Self-signed certificates or custom Certification Authorities

Tier: Free, Premium, Ultimate Offering: GitLab.com, Self-managed
History
  • Introduced in GitLab Runner 0.7.0

GitLab Runner provides two options to configure certificates to be used to verify TLS peers:

  • For connections to the GitLab server: the certificate file can be specified as detailed in the Supported options for self-signed certificates targeting the GitLab server section.

    This solves the x509: certificate signed by unknown authority problem when registering a runner.

    For existing Runners, the same error can be seen in Runner logs when trying to check the jobs: 

      Couldn't execute POST against https://hostname.tld/api/v4/jobs/request:
  Post https://hostname.tld/api/v4/jobs/request: x509: certificate signed by unknown authority

  • A more generic approach which also covers other scenarios such as user scripts, connecting to a cache server or an external Git LFS store: a certificate can be specified and installed on the container as detailed in the Trusting TLS certificates for Docker and Kubernetes executors section.

    An example job log error concerning a Git LFS operation that is missing a certificate: 

      LFS: Get https://object.hostname.tld/lfs-dev/c8/95/a34909dce385b85cee1a943788044859d685e66c002dbf7b28e10abeef20?X-Amz-Expires=600&X-Amz-Date=20201006T043010Z&X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=svcgitlabstoragedev%2F20201006%2Fus-east-1%2Fs3%2Faws4_request&X-Amz-SignedHeaders=host&X-Amz-Signature=012211eb0ff0e374086e8c2d37556f2d8ca4cc948763e90896f8f5774a100b55: x509: certificate signed by unknown authority

Supported options for self-signed certificates targeting the GitLab server

This section refers to the situation where only the GitLab server requires a custom certificate. If other hosts (e.g. object storage service without proxy download enabled) also require a custom certificate authority (CA), please see the next section.

GitLab Runner supports the following options:

  • Default - Read the system certificate: GitLab Runner reads the system certificate store and verifies the GitLab server against the certificate authorities (CA) stored in the system.

  • Specify a custom certificate file: GitLab Runner exposes the tls-ca-file option during registration (gitlab-runner register --tls-ca-file=/path), and in config.toml under the [[runners]] section. This allows you to specify a custom certificate file. This file will be read every time the Runner tries to access the GitLab server. If you are using GitLab Runner Helm chart, you will need to configure certificates as described in Providing a custom certificate for accessing GitLab.

  • Read a PEM certificate: GitLab Runner reads the PEM certificate (DER format is not supported) from a predefined file:

    • /etc/gitlab-runner/certs/gitlab.example.com.crt on *nix systems when GitLab Runner is executed as root.

      If your server address is https://gitlab.example.com:8443/, create the certificate file at: /etc/gitlab-runner/certs/gitlab.example.com.crt.

      You can use the openssl client to download the GitLab instance’s certificate to /etc/gitlab-runner/certs: 

      openssl s_client -showcerts -connect gitlab.example.com:443 -servername gitlab.example.com < /dev/null 2>/dev/null | openssl x509 -outform PEM > /etc/gitlab-runner/certs/gitlab.example.com.crt

      To verify that the file is correctly installed, you can use a tool like openssl. For example: 

      echo | openssl s_client -CAfile /etc/gitlab-runner/certs/gitlab.example.com.crt -connect gitlab.example.com:443 -servername gitlab.example.com
    • ~/.gitlab-runner/certs/gitlab.example.com.crt on *nix systems when GitLab Runner is executed as non-root.
    • ./certs/gitlab.example.com.crt on other systems. If running GitLab Runner as a Windows service, this will not work. Specify a custom certificate file instead.

