Skip to content

Security: subh05sus/porthole

Security

SECURITY.md

Security Policy

Porthole terminates processes, modifies OS firewall rules, and — if you opt in — can kill processes unsupervised via porthole daemon. Please take security reports about it seriously, and we will too.

Reporting a vulnerability

Please do not open a public GitHub issue for security vulnerabilities.

Instead, use GitHub's private vulnerability reporting: go to the Security tab on this repository and open a new draft advisory. If that's not available for you, open a regular issue asking for a private contact channel without describing the vulnerability itself.

Please include:

  • Porthole version (porthole --version) and OS/architecture
  • Steps to reproduce
  • What you expected vs. what happened
  • Impact, as you understand it (e.g. "kills a process the user didn't select," "escapes the firewall rule isolation group," "auto-kill daemon acts without the allow-list matching")

We'll acknowledge reports as promptly as we can and keep you updated as a fix is worked on.

Scope

In scope:

  • The kill ladder incorrectly signaling a process other than the one the user selected (in particular, any PID-reuse bypass)
  • porthole firewall rules affecting anything outside porthole's own isolated rule group/chain/anchor
  • porthole daemon acting outside its documented safety rails (disabled-by-default, dry-run-by-default without --live, matching outside the exact allow-list, skipping the pre-kill re-verification, ignoring the rate limit)
  • porthole serve's web dashboard: anything reachable from a non-localhost origin, or a cross-origin request able to trigger a kill
  • Privilege escalation of any kind
  • Memory-safety issues from the Windows PEB-reading code (internal/proc/peb_windows.go, peb_wow64_windows.go) — this code reads undocumented Windows internals and fails closed by design, but a bug there is still worth reporting

Out of scope:

  • A killed process the user explicitly targeted behaving as intended
  • Denial-of-service via scanning/killing ports you already have local-user permission to see and kill (porthole has no privilege beyond what the invoking user already has)
  • Issues requiring physical access or an already-compromised machine

Design notes for reviewers

If you're auditing this codebase, the places worth reading most carefully:

  • internal/kill/kill.go — the ladder's verifyAndSignal, the single choke point every signal passes through
  • internal/daemon/daemon.go — package doc comment lists every safety rail and where it's enforced
  • internal/firewall/*.go — RulePrefix-based ownership check, present in every List/Remove/RemoveAll implementation
  • internal/webui/webui.go — sameOrigin, the CSRF mitigation on the kill endpoint

Supported versions

Only the latest released version is supported with security fixes.

There aren't any published security advisories