Skip to content

dnscrypt-proxy refuses the internal zones, so no name under fzymgc.house leaves a resolver node

Date: 2026-10-06 Status: Accepted Deciders: Sean Brandt Supersedes: docs/adr/2026-09-25-untrusted-legs-resolve-public-dns-only.md, for the names under fzymgc.house only

Since 2026-10-06 one dnscrypt-proxy on each resolver node is the only path from the node to a public resolver. Both Blocky instances forward to it on 127.0.0.1:5300, and the node’s own /etc/resolv.conf names it second, on 127.0.0.2:53. It reaches every resolver through a relay of another operator. The relay sees the home address and no name. The resolver sees the name and the address of the relay.

The main Blocky instance never asks the daemon for a name under fzymgc.house. Its conditional.mapping sends the zone and both reverse zones to pdns, and fallbackUpstream: false keeps a NODATA from pdns on the node. The journals of both nodes confirm it: no .house query of the main instance went anywhere but pdns.

Two callers of the daemon do ask it for such names. blocky-public serves Guest and Telework and has no map to pdns, by the decision of 2026-09-25. A client on those legs that asks for nas.fzymgc.house sends the name through the daemon to a relay and a resolver. The node itself, while its Blocky is down, falls through to the daemon for every name, the internal ones included.

The decision of 2026-09-25 gave Guest and Telework the public answers of the Cloudflare zone fzymgc.house, so that id.fzymgc.house and cluster-docs.docs.fzymgc.house work on a phone on Guest. In the window measured on 2026-10-06, no client on those legs asked for any name under the zone.

The daemon refuses the internal zones. roles/dnscrypt-proxy renders a blocked_names file with fzymgc.house, 168.192.in-addr.arpa and 7.f.3.7.5.6.6.f.b.d.d.f.ip6.arpa, derived from the inventory values dns_zone_domain and dns_reverse_zones. A bare zone in that file matches the zone and every name under it. The daemon answers REFUSED before any upstream step, so no relay and no resolver sees the name.

  • Guest and Telework get REFUSED for every name under fzymgc.house, the public names of the zone included. Sean decided that those legs have no need to see the zone. A phone on Guest that needs id.fzymgc.house uses cellular data for it, as it did before 2026-09-25.
  • The node’s fallback path gets REFUSED for an internal name while Blocky is down. Before this decision it got a NODATA from the public zone, after the name had left the house.
  • The main Blocky instance is not affected. Nothing it sends to the daemon is under the zones.
  • The verify criterion DCP-06 of the role asks each listener for a name under each zone and requires REFUSED.
Alternative Verdict
forwarding_rules in the daemon, sending the zones to pdns on 127.0.0.1:1053 Rejected. The daemon is shared with blocky-public. A forward gives Guest and Telework a path to internal answers, and the fences of those legs turn a correct internal address into a connection that times out. That is the silent failure the ADR of 2026-08-28 named.
Block only lan.fzymgc.house and the reverse zones, and keep the public names of the apex on Guest and Telework Rejected by Sean. The names directly under the apex, such as nas.fzymgc.house, are the ones that matter, and the public zone is due for a review.
A denylist for the zone in blocky-public Rejected. It leaves the node’s own fallback path open, and the rule then lives in two places.
blocked_query_response = 'hinfo' Not needed. It exists for an Android quirk that no client here has shown. REFUSED names the fact.

A Guest or Telework client gets REFUSED for id.fzymgc.house, cluster-docs.docs.fzymgc.house and every other name under the zone. The other decisions of 2026-09-25 stay in force: the second instance, no conditional and no customDNS in it, rebinding protection, the trusted attribute, and the per-instance tracker chk_public_answer. The tracker asks for health.dns.invalid, which Blocky answers itself, so this decision does not touch it.

The ULA PTR query of a Guest or Telework client, which the ADR of 2026-09-25 noted goes upstream, now gets REFUSED on the node.

The public zone fzymgc.house in tf/cloudflare keeps its records, and nothing on the trusted legs ever saw them, because pdns is authoritative there. A review of that zone is a separate task.

  • openspec/changes/archive/2026-10-06-dnscrypt-refuses-internal-apex/: the change.
  • ansible/roles/dnscrypt-proxy/defaults/main.yml: dnscrypt_proxy_blocked_zones.
  • ansible/roles/dnscrypt-proxy/templates/blocked-names.txt.j2: the rendered file.
  • ansible/roles/dnscrypt-proxy/tasks/verify.yml: DCP-06.
  • docs/adr/2026-10-06-pair-resolves-public-names-through-anonymized-relays.md: the daemon.
  • Internal DNS runbook: the operational view.