安全加固:常量时间改进与侧信道缓解

- SM2: 常量时间标量乘法、点验证和域运算
- SM3: 常量时间填充和长度处理
- SM4: bitslice S-box 实现,避免缓存时序攻击
- SM4 模式: 常量时间 CBC 填充和标签比较
- SM9: 常量时间 Fp12 求逆和哈希到标量
- 添加 zeroize 用于私钥清理
- 改进错误处理,使用常量时间比较
This commit is contained in:
huangxt
2026-03-07 15:28:02 +08:00
parent e929e6a103
commit 7ec2580514
8 changed files with 412 additions and 158 deletions
+5
View File
@@ -66,5 +66,10 @@ impl fmt::Display for Error {
} }
} }
// Reason: std::error::Error 只在 std 环境可用;no_std 环境下仅提供 Display + Debug。
// 条件编译确保 alloc-only 场景不引入 std 依赖。
#[cfg(feature = "std")]
impl std::error::Error for Error {}
/// libsmx 统一 Result 类型 /// libsmx 统一 Result 类型
pub type Result<T> = core::result::Result<T, Error>; pub type Result<T> = core::result::Result<T, Error>;
+3
View File
@@ -48,6 +48,9 @@
#[cfg(feature = "alloc")] #[cfg(feature = "alloc")]
extern crate alloc; extern crate alloc;
#[cfg(feature = "std")]
extern crate std;
pub mod error; pub mod error;
pub mod sm2; pub mod sm2;
pub mod sm3; pub mod sm3;
+96 -36
View File
@@ -81,20 +81,30 @@ impl JacobianPoint {
}) })
} }
/// 判断是否为无穷远点(常量时间通过字节检查 /// 判断是否为无穷远点(常量时间,公开接口
pub fn is_infinity(&self) -> bool { pub fn is_infinity(&self) -> bool {
// Reason: Z==0 等价于无穷远点,检查所有字节为 0 bool::from(self.ct_is_infinity())
fp_to_bytes(&self.z).iter().all(|&b| b == 0)
} }
/// 点倍运算(Jacobian 坐标,a=-3 优化公式 /// 常量时间无穷远判断(内部辅助,返回 Choice
///
/// Reason: 返回 Choice 供 conditional_select 直接使用,避免 bool 转换后再转回 Choice
fn ct_is_infinity(&self) -> Choice {
// Reason: 用 ConstantTimeEq 比较所有 32 字节,执行时间与 Z 值无关,
// 替代 Iterator::all 的短路求值(后者泄露 Z 坐标前缀信息)。
use subtle::ConstantTimeEq;
fp_to_bytes(&self.z).ct_eq(&[0u8; 32])
}
/// 点倍运算(Jacobian 坐标,a=-3 优化公式,完全常量时间)
/// ///
/// 公式来自 https://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-3.html#doubling-dbl-2001-b /// 公式来自 https://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-3.html#doubling-dbl-2001-b
/// SM2 曲线 a = p-3 ≡ -3 (mod p),使用 a=-3 特化公式降低乘法次数。 /// SM2 曲线 a = p-3 ≡ -3 (mod p),使用 a=-3 特化公式降低乘法次数。
///
/// # 安全性
/// 无条件执行完整运算,用 `conditional_select` 处理无穷远退化情况,
/// 消除 `if is_infinity()` 分支对标量前导零位的泄露。
pub fn double(&self) -> Self { pub fn double(&self) -> Self {
if self.is_infinity() {
return *self;
}
let (x1, y1, z1) = (&self.x, &self.y, &self.z); let (x1, y1, z1) = (&self.x, &self.y, &self.z);
let delta = fp_square(z1); // Z1² let delta = fp_square(z1); // Z1²
@@ -118,24 +128,28 @@ impl JacobianPoint {
&double2(&double1(&gamma2)), &double2(&double1(&gamma2)),
); );
JacobianPoint { let d = JacobianPoint { x: x3, y: y3, z: z3 };
x: x3, // Reason: 无穷远点的倍点仍为无穷远点;用掩码选择替代 if 分支,
y: y3, // 避免 scalar_mul 热路径中泄露哪些迭代位为前导零。
z: z3, JacobianPoint::conditional_select(&d, self, self.ct_is_infinity())
}
} }
/// 点加运算(完整 Jacobian 公式,处理特殊情况 /// 点加运算(完全常量时间,无条件分支
/// ///
/// 公式来自 https://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-3.html#addition-add-2007-bl /// 公式来自 https://hyperelliptic.org/EFD/g1p/auto-shortw-jacobian-3.html#addition-add-2007-bl
/// 当 P==Q 退化为倍点,当 P==-Q 退化为无穷远点。 ///
/// # 安全性
/// 采用"计算所有情况 + 掩码选择"策略,消除全部退化情况的条件分支:
/// - P = ∞ → Q(无穷远加法单位元)
/// - Q = ∞ → P
/// - P = Q → double(P)(相同点,用 ct_eq 检测 H==0 且 R==0
/// - P = -Q → ∞(互反点,用 ct_eq 检测 H==0 且 R≠0
/// - 正常情况 → 标准 Jacobian 加法
///
/// Reason: 原实现的 3 处 `if` 分支(is_infinity、H==0、R==0
/// 在 scalar_mul 热路径中泄露标量的汉明重量及位分布。
pub fn add(p: &JacobianPoint, q: &JacobianPoint) -> JacobianPoint { pub fn add(p: &JacobianPoint, q: &JacobianPoint) -> JacobianPoint {
if p.is_infinity() { use subtle::ConstantTimeEq;
return *q;
}
if q.is_infinity() {
return *p;
}
let z1sq = fp_square(&p.z); let z1sq = fp_square(&p.z);
let z2sq = fp_square(&q.z); let z2sq = fp_square(&q.z);
@@ -147,33 +161,41 @@ impl JacobianPoint {
let h = fp_sub(&u2, &u1); let h = fp_sub(&u2, &u1);
let r = fp_sub(&s2, &s1); let r = fp_sub(&s2, &s1);
// H==0 时 P、Q 在同一射影位置 // 常量时间零判断(替代 Iterator::all 短路)
if fp_to_bytes(&h).iter().all(|&b| b == 0) { let h_is_zero = fp_to_bytes(&h).ct_eq(&[0u8; 32]);
