1// SPDX-License-Identifier: GPL-2.0
2#define DEBG(x)
3#define DEBG1(x)
4/* inflate.c -- Not copyrighted 1992 by Mark Adler
5 version c10p1, 10 January 1993 */
6
7/*
8 * Adapted for booting Linux by Hannu Savolainen 1993
9 * based on gzip-1.0.3
10 *
11 * Nicolas Pitre <nico@fluxnic.net>, 1999/04/14 :
12 * Little mods for all variable to reside either into rodata or bss segments
13 * by marking constant variables with 'const' and initializing all the others
14 * at run-time only. This allows for the kernel uncompressor to run
15 * directly from Flash or ROM memory on embedded systems.
16 */
17
18/*
19 Inflate deflated (PKZIP's method 8 compressed) data. The compression
20 method searches for as much of the current string of bytes (up to a
21 length of 258) in the previous 32 K bytes. If it doesn't find any
22 matches (of at least length 3), it codes the next byte. Otherwise, it
23 codes the length of the matched string and its distance backwards from
24 the current position. There is a single Huffman code that codes both
25 single bytes (called "literals") and match lengths. A second Huffman
26 code codes the distance information, which follows a length code. Each
27 length or distance code actually represents a base value and a number
28 of "extra" (sometimes zero) bits to get to add to the base value. At
29 the end of each deflated block is a special end-of-block (EOB) literal/
30 length code. The decoding process is basically: get a literal/length
31 code; if EOB then done; if a literal, emit the decoded byte; if a
32 length then get the distance and emit the referred-to bytes from the
33 sliding window of previously emitted data.
34
35 There are (currently) three kinds of inflate blocks: stored, fixed, and
36 dynamic. The compressor deals with some chunk of data at a time, and
37 decides which method to use on a chunk-by-chunk basis. A chunk might
38 typically be 32 K or 64 K. If the chunk is incompressible, then the
39 "stored" method is used. In this case, the bytes are simply stored as
40 is, eight bits per byte, with none of the above coding. The bytes are
41 preceded by a count, since there is no longer an EOB code.
42
43 If the data is compressible, then either the fixed or dynamic methods
44 are used. In the dynamic method, the compressed data is preceded by
45 an encoding of the literal/length and distance Huffman codes that are
46 to be used to decode this block. The representation is itself Huffman
47 coded, and so is preceded by a description of that code. These code
48 descriptions take up a little space, and so for small blocks, there is
49 a predefined set of codes, called the fixed codes. The fixed method is
50 used if the block codes up smaller that way (usually for quite small
51 chunks), otherwise the dynamic method is used. In the latter case, the
52 codes are customized to the probabilities in the current block, and so
53 can code it much better than the pre-determined fixed codes.
54
55 The Huffman codes themselves are decoded using a multi-level table
56 lookup, in order to maximize the speed of decoding plus the speed of
57 building the decoding tables. See the comments below that precede the
58 lbits and dbits tuning parameters.
59 */
60
61
62/*
63 Notes beyond the 1.93a appnote.txt:
64
65 1. Distance pointers never point before the beginning of the output
66 stream.
67 2. Distance pointers can point back across blocks, up to 32k away.
68 3. There is an implied maximum of 7 bits for the bit length table and
69 15 bits for the actual data.
70 4. If only one code exists, then it is encoded using one bit. (Zero
71 would be more efficient, but perhaps a little confusing.) If two
72 codes exist, they are coded using one bit each (0 and 1).
73 5. There is no way of sending zero distance codes--a dummy must be
74 sent if there are none. (History: a pre 2.0 version of PKZIP would
75 store blocks with no distance codes, but this was discovered to be
76 too harsh a criterion.) Valid only for 1.93a. 2.04c does allow
77 zero distance codes, which is sent as one code of zero bits in
78 length.
79 6. There are up to 286 literal/length codes. Code 256 represents the
80 end-of-block. Note however that the static length tree defines
81 288 codes just to fill out the Huffman codes. Codes 286 and 287
82 cannot be used though, since there is no length base or extra bits
83 defined for them. Similarly, there are up to 30 distance codes.
84 However, static trees define 32 codes (all 5 bits) to fill out the
85 Huffman codes, but the last two had better not show up in the data.
86 7. Unzip can check dynamic Huffman blocks for complete code sets.
87 The exception is that a single code would not be complete (see #4).
88 8. The five bits following the block type is really the number of
89 literal codes sent minus 257.
90 9. Length codes 8,16,16 are interpreted as 13 length codes of 8 bits
91 (1+6+6). Therefore, to output three times the length, you output
92 three codes (1+1+1), whereas to output four times the same length,
93 you only need two codes (1+3). Hmm.
94 10. In the tree reconstruction algorithm, Code = Code + Increment
95 only if BitLength(i) is not zero. (Pretty obvious.)
96 11. Correction: 4 Bits: # of Bit Length codes - 4 (4 - 19)
97 12. Note: length code 284 can represent 227-258, but length code 285
98 really is 258. The last length deserves its own, short code
99 since it gets used a lot in very redundant files. The length
100 258 is special since 258 - 3 (the min match length) is 255.
101 13. The literal/length and distance code bit lengths are read as a
102 single stream of lengths. It is possible (and advantageous) for
103 a repeat code (16, 17, or 18) to go across the boundary between
104 the two sets of lengths.
105 */
106#include <linux/compiler.h>
107#ifdef NO_INFLATE_MALLOC
108#include <linux/slab.h>
109#endif
110
111#ifdef RCSID
112static char rcsid[] = "#Id: inflate.c,v 0.14 1993/06/10 13:27:04 jloup Exp #";
113#endif
114
115#ifndef STATIC
116
117#if defined(STDC_HEADERS) || defined(HAVE_STDLIB_H)
118# include <sys/types.h>
119# include <stdlib.h>
120#endif
121
122#include "gzip.h"
123#define STATIC
124#endif /* !STATIC */
125
126#ifndef INIT
127#define INIT
128#endif
129
130#define slide window
131
132/* Huffman code lookup table entry--this entry is four bytes for machines
133 that have 16-bit pointers (e.g. PC's in the small or medium model).
