ntfs3/lib/decompress_common.h

0001 /* SPDX-License-Identifier: GPL-2.0-or-later */
0002 /*
0003  * decompress_common.h - Code shared by the XPRESS and LZX decompressors
0004  *
0005  * Copyright (C) 2015 Eric Biggers
0006  */
0007
0008 #ifndef _LINUX_NTFS3_LIB_DECOMPRESS_COMMON_H
0009 #define _LINUX_NTFS3_LIB_DECOMPRESS_COMMON_H
0010
0011 #include <linux/string.h>
0012 #include <linux/compiler.h>
0013 #include <linux/types.h>
0014 #include <linux/slab.h>
0015 #include <asm/unaligned.h>
0016
0017
0018 /* "Force inline" macro (not required, but helpful for performance)  */
0019 #define forceinline __always_inline
0020
0021 /* Enable whole-word match copying on selected architectures  */
0022 #if defined(__i386__) || defined(__x86_64__) || defined(__ARM_FEATURE_UNALIGNED)
0023 #  define FAST_UNALIGNED_ACCESS
0024 #endif
0025
0026 /* Size of a machine word  */
0027 #define WORDBYTES (sizeof(size_t))
0028
0029 static forceinline void
0030 copy_unaligned_word(const void *src, void *dst)
0031 {
0032     put_unaligned(get_unaligned((const size_t *)src), (size_t *)dst);
0033 }
0034
0035
0036 /* Generate a "word" with platform-dependent size whose bytes all contain the
0037  * value 'b'.
0038  */
0039 static forceinline size_t repeat_byte(u8 b)
0040 {
0041     size_t v;
0042
0043     v = b;
0044     v |= v << 8;
0045     v |= v << 16;
0046     v |= v << ((WORDBYTES == 8) ? 32 : 0);
0047     return v;
0048 }
0049
0050 /* Structure that encapsulates a block of in-memory data being interpreted as a
0051  * stream of bits, optionally with interwoven literal bytes.  Bits are assumed
0052  * to be stored in little endian 16-bit coding units, with the bits ordered high
0053  * to low.
0054  */
0055 struct input_bitstream {
0056
0057     /* Bits that have been read from the input buffer.  The bits are
0058      * left-justified; the next bit is always bit 31.
0059      */
0060     u32 bitbuf;
0061
0062     /* Number of bits currently held in @bitbuf.  */
0063     u32 bitsleft;
0064
0065     /* Pointer to the next byte to be retrieved from the input buffer.  */
0066     const u8 *next;
0067
0068     /* Pointer to just past the end of the input buffer.  */
0069     const u8 *end;
0070 };
0071
0072 /* Initialize a bitstream to read from the specified input buffer.  */
0073 static forceinline void init_input_bitstream(struct input_bitstream *is,
0074                          const void *buffer, u32 size)
0075 {
0076     is->bitbuf = 0;
0077     is->bitsleft = 0;
0078     is->next = buffer;
0079     is->end = is->next + size;
0080 }
0081
0082 /* Ensure the bit buffer variable for the bitstream contains at least @num_bits
0083  * bits.  Following this, bitstream_peek_bits() and/or bitstream_remove_bits()
0084  * may be called on the bitstream to peek or remove up to @num_bits bits.  Note
0085  * that @num_bits must be <= 16.
0086  */
0087 static forceinline void bitstream_ensure_bits(struct input_bitstream *is,
0088                           u32 num_bits)
0089 {
0090     if (is->bitsleft < num_bits) {
0091         if (is->end - is->next >= 2) {
0092             is->bitbuf |= (u32)get_unaligned_le16(is->next)
0093                     << (16 - is->bitsleft);
0094             is->next += 2;
0095         }
0096         is->bitsleft += 16;
0097     }
0098 }
0099
0100 /* Return the next @num_bits bits from the bitstream, without removing them.
0101  * There must be at least @num_bits remaining in the buffer variable, from a
0102  * previous call to bitstream_ensure_bits().
0103  */
0104 static forceinline u32
0105 bitstream_peek_bits(const struct input_bitstream *is, const u32 num_bits)
0106 {
0107     return (is->bitbuf >> 1) >> (sizeof(is->bitbuf) * 8 - num_bits - 1);
0108 }
0109
0110 /* Remove @num_bits from the bitstream.  There must be at least @num_bits
0111  * remaining in the buffer variable, from a previous call to
0112  * bitstream_ensure_bits().
