Question

I receive data, and use aes or blowfish to encrypt it before saving it to the database, so the encryption is done at the application level. If someone steals the database, the data will be relatively safe unless they stole the application also (where the key is stored/accessed).

I’m now looking into database encryption with libraries like ezNcrypt for MySQL, Encryption-GENERAL, or SQLCipher.

But I don’t quite understand how database encryption works. If the application only passes raw unaltered data to the database, and the database decrypts the data itself somehow, wouldn’t that make database-level encryption less secure if the database was stolen since 100% of the encryption component was stolen?

In my current situation, if a database is stolen, the attacker would have to steel the second component (the key which is at the application level) to decrypt the database. But with database encryption, the database itself has full responsibility of the encryption, so by stealing the database, wouldn’t the attacker have everything needed to decrypt the database?

Maybe I’m not clear on how database-level decryption works.

Answer 1

The encryption algorithm is applied at different points in your application stack, but at the end of the day the process is exactly the same regardless if you use an application or db-layer solution. The key has to be stored somewhere and if an attacker gets both the encrypted data and the key, then they have access to the data.

So, the real question boils down to how you store your key:

• Plaintext – if it’s in plaintext somewhere in the filesystem then that’s your weak point. Maybe it’s in the application code, maybe in a config file somewhere. In any case, someone who gains administrator access to the server (or simply steals the hard drive) has what they need, and obscurity is your only protection.
• Manually-entered – If you have a human user enter the key when the application/database/pc is started, then you mostly* alleviate the issue of a plaintext key. In exchange, you require a user to remember the key and you give up the ability to have a completely automated restart (bad news for a server that needs to stay up).

* I say mostly because technically the key is still available in plaintext somewhere in RAM. But that’s a lot harder to get at than something stored on disk (memory swapping notwithstanding).

Regarding MySQL specifically, you might find this thread helpful.