134 Valid extra bits are 0..13. e == 15 is EOB (end of block), e == 16
135 means that v is a literal, 16 < e < 32 means that v is a pointer to
136 the next table, which codes e - 16 bits, and lastly e == 99 indicates
137 an unused code. If a code with e == 99 is looked up, this implies an
138 error in the data. */
139struct huft {
140 uch e; /* number of extra bits or operation */
141 uch b; /* number of bits in this code or subcode */
142 union {
143 ush n; /* literal, length base, or distance base */
144 struct huft *t; /* pointer to next level of table */
145 } v;
146};
147
148
149/* Function prototypes */
150STATIC int INIT huft_build OF((unsigned *, unsigned, unsigned,
151 const ush *, const ush *, struct huft **, int *));
152STATIC int INIT huft_free OF((struct huft *));
153STATIC int INIT inflate_codes OF((struct huft *, struct huft *, int, int));
154STATIC int INIT inflate_stored OF((void));
155STATIC int INIT inflate_fixed OF((void));
156STATIC int INIT inflate_dynamic OF((void));
157STATIC int INIT inflate_block OF((int *));
158STATIC int INIT inflate OF((void));
159
160
161/* The inflate algorithm uses a sliding 32 K byte window on the uncompressed
162 stream to find repeated byte strings. This is implemented here as a
163 circular buffer. The index is updated simply by incrementing and then
164 ANDing with 0x7fff (32K-1). */
165/* It is left to other modules to supply the 32 K area. It is assumed
166 to be usable as if it were declared "uch slide[32768];" or as just
167 "uch *slide;" and then malloc'ed in the latter case. The definition
168 must be in unzip.h, included above. */
169/* unsigned wp; current position in slide */
170#define wp outcnt
171#define flush_output(w) (wp=(w),flush_window())
172
173/* Tables for deflate from PKZIP's appnote.txt. */
174static const unsigned border[] = { /* Order of the bit length code lengths */
175 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15};
176static const ush cplens[] = { /* Copy lengths for literal codes 257..285 */
177 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
178 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0};
179 /* note: see note #13 above about the 258 in this list. */
180static const ush cplext[] = { /* Extra bits for literal codes 257..285 */
181 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 2, 2, 2, 2,
182 3, 3, 3, 3, 4, 4, 4, 4, 5, 5, 5, 5, 0, 99, 99}; /* 99==invalid */
183static const ush cpdist[] = { /* Copy offsets for distance codes 0..29 */
184 1, 2, 3, 4, 5, 7, 9, 13, 17, 25, 33, 49, 65, 97, 129, 193,
185 257, 385, 513, 769, 1025, 1537, 2049, 3073, 4097, 6145,
186 8193, 12289, 16385, 24577};
187static const ush cpdext[] = { /* Extra bits for distance codes */
188 0, 0, 0, 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6,
189 7, 7, 8, 8, 9, 9, 10, 10, 11, 11,
190 12, 12, 13, 13};
191
192
193
194/* Macros for inflate() bit peeking and grabbing.
195 The usage is:
196
197 NEEDBITS(j)
198 x = b & mask_bits[j];
199 DUMPBITS(j)
200
201 where NEEDBITS makes sure that b has at least j bits in it, and
202 DUMPBITS removes the bits from b. The macros use the variable k
203 for the number of bits in b. Normally, b and k are register
204 variables for speed, and are initialized at the beginning of a
205 routine that uses these macros from a global bit buffer and count.
206
207 If we assume that EOB will be the longest code, then we will never
208 ask for bits with NEEDBITS that are beyond the end of the stream.
209 So, NEEDBITS should not read any more bytes than are needed to
210 meet the request. Then no bytes need to be "returned" to the buffer
211 at the end of the last block.
212
213 However, this assumption is not true for fixed blocks--the EOB code
214 is 7 bits, but the other literal/length codes can be 8 or 9 bits.
215 (The EOB code is shorter than other codes because fixed blocks are
216 generally short. So, while a block always has an EOB, many other
217 literal/length codes have a significantly lower probability of
218 showing up at all.) However, by making the first table have a
219 lookup of seven bits, the EOB code will be found in that first
220 lookup, and so will not require that too many bits be pulled from
221 the stream.
222 */
223
224STATIC ulg bb; /* bit buffer */
225STATIC unsigned bk; /* bits in bit buffer */
226
227STATIC const ush mask_bits[] = {
228 0x0000,
229 0x0001, 0x0003, 0x0007, 0x000f, 0x001f, 0x003f, 0x007f, 0x00ff,
230 0x01ff, 0x03ff, 0x07ff, 0x0fff, 0x1fff, 0x3fff, 0x7fff, 0xffff
231};
232
233#define NEXTBYTE() ({ int v = get_byte(); if (v < 0) goto underrun; (uch)v; })
234#define NEEDBITS(n) {while(k<(n)){b|=((ulg)NEXTBYTE())<<k;k+=8;}}
235#define DUMPBITS(n) {b>>=(n);k-=(n);}
236
237#ifndef NO_INFLATE_MALLOC
238/* A trivial malloc implementation, adapted from
239 * malloc by Hannu Savolainen 1993 and Matthias Urlichs 1994
240 */
241
242static unsigned long malloc_ptr;
243static int malloc_count;
244
245static void *malloc(int size)
246{
247 void *p;
248
249 if (size < 0)
250 error(m: "Malloc error");
251 if (!malloc_ptr)
252 malloc_ptr = free_mem_ptr;
253
254 malloc_ptr = (malloc_ptr + 3) & ~3; /* Align */
255
256 p = (void *)malloc_ptr;
257 malloc_ptr += size;
258
259 if (free_mem_end_ptr && malloc_ptr >= free_mem_end_ptr)
260 error(m: "Out of memory");
261
262 malloc_count++;
263 return p;
264}
265
266static void free(void *where)
267{
268 malloc_count--;
269 if (!malloc_count)
270 malloc_ptr = free_mem_ptr;
271}
272#else
273#define malloc(a) kmalloc(a, GFP_KERNEL)
274#define free(a) kfree(a)
