Trusty and Security Services in ACRN

This document provides an overview of the Trusty architecture for Linux-based system, what security services Trusty provides, and how Trusty works on top of the ACRN Hypervisor.

Trusty Architecture

Trusty is a set of software components supporting a Trusted Execution Environment (TEE) on embedded devices. It is a full software stack environment including OS, services, and APIs. As shown in Figure 178 below, it consists of:

• An operating system (the Trusty OS) that runs on a processor providing a TEE;
• Drivers for the kernel (Linux) to facilitate communication with applications running under the Trusty OS;
• A set of libraries for Android systems software to facilitate communication with trusted applications executed within the Trusty OS using the kernel drivers.

Figure 178 Trusty Architecture

Google provides an Android Open Source Project (AOSP) implementation of Trusty based on ARM TrustZone technology. Intel enables Trusty implementation on x86 based platforms with hardware virtualization technology (e.g. VT-x and VT-d). In Figure 178 above, the Secure Monitor is a VMM hypervisor. It could be any x86 hypervisor, and it is the customer’s responsibility to pick the right hypervisor for their product. Intel has developed a product-quality open source lightweight hypervisor reference implementation for customers to use; see https://github.com/intel/ikgt-core/tree/trusty.

The purpose of this secure monitor (hypervisor) is to isolate the normal and secure worlds, and to schedule Trusty OS in and out on demand. In the Trusty implementation, all the security services provided by Trusty OS in the secure world are event-driven. As long as there is no service request from normal world, Trusty OS won’t be scheduled in by the hypervisor. The normal world and secure world share the same processor resources, so this minimizes the context switching performance penalty.

In Trusty OS, the kernel is a derivative of the Little Kernel project, an embedded kernel supporting multi-thread, interrupt management, MMU, scheduling, and more. Google engineers added user-mode application support and a syscall layer to support privilege level isolation, so that each Trusted App can run in an isolated virtual address space to enhance application security. Intel added many more security enhancements such as SMEP (Supervisor Mode Execution Prevention), SMAP (Supervisor Mode Access Prevention), NX (Non-eXecution), ASLR (Address Space Layout Randomization), and stack overflow protector.

There are a couple of built-in Trusted Apps running in user mode of Trusty OS. However, an OEM can add more Trusted Apps in Trusty OS to serve any other customized security services.For security reasons and for serving early-boot time security requests (e.g. disk decryption), Trusty OS and Apps are typically started before Normal world OS.

In normal world OS, Trusty Driver is responsible for IPC communication with Trusty OS (over hypervisor) to exchange service request commands and messages. The IPC manager can support concurrent sessions for communications between Trusted App and Untrusted Client App. Typically, Trusty provides APIs for developing two classes of applications:

• Trusted applications or services that run on the TEE/Trusty OS in secure world;
• Untrusted applications running in normal world that use services provided by Trusted applications.

Software running in normal world can use Trusty client library APIs to connect to trusted applications and exchange arbitrary messages with them, just like a network service over IP. It is up to the application to determine the data format and semantics of these messages using an app-level protocol. Reliable delivery of messages is guaranteed by the underlying Trusty infrastructure (Trusty Drivers), and the communication is completely asynchronous.

Although this Trusty infrastructure is built by Google for Android OS, it can be applied to any normal world OS (typically a Linux-based OS). The Trusty OS infrastructure in secure world is normal world OS-agnostic. The differences truly depend on the security services that normal world OS would like to have.

Trusty Services

There are many uses for a Trusted Execution Environment such as mobile payments, secure banking, full-disk encryption or file-based encryption, multi-factor authentication, device reset protection, replay-protected persistent storage (secure storage), wireless display (“cast”) of protected content, secure PIN and fingerprint processing, and even malware detection.

In embedded products such as an automotive IVI system, the most important security services requested by customers are keystore and secure storage. In this article we will focus on these two services.

Keystore

Keystore (or Keymaster app in Trusty OS) provides the following services:

• Key generation
• Import and export of asymmetric keys (no key wrapping)
• Import of raw symmetric keys (no key wrapping)
• Asymmetric encryption and decryption with appropriate padding modes
• Asymmetric signing and verification with digesting and appropriate padding modes
• Symmetric encryption and decryption in appropriate modes, including an AEAD mode
• Generation and verification of symmetric message authentication codes

Protocol elements, such as purpose, mode and padding, as well as access control constraints, are specified when keys are generated or imported and are permanently bound to the key, ensuring the key cannot be used in any other way.

In addition to the list above, there is one more service that Keymaster implementations provide, but which is not exposed as an API: Random number generation. This is used internally for generation of keys, Initialization Vectors (IVs), random padding, and other elements of secure protocols that require randomness.

