rind 0.1.2

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
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293

// Copyright (c) 2014, Google Inc.
// SPDX-License-Identifier: ISC
//
// Poly1305 with NEON: aws-lc's `crypto/poly1305/poly1305_arm.c` at commit
// 91c93a8eb860e51482a871a8772cb9b1b310baea, in Rust, around the assembly
// `admin/import-aws-lc` imports beside it. Both come from the public
// domain neon2 version in SUPERCOP by D. J. Bernstein and Peter Schwabe,
// which computes two interleaved Poly1305 lanes in 26-bit limbs.
//
// Neither is formally verified: their correctness rests on their use and
// testing in BoringSSL and aws-lc, and on rind's tests of them against
// HACL*'s verified Poly1305, Wycheproof, and aws-lc-rs.

use super::cpu::HaveNeon;
use crate::low::{Zeroable, zeroise_value};

/// Two field elements in 26-bit limbs, interleaved: lane 0 in the even words
/// and lane 1 in the odd ones. The last two words only pad it to 16 bytes,
/// which the assembly's aligned loads need.
#[repr(C, align(16))]
#[derive(Clone, Copy)]
struct Fe1305x2 {
    v: [u32; 12],
}

static ZERO: Fe1305x2 = Fe1305x2 { v: [0; 12] };

unsafe extern "C" {
    /// `r = x * y + c`, per lane, modulo 2^130 - 5.
    #[link_name = aws_lc_symbol!("openssl_poly1305_neon2_addmulmod")]
    fn openssl_poly1305_neon2_addmulmod(
        r: *mut Fe1305x2,
        x: *const Fe1305x2,
        y: *const Fe1305x2,
        c: *const Fe1305x2,
    );

    /// Absorbs 32-byte pairs of blocks while more than 32 bytes remain, and
    /// returns how many remain.
    #[link_name = aws_lc_symbol!("openssl_poly1305_neon2_blocks")]
    fn openssl_poly1305_neon2_blocks(
        h: *mut Fe1305x2,
        precomp: *const [Fe1305x2; 2],
        input: *const u8,
        len: usize,
    ) -> usize;
}

// As in `aesv8_armx`: ARM mode with NEON, then the target's `.fpu` again.
core::arch::global_asm!(
    ".text",
    ".fpu neon",
    ".code 32",
    include_str!("aws_lc/poly1305_arm_asm.s"),
    ".fpu vfpv3",
    openssl_poly1305_neon2_blocks = sym openssl_poly1305_neon2_blocks,
    openssl_poly1305_neon2_addmulmod = sym openssl_poly1305_neon2_addmulmod,
);

fn addmulmod(r: *mut Fe1305x2, x: *const Fe1305x2, y: *const Fe1305x2, c: &Fe1305x2) {
    // SAFETY: every argument points to a live, aligned `Fe1305x2`, which the
    // routine allows `r` to share with `x`. Only a `HaveNeon` token builds the
    // state they come from.
    unsafe { openssl_poly1305_neon2_addmulmod(r, x, y, c) }
}

fn freeze(r: &mut Fe1305x2) {
    let mut x0 = r.v[0];
    let mut x1 = r.v[2];
    let mut x2 = r.v[4];
    let mut x3 = r.v[6];
    let mut x4 = r.v[8];

    for _ in 0..3 {
        x1 = x1.wrapping_add(x0 >> 26);
        x0 &= 0x3ffffff;
        x2 = x2.wrapping_add(x1 >> 26);
        x1 &= 0x3ffffff;
        x3 = x3.wrapping_add(x2 >> 26);
        x2 &= 0x3ffffff;
        x4 = x4.wrapping_add(x3 >> 26);
        x3 &= 0x3ffffff;
        x0 = x0.wrapping_add(5u32.wrapping_mul(x4 >> 26));
        x4 &= 0x3ffffff;
    }

