path: root/aesgcmanalysis.py

from binascii import unhexlify
import random, struct, hmac, itertools, math
from Crypto.Cipher import AES
import cryptography
from cryptography.hazmat.primitives.ciphers import Cipher, algorithms, modes
import numpy as np
import xarray as xr

## Computation in GF(2^128)/(x^128 + x^7 + x^2 + x^1 + 1)
## Elements are represented as an integer n where (n & (1 << i)) is the coefficient for x^i.

class X:
    """A helper class for specifying elements of GF(2^n)."""
    def __pow__(self, y):
        return 1 << y
x = X()
gcm_modulus = x**128 + x**7 + x**2 + x**1 + 1
gcm_modulus_degree = 128

def gf128_deg(x):
    """Compute the degree of x.
    >>> assert gf128_deg(x**20 + x**5 + 1) == 20
    >>> assert gf128_deg(x**90) == 90
    >>> assert gf128_deg(x**1) == 1
    >>> assert gf128_deg(1) == 0
    >>> assert gf128_deg(0) == -1
    """
    return x.bit_length() - 1

def gf128_add(a, b):
    """Compute a + b.
    >>> assert gf128_add(0, 0) == 0
    >>> assert gf128_add(x**7 + x**2, 0) == x**7 + x**2
    >>> assert gf128_add(x**7 + x**2, x**2 + x**1) == x**7 + x**1
    >>> assert gf128_add(x**7 + x**2, x**7 + x**2) == 0
    """
    return a ^ b

def gf128_mul(a, b):
    """Compute ab.
    >>> assert gf128_mul(x**20 + x**5 + 1, x**13 + x**2) == x**33 + x**22 + x**18 + x**13 + x**7 + x**2
    >>> assert gf128_mul(x**80, x**90) == x**49 + x**44 + x**43 + x**42
    >>> assert gf128_mul(x**23 + x**5, 1) == x**23 + x**5
    """
    if a == 0 or b == 0:
        return 0
    if a > b:
        return gf128_mul(b, a)
    p = 0
    deg_b = gf128_deg(b)
    while a > 0:
        if a & 1:
            p ^= b
        a = a >> 1
        b = b << 1
        deg_b += 1
        if deg_b == gcm_modulus_degree:
            b = b ^ gcm_modulus
            deg_b = gf128_deg(b)
    return p

def gf128_divmod(a, b):
    """Compute q, r such that a = qb + r; gf128_deg(r) < gf128_deg(b).
    >>> assert gf128_divmod(x**20 + x**5 + 1, x**13 + x**2) == (x**7, x**9 + x**5 + 1)
    >>> assert gf128_divmod(x**90, x**80) == (x**10, 0)
    """
    q, r = 0, a
    while gf128_deg(r) >= gf128_deg(b):
        d = gf128_deg(r) - gf128_deg(b)
        q = q ^ (1 << d)
        r = r ^ (b << d)
    return q, r

def gf128_mod(a, m=gcm_modulus):
    """Compute b such that a = b; gf128_deg(b) < gf128_deg(m).
    >>> assert gf128_mod(x**220 + x**5 + 1, x**50 + x**4) == x**36 + x**5 + 1
    >>> assert gf128_mod(x**130 + x**64 + x**2) == x**64 + x**9 + x**4 + x**3
    """
    q, r = gf128_divmod(a, m)
    return r

def gf128_egcd(a, b):
    """Compute g, x, y such that g | a, g | b, g = ax + by;
    for any h such that h | a and h | b, gf128_deg(g) >= gf128_deg(h).
    >>> assert gf128_egcd(x**2 + 1, x**1 + 1) == (x**1 + 1, 0, 1)
    >>> assert gf128_egcd(x**12 + x**4, x**5 + 1) == (x**1 + 1, x**3 + x**2 + 1, x**10 + x**9 + x**7 + x**5 + x**4 + x**1 + 1)
    >>> assert gf128_egcd(x**64 + x**2, x**37 + x**12 + 1) == (1, x**35 + x**34 + x**33 + x**31 + x**30 + x**28 + x**27 + x**25 + x**24 + x**21 + x**20 + x**18 + x**17 + x**15 + x**14 + x**12 + x**11 + x**9 + x**7 + x**6 + x**4 + x**3 + x**1 + 1, x**62 + x**61 + x**60 + x**58 + x**57 + x**55 + x**54 + x**52 + x**51 + x**48 + x**47 + x**45 + x**44 + x**42 + x**41 + x**39 + x**38 + x**37 + x**35 + x**34 + x**32 + x**31 + x**29 + x**28 + x**26 + x**25 + x**24 + x**22 + x**21 + x**19 + x**18 + x**16 + x**15 + x**13 + x**12 + x**11 + x**9 + x**8 + x**6 + x**5 + x**3 + x**2 + 1)
    """
    orig_a = a
    orig_b = b
    a_x = 1
    a_y = 0
    b_x = 0
    b_y = 1
    # Invariant = a_x*x + a_y*y = a, b_x*x + b_y*y = b, r_x*r + r_y*y = r
    r_x = 0
    r_y = 1
    if a == 0:
        return b, r_x, r_y
    while True:
        q, r = gf128_divmod(b, a)
        if r == 0:
            # FIX mul so mod isn't necessary; with zeros
            assert a == gf128_mod(gf128_add(gf128_mul(orig_a, r_x), gf128_mul(orig_b, r_y)))
            return a, r_x, r_y
        r_x = gf128_add(b_x, gf128_mul(q, a_x))
        r_y = gf128_add(b_y, gf128_mul(q, a_y))
        a, b = r, a
        b_x, b_y = a_x, a_y
        a_x, a_y = r_x, r_y

