rind 0.1.7

Pure-Rust cryptography for MoQ, forked from graviola
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206

// Written for Graviola by Joe Birr-Pixton, 2024.
// SPDX-License-Identifier: Apache-2.0 OR ISC OR MIT-0

//! SHA-256 compression with the SHA-256 instructions where the CPU has them,
//! and aws-lc's scalar `sha256-armv8` routine otherwise, imported by
//! `admin/import-aws-lc`. aws-lc's is not formally verified: its correctness
//! rests on its use and testing in OpenSSL, BoringSSL, and aws-lc, and on
//! rind's tests of it against the instructions, CAVP, and aws-lc-rs.

use core::arch::aarch64::*;

use crate::low::aarch64::cpu::{self, HaveCryptoExt};

pub(crate) fn sha256_compress_blocks(state: &mut [u32; 8], blocks: &[u8]) {
    debug_assert!(blocks.len().is_multiple_of(64));
    match HaveCryptoExt::check() {
        // SAFETY: the token proves the CPU has the `sha2` cpu feature, and
        // rind requires `neon`.
        Some(_) => unsafe { sha256(state, blocks) },
        // The routine runs its loop at least once, so zero blocks would read
        // past `blocks`.
        None if blocks.is_empty() => {}
        // SAFETY: `blocks` is `blocks.len() / 64` readable blocks.
        None => unsafe { sha256_block_data_order_nohw(state, blocks.as_ptr(), blocks.len() / 64) },
    }
}

/// The round constants, then a zero that ends the scalar routine's rounds.
#[repr(C, align(64))]
struct Table([u32; 65]);

static K256: Table = Table(include!("aws_lc/sha256-armv8.K256.rs"));

#[unsafe(naked)]
unsafe extern "C" fn sha256_block_data_order_nohw(
    state: *mut [u32; 8],
    data: *const u8,
    blocks: usize,
) {
    #[cfg(not(target_vendor = "apple"))]
    core::arch::naked_asm!(
        include_str!("aws_lc/sha256_block_data_order_nohw.linux.s"),
        K256 = sym K256,
    );
    #[cfg(target_vendor = "apple")]
    core::arch::naked_asm!(
        include_str!("aws_lc/sha256_block_data_order_nohw.apple.s"),
        K256 = sym K256,
    );
}

macro_rules! k {
    ($k:literal) => {
        // SAFETY: values for `$k` are compile-time constants
        unsafe { vld1q_u32(K.0.as_ptr().add($k * 4)) }
    };
}

macro_rules! round {
    ($msg0:ident, $msg1:ident, $msg2:ident, $msg3:ident, $state0:ident, $state1:ident, $k:ident) => {
        let t0 = vaddq_u32($msg0, $k);
        let t1 = $state0;
        $state0 = vsha256hq_u32($state0, $state1, t0);
        $state1 = vsha256h2q_u32($state1, t1, t0);
        let $msg0 = vsha256su0q_u32($msg0, $msg1);
        let $msg0 = vsha256su1q_u32($msg0, $msg2, $msg3);
    };

    ($msg:ident, $state0:ident, $state1:ident, $k:ident) => {
        let t0 = vaddq_u32($msg, $k);
        let t1 = $state0;
        $state0 = vsha256hq_u32($state0, $state1, t0);
        $state1 = vsha256h2q_u32($state1, t1, t0);
    };
}

#[target_feature(enable = "neon,sha2")]
fn sha256(state: &mut [u32; 8], blocks: &[u8]) {
    // SAFETY: `state` is readable, and 8 words in length
    let (mut state0, mut state1) = unsafe {
        (
            vld1q_u32(state[0..4].as_ptr()),
            vld1q_u32(state[4..8].as_ptr()),
        )
    };

    let k0 = k!(0);
    let k1 = k!(1);
    let k2 = k!(2);
    let k3 = k!(3);
    let k4 = k!(4);
    let k5 = k!(5);
    let k6 = k!(6);
    let k7 = k!(7);
    let k8 = k!(8);
    let k9 = k!(9);
    let k10 = k!(10);
    let k11 = k!(11);
    let k12 = k!(12);
    let k13 = k!(13);
    let k14 = k!(14);
    let k15 = k!(15);

    let (chunks, _) = blocks.as_chunks::<64>();
    for block in chunks {
        let state0_prev = state0;
        let state1_prev = state1;

