Data security in Boundary

Boundary has multiple mechanisms to ensure secure end-to-end behavior of the system. A key part of this is support for various Key Management Systems that protect the base encryption keys used for various functions. This page describes the various KMS key purposes that Boundary supports and how they are used within the system.

The worker-auth-storage KMS key

The worker-auth-storage KMS key is used by a PKI Worker for storage of authentication keys. It is optional for PKI workers; if not specified the authentication keys will not be encrypted on disk. This is not used by KMS workers.

The root KMS key and per-scope KEK/DEKs ^{Community Edition only}

Following best practices of using different encryption keys for different purposes, Boundary has a number of encryption keys generated within each scope. Each key further contains versions, which hold the key material used to encrypt data. You can create new key versions by rotating the keys. A key version can be destroyed, which will re-encrypt any existing data before destroying the key material permanently.

The root KMS key acts as a KEK (Key Encrypting Key) for the scope-specific KEKs (also referred to as the scope's root key). The scope's root KEK and the various DEKs (Data Encryption Keys) are created when a scope is created. The DEKs are encrypted with the scope's root KEK, and this is in turn encrypted with the KMS key marked for the root purpose.

The current scoped DEKs and their purposes are detailed below:

• audit: This is used to encrypt secret values in the event log. For more information about the event log, refer to the events config.

• database: This is the general-purpose DEK used to encrypt values within the database. Values that are encrypted are those generally considered to be secret, such as API keys, third-party tokens, certificate private keys, and so on.

• oidc: This is used to encrypt OIDC information in cookies and authentication requests.

• oplog: This is used for encrypting oplog (operation log) values for the given scope.

• tokens: This is used for encrypting tokens generated by auth methods within the given scope.

• sessions: This is used as a base key against which to derive session-specific encryption keys.

Key version lifecycle management

You can control the lifecycles of the per-scope KEK and DEKs via the CLI or API using the key endpoints under the scopes section. To rotate all the keys in a scope, use the rotate-keys endpoint:

$ boundary scopes rotate-keys -scope-id p_A4jfDjZ9jf

This endpoint creates new key versions for all keys in the scope p_A4jfDjZ9jf, and makes these key versions the active key versions for encrypting new data. The previous key version(s) will continue to be used for decrypting existing data. You can use the -rewrap flag to immediately rewrap all DEK versions with the new KEK version. Otherwise, the DEK versions remain encrypted by the prior KEK version from when they were created.

To list all keys in a scope and their versions, use the list-keys endpoint:

$ boundary scopes list-keys -scope-id p_A4jfDjZ9jf

To destroy a key version, use the destroy-key-version endpoint:

$ boundary scopes destroy-key-version -scope-id p_A4jfDjZ9jf -key-version-id kdkv_tr6ZN8opYr

The latest key version cannot be destroyed, so if you want to destroy the latest key version, you will need to call the rotate-keys endpoint first, to create a new set of key versions. The oplog purpose key versions also cannot be destroyed at this time.

Destroying a key version sometimes requires background work before it can be completed. This is because DEK versions that currently encrypt data necessitate re-encrypting that data with the latest DEK version for each purpose before they can be deleted. You can monitor the progress of this re-encryption job via the list-key-destruction-jobs endpoint:

$ boundary scopes list-key-version-destruction-jobs -scope-id p_A4jfDjZ9jf

Once the job disappears from this list, the associated key version will have been destroyed and any existing data will have been re-encrypted.

The bsr KMS key ^HCP/ENT

The bsr KMS key is required for session recording. If you do not add a bsr key to your controller configuration, you will receive an error when you attempt to enable session recording. The key is used for encrypting data and checking the integrity of recordings.

The previous-root KMS key ^{Community Edition only}

The previous-root KMS key is used when migrating to a new root key. Adding the previous-root KMS key to your configuration informs the Controller to use it for decrypting the existing information in the database, allowing you to rotate and rewrap the KEKs to complete the migration to the new root key.

The worker-auth KMS key ^{Community Edition only}

The worker-auth KMS key is a key shared by the Controller and Worker in order to authenticate a Worker to the Controller. Specifics of this mechanism can be found on the Connections/TLS page. If a worker is used with PKI authentication this is unnecessary.

The recovery KMS key ^{Community Edition only}

The recovery KMS key is used for rescue/recovery operations that can be used by a client to authenticate almost any operation within Boundary. Its mechanism of operation is very similar to the worker-auth flow in terms of using a shared KMS between the client and the Controller for authentication. A nonce and creation time are included as an encrypted payload, formatted as a token and sent to the Controller. The time and nonce are used to ensure that a value cannot be replayed by an adversary, and also to ensure that each operation must be individually authenticated by a client so that revoking access to the KMS has an immediate result.

On the client side, a user can use the -recovery-config flag with any operation on the CLI to specify a configuration file containing a suitable kms block. This functionality is also accessible via the Go SDK.

There are some other situations where this mechanism can be useful. For example, it is possible to use this mechanism, along with some defaults in the Terraform provider, to ensure that everything in Boundary is created through Terraform, with the exception of resources that cannot themselves be deleted. (This consists of built-in anonymous (u_anon), authenticated (u_auth), and recovery (u_recovery) users and the global scope.). By initializing Boundary with the options to skip creating default resources, Terraform can be used to create the specific resources needed instead, with the recovery KMS used to authenticate setting up the initial auth method(s).

The config KMS key ^{Community Edition only}

This key can be used to encrypt values within Boundary's configuration file. By sharing this block between Boundary and an operator, the operator can put sensitive or secret values (such as cloud API keys for KMSes) in Boundary's configuration file, run boundary config encrypt to encrypt the file, and then safely pass the file to a change control system. Only another operator or system with access to that KMS can decrypt the values. Boundary will check for a config KMS block on startup, and if it exists, will use it to decrypt any encrypted values found at startup time.