 0xffff));
459 0 : flush_pending(strm);
460 : /* If avail_out is zero, the application will call deflate again
461 : * to flush the rest.
462 : */
463 0 : s->noheader = -1; /* write the trailer only once! */
464 0 : return s->pending != 0 ? Z_OK : Z_STREAM_END;
465 : }
466 :
467 : /* ========================================================================= */
468 0 : int zlib_deflateEnd(
469 : z_streamp strm
470 : )
471 : {
472 0 : int status;
473 0 : deflate_state *s;
474 :
475 0 : if (strm == NULL || strm->state == NULL) return Z_STREAM_ERROR;
476 0 : s = (deflate_state *) strm->state;
477 :
478 0 : status = s->status;
479 0 : if (status != INIT_STATE && status != BUSY_STATE &&
480 : status != FINISH_STATE) {
481 : return Z_STREAM_ERROR;
482 : }
483 :
484 0 : strm->state = NULL;
485 :
486 0 : return status == BUSY_STATE ? Z_DATA_ERROR : Z_OK;
487 : }
488 :
489 : /* ===========================================================================
490 : * Read a new buffer from the current input stream, update the adler32
491 : * and total number of bytes read. All deflate() input goes through
492 : * this function so some applications may wish to modify it to avoid
493 : * allocating a large strm->next_in buffer and copying from it.
494 : * (See also flush_pending()).
495 : */
496 0 : static int read_buf(
497 : z_streamp strm,
498 : Byte *buf,
499 : unsigned size
500 : )
501 : {
502 0 : unsigned len = strm->avail_in;
503 :
504 0 : if (len > size) len = size;
505 0 : if (len == 0) return 0;
506 :
507 0 : strm->avail_in -= len;
508 :
509 0 : if (!DEFLATE_NEED_CHECKSUM(strm)) {}
510 0 : else if (!((deflate_state *)(strm->state))->noheader) {
511 0 : strm->adler = zlib_adler32(strm->adler, strm->next_in, len);
512 : }
513 0 : memcpy(buf, strm->next_in, len);
514 0 : strm->next_in += len;
515 0 : strm->total_in += len;
516 :
517 0 : return (int)len;
518 : }
519 :
520 : /* ===========================================================================
521 : * Initialize the "longest match" routines for a new zlib stream
522 : */
523 0 : static void lm_init(
524 : deflate_state *s
525 : )
526 : {
527 0 : s->window_size = (ulg)2L*s->w_size;
528 :
529 0 : CLEAR_HASH(s);
530 :
531 : /* Set the default configuration parameters:
532 : */
533 0 : s->max_lazy_match = configuration_table[s->level].max_lazy;
534 0 : s->good_match = configuration_table[s->level].good_length;
535 0 : s->nice_match = configuration_table[s->level].nice_length;
536 0 : s->max_chain_length = configuration_table[s->level].max_chain;
537 :
538 0 : s->strstart = 0;
539 0 : s->block_start = 0L;
540 0 : s->lookahead = 0;
541 0 : s->match_length = s->prev_length = MIN_MATCH-1;
542 0 : s->match_available = 0;
543 0 : s->ins_h = 0;
544 0 : }
545 :
546 : /* ===========================================================================
547 : * Set match_start to the longest match starting at the given string and
548 : * return its length. Matches shorter or equal to prev_length are discarded,
549 : * in which case the result is equal to prev_length and match_start is
550 : * garbage.
551 : * IN assertions: cur_match is the head of the hash chain for the current
552 : * string (strstart) and its distance is <= MAX_DIST, and prev_length >= 1
553 : * OUT assertion: the match length is not greater than s->lookahead.
554 : */
555 : /* For 80x86 and 680x0, an optimized version will be provided in match.asm or
556 : * match.S. The code will be functionally equivalent.
557 : */
558 0 : static uInt longest_match(
559 : deflate_state *s,
560 : IPos cur_match /* current match */
561 : )
562 : {
563 0 : unsigned chain_length = s->max_chain_length;/* max hash chain length */
564 0 : register Byte *scan = s->window + s->strstart; /* current string */
565 0 : register Byte *match; /* matched string */
566 0 : register int len; /* length of current match */
567 0 : int best_len = s->prev_length; /* best match length so far */
568 0 : int nice_match = s->nice_match; /* stop if match long enough */
569 0 : IPos limit = s->strstart > (IPos)MAX_DIST(s) ?
