Merge branch 'master' of https://source.denx.de/u-boot/custodians/u-boot-sunxi
[platform/kernel/u-boot.git] / lib / lz4.c
1 // SPDX-License-Identifier: GPL 2.0+ OR BSD-2-Clause
2 /*
3  * LZ4 - Fast LZ compression algorithm
4  * Copyright (C) 2011 - 2016, Yann Collet.
5  * BSD 2 - Clause License (http://www.opensource.org/licenses/bsd - license.php)
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions are
8  * met:
9  *      * Redistributions of source code must retain the above copyright
10  *        notice, this list of conditions and the following disclaimer.
11  *      * Redistributions in binary form must reproduce the above
12  * copyright notice, this list of conditions and the following disclaimer
13  * in the documentation and/or other materials provided with the
14  * distribution.
15  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
17  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
18  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
19  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
20  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
21  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
25  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26  * You can contact the author at :
27  *      - LZ4 homepage : http://www.lz4.org
28  *      - LZ4 source repository : https://github.com/lz4/lz4
29  */
30 #include <common.h>
31 #include <compiler.h>
32 #include <linux/kernel.h>
33 #include <linux/types.h>
34 #include <linux/bug.h>
35 #include <asm/unaligned.h>
36 #include <u-boot/lz4.h>
37
38 #define FORCE_INLINE inline __attribute__((always_inline))
39
40 static FORCE_INLINE u16 LZ4_readLE16(const void *src)
41 {
42         return get_unaligned_le16(src);
43 }
44
45 static FORCE_INLINE void LZ4_copy8(void *dst, const void *src)
46 {
47         put_unaligned(get_unaligned((const u64 *)src), (u64 *)dst);
48 }
49
50 typedef  uint8_t BYTE;
51 typedef uint16_t U16;
52 typedef uint32_t U32;
53 typedef  int32_t S32;
54 typedef uint64_t U64;
55 typedef uintptr_t uptrval;
56
57 static FORCE_INLINE void LZ4_write32(void *memPtr, U32 value)
58 {
59         put_unaligned(value, (U32 *)memPtr);
60 }
61
62 /**************************************
63 *  Reading and writing into memory
64 **************************************/
65
66 /* customized version of memcpy, which may overwrite up to 7 bytes beyond dstEnd */
67 static void LZ4_wildCopy(void* dstPtr, const void* srcPtr, void* dstEnd)
68 {
69     BYTE* d = (BYTE*)dstPtr;
70     const BYTE* s = (const BYTE*)srcPtr;
71     BYTE* e = (BYTE*)dstEnd;
72     do { LZ4_copy8(d,s); d+=8; s+=8; } while (d<e);
73 }
74
75
76 /**************************************
77 *  Common Constants
78 **************************************/
79 #define MINMATCH 4
80
81 #define WILDCOPYLENGTH 8
82 #define LASTLITERALS 5
83 #define MFLIMIT (WILDCOPYLENGTH + MINMATCH)
84
85 /*
86  * ensure it's possible to write 2 x wildcopyLength
87  * without overflowing output buffer
88  */
89 #define MATCH_SAFEGUARD_DISTANCE  ((2 * WILDCOPYLENGTH) - MINMATCH)
90
91 #define KB (1 <<10)
92
93 #define MAXD_LOG 16
94 #define MAX_DISTANCE ((1 << MAXD_LOG) - 1)
95
96 #define ML_BITS  4
97 #define ML_MASK  ((1U<<ML_BITS)-1)
98 #define RUN_BITS (8-ML_BITS)
99 #define RUN_MASK ((1U<<RUN_BITS)-1)
100
101 #define LZ4_STATIC_ASSERT(c)    BUILD_BUG_ON(!(c))
102
103 /**************************************
104 *  Local Structures and types
105 **************************************/
106 typedef enum { noDict = 0, withPrefix64k, usingExtDict } dict_directive;
107 typedef enum { endOnOutputSize = 0, endOnInputSize = 1 } endCondition_directive;
108 typedef enum { decode_full_block = 0, partial_decode = 1 } earlyEnd_directive;
109
110 #define DEBUGLOG(l, ...) {}     /* disabled */
111
112 #ifndef assert
113 #define assert(condition) ((void)0)
114 #endif
115
116 /*
117  * LZ4_decompress_generic() :
118  * This generic decompression function covers all use cases.
119  * It shall be instantiated several times, using different sets of directives.
120  * Note that it is important for performance that this function really get inlined,
121  * in order to remove useless branches during compilation optimization.
