[Anima] verification of manufacturer in BRSKI

[Michael Richardson <[e-mail]>]

{sorry if this is a duplicate, my draft folder says it did not go out}

Anoop Kumar Pandey <[e-mail]> wrote directly to the draft authors list,
and then gave me permission to share this on the list:
    Anoop> This is in context to the RFC detailing process of enrolling a pledge to a
    Anoop> domain. The major problem with the procedure is that the registrar doesn’t
    Anoop> verify the manufacturer. It simply verifies the voucher and enrols the
    Anoop> pledge. If pledge send a self-defined URL of manufacturer where provision for
    Anoop> issuing fake vouchers is in place, then the registrar can be fooled into
    Anoop> enrolment [assuming the registrar can’t know all the manufacturers
    Anoop> exhaustively] and later exploited from inside the domain. Additionally pledge
    Anoop> and a random manufacturer can also collaborate to do the same.

You raise a number of issues.

So let me name them so that we can discuss them easier, and make sure that I
understand you correctly.
Probably, we need to add some text to the Security Considerations, and I
would welcome your help in crafting that text!

    Anoop> In the similar way a registrar may collaborate with the manufacturer to lure
    Anoop> a device using fake voucher.

    Anoop> How can these problems be tackled?

I agree that the Pledge may point at any Manufacturer (in the form of a MASA)
that it wishes to.   The following identities exist:

1) The Pledge's Identity (PI) in the form an IDevID certificate.
   (It signs the voucher request to the JRC, and the Client side of the TLS
   connection from Pledge to JRC)
2) The Manufacturer's Identity (MI) which signs the IDevID certificate.
3) The MASA Identity (MASA) which signs the voucher.
4) The Join Registrar (JRC) which signs the voucher request to the MASA.
   It also signs the Server side of the TLS connection from Pledge to JRC.
5) The Domain PKI Certificate Authority, which signs the LDevID.

(The MI might be the same as the MASA, but in general the MASA
may be outsourced.  The Pledge SHOULD have a trusted anchor for both,
but it can be designed where it has a manufacturer CA which signs both MASA
and MI certificate)

Each of these in essence represent a private key!
I'd like start by ruling out of bounds for this discussion any scenario where
the attack requires that the private key be leaked, shared or used
incorrectly.  It's not that they can't happen, but rather that it is a
problem of TPMs, etc. and not protocol design.

The Pledge trusts network part works by:
a) MASA signs VOUCHER.
b) VOUCHER lists JRC in pinned-domain-cert.
c) Pledge uses MASA to validate VOUCHER, and therefore validates JRC.
d) JRC can also audit (verify signatures even) the voucher really comes
   from MASA, although unless there is a common CA, it may not be able to
   prove MASA and MI are same entity.

The JRC trusts Pledge part works by:
e) MI signs IDevID.
f) Pledge uses IDevID to identify to JRC.
g) JRC validates IDevID using MI certificate.

Now, to translate what you said:

problem 1.
    Anoop> The major problem with the procedure is that the registrar doesn’t
    Anoop> verify the manufacturer.

To translate, the JRC has no obvious way to verify that the "MI" key belongs
              to the manufacturer that they care about.

You actually hit the major reason this is not a problem when you assume:
   > assuming the registrar can’t know all the manufacturers exhaustively

We assume that in a managed network that the JRC *can* know all the
legitimate manufacturers.  The keys can come from sales channel integration
(via digital "packing slips" perhaps), can be manually loaded by humans, be
scanned from QR codes on the box, etc.  We believe that this is out of scope.

And if the JRC does see a MI that it does not know, then it can ask a human.
That's one reason we made sure that failures to enroll do not cause the
Pledge to never try again with that JRC.  It needs to try again later,
because maybe nobody has answered the "Yes/No" Dialog on the management
station yet.

And of course there is the WebPKI.  We aren't saying that MI keys
(or MASA server TLS keys) MUST be in the WebPKI, but we aren't saying that
they can't be.  That's not a panacea... between ComodoGate type situations
and certificates for "C1SC0" in a hard to distinguish font, many social
engineering attacks could get that "Yes" button pressed.


problem 2.
    Anoop> If pledge send a self-defined URL of manufacturer where provision for
    Anoop> issuing fake vouchers is in place, then the registrar can be fooled into
    Anoop> enrolment [assuming the registrar can’t know all the manufacturers
    Anoop> exhaustively] and later exploited from inside the domain.

The URL of the manufacturer is embedded in the IDevID certificate (section
2.3 defines the extension).  So the Pledge can't create this lie itself, it
requires collaboration with the MI to create an IDevID.  Such an MI can
point to any MASA it wishes, so long as that device can issue vouchers that
the Pledge can validate.

What this means is that JRC always knows the MI that created the Pledge.
If we can solve problem 1, then it's done.

problem 3.
    Anoop> Additionally pledge
    Anoop> and a random manufacturer can also collaborate to do the same.

I don't see how this is the case.  The pledge can only collaborate with
the manufacturer whose IDevID it has.

problem 4.
    Anoop> In the similar way a registrar may collaborate with the manufacturer to lure
    Anoop> a device using fake voucher.

The Pledge will only trust a voucher signed by MASA.
Any signature from a different entity will be rejected by the Pledge.
So a fake voucher is not possible.
Any voucher created by the real MASA is, by definition, not a fake voucher.
Can you explain this scenario clearer?
Or explain what text we need to change to clear up the mis-understanding?

--
]               Never tell me the odds!                 | ipv6 mesh networks [
]   Michael Richardson, Sandelman Software Works        | network architect  [
]     [e-mail]  http://www.sandelman.ca/        |   ruby on rails    [


-- 
Michael Richardson <[e-mail]>, Sandelman Software Works
 -= IPv6 IoT consulting =-