def gf128_inv(a):
    """Compute b such that ab = 1.
    >>> assert gf128_inv(x**1) == x**127 + x**6 + x**1 + 1
    >>> assert gf128_inv(x**4) == x**127 + x**125 + x**124 + x**6 + x**4 + x**3 + x**1 + 1
    """
    g, xx, yy = gf128_egcd(a, gcm_modulus)
    assert g == 1
    return xx

def gf128_exp(a, n):
    """Compute a^n.
    >>> assert gf128_exp(x**2 + 1, 0) == 1
    >>> assert gf128_exp(x**2 + 1, 1) == x**2 + 1
    >>> assert gf128_exp(x**2 + 1, 2) == x**4 + 1
    >>> assert gf128_exp(x**2 + 1, 10) == x**20 + x**16 + x**4 + 1
    """
    if n == 0:
        return 1
    rec = gf128_exp(a, n//2)
    rec = gf128_mul(rec, rec)
    if n % 2 == 1:
        rec = gf128_mul(rec, a)
    return rec

def gf128_sqrt(x):
    """Compute y such that y^2 = x: the inverse Frobenius automorphism.
    Reference: https://math.stackexchange.com/questions/943417/square-root-for-galois-fields-gf2m
    >>> assert gf128_sqrt(1) == 1
    >>> assert gf128_sqrt(x**2 + 1) == x**1 + 1
    >>> assert gf128_sqrt(x**3 + 1) == x**126 + x**123 + x**120 + x**117 + x**114 + x**111 + x**108 + x**105 + x**102 + x**99 + x**96 + x**93 + x**90 + x**87 + x**84 + x**81 + x**78 + x**75 + x**72 + x**69 + x**66 + x**63 + x**62 + x**60 + x**59 + x**57 + x**56 + x**54 + x**53 + x**51 + x**50 + x**48 + x**47 + x**45 + x**44 + x**42 + x**41 + x**39 + x**38 + x**36 + x**35 + x**33 + x**32 + x**30 + x**29 + x**27 + x**26 + x**24 + x**23 + x**21 + x**20 + x**18 + x**17 + x**15 + x**14 + x**12 + x**11 + x**9 + x**8 + x**6 + x**3 + 1
    """
    return gf128_exp(x, 2**127)

def gf128_to_vec(x):
    return [int(n) for n in bin(x)[2:].zfill(128)[::-1]]

def vec_to_gf128(vs):
    x = 0
    for i, v in enumerate(vs):
        if v:
            x += (1 << i)
    return x

def reverse_mask(b):
    # from https://stackoverflow.com/a/2602885
    b = (b & 0xF0) >> 4 | (b & 0x0F) << 4;
    b = (b & 0xCC) >> 2 | (b & 0x33) << 2;
    b = (b & 0xAA) >> 1 | (b & 0x55) << 1;
    return b;

def bytes_to_gf128(xs, st=None, en=None):
    if st is None and en is None:
        st = 0
        en = 16
    a, b = struct.unpack('<QQ', bytes(reverse_mask(xs[i]) for i in range(st, en)))
    return a + (b << 64)

def gf128_to_bytes(n):
    s = b''
    while n != 0:
        c = n & 0xff
        f = 0
        for i in range(8):
            if c & (1 << i):
                f += (1 << (7-i))
        s += int.to_bytes(f, 1, 'big')
        n >>= 8
    s += b'\x00' * (16 - len(s))
    assert len(s) == 16
    return s

def Ms():
    vs = []
    for i in range(0, 128):
        v = gf128_to_vec(gf128_mod(1 << (2*i)))
        vs.append(v)
    return np.array(vs).transpose()

def Mc(c):
    vs = []
    pr = c
    for i in range(0, 128):
        vs.append(gf128_to_vec(pr))
        pr = gf128_mul(x**1, pr)
    return np.array(vs).transpose()

# Adapted from Wiki
def rref_mod_2(M):
    M = M.copy().astype('int')

    lead = 0
    rowCount, columnCount = M.shape
    adj = np.eye(rowCount, rowCount).astype('int')

    for r in range(rowCount):
        if columnCount <= lead:
            return M, adj
        i = r
        while M[i][lead] == 0:
            i = i + 1
            if rowCount == i:
                i = r
                lead = lead + 1
                if columnCount == lead:
                    return M, adj
        if i != r:
            M[[i, r]] = M[[r, i]]
            adj[[i, r]] = adj[[r, i]]
        for j in range(rowCount):
            if M[j][lead] == 1 and j != r:
                M[j] ^= M[r]
                adj[j] ^= adj[r]
        lead = lead + 1
    return M, adj

def kernel(M, f):
    mt = M.transpose()
    N, adj = f(mt)
    basis = []
    for i, row in enumerate(N):
        if np.all(np.isclose(row, 0)):
            basis.append(adj[i])
    return N, adj, basis

## Computation in GF(2^128)[X]/(x^128 + x^7 + x^2 + x^1 + 1)
## Elements are represented as arrays where the ith element is the coefficient for x^i.