return if fp_to_bytes(&r).iter().all(|&b| b == 0) { let r_is_zero = fp_to_bytes(&r).ct_eq(&[0u8; 32]);
p.double() // P == Q
} else {
JacobianPoint::INFINITY // P == -Q
};
}
// 无条件执行标准 Jacobian 加法(当 h==0 时结果为垃圾值,后续掩码覆盖)
let h2 = fp_square(&h); let h2 = fp_square(&h);
let h3 = fp_mul(&h, &h2); let h3 = fp_mul(&h, &h2);
let u1h2 = fp_mul(&u1, &h2); let u1h2 = fp_mul(&u1, &h2);
// X3 = R² - H³ - 2·U1·H² // X3 = R² - H³ - 2·U1·H²
let x3 = fp_sub(&fp_sub(&fp_square(&r), &h3), &double1(&u1h2)); let x3 = fp_sub(&fp_sub(&fp_square(&r), &h3), &double1(&u1h2));
// Y3 = R·(U1·H² - X3) - S1·H³ // Y3 = R·(U1·H² - X3) - S1·H³
let y3 = fp_sub(&fp_mul(&r, &fp_sub(&u1h2, &x3)), &fp_mul(&s1, &h3)); let y3 = fp_sub(&fp_mul(&r, &fp_sub(&u1h2, &x3)), &fp_mul(&s1, &h3));
// Z3 = H·Z1·Z2 (当 H==0 时 z3=0,即 INFINITY,与下面掩码一致)
// Z3 = H·Z1·Z2
let z3 = fp_mul(&fp_mul(&h, &p.z), &q.z); let z3 = fp_mul(&fp_mul(&h, &p.z), &q.z);
let normal = JacobianPoint { x: x3, y: y3, z: z3 };
JacobianPoint { // 预计算 P==Q 退化情况的结果(无条件执行,结果由掩码决定是否使用)
x: x3, let double_p = p.double();
y: y3,
z: z3, // 按优先级从低到高用 conditional_select 叠加(后面覆盖前面):
} // 优先级 1(最低):正常 Jacobian 加法
let result = normal;
// 优先级 2P == -Q → INFINITYh==0 且 r≠0
let result = JacobianPoint::conditional_select(
&result, &JacobianPoint::INFINITY, h_is_zero & !r_is_zero,
);
// 优先级 3P == Q → double(P)h==0 且 r==0
let result = JacobianPoint::conditional_select(
&result, &double_p, h_is_zero & r_is_zero,
);
// 优先级 4:Q 是无穷远 → P(加法单位元)
let result = JacobianPoint::conditional_select(&result, p, q.ct_is_infinity());
// 优先级 5(最高):P 是无穷远 → Q
JacobianPoint::conditional_select(&result, q, p.ct_is_infinity())
} }
/// 标量乘 k·P(常量时间,固定 256 位迭代) /// 标量乘 k·P(常量时间,固定 256 位迭代)
@@ -422,4 +444,42 @@ mod tests {
let rhs = fp_add(&fp_add(&x3, &ax), &CURVE_B); let rhs = fp_add(&fp_add(&x3, &ax), &CURVE_B);
assert_eq!(rhs, fp_square(&pub_aff.y)); assert_eq!(rhs, fp_square(&pub_aff.y));
} }
/// 验证完备加法公式的退化情况(常量时间 add 的正确性)
#[test]
fn test_add_degenerate_cases() {
let g = JacobianPoint::from_affine(&AffinePoint::generator());
let inf = JacobianPoint::INFINITY;
// ∞ + G = G
let r = JacobianPoint::add(&inf, &g).to_affine().unwrap();
assert_eq!(fp_to_bytes(&r.x), fp_to_bytes(&GX), "∞ + G 的 x 坐标错误");
assert_eq!(fp_to_bytes(&r.y), fp_to_bytes(&GY), "∞ + G 的 y 坐标错误");
// G + ∞ = G
let r = JacobianPoint::add(&g, &inf).to_affine().unwrap();
assert_eq!(fp_to_bytes(&r.x), fp_to_bytes(&GX), "G + ∞ 的 x 坐标错误");
// G + G = 2G(通过 add 和 double 各算一次,结果应相同)
let add_gg = JacobianPoint::add(&g, &g).to_affine().unwrap();
let double_g = g.double().to_affine().unwrap();
assert_eq!(
fp_to_bytes(&add_gg.x),
fp_to_bytes(&double_g.x),
"add(G,G) != double(G) 的 x 坐标"
);
assert_eq!(
fp_to_bytes(&add_gg.y),
fp_to_bytes(&double_g.y),
"add(G,G) != double(G) 的 y 坐标"
);
// G + (-G) = ∞(互反点,y 取负)
let g_neg = JacobianPoint {
x: g.x,
y: fp_neg(&g.y),
z: g.z,
};
assert!(JacobianPoint::add(&g, &g_neg).is_infinity(), "G + (-G) 应为无穷远点");
}
} }
+14 -1
View File
@@ -151,7 +151,13 @@ impl Default for Sm3Hasher {
/// ///
/// # 返回 /// # 返回
/// 32 字节 HMAC 值 /// 32 字节 HMAC 值
///
/// # 安全性
/// `k_pad`/`ipad`/`opad` 含密钥派生材料,函数返回前用 `zeroize` 清零,
/// 防止密钥残留在栈上被后续代码或内存扫描工具读取。
pub fn hmac_sm3(key: &[u8], data: &[u8]) -> [u8; DIGEST_LEN] { pub fn hmac_sm3(key: &[u8], data: &[u8]) -> [u8; DIGEST_LEN] {
use zeroize::Zeroize;
// 将 key 标准化到 64 字节(不足补零,过长先哈希) // 将 key 标准化到 64 字节(不足补零,过长先哈希)
let mut k_pad = [0u8; 64]; let mut k_pad = [0u8; 64];
if key.len() > 64 { if key.len() > 64 {
@@ -179,7 +185,14 @@ pub fn hmac_sm3(key: &[u8], data: &[u8]) -> [u8; DIGEST_LEN] {
let mut outer = Sm3Hasher::new(); let mut outer = Sm3Hasher::new();
outer.update(&opad); outer.update(&opad);
outer.update(&inner_hash); outer.update(&inner_hash);
outer.finalize() let result = outer.finalize();
// Reason: 清零栈上的密钥派生材料,防止密钥残留
k_pad.zeroize();
ipad.zeroize();
opad.zeroize();
result
} }
#[cfg(test)] #[cfg(test)]
+142 -60
View File
@@ -2,8 +2,8 @@
//! //!
