Bilinear Cryptanalysis of Multivariate Schemes Pierre-Alain FOUQUE - - PowerPoint PPT Presentation

bilinear cryptanalysis of multivariate schemes
SMART_READER_LITE
LIVE PREVIEW

Bilinear Cryptanalysis of Multivariate Schemes Pierre-Alain FOUQUE - - PowerPoint PPT Presentation

Bilinear Cryptanalysis of Multivariate Schemes Pierre-Alain FOUQUE Crypto Team cole normale suprieure Joint work with Dubois, Stern, Shamir, Macario-Rat Summary Matsumoto-Imai (MI) cryptosystem Cryptanalysis of MI by Patarin


slide-1
SLIDE 1

Bilinear Cryptanalysis of Multivariate Schemes

Pierre-Alain FOUQUE Crypto Team École normale supérieure Joint work with Dubois, Stern, Shamir, Macario-Rat

slide-2
SLIDE 2

Summary

  • Matsumoto-Imai (MI) cryptosystem
  • Cryptanalysis of MI by Patarin
  • The SFLASH signature scheme
  • First attack against SFLASH (E07)
  • Second attack against NESSIE parameters
  • f SFLASH (C07)
  • Key Recovery on SFLASH (E08)
slide-3
SLIDE 3

Alternative to RSA

  • f(x)=xe mod N with N=pq
  • F(X)=X in the finite field GF(qn)
  • F permutation iff gcd(q+1,qn-1)
  • GF(qn) vector space over GF(q) ≅ GF(q)n
  • XX linear map on GF(q)n and F(X)=X!X

is quadratic over GF(q)n

  • F is described as n quadratic polynomials in

(x1,x2,...xn) where X=(x1,...,xn)

q+1 q q

slide-4
SLIDE 4

MI Cryptosystem E88

  • F system of n quadratic polynomials (f1,...,fn)

in X=(x1,x2,...,xn)∈(GF(q))n easily invertible

  • Linear masking: S and T two linear bijective

maps over GF(q)n

  • Public Key: P=T
  • -1oFooS where is an

isomorphism between GF(q)n and GF(qn)

  • P is also a set of n quadratic polynomials
  • Secret Key: S and T
slide-5
SLIDE 5

MI Encryption Scheme

  • For example: q=2 and n=128
  • Encryption: computing P(M) where M=(m1,...mn)

is the plaintext in bits is very simple and efficient even on low-cost smartcard

  • Decryption:
  • invert T
  • invert F(X): F(X)=Xh with h=(q+1)-1 mod qn-1

(similar to RSA decryption)

  • invert S
slide-6
SLIDE 6

Security of Multivariate Schemes

  • Solving a system of n polynomials over GF(2) is

a NP-hard problem

  • No known polynomial-time quantum algorithm

contrary to DL or RSA based systems

  • S and T hide an easy instance of this generic

problem

  • Generic Tool: Gröbner basis based algorithms
  • exponential complexity in Time/Space
slide-7
SLIDE 7

Cryptanalysis of Patarin C95

  • B=F(A)=A so B = A
  • AB - A B = 0
  • n bilinear relations over GF(q) between the

coordinates of A and B

  • Y B A X
  • n bilinear relations between X (plaintext)

and Y (ciphertext) bk(x1,...,xn,y1,...yn)=0 q+1 q-1 q2-1 q q2 T F S

slide-8
SLIDE 8

Cryptanalysis of Patarin

  • bk(x1,...xn,y1,...yn)=i,j i,j,k xiyj + i i,k xi + j j,k yj + k
  • Find n bilinear relations using (n+1)2 pairs

(X=plaintext, Y=ciphertext)

  • linear system in the unknowns , , ,

which has a n-dimensional kernel

  • To decrypt any ciphertext, it is sufficient to

solve a linear system in (x1,...,xn) since the , , , and y are known

slide-9
SLIDE 9

SFLASH Signature

  • Proposed by Patarin, Goubin, and Courtois (2000)
  • Idea: Removing some equations to the public key

(Shamir 93) (erase rows of T=Tr)

  • Also called the C*- scheme
  • The removed equations are kept in the secret key
  • to sign a message, try to invert the system with

random values for the missing coordinates

  • Usage for signature only if too many equations are

removed

slide-10
SLIDE 10

SFLASH

  • Claimed security: qr where r = number of

missing equations

  • NESSIE Parameters:

q n

  • r

log (Sec.) v2 128 37 11 11 77 v3 128 67 33 11 77

Primes n>3r

slide-11
SLIDE 11

First attack against SFLASH

(Eurocrypt 07)

slide-12
SLIDE 12

Main idea of the attacks

  • Reconstruct the missing polynomials
  • FoS system of n quadratic polynomials
  • T’s action: linear integer combinaisons
  • ver GF(q) of the FoS polynomials
  • Goal: Find other linear combinaisons of

FoS independent of those of the truncated public key ...

