Hacker News new | past | comments | ask | show | jobs | submit login
Proposed top-level domain string for private use: ".internal" (icann.org)
323 points by zacwest 11 months ago | hide | past | favorite | 229 comments



This makes me think of some AD best practices I read a long time ago. One of the practices was to never use made up local TLDs like .internal or .local because some day they might be real and get picked up by someone.

Essentially you should always use a domain you control both outside and inside, like a regular gTLD or ccTLD.

Pretty much every single company I've worked for with AD has broken this rule.


> Essentially you should always use a domain you control both outside and inside, like a regular gTLD or ccTLD.

Yep. We use two domains - everything on A is "public" and everything on "B" is internal. The root of "B" is a static page on AWS that is gated behind our VPN so it serves as a quick "can you hit our infra at all" check when troubleshooting, and catches the people who are convinced they've enabled their VPN.

One thing to note here that we haven't solved (We've only got two "infra" people, we're a small company) is how you handle an internal portal to an external service. If our public domain is maccard.com and private is maccard.dev, where does internal-admin.maccard.com _actually_ live? Our solve for this is we have an internal.maccard.com for this very specific use case, but I'd much rather it was admin.live.maccard.dev


We just use the same domain internally and externally with split DNS. Works fine.


We have enough issues with DNS that adding split DNS into the mix is a ball ache I don't want to contend with.

We actually have the DNS for our private domain set publicly, and all the actual work happens on a load balancer which is on the network. We're fully remote so this avoids the "my communal WiFi provider seems to have issues with the VPN" (which is what we had when we used split DNS)


I buried split DNS (and (for the most part) private CAs for that matter) with ACME DNS-01.


We use split DNS and the admins can't even do it right, they keep fucking it up and configuring one DNS view but not the other, so when I'm on VPN I randomly can't use certain domain names.

Also as another commenter mentioned, it is impossible to tell based on the name if it is an internal or external resource

I'm curious what split DNS offers that a separate internal zone wouldn't.


Until it doesn't.

And having a website on the domain.tld adds shenanigans.

One of many examples I had is when Outlook loses connection to Exchange (eg S2S VPN is down) it starts autodiscovery process, hits domain.tld (because users have email@domain.tld, duh) and complains to user with a scary messages (which are also blocks the process until the users hit something). Which is totally understandable, because the website is on some public hosting, so CN in the cert is from the public host at best and != domain.tld.

Using corp.domain.tld or even techdomain.tld solves this totally and also let you use public certs (LE in the current era) even on the 'local' side of the network.


Aside from all technical issues the biggest problem I have seen with such an approach is that is really hard for employees to remember what is external and what internal that way. Distinct domains help there


https://en.wikipedia.org/wiki/2021_Facebook_outage

made me think this wasnt such a great idea. particularly the part about facebook employees not being able to use their keycards to enter the buildings at the same time as the site outage.


The ".local" domain specifically is a bad choice as many platforms use MDNS instead of DNS for looking up those names. Leading to issues resolving names on some client devices. It's also very common due to Microsoft suggesting it as best practise in the early days of AD.


I'm wondering about one thing, now that I've read a few "may cause issues" and "is used for mDNS" replies: what the F is mDNS actually doing in the background?

Is it really going to assign "lancelot.roundtable.local" to my washing machine on a whim, which leaves the microwave unresolvable?

Can't I instruct the mDNS server running on my machine to respond to a particular name ending in .local?

Can't eg. dnsmasq insert itself into a conversation on 224.0.0.251 saying "Let me answer this question" for certain queries?


I'm a little fuzzy on this, but my understanding is that for mDNS to be reliable it is required that all .local hosts implement mDNS to allow for conflict resolution.


If you’ve set up x.local in your DNS for your dryer but your laptop uses mDNS, it’s possible that your lapatop’s mDNS will get a response from your microwave that it’s reachable at x.local. The solution (not an expert, please check this) is to set up the dryer in DNS as x.domain-thar-you-own or x.home.arpa


Thank Apple for this half-baked bullsh*t.


No. Fuck MS for encouraging their MVPs to use a reserved domain.


How so?


Maybe because of this?

> mDNS implementations

> RFC 6762 was authored by Apple Inc. employees Stuart Cheshire and Marc Krochmal, and Apple's Bonjour zeroconf networking software implements mDNS. That service will automatically resolve the private IP addresses of link-local Macintosh computers running MacOS and mobile devices running iOS if .local is appended to their hostnames. In addition, Bonjour devices will use those .local hostnames when advertising services to DNS Service Discovery clients.

> Most Linux distributions also incorporate and are configured to use zeroconf.

> ..The connection of Macintosh and Linux computers or zeroconf peripherals to Windows networks can be problematic if those networks include name servers that use .local as a search domain for internal devices.

https://en.wikipedia.org/wiki/.local


Kinda weird to blame the people who made an RFC, instead of the industry leader who recommended using .local completely on their own, without support from the wider industry. This is explained in the next couple paragraphs, where you stopped copying.


You're right, the confusion about the use of .local domain seems to be more due to Microsoft going back-and-forth about it.

> At one time, Microsoft at least suggested the use of .local as a pseudo-TLD for small private networks with internal DNS servers.

> ..However, more recent articles have cautioned or advised against such use of the .local TLD.

> Microsoft TechNet article 708159[7] suggested .local

> ..but later recommended against it.

> The Microsoft Learn article "Selecting the Forest Root Domain"[8] cautioned against using .local

> By default, a freshly installed Windows Server 2016 Essentials also adds .local as the default dns-prefix when a user doesn't select the advanced option, resulting in a domain with .local extension.


mDNS was announced after the Windows 2000 RTM date, so it would be unfair to blame Microsoft with regards to mDNS functionality.