0113  */
0114 static forceinline void
0115 bitstream_remove_bits(struct input_bitstream *is, u32 num_bits)
0116 {
0117     is->bitbuf <<= num_bits;
0118     is->bitsleft -= num_bits;
0119 }
0120
0121 /* Remove and return @num_bits bits from the bitstream.  There must be at least
0122  * @num_bits remaining in the buffer variable, from a previous call to
0123  * bitstream_ensure_bits().
0124  */
0125 static forceinline u32
0126 bitstream_pop_bits(struct input_bitstream *is, u32 num_bits)
0127 {
0128     u32 bits = bitstream_peek_bits(is, num_bits);
0129
0130     bitstream_remove_bits(is, num_bits);
0131     return bits;
0132 }
0133
0134 /* Read and return the next @num_bits bits from the bitstream.  */
0135 static forceinline u32
0136 bitstream_read_bits(struct input_bitstream *is, u32 num_bits)
0137 {
0138     bitstream_ensure_bits(is, num_bits);
0139     return bitstream_pop_bits(is, num_bits);
0140 }
0141
0142 /* Read and return the next literal byte embedded in the bitstream.  */
0143 static forceinline u8
0144 bitstream_read_byte(struct input_bitstream *is)
0145 {
0146     if (unlikely(is->end == is->next))
0147         return 0;
0148     return *is->next++;
0149 }
0150
0151 /* Read and return the next 16-bit integer embedded in the bitstream.  */
0152 static forceinline u16
0153 bitstream_read_u16(struct input_bitstream *is)
0154 {
0155     u16 v;
0156
0157     if (unlikely(is->end - is->next < 2))
0158         return 0;
0159     v = get_unaligned_le16(is->next);
0160     is->next += 2;
0161     return v;
0162 }
0163
0164 /* Read and return the next 32-bit integer embedded in the bitstream.  */
0165 static forceinline u32
0166 bitstream_read_u32(struct input_bitstream *is)
0167 {
0168     u32 v;
0169
0170     if (unlikely(is->end - is->next < 4))
0171         return 0;
0172     v = get_unaligned_le32(is->next);
0173     is->next += 4;
0174     return v;
0175 }
0176
0177 /* Read into @dst_buffer an array of literal bytes embedded in the bitstream.
0178  * Return either a pointer to the byte past the last written, or NULL if the
0179  * read overflows the input buffer.
0180  */
0181 static forceinline void *bitstream_read_bytes(struct input_bitstream *is,
0182                           void *dst_buffer, size_t count)
0183 {
0184     if ((size_t)(is->end - is->next) < count)
0185         return NULL;
0186     memcpy(dst_buffer, is->next, count);
0187     is->next += count;
0188     return (u8 *)dst_buffer + count;
0189 }
0190
0191 /* Align the input bitstream on a coding-unit boundary.  */
0192 static forceinline void bitstream_align(struct input_bitstream *is)
0193 {
0194     is->bitsleft = 0;
0195     is->bitbuf = 0;
0196 }
0197
0198 extern int make_huffman_decode_table(u16 decode_table[], const u32 num_syms,
0199                      const u32 num_bits, const u8 lens[],
0200                      const u32 max_codeword_len,
0201                      u16 working_space[]);
0202
0203
0204 /* Reads and returns the next Huffman-encoded symbol from a bitstream.  If the
0205  * input data is exhausted, the Huffman symbol is decoded as if the missing bits
0206  * are all zeroes.
0207  */
0208 static forceinline u32 read_huffsym(struct input_bitstream *istream,
0209                      const u16 decode_table[],
0210                      u32 table_bits,
0211                      u32 max_codeword_len)
0212 {
0213     u32 entry;
0214     u32 key_bits;
0215
0216     bitstream_ensure_bits(istream, max_codeword_len);
0217
0218     /* Index the decode table by the next table_bits bits of the input.  */
0219     key_bits = bitstream_peek_bits(istream, table_bits);
0220     entry = decode_table[key_bits];
0221     if (entry < 0xC000) {
0222         /* Fast case: The decode table directly provided the
0223          * symbol and codeword length.  The low 11 bits are the
0224          * symbol, and the high 5 bits are the codeword length.
