ltop is a local-first monitoring tool: the TUI runs unprivileged, reads local kernel interfaces, and — only when you enable it — uses a privileged eBPF helper plus active LAN probes. This document describes the supported versions, how to report vulnerabilities, and the boundaries of the threat model.
Only the latest release line receives security fixes:
| Version | Supported |
|---|---|
| 0.2.x (latest) | ✅ |
| older | ❌ — please upgrade |
Please use GitHub Private Vulnerability Reporting (Security → Report a vulnerability) — do not open a public issue for something exploitable. Include the affected target (Linux/macOS × x86_64/aarch64), a reproduction path, and your assessment of impact.
You will get a response within 7 days. Fixes land in the next patch/minor release; you will be credited in the release notes unless you prefer otherwise.
- The TUI runs unprivileged. All kernel data comes from interfaces any
process may read (
/proc,getifaddrs, PF_ROUTE/netlink dumps, ephemeral UDP sockets). Killing/relaunching is limited to your own uid by the kernel. ltop-bpfis an explicit, root-only helper (systemd service). It attaches kprobes, needsCAP_BPF+CAP_PERFMON(+CAP_SYS_RESOURCEon kernels < 5.11), and writes one snapshot file. The shipped unit sandbox it:NoNewPrivileges,ProtectSystem=strictwith only/run/ltopwritable,ProtectHome,PrivateTmp, kernel-interface protections, and a bounding set with noCAP_SYS_ADMIN. It holds no listening sockets and makes no network connections.
- Session logs (
~/.local/state/ltop/) record connection endpoints, process names and hostnames and are created 0600. /run/ltop/udp.jsonis world-readable by design (the unprivileged TUI reads the root-written snapshot). It contains per-process/per-socket byte totals only — on multi-user hosts this is visible to every local user; keep that in mind before deploying the helper on shared machines.- No telemetry: ltop never sends data anywhere beyond your own LAN probes. There are no update checks, no analytics, no phones-home.
The LAN box queries the local network: unicast mDNS, NetBIOS node-status and reverse-DNS lookups per neighbour, plus a configurable UDP "knock" sweep (one empty datagram to discard port 9 per subnet host, ≤ 1022 hosts, default every 60 s, toggleable). Replies are parsed defensively: transaction IDs are randomized per query and validated together with the source address; the parsers are bounds-checked pure Rust with fuzz-resistant fallbacks. The TUI renders untrusted strings through ratatui (control characters dropped), and the CLI output paths sanitize them explicitly.
--out/--csv/--replayaccept arbitrary paths within the calling user's permissions — they are features, not privilege boundaries.- Exported CSVs of process names can contain spreadsheet-hostile content;
fields are quoted and formula-prefixed (
=/+/-/@get a leading apostrophe), but treat exports as data, not code. - The eBPF
sk_inoLRU has a documented, low-probability mis-attribution window after SLAB address reuse (see the roadmap).
Releases are built from tagged commits with --locked (Cargo.lock is
committed). All GitHub Actions are pinned to commit SHAs; workflows run
least-privilege (CI is read-only; only the release job holds
contents: write, and the tap job's token never persists on the runner).
Dependabot (actions + crates) and secret scanning with push protection are
enabled on the repository.