Re: [Iot-onboarding] what can pinned-domain-cert actually pin?

[Michael Richardson <[e-mail]>]

Kent Watsen <[e-mail]> wrote:
    > The "private CA roll-over" scenario you mention is side-stepped by
    > enabling the Owner Certificate to itself be a partial chain (i.e.,
    > total-chain = partial-root-chain + partial-owner-cert-chain), hence the
    > partial-root-chain part can be protected by some offline HSM while
    > signing (e.g., annually) some middle-level CAs that are used to sign
    > the actual Owner Cert EE cert.

i.e. adding a layer of indirection solves the problem in the classic way.

    > Separately in SZTP, there is the notion the device dynamically (during
    > bootstrap) receiving the TA for a bootstrap server it it being
    > redirected to.  In this case, the bootstrap servers are presenting a
    > TLS-based service (specifically, HTTPS), and so EE certs MAY have a
    > public root TA.  It was discussed at one point that the TA cert may
    > expire and thus wouldn't it be better to enable the TA to be expressed
    > as a 2-tuple [ <long-lived TA cert (MAY be a public CA)>, <name of some
    > 'subject' matter in the TA-issued CA cert>].  But this became
    > complicated and, instead, we opted for introducing alerts/alarms for
    > when certificates are nearing expiration.

When certificates, or when vouchers?

    > Part of what made it complicated is that one can imagine an
    > organization obtaining a single top-level organization-wide commercial
    > CA certificate that it uses for a variety of reasons, only one of which
    > is to sign vouchers.

These organization-wide commerical CA certificates seem impossible to obtain
today.  At least, I haven't found a CA willing to sell that anymore, and I
would love to be told I'm wrong.

    > Thus, in such cases, the 2-tuple becomes a
    > 3-tuple (and perhaps even an N-tuple), e.g.: [ <long-lived TA cert (MAY
    > be a public CA)>, <name of some 'subject' matter in the TA-issued CA
    > cert>, <name of some 'subject' matter in the voucher-issuer's cert>].

    > Related, though not exactly, I regret that RFC 8366 states that the
    > pinned-domain-cert value is a single X.509 cert, as opposed to a
    > potential chain of certs.  One reason is so as to simplify support for
    > path-validations, as not all libraries support partial-chain
    > validation.

A revision could change it from single item to an array of items.
That would be relatively easy to figure out in code, but I'm not entirely
sure I understand the use case.  Are these independant anchors, or are the
intended to be related.

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