shadow.enabled in housecarl.yaml loads Housecarl's ruleset so that it
blocks, rejects and translates nothing: drops and rejects count and
accept (an accept only ends Housecarl's chain), policies accept and log
what they would drop, NAT only counts, MSS clamping and the flowtable
are left out, and raw rules that do more than accept, drop or reject are
left out with a warning. Comments mark what would happen ([would drop]).
Filter chains run after the other firewall's (+10) or before (-10); NAT
chains always first. Every change is checked by nftables in both forms.
No probation; takeovers and closing connections are refused. The boot
fallback filters nothing. internal/tablewatch puts the table back after
a flush ruleset. Mode in status, healthz, whoami, the web UI, the log
and the tracer; hc ruleset --enforcing. docs/shadow.md.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The boot unit fails closed: if ruleset.nft doesn't load, it loads
ruleset-fallback.nft (render.Fallback), saved beside it from the same
configuration: nothing new in from external zones, established, the
ICMPv6 IPv6 needs and DHCPv6 replies let in, the inside left alone and
masqueraded, and the unit marked failed with a critical message.
housecarld replaces it when it applies; status, the dashboard and
/healthz say while it is in force. no_external_zone now also says the
fallback closes everything without an external zone.
The boot ruleset has no flowtable: nftables on Debian 13's kernel
refuses one naming a device that doesn't exist yet (ppp0 before PPPoE),
and with it the whole file. internal/linkwatch listens to rtnetlink link
messages; housecarld applies again when a flowtable device appears or
comes back as a new interface (by ifindex), under the engine's lock.
docs/flowtable.md: what it does, when to use it and when not, how.
GET /healthz reports firewall (applied, unconfigured, fallback,
missing), changes on probation and the machine's warnings, with 503
unless in force, read at most every five seconds.
Renaming a zone, set or service changes what names it in the same
change (one history entry, one undo); dynamic sets keep their names.
The importer keeps a chain's head in order: once a would-be built-in
stays a rule, the ones after it do too, so the maintainer's ICMPv6
accepts stay behind the invalid drop as written. hc settings set
--flowtable none empties the list.
make test (nft 1.1.3) and make e2e (242 checks, with VLAN
sub-interfaces, the fallback and the flowtable) pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
internal/matrix walks each pair's rules in order, as nftables would, for
new connections: zone → zone (forward), zone → the firewall (input) and
back (output), with the built-ins that act on new connections. A rule
for the whole pair decides it; one for part of its traffic (addresses,
services, times, rates, a quota) is an exception; the rest falls to the
policy. Rules whose addresses all lie in networks connected to other
zones are left out of a pair. GET /matrix (with ?changeset, pending
changes), hc matrix [FROM TO], and a Matrix page whose cells open their
rules.
e2e: 166 checks, among them the matrix of the maintainer's imported
ruleset.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Undo is staged like any edit: in review mode the web UI's Undo, and hc
history undo with -C or HOUSECARL_CHANGESET, add a revert operation
that checks, when replayed, that nothing it touches changed since;
applied, History links the entries. Imports and takeovers still can't
be staged.
- The rules page sums up a chain's built-ins in one line; a window lists
them with what they do and their hits, and switches them.
- Warnings about the machine (dashboard, hc status): nftables.service
enabled (it loads /etc/nftables.conf at boot and flushes every table
when stopped), and the imported table still in force beside
Housecarl's. The docs say to disable nftables.service but not stop it.
- A raw rule with other fields set is refused (raw_conflict) instead of
silently ignoring them; the rule dialog's nftables section hides what
it replaces, and the importer's raw rules carry only their text.
- A dynamic set with configured entries is refused
(set_dynamic_entries): they were silently ignored. The set dialog
shows entries or the entry timeout, as fits.
- The rule dialog's chain choices are just the chains' names.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A change made over the network answers with a one-time confirm_token
and closes its connection (HTTP/2: a GOAWAY). The web UI and hc send the
token straight back to /confirm, which accepts it only for the newest
change, from the same user and API token, the same address, and over a
new connection: proof that the client can still reach the API through
the new ruleset (an established connection would get through anyway). A
change that cuts its client off can't be confirmed so, and rolls back by
itself. The setting manual_confirm (hc settings set --auto-confirm off)
and hc --no-confirm leave confirming to a person. Changes through the
local socket still need no confirmation. e2e: 136 checks, among them a
lockout change rolled back without anyone confirming.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Changesets, as in Jarl: configuration changes are operations that can be
staged in a changeset, checked as they are added (validation and nft -c),
reviewed with the diff of the ruleset, and applied together as one
history entry, then on probation like any change made over the network.
The web UI collects edits for review by default (pending rows, a bar with
Review and Apply now, a Changes page, a switch for live edits); hc stages
with -C ID or HOUSECARL_CHANGESET (checked like JARL_CHANGESET) and has
hc changeset create/list/show/diff/apply/discard/remove. Undo, import and
takeover refuse a changeset. --dry-run shows the ruleset's diff.
Bandwidth and quotas: rate limits in bytes, per source or destination;
quotas per rule or per address that reset daily, weekly, monthly on a
day, or by hand, with their usage carried across reloads, saved for
reboots, shown by hc quotas and on the rules page, and reset with hc rule
reset-quota. Rules that count traffic are checked before established
connections are accepted (so they see every packet) and can't accept.
Per-address limits group IPv6 by /64 by default (--group-v4/-v6).
Per-source rate meters use update. The importer turns meters, byte rates
and quotas into these fields.
The examples in the web UI, hc's help and the docs use documentation
addresses and generic names instead of the maintainer's network.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Zones, address sets (IPv4 and IPv6 mixed, dynamic sets with timeouts) and
services, ordered rules in the early, input, forward and output chains,
port forwards with hairpin NAT and IPv6 pinholes, source NAT, redirects,
and switchable built-ins (loopback, conntrack, ICMPv6, antispoofing, MSS
clamping, DHCPv6 client, flowtable). Every change is validated, checked
with nft -c and applied in one transaction to Housecarl's own table,
inet housecarl; changes made over the network roll back unless
confirmed, and the confirmed ruleset is loaded at boot by
housecarl-ruleset.service. An empty configuration applies nothing.
Also: hit counters and the drop log (nflog), 'hc import' of an existing
nftables ruleset and 'hc takeover', the hc CLI, the REST API and the web
UI, users, scoped tokens (sets:write:NAME), the audit log, the web UI's
certificate through Jarl, Debian packages, docs, unit tests with real
nft 1.1.3, and an e2e test with real traffic (71 checks).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
abrown38andClaude Opus 5.5