570 0 : s->strstart - (IPos)MAX_DIST(s) : NIL;
571 : /* Stop when cur_match becomes <= limit. To simplify the code,
572 : * we prevent matches with the string of window index 0.
573 : */
574 0 : Pos *prev = s->prev;
575 0 : uInt wmask = s->w_mask;
576 :
577 : #ifdef UNALIGNED_OK
578 : /* Compare two bytes at a time. Note: this is not always beneficial.
579 : * Try with and without -DUNALIGNED_OK to check.
580 : */
581 : register Byte *strend = s->window + s->strstart + MAX_MATCH - 1;
582 : register ush scan_start = *(ush*)scan;
583 : register ush scan_end = *(ush*)(scan+best_len-1);
584 : #else
585 0 : register Byte *strend = s->window + s->strstart + MAX_MATCH;
586 0 : register Byte scan_end1 = scan[best_len-1];
587 0 : register Byte scan_end = scan[best_len];
588 : #endif
589 :
590 : /* The code is optimized for HASH_BITS >= 8 and MAX_MATCH-2 multiple of 16.
591 : * It is easy to get rid of this optimization if necessary.
592 : */
593 0 : Assert(s->hash_bits >= 8 && MAX_MATCH == 258, "Code too clever");
594 :
595 : /* Do not waste too much time if we already have a good match: */
596 0 : if (s->prev_length >= s->good_match) {
597 0 : chain_length >>= 2;
598 : }
599 : /* Do not look for matches beyond the end of the input. This is necessary
600 : * to make deflate deterministic.
601 : */
602 0 : if ((uInt)nice_match > s->lookahead) nice_match = s->lookahead;
603 :
604 0 : Assert((ulg)s->strstart <= s->window_size-MIN_LOOKAHEAD, "need lookahead");
605 :
606 0 : do {
607 0 : Assert(cur_match < s->strstart, "no future");
608 0 : match = s->window + cur_match;
609 :
610 : /* Skip to next match if the match length cannot increase
611 : * or if the match length is less than 2:
612 : */
613 : #if (defined(UNALIGNED_OK) && MAX_MATCH == 258)
614 : /* This code assumes sizeof(unsigned short) == 2. Do not use
615 : * UNALIGNED_OK if your compiler uses a different size.
616 : */
617 : if (*(ush*)(match+best_len-1) != scan_end ||
618 : *(ush*)match != scan_start) continue;
619 :
620 : /* It is not necessary to compare scan[2] and match[2] since they are
621 : * always equal when the other bytes match, given that the hash keys
622 : * are equal and that HASH_BITS >= 8. Compare 2 bytes at a time at
623 : * strstart+3, +5, ... up to strstart+257. We check for insufficient
624 : * lookahead only every 4th comparison; the 128th check will be made
625 : * at strstart+257. If MAX_MATCH-2 is not a multiple of 8, it is
626 : * necessary to put more guard bytes at the end of the window, or
627 : * to check more often for insufficient lookahead.
628 : */
629 : Assert(scan[2] == match[2], "scan[2]?");
630 : scan++, match++;
631 : do {
632 : } while (*(ush*)(scan+=2) == *(ush*)(match+=2) &&
633 : *(ush*)(scan+=2) == *(ush*)(match+=2) &&
634 : *(ush*)(scan+=2) == *(ush*)(match+=2) &&
635 : *(ush*)(scan+=2) == *(ush*)(match+=2) &&
636 : scan < strend);
637 : /* The funny "do {}" generates better code on most compilers */
638 :
639 : /* Here, scan <= window+strstart+257 */
640 : Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
641 : if (*scan == *match) scan++;
642 :
643 : len = (MAX_MATCH - 1) - (int)(strend-scan);
644 : scan = strend - (MAX_MATCH-1);
645 :
646 : #else /* UNALIGNED_OK */
647 :
648 0 : if (match[best_len] != scan_end ||
649 0 : match[best_len-1] != scan_end1 ||
650 0 : *match != *scan ||
651 0 : *++match != scan[1]) continue;
652 :
653 : /* The check at best_len-1 can be removed because it will be made
654 : * again later. (This heuristic is not always a win.)
