Fix a log message.

[dnstt.git] / README

Userspace DNS tunnel with support for DoH and DoT
David Fifield <david@bamsoftware.com>
Public domain

dnstt is a DNS tunnel with these features:
 * Works over DNS over HTTPS (DoH) and DNS over TLS (DoT) as well as
   plaintext UDP DNS.
 * Embeds a sequencing and session protocol (KCP/smux), which means that
   the client does not have to wait for a response before sending more
   data, and any lost packets are automatically retransmitted.
 * Encrypts the contents of the tunnel and authenticates the server by
   public key.

It has these noteworthy limitations:
 * Requires intermediary resolvers to support large responses (1232 bytes,
   which is more than the mandated minimum of 512 bytes).

dnstt is an application-layer tunnel that runs in userspace. It doesn't
provide a TUN/TAP interface; it only hooks up a local TCP port with a
remote TCP port (like netcat or `ssh -L`) by way of a DNS resolver. It
does not itself provide a SOCKS or HTTP proxy interface, but you can get
the same effect by running a proxy on the tunnel server and having the
tunnel terminate at the proxy.

```
.------.               .--------.             .------.
|tunnel|-- DoH / DoT --|resolver|-- UDP DNS --|tunnel|
|client|               '--------'             |server|
'------'                                      '------'
```


## DNS zone setup

Because the server side of the tunnel acts like an authoritative name
server, you need to own a domain name and set up a subdomain for the
tunnel. Let's say your domain name is example.com and your server's IP
addresses are 203.0.113.2 and 2001:db8::2. Go to your name registrar and
add three new records:

```
A       tns.example.com points to 203.0.113.2
AAAA    tns.example.com points to 2001:db8::2
NS      t.example.com   is managed by tns.example.com
```

The labels `tns` and `t` can be anything you want, but the `tns` label
should not be a subdomain of the `t` label (that space is reserved for
the contents of the tunnel), and the `t` label should be short (because
there is limited space available in a DNS message, and the domain name
takes up part of that space).

Now, when a recursive DNS resolver receives a query for a name like
aaaa.t.example.com, it will forward the query to the tunnel server at
203.0.113.2 or 2001:db8::2.


## Tunnel server setup

Compile the server:
```
$ cd dnstt-server
$ go build
```

First you need to generate the server keypair that will be used to
authenticate the server and encrypt the tunnel.
```
$ ./dnstt-server -gen-key -privkey-file server.key -pubkey-file server.pub
privkey written to server.key
pubkey  written to server.pub
```

Run the server. You need to provide an address that will listen for UDP
DNS packets (`:5300`), the private key file (`server.key`), the root of
the DNS zone (`t.example.com`), and a TCP address to which incoming
tunnel stream will be forwarded (`127.0.0.1:8000`).
```
$ ./dnstt-server -udp :5300 -privkey-file server.key t.example.com 127.0.0.1:8000
```

The tunnel server needs to be able to receive packets on an external
port 53. You can have it listen on port 53 directly using `-udp :53`,
but that requires the program to run as root. It is better to run the
program as an ordinary user and have it listen on an unprivileged port
(`:5300` above), and port-forward port 53 to it. On Linux, use this
command to forward external port 53 to localhost port 5300:
```
# iptables -I INPUT -p udp --dport 5300 -j ACCEPT
# iptables -t nat -I PREROUTING -i eth0 -p udp --dport 53 -j REDIRECT --to-ports 5300
# ip6tables -I INPUT -p udp --dport 5300 -j ACCEPT
# ip6tables -t nat -I PREROUTING -i eth0 -p udp --dport 53 -j REDIRECT --to-ports 5300
```

You need to also run something for the tunnel server to connect to. It
can be a proxy server or anything else. For testing, you can use an
Ncat listener:
```
$ ncat -lkv 127.0.0.1 8000
```


## Tunnel client setup

Compile the client:
```
$ cd dnstt-client
$ go build
```

Copy the server.pub file from the server to the client. You don't need
server.key on the client; leave it on the server.

Choose a public DoH or DoT resolver. There is a list of DoH resolvers
here:
 * https://github.com/curl/curl/wiki/DNS-over-HTTPS#publicly-available-servers

And DoT resolvers here:
 * https://dnsprivacy.org/wiki/display/DP/DNS+Privacy+Public+Resolvers#DNSPrivacyPublicResolvers-DNS-over-TLS%28DoT%29
 * https://dnsencryption.info/imc19-doe.html

To run the tunnel client using DoH, you need to provide the URL of the
DoH resolver (`https://doh.example/dns-query`), the server's public key
files (`server.pub`), the root of the DNS zone (`t.example.com`), and
the local TCP port that will receive connections and forward them
through the tunnel (`127.0.0.1:7000`):
```
$ ./dnstt-client -doh https://doh.example/dns-query -pubkey-file server.pub t.example.com 127.0.0.1:7000
```

For DoT, it's the same, but use the `-dot` option instead:
```
$ ./dnstt-client -dot dot.example:853 -pubkey-file server.pub t.example.com 127.0.0.1:7000
```

Once the tunnel client is running, you can connect to the local end of
the tunnel, type something, and see it appear at the remote end.
```
$ ncat -v 127.0.0.1 7000
```

The client also has a plaintext UDP mode that can work through a
recursive resolver or directly to the tunnel server
(`-udp tns.example.com`), but it does not provide any covertness for the
tunnel and should only be used for testing.


