The encrypted storage engine was added to support a series of special use cases, mostly revolving around finance, retail, healthcare, education, and government.
We need to have encryption for the rest of our data if we have to comply to a set of regulations, including among others:
- PCI DSS for handling credit card information
- HIPAA for healthcare applications
- NIST for government
- FISMA for government
- STIG for government
This can be done in several ways and cloud services providers such as EC2 provide EBS storage volumes with built-in encryption.
Encrypted storage supports Intel's AES-NI equipped CPUs for acceleration of the encryption/decryption process.
The encryption algorithms supported are the following:
- AES-256, CBC (default)
- AES-256, GCM
- FIPS, FIPS-140-2
Encryption is supported at page level for better performance. When a change is made in a document, instead of re-encrypting/decrypting the entire underlying file, only the page that is affected gets modified.
Encryption key management is a huge aspect of encrypted storage security. Most specifications previously mentioned require key rotation at least once per year.
MongoDB's encrypted storage uses an internal database key per node. This key is wrapped by an external (master) key that must be used to start the node's mongod process. By using the underlying operating system's protection mechanisms such as mlock or VirtualLock, MongoDB can guarantee that the external key will never be leaked from memory to disk by page faults.
The external (master) key can be managed either by using the Key Management Interoperability Protocol (KMIP) or by using local key management via a keyfile.
MongoDB can achieve key rotation by performing rolling restarts of the replica set members. Using KMIP, MongoDB can rotate only the external key and not the underlying database files, delivering significant performance benefits.
Using MongoDB's encrypted storage provides the advantage of increased performance versus encrypted storage volumes. MongoDB's encrypted storage has an overhead of around 15% as compared to 25% or more for third-party encrypted storage solutions.
In most cases, if we need to use encrypted storage, we will know it way in advance from the application design phase and we can perform benchmarks against different solutions to choose the one that best fits our use case.