Since no browser today can do that, DNSSEC is a dead technology.
Since no browser today can do that, DNSSEC is a dead technology.
You’re thinking about DANE’s TLSA records, which supports trusting certificates not signed by the CAs browsers already trust. DANE does require DNSSEC to work but that’s not what the article is about.
DNSSEC doesn't matter; the parent commenter is right.
Some stats from this April 2019 presentation [1]:
* 91% of TLDs (.com, .org, .edu, etc.) are signed, including root, which was signed in 2010.
* about 20% of internet users are using validating resolvers, mostly thanks to Google and Cloudflare. I'd add Comcast too.
* Over 50% of the zones in .nl, .se and .no are DNSSEC-signed
What's actually damning is that DNS customers, including the ones with the savviest security teams, aren't opting in, and haven't for many, many years. They all have a choice whether to sign or not, and they choose not to. Unlike TLD signatures, that actually is a strong signal.
Signatures of European zones aren't interesting because they're done automatically by registrars. That's theater, not security.
It's not hard to check this for yourself. Just start with a list of the top domains on the Internet; I used the Moz Top 500, but Alexa will do fine. Create a flat text file with one domain per line, then do:
for N in `cat file` do; host -t ds $N; done
You'll see what I see:First, dozens of .GOV sites have DNSSEC delegations because, until a couple years ago, DNSSEC was mandatory in the USG; that mandate has since been rescinded.
Second, excluding those .GOV sites, the entire list of signed domains in the Moz Top 500 is as follows:
000webhost.com
addtoany.com
berkeley.edu
bund.de
cloudflare.com
cmu.edu
europa.eu
example.com
foxnews.com
icann.org
ietf.org
mediafire.com
mozilla.org
one.com
paypal.com
pixabay.com
stanford.edu
themeforest.net
time.com
welt.de
xfinity.com
I note with what I hope is plainly evident glee that while PAYPAL.COM is signed, none of Paypal's subsidiaries, like Braintree or Venmo, are.For everybody without a signed zone, it would be trivial for a resourceful adversary to pass the challenge by simply interposing to give a bogus answer to the query. Only people with DNSSEC actually get cryptographic security here.
Whether it's surprising that most people (where "most people" includes big banks for example) are comfortable with basically just relying on the honour system as good enough is I suppose a judgement about how smart you think people are or how much they actually care about security.
Key personnel from ISRG (the charity which builds Let's Encrypt) have HN accounts, but mostly steer clear of contradicting HN Fan Favourite Thomas Ptacek because it's not worth it even when he says things which are directly contradicted by the facts.
However I am a sucker and will respond to point out, once again, that in fact Let's Encrypt obeys the ACME RFC in this respect (as well it might given that two of the RFC's editors are key personnel for Let's Encrypt and both ACME and Let's Encrypt are bound together from the outset) and so yes of course it does DNSSEC validation.
The anti-DANE lobby from browsers (which caused a pretty nasty scene for the TLS WG) doesn't change the facts which made DNSSEC necessary and that's why the solution for problems keeps being DNSSEC no matter how much you say it mustn't or shouldn't. The same browsers are currently working on ESNI which, punts the authenticity problem back to DNSSEC because that's the only way it gets done. And they're working on DNS records to drive HTTP/3 without Alt-Svc (and throw in ESNI at the same time) and those punt authenticity to DNSSEC as well, for the same reason.
No matter how hard you shove your fingers into your ears, the authenticity problem has to be solved in DNS and that's what DNSSEC does.
I'm sorry that you aren't the only person in the world that knows people involved with CA operations, but, like you, I've been working in this field for "more than 10 years".
If you read my short comment more carefully, you'll see that I didn't say that LetsEncrypt doesn't do DNSSEC. It does, unlike other CAs that claimed to do DNSSEC and then stopped. My point, of course, is that it doesn't matter, because nobody signs their zones.
Since browsers are the only important customer of DANE, it's funny to see them referred to as "the anti-DANE lobby". Is it a "lobby" when it's the entire population? Were the Romans the "anti-Visigoth lobby"?
Your premise that browsers using DNS means browsers will have to use DNSSEC is obviously flawed; browsers rely on DNS today, and not a single one of them does DNSSEC. The ones that tried, stopped.
Make more persuasive arguments. It's not my fault, or the result of a message board popularity contest, that anyone can simply look at a list of the most popular and important sites on the Internet and see that virtually none of them are DNSSEC-signed, or observe that it's been 25 years of standardization effort with DNSSEC to no avail, or to poke around in their browser, the most important piece of Internet software they run, and see that it has no support for DNSSEC, or find articles from the TLS architects of browsers explaining why they can't get DANE to work. Those are just facts. They're not personal attacks on you.
ps
Here's a link from another person who doesn't want to wade into the orange site to make a point, about LetsEncrypt doing multi-perspective lookups, which protect DV challenges from spoofing and, of course, do not rely on DNSSEC, which nobody uses:
https://community.letsencrypt.org/t/acme-v1-v2-validating-ch...
pps
If ISRG is so bought into DNSSEC, why isn't LetsEncrypt.org signed?
