Scudo: Addressing Software Supply Chain Security in Uptane
- PURE: 3
- Title: Scudo: Addressing Software Supply Chain Security in Uptane
- Version: 1
- Last-Modified: 2023-02-28
- Authors: Marina Moore marinamoore@nyu.edu, Aditya Sirish aditya.sirish@nyu.edu, Lois Anne Delong lad278@nyu.edu
- Status: Accepted
- Content-Type: <text/markdown>
- Created: 2022-03-30
Abstract
Software supply chain attacks are on the rise. A 2022 Sonatype report found a 742% increase in the last three years. As a result, companies across all industry sectors have turned their attention to software supply chain security. The need to find solutions quickly is further fueled by rapidly increasing regulations and standards, including President Biden’s U.S.Executive Order 14028.
In this PURE, we propose Scudo, a framework that combines the compromise resilience and secure delivery protection of Uptane and the end-to-end software supply chain security guarantees offered by in-toto. It describes how in-toto metadata can be integrated into an automotive software supply chain that already uses Uptane. It further establishes verification responsibilities for ECUs based on the resources they possess and presents a stop-gap option for automobiles with no sufficiently capable ECUs.
Motivation
Software supply chain attacks are on the rise with the 2022 State of the Software Supply Chain report, noting a 742% increase over the course of the last three years. But, it was the 2020 Sunburst attack that really made both government and corporate CISOs take action. Affecting more than 100 companies and nine government agencies, the cause of the attack was Sunburst malware that had been uploaded via a routine update from Orion software produced by SolarWinds. The attack taught both public and private concerns some hard lessons, such as:
- The very thing designed to protect the security of software-patches and updates that fix routine vulnerabilities-can also be used to deliver malware that can damage or corrupt.
- As software products contain third-party components, program developers are not in complete control of the contents of their products.
- Malware can linger undetected for extended periods of time, allowing it to infect every system that touches it. The Sunburst malicious code was part of an update released by SolarWinds in the spring of 2020, yet it was not reported until December of that year.
Automotive software supply chains are particularly tempting targets for attackers. This is because automotive software runs on a large number of vehicles and updating all of them to recover quickly from an attack is a slow and difficult proposition. As such, a successful software supply chain attack means that thousands (or more) of vehicles are impacted. Any attack that targets a large number of vehicles on roads can do immeasurable damage. Further, while Uptane can securely deliver updates to vehicles to recover from the attack, ensuring every vehicle on the road receives the update immediately is very difficult given the circumstances of use of these vehicles. Therefore, strategies that can secure automotive software supply chains are vital to avoid such attacks altogether.
Governments around the world have recognized this fact. Regulations are forthcoming from the United Nations and other governing bodies, and standards have been published that specify how software running on automobiles must be secured. President Biden of the United States of America issued an Executive Order describing the need to improve the nation's cybersecurity. This EO, in particular, pushes for solutions like Software Bills of Materials (SBOMs) that list the components of software artifacts. This information is vital in ensuring developers are aware of the components in their supply chain so that they can fix vulnerabilities as and when they are discovered. For example, an SBOM would help developers react quickly to situations like 2021's log4j incident by informing them of the software products that require patches or updates.
Unfortunately, SBOMs do not provide end-to-end software supply chain security. They do not guarantee the integrity of either the operations performed as part of the software supply chain, or the artifacts they consume and produce. An SBOM also cannot help with verifying that each operation was performed by authorized actors, or indeed that all required operations were in fact performed. So, the use of SBOMs will not prevent attacks like that against SolarWinds, which used a compromised build server. A more comprehensive framework is necessary and is described in this document.
Reasoning
in-toto is a comprehensive framework that secures each step in the software supply chain. With in-toto, software supply chain owners can enforce policies pertaining to the flow of artifacts through the supply chain, and provide proof that authorized people or bots made changes to artifacts at any given step. Thus, in-toto can be used to secure software from the very first step all the way through the development pipeline. Furthermore, the framework plays well with other systems and tools that attempt to secure specific aspects of the software supply chain. in-toto is integrated into build tools and systems like Jenkins and Tekton, and is compatible with complementary software supply chain security efforts such as SLSA, Grafeas, Sigstore, and SBOMs.
Specification
Note: The reader is expected to be familiar with the in-toto Specification.
