/docker-letsencrypt-dns

Docker with Certbot + Lexicon to provide Let's Encrypt SSL certificates validated by DNS challenges

Primary LanguageShellMIT LicenseMIT

  adferrand/letsencrypt-dns

CircleCI

Container functionalities

This Docker is designed to manage Let's Encrypt SSL certificates based on DNS challenges.

  • Let's Encrypt wildcard and regular certificates generation by Certbot using DNS challenges,
  • Automated renewal of almost expired certificates using Cron Certbot task,
  • Standardized API throuh Lexicon library to insert the DNS challenge with various DNS providers,
  • Centralized configuration file to maintain several certificates,
  • Modification of container configuration without restart,
  • Automated restart of specific containers when a certificate is renewed,
  • Container built on top of Alpine Linux distribution to reduce the footprint (image size is below 125MB).

Why use this Docker ?

If you are reading these lines, you certainly want to secure all your dockerized services using Let's Encrypt SSL certificates, which are free and accepted everywhere.

If you want to secure Web services through HTTPS, there is already plenty of great tools. In the Docker world, one can check traefik, or nginx-proxy + letsencrypt-nginx-proxy-companion. Basically, theses tools will allow automated and dynamic generation/renewal of SSL certificates, based on TLS or HTTP challenges, on top of a reverse proxy to encrypt everything through HTTPS.

So far so good, but you may fall in one of the following categories:

  1. You are in a firewalled network, and your HTTP/80 and HTTPS/443 ports are not opened to the outside world.
  2. You want to secure non-Web services (like LDAP, IMAP, POP, etc.) were the HTTPS protocol is of no use.
  3. You want to generate a wildcard certificate, valid for any sub-domain of a given domain.

For the first case, ACME servers need to be able to access your website through HTTP (for HTTP challenges) or HTTPS (for TLS challenges) in order to validate the certificate. With a firewall these two challenges - which are widely used in HTTP proxy approaches - will not be usable: you need to ask a DNS challenge. Please note that traefik embed DNS challenges, but only for few DNS providers.

For the second case, there is no website to use TLS or HTTP challenges, and you should ask a DNS challenge. Of course you could create a "fake" website to validate the domain using a HTTP challenge, and reuse the certificate on the "real" service. But it is a workaround, and you have to implement a logic to propagate the certificate, including during its renewal. Indeed, most of the non-Web services will need to be restarted each time the certificate is renewed.

For the last case, the use of a DNS challenge is mandatory. Then the problems concerning certificates propagation that have been discussed in the second case will also occur.

The solution is a dedicated and specialized Docker service which handles the creation/renewal of Let's Encrypt certificates, and ensure their propagation in the relevant Docker services. It is the purpose of this container.

Preparation of the container

First of all, before using this container, two steps of configuration need to be done: describing all certificates to acquire and maintain, then configuring an access to your DNS zone to publish DNS challenges.

Configuring the SSL certificates

This container uses a file which must be put at /etc/letsencrypt/domains.conf in the container. It is a simple text file which follows theses rules:

  • each line represents a certificate,
  • one line may contain several domains separated by a space,
  • the first domain is the certificate main domain,
  • each following domain on a line is included in the SAN of the certificate, allowing it to be used for several domains.

Let's take an example. Our domain is example.com, and we want:

  • a certificate for smtp.example.com
  • a certificate for imap.example.com + pop.example.com
  • a certificate for ldap.example.com
  • a wildcard certificate for any sub-domain of example.com and for example.com itself

Then the domains.conf will look like this:

smtp.example.com
imap.example.com pop.example.com
ldap.example.com
*.example.com example.com

You need also to provide the mail which will be used to register your account on Let's Encrypt. Set the environment variable LETSENCRYPT_USER_MAIL (default: noreply@example.com) in the container for this purpose.

NB: For a wildcard certificate, specifying a sub-domain already covered by the wildcard will raise an error during Certbot certificate generation (eg. test.example.com cannot be put on the same line than *.example.com).

