Richard M. Smith wrote:
> < ... > What I still do not understand is how a Web site communicates
with the USB key device.
>
>
Token-aware browsers (Internet Explorer, Netscape, Mozilla, etc.) can be
configured with PKI credentials (e.g., private key and corresponding ID
certificate) that are stored either on the PC or on a smart card or USB
token. The client's private key is used during the optional (and server
controlled) "SSL client authentication" protocol sequence. The token
must be "online" at SSL/TLS connect time. The user is prompted to
enter his/her user PIN as needed to access the private key (which cannot
be read from the token under any circumstance). Use of certificates
that have been issued by an Enterprise CA, etc. can help simplify user
interaction. For example, the secure website can inform the remote
client/browser that it accepts client certificates issued by a certain
CA or list of CAs. In that case, other available client certificates,
if any, will not be part of the site's user-authentication process.
It's important to remember that, SSL connections can be "resumed" for
some (server-specified) period of time. The user can revisit the secure
site without additional token interaction. Normally, this resume
capability is invalidated immediately upon browser program close (since
the client's copy of previously negotiated session parameters is
deleted) or by expiration of the server/site defined time period. This
convenient but somewhat counter-intuitive property of the web browser &
underlying protocol operation must be factored into secure application
design.
So the web site communicates with a client machine and browser
software. Token operations occur indirectly through browser software,
token middleware, and USB device drivers.
A couple of fun? details for future reference ...
Per current standards, smart cards/tokens can be configured with a
"secondary authentication" PIN that must be presented each time a
particular signing key is used to perform it's digital signature
operation. This "Signing PIN" may be distinct from the User PIN that
controls general access to all protected objects on the card (private
keys, certificates, data objects, etc.). See the most recent version of
RSA's PKCS#11 standard for details. I believe Microsoft supports same
or simlar via their CAPI interface. Intended effect is "one digital
signature per Signing-PIN presentation".
Some experts believe that, in some environments, the user's PIN should
be entered through a specially designed reader that protects the PIN
from being captured (by keyloggers, etc.) -- see "protected
authentication path" in PKCS#11. Several vendors have fielded readers
with a PINPad capability that allows the PIN to be entered directly
through the reader unit (e.g., bypassing the client workstation).
If / when such capabilities will be sufficiently viable in commercial
products appears to be an open question.
HTH.
Best,
--dg
Dale Gustafson
Future Foundation Inc.
Secure Application Consulting & Design
+1 651-452-9033 office
+1 218-343-9724 mobile
+1 202-473-9481 onsite/DC