In the context that's exactly what they are? DANE proponents securing server-to-server email wanted to have an HSTS-style flag for TLS so they could use it for "OK this server says it always does DANE, if I try to connect in future and can't get DANE that's an attack, I'll just spool the mail". They wrote a simple draft which was then proposed for adoption by the working group. The honest thing for the anti-DANE lobby to do was reject adoption. The draft could proceed as an independent submission and live or die on its own. But instead tactically browser vendor participants supported adoption, and then stalled the process asking for endless tweaks and extra stuff even though they had no intention to deploy it. Basically they ensured it wouldn't get finished and adopted in a timely fashion. AIUI Kathleen chewed people out for abuse of process, everybody was very contrite but the harm can't be undone.
Can you address anything I actually said?
That all the browser developers as a bloc will pressure standards groups not to make more DNSSEC happen is not something I'm interested in litigating. Since they're the model consumers of DANE, it's weird to refer to them as a lobby; it's like talking about "the user lobby". It's a semantic point and not an important one, so I'm fine to stipulate that "almost everyone who will ever have to deal with DANE" constitutes "the anti-DANE lobby".
Here? Or in the Twitter thread where for some reason rather than quote what I wrote you decided to make stuff up about me claiming Let's Encrypt loves DNSSEC ?
Let's try this:
> it doesn't matter, because nobody signs their zones.
Last century when I spent a lot of my spare time working on a vanity site with friends that site didn't have HTTPS. These days the same site (now mostly historical) does have HTTPS.
Did HTTPS not matter last century? Did it start mattering? When and how did that happen?
No, HTTPS mattered all along but ordinary people don't really prioritise security. It's a long way behind like dental hygiene and remembering grandma's birthday, and like those it'll only get done if it's made as convenient as possible.
So that's what we have to do. Well, that's what we should be doing. Instead you're arguing for the status quo for the past half a decade or so.
> That all the browser developers as a bloc will pressure standards groups not to make more DNSSEC happen is not something I'm interested in litigating
DANE was undeployable. But what they learned with TLS 1.3 was that if something is undeployable (originally TLS 1.3 was undeployable) you can fix that through a mixture of encrypting more things and lying to middleboxes. So that's what you should be expecting.
The future isn't a closed web where everything is just Facebook posts but to a middlebox it may look that way because it turns out that middleboxes allow Facebook posts unmolested (People who sign off on buying a middlebox don't know what TCP/IP is but they use Facebook and they'll be angry if it stops working) and so it's easier to just disguise say a DNS request as a Facebook post and then unwrap and decrypt it at the destination. Only half joking.
But as you well know, virtually none of the DV challenges done by Lets Encrypt use DNSSEC in any way. Lets Encrypt is not operationally dependent on DNSSEC. If the root DNSSEC keys landed on Pastebin tomorrow, literally nothing would change at Lets Encrypt. That's not hyperbole; if it's inaccurate in any way, it'll be in some nitpicky way that will probably just prove my point.
We apparently both have, in some sense, sources from ISRG. They're apparently saying conflicting things, because the ISRG people you've heard from would speak up for DNSSEC if it weren't for their fear of me arguing with them. I have less faith in my ability to shout down Lets Encrypt staff than you do, and also people telling me the exact opposite of what you are about their take on DNSSEC; what I hear is "they feel they have to support it, and as long as they have to support it they're going to do it right, unlike other CAs".
But rather than accepting these weird secondhand arguments, we can just look at the actual behavior of Lets Encrypt:
* LetsEncrypt.org isn't DNSSEC-signed.
* The ISRG's own site isn't DNSSEC-signed.
* LetsEncrypt is investing in non-DNSSEC DNS security tools.
* LetsEncrypt doesn't even suggest that users enable DNSSEC.
Make a more persuasive argument.
Meanwhile: TLS has always mattered, because TLS actually does something meaningful to protect traffic. DNSSEC does not. That is the reason we had a push to "encrypt the web" that involved getting everyone onto TLS, and no corresponding effort to do the same with the DNS. Another thing you're well aware of was that one of the key design goals for SSL/TLS was not to depend on the DNS for security.
I'm still not clear on whether you're talking about chain-extensions or something else. Chain-extensions is pretty funny. I'm happy to drill into it if you want.
Later
I guess we don't have to take secondhand arguments for this anymore after all:
Did you hear that from @cpu too? Is it an actual quote, or is this you paraphrasing again and it might be completely wrong?
Regardless this did prompt me to go back and look at what became https://github.com/letsencrypt/boulder/pull/3726 and confirm that yeah, they still don't get it. Must try harder.
