4 minute read Published:

Building an API for the BIND9 DNS server to solve ACME DNS challenges

I manage most of my domains using my own nameservers, running BIND9 on two Debian VPS located in Italy (master) and France (slave). Until now, I’ve been changing the DNS records by SSHing into the machine and editing the zone file by hand. This worked fine since I rarely needed to change any DNS records. Then earlier this year, Let’s Encrypt put the ACME v2 endpoint into production which allows users to obtain wildcard certificates using the DNS challenge. This put me into a situation where I needed to create, update and delete DNS records automatically.

The ACME HTTP challenge requires the user to make the challenge flag available via HTTP under This way, the ACME endpoint can only verify ownership over a specific subdomain ( in this case). The DNS challenge looks for the flag in the TXT record This allows the ACME endpoint to validate ownership over the whole domain and it is possible to issue a wildcard certificate for *

Since DNS setups vary depending on the domain provider or used DNS server, certbot can use manual auth and cleanup hooks, that receive the domain name and challenge flag via the environment variables $CERTBOT_DOMAIN and $CERTBOT_VALIDATION respectively.

Once the challenge mechanism was understood, I needed a way to programmatically create and delete records on my BIND9 server. I decided to implement a REST-like webservice to run on the same machine as BIND9 and modify records using the nsupdate command.

The REST API offers two methods:

POST /record
X-Api-Token: <api-token>

    "name": "",
    "value": "<challenge flag>",
    "record": "TXT",
    "ttl": 1337
DELETE /record
X-Api-Token: <api-token>

    "name": "",
    "record": "TXT"

The X-Api-Token header contains the SHA256-HMAC over the request body using a pre-shared secret to prevent unauthenticated use of the API but this still does not protect against replay attacks. If an attacker managed to intercept an request to the API, (s)he would be able to resend the same request to the server and re-execute the command. To prevent this, the API server has to be placed behind a reverse proxy like nginx to encrypt the requests using TLS or as I am doing it, make the server listen on a private IP address inside an encrypted VLAN (tinc in my case).

Once the body was verified using the pre-shared secret, nsupdate is invoked and the following update or delete scripts are passed via stdin:

update add 1337 TXT <challenge flag>
update delete TXT

For the implementation of the API and the client, I chose to use Rust with the actix-web framework for the server and reqwest to make HTTP requests on the client side. The implementation along with installation instructions can be found on Github or my Gitea instance. I have already worked with the Rocket web framework for my Bachelor thesis but it depends on the nightly branch of the compiler and is a pain to maintain over a longer period of time due to breaking changes in the nightly compiler. Also actix-web is really fast1. Further crates that were used and should be mentioned include ring for cryptographic operations, serde for (de)serialization of data and proptest to verify some properties of my code (e.g. verify_signature(key, msg, sign(key, msg)) must be true for every input of key and msg). Rust made it easy to exchange data between the client and the server in a typesafe manner and actix-web offers an well designed API to build fast web applications. While actix-web lacks the incredible ergonomics of Rocket (it’s not bad, just not as good as Rocket), I prioritize using the stable compiler branch over API ergonomics.

The client itself is independent of the way, certbot works and the integration into the workflow is archived by bash scripts inspired by these INWX certbot hooks.

For the server to work, a DNS key has to be generated as described in the repository to be able to modify the records using nsupdate. I start the API server using a systemd service:

Description=BIND9 API

ExecStart=/usr/local/bin/bind9-api -k /etc/bind/dnskey -h -t <api secret>
ExecStop=pkill bind9-api


The client is configured using the configuration file /etc/bind9apiclient.toml that contains the API URL and secret.

# API server host
host = ""
# API secret
secret = "topsecret"

The final binaries, I use in production, are compiled using the ekidd/rust-musl-builder Docker image to build completely static binaries by linking against the musl libc (Linking against the default glibc target, produces dynamically linked binaries that depend to the systems glibc and OpenSSL version).

After placing the client somewhere in $PATH and putting the certbot hooks on the machine that should obtain the certificates, I can invoke certbot like followed:

certbot certonly -n --agree-tos --server \ --preferred-challenges=dns-01 \
--manual --manual-auth-hook /usr/lib/letsencrypt-bind9/certbot-bind9-auth \
--manual-cleanup-hook /usr/lib/letsencrypt-bind9/certbot-bind9-cleanup \
--manual-public-ip-logging-ok -d -d '*'

I already obtained a wildcard certificate for my domain, even if I’m using only a single subdomain, to test my code. Obtaining the certificate worked fine, and I guess renewal won’t pose any problems either.