I would also note that a lack of copy-and-paste need not necessarily be a
barrier to sending around MAC keys.

First (and Karthik may punch me in the face for saying this) but I think
even with a short (say 8-character) MAC key, you're better off than having
a bearer token.

Second, you can make base64url-encoded strings that have non-trivial
entropy and are still easy to type.  For instance,
"lorem-ipsum-dolor-sit-amet" is a valid base64url string (corresponding to
"968ade9be8a9b2e9be768968afeb22b7e6a67a" in hex).  That wouldn't get you
the full entropy for the length of the string, but with sufficiently-large
word list, you could get up to ~64 bits.

On Wed, Nov 23, 2016 at 1:59 AM, Karthikeyan Bhargavan <
[email protected]> wrote:

> Thanks for the clarification, Ray.
>
> I am guessing not having copy/paste means that the customer should be able
> to type in the token into the ACME client?
>
> If so, a second (less desirable, but still) option for non-copy-paste
> scenarios would be to generate a MAC key from the token,
> using a strong modern PBKDF. ACME clients would not normally implement
> this primitive, because they have no need for it,
> so you’d either need to extend the ACME client or ask customers to use a
> command-line key derivation function.
>
> I think everyone gets this, but just to be clear… There may well be some
> use cases where bearer tokens are the right light-weight solution.
> The main concern is about high-value secret tokens, where knowing a token
> would allow an attacker to take over an account
> (and/or link a CA website account to the attacker’s account key). The ACME
> threat model explicitly states that the network
> connection between the ACME client and server should not be relied on for
> confidentiality (e.g. the server may be hosted by a CDN)
> and this isn’t compatible with a secret bearer token.
>
> On 23 Nov 2016, at 02:57, Ray Cheng <[email protected]> wrote:
>
> Richard and Karthikeyan,
>
> We had to support not making copy/paste a hard requirement – desirable
> perhaps but not mandatory.
> Our "ca-account-secret" put us in the copy/paste desirable space, but
> still do-able without it.  A MAC key is realistically  in the do-able only
> with copy/paste spectrum.
>
> If you assume copy/paste is available, then the usability impact of
> introducing a MAC key is lessened as Richard has described.
> This is a  reasonable assumption in most cases so
> https://github.com/ietf-wg-acme/acme/pull/212 is a good step forward for
> these more general use cases.
>
> I believe we still have the option of using  https://github.com/ietf-wg-
> acme/acme/pull/211 to register custom fields for use cases not covered by
> https://github.com/ietf-wg-acme/acme/pull/212. These two PRs together
> should  give us the security and flexibility that we need.
>
> Ray
>
>
> *From:* Richard Barnes [mailto:[email protected] <[email protected]>]
> *Sent:* Tuesday, November 22, 2016 4:55 PM
> *To:* Ray Cheng <[email protected]>
> *Cc:* Karthikeyan Bhargavan <[email protected]>; [email protected];
> Jacob Hoffman-Andrews <[email protected]>
> *Subject:* Re: [Acme] Proposal for adding arbitrary CA-specific
> name-value pairs to registration object
>
>
>
> On Tue, Nov 22, 2016 at 3:37 PM, Richard Barnes <[email protected]> wrote:
>
> On Tue, Nov 22, 2016 at 1:33 PM, Ray Cheng <[email protected]>
> wrote:
>
> We are using a "ca-account-secret" with the "new-reg" request to make an
> association between the ACME account key and a CA account.
>
> We have implemented additional security measures around our
> “ca-account-secret” beyond ACME to satisfy our own security requirements,
> for example:
> 1. Secret expiry
> 2. Account lockout
> 3. Human approval post association
> 4. Monitoring
>
> At least for our CA use case, the logistics of having to distribute a MAC
> key out of band would make it impractical enough that we would likely
> choose other options. For example, we could just have someone manually
> provision their ACME account public key into our CA account system to make
> the association. We didn't choose this route as it was deemed less
> operator-friendly.
>
>
> I'm a little confused here.  The only thing that "distributing a MAC key
> out of band" requires is that you give the customer two strings instead of
> one.  Assuming client software had a place to copy/paste in each of these,
