For many storage encryption applications, the ciphertext must be the same size as the plaintext; generally this matches the disk sector size of either 512 or 4096 bytes. This means that standard approaches like AES-GCM or RFC7539 cannot be applied. The standard solution is AES-XTS, but this has two disadvantages:
If AES hardware is absent, AES is relatively slow, especially constant-time implementations
Using XTS, a one-bit change to the plaintext means only a 16-byte change to the ciphertext, revealing more to the attacker than necessary.
Adiantum uses a fast hash (NH + Poly1305) and a fast stream cipher (XChaCha12) to build a construction which encrypts an entire sector at a time. On an ARM Cortex-A7 processor, Adiantum decrypts 4096-byte messages at 10.6 cycles per byte, over five times faster than AES-256-XTS. It is a "super pseudorandom permutation" over the whole sector, which means that any change to the plaintext of the sector results in an unrecognizably different ciphertext sector and vice versa.
Adiantum appears in IACR Transactions on Symmetric Cryptology, Volume 2018, Issue 4. We also document HPolyC, our first such proposal, which is slower on large messages but simpler and more key agile for small messages.
specification/: LaTeX sources for our paper presenting Adiantum
test_vectors/other: Test vectors we use to validate our implementations of other primitives
test_vectors/ours: Test vectors we generate, in JSON format
python/: Python implementation and test vector generation
benchmark/: software we used to generate the benchmarks in our paper
third_party/: derived works covered by a different license than our main MIT license
third_party/ includes derived works not covered by the MIT license;
specifically software derived from the Linux kernel and licensed under GPLv2.
We include here a variety of algorithms and implementations; we make no guarantee they are suitable for production use.
This is not an officially supported Google product.