def gf128poly_lead(x):
    """Compute the leading coefficient of x.
    >>> assert gf128poly_lead([x**3, x**1, 1]) == 1
    >>> assert gf128poly_lead([x**3, x**1, x**2, x**4]) == x**4
    >>> assert gf128poly_lead([1]) == 1
    """
    assert len(x) > 0
    return x[-1]

def collapse(x):
    if len(x) == 0:
        return []
    if gf128poly_lead(x) == 0:
        return collapse(x[:-1])
    return [gf128_mod(y, gcm_modulus) for y in x]

def gf128poly_deg(x):
    """Compute the degree of x.
    >>> assert gf128poly_deg([x**3, x**1, 1]) == 2
    >>> assert gf128poly_deg([x**2, x**3]) == 1
    >>> assert gf128poly_deg([x**15]) == 0
    >>> assert gf128poly_deg([1]) == 0
    >>> assert gf128poly_deg([]) == -1
    """
    return len(x) - 1

def gf128poly_add(a, b):
    """Compute a + b.
    >>> assert gf128poly_add([x**3, x**1, 1], [x**2 + 1, x**3]) == [x**3 + x**2 + 1, x**1 + x**3, 1]
    >>> assert gf128poly_add([x**2, x**3], [x**5, x**12, x**2, x**4]) == [x**5 + x**2, x**12 + x**3, x**2, x**4]
    >>> assert gf128poly_add([x**15], []) == [x**15]
    >>> assert gf128poly_add([x**15], [x**15]) == []
    """
    return collapse([gf128_add(x, y) for x, y in itertools.zip_longest(a, b, fillvalue=0)])

def gf128poly_mul(a, b):
    """Compute ab.
    >>> assert gf128poly_mul([x**3, x**1, 1], [x**2 + 1, x**3]) == [x**5 + x**3, x**6 + x**3 + x**1, x**4 + x**2 + 1, x**3]
    >>> assert gf128poly_mul([x**2, x**3], [x**5, x**12, x**2, x**4]) == [x**7, x**14 + x**8, x**15 + x**4, x**6 + x**5, x**7]
    >>> assert gf128poly_mul([x**15], []) == []
    >>> assert gf128poly_mul([x**15], [1]) == [x**15]
    >>> assert gf128poly_mul([0, x**15], [0, x**15]) == [0, 0, x**30]
    >>> assert gf128poly_mul([x**70], [1, x**70, 1]) == [x**70, x**19 + x**14 + x**13 + x**12, x**70]
    """
    p = []
    while gf128poly_deg(a) >= 0:
        p = gf128poly_add(p, [gf128_mul(x, a[0]) for x in b])
        a = a[1:]
        b = [0] + b
    return collapse(p)

def gf128poly_divmod(a, b):
    """Compute q, r such that a = qb + r; gf128poly_deg(r) < gf128poly_deg(b).
    >>> assert gf128poly_divmod([x**5 + x**3, x**6 + x**3 + x**1, x**4 + x**2 + 1, x**3], [x**2 + 1, x**3]) == ([x**3, x**1, 1], [])
    >>> assert gf128poly_divmod([x**7, x**14 + x**8, x**15 + x**4, x**6 + x**5, x**7], [x**2, x**3]) == ([x**5, x**12, x**2, x**4], [])
    >>> assert gf128poly_divmod([x**15], [1]) == ([x**15], [])
    >>> assert gf128poly_divmod([x**15], [x**15]) == ([1], [])
    >>> assert gf128poly_divmod([0, 0, x**30], [0, x**15]) == ([0, x**15], [])
    >>> assert gf128poly_divmod([x**70, x**19 + x**14 + x**13 + x**12, x**70], [1, x**70, 1]) == ([x**70], [])
    >>> assert gf128poly_divmod([x**18, x**18, x**3], [x**3, x**3]) == ([x**15 + 1, 1], [x**3])
    """
    assert b != [], 'divide by zero'
    q = []
    while gf128poly_deg(a) >= gf128poly_deg(b):
        highA = gf128poly_lead(a)
        highB = gf128poly_lead(b)
        highBinv = 1 if highB == 1 else gf128_inv(highB)
        qq = gf128_mul(highA, highBinv)
        de = gf128poly_deg(a) - gf128poly_deg(b)
        new_term = [0]*(de+1)
        new_term[de] = qq
        q = gf128poly_add(q, new_term)
        fact = gf128poly_mul(b, new_term)
        a = gf128poly_add(a, fact)
    return collapse(q), collapse(a)

def gf128poly_sqrt(p):
    """Compute q such that q^2 = p for p such that p' = 0.
    >>> assert gf128poly_sqrt([1, 0, 1, 0, 1, 0, 1]) == [1, 1, 1, 1]
    >>> assert gf128poly_sqrt([1, 0, 1, 0, 0, 0, 1]) == [1, 1, 0, 1]
    >>> assert gf128poly_sqrt([1, 0, x**4]) == [1, x**2]
    >>> assert gf128poly_sqrt([1, 0, x**3]) == [1, x**126 + x**123 + x**120 + x**117 + x**114 + x**111 + x**108 + x**105 + x**102 + x**99 + x**96 + x**93 + x**90 + x**87 + x**84 + x**81 + x**78 + x**75 + x**72 + x**69 + x**66 + x**63 + x**62 + x**60 + x**59 + x**57 + x**56 + x**54 + x**53 + x**51 + x**50 + x**48 + x**47 + x**45 + x**44 + x**42 + x**41 + x**39 + x**38 + x**36 + x**35 + x**33 + x**32 + x**30 + x**29 + x**27 + x**26 + x**24 + x**23 + x**21 + x**20 + x**18 + x**17 + x**15 + x**14 + x**12 + x**11 + x**9 + x**8 + x**6 + x**3]
    """
    assert gf128poly_formal_derivative(p) == []
    q = []
    for x in p:
        if x == 0 or x == 1:
            q.append(x)
        else:
            q.append(gf128_sqrt(x))
    for i in range(2, len(q), 2):
        q[i//2] = q[i]
        q[i] = 0
    return collapse(q)

