xref: /openbmc/linux/fs/ntfs3/lib/decompress_common.h (revision f05643a0)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3  * decompress_common.h - Code shared by the XPRESS and LZX decompressors
4  *
5  * Copyright (C) 2015 Eric Biggers
6  */
7 
8 #ifndef _LINUX_NTFS3_LIB_DECOMPRESS_COMMON_H
9 #define _LINUX_NTFS3_LIB_DECOMPRESS_COMMON_H
10 
11 #include <linux/string.h>
12 #include <linux/compiler.h>
13 #include <linux/types.h>
14 #include <linux/slab.h>
15 #include <asm/unaligned.h>
16 
17 
18 /* "Force inline" macro (not required, but helpful for performance)  */
19 #define forceinline __always_inline
20 
21 /* Enable whole-word match copying on selected architectures  */
22 #if defined(__i386__) || defined(__x86_64__) || defined(__ARM_FEATURE_UNALIGNED)
23 #  define FAST_UNALIGNED_ACCESS
24 #endif
25 
26 /* Size of a machine word  */
27 #define WORDBYTES (sizeof(size_t))
28 
29 static forceinline void
30 copy_unaligned_word(const void *src, void *dst)
31 {
32 	put_unaligned(get_unaligned((const size_t *)src), (size_t *)dst);
33 }
34 
35 
36 /* Generate a "word" with platform-dependent size whose bytes all contain the
37  * value 'b'.
38  */
39 static forceinline size_t repeat_byte(u8 b)
40 {
41 	size_t v;
42 
43 	v = b;
44 	v |= v << 8;
45 	v |= v << 16;
46 	v |= v << ((WORDBYTES == 8) ? 32 : 0);
47 	return v;
48 }
49 
50 /* Structure that encapsulates a block of in-memory data being interpreted as a
51  * stream of bits, optionally with interwoven literal bytes.  Bits are assumed
52  * to be stored in little endian 16-bit coding units, with the bits ordered high
53  * to low.
54  */
55 struct input_bitstream {
56 
57 	/* Bits that have been read from the input buffer.  The bits are
58 	 * left-justified; the next bit is always bit 31.
59 	 */
60 	u32 bitbuf;
61 
62 	/* Number of bits currently held in @bitbuf.  */
63 	u32 bitsleft;
64 
65 	/* Pointer to the next byte to be retrieved from the input buffer.  */
66 	const u8 *next;
67 
68 	/* Pointer to just past the end of the input buffer.  */
69 	const u8 *end;
70 };
71 
72 /* Initialize a bitstream to read from the specified input buffer.  */
73 static forceinline void init_input_bitstream(struct input_bitstream *is,
74 					     const void *buffer, u32 size)
75 {
76 	is->bitbuf = 0;
77 	is->bitsleft = 0;
78 	is->next = buffer;
79 	is->end = is->next + size;
80 }
81 
82 /* Ensure the bit buffer variable for the bitstream contains at least @num_bits
83  * bits.  Following this, bitstream_peek_bits() and/or bitstream_remove_bits()
84  * may be called on the bitstream to peek or remove up to @num_bits bits.  Note
85  * that @num_bits must be <= 16.
86  */
87 static forceinline void bitstream_ensure_bits(struct input_bitstream *is,
88 					      u32 num_bits)
89 {
90 	if (is->bitsleft < num_bits) {
91 		if (is->end - is->next >= 2) {
92 			is->bitbuf |= (u32)get_unaligned_le16(is->next)
93 					<< (16 - is->bitsleft);
94 			is->next += 2;
95 		}
96 		is->bitsleft += 16;
97 	}
98 }
99 
100 /* Return the next @num_bits bits from the bitstream, without removing them.
101  * There must be at least @num_bits remaining in the buffer variable, from a
102  * previous call to bitstream_ensure_bits().
103  */
104 static forceinline u32
105 bitstream_peek_bits(const struct input_bitstream *is, const u32 num_bits)
106 {
107 	return (is->bitbuf >> 1) >> (sizeof(is->bitbuf) * 8 - num_bits - 1);
108 }
109 
110 /* Remove @num_bits from the bitstream.  There must be at least @num_bits
111  * remaining in the buffer variable, from a previous call to
112  * bitstream_ensure_bits().
