Commit | Line | Data |
---|---|---|
f4bcbe79 GS |
1 | #ifndef __LINUX_STRINGHASH_H |
2 | #define __LINUX_STRINGHASH_H | |
3 | ||
fcfd2fbf GS |
4 | #include <linux/compiler.h> /* For __pure */ |
5 | #include <linux/types.h> /* For u32, u64 */ | |
703b5faf | 6 | #include <linux/hash.h> |
f4bcbe79 GS |
7 | |
8 | /* | |
9 | * Routines for hashing strings of bytes to a 32-bit hash value. | |
10 | * | |
11 | * These hash functions are NOT GUARANTEED STABLE between kernel | |
12 | * versions, architectures, or even repeated boots of the same kernel. | |
13 | * (E.g. they may depend on boot-time hardware detection or be | |
14 | * deliberately randomized.) | |
15 | * | |
16 | * They are also not intended to be secure against collisions caused by | |
17 | * malicious inputs; much slower hash functions are required for that. | |
18 | * | |
19 | * They are optimized for pathname components, meaning short strings. | |
20 | * Even if a majority of files have longer names, the dynamic profile of | |
21 | * pathname components skews short due to short directory names. | |
22 | * (E.g. /usr/lib/libsesquipedalianism.so.3.141.) | |
23 | */ | |
24 | ||
25 | /* | |
26 | * Version 1: one byte at a time. Example of use: | |
27 | * | |
28 | * unsigned long hash = init_name_hash; | |
29 | * while (*p) | |
30 | * hash = partial_name_hash(tolower(*p++), hash); | |
31 | * hash = end_name_hash(hash); | |
32 | * | |
33 | * Although this is designed for bytes, fs/hfsplus/unicode.c | |
34 | * abuses it to hash 16-bit values. | |
35 | */ | |
36 | ||
37 | /* Hash courtesy of the R5 hash in reiserfs modulo sign bits */ | |
8387ff25 | 38 | #define init_name_hash(salt) (unsigned long)(salt) |
f4bcbe79 GS |
39 | |
40 | /* partial hash update function. Assume roughly 4 bits per character */ | |
41 | static inline unsigned long | |
42 | partial_name_hash(unsigned long c, unsigned long prevhash) | |
43 | { | |
44 | return (prevhash + (c << 4) + (c >> 4)) * 11; | |
45 | } | |
46 | ||
47 | /* | |
48 | * Finally: cut down the number of bits to a int value (and try to avoid | |
703b5faf GS |
49 | * losing bits). This also has the property (wanted by the dcache) |
50 | * that the msbits make a good hash table index. | |
f4bcbe79 GS |
51 | */ |
52 | static inline unsigned long end_name_hash(unsigned long hash) | |
53 | { | |
703b5faf | 54 | return __hash_32((unsigned int)hash); |
f4bcbe79 GS |
55 | } |
56 | ||
57 | /* | |
58 | * Version 2: One word (32 or 64 bits) at a time. | |
59 | * If CONFIG_DCACHE_WORD_ACCESS is defined (meaning <asm/word-at-a-time.h> | |
60 | * exists, which describes major Linux platforms like x86 and ARM), then | |
61 | * this computes a different hash function much faster. | |
62 | * | |
63 | * If not set, this falls back to a wrapper around the preceding. | |
64 | */ | |
8387ff25 | 65 | extern unsigned int __pure full_name_hash(const void *salt, const char *, unsigned int); |
f4bcbe79 GS |
66 | |
67 | /* | |
68 | * A hash_len is a u64 with the hash of a string in the low | |
69 | * half and the length in the high half. | |
70 | */ | |
71 | #define hashlen_hash(hashlen) ((u32)(hashlen)) | |
72 | #define hashlen_len(hashlen) ((u32)((hashlen) >> 32)) | |
73 | #define hashlen_create(hash, len) ((u64)(len)<<32 | (u32)(hash)) | |
74 | ||
fcfd2fbf | 75 | /* Return the "hash_len" (hash and length) of a null-terminated string */ |
8387ff25 | 76 | extern u64 __pure hashlen_string(const void *salt, const char *name); |
fcfd2fbf | 77 | |
f4bcbe79 | 78 | #endif /* __LINUX_STRINGHASH_H */ |