Adding upstream version 1:7.0.3.upstream/1%7.0.3

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

8 files changed, 1177 insertions, 0 deletions

diff --git a/rust/vendor/polyval/src/backend.rs b/rust/vendor/polyval/src/backend.rs
new file mode 100644
index 0000000..2bd0c28
--- /dev/null
+++ b/rust/vendor/polyval/src/backend.rs
@@ -0,0 +1,24 @@
+//! POLYVAL backends
+
+#[cfg_attr(not(target_pointer_width = "64"), path = "backend/soft32.rs")]
+#[cfg_attr(target_pointer_width = "64", path = "backend/soft64.rs")]
+mod soft;
+
+use cfg_if::cfg_if;
+
+cfg_if! {
+    if #[cfg(all(target_arch = "aarch64", feature = "armv8", not(feature = "force-soft")))] {
+        mod autodetect;
+        mod pmull;
+        pub use crate::backend::autodetect::Polyval;
+    } else if #[cfg(all(
+        any(target_arch = "x86_64", target_arch = "x86"),
+        not(feature = "force-soft")
+    ))] {
+        mod autodetect;
+        mod clmul;
+        pub use crate::backend::autodetect::Polyval;
+    } else {
+        pub use crate::backend::soft::Polyval;
+    }
+}
diff --git a/rust/vendor/polyval/src/backend/autodetect.rs b/rust/vendor/polyval/src/backend/autodetect.rs
new file mode 100644
index 0000000..4d4ee97
--- /dev/null
+++ b/rust/vendor/polyval/src/backend/autodetect.rs
@@ -0,0 +1,111 @@
+//! Autodetection for CPU intrinsics, with fallback to the "soft" backend when
+//! they are unavailable.
+
+use crate::{backend::soft, Block, Key};
+use core::mem::ManuallyDrop;
+use universal_hash::{consts::U16, NewUniversalHash, Output, UniversalHash};
+
+#[cfg(all(target_arch = "aarch64", feature = "armv8"))]
+use super::pmull as intrinsics;
+
+#[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
+use super::clmul as intrinsics;
+
+#[cfg(all(target_arch = "aarch64", feature = "armv8"))]
+cpufeatures::new!(mul_intrinsics, "aes"); // `aes` implies PMULL
+
+#[cfg(any(target_arch = "x86_64", target_arch = "x86"))]
+cpufeatures::new!(mul_intrinsics, "pclmulqdq", "sse4.1");
+
+/// **POLYVAL**: GHASH-like universal hash over GF(2^128).
+pub struct Polyval {
+    inner: Inner,
+    token: mul_intrinsics::InitToken,
+}
+
+union Inner {
+    intrinsics: ManuallyDrop<intrinsics::Polyval>,
+    soft: ManuallyDrop<soft::Polyval>,
+}
+
+impl NewUniversalHash for Polyval {
+    type KeySize = U16;
+
+    /// Initialize POLYVAL with the given `H` field element
+    fn new(h: &Key) -> Self {
+        let (token, has_intrinsics) = mul_intrinsics::init_get();
+
+        let inner = if has_intrinsics {
+            Inner {
+                intrinsics: ManuallyDrop::new(intrinsics::Polyval::new(h)),
+            }
+        } else {
+            Inner {
+                soft: ManuallyDrop::new(soft::Polyval::new(h)),
+            }
+        };
+
+        Self { inner, token }
+    }
+}
+
+impl UniversalHash for Polyval {
+    type BlockSize = U16;
+
+    /// Input a field element `X` to be authenticated
+    #[inline]
+    fn update(&mut self, x: &Block) {
+        if self.token.get() {
+            unsafe { (*self.inner.intrinsics).update(x) }
+        } else {
+            unsafe { (*self.inner.soft).update(x) }
+        }
+    }
+
+    /// Reset internal state
+    fn reset(&mut self) {
+        if self.token.get() {
+            unsafe { (*self.inner.intrinsics).reset() }
+        } else {
+            unsafe { (*self.inner.soft).reset() }
+        }
+    }
+
+    /// Get POLYVAL result (i.e. computed `S` field element)
+    fn finalize(self) -> Output<Self> {
+        let output_bytes = if self.token.get() {
+            unsafe {
+                ManuallyDrop::into_inner(self.inner.intrinsics)
+                    .finalize()
+                    .into_bytes()
+            }
+        } else {
+            unsafe {
+                ManuallyDrop::into_inner(self.inner.soft)
+                    .finalize()
+                    .into_bytes()
+            }
+        };
+
+        Output::new(output_bytes)
+    }
+}
+
+impl Clone for Polyval {
+    fn clone(&self) -> Self {
+        let inner = if self.token.get() {
+            Inner {
+                intrinsics: ManuallyDrop::new(unsafe { (*self.inner.intrinsics).clone() }),
+            }
+        } else {
+            Inner {
+                soft: ManuallyDrop::new(unsafe { (*self.inner.soft).clone() }),
+            }
+        };
+
+        Self {
+            inner,
+            token: self.token,
+        }
+    }
+}
diff --git a/rust/vendor/polyval/src/backend/clmul.rs b/rust/vendor/polyval/src/backend/clmul.rs
new file mode 100644
index 0000000..d5d3c2b
--- /dev/null
+++ b/rust/vendor/polyval/src/backend/clmul.rs
@@ -0,0 +1,142 @@
+//! Intel `CLMUL`-accelerated implementation for modern x86/x86_64 CPUs
+//! (i.e. Intel Sandy Bridge-compatible or newer)
+
+use crate::{Block, Key};
+use universal_hash::{consts::U16, NewUniversalHash, Output, UniversalHash};
+
+#[cfg(target_arch = "x86")]
+use core::arch::x86::*;
+#[cfg(target_arch = "x86_64")]
+use core::arch::x86_64::*;
+
+/// **POLYVAL**: GHASH-like universal hash over GF(2^128).
