Re: OpenSSH certified keys

[Damien Miller <djm%mindrot.org@localhost>]

On Wed, 17 Mar 2010, Roumen Petrov wrote:

> I think that one of issue to previous drafts (O. Saarenmaa and J.
> Galbraith) as Damien point to this list was that ssh server has to
> parse asn.1.

At present, the SSH server doesn't have to parse ASN.1. The ssh-rsa
signature format does *use* ASN.1 encoding, but it is enough of
a canonical representation that a general parser is not needed.
Avoiding the use of ASN.1 here has saved OpenSSH from a few bugs in
OpenSSL's ASN.1 routines over the years. BTW, that several other ASN.1
implementations have suffered from multiple bugs over the years is (to
me at least) a good signal that the format itself is problematic.

See http://anoncvs.mindrot.org/index.cgi/openssh/ssh-rsa.c?revision=HEAD
for more details.

> If I understand Damien most acceptable is solution is if ssh could
> read without to process(parse) certificate(s) data and to pass to
> external system for further processing. Damien ?

I don't think there is much getting away from the fact that there is
a lot of complexity in the pre-auth path when adding X.509. While
delegating verification and authorisation decisions to a sandboxed agent
outside the SSH server is an improvement, the complexity is still in the
critical path.

-d