wan24-Crypto-TPM 2.3.0

wan24-Crypto-TPM

WARNING: The code has not been tested with a real TPM running on Linux yet. I'd appreciate if someone would give me some feedback, if it worked for them. Anyway the tests use the Microsoft TPM simulator and did run successfully (and also with a real TPM device on Windows 11).

This library contains some helpers for easy TPM(2) usage. It does way not implement everything that a TPM offers - these are the selected features, which include the TPM into your security model:

• Determine if TPM2 can be accessed
• Determine the max. supported digest (size)
• Random number generation (TpmRng)
• HMAC-SHA-1/256/384/512 (Tpm2Helper and MacTpmHmacSha*Algorithm)
• TpmSymmetricKeySuite which implements ISymmetricKeySuite
• TpmSecuredValue which works like SecureValue
• TpmSharedSecret for usage with a remote key storage
• TpmValueProtection to extend ValueProtection

As you can see, the number of features is quiet clear.

This library extends the wan24-Crypto library with these algorithms:

CAUTION: TPM secured information won't be usable anymore, if the TPM (or even the TPM owner) changes!

The goal of this library is to make TPM usable for everyone in a simple way, without having to fight with a firmware and complex/missing documentation. It's an ideal extension to the existing wan24-Crypto infrastructure.

NOTE: There are no provisioning functionaliies implemented in this library. A TPM which is going to be used needs to be provisioned manually or from the host OS (Windows f.e. does provision a TPM automatically).

How to get it

This library is available as NuGet package.

Usage

In case you don't use the wan24-Core bootstrapper logic, you need to initialize the TPM2 extension first, before you can use it:

wan24.Crypto.TPM.Bootstrap.Boot();

This will register the algorithms to the wan24-Crypto library.

NOTE: All algorithms will be registered, no matter if there's even a TPM available or not, or if the algorithm is supported by an available TPM. This is because the TPM options support configuring a simulator or to choose between multiple available TPMs. So the bootstrapper can't really know which algorithms are going to be available (or used).

In case you work with dependency injection (DI), you may want to add some services:

builder.Services.AddWan24CryptoTpm();

This will register transient Tpm2Options (using Tpm2Helper.DefaultOptions) and Tpm2 (using Tpm2Helper.CreateEngine) service objects.

JSON configuration

You could implement a JSON configuration file using the AppConfig logic from wan24-Core, and the TpmCryptoAppConfig. There it's possible to define disabled algorithms, which makes it possible to react to an unwanted algorithm very fast, at any time and without having to update your app, for example. If you use an AppConfig, it could look like this:

public class YourAppConfig : AppConfig
{
    public YourAppConfig() : base() { }

    [AppConfig(AfterBootstrap = true, Priority = 20)]
    public CryptoAppConfig? Crypto { get; set; }

    [AppConfig(AfterBootstrap = true, Priority = 10)]
    public TpmCryptoAppConfig? Tpm { get; set; }
}

await AppConfig.LoadAsync<YourAppConfig>();

NOTE: A TpmCryptoAppConfig should be applied before a CryptoAppConfig. For this reason the example defines a priority in the AppConfigAttribute.

In the config.json in your app root folder:

{
    "Tpm":{
        ...
    }
}

Anyway, you could also place and load a TpmCryptoAppConfig in any configuration which supports using that custom type.

Tpm2Engine fixes multithreading bugs

Using a Tpm2 instance for each thread still has multithreading problems in the MS.TSS .NET library, that's why a Tpm2Engine should be used in multithreading environments. It ensures that

• only one Tpm2 instance is being used at a time
• only one thread can use the Tpm2 instance at a time

Example:

// Creating a Tpm2Engine uses static thread synchronization (a Tpm2Engine instance should be singleton)
using Tpm2Engine engine = Tmp2Engine.Create();

// Using per-engine thread synchronization (optional, to use one Tpm2Engine instance from multiple threads)
using SemaphoreSyncContext ssc = engine.Sync;

// Now you can perform a TPM operation using the engine.TPM property, which hosts the Tpm2 instance

NOTE: This is only required unless the multithreading bugs in the MS.TSS .NET library has been fixed by its vendor. In theory it should be possible to use a Tpm2 instance per thread without static thread locking (while multithreaded access to a Tpm2 instance still requires thread synchronization).

