Symmetric-key primitives, especially block ciphers, constitute a major building block in many
cryptographic applications. In contrast with asymmetric-key primitives whose security is
often provably reduced to some hard mathematical problems, the security of symmetrickey
primitives is often based on the empirical non-existence of successful attacks. Hence,
cryptanalytic results are fundamental for evaluating their security. However, by only evaluating
the security against well-known attacks, there is no guaranty against yet-to-be-discovered
attacks. It would be preferable to seek for provable security. This is a goal of Decorrelation
Theory, put forward by Vaudenay. This theory provides useful tools to design provable secure
block ciphers against a big set of statistical attacks.
The contribution of this dissertation is two-fold: we study provable security of block ciphers
with a focus on Decorrelation Theory, and we evaluate the security of two block ciphers against
statistical attacks.
Firstly, we focus on Decorrelation Theory. We study iterated distinguishers which comprise
iterating adversaries who can make d plaintext queries, keep one bit information for each
query, and they try to distinguish a random cipher C from the perfect cipher C¤ based on
the collected bits. We revisit a previous result about non-adaptive iterated attacks of order
d which was showing some sufficient conditions for a cipher to resist to these attacks. We
show that these conditions are somehow necessary, which was an open question since 1999.
Furthermore, we extend the applicability of these results to a bigger set of iterated attacks,
i.e., adaptive iterated attacks. In addition to these results, we concentrate on proving the
security of block ciphers against both boomerang and differential-linear attacks using the
tools provided by Decorrelation Theory.
Secondly,we provide cryptanalysis of two block ciphers; namely, Triple-DES and the lightweight
block cipher MIBS. For Triple-DES, based on finding fixed points, a related-key attack against
two-key and three-key triple encryptions is proposed. For the attack on two-key triple encryption,
it has exactly the same performance as the meet-in-the-middle attack on double
encryption and comparable to theMerkle-Hellman attack (except related keys). In addition,
the attack on three-key triple encryption has a higher complexity than the Kelsey-Schneier-
Wagner attack; nevertheless, it has the advantage that it is feasible with known plaintexts. Our
final contribution is a cryptanalyzing ofMIBS by finding out its vulnerability against linear
cryptanalysis together with using several linear approximations at the same time. We show
that 19 rounds out of 32 rounds of MIBS are theoretically broken by the multidimensional
linear attack.
Keywords: Symmetric cryptography, block ciphers, Decorrelation Theory, statistical attacks,
iterated attacks, linear attacks, boomerang attacks, differential-linear attacks, related-key
attacks, multidimensional linear attacks, MIBS, Triple-DES |