- From: Erik Nygren <erik+ietf@nygren.org>
- Date: Tue, 18 Jul 2017 21:26:36 -0400
- To: Watson Ladd <watson@cloudflare.com>
- Cc: Mike Bishop <Michael.Bishop@microsoft.com>, Ryan Hamilton <rch@google.com>, Martin Thomson <martin.thomson@gmail.com>, "ietf-http-wg@w3.org" <ietf-http-wg@w3.org>, Piotr Sikora <piotrsikora@google.com>
- Message-ID: <CAKC-DJixdMSF626OGOE86AU8+ujQKK8myLYoBXaTHLWvG8P7WA@mail.gmail.com>
On Tue, Jul 18, 2017 at 7:29 PM, Watson Ladd <watson@cloudflare.com> wrote: > On Tue, Jul 18, 2017 at 2:23 PM, Mike Bishop <Michael.Bishop@microsoft.com > > wrote: > >> Unless the cdnjs.cloudflare.com is in the same cert, you’d also need the >> Secondary Certificates extension to support that. Once it’s there, though, >> I think the discussion is the same whether we have multiple certificates or >> one – we have a single factor (the certificate) that claims authority for >> some other name. How do we reach a sufficient level of trust? DNS is a >> weak second factor; are there other second factors we would be willing to >> consider as equivalent? >> > > We can put it in the same cert for a lot of our customers and Secondary > Certificates is also on the roadmap for this. > > It depends on what we want to guard against. If it is just mississuance, > presence of the certificate in CT logs is fine. If we're concerned about an > attacker with the private key of the certificate, then presence of the > certificate in the CT logs isn't enough and DNS provides an extra fig-leaf. > > A lot of the proposals being discussed aren't detailed enough for me to > say if they would work or not for us. Would it be enough to have every site > Alt-Svc cloudflare.com (easy), or would the Alt-Svc proposals end up N^2 > scaling (nonstarter)? Our IPs aren't shared between sites in any sort of > regular pattern and which one is used by customer.example.com can change > for various reasons. > > We can't have cdnjs.cloudflare.com advertise every single customer as an > Alt-Svc. > >> Strawperson: if every one of a CDN's sites (including cdnjs.example.com) did an: Alt-Svc: h2="cdn.example.net:443"; ma=7200; coalesce=1 then one might argue that there is a positive indication from a strongly authenticated origin (where DNS was followed initially to receive the Alt-Svc) saying that connections to cdn.example.net can safely coalesce as long as there is cert coverage. ie, if the cert presented when connecting to cdn.example.net and sending SNI="www.example.com" covers "www.example.com", or if such a cert is pushed via Secondary Certificates down the road. It may be that to make this really useful, clients may want to be able to probe for a Secondary Cert + ORIGIN to be sent to them. The risk exposures in this case seem to be from the Alt-Svc MA TTL extending too long (ie, long past a typical DNS TTL, so perhaps there should be a cap on the MA allowed before re-authenticating with the origin), or if the Alt-Svc is learned from an Origin using a MitM CA cert (which we may want to explictly prohibit being used in combination here as otherwise this could be used to shove everything through a MitM appliance that can construct certs, unless that is a desired behavior which I doubt is the case in most places). Erik
Received on Wednesday, 19 July 2017 01:27:05 UTC