News for hackers. Stuff that matters.
Re:Towards HTTPS by Default by janosdebugs on Wednesday August 16, 2023 @05:46PM (#37154327)
Several problems with this proposal:

1. Who defines what the local network is and why should this environment be trusted more? The IP address doesn't guarantee that the packets won't be going round the globe 3 times.

2. You are proposing a whole new, as of yet unresearched verification mechanism to TLS, which won't be supported by a whole lot of devices for a long time. I just had to downgrade a server of mine to TLS 1.2 because a high profile software couldn't deal with 1.3.

3. DHCP is inherently insecure and setting up a rogue AP is also not too difficult outside of an enterprise environment where certificate rollout is already not a problem. Ideal for hijacking an intranet connection covertly.

4. The proposed CA would not have the benefit of public transparency logs, so it would not meet the minimum requirements currently set forth by the browsers to trust a CA.

5. What happens if you are connected to more than one network at a time? Which one gets authority over the special TLD?

6. Who guarantees that the connected client uses the mandated DNS servers to resolve the special domain?

This is not a viable proposal for more than a handful of cases, but makes it a lot less secure for everyone else.

Re:Towards HTTPS by Default by xg15 on Wednesday August 16, 2023 @06:59PM (#37155006)
1 + 3:

DHCP, wifi identification or what you can reach with a broadcast would be examples. If that's too insecure, out-of-band identification would also be possible, i.e. a QR code that contains a SSID, a hostname/IP address where the CA cert can be downloaded and a hash of the cert.

I don't see why it's important how the packet is actually routed as long as you have established a root of trust.

2: Yeah. So?

The worst that can happen to a device that doesn't have support is that it sees an invalid certificate.

I don't think widespread support would be necessary. The only software where support would be essential would be browsers.

4: Neither is a local root CA today. The entire point of the proposal is that it doesn't require any third parties. To make up for this, the CA is only valid for local domains, so you can't generate rogue certificates for google.com anyway.

5: That's a good point, but strictly speaking, the problem already exists today: what happens if you're connected to two networks which route the same IP address range? Or which reuse the same hostnames?

One option would be to add some sort of network identifier to the hostnames to make them unique. This might make the domains unwieldy to use though.

6: The same thing that "guarantees" that the client uses the network's CA for validation: The client itself, that hopefully correctly implements the spec.

You have the same problem with the "split DNS + LE with DNS validation" approach: If a client has 1.1.1.1 hardwired for DNS resolution and you only serve A/AAAA records inside the network, the client won't see your server. That's why even DoH will fall back to the network's DNS servers if it can't resolve a domain over the DoH server.

> This is not a viable proposal for more than a handful of cases, but makes it a lot less secure for everyone else.

Local-only hosts may be less secure than internet hosts, granted - but how would this reduce the security of everyone else?

Re:Towards HTTPS by Default by xnyan on Wednesday August 16, 2023 @08:05PM (#37155527)
>out-of-band identification would also be possible, i.e. a QR code

Any out-of-band method of enabling what you are talking about is necessarily going to prompt the user with the same kind of serious-looking security prompt they would see when adding a self-signed cert to the OS certificate store.

Re:Towards HTTPS by Default by janosdebugs on Wednesday August 16, 2023 @11:32PM (#37156832)
You are making excellent points for using globally resolvable DNS + Letsencrypt. :) None of these problems exist with that solution. Unless you enable AXFR, your hostnames won't be easily visible to the outside world and wildcard certificates are a thing too.

Specifically:

1,3,4,6: You want a seamless user for the user, assuming that the user is capable of making the decision to trust a network operator. This is not a decision an average user can make. Say, I set ip a rogue AP for your university network which uses this special TLD. You connect to my AP, log in to your uni account, and now I have your credentials because you went through my reverse proxy. Very easy to pull off, almost impossible to reconstruct after the fact. QR codes, etc don't offer any protection here.

