Column Level Encryption

On this page

New in v23.2: CockroachDB supports column level encryption. Using this feature, you can encrypt one or more of the columns in each row of a database table.

Column level encryption can be useful for compliance scenarios such as adhering to PCI.

Once a column is encrypted, SQL users are required to pass an encryption key when reading from or writing to that column. This prevents users without the encryption key from taking actions such as reading sensitive information (like personally identifiable information (PII)) when accessing the tables.

Available Functions

encrypt and encrypt_iv

The encrypt and encrypt_iv functions encrypt a column's data with a given key and cipher method. For more information, see Cryptographic functions.

For usage examples, see:

• Encrypt using the encrypt function.
• Encrypt using the encrypt_iv function.

These functions require an Enterprise license. If you do not have an Enterprise license, you will see an error message like the following if you try to use them:

ERROR: encrypt(): use of this cryptographic function (https://www.cockroachlabs.com/docs/stable/functions-and-operators#cryptographic-functions) requires an enterprise license. see https://cockroachlabs.com/pricing for details on how to enable enterprise features

decrypt and decrypt_iv

The decrypt and decrypt_iv functions decrypt an encrypted column's data with a given key and cipher method. For more information, see Cryptographic functions.

For usage examples, see:

• Decrypt using the decrypt function.
• Decrypt using the decrypt_iv function.

These functions require an Enterprise license. If you do not have an Enterprise license, you will see an error message like the following if you try to use them:

ERROR: decrypt(): use of this cryptographic function (https://www.cockroachlabs.com/docs/stable/functions-and-operators#cryptographic-functions) requires an enterprise license. see https://cockroachlabs.com/pricing for details on how to enable enterprise features

Security considerations

Whether to use encrypt and decrypt or encrypt_iv and decrypt_iv

Both encrypt and decrypt have *_iv variants: encrypt_iv and decrypt_iv. You will need to assess your risk profile to determine which functions to use.

The benefits of using encrypt_iv and decrypt_iv include:

• Avoid repetition: If you use the same encryption key and method to encrypt the same plaintext multiple times without an initialization vector (IV), you'll get the same ciphertext every time. This repetition can provide a point of attack for someone trying to break the encryption. By using an IV, even the same plaintext will produce different ciphertexts, provided a different IV is used each time.
• Defend against pattern analysis: Without an IV, if two users have the same piece of data (such as an SSN), their encrypted values will also be the same. An attacker can exploit these patterns. By using different IVs for each encryption, the encrypted values will be different even if the plaintext values are the same.
• Cipher block chaining (CBC) mode: Many encryption algorithms, like AES, operate on blocks of data. In modes like CBC, the previous block of ciphertext is used as an IV for the encryption of the next block. This means that even if there are patterns in the plaintext, they won't appear in the ciphertext. However, for the first block, there is no previous block of ciphertext, so an IV is used. This is another way IVs help in breaking up patterns in the ciphertext.
• Mitigate replay attacks: Since the IV is typically random and changed for every encryption, it makes replay attacks more difficult. An attacker can't simply take an old piece of encrypted data and send it again, as the IV will likely have changed.

The drawbacks of using encrypt_iv and decrypt_iv include:

• Storage: You need to store the IV alongside the ciphertext. It's common practice to prepend or append the IV to the ciphertext before storing it. Unlike the encryption key, the IV doesn't need to be kept secret, but it does need to be known for decryption.
• Randomness: It's crucial that IVs are random and not predictable. If an attacker can predict the next IV, some of the security benefits are negated.
• Unique IVs with the Same Key: While IVs need to be random, it's also essential that the same IV isn't used twice with the same encryption key. Doing so can leak information about the plaintext.

How AES variants are determined

The actual AES variant (AES-128, AES-192, or AES-256) is determined by the length of the encryption key you provide in the functions:

• AES-128: 16-byte key
• AES-192: 24-byte key
• AES-256: 32-byte key

Internally, keys have to be 16-, 24-, or 32-byte lengths and map to the corresponding AES encryption strength. If you use a key with a different length, the key is zero-padded up to the next valid key length. If the key is more than 32 bytes long, it is truncated to 32 bytes.

Performance considerations

Use of the encrypt built-in function can have anywhere from 10-40% overhead depending on the length of the data being encrypted and the hardware provisioned for CockroachDB.

Cockroach Labs measured baseline performance in a 3-node CockroachDB cluster running on three n1-standard-4 machines on GCP.

