[Freeipa-devel] Thesis - Gnome Keyring Key Storage in Vault/KRA

[Simo Sorce]

On Tue, 14 Oct 2014 13:21:53 +0200
Martin Kosek <mkosek at redhat.com> wrote:

> On 10/13/2014 07:37 PM, Simo Sorce wrote:
> > On Mon, 13 Oct 2014 13:24:10 +0200
> > Martin Kosek <mkosek at redhat.com> wrote:
> > 
> >> Hello all,
> >>
> >> Last week me, Jakub and Stef discussed a design for a candidate
> >> for a FreeIPA&Gnome keyring related thesis:
> >>
> >> https://thesis-managementsystem.rhcloud.com/topic/show/219/gnome-keyring-storage-in-freeipa
> >>
> >> Apparently, there was a misunderstanding when crafting the topic
> >> proposal, it is not about storing Gnome Keyring *secrets* in
> >> FreeIPA tree, but only about storing a *master key*/AUTHTOK in
> >> FreeIPA so that keyring can be unlocked by SSSD on user log on
> >> even in non-password authentication case (like SSO).
> >>
> >> With SSO, Gnome keyring cannot grab the long term secret to unlock
> >> the keyring, the keyring also cannot re-encrypt itself when the
> >> long term password changes on the server (this is no problem
> >> locally as Gnome keyring is hooked to PAM password change module).
> >> Thus, there is the idea to store the master key/AUTHTOK centrally
> >> on the server.
> >>
> >> Given the sensitivity of the password, it would be best to store it
> >> in the upcoming FreeIPA Password Vault/KRA:
> >>
> >> http://www.freeipa.org/page/V4/Password_Vault
> >>
> >> However, here comes the problem with how to authorize the operation
> >> on SSSD level. To make it working, SSSD,
> >> host/client.example.com at EXAMPLE.COM would need to be able to
> >> retrieve and decipher a key from (any) user's Vault, which is
> >> probably not what we want to allow. We may add S4U2Proxy rules so
> >> that host/client.example.com at EXAMPLE.COM can get
> >> vault/server.example.com at EXAMPLE.COM for the client, but then SSSD
> >> could grab any user's secret when he logs in.
> >>
> >> Are there any ideas how to overcome this issue? Otherwise, it seems
> >> that the Vault idea is not the right way. Maybe centralizing access
> >> to Gnome keyring is not a good idea at all, we will see.
> > 
> > SSSD has access to the user credentials at authentication, that's
> > what it should use to retrieve the user's master key.
> 
> I am confused. I thought this would only work if someone
> authenticates with user+password or sends his TGT, right? Otherwise
> SSSD cannot use user credentials...

Well isn't this always the case when you do a GDM login ?
You are not really going to use the gnome-keyring when you log in via
SSH.

> It is true that authenticating with user+password will still be the
> most common authentication method on desktop interactive login, so it
> should cover the most scenarios. Question is if we are even
> interested in use cases like unlocking GNOME keyring after ssh-ing to
> a host with Kerberos.

No, it doesn't happen today either as most people log in via ssh keys
so gnome-keyring would have no password there either even if it were
hooked in the sshd PAM stack (which is not).

Also I think we could offer an option to cache this key (maybe
encrypted in the host keytab (not that it helps much if someone get
hold of the machine), so that it always works for offline or
password-less logins (think swiping the fingeprint and bypassing SSSD
auth altogether).

Simo.

-- 
Simo Sorce * Red Hat, Inc * New York