I thought I did a pretty good job of spelling out what needs to happen to have confidence here, using an example from EMV of what not to do - but it didn't have the desired effect and this is still the same "parse first, regret later" approach that is ultimately why that PR even exists in the first place (Let's Encrypt were mis-parsing CAA records and discovered in post mortem their records weren't good enough to actually figure out the impact from the incident).
> Another thing you're well aware of was that one of the key design goals for SSL/TLS was not to depend on the DNS for security.
I can't figure out any way to squint at this which makes it both true and non-trivial.
In practice what Netscape did was punt to the existing Certificate Authorities and carefully avoid asking awkward questions about how they did it. This achieves a goal "not to depend on the DNS" only in the same sense that that chap the other day was able "not to depend on a mobile phone" by just getting a friend to receive all their calls for them and pass on messages...
For many years what the Certificate Authorities actually did was not officially documented anywhere. But we do have some amazing patents they filed to protect their "innovative" solutions. They're uh... they're very bad. Go read some, you can laugh now because none of this is permitted any more. What we know of the unpatented stuff they did was mostly worse. Would you be surprised that OCRing the images from anti-spammer WHOIS websites was an actual thing that an actual trusted Certificate Authority was doing to figure out the contact email address for a domain? Guess how reliable that was...
I'm not sure how your argument hasn't been comprehensively refuted at this point.
I would love to take a merry walk with you through the tls-wg mailing list posts about dnssec-chain-extension, which I feel fairly confident I can show you've mischaracterized. The whole thing is really funny.
I don't think this has been an especially productive exchange on HN itself, but since we both appear on opposite sides of the DNSSEC issue every time it comes up here, and your arguments tend to involve subtleties of what Lets Encrypt does or doesn't do, I appreciate the opportunity and motivation to read up on specifics, so that I can more quickly shut down this bad argument when it's brought up again. Thank you!
Anyway I wasn't talking about Melinda, who as a co-author would be in your terminology one of the proponents that you think she's directing that message at, I was talking about Kathleen Moriarty at IETF 103.
The chair's choice of phrasing "This was not our finest hour" undersells it somewhat.
Links to your messages are welcome.
https://news.ycombinator.com/newsguidelines.html
It's a pity you posted that way, because users flagged it, which eventually removed it from view. That makes sense, because protecting the container here (the commons—not any one user) is more important. But it's a pity because it takes your argument out of the mix. Even if users hadn't flagged it, crossing into snark/flamewar/personal attack discredits your points anyway, so please don't do that.
I know nothing about the issue here, but it seems to me that if your post had begun at "Let's Encrypt obeys the ACME RFC" and ended at "for the same reason", it would have been fine for HN.
(If you care, you'd be welcome to email hn@ycombinator.com and we could let you edit the post.)
DNSSEC also enables them to verify whether empty CAA records are intentional or the result of something malicious.
From https://tools.ietf.org/html/rfc6844#section-4.1
Use of DNSSEC to authenticate CAA RRs is strongly RECOMMENDED but not required. An issuer MUST NOT issue certificates if doing so would conflict with the relevant CAA Resource Record set, irrespective of whether the corresponding DNS records are signed.
DNSSEC provides a proof of non-existence for both DNS domains and RR set within domains. DNSSEC verification thus enables an issuer to determine if the answer to a CAA record query is empty because the RR set is empty or if it is non-empty but the response has been suppressed.
Use of DNSSEC allows an issuer to acquire and archive a proof that they were authorized to issue certificates for the domain.
Verification of such archives MAY be an audit requirement to verify CAA record processing compliance. Publication of such archives MAY be a transparency requirement to verify CAA record processing compliance.
Once again, it is my claim that Lets Encrypt does not operationally rely on DNSSEC. Nick Lamb disagreed, and suggested that ISRG staff would stick up for DNSSEC but for my presence on the thread. I reply with a quote from a Lets Encrypt developer: "I think DNSSEC is rubbish and so does the whole Boulder development team".
I think you should take this up with them.
That’s not the point, although I doubt most people on this thread knew that the use of DNSSEC is strongly recommended for issuers of certificates.
I’m pretty sure Lets Encrypt doesn’t operationally depend on DNSSEC, but if they’re issuing certificates to a zone with DNSSEC-signed CAA resource records, that’s irrefutable proof of the intent of the zone operator, something that Lets Encrypt and other issuers of certificates should probably pay attention to, regardless of the personal feelings of their engineers about any particular standard.
Without DNSSEC, there’s no way to know if the CAA records have been tampered with when using something like ACME.
Up until now, you only had to intercept one path:
> On Wednesday February 19th, 2020 we’ll turn on stricter validation requirements 30 in production. We’ll make multiple validation requests from different network perspectives.
That doesn’t sound dead to me.