Not that Microsoft should have made that recommendation, but hindsight and all that.


I understand that .home.arpa is the standard TLD for home networks as per RFC-8375

https://www.rfc-editor.org/rfc/rfc8375


I tried to use that on my Turris Omnia and it broke a lot of things. I ended up using .lan


I never heard of home.arpa so I'm also curious what broke.

The only thing I can imagine would be if you accidentally created a root for .arpa.


What broke?


Yeah, but it's basically a domain as opposed to a TLD and it's hugely impractical to work with. We need a short TLD like .lan for that!


The following TLDs are fine and reserved:

                .test
                .example
                .invalid
              .localhost
RFC2606 (https://datatracker.ietf.org/doc/html/rfc2606)


Test is best used in temporary setups that you use for testing things.

Example is best used in documentation only.

Invalid is weird and confusing.

Localhost as a TLD should still be on the machine itself. Keeping in mind that 127.0.0.1 is not the only loopback address at your disposal – you have the whole /8. You could bind different services on different loopback ip addresses and then assign host names in the .localhost tld to those.

So for example you could run different Wordpress instances locally on your machine on ips 127.87.80.2, 127.87.80.3, and 127.87.80.4, so that each can run on port 80 without colliding with one another, and without resorting to having non-80/non-443 ports.

Then have corresponding entries in /etc/hosts

127.87.80.2 dogblog.localhost

127.87.80.3 cooking.localhost

127.87.80.4 travel.localhost

And use those domains to access each of those from your browser. Then you don’t even need to keep all services behind the same Nginx instance for example, as you otherwise would do if you had different domain names but was using 127.0.0.1 and port 80 for all of them.

Whereas having the localhost tld refer to hosts elsewhere on a physical network.. that’s about equally as weird and confusing as “invalid”.


> Keeping in mind that 127.0.0.1 is not the only loopback address at your disposal – you have the whole /8.

Tell that to Google Chrome developers, who are so arrogant they think they know better than the operating system, and force-resolve *.localhost. to 127.0.0.1


‘The ".localhost" TLD has traditionally been statically defined in host DNS implementations as having an A record pointing to the loop back IP address and is reserved for such use. Any other use would conflict with widely deployed code which assumes this use.’

Reading that sentence implies a single loop back address. I appreciate that IPv6 “fixed” this confusion by having a single loop back address of ::1. While you are correct that IPv4 reserved the entire /8 in practice the expectation is generally that 127.0.0.1 is the loopback address, and using other addresses in the 127/8 space tends to lead to unexpected issues across many pieces of software. The intention of the .localhost domain is to ensure that DNS resolutions never resolves to something external, for security reasons.

See also, RFC 5735: “… This is ordinarily implemented using only 127.0.0.1/32 for loopback.” — https://datatracker.ietf.org/doc/html/rfc5735


Then use a non-brain dead browser :)


Looks like someone might have been trying to prepare the ground for a hijack of 127.0.0.0/8...


Yes, but who would want .test, .example or .invalid as an internal domain? Also, they are too long (yes, that matters).

What I've seen lately is '.int' for internal usage. While this is a valid TLD, it is only for international organizations, and it is not possible for "normal" people to reserve a domain with that TLD, so unless your company is called "WHO" or similar, you shouldn't have any problems...


People thought .dev was safe until it wasn't. Bear in mind they can still do something like "enroll all of .int in HSTS preload" (like was done for .dev), and suddenly your browsers will permanently refuse to load any of your internal sites.


No one competent would ever have thought dev. was safe. It's quite simple: if you don't own the domain, or it hasn't been reserved (e.g., home.arpa.), don't use it!


.Corp exist and is reserved for private use.


They are reserved for Testing, & Documentation Examples.

You should not used them in production.


The only one of those appropriate for accessing actual hosts would be .test, and obviously using .test for non-testing purposes would also not be appropriate.


.domain, .lan, .home, .host, and .Corp

Are all options as well.


I worked for Merrill Lynch around 2000 and they used other people’s public IP addresses on their (theoretically) internal network.

Naturally, this went terribly as soon any of their sites had their own connection to the internet.


I had a very large client that was squatting someone else's address space in the pre-WWW days. When I pointed out the problem this would eventually cause and that RFC1918 was the (then) answer, the CTO said "no one here needs to get to the University of Tokyo". I was long gone by the time the SHTF.

I had another much smaller client who's leadership insisted that because they were using it the owners would have to just have to not use it. Some sort of imagined IP squatters rights, I guess. I doubt the DoD accommodated them.


Why not use the designated private IP ranges? There are more than enough addresses there, unless it's some crazy application with millions of instances each needing an IP. But then use IPv6?


I’ve been told that this is still a problem at large organizations in that industry.


> made up local TLDs like .internal or .local

.local actually exists. It is used for mDNS.


TIL, woops!

I use this for all my internal domains, haha. I guess this is why we need a proper tld like this.


You could consider making them real mDNS domains then, and using mDNS to resolve them to your desired IPs.


The following TLDs are fine and reserved:

                .test
                .example
                .invalid
              .localhost
RFC2606 (https://datatracker.ietf.org/doc/html/rfc2606)



I saw a successful attack on a .dev domain doing exactly this. Links on PCs worked correctly, but phones showed a scam site, so emailed links were attacked.

It was hard to fix because they couldn’t get the spoofed domain, and there were so many copies of bad links everywhere.


Another fun fact about .dev which I recently found out while working on a side project:

You NEED to use https when visiting any .dev domain. Google has put it on the HSTS preload list.

It took me a while to find out why my browser kept redirecting me to https when I wanted to use http (local development). Curl worked fine…


> Google has put it on the HSTS preload list.

Just for the record, anybody can add their own domain to HSTS via the submission site[1].