113  */
114 static forceinline void
115 bitstream_remove_bits(struct input_bitstream *is, u32 num_bits)
116 {
117 	is->bitbuf <<= num_bits;
118 	is->bitsleft -= num_bits;
119 }
120 
121 /* Remove and return @num_bits bits from the bitstream.  There must be at least
122  * @num_bits remaining in the buffer variable, from a previous call to
123  * bitstream_ensure_bits().
124  */
125 static forceinline u32
126 bitstream_pop_bits(struct input_bitstream *is, u32 num_bits)
127 {
128 	u32 bits = bitstream_peek_bits(is, num_bits);
129 
130 	bitstream_remove_bits(is, num_bits);
131 	return bits;
132 }
133 
134 /* Read and return the next @num_bits bits from the bitstream.  */
135 static forceinline u32
136 bitstream_read_bits(struct input_bitstream *is, u32 num_bits)
137 {
138 	bitstream_ensure_bits(is, num_bits);
139 	return bitstream_pop_bits(is, num_bits);
140 }
141 
142 /* Read and return the next literal byte embedded in the bitstream.  */
143 static forceinline u8
144 bitstream_read_byte(struct input_bitstream *is)
145 {
146 	if (unlikely(is->end == is->next))
147 		return 0;
148 	return *is->next++;
149 }
150 
151 /* Read and return the next 16-bit integer embedded in the bitstream.  */
152 static forceinline u16
153 bitstream_read_u16(struct input_bitstream *is)
154 {
155 	u16 v;
156 
157 	if (unlikely(is->end - is->next < 2))
158 		return 0;
159 	v = get_unaligned_le16(is->next);
160 	is->next += 2;
161 	return v;
162 }
163 
164 /* Read and return the next 32-bit integer embedded in the bitstream.  */
165 static forceinline u32
166 bitstream_read_u32(struct input_bitstream *is)
167 {
168 	u32 v;
169 
170 	if (unlikely(is->end - is->next < 4))
171 		return 0;
172 	v = get_unaligned_le32(is->next);
173 	is->next += 4;
174 	return v;
175 }
176 
177 /* Read into @dst_buffer an array of literal bytes embedded in the bitstream.
178  * Return either a pointer to the byte past the last written, or NULL if the
179  * read overflows the input buffer.
180  */
181 static forceinline void *bitstream_read_bytes(struct input_bitstream *is,
182 					      void *dst_buffer, size_t count)
183 {
184 	if ((size_t)(is->end - is->next) < count)
185 		return NULL;
186 	memcpy(dst_buffer, is->next, count);
187 	is->next += count;
188 	return (u8 *)dst_buffer + count;
189 }
190 
191 /* Align the input bitstream on a coding-unit boundary.  */
192 static forceinline void bitstream_align(struct input_bitstream *is)
193 {
194 	is->bitsleft = 0;
195 	is->bitbuf = 0;
196 }
197 
198 extern int make_huffman_decode_table(u16 decode_table[], const u32 num_syms,
199 				     const u32 num_bits, const u8 lens[],
200 				     const u32 max_codeword_len,
201 				     u16 working_space[]);
202 
203 
204 /* Reads and returns the next Huffman-encoded symbol from a bitstream.  If the
205  * input data is exhausted, the Huffman symbol is decoded as if the missing bits
206  * are all zeroes.
207  */
208 static forceinline u32 read_huffsym(struct input_bitstream *istream,
209 					 const u16 decode_table[],
210 					 u32 table_bits,
211 					 u32 max_codeword_len)
212 {
213 	u32 entry;
214 	u32 key_bits;
215 
216 	bitstream_ensure_bits(istream, max_codeword_len);
217 
218 	/* Index the decode table by the next table_bits bits of the input.  */
219 	key_bits = bitstream_peek_bits(istream, table_bits);
220 	entry = decode_table[key_bits];
221 	if (entry < 0xC000) {
222 		/* Fast case: The decode table directly provided the
223 		 * symbol and codeword length.  The low 11 bits are the
224 		 * symbol, and the high 5 bits are the codeword length.
225 		 */
226 		bitstream_remove_bits(istream, entry >> 11);
227 		return entry & 0x7FF;
228 	}
229 	/* Slow case: The codeword for the symbol is longer than
230 	 * table_bits, so the symbol does not have an entry
231 	 * directly in the first (1 << table_bits) entries of the
232 	 * decode table.  Traverse the appropriate binary tree
233 	 * bit-by-bit to decode the symbol.