275#endif
276
277/*
278 Huffman code decoding is performed using a multi-level table lookup.
279 The fastest way to decode is to simply build a lookup table whose
280 size is determined by the longest code. However, the time it takes
281 to build this table can also be a factor if the data being decoded
282 is not very long. The most common codes are necessarily the
283 shortest codes, so those codes dominate the decoding time, and hence
284 the speed. The idea is you can have a shorter table that decodes the
285 shorter, more probable codes, and then point to subsidiary tables for
286 the longer codes. The time it costs to decode the longer codes is
287 then traded against the time it takes to make longer tables.
288
289 This results of this trade are in the variables lbits and dbits
290 below. lbits is the number of bits the first level table for literal/
291 length codes can decode in one step, and dbits is the same thing for
292 the distance codes. Subsequent tables are also less than or equal to
293 those sizes. These values may be adjusted either when all of the
294 codes are shorter than that, in which case the longest code length in
295 bits is used, or when the shortest code is *longer* than the requested
296 table size, in which case the length of the shortest code in bits is
297 used.
298
299 There are two different values for the two tables, since they code a
300 different number of possibilities each. The literal/length table
301 codes 286 possible values, or in a flat code, a little over eight
302 bits. The distance table codes 30 possible values, or a little less
303 than five bits, flat. The optimum values for speed end up being
304 about one bit more than those, so lbits is 8+1 and dbits is 5+1.
305 The optimum values may differ though from machine to machine, and
306 possibly even between compilers. Your mileage may vary.
307 */
308
309
310STATIC const int lbits = 9; /* bits in base literal/length lookup table */
311STATIC const int dbits = 6; /* bits in base distance lookup table */
312
313
314/* If BMAX needs to be larger than 16, then h and x[] should be ulg. */
315#define BMAX 16 /* maximum bit length of any code (16 for explode) */
316#define N_MAX 288 /* maximum number of codes in any set */
317
318
319STATIC unsigned hufts; /* track memory usage */
320
321
322STATIC int INIT huft_build(
323 unsigned *b, /* code lengths in bits (all assumed <= BMAX) */
324 unsigned n, /* number of codes (assumed <= N_MAX) */
325 unsigned s, /* number of simple-valued codes (0..s-1) */
326 const ush *d, /* list of base values for non-simple codes */
327 const ush *e, /* list of extra bits for non-simple codes */
328 struct huft **t, /* result: starting table */
329 int *m /* maximum lookup bits, returns actual */
330 )
331/* Given a list of code lengths and a maximum table size, make a set of
332 tables to decode that set of codes. Return zero on success, one if
333 the given code set is incomplete (the tables are still built in this
334 case), two if the input is invalid (all zero length codes or an
335 oversubscribed set of lengths), and three if not enough memory. */
336{
337 unsigned a; /* counter for codes of length k */
338 unsigned f; /* i repeats in table every f entries */
339 int g; /* maximum code length */
340 int h; /* table level */
341 register unsigned i; /* counter, current code */
342 register unsigned j; /* counter */
343 register int k; /* number of bits in current code */
344 int l; /* bits per table (returned in m) */
345 register unsigned *p; /* pointer into c[], b[], or v[] */
346 register struct huft *q; /* points to current table */
347 struct huft r; /* table entry for structure assignment */
348 register int w; /* bits before this table == (l * h) */
349 unsigned *xp; /* pointer into x */
350 int y; /* number of dummy codes added */
351 unsigned z; /* number of entries in current table */
352 struct {
353 unsigned c[BMAX+1]; /* bit length count table */
354 struct huft *u[BMAX]; /* table stack */
355 unsigned v[N_MAX]; /* values in order of bit length */
356 unsigned x[BMAX+1]; /* bit offsets, then code stack */
357 } *stk;
358 unsigned *c, *v, *x;
359 struct huft **u;
360 int ret;
361
362DEBG("huft1 ");
363
364 stk = malloc(size: sizeof(*stk));
365 if (stk == NULL)
366 return 3; /* out of memory */
367
368 c = stk->c;
369 v = stk->v;
370 x = stk->x;
371 u = stk->u;
372
373 /* Generate counts for each bit length */
374 memzero(stk->c, sizeof(stk->c));
375 p = b; i = n;
376 do {
377 Tracecv(*p, (stderr, (n-i >= ' ' && n-i <= '~' ? "%c %d\n" : "0x%x %d\n"),
378 n-i, *p));
379 c[*p]++; /* assume all entries <= BMAX */
380 p++; /* Can't combine with above line (Solaris bug) */
381 } while (--i);
382 if (c[0] == n) /* null input--all zero length codes */
383 {
384 *t = (struct huft *)NULL;
385 *m = 0;
386 ret = 2;
387 goto out;
388 }
389
390DEBG("huft2 ");
391
392 /* Find minimum and maximum length, bound *m by those */
393 l = *m;
394 for (j = 1; j <= BMAX; j++)
395 if (c[j])
396 break;
397 k = j; /* minimum code length */
398 if ((unsigned)l < j)
399 l = j;
