Skip to main content

ff/
lib.rs

1//! This crate provides traits for working with finite fields.
2
3#![no_std]
4#![cfg_attr(docsrs, feature(doc_cfg))]
5// Catch documentation errors caused by code changes.
6#![deny(rustdoc::broken_intra_doc_links)]
7#![forbid(unsafe_code)]
8
9#[cfg(feature = "alloc")]
10extern crate alloc;
11
12mod batch;
13pub use batch::*;
14
15pub mod helpers;
16
17#[cfg(feature = "derive")]
18#[cfg_attr(docsrs, doc(cfg(feature = "derive")))]
19pub use ff_derive::PrimeField;
20
21#[cfg(feature = "bits")]
22#[cfg_attr(docsrs, doc(cfg(feature = "bits")))]
23pub use bitvec::view::BitViewSized;
24
25#[cfg(feature = "bits")]
26use bitvec::{array::BitArray, order::Lsb0};
27
28use core::fmt;
29use core::iter::{Product, Sum};
30use core::ops::{Add, AddAssign, Mul, MulAssign, Neg, Sub, SubAssign};
31
32use rand_core::{Rng, TryRng};
33use subtle::{Choice, ConditionallySelectable, ConstantTimeEq, CtOption};
34
35/// Bit representation of a field element.
36#[cfg(feature = "bits")]
37#[cfg_attr(docsrs, doc(cfg(feature = "bits")))]
38pub type FieldBits<V> = BitArray<V, Lsb0>;
39
40/// This trait represents an element of a field.
41pub trait Field:
42    Sized
43    + Eq
44    + Copy
45    + Clone
46    + Default
47    + Send
48    + Sync
49    + fmt::Debug
50    + 'static
51    + ConditionallySelectable
52    + ConstantTimeEq
53    + Neg<Output = Self>
54    + Add<Output = Self>
55    + Sub<Output = Self>
56    + Mul<Output = Self>
57    + Sum
58    + Product
59    + for<'a> Add<&'a Self, Output = Self>
60    + for<'a> Sub<&'a Self, Output = Self>
61    + for<'a> Mul<&'a Self, Output = Self>
62    + for<'a> Sum<&'a Self>
63    + for<'a> Product<&'a Self>
64    + AddAssign
65    + SubAssign
66    + MulAssign
67    + for<'a> AddAssign<&'a Self>
68    + for<'a> SubAssign<&'a Self>
69    + for<'a> MulAssign<&'a Self>
70{
71    /// The zero element of the field, the additive identity.
72    const ZERO: Self;
73
74    /// The one element of the field, the multiplicative identity.
75    const ONE: Self;
76
77    /// Returns an element chosen uniformly at random using a user-provided infallible RNG.
78    ///
79    /// This is a convenience wrapper around [`Field::try_random`] for RNGs that cannot
80    /// fail. Use [`Field::try_random`] if your RNG may fail (for example, an OS-backed
81    /// entropy source).
82    fn random<R: Rng + ?Sized>(rng: &mut R) -> Self {
83        let Ok(out) = Self::try_random(rng);
84        out
85    }
86
87    /// Returns an element chosen uniformly at random using a user-provided fallible RNG.
88    ///
89    /// Returns `Err` propagating the RNG's error if the underlying RNG fails to produce
90    /// the randomness required to sample an element. Implementors of `Field` must
91    /// provide this method; [`Field::random`] is derived from it for infallible RNGs.
92    fn try_random<R: TryRng + ?Sized>(rng: &mut R) -> Result<Self, R::Error>;
93
94    /// Returns true iff this element is zero.
95    fn is_zero(&self) -> Choice {
96        self.ct_eq(&Self::ZERO)
97    }
98
99    /// Returns true iff this element is zero.
100    ///
101    /// # Security
102    ///
103    /// This method provides **no** constant-time guarantees. Implementors of the
104    /// `Field` trait **may** optimise this method using non-constant-time logic.
105    fn is_zero_vartime(&self) -> bool {
106        self.is_zero().into()
107    }
108
109    /// Squares this element.
110    #[must_use]
111    fn square(&self) -> Self;
112
113    /// Cubes this element.
