Channel Bindings questions

[Sam Hartman]

>>>>> "Larry" == Liqiang(Larry) Zhu <lzhu at windows.microsoft.com> writes:

    Larry> Folks, can you comment on the following findings by one of
    Larry> our customers.

The assertion that channel bindings are useless is false.  Remember
they are authenticated; the server knows whether a client intended to
include them.

However I agree that the MIT behavior is wrong; see bug #2591 at
http://krbdev.mit.edu/rt/ (login guest, password guest).  

Here's a description of the bug:


Based on discussion on kerberos at mit.edu, the decision to allow null
channel bindings from a client to match even when server channel
bindings are supplied is flawed.  This decision assumes that we cannot
get server implementations to change even though we are able to deploy
a new Kerberos implementation on the server.  In practice the server
implementations in question have actually changed and so the only part
of revision 1.54 of accept_sec_context.c we actually need is the code
to ignore channel bindings if null channel bindings are passed into
the server.  Thus the change to allow null channel bindings from the
client to match against any channel bindings on the server is backed
out.


--Sam