655 : * It is not necessary to compare scan[2] and match[2] since they
656 : * are always equal when the other bytes match, given that
657 : * the hash keys are equal and that HASH_BITS >= 8.
658 : */
659 0 : scan += 2, match++;
660 0 : Assert(*scan == *match, "match[2]?");
661 :
662 : /* We check for insufficient lookahead only every 8th comparison;
663 : * the 256th check will be made at strstart+258.
664 : */
665 0 : do {
666 0 : } while (*++scan == *++match && *++scan == *++match &&
667 0 : *++scan == *++match && *++scan == *++match &&
668 0 : *++scan == *++match && *++scan == *++match &&
669 0 : *++scan == *++match && *++scan == *++match &&
670 0 : scan < strend);
671 :
672 0 : Assert(scan <= s->window+(unsigned)(s->window_size-1), "wild scan");
673 :
674 0 : len = MAX_MATCH - (int)(strend - scan);
675 0 : scan = strend - MAX_MATCH;
676 :
677 : #endif /* UNALIGNED_OK */
678 :
679 0 : if (len > best_len) {
680 0 : s->match_start = cur_match;
681 0 : best_len = len;
682 0 : if (len >= nice_match) break;
683 : #ifdef UNALIGNED_OK
684 : scan_end = *(ush*)(scan+best_len-1);
685 : #else
686 0 : scan_end1 = scan[best_len-1];
687 0 : scan_end = scan[best_len];
688 : #endif
689 : }
690 0 : } while ((cur_match = prev[cur_match & wmask]) > limit
691 0 : && --chain_length != 0);
692 :
693 0 : if ((uInt)best_len <= s->lookahead) return best_len;
694 : return s->lookahead;
695 : }
696 :
697 : #ifdef DEBUG_ZLIB
698 : /* ===========================================================================
699 : * Check that the match at match_start is indeed a match.
700 : */
701 : static void check_match(
702 : deflate_state *s,
703 : IPos start,
704 : IPos match,
705 : int length
706 : )
707 : {
708 : /* check that the match is indeed a match */
709 : if (memcmp((char *)s->window + match,
710 : (char *)s->window + start, length) != EQUAL) {
711 : fprintf(stderr, " start %u, match %u, length %d\n",
712 : start, match, length);
713 : do {
714 : fprintf(stderr, "%c%c", s->window[match++], s->window[start++]);
715 : } while (--length != 0);
716 : z_error("invalid match");
717 : }
718 : if (z_verbose > 1) {
719 : fprintf(stderr,"\\[%d,%d]", start-match, length);
720 : do { putc(s->window[start++], stderr); } while (--length != 0);
721 : }
722 : }
723 : #else
724 : # define check_match(s, start, match, length)
725 : #endif
726 :
727 : /* ===========================================================================
728 : * Fill the window when the lookahead becomes insufficient.
729 : * Updates strstart and lookahead.
730 : *
731 : * IN assertion: lookahead < MIN_LOOKAHEAD
732 : * OUT assertions: strstart <= window_size-MIN_LOOKAHEAD
733 : * At least one byte has been read, or avail_in == 0; reads are
734 : * performed for at least two bytes (required for the zip translate_eol
735 : * option -- not supported here).
736 : */
737 0 : static void fill_window(
738 : deflate_state *s
739 : )
740 : {
741 0 : register unsigned n, m;
742 0 : register Pos *p;
743 0 : unsigned more; /* Amount of free space at the end of the window. */
744 0 : uInt wsize = s->w_size;
745 :
746 0 : do {
747 0 : more = (unsigned)(s->window_size -(ulg)s->lookahead -(ulg)s->strstart);
748 :
749 : /* Deal with !@#$% 64K limit: */
750 0 : if (more == 0 && s->strstart == 0 && s->lookahead == 0) {
751 : more = wsize;
752 :
753 0 : } else if (more == (unsigned)(-1)) {
754 : /* Very unlikely, but possible on 16 bit machine if strstart == 0
755 : * and lookahead == 1 (input done one byte at time)
756 : */
757 : more--;
758 :
759 : /* If the window is almost full and there is insufficient lookahead,
760 : * move the upper half to the lower one to make room in the upper half.
