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
29static forceinline void
30copy_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 */
39static 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 */
55struct 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.  */
73static 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 */
87static 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 */
104static forceinline u32
105bitstream_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 */
114static forceinline void
115bitstream_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 */
125static forceinline u32
126bitstream_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.  */
135static forceinline u32
136bitstream_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.  */
143static forceinline u8
144bitstream_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.  */
152static forceinline u16
153bitstream_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.  */
165static forceinline u32
166bitstream_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 */
181static 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.  */
192static forceinline void bitstream_align(struct input_bitstream *is)
193{
194	is->bitsleft = 0;
195	is->bitbuf = 0;
196}
197
198extern 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 */
208static 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 */
254static 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