5. AFAIK the problem doesn't exist as far as domain names are concerned, search domains work across multiple networks. mDNS can also work across multiple networks to resolve host names. For IP addresses this is a problem, which the abundance of possible locally admininstered addresses with IPv6 will hopefully solve (ULA address ranges).

Re:Towards HTTPS by Default by xg15 on Saturday August 19, 2023 @09:14AM (#37188444)
Yeah, you have a point. The fundamental problem is always: How can I keep an adversary from cloning/proxying a legitimate site then trickung users into visiting it. It's a hard problem even if there is a centralized authority and pretty much still open for decentralized systems.

I still think however that there are some properties of local networks that make this less risky:

- an attacker needs to be physically close to their victims in some way (or at least get devices into physical proximity): The rogue access point has to be located somewhere, someone has to tape over the QR codes, someone has to plug into the local network or connect to WLAN to send spoofed ARP/DHCP/DNS replies, etc. Compare that to phishing/typosquatting attacks on the internet where an attacker can be anywhere on the globe and can make their site look legitimate with a simple LE cert.

- an attacker is more at risk of being exposed. i.e. if someone keeps spoofing the university network, people might start to try and locate the wifi signal and knock on doors. That's a lot more risk than phishing/typosquatting for a lot less reward: Even if the attack works perfectly, you can only impersonate the local-only domains, not internet domains. (In fact, we don't currently live in an HTTPS-only world, so there would actually be some value for an attacker to spoof some campus wifi right now. I haven't heard of any incidents in that regard, so that leads me to believe that the risks are to high compared to the rewards)

- there are lots of situations where spoofing the CA information would require you to do a full supply-chain attack, such if I connect directly to a device using a cable or NFC connection; QR code stickers put directly on the device; if I manually add the device of another person to a wifi network I know, etc. (Those are slightly different use cases than "art installation" etc. I'd say they are still valuable though, e.g. for frictionless setup of local-only IoT devices)

> If you want to have full control over your local network environment, that's another thing, but you can't then expect everyone else's devices to automatically trust your environment

But that's the point, I don't want them to fully trust my environment. It's perfectly reasonable that there is no way to override the CA of say, google.com without an extremely scary security warning.

However, I'd like a way to offer some sort of website from a local network without causing a huge hassle for everyone who wants to use it or having to pretend I it's an internet domain.

Re:Towards HTTPS by Default by janosdebugs on Saturday August 19, 2023 @09:57AM (#37188850)
The problem is that even if we assume this (or anything similar) is a good thing, we can't simply wish it into existence. The world is going into a different direction, everything is ever-more connected and apart from a very small minority, most local network admins are fine with Letsencrypt certificates or rolling their own CA.

If someone would want to put in the advocacy and development work, this would need to be written up in a formal fashion, probably in a scientific journal and/or an RFC published. Then there would need to be an implementation for at least one of the major OSes so security researchers can get their hands dirty with it and iron out the bugs. Once all this is done, you still need to convince Microsoft, Apple and Google to ship it in their OS because this is not a browser-only solution and needs OS support. And at this point the thing is probably pretty dead because their big customers are happy to run their own CA infrastructure, which leaves only a tiny minority of sysadmins for small systems that could ask for it. Not enough for these big companies to care because these small customers moving to Linux won't affect their bottom line. I would rather spend my time to advocate for simpler CA management or other verification methods with LE.

Honestly, I think the FQDN and Letsencrypt solution is fine. I've been using FQDNs for server names for a good 15 years now and I never had a problem with it. The people who didn't, however, had a bunch of pain to deal with, from having to merge networks to weird trust issues. I still remember the headless running around when .dev became a TLD and suddenly there was a conflict with oh so many folks' internal hostnames. Nothing about DNS says it has to be on the Internet. If you really want, you can even run a split DNS to make it only available in your own network, you just need to make the Letsencrypt verification work over the Internet.