Google Cloud secrets engine

The Google Cloud Vault secrets engine dynamically generates Google Cloud service account keys and OAuth tokens based on IAM policies. This enables users to gain access to Google Cloud resources without needing to create or manage a dedicated service account.

The benefits of using this secrets engine to manage Google Cloud IAM service accounts are:

Automatic cleanup of GCP IAM service account keys - each Service Account key is associated with a Vault lease. When the lease expires (either during normal revocation or through early revocation), the service account key is automatically revoked.
Quick, short-term access - users do not need to create new GCP Service Accounts for short-term or one-off access (such as batch jobs or quick introspection).
Multi-cloud and hybrid cloud applications - users authenticate to Vault using a central identity service (such as LDAP) and generate GCP credentials without the need to create or manage a new Service Account for that user.

NOTE: Deprecation of access_token Leases: In previous versions of this secrets engine (released with Vault <= 0.11.1), a lease was generated with access tokens. If you're using an old version of the plugin, please upgrade. Read more in the upgrade guide

Setup

Most secrets engines must be configured in advance before they can perform their functions. These steps are usually completed by an operator or configuration management tool.

Enable the Google Cloud secrets engine:
```
$ vault secrets enable gcp
Success! Enabled the gcp secrets engine at: gcp/
```
By default, the secrets engine will mount at the name of the engine. To enable the secrets engine at a different path, use the -path argument.
Configure the secrets engine with account credentials, or leave blank or unwritten to use Application Default Credentials.
```
$ vault write gcp/config credentials=@my-credentials.json
Success! Data written to: gcp/config
```
If you are running Vault from inside Google Compute Engine or Google Kubernetes Engine, the instance or pod service account can be used in place of specifying the credentials JSON file. For more information on authentication, see the authentication section below.
In some cases, you cannot set sensitive IAM security credentials in your Vault configuration. For example, your organization may require that all security credentials are short-lived or explicitly tied to a machine identity.
To provide IAM security credentials to Vault, we recommend using Vault plugin workload identity federation (WIF) as shown below.

Alternatively, configure the audience claim value and the service account email to assume for plugin workload identity federation:

$ vault write gcp/config \
    identity_token_audience="<TOKEN AUDIENCE>" \
    service_account_email="<SERVICE ACCOUNT EMAIL>"

Vault's identity token provider signs the plugin identity token JWT internally. If a trust relationship exists between Vault and GCP through WIF, the secrets engine can exchange the Vault identity token for a federated access token.

To configure a trusted relationship between Vault and GCP:

- You must configure the [identity token issuer backend](/vault/api-docs/secret/identity/tokens#configure-the-identity-tokens-backend)
  for Vault.
- GCP must have a
  [workload identity pool and provider](https://cloud.google.com/iam/docs/manage-workload-identity-pools-providers)
  configured with information about the fully qualified and network-reachable
  issuer URL for the Vault plugin's
  [identity token provider](/vault/api-docs/secret/identity/tokens#read-plugin-identity-well-known-configurations).

Establishing a trusted relationship between Vault and GCP ensures that GCP can fetch JWKS public keys and verify the plugin identity token signature.

Configure rolesets or static accounts. See the relevant sections below.

Rolesets

A roleset consists of a Vault managed GCP Service account along with a set of IAM bindings defined for that service account. The name of the service account is generated based on the time of creation or update. You should not depend on the name of the service account being fixed and should manage all IAM bindings for the service account through the bindings parameter when creating or updating the roleset.

For more information on the differences between rolesets and static accounts, see the things to note section below.

Roleset policy considerations

Starting with Vault 1.8.0, existing permissive policies containing globs for the GCP Secrets Engine may grant additional privileges due to the introduction of /gcp/roleset/:roleset/token and /gcp/roleset/:roleset/key endpoints.