0225          */
0226         bitstream_remove_bits(istream, entry >> 11);
0227         return entry & 0x7FF;
0228     }
0229     /* Slow case: The codeword for the symbol is longer than
0230      * table_bits, so the symbol does not have an entry
0231      * directly in the first (1 << table_bits) entries of the
0232      * decode table.  Traverse the appropriate binary tree
0233      * bit-by-bit to decode the symbol.
0234      */
0235     bitstream_remove_bits(istream, table_bits);
0236     do {
0237         key_bits = (entry & 0x3FFF) + bitstream_pop_bits(istream, 1);
0238     } while ((entry = decode_table[key_bits]) >= 0xC000);
0239     return entry;
0240 }
0241
0242 /*
0243  * Copy an LZ77 match at (dst - offset) to dst.
0244  *
0245  * The length and offset must be already validated --- that is, (dst - offset)
0246  * can't underrun the output buffer, and (dst + length) can't overrun the output
0247  * buffer.  Also, the length cannot be 0.
0248  *
0249  * @bufend points to the byte past the end of the output buffer.  This function
0250  * won't write any data beyond this position.
0251  *
0252  * Returns dst + length.
0253  */
0254 static forceinline u8 *lz_copy(u8 *dst, u32 length, u32 offset, const u8 *bufend,
0255                    u32 min_length)
0256 {
0257     const u8 *src = dst - offset;
0258
0259     /*
0260      * Try to copy one machine word at a time.  On i386 and x86_64 this is
0261      * faster than copying one byte at a time, unless the data is
0262      * near-random and all the matches have very short lengths.  Note that
0263      * since this requires unaligned memory accesses, it won't necessarily
0264      * be faster on every architecture.
0265      *
0266      * Also note that we might copy more than the length of the match.  For
0267      * example, if a word is 8 bytes and the match is of length 5, then
0268      * we'll simply copy 8 bytes.  This is okay as long as we don't write
0269      * beyond the end of the output buffer, hence the check for (bufend -
0270      * end >= WORDBYTES - 1).
0271      */
0272 #ifdef FAST_UNALIGNED_ACCESS
0273     u8 * const end = dst + length;
0274
0275     if (bufend - end >= (ptrdiff_t)(WORDBYTES - 1)) {
0276
0277         if (offset >= WORDBYTES) {
0278             /* The source and destination words don't overlap.  */
0279
0280             /* To improve branch prediction, one iteration of this
0281              * loop is unrolled.  Most matches are short and will
0282              * fail the first check.  But if that check passes, then
0283              * it becomes increasing likely that the match is long
0284              * and we'll need to continue copying.
0285              */
0286
0287             copy_unaligned_word(src, dst);
0288             src += WORDBYTES;
0289             dst += WORDBYTES;
0290
0291             if (dst < end) {
0292                 do {
0293                     copy_unaligned_word(src, dst);
0294                     src += WORDBYTES;
0295                     dst += WORDBYTES;
0296                 } while (dst < end);
0297             }
0298             return end;
0299         } else if (offset == 1) {
0300
0301             /* Offset 1 matches are equivalent to run-length
0302              * encoding of the previous byte.  This case is common
0303              * if the data contains many repeated bytes.
0304              */
0305             size_t v = repeat_byte(*(dst - 1));
0306
0307             do {
0308                 put_unaligned(v, (size_t *)dst);
0309                 src += WORDBYTES;
0310                 dst += WORDBYTES;
0311             } while (dst < end);
0312             return end;
0313         }
0314         /*
0315          * We don't bother with special cases for other 'offset <
0316          * WORDBYTES', which are usually rarer than 'offset == 1'.  Extra
0317          * checks will just slow things down.  Actually, it's possible
0318          * to handle all the 'offset < WORDBYTES' cases using the same
0319          * code, but it still becomes more complicated doesn't seem any
0320          * faster overall; it definitely slows down the more common
0321          * 'offset == 1' case.
0322          */
0323     }
0324 #endif /* FAST_UNALIGNED_ACCESS */
0325
0326     /* Fall back to a bytewise copy.  */
0327
0328     if (min_length >= 2) {
0329         *dst++ = *src++;
0330         length--;
0331     }
0332     if (min_length >= 3) {
0333         *dst++ = *src++;
0334         length--;
0335     }
0336     do {
0337         *dst++ = *src++;
0338     } while (--length);
0339
0340     return dst;
0341 }
0342
0343 #endif /* _LINUX_NTFS3_LIB_DECOMPRESS_COMMON_H */