400 for (i = BMAX; i; i--)
401 if (c[i])
402 break;
403 g = i; /* maximum code length */
404 if ((unsigned)l > i)
405 l = i;
406 *m = l;
407
408DEBG("huft3 ");
409
410 /* Adjust last length count to fill out codes, if needed */
411 for (y = 1 << j; j < i; j++, y <<= 1)
412 if ((y -= c[j]) < 0) {
413 ret = 2; /* bad input: more codes than bits */
414 goto out;
415 }
416 if ((y -= c[i]) < 0) {
417 ret = 2;
418 goto out;
419 }
420 c[i] += y;
421
422DEBG("huft4 ");
423
424 /* Generate starting offsets into the value table for each length */
425 x[1] = j = 0;
426 p = c + 1; xp = x + 2;
427 while (--i) { /* note that i == g from above */
428 *xp++ = (j += *p++);
429 }
430
431DEBG("huft5 ");
432
433 /* Make a table of values in order of bit lengths */
434 p = b; i = 0;
435 do {
436 if ((j = *p++) != 0)
437 v[x[j]++] = i;
438 } while (++i < n);
439 n = x[g]; /* set n to length of v */
440
441DEBG("h6 ");
442
443 /* Generate the Huffman codes and for each, make the table entries */
444 x[0] = i = 0; /* first Huffman code is zero */
445 p = v; /* grab values in bit order */
446 h = -1; /* no tables yet--level -1 */
447 w = -l; /* bits decoded == (l * h) */
448 u[0] = (struct huft *)NULL; /* just to keep compilers happy */
449 q = (struct huft *)NULL; /* ditto */
450 z = 0; /* ditto */
451DEBG("h6a ");
452
453 /* go through the bit lengths (k already is bits in shortest code) */
454 for (; k <= g; k++)
455 {
456DEBG("h6b ");
457 a = c[k];
458 while (a--)
459 {
460DEBG("h6b1 ");
461 /* here i is the Huffman code of length k bits for value *p */
462 /* make tables up to required level */
463 while (k > w + l)
464 {
465DEBG1("1 ");
466 h++;
467 w += l; /* previous table always l bits */
468
469 /* compute minimum size table less than or equal to l bits */
470 z = (z = g - w) > (unsigned)l ? l : z; /* upper limit on table size */
471 if ((f = 1 << (j = k - w)) > a + 1) /* try a k-w bit table */
472 { /* too few codes for k-w bit table */
473DEBG1("2 ");
474 f -= a + 1; /* deduct codes from patterns left */
475 xp = c + k;
476 if (j < z)
477 while (++j < z) /* try smaller tables up to z bits */
478 {
479 if ((f <<= 1) <= *++xp)
480 break; /* enough codes to use up j bits */
481 f -= *xp; /* else deduct codes from patterns */
482 }
483 }
484DEBG1("3 ");
485 z = 1 << j; /* table entries for j-bit table */
486
487 /* allocate and link in new table */
488 if ((q = (struct huft *)malloc(size: (z + 1)*sizeof(struct huft))) ==
489 (struct huft *)NULL)
490 {
491 if (h)
492 huft_free(u[0]);
493 ret = 3; /* not enough memory */
494 goto out;
495 }
496DEBG1("4 ");
497 hufts += z + 1; /* track memory usage */
498 *t = q + 1; /* link to list for huft_free() */
499 *(t = &(q->v.t)) = (struct huft *)NULL;
500 u[h] = ++q; /* table starts after link */
501
502DEBG1("5 ");
503 /* connect to last table, if there is one */
504 if (h)
505 {
506 x[h] = i; /* save pattern for backing up */
507 r.b = (uch)l; /* bits to dump before this table */
508 r.e = (uch)(16 + j); /* bits in this table */
509 r.v.t = q; /* pointer to this table */
510 j = i >> (w - l); /* (get around Turbo C bug) */
511 u[h-1][j] = r; /* connect to last table */
512 }
513DEBG1("6 ");
514 }
515DEBG("h6c ");
516
517 /* set up table entry in r */
518 r.b = (uch)(k - w);
519 if (p >= v + n)
520 r.e = 99; /* out of values--invalid code */
521 else if (*p < s)
522 {
523 r.e = (uch)(*p < 256 ? 16 : 15); /* 256 is end-of-block code */
524 r.v.n = (ush)(*p); /* simple code is just the value */
525 p++; /* one compiler does not like *p++ */
526 }
527 else
528 {
529 r.e = (uch)e[*p - s]; /* non-simple--look up in lists */
530 r.v.n = d[*p++ - s];
531 }
532DEBG("h6d ");
533
534 /* fill code-like entries with r */
535 f = 1 << (k - w);
536 for (j = i >> w; j < z; j += f)
537 q[j] = r;
538
539 /* backwards increment the k-bit code i */
540 for (j = 1 << (k - 1); i & j; j >>= 1)
541 i ^= j;
542 i ^= j;
543
544 /* backup over finished tables */
545 while ((i & ((1 << w) - 1)) != x[h])
546 {
547 h--; /* don't need to update q */
548 w -= l;
549 }
550DEBG("h6e ");
551 }
552DEBG("h6f ");
553 }
554
555DEBG("huft7 ");
556
557 /* Return true (1) if we were given an incomplete table */
558 ret = y != 0 && g != 1;
559
560 out:
561 free(where: stk);
562 return ret;
563}
564
565
566
567STATIC int INIT huft_free(
568 struct huft *t /* table to free */
569 )
570/* Free the malloc'ed tables built by huft_build(), which makes a linked
571 list of the tables it made, with the links in a dummy first entry of
572 each table. */
573{
574 register struct huft *p, *q;
575
576
577 /* Go through linked list, freeing from the malloced (t[-1]) address. */
578 p = t;
579 while (p != (struct huft *)NULL)
580 {
581 q = (--p)->v.t;
582 free(where: (char*)p);
583 p = q;
584 }
585 return 0;
586}
587
588
589STATIC int INIT inflate_codes(
590 struct huft *tl, /* literal/length decoder tables */
591 struct huft *td, /* distance decoder tables */
592 int bl, /* number of bits decoded by tl[] */
593 int bd /* number of bits decoded by td[] */
594 )
595/* inflate (decompress) the codes in a deflated (compressed) block.