[1] https://hstspreload.org/


Pretty much the first I do on any fresh server is to disable http, no matter what it is used for.

IMHO http has no use in the modern world.


My work used .local which means mDNS and service discovery etc doesn't work. Very annoying. What are they teaching these network admins? Why don't they even know what a domain name is for?


Similar issue with modern devs that don’t know what a page of memory is.


Or don't even know the bandgap of the transistors the computer is running off.


Yeah, developers nowadays don’t know their Pauli exclusion principle.


or a database.


Microsoft's recommendation in the early 2000's was to use .local


Even if they know now, that doesn’t mean they knew when setting up the AD domain back in 2000.


Thank Apple for designing the protocol the Apple way, that is being not interoperable with the world.


Yes, the classic not-interoperable approach of publishing the protocol as Standards Track RFCs with the IETF.


The timeline was IIRC

1. MS made recommendations to use .local

2. Apple networks started to use Bonjour and used .local, breaking on networks that used .local already

3. RFC was submitted

4. MS recommended against using .local or any domain you do not own fully


This happened with the .dev situation. I switched to .d ever since. I’ve read somewhere that new TLDs are required to be at least 2 letters so I guess I’m safe this way.


You can use .lan, . domain, .home, .host or .Corp as they are all reserved TLDs


they aren't reserved that I can see? https://www.iana.org/assignments/special-use-domain-names/sp...

Where did you see that? would be interesting to have reference for a couple people


.d is easier to type.


Devil's advocate: using an Internet-public domain for internal purposes will publicly expose your internal hostnames if you enable DNSSEC on the Internet-public domain. This is a problem if you're required to enable DNSSEC, e.g. for FedRAMP compliance.

The number of cases where this is actually a legitimate concern, IMO, is extremely small, and I'm personally of the opinion that using Internet-public domains for internal purposes is generally fine. But it's still important to point out that the number of cases is not zero.


You can use a public domain but a local/private DNS server


I believe the advice about using a TLD you control outside and inside is mostly to prevent takeover on the outside that could affect the inside.

But you can still have completely separate DNS for the inside. Using a shared DB for both would probably be recommended to avoid conflicts.


Every security auditor in ever regs regime will flag zone transfers.


> expose your internal hostnames if you enable DNSSEC

Zone content enumeration in DNSsec was fixed by NSEC3 records (RFC5155, March 2008)


No it wasn't! NSEC3 is crackable the same way a 1990s Unix password file is. This was such a big problem that two competing approaches were introduced to defeat it: "whitelies", which I perceive as the "best practices standard" answer, requires servers to operate as online signers (they should have been all along) so they can generate dynamic chaff records to foil enumeration, and NSEC5.


Hmm. Interesting. I wasn't aware of that, thanks! … also, https://github.com/CyberCX-STA/NSEC-3-Walker


>Pretty much every single company I've worked for with AD has broken this rule.

It's a lot better now. Ever since companies started moving from on-prem Exchange to O365 in droves I've noticed that most orgs I work with (painfully) updated their domain so their user principals align w/ their O365 mailbox.

There's only one customer I have that still uses a ".local" domain for AD, and they got bought out last year. (By an org that uses a real FQDN.)


I joke that I would use .icannsux because there is just no way they would assign it.


`.sucks` is a real TLD though, so you can register `icann.sucks` today!


https://www.hover.com/domains/results?utf8=%E2%9C%93&q=icann... alleges it is $219,999.99 and renews at $2599.99/yr due to "This is a tiered price domain."

Also, yeah, they almost had me with those .99 figures :-/


Hah that's a funny idea that you'd be relatively safer if you used something outrageous like godsavethequeen or crappymcfartlegs. But you never know what promotional TLDs are created in the future. Or how free the process might become, as remembering names becomes less and less important to a majority of internet users.


We had a bunch of .dev urls break thanks to google and our lack of foresight.


We used .dev before google registered it so we migrated to another paid domain so we don't need to pay huge amount per year to google


To this day this story still blows my mind:

>Register a TLD which is informal standard for development websites and environments (.dev)

>Charge an excessive amount of money for it

>To make sure you ruin everyone's day, put it on a HSTS preload list

>Refuse to elaborate


oh. its worse. the original request for the TLD was only granted because in the application Google specifically mentioned that it should be reserved due to its unofficial use by developers and that if anyone else got it then they might put real domains on it.

You cant make it up.


>Sell your domain business and force your customers to migrate.


... to a provider that doesn't even offer all the same services.


I specifically use some .dev domains because of HSTS. Some of us don't cling to http and I prefer an error rather than transparent fallback to unsafe protocol if I screw up the config.

The first point is valid but that is mostly ICANN's fault, they should have proposed it as reserved instead of selling it.


> informal standard

FAFO, as they say.


Just my person experience, but I got my relatively uncommon name on .dev for $12/yr so I'm pretty happy with that. While the situation worked in my favor, I agree .dev probably should have been the official internal only TLD.


The last time I set up DNS at home, I decided to use a fictitious and undelegated subdomain under my ISP's domain name. This structure did not create any "extra" problems in the short term.

But I suppose that I still ran the risk that the subdomain could "become real", or draw attention from security admins, or I would change ISPs.


> draw attention from security admins

How? I’m not saying this is great practice — there are certainly better options — but no one outside your network will ever know about it. It also won’t matter if you switch ISPs.


It's not a problem (well, most of the time), but you would see the requests for 'internal' resources in DNS (ie your machine is not on your network but tries to resolve the internal DNS records) and in certificate checks even for non-public PKIs


I've sold IT stuff to a city that had some expensive consultants configure their ActiveDirectory as <name_of_city>.ad, as in the ccTLD of Andorra.