Configuring DNS provider and authentication to DNS API

With environment variables

When using a DNS challenge, a TXT entry must be inserted in the DNS zone which manage the certificate domain. This TXT entry must contain a unique hash calculated by Certbot, and the ACME servers will check it before delivering the certificate.

This container will do the hard work for you, thanks to the association between Certbot and Lexicon: DNS provider API will be called automatically to insert the TXT record when needed. All you have to do is to define for Lexicon the DNS provider to use, and the API access key.

Following DNS provider are supported: AuroraDNS, AWS Route53, Cloudflare, ClouDNS, CloudXNS, Constellix, DigitalOcean, DNSimple, DnsMadeEasy, DNSPark, DNSPod, EasyDNS, ExoScale, Gandi, Gehirn, Glesys, GoDaddy, Google Cloud DNS, Linode, Linode V4, LuaDNS, Memset, Namecheap, Namesilo, NS1, OnApp, Online, OVH, PointHQ, PowerDNS, Rackspace, Rage4, Sakura Cloud, SoftLayer, Subreg, Transip, Vultr, Yandex, Zeit, Zonomi.

The DNS provider is choosen by setting an environment variable passed to the container: LEXICON_PROVIDER (default: cloudflare).

Most of the DNS APIs requires a user and a unique access token delivered by the DNS provider. See the documentation of your provider to check how to get these (see the DNS providers list on Lexicon documentation. Once done, authentication stuff can be set using one of the three following approach:

  • using environment variables in the form of LEXICON_[PROVIDER]_[OPTION] for provider parameters in the form of --[option] (for instance, LEXICON_CLOUDFLARE_AUTH_USERNAME with the CloudFlare provider for --auth-username option)
  • using environment variable LEXICON_PROVIDER_OPTIONS (default empty) which will be append directly to the lexicon binary (for instance, LEXICON_PROVIDER_OPTIONS could be set to --auth-token=my-token ...)

For instance, if the provider is CloudFlare, the username is my_user and the access token is my_secret_token, following environment variables must be passed to the container:

LEXICON_CLOUDFLARE_AUTH_USERNAME=my_cloudflare_email
LEXICON_CLOUDFLARE_AUTH_TOKEN=my_cloudflare_global_api_key

Or alternatively:

LEXICON_PROVIDER=cloudflare
LEXICON_PROVIDER_OPTIONS=--auth-username=my_cloudflare_email --auth-token=my_cloudflare_global_api_key

Some providers (like OVH) need more specific environment variables. First, run following command to get the Lexicon help for this DNS provider:

docker run -it --rm adferrand/letsencrypt-dns lexicon ovh --help

Once done, you will see authentication parameters of the form --auth-somevar. Theses parameters must be setted using environment variables of the form LEXICON_[PROVIDER]_AUTH_SOMEVAR.

For example with OVH, authentication parameters are --auth-entrypoint, --auth-application-key, --auth-application-secret and --auth-consumer-key. Corresponding environment variables are LEXICON_OVH_AUTH_ENTRYPOINT, LEXICON_OVH_AUTH_APPLICATION_KEY, LEXICON_OVH_AUTH_APPLICATION_SECRET and LEXICON_OVH_AUTH_CONSUMER_KEY. Or alternatively, set the LEXICON_PROVIDER_OPTIONS to --auth-entrypoint=my_entrypoint --auth-application-key=my_application_key --auth-application-secret=my_application_secret --auth-consumer-key=my_consumer_key.

Finally there is some options specific to Lexicon itself, not related to the authentication on a particular provider (like --delegate). You can specify this kind of options (eg. domain for Cloudns) via the LEXICON_OPTIONS (default empty) environment variable.

With YAML configuration files

Starting with version 2.7.0 that uses Lexicon 3.x, DNS providers and Lexicon can be configured using YAML configuration files. It means that a unique lexicon.yml, located on the persisted storage /etc/letsencrypt, can contain all parameters for Lexicon and several DNS providers. Each entry corresponds to a Lexicon option, or a provider name. Under a provider name, the relevant parameters for this provider can be set. As a convention name for the parameters, a given CLI argument --auth-username will be written in the YAML configuration file as auth_username.

