Hacker News Reader

Comment by rsync | Hacker News Reader

"they seem to acknowledge that they just don't know of anything better in that case"

They don't know you can run your own DNS server ?

It's technically trivial (for this audience) and it's basically free (very low traffic and resource instance or droplet or VM ...).

View on HN

I disagree that it's technically trivial for the audience this article is written for. The suggestions on the page are all as simple as using a particular browser, search engine, or browser plugin.

Where would you point the upstream? Doesn't it just push the problem up one server?

If you run your own recursing DNS server, your "upstream" is ICANN and the various registries that run the top level domains.

suppose a local government blocks a global gambling or betting platform, will ICANN happily serve the IP one seeks to a end-user installed DNS server? What is then the legal perspective? Is ICANN then in contempt of local law?

Because of how recursing works[1], ICANN only serves the IP for the dns server responsible for the tld. I doubt that would run afoul of any laws because ICANN doesn't even know what the actual domain you're looking up is.

[1] https://en.wikipedia.org/wiki/File:Example_of_an_iterative_D...

> I doubt that would run afoul of any laws

Sure it does. If the law says that domain X is not accessible in country Y, then ICANN has to scrub domain X from the list of domains when queried from country Y. Whether country Y has the jurisdiction to demand ICANN comply with their laws is a different matter, but I believe that technically serving "blocked" domains is probably against the relevant laws.

>Sure it does. If the law says that domain X is not accessible in country Y, then ICANN has to scrub domain X from the list of domains when queried from country Y.

But thing is that ICANN doesn't have a list of domains. All it has is a list of tlds and the dns servers for them. So if www.example.com. was blocked in some random country, the conversation with ICANN would go something like this:

    client: what's the IP for example.com?

    ICANN: I don't know, but you should ask verisign (the operator of the .com TLD)

You could argue that ICANN should have responded with "I don't know" (which will cause the recursion to stop), but that's sort of pointless because if you really wanted to know, you could ask ICANN to resolve some random.com domain and you would get the response that you need ("go ask verisign"). The company you want to block/sanction/threaten here is verisign (the actual company that controls the .com zone), not ICANN (the company that controls the root zone).

how much daily traffic does running your own DNS server generate? (to keep up to date with the domain-IP binding?)

I am really interested in trying this...

The DNS requests corresponding to typical HTTP traffic would be negligible compared to that HTTP traffic. DNS servers generally don't generate any traffic except when you perform a query.

Should be <512 bytes per request and after a particular domain name is queried it doesn't need to be queried again for some time (depending on TTL).

so its really only somewhat anonymizing if many people use it? is there some kind of distributed DNS table through TOR or something?

If you were to run your own recursive DNS resolver, you'd probably do 3-10x more DNS traffic than you did before, due to the number of requests increasing from 1 to multiple.

That said, each DNS request & response is usually under 1KB, so the background traffic doesn't really stop being negligible.

Don't run your resolver open to the wider internet, however. Only accept recursive requests from your local network segment, or you'll unwittingly become part of DNS amplification attacks.

oh, I thought a DNS server meant I would be storing my own copy of table of URL's to IP adresses?

So the privacy enhancement is tracking by the DNS server? but if my local DNS server forwards these requests, how exactly does it improve my privacy if it does not store a bulk "DNS table" in my amateur speak?

DNS servers deal with domain names, not URLs. For an HTTPS request, the rest of the URL (path etc.) is only ever sent to the target site, over an encrypted connection. From an information privacy perspective, DNS is a weak link in this mechanism.

The reason running your own DNS improves your privacy has to do with how centralized or decentralized your DNS traffic is, and who gets to see it as a result.

If you use a DNS controlled by someone else, the DNS operator can see, and therefore (in principle) track, all of the domain name lookups you make that aren't served by your local machine or router's cache.

In contrast, if you run your own DNS server that's not just a slave to e.g. an ISP's server, then external domain lookups are made directly against the authoritative DNS servers for the domain you're looking up. There's no longer a single central server, other than your own, that's seeing all your lookups.

So if you look up news.ycombinator.com, and the name isn't already in your cache, the request goes direct from your DNS server to an AWS DNS server, because that's who hosts the ycombinator.com DNS. Your non-cached DNS requests are thus spread out across all the DNS servers for all the different domains you're accessing, and the only DNS servers that see your requests are the servers for the domain you're accessing.

There are some caveats to this.

First, if your ISP is unscrupulous, it can examine your DNS traffic anyway, since most DNS traffic is currently unencrypted (this is changing, but slowly.) To get around that you'd need an encrypted DNS proxy hosted outside your provider's network.

Second, large DNS providers like Verisign, GoDaddy, AWS etc. will tend to get a large subset of your requests anyway. However, most of those big companies are publicly focused on information security, and it would be big news if it turned out they were tracking your DNS lookups. Some of them provide audited guarantees that they don't do this.

Let's say you look up "porn.example.com" and "dissidents.github.io".

. (the ICANN DNS root) sees "io" and "com".

com sees "example.com" but not which subdomain or anything to do with io.

io sees "github.io" but not which subdomain or anything to do with com.

example.com sees "porn.example.com" but doesn't know anything about "dissidents.github.io".

github.io sees "dissidents.github.io" but doesn't know anything about "porn.example.com".

You leak the same total information, but no single entity has a convenient list of every DNS request you make.

That's wrong though, the ICANN sees the whole domain, dissidents.github.io and replies with "I don't know, but ask the owner of .IO which is xx.xx.xx.xx"

Well, unless the servers implement https://tools.ietf.org/html/rfc7816, which is experimental and quite recent.

Unbound seems to support it, which is great (and that makes you right, actually, if the user installs this software). I don't know for others.

Actually, this appears to be implementation-specific? The resolver I looked at a while ago[0] did something like:

  @. io NS -> ns1.io # (for example[1])
  @. ns1.io A -> x.x.x.x
  @x.x.x.x github.io NS -> ns1.github.io
  @x.x.x.x ns1.github.io A -> y.y.y.y
  @y.y.y.y dissidents.github.io A -> z.z.z.z
  return z.z.z.z

caching every step of the way. Admittedly, I don't know that it wasn't going out of it's way to avoid leaking, but this was circa 2012 or so, so it predates rfc7816 at least.

0: found by googling "dns recursive resolver example code" or something to that effect

1: There's actually several NS entries, with retrying/failover.

Reply on news.ycombinator.com