Re: [websec] Decouple HSTS's two orthogonal effects?

[Adam Barth <[e-mail]>]

On Tue, Mar 29, 2011 at 2:29 PM, Tom Ritter <[e-mail]> wrote:
> On Tue, Mar 29, 2011 at 4:58 PM, Adam Barth <[e-mail]> wrote:
>> There's no coupling between HSTS and the particular algorithm a UA
>> uses to verify certificates.  The UA is free to use whatever
>> verification mechanism it desires.
>
> This is good, but perhaps some clarification to the draft would be in order:
>
> Section 2.2 states:
>
>   2.  The UA terminates, without user recourse, any secure transport
>       connection attempts upon any and all secure transport errors or
>       warnings, including those caused by a site presenting self-signed
>       certificates

If a self-signed certificate does not cause a secure transport error,
then you're all set.  For example, it's fine for a self-signed
certificate to be in the list of explicitly trusted certificates.  In
that case, no secure transport error is generated.  Try it.  :)

> Knowing that HSTS allows any validation method a posteriori allows you
> interpret this correctly - that self-signed certs *may* be allowed
> under HSTS, if the user has added them to their store.  But without
> that, it may be interpretted incorrectly - that no self-signed certs
> would be allowed.

That's not what it says.

> Furthermore, I'm not sure, but "any and all secure
> transport errors or warnings" may be ambiguous.  I don't know if it's
> an existing standard to enter a warning or error state in event of
> (for example) a revocation check failure - although we do know that
> most browsers do not present any warning or error.  There's more on
> that in Adam Langley's thread.   If HSTS does not define whether or
> not a revocation check failure is an error condition, I think it
> should.

Indeed.  A reference there would be helpful.

> Also Section 9 recommends distributing root CA certs to users'
> browsers, and does not mention the possibly of distributing the leaf
> certs instead.  Less related, but I prefer to trust organizations leaf
> certs individually than their root cert.

I don't have a problem with also recommending leaf certs, but you
should check with =JeffH.

Adam