A CI worker may preempt an in-flight ci-local.sh run (SIGTERM, then SIGKILL
after a grace period) to restart on a newer commit. For that kill to be safe,
the script must clean up after itself and never let a dying run collide with
its restart. It previously had no signal handling, shared the default compose
project name across runs, and tore down each suite only at the suite end.
- Derive a per-run id (honoring FIPS_CI_RUN_ID, else short-sha+random) and
namespace every docker resource to it: a fipsci_<run>_<suite> compose project
per suite and per parallel chaos child, and per-run image tags
(fips-test:<run>, fips-test-app:<run>) retagged to :latest only after both
builds succeed so :latest never points at a half-built image.
- Install a bounded, idempotent teardown trap on SIGTERM/SIGINT (+ EXIT): reap
parallel chaos children, then force-remove this run's docker resources via
the new ci-cleanup.sh, wrapped in timeout so a stuck down cannot wedge it.
- Exit 143 (SIGTERM) / 130 (SIGINT), distinct from 0 (pass) / 1 (failed), so a
preempting worker tells a cancelled run from a real failure.
- Add ci-cleanup.sh (also ci-local.sh --reap): force-removes leftover CI
resources by the com.corganlabs.fips-ci=1 label and the fipsci_ project
prefix, robust to however a prior run died.
- Label every per-suite docker resource so the label sweep reaps it after a
SIGKILL regardless of network name: direct docker run/network resources, the
sidecar compose services, and every per-suite compose network (acl-allowlist,
boringtun, firewall, nat, static, both tor suites, and the chaos generator
template). Parametrize the static/sidecar compose image refs so the per-run
tags are honored.
- Give each parallel chaos child a unique /24 from 10.30.x (a new --subnet
override on the sim CLI, assigned per-child in ci-local.sh) so parallel
children never collide on a shared docker subnet, and a chaos net can never
span a fixed-subnet suite (sidecar/static in 172.20.x). 10.30.x sits outside
docker's default-address-pool range, so an auto-assigned net cannot land on
it either; node IPs derive from the subnet, so no scenario config changes.
Replace the fixed post-first-rekey and post-second-rekey settle sleeps in
the rekey integration test, and the fixed-deadline baseline wait in the
interop test, with the deterministic wait_until_connected progress-aware
polling helper already used for the initial baseline convergence.
Each converted site fails fast with structured diagnostics when the mesh
is stuck and extends its deadline only while pairwise reachability is
still climbing, removing the wall-clock settle windows that produced
intermittent connectivity failures after a rekey.
Add wait_until_connected to the shared convergence helpers: it polls a
suite's own pairwise pings (the signal it actually asserts on), returns
as soon as every pair is reachable, extends its deadline while the
reachable-pair count is still climbing, and gives up only when progress
stalls.
Use it in the rekey, static-mesh, and sidecar suites in place of the
fixed wall-clock baseline timeout and the blind sleep, which timed out
under concurrent CI load while the mesh was still converging.
New integration scenario verifying the early-gate silent-drop behavior
of the inbound max_peers admission check at sustained scale, using the
existing 5-node mesh topology with one node's node.limits.max_peers
lowered to 1. This forces 2 of the cap'd node's 3 configured peers
into a sustained denied state, and asserts via tcpdump that no Msg2
responses go back to those denied peers across a 60s capture window.
A background load-driver restarts the denied peer containers every 15s
to reset their auto-reconnect exponential backoff (5s base / 300s cap),
producing fresh Msg1 bursts each cycle. Without this loop the gate
fires ~3-4 times per denied peer in a 60s window; with restarts the
observed rate is 15 per denied peer (~30 total firings), high enough
that any Msg2 leakage would be caught with strong statistical
confidence.
Local run on this branch: cap'd node-c converged to peer_count=1 with
node-b admitted; nodes d and e sustained-retried as denied; tcpdump
captured 30 inbound Msg1 (len 84) packets from the denied pair and 0
outbound Msg2 (len 104) packets, with final peer_count unchanged.