The .local bullshit gate so much headache. When I was working on Ubuntu based POS systems we had to make sure avahi or any kind of mDNS wouldn’t be installed. It became a running joke “we got another MS MVP here” every time we had issues with a client with a .local domain.


My friends and I own the domain name “dev.host” and you wouldn’t believe the traffic we get.


AD?

[edit] Oh, Active Directory


.corp, .home and .mail should also be perfectly viable for private use after ICANN eventualy decided to cease all processing of applications for those TLDs.

    "Whereas, on 30 July 2014, the ICANN Board New gTLD Program Committee adopted the Name Collision Management Framework. In the Framework, .CORP, .HOME, and .MAIL were noted as high-risk strings whose delegation should be deferred indefinitely".[1]
[1] https://www.icann.org/en/board-activities-and-meetings/mater...


Deferred indefinitely is not quite the same thing as reserved for private use. For home use cases it's probably good enough, but a corporation will want more assurance than "the current ICANN has stopped processing applications for this TLD."


Yeah, with Google getting .dev we can't count on ICANN being reasonable.


Actually, if history is any indicator, we can’t trust any of these companies with this. Remember the org tld private equity fiasco? ICANN only did something after public outcry and EFF interest.


ICANN persisted with their antics until the California attorney general announced an inquiry into the deal. Only then did ICANN change their plans.


I trust ICANN to be the opposite of reasonable if they can make money off of it.


Another possibility is `.zz`, which technically can be a ccTLD but it's a user-assigned ISO 3166-1 alpha-2 code, and its last position makes extremely impossible for it to be repurposed as a valid code even in that setting. In comparison, some user-assigned codes like `XZ` are often used for temporary country codes so `.xz` would be less appropriate.

It seems that ICANN did consider this choice among others, but reject for the lack of meaningfulness:

> The qualitative assessment on the latter properties was performed in the six United Nations languages (Arabic, Chinese, English, French, Russian and Spanish). [...] Many candidate strings were deemed unsuitable due to their lack of meaningfulness. [...] In this evaluation, only two candidates emerged as broadly meeting the assessment criteria across the assessed languages. These were “INTERNAL” and “PRIVATE”. Some weaknesses were identified for both of these candidates. [...]

I wonder if this means that they only scored the highest among others and all candidate strings were indeed unsuitable, but that they had to pick one anyway. I'm not even sure that laypersons can relate `.internal` with the stuff for "internal" uses.


I opposite to use the `.zz`. Instead, I think this should be assigned as new TLD domain for all people live in any country, because it is a reserved ISO 3166-1 alpha-2 code, and will not be assigned to any country. It's the best convenient TLD for human on the earth.


There is not much to gain from having a new two-letter gTLD instead of existing all-purpose three-letter TLDs (say, `.com`, `.ooo` or `.xyz`). Especially given that every TLD needs a registry, and `.zz` would be a particularly extra-special gTLD that needs a special procedure to select the registry. Better to make it a reserved TLD if it is ever going to be used.


I consider .xyz to be a good TLD for that purpose already, but .zz would be nice too.


Also worth noting that "home.arpa" is already reserved and specifically designed for residential use. It will never conflict with anything.


And is nine characters long, which is cumbersome for those using it in an IOT context where data entry is often not straightforward.


so, exactly the same as .internal


You didn't quite count that right with the dots ;)


Huh?


Typing it on your TV with a laggy UI where you can only select letters using arrow keys. Which only work half the time, and also it isn’t QWERTY.


It's so sad this is still necessary. We've had service discovery via avahi/zeroconf for years. Why does it seem like uPNP has been and gone? It shouldn't be necessary to type hostnames any more.


I use it and it’s very confusing for my guests.


I use this, mostly for the .arpa vibes.


I avoid it for the same reason. The ARPANET branch of the US Department of Defense feels anything but home to me.


Would it relax you to know that it has been retconned into "Address Routing Paramater Area"?


After reading through the threads, I still think that '.lan' is a better non-reserved suffix to use for this than '.internal'; however, my opinion rarely has significant weight in the grand scheme of things.


And when you're using it to connect to devices not physically on your LAN, like servers reached via tailscale? `.internal` implies your internal network, however it may be setup, `.lan` implies your "local" area network.


Do you have a reason why you think '.lan' is better then '.internal' besides being shorter? I am curious why.


It is already in use many places.

Isn't it the default domain name for OpenWRT? Just like the default ip addresses are 192.168.*


lan = LAN or Local Area Network


It would be nice to see this paired with more widespread support for the Name Constraints TLS extension, which would in theory allow internal CAs to be restricted to issuing certificates for .internal domains. That would open up a lot of very interesting applications in terms of streamlining HTTPS on local networks, for example, ACME on openWRT routers.


Absolutely. We're getting closer, but it's hard to measure what actually supports it as bettertls (the caniuse equivalent in this space) doesn't track it.

https://github.com/Netflix/bettertls/issues/19


Note there was a W3C community group that explored the "HTTPS on local networks" question. It's closed now but they wrote use cases and proposals:

https://www.w3.org/community/httpslocal/ https://httpslocal.github.io/proposals/ https://httpslocal.github.io/usecases/

I wish some of this work would continue as well.


As others have mentioned there already is the ".home.arpa" TLD but I definitely think ".internal" is a step up in terms of clarity. That said, for my internal network I just put things under a subdomain of a domain I own so I can use HTTPS with a proper SSL cert


> I just put things under a subdomain of a domain I own

Yup, same here. Great in combination with ACME DNS-01 so your DNS server can request all those certificates and then push them out to your devices. (Otherwise the hostnames need to be externally accessible, which means either exposing the internal devices, or mucking around with split-view DNS. The former is a terrible idea, the latter is also DNS server complexity and worse than doing DNS-01 IMHO.)