Using Android as an example, Keystore functions are explained in greater details in this Android keymaster functions document

As shown in Figure 179 above, the Keymaster HAL is a dynamically-loadable library used by the Keystore service to provide hardware-backed cryptographic services. To keep things secure, HAL implementations don’t perform any security sensitive operations/algorithms in user space, or even in kernel space. Sensitive operations are delegated to a secure world TEE (Trusty OS) reached through a kernel interface. The purpose of the Keymaster HAL is only to marshal and unmarshal requests to the secure world.

Secure Storage (SS)

Trusty implements a secure storage services (in Secure Storage TA) based on RPMB (Replay Protected Memory Block) partition in eMMC or UFS flash storage. The details of how RPMB works are out of scope in this article. You can read the eMMC/UFS JEDEC specification to understand that.

This secure storage can provide data confidentiality, integrity, and anti-replay protection.Confidentiality is guaranteed by data encryption with a root key derived from the platform chipset’s unique key/secret.

RPMB partition is a fixed size partition (128KB ~ 16MB) in eMMC (or UFS) drive. Users can not change its size after buying an eMMC flash drive from vendor.

This secure storage could be used for anti-rollback in verified boot, for saving authentication (e.g. password/pin) retry attempt failure record to prevent brute-force attacks, for storing Android attestation keybox, or for storing customer’s credential/secrets (e.g. OEM image encryption key). See Android Key and ID Attestation for details.

In Trusty, the secure storage architecture is shown in the figure below. In the secure world, there is a SS (Secure Storage) TA, which has an RPMB authentication key (AuthKey, an HMAC key) and uses this Authkey to talk with the RPMB controller in the eMMC device. Since the eMMC device is controlled by normal world driver, Trusty needs to send an RPMB data frame ( encrypted by hardware-backed unique encryption key and signed by AuthKey) over Trusty IPC channel to Trusty SS proxy daemon, which then forwards RPMB data frame to physical RPMB partition in eMMC.

As shown in Figure 180 above, Trusty SS TA provides two different services simultaneously:

• TD (Tamper-Detection): The Trusty secure file system metadata is stored in RPMB, while the user data (after encrypted with hardware-backed encryption key), is stored in Linux-backed file system in user data partition of eMMC (as shown in Figure above). This type of service supports large amount of data storage. Because of potential data deletion/modification, Trusty OS SS TA provides a mechanism to detect such tampering behaviors (deletion/modification, etc.)
• TP (Tamper-Proof): This is a tamper-resistant secure storage service with much higher level of data protection. In this service, the file system metadata and user data (encrypted) are both stored in RPMB. And both can survive after a factory reset or user data partition wipe. As previously mentioned though, the amount of data storage depends on the eMMC RPMB partition size.

We’ve discussed how this secure storage architecture looks, and what secure storage services Trusty SS TA can provide. Now let’s briefly take a look at how it can be used.

As Figure 181 below shows, an OEM can develop a client App in normal world and a Trusted App (TA) in Trusty OS. The OEM TA then can talk with either TD or TP (or both) of SS TA through Trusty internal process IPC to request TA-specific secure file open/creation/deletion/read/write operations.

Here is a simple example showing data signing:

1. An OEM Client App sends the message that needs signing to the OEM Trusted App in TEE/secure world.
2. The OEM Trusted App retrieves the signing key (that was previously saved into SS TA) from SS TA, and uses it for signing the message, then discard the signing key.
3. The OEM Trusted App sends the signed message (with signature) back to OEM Client App.

In this entire process, the secret signing key is never released outside of secure world.

Trusty in ACRN

ACRN is a flexible, lightweight reference hypervisor, built with real-time and safety-criticality in mind, optimized to streamline embedded development through an open source platform. In this section, we’ll focus on two major components:

• one is the basic idea of secure world and unsecure world isolation (so called one-vm, two-worlds),
• the other one is the secure storage virtualization in ACRN.

See Trusty TEE for additional details of Trusty implementation in ACRN.

One-VM, Two-Worlds

As previously mentioned, Trusty Secure Monitor could be any hypervisor. In the ACRN project the ACRN hypervisor will behave as the secure monitor to schedule in/out Trusty secure world.

As shown in Figure 182 above, the hypervisor creates an isolated secure world UOS to support a Trusty OS running in a UOS on ACRN.

Figure 183 below shows further implementation details. The RHS (right-hand system) is such a secure world in which the Trusty OS runs. The LHS (left-hand system) is the non-secure world system in which a Linux-based system (e.g. Android) runs.