def gf128poly_monic(p):
    """Compute p/p_lead for p such that gf128poly_deg(p) >= 0.
    >>> assert gf128poly_monic([x**54, x**2]) == [x**52, 1]
    >>> assert gf128poly_monic([x**2, x**3]) == [x**127 + x**6 + x**1 + 1, 1]
    """
    assert len(p) > 0
    lead = gf128poly_lead(p)
    if lead == 1:
        return p
    lead_inv = gf128_inv(lead)
    return [gf128_mul(x, lead_inv) for x in p]

def gf128poly_formal_derivative(p):
    """Compute p'.
    >>> assert gf128poly_formal_derivative([]) == []
    >>> assert gf128poly_formal_derivative([x**25]) == []
    >>> assert gf128poly_formal_derivative([x**25, x**3 + x**1, x**10, x**7]) == [x**3 + x**1, 0, x**7]
    """
    q = []
    for i, x in enumerate(p):
        if i == 0:
            continue
        if i % 2 == 0:
            e = 0
        else:
            e = x
        q.append(e)
    return collapse(q)

def gf128poly_modexp(a, n, m):
    """Compute a^n (mod m).
    >>> assert gf128poly_modexp([x**2, x**3, x**4], 0, [x**3, x**2, 1]) == [1]
    >>> assert gf128poly_modexp([x**2, x**3, x**4], 1, [x**3, x**2, 1, x**5]) == [x**2, x**3, x**4]
    >>> assert gf128poly_modexp([x**2, x**3, x**4], 5, [x**3, x**2, 1]) == [x**39 + x**38 + x**37 + x**36 + x**35 + x**33 + x**32 + x**30 + x**27 + x**26 + x**25 + x**24 + x**21 + x**20 + x**15 + x**10, x**38 + x**36 + x**35 + x**33 + x**32 + x**31 + x**26 + x**24 + x**23 + x**22 + x**21 + x**20 + x**14 + x**11]
    """
    def modmul(a, b, m):
        q, r = gf128poly_divmod(gf128poly_mul(a, b), m)
        return r

    if n == 0:
        return [1]
    rec = gf128poly_modexp(a, n//2, m)
    rec = modmul(rec, rec, m)
    if n % 2 == 1:
        rec = modmul(rec, a, m)
    return rec

def gf128poly_monic_gcd(a, b):
    """Compute g such that g | a, g | b, gf128poly_lead(g) = 1;
    for any h such that h | a and h | b, gf128poly_deg(g) >= gf128poly_deg(h).
    >>> assert gf128poly_monic_gcd([x**3, x**2, x**4], [x**3, x**2, x**4]) == [x**127 + x**6 + x**1 + 1, x**127 + x**126 + x**6 + x**5 + x**1, 1]
    >>> assert gf128poly_monic_gcd([1], [1]) == [1]
    >>> assert gf128poly_monic_gcd([x**5 + x**2], []) == [x**5 + x**2]
    >>> assert gf128poly_monic_gcd([x**3, 0, 0, x**6], [x**3, x**8 + x**4, x**9 + x**7, x**8]) == [x**127 + x**6 + x**1 + 1, 1]
    """
    assert gf128poly_deg(a) >= 0 or gf128poly_deg(b) >= 0
    if gf128poly_deg(a) < 0:
        return b
    if gf128poly_deg(b) < 0:
        return a
    q, r = gf128poly_divmod(b, a)
    ret = gf128poly_monic_gcd(r, a)
    return gf128poly_monic(ret)

def gf128poly_square_free_factorization(f):
    """Compute the square-free factorization of a monic polynomial f.
    Reference: https://en.wikipedia.org/wiki/Factorization_of_polynomials_over_finite_fields#Square-free_factorization
    >>> a = [x**1, x**2, 1]
    >>> b = [x**2, x**3+1, 1]
    >>> c = [x**3, x**4, x**5, 1]
    >>> asq = gf128poly_mul(a, a)
    >>> acb = gf128poly_mul(asq, a)
    >>> bsq = gf128poly_mul(b, b)
    >>> acbbsq = gf128poly_mul(acb, bsq)
    >>> p = gf128poly_mul(acbbsq, c)
    >>> assert gf128poly_square_free_factorization([1]) == [([1], 1)]
    >>> assert gf128poly_square_free_factorization(p) == [([x**3, x**4, x**5, 1], 1), ([x**1, x**2, 1], 3), ([x**2, x**3 + 1, 1], 2)]
    """
    R = []
    assert f[-1] == 1, 'monic'
    if f == [1]:
        return [([1], 1)]
    fprime = gf128poly_formal_derivative(f)
    c = gf128poly_monic_gcd(f, fprime)
    w, r = gf128poly_divmod(f, c)
    assert r == []
    i = 1
    while w != [1]:
        y = gf128poly_monic_gcd(w, c)
        fac, r = gf128poly_divmod(w, y)
        assert r == []
        R.append((fac, i))
        w = y
        c, r = gf128poly_divmod(c, y)
        assert r == []
        i = i + 1
    if c != [1]:
        c = gf128poly_sqrt(c)
        c = gf128poly_monic(c)
        Rrec = gf128poly_square_free_factorization(c)
        R += [(fac, i*2) for (fac, i) in Rrec]
    return [(fac, mult) for fac, mult in R if fac != [1]]