122  */
123 static FORCE_INLINE int LZ4_decompress_generic(
124          const char * const src,
125          char * const dst,
126          int srcSize,
127                 /*
128                  * If endOnInput == endOnInputSize,
129                  * this value is `dstCapacity`
130                  */
131          int outputSize,
132          /* endOnOutputSize, endOnInputSize */
133          endCondition_directive endOnInput,
134          /* full, partial */
135          earlyEnd_directive partialDecoding,
136          /* noDict, withPrefix64k, usingExtDict */
137          dict_directive dict,
138          /* always <= dst, == dst when no prefix */
139          const BYTE * const lowPrefix,
140          /* only if dict == usingExtDict */
141          const BYTE * const dictStart,
142          /* note : = 0 if noDict */
143          const size_t dictSize
144          )
145 {
146         const BYTE *ip = (const BYTE *) src;
147         const BYTE * const iend = ip + srcSize;
148
149         BYTE *op = (BYTE *) dst;
150         BYTE * const oend = op + outputSize;
151         BYTE *cpy;
152
153         const BYTE * const dictEnd = (const BYTE *)dictStart + dictSize;
154         static const unsigned int inc32table[8] = {0, 1, 2, 1, 0, 4, 4, 4};
155         static const int dec64table[8] = {0, 0, 0, -1, -4, 1, 2, 3};
156
157         const int safeDecode = (endOnInput == endOnInputSize);
158         const int checkOffset = ((safeDecode) && (dictSize < (int)(64 * KB)));
159
160         /* Set up the "end" pointers for the shortcut. */
161         const BYTE *const shortiend = iend -
162                 (endOnInput ? 14 : 8) /*maxLL*/ - 2 /*offset*/;
163         const BYTE *const shortoend = oend -
164                 (endOnInput ? 14 : 8) /*maxLL*/ - 18 /*maxML*/;
165
166         DEBUGLOG(5, "%s (srcSize:%i, dstSize:%i)", __func__,
167                  srcSize, outputSize);
168
169         /* Special cases */
170         assert(lowPrefix <= op);
171         assert(src != NULL);
172
173         /* Empty output buffer */
174         if ((endOnInput) && (unlikely(outputSize == 0)))
175                 return ((srcSize == 1) && (*ip == 0)) ? 0 : -1;
176
177         if ((!endOnInput) && (unlikely(outputSize == 0)))
178                 return (*ip == 0 ? 1 : -1);
179
180         if ((endOnInput) && unlikely(srcSize == 0))
181                 return -1;
182
183         /* Main Loop : decode sequences */
184         while (1) {
185                 size_t length;
186                 const BYTE *match;
187                 size_t offset;
188
189                 /* get literal length */
190                 unsigned int const token = *ip++;
191                 length = token>>ML_BITS;
192
193                 /* ip < iend before the increment */
194                 assert(!endOnInput || ip <= iend);
195
196                 /*
197                  * A two-stage shortcut for the most common case:
198                  * 1) If the literal length is 0..14, and there is enough
199                  * space, enter the shortcut and copy 16 bytes on behalf
200                  * of the literals (in the fast mode, only 8 bytes can be
201                  * safely copied this way).
202                  * 2) Further if the match length is 4..18, copy 18 bytes
203                  * in a similar manner; but we ensure that there's enough
204                  * space in the output for those 18 bytes earlier, upon
205                  * entering the shortcut (in other words, there is a
206                  * combined check for both stages).
207                  *
208                  * The & in the likely() below is intentionally not && so that
209                  * some compilers can produce better parallelized runtime code
210                  */
211                 if ((endOnInput ? length != RUN_MASK : length <= 8)
212                    /*
213                     * strictly "less than" on input, to re-enter
214                     * the loop with at least one byte
215                     */
216                    && likely((endOnInput ? ip < shortiend : 1) &
217                              (op <= shortoend))) {
218                         /* Copy the literals */
219                         memcpy(op, ip, endOnInput ? 16 : 8);
220                         op += length; ip += length;
221
222                         /*
223                          * The second stage:
224                          * prepare for match copying, decode full info.
225                          * If it doesn't work out, the info won't be wasted.
226                          */
227                         length = token & ML_MASK; /* match length */
228                         offset = LZ4_readLE16(ip);
229                         ip += 2;
230                         match = op - offset;
231                         assert(match <= op); /* check overflow */
232
233                         /* Do not deal with overlapping matches. */
234                         if ((length != ML_MASK) &&
235                             (offset >= 8) &&
236                             (dict == withPrefix64k || match >= lowPrefix)) {
237                                 /* Copy the match. */
238                                 memcpy(op + 0, match + 0, 8);
239                                 memcpy(op + 8, match + 8, 8);
240                                 memcpy(op + 16, match + 16, 2);
241                                 op += length + MINMATCH;
242                                 /* Both stages worked, load the next token. */
243                                 continue;
244                         }
245
246                         /*
247                          * The second stage didn't work out, but the info
248                          * is ready. Propel it right to the point of match
249                          * copying.