+#[derive(Clone)]
+pub struct Polyval {
+    h: __m128i,
+    y: __m128i,
+}
+
+impl NewUniversalHash for Polyval {
+    type KeySize = U16;
+
+    /// Initialize POLYVAL with the given `H` field element
+    fn new(h: &Key) -> Self {
+        unsafe {
+            // `_mm_loadu_si128` performs an unaligned load
+            #[allow(clippy::cast_ptr_alignment)]
+            Self {
+                h: _mm_loadu_si128(h.as_ptr() as *const __m128i),
+                y: _mm_setzero_si128(),
+            }
+        }
+    }
+}
+
+impl UniversalHash for Polyval {
+    type BlockSize = U16;
+
+    #[inline]
+    fn update(&mut self, x: &Block) {
+        unsafe {
+            self.mul(x);
+        }
+    }
+
+    /// Reset internal state
+    fn reset(&mut self) {
+        unsafe {
+            self.y = _mm_setzero_si128();
+        }
+    }
+
+    /// Get GHASH output
+    fn finalize(self) -> Output<Self> {
+        unsafe { core::mem::transmute(self.y) }
+    }
+}
+
+impl Polyval {
+    #[inline]
+    #[target_feature(enable = "pclmulqdq")]
+    #[target_feature(enable = "sse4.1")]
+    unsafe fn mul(&mut self, x: &Block) {
+        let h = self.h;
+
+        // `_mm_loadu_si128` performs an unaligned load
+        #[allow(clippy::cast_ptr_alignment)]
+        let x = _mm_loadu_si128(x.as_ptr() as *const __m128i);
+        let y = _mm_xor_si128(self.y, x);
+
+        let h0 = h;
+        let h1 = _mm_shuffle_epi32(h, 0x0E);
+        let h2 = _mm_xor_si128(h0, h1);
+        let y0 = y;
+
+        // Multiply values partitioned to 64-bit parts
+        let y1 = _mm_shuffle_epi32(y, 0x0E);
+        let y2 = _mm_xor_si128(y0, y1);
+        let t0 = _mm_clmulepi64_si128(y0, h0, 0x00);
+        let t1 = _mm_clmulepi64_si128(y, h, 0x11);
+        let t2 = _mm_clmulepi64_si128(y2, h2, 0x00);
+        let t2 = _mm_xor_si128(t2, _mm_xor_si128(t0, t1));
+        let v0 = t0;
+        let v1 = _mm_xor_si128(_mm_shuffle_epi32(t0, 0x0E), t2);
+        let v2 = _mm_xor_si128(t1, _mm_shuffle_epi32(t2, 0x0E));
+        let v3 = _mm_shuffle_epi32(t1, 0x0E);
+
+        // Polynomial reduction
+        let v2 = xor5(
+            v2,
+            v0,
+            _mm_srli_epi64(v0, 1),
+            _mm_srli_epi64(v0, 2),
+            _mm_srli_epi64(v0, 7),
+        );
+
+        let v1 = xor4(
+            v1,
+            _mm_slli_epi64(v0, 63),
+            _mm_slli_epi64(v0, 62),
+            _mm_slli_epi64(v0, 57),
+        );
+
+        let v3 = xor5(
+            v3,
+            v1,
+            _mm_srli_epi64(v1, 1),
+            _mm_srli_epi64(v1, 2),
+            _mm_srli_epi64(v1, 7),
+        );
+
+        let v2 = xor4(
+            v2,
+            _mm_slli_epi64(v1, 63),
+            _mm_slli_epi64(v1, 62),
+            _mm_slli_epi64(v1, 57),
+        );
+
+        self.y = _mm_unpacklo_epi64(v2, v3);
+    }
+}
+
+#[cfg(feature = "zeroize")]
+impl Drop for Polyval {
+    fn drop(&mut self) {
+        use zeroize::Zeroize;
+        self.h.zeroize();
+        self.y.zeroize();
+    }
+}
+
+#[inline(always)]
+unsafe fn xor4(e1: __m128i, e2: __m128i, e3: __m128i, e4: __m128i) -> __m128i {
+    _mm_xor_si128(_mm_xor_si128(e1, e2), _mm_xor_si128(e3, e4))
+}
+
+#[inline(always)]
+unsafe fn xor5(e1: __m128i, e2: __m128i, e3: __m128i, e4: __m128i, e5: __m128i) -> __m128i {
+    _mm_xor_si128(
+        e1,
+        _mm_xor_si128(_mm_xor_si128(e2, e3), _mm_xor_si128(e4, e5)),
+    )
+}
diff --git a/rust/vendor/polyval/src/backend/pmull.rs b/rust/vendor/polyval/src/backend/pmull.rs
new file mode 100644
index 0000000..01d3626
--- /dev/null
+++ b/rust/vendor/polyval/src/backend/pmull.rs
@@ -0,0 +1,116 @@
+//! ARMv8 `PMULL`-accelerated implementation of POLYVAL.
+//!
+//! Based on this C intrinsics implementation:
+//! <https://github.com/noloader/AES-Intrinsics/blob/master/clmul-arm.c>
+//!
+//! Original C written and placed in public domain by Jeffrey Walton.
+//! Based on code from ARM, and by Johannes Schneiders, Skip Hovsmith and
+//! Barry O'Rourke for the mbedTLS project.
+//!
+//! For more information about PMULL, see:
+//! - <https://developer.arm.com/documentation/100069/0608/A64-SIMD-Vector-Instructions/PMULL--PMULL2--vector->
+//! - <https://eprint.iacr.org/2015/688.pdf>
+
+use crate::{Block, Key};
+use core::{arch::aarch64::*, mem};
+use universal_hash::{consts::U16, NewUniversalHash, Output, UniversalHash};
+
+/// **POLYVAL**: GHASH-like universal hash over GF(2^128).
+#[derive(Clone)]
+pub struct Polyval {
+    h: uint8x16_t,
+    y: uint8x16_t,
+}
+
+impl NewUniversalHash for Polyval {
+    type KeySize = U16;
+
+    /// Initialize POLYVAL with the given `H` field element
+    fn new(h: &Key) -> Self {
+        unsafe {
+            Self {
+                h: vld1q_u8(h.as_ptr()),
+                y: vdupq_n_u8(0), // all zeroes
+            }
+        }
+    }
+}
+
+impl UniversalHash for Polyval {
+    type BlockSize = U16;
+
+    #[inline]
+    fn update(&mut self, x: &Block) {
+        unsafe {
+            self.mul(x);
+        }
+    }
+
+    /// Reset internal state
+    fn reset(&mut self) {
+        unsafe {
+            self.y = vdupq_n_u8(0);
+        }
+    }
+
+    /// Get GHASH output
+    fn finalize(self) -> Output<Self> {
+        unsafe { mem::transmute(self.y) }
+    }
+}
+
+impl Polyval {
+    /// Mask value used when performing reduction.