    let mut y0 = x0.wrapping_add(5);
    let mut y1 = x1.wrapping_add(y0 >> 26);
    y0 &= 0x3ffffff;
    let mut y2 = x2.wrapping_add(y1 >> 26);
    y1 &= 0x3ffffff;
    let mut y3 = x3.wrapping_add(y2 >> 26);
    y2 &= 0x3ffffff;
    let mut y4 = x4.wrapping_add(y3 >> 26);
    y3 &= 0x3ffffff;
    let swap = 0u32.wrapping_sub(y4 >> 26);
    y4 &= 0x3ffffff;

    y0 ^= x0;
    y1 ^= x1;
    y2 ^= x2;
    y3 ^= x3;
    y4 ^= x4;

    y0 &= swap;
    y1 &= swap;
    y2 &= swap;
    y3 &= swap;
    y4 &= swap;

    y0 ^= x0;
    y1 ^= x1;
    y2 ^= x2;
    y3 ^= x3;
    y4 ^= x4;

    r.v[0] = y0;
    r.v[2] = y1;
    r.v[4] = y2;
    r.v[6] = y3;
    r.v[8] = y4;
}

fn load32(t: &[u8]) -> u32 {
    u32::from_le_bytes([t[0], t[1], t[2], t[3]])
}

fn fe1305x2_tobytearray(r: &mut [u8; 16], x: &Fe1305x2) {
    let mut x0 = x.v[0];
    let mut x1 = x.v[2];
    let mut x2 = x.v[4];
    let mut x3 = x.v[6];
    let mut x4 = x.v[8];

    x1 = x1.wrapping_add(x0 >> 26);
    x0 &= 0x3ffffff;
    x2 = x2.wrapping_add(x1 >> 26);
    x1 &= 0x3ffffff;
    x3 = x3.wrapping_add(x2 >> 26);
    x2 &= 0x3ffffff;
    x4 = x4.wrapping_add(x3 >> 26);
    x3 &= 0x3ffffff;

    r[0..4].copy_from_slice(&x0.wrapping_add(x1 << 26).to_le_bytes());
    r[4..8].copy_from_slice(&(x1 >> 6).wrapping_add(x2 << 20).to_le_bytes());
    r[8..12].copy_from_slice(&(x2 >> 12).wrapping_add(x3 << 14).to_le_bytes());
    r[12..16].copy_from_slice(&(x3 >> 18).wrapping_add(x4 << 8).to_le_bytes());
}

/// Loads up to two blocks from `x`, at most 32 bytes, each with its final 1
/// bit; lane 1 is zero when `x` has 16 bytes or fewer.
fn fe1305x2_frombytearray(r: &mut Fe1305x2, x: &[u8]) {
    let (first, second) = x.split_at(x.len().min(16));
    load_lane(r, 0, first);
    if second.is_empty() {
        for i in [1, 3, 5, 7, 9] {
            r.v[i] = 0;
        }
    } else {
        load_lane(r, 1, second);
    }
}

/// Loads `block`, at most 16 bytes, and its final 1 bit into `lane`.
fn load_lane(r: &mut Fe1305x2, lane: usize, block: &[u8]) {
    let mut t = [0u8; 17];
    t[..block.len()].copy_from_slice(block);
    t[block.len()] = 1;

    r.v[lane] = 0x3ffffff & load32(&t[0..]);
    r.v[lane + 2] = 0x3ffffff & (load32(&t[3..]) >> 2);
    r.v[lane + 4] = 0x3ffffff & (load32(&t[6..]) >> 4);
    r.v[lane + 6] = 0x3ffffff & (load32(&t[9..]) >> 6);
    r.v[lane + 8] = load32(&t[13..]);
}

pub(crate) struct Poly1305Neon {
    r: Fe1305x2,
    h: Fe1305x2,
    c: Fe1305x2,
    /// r^2 and r^4, in both lanes.
    precomp: [Fe1305x2; 2],
    buf: [u8; 32],
    buf_used: usize,
    key: [u8; 16],
}