IMHO if you are already doing some process of "push certificates out to devices," you'll likely be much happier with getting a wildcard cert using DNS-01 and change that update process from "all devices all the time according to their schedule" over to "all devices but once every 80 days"

I do appreciate the threat model of one device getting owned leaks all your certs but security is always a trade-off between security and convenience. It also lowers the load upon the LE servers, for what that's worth


Not sure everything updating at the same time is more "convenient" than staggered failures. For one, if multiple things break at the same time, it's easier to lock yourself out of things in more complicated ways. Also it's generally the first refresh that breaks, and everything at once only helps when you freshly roll out certs to a whole bunch of devices… if you add things incrementally (e.g. either because you finally get around to it, or you just bought something new) it makes no difference if it's all in the same cycle. Except now you have a wildcard cert floating around…


That seems like a great way to go - get it signed the normal way and TLS will 'just work', no messing about configuring trust on your devices.

Will this be possible with .internal ?


The TLS cert will either be self-signed or you'll need to run a private CA. A public CA won't issue you a cert as you can't own any .internal domain.


You can use a proper certificate with any domain you like.


except not with .home.arpa .internal .lan or whatever else, since you don't have "domain ownership"


I think I would have preferred .intra (Unless it's already used somehow ?).

Just 5 letters is less annoying to type repeatedly than .internal, while still conveying the overall purpose relatively well.

It might just be my laziness talking though.


Counterpoint: .internal is much easier to understrand, .intra sounds like a reference to "intranet" as opposed to "internet". That's terminology networking people would use, but .internal is likely something both non-tech and tech people would intuitively understand.


I don't think `.internal` to be intuitively understandable either. In spite of other shortcomings, `.home` and `.corp` convey its private-use natures much better than `.internal`. (But they won't be suitable for anything other than that.)


You are not alone. Typing matters, 'internal' is just too long. I'd much prefer if they for instance would reserve '.lan' for internal usage.


Isn’t it possible to setup a default search domain. So most places you could just type something like https://site/ and your dns config knows to look for https://site.internal/


so, for example, an entirely internal zero trust network across all locations, should be .lan and not .internal?

.lan is so "back to office"


zero trust means no distinction between LAN and WAN, what do you mean?


I liked this idea at first, but the more I think about it the less I'm sure it works. Intra seems like a word which denotes physical boundaries to be within. It prefixes words like intramural, intramuscular, and intracellular. It's used for intranet as well, and while this usage of the prefix works, it has more broad use-cases than that. On the other hand, internal seems very clear and direct, and might suit a more digital, organizational context with less of a physical or spatial aspect to it.

Also, since intra is typically used as a prefix, it seems strange to use it bare. People familiar with intranets will probably make (or assume) the connection, but others might find it unclear while "internal" would likely not be.

Even so, I do prefer .intra for selfish reasons.


ICANN: [Proposed Top-Level Domain String for Private Use](https://www.icann.org/en/public-comment/proceeding/proposed-...) "The Internet Assigned Numbers Authority (IANA) has made a provisional determination that “.INTERNAL” should be reserved for private-use and internal network applications(...)"

... possibly a better link.


Didn't .local start out this way until it was co-opted by Apple for some network abomination? Any new private domain is just going to get co-opted by something else soon enough. Browser authors and network service authors are going to start using it for random, incompatible purposes and break everything.

If you need DNS, register and use a real domain name. Everything else is going to be a hack. Anyone tech-savvy enough to know what an internal, unroutable TLD is, and have a use for one, is going to be just as comfortable and capable of managing a real domain.

I support the idea of something like .internal, but I'm certain it will be made useless for its intended purpose in short order.


Zeroconf is only considered an abomination by folks who view any naming system other than DNS to be an abomination.

Domain names are not slaved to DNS, they predate DNS by over a decade, and there is room for more than one naming system.

If DNS has to surrender one of their precious, precious, TLDs (lord knows there are soooo few to choose from) which, again, existed BEFORE DNS was even a twinkle in Paul Mockapetris' eye, so be it.


I can see one benefit to this approach.

Messing up your .internal configuration won’t result in leaking queries to the public.

And maybe secondarily this may encourage tool development for supporting internal names and make it easier for setting up informal or per department configurations.


I'd revently ran into this, after using .local for a long time, and installing something with mdns. Nslookup gave the correct ip, but ping got confused.

A quick google did not deliver a decent reserved domain, but multiple people suggested .home


I use .localnet to go with the name of localhost, as this has been suggested by ... one of the RFCs, but I can't remember which.

If .localnet ever becomes a real TLD, well, I'm pretty sure the entire global infra is going to collapse and not necessarily be my problem.

Edit: And to be clear, I'm doing this for my house, not some enterprise setup; using real actual FQDN for internal services at a company, especially one that is multi-site/cloud, is still the best advice.


I’ve seen localhost.localdomain often but I’m not sure whether it’s actually reserved.


At last (or at least soon) I can stop using the special reserved example.com for things. :p

https://en.wikipedia.org/wiki/Example.com


I use a subdomain of “int” for all internal hosts: host.int.example.com. My internal machines have int.example.com and example.com in the search path.


This is exactly how a committee would design it if none of the participants had actually used an internal domain.

For example, in Google, https://go/foo had "go" as technically a TLD, and the memorable suffix that followed was already part of the path and not the domain name. It made it easy to type or include anywhere, including chats, posters, presentation slides, etc.

If they were to follow this proposal instead, you'd be typing or including https://go.internal/foo , which while more explicit largely defeats the point of the short URL.



Are you sure go wasn't just the hostname?

This was very common in Windows shops back in the NT and even post-NT4 days to leverage the hostname as the URL (http://exchange, http://sharepoint, and so on).