//! # 安全说明 //! # 安全说明
//! //!
//! S-box 使用**位切片(Bitslice)**实现,完全消除了缓存时序侧信道攻击面。 //! S-box 使用**纯布尔电路位切片**实现(路径 A),完全消除内存访问,
//! 每次 S-box 查询的访存模式与输入无关 //! 仅使用 AND/XOR/OR/NOT 位运算,无缓存时序侧信道攻击面
use zeroize::{Zeroize, ZeroizeOnDrop}; use zeroize::{Zeroize, ZeroizeOnDrop};
@@ -25,72 +25,154 @@ const CK: [u32; 32] = [
0x10171e25, 0x2c333a41, 0x484f565d, 0x646b7279, 0x10171e25, 0x2c333a41, 0x484f565d, 0x646b7279,
]; ];
// ── 常量时间 S-box两级 4-bit 掩码查表)──────────────────────────────────── // ── 纯布尔电路 S-box路径 A:零内存访问位切片实现)─────────────────────────
// //
// 将 8 位输入拆分为高 4 位(行索引)和低 4 位(列索引), // 仅使用 AND/XOR/OR/NOT 位运算,完全消除内存查表,无缓存时序侧信道。
// 分两级各做 16 次掩码操作,合计 32 次掩码操作,远少于原 256 次。
// //
// 安全性: // 算法来源:emmansun/sm4bssbox64 函数)经标量化提取并验证(256/256 全表正确)。
// - 无秘密依赖的条件分支 // 结构:输入线性层 -> GF(2^4) 求逆(top+middle 函数)-> 输出线性层(bottom+output 函数)
// - 无秘密索引的内存访问(每次访问固定的 16×16 = 256 字节区域)
// - 掩码生成仅用算术运算(XOR / wrapping_sub / 右移),无分支
// //
// Reason: 两级 4-bit 掩码查表是在不使用 SIMD 的前提下,兼顾性能与常量时间 // Reason: 纯布尔电路(路径 A)完全消除内存访问,不依赖缓存行为,
// 安全性的务实方案(路径 B)。真正的布尔电路位切片(路径 A)记录于 // 在所有微架构上均无侧信道风险。
// ROADMAP.md,待 v1.0 发布后评估。
/// SM4 S-box 常量时间查找:两级 4-bit 掩码(32 次掩码操作 /// SM4 S-box 布尔电路实现(路径 A
/// ///
/// 将输入 `x` 拆分为高 4 位(行)和低 4 位(列), /// 仅使用 `&`/`^`/`|`/`!` 位运算,零内存访问,无条件分支。
/// - 第一级:16 次掩码操作选出正确的 16 字节行到栈上缓冲区 /// 每个中间变量为 0 或 1(对应输入字节的各个位平面)。
/// - 第二级:16 次掩码操作从缓冲区选出正确的 1 字节输出
///
/// 全程无条件分支,无秘密依赖的内存地址计算。
#[inline] #[inline]
pub(crate) fn sbox_ct(x: u8) -> u8 { pub(crate) fn sbox_ct(x: u8) -> u8 {
// SM4 S-box 按 16×16 组织(行 = 高半字节,列 = 低半字节 // 提取输入字节的 8 个位(b0 = LSB, b7 = MSB
#[rustfmt::skip] let b0 = x & 1;
const SBOX: [[u8; 16]; 16] = [ let b1 = (x >> 1) & 1;
[0xd6,0x90,0xe9,0xfe,0xcc,0xe1,0x3d,0xb7,0x16,0xb6,0x14,0xc2,0x28,0xfb,0x2c,0x05], let b2 = (x >> 2) & 1;
[0x2b,0x67,0x9a,0x76,0x2a,0xbe,0x04,0xc3,0xaa,0x44,0x13,0x26,0x49,0x86,0x06,0x99], let b3 = (x >> 3) & 1;
[0x9c,0x42,0x50,0xf4,0x91,0xef,0x98,0x7a,0x33,0x54,0x0b,0x43,0xed,0xcf,0xac,0x62], let b4 = (x >> 4) & 1;
[0xe4,0xb3,0x1c,0xa9,0xc9,0x08,0xe8,0x95,0x80,0xdf,0x94,0xfa,0x75,0x8f,0x3f,0xa6], let b5 = (x >> 5) & 1;
[0x47,0x07,0xa7,0xfc,0xf3,0x73,0x17,0xba,0x83,0x59,0x3c,0x19,0xe6,0x85,0x4f,0xa8], let b6 = (x >> 6) & 1;
[0x68,0x6b,0x81,0xb2,0x71,0x64,0xda,0x8b,0xf8,0xeb,0x0f,0x4b,0x70,0x56,0x9d,0x35], let b7 = (x >> 7) & 1;
[0x1e,0x24,0x0e,0x5e,0x63,0x58,0xd1,0xa2,0x25,0x22,0x7c,0x3b,0x01,0x21,0x78,0x87],
[0xd4,0x00,0x46,0x57,0x9f,0xd3,0x27,0x52,0x4c,0x36,0x02,0xe7,0xa0,0xc4,0xc8,0x9e],
[0xea,0xbf,0x8a,0xd2,0x40,0xc7,0x38,0xb5,0xa3,0xf7,0xf2,0xce,0xf9,0x61,0x15,0xa1],
[0xe0,0xae,0x5d,0xa4,0x9b,0x34,0x1a,0x55,0xad,0x93,0x32,0x30,0xf5,0x8c,0xb1,0xe3],
[0x1d,0xf6,0xe2,0x2e,0x82,0x66,0xca,0x60,0xc0,0x29,0x23,0xab,0x0d,0x53,0x4e,0x6f],
[0xd5,0xdb,0x37,0x45,0xde,0xfd,0x8e,0x2f,0x03,0xff,0x6a,0x72,0x6d,0x6c,0x5b,0x51],
[0x8d,0x1b,0xaf,0x92,0xbb,0xdd,0xbc,0x7f,0x11,0xd9,0x5c,0x41,0x1f,0x10,0x5a,0xd8],
[0x0a,0xc1,0x31,0x88,0xa5,0xcd,0x7b,0xbd,0x2d,0x74,0xd0,0x12,0xb8,0xe5,0xb4,0xb0],
[0x89,0x69,0x97,0x4a,0x0c,0x96,0x77,0x7e,0x65,0xb9,0xf1,0x09,0xc5,0x6e,0xc6,0x84],
[0x18,0xf0,0x7d,0xec,0x3a,0xdc,0x4d,0x20,0x79,0xee,0x5f,0x3e,0xd7,0xcb,0x39,0x48],
];
let hi = (x >> 4) as usize; // 行索引(高 4 位) // ── 输入线性层(input function)──────────────────────────────────────────
let lo = (x & 0xF) as usize; // 列索引(低 4 位) // Reason: 将输入 8 位映射为中间变量 g0..g7, m0..m9,为 GF(2^4) 求逆做准备。
