Additional authenticated data (AAD) is any string that you pass to Cloud Key Management Service as part of an encrypt or decrypt request. AAD is used as an integrity check and can help protect your data from a confused deputy attack. The AAD string is base64-encoded and must be no larger than 64 KiB.
Cloud KMS will not decrypt ciphertext unless the same AAD value is used for both encryption and decryption.
AAD is bound to the encrypted data, because you cannot decrypt the ciphertext unless you know the AAD, but it is not stored as part the ciphertext. AAD also does does not increase the cryptographic strength of the ciphertext. Instead it is an additional check by Cloud KMS to authenticate a decryption request.
In Cloud KMS, AAD is always present when you make a call to encrypt or decrypt - if you do not provide a value for AAD, then an empty string is used. If an empty string is used as the AAD for the encryption of plaintext, then only an empty string will allow decryption of the ciphertext.
AAD is not logged by Cloud Audit Logging.
When to use AAD
One example of using AAD is when your application serves as a wrap/unwrap proxy with a single key and an unbounded number of clients, with each client in distinct security boundaries. For example, the application could be a diary application that allows users to maintain a private diary. When a user needs to view a private diary entry, the application can use the unique user name as the AAD in the unwrap (decrypt) request to explicitly authenticate the user. In this scenario you can use a single key to serve multiple (unbounded) users. A primary benefit is you don't need to maintain state for individual users.
Another example is if your application needs to use bearer tokens that contain some private information, such as an email address. Inputs to the bearer token would be the authenticated data used for a bearer token plus the plaintext email address. These inputs would then get encrypted so the bearer token that gets exchanged is in the form of Additional Encrypted Authenticated Data (AEAD).
Confused deputy attack example
This example shows how an application could be tricked into decrypting ciphertext on behalf of a malicious user. The application is the confused deputy in this example because the application is unaware the attacker has tricked the application into misusing its authority. The result is the attacker is able to view decrypted data that originally was encrypted for another user. Note in this attack, the attacker does not need to know the encryption key, because it relies on the confused deputy to perform the decryption.
A diary application allows users to maintain a private diary. Each diary entry is encrypted and intended to be decrypted only by the user that created the diary entry.
Alice creates a diary entry. The application encrypts the diary entry and then stores the encrypted diary entry at a location reserved for diary entries that belong to Alice.
Alice sends a request to view her diary entry. Because the encrypted diary entry is at a location reserved for Alice, the application decrypts the data and returns it as the response to Alice's request. This is the intended behaviour of the application.
Mallory copies Alice's diary entry from the location reserved for Alice to the location reserved for Mallory.
Mallory sends a request to view Mallory's copy of Alice's diary entry. Because the copy of Alice's diary entry is at a location reserved for Mallory, the application decrypts the diary entry and returns the plaintext as the response to Mallory's request. Mallory can then view Alice's diary entry, which is not the intended behaviour of the application.
To defend against this type of attack, the application can use a non-empty string as AAD for encryption and decryption. The AAD provides an extra check for decryption requests. When Mallory sends the decryption request to view Mallory's copy of Alice's diary entry, Mallory's request will not succeed unless Mallory can also trick the application into using the correct AAD.
Storing or reproducing AAD
Before encrypting with AAD, decide whether you will store the AAD side-by-side with the encrypted data, or reproduce the AAD, for subsequent decryption.
A reason to store AAD is to make it straightforward to use the AAD when you need to decrypt ciphertext. For example, a database row could contain both the ciphertext and the AAD used when the plaintext was encrypted. When a decrypt request is received, the application can retrieve both the AAD and the ciphertext from the database. The application can then use the AAD for the ciphertext decryption process.
A reason to reproduce AAD is to verify any non-private data while also avoiding storage of the AAD. For example, if you want to ensure that the filepath and filename of an encrypted file has not changed, you can include the filepath and filename as AAD when you encrypt the file. Then, when a decrypt request is sent for the file, you can reproduce the AAD by examining the filepath and filename.
Storing AAD means AAD is saved and then readily available by your application for future use. For example, a database table could contain a column for ciphertext and a column for the AAD used when the ciphertext was created. When it is time to decrypt the ciphertext, the application retrieves the AAD and uses it for the decryption.
Consider a diary application that is designed to show a diary entry only to the
user that created it. When a diary entry is created, it is encrypted and then
saved to a database, in column
ENCRYPTED_DIARY_ENTRY. For each request to view
a diary entry, the application authenticates the user and then provides the
diary entry to the user.
Suppose no AAD is used (other than the default empty string) and hypothetically
Mallory was able to copy Alice's
ENCRYPTED_DIARY_ENTRY data to Mallory's
ENCRYPTED_DIARY_ENTRY data. When Mallory sends a decrypt request for Mallory's
ENCRYPTED_DIARY_ENTRY data (which was copied from Alice's data), the
application performs the decryption without realizing it has been tricked.
Mallory is able to see Alice's diary entry as plaintext.
Now let's suppose the user email address is used as the AAD when a diary entry
is encrypted. When Alice creates a diary entry, the application stores her
unencrypted email address in the column
ENCRYPTED_DIARY_ENTRY data. Let's suppose again that Mallory was able to copy
ENCRYPTED_DIARY_ENTRY data to Mallory's
When Mallory sends a decrypt request, the application retrieves Mallory's email
address from the
Reproducing AAD means it isn't stored anywhere, but you can reproduce when it is time for decryption.
As an example, you can use a filepath and filename as AAD. During the encryption
process, suppose the encrypted file was saved to
"MY_PATH\MY_FILE1.enc" was used as the AAD. This AAD is not stored. When a
decrypt request is received, the application reproduces the AAD by examining the
filepath and filename of the file to decrypt. If the encrypted file has not been
moved to another location,
"MY_PATH\MY_FILE1.enc" will be used as the AAD
during decryption, which is the same as the AAD used during encryption, so the
decryption can proceed.
If the encrypted file has been moved, for example to
"SOME_OTHER_PATH\MY_FILE1.enc" will be
used as the AAD for decryption. This AAD value does not match the AAD value used
for encryption, so the decryption will fail.