The fact that the technique relies on a CNAME or some other DNS indirection seems to suggest that the ad server or tracker will have a different IP from the website. That may be another weak point in any effort to conceal the fact that some resources referenced in the page or Javascript files are only necessary for advertising purposes. If both content and ad cruft were being served from a single IP, then that might pose more of a challenge in deciphering what to retreieve. I have yet to see that and doubt I ever will.
I am a believer that ultimately whitelisting is more effective than blacklisting. Request what you want, leave the rest. As opposed to letting a browser request everything according to a web developer's wishes, and then you try to block stuff. With extensions, third party assistance, etc.
For interest, what do you use? A standard browser with plugins, or a specialised client?
Today's "standard browser" that runs Javascript is an omnibus, overly complex, kitchen sink program that is inextricably linked to the online advertising industry. Online ads and tracking generally do not work without the help of one of these so-called "standard" browsers.
So does URL based content-blocking. I recently want to block Youtube/Twitter ads on my own, to my dismay, the ads were buried in some deep JSON response. And ads resource URLs are not easily distinguishable from real content.
Correct, because first-party / third-party is not a technical difference, but a social/commercial one. The app.example.com may run in a different cloud and be part of the same first-party service.
AFAIK, cloudflare only flattens CNAMEs at the root level, and that’s because CNAMEs at the root are not a standard. They have to convert it to an A record to be standards compliant.
Cloudflare nameservers also flatten CNAMEs pointing to workers.dev (a domain they own), for example.
My intention was to point out that nameservers that flatten CNAMEs render DNS-based blocking ineffective.
Besides CNAMEs breaking all sorts of assumptions a client software makes (and hence also causing security headaches in the process as outlined in the paper), there are a couple other DNS cloaking techniques that the paper doesn't discuss:
1. ALIAS records (not standardized? popularized by Route53) hide CNAME-like pointers. Another variant of this is, some DNS nameservers (like Cloudflare) flatten CNAME records (aka transparently ALIAS endpoints): CNAMEs aren't sent with the answer, that is, you're straight up served the A/AAAA record with IPs (which could easily be third-party). DNSSEC doesn't help here, afaik.
2. The shiny new SVCB/HTTPS records open up another avenue for DNS cloaking. For example, consider this (unverified if correct) record with a chain of pointers:
(the above is missing the example where targets follow "port prefix naming" viz. _443._https.example.com)Though it remains trivial to uncloak domains hiding behind SVCB/HTTPS records, implementations have to be careful about what they let through. Flattened CNAMEs and ALIAS records; however, to my knowledge, remain undetectable.
But: All indications are that it is foolish to rely on DNS to discern between first-party and third-party. I mean, I can already run www.example.com on Netlify, app.example.com on Vercel, api.example.com on AWS, and cdn.example.com on Cloudflare... and those endpoints could very well be running anything the cloud providers want (third party).
IP based firewall doesn't suffer these shortcomings, but then, enforcing IP blocks are complicated by Virtual Hosting (multiple web services behind a single IP) and IPv6 (too many addresses to curate and block).