impl Zeroable for Poly1305Neon {}

impl Poly1305Neon {
    pub(crate) fn new(key: &[u8; 32], _: HaveNeon) -> Self {
        let mut r = Fe1305x2 { v: [0; 12] };
        r.v[0] = 0x3ffffff & load32(&key[0..]);
        r.v[2] = 0x3ffff03 & (load32(&key[3..]) >> 2);
        r.v[4] = 0x3ffc0ff & (load32(&key[6..]) >> 4);
        r.v[6] = 0x3f03fff & (load32(&key[9..]) >> 6);
        r.v[8] = 0x00fffff & (load32(&key[12..]) >> 8);
        for i in [0, 2, 4, 6, 8] {
            r.v[i + 1] = r.v[i];
        }

        let mut st = Self {
            r,
            h: ZERO,
            c: ZERO,
            precomp: [ZERO; 2],
            buf: [0; 32],
            buf_used: 0,
            key: core::array::from_fn(|i| key[16 + i]),
        };
        let precomp = st.precomp.as_mut_ptr();
        addmulmod(precomp, &st.r, &st.r, &ZERO);
        // SAFETY: `precomp` has two elements.
        addmulmod(unsafe { precomp.add(1) }, precomp, precomp, &ZERO);
        st
    }

    pub(crate) fn add_bytes(&mut self, mut input: &[u8]) {
        if self.buf_used != 0 {
            let todo = (self.buf.len() - self.buf_used).min(input.len());
            let (head, rest) = input.split_at(todo);
            self.buf[self.buf_used..self.buf_used + todo].copy_from_slice(head);
            self.buf_used += todo;
            input = rest;

            if self.buf_used == self.buf.len() && !input.is_empty() {
                let h: *mut Fe1305x2 = &mut self.h;
                addmulmod(h, h, &self.precomp[0], &ZERO);
                fe1305x2_frombytearray(&mut self.c, &self.buf);
                for i in 0..10 {
                    self.h.v[i] = self.h.v[i].wrapping_add(self.c.v[i]);
                }
                self.buf_used = 0;
            }
        }

        while input.len() > 32 {
            let len = input.len().min(1 << 20);
            // SAFETY: `input` has `len` readable bytes, `self.h` is aligned,
            // and only a `HaveNeon` token builds `self`.
            let left = unsafe {
                openssl_poly1305_neon2_blocks(&mut self.h, &self.precomp, input.as_ptr(), len)
            };
            input = &input[len - left..];
        }

        if !input.is_empty() {
            self.buf[..input.len()].copy_from_slice(input);
            self.buf_used = input.len();
        }
    }

    pub(crate) fn finish(mut self) -> [u8; 16] {
        let h: *mut Fe1305x2 = &mut self.h;
        addmulmod(h, h, &self.precomp[0], &ZERO);

        if self.buf_used > 16 {
            fe1305x2_frombytearray(&mut self.c, &self.buf[..self.buf_used]);
            for i in [1, 3, 5, 7, 9] {
                self.precomp[0].v[i] = self.r.v[i];
            }
            addmulmod(h, h, &self.precomp[0], &self.c);
        } else if self.buf_used > 0 {
            fe1305x2_frombytearray(&mut self.c, &self.buf[..self.buf_used]);
            self.r.v[1] = 1;
            for i in [3, 5, 7, 9] {
                self.r.v[i] = 0;
            }
            addmulmod(h, h, &self.r, &self.c);
        }

        for i in [0, 2, 4, 6, 8] {
            self.h.v[i] = self.h.v[i].wrapping_add(self.h.v[i + 1]);
        }
        freeze(&mut self.h);

        fe1305x2_frombytearray(&mut self.c, &self.key);
        self.c.v[8] ^= 1 << 24;

        for i in [0, 2, 4, 6, 8] {
            self.h.v[i] = self.h.v[i].wrapping_add(self.c.v[i]);
        }
        let mut mac = [0u8; 16];
        fe1305x2_tobytearray(&mut mac, &self.h);
        mac
    }
}

impl Drop for Poly1305Neon {
    fn drop(&mut self) {
        zeroise_value(self);
    }
}