1 | // Copyright 2019 The CryptoCorrosion Contributors |
2 | // Copyright 2020 Developers of the Rand project. |
3 | // |
4 | // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or |
5 | // https://www.apache.org/licenses/LICENSE-2.0> or the MIT license |
6 | // <LICENSE-MIT or https://opensource.org/licenses/MIT>, at your |
7 | // option. This file may not be copied, modified, or distributed |
8 | // except according to those terms. |
9 | |
10 | //! The ChaCha random number generator. |
11 | |
12 | use ppv_lite86::{dispatch, dispatch_light128}; |
13 | |
14 | pub use ppv_lite86::Machine; |
15 | use ppv_lite86::{vec128_storage, ArithOps, BitOps32, LaneWords4, MultiLane, StoreBytes, Vec4}; |
16 | |
17 | pub(crate) const BLOCK: usize = 64; |
18 | pub(crate) const BLOCK64: u64 = BLOCK as u64; |
19 | const LOG2_BUFBLOCKS: u64 = 2; |
20 | const BUFBLOCKS: u64 = 1 << LOG2_BUFBLOCKS; |
21 | pub(crate) const BUFSZ64: u64 = BLOCK64 * BUFBLOCKS; |
22 | pub(crate) const BUFSZ: usize = BUFSZ64 as usize; |
23 | |
24 | const STREAM_PARAM_NONCE: u32 = 1; |
25 | const STREAM_PARAM_BLOCK: u32 = 0; |
26 | |
27 | #[derive (Clone, PartialEq, Eq)] |
28 | pub struct ChaCha { |
29 | pub(crate) b: vec128_storage, |
30 | pub(crate) c: vec128_storage, |
31 | pub(crate) d: vec128_storage, |
32 | } |
33 | |
34 | #[derive (Clone)] |
35 | pub struct State<V> { |
36 | pub(crate) a: V, |
37 | pub(crate) b: V, |
38 | pub(crate) c: V, |
39 | pub(crate) d: V, |
40 | } |
41 | |
42 | #[inline (always)] |
43 | pub(crate) fn round<V: ArithOps + BitOps32>(mut x: State<V>) -> State<V> { |
44 | x.a += x.b; |
45 | x.d = (x.d ^ x.a).rotate_each_word_right16(); |
46 | x.c += x.d; |
47 | x.b = (x.b ^ x.c).rotate_each_word_right20(); |
48 | x.a += x.b; |
49 | x.d = (x.d ^ x.a).rotate_each_word_right24(); |
50 | x.c += x.d; |
51 | x.b = (x.b ^ x.c).rotate_each_word_right25(); |
52 | x |
53 | } |
54 | |
55 | #[inline (always)] |
56 | pub(crate) fn diagonalize<V: LaneWords4>(mut x: State<V>) -> State<V> { |
57 | x.b = x.b.shuffle_lane_words3012(); |
58 | x.c = x.c.shuffle_lane_words2301(); |
59 | x.d = x.d.shuffle_lane_words1230(); |
60 | x |
61 | } |
62 | #[inline (always)] |
63 | pub(crate) fn undiagonalize<V: LaneWords4>(mut x: State<V>) -> State<V> { |
64 | x.b = x.b.shuffle_lane_words1230(); |
65 | x.c = x.c.shuffle_lane_words2301(); |
66 | x.d = x.d.shuffle_lane_words3012(); |
67 | x |
68 | } |
69 | |
70 | impl ChaCha { |
71 | #[inline (always)] |
72 | pub fn new(key: &[u8; 32], nonce: &[u8]) -> Self { |
73 | init_chacha(key, nonce) |
74 | } |
75 | |
76 | #[inline (always)] |
77 | fn pos64<M: Machine>(&self, m: M) -> u64 { |
78 | let d: M::u32x4 = m.unpack(self.d); |
79 | ((d.extract(1) as u64) << 32) | d.extract(0) as u64 |
80 | } |
81 | |
82 | /// Produce 4 blocks of output, advancing the state |
83 | #[inline (always)] |
84 | pub fn refill4(&mut self, drounds: u32, out: &mut [u8; BUFSZ]) { |
85 | refill_wide(self, drounds, out) |
86 | } |
87 | |
88 | #[inline (always)] |
89 | pub fn set_block_pos(&mut self, value: u64) { |
90 | set_stream_param(self, STREAM_PARAM_BLOCK, value) |
91 | } |
92 | |
