On 02/02/15 03:53, Damien Miller wrote:
On Mon, 2 Feb 2015, ?ngel Gonz?lez wrote:
IMHO the way to go is not teach ssh the agent protocol, but modify the agent
protocol so that each request gets prepended the hostname requesting it
(forwarded connections would contain the full chain)
Then you have to modify all of ssh, sshd and ssh-agent and doesn't
work until they are all upgraded.
Only ssh-agent and ssh (and the change to the former could be trivial)
Moreover, unless you include signing (by the hostkey) for forwarded hops
and verification of same at the agent side, then you can't trust anything
past the first hop.
I wasn't attempting to go that far. Just accountability, similar to how
Received:
headers work in SMTP. And yes, you can't trust anything past the first
evil hop.
Still, I see many benefits compared to the current all-or-nothing agent
trust.
(Of course, to be really sure that nobody intercepts the agent request,
you MUST
perform the ssh connection locally, with a ProxyCommand. Full Stop)
_______________________________________________
openssh-unix-dev mailing list
openssh-unix-dev@xxxxxxxxxxx
https://lists.mindrot.org/mailman/listinfo/openssh-unix-dev