blob: 908d348ff7776b3ae966d86dd7372d7f367337a7 [file] [log] [blame]
Jason A. Donenfeld53e054b2019-08-17 00:01:12 +01001 SipHash - a short input PRF
2-----------------------------------------------
3Written by Jason A. Donenfeld <jason@zx2c4.com>
4
5SipHash is a cryptographically secure PRF -- a keyed hash function -- that
6performs very well for short inputs, hence the name. It was designed by
7cryptographers Daniel J. Bernstein and Jean-Philippe Aumasson. It is intended
8as a replacement for some uses of: `jhash`, `md5_transform`, `sha_transform`,
9and so forth.
10
11SipHash takes a secret key filled with randomly generated numbers and either
12an input buffer or several input integers. It spits out an integer that is
13indistinguishable from random. You may then use that integer as part of secure
14sequence numbers, secure cookies, or mask it off for use in a hash table.
15
161. Generating a key
17
18Keys should always be generated from a cryptographically secure source of
19random numbers, either using get_random_bytes or get_random_once:
20
21siphash_key_t key;
22get_random_bytes(&key, sizeof(key));
23
24If you're not deriving your key from here, you're doing it wrong.
25
262. Using the functions
27
28There are two variants of the function, one that takes a list of integers, and
29one that takes a buffer:
30
31u64 siphash(const void *data, size_t len, const siphash_key_t *key);
32
33And:
34
35u64 siphash_1u64(u64, const siphash_key_t *key);
36u64 siphash_2u64(u64, u64, const siphash_key_t *key);
37u64 siphash_3u64(u64, u64, u64, const siphash_key_t *key);
38u64 siphash_4u64(u64, u64, u64, u64, const siphash_key_t *key);
39u64 siphash_1u32(u32, const siphash_key_t *key);
40u64 siphash_2u32(u32, u32, const siphash_key_t *key);
41u64 siphash_3u32(u32, u32, u32, const siphash_key_t *key);
42u64 siphash_4u32(u32, u32, u32, u32, const siphash_key_t *key);
43
44If you pass the generic siphash function something of a constant length, it
45will constant fold at compile-time and automatically choose one of the
46optimized functions.
47
483. Hashtable key function usage:
49
50struct some_hashtable {
51 DECLARE_HASHTABLE(hashtable, 8);
52 siphash_key_t key;
53};
54
55void init_hashtable(struct some_hashtable *table)
56{
57 get_random_bytes(&table->key, sizeof(table->key));
58}
59
60static inline hlist_head *some_hashtable_bucket(struct some_hashtable *table, struct interesting_input *input)
61{
62 return &table->hashtable[siphash(input, sizeof(*input), &table->key) & (HASH_SIZE(table->hashtable) - 1)];
63}
64
65You may then iterate like usual over the returned hash bucket.
66
674. Security
68
69SipHash has a very high security margin, with its 128-bit key. So long as the
70key is kept secret, it is impossible for an attacker to guess the outputs of
71the function, even if being able to observe many outputs, since 2^128 outputs
72is significant.
73
74Linux implements the "2-4" variant of SipHash.
75
765. Struct-passing Pitfalls
77
78Often times the XuY functions will not be large enough, and instead you'll
79want to pass a pre-filled struct to siphash. When doing this, it's important
80to always ensure the struct has no padding holes. The easiest way to do this
81is to simply arrange the members of the struct in descending order of size,
82and to use offsetendof() instead of sizeof() for getting the size. For
83performance reasons, if possible, it's probably a good thing to align the
84struct to the right boundary. Here's an example:
85
86const struct {
87 struct in6_addr saddr;
88 u32 counter;
89 u16 dport;
90} __aligned(SIPHASH_ALIGNMENT) combined = {
91 .saddr = *(struct in6_addr *)saddr,
92 .counter = counter,
93 .dport = dport
94};
95u64 h = siphash(&combined, offsetofend(typeof(combined), dport), &secret);
96
976. Resources
98
99Read the SipHash paper if you're interested in learning more:
100https://131002.net/siphash/siphash.pdf
Jason A. Donenfeld175a4072019-08-17 00:01:19 +0100101
102
103~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~=~
104
105HalfSipHash - SipHash's insecure younger cousin
106-----------------------------------------------
107Written by Jason A. Donenfeld <jason@zx2c4.com>
108
109On the off-chance that SipHash is not fast enough for your needs, you might be
110able to justify using HalfSipHash, a terrifying but potentially useful
111possibility. HalfSipHash cuts SipHash's rounds down from "2-4" to "1-3" and,
112even scarier, uses an easily brute-forcable 64-bit key (with a 32-bit output)
113instead of SipHash's 128-bit key. However, this may appeal to some
114high-performance `jhash` users.
115
116Danger!
117
118Do not ever use HalfSipHash except for as a hashtable key function, and only
119then when you can be absolutely certain that the outputs will never be
120transmitted out of the kernel. This is only remotely useful over `jhash` as a
121means of mitigating hashtable flooding denial of service attacks.
122
1231. Generating a key
124
125Keys should always be generated from a cryptographically secure source of
126random numbers, either using get_random_bytes or get_random_once:
127
128hsiphash_key_t key;
129get_random_bytes(&key, sizeof(key));
130
131If you're not deriving your key from here, you're doing it wrong.
132
1332. Using the functions
134
135There are two variants of the function, one that takes a list of integers, and
136one that takes a buffer:
137
138u32 hsiphash(const void *data, size_t len, const hsiphash_key_t *key);
139
140And:
141
142u32 hsiphash_1u32(u32, const hsiphash_key_t *key);
143u32 hsiphash_2u32(u32, u32, const hsiphash_key_t *key);
144u32 hsiphash_3u32(u32, u32, u32, const hsiphash_key_t *key);
145u32 hsiphash_4u32(u32, u32, u32, u32, const hsiphash_key_t *key);
146
147If you pass the generic hsiphash function something of a constant length, it
148will constant fold at compile-time and automatically choose one of the
149optimized functions.
150
1513. Hashtable key function usage:
152
153struct some_hashtable {
154 DECLARE_HASHTABLE(hashtable, 8);
155 hsiphash_key_t key;
156};
157
158void init_hashtable(struct some_hashtable *table)
159{
160 get_random_bytes(&table->key, sizeof(table->key));
161}
162
163static inline hlist_head *some_hashtable_bucket(struct some_hashtable *table, struct interesting_input *input)
164{
165 return &table->hashtable[hsiphash(input, sizeof(*input), &table->key) & (HASH_SIZE(table->hashtable) - 1)];
166}
167
168You may then iterate like usual over the returned hash bucket.
169
1704. Performance
171
172HalfSipHash is roughly 3 times slower than JenkinsHash. For many replacements,
173this will not be a problem, as the hashtable lookup isn't the bottleneck. And
174in general, this is probably a good sacrifice to make for the security and DoS
175resistance of HalfSipHash.