Files:
testing/static/scripts/admission-cap-test.sh — new test script with
inject-config subcommand (sets node.limits.max_peers) and a
3-phase test driver (converge, capture-with-load, per-peer assert)
testing/ci-local.sh — register admission-cap as a new suite category
(ADMISSION_SUITES), wire run_admission_cap function, add to
run_suite dispatch, list_suites, and the default integration sweep
Together with the existing unit-level coverage in src/node/tests/unit.rs
(handle_msg1_silent_drops_at_cap_for_new_peer with mock-transport Msg2
discriminator, and handle_msg1_admits_existing_peer_at_cap as the
bypass regression guard), the gate's silent-drop behavior is now
verified both at single-firing wire-observable resolution and at
sustained multi-firing cross-process scale.
Closes the eventually-consistent gap in spanning-tree state
distribution. Every existing send_tree_announce_to_all call site
gates on a local state-change event (parent switch, self-root
promotion, ancestry change, peer promotion, parent loss). Once a
partition latches — for example a parent-switch announce stranded
in the brief cross-init handshake swap window, where the announce
arrives on a session-index whose decrypt-worker entry has been
unregistered — neither side's state changes again, so neither
side ever re-broadcasts. The existing 60 s check_periodic_parent_reeval
was a re-evaluation, not a re-broadcast: it short-circuited
silently on no-change. Production-side healing depended on
incidental link churn; lab harnesses with stable docker-bridge
links had no equivalent path.
Add a final else branch that fires send_tree_announce_to_all
unconditionally on the no-change path, alongside the existing
switch and self-promote arms. Receivers coalesce by sequence
comparison (ParentDeclaration::is_fresher_than) and short-circuit
at the `if !updated` gate in handle_tree_announce; same-sequence
repeats drop silently with no cascade. The per-peer 500 ms
rate-limiter is well below this 60 s cadence and does not suppress
the heartbeat broadcast.
The fix is a general protocol-robustness improvement: it addresses
any in-flight TreeAnnounce loss class, not only the specific
cross-init swap-window drop site.
testing/static/scripts/rekey-test.sh BASELINE_CONVERGENCE_TIMEOUT
60 -> 65 so a partition healed by the periodic broadcast at T+60
lands inside the convergence window. wait_for_full_baseline
early-exits on PASS, so successful reps see no extra wall-clock.
An FSP session rekey could leave the two endpoints holding different
key sets for a brief window: if a handshake message was lost in
transit, one side rotated to the new keys while the other did not.
Traffic sealed in one key epoch then reached a peer still on the
other epoch and failed to decrypt, producing bursts of AEAD
decryption failures and dropped connectivity until a later rekey
cycle reconverged the pair. Choreographing the cutover order cannot
close this window: any fixed ordering still leaves a skew that
packet reordering widens.
Make rekey correctness independent of cutover timing by overlapping
the key epochs on the receive path. During a rekey transition the
receiver trial-decrypts each frame against every live session it
holds: current, the not-yet-promoted pending session, and the
draining previous session. The K-bit becomes a hint that orders the
trial-decrypt cascade rather than a hard gate, and a frame that
authenticates against the pending session is itself the cutover
signal. No rotation ordering and no packet reordering can then cause
a decryption failure.
The pre-rekey Noise session is held in the `previous` slot until the
peer has demonstrably moved off it. Its drain deadline is anchored
on the most recent frame the peer authenticated against that slot,
refreshed each time the trial-decrypt cascade lands there, rather
than on a fixed wall-clock timer started unilaterally at the local
cutover. A peer that never received the new keys keeps authenticating
against `previous` and the slot stays live; without this, a fixed
timer would erase the only key set that could decrypt the peer's
frames, producing a permanent silent decrypt failure on a live data
path. A peer that never catches up is handled by the existing FSP
session liveness path rather than by silent decrypt failure.
The lost-handshake liveness gap is closed separately by retransmitting
the third rekey handshake message until the peer is confirmed on the
new keys, with a bounded retry budget after which the rekey cycle is
cleanly abandoned and retried on the next timer.
Adds unit tests covering the trial-decrypt cascade (epoch selection,
promotion on pending decrypt, reordered old-epoch stragglers after
cutover, per-slot replay-window integrity), the msg3 retransmission
lifecycle, and the peer-progress-aware drain retirement.