250                          */
251                         goto _copy_match;
252                 }
253
254                 /* decode literal length */
255                 if (length == RUN_MASK) {
256                         unsigned int s;
257
258                         if (unlikely(endOnInput ? ip >= iend - RUN_MASK : 0)) {
259                                 /* overflow detection */
260                                 goto _output_error;
261                         }
262                         do {
263                                 s = *ip++;
264                                 length += s;
265                         } while (likely(endOnInput
266                                 ? ip < iend - RUN_MASK
267                                 : 1) & (s == 255));
268
269                         if ((safeDecode)
270                             && unlikely((uptrval)(op) +
271                                         length < (uptrval)(op))) {
272                                 /* overflow detection */
273                                 goto _output_error;
274                         }
275                         if ((safeDecode)
276                             && unlikely((uptrval)(ip) +
277                                         length < (uptrval)(ip))) {
278                                 /* overflow detection */
279                                 goto _output_error;
280                         }
281                 }
282
283                 /* copy literals */
284                 cpy = op + length;
285                 LZ4_STATIC_ASSERT(MFLIMIT >= WILDCOPYLENGTH);
286
287                 if (((endOnInput) && ((cpy > oend - MFLIMIT)
288                         || (ip + length > iend - (2 + 1 + LASTLITERALS))))
289                         || ((!endOnInput) && (cpy > oend - WILDCOPYLENGTH))) {
290                         if (partialDecoding) {
291                                 if (cpy > oend) {
292                                         /*
293                                          * Partial decoding :
294                                          * stop in the middle of literal segment
295                                          */
296                                         cpy = oend;
297                                         length = oend - op;
298                                 }
299                                 if ((endOnInput)
300                                         && (ip + length > iend)) {
301                                         /*
302                                          * Error :
303                                          * read attempt beyond
304                                          * end of input buffer
305                                          */
306                                         goto _output_error;
307                                 }
308                         } else {
309                                 if ((!endOnInput)
310                                         && (cpy != oend)) {
311                                         /*
312                                          * Error :
313                                          * block decoding must
314                                          * stop exactly there
315                                          */
316                                         goto _output_error;
317                                 }
318                                 if ((endOnInput)
319                                         && ((ip + length != iend)
320                                         || (cpy > oend))) {
321                                         /*
322                                          * Error :
323                                          * input must be consumed
324                                          */
325                                         goto _output_error;
326                                 }
327                         }
328
329                         /*
330                          * supports overlapping memory regions; only matters
331                          * for in-place decompression scenarios
332                          */
333                         memmove(op, ip, length);
334                         ip += length;
335                         op += length;
336
337                         /* Necessarily EOF, due to parsing restrictions */
338                         if (!partialDecoding || (cpy == oend))
339                                 break;
340                 } else {
341                         /* may overwrite up to WILDCOPYLENGTH beyond cpy */
342                         LZ4_wildCopy(op, ip, cpy);
343                         ip += length;
344                         op = cpy;
345                 }
346
347                 /* get offset */
348                 offset = LZ4_readLE16(ip);
349                 ip += 2;
350                 match = op - offset;
351
352                 /* get matchlength */
353                 length = token & ML_MASK;
354
355 _copy_match:
356                 if ((checkOffset) && (unlikely(match + dictSize < lowPrefix))) {
357                         /* Error : offset outside buffers */
358                         goto _output_error;
359                 }
360
361                 /* costs ~1%; silence an msan warning when offset == 0 */
362                 /*
363                  * note : when partialDecoding, there is no guarantee that
364                  * at least 4 bytes remain available in output buffer
365                  */
366                 if (!partialDecoding) {
367                         assert(oend > op);
368                         assert(oend - op >= 4);
369
370                         LZ4_write32(op, (U32)offset);
371                 }
372
373                 if (length == ML_MASK) {
374                         unsigned int s;
375
376                         do {
377                                 s = *ip++;
378
379                                 if ((endOnInput) && (ip > iend - LASTLITERALS))
380                                         goto _output_error;
381
382                                 length += s;
383                         } while (s == 255);
384
385                         if ((safeDecode)
386                                 && unlikely(
387                                         (uptrval)(op) + length < (uptrval)op)) {
388                                 /* overflow detection */
389                                 goto _output_error;
390                         }
391                 }
392
393                 length += MINMATCH;
394