We describe how in-toto can be used to achieve software supply chain security alongside the Uptane deployment. In such a model, in-toto can be backed by the Uptane root of trust, and use Uptane's delegation structure to distribute all of in-toto's metadata. Essentially, in-toto metadata is used to attest to the steps performed in the software supply chain itself, while Uptane is used to securely distribute that metadata alongside the ECU images. This combination of in-toto for software supply chain guarantees and Uptane for secure distribution of ECU images and in-toto metadata is called Scudo.
Modifications to the Repositories and Vehicle Interactions
To implement Scudo, Uptane's Image repository will need to be updated to enable it to store in-toto metadata alongside the images. While the Director repository can also be used, the Image repository is recommended as its contents are typically signed for using offline keys.
When an image is submitted to the Image repository, it MUST be accompanied by its in-toto metadata. The in-toto layout and link metadata for every image are noted in the opaque "custom" field of the image's entry in Targets metadata. This way, there is no ambiguity about the in-toto metadata to be used when an image is fetched and verified. Further, as it MAY not be practical for the top level Targets role to issue new metadata for every update, Uptane's delegations model can be used to make this entire process more scalable. Instead, delegated roles can be used to sign for the image itself and its in-toto link metadata. However, in-toto layouts and the key artifacts used to verify them SHOULD be signed for by roles that use offline keys. An example of such a role in typical deployments is the top level Targets role.
As the Image repository handles in-toto metadata for the images, the Director repository has no additional responsibilities than those specified in the Uptane standard. Once the Director repository decides which images are required and signs Targets metadata for them, the vehicle MUST fetch the images and all the corresponding metadata using the interfaces provided by both the repositories. From the Director repository, the automobile receives Uptane metadata only. From the Image repository, the automobile receives Uptane metadata, the image itself, and its corresponding in-toto metadata.
Full Verification ECUs
The Primary ECU receives images and metadata from the repositories, and sends them on to the other ECUs that it updates. Each Full Verification ECU is extended to perform independent in-toto verification that will be secure even if the Primary ECU is compromised. Some Full Verification ECUs, for example the Primary, are also expected to perform in-toto verification on behalf of Partial Verification ECUs that are resource constrained.
For every target image the client needs to install, the corresponding in-toto layout, links, and keys are identified via the custom field of the existing Uptane metadata. These files are available as targets on the Image repository. Therefore, for each installable target, the updater will need to download and verify the Uptane metadata that signs for the in-toto metadata, either the Targets role or a delegated role. If verification passes, the actual in-toto metadata files—layout, links, and public keys—MUST be downloaded and their hashes MUST be compared with the Uptane metadata. With these two checks, Uptane’s secure delivery properties are extended to the in-toto files. These files can then be used with confidence by the client to perform in-toto verification.
Resource Constrained ECUs
There are a number of different types of resource constraints that accompany the addition of in-toto metadata. Some automobiles MAY have no ECUs capable of providing even a basic level of software supply chain security due to bandwidth or memory constraints. Therefore, this PURE presents a "stop-gap" option. In this configuration, the repository verifies the software supply chain of the image and signs an in-toto summary link with the keys used to sign either the top level Targets or delegated signer role. This way, the key used to sign the summary link metadata is securely bootstrapped and can be rotated in case of a compromise.
For each target image, the custom field points to the summary link rather than all the in-toto metadata. The client MUST download and verify the Targets metadata for the summary link and, if this passes, download and verify the hash and signature on the summary link. Once the authenticity of the summary link is established, the client MUST verify that the hashes in the products field of the summary link match those of the image.
The summary link does offer some evidence that the repository performed in-toto verification on behalf of the automobile, but it is not a sufficiently strong replacement for actual verification on the vehicle. A key issue with the stop-gap option is that it turns the Image repository into a single point of failure and therefore can only be trusted as long as the repository remains secure. As such, this model MUST only be adopted as a stop-gap option until the update system can be transitioned to the full model in which in-toto verification happens on individual vehicles.
Another resource that MAY be constrained is the amount of in-vehicle bandwidth available to transfer data between the different ECUs. The addition of in-toto metadata greatly increases the amount of information communicated between the Primary ECU and the other ECUs. In such vehicles, the Primary ECU MUST receive the pertinent in-toto metadata in addition to the ECU image, perform in-toto verification, and then forward the image and (optionally) the relevant Uptane metadata. If the vehicle is configured such that the Uptane metadata is forwarded to the target ECU, then the ECU can perform just the Uptane verification workflow, either the full or partial variant depending on the resources it possesses. However, as the Primary ECU MUST use the Uptane metadata to verify the in-toto metadata it received for the image, it might be sufficient to have it perform both Uptane and in-toto verification on behalf of the target ECU. All non-primary ECUs can then perform just the partial verification workflow. This is an implementation detail left to adopters to decide.