596 Return an error code or zero if it all goes ok. */
597{
598 register unsigned e; /* table entry flag/number of extra bits */
599 unsigned n, d; /* length and index for copy */
600 unsigned w; /* current window position */
601 struct huft *t; /* pointer to table entry */
602 unsigned ml, md; /* masks for bl and bd bits */
603 register ulg b; /* bit buffer */
604 register unsigned k; /* number of bits in bit buffer */
605
606
607 /* make local copies of globals */
608 b = bb; /* initialize bit buffer */
609 k = bk;
610 w = wp; /* initialize window position */
611
612 /* inflate the coded data */
613 ml = mask_bits[bl]; /* precompute masks for speed */
614 md = mask_bits[bd];
615 for (;;) /* do until end of block */
616 {
617 NEEDBITS((unsigned)bl)
618 if ((e = (t = tl + ((unsigned)b & ml))->e) > 16)
619 do {
620 if (e == 99)
621 return 1;
622 DUMPBITS(t->b)
623 e -= 16;
624 NEEDBITS(e)
625 } while ((e = (t = t->v.t + ((unsigned)b & mask_bits[e]))->e) > 16);
626 DUMPBITS(t->b)
627 if (e == 16) /* then it's a literal */
628 {
629 slide[w++] = (uch)t->v.n;
630 Tracevv((stderr, "%c", slide[w-1]));
631 if (w == WSIZE)
632 {
633 flush_output(w);
634 w = 0;
635 }
636 }
637 else /* it's an EOB or a length */
638 {
639 /* exit if end of block */
640 if (e == 15)
641 break;
642
643 /* get length of block to copy */
644 NEEDBITS(e)
645 n = t->v.n + ((unsigned)b & mask_bits[e]);
646 DUMPBITS(e);
647
648 /* decode distance of block to copy */
649 NEEDBITS((unsigned)bd)
650 if ((e = (t = td + ((unsigned)b & md))->e) > 16)
651 do {
652 if (e == 99)
653 return 1;
654 DUMPBITS(t->b)
655 e -= 16;
656 NEEDBITS(e)
657 } while ((e = (t = t->v.t + ((unsigned)b & mask_bits[e]))->e) > 16);
658 DUMPBITS(t->b)
659 NEEDBITS(e)
660 d = w - t->v.n - ((unsigned)b & mask_bits[e]);
661 DUMPBITS(e)
662 Tracevv((stderr,"\\[%d,%d]", w-d, n));
663
664 /* do the copy */
665 do {
666 n -= (e = (e = WSIZE - ((d &= WSIZE-1) > w ? d : w)) > n ? n : e);
667#if !defined(NOMEMCPY) && !defined(DEBUG)
668 if (w - d >= e) /* (this test assumes unsigned comparison) */
669 {
670 memcpy(slide + w, slide + d, e);
671 w += e;
672 d += e;
673 }
674 else /* do it slow to avoid memcpy() overlap */
675#endif /* !NOMEMCPY */
676 do {
677 slide[w++] = slide[d++];
678 Tracevv((stderr, "%c", slide[w-1]));
679 } while (--e);
680 if (w == WSIZE)
681 {
682 flush_output(w);
683 w = 0;
684 }
685 } while (n);
686 }
687 }
688
689
690 /* restore the globals from the locals */
691 wp = w; /* restore global window pointer */
692 bb = b; /* restore global bit buffer */
693 bk = k;
694
695 /* done */
696 return 0;
697
698 underrun:
699 return 4; /* Input underrun */
700}
701
702
703
704STATIC int INIT inflate_stored(void)
705/* "decompress" an inflated type 0 (stored) block. */
706{
707 unsigned n; /* number of bytes in block */
708 unsigned w; /* current window position */
709 register ulg b; /* bit buffer */
710 register unsigned k; /* number of bits in bit buffer */
711
712DEBG("<stor");
713
714 /* make local copies of globals */
715 b = bb; /* initialize bit buffer */
716 k = bk;
717 w = wp; /* initialize window position */
718
719
720 /* go to byte boundary */
721 n = k & 7;
722 DUMPBITS(n);
723
724
725 /* get the length and its complement */
726 NEEDBITS(16)
727 n = ((unsigned)b & 0xffff);
728 DUMPBITS(16)
729 NEEDBITS(16)
730 if (n != (unsigned)((~b) & 0xffff))
731 return 1; /* error in compressed data */
732 DUMPBITS(16)
733
734
735 /* read and output the compressed data */
736 while (n--)
737 {
738 NEEDBITS(8)
739 slide[w++] = (uch)b;
740 if (w == WSIZE)
741 {
742 flush_output(w);
743 w = 0;
744 }
745 DUMPBITS(8)
746 }
747
748
749 /* restore the globals from the locals */
750 wp = w; /* restore global window pointer */
751 bb = b; /* restore global bit buffer */
752 bk = k;
753
754 DEBG(">");
755 return 0;
756
757 underrun:
758 return 4; /* Input underrun */
759}
760
761
762/*
763 * We use `noinline' here to prevent gcc-3.5 from using too much stack space
764 */
765STATIC int noinline INIT inflate_fixed(void)
766/* decompress an inflated type 1 (fixed Huffman codes) block. We should
767 either replace this with a custom decoder, or at least precompute the
768 Huffman tables. */
769{
770 int i; /* temporary variable */
