Deriving keys

[Wyllys Ingersoll]

Sam Hartman wrote:
>  Unfortunately, expanding the lucid context and making krb5_derive_key
>  public both seem unacceptable.
>
>  The reason is the same.  The representation of a derived key in RFC
>  3961 is crypto system specific.  For some crypto systems you might
>  just store a key schedule in some very implementation specific form.
>  There's no reason to believe that all crypto systems will even call
>  or support krb5_derive_key.
>
>
>  I think that you really do need a fairly full RFC 3961 implementation
>  in the kernel if you hope to avoid significant implementation
>  dependence.



I think that's the conclusion that we came to as well.

In Solaris, the GSSAPI/KRB5 support in the kernel includes most of the
critical crypto APIs.     I think its significantly easier to do it this 
way than to
try and design a secure method for deriving keys in userland and then
somehow passing them into the kernel via some new interface.

You can actually see how we split our code by following the links on 
this page:
http://www.opensolaris.org/os/community/security


-Wyllys