> could you work with that?
>
>
> Here's a PR implementing this idea, for concreteness:
>
> https://github.com/ietf-wg-acme/acme/pull/212
>
>
>
>
> --Richard
>
>
>
>
>
> Regards,
>
> Ray
>
> > -----Original Message-----
> > From: Karthikeyan Bhargavan [mailto:[email protected]]
> > Sent: Saturday, November 19, 2016 2:23 AM
> > To: Ray Cheng <[email protected]>
> > Cc: [email protected]; Jacob Hoffman-Andrews <[email protected]>
> > Subject: Re: [Acme] Proposal for adding arbitrary CA-specific name-value
> > pairs to registration object
> >
> > It is worth noting that “external_secret”, “ca-account-secret”, and
> > “token” are all secret bearer tokens and, as such, offer *significantly*
> > less security than the default ACME mechanisms, and in some cases may
> > defeat the purpose of having account keys.
> >
> > I can see that different vendors may have need for symmetric secrets like
> > these for their own use, but if we were to extend ACME to allow such
> > “external” secrets, we should standardize their secure use.
> >
> > For example, a much safer way to embed “external_secret”, “ca-account-
> > secret”, and “token” would be to have an “external-key-id” field, that
> > indicates a MAC key that has been exchanged out-of-band; we would then
> use
> > MAC-based authentication to bind the MAC key to the relevant JWS request,
> > using the standard nested authentication mechanism we are using in roll-
> > over, something like:
> >
> >       Sign (account-key, (request, external-key-id, MAC (external_key,
> > Hash(account_key))))
> >
> > This would ensure that the external secret is never sent on the wire and
> > is not a bearer token, and the compound authentication guarantees of the
> > resulting protocol are quite strong.
> >
> > More generally, ACME is not a browser-based protocol and there is no need
> > for us to rely on obsolete cookie-like mechanisms; ACME clients and
> > servers are already doing crypto, so why not use MACs instead of passing
> > bearer tokens in URLs or in the request?
> >
> > Best,
> > Karthik
> >
> > > On 31 Oct 2016, at 16:43, Ray Cheng <[email protected]>
> > wrote:
> > >
> > > Hi Jacob,
> > >
> > > From: Jacob Hoffman-Andrews, Thursday, October 27, 2016 5:28 PM:
> > >>
> > >> On 10/04/2016 06:11 AM, Ray Cheng wrote:
> > >>> One way to accomplish this in the protocol is to simply add a "ca-
> > >> extension" object to the registration object, where the "ca-extension"
> > >> object is an array of name-value pairs of strings. For example:
> > >> I think this makes a lot of sense, and is in the spirit of the other
> > >> places we've intentional left hooks for CA-specific customization,
> > >> like OOB challenges. I'm inclined to accept it.
> > >>
> > >> Additionally, your experience as a commercial implementer of ACME may
> > >> be valuable to spot places where the current protocol is lacking. For
> > >> instance, you need an "account" field, and StartCom needs
> > >> (https://github.com/ietf-wg-acme/acme/pull/172) a "token" field. Can
> > >> we support those use cases natively in ACME so they don't need to be
> > >> extensions? What do you put in the "account" field, and how do you
> > >> authenticate the link between the ACME account and the Entrust
> account.
> > >>
> > >
> > > We are effectively using "ca-account-id" and "ca-account-secret" fields
> > to authenticate the link to the Entrust account. Since these fields do
> not
> > currently exist in draft-03 and certbot, we are embedding them in the URL
> > that a particular customer uses to access our ACME server.
> > >
> > > The "external_secret"/"token" is slightly different but is also useful
> > as an "API key" type of authentication.
> > >
> > > Although we see value in a "ca-extension" object, we also support
> adding
> > explicit fields in ACME. There are some advantages to explicit fields -
> > one in particular is the special handling of designated secret fields by
> > clients.
> > >
> > > Thanks for your feedback.
> > >
> > >
> > > _______________________________________________
> > > Acme mailing list
> > > [email protected]
> > > https://www.ietf.org/mailman/listinfo/acme
> >
>
> _______________________________________________
> Acme mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/acme
>
>
>
_______________________________________________
Acme mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/acme

Reply via email to