In case you're using Tpm2Helper.DefaultEngine, the Tpm2Engine usage is slightly different:

// Creating a Tpm2Engine uses static thread synchronization (a Tpm2Engine instance should be singleton)
using Tpm2Engine engine = new();// The empty constructor will use the Tpm2Helper.DefaultEngine and Tpm2Helper.DefaultEngineSync

// Using per-engine thread synchronization
using SemaphoreSyncContext ssc = engine.Sync;

// Now you can perform a TPM operation using the engine.TPM property, which hosts the Tpm2 instance

Implemented types support using a Tpm2Engine also, which will then not be disposed, but used for synchronizing the TPM access.

TPM2 options

In the Tpm2Options you can define how to connect to the TPM. You may also specify

• a resource handle (currently used for finalizing a HMAC)
• an algorithm (currently used for creating a HMAC)
• a tagged object (which will be cloned, if it implements ICloneable, and the GetCopy method of the Tpm2Options instance has been called)

Using the With* methods you can configure options fluent.

Determine if TPM2 can be accessed

bool canAccessTpm2 = Tpm2Helper.IsAvailable();

Because on a Linux system some file IO operations may run, there's an IsAvailableAsync method, too.

Tpm2 instance creation

using Tpm2 engine = Tpm2Helper.CreateEngine();

The Tpm2 instance is a connected TPM2 TSS, which allows to do whatever the TSS offers. By giving Tpm2Options to the CreateEngine method, you can define which TPM to use, and optional set an Initializer delegate, which may bring the TPM into the desired state, before running any other operation.

The CreateEngine method is being called internal, whenever you use a TPM functionality without giving an existing Tpm2 instance to the called method. And if you didn't specify the Tpm2Options, the Tpm2Helper.DefaultOptions will be used, which you may preset, if required.

The TryCreateEngine does the same as CreateEngine, but won't throw on error.

Maximum supported digest (size)

int maxDigestSize = Tpm2Helper.GetMaxDigestSize();// Size in byte
TpmAlgId maxDigest = Tpm2Helper.GetDigestAlgorithm(maxDigestSize);

The max. supported digest size limits the output of the random number generator, and it also defines the possible digest algorithms.

NOTE: TpmRng doesn't limit the random number count being generated in any way.

Random number generator

CAUTION: The example code is actually a negative example - see "Best practices" for a better solution suggestion!

RND.Generator = new TpmRng();// If not used as singleton, an instance should be disposed!

The TpmRng implements the IRng interface, which allows to use the TPM as RNG for wan24-Crypto. Internal it uses the Tpm2Helper.CreateRandomData helper method, which is restricted to the TPMs random number output length, while the RNG implementation doesn't restrict the length of the generated random data.

HMAC-SHA-1/256/384/512

byte[] hmac = Tpm2Helper.Hmac(anyAuthMessage);

NOTE: The owner resource handle will be used per default.

Using the Tpm2Helper.Hmac method you can create a HMAC-SHA-1/256/384/512 using the TPM. These HMACs can only be re-created using the same TPM. Specifying an additional MAC key is optional.

NOTE: Not every TPM implements all algorithms. HMAC-SHA-256 seems to be implemented by most TPMs. If you don't specify an algorithm to the Hmac method, it'll determine and use the maximum supported algorithm.

CAUTION: If you change your TPM hardware, you won't be able to re-create a HMAC! This also applies even only the TPM owner changes.

You can also use the wan24-Crypto registered HMAC algorithms during encryption, for example. Then cipher data couldn't be decrypted on any other computer than the one that encrypted it.