def gf128poly_distinct_degree_factorization(f):
    """Compute the distinct-degree factorization of a square-free monic polynomial f.
    Reference: https://en.wikipedia.org/wiki/Factorization_of_polynomials_over_finite_fields#Distinct-degree_factorization
    >>> a = [x**1, 1]
    >>> b = [x**2, 1]
    >>> ab = gf128poly_mul(a, b)
    >>> assert gf128poly_distinct_degree_factorization(ab) == {((x**3, x**2 + x**1, 1), 1)}
    """
    i = 1
    S = set()
    f_ = f.copy()
    while gf128poly_deg(f_) >= 2*i:
        qq = gf128poly_modexp([0, 1], 2**(128*i), f_)
        qq = gf128poly_add(qq, [0, 1])
        g = gf128poly_monic_gcd(f_, qq)
        if g != [1]:
            S.add((tuple(g), i))
            q, r = gf128poly_divmod(f_, g)
            assert r == []
            f_ = q
        i += 1
    if f_ != [1]:
        S.add((tuple(f_), gf128poly_deg(f_)))
    if len(S) == 0:
        return {(tuple(f), 1)}
    else:
        return S

def gf128poly_equal_degree_factorization(f, d):
    """Compute the equal-degree factorization of a square-free monic polynomial f whose factors are all of degree d.
    Uses the Cantor-Zassenhaus algorithm.
    Reference: https://en.wikipedia.org/wiki/Factorization_of_polynomials_over_finite_fields#Equal-degree_factorization
    >>> a = [x**1, 1]
    >>> b = [x**2, 1]
    >>> ab = gf128poly_mul(a, b)
    >>> assert gf128poly_equal_degree_factorization(ab, 1) == {(x**2, 1), (x**1, 1)}
    """
    def gf128_rand():
        return random.randint(0, 2**128-1)

    def gf128poly_rand(degree):
        return [gf128_rand() for _ in range(degree+1)]

    exp = (2**128-1)//3
    n = gf128poly_deg(f)
    r = n//d
    S = {tuple(f)}
    while len(S) < r:
        h = gf128poly_rand(n - 1)
        g = gf128poly_monic_gcd(h, f)
        if g == [1]:
            g = gf128poly_add(gf128poly_modexp(h, exp, f), [1])
        for u in list(S):
            u = list(u)
            if gf128poly_deg(u) == d:
                continue
            gugcd = gf128poly_monic_gcd(g, u)
            if gugcd != [1] and gugcd != u:
                S.remove(tuple(u))
                S.add(tuple(gugcd))
                qq, rr = gf128poly_divmod(u, gugcd)
                assert rr == []
                S.add(tuple(qq))
    return S

def gf128poly_factorize(f, degree=None):
    """Compute the factors of a polynomial f. Does not return multiplicity.
    If degree is specified, only returns factors of that degree.
    """
    factors = set()
    f = gf128poly_monic(f)
    fs = gf128poly_square_free_factorization(f)
    for p, _ in fs:
        qs = gf128poly_distinct_degree_factorization(p)
        for q, d in qs:
            if degree is not None and d != degree:
                continue
            rs = gf128poly_equal_degree_factorization(list(q), d)
            factors |= rs
    return factors

## AES-GCM

def pad(xs):
    return xs + b'\x00' * ((16 - len(xs)) % 16)

def build_blocks(aad, c):
    padded_aad = pad(aad)
    padded_c = pad(c)
    length = struct.pack('>QQ', len(aad)*8, len(c)*8)
    return padded_aad + padded_c + length

def gmac(h, s, aad, c):
    blocks = build_blocks(aad, c)
    g = 0
    for i in range(len(blocks)//16):
        block = bytes_to_gf128(blocks, i*16, (i + 1)*16)
        g = gf128_add(g, block)
        g = gf128_mul(g, h)
    t = gf128_add(g, s)
    return gf128_to_bytes(t)

def ecb_encrypt(key, xs):
    c = AES.new(key, AES.MODE_ECB)
    return c.encrypt(xs)

def xor(a, b):
    return bytes([(x^y) for (x, y) in zip(a, b)])

def gctr(k, nonce, x):
    y = b''
    ctr = (int.from_bytes(nonce, 'big') << 32) + 2
    i = 0
    while len(y) < len(x):
        y += xor(x[i*16:(i + 1)*16], ecb_encrypt(k, ctr.to_bytes(16, 'big')))
        ctr += 1
        i += 1
    return y

def authentication_key(k):
    return bytes_to_gf128(ecb_encrypt(k, b'\x00'*16))

def blind(k, nonce):
    return bytes_to_gf128(ecb_encrypt(k, nonce + b'\x00\x00\x00\x01'))

def gcm_encrypt(k, nonce, aad, m, mac_bytes=16):
    c = gctr(k, nonce, m)
    mac = gmac(authentication_key(k), blind(k, nonce), aad, c)
    return c, mac[:mac_bytes]

def gcm_decrypt(k, nonce, aad, c, mac, mac_bytes=16):
    m = gctr(k, nonce, c)
    expected_mac = gmac(authentication_key(k), blind(k, nonce), aad, c)[:mac_bytes]
    assert hmac.compare_digest(mac, expected_mac)
    return m