395                 /* match starting within external dictionary */
396                 if ((dict == usingExtDict) && (match < lowPrefix)) {
397                         if (unlikely(op + length > oend - LASTLITERALS)) {
398                                 /* doesn't respect parsing restriction */
399                                 if (!partialDecoding)
400                                         goto _output_error;
401                                 length = min(length, (size_t)(oend - op));
402                         }
403
404                         if (length <= (size_t)(lowPrefix - match)) {
405                                 /*
406                                  * match fits entirely within external
407                                  * dictionary : just copy
408                                  */
409                                 memmove(op, dictEnd - (lowPrefix - match),
410                                         length);
411                                 op += length;
412                         } else {
413                                 /*
414                                  * match stretches into both external
415                                  * dictionary and current block
416                                  */
417                                 size_t const copySize = (size_t)(lowPrefix - match);
418                                 size_t const restSize = length - copySize;
419
420                                 memcpy(op, dictEnd - copySize, copySize);
421                                 op += copySize;
422                                 if (restSize > (size_t)(op - lowPrefix)) {
423                                         /* overlap copy */
424                                         BYTE * const endOfMatch = op + restSize;
425                                         const BYTE *copyFrom = lowPrefix;
426
427                                         while (op < endOfMatch)
428                                                 *op++ = *copyFrom++;
429                                 } else {
430                                         memcpy(op, lowPrefix, restSize);
431                                         op += restSize;
432                                 }
433                         }
434                         continue;
435                 }
436
437                 /* copy match within block */
438                 cpy = op + length;
439
440                 /*
441                  * partialDecoding :
442                  * may not respect endBlock parsing restrictions
443                  */
444                 assert(op <= oend);
445                 if (partialDecoding &&
446                     (cpy > oend - MATCH_SAFEGUARD_DISTANCE)) {
447                         size_t const mlen = min(length, (size_t)(oend - op));
448                         const BYTE * const matchEnd = match + mlen;
449                         BYTE * const copyEnd = op + mlen;
450
451                         if (matchEnd > op) {
452                                 /* overlap copy */
453                                 while (op < copyEnd)
454                                         *op++ = *match++;
455                         } else {
456                                 memcpy(op, match, mlen);
457                         }
458                         op = copyEnd;
459                         if (op == oend)
460                                 break;
461                         continue;
462                 }
463
464                 if (unlikely(offset < 8)) {
465                         op[0] = match[0];
466                         op[1] = match[1];
467                         op[2] = match[2];
468                         op[3] = match[3];
469                         match += inc32table[offset];
470                         memcpy(op + 4, match, 4);
471                         match -= dec64table[offset];
472                 } else {
473                         LZ4_copy8(op, match);
474                         match += 8;
475                 }
476
477                 op += 8;
478
479                 if (unlikely(cpy > oend - MATCH_SAFEGUARD_DISTANCE)) {
480                         BYTE * const oCopyLimit = oend - (WILDCOPYLENGTH - 1);
481
482                         if (cpy > oend - LASTLITERALS) {
483                                 /*
484                                  * Error : last LASTLITERALS bytes
485                                  * must be literals (uncompressed)
486                                  */
487                                 goto _output_error;
488                         }
489
490                         if (op < oCopyLimit) {
491                                 LZ4_wildCopy(op, match, oCopyLimit);
492                                 match += oCopyLimit - op;
493                                 op = oCopyLimit;
494                         }
495                         while (op < cpy)
496                                 *op++ = *match++;
497                 } else {
498                         LZ4_copy8(op, match);
499                         if (length > 16)
500                                 LZ4_wildCopy(op + 8, match + 8, cpy);
501                 }
502                 op = cpy; /* wildcopy correction */
503         }
504
505         /* end of decoding */
506         if (endOnInput) {
507                 /* Nb of output bytes decoded */
508                 return (int) (((char *)op) - dst);
509         } else {
510                 /* Nb of input bytes read */
511                 return (int) (((const char *)ip) - src);
512         }
513
514         /* Overflow error detected */
515 _output_error:
516         return (int) (-(((const char *)ip) - src)) - 1;
517 }
518
519 int LZ4_decompress_safe(const char *source, char *dest,
520         int compressedSize, int maxDecompressedSize)
521 {
522         return LZ4_decompress_generic(source, dest,
523                                       compressedSize, maxDecompressedSize,
524                                       endOnInputSize, decode_full_block,
525                                       noDict, (BYTE *)dest, NULL, 0);
526 }
527
528 int LZ4_decompress_safe_partial(const char *src, char *dst,
529         int compressedSize, int targetOutputSize, int dstCapacity)
530 {
531         dstCapacity = min(targetOutputSize, dstCapacity);
532         return LZ4_decompress_generic(src, dst, compressedSize, dstCapacity,
533                                       endOnInputSize, partial_decode,
534                                       noDict, (BYTE *)dst, NULL, 0);
535 }