The following policy grants a user the ability to read all rolesets, but would also allow them to generate tokens and keys. This type of policy is not recommended:

# DO NOT USE
path "/gcp/roleset/*" {
    capabilities = ["read"]
}

The following example demonstrates how a wildcard can instead be used in a roleset policy to adhere to the principle of least privilege:

path "/gcp/roleset/+" {
    capabilities = ["read"]
}

For more more information on policy syntax, see the policy documentation.

Examples

To configure a roleset that generates OAuth2 access tokens (preferred):

$ vault write gcp/roleset/my-token-roleset \
    project="my-project-id" \
    secret_type="access_token"  \
    token_scopes="https://www.googleapis.com/auth/cloud-platform" \
    bindings=-<<EOF
      resource "//cloudresourcemanager.googleapis.com/projects/my-project-id" {
        roles = ["roles/viewer"]
      }
    EOF

To configure a roleset that generates GCP Service Account keys:

$ vault write gcp/roleset/my-key-roleset \
    project="my-project" \
    secret_type="service_account_key"  \
    bindings=-<<EOF
      resource "//cloudresourcemanager.googleapis.com/projects/my-project" {
        roles = ["roles/viewer"]
      }
    EOF

Alternatively, provide a file for the bindings argument like so:

$ vault write gcp/roleset/my-roleset
    bindings=@mybindings.hcl
    ...

For more information on role bindings and sample role bindings, please see the bindings section below.

For more information on the differences between OAuth2 access tokens and Service Account keys, see the things to note section below.

For more information on creating and managing rolesets, see the GCP secrets engine API docs docs.

Static accounts

Static accounts are GCP service accounts that are created outside of Vault and then provided to Vault to generate access tokens or keys. You can also use Vault to optionally manage IAM bindings for the service account.

For more information on the differences between rolesets and static accounts, see the things to note section below.

Examples

Before configuring a static account, you need to create a Google Cloud Service Account. Take note of the email address of the service account you have created. Service account emails are of the format <service-account-id>@<project-id>.iam.gserviceaccount.com.

To configure a static account that generates OAuth2 access tokens (preferred):

$ vault write gcp/static-account/my-token-account \
    service_account_email="account@my-project.iam.gserviceaccount.com" \
    secret_type="access_token"  \
    token_scopes="https://www.googleapis.com/auth/cloud-platform" \
    bindings=-<<EOF
      resource "//cloudresourcemanager.googleapis.com/projects/my-project" {
        roles = ["roles/viewer"]
      }
    EOF

To configure a static account that generates GCP Service Account keys:

$ vault write gcp/static-account/my-key-account \
    service_account_email="account@my-project.iam.gserviceaccount.com" \
    secret_type="service_account_key"  \
    bindings=-<<EOF
      resource "//cloudresourcemanager.googleapis.com/projects/my-project" {
        roles = ["roles/viewer"]
      }
    EOF

Alternatively, provide a file for the bindings argument like so:

$ vault write gcp/static-account/my-account
    bindings=@mybindings.hcl
    ...

For more information on role bindings and sample role bindings, please see the bindings section below.

For more information on the differences between OAuth2 access tokens and Service Account keys, see the things to note section below.

For more information on creating and managing static accounts, see the GCP secrets engine API docs docs.

Impersonated accounts

Impersonated accounts are a way to generate an OAuth2 access token that is granted the permissions and accesses of another given service account. These access tokens do not have the same 10-key limit as service account keys do, yet they retain their short-lived nature. By default, their TTL in GCP is 1 hour, but this may be configured to be up to 12 hours as explained in Google's short-lived credentials documentation.

For more information regarding service account impersonation in GCP, consider starting with their documentation available here.

Examples

To configure a Vault role that impersonates the administrator on the Google Cloud project with the cloud platform and compute scopes:

$ vault write gcp/impersonated-account/my-token-impersonate \
    service_account_email="projectAdmin@my-project.iam.gserviceaccount.com" \
    token_scopes="https://www.googleapis.com/auth/cloud-platform,https://www.googleapis.com/auth/compute" \
    ttl="6h"

Usage

After the secrets engine is configured and a user/machine has a Vault token with the proper permission, it can generate credentials. Depending on how the Vault role was configured, you can generate OAuth2 tokens or service account keys.