| Type of ECU | Uptane Responsibility | in-toto Responsibility |
|---|---|---|
| Primary ECU | Full Uptane verification for its images, for all in-toto metadata, and (optionally) other ECU images | All in-toto verification |
| Secondary Full Verification ECU | (Optional) Full / partial Uptane verification for its images | No in-toto verification |
| Resource Constrained ECU | (Optional) Partial Uptane verification for its images | No in-toto verification |
Table: Comparison of responsibilities expected of different ECUs deployed with limited in-vehicle bandwidth. The in-toto verification process used can be either the full workflow or the stop-gap option, as this choice does not affect individual ECUs, but only the setup as a whole.
Finally, even if the additional in-vehicle bandwidth overhead is not a concern, a subset of ECUs MAY lack the resources to perform all of Scudo's verification workflows. Note that in the stop-gap option, all ECUs lacked the necessary resources. Uptane already accounts for such secondary ECUs by providing a separate Partial Verification workflow. Scudo assumes that these Partial Verification ECUs are also incapable of performing in-toto verification. In addition, there MAY be Uptane Full Verification ECUs that also lack the resources to perform the added in-toto verification.
For these ECUs, Scudo specifies that other capable ECUs, perhaps the Primary ECU, MUST perform in-toto verification on behalf of the resource constrained ECUs. In this model, while every ECU MAY NOT be verifying the software supply chain of the images it is expected to install, verification is still happening on the individual automobile. Essentially, there is no difference in the update workflow of resource constrained ECUs but they are more robust because of by the supply chain verification happening on the automobile.
| Type of ECU | Uptane Responsibility | in-toto Responsibility |
|---|---|---|
| Primary ECU (Full Verification) | Full Uptane verification for its images and all secondaries | in-toto verification of its images as well as images of resource constrained ECUs |
| Secondary Full Verification ECU | Full Uptane verification for its images | in-toto verification for its images |
| Resource Constrained ECU | Partial Uptane verification for its images | No in-toto verification |
Table: Comparison of responsibilities expected of different ECUs where in-vehicle communication bandwidth is not constrained. As before, the in-toto verification process used can be either the full workflow or the stop-gap option, as this choice does not affect individual ECUs, but only the setup as a whole.
Note that the stop-gap option can be employed in conjunction with the solutions for the other constraints. A vehicle MAY simultaneously have no ECUs capable of full in-toto verification and limited bandwidth for in-vehicle communication. Similarly, a vehicle MAY have no ECUs capable of full in-toto verification and some of these ECUs MAY be incapable of any in-toto verification at all.
Security Analysis
Scudo is an extension for Uptane that adds in-toto's software supply chain security properties. Therefore, its threat model is a combination of the threat models of Uptane for the delivery of the software artifacts, and in-toto for the development of the image as it passes through the software supply chain.
Prototype
in-toto has previously been deployed with Uptane's parent project, The Update Framework (TUF), as described here in Datadog's pipelines. This deployment has been documented as two in-toto Enhancements (ITEs), [ITE-2] and [ITE-3]. Datadog has also published a blog post describing their setup and released open source tooling that combines in-toto's and TUF's verification workflows.
Backwards Compatibility
Backwards compatibility is a key factor when introducing in-toto to an existing Uptane deployment. Merely adding in-toto metadata to Uptane Targets metadata does not change Uptane's verification workflow. Older Uptane clients can continue to performing just Uptane's verification workflow for the images and safely ignore references to in-toto metadata.
In addition, adopters MUST be careful with how they switch their devices over from performing only Uptane verification to also including in-toto verification for their ECU images. This is because there MAY be older images without in-toto metadata that are still the latest for their respective ECUs. In these cases, the workflow MUST NOT expect in-toto metadata. Similarly, the workflow MUST NOT be fooled into skipping in-toto verification for newer images that actually have in-toto metadata.
Finally, as always, in-toto and Uptane implementations SHOULD NOT make backward-incompatible changes to how signatures are generated. This will ensure that previous package managers are able to continue to install new packages. Note that Uptane can otherwise be used to safely rotate the keys for the entire system, including those using different key types, key sizes, signature schemes, and cryptographic hashes. Any backwards incompatible changes that are unavoidable SHOULD be handled with a clear plan.
Copyright
This document has been placed in the public domain.
Acknowledgements
This PURE was inspired by two in-toto Enhancements, ITE-2 and ITE-3, that describe how TUF, Uptane's parent project, and in-toto can be combined.