TIP: If you use a TPM HMAC of your encryption password, you can ensure that the cipher data can only be decrypted from the same computer that was used to encrypt it!

There are also TpmHmac* extension methods for a byte[] and (ReadOnly)Span<byte>.

TPM symmetric key suite

using TpmSymmetricKeySuite tpmAuth = new(key);

The TpmSymmetricKeySuite works as the SymmetricKeySuite, but uses a TPM HMAC for calculating the final key (and identifier, if any).

TPM secured value

The TpmSecuredValue works as SecureValue and protects a value using the TPM.

If you'd like TPM only if available, you can set the constructor parameter value of requireTpm to false. If TPM is not available, the constructor won't throw, and TpmSecuredValue will just work as SecureValue as a fallback solution.

En-/decrypting a private key suite

Using the TpmEncrypt extension method you can encrypt a PrivateKeySuite using a TPM flavored key. With Tpm2Helper.DecryptPrivateKeySuite you can decrypt it.

CAUTION: If you change your TPM hardware, there's no way to decrypt the private key suite anymore! The cipher data can only be decrypted using the same TPM hardware that was used for encryption. This also applies even only the TPM owner changes.

TPM shared secret

The TpmSharedSecret is a helper for deriving a TPM secured key from a remote key storage.

NOTE: The following examples assume that your remote key storage requires sending a secret for receiving a secret. This may be different per each remote key storage.

Example how to initialize a new secret:

using Tpm2 engine = Tpm2Helper.CreateEngine();
byte[] token = RND.GetBytes(Tpm2Helper.GetMaxDigestLength(engine)),
	remoteSecret = RND.GetBytes(token.Length);
// Store the token somewhere for restoring the secret later
using TpmSharedSecret tpmSecret = new(token, engine: engine);
tpmSecret.ProtectRemoteSecret(remoteSecret);
// Send tpmSecret.Secret.Array and remoteSecret to the remote key storage
byte[] secret = tpmSecret + remoteSecret;

CAUTION: NEVER store remoteSecret persistent outside of the remote key storage! NEVER store tpmSecret.Secret.Array anywhere!

tpmSecret.Secret.Array is used to authenticate for receiving the value of remoteSecret from the remote key storage later.

NOTE: token may be stored plain, maybe protected using the OS capabilities (like the file system ACL, f.e.).

Example how to restore a previously initialized secret:

// Load the token from where it was saved during secret initialization
using TpmSharedSecret tpmSecret = new(token);
// Send tpmSecret.Secret.Array to the remote key storage and receive remoteSecret
byte[] secret = tpmSecret + remoteSecret;

The TpmSharedSecret also supports including an additional secret (for user authentication f.e.).

TPM value protection

The TpmValueProtection uses the TPM for protecting a value as ValueProtection does without the TPM. For this the scope keys will be used as value for a TPM HMAC, which will then be the final key being used for the value encryption (the max. TPM supported HMAC algorithm will be used).

NOTE: The TpmValueProtection uses the scope keys from ValueProtection and uses the default TPM state for creating the HMAC. That means in particular you'll still have to ensure a restorable user scope key, while you don't have to take care the system scope key anymore.

You may replace the ValueProtection protect/unprotect handlers:

TpmValueProtection.Enable();

NOTE: The TpmValueProtection protect/unprotect handlers will connect to the TPM for every call, which is an overhead and may impact the performance of your application. If you don't want that, you may simply replace the ValueProtection user/system scope keys with TPM HMACs, probably including an user secret for the user scope key.

Or you can use both, the ValueProtection and the TpmValueProtection, as it is suitable for your application, separately.

Extension methods

The TpmExtensions class exports some extension methods to make life more easy, when working with TPM types and wan24-Crypto. There are also extensions for the PrivateKeySuite, byte[] and (ReadOnly)Span<byte> (TPM HMAC creation). Using the CryptoOptions.WithTpmHmac extension method, you can set the max. supported TPM HMAC algorithm for any crypto application which requires to compute a MAC.