+    /// This corresponds to POLYVAL's polynomial with the highest bit unset.
+    const MASK: u128 = 1 << 127 | 1 << 126 | 1 << 121 | 1;
+
+    /// POLYVAL carryless multiplication.
+    // TODO(tarcieri): investigate ordering optimizations and fusions e.g.`fuse-crypto-eor`
+    #[inline]
+    #[target_feature(enable = "neon")]
+    unsafe fn mul(&mut self, x: &Block) {
+        let h = self.h;
+        let y = veorq_u8(self.y, vld1q_u8(x.as_ptr()));
+
+        // polynomial multiply
+        let z = vdupq_n_u8(0);
+        let r0 = pmull::<0, 0>(h, y);
+        let r1 = pmull::<1, 1>(h, y);
+        let t0 = pmull::<0, 1>(h, y);
+        let t1 = pmull::<1, 0>(h, y);
+        let t0 = veorq_u8(t0, t1);
+        let t1 = vextq_u8(z, t0, 8);
+        let r0 = veorq_u8(r0, t1);
+        let t1 = vextq_u8(t0, z, 8);
+        let r1 = veorq_u8(r1, t1);
+
+        // polynomial reduction
+        let p = mem::transmute(Self::MASK);
+        let t0 = pmull::<0, 1>(r0, p);
+        let t1 = vextq_u8(t0, t0, 8);
+        let r0 = veorq_u8(r0, t1);
+        let t1 = pmull::<1, 1>(r0, p);
+        let r0 = veorq_u8(r0, t1);
+
+        self.y = veorq_u8(r0, r1);
+    }
+}
+
+/// Wrapper for the ARM64 `PMULL` instruction.
+#[inline(always)]
+unsafe fn pmull<const A_LANE: i32, const B_LANE: i32>(a: uint8x16_t, b: uint8x16_t) -> uint8x16_t {
+    mem::transmute(vmull_p64(
+        vgetq_lane_u64(vreinterpretq_u64_u8(a), A_LANE),
+        vgetq_lane_u64(vreinterpretq_u64_u8(b), B_LANE),
+    ))
+}
+
+// TODO(tarcieri): zeroize support
+// #[cfg(feature = "zeroize")]
+// impl Drop for Polyval {
+//     fn drop(&mut self) {
+//         use zeroize::Zeroize;
+//         self.h.zeroize();
+//         self.y.zeroize();
+//     }
+// }
diff --git a/rust/vendor/polyval/src/backend/soft32.rs b/rust/vendor/polyval/src/backend/soft32.rs
new file mode 100644
index 0000000..6a09c58
--- /dev/null
+++ b/rust/vendor/polyval/src/backend/soft32.rs
@@ -0,0 +1,281 @@
+//! Constant-time software implementation of POLYVAL for 32-bit architectures
+//! Adapted from BearSSL's `ghash_ctmul32.c`:
+//!
+//! <https://bearssl.org/gitweb/?p=BearSSL;a=blob;f=src/hash/ghash_ctmul32.c;hb=4b6046412>
+//!
+//! Copyright (c) 2016 Thomas Pornin <pornin@bolet.org>
+//!
+//! This implementation uses 32-bit multiplications, and only the low
+//! 32 bits for each multiplication result. This is meant primarily for
+//! the ARM Cortex M0 and M0+, whose multiplication opcode does not yield
+//! the upper 32 bits; but it might also be useful on architectures where
+//! access to the upper 32 bits requires use of specific registers that
+//! create contention (e.g. on i386, "mul" necessarily outputs the result
+//! in edx:eax, while "imul" can use any registers but is limited to the
+//! low 32 bits).
+//!
+//! The implementation trick that is used here is bit-reversing (bit 0
+//! is swapped with bit 31, bit 1 with bit 30, and so on). In GF(2)[X],
+//! for all values x and y, we have:
+//!
+//! ```text
+//! rev32(x) * rev32(y) = rev64(x * y)
+//! ```
+//!
+//! In other words, if we bit-reverse (over 32 bits) the operands, then we
+//! bit-reverse (over 64 bits) the result.
+
+use crate::{Block, Key};
+use core::{
+    convert::TryInto,
+    num::Wrapping,
+    ops::{Add, Mul},
+};
+use universal_hash::{consts::U16, NewUniversalHash, Output, UniversalHash};
+
+#[cfg(feature = "zeroize")]
+use zeroize::Zeroize;
+
+/// **POLYVAL**: GHASH-like universal hash over GF(2^128).
+#[derive(Clone)]
+pub struct Polyval {
+    /// GF(2^128) field element input blocks are multiplied by
+    h: U32x4,
+
+    /// Field element representing the computed universal hash
+    s: U32x4,
+}
+
+impl NewUniversalHash for Polyval {
+    type KeySize = U16;
+
+    /// Initialize POLYVAL with the given `H` field element
+    fn new(h: &Key) -> Self {
+        Self {
+            h: h.into(),
+            s: U32x4::default(),
+        }
+    }
+}
+
+impl UniversalHash for Polyval {
+    type BlockSize = U16;
+
+    /// Input a field element `X` to be authenticated
+    fn update(&mut self, x: &Block) {
+        let x = U32x4::from(x);
+        self.s = (self.s + x) * self.h;
+    }
+
+    /// Reset internal state
+    fn reset(&mut self) {
+        self.s = U32x4::default();
+    }
+
+    /// Get POLYVAL result (i.e. computed `S` field element)
+    fn finalize(self) -> Output<Self> {
+        let mut block = Block::default();
+
+        for (chunk, i) in block
+            .chunks_mut(4)
+            .zip(&[self.s.0, self.s.1, self.s.2, self.s.3])
+        {
+            chunk.copy_from_slice(&i.to_le_bytes());
+        }
+
+        Output::new(block)
+    }
+}
+
+#[cfg(feature = "zeroize")]
+impl Drop for Polyval {
+    fn drop(&mut self) {
+        self.h.zeroize();
+        self.s.zeroize();
+    }
+}
+
+/// 4 x `u32` values
+#[derive(Copy, Clone, Debug, Default, Eq, PartialEq)]
+struct U32x4(u32, u32, u32, u32);
+
+impl From<&Block> for U32x4 {
+    fn from(bytes: &Block) -> U32x4 {
+        U32x4(
+            u32::from_le_bytes(bytes[..4].try_into().unwrap()),
+            u32::from_le_bytes(bytes[4..8].try_into().unwrap()),
+            u32::from_le_bytes(bytes[8..12].try_into().unwrap()),
+            u32::from_le_bytes(bytes[12..].try_into().unwrap()),
+        )
+    }
+}
+
+#[allow(clippy::suspicious_arithmetic_impl)]
+impl Add for U32x4 {
+    type Output = Self;
+
+    /// Adds two POLYVAL field elements.