114    #[must_use]
115    fn cube(&self) -> Self {
116        self.square() * self
117    }
118
119    /// Doubles this element.
120    #[must_use]
121    fn double(&self) -> Self;
122
123    /// Computes the multiplicative inverse of this element,
124    /// failing if the element is zero.
125    fn invert(&self) -> CtOption<Self>;
126
127    /// Computes:
128    ///
129    /// - $(\textsf{true}, \sqrt{\textsf{num}/\textsf{div}})$, if $\textsf{num}$ and
130    ///   $\textsf{div}$ are nonzero and $\textsf{num}/\textsf{div}$ is a square in the
131    ///   field;
132    /// - $(\textsf{true}, 0)$, if $\textsf{num}$ is zero;
133    /// - $(\textsf{false}, 0)$, if $\textsf{num}$ is nonzero and $\textsf{div}$ is zero;
134    /// - $(\textsf{false}, \sqrt{G_S \cdot \textsf{num}/\textsf{div}})$, if
135    ///   $\textsf{num}$ and $\textsf{div}$ are nonzero and $\textsf{num}/\textsf{div}$ is
136    ///   a nonsquare in the field;
137    ///
138    /// where $G_S$ is a non-square.
139    ///
140    /// # Warnings
141    ///
142    /// - The choice of root from `sqrt` is unspecified.
143    /// - The value of $G_S$ is unspecified, and cannot be assumed to have any specific
144    ///   value in a generic context.
145    fn sqrt_ratio(num: &Self, div: &Self) -> (Choice, Self);
146
147    /// Equivalent to `Self::sqrt_ratio(self, one())`.
148    ///
149    /// The provided method is implemented in terms of [`Self::sqrt_ratio`].
150    fn sqrt_alt(&self) -> (Choice, Self) {
151        Self::sqrt_ratio(self, &Self::ONE)
152    }
153
154    /// Returns the square root of the field element, if it is
155    /// quadratic residue.
156    ///
157    /// The provided method is implemented in terms of [`Self::sqrt_ratio`].
158    fn sqrt(&self) -> CtOption<Self> {
159        let (is_square, res) = Self::sqrt_ratio(self, &Self::ONE);
160        CtOption::new(res, is_square)
161    }
162
163    /// Exponentiates `self` by `exp`, where `exp` is a little-endian order integer
164    /// exponent.
165    ///
166    /// # Guarantees
167    ///
168    /// This operation is constant time with respect to `self`, for all exponents with the
169    /// same number of digits (`exp.as_ref().len()`). It is variable time with respect to
170    /// the number of digits in the exponent.
171    fn pow<S: AsRef<[u64]>>(&self, exp: S) -> Self {
172        let mut res = Self::ONE;
173        for e in exp.as_ref().iter().rev() {
174            for i in (0..64).rev() {
175                res = res.square();
176                let mut tmp = res;
177                tmp *= self;
178                res.conditional_assign(&tmp, (((*e >> i) & 1) as u8).into());
179            }
180        }
181        res
182    }
183
184    /// Exponentiates `self` by `exp`, where `exp` is a little-endian order integer
185    /// exponent.
186    ///
187    /// # Guarantees
188    ///
189    /// **This operation is variable time with respect to `self`, for all exponent.** If
190    /// the exponent is fixed, this operation is effectively constant time. However, for
191    /// stronger constant-time guarantees, [`Field::pow`] should be used.
192    fn pow_vartime<S: AsRef<[u64]>>(&self, exp: S) -> Self {
193        let mut res = Self::ONE;
194        for e in exp.as_ref().iter().rev() {
195            for i in (0..64).rev() {
196                res = res.square();
197
198                if ((*e >> i) & 1) == 1 {
199                    res.mul_assign(self);
200                }
201            }
202        }
203
204        res
205    }
206}
207
208/// This represents an element of a non-binary prime field.
209pub trait PrimeField: Field + From<u64> {
210    /// The prime field can be converted back and forth into this binary
211    /// representation.
212    type Repr: Copy + Default + Send + Sync + 'static + AsRef<[u8]> + AsMut<[u8]>;
213
214    /// Interpret a string of numbers as a (congruent) prime field element.
