Re: [Masque] FW: New Version Notification for draft-kuehlewind-quic-proxy-discovery-00.txt

[Eric Rescorla <[e-mail]>]

On Wed, Jan 22, 2020 at 5:54 AM Mirja Kuehlewind <mirja.kuehlewind=
[e-mail]> wrote:

> Hi Partick, hi Lucus, hi all,
>
>
>
> I’m currently updating the discovery draft and therefore looking at this
> email thread.
>
>
>
> So looking at WPAD, it also uses DHCP and/or DNS to get a config file. In
> the draft we mentioned also the use of Provisioning Domains (PvDs) to get
> information about local configuration. This seems to be the more
> appropriate and soon to be standardized alternative, so I don’t think
> mentioning WPAD will add much. Lucas?
>
>
>
> Mentioning WPAD could maybe lead to a discussion about who this
> information can be access by an application/browser (if needed because it
> could also be the OS that handles the proxying transparently for the app),
> however, this seems a bit out of scope for this document and probably is
> mostly covered by taps for PvDs.
>
>
>
> Patrick, you said below that “series of CONNECT tunnels to the proxy are
> giving up the end to end security of transport features that QUIC“. In
> the setup we have in mind for use with MASQUE, we never touch the encrypted
> QUIC end-to-end connection and just forward those packets. Other than for
> HTTP/TCP proxies, the idea is to always have an outer QUIC connection
> between the client and the proxy and an inner QUIC connection between the
> client and the server. So I think if you only look for a forwarding service
> (+ address translation/anonymization), it’s actually okay to rely on
> potentially unauthenticated local discovery as you basically only have to
> trust the proxy as much as you trust any router on the path.
>

Well,  one of the most common reasons why you would want some sort of
tunnel proxy is to conceal your behavior from the local network, which is
clearly inconsistent with having the local network supply the proxy
information.

More generally, I find the whole premise of this draft somewhat confusing.
If we look at TLS MITM proxies as a loose analogy, there are two kinds of
configuration one might want:

1. To tell the client to trust the proxy
2. To tell the client to connect to the proxy

However, as a practical matter, TLS MITM proxies don't do (2) and just rely
on network mechanisms to make that work. They only configure trust anchors
to let the client trust the proxy. Similarly here, the challenge with QUIC
is that you want the client to engage in some behavior that has trust
implications, but for this you need some discovery mechanism that is
secure, which the ones you propose are not.

-Ekr