Skip to content

Security: sara-star-quant/oxfuse

SECURITY.md

Security

Oxfuse reads graph files, ONNX models and input files, so the parsing and resource-limit code is the part that matters.

What is in scope

  • Crashes, panics, hangs or unbounded memory use caused by a crafted graph file, ONNX model or inputs file. Files are untrusted: sizes, value counts, tensor ranks and element counts are bounded before anything is allocated (see the limits in crates/core/src/limits.rs).
  • Any case where an optimization pass changes the result of a graph.
  • Incorrect results from an operation compared with its documented semantics.

The code of every oxfuse crate is forbid(unsafe_code), makes no network calls and loads no native code from a model. Its dependencies are another matter: wgpu and fearless_simd (the SIMD kernels) use unsafe inside, and their advisories count as ours.

What is not

This is an experimental project and not meant for safety-critical or production use. A graph that is merely slow or heavy within the documented limits is not a vulnerability; a crafted file that hangs the loader or gets past the limits is.

Reporting

Please use GitHub's private vulnerability reporting on this repository ("Security" tab, "Report a vulnerability") instead of a public issue. Reports are handled on a best-effort basis; there is no service-level agreement. Fixes go into the latest release and main; older releases are not patched.

Report at https://github.com/sara-star-quant/oxfuse/security/advisories/new. Past advisories and the list of supported versions live at https://github.com/sara-star-quant/oxfuse/security. Dependency advisories are checked in CI with cargo deny (see CONTRIBUTING.md). A first reply usually comes within a week.

There aren't any published security advisories