plugin-bpf

Version	2026.243.0001
Repo	box/github.com/opencharly/plugin-bpf:v2026.260.1008
Plugin	yes — see the plugin reference

OUT-OF-TREE GENERIC charly BPF/eBPF kernel plugin — the one canonical surface for BPF kernel-feature readiness on any venue (bare host or VM guest): command:bpf (status | lsm | config | probe) plus the verb:bpf declarative check steps any candy can bake into its plan. Reads the kernel state directly and READ-ONLY: /sys/kernel/security/lsm (the BPF-LSM gate cardwire-class tools need), /sys/kernel/btf/vmlinux, CONFIG_BPF_LSM/CONFIG_DEBUG_INFO_BTF, the BPF sysctl knobs, memlock and lockdown. Read-only by default; probe is a verify-only attach-requirements report (a real transient attach probe requires an explicit –attach flag + root + bpftool). Dual-class (plugin-mcp precedent): command:bpf is CLI-dispatched (charly fork/execs the binary), verb:bpf is served over gRPC with a self-contained #BpfInput schema. First consumer: plugin-cardwire’s status gates on the same BPF-LSM facts (R3: one canonical surface, no ad-hoc copies).

Acceptance plan

This candy’s plan: — the runnable spec charly check executes against a live deployment. check: steps are idempotent probes; run: steps change state.

Intent	Step
check	the bpf plugin ships a buildable Go module the host can build + fork/exec (CLI) and compile in (verb); the deterministic end-to-end (bpf status / bpf lsm exit-code contracts) is exercised by the check-bpf-local bed in opencharly/charly