Notes:

  • If your GitLab server certificate is signed by your CA, use your CA certificate (not your GitLab server signed certificate). You might need to add the intermediates to the chain as well. For example, if you have a primary, intermediate, and root certificate, you can put all of them into one file: 

      -----BEGIN CERTIFICATE-----
  (Your primary SSL certificate: your_domain_name.crt)
  -----END CERTIFICATE-----
  -----BEGIN CERTIFICATE-----
  (Your intermediate certificate)
  -----END CERTIFICATE-----
  -----BEGIN CERTIFICATE-----
  (Your root certificate)
  -----END CERTIFICATE-----
  • If you are updating the certificate for an existing Runner, restart it.
  • If you already have a Runner configured through HTTP, update your instance path to the new HTTPS URL of your GitLab instance in your config.toml.
  • As a temporary and insecure workaround, to skip the verification of certificates, in the variables: section of your .gitlab-ci.yml file, set the CI variable GIT_SSL_NO_VERIFY to true.

Git cloning

The Runner injects missing certificates to build the CA chain by using CI_SERVER_TLS_CA_FILE. This allows git clone and artifacts to work with servers that do not use publicly trusted certificates.

This approach is secure, but makes the Runner a single point of trust.

Trusting TLS certificates for Docker and Kubernetes executors

There are two contexts that need to be taken into account when we consider registering a certificate on a container:

  • The user image, which is used to run the user script. For scenarios that involve trusting the certificate for user scripts, the user must take ownership regarding how to install a certificate, since this is highly dependent on the image itself, and the Runner has no way of knowing how to install a certificate in each possible scenario.
  • The Runner helper image, which is used to handle Git, artifacts, and cache operations. For scenarios that involve trusting the certificate for other CI/CD stages, the user only needs to make a certificate file available at a specific location (for example, /etc/gitlab-runner/certs/ca.crt), and the Docker container will automatically install it for the user.

Trusting the certificate for user scripts

If your build script needs to communicate with peers through TLS and needs to rely on a self-signed certificate or custom Certificate Authority, you will need to perform the certificate installation in the build job, as the Docker container running the user scripts doesn’t have the certificate files installed by default. This might be required to use a custom cache host, perform a secondary git clone, or fetch a file through a tool like wget, for example.

To install the certificate:

  1. Map the necessary files as a Docker volume so that the Docker container that will run the scripts can see them. Do this by adding a volume inside the respective key inside the [runners.docker] in the config.toml file, for example:

    • Linux: 

        [[runners]]
   name = "docker"
   url = "https://example.com/"
   token = "TOKEN"
   executor = "docker"

   [runners.docker]
     image = "ubuntu:latest"

     # Add path to your ca.crt file in the volumes list
     volumes = ["/cache", "/path/to-ca-cert-dir/ca.crt:/etc/gitlab-runner/certs/ca.crt:ro"]

  2. Linux-only: Use the mapped file (e.g ca.crt) in a pre_build_script that:

    1. Copies it to /usr/local/share/ca-certificates/ca.crt inside the Docker container.

    2. Installs it by running update-ca-certificates --fresh. For example (commands vary based on the distribution you’re using):

      • On Ubuntu: 

          [[runners]]
    name = "docker"
    url = "https://example.com/"
    token = "TOKEN"
    executor = "docker"

    # Copy and install CA certificate before each job
    pre_build_script = """
    apt-get update -y > /dev/null
    apt-get install -y ca-certificates > /dev/null

    cp /etc/gitlab-runner/certs/ca.crt /usr/local/share/ca-certificates/ca.crt
    update-ca-certificates --fresh > /dev/null
    """

      • On Alpine: 

          [[runners]]
    name = "docker"
    url = "https://example.com/"
    token = "TOKEN"
    executor = "docker"

    # Copy and install CA certificate before each job
    pre_build_script = """
    apk update >/dev/null
    apk add ca-certificates > /dev/null
    rm -rf /var/cache/apk/*

    cp /etc/gitlab-runner/certs/ca.crt /usr/local/share/ca-certificates/ca.crt
    update-ca-certificates --fresh > /dev/null
    """

If you just need the GitLab server CA cert that can be used, you can retrieve it from the file stored in the CI_SERVER_TLS_CA_FILE variable: 

curl --cacert "${CI_SERVER_TLS_CA_FILE}"  ${URL} -o ${FILE}

Trusting the certificate for the other CI/CD stages

You can map a certificate file to /etc/gitlab-runner/certs/ca.crt on Linux, or C:\GitLab-Runner\certs\ca.crt on Windows. The Runner helper image installs this user-defined ca.crt file at start-up, and uses it when performing operations like cloning and uploading artifacts, for example.