Access tokens

To generate OAuth2 access tokens, read from the gcp/.../token API. If using a roleset or static account, it must have been created with a secret_type of access_token. Impersonated accounts will generate OAuth2 tokens by default.

Roleset:

$ vault read gcp/roleset/my-token-roleset/token

Key                Value
---                -----
expires_at_seconds    1537402548
token                 ya29.c.ElodBmNPwHUNY5gcBpnXcE4ywG4w1k...
token_ttl             3599

Static account:

$ vault read gcp/static-account/my-token-account/token

Key                Value
---                -----
expires_at_seconds    1672231587
token                 ya29.c.b0Aa9VdykAdYoW9S1ImtPZykF_oTi9...
token_ttl             3599

Impersonated account:

$ vault read gcp/impersonated-account/my-token-impersonate/token

Key                Value
---                -----
expires_at_seconds    1671667844
token                 ya29.c.b0AT7lpjBRmO7ghBEyMV18evd016hq...
token_ttl             59m59s

This endpoint generates a non-renewable, non-revocable static OAuth2 access token with a max lifetime of one hour, where token_ttl is given in seconds and the expires_at_seconds is the expiry time for the token, given as a Unix timestamp. The token value then can be used as a HTTP Authorization Bearer token in requests to GCP APIs:

$ curl -H "Authorization: Bearer ya29.c.ElodBmNPwHUNY5gcBpnXcE4ywG4w1k..."

Service account keys

To generate service account keys, read from gcp/.../key. Vault returns the service account key data as a base64-encoded string in the private_key_data field. This can be read by decoding it using base64 --decode "ewogICJ0e..." or another base64 tool of your choice. The roleset or static account must have been created as type service_account_key:

$ vault read gcp/roleset/my-key-roleset/key

Key                 Value
---                 -----
lease_id            gcp/key/my-key-roleset/ce563a99-5e55-389b...
lease_duration      30m
lease_renewable     true
key_algorithm       KEY_ALG_RSA_2048
key_type            TYPE_GOOGLE_CREDENTIALS_FILE
private_key_data    ewogICJ0eXBlIjogInNlcnZpY2VfYWNjb3VudCIsC...

This endpoint generates a new GCP IAM service account key associated with the role's Service Account. When the lease expires (or is revoked early), the Service Account key will be deleted.

There is a default limit of 10 keys per Service Account. For more information on this limit and recommended mitigation, please see the things to note section below.

Bindings

Roleset or static account bindings define a list of resources and the associated IAM roles on that resource. Bindings are used as the binding argument when creating or updating a roleset or static account and are specified in the following format using HCL:

resource NAME {
  roles = [ROLE, [ROLE...]]
}

For example:

resource "buckets/my-bucket" {
  roles = [
    "roles/storage.objectAdmin",
    "roles/storage.legacyBucketReader",
  ]
}

# At instance level, using self-link
resource "https://www.googleapis.com/compute/v1/projects/my-project/zone/my-zone/instances/my-instance" {
  roles = [
    "roles/compute.instanceAdmin.v1"
  ]
}

# At project level
resource "//cloudresourcemanager.googleapis.com/projects/my-project" {
  roles = [
    "roles/compute.instanceAdmin.v1",
    "roles/iam.serviceAccountUser",  # required if managing instances that run as service accounts
  ]
}

# At folder level
resource "//cloudresourcemanager.googleapis.com/folders/123456" {
  roles = [
    "roles/compute.viewer",
    "roles/deploymentmanager.viewer",
  ]
}

The top-level resource block defines the resource or resource path for which IAM policy information will be bound. The resource path may be specified in a few different formats:

Project-level self-link - a URI with scheme and host, generally corresponding to the self_link attribute of a resource in GCP. This must include the resource nested in the parent project.
```
# compute alpha zone
https://www.googleapis.com/compute/alpha/projects/my-project/zones/us-central1-c
```

Full resource name - a schema-less URI consisting of a DNS-compatible API service name and resource path. See the full resource name API documentation for more information.

# Compute snapshot
//compute.googleapis.com/project/my-project/snapshots/my-compute-snapshot

# Pubsub snapshot
//pubsub.googleapis.com/project/my-project/snapshots/my-pubsub-snapshot

# BigQuery dataset
//bigquery.googleapis.com/projects/my-project/datasets/mydataset

# Resource manager
//cloudresourcemanager.googleapis.com/projects/my-project"

Relative resource name - A path-noscheme URI path, usually as accepted by the API. Use this if the version or service are apparent from the resource type. Please see the relative resource name API documentation for more information.
```
# Storage bucket objects
buckets/my-bucket
buckets/my-bucket/objects/my-object

# PubSub topics
projects/my-project/topics/my-pubsub-topic
```

The nested roles attribute is an array of strings names of GCP IAM roles. The roles may be specified in the following formats:

Global role name - these are global roles built into Google Cloud. For the full list of available roles, please see the list of predefined GCP roles.
```
roles/viewer
roles/bigquery.user
roles/billing.admin
```
Organization-level custom role - these are roles that are created at the organization level by organization owners.
```
organizations/my-organization/roles/my-custom-role
```
For more information, please see the documentation on GCP custom roles.
Project-level custom role - these are roles that are created at a per-project level by project owners.
```
projects/my-project/roles/my-custom-role
```
For more information, please see the documentation on GCP custom roles.

Authentication

The Google Cloud Vault secrets backend uses the official Google Cloud Golang SDK. This means it supports the common ways of providing credentials to Google Cloud. In addition to specifying credentials directly via Vault configuration, you can also get configuration from the following values on the Vault server:

The GOOGLE_APPLICATION_CREDENTIALS environment variable. This is specified as the path to a Google Cloud credentials file, typically for a service account. If this environment variable is present, the resulting credentials are used. If the credentials are invalid, an error is returned.
The identity of a Google Cloud workload. When Vault server is running on a Google workload like Google Compute Engine or Google Kubernetes Engine, identity associated with the workload is automatically used. To configure Google Compute Engine with an identity, see attached service accounts. To configure Google Kubernetes Engine with an identity, see GKE workload identity.

For more information on service accounts, please see the Google Cloud Service Accounts documentation.

To use this secrets engine, the service account must have the following minimum scope(s):

https://www.googleapis.com/auth/cloud-platform

Required permissions

The credentials given to Vault must have the following permissions when using rolesets at the project level:

# Service account + key admin
iam.serviceAccounts.create
iam.serviceAccounts.delete
iam.serviceAccounts.get
iam.serviceAccounts.list
iam.serviceAccounts.update
iam.serviceAccountKeys.create
iam.serviceAccountKeys.delete
iam.serviceAccountKeys.get
iam.serviceAccountKeys.list

When using static accounts or impersonated accounts, Vault must have the following permissions at the service account level:

# For `access_token` secrets and impersonated accounts
iam.serviceAccounts.getAccessToken

# For `service_account_keys` secrets
iam.serviceAccountKeys.create
iam.serviceAccountKeys.delete
iam.serviceAccountKeys.get
iam.serviceAccountKeys.list

When using rolesets or static accounts with bindings, Vault must have the following permissions:

# IAM policy changes
<service>.<resource>.getIamPolicy
<service>.<resource>.setIamPolicy

where <service> and <resource> correspond to permissions which will be granted, for example:

# Projects
resourcemanager.projects.getIamPolicy
resourcemanager.projects.setIamPolicy

# All compute
compute.*.getIamPolicy
compute.*.setIamPolicy

# BigQuery datasets
bigquery.datasets.get
bigquery.datasets.update

You can either:

Create a custom role using these permissions, and assign this role at a project-level
Assign the set of roles required to get resource-specific getIamPolicy/setIamPolicy permissions. At a minimum you will need to assign roles/iam.serviceAccountAdmin and roles/iam.serviceAccountKeyAdmin so Vault can manage service accounts and keys.
Notice that BigQuery requires different permissions than other resource. This is because BigQuery currently uses legacy ACL instead of traditional IAM permissions. This means to update access on the dataset, Vault must be able to update the dataset's metadata.

Plugin Workload Identity Federation (WIF)

This feature requires Vault Enterprise(opens in new tab).

The GCP secrets engine supports the plugin WIF workflow and has a source of identity called a plugin identity token. The plugin identity token is a JWT that is signed internally by Vault's plugin identity token issuer.

If there is a trust relationship configured between Vault and GCP through workload identity federation, the secrets engine can exchange its identity token for short-lived access tokens needed to perform its actions.

Exchanging identity tokens for access tokens lets the GCP secrets engine operate without configuring explicit access to sensitive IAM security credentials.

To configure the secrets engine to use plugin WIF:

Ensure that Vault openid-configuration and public JWKS APIs are network-reachable by GCP. We recommend using an API proxy or gateway if you need to limit Vault API exposure.
Create a workload identity pool and provider in GCP.
1. The provider URL must point at your Vault plugin identity token issuer with the /.well-known/openid-configuration suffix removed. For example: https://host:port/v1/identity/oidc/plugins.
2. Uniquely identify the recipient of the plugin identity token as the audience. You can use the default audience for the identity pool or a custom value less than 256 characters.
Authenticate a workload in GCP by granting the identity pool access to a dedicated service account using service account impersonation. Filter requests using the unique sub claim issued by plugin identity tokens so the GCP Auth engine can impersonate the service account. sub claims have the form: plugin-identity:<NAMESPACE>:secret:<GCP_SECRETS_MOUNT_ACCESSOR>.

Configure the GCP secrets engine with the OIDC audience value and service account email.

$ vault write gcp/config \
  identity_token_audience="//iam.googleapis.com/projects/410449834127/locations/global/workloadIdentityPools/vault-gcp-secrets-43777a63/providers/vault-gcp-secrets-wif-provider" \
  service_account_email="vault-plugin-wif-secrets@hc-b712f250b4e04cacbadd258a90b.iam.gserviceaccount.com"

Your secrets engine can now use plugin WIF for its configuration credentials. By default, WIF credentials have a time-to-live of 1 hour and automatically refresh when they expire.

Please see the API documentation for more details on the fields associated with plugin WIF.

Root credential rotation

If the mount is configured with credentials directly, the credential's key may be rotated to a Vault-generated value that is not accessible by the operator. For more details on this operation, please see the Root Credential Rotation API docs.

Things to note

Rolesets vs. static accounts

Advantages of rolesets:

Service accounts and IAM bindings are fully managed by Vault

Disadvantages of rolesets:

Cannot easily decouple IAM bindings from the ones managed in Vault
Vault requires permissions to manage IAM bindings and service accounts

Advantages of static accounts:

Can manage IAM bindings independently from the ones managed in Vault
Vault does not require permissions to IAM bindings and service accounts and only permissions related to the keys of the service account

Disadvantages of static accounts:

Self management of service accounts is necessary.

Access tokens vs. service account keys

Advantages of access_tokens:

Can generate infinite number of tokens per roleset

Disadvantages of access_tokens:

Cannot be used with some client libraries or tools
Have a static life-time of 1 hr that cannot be modified, revoked, or extended.