Moves both AEAD layers (ChaCha20-Poly1305, one round per layer per
packet) plus the sendmsg syscall off the rx_loop task onto a per-shard
worker pool, adds per-peer connect(2)-ed UDP with SO_REUSEPORT, and
uses Linux UDP GSO (sendmsg+UDP_SEGMENT — kernel splits one super-skb
into N on-the-wire datagrams in a single TX-stack walk) when packets
in a batch are uniform-size. Same kernel primitive WireGuard's
in-kernel module and BoringTun use to hit 2.5–3.2 Gbps single-stream.
Single TCP stream on a 5-node docker-bridge mesh, 5 x 15 s x P=1:
A→D: 1379 → 2708 Mbps (1.96x, RTT +0.12 ms)
A→E: 1394 → 2663 Mbps (1.91x, RTT +0.11 ms)
E→A: 1406 → 2624 Mbps (1.87x, RTT +0.19 ms)
Static-peer pairs only — every CoV under 3%, 0 outliers, 0% ICMP
loss. The ~+100 µs RTT is the worker queue handoff cost; AEAD +
sendmmsg now run on a separate core in exchange.
What lands:
- src/node/encrypt_worker.rs: std::thread + crossbeam_channel
workers; hash-by-destination dispatch pins a TCP flow to one
worker so wire ordering is preserved; per-worker sendmmsg(2)
batching up to 32; Linux uses sendmsg(2)+UDP_SEGMENT when
packets in a group are uniform-size.
- src/node/decrypt_worker.rs: receive-side mirror. Each shard owns
its session's recv cipher + replay window in a thread-local
HashMap (no shared RwLock/Mutex). Sessions are handed off at
promote_connection and re-registered on K-bit flip / rekey
cutover.
- src/node/handlers/session.rs try_send_session_data_pipelined:
FSP+FMP both seal in-place in the worker on one wire-buffer
alloc; no intermediate inner_plaintext / fsp_payload Vecs.
- src/transport/udp/connected_peer.rs + peer_drain.rs: per-peer
connect(2)-ed UDP socket with SO_REUSEPORT (set on the listen
socket too — without that, EADDRINUSE on activation and every
packet falls back to the wildcard path); the worker sends with
msg_name=NULL and the kernel uses its cached 5-tuple. Tick-
driven activation in handlers/connected_udp.rs, idempotent.
- src/transport/udp/mod.rs: mem::replace the recvmmsg backing buffer
instead of buf.to_vec() per packet — single pointer swap, no
MTU-sized memcpy.
- src/protocol/link.rs SessionDatagramRef: zero-copy borrowed view
used by handle_session_datagram for the bulk local-delivery
path; handle_session_payload takes the borrowed payload
directly (no payload[35..].to_vec()).
- src/transport/mod.rs TransportAddr::from_socket_addr: collapses
the two-alloc from_string(addr.to_string()) pattern to one.
- src/node/handlers/rx_loop.rs: decrypt-fallback drain promoted
ahead of packet_rx in the select! (TCP ACK starvation fix);
interleaved fallback drain every 32 packets inside the rx burst
loop.
- noise::Session: send_cipher_clone / recv_cipher_clone /
recv_replay_snapshot_owned / take_send_counter / accept_replay
so off-task workers can hold a cloned cipher + reserved counter
while the dispatcher keeps replay/counter sequencing serial.
CipherState::cipher_clone returns a refcount-bumped LessSafeKey.
AsyncUdpSocket: AsRawFd so workers issue raw sendmmsg / sendmsg
without going through the tokio reactor.
- Worker pool sizing: both default to num_cpus, overridable via
FIPS_ENCRYPT_WORKERS=N / FIPS_DECRYPT_WORKERS=N. Per-peer
connected UDP can be disabled via FIPS_CONNECTED_UDP=0.
- src/perf_profile.rs: optional per-stage timing reporter under
FIPS_PERF=1 (or FIPS_PIPELINE_TRACE=1). Off by default; zero
overhead when disabled.
- All cfg(unix)-gated. Windows continues on the existing tokio-
based send/recv.