## Forbidden Attack

def compute_forbidden_polynomial(aad1, aad2, c1, c2, mac1, mac2):
    bs1 = build_blocks(aad1, c1)
    bs2 = build_blocks(aad2, c2)
    if len(bs1) < len(bs2):
        bs1 = b'\x00'*(len(bs2)-len(bs1)) + bs1
    else:
        bs2 = b'\x00'*(len(bs1)-len(bs2)) + bs2
    assert len(bs1) == len(bs2)
    f = []
    N = len(bs1)//16
    for i in range(N):
        b1 = bytes_to_gf128(bs1[i*16:(i+1)*16])
        b2 = bytes_to_gf128(bs2[i*16:(i+1)*16])
        f.append(gf128_add(b1, b2))
    f.append(gf128_add(bytes_to_gf128(mac1), bytes_to_gf128(mac2)))
    return collapse(list(reversed(f)))

def nonce_reuse_recover_secrets(nonce, aad1, aad2, c1, c2, mac1, mac2):
    f = compute_forbidden_polynomial(aad1, aad2, c1, c2, mac1, mac2)
    factors = gf128poly_factorize(f, degree=1)
    secrets = []
    for factor in factors:
        if gf128poly_deg(factor) == 1:
            h = factor[0]
            zero_tag = gmac(h, 0, aad1, c1)
            s = gf128_add(bytes_to_gf128(mac1), bytes_to_gf128(zero_tag))
            secrets.append((h, s))
    return secrets

## MAC Truncation Attack

def gen_blocks(n, js):
    blocks = b'\x00'*16*n
    for j in js:
        whichbyte = j // 8
        whichbit = j % 8
        newbyte = blocks[whichbyte] ^ (1 << (7-whichbit))
        blocks = blocks[:whichbyte] + bytes([newbyte]) + blocks[whichbyte+1:]
    return blocks

squarer = np.array(Ms())
use_numpy = True
if use_numpy:
    mcsqlookup = np.load(open('square-basis.np', 'rb'))
else:
    mcsqlookup = xr.open_dataarray('square-basis.nc')

def mc_squared(c, j):
    return sum(mcsqlookup[i, j] if use_numpy else mcsqlookup[i, j].to_numpy() for i in range(128) if 1 == (c >> i) & 1) % 2

def gen_ad(blocks):
    matret = np.zeros((128, 128))
    for i in range(len(blocks)//16):
        block = blocks[(i*16):(i+1)*16]
        if block == b'\x00'*16:
            continue
        j = i + 1 # first is taken up by length block
        mat = None
        d = bytes_to_gf128(block)
        try:
            mat = mc_squared(d, j)
        except IndexError:
            matsq = np.linalg.matrix_power(squarer, j) % 2
            mat = Mc(d) @ matsq
        matret += mat
    return matret % 2 # Because the elements of Ad are in GF2 we can mod 2

def gen_t(n, macbytes, X=None, minrows=8):
    T = []
    for j in range(n*128):
        blocks = gen_blocks(n, [j])
        Ad = gen_ad(blocks)

        if X is not None:
            rows = min((n*128)//(X.shape[1]) - 1, macbytes*8-minrows)
        else:
            rows = min((n*128)//(Ad.shape[1]) - 1, macbytes*8-minrows)

        if X is not None:
            Ad = (Ad[:rows] @ X) % 2
        z = np.concatenate(Ad[:rows])
        T.append(z)
    return (np.array(T)).transpose().astype('int')

def gen_flips(b):
    return np.nonzero(b)[0]

def compute_n(ct):
    num_blocks = len(ct)//16
    ns = [1] # 1-indexed
    while (ns[-1]*2)<=(num_blocks):
        ns.append(2*ns[-1])
    ns = [n-1 for n in ns] # 0-indexed into c
    n = len(ns) - 1
    return n

def chunk(xs, n=16):
    return [xs[i*n:(i+1)*n] for i in range(len(xs)//16)]

def find_b(n, basis, ct, mac, nonce, aad, oracle):
    orig_base = bytearray(ct).copy()
    base = bytearray(ct)
    idx = 0
    while True:
        choice = random.sample(basis, random.randint(1, 10))
        b = sum(choice) % 2
        flips = gen_flips(b)
        blocks = gen_blocks(n, flips)
        for i, block in enumerate(chunk(blocks)):
            j = (len(base)//16)-(2**(i+1)-1)
            base[j*16:(j+1)*16] = xor(base[j*16:(j+1)*16], block)
        try:
            oracle(base[len(aad):], base[:len(aad)], mac, nonce)
            return b
        except (cryptography.exceptions.InvalidTag, ValueError):
            base = orig_base.copy()
        idx += 1