Without using encrypt or decrypt, the following statement generally ran in 60-80 ms:

WITH
    a
        AS (
            SELECT
                lpad(
                    (1000 * random())::INT8::STRING,
                    3,
                    '0'
                )
                || '-'
                || lpad(
                        (100 * random())::INT8::STRING,
                        2,
                        '0'
                    )
                || '-'
                || lpad(
                        (10000 * random())::INT8::STRING,
                        4,
                        '0'
                    )
                    AS ssn
            FROM
                ROWS FROM (generate_series(1, 1.0E+04))
        )
SELECT
    ssn
FROM
    a

Using both encrypt and decrypt, the following statement generally ran in 80-100 ms:

WITH
    a
        AS (
            SELECT
                lpad(
                    (1000 * random())::INT8::STRING,
                    3,
                    '0'
                )
                || '-'
                || lpad(
                        (100 * random())::INT8::STRING,
                        2,
                        '0'
                    )
                || '-'
                || lpad(
                        (10000 * random())::INT8::STRING,
                        4,
                        '0'
                    )
                    AS ssn
            FROM
                ROWS FROM (generate_series(1, 1.0E+04))
        )
SELECT
    convert_from(
        decrypt(
            encrypt(
                ssn::BYTES,
                e'\\xd54a43d2a4caf8d3fbe4e4f711b39d4a0fedf26ac0dcdfb0811c2078a6a9cd147e77da38e35e14cacfc79c7e11a052c4bc9449e1d6fa280dcdc45bb4004f1648',
                'aes'
            ),
            e'\\xd54a43d2a4caf8d3fbe4e4f711b39d4a0fedf26ac0dcdfb0811c2078a6a9cd147e77da38e35e14cacfc79c7e11a052c4bc9449e1d6fa280dcdc45bb4004f1648',
            'aes'
        ),
        'UTF8'
    )
FROM
    a

With encrypt only, the following statement generally ran in 80-100 ms:

WITH
    a
        AS (
            SELECT
                lpad(
                    (1000 * random())::INT8::STRING,
                    3,
                    '0'
                )
                || '-'
                || lpad(
                        (100 * random())::INT8::STRING,
                        2,
                        '0'
                    )
                || '-'
                || lpad(
                        (10000 * random())::INT8::STRING,
                        4,
                        '0'
                    )
                    AS ssn
            FROM
                ROWS FROM (generate_series(1, 1.0E+04))
        )
SELECT
    encrypt(
        ssn::BYTES,
        e'\\xd54a43d2a4caf8d3fbe4e4f711b39d4a0fedf26ac0dcdfb0811c2078a6a9cd147e77da38e35e14cacfc79c7e11a052c4bc9449e1d6fa280dcdc45bb4004f1648',
        'aes'
    )
FROM
    a

Examples

Setup

The examples in this section operate on the following table.

CREATE TABLE IF NOT EXISTS users (
    user_id UUID DEFAULT uuid_generate_v4() PRIMARY KEY,
    name VARCHAR(255),
    encrypted_ssn BYTES,
    ssn_iv BYTES
);

Encrypt using the encrypt function

INSERT
INTO
    users (name, encrypted_ssn)
VALUES
    (
        'John Doe',
        encrypt(
            '123-45-6789'::BYTES,
            'your_secret_key'::BYTES,
            'aes'
        )
    );

Encrypt using the encrypt_iv function

WITH
    iv AS (SELECT gen_random_bytes(16) AS iv)
INSERT
INTO
    users (name, encrypted_ssn, ssn_iv)
SELECT
    'Jane Doe',
    encrypt_iv(
        '987-65-4321'::BYTES,
        'your_secret_key'::BYTES,
        iv,
        'aes'
    ),
    iv
FROM
    iv;

Decrypt using the decrypt function

SELECT
    name,
    convert_from(
        decrypt(
            encrypted_ssn,
            'your_secret_key'::BYTES,
            'aes'
        ),
        'UTF8'
    )
        AS ssn
FROM
    users
WHERE
    name = 'John Doe';

Decrypt using the decrypt_iv function

SELECT
    name,
    convert_from(
        decrypt_iv(
            encrypted_ssn,
            'your_secret_key'::BYTES,
            ssn_iv,
            'aes'
        ),
        'UTF8'
    )
        AS ssn
FROM
    users
WHERE
    name = 'Jane Doe';

See also

• Encryption at Rest
• Cryptographic functions

Was this helpful?