+    fn add(self, rhs: Self) -> Self::Output {
+        U32x4(
+            self.0 ^ rhs.0,
+            self.1 ^ rhs.1,
+            self.2 ^ rhs.2,
+            self.3 ^ rhs.3,
+        )
+    }
+}
+
+#[allow(clippy::suspicious_arithmetic_impl)]
+impl Mul for U32x4 {
+    type Output = Self;
+
+    /// Computes carryless POLYVAL multiplication over GF(2^128) in constant time.
+    ///
+    /// Method described at:
+    /// <https://www.bearssl.org/constanttime.html#ghash-for-gcm>
+    ///
+    /// POLYVAL multiplication is effectively the little endian equivalent of
+    /// GHASH multiplication, aside from one small detail described here:
+    ///
+    /// <https://crypto.stackexchange.com/questions/66448/how-does-bearssls-gcm-modular-reduction-work/66462#66462>
+    ///
+    /// > The product of two bit-reversed 128-bit polynomials yields the
+    /// > bit-reversed result over 255 bits, not 256. The BearSSL code ends up
+    /// > with a 256-bit result in zw[], and that value is shifted by one bit,
+    /// > because of that reversed convention issue. Thus, the code must
+    /// > include a shifting step to put it back where it should
+    ///
+    /// This shift is unnecessary for POLYVAL and has been removed.
+    fn mul(self, rhs: Self) -> Self {
+        let hw = [self.0, self.1, self.2, self.3];
+        let yw = [rhs.0, rhs.1, rhs.2, rhs.3];
+        let hwr = [rev32(hw[0]), rev32(hw[1]), rev32(hw[2]), rev32(hw[3])];
+
+        // We are using Karatsuba: the 128x128 multiplication is
+        // reduced to three 64x64 multiplications, hence nine
+        // 32x32 multiplications. With the bit-reversal trick,
+        // we have to perform 18 32x32 multiplications.
+
+        let mut a = [0u32; 18];
+
+        a[0] = yw[0];
+        a[1] = yw[1];
+        a[2] = yw[2];
+        a[3] = yw[3];
+        a[4] = a[0] ^ a[1];
+        a[5] = a[2] ^ a[3];
+        a[6] = a[0] ^ a[2];
+        a[7] = a[1] ^ a[3];
+        a[8] = a[6] ^ a[7];
+        a[9] = rev32(yw[0]);
+        a[10] = rev32(yw[1]);
+        a[11] = rev32(yw[2]);
+        a[12] = rev32(yw[3]);
+        a[13] = a[9] ^ a[10];
+        a[14] = a[11] ^ a[12];
+        a[15] = a[9] ^ a[11];
+        a[16] = a[10] ^ a[12];
+        a[17] = a[15] ^ a[16];
+
+        let mut b = [0u32; 18];
+
+        b[0] = hw[0];
+        b[1] = hw[1];
+        b[2] = hw[2];
+        b[3] = hw[3];
+        b[4] = b[0] ^ b[1];
+        b[5] = b[2] ^ b[3];
+        b[6] = b[0] ^ b[2];
+        b[7] = b[1] ^ b[3];
+        b[8] = b[6] ^ b[7];
+        b[9] = hwr[0];
+        b[10] = hwr[1];
+        b[11] = hwr[2];
+        b[12] = hwr[3];
+        b[13] = b[9] ^ b[10];
+        b[14] = b[11] ^ b[12];
+        b[15] = b[9] ^ b[11];
+        b[16] = b[10] ^ b[12];
+        b[17] = b[15] ^ b[16];
+
+        let mut c = [0u32; 18];
+
+        for i in 0..18 {
+            c[i] = bmul32(a[i], b[i]);
+        }
+
+        c[4] ^= c[0] ^ c[1];
+        c[5] ^= c[2] ^ c[3];
+        c[8] ^= c[6] ^ c[7];
+
+        c[13] ^= c[9] ^ c[10];
+        c[14] ^= c[11] ^ c[12];
+        c[17] ^= c[15] ^ c[16];
+
+        let mut zw = [0u32; 8];
+
+        zw[0] = c[0];
+        zw[1] = c[4] ^ rev32(c[9]) >> 1;
+        zw[2] = c[1] ^ c[0] ^ c[2] ^ c[6] ^ rev32(c[13]) >> 1;
+        zw[3] = c[4] ^ c[5] ^ c[8] ^ rev32(c[10] ^ c[9] ^ c[11] ^ c[15]) >> 1;
+        zw[4] = c[2] ^ c[1] ^ c[3] ^ c[7] ^ rev32(c[13] ^ c[14] ^ c[17]) >> 1;
+        zw[5] = c[5] ^ rev32(c[11] ^ c[10] ^ c[12] ^ c[16]) >> 1;
+        zw[6] = c[3] ^ rev32(c[14]) >> 1;
+        zw[7] = rev32(c[12]) >> 1;
+
+        for i in 0..4 {
+            let lw = zw[i];
+            zw[i + 4] ^= lw ^ (lw >> 1) ^ (lw >> 2) ^ (lw >> 7);
+            zw[i + 3] ^= (lw << 31) ^ (lw << 30) ^ (lw << 25);
+        }
+
+        U32x4(zw[4], zw[5], zw[6], zw[7])
+    }
+}
+
+#[cfg(feature = "zeroize")]
+impl Zeroize for U32x4 {
+    fn zeroize(&mut self) {
+        self.0.zeroize();
+        self.1.zeroize();
+        self.2.zeroize();
+        self.3.zeroize();
+    }
+}
+
+/// Multiplication in GF(2)[X], truncated to the low 32-bits, with “holes”
+/// (sequences of zeroes) to avoid carry spilling.
+///
+/// When carries do occur, they wind up in a "hole" and are subsequently masked
+/// out of the result.