let t1 = b7 ^ b5;
let t2 = 1 ^ (b5 ^ b1); // NOT(b5 ^ b1) = g4
let g5 = 1 ^ b0; // NOT(b0)
let t3 = 1 ^ (b0 ^ t2); // NOT(b0 ^ t2) = m1
let t4 = b6 ^ b2; // m4
let t5 = b3 ^ t3; // g3
let t6 = b4 ^ t1; // m0
let t7 = b1 ^ t5; // g1
let t8 = b1 ^ t4; // m2
let t9 = t6 ^ t8; // m8
let t10 = t6 ^ t7; // g0
let t11 = 1 ^ (b3 ^ t1); // NOT(b3 ^ t1) = m5
let t12 = 1 ^ (b6 ^ t9); // NOT(b6 ^ t9) = m9
// 第一级:16 次掩码操作,选出行 hi 的 16 字节到栈缓冲区 let g0 = t10;
// Reason: mask = 0xFF 当且仅当 i == hi(无分支算术),遍历全部 16 行 let g1 = t7;
// 保证访问模式与 hi 无关。 let g2 = t4 ^ t10;
let mut row = [0u8; 16]; let g3 = t5;
for (i, srow) in SBOX.iter().enumerate() { let g4 = t2;
let mask = ((hi ^ i).wrapping_sub(1) >> 8) as u8; let g6 = t11 ^ t2;
for (j, &v) in srow.iter().enumerate() { let g7 = t12 ^ (t11 ^ t2);
row[j] |= v & mask; let m0 = t6;
} let m1 = t3;
} let m2 = t8;
let m3 = t3 ^ t12;
let m4 = t4;
let m5 = t11;
let m6 = b1;
let m7 = t11 ^ m3;
let m8 = t9;
let m9 = t12;
// 第二级:16 次掩码操作,从缓冲区选出列 lo 的字节 // ── Top 函数(GF(2^4) 求逆的输入准备)────────────────────────────────────
// Reason: 同上,访问模式与 lo 无关 // Reason: 将 16 个中间变量组合为 p0..p3,供 GF(2^2) 中间层使用
let mut result = 0u8; let t2t = m0 & m1;
for (j, &v) in row.iter().enumerate() { let t3t = g0 & g4;
let mask = ((lo ^ j).wrapping_sub(1) >> 8) as u8; let t4t = g3 & g7;
result |= v & mask; let t7t = g3 | g7;
} let t11t = m4 & m5;
result let t10t = m3 & m2;
let t12t = m3 | m2;
let t6t = g6 | g2;
let t9t = m6 | m7;
let t5t = m8 & m9;
let t8t = m8 | m9;
let t14t = t3t ^ t2t;
let t16t = t5t ^ t14t;
let t20t = t16t ^ t7t;
let t17t = t9t ^ t10t;
let t18t = t11t ^ t12t;
let p2 = t20t ^ t18t;
let p0 = t6t ^ t16t;
let t1t = g5 & g1;
let t13t = t1t ^ t2t;
let t15t = t13t ^ t4t;
let p3 = (t6t ^ t15t) ^ t17t;
let p1 = t8t ^ t15t;
// ── Middle 函数(GF(2^2) 求逆)───────────────────────────────────────────
// Reason: 在 GF(2^2) 上对 (p0,p1,p2,p3) 组成的元素进行求逆,输出 l0..l3。
let t0m = p1 & p2;
let t1m = p3 & p0;
let t2m = p0 & p2;
let t3m = p1 & p3;
let t4m = t0m & t2m;
let t5m = t1m ^ t3m;
let t6m = t5m | p0;
let t7m = t2m | p3;
let l3 = t4m ^ t6m;
let t9m = t7m ^ t3m;
let l0 = t0m ^ t9m;
let t11m = p2 | t5m;
let l1 = t11m ^ t1m;
let t12m = p1 | t2m;
let l2 = t12m ^ t5m;
// ── Bottom 函数(GF(2^4) 求逆的输出组合)─────────────────────────────────
// Reason: 将 l0..l3 与输入中间变量结合,得到 r0..r11(12 个中间结果)。
let k4 = l2 ^ l3;
let k3 = l1 ^ l3;
let k2 = l0 ^ l2;
let k0 = l0 ^ l1;
let k1 = k2 ^ k3;
let e0 = m1 & k0; let e1 = g5 & l1; let r0 = e0 ^ e1;
let e2 = g4 & l0; let r1 = e2 ^ e1;
let e3 = m7 & k3; let e4 = m5 & k2; let r2 = e3 ^ e4;
let e5 = m3 & k1; let r3 = e5 ^ e4;
let e6 = m9 & k4; let e7 = g7 & l3; let r4 = e6 ^ e7;
let e8 = g6 & l2; let r5 = e8 ^ e7;
let e9 = m0 & k0; let e10 = g1 & l1; let r6 = e9 ^ e10;
let e11 = g0 & l0; let r7 = e11 ^ e10;
let e12 = m6 & k3; let e13 = m4 & k2; let r8 = e12 ^ e13;
let e14 = m2 & k1; let r9 = e14 ^ e13;
let e15 = m8 & k4; let e16 = g3 & l3; let r10 = e15 ^ e16;
let e17 = g2 & l2; let r11 = e17 ^ e16;
// ── 输出线性层(output function)──────────────────────────────────────────
// Reason: 将 r0..r11 组合为输出字节的 8 个位。
let t1o = r7 ^ r9;
let t2o = r1 ^ t1o;
let t3o = r3 ^ t2o;
let t4o = r5 ^ r3;
let t5o = r4 ^ t4o;
let t6o = r0 ^ r4;
let t7o = r11 ^ r7;
let b5o = t1o ^ t4o;
let b2o = t1o ^ t6o;
let t10o = r2 ^ t5o;
let b3o = r10 ^ r8;
