dpapi-ng
DPAPI-NG (CNG "group protection") and MS-GKDI key derivation, in pure Rust. No FFI, no
Windows host, no bcrypt.dll.
Windows LAPS (msLAPS-EncryptedPassword), gMSA and dMSA managed passwords are all protected
the same way: a CMS EnvelopedData whose content key is wrapped under a KEK derived from a
KDS root key that lives on the domain controllers. The blob carries no key — only a key
identifier: which root key, and which node of its L0/L1/L2 tree. A principal allowed to read
the secret asks a DC for that node (GetKey over MS-GKDI) and derives downwards.
This crate does everything except the RPC call.
use ;
use GroupKeyEnvelope;
// 1. Split the LAPS attribute and parse the CMS blob.
let laps = parse?;
let blob = laps.protected?;
// 2. Ask a DC for the node named in blob.key_identifier (your DCE/RPC, not ours).
let envelope = parse?;
// 3. Derive down the tree and open the content.
let plaintext = decrypt?;
println!;
What is in here
gkdi::KeyIdentifier— which root key and tree node a blob was protected for.gkdi::GroupKeyEnvelope— whatGetKeyreturns.gkdi::compute_l2_key— the L0/L1/L2 tree walk. Pure computation, no network.kdf— the SP800-108 counter-mode KDF over SHA-1/256/384/512.cms::ProtectedBlob— the CMSEnvelopedData, itsKEKRecipientInfoand the AES-GCM content parameters, on a small hand-written DER reader rather than a codegen'd ASN.1 stack.LapsBlob+laps_password_from_json— the Windows LAPS attribute header and its UTF-16LE JSON payload.
Transport is deliberately out of scope
GetKey is one MS-GKDI RPC call over \pipe\lsass with authentication and sealing, and every
project that needs this already has a DCE/RPC stack. Make the call, hand the response bytes to
GroupKeyEnvelope::parse, and keep this crate free of a transport dependency.
Status
The structures, the KDF, the tree walk and the CMS path are covered by tests, including an
end-to-end build-then-decrypt of a real EnvelopedData (AES key-wrap plus AES-256-GCM).
One detail is still pinned down against live data rather than by specification: Windows has
used more than one KEK derivation from the L2 seed key across versions, so
decrypt_with_seed_key tries the known candidates in order. This is safe rather than sloppy —
AES key-wrap and AES-GCM are both authenticated, so a wrong KEK fails loudly instead of
producing plausible garbage. If you know the derivation your environment uses, call
decrypt_with_kek directly.
License
MIT