93 | #[inline (always)] |
94 | pub fn get_block_pos(&self) -> u64 { |
95 | get_stream_param(self, STREAM_PARAM_BLOCK) |
96 | } |
97 | |
98 | #[inline (always)] |
99 | pub fn set_nonce(&mut self, value: u64) { |
100 | set_stream_param(self, STREAM_PARAM_NONCE, value) |
101 | } |
102 | |
103 | #[inline (always)] |
104 | pub fn get_nonce(&self) -> u64 { |
105 | get_stream_param(self, STREAM_PARAM_NONCE) |
106 | } |
107 | |
108 | #[inline (always)] |
109 | pub fn get_seed(&self) -> [u8; 32] { |
110 | get_seed(self) |
111 | } |
112 | } |
113 | |
114 | #[allow (clippy::many_single_char_names)] |
115 | #[inline (always)] |
116 | fn refill_wide_impl<Mach: Machine>( |
117 | m: Mach, state: &mut ChaCha, drounds: u32, out: &mut [u8; BUFSZ], |
118 | ) { |
119 | let k = m.vec([0x6170_7865, 0x3320_646e, 0x7962_2d32, 0x6b20_6574]); |
120 | let mut pos = state.pos64(m); |
121 | let d0: Mach::u32x4 = m.unpack(state.d); |
122 | pos = pos.wrapping_add(1); |
123 | let d1 = d0.insert((pos >> 32) as u32, 1).insert(pos as u32, 0); |
124 | pos = pos.wrapping_add(1); |
125 | let d2 = d0.insert((pos >> 32) as u32, 1).insert(pos as u32, 0); |
126 | pos = pos.wrapping_add(1); |
127 | let d3 = d0.insert((pos >> 32) as u32, 1).insert(pos as u32, 0); |
128 | |
129 | let b = m.unpack(state.b); |
130 | let c = m.unpack(state.c); |
131 | let mut x = State { |
132 | a: Mach::u32x4x4::from_lanes([k, k, k, k]), |
133 | b: Mach::u32x4x4::from_lanes([b, b, b, b]), |
134 | c: Mach::u32x4x4::from_lanes([c, c, c, c]), |
135 | d: m.unpack(Mach::u32x4x4::from_lanes([d0, d1, d2, d3]).into()), |
136 | }; |
137 | for _ in 0..drounds { |
138 | x = round(x); |
139 | x = undiagonalize(round(diagonalize(x))); |
140 | } |
141 | let mut pos = state.pos64(m); |
142 | let d0: Mach::u32x4 = m.unpack(state.d); |
143 | pos = pos.wrapping_add(1); |
144 | let d1 = d0.insert((pos >> 32) as u32, 1).insert(pos as u32, 0); |
145 | pos = pos.wrapping_add(1); |
146 | let d2 = d0.insert((pos >> 32) as u32, 1).insert(pos as u32, 0); |
147 | pos = pos.wrapping_add(1); |
148 | let d3 = d0.insert((pos >> 32) as u32, 1).insert(pos as u32, 0); |
149 | pos = pos.wrapping_add(1); |
150 | let d4 = d0.insert((pos >> 32) as u32, 1).insert(pos as u32, 0); |
151 | |
152 | let (a, b, c, d) = ( |
153 | x.a.to_lanes(), |
154 | x.b.to_lanes(), |
155 | x.c.to_lanes(), |
156 | x.d.to_lanes(), |
157 | ); |
158 | let sb = m.unpack(state.b); |
159 | let sc = m.unpack(state.c); |
160 | let sd = [m.unpack(state.d), d1, d2, d3]; |
161 | state.d = d4.into(); |
162 | let mut words = out.chunks_exact_mut(16); |
163 | for ((((&a, &b), &c), &d), &sd) in a.iter().zip(&b).zip(&c).zip(&d).zip(&sd) { |
164 | (a + k).write_le(words.next().unwrap()); |
165 | (b + sb).write_le(words.next().unwrap()); |
166 | (c + sc).write_le(words.next().unwrap()); |
167 | (d + sd).write_le(words.next().unwrap()); |
168 | } |
169 | } |
170 | |
171 | dispatch!(m, Mach, { |
172 | fn refill_wide(state: &mut ChaCha, drounds: u32, out: &mut [u8; BUFSZ]) { |
173 | refill_wide_impl(m, state, drounds, out); |
174 | } |
175 | }); |
176 | |