761 : */
762 0 : } else if (s->strstart >= wsize+MAX_DIST(s)) {
763 :
764 0 : memcpy((char *)s->window, (char *)s->window+wsize,
765 : (unsigned)wsize);
766 0 : s->match_start -= wsize;
767 0 : s->strstart -= wsize; /* we now have strstart >= MAX_DIST */
768 0 : s->block_start -= (long) wsize;
769 :
770 : /* Slide the hash table (could be avoided with 32 bit values
771 : at the expense of memory usage). We slide even when level == 0
772 : to keep the hash table consistent if we switch back to level > 0
773 : later. (Using level 0 permanently is not an optimal usage of
774 : zlib, so we don't care about this pathological case.)
775 : */
776 0 : n = s->hash_size;
777 0 : p = &s->head[n];
778 0 : do {
779 0 : m = *--p;
780 0 : *p = (Pos)(m >= wsize ? m-wsize : NIL);
781 0 : } while (--n);
782 :
783 0 : n = wsize;
784 0 : p = &s->prev[n];
785 0 : do {
786 0 : m = *--p;
787 0 : *p = (Pos)(m >= wsize ? m-wsize : NIL);
788 : /* If n is not on any hash chain, prev[n] is garbage but
789 : * its value will never be used.
790 : */
791 0 : } while (--n);
792 0 : more += wsize;
793 : }
794 0 : if (s->strm->avail_in == 0) return;
795 :
796 : /* If there was no sliding:
797 : * strstart <= WSIZE+MAX_DIST-1 && lookahead <= MIN_LOOKAHEAD - 1 &&
798 : * more == window_size - lookahead - strstart
799 : * => more >= window_size - (MIN_LOOKAHEAD-1 + WSIZE + MAX_DIST-1)
800 : * => more >= window_size - 2*WSIZE + 2
801 : * In the BIG_MEM or MMAP case (not yet supported),
802 : * window_size == input_size + MIN_LOOKAHEAD &&
803 : * strstart + s->lookahead <= input_size => more >= MIN_LOOKAHEAD.
804 : * Otherwise, window_size == 2*WSIZE so more >= 2.
805 : * If there was sliding, more >= WSIZE. So in all cases, more >= 2.
806 : */
807 0 : Assert(more >= 2, "more < 2");
808 :
809 0 : n = read_buf(s->strm, s->window + s->strstart + s->lookahead, more);
810 0 : s->lookahead += n;
811 :
812 : /* Initialize the hash value now that we have some input: */
813 0 : if (s->lookahead >= MIN_MATCH) {
814 0 : s->ins_h = s->window[s->strstart];
815 0 : UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
816 : #if MIN_MATCH != 3
817 : Call UPDATE_HASH() MIN_MATCH-3 more times
818 : #endif
819 : }
820 : /* If the whole input has less than MIN_MATCH bytes, ins_h is garbage,
821 : * but this is not important since only literal bytes will be emitted.
822 : */
823 :
824 0 : } while (s->lookahead < MIN_LOOKAHEAD && s->strm->avail_in != 0);
825 : }
826 :
827 : /* ===========================================================================
828 : * Flush the current block, with given end-of-file flag.
829 : * IN assertion: strstart is set to the end of the current match.
830 : */
831 : #define FLUSH_BLOCK_ONLY(s, eof) { \
832 : zlib_tr_flush_block(s, (s->block_start >= 0L ? \
833 : (char *)&s->window[(unsigned)s->block_start] : \
834 : NULL), \
835 : (ulg)((long)s->strstart - s->block_start), \
836 : (eof)); \
837 : s->block_start = s->strstart; \
838 : flush_pending(s->strm); \
839 : Tracev((stderr,"[FLUSH]")); \
840 : }
841 :
842 : /* Same but force premature exit if necessary. */
843 : #define FLUSH_BLOCK(s, eof) { \
844 : FLUSH_BLOCK_ONLY(s, eof); \
845 : if (s->strm->avail_out == 0) return (eof) ? finish_started : need_more; \
846 : }
847 :
848 : /* ===========================================================================
849 : * Copy without compression as much as possible from the input stream, return
850 : * the current block state.
851 : * This function does not insert new strings in the dictionary since
852 : * uncompressible data is probably not useful. This function is used
853 : * only for the level=0 compression option.
854 : * NOTE: this function should be optimized to avoid extra copying from
855 : * window to pending_buf.