Using a singleton TPM2 connection

By setting a Tpm2 instance to the Tpm2Helper.DefaultEngine property, you can specify a singleton connection to use from Tpm2Helper methods. Use the Tmp2Helper.DefaultEngineSync to synchronize multithreaded connection usage:

// Set a singleton default TPM engine
Tpm2Helper.DefaultEngine = Tpm2Helper.CreateEngine();

// Synchronize the default TPM engine access before performing any Tpm2Helper operation
using SemaphoreSyncContext ssc = Tpm2Helper.DefaultEngineSync;
// Now you can perform any Tpm2Helper operation in a multithreaded environment using the singleton Tpm2Helper.DefaultEngine

The Tpm2Helper.DefaultEngine value will be set to the engine parameter of Tpm2Helper methods, if no value was given.

Why not support TPM PKI/signing/sealing/etc.?

If you followed the TPM development process until today you know that TPM2 is fully incompatible with TPM1. I try to concentrate on the absolute minimum that TPM offers, to stay (hopefully) compatible with TPM3 (or any future TPM version). With the HMAC function you should have everything that is required at minimum, for implementing everything else using wan24-Crypto (which offers way more functionality than TPM does). The ExpandedKey of a TpmSymmetricKeySuite can be used for any encryption, and it's bound to the available TPM, so you could encrypt a PrivateKeySuite, for example, which can then only be decrypted using the same TPM. And you're not bound to the TPM implemented algorithms, as you have the free choice to use any wan24-Crypto implemented cryptographic algorithm, and optional combine them with the provided TPM functionality.

To sum it up - the reasons for not using all of the TPM capabilities:

• TPM doesn't implement the cryptographic algorithms that you need to use
• The TPM processing speed is decreased because of KDF usage in places where you don't want (need) to use KDF (at all)
• Future security developments require new TPM hardware, which will mess up your PKI
• TPM is way not the answer to all crypto related questions
• TPMx may fully break TPM2 key capabilities (again), while the implemented features of wan24-Crypto-TPM may still be supported ('cause they're the absolute basics, which should be valid for at last the next decade from now)

There are many good reasons to use only the absolute basics of the offered TPM features, and only a few applications which are really enriched by the TPM, which is usually being used in normal devices.

Someone might argue that TPM can encrypt/decrypt (seal/unseal) data independent from the OS and other hardware - yes, that's true. If AES-128 does still fit your security policy in 2023+, you'd be fine with it (use Tpm2). But remember that the in-TPM en-/decryption is only suitable for small blobs! This in combination limits the application in a way which is not acceptable for the most use cases for cryptography: If you want to process larger blobs, you have to DIY. If you need AES-256 (or any other algorithm than the TPM implemented ones), you have to DIY. Instead of using the TPM lockout, DIY and use KDF in addition. That's enough DIY to skip implementing support for the TPM offered functionality into wan24-Crypto-TPM and sticking to the TPMs HMAC only, which is enough already (and not to forget the RNG also). Brute Force will always stay possible, no matter if you use TPM or not - remember that.

However, if you need all the TPM functionality (if your boss or a customer is obsessed with TPM and no technical argument seems to count anymore - I know something like that...), you're free to use Tpm2Helper.CreateEngine and work with the Tpm2 object directly and without any limit.

From my sight there's only one reason for sticking to the TPM implemented functionality: Private keys will be used for crypto/signature witin the TPM only, which allows protecting/authenticating sensitive information within an isolated processor, which runs independent from the rest of the system. But since the rest of the system controls the TPM, it's nothing more than a piece of hardware which can be used to identify a device. Remember that there's still software (the TSS and the firmware), which is required to be implemented, and is a point of failure for the TPM offered security stack. Once that software was attacked with success, your software has been broken, too. So even the identification of a device using TPM isn't 100% trustable.

Supported platforms

All platforms which support TPM should be supported by this library. Anyway, Apple devices often don't contain a TPM, but a T2 (which is similar to TPM) instead (which may be called T8012, too).