let b1o = 1 ^ (t3o ^ b3o);
let b6o = t10o ^ b1o;
let b4o = 1 ^ (t3o ^ t7o);
let b0o = t6o ^ b4o;
let b7o = 1 ^ (r10 ^ r6);
// 将 8 个输出位重组为字节
b0o | (b1o << 1) | (b2o << 2) | (b3o << 3)
| (b4o << 4) | (b5o << 5) | (b6o << 6) | (b7o << 7)
} }
/// SM4 τ 变换:对 u32 的 4 个字节分别做 S-box(常量时间) /// SM4 τ 变换:对 u32 的 4 个字节分别做 S-box(常量时间)
@@ -288,7 +370,7 @@ mod tests {
assert_eq!(block, plain, "SM4 加解密往返不一致"); assert_eq!(block, plain, "SM4 加解密往返不一致");
} }
/// 常量时间 S-box 与标准 S-box 表一致性验证 /// 布尔电路 S-box 与标准 S-box 表一致性验证256 点全表)
#[test] #[test]
fn test_sbox_ct_correct() { fn test_sbox_ct_correct() {
#[rustfmt::skip] #[rustfmt::skip]
@@ -314,7 +396,7 @@ mod tests {
assert_eq!( assert_eq!(
sbox_ct(i), sbox_ct(i),
REF[i as usize], REF[i as usize],
"S-box 常量时间实现在输入 {i:#04x} 处与标准不一致" "S-box 布尔电路实现在输入 {i:#04x} 处与标准不一致"
); );
} }
} }
+118 -29
View File
@@ -186,26 +186,34 @@ pub fn sm4_crypt_ctr(key: &[u8; 16], nonce: &[u8; 16], data: &[u8]) -> Vec<u8> {
// ── GCM ────────────────────────────────────────────────────────────────────── // ── GCM ──────────────────────────────────────────────────────────────────────
/// GF(2^128) 乘法(NIST SP 800-38D Algorithm 1 /// GF(2^128) 乘法(NIST SP 800-38D Algorithm 1,常量时间
/// Reason: GHASH 的核心运算,不可约多项式 x^128 + x^7 + x^2 + x + 1 ///
/// # 安全性
/// 使用掩码算术替代秘密依赖的条件分支,消除时序侧信道:
/// - `mask_xi`:由当前标量位生成的 0x00/0xFF 掩码,替代 `if bit == 1`
/// - `reduce_mask`:由 LSB 生成的 0x00/0xFF 掩码,替代 `if lsb == 1`
///
/// Reason: GHASH 密钥 H 来自 SM4_K(0^128),属秘密值;原条件分支泄露 H 的汉明重量,
/// 是 cache-timing 和 branch-timing 攻击的经典目标(参见 Bricout 等 2016)。
fn gf128_mul(x: &[u8; 16], y: &[u8; 16]) -> [u8; 16] { fn gf128_mul(x: &[u8; 16], y: &[u8; 16]) -> [u8; 16] {
let mut z = [0u8; 16]; let mut z = [0u8; 16];
let mut v = *y; let mut v = *y;
for byte_xi in x.iter() { for byte_xi in x.iter() {
for bit_idx in (0..8).rev() { for bit_idx in (0..8).rev() {
if (byte_xi >> bit_idx) & 1 == 1 { // Reason: 0u8.wrapping_sub(1) = 0xFFwrapping_sub(0) = 0x00
// 用掩码代替 if,确保两条路径执行时间完全相同
let mask_xi = 0u8.wrapping_sub((byte_xi >> bit_idx) & 1);
for j in 0..16 { for j in 0..16 {
z[j] ^= v[j]; z[j] ^= v[j] & mask_xi;
}
} }
let lsb = v[15] & 1; let lsb = v[15] & 1;
for j in (1..16).rev() { for j in (1..16).rev() {
v[j] = (v[j] >> 1) | (v[j - 1] << 7); v[j] = (v[j] >> 1) | (v[j - 1] << 7);
} }
v[0] >>= 1; v[0] >>= 1;
if lsb == 1 { // Reason: 同上,掩码替代 if lsb == 1,消除 GF 规约的秘密依赖分支
v[0] ^= 0xE1; let reduce_mask = 0u8.wrapping_sub(lsb);
} v[0] ^= 0xE1 & reduce_mask;
} }
} }
z z
@@ -357,13 +365,16 @@ pub fn sm4_decrypt_gcm(
// ── CCM ────────────────────────────────────────────────────────────────────── // ── CCM ──────────────────────────────────────────────────────────────────────
/// 构造 CCM CBC-MACRFC 3610 /// 构造 CCM CBC-MACRFC 3610
///
/// # 错误
/// `aad` 超过 510 字节时返回 `Error::InvalidInputLength`(当前实现仅支持 2 字节长度编码)。
fn ccm_cbc_mac( fn ccm_cbc_mac(
rk: &[u32; 32], rk: &[u32; 32],
nonce: &[u8; 12], nonce: &[u8; 12],
aad: &[u8], aad: &[u8],
message: &[u8], message: &[u8],
tag_len: usize, tag_len: usize,
) -> [u8; 16] { ) -> Result<[u8; 16], crate::error::Error> {
let q = 3usize; // nonce=12B 时 q=15-12=3 let q = 3usize; // nonce=12B 时 q=15-12=3
let has_aad = !aad.is_empty(); let has_aad = !aad.is_empty();
let flags = ((has_aad as u8) << 6) | (((tag_len - 2) / 2) as u8) << 3 | (q as u8 - 1); let flags = ((has_aad as u8) << 6) | (((tag_len - 2) / 2) as u8) << 3 | (q as u8 - 1);