        // prefetch next block
        // SAFETY: prefetches don't constitute an access and are not
        // archtecturally visible.
        unsafe { cpu::prefetch_ro(block.as_ptr().add(64)) };

        // SAFETY: `block` is 64 bytes long and readable, via `chunks_exact`
        let (msg0, msg1, msg2, msg3) = unsafe {
            (
                vld1q_u32(block[0..].as_ptr().cast()),
                vld1q_u32(block[16..].as_ptr().cast()),
                vld1q_u32(block[32..].as_ptr().cast()),
                vld1q_u32(block[48..].as_ptr().cast()),
            )
        };

        let msg0 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(msg0)));
        let msg1 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(msg1)));
        let msg2 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(msg2)));
        let msg3 = vreinterpretq_u32_u8(vrev32q_u8(vreinterpretq_u8_u32(msg3)));

        round!(msg0, msg1, msg2, msg3, state0, state1, k0);
        round!(msg1, msg2, msg3, msg0, state0, state1, k1);
        round!(msg2, msg3, msg0, msg1, state0, state1, k2);
        round!(msg3, msg0, msg1, msg2, state0, state1, k3);

        round!(msg0, msg1, msg2, msg3, state0, state1, k4);
        round!(msg1, msg2, msg3, msg0, state0, state1, k5);
        round!(msg2, msg3, msg0, msg1, state0, state1, k6);
        round!(msg3, msg0, msg1, msg2, state0, state1, k7);

        round!(msg0, msg1, msg2, msg3, state0, state1, k8);
        round!(msg1, msg2, msg3, msg0, state0, state1, k9);
        round!(msg2, msg3, msg0, msg1, state0, state1, k10);
        round!(msg3, msg0, msg1, msg2, state0, state1, k11);

        round!(msg0, state0, state1, k12);
        round!(msg1, state0, state1, k13);
        round!(msg2, state0, state1, k14);
        round!(msg3, state0, state1, k15);

        state0 = vaddq_u32(state0, state0_prev);
        state1 = vaddq_u32(state1, state1_prev);
    }

    // SAFETY: `state` is writable, and 8 words in length
    unsafe {
        vst1q_u32(state[0..4].as_mut_ptr(), state0);
        vst1q_u32(state[4..8].as_mut_ptr(), state1);
    }
}

#[repr(align(16))]
struct Aligned([u32; 64]);

static K: Aligned = Aligned([
    0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5, 0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5,
    0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3, 0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174,
    0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc, 0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da,
    0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7, 0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967,
    0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13, 0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85,
    0xa2bfe8a1, 0xa81a664b, 0xc24b8b70, 0xc76c51a3, 0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070,
    0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5, 0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3,
    0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208, 0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2,
]);

#[cfg(test)]
#[cfg_attr(coverage_nightly, coverage(off))]
mod tests {
    use super::*;

    // Every other test sees only the routine the CPU picks, so the two are
    // checked against each other here: odd and even block counts, with
    // distinct bytes in every block.
    #[test]
    fn hw_matches_nohw() {
        if HaveCryptoExt::check().is_none() {
            return;
        }
        let blocks: Vec<u8> = (0..5 * 64u32)
            .map(|i| (i.wrapping_mul(0x9e3779b9) >> 24) as u8)
            .collect();
        for n in 1..=5 {
            let blocks = &blocks[..n * 64];
            let mut want = [
                0x6a09e667, 0xbb67ae85, 0x3c6ef372, 0xa54ff53a, 0x510e527f, 0x9b05688c, 0x1f83d9ab,
                0x5be0cd19,
            ];
            let mut got = want;
            // SAFETY: `blocks` is `n` blocks, and the CPU has the SHA-256
            // instructions.
            unsafe {
                sha256(&mut want, blocks);
                sha256_block_data_order_nohw(&mut got, blocks.as_ptr(), n);
            }
            assert_eq!(got, want, "{n} blocks");
        }
    }
}