Use HashiCorp Vault secrets in GitLab CI/CD

Tier: Premium, Ultimate Offering: GitLab.com, Self-managed, GitLab Dedicated
caution
Authenticating with CI_JOB_JWT was deprecated in GitLab 15.9 and the token is scheduled to be removed in GitLab 18.0. Use ID tokens to authenticate with HashiCorp Vault instead, as demonstrated on this page.
note
Starting in Vault 1.17, JWT auth login requires bound audiences on the role when the JWT contains an aud claim. The aud claim can be a single string or a list of strings.

This tutorial demonstrates how to authenticate, configure, and read secrets with HashiCorp’s Vault from GitLab CI/CD.

Prerequisites

This tutorial assumes you are familiar with GitLab CI/CD and Vault.

To follow along, you must have:

  • An account on GitLab.
  • Access to a running Vault server (at least v1.2.0) to configure authentication and to create roles and policies. For HashiCorp Vaults, this can be the Open Source or Enterprise version.
note
You must replace the vault.example.com URL below with the URL of your Vault server, and gitlab.example.com with the URL of your GitLab instance.

How it works

ID tokens are JSON Web Tokens (JWTs) used for OIDC authentication with third-party services. If a job has at least one ID token defined, the secrets keyword automatically uses that token to authenticate with Vault.

The following fields are included in the JWT:

Field When Description
jti Always Unique identifier for this token
iss Always Issuer, the domain of your GitLab instance
iat Always Issued at
nbf Always Not valid before
exp Always Expires at
sub Always Subject (job ID)
namespace_id Always Use this to scope to group or user level namespace by ID
namespace_path Always Use this to scope to group or user level namespace by path
project_id Always Use this to scope to project by ID
project_path Always Use this to scope to project by path
user_id Always ID of the user executing the job
user_login Always Username of the user executing the job
user_email Always Email of the user executing the job
pipeline_id Always ID of this pipeline
pipeline_source Always Pipeline source
job_id Always ID of this job
ref Always Git ref for this job
ref_type Always Git ref type, either branch or tag
ref_path Always Fully qualified ref for the job. For example, refs/heads/main. Introduced in GitLab 16.0.
ref_protected Always true if this Git ref is protected, false otherwise
environment Job specifies an environment Environment this job specifies
groups_direct User is a direct member of 0 to 200 groups The paths of the user’s direct membership groups. Omitted if the user is a direct member of more than 200 groups. (Introduced in GitLab 16.11).
environment_protected Job specifies an environment true if specified environment is protected, false otherwise
deployment_tier Job specifies an environment Deployment tier of environment this job specifies (introduced in GitLab 15.2)
environment_action Job specifies an environment Environment action (environment:action) specified in the job. (Introduced in GitLab 16.5)

Example JWT payload:

{
  "jti": "c82eeb0c-5c6f-4a33-abf5-4c474b92b558",
  "iss": "gitlab.example.com",
  "iat": 1585710286,
  "nbf": 1585798372,
  "exp": 1585713886,
  "sub": "job_1212",
  "namespace_id": "1",
  "namespace_path": "mygroup",
  "project_id": "22",
  "project_path": "mygroup/myproject",
  "user_id": "42",
  "user_login": "myuser",
  "user_email": "myuser@example.com",
  "pipeline_id": "1212",
  "pipeline_source": "web",
  "job_id": "1212",
  "ref": "auto-deploy-2020-04-01",
  "ref_type": "branch",
  "ref_path": "refs/heads/auto-deploy-2020-04-01",
  "ref_protected": "true",
  "groups_direct": ["mygroup/mysubgroup", "myothergroup/myothersubgroup"],
  "environment": "production",
  "environment_protected": "true",
  "environment_action": "start"
}

The JWT is encoded by using RS256 and signed with a dedicated private key. The expire time for the token is set to job’s timeout, if specified, or 5 minutes if it is not. The key used to sign this token may change without any notice. In such case retrying the job generates new JWT using the current signing key.

You can use this JWT for authentication with a Vault server that is configured to allow the JWT authentication method. Provide your GitLab instance’s base URL (for example https://gitlab.example.com) to your Vault server as the oidc_discovery_url. The server can then retrieve the keys for validating the token from your instance.

When configuring roles in Vault, you can use bound claims to match against the JWT claims and restrict which secrets each CI/CD job has access to.

To communicate with Vault, you can use either its CLI client or perform API requests (using curl or another client).

Example

caution
JWTs are credentials, which can grant access to resources. Be careful where you paste them!

Let’s say you have the passwords for your staging and production databases stored in a Vault server that is running on http://vault.example.com:8200. Your staging password is pa$$w0rd and your production password is real-pa$$w0rd.

$ vault kv get -field=password secret/myproject/staging/db
pa$$w0rd

$ vault kv get -field=password secret/myproject/production/db
real-pa$$w0rd

To configure your Vault server, start by enabling the JWT Auth method:

$ vault auth enable jwt
Success! Enabled jwt auth method at: jwt/

Then create policies that allow you to read these secrets (one for each secret):

$ vault policy write myproject-staging - <<EOF
# Policy name: myproject-staging
#
# Read-only permission on 'secret/myproject/staging/*' path
path "secret/myproject/staging/*" {
  capabilities = [ "read" ]
}
EOF
Success! Uploaded policy: myproject-staging

$ vault policy write myproject-production - <<EOF
# Policy name: myproject-production
#
# Read-only permission on 'secret/myproject/production/*' path
path "secret/myproject/production/*" {
  capabilities = [ "read" ]
}
EOF
Success! Uploaded policy: myproject-production

You also need roles that link the JWT with these policies.

One for staging named myproject-staging:

$ vault write auth/jwt/role/myproject-staging - <<EOF
{
  "role_type": "jwt",
  "policies": ["myproject-staging"],
  "token_explicit_max_ttl": 60,
  "user_claim": "user_email",
  "bound_audiences": "https://vault.example.com",
  "bound_claims": {
    "project_id": "22",
    "ref": "master",
    "ref_type": "branch"
  }
}
EOF

And one for production named myproject-production:

$ vault write auth/jwt/role/myproject-production - <<EOF
{
  "role_type": "jwt",
  "policies": ["myproject-production"],
  "token_explicit_max_ttl": 60,
  "user_claim": "user_email",
  "bound_audiences": "https://vault.example.com",
  "bound_claims_type": "glob",
  "bound_claims": {
    "project_id": "22",
    "ref_protected": "true",
    "ref_type": "branch",
    "ref": "auto-deploy-*"
  }
}
EOF

This example uses bound claims to specify that only a JWT with matching values for the specified claims is allowed to authenticate.

Combined with protected branches, you can restrict who is able to authenticate and read the secrets.

Any of the claims included in the JWT can be matched against a list of values in the bound claims. For example:

"bound_claims": {
  "user_login": ["alice", "bob", "mallory"]
}

"bound_claims": {
  "ref": ["main", "develop", "test"]
}

"bound_claims": {
  "namespace_id": ["10", "20", "30"]
}

"bound_claims": {
  "project_id": ["12", "22", "37"]
}
  • If only namespace_id is used, all projects in the namespace are allowed. Nested projects are not included, so their namespace IDs must also be added to the list if needed.
  • If both namespace_id and project_id are used, Vault first checks if the project’s namespace is in namespace_id then checks if the project is in project_id.

token_explicit_max_ttl specifies that the token issued by Vault, upon successful authentication, has a hard lifetime limit of 60 seconds.

user_claim specifies the name for the Identity alias created by Vault upon a successful login.

bound_claims_type configures the interpretation of the bound_claims values. If set to glob, the values are interpreted as globs, with * matching any number of characters.

The claim fields listed in the table above can also be accessed for Vault’s policy path templating purposes by using the accessor name of the JWT auth in Vault. The mount accessor name (ACCESSOR_NAME in the example below) can be retrieved by running vault auth list.

Policy template example making use of a named metadata field named project_path:

path "secret/data/{{identity.entity.aliases.ACCESSOR_NAME.metadata.project_path}}/staging/*" {
  capabilities = [ "read" ]
}

Role example to support the templated policy above, mapping the claim field project_path as a metadata field through use of claim_mappings configuration:

{
  "role_type": "jwt",
  ...
  "claim_mappings": {
    "project_path": "project_path"
  }
}

For the full list of options, see Vault’s Create Role documentation.

caution
Always restrict your roles to project or namespace by using one of the provided claims (for example, project_id or namespace_id). Otherwise any JWT generated by this instance may be allowed to authenticate using this role.

Now, configure the JWT Authentication method:

$ vault write auth/jwt/config \
    oidc_discovery_url="https://gitlab.example.com" \
    bound_issuer="https://gitlab.example.com"

bound_issuer specifies that only a JWT with the issuer (that is, the iss claim) set to gitlab.example.com can use this method to authenticate, and that the oidc_discovery_url (https://gitlab.example.com) should be used to validate the token.

For the full list of available configuration options, see Vault’s API documentation.

In GitLab, create the following CI/CD variables to provide details about your Vault server:

  • VAULT_SERVER_URL - The URL of your Vault server, for example https://vault.example.com:8200.
  • VAULT_AUTH_ROLE - Optional. The role to use when attempting to authenticate. If no role is specified, Vault uses the default role specified when the authentication method was configured.
  • VAULT_AUTH_PATH - Optional. The path where the authentication method is mounted. Default is jwt.
  • VAULT_NAMESPACE - Optional. The Vault Enterprise namespace to use for reading secrets and authentication. If no namespace is specified, Vault uses the root (/) namespace. The setting is ignored by Vault Open Source.

The following job, when run for the default branch, can read secrets under secret/myproject/staging/, but not the secrets under secret/myproject/production/:

job_with_secrets:
  id_tokens:
    VAULT_ID_TOKEN:
      aud: https://vault.example.com
  secrets:
    STAGING_DB_PASSWORD:
      vault: secret/myproject/staging/db/password@secrets # authenticates using $VAULT_ID_TOKEN
  script:
    - access-staging-db.sh --token $STAGING_DB_PASSWORD

In this example:

  • id_tokens - The JSON Web Token (JWT) used for OIDC authentication. The aud claim is set to match the bound_audiences parameter of the Vault JWT authentication method.
  • @secrets - The vault name, where your Secrets Engines are enabled.
  • secret/myproject/staging/db - The path location of the secret in Vault.
  • password The field to be fetched in the referenced secret.

Limit token access to Vault secrets

You can control ID token access to Vault secrets by using Vault protections and GitLab features. For example, restrict the token by:

  • Using Vault bound audiences for specific ID token aud claims.
  • Using Vault bound claims for specific groups using group_claim.
  • Hard coding values for Vault bound claims based on the user_login and user_email of specific users.
  • Setting Vault time limits for TTL of the token as specified in token_explicit_max_ttl, where the token expires after authentication.
  • Scoping the JWT to GitLab protected branches that are restricted to a subset of project users.
  • Scoping the JWT to GitLab protected tags, that are restricted to a subset of project users.