Advantages of service_account_keys:

Controllable life-time through Vault, allowing for longer access
Can be used by all normal GCP tooling

Disadvantages of service_account_keys:

Infinite lifetime in GCP (i.e. if they are not managed properly, leaked keys can live forever)
Limited to 10 per roleset/service account.

When generating OAuth access tokens, Vault will still generate a dedicated service account and key. This private key is stored in Vault and is never accessible to other users, and the underlying key can be rotated. See the GCP API documentation for more information on rotation.

Service accounts are tied to rolesets

Service Accounts are created when the roleset is created (or updated) rather than each time a secret is generated. This may be different from how other secrets engines behave, but it is for good reasons:

IAM Service Account creation and permission propagation can take up to 60 seconds to complete. By creating the Service Account in advance, we speed up the timeliness of future operations and reduce the flakiness of automated workflows.
Each GCP project has a limit on the number of IAM Service Accounts. You can request additional quota. The quota increase is processed by humans, so it is best to request this additional quota in advance. This limit is currently 100, including system-managed Service Accounts. If Service Accounts were created per secret, this quota limit would reduce the number of secrets that can be generated.

Service account keys quota limits

GCP IAM has a hard limit (currently 10) on the number of Service Account keys. Attempts to generate more keys will result in an error. If you find yourself running into this limit, consider the following:

Have shorter TTLs or revoke access earlier. If you are not using past Service Account keys, consider rotating and freeing quota earlier.
Create additional rolesets which share the same set of permissions. Additional rolesets can be created with the same set of permissions. This will create a new service account and increases the number of keys you can create.
Where possible, use OAuth2 access tokens instead of Service Account keys.

Resources in IAM bindings must exist at roleset or static account creation

Because the bindings for the Service Account are set during roleset/static account creation, resources that do not exist will fail the getIamPolicy API call.

Roleset creation may partially fail

Every Service Account creation, key creation, and IAM policy change is a GCP API call per resource. If an API call to one of these resources fails, the roleset creation fails and Vault will attempt to rollback.

These rollbacks are API calls, so they may also fail. The secrets engine uses a WAL to ensure that unused bindings are cleaned up. In the case of quota limits, you may need to clean these up manually.

Do not modify vault-owned IAM accounts

While Vault will initially create and assign permissions to IAM service accounts, it is possible that an external user deletes or modifies this service account. These changes are difficult to detect, and it is best to prevent this type of modification through IAM permissions.

Vault roleset Service Accounts will have emails in the format:

vault<roleset-prefix>-<creation-unix-timestamp>@...

Communicate with your teams (or use IAM permissions) to not modify these resources.

Help & support

The Google Cloud Vault secrets engine is written as an external Vault plugin and thus exists outside the main Vault repository. It is automatically bundled with Vault releases, but the code is managed separately.

Please report issues, add feature requests, and submit contributions to the vault-plugin-secrets-gcp repo on GitHub.

API

The GCP secrets engine has a full HTTP API. Please see the GCP secrets engine API docs for more details.

Upgrade guides

Deprecation of access token leases

NOTE: This deprecation only affects access tokens. There is no change to the service_account_key secret type.

Previous versions of this secrets engine (Vault <= 0.11.1) created a lease for each access token secret. We have removed them after discovering that these tokens, specifically Google OAuth2 tokens for IAM service accounts, are non-revocable and have a static 60 minute lifetime. To match the current limitations of the GCP APIs, the secrets engine will no longer allow for revocation or manage the token TTL - more specifically, the access_token response will no longer include lease_id or other lease information. This change does not reflect any change to the actual underlying OAuth tokens or GCP service accounts.

To upgrade:

Remove references from lease_id, lease_duration or other lease_* attributes when reading responses for the access tokens secrets endpoint (i.e. from gcp/token/$roleset). See the documentation for access tokens to see the new format for the response.
Be aware of leftover leases from previous versions. While these old leases will still be revocable, they will not actually invalidate their associated access token, and that token will still be useable for up to one hour.