771 struct huft *tl; /* literal/length code table */
772 struct huft *td; /* distance code table */
773 int bl; /* lookup bits for tl */
774 int bd; /* lookup bits for td */
775 unsigned *l; /* length list for huft_build */
776
777DEBG("<fix");
778
779 l = malloc(size: sizeof(*l) * 288);
780 if (l == NULL)
781 return 3; /* out of memory */
782
783 /* set up literal table */
784 for (i = 0; i < 144; i++)
785 l[i] = 8;
786 for (; i < 256; i++)
787 l[i] = 9;
788 for (; i < 280; i++)
789 l[i] = 7;
790 for (; i < 288; i++) /* make a complete, but wrong code set */
791 l[i] = 8;
792 bl = 7;
793 if ((i = huft_build(b: l, n: 288, s: 257, d: cplens, e: cplext, t: &tl, m: &bl)) != 0) {
794 free(where: l);
795 return i;
796 }
797
798 /* set up distance table */
799 for (i = 0; i < 30; i++) /* make an incomplete code set */
800 l[i] = 5;
801 bd = 5;
802 if ((i = huft_build(b: l, n: 30, s: 0, d: cpdist, e: cpdext, t: &td, m: &bd)) > 1)
803 {
804 huft_free(t: tl);
805 free(where: l);
806
807 DEBG(">");
808 return i;
809 }
810
811
812 /* decompress until an end-of-block code */
813 if (inflate_codes(tl, td, bl, bd)) {
814 free(where: l);
815 return 1;
816 }
817
818 /* free the decoding tables, return */
819 free(where: l);
820 huft_free(t: tl);
821 huft_free(t: td);
822 return 0;
823}
824
825
826/*
827 * We use `noinline' here to prevent gcc-3.5 from using too much stack space
828 */
829STATIC int noinline INIT inflate_dynamic(void)
830/* decompress an inflated type 2 (dynamic Huffman codes) block. */
831{
832 int i; /* temporary variables */
833 unsigned j;
834 unsigned l; /* last length */
835 unsigned m; /* mask for bit lengths table */
836 unsigned n; /* number of lengths to get */
837 struct huft *tl; /* literal/length code table */
838 struct huft *td; /* distance code table */
839 int bl; /* lookup bits for tl */
840 int bd; /* lookup bits for td */
841 unsigned nb; /* number of bit length codes */
842 unsigned nl; /* number of literal/length codes */
843 unsigned nd; /* number of distance codes */
844 unsigned *ll; /* literal/length and distance code lengths */
845 register ulg b; /* bit buffer */
846 register unsigned k; /* number of bits in bit buffer */
847 int ret;
848
849DEBG("<dyn");
850
851#ifdef PKZIP_BUG_WORKAROUND
852 ll = malloc(sizeof(*ll) * (288+32)); /* literal/length and distance code lengths */
853#else
854 ll = malloc(size: sizeof(*ll) * (286+30)); /* literal/length and distance code lengths */
855#endif
856
857 if (ll == NULL)
858 return 1;
859
860 /* make local bit buffer */
861 b = bb;
862 k = bk;
863
864
865 /* read in table lengths */
866 NEEDBITS(5)
867 nl = 257 + ((unsigned)b & 0x1f); /* number of literal/length codes */
868 DUMPBITS(5)
869 NEEDBITS(5)
870 nd = 1 + ((unsigned)b & 0x1f); /* number of distance codes */
871 DUMPBITS(5)
872 NEEDBITS(4)
873 nb = 4 + ((unsigned)b & 0xf); /* number of bit length codes */
874 DUMPBITS(4)
875#ifdef PKZIP_BUG_WORKAROUND
876 if (nl > 288 || nd > 32)
877#else
878 if (nl > 286 || nd > 30)
879#endif
880 {
881 ret = 1; /* bad lengths */
882 goto out;
883 }
884
885DEBG("dyn1 ");
886
887 /* read in bit-length-code lengths */
888 for (j = 0; j < nb; j++)
889 {
890 NEEDBITS(3)
891 ll[border[j]] = (unsigned)b & 7;
892 DUMPBITS(3)
893 }
894 for (; j < 19; j++)
895 ll[border[j]] = 0;
896
897DEBG("dyn2 ");
898
899 /* build decoding table for trees--single level, 7 bit lookup */
900 bl = 7;
901 if ((i = huft_build(b: ll, n: 19, s: 19, NULL, NULL, t: &tl, m: &bl)) != 0)
902 {
903 if (i == 1)
904 huft_free(t: tl);
905 ret = i; /* incomplete code set */
906 goto out;
907 }
908
909DEBG("dyn3 ");
910
911 /* read in literal and distance code lengths */
912 n = nl + nd;
913 m = mask_bits[bl];
914 i = l = 0;
915 while ((unsigned)i < n)
916 {
917 NEEDBITS((unsigned)bl)
918 j = (td = tl + ((unsigned)b & m))->b;
919 DUMPBITS(j)
920 j = td->v.n;
921 if (j < 16) /* length of code in bits (0..15) */
922 ll[i++] = l = j; /* save last length in l */
923 else if (j == 16) /* repeat last length 3 to 6 times */
924 {
925 NEEDBITS(2)
926 j = 3 + ((unsigned)b & 3);
927 DUMPBITS(2)
928 if ((unsigned)i + j > n) {
929 ret = 1;
930 goto out;
931 }
932 while (j--)
933 ll[i++] = l;
934 }
935 else if (j == 17) /* 3 to 10 zero length codes */
