[Cryptography] Against against DNS (Re: New SSL/TLS certs to each live no longer than 47) days by 2029

[Christian Huitema]

On 4/25/2025 8:38 PM, Viktor Dukhovni wrote:

> The obstacle to DNSSEC in browsers is fundamentally a last-mile problem
> with CPE devices (routers, cable boxes, ISP resolvers, ...) often enough
> making it impossible to get signed answers.  The landscape has changed
> with DoH, ... creating a pathway to bypass all the impedance, but it is
> not yet pervasive to the point of making DNSSEC+DANE available to edge
> consumer devices.

There could be something similar happening with Encrypted Client Hello. 
ECH relies on the client securely obtaining the HTTPS resource record of 
the target server. Since ECH is used to hide the "real" target from the 
local network, it makes sense to get that record without using the local 
DNS servers. So I would expect the client to use DoH to acquire the 
HTTPS record, and then using the default DNS service to get the address 
of the fronting server. So, bifurcated DNS, with data acquired from 
multiple servers.

Whether that's an argument for DNSSEC is a bit debatable. Yes, it would 
be nice if the HTTPS record was properly signed, etc. But DoH (or DoT) 
already protects against the "last mile" attacks, so we are left with 
attacks on the name resolution by the DoH server. Also, there is a high 
correlation between the "authoritative DNS" for a domain and the "CDN 
server" for that domain. See for example 
https://ithi.research.icann.org/graph-m9.html, which shows that the most 
popular DNS servers for big domains are Cloudflare, AWS, Google and 
Akamai. If you are making a DoH connection to Cloudflare and the 
authoritative server for the domain is also Cloudflare, DNSSEC does not 
bring a lot of value...

-- Christian Huitema