856 : */
857 0 : static block_state deflate_stored(
858 : deflate_state *s,
859 : int flush
860 : )
861 : {
862 : /* Stored blocks are limited to 0xffff bytes, pending_buf is limited
863 : * to pending_buf_size, and each stored block has a 5 byte header:
864 : */
865 0 : ulg max_block_size = 0xffff;
866 0 : ulg max_start;
867 :
868 0 : if (max_block_size > s->pending_buf_size - 5) {
869 0 : max_block_size = s->pending_buf_size - 5;
870 : }
871 :
872 : /* Copy as much as possible from input to output: */
873 0 : for (;;) {
874 : /* Fill the window as much as possible: */
875 0 : if (s->lookahead <= 1) {
876 :
877 : Assert(s->strstart < s->w_size+MAX_DIST(s) ||
878 0 : s->block_start >= (long)s->w_size, "slide too late");
879 :
880 0 : fill_window(s);
881 0 : if (s->lookahead == 0 && flush == Z_NO_FLUSH) return need_more;
882 :
883 0 : if (s->lookahead == 0) break; /* flush the current block */
884 : }
885 0 : Assert(s->block_start >= 0L, "block gone");
886 :
887 0 : s->strstart += s->lookahead;
888 0 : s->lookahead = 0;
889 :
890 : /* Emit a stored block if pending_buf will be full: */
891 0 : max_start = s->block_start + max_block_size;
892 0 : if (s->strstart == 0 || (ulg)s->strstart >= max_start) {
893 : /* strstart == 0 is possible when wraparound on 16-bit machine */
894 0 : s->lookahead = (uInt)(s->strstart - max_start);
895 0 : s->strstart = (uInt)max_start;
896 0 : FLUSH_BLOCK(s, 0);
897 : }
898 : /* Flush if we may have to slide, otherwise block_start may become
899 : * negative and the data will be gone:
900 : */
901 0 : if (s->strstart - (uInt)s->block_start >= MAX_DIST(s)) {
902 0 : FLUSH_BLOCK(s, 0);
903 : }
904 : }
905 0 : FLUSH_BLOCK(s, flush == Z_FINISH);
906 0 : return flush == Z_FINISH ? finish_done : block_done;
907 : }
908 :
909 : /* ===========================================================================
910 : * Compress as much as possible from the input stream, return the current
911 : * block state.
912 : * This function does not perform lazy evaluation of matches and inserts
913 : * new strings in the dictionary only for unmatched strings or for short
914 : * matches. It is used only for the fast compression options.
915 : */
916 0 : static block_state deflate_fast(
917 : deflate_state *s,
918 : int flush
919 : )
920 : {
921 0 : IPos hash_head = NIL; /* head of the hash chain */
922 0 : int bflush; /* set if current block must be flushed */
923 :
924 0 : for (;;) {
925 : /* Make sure that we always have enough lookahead, except
926 : * at the end of the input file. We need MAX_MATCH bytes
927 : * for the next match, plus MIN_MATCH bytes to insert the
928 : * string following the next match.
929 : */
930 0 : if (s->lookahead < MIN_LOOKAHEAD) {
931 0 : fill_window(s);
932 0 : if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
933 : return need_more;
934 : }
935 0 : if (s->lookahead == 0) break; /* flush the current block */
936 : }
937 :
938 : /* Insert the string window[strstart .. strstart+2] in the
939 : * dictionary, and set hash_head to the head of the hash chain:
940 : */
941 0 : if (s->lookahead >= MIN_MATCH) {
942 0 : INSERT_STRING(s, s->strstart, hash_head);
943 : }
944 :
945 : /* Find the longest match, discarding those <= prev_length.
946 : * At this point we have always match_length < MIN_MATCH
947 : */
948 0 : if (hash_head != NIL && s->strstart - hash_head <= MAX_DIST(s)) {
949 : /* To simplify the code, we prevent matches with the string
950 : * of window index 0 (in particular we have to avoid a match
951 : * of the string with itself at the start of the input file).
952 : */
953 0 : if (s->strategy != Z_HUFFMAN_ONLY) {
954 0 : s->match_length = longest_match (s, hash_head);
955 : }
956 : /* longest_match() sets match_start */
957 : }
958 0 : if (s->match_length >= MIN_MATCH) {
959 0 : check_match(s, s->strstart, s->match_start, s->match_length);
960 :
961 0 : bflush = zlib_tr_tally(s, s->strstart - s->match_start,
962 : s->match_length - MIN_MATCH);
963 :
964 0 : s->lookahead -= s->match_length;
965 :
966 : /* Insert new strings in the hash table only if the match length
967 : * is not too large. This saves time but degrades compression.