Decrypt worker session lifecycle:
- Node::unregister_decrypt_worker_session mirrors the existing
register helper. Wired at the two natural sites that already
iterate peers_by_index: the rekey drain-completion block in
handlers/rekey.rs (drops the worker entry for the old our_index
once the drain window has expired and the cache_key is
unreachable to any in-flight OLD-K packet), and remove_active_peer
in handlers/dispatch.rs (drops the worker entry for each of the
four index slots: current, rekey, pending, previous). Only
our_index is normally registered; unregister_session is fire-
and-forget for missing entries, so calling unconditionally on
all four slots is correct and bounds the cleanup without per-
slot accounting. Without these callers the per-worker sessions
HashMap and the Node's decrypt_registered_sessions set would
grow monotonically per rekey on long-lived peers.
Testing:
- testing/static/scripts/bench-multirun.sh: multi-run iperf3 +
ping bench. N reruns (default 5), median / min / max / CoV % /
per-run outlier flag, avg ping RTT, ICMP loss %, TCP retransmit
total. Plain client→dest labels + topology header. Pre-bench
peer-convergence check (FIPS_BENCH_CONVERGE_SECS, default 15);
per-path route verification via stats.bytes_sent deltas — fails
fast if traffic exits via a non-static-peer link.
- testing/static/docker-compose.yml: passes FIPS_ENCRYPT_WORKERS /
FIPS_DECRYPT_WORKERS / FIPS_PERF through to containers for A/B
benchmarking without rebuilds.
- testing/static/scripts/iperf-test.sh: same plain client→dest
labels + topology header (was multihop/direct/N hop, which
conflated topology distance with on-wire path).
- .config/nextest.toml: synthetic UDP node tests serialized
through a max-threads=1 test group. Localhost handshakes drop
on shared CI runners under parallel load; one-at-a-time keeps
assertions reliable.
- src/node/tests/spanning_tree.rs: repair_missing_edge_handshakes
— retries up to 5 times for synthetic edges whose msg1 was
dropped, with a drain after each edge retry instead of after
each attempt's full burst.
- src/node/decrypt_worker.rs::tests: two unit tests asserting
WorkerMsg::UnregisterSession removes the worker-thread session
HashMap entry (handle_msg_unregister_session_removes_entry) and
is a no-op for never-seen cache_keys
(handle_msg_unregister_session_idempotent_on_unknown_key), which
is the safety invariant the unconditional unregister calls at
the four index slots in remove_active_peer rely on.
- src/node/encrypt_worker.rs::unix_tests
pipelined_send_wire_layout_roundtrips_canonical_decoders: mirrors
the encoder geometry of try_send_session_data_pipelined (no
coords, the common established-session path), runs the worker's
real seal + send via flush_direct_batch_sync, and decodes the
resulting wire packet using only canonical receive-side decoders
(EncryptedHeader::parse, SessionDatagramRef::decode, FSP header
parse, noise::open). Any divergence between the hand-rolled
encoder offsets (fsp_aad_offset, fsp_plaintext_offset) and the
decoders fails at one of the parse / open / decode steps before
the inner-plaintext assertion fires. Complements the existing
fsp_preseal_runs_before_outer_fmp_seal test which covers the
seal-ordering invariant with synthetic headers but does not
exercise the wire-layout invariant.
CHANGELOG.md [Unreleased] # Changed entry added describing the
worker-pool threading model, hash-by-destination dispatch,
sendmmsg/UDP_GSO, per-peer connected UDP, the operator-facing env
vars, and the bench numbers above.
Cherry-picks from mmalmi/master (paths translated from
crates/fips-core/src/ to src/): 9b7c723, 0deb5cb, 13f7339, e036c0e,
3740a68, 3792f83, 8510193, 4910b07, e53f545, e4e2896, 5fe4af5,
1d01ada, 8c37008, e12469e, 6eb2860.