+fn bmul32(x: u32, y: u32) -> u32 {
+    let x0 = Wrapping(x & 0x1111_1111);
+    let x1 = Wrapping(x & 0x2222_2222);
+    let x2 = Wrapping(x & 0x4444_4444);
+    let x3 = Wrapping(x & 0x8888_8888);
+    let y0 = Wrapping(y & 0x1111_1111);
+    let y1 = Wrapping(y & 0x2222_2222);
+    let y2 = Wrapping(y & 0x4444_4444);
+    let y3 = Wrapping(y & 0x8888_8888);
+
+    let mut z0 = ((x0 * y0) ^ (x1 * y3) ^ (x2 * y2) ^ (x3 * y1)).0;
+    let mut z1 = ((x0 * y1) ^ (x1 * y0) ^ (x2 * y3) ^ (x3 * y2)).0;
+    let mut z2 = ((x0 * y2) ^ (x1 * y1) ^ (x2 * y0) ^ (x3 * y3)).0;
+    let mut z3 = ((x0 * y3) ^ (x1 * y2) ^ (x2 * y1) ^ (x3 * y0)).0;
+
+    z0 &= 0x1111_1111;
+    z1 &= 0x2222_2222;
+    z2 &= 0x4444_4444;
+    z3 &= 0x8888_8888;
+
+    z0 | z1 | z2 | z3
+}
+
+/// Bit-reverse a 32-bit word in constant time.
+fn rev32(mut x: u32) -> u32 {
+    x = ((x & 0x5555_5555) << 1) | (x >> 1 & 0x5555_5555);
+    x = ((x & 0x3333_3333) << 2) | (x >> 2 & 0x3333_3333);
+    x = ((x & 0x0f0f_0f0f) << 4) | (x >> 4 & 0x0f0f_0f0f);
+    x = ((x & 0x00ff_00ff) << 8) | (x >> 8 & 0x00ff_00ff);
+    (x << 16) | (x >> 16)
+}
diff --git a/rust/vendor/polyval/src/backend/soft64.rs b/rust/vendor/polyval/src/backend/soft64.rs
new file mode 100644
index 0000000..fe159a4
--- /dev/null
+++ b/rust/vendor/polyval/src/backend/soft64.rs
@@ -0,0 +1,205 @@
+//! Constant-time software implementation of POLYVAL for 64-bit architectures.
+//! Adapted from BearSSL's `ghash_ctmul64.c`:
+//!
+//! <https://bearssl.org/gitweb/?p=BearSSL;a=blob;f=src/hash/ghash_ctmul64.c;hb=4b6046412>
+//!
+//! Copyright (c) 2016 Thomas Pornin <pornin@bolet.org>
+
+use crate::{Block, Key};
+use core::{
+    convert::TryInto,
+    num::Wrapping,
+    ops::{Add, Mul},
+};
+use universal_hash::{consts::U16, NewUniversalHash, Output, UniversalHash};
+
+#[cfg(feature = "zeroize")]
+use zeroize::Zeroize;
+
+/// **POLYVAL**: GHASH-like universal hash over GF(2^128).
+#[derive(Clone)]
+pub struct Polyval {
+    /// GF(2^128) field element input blocks are multiplied by
+    h: U64x2,
+
+    /// Field element representing the computed universal hash
+    s: U64x2,
+}
+
+impl NewUniversalHash for Polyval {
+    type KeySize = U16;
+
+    /// Initialize POLYVAL with the given `H` field element
+    fn new(h: &Key) -> Self {
+        Self {
+            h: h.into(),
+            s: U64x2::default(),
+        }
+    }
+}
+
+impl UniversalHash for Polyval {
+    type BlockSize = U16;
+
+    /// Input a field element `X` to be authenticated
+    fn update(&mut self, x: &Block) {
+        let x = U64x2::from(x);
+        self.s = (self.s + x) * self.h;
+    }
+
+    /// Reset internal state
+    fn reset(&mut self) {
+        self.s = U64x2::default();
+    }
+
+    /// Get POLYVAL result (i.e. computed `S` field element)
+    fn finalize(self) -> Output<Self> {
+        let mut block = Block::default();
+
+        for (chunk, i) in block.chunks_mut(8).zip(&[self.s.0, self.s.1]) {
+            chunk.copy_from_slice(&i.to_le_bytes());
+        }
+
+        Output::new(block)
+    }
+}
+
+#[cfg(feature = "zeroize")]
+impl Drop for Polyval {
+    fn drop(&mut self) {
+        self.h.zeroize();
+        self.s.zeroize();
+    }
+}
+
+/// 2 x `u64` values
+#[derive(Copy, Clone, Debug, Default, Eq, PartialEq)]
+struct U64x2(u64, u64);
+
+impl From<&Block> for U64x2 {
+    fn from(bytes: &Block) -> U64x2 {
+        U64x2(
+            u64::from_le_bytes(bytes[..8].try_into().unwrap()),
+            u64::from_le_bytes(bytes[8..].try_into().unwrap()),
+        )
+    }
+}
+
+#[allow(clippy::suspicious_arithmetic_impl)]
+impl Add for U64x2 {
+    type Output = Self;
+
+    /// Adds two POLYVAL field elements.
+    fn add(self, rhs: Self) -> Self::Output {
+        U64x2(self.0 ^ rhs.0, self.1 ^ rhs.1)
+    }
+}
+
+#[allow(clippy::suspicious_arithmetic_impl)]
+impl Mul for U64x2 {
+    type Output = Self;
+
+    /// Computes carryless POLYVAL multiplication over GF(2^128) in constant time.
+    ///
+    /// Method described at:
+    /// <https://www.bearssl.org/constanttime.html#ghash-for-gcm>
+    ///
+    /// POLYVAL multiplication is effectively the little endian equivalent of
+    /// GHASH multiplication, aside from one small detail described here:
+    ///
+    /// <https://crypto.stackexchange.com/questions/66448/how-does-bearssls-gcm-modular-reduction-work/66462#66462>
+    ///
+    /// > The product of two bit-reversed 128-bit polynomials yields the
+    /// > bit-reversed result over 255 bits, not 256. The BearSSL code ends up
+    /// > with a 256-bit result in zw[], and that value is shifted by one bit,
+    /// > because of that reversed convention issue. Thus, the code must
+    /// > include a shifting step to put it back where it should
+    ///
+    /// This shift is unnecessary for POLYVAL and has been removed.