To select the provider to use, the environment variable LEXICON_PROVIDER still need to be set.

Taking the OVH example in the previous version, with a specific configuration to provide for Lexicon itself (--delegate), the lexicon.yml will take the following form:

# Content of /etc/letsencrypt/lexicon.yml
delegate: subdomain
ovh:
  auth_entrypoint: ovh-eu
  auth_application_key: my_application_key
  auth_application_secret: my_application_secret
  auth_consumer_key: my_consumer_key

The YAML configuration file approach is particularly well suited for the auto DNS provider, that is able to resolve the actual DNS provider for a given domain. Indeed, lexicon.yml will be a convenient place to put the configuration of several providers.

NB: A configuration file can be set for one particular provider. In this case, the filename must be lexicon_[provider_name].yml (eg. lexicon_ovh.yml for OVH). Theses files must be located in /etc/letsencrypt. There root content is directly the parameters of the relevant provider, like so:

# Content of /etc/letsencrypt/lexicon_cloudflare.yml
auth_username: my_cloudflare_email
auth_token: my_cloudflare_global_api_key

Run the container

Once preparation is done, the container can be run. As said, domains.conf must be mounted in the container, and API authentication variables must be passed as environment variables to the container.

For Cloudflare, with example described during preparation, run :

docker run \
	--name letsencrypt-dns \
	--volume /etc/letsencrypt/domains.conf:/etc/letsencrypt/domains.conf \
	--env 'LETSENCRYPT_USER_MAIL=admin@example.com' \
	--env 'LEXICON_PROVIDER=cloudflare' \
	--env 'LEXICON_CLOUDFLARE_USERNAME=my_cloudflare_email' \
	--env 'LEXICON_CLOUDFLARE_TOKEN=my_cloudflare_global_api_key' \
	adferrand/letsencrypt-dns

At start, the container will look to domains.conf and generate the certificates if needed. Then a cron task is launched twice a day to regenerate certificates if needed. The certificates are located in the container at /etc/letsencrypt/live/.

Data persistency

This container declares /etc/letsencrypt as a volume. Consequently generated certificates will not be destroyed if the container is destroyed. Upon re-creation, certificates will be available again.

Share certificates with the host

If you want to share the generated certificates to the host (eg. in /var/docker-data/letsencrypt), you can use a host mount:

docker run \
	--name letsencrypt-dns \
	--volume /etc/letsencrypt/domains.conf:/etc/letsencrypt/domains.conf \
	--volume /var/docker-data/letsencrypt:/etc/letsencrypt \
    --env 'LETSENCRYPT_USER_MAIL=admin@example.com' \
	--env 'LEXICON_PROVIDER=cloudflare' \
	--env 'LEXICON_CLOUDFLARE_USERNAME=my_cloudflare_email' \
	--env 'LEXICON_CLOUDFLARE_TOKEN=my_cloudflare_global_api_key' \
	adferrand/letsencrypt-dns

Share certificates with other containers

If you want to share the generated certificates with other containers, mount the container volume /etc/letsencrypt into the target containers. For example:

docker run \
	--name letsencrypt-dns \
	--volume /etc/letsencrypt/domains.conf:/etc/letsencrypt/domains.conf \
	--volume /var/docker-data/letsencrypt:/etc/letsencrypt \
	--env 'LETSENCRYPT_USER_MAIL=admin@example.com' \
	--env 'LEXICON_PROVIDER=cloudflare' \
	--env 'LEXICON_CLOUDFLARE_USERNAME=my_cloudflare_email' \
	--env 'LEXICON_CLOUDFLARE_TOKEN=my_cloudflare_global_api_key' \
	adferrand/letsencrypt-dns

docker run \
	--volumes-from letsencrypt-dns \
	--env 'KEY_PATH=/etc/letsencrypt/live/smtp.example.com/privkey.pem' \
	--env 'CERTIFICATE_PATH=/etc/letsencrypt/live/smtp.example.com/cert.pem' \
	namshi/smtp

The volume /etc/letsencrypt will be available for the SMTP container, which can use a generated certificate for its own concern (here, securing the SMTP protocol).