215    /// Does not accept unnecessary leading zeroes or a blank string.
216    ///
217    /// # Security
218    ///
219    /// This method provides **no** constant-time guarantees.
220    fn from_str_vartime(s: &str) -> Option<Self> {
221        if s.is_empty() {
222            return None;
223        }
224
225        if s == "0" {
226            return Some(Self::ZERO);
227        }
228
229        let mut res = Self::ZERO;
230
231        let ten = Self::from(10);
232
233        let mut first_digit = true;
234
235        for c in s.chars() {
236            match c.to_digit(10) {
237                Some(c) => {
238                    if first_digit {
239                        if c == 0 {
240                            return None;
241                        }
242
243                        first_digit = false;
244                    }
245
246                    res.mul_assign(&ten);
247                    res.add_assign(&Self::from(u64::from(c)));
248                }
249                None => {
250                    return None;
251                }
252            }
253        }
254
255        Some(res)
256    }
257
258    /// Obtains a field element congruent to the integer `v`.
259    ///
260    /// For fields where `Self::CAPACITY >= 128`, this is injective and will produce a
261    /// unique field element.
262    ///
263    /// For fields where `Self::CAPACITY < 128`, this is surjective; some field elements
264    /// will be produced by multiple values of `v`.
265    ///
266    /// If you want to deterministically sample a field element representing a value, use
267    /// [`FromUniformBytes`] instead.
268    fn from_u128(v: u128) -> Self {
269        let lower = v as u64;
270        let upper = (v >> 64) as u64;
271        let mut tmp = Self::from(upper);
272        for _ in 0..64 {
273            tmp = tmp.double();
274        }
275        tmp + Self::from(lower)
276    }
277
278    /// Attempts to convert a byte representation of a field element into an element of
279    /// this prime field, failing if the input is not canonical (is not smaller than the
280    /// field's modulus).
281    ///
282    /// The byte representation is interpreted with the same endianness as elements
283    /// returned by [`PrimeField::to_repr`].
284    fn from_repr(repr: Self::Repr) -> CtOption<Self>;
285
286    /// Attempts to convert a byte representation of a field element into an element of
287    /// this prime field, failing if the input is not canonical (is not smaller than the
288    /// field's modulus).
289    ///
290    /// The byte representation is interpreted with the same endianness as elements
291    /// returned by [`PrimeField::to_repr`].
292    ///
293    /// # Security
294    ///
295    /// This method provides **no** constant-time guarantees. Implementors of the
296    /// `PrimeField` trait **may** optimise this method using non-constant-time logic.
297    fn from_repr_vartime(repr: Self::Repr) -> Option<Self> {
298        Self::from_repr(repr).into()
299    }
300
301    /// Converts an element of the prime field into the standard byte representation for
302    /// this field.
303    ///
304    /// The endianness of the byte representation is implementation-specific. Generic
305    /// encodings of field elements should be treated as opaque.
306    fn to_repr(&self) -> Self::Repr;
307
308    /// Returns true iff this element is odd.
309    fn is_odd(&self) -> Choice;
310
311    /// Returns true iff this element is even.
312    #[inline(always)]
313    fn is_even(&self) -> Choice {
314        !self.is_odd()
315    }
316
317    /// Modulus of the field written as a string for debugging purposes.
318    ///
319    /// The encoding of the modulus is implementation-specific. Generic users of the
320    /// `PrimeField` trait should treat this string as opaque.
321    const MODULUS: &'static str;
322
323    /// How many bits are needed to represent an element of this field.
324    const NUM_BITS: u32;
325
326    /// How many bits of information can be reliably stored in the field element.
327    ///
328    /// This is usually `Self::NUM_BITS - 1`.
329    const CAPACITY: u32;
330
331    /// Inverse of $2$ in the field.
332    const TWO_INV: Self;
333
334    /// A fixed multiplicative generator of `modulus - 1` order. This element must also be
335    /// a quadratic nonresidue.
336    ///
337    /// It can be calculated using [SageMath] as `GF(modulus).primitive_element()`.