936 {
937 NEEDBITS(3)
938 j = 3 + ((unsigned)b & 7);
939 DUMPBITS(3)
940 if ((unsigned)i + j > n) {
941 ret = 1;
942 goto out;
943 }
944 while (j--)
945 ll[i++] = 0;
946 l = 0;
947 }
948 else /* j == 18: 11 to 138 zero length codes */
949 {
950 NEEDBITS(7)
951 j = 11 + ((unsigned)b & 0x7f);
952 DUMPBITS(7)
953 if ((unsigned)i + j > n) {
954 ret = 1;
955 goto out;
956 }
957 while (j--)
958 ll[i++] = 0;
959 l = 0;
960 }
961 }
962
963DEBG("dyn4 ");
964
965 /* free decoding table for trees */
966 huft_free(t: tl);
967
968DEBG("dyn5 ");
969
970 /* restore the global bit buffer */
971 bb = b;
972 bk = k;
973
974DEBG("dyn5a ");
975
976 /* build the decoding tables for literal/length and distance codes */
977 bl = lbits;
978 if ((i = huft_build(b: ll, n: nl, s: 257, d: cplens, e: cplext, t: &tl, m: &bl)) != 0)
979 {
980DEBG("dyn5b ");
981 if (i == 1) {
982 error(m: "incomplete literal tree");
983 huft_free(t: tl);
984 }
985 ret = i; /* incomplete code set */
986 goto out;
987 }
988DEBG("dyn5c ");
989 bd = dbits;
990 if ((i = huft_build(b: ll + nl, n: nd, s: 0, d: cpdist, e: cpdext, t: &td, m: &bd)) != 0)
991 {
992DEBG("dyn5d ");
993 if (i == 1) {
994 error(m: "incomplete distance tree");
995#ifdef PKZIP_BUG_WORKAROUND
996 i = 0;
997 }
998#else
999 huft_free(t: td);
1000 }
1001 huft_free(t: tl);
1002 ret = i; /* incomplete code set */
1003 goto out;
1004#endif
1005 }
1006
1007DEBG("dyn6 ");
1008
1009 /* decompress until an end-of-block code */
1010 if (inflate_codes(tl, td, bl, bd)) {
1011 ret = 1;
1012 goto out;
1013 }
1014
1015DEBG("dyn7 ");
1016
1017 /* free the decoding tables, return */
1018 huft_free(t: tl);
1019 huft_free(t: td);
1020
1021 DEBG(">");
1022 ret = 0;
1023out:
1024 free(where: ll);
1025 return ret;
1026
1027underrun:
1028 ret = 4; /* Input underrun */
1029 goto out;
1030}
1031
1032
1033
1034STATIC int INIT inflate_block(
1035 int *e /* last block flag */
1036 )
1037/* decompress an inflated block */
1038{
1039 unsigned t; /* block type */
1040 register ulg b; /* bit buffer */
1041 register unsigned k; /* number of bits in bit buffer */
1042
1043 DEBG("<blk");
1044
1045 /* make local bit buffer */
1046 b = bb;
1047 k = bk;
1048
1049
1050 /* read in last block bit */
1051 NEEDBITS(1)
1052 *e = (int)b & 1;
1053 DUMPBITS(1)
1054
1055
1056 /* read in block type */
1057 NEEDBITS(2)
1058 t = (unsigned)b & 3;
1059 DUMPBITS(2)
1060
1061
1062 /* restore the global bit buffer */
1063 bb = b;
1064 bk = k;
1065
1066 /* inflate that block type */
1067 if (t == 2)
1068 return inflate_dynamic();
1069 if (t == 0)
1070 return inflate_stored();
1071 if (t == 1)
1072 return inflate_fixed();
1073
1074 DEBG(">");
1075
1076 /* bad block type */
1077 return 2;
1078
1079 underrun:
1080 return 4; /* Input underrun */
1081}
1082
1083
1084
1085STATIC int INIT inflate(void)
1086/* decompress an inflated entry */
1087{
1088 int e; /* last block flag */
1089 int r; /* result code */
1090 unsigned h; /* maximum struct huft's malloc'ed */
1091
1092 /* initialize window, bit buffer */
1093 wp = 0;
1094 bk = 0;
1095 bb = 0;
1096
1097
1098 /* decompress until the last block */
1099 h = 0;
1100 do {
1101 hufts = 0;
1102#ifdef ARCH_HAS_DECOMP_WDOG
1103 arch_decomp_wdog();
1104#endif
1105 r = inflate_block(e: &e);
1106 if (r)
1107 return r;
1108 if (hufts > h)
1109 h = hufts;
1110 } while (!e);
1111
1112 /* Undo too much lookahead. The next read will be byte aligned so we
1113 * can discard unused bits in the last meaningful byte.
1114 */
1115 while (bk >= 8) {
1116 bk -= 8;
1117 inptr--;
1118 }
1119
1120 /* flush out slide */
1121 flush_output(wp);
1122
1123
1124 /* return success */
1125#ifdef DEBUG
1126 fprintf(stderr, "<%u> ", h);
1127#endif /* DEBUG */
1128 return 0;
1129}
1130
1131/**********************************************************************
1132 *
1133 * The following are support routines for inflate.c
1134 *
1135 **********************************************************************/
1136
1137static ulg crc_32_tab[256];
1138static ulg crc; /* initialized in makecrc() so it'll reside in bss */
1139#define CRC_VALUE (crc ^ 0xffffffffUL)
1140
1141/*
1142 * Code to compute the CRC-32 table. Borrowed from
1143 * gzip-1.0.3/makecrc.c.
