agora inbox for postgres@postgres.berkeley.edu  
help / color / mirror / Atom feed
From: Kai Petzke <wpp@marie.physik.tu-berlin.de>
Subject: (NET) authentification over a network
Date: 12 May 94 07:35:08 GMT
Message-ID: <wpp.768728108@marie> (raw)


Hi,

I am looking for a good scheme to do login authentification
over a network.  I want to use it with the database Postgres,
but if I have to write new code for it, I will put it into
an library, which should be of use with other STREAM type
connections as well.

In general:

A server S is offering some service to clients C.  S wants to be
sure, that C is allowed to use that service.  Therefore, they
have agreed on a secret password some time ago.  Now C connects
to S.  How do we check, that the password of C is correct?

The two concepts for this authentifictation, which I am aware of
right now, are:

- Shadow password.  The password is used as a key to encrypt
  the value 0.  The encrypted value is stored in a hidden file
  or database available to S.  To check the password, C has
  to send the password, and S repeats the encryption process
  with that password, and compares the result with the stored
  value.

  The problem is, that the password is transferred unencrypted.
  Charlie and his packet sniffer do a good job in compromising.

- Kerberos.  The Kerberos server runs a database, which holds the
  passwords in cleartext.  To log in, you request a "ticket" from
  that server, which contains (among other information) a random
  session key, which is encrypted with your password.

  If you type your password correctly, then you can decrypt the
  session key.  This can then be used to request further tickets,
  like one, that gives C access to S.

  However, Kerberos is not easy to install, and you have to
  dedicate hardware to it, because the kerberos server machine
  is very vulnerable.


I am looking for something, which fits the following design goals:

	a) Withstands the packet sniffer.
	b) Does not store the plaintext passwords in a database.
	c) Does not require additional servers.

Do you know of free software, which serves these aspects?  If
not, I might want to implement the following concept:

- S holds a database with two values for each user:
  * a "request code" q
  * a "reply code" r

- q has two parts:
  * one random number.
  * one number, which must be incremented after access was granted
    to the server.

- r is the same as q encrypted by a session key k:

	r = E(q, k)

  E(plaintext, key) is an encryption "function" like DES.  The
  key k is not stored in the server.  Access is only granted, if
  the client is able to present the correct value for k.

- The session key itself is derived from the secret password p
  and the "request code":

	k = E(q, p)

- When C connects to S, S transmits q to C.

- C requests the password p from the user, then calculates k and
  r using p and q.  Furthermore, it creates a new number q' by
  incrementing the enumerative part and doing some random things
  (like encryption with the current date) on the other half.  k'
  and r' are derived from q'.

- C encrypts the values k and r' using r as key:

	  ke = E(k, r); re' = E(r', r);

  ke, re' and q' are put into a block and transferred back to S.

- S uses r to decrypt k and r' again.  The value of k is checked
  by performing: r == f(q, k).  It is also tested, that the
  enumerative half of q' was really incremented.

- If both tests went ok, access is granted.  q' and r' are stored
  as new values for q and r.


The last things drawn as a picture:


    +--------+            q              +--------+   q, r    +-------+
    |        |   <-------------------    |        | <-------- | data- |
    | client |                           | server |           | base  |
    |        |   ------------------->    |        | --------> |       |
    +--------+        q', k, r'          +--------+   q', r'  +-------+
         ^                ^  ^
         |                |  |
         |            encrypted with r
      Password


I am not a crypto expert.  My question is, whether the following
statements are true, or whether I have overseen some weakness:

- Mr. Packet Sniffer (P) has quite a complicated life.  All he
  gets, is:

  * a random number q
  * a reply, which contains:
    . another random number q',
    . ke = E(k, r) = E(k, E(q, k))
    . re' = E(r', r) = E(E(q', k), E(q, k))

  He will be able to deduce the session k by using a brute force
  attack, but that requires to break at least a combination of
  two ciphers.

- After P broke into one session key k, he can calculate all future
  reply codes r, if he can listen to every connection made.  He is
  not able, though, to derive the future session keys k without
  doing another attack:

	r = E(q, k).

- Hacker H, who already achieved read access to the S' database,
  can use a known plaintext attack to derive a session k:

	r = E(q, k).

- If H and P meet, they form quite a good time.  H knows r, and P
  can spy ke, when it is transmitted.  So they can find out each
  session key k just after it has been used.  Nevertheless,
  session keys are throw away goods, and mostly worthless after
  their use.

- If they have the session key for the next session, they can get
  access to S.  However, they will not be able to create the new
  session key k' from the q' they choose.  But they can use their
  own value instead.  So they will have further access, while the
  old password will no longer work.  The next time, that the
  correct client C connects, he will get repeated login failures.
  He cannot avoid to notice, that something bad happened.

- H(acker) and P(acket Sniffer) can try to deduct the password p
  from k and q.  This requires an attack of the cipher:

	k = E(q, p)

  C is completely indepandent in its choice of the function E.
  If the password is long enough, C can discard some of the
  information from k, before it is transmitted.  Then H and P
  cannot reconstruct the full password from only one session key.



Kai
-- 
Kai Petzke                      | How fast can computers get?
Technical University of Berlin  |
Berlin, Germany                 | Sol 9, of course, on Star Trek.
wpp@marie.physik.tu-berlin.de   |



reply

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Reply to all the recipients using the --to and --cc options:
  reply via email

  To: postgres@postgres.berkeley.edu
  Cc: wpp@marie.physik.tu-berlin.de
  Subject: Re: (NET) authentification over a network
  In-Reply-To: <wpp.768728108@marie>

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

This inbox is served by agora; see mirroring instructions
for how to clone and mirror all data and code used for this inbox