338    ///
339    /// Implementations of this trait MUST ensure that this is the generator used to
340    /// derive `Self::ROOT_OF_UNITY`.
341    ///
342    /// [SageMath]: https://www.sagemath.org/
343    const MULTIPLICATIVE_GENERATOR: Self;
344
345    /// An integer `s` satisfying the equation `2^s * t = modulus - 1` with `t` odd.
346    ///
347    /// This is the number of leading zero bits in the little-endian bit representation of
348    /// `modulus - 1`.
349    const S: u32;
350
351    /// The `2^s` root of unity.
352    ///
353    /// It can be calculated by exponentiating `Self::MULTIPLICATIVE_GENERATOR` by `t`,
354    /// where `t = (modulus - 1) >> Self::S`.
355    const ROOT_OF_UNITY: Self;
356
357    /// Inverse of [`Self::ROOT_OF_UNITY`].
358    const ROOT_OF_UNITY_INV: Self;
359
360    /// Generator of the `t-order` multiplicative subgroup.
361    ///
362    /// It can be calculated by exponentiating [`Self::MULTIPLICATIVE_GENERATOR`] by `2^s`,
363    /// where `s` is [`Self::S`].
364    const DELTA: Self;
365}
366
367/// The subset of prime-order fields such that `(modulus - 1)` is divisible by `N`.
368///
369/// If `N` is prime, there will be `N - 1` valid choices of [`Self::ZETA`]. Similarly to
370/// [`PrimeField::MULTIPLICATIVE_GENERATOR`], the specific choice does not matter, as long
371/// as the choice is consistent across all uses of the field.
372pub trait WithSmallOrderMulGroup<const N: u8>: PrimeField {
373    /// A field element of small multiplicative order $N$.
374    ///
375    /// The presence of this element allows you to perform (certain types of)
376    /// endomorphisms on some elliptic curves.
377    ///
378    /// It can be calculated using [SageMath] as
379    /// `GF(modulus).primitive_element() ^ ((modulus - 1) // N)`.
380    /// Choosing the element of order $N$ that is smallest, when considered
381    /// as an integer, may help to ensure consistency.
382    ///
383    /// [SageMath]: https://www.sagemath.org/
384    const ZETA: Self;
385}
386
387/// Trait for constructing a [`PrimeField`] element from a fixed-length uniform byte
388/// array.
389///
390/// "Uniform" means that the byte array's contents must be indistinguishable from the
391/// [discrete uniform distribution]. Suitable byte arrays can be obtained:
392/// - from a cryptographically-secure randomness source (which makes this constructor
393///   equivalent to [`Field::random`]).
394/// - from a cryptographic hash function output, which enables a "random" field element to
395///   be selected deterministically. This is the primary use case for `FromUniformBytes`.
396///
397/// The length `N` of the byte array is chosen by the trait implementer such that the loss
398/// of uniformity in the mapping from byte arrays to field elements is cryptographically
399/// negligible.
400///
401/// [discrete uniform distribution]: https://en.wikipedia.org/wiki/Discrete_uniform_distribution
402///
403/// # Examples
404///
405/// ```
406/// # #[cfg(feature = "derive")] {
407/// # // Fake this so we don't actually need a dev-dependency on bls12_381.
408/// # mod bls12_381 {
409/// #     use ff::{Field, PrimeField};
410/// #
411/// #     #[derive(PrimeField)]
412/// #     #[PrimeFieldModulus = "52435875175126190479447740508185965837690552500527637822603658699938581184513"]
413/// #     #[PrimeFieldGenerator = "7"]
414/// #     #[PrimeFieldReprEndianness = "little"]
415/// #     pub struct Scalar([u64; 4]);
416/// #
417/// #     impl ff::FromUniformBytes<64> for Scalar {
418/// #         fn from_uniform_bytes(_bytes: &[u8; 64]) -> Self {
419/// #             // Fake impl for doctest
420/// #             Scalar::ONE
421/// #         }
422/// #     }
423/// # }
424/// #
425/// use blake2b_simd::blake2b;
426/// use bls12_381::Scalar;
427/// use ff::FromUniformBytes;
428///
429/// // `bls12_381::Scalar` implements `FromUniformBytes<64>`, and BLAKE2b (by default)
430/// // produces a 64-byte hash.
