1998-12-13 - passwords and hashes

Header Data

From: Alexander Kjeldaas <astor@guardian.no>
To: cypherpunks@toad.com
Message Hash: 2beed07e78e9406704dd21ccf83542fd4b5cd5bde48a1a375f412302f61d0f08
Message ID: <19981213234401.A25332@lucifer.guardian.no>
Reply To: N/A
UTC Datetime: 1998-12-13 22:44:11 UTC
Raw Date: Sun, 13 Dec 1998 14:44:11 -0800 (PST)

Raw message

From: Alexander Kjeldaas <astor@guardian.no>
Date: Sun, 13 Dec 1998 14:44:11 -0800 (PST)
To: cypherpunks@toad.com
Subject: passwords and hashes
Message-ID: <19981213234401.A25332@lucifer.guardian.no>
MIME-Version: 1.0
Content-Type: text/plain
Content-Length: 1215
Lines: 32


I'm looking into how passwords should be transferred into keys for the
loopback filesystem in Linux.  Currently, what happens is that you
take the SHA or RIPEMD-160 hash of the password string and use that as
a key.  If the cipher only uses a 128-bit key, the last 32 bits of the
hash is unused.  I have some questions about this scheme:

o In the case of a 128-bit cipher, do I lose any information by not
  using the last 32 bits of information?  Should the last 32 bits be
  xored with the first 128 in order to not lose any info?

o Is there any advantage to _not_ using a 256-bit hash function? (i.e
  - use a 128-bit hash for 128-bit ciphers).  Currently there are lots
  of AES ciphers that don't get fed a 256-bit key because we only have
  a 160-bit hash.

o Is there a good 256-bit hash function?  I don't know of any other
  than snefru.  In this application, speed doesn't matter much.
  Should I use a 256-bit cipher instead maybe?

o Is it safe to always take the hash of the password, or is it better to
  use the password directly as the key if it is less than 16
  characters for 128-bit cipher?

astor

-- 
 Alexander Kjeldaas, Guardian Networks AS, Trondheim, Norway
 http://www.guardian.no/





Thread