Hacker News new | comments | show | ask | jobs | submit login
Certificate Authority Authorization (CAA) Research Study (caastudy.github.io)
40 points by okket 6 months ago | hide | past | web | favorite | 10 comments



Whomever controls DNS controls Web PKI. True or false?

Please explain your answer.

You may assume "controls DNS" means potential control over delivery or content of unencrypted DNS packets, e.g. the operator of a parent nameserver, the operator of an open resolver, a delegated authoritative nameserver (e.g., Cloudflare), someone else intercepting and modifying packets, etc.

Rephrased question: In what ways, if any, is Web PKI dependent on DNS?


> control over delivery or content of unencrypted DNS packets

And thus we are back to asking for DNSSec. With DNSSec all Records are signed with RRSIGs, and can be validated as being legitimate answers. Your client needs to support validating these of course.

Now, there are a lot of people who don't like DNSSec, and I don't blame them. It's a hot mess in terms of deployment, etc.

It feels like with CAA, TLSA, and other record types we're getting closer to some of the security people want in DNS. We also now have a pretty decent RFC for DNS over TLS, which provides privacy (where needed) in addition to the signed records.

We should be able to come up with a solution to securing DNS by utilizing the existing CA systems out there. Where we could sign zones, and verify them, independent of the chain of DNSKEY to DS records all the way back to the root (this would be different from the current DNSSec validation). I'm not aware of any existing RFCs for something like this.


"> control over delivery or content of unencrypted DNS packets

And thus we are back to asking for DNSSec."

DNSSEC does not encrypt DNS packets.

(Nor does TLS encrypt UDP DNS packets per packet. It encrypts a TCP stream.)

Not interested in a debate over DNSSEC, but please note I used the word "unencrypted" for a reason.

Not only can the contents of unencrypted packets be tampered with, but, if I am not mistaken, djb has suggested DNSSEC signed RRs are still vulnerable to forgery.


DNS does not only operate over UDP, it also uses TCP.

> (Nor does TLS encrypt UDP DNS packets per packet. It encrypts a TCP stream.)

DNS over TLS is for TCP: https://tools.ietf.org/html/rfc7858

It is an encrypted private stream.

For UDP encryption over DTLS: https://tools.ietf.org/html/rfc8094

> Not interested in a debate over DNSSEC, but please note I used the word "unencrypted" for a reason.

If you want privacy you use DNS over TLS or DNSCrypt. If you want assurance that the records can be trusted, the signatures are required. Because DNS relies on so many intermediary nodes for caching, etc, you need both. You need the RRSIG to validate that the zone actually created and manages that record. If you don't want people snooping on your queries, then you need a secure encrypted connection.

You need both.

> djb has suggested DNSSEC signed RRs are still vulnerable to forgery.

can you find that? I know he's very much against DNSSEC, but I haven't seen comments on forgery specifically.


"You need both."

Me, personally?

I do not use shared caches.

I do not use local or remote caches.

Wrote own nonrecursive (i.e. no RD bit ever set) stub resolver.

As such, I have no use for DNSCrypt.

I use trusted sources for lists of authoritative servers I need. I store IP addresses for those servers permanently and track changes in them, if any, over time.

I have little interest in what ICANN certifies as a "valid" or "invalid" name. DNSSEC therefore does not appeal to me.

IMO, better Web PKI could be focused on IP addresses and user-generated public keys (maybe combined with user-chosen "pet names"), something like SSH. Instead it appears to be solely focused on names issued by a third party and certificates based on those names, also issued by a third party.

Probably "You need both" was a figure of speech. If so, pay no mind.

"can you find that?"

http://cr.yp.to/talks/2016.12.08/slides-djb-20161208-dnssec-... and other DNS talks on that page

Pages 11, 32.


DNS, or whois (43/tcp), or the ability to retrieve/fabricate government documents. There are plenty of documented ways that the Web PKI can be subverted.


Last time I checked Cloudflare had incomplete support for CAA (no issuewild or flags).

OHV also mentioned they are working on it but that was half a year ago and so nothing.


They definitely support issuewild now, but annoyingly will inject their own extremely broad CAA records into your zone unless you go into a different UI and hit "Disable Universal SSL" down the bottom of the page, even if you don't use their proxy service. I tried submitting feedback about this but I assume it got ignored by the support rep.

Not a huge fan of the CAA UI. I would prefer the option to enter the RRs in a raw format (like TXT records), even if it's not the default UI.


That they generate certificates with my name was such a bewildering move (I knew as I monitor CT logs). I had to contact support for them to stop as they admitted there was no UI for that.

> Not a huge fan of the CAA UI. I would prefer the option to enter the RRs in a raw format (like TXT records), even if it's not the default UI.

Couldn't agree more. No SSHFP too. I'm waiting for OVH to add CAA records and I migrate back to them.


I sent a support request a few weeks ago asking for CAA recently too, they didn't even indicate they were planning to add that.




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: