Re: AES as hash function and PRNG

Hello,

Sam Trenholme wrote:

> * Make the 16 secret bytes the key for the AES encryption
> * If you have less than 16 bytes to hash, just pad the key with ASCII
> nulls
> * Choose any input as the plaintext to run through AES. Something
> with a random pattern (binary representations of pi are popular)
> will speed up AES' diffusion a wee bit, but all NULLS is fine.
> * The resulting ciphertext is your hash
> * Make sure you only hash up to 16 bytes this way. If you want to
> hash more bytes, use a 128-bit variant of, say, Bram Cohen's
> AES-hash.
>
> This will generate 16 random bytes; as long as the key is unknown, we
> don't know how we got the bytes.

Usiung the secret bytes as key sounds good. I think I'll do it this way.
Thank you.

> Now, if you only hash six bytes, and someone knows the IV (the
> plaintext), then it is possible for an attacker to guess the six bytes
> being hashed via brute force. If this is not desirable, make the IV
> secret; this results in what is known as a MAC (message authentication
> code).

Well, I know. But a secret IV is not possible here.

> I will assume you can somehow get 16 measly truly random bytes. If
> not, the system is insecure, and people will break the bank when
> playing Keno with your implementation. I don't know enough about your
> hardware to know how to squeeze 16 random bytes out of it, so I'll
> gloss over this step. (Timing how fast people type is one way to get
> good random numbers)

In the worst case I'll get only one random byte between generating two
session keys. That's the reason I wanted to use a secret (here ist is
possible) random IV. Would this be secure? (I think it would.)

> Once you have these 16 random bytes, make them the key for the AES
> cipher, then run the cipher in either CTR (counter) or OFB mode to
> generate secure pseudo-random bytes.
>
> http://en.wikipedia.org/wiki/Block_cipher_modes_of_operation

IMHO OFB mode would be the best choice for my purpose. I suppose,
I don't need a plaintext here. Am I right?

> And, oh, convince your boss to buy you a book. "Applied Crpytography"
> by Bruce Schneier is an excellent, albeit slightly dated, book.

Done. I'll get it tomorrow.

Thanks again
Mike
Received on Thu Sep 29 21:43:52 2005