NetNews Usenet Archive 1992 #26/NN_1992_26.iso/spool/sci/crypt/4526

Text File | 1992-11-09 | 1.8 KB | 38 lines

Newsgroups: sci.crypt
Path: sparky!uunet!walter!qualcom.qualcomm.com!servo.qualcomm.com!karn
From: [e-mail] (Phil Karn)
Subject: Re: MDC cipher code (long)
Message-ID: <[e-mail]>
Sender: [e-mail]
Nntp-Posting-Host: servo.qualcomm.com
Organization: Qualcomm, Inc
References: <[e-mail]> <[e-mail]>
Date: Sun, 1 Nov 1992 22:38:36 GMT
Lines: 25

In article <[e-mail]> [e-mail] (Eric Backus) writes:
>I was under the impression that a secure cipher could be obtained from
>a message digest algorithm, by using several passes of a message
>digest algorithm and using the user's key as part of the input to the
>digest algorithm.  I don't think the message digest algorithm needs to
>be altered in any way for this.  I think Phil Karn proposed something
>like this awhile back.  (Phil, did you make this work?)

I proposed it, but I never actually followed through with the
definitive implementation. But you're right, all you need is three
Feistel rounds with the nonlinear transformation 'f' in each being a
strong hash function like MD-5.  It is important that each round be
separately keyed, although the three subkeys can be derived from a
single input key (e.g., by appending the input key to a
round-dependent constant and hashing).

MD-5 does have a lot of baggage from its role as a hash function that
can be removed when it is to be used as the nonlinear primitive in a
cipher. E.g., there's no need to pad out each input block with the
input length. All you need is the internal "transform" function, the
routine that crunches 64 bytes to 16 bytes on each iteration.  You
design your cipher to work with these block sizes, which admittedly
are inconveniently large for some applications.

Phil