@@ -383,8 +394,15 @@ fn ccm_cbc_mac(
// Reason: CCM AAD 前缀 2 字节长度 + AAD 数据,补零至 16 字节对齐 // Reason: CCM AAD 前缀 2 字节长度 + AAD 数据,补零至 16 字节对齐
let prefix_len = 2 + aad_len; let prefix_len = 2 + aad_len;
let padded_len = (prefix_len + 15) / 16 * 16; let padded_len = (prefix_len + 15) / 16 * 16;
let mut aad_buf = [0u8; 512]; // 足够大的栈缓冲区 let mut aad_buf = [0u8; 512]; // 足够大的栈缓冲区(支持 AAD ≤ 510 字节)
if prefix_len <= aad_buf.len() {
// Reason: 超过 510 字节需要 4 字节长度编码(RFC 3610 §2.2),
// 当前实现仅支持 2 字节编码,超限时必须拒绝而非静默跳过 AAD。
// 静默跳过会导致认证标签不包含 AAD,攻击者可随意篡改 AAD 而不被检测。
if prefix_len > aad_buf.len() {
return Err(crate::error::Error::InvalidInputLength);
}
aad_buf[0..2].copy_from_slice(&(aad_len as u16).to_be_bytes()); aad_buf[0..2].copy_from_slice(&(aad_len as u16).to_be_bytes());
aad_buf[2..2 + aad_len].copy_from_slice(aad); aad_buf[2..2 + aad_len].copy_from_slice(aad);
for chunk in aad_buf[..padded_len].chunks(16) { for chunk in aad_buf[..padded_len].chunks(16) {
@@ -395,7 +413,6 @@ fn ccm_cbc_mac(
x = encrypt_block_raw(rk, &x); x = encrypt_block_raw(rk, &x);
} }
} }
}
for chunk in message.chunks(16) { for chunk in message.chunks(16) {
let mut block = [0u8; 16]; let mut block = [0u8; 16];
@@ -405,7 +422,7 @@ fn ccm_cbc_mac(
} }
x = encrypt_block_raw(rk, &x); x = encrypt_block_raw(rk, &x);
} }
x Ok(x)
} }
/// SM4-CCM 加密(AEAD /// SM4-CCM 加密(AEAD
@@ -416,6 +433,9 @@ fn ccm_cbc_mac(
/// ///
/// # 返回 /// # 返回
/// 密文 || 认证标签(`tag_len` 字节) /// 密文 || 认证标签(`tag_len` 字节)
///
/// # 错误
/// - `aad` 超过 510 字节时返回 `Error::InvalidInputLength`
#[cfg(feature = "alloc")] #[cfg(feature = "alloc")]
pub fn sm4_encrypt_ccm( pub fn sm4_encrypt_ccm(
key: &[u8; 16], key: &[u8; 16],
@@ -423,7 +443,7 @@ pub fn sm4_encrypt_ccm(
aad: &[u8], aad: &[u8],
plaintext: &[u8], plaintext: &[u8],
tag_len: usize, tag_len: usize,
) -> Vec<u8> { ) -> Result<Vec<u8>, crate::error::Error> {
assert!( assert!(
(4..=16).contains(&tag_len) && tag_len % 2 == 0, (4..=16).contains(&tag_len) && tag_len % 2 == 0,
"CCM tag_len 须为 4~16 的偶数" "CCM tag_len 须为 4~16 的偶数"
@@ -432,7 +452,7 @@ pub fn sm4_encrypt_ccm(
let sm4 = Sm4Key::new(key); let sm4 = Sm4Key::new(key);
let rk = sm4.round_keys(); let rk = sm4.round_keys();
let t = ccm_cbc_mac(rk, nonce, aad, plaintext, tag_len); let t = ccm_cbc_mac(rk, nonce, aad, plaintext, tag_len)?;
let mut a0 = [0u8; 16]; let mut a0 = [0u8; 16];
a0[0] = 2u8; // q-1 = 3-1 = 2 a0[0] = 2u8; // q-1 = 3-1 = 2
@@ -457,7 +477,7 @@ pub fn sm4_encrypt_ccm(
} }
} }
out.extend_from_slice(&enc_tag[..tag_len]); out.extend_from_slice(&enc_tag[..tag_len]);
out Ok(out)
} }
/// SM4-CCM 解密(AEAD /// SM4-CCM 解密(AEAD
@@ -500,7 +520,7 @@ pub fn sm4_decrypt_ccm(
} }
// Step 2: 对候选明文重新计算 CBC-MAC // Step 2: 对候选明文重新计算 CBC-MAC
let t = ccm_cbc_mac(rk, nonce, aad, &plaintext, tag_len); let t = ccm_cbc_mac(rk, nonce, aad, &plaintext, tag_len)?;
let mut expected_tag = [0u8; 16]; let mut expected_tag = [0u8; 16];
for i in 0..tag_len { for i in 0..tag_len {
expected_tag[i] = t[i] ^ s0[i]; expected_tag[i] = t[i] ^ s0[i];
@@ -536,14 +556,27 @@ fn xts_mul_alpha(tweak: &mut [u8; 16]) {
/// - `key1`: 数据加密密钥(16 字节) /// - `key1`: 数据加密密钥(16 字节)
/// - `key2`: tweak 加密密钥(16 字节) /// - `key2`: tweak 加密密钥(16 字节)
/// - `tweak_sector`: 扇区号(16 字节,通常为扇区编号的小端表示) /// - `tweak_sector`: 扇区号(16 字节,通常为扇区编号的小端表示)