431/// let hash = blake2b(b"Some message");
432/// let val = Scalar::from_uniform_bytes(hash.as_array());
433/// # }
434/// ```
435///
436/// # Implementing `FromUniformBytes`
437///
438/// [`Self::from_uniform_bytes`] should always be implemented by interpreting the provided
439/// byte array as the little endian unsigned encoding of an integer, and then reducing that
440/// integer modulo the field modulus.
441///
442/// For security, `N` must be chosen so that `N * 8 >= Self::NUM_BITS + 128`. A larger
443/// value of `N` may be chosen for convenience; for example, for a field with a 255-bit
444/// modulus, `N = 64` is convenient as it matches the output length of several common
445/// cryptographic hash functions (such as SHA-512 and BLAKE2b).
446///
447/// ## Trait design
448///
449/// This trait exists because `PrimeField::from_uniform_bytes([u8; N])` cannot currently
450/// exist (trait methods cannot use associated constants in the const positions of their
451/// type signature, and we do not want `PrimeField` to require a generic const parameter).
452/// However, this has the side-effect that `FromUniformBytes` can be implemented multiple
453/// times for different values of `N`. Most implementations of [`PrimeField`] should only
454/// need to implement `FromUniformBytes` trait for one value of `N` (chosen following the
455/// above considerations); if you find yourself needing to implement it multiple times,
456/// please [let us know about your use case](https://github.com/zkcrypto/ff/issues/new) so
457/// we can take it into consideration for future evolutions of the `ff` traits.
458pub trait FromUniformBytes<const N: usize>: PrimeField {
459    /// Returns a field element that is congruent to the provided little endian unsigned
460    /// byte representation of an integer.
461    fn from_uniform_bytes(bytes: &[u8; N]) -> Self;
462}
463
464/// This represents the bits of an element of a prime field.
465#[cfg(feature = "bits")]
466#[cfg_attr(docsrs, doc(cfg(feature = "bits")))]
467pub trait PrimeFieldBits: PrimeField {
468    /// The backing store for a bit representation of a prime field element.
469    type ReprBits: BitViewSized + Send + Sync;
470
471    /// Converts an element of the prime field into a little-endian sequence of bits.
472    fn to_le_bits(&self) -> FieldBits<Self::ReprBits>;
473
474    /// Returns the bits of the field characteristic (the modulus) in little-endian order.
475    fn char_le_bits() -> FieldBits<Self::ReprBits>;
476}
477
478/// Functions and re-exported crates used by the [`PrimeField`] derive macro.
479#[cfg(feature = "derive")]
480#[cfg_attr(docsrs, doc(cfg(feature = "derive")))]
481pub mod derive {
482    pub use crate::arith_impl::*;
483
484    pub use {byteorder, rand_core, subtle};
485
486    #[cfg(feature = "bits")]
487    pub use bitvec;
488}
489
490#[cfg(feature = "derive")]
491mod arith_impl {
492    /// Computes `a - (b + borrow)`, returning the result and the new borrow.
493    #[inline(always)]
494    pub const fn sbb(a: u64, b: u64, borrow: u64) -> (u64, u64) {
495        let ret = (a as u128).wrapping_sub((b as u128) + ((borrow >> 63) as u128));
496        (ret as u64, (ret >> 64) as u64)
497    }
498
499    /// Computes `a + b + carry`, returning the result and the new carry over.
500    #[inline(always)]
501    pub const fn adc(a: u64, b: u64, carry: u64) -> (u64, u64) {
502        let ret = (a as u128) + (b as u128) + (carry as u128);
503        (ret as u64, (ret >> 64) as u64)
504    }
505
506    /// Computes `a + (b * c) + carry`, returning the result and the new carry over.
507    #[inline(always)]
508    pub const fn mac(a: u64, b: u64, c: u64, carry: u64) -> (u64, u64) {
509        let ret = (a as u128) + ((b as u128) * (c as u128)) + (carry as u128);
510        (ret as u64, (ret >> 64) as u64)
511    }
512}