1144 */
1145
1146static void INIT
1147makecrc(void)
1148{
1149/* Not copyrighted 1990 Mark Adler */
1150
1151 unsigned long c; /* crc shift register */
1152 unsigned long e; /* polynomial exclusive-or pattern */
1153 int i; /* counter for all possible eight bit values */
1154 int k; /* byte being shifted into crc apparatus */
1155
1156 /* terms of polynomial defining this crc (except x^32): */
1157 static const int p[] = {0,1,2,4,5,7,8,10,11,12,16,22,23,26};
1158
1159 /* Make exclusive-or pattern from polynomial */
1160 e = 0;
1161 for (i = 0; i < sizeof(p)/sizeof(int); i++)
1162 e |= 1L << (31 - p[i]);
1163
1164 crc_32_tab[0] = 0;
1165
1166 for (i = 1; i < 256; i++)
1167 {
1168 c = 0;
1169 for (k = i | 256; k != 1; k >>= 1)
1170 {
1171 c = c & 1 ? (c >> 1) ^ e : c >> 1;
1172 if (k & 1)
1173 c ^= e;
1174 }
1175 crc_32_tab[i] = c;
1176 }
1177
1178 /* this is initialized here so this code could reside in ROM */
1179 crc = (ulg)0xffffffffUL; /* shift register contents */
1180}
1181
1182/* gzip flag byte */
1183#define ASCII_FLAG 0x01 /* bit 0 set: file probably ASCII text */
1184#define CONTINUATION 0x02 /* bit 1 set: continuation of multi-part gzip file */
1185#define EXTRA_FIELD 0x04 /* bit 2 set: extra field present */
1186#define ORIG_NAME 0x08 /* bit 3 set: original file name present */
1187#define COMMENT 0x10 /* bit 4 set: file comment present */
1188#define ENCRYPTED 0x20 /* bit 5 set: file is encrypted */
1189#define RESERVED 0xC0 /* bit 6,7: reserved */
1190
1191/*
1192 * Do the uncompression!
1193 */
1194static int INIT gunzip(void)
1195{
1196 uch flags;
1197 unsigned char magic[2]; /* magic header */
1198 char method;
1199 ulg orig_crc = 0; /* original crc */
1200 ulg orig_len = 0; /* original uncompressed length */
1201 int res;
1202
1203 magic[0] = NEXTBYTE();
1204 magic[1] = NEXTBYTE();
1205 method = NEXTBYTE();
1206
1207 if (magic[0] != 037 ||
1208 ((magic[1] != 0213) && (magic[1] != 0236))) {
1209 error(m: "bad gzip magic numbers");
1210 return -1;
1211 }
1212
1213 /* We only support method #8, DEFLATED */
1214 if (method != 8) {
1215 error(m: "internal error, invalid method");
1216 return -1;
1217 }
1218
1219 flags = (uch)get_byte();
1220 if ((flags & ENCRYPTED) != 0) {
1221 error(m: "Input is encrypted");
1222 return -1;
1223 }
1224 if ((flags & CONTINUATION) != 0) {
1225 error(m: "Multi part input");
1226 return -1;
1227 }
1228 if ((flags & RESERVED) != 0) {
1229 error(m: "Input has invalid flags");
1230 return -1;
1231 }
1232 NEXTBYTE(); /* Get timestamp */
1233 NEXTBYTE();
1234 NEXTBYTE();
1235 NEXTBYTE();
1236
1237 (void)NEXTBYTE(); /* Ignore extra flags for the moment */
1238 (void)NEXTBYTE(); /* Ignore OS type for the moment */
1239
1240 if ((flags & EXTRA_FIELD) != 0) {
1241 unsigned len = (unsigned)NEXTBYTE();
1242 len |= ((unsigned)NEXTBYTE())<<8;
1243 while (len--) (void)NEXTBYTE();
1244 }
1245
1246 /* Get original file name if it was truncated */
1247 if ((flags & ORIG_NAME) != 0) {
1248 /* Discard the old name */
1249 while (NEXTBYTE() != 0) /* null */ ;
1250 }
1251
1252 /* Discard file comment if any */
1253 if ((flags & COMMENT) != 0) {
1254 while (NEXTBYTE() != 0) /* null */ ;
1255 }
1256
1257 /* Decompress */
1258 if ((res = inflate())) {
1259 switch (res) {
1260 case 0:
1261 break;
1262 case 1:
1263 error(m: "invalid compressed format (err=1)");
1264 break;
1265 case 2:
1266 error(m: "invalid compressed format (err=2)");
1267 break;
1268 case 3:
1269 error(m: "out of memory");
1270 break;
1271 case 4:
1272 error(m: "out of input data");
1273 break;
1274 default:
1275 error(m: "invalid compressed format (other)");
1276 }
1277 return -1;
1278 }
1279
1280 /* Get the crc and original length */
1281 /* crc32 (see algorithm.doc)
1282 * uncompressed input size modulo 2^32
1283 */
1284 orig_crc = (ulg) NEXTBYTE();
1285 orig_crc |= (ulg) NEXTBYTE() << 8;
1286 orig_crc |= (ulg) NEXTBYTE() << 16;
1287 orig_crc |= (ulg) NEXTBYTE() << 24;
1288
1289 orig_len = (ulg) NEXTBYTE();
1290 orig_len |= (ulg) NEXTBYTE() << 8;
1291 orig_len |= (ulg) NEXTBYTE() << 16;
1292 orig_len |= (ulg) NEXTBYTE() << 24;
1293
1294 /* Validate decompression */
1295 if (orig_crc != CRC_VALUE) {
1296 error(m: "crc error");
1297 return -1;
1298 }
1299 if (orig_len != bytes_out) {
1300 error(m: "length error");
1301 return -1;
1302 }
1303 return 0;
1304
1305 underrun: /* NEXTBYTE() goto's here if needed */
1306 error(m: "out of input data");
1307 return -1;
1308}
1309
1310
1311

source code of linux/lib/inflate.c