## How to make a proxy

You can make the tunnel into a general-purpose proxy by running a proxy
server and connecting the server end of the tunnel to it. For example,
Ncat has a built-in simple HTTP server:
```
$ ncat -lkv --proxy-type http 127.0.0.1 8000
$ ./dnstt-server -udp :5300 -privkey-file server.key t.example.com 127.0.0.1:8000
```

On the client, have the tunnel client listen on 127.0.0.1:7000, and configure
your applications to use http://127.0.0.1:7000/ as an HTTP proxy.

```
$ ./dnstt-client -doh https://doh.example/dns-query -pubkey-file server.pub t.example.com 127.0.0.1:7000
$ curl -x http://127.0.0.1:7000/ http://example.com/
```


## Covertness

Support for DoH and DoT is only to make it more difficult for a local
observer to see that a DNS tunnel is being used, not for the overall
security of the connection. There is a separate encryption layer inside
the tunnel that protects the contents of the tunnel from the resolver
itself.

The encryption of DoH or DoT prevents a network observer between the
tunnel client and the resolver from seeing the remote destination of the
tunnel. An observer can see that the tunnel client is connecting to a
resolver, but cannot see where the resolver is forwarding its queries.
An observer can probably infer, based on volume and other traffic
characteristics, that a tunnel is being used, though it cannot tell
where the remote end of the tunnel is, nor what the contents of the
tunnel are. If the tunnel client is not using DoH or DoT but instead UDP
(`-udp` option), then even an observer between the tunnel client and the
resolver can see that a tunnel is being used and where the remote end of
the tunnel is.

An observer between the resolver and the tunnel server (this includes
the resolver itself) can easily tell that a tunnel is being used and
where the remote end of the tunnel is, because there is no DoH or DoT
encryption at that point. This kind of observer still cannot read the
contents of the tunnel, because there is an additional layer of
end-to-end encryption between the tunnel client and the tunnel server.

An observer who watches what leaves the tunnel server will be able to
see anything that the tunnel server forwards to some other host (if the
tunnel server is acting as a proxy, for example), unless that data has
been separately encrypted before being sent through the tunnel.


## Encryption and authentication

The tunnel uses a Noise protocol (https://noiseprotocol.org/noise.html)
for end-to-end security between the tunnel client and tunnel server.
This protocol is independent of the DoH or DoT encryption between the
tunnel client and resolver. The specific protocol is Noise_NK_25519_ChaChaPoly_BLAKE2s
(https://noiseprotocol.org/noise.html#protocol-names-and-modifiers).
The NK handshake pattern authenticates the server but not the client.

The Noise layer is sandwiched between two other protocol layers: KCP
(https://github.com/xtaci/kcp-go) which creates a reliable stream on top
of unreliable datagrams, and smux (https://github.com/xtaci/smux) which
provides stream multiplexing and session features. An observer (such as
the intermediary resolver) may read the headers of the KCP layer, but not
of the smux layer nor of the streams that are inside. The model is
similar to what you would get with TLS or SSH over TCP: an observer can
see TCP-level ACKs and sequence numbers, but cannot read the stream data
inside.

```
application data
smux
Noise
KCP
DNS messages
DoH / DoT / UDP DNS
```

When you run `dnstt-server -gen-key`, you can save the private and
public keys to a file using the `-privkey-file` and `-pubkey-file`
options. You can then load the keys later using `-privkey-file` on the
server and `-pubkey-file` on the client. Alternatively, you can deal
with the keys as literal hexadecimal strings rather than files. If you
run `dnstt-server -gen-key` without the `-privkey-file` and
`-pubkey-file` options, it will display the keys rather than save them
to files. You can then use the keys with `-privkey` on the server and
`-pubkey` on the client.
```
$ ./dnstt-server -gen-key
privkey 0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef
pubkey  0000111122223333444455556666777788889999aaaabbbbccccddddeeeeffff
$ ./dnstt-server -udp :5300 -privkey 0123456789abcdef0123456789abcdef0123456789abcdef0123456789abcdef t.example.com 127.0.0.1:8000
$ ./dnstt-client -dot dot.example:853 -pubkey 0000111122223333444455556666777788889999aaaabbbbccccddddeeeeffff t.example.com 127.0.0.1:7000
```
If you run the server without `-privkey-file` or `-privkey`, it will
generate a temporary keypair and print the public key in the log. But
the key will be different the next time you restart the server, and you
will have to reconfigure clients.


## Payload sizes

In the client, the available space for user data per query depends on
the length of the domain name in use. Shorter domain names leave more
space for user data.

In the server, the available space for user data per response depends on
the maximum UDP payload size. The larger the UDP payload size, the more
space there is for user data. You want to use as large a UDP payload
size as possible, but not larger than what is supported by the resolver
you are using. Values above 1452 may cause IP fragmentation which can
reduce performance. You can control the maximum UDP payload size with
the `-mtu` option. The default is 1232 bytes; this ought to be supported
by most resolvers that understand EDNS(0) (RFC 6891). For maximum
compatibility, set the maximum to 512, but know that doing so will
reduce downstream bandwidth.
```
$ ./dnstt-client -mtu 512 -doh https://doh.example/dns-query -pubkey-file server.pub t.example.com 127.0.0.1:7000
```

The client and server emit an "effective MTU" log line when starting up
that shows how much space is available for user data in each query or
response. For the server, there may be more space available in some
responses and less in others (depending on the size of the corresponding
query); the logged value is the minimum that is guaranteed to be
supported in any response.