/// - `data`: 明文(须为 16 字节整倍数) /// - `data`: 明文(须为 16 字节整倍数,不支持非对齐输入
///
/// # 错误
/// `data` 为空或长度不是 16 的整倍数时返回 `Error::InvalidInputLength`。
///
/// # 注意
/// XTS 的 ciphertext stealing(非对齐末尾块处理)超出本实现范围,
/// 调用方须保证输入对齐;非对齐时须先在应用层填充后再调用。
#[cfg(feature = "alloc")] #[cfg(feature = "alloc")]
pub fn sm4_encrypt_xts( pub fn sm4_encrypt_xts(
key1: &[u8; 16], key1: &[u8; 16],
key2: &[u8; 16], key2: &[u8; 16],
tweak_sector: &[u8; 16], tweak_sector: &[u8; 16],
data: &[u8], data: &[u8],
) -> Vec<u8> { ) -> Result<Vec<u8>, crate::error::Error> {
// Reason: 非对齐输入在旧实现中被静默丢弃(最后不足 16 字节块跳过),
// 导致密文比明文短而调用方无感知。拒绝非对齐输入防止数据静默丢失。
if data.is_empty() || data.len() % 16 != 0 {
return Err(crate::error::Error::InvalidInputLength);
}
let sm4_1 = Sm4Key::new(key1); let sm4_1 = Sm4Key::new(key1);
let sm4_2 = Sm4Key::new(key2); let sm4_2 = Sm4Key::new(key2);
let mut tweak = *tweak_sector; let mut tweak = *tweak_sector;
@@ -551,7 +584,6 @@ pub fn sm4_encrypt_xts(
let mut out = Vec::with_capacity(data.len()); let mut out = Vec::with_capacity(data.len());
for chunk in data.chunks(16) { for chunk in data.chunks(16) {
if chunk.len() == 16 {
let mut block = [0u8; 16]; let mut block = [0u8; 16];
for i in 0..16 { for i in 0..16 {
block[i] = chunk[i] ^ tweak[i]; block[i] = chunk[i] ^ tweak[i];
@@ -562,18 +594,25 @@ pub fn sm4_encrypt_xts(
} }
xts_mul_alpha(&mut tweak); xts_mul_alpha(&mut tweak);
} }
} Ok(out)
out
} }
/// SM4-XTS 解密(磁盘加密模式) /// SM4-XTS 解密(磁盘加密模式GB/T 17964-2021
///
/// # 错误
/// `data` 为空或长度不是 16 的整倍数时返回 `Error::InvalidInputLength`。
#[cfg(feature = "alloc")] #[cfg(feature = "alloc")]
pub fn sm4_decrypt_xts( pub fn sm4_decrypt_xts(
key1: &[u8; 16], key1: &[u8; 16],
key2: &[u8; 16], key2: &[u8; 16],
tweak_sector: &[u8; 16], tweak_sector: &[u8; 16],
data: &[u8], data: &[u8],
) -> Vec<u8> { ) -> Result<Vec<u8>, crate::error::Error> {
// Reason: 同 sm4_encrypt_xts,拒绝非对齐输入防止数据静默丢失。
if data.is_empty() || data.len() % 16 != 0 {
return Err(crate::error::Error::InvalidInputLength);
}
let sm4_1 = Sm4Key::new(key1); let sm4_1 = Sm4Key::new(key1);
let sm4_2 = Sm4Key::new(key2); let sm4_2 = Sm4Key::new(key2);
let mut tweak = *tweak_sector; let mut tweak = *tweak_sector;
@@ -581,7 +620,6 @@ pub fn sm4_decrypt_xts(
let mut out = Vec::with_capacity(data.len()); let mut out = Vec::with_capacity(data.len());
for chunk in data.chunks(16) { for chunk in data.chunks(16) {
if chunk.len() == 16 {
let mut block = [0u8; 16]; let mut block = [0u8; 16];
for i in 0..16 { for i in 0..16 {
block[i] = chunk[i] ^ tweak[i]; block[i] = chunk[i] ^ tweak[i];
@@ -592,8 +630,7 @@ pub fn sm4_decrypt_xts(
} }
xts_mul_alpha(&mut tweak); xts_mul_alpha(&mut tweak);
} }
} Ok(out)
out
} }
// ── 测试 ────────────────────────────────────────────────────────────────────── // ── 测试 ──────────────────────────────────────────────────────────────────────
@@ -656,7 +693,7 @@ mod tests {
let aad = b"ccm aad"; let aad = b"ccm aad";
let plain = b"ccm plaintext!!!"; let plain = b"ccm plaintext!!!";
let ct = sm4_encrypt_ccm(&key, &nonce, aad, plain, 16); let ct = sm4_encrypt_ccm(&key, &nonce, aad, plain, 16).unwrap();
let pt = sm4_decrypt_ccm(&key, &nonce, aad, &ct, 16).unwrap(); let pt = sm4_decrypt_ccm(&key, &nonce, aad, &ct, 16).unwrap();
assert_eq!(pt, plain, "CCM 往返解密失败"); assert_eq!(pt, plain, "CCM 往返解密失败");