I'm pretty sure it was go.corp.google.com, with corp.google.com configured as one of the search domains.


I think it was both, because I think it also worked even when it wasn't a search domain. Can anyone still at Google check?


I think imternal, .lan, .inside should all be reserved and established to never allow registration with OCANN.


This may generate confusion with the .int gTLD used for international organizations like UN agencies.


If you have a customer and DNS seems to work strangely, have a look if you don't have a so called single-label domain for Active Directory, such as host1.internal or crusty.internal if you have more creative admins or even worse, db.local. On Windows, the name resolution on single-label domains behaves a bit differently, using NetBIOS resolution which can prevent you from e.g. adding a new host to the domain from a different subnet - you might see this as DNS failure. Of course, it is not DNS's fault, if it wasn't asked in the first place.

Here are some more details: https://support.microsoft.com/en-us/help/300684/deployment-a... and https://admx.help/?Category=Windows_10_2016&Policy=Microsoft... which does the DNS resolution even for these less than ideal domains.


The server authentication story is fairly weak. Multiple companies may use the same .internal domain name and none of them can get a TLS certificate from a public certificate authority. This means they'll each need to operate a private CA if they want to authenticate connections to the server (and encrypt with HTTPS). A major problem with this approach is that computers (especially laptops) travel between networks and can end up trusting more than one private CA. This means that you can have multiple servers using the same domain name, but operated by different orgs, and each appears valid to the end user. Session cookies and other data can leak when this happens.

I think the right solution is that we should require domain registration (google.internal, microsoft.internal, etc.) to avoid these conflicts. A public CA may be able to verify ownership, avoiding the need for private CAs.

I built a service [1] that does this and is compatible with Let's Encrypt. The trick is that I only allow users to set ACME-DNS01 TXT records, not A/AAAA/CNAME records. So you'll still need to run internal DNS for those.

[1] https://www.getlocalcert.net/


What's the benefit over just buying a regular domain for my internal stuff? The result would be about the same, and I wouldn't be dependent on you never going out of business.


Even with a TLD, going out of business is a concern. Take Freenom as an example, they had rampant abuse and ran into funding issues. People who previously used that service for free internal domains have been looking for a new home.

What's different is that the public suffixes I operate cannot publicly host content, which should protect the service from the abuse concerns that plagued Freenom and other free public suffixes. That reduces cost and should keep the site running.

I recommend buying your own domain if you don't mind the cost. A free solution for domains with TLS on internal networks is valuable to many.


since ".dev" was hijacked ive been using ".lan" and ".lab"


To be fair, .dev TLD seems to imply developer or development. .internal is a broader name and even if .dev was an option it probably wouldn't be selected.


I also use .lan


So sad that a big tech company stole the dot dev thanks to the ICANN greediness...



.example (for examples), .invalid (invalid domains), .localhost (avoids conflict with localhost hostname), .test (for tests), .local (link-local names), .onion (for Tor)


Also good to reserve localhost since some system resolvers will actually resolve any subdomain of localhost as 127.0.0.1. (I think systemd-resolved does, but I know for sure glibc NSS with the nss-myhostname module does.)


That's the speced behavior of localhost:

> The domain "localhost." and any names falling within ".localhost." are special in the following ways:

> 4. [---] Instead, caching DNS servers SHOULD, for all such address queries, generate an immediate positive response giving the IP loopback address

https://datatracker.ietf.org/doc/html/rfc6761#section-6.3


Couple things here:

The GP describes resolver software, which corresponds to item 6.3.3, not a caching server. This does specify the same behavior.

In RFC terminology, "MUST" > "SHOULD" > "MAY", so there is some wiggle room there.

6.3.2 permits Chromium and other apps to hardcode localhost names as such, instead of using a resolver.

A very popular vector for adware/malware is to take over the system resolver, or replace the DNS client configuration, so this is one reason Chromium jealously guards 127.0.0.0/8


I haven't figured out how to get unbound to do this. It's easy to configure dnsmasq to do it, though

    address=/localhost/127.0.0.1
    address=/localhost/::1
I do this on my network because not every system resolver does it itself (annoyingly).


I actually did not know this was specified by a standard. That does indeed make it perplexing that not all resolvers do it.


what about .localdomain? I haven't been able to find anything standards-wise about it.



That's too long. I just bastardize an existing tld on local network like home.net. Some browsers don't even allow made up names. Internal is too long to type.


Stumbled across this thread from 2020 - https://news.ycombinator.com/item?id=24606723


So as others have mentioned there already is the ".home.arpa" TLD;

Would the only difference then be the name ".internal" or is there another difference/advantage versus ".home.arpa"?


Eh, .internal is fine, but it's 8 characters to type. I'll probably keep using .lan until someone else takes it.


You still can provide your input on the matter till march; so it's not too late (yet): https://www.icann.org/en/public-comment/proceeding/proposed-...


The problem is that .internal is the bureaucratically correct answer, and not all internal networks are LANs.

Maybe they can reserve both .interNAL and its convenient abbreviation, which just happens to be written backwards because it's DNS.


Answer is simple and straightforward - nothing like this is going to happen because short abbreviations and any potential short alternatives can be sold and sweet sweet money can be extracted


Ha, excellent


Submitted to ICANN; everybody keep yelling .interNAL until it happens!


Not all use cases for .internal involve a LAN though.


Wait, so all the '.io' domains are actually registered by people operating in the British Indian Ocean Territory?

Seriously, .lan is just a convention. For all I care they could come up with any three-letter thingy, as long as there's a mutual understanding that no global DNS will ever resolve it.


Per my other comment, we could define LAN to be the abbreviation for interNAL, because INT was already taken.