Certificates files permissions

By default certificates files (cert.pem, privkey.pem, etc.) are accessible only to the user/group owning /etc/letsencrypt, which is root by default. It means that generated certificates cannot be used by non-root processes (in other containers or on the host).

You can modify file mode of /etc/letsencrypt/archive and /etc/letsencrypt/live folders and their content to allow non-root processes to access the certificates. Set environment variables CERTS_DIRS_MODE (default: 0750) and CERTS_FILES_MODE (default: 0640) to modify directories and files mode respectivly. For example, a file mode of 0644 and directory mode of 0755 will open access to everyone.

Alternatively or cumulatively you may need to change the owner user/group of /etc/letsencrypt/archive and /etc/letsencrypt/live folders and their content. To do so, specify user/group name or uid/gid in the relevant environment variables: CERTS_USER_OWNER (default: root) and CERTS_GROUP_OWNER (default: root).

(Warning) Certificates files permissions, introduced in container version 1.4, will modify default permissions for certificates. Previously, /etc/letsencrypt/live and /etc/letsencrypt/archive were 0750, their sub-folders where 0755 and contained files were 0644. Now theses folders and their sub-folders are 0750 while contained files are 0640: this should not lead to any regression, as the parent folders have a more restrictive permission than their content, leading certs files to be unaccessible to non-root processes. However for pathological cases you will need to set environment variable CERTS_DIRS_MODE and CERTS_FILES_MODE appropriately.

Runtime operations

Certificates reconfiguration at runtime

If you want to add a new certificate, remove one, or extend existing one to other domains, you just need to modify the domains.conf file from the host. Once saved, the container will automatically mirror the modifications in /etc/letsencrypt volume. If new certificates need to be generated, please note that approximately 30 seconds are required for each generation before modifications are visible.

Please check the container logs to follow the operations.

Restart containers when a certificate is renewed

As said in introduction, most of the non-Web services require a restart when the certificate is changed. And this will occur at least once each two months. To ensure correct propagation of the new certificates in your Docker services, one special entry can be added at the end of a line for the concerned certificate in domains.conf.

This entry takes the form of autorestart-containers=container1,container2,container3 where containerX is the name of a container running on the same Docker instance than letsencrypt-dns.

You need also to mount the Docker socket of the host /var/run/docker.sock in the letsencrypt-dns container.

Once done, all specified containers will be restarted when the target certificate is renewed.

For example, we want to restart the container named smtp when the certificate smtp.example.com is renewed. Construct the following domains.conf file:

smtp.example.com imap.example.com autorestart-containers=smtp
auth.example.com

Then execute following commands:

docker run \
	--name letsencrypt-dns \
	--volume /etc/letsencrypt/domains.conf:/etc/letsencrypt/domains.conf \
	--volume /var/docker-data/letsencrypt:/etc/letsencrypt \
	--volume /var/run/docker.sock:/var/run/docker.sock \
	--env 'LETSENCRYPT_USER_MAIL=admin@example.com' \
	--env 'LEXICON_PROVIDER=cloudflare' \
	--env 'LEXICON_CLOUDFLARE_USERNAME=my_cloudflare_email' \
	--env 'LEXICON_CLOUDFLARE_TOKEN=my_cloudflare_global_api_key' \
	adferrand/letsencrypt-dns

docker run \
	--name smtp \
	--volumes-from letsencrypt-dns \
	--env 'KEY_PATH=/etc/letsencrypt/live/smtp.example.com/privkey.pem' \
	--env 'CERTIFICATE_PATH=/etc/letsencrypt/live/smtp.example.com/cert.pem' \
	namshi/smtp

If the certificate smtp.example.com is renewed, the container named smtp will be restarted. Renewal of auth.example.com will not restart anything.

Call a reload command on containers when a certificate is renewed

Restarting a container when a certificate is renewed is sufficient for all cases. However one drawback is that the target processes will stop during a little time, and consequently the services provided are not continuous. This may be ok for non critical services, but problematic for things like authentication services or database servers.