Co-authored-by: Johnathan Corgan <johnathan@corganlabs.com>
The Phase 1, Phase 3, and Phase 5 strict asserts each fire a
single ping per directed pair. Under low-level packet loss
(e.g. 1% i.i.d. per-direction loss from a CI runner under
pressure), a single-shot round-trip fails at ~2% per pair, so a
20-pair strict assert misses with probability
1 - (0.98)^20 = ~33% per phase from ICMP noise alone, well above
the routing-state signal the asserts are meant to catch.
ping_one gains a max_attempts parameter (default 1, preserving
existing call sites). On failure it retries up to
MAX_PING_ATTEMPTS-1 additional times with PING_RETRY_DELAY
seconds between attempts. Per-pair worst case under the defaults
(4 attempts, 1 s spacing, 5 s ping6 -W timeout) is 4*5 + 3 = 23 s;
per-rep worst case scales with the failing-pair count.
Successful retries log "OK (RTT, attempt N)"; exhausted retries
log "FAIL (after N attempts)".
The retry budget is wired into all three strict asserts:
- Phase 1 final ping_all (after wait_for_full_baseline converges)
- Phase 3 ping_all (post-first-rekey)
- Phase 5 ping_all (post-second-rekey)
The wait_for_full_baseline convergence loop itself stays
single-shot. Its job is to detect when the mesh first sees a
fully clean 20-pair batch, and retries inside the loop would
conflate transient ping loss with still-converging routing
state.
No daemon code changes.
New testing/boringtun/ harness runs two Cloudflare BoringTun userspace
WireGuard containers with iperf3 between them, giving a single-hop
userspace tunnel baseline for comparison against FIPS throughput
numbers. Local WG key generation runs through the harness image so the
host needs no wireguard-tools.
New testing/static/scripts/iperf-compare-refs.sh builds two git refs
into separate fips-test:* images via git worktree and runs the same
static iperf topology against both, with RUNS-based repetition and
aggregate avg/min/max reporting.
testing/static/scripts/iperf-test.sh gains DURATION, PARALLEL,
SETTLE_SECONDS, IPERF_TIMEOUT env knobs and a per-path iperf timeout.
testing/static/docker-compose.yml selects the image under test via
FIPS_TEST_IMAGE; testing/scripts/build.sh respects CARGO_TARGET_DIR.
Author benchmark on aarch64 Docker Desktop:
boringtun bob -> alice : 1000.13 Mbits/sec
Phase 5's per-pair connectivity check ran immediately after the second
rekey cycle, with no settle for routing reconvergence. Under
GitHub-runner CPU contention, post-rekey parent-switches and
coord-cache flushes can take longer than the per-ping 5s timeout for
a small fraction of pairs (1-3 of 20 typically), even though the
rekey mechanism itself completes cleanly. Phase 6 log analysis stays
all-green on these failed runs; the failure is purely connectivity
timing.
Mirror Phase 3's existing 12-second settle pattern: reuse REKEY_SETTLE
and emit the same banner shape. Two-line change. Cost on the success
path is a fixed 12s per suite run.
The Phase 1 pre-rekey baseline in `wait_for_full_baseline`
occasionally times out on GitHub-hosted runners with one ping pair
failing to converge inside the BASELINE_CONVERGENCE_TIMEOUT window.
Phases 2–6 always pass cleanly when this happens — the rekey itself
is fine, the mesh just hasn't finished spanning-tree + bloom-filter
convergence by the time Phase 1 starts pinging.
The wait loop returns as soon as all 20 pairs converge, so the cost
on the success path is unchanged (typical local CI returns well
under the old 36s). The bump only adds headroom on the failure
path. Operators previously worked around this with
`gh run rerun --failed`; this aims to retire that workaround for
the IK/XK lines.
Distinct from the next-branch XX rekey dual-init race, which is a
real protocol bug tracked separately.
The sidecar entrypoint hardcoded `udp.mtu: 1472`, the Docker-bridge
IPv4 maximum (1500 MTU - 8 UDP - 20 IPv4 header). Promote it to
`FIPS_UDP_MTU` (defaulting to 1472, preserving behavior) so non-Docker
reuses of the example can override without editing the script. Plumb
the env var through `docker-compose.yml` so a host-level setting
reaches the container, and add a row to the README's env-var table.
Also annotate the static-CI node template with a comment explaining
the same 1472 rationale and the daemon's 1280 default. The template
keeps 1472 as the literal value since the CI suite runs on a Docker
bridge where that's correct.
No behavior change unless the host explicitly sets FIPS_UDP_MTU.