I've successfully run the tests on a Windows 11 computer only so far, since at the moment I don't own a Linux device with a TPM. But Linux supports TPM, and the underlaying TSS.MSR .NET library supports Linux, finally.

So the supported platform list may be:

• Windows (10+)
• Linux
• (MAC OSX)

There seems to be no .NET library for Apples T2 chip, and I'm not going to implement one. You could use the MAC OSX API for the T2 chip directly by using interop, but however, since HMAC seems not to be supported, I'd use a T2 as a better HWRNG only.

For Apple iOS (and others != OSX) there is a "Security Enclave", which is a SoC like TPM - but also without HMAC support, so it can be seen as a better HWRNG, too.

On an Android device you'd use the KeyChain or TEE API usually, but there could also be a TPM being supported. However, it's not supported by the TSS.MSR, so this library can't offer support, too.

To sum it up: Forget about Apple and Android, and concentrate on Windows and Linux, if you'd like to use this TPM library.

Best practice

TpmRng usage

Random numbers are security critical, and it may be a bad idea to rely on one entropy source or RNG only. For this I suggest to use the TpmRng together with other RNGs, and combine their generated random numbers using a XorRng.

TPM encrypted PrivateKeySuite

When your app requires a TPM protected private key suite, you can create one with these steps:

1. Create a PrivateKeySuite
2. Encrypt the PrivateKeySuite using the TpmEncrypt extension method
3. Store the cipher data in a file
4. Dispose the PrivateKeySuite when not in use anymore!

To load it when your app starts again:

1. Decrypt the PrivateKeySuite cipher data from the file using Tpm2Helper.DecryptPrivateKeySuite
2. Dispose the PrivateKeySuite when not in use anymore!

Persistent secret storage

Different OS offer different secret storage solutions - but none of them seem to offer a real security benefit. There's only one thing, which could enhance security (really): Storing a part of a secret at another system.

To make this process combinable with TPM, there's the TpmSharedSecret helper class, which makes it possible to restore an (user) secret using a TPM bound token, which may be stored in plain on the processing system, but requires using a remote key storage to provide a partial key for a TPM processed token value, which acts as a shared secret.

When storing a mashine scope secret, it ensures that

• the ciphered data on that mashine can be remote-deleted by simply deleting the remote stored secret
• someone which could access any key part (or both), but isn't able to access the TPM, can't get to the final key

When storing an user scope secret, it ensures in addition that

• even when having both key parts and access to the TPM, a dictionary or Brute Force attack on the user password isn't practicable, when the user password is secure (has been KDF processed)

These benefits apply to both sides: The local system, and the remote key storage. By the way the remote key storage should store the provided secret encrypted using the shared secret, and never store the shared secret anywhere.

To get a final key

• the plain stored token must be available (1st part of the key)
• access to the TPM must be available (for computing the shared secret)
• the remote storage must reply the 2nd part of the key for the provided shared secret
• another TPM access must be possible to combine both key parts

An attacker can't use the plain stored token (1st key part) alone. He even can't request the 2nd key part from the remote key storage, if the TPM can't be accessed. Also the 2nd key part alone doesn't offer any success for an attacker, if the TPM can't be accessed, too - and even when having both key parts, the TPM access is a required component to get to the final key. To break the security, an attacker requires both key parts and having access to the TPM. This applies to a mashine scope secret.

For an user scope secret, an attacker would then still need the user password. To secure the user password, you should pre-process it using KDF before you use it as the key parameter in the TpmSharedSecret constructor. This I'd call an almost perfect solution in 2023 then.

Anyway, there are some pitfalls with that solution: IF

• the TPM (owner) changed, or the TPM is broken, then ciphered data is lost
• there is no connection to the remote key storage possible, ciphered data can't be accessed unless the connection problem was solved
• the remote stored key part got lost, then ciphered data is lost, too

It's important to have that in mind and implement emergency solutions for such (worst case) scenarios to avoid a data loss.