If a target process allows it, the letsencrypt-dns container can call a reload configuration command on the target container when a certificate is renewed. In this case, service is not stopped and immediatly takes into account the new config, including the new certificate. Apache2 for example (example only, as an http challenge will be a better option here) can see its configuration to be hot-reloaded by invoking the command apachectl graceful in the target container.

To specify which command to launch on which container when a certificate is renewed, one will put at the end of the relevant line of domains.conf a special entry which takes the form of autocmd-containers=container1:command1,container2:command2 arg2a arg2b,container3:command3 arg3a. Comma , separates each container/command configuration, colon : separates the container name from the command to launch. Commands must be executable files, located in the $PATH of the target container, or accessed by their full path.

Similarly to the automatic containers restart functionality, you need to mount the Docker socket of the host /var/run/docker.sock in the letsencrypt-dns container (with the docker executable, using the command-line parameter --volume /var/run/docker.sock:/var/run/docker.sock).

In the case of an Apache2 server embedded in a container named my-apache to be reloaded when certificate web.example.com is renewed, put following entry in domains.conf:

web.example.com autocmd-containers=my-apache:apachectl graceful

If the certificate web.example.com is renewed, command apachectl graceful will be invoked on container my-apache, and the apache2 service will use the new certificate without killing any HTTP session.

(Limitations on invokable commands) The option autocmd-container is intended to call a simple executable file with few potential arguments. It is not made to call some advanced bash script, and would likely fail if you do so. In fact, the command is not executed in a shell on the target, and variables will be resolved against the lets-encrypt container environment. If you want to operate advanced scripting, put an executable script in the target container, and use its path in autocmd-container option.

Miscellaneous and testing

Using ACME v1 servers

Starting to version 2.0.0, this container uses the ACME v2 servers (production & staging) to allow wildcard certificates generation. If for any reason you want to continue to use old ACME v1 servers, you can set the environment variable LETSENCRYPT_ACME_V1 (default: false) to true. In this case, ACME v1 servers will be used to any certificate generation, but wildcard certificates will not be supported.

NB: During a certificate renewal, the server (and authentication) used for the certificate generation will be reused, independently of the LETSENCRYPT_ACME_V1 environment variable value. If you want to change the server used for a particular certificate, you will need first to revoke it by removing the relevant entry from domains.txt file before recreating it.

Activating staging ACME servers

During development it is not advised to generate certificates against production ACME servers, as one could reach easily the weekly limit of Let's Encrypt and could not generate certificates for a certain period of time. Staging ACME servers do not have this limit. To use them, set the environment variable LETSENCRYPT_STAGING (default: false) to true.

You will need to wipe content of /etc/letsencrypt volume before container re-creation when enabling or disabling staging.

Auto-export certificates in PFX format

Some services need the SSL key and certificate stored together in PFX format (also known as PKCS#12) whose extension is .pfx (or .p12). For this purpose one can set the container environment variable PFX_EXPORT (default: false) to true: in this case, the container will ensure that every certificate handled by Certbot is exported in PFX format during certificate creation, renewal or container start/restart.

The PFX certificate for a given primary domain is located in the container on /etc/letsencrypt/[DOMAIN]/cert.pks: it contains the key, certificate and all intermediate certificates.

By default, the PFX certificates are not protected by a passphrase. You can define one using the environment variable PFX_EXPORT_PASSPHRASE.

Sleep time

During a DNS challenge, a sleep must be done after TXT entry insertions in order to let DNS zone updates be propagated correctly and ensure that ACME servers will see them. Default value is 30 seconds: if this value does not suit your needs, you can modify it by setting the environment variable LEXICON_SLEEP_TIME (default: 30).

Shell access

For debugging and maintenance purpose, you may need to start a shell in your running container. With a Docker of version 1.3.0 or higher, you can do:

docker exec -it letsencrypt-dns /bin/sh

You will obtain a shell with the standard tools of an Alpine distribution.