  • Then, apply Patarin’s attack
slide-13
SLIDE 13

Final stage of the attack

  • Goal: Find linear combinations of FoS

independent of those of the truncated PK, Pr

  • Find N such that N=S-1MS where M is a

matrix of the multiplication by in GF(qn)

  • ProN=(TroFoS)o(S-1oMoS)=ToFoMoS
  • FoM=MF()F since F(X)=F()F(X)
  • ProN=(TroMF())oFoS gives (n-r) linear

combinations of FoS, some independent of those of Pr provided ∉GF(q)

slide-14
SLIDE 14

Differential Cryptanalysis

  • DF(X,Y)=F(X+Y)-F(X)-F(Y)+F(0)
  • Since F is quadratic, DF is bilinear !
  • DF(X,Y)=X Y+XY symmetric bilinear map
  • DP(X,Y)=T(DF(S(X),S(Y)))
  • Each coordinate of the PK is a bilinear map
  • Multiplicative Property:
  • DF(X,Y)+DF(X,Y)=(+ )DF(X,Y)

q q q

slide-15
SLIDE 15

Studying bilinear maps

  • Mathematicians usually study linear maps called

skew-symmetric maps for bilinear maps objects

  • L is skew-symmetric map for a bilinear map B(x,y)

iff B(Lx,y)=B(x,Ly)

  • Skey-symmetric maps for DF are exactly some

multiplications M by since F is multiplicative and if gcd(n,)>1, there exists s.t. + = 0

  • For DP

, they are conjuguates of M, N=S-1MS q

slide-16
SLIDE 16

Second Attack on NESSIE Parameters

(Crypto 07)

slide-17
SLIDE 17

NESSIE Parameters

  • Since n is prime, there is no element ∉GF(q)

s.t. + = 0

  • We have to find other (or any) multiplications
  • We still have the equation:
  • DF(X,Y)+DF(X,Y)=( + )DF(X,Y)
  • DP(N(X),

Y) + DP(X, N(Y))

  • = (T(+ )T-1)DP(X,Y)
  • = LDP(X,Y)

q q q

slide-18
SLIDE 18

SFLASH Attack

  • Let B be the vector space of the symmetric

bilinear forms of dimension n(n-1)/2

  • Let the n-dimensional vector space

Vect (DP1,...,DPn) spanned by the n forms of PK

  • If N is a multiplication, for each coordinate i
  • DPi(N(X),Y)+DPi(X,N(Y)) ∈

Vect(DP1,...,DPn)

  • n(n-1)/2 linear relations on the N’s unknowns
slide-19
SLIDE 19

Problem of a truncated public key

  • We don’t know

Vect(DP1,...,DPn) since we have

  • nly (n-r) polynomials in PK,
  • ... but we know the subspace

Vect(DP1,...,DPn-r)

  • One bilinear form DPi(N(X),Y)+DPi(X,N(Y))

will be in Vect(DP1,...,DPn-r) with proba. 1/qr

  • No characterization property holds: N are not

always conjuguates of multiplications and not all

  • but 3 equa. characterize somes conjuguates and

if n-3r>1, and we find only conjuguates of mult.

slide-20
SLIDE 20

Recovering the secret keys (E08)

  • N is one conjuguate of a multiplication

M:XX by the secret matrix S: N=S-1MS

  • If M is known, we can linearize the system in

SN=MS and look at S

  • The minimal polynomial of N has as root

and all the conjuguates of

  • Any conjuguate will give a possible M and an

equivalent secret key

slide-21
SLIDE 21

Conclusion and open problems

  • Bilinear algebra appears to be well suited to

cryptanalyze multivariate cryptosystems

  • Breaking HFE where the monomial X is

remplaced by a quadratic polynomial of small degree i,j<D pi,j X ?

q+1 qi+qj