On Mon, Jan 10, 2022 at 7:51 AM Fabian Stelzer <fs at gigacodes.de>
wrote:>
> On 09.01.2022 15:56, ?ngel wrote:
> >On 2022-01-07 at 18:51 +0000, Morgan, Iain (ARC-TN)[InuTeq, LLC] wrote:
> >> What we would want is some way of associating the restrictions
with a
> >> key so that they are automatically applied when the key is loaded
> >> into the agent, either by ssh-add or when triggered by
> >> AddKeysToAgent. That would either mean adding the restrictions to
the
> >> private key, or adding another file that is associated with the
key
> >> which contains the restrictions. The latter would be easier to
> >> implement from a backwards-compatibility standpoint, but I'm
not
> >> particularly happy about adding another file type.
> >>
> >> The file containing the options would be assumed to be in the same
> >> directory as the private key, and would be formed by adding an
> >> extension such as ".opts" or ".cfg" to the
name of the private key.
> >>
> >> I don't have any specific ideas as to how to express the
> >> restrictions, but we would probably want one line for each allowed
> >> path. Also, the ssh-add command line should be able to override
the
> >> restrictions. Perhaps the syntax might be something like:
> >>
> >> AllowedPath: foo.example.com > bar.example.com ...
> >>
> >> Alternatively, perhaps it would be possible to include the
> >> restrictions as plain-text in the private key, before the
"-----
> >> BEGIN" line. That assumes that any text outside of the BEGIN
... END
> >> sequence is ignored by existing implementations.
> >>
> >> As an aside, either approach should allow for adding support for
> >> other kinds of restrictions. For example, it might be convenient
to
> >> add a Timeout parameter that would correspond to the -t option
with
> >> ssh-add.
> >>
> >> --
> >> Iain
> >
> >Good point. A key should probably always have the same restrictions
> >(after some testing to get those right).
> >
> >I would include them as armor headers, though:
> >
> >-----BEGIN ALGO PRIVATE KEY-----
> >Proc-Type: 4,ENCRYPTED
> >DEK-Info: AES-128-CBC,AABBCC...
> >Destination-constraint: perseus at cetus.example.org
> >Destination-constraint: scylla.example.org
> >Destination-constraint: scylla.example.org>medea at
charybdis.example.orgscylla.example.org
> >
> >SGVsbG8gd29ybGQhCg...
> >
>
> I think the trend is (and should be) going towards keeping private keys in
> hardware tokens and no longer in files on disk. FIDO, PIV Smartcards, GPG
> Smartcards, HSMs, TPMs all allow for very safe private key storage. If you
> allow these constraints to be specified in a config file either referenced
> under a key fingerprint (probably easiest):
There are many trends Many, many tools and systems rely on old
features and would break very badly indeed if support for private keys
is lost. And there are some features of locally stured keys, such as
using multiple keys simultaneously to provide signed keys such as
those supported by Hashicorp Vault, that do not work well with other
key storage formats
(https://www.vaultproject.io/docs/secrets/ssh/signed-ssh-certificates
)
> Key SHA:256INGERPRINT:
> DestinationConstraint: perseus at cetus.example.org
> DestinationConstraint: scylla.example.org
> DestinationConstraint:
> scylla.example.org>medea at charybdis.example.orgscylla.example.org
>
> Or maybe under a Host section:
> Host cetus.example.org
> ForwardAgent yes
> ForwardKey SHA256:xxx
> ForwardKeyConstraint scylla.example.org
>
> then you can still auto load these instead of writing ssh-add scripts. I
> think most ssh users who would use this feature probably already have some
> mechanism of managing their .ssh/config.
For keys such as signed SSH certificates, I'd want to support
providing multiple forwarding keys. I'm not working with those
personally right now, but the linking of only single keyes to SSH
access inerferes with the otherwise supported signed key access.