.local already does this


> .local already does this

Best to avoid it for that purpose:

1. "Using '.local' as a private top-level domain conflicts with Multicast DNS and may cause problems for users." https://www.rfc-editor.org/rfc/rfc6762#appendix-G

2. ".local has since been designated for use in link-local networking, in applications of multicast DNS (mDNS) and zero-configuration networking (zeroconf) so that DNS service may be established without local installations of conventional DNS infrastructure on local area networks." https://en.wikipedia.org/wiki/.local

3. "PSA: Don't use domain.local" https://old.reddit.com/r/sysadmin/comments/a9sfks/psa_dont_u...

4. "Why using .local as your domain name extension is a BAD idea!" https://community.veeam.com/blogs-and-podcasts-57/why-using-...


That's a good list of reasons, however it seems the biggest concern is if you run a dedicated DNS service on your network.

For a simple home network setup, as long as naming conflicts can be managed, it looks like mDNS is quite handy.

On a side note, I find .local to be best suited for the purpose, since from the language perspective it's easier on international users than .localhost

The newly proposed .internal comes close, but .local still looks more semantically flexible or maybe this is a cognitive bias of mine.


.local is used by mdns so your .local machines can conflict with discoverable devices and services. I guess they wanted something that wouldn't conflict?


Only to the same extent that .test and .example do, which is to say it does not.


I can't find it now, but wasn't there a story about microsoft using a 'dummy' url in some internal documents that later became real?


Yes:

Microsoft Buys Corp.com So Bad Guys Can’t https://krebsonsecurity.com/2020/04/microsoft-buys-corp-com-...


What about ssl? will that work with i.e. letsencrypt?


It shouldn't. There is no way to prove ownership of a domain, because everyone owns it. Both a genuine company and their attacker have the right to use the .internal TLD, so both should be granted a certificate. This makes it completely trivial for the attacker to MitM the company's TLS connections.

The only option to somewhat-securely run TLS would be to have the company run their own internal CA, and trust its root certificate on all internal clients.


I suspect you'd need to generate your own, unless they intend on allowing people to register them. It's hard to provide a SSL for the 100,000 different "tv.internal".


host.docker.internal


I just use i.slow.network as my internal domain. Most things support search domains to avoid having to type too much.


If it doesn't get accepted I'll use .arpanet because nobody's registering that


.lan thanks.


it seems to cover the whole RFC 1918 (IPv4) and 4193 (IPv6), so not only 192.168.0.0/16 like some media indicated.


.intranet ?


Why did this take so long?


"foo.int/ernal" lookalike attacks in 3… 2… 1…

(to be fair, you generally can't get an .int domain registered. "int is considered to have the strictest application policies of all TLDs, as it implies that the holder is a subject of international law.")

… now that I think about it, "foo.in/ternal" makes so much more sense …


But you could do that with any domain, what's special in this case?


> what's special in this case

There's an existing TLD that is a string prefix of the new TLD.

Apart from lookalike attacks I'm also wondering if this will do weird things while you type in an address, e.g. if you try to type "foo.internal" and pause for a second after "foo.int"… your application may run off and do lookups or even prefetches.


That happens pretty regularly anyway, e.g. `example.com` vs. `example.co/m` or `example.net` vs. `example.ne/t` (and both `.co` and `.ne` allow SLD registrations).


Please allow self-signed certificates for ".internal" by default


If all self-signed certs are allowed by default, then the crypto means nothing at all -- that's not good.


Crypto is/could be still fine, it will still make it harder for others to see the contents of the conversation that is going on between the two machines.

What you are losing is the modicum of confidence that a website is the real deal because another party (that browser and OS manufacturers trust) took the time to, at the very least, verify ownership of the domain.


Automatically trusting any cert raises the bar for attackers from passive snooping to MITM. On a home LAN/wifi, MITM is pretty much as likely as passive listening. At the very least you need TOFU (trust-on-first-use) for any kind of real attack prevention.


The reality is that most people have unencrypted communication with their home appliances because the certificate isn't trusted.


So rather no crypto at all?


Currently you can get encryption if you care to configure it, so some have crypto and some don't; but if all certs are trusted, then everybody gets compromised crypto, which isn't much better than no crypto, so it's IMHO a downgrade.


The problem is ownership and trust roots. If you don't uniquely own a domain name then you can't get a TLS cert for it from a public CA. Private CAS still work, but are challenging.

I've been exploring private-use-only domain registration at https://www.getlocalcert.net/, which is compatible with LetsEncrypt.


No, doing this properly isn’t hard.


From the article

“ICANN has picked the TLD string that it will recommend for safe use behind corporate firewalls on the basis that it will never, ever be delegated.

The string is .internal, and the choice is now open for public comment”

Saved you a click :)



Personally I use the HOSTS file instead of DNS.

Alternatively I use a map file loaded into the memory of a loopback-bound forward proxy. No DNS.

I also use loopback-bound authoritative DNS to a limited extent as it provides wildcards.

There are ways to avoid using DNS.

Most web developers do not understand DNS, or at least dislike it, and some get annoyed by the HOSTS file. Quite funny. But I'm not a developer. DNS is something I understand well enough, I like it, and, in addition, the HOSTS file is useful for me. But sometimes it's most useful for me to avoid DNS.


How do you synchronize hosts file between clients?


You can set up a server listening to port 53 that will return the corresponding entry from a "hosts file" if you query it for an A record. To avoid the file growing uncomfortably big, it can be split up amongst hierarchically arranged servers.


How would you call this hypothetical system? It’s almost like a distributed system for looking up names.


An RFC might have been submitted for this already, let me check!