def mac_truncation_recover_secrets(ct, mac, nonce, mac_bytes, aad, oracle, compute_T_once=False):
    orig_ct = ct
    ct = aad + ct
    n = compute_n(ct)
    assert n > (mac_bytes*8//2)
    assert len(ct) % 16 == 0
    assert len(aad) % 16 == 0
    X = None
    K = None
    basisKerK = None
    if compute_T_once:
        T = gen_t(n, mac_bytes, X, minrows=7)
        _, _, basisKerT = kernel(T, rref_mod_2)
    while K is None or (basisKerK is None or len(basisKerK) > 1):
        if not compute_T_once:
            T = gen_t(n, mac_bytes, X, minrows=7)
            _, _, basisKerT = kernel(T, rref_mod_2)
            assert len(basisKerT[0]) == n*128

        b = find_b(n, basisKerT, ct, mac, nonce, aad, oracle)
        flips = gen_flips(b)
        blocks = gen_blocks(n, flips)
        Ad = gen_ad(blocks)

        if X is not None:
            AdRelevant = ((Ad @ X) % 2)[:mac_bytes*8]
        else:
            AdRelevant = Ad[:mac_bytes*8]
        incrK = Ad[:mac_bytes*8][np.any(AdRelevant, axis=1)]
        if K is None:
            K = incrK
        else:
            K = np.concatenate([K, incrK])
        _, _, basisKerK = kernel(K, rref_mod_2)
        if not compute_T_once:
            X = np.array(basisKerK).transpose()
    _, _, kerK = kernel(K, rref_mod_2)
    assert len(kerK) == 1, len(kerK)
    h = kerK[0]

    zero_tag = gf128_to_vec(bytes_to_gf128(gmac(vec_to_gf128(h), 0, aad, orig_ct)))[:mac_bytes*8]
    gf128_mac = 0
    i = 0
    for b in mac:
        for j in range(8):
            if b & (1 << (7-j)):
                gf128_mac += (1<<i)
            i += 1
    def small_gf128_to_vec(x):
        return [int(n) for n in bin(x)[2:].zfill(mac_bytes*8)[::-1]]
    mac_vec = small_gf128_to_vec(gf128_mac)
    s = (np.array(zero_tag) - np.array(mac_vec)) % 2
    return vec_to_gf128(h), vec_to_gf128(s)

# Key Commitment Attack

def gmac_blind(k, nonce):
    return bytes_to_gf128(ecb_encrypt(k, nonce + b'\x00\x00\x00\x01'))
def encode_lengths(ad_length, ct_length):
    return struct.pack('>QQ', ad_length*8, ct_length*8)
def collide(k1, k2, nonce, c):
    h1 = gmac_key(k1)
    h2 = gmac_key(k2)
    p1 = gmac_blind(k1, nonce)
    p2 = gmac_blind(k2, nonce)
    assert len(c) % 16 == 0
    mlen = len(c)//16+1
    lens = bytes_to_gf128(encode_lengths(0, len(c) + 16))
    acc = gf128_mul(lens, gf128_add(h1, h2))
    acc = gf128_add(acc, gf128_add(p1, p2))
    for i in range(1, mlen):
        hi = gf128_add(gf128_exp(h1, mlen+2-i), gf128_exp(h2, mlen+2-i))
        acc = gf128_add(acc, gf128_mul(bytes_to_gf128(c[(i-1)*16:((i-1)+1)*16]), hi))
    inv = gf128_inv(gf128_add(gf128_mul(h1, h1), gf128_mul(h2, h2)))
    c_append = gf128_mul(acc, inv)
    c_ = c + gf128_to_bytes(c_append)
    mac = gmac(h1, p1, b'', c_)
    return (c_, mac)

def collide_penultimate(k1, k2, nonce, c):
    h1 = authentication_key(k1)
    h2 = authentication_key(k2)
    p1 = gmac_blind(k1, nonce)
    p2 = gmac_blind(k2, nonce)
    assert len(c) % 16 == 0
    mlen = len(c)//16
    lens = bytes_to_gf128(encode_lengths(0, len(c)))
    acc = gf128_mul(lens, gf128_add(h1, h2))
    acc = gf128_add(acc, gf128_add(p1, p2))
    n=4
    h1Running = gf128_exp(h1, 4)
    h2Running = gf128_exp(h2, 4)
    for i in reversed(range(0, mlen-2)):
        # print(mlen+1-(i))
        # i = mlen-2-1-i
        hi = gf128_add(h1Running, h2Running)
        h1Running = gf128_mul(h1Running, h1)
        h2Running = gf128_mul(h2Running, h2)
        n+=1
        # hi = gf128_add(gf128_exp(h1, mlen+1-i), gf128_exp(h2, mlen+1-i))
        #print('block', i, pt[i*16:(i+1)*16], 'exp', mlen+1-i)
        acc = gf128_add(acc, gf128_mul(bytes_to_gf128(c[i*16:(i+1)*16]), hi))
    # for i in range(0, mlen-2):
    #     print(i,mlen+1-i)
    #     hi = gf128_add(gf128_exp(h1, mlen+1-i), gf128_exp(h2, mlen+1-i))
    #     acc = gf128_add(acc, gf128_mul(bytes_to_gf128(c[i*16:(i+1)*16]), hi))
    hi = gf128_add(gf128_exp(h1, 2), gf128_exp(h2, 2))
    i = mlen-1
    #print('block', i, pt[i*16:(i+1)*16], 'exp', 2)
    acc = gf128_add(acc, gf128_mul(bytes_to_gf128(c[i*16:(i+1)*16]), hi))
    inv = gf128_inv(gf128_add(gf128_exp(h1, 3), gf128_exp(h2, 3)))
    c_append = gf128_mul(acc, inv)
    c_ = c[:-32] + gf128_to_bytes(c_append) + c[-16:]
    mac = gmac(h1, p1, b'', c_)
    return (c_, mac)

