# Always Encrypted cryptography

**Applies to:**
SQL Server
Azure SQL Database
Azure SQL Managed Instance

This document describes encryption algorithms and mechanisms to derive cryptographic material used in the Always Encrypted feature in SQL Server and Azure SQL Database.

## Keys, key stores, and key encryption algorithms

Always Encrypted uses two types of keys: Column master keys and column encryption keys.

A column master key (CMK) is a key encrypting key (for example, a key that is used to encrypt other keys) that is always in a client's control, and is stored in an external key store. An Always Encrypted-enabled client driver interacts with the key store via a CMK store provider, which can be either part of the driver library (a Microsoft/system provider) or part of the client application (a custom provider). Client driver libraries currently include Microsoft key store providers for Windows Certificate Store and hardware security modules (HSMs). For the current list of providers, see CREATE COLUMN MASTER KEY (Transact-SQL). An application developer can supply a custom provider for an arbitrary store.

A column encryption key (CEK), is a content encryption key (for example, a key that is used to protect data) that is protected by a CMK.

All Microsoft CMK store providers encrypt CEKs by using RSA with Optimal Asymmetric Encryption Padding (RSA-OAEP). The key store provider that supports Microsoft Cryptography API: Next Generation (CNG) in .NET Framework (SqlColumnEncryptionCngProvider Class) uses the default parameters specified by RFC 8017 in Section A.2.1. Those default parameters are using a hash function of SHA-1 and a mask generation function of MGF1 with SHA-1. All other key store providers use SHA-256.

Always Encrypted internally uses FIPS 140-2 validated cryptographic modules.

## Data Encryption Algorithm

Always Encrypted uses the **AEAD_AES_256_CBC_HMAC_SHA_256** algorithm to encrypt data in the database.

**AEAD_AES_256_CBC_HMAC_SHA_256** is derived from the specification draft at https://tools.ietf.org/html/draft-mcgrew-aead-aes-cbc-hmac-sha2-05. It uses an Authenticated Encryption scheme with Associated Data, following an Encrypt-then-MAC approach. That is, the plaintext is first encrypted, and the MAC is produced based on the resulting ciphertext.

In order to conceal patterns, **AEAD_AES_256_CBC_HMAC_SHA_256** uses the Cipher Block Chaining (CBC) mode of operation, where an initial value is fed into the system named the initialization vector (IV). The full description of the CBC mode can be found at https://csrc.nist.gov/publications/nistpubs/800-38a/sp800-38a.pdf.

**AEAD_AES_256_CBC_HMAC_SHA_256** computes a ciphertext value for a given plaintext value using the following steps.

### Step 1: Generating the initialization vector (IV)

Always Encrypted supports two variations of **AEAD_AES_256_CBC_HMAC_SHA_256**:

Randomized

Deterministic

For randomized encryption, the IV is randomly generated. As a result, each time the same plaintext is encrypted, a different ciphertext is generated, which prevents any information disclosure.

```
When using randomized encryption: IV = Generate cryptographicaly random 128bits
```

If there's deterministic encryption, the IV isn't randomly generated, but instead it's derived from the plaintext value using the following algorithm:

```
When using deterministic encryption: IV = HMAC-SHA-256( iv_key, cell_data ) truncated to 128 bits.
```

Where iv_key is derived from the CEK in the following way:

```
iv_key = HMAC-SHA-256(CEK, "Microsoft SQL Server cell IV key" + algorithm + CEK_length)
```

The HMAC value truncation is performed to fit one block of data as needed for the IV. As a result, deterministic encryption always produces the same ciphertext for a given plaintext value, which enables inferring whether two plaintext values are equal by comparing their corresponding ciphertext values. This limited information disclosure allows the database system to support equality comparison on encrypted column values.

Deterministic encryption is more effective in concealing patterns, compared to alternatives, such as using a pre-defined IV value.

### Step 2: Computing AES_256_CBC Ciphertext

After computing the IV, the **AES_256_CBC** ciphertext is generated:

```
aes_256_cbc_ciphertext = AES-CBC-256(enc_key, IV, cell_data) with PKCS7 padding.
```

Where the encryption key (enc_key) is derived from the CEK as follows.

```
enc_key = HMAC-SHA-256(CEK, "Microsoft SQL Server cell encryption key" + algorithm + CEK_length )
```

### Step 3: Computing MAC

Subsequently, the MAC is computed using the following algorithm:

```
MAC = HMAC-SHA-256(mac_key, versionbyte + IV + Ciphertext + versionbyte_length)
```

Where:

```
versionbyte = 0x01 and versionbyte_length = 1
mac_key = HMAC-SHA-256(CEK, "Microsoft SQL Server cell MAC key" + algorithm + CEK_length)
```

### Step 4: Concatenation

Finally, the encrypted value is produced by concatenating the algorithm version byte, the MAC, the IV, and the AES_256_CBC ciphertext:

```
aead_aes_256_cbc_hmac_sha_256 = versionbyte + MAC + IV + aes_256_cbc_ciphertext
```

## Ciphertext Length

The lengths (in bytes) of particular components of **AEAD_AES_256_CBC_HMAC_SHA_256** ciphertext are:

versionbyte: 1

MAC: 32

IV: 16

aes_256_cbc_ciphertext:

`(FLOOR (DATALENGTH(cell_data)/ block_size) + 1)* block_size`

, where:block_size is 16 bytes

cell_data is a plaintext value

Therefore, the minimal size of aes_256_cbc_ciphertext is 1 block, which is 16 bytes.

Thus, the length of ciphertext, resulting from encrypting a given plaintext values (cell_data), can be calculated using the following formula:

```
1 + 32 + 16 + (FLOOR(DATALENGTH(cell_data)/16) + 1) * 16
```

For example:

A 4-byte long

**int**plaintext value becomes a 65-byte long binary value after encryption.A 2,000-byte long

**nchar(1000)**plaintext values becomes a 2,065-byte long binary value after encryption.

The following table contains a complete list of data types and the length of ciphertext for each type.

Data Type | Ciphertext Length [bytes] |
---|---|

bigint |
65 |

binary |
Varies. Use the formula above. |

bit |
65 |

char |
Varies. Use the formula above. |

date |
65 |

datetime |
65 |

datetime2 |
65 |

datetimeoffset |
65 |

decimal |
81 |

float |
65 |

geography |
N/A (not supported) |

geometry |
N/A (not supported) |

hierarchyid |
N/A (not supported) |

image |
N/A (not supported) |

int |
65 |

money |
65 |

nchar |
Varies. Use the formula above. |

ntext |
N/A (not supported) |

numeric |
81 |

nvarchar |
Varies. Use the formula above. |

real |
65 |

smalldatetime |
65 |

smallint |
65 |

smallmoney |
65 |

sql_variant |
N/A (not supported) |

sysname |
N/A (not supported) |

text |
N/A (not supported) |

time |
65 |

timestamp( rowversion) |
N/A (not supported) |

tinyint |
65 |

uniqueidentifier |
81 |

varbinary |
Varies. Use the formula above. |

varchar |
Varies. Use the formula above. |

xml |
N/A (not supported) |

## .NET Reference

For details about the algorithms, discussed in this document, see the **SqlAeadAes256CbcHmac256Algorithm.cs**, **SqlColumnEncryptionCertificateStoreProvider.cs**, and **SqlColumnEncryptionCertificateStoreProvider.cs** files in the .NET Reference.

## See also

## Feedback

https://aka.ms/ContentUserFeedback.

Coming soon: Throughout 2024 we will be phasing out GitHub Issues as the feedback mechanism for content and replacing it with a new feedback system. For more information see:Submit and view feedback for