968 : */
969 0 : if (s->match_length <= s->max_insert_length &&
970 : s->lookahead >= MIN_MATCH) {
971 0 : s->match_length--; /* string at strstart already in hash table */
972 0 : do {
973 0 : s->strstart++;
974 0 : INSERT_STRING(s, s->strstart, hash_head);
975 : /* strstart never exceeds WSIZE-MAX_MATCH, so there are
976 : * always MIN_MATCH bytes ahead.
977 : */
978 0 : } while (--s->match_length != 0);
979 0 : s->strstart++;
980 : } else {
981 0 : s->strstart += s->match_length;
982 0 : s->match_length = 0;
983 0 : s->ins_h = s->window[s->strstart];
984 0 : UPDATE_HASH(s, s->ins_h, s->window[s->strstart+1]);
985 : #if MIN_MATCH != 3
986 : Call UPDATE_HASH() MIN_MATCH-3 more times
987 : #endif
988 : /* If lookahead < MIN_MATCH, ins_h is garbage, but it does not
989 : * matter since it will be recomputed at next deflate call.
990 : */
991 : }
992 : } else {
993 : /* No match, output a literal byte */
994 0 : Tracevv((stderr,"%c", s->window[s->strstart]));
995 0 : bflush = zlib_tr_tally (s, 0, s->window[s->strstart]);
996 0 : s->lookahead--;
997 0 : s->strstart++;
998 : }
999 0 : if (bflush) FLUSH_BLOCK(s, 0);
1000 : }
1001 0 : FLUSH_BLOCK(s, flush == Z_FINISH);
1002 0 : return flush == Z_FINISH ? finish_done : block_done;
1003 : }
1004 :
1005 : /* ===========================================================================
1006 : * Same as above, but achieves better compression. We use a lazy
1007 : * evaluation for matches: a match is finally adopted only if there is
1008 : * no better match at the next window position.
1009 : */
1010 0 : static block_state deflate_slow(
1011 : deflate_state *s,
1012 : int flush
1013 : )
1014 : {
1015 0 : IPos hash_head = NIL; /* head of hash chain */
1016 0 : int bflush; /* set if current block must be flushed */
1017 :
1018 : /* Process the input block. */
1019 0 : for (;;) {
1020 : /* Make sure that we always have enough lookahead, except
1021 : * at the end of the input file. We need MAX_MATCH bytes
1022 : * for the next match, plus MIN_MATCH bytes to insert the
1023 : * string following the next match.
1024 : */
1025 0 : if (s->lookahead < MIN_LOOKAHEAD) {
1026 0 : fill_window(s);
1027 0 : if (s->lookahead < MIN_LOOKAHEAD && flush == Z_NO_FLUSH) {
1028 : return need_more;
1029 : }
1030 0 : if (s->lookahead == 0) break; /* flush the current block */
1031 : }
1032 :
1033 : /* Insert the string window[strstart .. strstart+2] in the
1034 : * dictionary, and set hash_head to the head of the hash chain:
1035 : */
1036 0 : if (s->lookahead >= MIN_MATCH) {
1037 0 : INSERT_STRING(s, s->strstart, hash_head);
1038 : }
1039 :
1040 : /* Find the longest match, discarding those <= prev_length.
1041 : */
1042 0 : s->prev_length = s->match_length, s->prev_match = s->match_start;
1043 0 : s->match_length = MIN_MATCH-1;
1044 :
1045 0 : if (hash_head != NIL && s->prev_length < s->max_lazy_match &&
1046 0 : s->strstart - hash_head <= MAX_DIST(s)) {
1047 : /* To simplify the code, we prevent matches with the string
1048 : * of window index 0 (in particular we have to avoid a match
1049 : * of the string with itself at the start of the input file).
