From: Marc Horowitz <marc@MIT.EDU>
Date: Sun, 29 Nov 92 13:27:17 PST
To: cypherpunks@toad.com
Subject: crypto dongle design problems
Message-ID: <9211292126.AA22394@deathtongue.MIT.EDU>
MIME-Version: 1.0
Content-Type: text/plain


A few comments on "Detailed Scenario of Dongle Usage":

>> So, you walk up to ANY terminal directly connected to a host, or to a
>> terminal server, or a personal computer of any kind connected through a
>> modem, or borrow someone's laptop connected to a cellular modem.

As several people have pointed out, terminals are passe'.  In general,
my mail never goes over a serial line.  Here at MIT, most people have
unix machines on their desks.  Mail is composed on the unix machine,
sent via ethernet to the mail hub, and received via POP or NFS from
the mailhub.  To make life worse, often, the machines are in public
clusters where anyone can log in.  And, (sit down here), the root
password for these public machines is well known.  (The servers' root
passwords are a closely guarded secret.)  So, for me, the
crypto-dongle is almost completely useless. 

>> Call up, or telnet, dial up a bbs, or connect through any other means,
>> to your host, and log in.

Except IP.  One of my machine at work runs SLIP over its serial port.
What am I supposed to do now?

>> Any time an encrypted message is displayed whose public key ID matches
>> one of the private keys on your keyring, the dongle temporarily buffers
>> the message it into its RAM, flashes the "decrypt" LED, while
>> decrypting the message. 

Someone else mentioned this, but I didn't understand your response.  I
read mail in emacs, a full-screen editor.  it displays the first
screenfull, and I hit space to scroll down.  Does the dongle recognize
emacs and hit space, or what?

>> Now, press the "encrypt" button on the dongle.  It presents you with a
>> simple line editor, that works with any terminal or terminal emulation,
>> but is reasonably easy to use (something like most bbs-es use for
>> composing messages).  You type your message, or if it was prepared
>> ahead of time on the local equipment, you transmit the text.

So, I'm trusting the local equipment with my private message.  Even if
I'm only "typing directly to the serial port", how do I know if my
friend's borrowed laptop has been (maybe unknowingly) modified to
store the data?

Similar problems arise with all the other sample applications you
describe.

More personal comments:

I've had several years of experience designing large-scale systems
incorporating security features.  One of the most valuable lessons
learned is that security and privacy must be designed into a system
from the beginning, not added later.  Ease of use comes from design,
not hacks, which inevitably fail.

My view of a useful crypto-dongle is more like Phil's.  A device with
one serial port, which stores my private keyring.  It has functions to
list the keys it has (names and hashes, not keys), to decrypt a
message (while displaying some external indication of having done so),
and to sign a message (again, with display).  Programs would have to
be modified to use the dongle to do the appropriate very sensitive
crypto stuff, and I can do the DES or other work on my desktop
machine.  I forsee the dongle being made of a well-known CPU or MCU
(easily verifiable), an EPROM (software), a static RAM (long-term,
encrypted storage of keys), a DRAM (short-term cleartext storage and
buffering), a UART, an input keypad of some sort, a small (say 1x20
LCD display for simple output, and perhaps a special chip for the
math.  Sources to the EPROM would be available, of course.  Paranoid
people could burn their own chip (assuming they trusted the programmer
:-).  Yeah, this isn't much use on a dumb terminal, but dumb terminals
aren't of much use nowadays, anyway.

		Marc