Independent Audit of Ionspin’s KMP-Libsodium

Published on

The kotlin-multiplatform-libsodium library from developer Ugljesa Jovanovic, aka ionspin, is a Kotlin Multiplatform library that gives access to the many cryptographic APIs of Libsodium. It currently supports most, but not all, of the platform targets offered by Kotlin.

This library claims to be “just a wrapper around the well known Libsodium library”.
We’re going to test that!

Summary of Findings

The goal of this audit was to verify the functionality in the kotlin-multiplatform-libsodium library. During the course of this audit, no major concerns were found. While there were some areas for improvement and monitoring in future releases, this library appears to do as claimed and provide access to Libsodium on Kotlin Multiplatform.

Versions Audited:

About the Auditor

As a disclaimer up-front: This audit was done by myself in my personal time and was not done in a professional context. While I do work with some cryptography professionally, this audit was not done within the scope or course of my employment, and does not reflect the views of my employer in any way.

For my personal experience: I have been developing software for over 15 years. I have experience with many of the cryptographic APIs that Libsodium uses from a practical context. However, I am not a security analyst by trade, nor have I made material contributions to Libsodium or related cryptography libraries at the time of writing. I do have extensive experience with Kotlin Multiplatform code, making this wrapping library a good candidate for me to audit, since it is expected to be primarily a consumer of the existing Libsodium APIs adapted to Kotlin Multiplatform.

Scope and Methodology

For this audit I have looked at the 0.9.2 tag of the kotlin-multiplatform-libsodium library. I will update this article if and when future audits are conducted.

This audit focused primarily on manual review of files, checking the source code for obvious errors, validating the claim that this is merely a wrapper of the Libsodium library, analyzing dependencies, and verifying the build pipeline for any suspect operations.

This audit does not put the library through any automated security analysis tools.

This is not an audit of Libsodium itself. This library has already been audited by engineers much more qualified than me.

For the purposes of this audit, the sample project included with the repository has not been reviewed.

Findings - Code Analysis

A wrapper library should be very straightforward to review. Ideally, the library would do little more than proxy calls to the underlying library. This can be more complex in practicality. The primary concerns this code analysis is focused on are:
Is the logic implemented in this library warranted?
Does the library implement parts of Libsodium on its own rather than deferring to the underlying library?

The Libsodium methods do appear to be proxied directly to the underlying library as advertised. Little is done beyond transforming the data from a type convenient for use in Kotlin into the arguments needed to be provided to Libsodium.

There are, however, extra utilities provided in this library that are not provided by Libsodium. Examples of this can include simple utility methods, such as the byte array extensions found in StringUtil.kt as well as supplemental properties such as transforming results into readable strings found in ArgonResult. While this is additional functionality on top of the wrapping library, these methods seem largely limited to transforming data for common use cases in Kotlin. These utilities aren’t typically used in the library itself, so they can be avoided if desired without affecting the functionality. I do not see any reason these should raise concern.

One utility method of dubious quality appears in the common code as an extension on UByteArray called fromLittleEndianUByteArrayToBigEndianUByteArray. This method simply reverses the array. This is not sufficient to convert the array as claimed. This method should be avoided. This method does not appear to be used in the library itself and is only provided for convenience.

Area for Concern - Return Codes

My primary area of concern for this library is the way return codes from Libsodium are currently handled. While I did not immediately find any errors, the strategy here seems highly prone to error, and appears very brittle. This area deserves the maintainer's focus.

The return codes for libsodium are fairly straightforward and documented. However, in ionspin’s wrapper library, I find multiple different strategies for checking the return codes: [1] [2] [3].
Multiple of these methods are checking for a specific error code, rather than checking for a success code. Meaning, these methods may incorrectly pass should an unexpected code be returned. This, combined with the variance in strategies, makes this code very brittle and susceptible to breakage.

Findings - Dependency Analysis

The wrapper library has very few external dependencies. This is great news for those concerned with supply chain attacks. The library primarily only depends on language-level jetbrains/kotlin dependencies directly, with a few exceptions.

The primary dependency of concern is a resource-loader library. This library is used by the wrapper to load the Libsodium .so files in Java. While the use case of this seems reasonable, this library has very few maintainers, and does not seem particularly widely used.

The second dependency for concern is the most obvious one, Libsodium itself. The library embeds the Libsodium library directly. However, in many cases the library is being built from a forked repository of ionspin’s as a submodule of the repository. This fork appears to be based on the 1.0.19 version of Libsodium. At the time of writing, this is only one release out of date. However, the snapshot has some *extra* commits included that are not present in Libsodium’s 1.0.19 release.

Four of these extra changes (commits 97f7722, 86a53a9, 0a266e0, ac6d390) can be verified as commits in the Libsodium repository, but were made after the 1.0.19 release, and are likely not verified independently of the subsequent release of 1.0.20. This is likely to include the last commit (ac6d390), which may fix some platform targets for apple devices. However, this should be updated to a tagged version of the library as soon as possible.

The last commit included in the Libsodium fork is by ionspin (49c37e8), and appears to disable the Apple Vision platform as a build target. It’s not clear that this change was entirely necessary. It seems that there would be a benefit to remove this change so that the library can rely on an unmodified commit of Libsodium instead. However, the change does not appear to modify any source files of Libsodium itself.

Conclusion

Overall this library appears to do as claimed, and is a great addition to the Kotlin Multiplatform community. The project is not yet fully stable, but appears to be actively maintained. During this audit I even managed to create a pull-request for a very minor bug, for which the maintainer was responsive and promptly merged. I hope to contribute more to the project as I use it more. I plan to keep an eye on future changes to this project, and will update findings here.