1050 : */
1051 0 : if (s->strategy != Z_HUFFMAN_ONLY) {
1052 0 : s->match_length = longest_match (s, hash_head);
1053 : }
1054 : /* longest_match() sets match_start */
1055 :
1056 0 : if (s->match_length <= 5 && (s->strategy == Z_FILTERED ||
1057 0 : (s->match_length == MIN_MATCH &&
1058 0 : s->strstart - s->match_start > TOO_FAR))) {
1059 :
1060 : /* If prev_match is also MIN_MATCH, match_start is garbage
1061 : * but we will ignore the current match anyway.
1062 : */
1063 0 : s->match_length = MIN_MATCH-1;
1064 : }
1065 : }
1066 : /* If there was a match at the previous step and the current
1067 : * match is not better, output the previous match:
1068 : */
1069 0 : if (s->prev_length >= MIN_MATCH && s->match_length <= s->prev_length) {
1070 0 : uInt max_insert = s->strstart + s->lookahead - MIN_MATCH;
1071 : /* Do not insert strings in hash table beyond this. */
1072 :
1073 0 : check_match(s, s->strstart-1, s->prev_match, s->prev_length);
1074 :
1075 0 : bflush = zlib_tr_tally(s, s->strstart -1 - s->prev_match,
1076 : s->prev_length - MIN_MATCH);
1077 :
1078 : /* Insert in hash table all strings up to the end of the match.
1079 : * strstart-1 and strstart are already inserted. If there is not
1080 : * enough lookahead, the last two strings are not inserted in
1081 : * the hash table.
1082 : */
1083 0 : s->lookahead -= s->prev_length-1;
1084 0 : s->prev_length -= 2;
1085 0 : do {
1086 0 : if (++s->strstart <= max_insert) {
1087 0 : INSERT_STRING(s, s->strstart, hash_head);
1088 : }
1089 0 : } while (--s->prev_length != 0);
1090 0 : s->match_available = 0;
1091 0 : s->match_length = MIN_MATCH-1;
1092 0 : s->strstart++;
1093 :
1094 0 : if (bflush) FLUSH_BLOCK(s, 0);
1095 :
1096 0 : } else if (s->match_available) {
1097 : /* If there was no match at the previous position, output a
1098 : * single literal. If there was a match but the current match
1099 : * is longer, truncate the previous match to a single literal.
1100 : */
1101 0 : Tracevv((stderr,"%c", s->window[s->strstart-1]));
1102 0 : if (zlib_tr_tally (s, 0, s->window[s->strstart-1])) {
1103 0 : FLUSH_BLOCK_ONLY(s, 0);
1104 : }
1105 0 : s->strstart++;
1106 0 : s->lookahead--;
1107 0 : if (s->strm->avail_out == 0) return need_more;
1108 : } else {
1109 : /* There is no previous match to compare with, wait for
1110 : * the next step to decide.
1111 : */
1112 0 : s->match_available = 1;
1113 0 : s->strstart++;
1114 0 : s->lookahead--;
1115 : }
1116 : }
1117 0 : Assert (flush != Z_NO_FLUSH, "no flush?");
1118 0 : if (s->match_available) {
1119 0 : Tracevv((stderr,"%c", s->window[s->strstart-1]));
1120 0 : zlib_tr_tally (s, 0, s->window[s->strstart-1]);
1121 0 : s->match_available = 0;
1122 : }
1123 0 : FLUSH_BLOCK(s, flush == Z_FINISH);
1124 0 : return flush == Z_FINISH ? finish_done : block_done;
1125 : }
1126 :
1127 0 : int zlib_deflate_workspacesize(int windowBits, int memLevel)
1128 : {
1129 0 : if (windowBits < 0) /* undocumented feature: suppress zlib header */
1130 : windowBits = -windowBits;
1131 :
1132 : /* Since the return value is typically passed to vmalloc() unchecked... */
1133 0 : BUG_ON(memLevel < 1 || memLevel > MAX_MEM_LEVEL || windowBits < 9 ||
1134 : windowBits > 15);
1135 :
1136 0 : return sizeof(deflate_workspace)
1137 0 : + zlib_deflate_window_memsize(windowBits)
1138 0 : + zlib_deflate_prev_memsize(windowBits)
1139 0 : + zlib_deflate_head_memsize(memLevel)
1140 0 : + zlib_deflate_overlay_memsize(memLevel);
1141 : }
1142 :
1143 0 : int zlib_deflate_dfltcc_enabled(void)
1144 : {
1145 0 : return DEFLATE_DFLTCC_ENABLED();
1146 : }
|
