rustnet-sandbox
Post-initialization sandboxing and root privilege dropping for rustnet, usable by any front-end (the TUI binary or a headless tool) built on the rustnet crates.
One apply_sandbox(&SandboxConfig) -> Result<SandboxReport> entry point with
a platform backend behind it:
- Linux:
PR_SET_NO_NEW_PRIVS, capability drops (CAP_NET_RAW, CAP_BPF, CAP_PERFMON), the root uid drop, and Landlock filesystem/network/scope restrictions (behind thelandlockfeature). - macOS: the root uid drop, then a Seatbelt profile blocking outbound
network, credential reads, and writes outside configured output paths
(behind the
macos-sandboxfeature). - Windows: dangerous token privileges removed and a job object blocking child process creation.
- FreeBSD: the root uid drop (Capsicum is planned).
The privdrop module (Linux/macOS/FreeBSD) resolves the drop target from
SUDO_UID/SUDO_GID (falling back to nobody), chowns pre-created output
files to it, and performs the verified, irreversible drop.
Sandboxing is a post-initialization step: open capture handles, load eBPF programs, and pre-create output files first, apply the sandbox on the main thread, and only then spawn the worker threads that should inherit the restrictions. See the crate docs for the full ordering contract.
The crate deliberately depends on no other rustnet crate: callers pass in paths and an optional drop target, nothing else.