Operator-facing IPs in user-visible configs/docs (examples, tutorials,
packaging, sidecar templates) are now the resolvable hostnames of the
public test fleet (test-us01.fips.network, etc.) so they keep working
without baking specific addresses into examples.
Doc-comment and test fixtures in src/config/transport.rs use RFC 5737
TEST-NET-2 (198.51.100.1) so they cannot accidentally point at a real
host.
Also resyncs the openwrt-ipk fips.yaml with the common reference
(merge from master) and applies the same DNS-name swap there.
Extend the gateway integration suite with three previously unexercised
runtime paths. All three share testing/static/scripts/gateway-test.sh
and testing/static/docker-compose.yml so they land as one commit.
6A — UDP port forwarding runtime path. Add udp 18081 -> [fd02::20]:8081
to inject_gateway_config() and a phase-7 case where gw-client runs an
inline Python UDP echo server bound [::]:8081 and gw-server sends a
UDP probe to [GW_MESH_IP]:18081 via inline python3, asserting the
echoed payload prefix. The config layer already accepted proto: udp
(test_port_forwards_same_port_different_proto_ok) but the UDP NAT
rule shape and conntrack handling differ from TCP and were unverified.
Uses Python rather than socat because fips-test:latest does not ship
socat; nc -u IPv6 round-trip semantics are messier than a Python
one-liner.
6B — Second simultaneous TCP forward. Add tcp 18082 -> [fd02::20]:8081
alongside the existing 18080 forward. Phase 7 now greps the daemon's
nft DNAT table for all three rules (18080, 18082, 18081) and runs
HTTP fetches through both TCP forwards with distinct backend payloads
(inbound-forward-ok vs inbound-forward-ok-2) so a misrouted response
fails the assertion.
11A — Concurrent multi-client flows. Add gw-client-2 service to
docker-compose mirroring gw-client (IPv6 fd02::21, IPv4 172.20.1.21).
Phase 3 sets the fd01::/112 route on both. Phase 4 issues DNS lookups
from both, asserts they receive distinct virtual IPs, and queries the
gateway control socket (show_mappings) to confirm exactly 2 active
mappings (5-attempt retry loop tolerates snapshot-publish lag). Phase
5 launches both curl requests concurrently as background processes,
asserts each response. Validates concurrent NAT mappings, pool
contention, proxy NDP under simultaneous LAN-client traffic — all
real-world deployment shape that was not pinned.
Phase 8 reclamation timing unchanged (TTL=5s, grace=5s, 25s wait
covers both mapping ticks generously even with a slight stagger).
Companion to the ethernet `accept_connections: false` rekey-deadlock
fix from earlier this release: the same dual-init failure mode shows
up over UDP when peers register by hostname, and the existing
addr_to_link-only carve-out in `should_admit_msg1` doesn't cover it.
The carve-out's first predicate keys `addr_to_link` by the literal
`TransportAddr` that `initiate_connection` inserted, which is the
hostname-form when a peer config carries a hostname (e.g.,
`core-vm.tail65015.ts.net:2121`). Inbound packets always arrive with
numeric source addrs because `udp_receive_loop` builds the
`TransportAddr` from the `SocketAddr` the kernel reports via
`recvfrom`. `TransportAddr` equality is byte-exact, so the two forms
don't match and the lookup misses. With `udp.accept_connections:
false` (or `udp.outbound_only: true`, which forces it false) the
gate then rejects the rekey msg1 from an established peer. The
dual-init tie-breaker stalls because the loser side never produces
msg2; both sides retry indefinitely and the winner side keeps
logging "Dual rekey initiation: we win, dropping their msg1" at 1Hz.
The earlier ethernet fix didn't generalize to this variant because
ethernet TransportAddrs are always numeric MAC bytes — both the
config-time form and the inbound-arrival form match identically.
Add a second predicate to `should_admit_msg1`: an active peer's
`current_addr()` matching `(transport_id, remote_addr)`.
`current_addr` is updated and refreshed from inbound encrypted-frame
source addrs (`handlers/encrypted.rs`), which are always numeric
`SocketAddr`-form, so this catches the established peer regardless
of how its `addr_to_link` key was originally inserted. The fast
`addr_to_link` check stays first; the iteration over peers is
bounded by peer count and only runs when the first predicate misses.