+    fn mul(self, rhs: Self) -> Self {
+        let h0 = self.0;
+        let h1 = self.1;
+        let h0r = rev64(h0);
+        let h1r = rev64(h1);
+        let h2 = h0 ^ h1;
+        let h2r = h0r ^ h1r;
+
+        let y0 = rhs.0;
+        let y1 = rhs.1;
+        let y0r = rev64(y0);
+        let y1r = rev64(y1);
+        let y2 = y0 ^ y1;
+        let y2r = y0r ^ y1r;
+        let z0 = bmul64(y0, h0);
+        let z1 = bmul64(y1, h1);
+
+        let mut z2 = bmul64(y2, h2);
+        let mut z0h = bmul64(y0r, h0r);
+        let mut z1h = bmul64(y1r, h1r);
+        let mut z2h = bmul64(y2r, h2r);
+
+        z2 ^= z0 ^ z1;
+        z2h ^= z0h ^ z1h;
+        z0h = rev64(z0h) >> 1;
+        z1h = rev64(z1h) >> 1;
+        z2h = rev64(z2h) >> 1;
+
+        let v0 = z0;
+        let mut v1 = z0h ^ z2;
+        let mut v2 = z1 ^ z2h;
+        let mut v3 = z1h;
+
+        v2 ^= v0 ^ (v0 >> 1) ^ (v0 >> 2) ^ (v0 >> 7);
+        v1 ^= (v0 << 63) ^ (v0 << 62) ^ (v0 << 57);
+        v3 ^= v1 ^ (v1 >> 1) ^ (v1 >> 2) ^ (v1 >> 7);
+        v2 ^= (v1 << 63) ^ (v1 << 62) ^ (v1 << 57);
+
+        U64x2(v2, v3)
+    }
+}
+
+#[cfg(feature = "zeroize")]
+impl Zeroize for U64x2 {
+    fn zeroize(&mut self) {
+        self.0.zeroize();
+        self.1.zeroize();
+    }
+}
+
+/// Multiplication in GF(2)[X], truncated to the low 64-bits, with “holes”
+/// (sequences of zeroes) to avoid carry spilling.
+///
+/// When carries do occur, they wind up in a "hole" and are subsequently masked
+/// out of the result.
+fn bmul64(x: u64, y: u64) -> u64 {
+    let x0 = Wrapping(x & 0x1111_1111_1111_1111);
+    let x1 = Wrapping(x & 0x2222_2222_2222_2222);
+    let x2 = Wrapping(x & 0x4444_4444_4444_4444);
+    let x3 = Wrapping(x & 0x8888_8888_8888_8888);
+    let y0 = Wrapping(y & 0x1111_1111_1111_1111);
+    let y1 = Wrapping(y & 0x2222_2222_2222_2222);
+    let y2 = Wrapping(y & 0x4444_4444_4444_4444);
+    let y3 = Wrapping(y & 0x8888_8888_8888_8888);
+
+    let mut z0 = ((x0 * y0) ^ (x1 * y3) ^ (x2 * y2) ^ (x3 * y1)).0;
+    let mut z1 = ((x0 * y1) ^ (x1 * y0) ^ (x2 * y3) ^ (x3 * y2)).0;
+    let mut z2 = ((x0 * y2) ^ (x1 * y1) ^ (x2 * y0) ^ (x3 * y3)).0;
+    let mut z3 = ((x0 * y3) ^ (x1 * y2) ^ (x2 * y1) ^ (x3 * y0)).0;
+
+    z0 &= 0x1111_1111_1111_1111;
+    z1 &= 0x2222_2222_2222_2222;
+    z2 &= 0x4444_4444_4444_4444;
+    z3 &= 0x8888_8888_8888_8888;
+
+    z0 | z1 | z2 | z3
+}
+
+/// Bit-reverse a `u64` in constant time
+fn rev64(mut x: u64) -> u64 {
+    x = ((x & 0x5555_5555_5555_5555) << 1) | ((x >> 1) & 0x5555_5555_5555_5555);
+    x = ((x & 0x3333_3333_3333_3333) << 2) | ((x >> 2) & 0x3333_3333_3333_3333);
+    x = ((x & 0x0f0f_0f0f_0f0f_0f0f) << 4) | ((x >> 4) & 0x0f0f_0f0f_0f0f_0f0f);
+    x = ((x & 0x00ff_00ff_00ff_00ff) << 8) | ((x >> 8) & 0x00ff_00ff_00ff_00ff);
+    x = ((x & 0xffff_0000_ffff) << 16) | ((x >> 16) & 0xffff_0000_ffff);
+    (x << 32) | (x >> 32)
+}
diff --git a/rust/vendor/polyval/src/lib.rs b/rust/vendor/polyval/src/lib.rs
new file mode 100644
index 0000000..146776e
--- /dev/null
+++ b/rust/vendor/polyval/src/lib.rs
@@ -0,0 +1,110 @@
+//! **POLYVAL** is a GHASH-like universal hash over GF(2^128) useful for
+//! implementing [AES-GCM-SIV] or [AES-GCM/GMAC].
+//!
+//! From [RFC 8452 Section 3] which defines POLYVAL for use in AES-GCM-SIV:
+//!
+//! > "POLYVAL, like GHASH (the authenticator in AES-GCM; ...), operates in a
+//! > binary field of size 2^128.  The field is defined by the irreducible
+//! > polynomial x^128 + x^127 + x^126 + x^121 + 1."
+//!
+//! By multiplying (in the finite field sense) a sequence of 128-bit blocks of
+//! input data data by a field element `H`, POLYVAL can be used to authenticate
+//! the message sequence as powers (in the finite field sense) of `H`.
+//!
+//! # Minimum Supported Rust Version
+//! Rust **1.49** or higher.
+//!
+//! In the future the minimum supported Rust version may be changed, but it
+//! be will be accompanied with a minor version bump.
+//!
+//! # Supported backends
+//! This crate provides multiple backends including a portable pure Rust
+//! backend as well as ones based on CPU intrinsics.
+//!
+//! ## "soft" portable backend
+//! As a baseline implementation, this crate provides a constant-time pure Rust
+//! implementation based on [BearSSL], which is a straightforward and
+//! compact implementation which uses a clever but simple technique to avoid
+//! carry-spilling.
+//!
