Firmware Manifest Format

A firmware update mechanism is an essential security feature for IoT devices to deal with vulnerabilities. While the transport of firmware images to the devices themselves is important there are already various techniques available, such as the Lightweight Machine-to-Machine (LwM2M) protocol offering device management of IoT devices. Equally important is the inclusion of meta-data about the conveyed firmware image (in the form of a manifest) and the use of end-to-end security protection to detect modifications and (optionally) to make reverse engineering more difficult. End-to-end security allows the author, who builds the firmware image, to be sure that no other party (including potential adversaries) to install firmware updates on IoT devices with adequate privileges. This authorization process is ensured by the use of dedicated asymmetric keys installed on the IoT device: for use cases where only integrity protection is required it is sufficient to install a trust anchor on the IoT device. For confidentiality protected firmware images it is additionally required to install either one or multiple symmetric or asymmetric keys on the IoT device. Starting security protection by the author is a risk mitigation technique so firmware images and manifests can be stored on untrusted respositories. It is assumed that the reader is familiar with the high-level firmware update architecture . This document is structured as follows: In we describe the main building blocks of the manifest and contains the description of the ASN.1 of the manifest.

The key components of a manifest are shown in and are explained in the sub-sections below.

The Manifest structure is the top-level construct that ties all other structures together. In addition to the structures explained in subsections below it contains: a version number (in the ‘manifestVersion’ attribute) a textual description about the update, including the version / vendor / model of the device (in the ‘text’ attribute). This information is optional. a timestamp indicating when the manifest was created (in the ‘timestamp’ attribute).

The PayloadInfo structure contains information about the firmware image. The ‘format’ attribute contains the firmware image type (such as rawBinary, hexLocationLengthData, ELF). The ‘size’ attribute offers information about the size of the firmware image in bytes. If the size of the obtained firmware image differs from the size stated in the manifest then the obtained image MUST be consider corrupted. The ‘nonce’ attribute contains a (short) random value to ensure that a given manifest is unique. This separates the function of the timestamp, which is provided for rollback protection, from the function of the nonce, which is for uniqueness. Keeping these functions separate ensures that a number of edge cases are catered for, for example: the creation of manifests quickly enough that they have the same timestamp. The ‘storageIdentifier’ attribute indicates where the image should be placed on the device. This value useful, for example, when an IoT device contains multiple microcontrollers (MCUs) and the decision needs to be made to which MCU to send which firmware image. Most importantly, however, the PayloadInfo structure contains a reference to the firmware image (in the ‘reference’ attribute) or the image is embedded inside the PayloadInfo structure (within the ‘integrated’ attribute). A referenced image first needs to be fetched by the device before the update can be applied. The ‘reference’ attribute contains a ‘hash’ and a ‘uri’ attribute: the value in the ‘hash’ attribute allows the device to determine whether it has already obtained this firmware image and, since it is included in the digitally signed manifest, it protects the firmware image against modifications. The ‘uri’ attribute references the image. Finally, a firmware image may be encrypted and information about how to decrypt is provided in this payload in the ‘encryptionInfo’ attribute. The following options are provided: No encryption (mode=”none”). In this case the firmware image is not encrypted and only integrity protected. Encryption using a symmetric key (mode=”preSharedKey”). The assumption is that the symmetric key is pre-provisioned (in an out-of-band fashion) on the IoT device and also available to the developer. Encryption using a symmetric key derived via a key derivation function (mode=”preSharedKeyKdf”). This option is a variation of the symmetric key encryption mode whereby a key derivation function is applied to the pre-provisioned key before it is used for encrypting the firmware image. Encryption using a symmetric key found in the ‘KeyTable’ attribute (mode=”keyTable”). This mode is tailored to use cases where a single encrypted firmware image is transmitted to many IoT devices. Depending on the selected mode different information has to be conveyed in the manifest. When encryption using a symmetric key is selected then the ‘KeyId’ attributes provides information for identifying the appropriate symmetric key. When encryption using a symmetric key derived via a key derivation function is selected then the following three parameters are provided by the ‘KdfParameters’ attribute: KDF algorithm, nonce, and a key id. The computed function KDF(key, nonce). When encryption using the key table is selected then the ‘KeyTable’ attribute is used. shows the concept graphically where the firmware image is encrypted by a symmetric key and this symmetric key is encrypted with the public key of each of the devices.

* (encrypted * . | | . * with key K) * . | +----------------+ | . * * . | |{K}Pub(Device B)| | . ***************** . | +----------------+ | . . +----------------------+ . . . +.............................+ ]]>

The Condition and the Directive structures together allow “If <…> Then <…>” rules to be expressed. It offers the following functionality: Apply an update before a given date only (Directive.applyAfter) Apply an update immediately (Directive.applyImmediately) Apply an update only to devices that match the vendorId, classId, deviceId attributes Apply an update only if the device system time is before the time indicated in the Condition.lastApplicationTime.

In some situations an IoT device may require more than a single firmware update image. To express the requirement that more than a single image has to be installed on a device the dependencies structure is used, which is of type ResourceReference (as used by the PayloadInfo structure). Aliases are used to refer to alternative locations of firmware images. This is useful in environments where organizations cache firmware images (and their corresponding manifests) on premise to avoid the need to fetch imagines from repositories maintained by the developer’s organizations (such a device manufacturer or an OEM).

A device is identified by at least three identifiers: A vendor identifier A device class identifier A device identifier

The vendor ID is a 128-bit number that conforms to RFC-4122, type 5. This number is used by the device to verify manifests. The Vendor ID should be derived from the manufacturer’s domain name using the algorithm defined in Section 4.3 of RFC-4122. A vendor ID is typically compiled into a firmware image since it is static for the lifetime of the firmware.

The device class is a 128-bit number that conforms to RFC-4122, type 5. This number is used by the client to verify manifests. The Device Class ID SHOULD use the Vendor ID as the namespace, but the ID within the namespace can be arbitrary. A class ID is also typically compiled into a firmware image since it is static for the lifetime of the firmware.

The device ID is also a 128-bit number that conforms to RFC-4122. The device ID can come from a variety of sources. For example, a device may obtain this identifier during the manufacturing phase (together with other configuration information and manufacturer-provided credentials). In this case, we recommend using RFC-4122, type 1, where the node ID is the factory tool ID, which provides traceability of a device back to the origin of manufacture. A device ID can also come from on-device resources, such as device unique-ID registers or device identifiers in CPUs. Our recommendation is to provide unique CPU resources to a generator function similar to the one used for the class_id. In this example, the device_info may be a combination of several components, such as: MAC address Device unique identifier Where multiple sources of unique identity are available, they should all be provided to the UUID function, since it combines them to create a single, unique identifier.