Posting because OpenWRT, possibly others as well, assign .lan names to devices on the LAN by default. People who make use of this might want to follow the application, and, if it goes through, either change the name or make sure queries can't get incorrectly get sent to upstream resolvers.
Regarding that last point, I've had issues with dnsmasq in the past where it was remotely resolving domains even when they were configured to resolve locally. For .lan domains, this could be disastrous because I often send plaintext traffic to them. I did submit a fix/workaround[0], but it's still something to look out for. If anyone knows more about this issue or other ways queries could leak, please share!
It looks like as part of the process, they delegate the prefix in global DNS and see how much traffic it currently gets; if it gets too much it will be classified as "high-risk" which at least makes things harder. Which is to say, we want as much leakage as possible to hopefully make ICANN think twice about approving this.
This test is poorly designed for .lan because the opposite problem is more likely: that many people using .lan will have any resolution attempts swallowed by their routers. This is certainly true for OpenWrt and GL.iNet. It may be true for Unifi AmpliFi, which reserves .lan by default. It's potentially true for other vendors when .lan is configured as the local domain (which many online guides suggest people do).
The same thing happened to AVM recently, who make a popular (in Germany) series of home routers that assign <hostname>.fritz.box hostnames to devices by default. `ssh foo.fritz.box` would conveniently work – until .box became a registerable TLD and somebody snatched fritz.box and directed it somewhere else, concerningly also running SSH...
Arguably that's on them, but .lan seems much more controversial and I really hope ICANN does the right thing here and at least declares it permanently reserved/non-registerable.
Once that happens, browsers and other stakeholders could then consider allowing self-signed TLS certificates, maybe with TOFU semantics, for .lan hosts, ideally together with .local mDNS ones.
RFCs were originally formalization of what was being used in practice, home.arpa was chosen apparently for reasons of beauracratic convenience rather than what would best serve existing users, who all would prefer .lan and continue to do so since that was published in 2018
- Originally, IETF specified .home, but never went through the process to add it to the list of reserved names.
- Someone applied for the .home gTLD, and while it ultimately didn't go through there was a while while it was up in the air. Due to this the homenet working group had to change gTLDs.
- They decided to switch to .homenet. However, because of DNSSEC, whatever they used would need an insecure delegation in the root zone or validating resolvers wouldn't be able to resolve it. Since IAB controls .arpa but IANNA controls the root and there was no process to ask IANNA for this, they eventually decided to use home.arpa instead.
> RFCs were originally formalization of what was being used in practice
You may be confusing it with the IETF’s policy.
In early days of the RFCs, most started out as proposals, often but not always with some existing implementation as a jumping-off point to conversation (hence the name). I no longer remember why they started to be numbered and tracked, as the process naturally preceded that.
You can verify what I say by just reading some old ones at the rfc editor site.
- WTF is arpa? (in reality, I know, but there are better names than an obscure reference to internet prehistory)
- it is a second-level domain, why?
- 9 characters (home.arpa) is a lot
I kind of understand the motivation, it is a "technical" domain since it doesn't represent something in the global DNS registry, so it gets the .arpa TLD. However, that's reasoning that it out of touch from normal users. Normal users don't want to be exposed to the technicalities of DNS when they enter something in the address bar, and "myserver.lan" is more meaningful than "myserver.home.arpa", and giving meaningful names is the whole point of DNS.
I'm on that side that it was not a good idea right from the beginning.
Because this problem exists in small enterprises since ages and the workaround always was to use a domain you own.
At home I use home.arpa since 10 or 8 years or something.
Your DHCP/SLAAC RA tells clients "here is a list of search domains, when you attempt to query a bare name, also look for the name plus any suffixes listed in the search domain list". So on a system with "home.arpa" in the search domains, you can do something like 'ping myhost' and your system should attempt to resolve 'myhost.home.arpa'.
Next time may narrow down the search space and look for those keywords combinend.
The DHCP spec includes the search domain so it's not that hard to find and follow.
Of course without any context this search would be not reveal anything useful.
Ubiquiti aren't the only ones. OpenWrt (and, by extension, GL.iNet) explicitly refuses to forward .lan DNS queries to public resolvers. This is a good thing: "LAN" has a singular, unambiguous meaning that directly contradicts its use as a public TLD. The fact that we're even debating this is outrageous.
But why is it that nowadays, it's enough for a vulture corporation to have enough money to be able to privatize a part of the "web" that is in common usage (in the sense of common good) since so long?
It should be logical even without thinking that the request have to be rejected.
If you actually think it's a trivial affair to change a well-established default with more than 20 years of history that is, on top of all other difficulties that such a change typically encompasses, used to identify and name things, I hereby beg you to never design or provide any kind of infrastructure.
Looking at your links, subjective and legacy considerations aside, .alt per RFC 9476[1] seems as good a choice as .lan.