+//! ## ARMv8 intrinsics (`PMULL`, nightly-only)
+//! On `aarch64` targets including `aarch64-apple-darwin` (Apple M1) and Linux
+//! targets such as `aarch64-unknown-linux-gnu` and `aarch64-unknown-linux-musl`,
+//! support for using the `PMULL` instructions in ARMv8's Cryptography Extensions
+//! is available when using the nightly compiler, and can be enabled using the
+//! `armv8` crate feature.
+//!
+//! On Linux and macOS, when the `armv8` feature is enabled support for AES
+//! intrinsics is autodetected at runtime. On other platforms the `crypto`
+//! target feature must be enabled via RUSTFLAGS.
+//!
+//! ## `x86`/`x86_64` intrinsics (`CMLMUL`)
+//! By default this crate uses runtime detection on `i686`/`x86_64` targets
+//! in order to determine if `CLMUL` is available, and if it is not, it will
+//! fallback to using a constant-time software implementation.
+//!
+//! For optimal performance, set `target-cpu` in `RUSTFLAGS` to `sandybridge`
+//! or newer:
+//!
+//! Example:
+//!
+//! ```text
+//! $ RUSTFLAGS="-Ctarget-cpu=sandybridge" cargo bench
+//! ```
+//!
+//! # Relationship to GHASH
+//! POLYVAL can be thought of as the little endian equivalent of GHASH, which
+//! affords it a small performance advantage over GHASH when used on little
+//! endian architectures.
+//!
+//! It has also been designed so it can also be used to compute GHASH and with
+//! it GMAC, the Message Authentication Code (MAC) used by AES-GCM.
+//!
+//! From [RFC 8452 Appendix A]:
+//!
+//! > "GHASH and POLYVAL both operate in GF(2^128), although with different
+//! > irreducible polynomials: POLYVAL works modulo x^128 + x^127 + x^126 +
+//! > x^121 + 1 and GHASH works modulo x^128 + x^7 + x^2 + x + 1.  Note
+//! > that these irreducible polynomials are the 'reverse' of each other."
+//!
+//! [AES-GCM-SIV]: https://en.wikipedia.org/wiki/AES-GCM-SIV
+//! [AES-GCM/GMAC]: https://en.wikipedia.org/wiki/Galois/Counter_Mode
+//! [BearSSL]: https://www.bearssl.org/constanttime.html#ghash-for-gcm
+//! [RFC 8452 Section 3]: https://tools.ietf.org/html/rfc8452#section-3
+//! [RFC 8452 Appendix A]: https://tools.ietf.org/html/rfc8452#appendix-A
+
+#![no_std]
+#![cfg_attr(docsrs, feature(doc_cfg))]
+#![cfg_attr(
+    all(feature = "armv8", target_arch = "aarch64"),
+    feature(stdsimd, aarch64_target_feature)
+)]
+#![doc(
+    html_logo_url = "https://raw.githubusercontent.com/RustCrypto/media/8f1a9894/logo.svg",
+    html_favicon_url = "https://raw.githubusercontent.com/RustCrypto/media/8f1a9894/logo.svg",
+    html_root_url = "https://docs.rs/polyval/0.5.3"
+)]
+#![warn(missing_docs, rust_2018_idioms)]
+
+mod backend;
+mod mulx;
+
+pub use crate::{backend::Polyval, mulx::mulx};
+pub use universal_hash;
+
+opaque_debug::implement!(Polyval);
+
+/// Size of a POLYVAL block in bytes
+pub const BLOCK_SIZE: usize = 16;
+
+/// Size of a POLYVAL key in bytes
+pub const KEY_SIZE: usize = 16;
+
+/// POLYVAL keys (16-bytes)
+pub type Key = universal_hash::Key<Polyval>;
+
+/// POLYVAL blocks (16-bytes)
+pub type Block = universal_hash::Block<Polyval>;
+
+/// POLYVAL tags (16-bytes)
+pub type Tag = universal_hash::Output<Polyval>;
diff --git a/rust/vendor/polyval/src/mulx.rs b/rust/vendor/polyval/src/mulx.rs
new file mode 100644
index 0000000..2abd868
--- /dev/null
+++ b/rust/vendor/polyval/src/mulx.rs
@@ -0,0 +1,188 @@
+//! The `mulX_POLYVAL()` function.
+
+use crate::Block;
+
+/// The `mulX_POLYVAL()` function as defined in [RFC 8452 Appendix A][1].
+///
+/// Performs a doubling (a.k.a. "multiply by x") over GF(2^128).
+/// This is useful for implementing GHASH in terms of POLYVAL.
+///
+/// [1]: https://tools.ietf.org/html/rfc8452#appendix-A
+pub fn mulx(block: &Block) -> Block {
+    let mut v = u128::from_le_bytes((*block).into());
+    let v_hi = v >> 127;
+
+    v <<= 1;
+    v ^= v_hi ^ (v_hi << 127) ^ (v_hi << 126) ^ (v_hi << 121);
+    v.to_le_bytes().into()
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{mulx, Block};
+    use hex_literal::hex;
+
+    /// Test vector given in RFC 8452 Appendix A.
+    ///
+    /// NOTE: the vector in the RFC actually contains a typo which has been
+    /// reported (and accepted) as RFC errata, so we use the vector from the
+    /// errata instead:
+    ///
+    /// <https://www.rfc-editor.org/errata_search.php?rfc=8452>
+    #[test]
+    fn rfc8452_vector() {
+        let input = Block::from(hex!("9c98c04df9387ded828175a92ba652d8"));
+        let expected_output = Block::from(hex!("3931819bf271fada0503eb52574ca572"));
+        let actual_output = mulx(&input);
+        assert_eq!(expected_output, actual_output);
+    }
+
+    /// Test against the `MULX_TEST_VECTORS` given below, which cover the full
+    /// size of a POLYVAL field element.
+    #[test]
+    fn mulx_vectors() {
+        // One
+        let mut r = Block::from(hex!("01000000000000000000000000000000"));
+
+        for vector in MULX_TEST_VECTORS {
+            r = mulx(&r);
+            assert_eq!(&r, Block::from_slice(vector));
+        }
+    }
+
+    /// `mulX_POLYVAL()` test vectors.
+    ///
+    /// These were generated by this crate when in a known-correct state,
+    /// verified by a GHASH implementation based on a POLYVAL core successfully
+    /// passing the NIST test vectors.