177 | // Single-block, rounds-only; shared by try_apply_keystream for tails shorter than BUFSZ |
178 | // and XChaCha's setup step. |
179 | dispatch!(m, Mach, { |
180 | fn refill_narrow_rounds(state: &mut ChaCha, drounds: u32) -> State<vec128_storage> { |
181 | let k: Mach::u32x4 = m.vec([0x6170_7865, 0x3320_646e, 0x7962_2d32, 0x6b20_6574]); |
182 | let mut x = State { |
183 | a: k, |
184 | b: m.unpack(state.b), |
185 | c: m.unpack(state.c), |
186 | d: m.unpack(state.d), |
187 | }; |
188 | for _ in 0..drounds { |
189 | x = round(x); |
190 | x = undiagonalize(round(diagonalize(x))); |
191 | } |
192 | State { |
193 | a: x.a.into(), |
194 | b: x.b.into(), |
195 | c: x.c.into(), |
196 | d: x.d.into(), |
197 | } |
198 | } |
199 | }); |
200 | |
201 | dispatch_light128!(m, Mach, { |
202 | fn set_stream_param(state: &mut ChaCha, param: u32, value: u64) { |
203 | let d: Mach::u32x4 = m.unpack(state.d); |
204 | state.d = d |
205 | .insert((value >> 32) as u32, (param << 1) | 1) |
206 | .insert(value as u32, param << 1) |
207 | .into(); |
208 | } |
209 | }); |
210 | |
211 | dispatch_light128!(m, Mach, { |
212 | fn get_stream_param(state: &ChaCha, param: u32) -> u64 { |
213 | let d: Mach::u32x4 = m.unpack(state.d); |
214 | ((d.extract((param << 1) | 1) as u64) << 32) | d.extract(param << 1) as u64 |
215 | } |
216 | }); |
217 | |
218 | dispatch_light128!(m, Mach, { |
219 | fn get_seed(state: &ChaCha) -> [u8; 32] { |
220 | let b: Mach::u32x4 = m.unpack(state.b); |
221 | let c: Mach::u32x4 = m.unpack(state.c); |
222 | let mut key = [0u8; 32]; |
223 | b.write_le(&mut key[..16]); |
224 | c.write_le(&mut key[16..]); |
225 | key |
226 | } |
227 | }); |
228 | |
229 | fn read_u32le(xs: &[u8]) -> u32 { |
230 | assert_eq!(xs.len(), 4); |
231 | u32::from(xs[0]) | (u32::from(xs[1]) << 8) | (u32::from(xs[2]) << 16) | (u32::from(xs[3]) << 24) |
232 | } |
233 | |
234 | dispatch_light128!(m, Mach, { |
235 | fn init_chacha(key: &[u8; 32], nonce: &[u8]) -> ChaCha { |
236 | let ctr_nonce = [ |
237 | 0, |
238 | if nonce.len() == 12 { |
239 | read_u32le(&nonce[0..4]) |
240 | } else { |
241 | 0 |
242 | }, |
243 | read_u32le(&nonce[nonce.len() - 8..nonce.len() - 4]), |
244 | read_u32le(&nonce[nonce.len() - 4..]), |
245 | ]; |
246 | let key0: Mach::u32x4 = m.read_le(&key[..16]); |
247 | let key1: Mach::u32x4 = m.read_le(&key[16..]); |
248 | ChaCha { |
249 | b: key0.into(), |
250 | c: key1.into(), |
251 | d: ctr_nonce.into(), |
252 | } |
253 | } |
254 | }); |
255 | |
256 | dispatch_light128!(m, Mach, { |
257 | fn init_chacha_x(key: &[u8; 32], nonce: &[u8; 24], rounds: u32) -> ChaCha { |
258 | let key0: Mach::u32x4 = m.read_le(&key[..16]); |
259 | let key1: Mach::u32x4 = m.read_le(&key[16..]); |
260 | let nonce0: Mach::u32x4 = m.read_le(&nonce[..16]); |
261 | let mut state = ChaCha { |
262 | b: key0.into(), |
263 | c: key1.into(), |
264 | d: nonce0.into(), |
265 | }; |
266 | let x = refill_narrow_rounds(&mut state, rounds); |
267 | let ctr_nonce1 = [0, 0, read_u32le(&nonce[16..20]), read_u32le(&nonce[20..24])]; |
268 | state.b = x.a; |
269 | state.c = x.d; |
270 | state.d = ctr_nonce1.into(); |
271 | state |
272 | } |
273 | }); |
274 | |