»Transit Secrets Engine
The transit secrets engine handles cryptographic functions on data in-transit. Vault doesn't store the data sent to the secrets engine. It can also be viewed as "cryptography as a service" or "encryption as a service". The transit secrets engine can also sign and verify data; generate hashes and HMACs of data; and act as a source of random bytes.
The primary use case for
transit is to encrypt data from applications while
still storing that encrypted data in some primary data store. This relieves the
burden of proper encryption/decryption from application developers and pushes
the burden onto the operators of Vault.
Key derivation is supported, which allows the same key to be used for multiple purposes by deriving a new key based on a user-supplied context value. In this mode, convergent encryption can optionally be supported, which allows the same input values to produce the same ciphertext.
Datakey generation allows processes to request a high-entropy key of a given bit length be returned to them, encrypted with the named key. Normally this will also return the key in plaintext to allow for immediate use, but this can be disabled to accommodate auditing requirements.
»Working Set Management
The Transit engine supports versioning of keys. Key versions that are earlier
than a key's specified
min_decryption_version gets archived, and the rest of
the key versions belong to the working set. This is a performance consideration
to keep key loading fast, as well as a security consideration: by disallowing
decryption of old versions of keys, found ciphertext corresponding to obsolete
(but sensitive) data can not be decrypted by most users, but in an emergency
min_decryption_version can be moved back to allow for legitimate
Currently this archive is stored in a single storage entry. With some storage backends, notably those using Raft or Paxos for HA capabilities, frequent rotation may lead to a storage entry size for the archive that is larger than the storage backend can handle. For frequent rotation needs, using named keys that correspond to time bounds (e.g. five-minute periods floored to the closest multiple of five) may provide a good alternative, allowing for several keys to be live at once and a deterministic way to decide which key to use at any given time.
»NIST Rotation Guidance
Periodic rotation of the encryption keys is recommended, even in the absence of compromise. For AES-GCM keys, rotation should occur before approximately 232 encryptions have been performed by a key version, following the guidelines of NIST publication 800-38D. It is recommended that operators estimate the encryption rate of a key and use that to determine a frequency of rotation that prevents the guidance limits from being reached. For example, if one determines that the estimated rate is 40 million operations per day, then rotating a key every three months is sufficient.
As of now, the transit secrets engine supports the following key types (all key types also generate separate HMAC keys):
aes128-gcm96: AES-GCM with a 128-bit AES key and a 96-bit nonce; supports encryption, decryption, key derivation, and convergent encryption
aes256-gcm96: AES-GCM with a 256-bit AES key and a 96-bit nonce; supports encryption, decryption, key derivation, and convergent encryption (default)
chacha20-poly1305: ChaCha20-Poly1305 with a 256-bit key; supports encryption, decryption, key derivation, and convergent encryption
ed25519: Ed25519; supports signing, signature verification, and key derivation
ecdsa-p256: ECDSA using curve P-256; supports signing and signature verification
ecdsa-p384: ECDSA using curve P-384; supports signing and signature verification
ecdsa-p521: ECDSA using curve P-521; supports signing and signature verification
rsa-2048: 2048-bit RSA key; supports encryption, decryption, signing, and signature verification
rsa-3072: 3072-bit RSA key; supports encryption, decryption, signing, and signature verification
rsa-4096: 4096-bit RSA key; supports encryption, decryption, signing, and signature verification
Convergent encryption is a mode where the same set of plaintext+context always result in the same ciphertext. It does this by deriving a key using a key derivation function but also by deterministically deriving a nonce. Because these properties differ for any combination of plaintext and ciphertext over a keyspace the size of 2^256, the risk of nonce reuse is near zero.
This has many practical uses. One common usage mode is to allow values to be stored encrypted in a database, but with limited lookup/query support, so that rows with the same value for a specific field can be returned from a query.
To accommodate for any needed upgrades to the algorithm, different versions of convergent encryption have historically been supported:
- Version 1 required the client to provide their own nonce, which is highly flexible but if done incorrectly can be dangerous. This was only in Vault 0.6.1, and keys using this version cannot be upgraded.
- Version 2 used an algorithmic approach to deriving the parameters. However, the algorithm used was susceptible to offline plaintext-confirmation attacks, which could allow attackers to brute force decryption if the plaintext size was small. Keys using version 2 can be upgraded by simply performing a rotate operation to a new key version; existing values can then be rewrapped against the new key version and will use the version 3 algorithm.
- Version 3 uses a different algorithm designed to be resistant to offline plaintext-confirmation attacks. It is similar to AES-SIV in that it uses a PRF to generate the nonce from the plaintext.
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 Transit secrets engine:
$ vault secrets enable transitSuccess! Enabled the transit secrets engine at: transit/
By default, the secrets engine will mount at the name of the engine. To enable the secrets engine at a different path, use the
Create a named encryption key:
$ vault write -f transit/keys/my-keySuccess! Data written to: transit/keys/my-key
Usually each application has its own encryption key.
After the secrets engine is configured and a user/machine has a Vault token with the proper permission, it can use this secrets engine.
Encrypt some plaintext data using the
/encryptendpoint with a named key:
NOTE: All plaintext data must be base64-encoded. The reason for this requirement is that Vault does not require that the plaintext is "text". It could be a binary file such as a PDF or image. The easiest safe transport mechanism for this data as part of a JSON payload is to base64-encode it.
$ vault write transit/encrypt/my-key plaintext=$(base64 <<< "my secret data") Key Value--- -----ciphertext vault:v1:8SDd3WHDOjf7mq69CyCqYjBXAiQQAVZRkFM13ok481zoCmHnSeDX9vyf7w==
The returned ciphertext starts with
vault:v1:. The first prefix (
vault) identifies that it has been wrapped by Vault. The
v1indicates the key version 1 was used to encrypt the plaintext; therefore, when you rotate keys, Vault knows which version to use for decryption. The rest is a base64 concatenation of the initialization vector (IV) and ciphertext.
Note that Vault does not store any of this data. The caller is responsible for storing the encrypted ciphertext. When the caller wants the plaintext, it must provide the ciphertext back to Vault to decrypt the value.
Vault HTTP API imposes a maximum request size of 32MB to prevent a denial of service attack. This can be tuned per
listenerblock in the Vault server configuration.
Decrypt a piece of data using the
/decryptendpoint with a named key:
$ vault write transit/decrypt/my-key ciphertext=vault:v1:8SDd3WHDOjf7mq69CyCqYjBXAiQQAVZRkFM13ok481zoCmHnSeDX9vyf7w== Key Value--- -----plaintext bXkgc2VjcmV0IGRhdGEK
The resulting data is base64-encoded (see the note above for details on why). Decode it to get the raw plaintext:
$ base64 --decode <<< "bXkgc2VjcmV0IGRhdGEK"my secret data
It is also possible to script this decryption using some clever shell scripting in one command:
$ vault write -field=plaintext transit/decrypt/my-key ciphertext=... | base64 --decodemy secret data
Using ACLs, it is possible to restrict using the transit secrets engine such that trusted operators can manage the named keys, and applications can only encrypt or decrypt using the named keys they need access to.
Rotate the underlying encryption key. This will generate a new encryption key and add it to the keyring for the named key:
$ vault write -f transit/keys/my-key/rotateSuccess! Data written to: transit/keys/my-key/rotate
Future encryptions will use this new key. Old data can still be decrypted due to the use of a key ring.
Upgrade already-encrypted data to a new key. Vault will decrypt the value using the appropriate key in the keyring and then encrypted the resulting plaintext with the newest key in the keyring.
$ vault write transit/rewrap/my-key ciphertext=vault:v1:8SDd3WHDOjf7mq69CyCqYjBXAiQQAVZRkFM13ok481zoCmHnSeDX9vyf7w== Key Value--- -----ciphertext vault:v2:0VHTTBb2EyyNYHsa3XiXsvXOQSLKulH+NqS4eRZdtc2TwQCxqJ7PUipvqQ==
This process does not reveal the plaintext data. As such, a Vault policy could grant almost an untrusted process the ability to "rewrap" encrypted data, since the process would not be able to get access to the plaintext data.
Refer to the Encryption as a Service: Transit Secrets Engine guide for a step-by-step tutorial.
The Transit secrets engine has a full HTTP API. Please see the Transit secrets engine API for more details.