+    const MULX_TEST_VECTORS: &[[u8; 16]] = &[
+        hex!("02000000000000000000000000000000"),
+        hex!("04000000000000000000000000000000"),
+        hex!("08000000000000000000000000000000"),
+        hex!("10000000000000000000000000000000"),
+        hex!("20000000000000000000000000000000"),
+        hex!("40000000000000000000000000000000"),
+        hex!("80000000000000000000000000000000"),
+        hex!("00010000000000000000000000000000"),
+        hex!("00020000000000000000000000000000"),
+        hex!("00040000000000000000000000000000"),
+        hex!("00080000000000000000000000000000"),
+        hex!("00100000000000000000000000000000"),
+        hex!("00200000000000000000000000000000"),
+        hex!("00400000000000000000000000000000"),
+        hex!("00800000000000000000000000000000"),
+        hex!("00000100000000000000000000000000"),
+        hex!("00000200000000000000000000000000"),
+        hex!("00000400000000000000000000000000"),
+        hex!("00000800000000000000000000000000"),
+        hex!("00001000000000000000000000000000"),
+        hex!("00002000000000000000000000000000"),
+        hex!("00004000000000000000000000000000"),
+        hex!("00008000000000000000000000000000"),
+        hex!("00000001000000000000000000000000"),
+        hex!("00000002000000000000000000000000"),
+        hex!("00000004000000000000000000000000"),
+        hex!("00000008000000000000000000000000"),
+        hex!("00000010000000000000000000000000"),
+        hex!("00000020000000000000000000000000"),
+        hex!("00000040000000000000000000000000"),
+        hex!("00000080000000000000000000000000"),
+        hex!("00000000010000000000000000000000"),
+        hex!("00000000020000000000000000000000"),
+        hex!("00000000040000000000000000000000"),
+        hex!("00000000080000000000000000000000"),
+        hex!("00000000100000000000000000000000"),
+        hex!("00000000200000000000000000000000"),
+        hex!("00000000400000000000000000000000"),
+        hex!("00000000800000000000000000000000"),
+        hex!("00000000000100000000000000000000"),
+        hex!("00000000000200000000000000000000"),
+        hex!("00000000000400000000000000000000"),
+        hex!("00000000000800000000000000000000"),
+        hex!("00000000001000000000000000000000"),
+        hex!("00000000002000000000000000000000"),
+        hex!("00000000004000000000000000000000"),
+        hex!("00000000008000000000000000000000"),
+        hex!("00000000000001000000000000000000"),
+        hex!("00000000000002000000000000000000"),
+        hex!("00000000000004000000000000000000"),
+        hex!("00000000000008000000000000000000"),
+        hex!("00000000000010000000000000000000"),
+        hex!("00000000000020000000000000000000"),
+        hex!("00000000000040000000000000000000"),
+        hex!("00000000000080000000000000000000"),
+        hex!("00000000000000010000000000000000"),
+        hex!("00000000000000020000000000000000"),
+        hex!("00000000000000040000000000000000"),
+        hex!("00000000000000080000000000000000"),
+        hex!("00000000000000100000000000000000"),
+        hex!("00000000000000200000000000000000"),
+        hex!("00000000000000400000000000000000"),
+        hex!("00000000000000800000000000000000"),
+        hex!("00000000000000000100000000000000"),
+        hex!("00000000000000000200000000000000"),
+        hex!("00000000000000000400000000000000"),
+        hex!("00000000000000000800000000000000"),
+        hex!("00000000000000001000000000000000"),
+        hex!("00000000000000002000000000000000"),
+        hex!("00000000000000004000000000000000"),
+        hex!("00000000000000008000000000000000"),
+        hex!("00000000000000000001000000000000"),
+        hex!("00000000000000000002000000000000"),
+        hex!("00000000000000000004000000000000"),
+        hex!("00000000000000000008000000000000"),
+        hex!("00000000000000000010000000000000"),
+        hex!("00000000000000000020000000000000"),
+        hex!("00000000000000000040000000000000"),
+        hex!("00000000000000000080000000000000"),
+        hex!("00000000000000000000010000000000"),
+        hex!("00000000000000000000020000000000"),
+        hex!("00000000000000000000040000000000"),
+        hex!("00000000000000000000080000000000"),
+        hex!("00000000000000000000100000000000"),
+        hex!("00000000000000000000200000000000"),
+        hex!("00000000000000000000400000000000"),
+        hex!("00000000000000000000800000000000"),
+        hex!("00000000000000000000000100000000"),
+        hex!("00000000000000000000000200000000"),
+        hex!("00000000000000000000000400000000"),
+        hex!("00000000000000000000000800000000"),
+        hex!("00000000000000000000001000000000"),
+        hex!("00000000000000000000002000000000"),
+        hex!("00000000000000000000004000000000"),
+        hex!("00000000000000000000008000000000"),
+        hex!("00000000000000000000000001000000"),
+        hex!("00000000000000000000000002000000"),
+        hex!("00000000000000000000000004000000"),
+        hex!("00000000000000000000000008000000"),
+        hex!("00000000000000000000000010000000"),
+        hex!("00000000000000000000000020000000"),
+        hex!("00000000000000000000000040000000"),
+        hex!("00000000000000000000000080000000"),
+        hex!("00000000000000000000000000010000"),
+        hex!("00000000000000000000000000020000"),
+        hex!("00000000000000000000000000040000"),
+        hex!("00000000000000000000000000080000"),
+        hex!("00000000000000000000000000100000"),
+        hex!("00000000000000000000000000200000"),
+        hex!("00000000000000000000000000400000"),
+        hex!("00000000000000000000000000800000"),
+        hex!("00000000000000000000000000000100"),
+        hex!("00000000000000000000000000000200"),
+        hex!("00000000000000000000000000000400"),
+        hex!("00000000000000000000000000000800"),
+        hex!("00000000000000000000000000001000"),
+        hex!("00000000000000000000000000002000"),
+        hex!("00000000000000000000000000004000"),
+        hex!("00000000000000000000000000008000"),
+        hex!("00000000000000000000000000000001"),
+        hex!("00000000000000000000000000000002"),
+        hex!("00000000000000000000000000000004"),
+        hex!("00000000000000000000000000000008"),
+        hex!("00000000000000000000000000000010"),
+        hex!("00000000000000000000000000000020"),
+        hex!("00000000000000000000000000000040"),
+        hex!("00000000000000000000000000000080"),
+        hex!("010000000000000000000000000000c2"),
+    ];
+}