Regression coverage in this commit:
- Unit test `test_should_admit_msg1_admits_rekey_when_addr_form_differs`
in `src/node/tests/handshake.rs`. Constructs the failing scenario
in-process: `addr_to_link` populated with hostname-form key, peer's
`current_addr` at the resolved numeric form, query with numeric form.
Without the new predicate this fails immediately.
- New integration topology `rekey-outbound-only` plus matching
docker-compose profile. Same 5-node mesh shape as `rekey-accept-off`
but `inject-config` sets `udp.outbound_only: true` on node-b and
rewrites node-b's peer-c address from the numeric docker IP to the
docker hostname (`node-c:2121`), reproducing the production
hostname-vs-numeric mismatch. The test asserts no sustained
"Dual rekey initiation: we win" log lines on any node (>10 = bug)
and the existing rekey health checks catch the connectivity loss
the loop produces.
- `testing/ci-local.sh` and `.github/workflows/ci.yml` extended to
run the new variant in the local sweep and the GitHub CI integration
matrix alongside `rekey` and `rekey-accept-off`.
Verified locally: full `bash testing/ci-local.sh` sweep passes 29/29
suites (23m 12s) with the new variant green; 1084 unit tests pass.
Single combined commit covering five interlocking pieces of test and
CI work that landed during the v0.3.0-prep cycle.
## fips-gateway robustness
- src/bin/fips-gateway.rs DNS upstream probe converted from a 3-second
hard-fail to a bounded retry loop (5 attempts × 1s timeout, 1s sleep
between attempts; ~10s worst case). Covers the cold-boot race where
the daemon's TUN is up but the DNS responder at [::1]:5354 is still
binding. Each failed attempt logs at INFO. In production the binary's
retry is the live recovery mechanism; with retry it recovers silently
instead of relying on Restart=on-failure (~5s blip + spurious ERROR
per cycle).
- packaging/debian/fips-gateway.service `ExecStartPre` now waits up to
30 seconds for the daemon's `fips0` TUN to appear before exec'ing
the gateway binary. Eliminates the cold-boot race where the gateway
exits with `fips0 interface not found` and recovers via
`Restart=on-failure`, producing a 5-second blip and a spurious error
log per restart cycle.
- testing/docker/entrypoint.sh gateway-mode waits up to 30s for the
daemon's DNS responder to bind [::1]:5354 (probes once per second
with `dig @::1 -p 5354 ... test.fips`) before exec'ing fips-gateway.
Belt-and-suspenders with the binary's own retry: in CI we want
deterministic startup ordering. On timeout, fall through so the
binary's probe reports the definitive error.
## Test infrastructure DNS bind migration to ::1
After session 359's daemon DNS-bind default flipped from `127.0.0.1`
to `::1` (the production fix for ISSUE-2026-0002), the static-test
infrastructure was carrying a stale workaround that overrode the
default back to IPv4 loopback. The fips-gateway integration test
exposed the divergence: the gateway probes its DNS upstream at
`[::1]:5354` (production default) while the daemon was binding
`127.0.0.1:5354` from the template override — IPv6-explicit sockets
do not accept v4-mapped traffic, so the upstream probe exhausted
retries and the gateway exited.
- Drop the explicit `bind_addr: "127.0.0.1"` line from every test
config that emits it: testing/static/configs/node.template.yaml,
testing/chaos/configs/node.template.yaml, the sidecar heredoc in
testing/docker/entrypoint.sh, testing/acl-allowlist/generate-configs.sh
(six per-node blocks), testing/nat/scripts/generate-configs.sh, and
the four tor templates under testing/tor/. Daemon picks up its
production `::1` default.
- Flip the dnsmasq forwarder for `.fips` in testing/docker/Dockerfile
from `127.0.0.1#5354` to `::1#5354` so dnsmasq on the shared test
image continues to reach the daemon. Template and Dockerfile must
move together since most static suites resolve `<npub>.fips` via
the test-image dnsmasq.