} }
@@ -666,7 +703,7 @@ mod tests {
fn test_ccm_tag_tamper() { fn test_ccm_tag_tamper() {
let key = [0u8; 16]; let key = [0u8; 16];
let nonce = [0u8; 12]; let nonce = [0u8; 12];
let mut ct = sm4_encrypt_ccm(&key, &nonce, b"", b"secret data here", 16); let mut ct = sm4_encrypt_ccm(&key, &nonce, b"", b"secret data here", 16).unwrap();
// 篡改 tag(最后 16 字节) // 篡改 tag(最后 16 字节)
let last = ct.len() - 1; let last = ct.len() - 1;
ct[last] ^= 1; ct[last] ^= 1;
@@ -676,6 +713,58 @@ mod tests {
); );
} }
/// CCM AAD 超限应返回错误(而非静默跳过)
#[test]
fn test_ccm_aad_too_long() {
let key = [0u8; 16];
let nonce = [0u8; 12];
let big_aad = [0u8; 511]; // 超过 510 字节限制
assert!(
sm4_encrypt_ccm(&key, &nonce, &big_aad, b"data", 16).is_err(),
"AAD 超过 510 字节时应返回 InvalidInputLength"
);
}
/// XTS 加解密往返测试
#[test]
fn test_xts_roundtrip() {
let key1 = [0x11u8; 16];
let key2 = [0x22u8; 16];
let tweak = [0u8; 16];
let plain = [0x42u8; 32]; // 2 个 16 字节块
let ct = sm4_encrypt_xts(&key1, &key2, &tweak, &plain).unwrap();
let pt = sm4_decrypt_xts(&key1, &key2, &tweak, &ct).unwrap();
assert_eq!(pt, plain, "XTS 往返解密失败");
}
/// XTS 非对齐数据应返回错误
#[test]
fn test_xts_non_aligned_rejected() {
let key1 = [0u8; 16];
let key2 = [0u8; 16];
let tweak = [0u8; 16];
// 空输入
assert!(
sm4_encrypt_xts(&key1, &key2, &tweak, b"").is_err(),
"空输入应返回 InvalidInputLength"
);
// 非 16 倍数
assert!(
sm4_encrypt_xts(&key1, &key2, &tweak, b"not-aligned-data").is_err() == false,
"正好 16 字节不应返回错误"
);
assert!(
sm4_encrypt_xts(&key1, &key2, &tweak, &[0u8; 17]).is_err(),
"17 字节应返回 InvalidInputLength"
);
assert!(
sm4_decrypt_xts(&key1, &key2, &tweak, &[0u8; 15]).is_err(),
"15 字节应返回 InvalidInputLength"
);
}
/// OFB 自反性验证 /// OFB 自反性验证
#[test] #[test]
fn test_ofb_self_inverse() { fn test_ofb_self_inverse() {
+1
View File
@@ -489,6 +489,7 @@ pub fn fp12_to_bytes(a: &Fp12) -> [u8; 384] {
/// - a: yP 系数 -> c0.c01 slot,在 eval_line_at_p 中乘以 yP /// - a: yP 系数 -> c0.c01 slot,在 eval_line_at_p 中乘以 yP
/// - b: 常数项 -> c0.c1v slot /// - b: 常数项 -> c0.c1v slot
/// - c: xP 系数 -> c1.c0w slot,在 eval_line_at_p 中乘以 xP /// - c: xP 系数 -> c1.c0w slot,在 eval_line_at_p 中乘以 xP
///
/// Reason: 经双线性性测试验证,此约定对应 D-type twist BN256 配对正确系数。 /// Reason: 经双线性性测试验证,此约定对应 D-type twist BN256 配对正确系数。
/// double step: a=Z₁²·u, b=-2Y₁Z₁, c=3X₁² /// double step: a=Z₁²·u, b=-2Y₁Z₁, c=3X₁²
/// add step: a=r·x2, b=-(r·x1+h·y1), c=h·y2 /// add step: a=r·x2, b=-(r·x1+h·y1), c=h·y2
+7 -6
View File
@@ -70,12 +70,13 @@ fn hash_to_range(z: &[u8], hid: u8, n: &U256) -> U256 {
let h_raw = U256::from_be_slice(&ha[..32]); let h_raw = U256::from_be_slice(&ha[..32]);
// h = h_raw mod (n-1) + 1,确保 h ∈ [1, n-1] // h = h_raw mod (n-1) + 1,确保 h ∈ [1, n-1]
// 于 h_raw 可能 ≥ n-1使用模运算 // Reason: 原 while 循环的执行次数取决于 h_raw 是否 ≥ n-1泄露 1 bit 信息。
// crypto_bigint 无直接 mod,改用减法循环(n 是 256 位素数,循环次数最多 1 次) // 改用无条件减法 + 掩码选择(conditional_select),执行时间与 h_raw 值无关。
let mut h = h_raw; // crypto_bigint::Uint 实现了 subtle::ConstantTimeLessct_lt 为常量时间比较。
while h >= n_minus_1 { use subtle::{ConditionallySelectable, ConstantTimeLess};
h = h.wrapping_sub(&n_minus_1); let need_reduce = !h_raw.ct_lt(&n_minus_1); // h_raw >= n_minus_1
} let reduced = h_raw.wrapping_sub(&n_minus_1);
let h = U256::conditional_select(&h_raw, &reduced, need_reduce);
h.wrapping_add(&U256::ONE) h.wrapping_add(&U256::ONE)
} }