While I have no need to lookup an address in local computer A's HOSTS file from local computer B as this is not how I use the HOSTS file, in the event that I did want computer B to look up addresses in computer A's HOSTS file, it seems there are many possible options. I could not even come close to listing all of them. (The discussion here is computers on a local network. Is there a need for a "distributed hierarchical system".)

pdns_recursor can serve /etc/hosts

https://docs.powerdns.com/recursor/settings.html

unbound can serve "local data"

https://unbound.docs.nlnetlabs.nl/en/latest/manpages/unbound...

dnsd from busybox can serve a HOSTS file

   (exec 2>/dev/null;while read a b c ;do echo "$b $a";done < /etc/hosts|busybox dnsd -c /dev/stdin -p1153 -i 10.21.66.4)
tinydns would work too

   echo . > data
   while read a b c;do echo =$b:$a:1;done < /etc/hosts >> data
   tinydns-data
   ROOT=. IP=10.21.66.4 GID=0 UID=0 tinydns
Or just transfer the file instead;

rsync/mrsync could be used to keep computer A and computer B's HOSTS files the same

Or ssh to transfer the HOSTS file from A to B.

   echo cat /etc/hosts |ssh -T computerB "cat > /etc/hosts"
Or use ssh to query computer A's HOSTS file from computer B

   echo getent hosts name|ssh -T computerB
Or some small httpd, e.g., darkhttpd

   httpd /etc --port 1153 --addr 10.21.66.4
   tnftp -4o'|grep name' http://10.21.66.4:1153/hosts   
It goes on and on.

But I'm not interested in using the HOSTS file in this way on the local network. I'm more interested in IP addresses than "domain names". I am not a fan of web browsers; I make HTTP requests from the shell command prompt and from shell scripts. For example, I like to create shortcuts for certain IP addresses so I do not have to type them, e.g., when using netcat. For me, the HOSTS file works perfectly for that purpose. I use this functionality every day.

Not every computer on my local network has the same ability to lookup names and IP addresses. Most have zero access to DNS data. No lookups. Some may only be able to lookup a few remote addresses. I might put those in the computer's HOSTS file.


There is a fear or hatred of /etc/hosts amongst web developers. A regurgitated origin story about DNS and a perpetuated myth about the HOSTS file, having to do with constantly dynamic IP addresses. But the truth is that the conditions of the internet have changed. As someone who uses static, stored DNS data, and as such possesses a large chunk of historical DNS data, I have proof that, for the websites I may visit, most IP addresses do not change frequently.

Domain names are overrated. Web marketing hype. For example, no one uses a domain name to log into their router. But no one at home is getting internet access without typing an IP address at least once to set up a router. If I want to type a short, memorable name instead of an IP number to reach a computer on the local network I can make an entry in /etc/hosts. Using computers that have no /etc/hosts and no control over DNS sucks. Let web developers use those computers.

How many times have I seen developers copy entire portions of RFC 1035 into their code as a "comment". Too many to count. They will always struggle to understand it.


Funny I do use a name to log into my router at home. That worked right away and the name is printed in large friendly letters on the router.


Why not tell us the name of the router so we can learn about something different. Would be interesting to see a router with a built-in DNS server. Most home routers I have seen require people to type 192.168.0.1, 10.0.0.1 or whatever to set it up or to change settings. If anyone wants to try to argue otherwise, I could provide links to countless PDFs of manuals online showing this step in the set up instructions.


The router I'm referring to is specific to a provider. Wouldn't be very useful. I've also observed Fritzbox routers using the fritz.box domain for internal networking out of the box.


For example,

https://assets.avm.de/files/docs/fritzbox/fritzbox-7590-ax-v...

"You can also enter the IP address of the FRITZ!Box or the emergency IP.

The preset IP address is: http://192.168.178.1

The emergency IP address is: http://169.254.1.1"

Typing "fritz.box" into a browser is useless unless one already has an internet connection.

If it's a remote DNS query then is that really "internal networking".

Looks like there's an ad for NFTs at fritz.box along with some links to Javascript files and nothing else. No content. I think I'd rather just use a local address.


Somehow you skipped step 2 that says: "Enter the address http://fritz.box" which is the expected way to connect to the device and that worked for me even before Internet is set up. What you quote is the alternative method.

It does look like the domain fritz.box is not owned by AVM, the manufacturer of the device. Apparently they didn't manage to register this domain once .box became available. So in the future they might want to use fritzbox.internal, if this proposal gets approved.


Ask yourself: Why would an alternative method be required.

Typing "fritz.box" into a browser without an internet connection will not accomplish anything. To even get an IP address for "fritz.box" there needs to be either (a) an appropriate entry in /etc/hosts assuming the browser does not ignore /etc/hosts, (b) a DNS server on a loopback address or (c) a DNS server on the local network, _and_ the DNS server needs to have the IP address for "fritz.box" (so the person would have to know the address already, before connecting to the internet) _and_ the browser has been configured to use that local DNS server.

To demonstrate, assuming there is not a DNS server listening on the loopback address 127.23.59.88, change the DNS settings in the operating system to 127.23.59.88. Then try typing "fritz.box" to set up the router.

The reason an "alternative" method is provided in the instructions is because without an internet connection, typing "fritz.box" will not accomplish anything except generating an error.

The reason the company suggests that people type "fritz.box" instead of an IP address is likely because the operator of "fritz.box" website is advertising NFTs for sale. Not to mention the data collection the operator of fritz.box and potentially their marketing partners will gain with respect to people who own these routers. For example, everytime someone is configuring their router and types "fritz.box" the operator of "fritz.box" website gets to know about it.

The "alternative" method is the most reliable method, and the most private one. If we review the manuals for thousands of routers, we learn it is the most common method.


Yea sure if people change their resolver away from the one provided by DHCP, they're on their own. But when they get their connection configured by the router, it works, because the router will act as resolver and resolve fritz.box to itself. As I said, worked fine for me.




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

Search: