mirror of
https://github.com/bitcoin/bips.git
synced 2026-02-09 15:23:09 +00:00
* Add Chaincode Delegation BIP * Update license to BSD-3-Clause and expand blinded signing documentation * Address initial PR comments * Update with BIP number assignment * Fix delegator_sign test vector * Upgrade secp256k1lab and add license file - Upgrade vendored secp256k1lab to commit a265da1 (adds type annotations) - Add COPYING file to satisfy MIT license requirements - Document secp256k1lab commit reference in BIP text * Fix type checker and linter issues in reference implementation - Fix TweakContext to use Scalar types for gacc/tacc - Replace HashFunction enum with Callable type alias - Fix bytearray to bytes conversion in blind_sign - Move imports to top of file - Fix boolean comparison style (use 'not' instead of '== False') - Add proper type annotations and casts for dict handling - Remove unused imports and type ignore comments * Address PR review comments on terminology and clarity - Add intro explaining delegation naming (chain code is delegated, not signing authority) - Reorder terminology to list Delegator before Delegatee - Replace "quorum" with clearer "can co-sign for UTXOs" language - Clarify derivation constraints in terms of delegatee's extended key - Rename "Delegatee Signing" section to "Signing Modes" - Fix "delegatee can apply" to "delegator can produce" (line 112) - Replace undefined "caller" with "delegatee" (line 173) - Clarify "Change outputs" to "Tweaks for change outputs" (line 98) - Add note that message is separate from CCD bundle - Add note on application-specific verification (addresses, amounts) - Add transition sentence clarifying non-concurrent protocol scope * Add changelog entry for 0.1.3 * Fix header: use Authors (plural) for multiple authors * Fix BIP header format for CI compliance - Change Type from 'Standards Track' to 'Specification' (valid type) - Change 'Created' to 'Assigned' (correct field name per BIP format) - Change 'Post-History' to 'Discussion' (recognized field in buildtable.pl) * Apply suggestion from @murchandamus --------- Co-authored-by: Jesse Posner <jesse.posner@gmail.com>
484 lines
17 KiB
Python
484 lines
17 KiB
Python
# Copyright (c) 2022-2023 The Bitcoin Core developers
|
|
# Distributed under the MIT software license, see the accompanying
|
|
# file COPYING or http://www.opensource.org/licenses/mit-license.php.
|
|
|
|
"""Test-only implementation of low-level secp256k1 field and group arithmetic
|
|
|
|
It is designed for ease of understanding, not performance.
|
|
|
|
WARNING: This code is slow and trivially vulnerable to side channel attacks. Do not use for
|
|
anything but tests.
|
|
|
|
Exports:
|
|
* FE: class for secp256k1 field elements
|
|
* GE: class for secp256k1 group elements
|
|
* G: the secp256k1 generator point
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
from typing import Self
|
|
|
|
# TODO Docstrings of methods still say "field element"
|
|
class APrimeFE:
|
|
"""Objects of this class represent elements of a prime field.
|
|
|
|
They are represented internally in numerator / denominator form, in order to delay inversions.
|
|
"""
|
|
|
|
# The size of the field (also its modulus and characteristic).
|
|
SIZE: int
|
|
|
|
def __init__(self, a: int | Self = 0, b: int | Self = 1) -> None:
|
|
"""Initialize a field element a/b; both a and b can be ints or field elements."""
|
|
if isinstance(a, type(self)):
|
|
num = a._num
|
|
den = a._den
|
|
else:
|
|
assert isinstance(a, int)
|
|
num = a % self.SIZE
|
|
den = 1
|
|
if isinstance(b, type(self)):
|
|
den = (den * b._num) % self.SIZE
|
|
num = (num * b._den) % self.SIZE
|
|
else:
|
|
assert isinstance(b, int)
|
|
den = (den * b) % self.SIZE
|
|
assert den != 0
|
|
if num == 0:
|
|
den = 1
|
|
self._num: int = num
|
|
self._den: int = den
|
|
|
|
def __add__(self, a: int | Self) -> Self:
|
|
"""Compute the sum of two field elements (second may be int)."""
|
|
if isinstance(a, type(self)):
|
|
return type(self)(self._num * a._den + self._den * a._num, self._den * a._den)
|
|
if isinstance(a, int):
|
|
return type(self)(self._num + self._den * a, self._den)
|
|
return NotImplemented
|
|
|
|
def __radd__(self, a: int) -> Self:
|
|
"""Compute the sum of an integer and a field element."""
|
|
return type(self)(a) + self
|
|
|
|
@classmethod
|
|
def sum(cls, *es: Self) -> Self:
|
|
"""Compute the sum of field elements.
|
|
|
|
sum(a, b, c, ...) is identical to (0 + a + b + c + ...)."""
|
|
return sum(es, start=cls(0))
|
|
|
|
def __sub__(self, a: int | Self) -> Self:
|
|
"""Compute the difference of two field elements (second may be int)."""
|
|
if isinstance(a, type(self)):
|
|
return type(self)(self._num * a._den - self._den * a._num, self._den * a._den)
|
|
if isinstance(a, int):
|
|
return type(self)(self._num - self._den * a, self._den)
|
|
return NotImplemented
|
|
|
|
def __rsub__(self, a: int) -> Self:
|
|
"""Compute the difference of an integer and a field element."""
|
|
return type(self)(a) - self
|
|
|
|
def __mul__(self, a: int | Self) -> Self:
|
|
"""Compute the product of two field elements (second may be int)."""
|
|
if isinstance(a, type(self)):
|
|
return type(self)(self._num * a._num, self._den * a._den)
|
|
if isinstance(a, int):
|
|
return type(self)(self._num * a, self._den)
|
|
return NotImplemented
|
|
|
|
def __rmul__(self, a: int) -> Self:
|
|
"""Compute the product of an integer with a field element."""
|
|
return type(self)(a) * self
|
|
|
|
def __truediv__(self, a: int | Self) -> Self:
|
|
"""Compute the ratio of two field elements (second may be int)."""
|
|
if isinstance(a, type(self)) or isinstance(a, int):
|
|
return type(self)(self, a)
|
|
return NotImplemented
|
|
|
|
def __pow__(self, a: int) -> Self:
|
|
"""Raise a field element to an integer power."""
|
|
return type(self)(pow(self._num, a, self.SIZE), pow(self._den, a, self.SIZE))
|
|
|
|
def __neg__(self) -> Self:
|
|
"""Negate a field element."""
|
|
return type(self)(-self._num, self._den)
|
|
|
|
def __int__(self) -> int:
|
|
"""Convert a field element to an integer in range 0..SIZE-1. The result is cached."""
|
|
if self._den != 1:
|
|
self._num = (self._num * pow(self._den, -1, self.SIZE)) % self.SIZE
|
|
self._den = 1
|
|
return self._num
|
|
|
|
def sqrt(self) -> Self | None:
|
|
"""Compute the square root of a field element if it exists (None otherwise)."""
|
|
raise NotImplementedError
|
|
|
|
def is_square(self) -> bool:
|
|
"""Determine if this field element has a square root."""
|
|
# A more efficient algorithm is possible here (Jacobi symbol).
|
|
return self.sqrt() is not None
|
|
|
|
def is_even(self) -> bool:
|
|
"""Determine whether this field element, represented as integer in 0..SIZE-1, is even."""
|
|
return int(self) & 1 == 0
|
|
|
|
def __eq__(self, a: object) -> bool:
|
|
"""Check whether two field elements are equal (second may be an int)."""
|
|
if isinstance(a, type(self)):
|
|
return (self._num * a._den - self._den * a._num) % self.SIZE == 0
|
|
elif isinstance(a, int):
|
|
return (self._num - self._den * a) % self.SIZE == 0
|
|
return False # for other types
|
|
|
|
def to_bytes(self) -> bytes:
|
|
"""Convert a field element to a 32-byte array (BE byte order)."""
|
|
return int(self).to_bytes(32, 'big')
|
|
|
|
@classmethod
|
|
def from_int_checked(cls, v: int) -> Self:
|
|
"""Convert an integer to a field element (no overflow allowed)."""
|
|
if v >= cls.SIZE:
|
|
raise ValueError
|
|
return cls(v)
|
|
|
|
@classmethod
|
|
def from_int_wrapping(cls, v: int) -> Self:
|
|
"""Convert an integer to a field element (reduced modulo SIZE)."""
|
|
return cls(v % cls.SIZE)
|
|
|
|
@classmethod
|
|
def from_bytes_checked(cls, b: bytes) -> Self:
|
|
"""Convert a 32-byte array to a field element (BE byte order, no overflow allowed)."""
|
|
v = int.from_bytes(b, 'big')
|
|
return cls.from_int_checked(v)
|
|
|
|
@classmethod
|
|
def from_bytes_wrapping(cls, b: bytes) -> Self:
|
|
"""Convert a 32-byte array to a field element (BE byte order, reduced modulo SIZE)."""
|
|
v = int.from_bytes(b, 'big')
|
|
return cls.from_int_wrapping(v)
|
|
|
|
def __str__(self) -> str:
|
|
"""Convert this field element to a 64 character hex string."""
|
|
return f"{int(self):064x}"
|
|
|
|
def __repr__(self) -> str:
|
|
"""Get a string representation of this field element."""
|
|
return f"{type(self).__qualname__}(0x{int(self):x})"
|
|
|
|
|
|
class FE(APrimeFE):
|
|
SIZE = 2**256 - 2**32 - 977
|
|
|
|
def sqrt(self) -> Self | None:
|
|
# Due to the fact that our modulus p is of the form (p % 4) == 3, the Tonelli-Shanks
|
|
# algorithm (https://en.wikipedia.org/wiki/Tonelli-Shanks_algorithm) is simply
|
|
# raising the argument to the power (p + 1) / 4.
|
|
|
|
# To see why: (p-1) % 2 = 0, so 2 divides the order of the multiplicative group,
|
|
# and thus only half of the non-zero field elements are squares. An element a is
|
|
# a (nonzero) square when Euler's criterion, a^((p-1)/2) = 1 (mod p), holds. We're
|
|
# looking for x such that x^2 = a (mod p). Given a^((p-1)/2) = 1, that is equivalent
|
|
# to x^2 = a^(1 + (p-1)/2) mod p. As (1 + (p-1)/2) is even, this is equivalent to
|
|
# x = a^((1 + (p-1)/2)/2) mod p, or x = a^((p+1)/4) mod p.
|
|
v = int(self)
|
|
s = pow(v, (self.SIZE + 1) // 4, self.SIZE)
|
|
if s**2 % self.SIZE == v:
|
|
return type(self)(s)
|
|
return None
|
|
|
|
|
|
class Scalar(APrimeFE):
|
|
"""TODO Docstring"""
|
|
SIZE = 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFEBAAEDCE6AF48A03BBFD25E8CD0364141
|
|
|
|
@classmethod
|
|
def from_int_nonzero_checked(cls, v: int) -> Self:
|
|
"""Convert an integer to a scalar (no zero or overflow allowed)."""
|
|
if not (0 < v < cls.SIZE):
|
|
raise ValueError
|
|
return cls(v)
|
|
|
|
@classmethod
|
|
def from_bytes_nonzero_checked(cls, b: bytes) -> Self:
|
|
"""Convert a 32-byte array to a scalar (BE byte order, no zero or overflow allowed)."""
|
|
v = int.from_bytes(b, 'big')
|
|
return cls.from_int_nonzero_checked(v)
|
|
|
|
|
|
class GE:
|
|
"""Objects of this class represent secp256k1 group elements (curve points or infinity)
|
|
|
|
GE objects are immutable.
|
|
|
|
Normal points on the curve have fields:
|
|
* x: the x coordinate (a field element)
|
|
* y: the y coordinate (a field element, satisfying y^2 = x^3 + 7)
|
|
* infinity: False
|
|
|
|
The point at infinity has field:
|
|
* infinity: True
|
|
"""
|
|
|
|
# TODO The following two class attributes should probably be just getters as
|
|
# classmethods to enforce immutability. Unfortunately Python makes it hard
|
|
# to create "classproperties". `G` could then also be just a classmethod.
|
|
|
|
# Order of the group (number of points on the curve, plus 1 for infinity)
|
|
ORDER = Scalar.SIZE
|
|
|
|
# Number of valid distinct x coordinates on the curve.
|
|
ORDER_HALF = ORDER // 2
|
|
|
|
@property
|
|
def infinity(self) -> bool:
|
|
"""Whether the group element is the point at infinity."""
|
|
return self._infinity
|
|
|
|
@property
|
|
def x(self) -> FE:
|
|
"""The x coordinate (a field element) of a non-infinite group element."""
|
|
assert not self.infinity
|
|
return self._x
|
|
|
|
@property
|
|
def y(self) -> FE:
|
|
"""The y coordinate (a field element) of a non-infinite group element."""
|
|
assert not self.infinity
|
|
return self._y
|
|
|
|
def __init__(self, x: int | FE | None = None, y: int | FE | None = None) -> None:
|
|
"""Initialize a group element with specified x and y coordinates, or infinity."""
|
|
if x is None:
|
|
# Initialize as infinity.
|
|
assert y is None
|
|
self._infinity = True
|
|
else:
|
|
# Initialize as point on the curve (and check that it is).
|
|
assert x is not None
|
|
assert y is not None
|
|
fx = FE(x)
|
|
fy = FE(y)
|
|
assert fy**2 == fx**3 + 7
|
|
self._infinity = False
|
|
self._x = fx
|
|
self._y = fy
|
|
|
|
def __add__(self, a: GE) -> GE:
|
|
"""Add two group elements together."""
|
|
# Deal with infinity: a + infinity == infinity + a == a.
|
|
if self.infinity:
|
|
return a
|
|
if a.infinity:
|
|
return self
|
|
if self.x == a.x:
|
|
if self.y != a.y:
|
|
# A point added to its own negation is infinity.
|
|
assert self.y + a.y == 0
|
|
return GE()
|
|
else:
|
|
# For identical inputs, use the tangent (doubling formula).
|
|
lam = (3 * self.x**2) / (2 * self.y)
|
|
else:
|
|
# For distinct inputs, use the line through both points (adding formula).
|
|
lam = (self.y - a.y) / (self.x - a.x)
|
|
# Determine point opposite to the intersection of that line with the curve.
|
|
x = lam**2 - (self.x + a.x)
|
|
y = lam * (self.x - x) - self.y
|
|
return GE(x, y)
|
|
|
|
@staticmethod
|
|
def sum(*ps: GE) -> GE:
|
|
"""Compute the sum of group elements.
|
|
|
|
GE.sum(a, b, c, ...) is identical to (GE() + a + b + c + ...)."""
|
|
return sum(ps, start=GE())
|
|
|
|
@staticmethod
|
|
def batch_mul(*aps: tuple[Scalar, GE]) -> GE:
|
|
"""Compute a (batch) scalar group element multiplication.
|
|
|
|
GE.batch_mul((a1, p1), (a2, p2), (a3, p3)) is identical to a1*p1 + a2*p2 + a3*p3,
|
|
but more efficient."""
|
|
# Reduce all the scalars modulo order first (so we can deal with negatives etc).
|
|
naps = [(int(a), p) for a, p in aps]
|
|
# Start with point at infinity.
|
|
r = GE()
|
|
# Iterate over all bit positions, from high to low.
|
|
for i in range(255, -1, -1):
|
|
# Double what we have so far.
|
|
r = r + r
|
|
# Add then add the points for which the corresponding scalar bit is set.
|
|
for (a, p) in naps:
|
|
if (a >> i) & 1:
|
|
r += p
|
|
return r
|
|
|
|
def __rmul__(self, a: int | Scalar) -> GE:
|
|
"""Multiply an integer or scalar with a group element."""
|
|
if self == G:
|
|
return FAST_G.mul(Scalar(a))
|
|
return GE.batch_mul((Scalar(a), self))
|
|
|
|
def __neg__(self) -> GE:
|
|
"""Compute the negation of a group element."""
|
|
if self.infinity:
|
|
return self
|
|
return GE(self.x, -self.y)
|
|
|
|
def __sub__(self, a: GE) -> GE:
|
|
"""Subtract a group element from another."""
|
|
return self + (-a)
|
|
|
|
def __eq__(self, a: object) -> bool:
|
|
"""Check if two group elements are equal."""
|
|
if not isinstance(a, type(self)):
|
|
return False
|
|
return (self - a).infinity
|
|
|
|
def has_even_y(self) -> bool:
|
|
"""Determine whether a non-infinity group element has an even y coordinate."""
|
|
assert not self.infinity
|
|
return self.y.is_even()
|
|
|
|
def to_bytes_compressed(self) -> bytes:
|
|
"""Convert a non-infinite group element to 33-byte compressed encoding."""
|
|
assert not self.infinity
|
|
return bytes([3 - self.y.is_even()]) + self.x.to_bytes()
|
|
|
|
def to_bytes_compressed_with_infinity(self) -> bytes:
|
|
"""Convert a group element to 33-byte compressed encoding, mapping infinity to zeros."""
|
|
if self.infinity:
|
|
return 33 * b"\x00"
|
|
return self.to_bytes_compressed()
|
|
|
|
def to_bytes_uncompressed(self) -> bytes:
|
|
"""Convert a non-infinite group element to 65-byte uncompressed encoding."""
|
|
assert not self.infinity
|
|
return b'\x04' + self.x.to_bytes() + self.y.to_bytes()
|
|
|
|
def to_bytes_xonly(self) -> bytes:
|
|
"""Convert (the x coordinate of) a non-infinite group element to 32-byte xonly encoding."""
|
|
assert not self.infinity
|
|
return self.x.to_bytes()
|
|
|
|
@staticmethod
|
|
def lift_x(x: int | FE) -> GE:
|
|
"""Return group element with specified field element as x coordinate (and even y)."""
|
|
y = (FE(x)**3 + 7).sqrt()
|
|
if y is None:
|
|
raise ValueError
|
|
if not y.is_even():
|
|
y = -y
|
|
return GE(x, y)
|
|
|
|
@staticmethod
|
|
def from_bytes_compressed(b: bytes) -> GE:
|
|
"""Convert a compressed to a group element."""
|
|
assert len(b) == 33
|
|
if b[0] != 2 and b[0] != 3:
|
|
raise ValueError
|
|
x = FE.from_bytes_checked(b[1:])
|
|
r = GE.lift_x(x)
|
|
if b[0] == 3:
|
|
r = -r
|
|
return r
|
|
|
|
@staticmethod
|
|
def from_bytes_compressed_with_infinity(b: bytes) -> GE:
|
|
"""Convert a compressed to a group element, mapping zeros to infinity."""
|
|
if b == 33 * b"\x00":
|
|
return GE()
|
|
else:
|
|
return GE.from_bytes_compressed(b)
|
|
|
|
@staticmethod
|
|
def from_bytes_uncompressed(b: bytes) -> GE:
|
|
"""Convert an uncompressed to a group element."""
|
|
assert len(b) == 65
|
|
if b[0] != 4:
|
|
raise ValueError
|
|
x = FE.from_bytes_checked(b[1:33])
|
|
y = FE.from_bytes_checked(b[33:])
|
|
if y**2 != x**3 + 7:
|
|
raise ValueError
|
|
return GE(x, y)
|
|
|
|
@staticmethod
|
|
def from_bytes(b: bytes) -> GE:
|
|
"""Convert a compressed or uncompressed encoding to a group element."""
|
|
assert len(b) in (33, 65)
|
|
if len(b) == 33:
|
|
return GE.from_bytes_compressed(b)
|
|
else:
|
|
return GE.from_bytes_uncompressed(b)
|
|
|
|
@staticmethod
|
|
def from_bytes_xonly(b: bytes) -> GE:
|
|
"""Convert a point given in xonly encoding to a group element."""
|
|
assert len(b) == 32
|
|
x = FE.from_bytes_checked(b)
|
|
r = GE.lift_x(x)
|
|
return r
|
|
|
|
@staticmethod
|
|
def is_valid_x(x: int | FE) -> bool:
|
|
"""Determine whether the provided field element is a valid X coordinate."""
|
|
return (FE(x)**3 + 7).is_square()
|
|
|
|
def __str__(self) -> str:
|
|
"""Convert this group element to a string."""
|
|
if self.infinity:
|
|
return "(inf)"
|
|
return f"({self.x},{self.y})"
|
|
|
|
def __repr__(self) -> str:
|
|
"""Get a string representation for this group element."""
|
|
if self.infinity:
|
|
return "GE()"
|
|
return f"GE(0x{int(self.x):x},0x{int(self.y):x})"
|
|
|
|
def __hash__(self) -> int:
|
|
"""Compute a non-cryptographic hash of the group element."""
|
|
if self.infinity:
|
|
return 0 # 0 is not a valid x coordinate
|
|
return int(self.x)
|
|
|
|
|
|
# The secp256k1 generator point
|
|
G = GE.lift_x(0x79BE667EF9DCBBAC55A06295CE870B07029BFCDB2DCE28D959F2815B16F81798)
|
|
|
|
|
|
class FastGEMul:
|
|
"""Table for fast multiplication with a constant group element.
|
|
|
|
Speed up scalar multiplication with a fixed point P by using a precomputed lookup table with
|
|
its powers of 2:
|
|
|
|
table = [P, 2*P, 4*P, (2^3)*P, (2^4)*P, ..., (2^255)*P]
|
|
|
|
During multiplication, the points corresponding to each bit set in the scalar are added up,
|
|
i.e. on average ~128 point additions take place.
|
|
"""
|
|
|
|
def __init__(self, p: GE) -> None:
|
|
self.table: list[GE] = [p] # table[i] = (2^i) * p
|
|
for _ in range(255):
|
|
p = p + p
|
|
self.table.append(p)
|
|
|
|
def mul(self, a: Scalar | int) -> GE:
|
|
result = GE()
|
|
a_ = int(a)
|
|
for bit in range(a_.bit_length()):
|
|
if a_ & (1 << bit):
|
|
result += self.table[bit]
|
|
return result
|
|
|
|
# Precomputed table with multiples of G for fast multiplication
|
|
FAST_G = FastGEMul(G)
|