17 KiB
layout | page_title | sidebar_title | sidebar_current | description |
---|---|---|---|---|
docs | Google Cloud - Secrets Engines | Google Cloud | docs-secrets-gcp | The Google Cloud secrets engine for Vault dynamically generates Google Cloud service account keys and OAuth tokens based on IAM policies. |
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 or specifying the credentials JSON file. For more information on authentication, see the authentication section below.
-
Configure a roleset. Rolesets determine the permissions that Service Account credentials generated by Vault will have on GCP resources.
To configure a roleset that generates OAuth2 access tokens (preferred):
$ vault write gcp/roleset/my-token-roleset \ project="my-project" \ 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 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 roleset bindings section below.
For more information on the differences between OAuth2 access tokens and Service Account keys, see the things to note section below.
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 roleset was configured, you can generate OAuth2 tokens or service account keys.
Access Tokens
To generate OAuth2 tokens, read from gcp/token/...
. The roleset must have been
created as type access_token
:
$ vault read gcp/token/my-token-roleset
Key Value
--- -----
expires_at_seconds 1537402548
token ya29.c.ElodBmNPwHUNY5gcBpnXcE4ywG4w1k...
token_ttl 3599
This endpoint generates a non-renewable, non-revocable static OAuth2 access token
with a 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/...
. The roleset must have
been created as type service_account_key
:
$ vault read gcp/key/my-key-roleset
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 roleset'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.
Roleset Bindings
Roleset bindings define a list of resources and the associated IAM roles on that
resource. Roleset bindings are used as the binding
argument when creating or
updating a rolset 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",
]
}
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 zome 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
-
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 environment variables
GOOGLE_APPLICATION_CREDENTIALS
. 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. -
Default instance credentials. When no environment variable is present, the default service account credentials are used. This is useful when running Vault on Google Compute Engine or Google Kubernetes Engine
For more information on service accounts, please see the Google Cloud Service Accounts documentation.
To use this storage backend, the service account must have the following minimum scope(s):
https://www.googleapis.com/auth/cloud-platform
https://www.googleapis.com/auth/iam
Required Permissions
The credentials given to Vault must have the following permissions:
# 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
# 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
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 assignroles/iam.serviceAccountAdmin
androles/iam.serviceAccountKeyAdmin
so Vault can manage service accounts and keys.
Things to Note
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 follwing:
-
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 Must Exist at Roleset Creation
Because we the bindings for the Service Account are set during roleset 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 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 otherlease_*
attributes when reading responses for the access tokens secrets endpoint (i.e. fromgcp/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.