Take the 2-minute tour ×
Cryptography Stack Exchange is a question and answer site for software developers, mathematicians and others interested in cryptography. It's 100% free, no registration required.

I'm looking for information on the MJH double-block-length hash function, but the best free source I could find is the diagram on page 18 of Efficient hashing using the AES Instruction Set (submitted to the ECRYPT II Hash Workshop 2011).

Is source code available for this function? What are the standard symbols used (M, V, Z) when describing hash functions?

share|improve this question
    
I'll elaborate this to an answer later (if not someone else does it first), but here about the meanings of the symbols used in the diagram (as I understand them): n is the block size of the block cipher, k the key size. M and V are inputs to the compression function (M typically being one message block, V the state of the hash function), Z is the output (same size as V, and the new state of the hash function). –  Paŭlo Ebermann Sep 11 '11 at 1:01
add comment

1 Answer

up vote 2 down vote accepted

I had a look at the paper introducing the MJH construction (MJH: A Faster Alternative to MDC-2).

It actually presents a different diagram than the one in the paper Efficient hashing using the AES Instruction Set your referenced – there $V_2$ and $M$ are swapped. I'll describe the original one here, and the variation below.

The core of MJH is a compression function, using a similar construction like the JH hash function (one of the SHA-3 finalists), hence the name.

The JH construction

Given a (non-compressing) function $F : \{ 0,1\}^{2n} \to \{0,1\}^{2n}$, we define the compression function $JH[F] : \{0,1\}^{3n} \to \{0,1\}^{2n}$ as $JH[F](V_1, V_2, M) := (Z_1, Z_2)$, with $(X_1, X_2) = (V_1, V_2 \oplus M)$, $(Y_1, Y_2) = F(X_1, X_2)$ and $(Z_1, Z_2) := (Y_1 \oplus M, Y_2)$.

diagram

In the JH hash function, F is a specially-made (fixed) permutation.

MJH's $F[\sigma, \theta]$

Given a $n$-bit-blockcipher $E$ with keysize $k = n$, $\sigma : \{0,1\}^{n} \to \{0,1\}^n$ an involution (i.e. $\sigma \circ \sigma = \mathrm{id}$) without fixed point (i.e. $\sigma(X) \neq X$) – an example would be an XOR with a non-zero constant –, and $\theta \in \mathbb{F}_{2^n} \setminus\{0,1\}$ a constant (so multiplication with $\theta$ is another non-trivial permutation).

We define $F[\sigma, \theta] : \{0,1\}^{2n} \to \{0,1\}^{2n}$ as $F(X, K) = (L, R)$, with $L = E_K(X) \oplus X$ and $R := \theta(E_K(\sigma(X)) \oplus \sigma(X) ) \oplus X$.

enter image description here

(The $E_K(X) \oplus X$ part is essentially the Davies-Meyer construction to make it one-way, even if $K$ is known. It is actually used twice here.)

The final Hash function

Combining these ideas, we get our compression function $\tilde F[\sigma, \theta] := JH[F[\sigma, \theta]] : \{0,1\}^{3n} \to \{0,1\}^{2n}$.

enter image description here

We then apply the known Merkle–Damgård caining construction on this compression function, receiving our final hash function $H[\sigma, \theta] = MD[\tilde F[\sigma, \theta]]$.

enter image description here
(This image is from Wikimedia Commons, the others are made by me.)

In practice we now also have to choose some specific $\sigma$ and $\theta$, a specific block cipher, as well as an initialization vector and padding for MD - the security proof still applies for all of them (if the block cipher is good).

Longer key variant

The paper also describes a variant of the compression function to be used when the key length $k$ is greater than the block size $n$. Then we use message blocks of size $k$, and split these blocks into two parts $z, z'$. $z$ (of size $n$)is used as before to be XOR'ed into the left half of the state before encryption and right half after, while $z'$ (of size $k-n$) is appended to the right half of the state to form the key for the block cipher.

By using a larger (hash) block size, this longer variant can be more efficient than the original one.

The variant depicted in Efficient Hashing

For $n = k$, this is the same compression function as the original one, but it is used in a different way in the MD-construction: Here the message blocks are passed in the key position, instead of being XOR'ed into the parts of the state before and after encryption. (This XOR'ing instead uses one half of the original state.)

This gives a more obvious generalization into the $k \geq n$ case, as we don't have to split the message block and compose the key, but simply can pass the longer key to the block cipher.

I didn't check if the proofs given in the MJH paper also apply to this variant, though, and I don't know which version the authors of Efficient Hashing actually measured.

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.