Third-party crates

Introduction

Rust provides a package manager and build system called Cargo. Rust also provides crates.io, its default package registry. In userspace, all these form a solution that a lot of open-source Rust projects use.

Some of those open-source libraries are potentially usable in the kernel because they only depend on core and alloc (rather than std), or because they only provide macro facilities.

Thus it is natural to consider whether some of these libraries could be reused for the kernel.

Suitability of a crate

Even if a library only depends on core and alloc, it may still not be usable within the kernel for other reasons.

For instance, its license may be incompatible, it may not support fallible allocations, the kernel may only need a very small subset of what it supports (even if it supports configuring out some of its features), the kernel may already provide the same functionality on the C side (which could be abstracted), etc.

On top of that, the code of a crate may require some changes to be adapted for the kernel anyway. For instance, adding SPDX license identifiers, removing a dependency, tweaking some code, enabling an unstable feature, etc.

Moreover, new code arriving to the kernel should be maintained; and thus somebody needs to step up for that role.

Therefore, in general, whether a third-party crate is suitable for the kernel needs to be decided on a case-by-case basis.

Importing crates

The kernel currently integrates some dependencies (e.g. some of the compression algorithms or our Rust alloc fork) by importing the files into its source tree, adapted as needed. In other words, they are not fetched/patched on demand.

There have been discussions about potentially incorporating a system where crates/libraries are fetched dynamically given a list of crates, versions, hashes, etc.; however, it remains to be seen whether such a system would be preferred and accepted.

Supporting out-of-tree modules

The project is focused on getting features upstreamed, i.e. available for everybody. Therefore, if mainline does not support third-party crates and/or a system to fetch them dynamically, then it is unlikely it will be supported for out-of-tree modules.

Experiment

Experimental integration for a few popular crates has been provided for interested users, e.g. PR #1007 adds support for proc-macro2, quote, syn, serde and serde_derive.

Feedback

We are looking for feedback from other kernel developers, maintainers and companies on which third-party crates would be most useful to have in the kernel.