| 1 | /* SPDX-License-Identifier: GPL-2.0 */ | 
|---|
| 2 | #ifndef __LINUX_STRINGHASH_H | 
|---|
| 3 | #define __LINUX_STRINGHASH_H | 
|---|
| 4 |  | 
|---|
| 5 | #include <linux/compiler.h>	/* For __pure */ | 
|---|
| 6 | #include <linux/types.h>	/* For u32, u64 */ | 
|---|
| 7 | #include <linux/hash.h> | 
|---|
| 8 |  | 
|---|
| 9 | /* | 
|---|
| 10 | * Routines for hashing strings of bytes to a 32-bit hash value. | 
|---|
| 11 | * | 
|---|
| 12 | * These hash functions are NOT GUARANTEED STABLE between kernel | 
|---|
| 13 | * versions, architectures, or even repeated boots of the same kernel. | 
|---|
| 14 | * (E.g. they may depend on boot-time hardware detection or be | 
|---|
| 15 | * deliberately randomized.) | 
|---|
| 16 | * | 
|---|
| 17 | * They are also not intended to be secure against collisions caused by | 
|---|
| 18 | * malicious inputs; much slower hash functions are required for that. | 
|---|
| 19 | * | 
|---|
| 20 | * They are optimized for pathname components, meaning short strings. | 
|---|
| 21 | * Even if a majority of files have longer names, the dynamic profile of | 
|---|
| 22 | * pathname components skews short due to short directory names. | 
|---|
| 23 | * (E.g. /usr/lib/libsesquipedalianism.so.3.141.) | 
|---|
| 24 | */ | 
|---|
| 25 |  | 
|---|
| 26 | /* | 
|---|
| 27 | * Version 1: one byte at a time.  Example of use: | 
|---|
| 28 | * | 
|---|
| 29 | * unsigned long hash = init_name_hash; | 
|---|
| 30 | * while (*p) | 
|---|
| 31 | *	hash = partial_name_hash(tolower(*p++), hash); | 
|---|
| 32 | * hash = end_name_hash(hash); | 
|---|
| 33 | * | 
|---|
| 34 | * Although this is designed for bytes, fs/hfsplus/unicode.c | 
|---|
| 35 | * abuses it to hash 16-bit values. | 
|---|
| 36 | */ | 
|---|
| 37 |  | 
|---|
| 38 | /* Hash courtesy of the R5 hash in reiserfs modulo sign bits */ | 
|---|
| 39 | #define init_name_hash(salt)		(unsigned long)(salt) | 
|---|
| 40 |  | 
|---|
| 41 | /* partial hash update function. Assume roughly 4 bits per character */ | 
|---|
| 42 | static inline unsigned long | 
|---|
| 43 | partial_name_hash(unsigned long c, unsigned long prevhash) | 
|---|
| 44 | { | 
|---|
| 45 | return (prevhash + (c << 4) + (c >> 4)) * 11; | 
|---|
| 46 | } | 
|---|
| 47 |  | 
|---|
| 48 | /* | 
|---|
| 49 | * Finally: cut down the number of bits to a int value (and try to avoid | 
|---|
| 50 | * losing bits).  This also has the property (wanted by the dcache) | 
|---|
| 51 | * that the msbits make a good hash table index. | 
|---|
| 52 | */ | 
|---|
| 53 | static inline unsigned int end_name_hash(unsigned long hash) | 
|---|
| 54 | { | 
|---|
| 55 | return hash_long(hash, 32); | 
|---|
| 56 | } | 
|---|
| 57 |  | 
|---|
| 58 | /* | 
|---|
| 59 | * Version 2: One word (32 or 64 bits) at a time. | 
|---|
| 60 | * If CONFIG_DCACHE_WORD_ACCESS is defined (meaning <asm/word-at-a-time.h> | 
|---|
| 61 | * exists, which describes major Linux platforms like x86 and ARM), then | 
|---|
| 62 | * this computes a different hash function much faster. | 
|---|
| 63 | * | 
|---|
| 64 | * If not set, this falls back to a wrapper around the preceding. | 
|---|
| 65 | */ | 
|---|
| 66 | extern unsigned int __pure full_name_hash(const void *salt, const char *, unsigned int); | 
|---|
| 67 |  | 
|---|
| 68 | /* | 
|---|
| 69 | * A hash_len is a u64 with the hash of a string in the low | 
|---|
| 70 | * half and the length in the high half. | 
|---|
| 71 | */ | 
|---|
| 72 | #define hashlen_hash(hashlen) ((u32)(hashlen)) | 
|---|
| 73 | #define hashlen_len(hashlen)  ((u32)((hashlen) >> 32)) | 
|---|
| 74 | #define hashlen_create(hash, len) ((u64)(len)<<32 | (u32)(hash)) | 
|---|
| 75 |  | 
|---|
| 76 | /* Return the "hash_len" (hash and length) of a null-terminated string */ | 
|---|
| 77 | extern u64 __pure hashlen_string(const void *salt, const char *name); | 
|---|
| 78 |  | 
|---|
| 79 | #endif	/* __LINUX_STRINGHASH_H */ | 
|---|
| 80 |  | 
|---|