Personally, I just use a registered domain for this purpose, as they're cheap enough to not care (last I checked, I pay $10–20/domain/year for registrations, depending on registrar and TLD, and $0.21/zone/month for hosting public zones on Google Cloud).
A registered domain has the added benefit that you can use the same namespace for externally accessible services. I personally do this (using Cloudflare Tunnels, not open ports).
This reminds me of when I used to do IT work for small businesses in college. One printing company I worked for, had about 100 computers on their network, and was using public ipv4 addresses, that they did not own, on their internal network. I forget what range they were using now. But imagine seeing a DHCP server handing out addresses like 142.250.110.1/16 on a LAN and the public ip being something totally different.
It was funny, because when I brought it up to them, it was hard to articulate why it was a problem and I couldn't convince them it was worth the effort of trying to fix. They never ran into a specific issue due to this while I was there but it felt so gross.
It's always hard because when you contrive possible examples of how it goes wrong, every single example sounds contrived because of course they are contrived.
Sure one day your printer might start spewing random json code meant for some microservice of the rightful IP owner.
Sure the IP's might be owned by the Air Force and one day they might start getting traffic from your pos ipad that they decide looks like an attempt to attack one of their internal secret networks...
Sure one day traffic meant to go to your printer ends up flooding and dossing a windmill controller, preventing the rightful operators from turning it the right direction during bad weather and causing $25M damage...
And of course the real failures are more like, only people from the Maldives can't send email to your email server, a failure with no impact.
The problem is that they wouldn't be able to talk to any internet service that legitimately used those addresses. Whether that was likely to be a problem very much depends on whose addresses they were.
There's a good reason to do this if you can't be certain what reserved subnets are used in a given network and you absolutely need a static ip for some reason.
At least that's the conclusion I've arrived at at some point, but I don't remember what was the exact use case anymore.
However there are still several global IPv4 ranges that are not local reserved ranges but are effectively reserved and you could use them if you really want to without any issues.
I used to do consulting for companies that weren't connected to the Internet yet. This was the early-mid-1990s and "Private IP Network" assignments literally did not exist until 1996.
Therefore, people who weren't on the Internet yet, or didn't expect to be connected to any particular WAN, or even had their own intranet WAN without public access, they improvised like crazy. Like, whatever IP address ranges worked for them, they just plugged 'em in.
Since it was indeed early, and IPv4 networks were being handed out like candy, the savvy companies just applied for their own network space, hopefully of an appropriate size. The applications at the time didn't specify or require that they route the traffic or put it into BGP, so registering your intranet with IANA was the right thing to do, and future-proof. Unfortunately, in the wild-west early days before CIDR, IANA needed to find either a chunk of class C, or class B without significant nuance. And that is how Apple, HP, and Ford came to control gigantic swaths of "Class A" space.
Renumbering private networks was one of the nightmare scenarios of consulting firms like mine. And it hasn't really stopped, I suppose. If your company does an M/A their private LANs may clash, and they either renumber or they NAT.
By the way, OpenWRT isn't supposed to be using ".lan". That's a ridiculous non-compliant behavior. Someone should've stopped that, long ago.
There are ways to subdomain properly and IETF has provided dedicated TLDs for it even now, but back in the day, it was not easy, and I get it. When I needed DNS for my homelab, I discovered that the best practice among bad practices was to subdomain your ISP's. Like, make up a fictitious subdomain and hang it off their real one. Then your DNS had an anchor in reality, even if it was all made up, never delegated, unreachable, unresolvable from the outside.
When I was working with Linux based kiosks and PoS systems I had a good share of issues with the Microsoft MVP signature use of .local on their forests. Back then their training material recommended .local for Active Directory services.
Specifically, either a registered name or a subdomain of a registered name reserved for this purpose, because the A records for the apex should or must — I don't remember which, but it's definitely the default configuration — point to the domain controllers, and you probably don't want to host your public web site on the same addresses as your DCs (or for that matter, your internal and public records necessarily hosted on the same server).
Historically (20+ years ago), some Microsoft documentation suggested .local as an example of an unregistered domain that could be used for this purpose, which was problematic when .local was subsequently reserved for multicast DNS.
Meanwhile, I watched a very prominent Microsoft consultancy shop install a client's ActiveDirectory as <nameoforganization>.ad. You know, the country of Andorra? That has an actual real ccTLD?
Given that there don't appear to be any BGP announcements for 25.anything, is it possible that the MoD uses this address range internally, or not at all, and has agreed to continue to do so?
Given that the plain text meaning of .lan is even more unambiguous than .home, it's overwhelmingly likely that .lan will suffer the same fate, where delegation was "indefinitely deferred".
Regarding that last point, I've had issues with dnsmasq in the past where it was remotely resolving domains even when they were configured to resolve locally. For .lan domains, this could be disastrous because I often send plaintext traffic to them. I did submit a fix/workaround[0], but it's still something to look out for. If anyone knows more about this issue or other ways queries could leak, please share!
[0]: https://github.com/openwrt/openwrt/pull/18610