The secure world is configured by the hypervisor so it has read/write access to a non-secure world’s memory space. But non-secure worlds do not have access to a secure world’s memory. This is guaranteed by switching different EPT tables when a world switch (WS) Hypercall is invoked. The WS Hypercall has parameters to specify the services cmd ID requested from the non-secure world.

In the ACRN hypervisor design of the “one VM, two worlds” architecture, there is a single UOS/VM structure per-UOS in the Hypervisor, but two vCPU structures that save the LHS/RHS virtual logical processor states respectively.

Whenever there is a WS (world switch) Hypercall from LHS, the hypervisor copies the LHS CPU contexts from Guest VMCS to the LHS-vCPU structure for saving contexts, and then copies the RHS CPU contexts from RHS-vCPU structure to Guest VMCS. It then does a VMRESUME to RHS, and vice versa! In addition, the EPTP pointer will be updated accordingly in the VMCS (not shown in the picture above).

Secure Storage Virtualization

As previously mentioned, secure storage is one of the security services provided by secure world (TEE/Trusty). In the current ACRN implementation, secure storage is built in the RPMB partition in eMMC (or UFS storage).

Currently the eMMC in the APL SoC platform only has a single RPMB partition for tamper-resistant and anti-replay secure storage. The secure storage (RPMB) is virtualized to support multiple guest UOS VMs. Although newer generations of flash storage (e.g. UFS 3.0, and NVMe) support multiple RPMB partitions, this article only discusses the virtualization solution for single-RPMB flash storage device in APL SoC platform.

Figure 184 shows an overview of the virtualization of secure storage high-level architecture.

In Figure 184, the rKey (RPMB AuthKey) is the physical RPMB authentication key used for data authenticated read/write access between SOS kernel and physical RPMB controller in eMMC device. The VrKey is the virtual RPMB authentication key used for authentication between SOS DM module and its corresponding UOS secure software. Each UOS (if secure storage is supported) has its own VrKey, generated randomly when the DM process starts, and is securely distributed to UOS secure world for each reboot. The rKey is fixed on a specific platform unless the eMMC is replaced with another one.

In the current ACRN project implementation on an APL platform, the rKey is provisioned by the BIOS (SBL) near the end of the platform’s manufacturing process. (The details of physical RPMB key (rKey) provisioning are out of scope for this document.)

For each reboot, the BIOS/SBL retrieves the rKey from CSE FW (or generated from a special unique secret that is retrieved from CSE FW), and SBL hands it off to the ACRN hypervisor, and the hypervisor in turn sends the key to the SOS kernel.

As an example, secure storage virtualization workflow for data write access is like this:

1. UOS Secure world (e.g. Trusty) packs the encrypted data and signs it with the vRPMB authentication key (VrKey), and sends the data along with its signature over the RPMB FE driver in UOS non-secure world.
2. After DM process in SOS receives the data and signature, the vRPMB module in DM verifies them with the shared secret (vRPMB authentication key, VrKey),
3. If verification is success, the vRPMB module does data address remapping (remembering that the multiple UOS VMs share a single physical RPMB partition), and forwards those data to SOS kernel, then kernel packs the data and signs it with the physical RPMB authentication key (rKey). Eventually, the data and its signature will be sent to physical eMMC device.
4. If the verification is successful in the eMMC RPMB controller, the data will be written into the storage device.

The work flow of authenticated data read is very similar to this flow above in reverse order.

Note that there are some security considerations in this architecture:

• The rKey protection is very critical in this system. If the key is leaked, an attacker can change/overwrite the data on RPMB, bypassing the “tamper-resistant & anti-replay” capability.
• Typically, the vRPMB module in DM process of SOS system can filter data access, i.e. it doesn’t allow one UOS to perform read/write access to the data from another UOS VM. If the vRPMB module in DM process is compromised, a UOS could change/overwrite the secure data of other UOSs.

Keeping SOS system as secure as possible is a very important goal in the system security design. In practice, the SOS designer and implementer should obey these following rules (and more):

• Make sure the SOS is a closed system and doesn’t allow users to install any unauthorized 3rd party software or components.
• External peripherals are constrained.
• Enable kernel-based hardening techniques, e.g., dm-verity (to make sure integrity of DM and vBIOS/vOSloaders), kernel module signing, etc.
• Enable system level hardening such as MAC (Mandatory Access Control).

Detailed configurations and policies are out of scope in this article. Good references for OS system security hardening and enhancement include: AGL security and Android security

References:

• Trusty TEE | Android Open Source Project
• Secure Storage (Tamper-resistant and Anti-replay)
• Eddie Dong, ACRN: A Big Little Hypervisor for IoT Development