Docker

  • Linux: 

      [[runners]]
   name = "docker"
   url = "https://example.com/"
   token = "TOKEN"
   executor = "docker"

   [runners.docker]
     image = "ubuntu:latest"

     # Add path to your ca.crt file in the volumes list
     volumes = ["/cache", "/path/to-ca-cert-dir/ca.crt:/etc/gitlab-runner/certs/ca.crt:ro"]

  • Windows: 

      [[runners]]
   name = "docker"
   url = "https://example.com/"
   token = "TOKEN"
   executor = "docker"

   [runners.docker]
     image = "mcr.microsoft.com/windows/servercore:21H2"

     # Add directory holding your ca.crt file in the volumes list
     volumes = ["c:\\cache", "c:\\path\\to-ca-cert-dir:C:\\GitLab-Runner\\certs:ro"]

Kubernetes

To provide a certificate file to jobs running in Kubernetes:

  1. Store the certificate as a Kubernetes secret in your namespace: 

    kubectl create secret generic <SECRET_NAME> --namespace <NAMESPACE> --from-file=<CERT_FILE>

  2. Mount the secret as a volume in your runner, replacing <SECRET_NAME> and <LOCATION> with appropriate values: 

    gitlab-runner:
  runners:
   config: |
     [[runners]]
       [runners.kubernetes]
         namespace = "{{.Release.Namespace}}"
         image = "ubuntu:latest"
       [[runners.kubernetes.volumes.secret]]
           name = "<SECRET_NAME>"
           mount_path = "<LOCATION>"

    The mount_path is the directory in the container where the certificate is stored. If you used /etc/gitlab-runner/certs/ as the mount_path and ca.crt as your certificate file, your certificate is available at /etc/gitlab-runner/certs/ca.crt inside your container.

  3. As part of the job, install the mapped certificate file to the system certificate store. For example, in an Ubuntu container: 

    script:
  - cp /etc/gitlab-runner/certs/ca.crt /usr/local/share/ca-certificates/
  - update-ca-certificates

Due to a known issue in the Kubernetes executor’s handling of the helper image’s ENTRYPOINT, the mapped certificate file isn’t automatically installed to the system certificate store.

Troubleshooting

Refer to the general SSL troubleshooting documentation.

In addition, you can use the tlsctl tool to debug GitLab certificates from the Runner’s end.

x509: certificate signed by unknown authority while trying to pull executor images from private registry

This error occurs when the Docker host or Kubernetes node where the runner schedules the executors does not trust the certificate used by the private registry. To fix the error, add the relevant root certificate authority or certificate chain to the system’s trust store and restart the container service.

If you’re on Ubuntu or Alpine, run the following commands: 

cp ca.crt /usr/local/share/ca-certificates/ca.crt
update-ca-certificates
systemctl restart docker.service

If you’re not on Ubuntu or Alpine, check your operating system and the Docker documentation for the commands to run.

Depending on your version of GitLab Runner and the Docker host environment, you might also have to disable the FF_RESOLVE_FULL_TLS_CHAIN feature flag.

apt-get: not found errors in jobs

The pre_build_script commands are executed before every job a runner executes. If you add commands that are specific to certain distributions, like apk or apt-get when you install a certificate for user scripts, your CI jobs might fail if they use images based on different distributions.

For example, if some of your CI jobs run Ubuntu-based images and some run Alpine-based images and you add Ubuntu commands, the apt-get: not found error occurs in jobs with Alpine-based images. To resolve this, either:

  • Write your pre_build_script so that it is distribution-agnostic.
  • Use tags to ensure runners only pick up jobs with compatible images.