def gctr_oneblock(k, pt, nonce):
    enckeyval = nonce + b'\x00\x00\x00\x02'
    stream = ecb_encrypt(k, enckeyval)
    return xor(pt, stream)

def key_search(nonce, init_bytes1, init_bytes2):
    seen1 = dict()
    seen2 = dict()
    while True:
        k1 = secrets.token_bytes(16)
        k2 = secrets.token_bytes(16)
        ct1 = gctr_oneblock(k1, init_bytes1, nonce)
        ct2 = gctr_oneblock(k2, init_bytes2, nonce)
        seen1[ct1] = k1
        seen2[ct2] = k2
        if ct1 in seen2:
            return k1, seen2[ct1]
        if ct2 in seen1:
            return seen1[ct2], k2

def att_merge_jpg_bmp(jpg, bmp, aad):
    # Precomputed with key_search; works for any files
    k1 = unhexlify('8007941455b5af579bb12fff92ef31a3')
    k2 = unhexlify('14ef746e8b1792e52b1d22ef124fae97')
    nonce = b'JORGELBORGES'

    total_len = 6 + (0xffff) + len(jpg) + 0xff # get some extra
    jpgstream, _ = gcm_encrypt(k1, nonce, aad, b'\x00'*total_len)
    bmpstream, _ = gcm_encrypt(k2, nonce, aad, b'\x00'*total_len)

    # 6 bytes
    r = xor(jpgstream, b'\xff\xd8\xff\xfe\xff\xff')

    # len(bmp) bytes
    bmpenc = xor(bmp[6:], bmpstream[6:6+len(bmp)])
    r += bmpenc

    comlen = 0xffff

    # finish comment with padding
    r += b'\x00'*(comlen - len(bmpenc))

    # jpg
    r += xor(jpg[2:-2], jpgstream[6+comlen:])

    # comment; include penultimate block to be overwritten; therefore must be at least 3 blocks long
    # also serves to block-align to 14 bytes so the final ciphertext will be complete blocks

    endcomlen = (28 - (len(r) % 16)) + 16 + 14

    tail = b'\xff\xfe' + struct.pack('>H', endcomlen) + b'\x00'*endcomlen + b'\xff\xd9'
    tailx = xor(tail, jpgstream[6+comlen+len(jpg)-4:])
    r += tailx
    assert len(r) % 16 == 0

    cfin, macfin = collide_penultimate(k1, k2, nonce, r)

    return cfin, macfin

# Demos

def forbidden_attack_demo():
    k = b"tlonorbistertius"
    nonce = b"jorgelborges"
    m1 = b"The universe (which others call the Library)"
    aad1 = b"The Anatomy of Melancholy"
    m2 = b"From any of the hexagons one can see, interminably"
    aad2 = b"Letizia Alvarez de Toledo"

    c1, mac1 = gcm_encrypt(k, nonce, aad1, m1)
    c2, mac2 = gcm_encrypt(k, nonce, aad2, m2)

    # Recover the authentication key and blind from public information
    possible_secrets = nonce_reuse_recover_secrets(nonce, aad1, aad2, c1, c2, mac1, mac2)

    # Forge the ciphertext
    m_forged = b"As was natural, this inordinate hope"
    assert len(m_forged) <= len(m1)
    c_forged = xor(c1, xor(m1, m_forged))
    aad_forged = b"You who read me, are You sure of understanding my language?"

    # Check possible candidates for authentication key
    succeeded = False
    for h, s in possible_secrets:
        mac_forged = gmac(h, s, aad_forged, c_forged)
        try:
            assert gcm_decrypt(k, nonce, aad_forged, c_forged, mac_forged) == m_forged
            succeeded = True
            print(c_forged.hex(), mac_forged.hex())
        except AssertionError:
            pass
    assert succeeded

def mac_truncation_demo():
    # Doesn't work with non-block size multiples.
    # Need to modify to consider padding, but we can't mess with the bits in the padding,
    # nor can we extend ad/ct unless we also change length block.
    k = b'tlonorbistertius'
    aad = b''
    mac_bytes=1
    pt = b'celerypatchworks'*(2**5)
    nonce = b'jorgelborges'
    ct, mac = gcm_encrypt(k, nonce, aad, pt, mac_bytes=mac_bytes)
    def oracle(base, aad, mac, nonce):
        decryptor = Cipher(
            algorithms.AES(k),
            modes.GCM(nonce, mac, min_tag_length=mac_bytes),
        ).decryptor()
        decryptor.authenticate_additional_data(aad)
        decryptor.update(base) + decryptor.finalize()

    h, s = mac_truncation_recover_secrets(ct, mac, nonce, mac_bytes, aad, oracle, compute_T_once=mac_bytes==1)
    assert h == authentication_key(k)

if __name__ == "__main__":
    pass
    # mac_truncation_demo()
    jpg = open('static/axolotl.jpg', 'rb').read()
    bmp = open('static/kitten.bmp', 'rb').read()
    c, mac = att_merge_jpg_bmp(jpg, bmp, aad=b"")
    print(mac.hex())
    f = open('c.txt', 'wb')
    f.write(c)
    f.write(mac)
    f.close()