234 	 */
235 	bitstream_remove_bits(istream, table_bits);
236 	do {
237 		key_bits = (entry & 0x3FFF) + bitstream_pop_bits(istream, 1);
238 	} while ((entry = decode_table[key_bits]) >= 0xC000);
239 	return entry;
240 }
241 
242 /*
243  * Copy an LZ77 match at (dst - offset) to dst.
244  *
245  * The length and offset must be already validated --- that is, (dst - offset)
246  * can't underrun the output buffer, and (dst + length) can't overrun the output
247  * buffer.  Also, the length cannot be 0.
248  *
249  * @bufend points to the byte past the end of the output buffer.  This function
250  * won't write any data beyond this position.
251  *
252  * Returns dst + length.
253  */
254 static forceinline u8 *lz_copy(u8 *dst, u32 length, u32 offset, const u8 *bufend,
255 			       u32 min_length)
256 {
257 	const u8 *src = dst - offset;
258 
259 	/*
260 	 * Try to copy one machine word at a time.  On i386 and x86_64 this is
261 	 * faster than copying one byte at a time, unless the data is
262 	 * near-random and all the matches have very short lengths.  Note that
263 	 * since this requires unaligned memory accesses, it won't necessarily
264 	 * be faster on every architecture.
265 	 *
266 	 * Also note that we might copy more than the length of the match.  For
267 	 * example, if a word is 8 bytes and the match is of length 5, then
268 	 * we'll simply copy 8 bytes.  This is okay as long as we don't write
269 	 * beyond the end of the output buffer, hence the check for (bufend -
270 	 * end >= WORDBYTES - 1).
271 	 */
272 #ifdef FAST_UNALIGNED_ACCESS
273 	u8 * const end = dst + length;
274 
275 	if (bufend - end >= (ptrdiff_t)(WORDBYTES - 1)) {
276 
277 		if (offset >= WORDBYTES) {
278 			/* The source and destination words don't overlap.  */
279 
280 			/* To improve branch prediction, one iteration of this
281 			 * loop is unrolled.  Most matches are short and will
282 			 * fail the first check.  But if that check passes, then
283 			 * it becomes increasing likely that the match is long
284 			 * and we'll need to continue copying.
285 			 */
286 
287 			copy_unaligned_word(src, dst);
288 			src += WORDBYTES;
289 			dst += WORDBYTES;
290 
291 			if (dst < end) {
292 				do {
293 					copy_unaligned_word(src, dst);
294 					src += WORDBYTES;
295 					dst += WORDBYTES;
296 				} while (dst < end);
297 			}
298 			return end;
299 		} else if (offset == 1) {
300 
301 			/* Offset 1 matches are equivalent to run-length
302 			 * encoding of the previous byte.  This case is common
303 			 * if the data contains many repeated bytes.
304 			 */
305 			size_t v = repeat_byte(*(dst - 1));
306 
307 			do {
308 				put_unaligned(v, (size_t *)dst);
309 				src += WORDBYTES;
310 				dst += WORDBYTES;
311 			} while (dst < end);
312 			return end;
313 		}
314 		/*
315 		 * We don't bother with special cases for other 'offset <
316 		 * WORDBYTES', which are usually rarer than 'offset == 1'.  Extra
317 		 * checks will just slow things down.  Actually, it's possible
318 		 * to handle all the 'offset < WORDBYTES' cases using the same
319 		 * code, but it still becomes more complicated doesn't seem any
320 		 * faster overall; it definitely slows down the more common
321 		 * 'offset == 1' case.
322 		 */
323 	}
324 #endif /* FAST_UNALIGNED_ACCESS */
325 
326 	/* Fall back to a bytewise copy.  */
327 
328 	if (min_length >= 2) {
329 		*dst++ = *src++;
330 		length--;
331 	}
332 	if (min_length >= 3) {
333 		*dst++ = *src++;
334 		length--;
335 	}
336 	do {
337 		*dst++ = *src++;
338 	} while (--length);
339 
340 	return dst;
341 }
342 
343 #endif /* _LINUX_NTFS3_LIB_DECOMPRESS_COMMON_H */
344