## rekey-accept-off integration variant + UDP unit test
- New `rekey-accept-off` topology and docker-compose profile under
testing/static/. 2-node variant where node-b runs with
`udp.accept_connections: false`. Pins the regression class that
ISSUE-2026-0004 fixed (cross-connection winner's rekey msg1 was
being filtered by the accept_connections gate, breaking rekey).
- testing/static/scripts/rekey-test.sh accepts REKEY_TOPOLOGY and
REKEY_ACCEPT_OFF_NODES env vars; its inject-config subcommand
applies the per-node `udp.accept_connections: false` edit, and
the test asserts no sustained "Dual rekey initiation" log lines.
- New UDP variant of `should_admit_msg1` admit-rekey unit test in
src/node/tests/handshake.rs.
## ci-local.sh full integration coverage
- New runner functions and dispatcher entries for `acl-allowlist`,
`nat-cone` / `nat-symmetric` / `nat-lan`, `rekey-accept-off`,
`dns-resolver`, `deb-install`. Each integrates with the existing
summary tracking via `record`.
- New `--with-tor` flag (off by default) gates `tor-socks5-outbound`
and `tor-directory-mode` runners. Tor stays opt-in because both
harnesses depend on the live Tor network and would introduce a
flake source unrelated to the FIPS code.
- New suite arrays (`ACL_SUITES`, `NAT_SUITES`, `DNS_RESOLVER_SUITES`,
`DEB_INSTALL_SUITES`, `TOR_SUITES`) drive both the default sweep
and `--list` output.
- `run_suite` extended to accept the new suite names for `--only`
invocations.
## GitHub CI matrix expansions
- `gateway` matrix entry runs testing/static/scripts/gateway-test.sh
against the existing docker-compose `gateway` profile.
- `rekey-accept-off` matrix entry exercises the new topology with
REKEY_ACCEPT_OFF_NODES=b.
- `deb-install` matrix (debian12 + ubuntu24 + ubuntu26) runs
testing/deb-install/test.sh with privileged systemd containers.
~5-7 min cold cache, ~2 min warm per distro. Self-contained: builds
its own .deb in a Debian 12 cargo-deb builder image; does not
depend on the build job's pre-built artifact.
- `dns-resolver` matrix entry runs the full 13-scenario harness
(per-distro systemd resolver-backend tests + real-fips end-to-end
scenarios) in a single job. Pins the production DNS bind path that
ISSUE-2026-0002 lived in. ~7-12 min warm, ~12-15 min cold.
Verified locally: full `bash testing/ci-local.sh` sweep passes,
including 5/5 deb-install distros and all 13 dns-resolver scenarios.
Tor-inclusive sweep (`--with-tor`) verified in a follow-up run.
- testing/static/scripts/rekey-test.sh: bump REKEY_SETTLE from 5s to
12s so post-rekey ping_all samples are taken after the 10s old-
session drain window has closed (DRAIN_WINDOW_SECS in
handlers/rekey.rs). The 5s sample window straddled the drain,
occasionally catching old-session decrypts mid-cutover and producing
spurious ping failures.
- testing/lib/log_analysis.py: discovery-succeeded match string is
"proof verified, route cached" (the wording in handlers/discovery.rs);
the regex looked for the older "caching route" wording, so the chaos
analysis summary always reported `Succeeded: 0` regardless of how
many successes the raw node logs actually showed.
- testing/scripts/build.sh: drop the second cargo invocation that
passed `--features gateway --bin fips-gateway`. PR #79 removed the
`gateway` cargo feature in favor of platform cfg gates, so this
invocation now fails with `the package 'fips' does not contain this
feature: gateway`. The first cargo build already produces
fips-gateway as a workspace binary on every supported target.
Mesh peers can now reach a configured host:port on the gateway's LAN
via static port-forward rules on fips-gateway. Mirror of the outbound
LAN gateway (IDEA-0079 / TASK-2026-0056).
Config: new gateway.port_forwards list of { listen_port, proto,
target } entries. Targets are SocketAddrV6 — IPv4 is rejected at
parse time. Validation rejects zero listen ports and duplicate
(listen_port, proto) pairs.
NAT: NatManager gains a port_forwards field and set_port_forwards()
setter, rebuilt in the same atomic rustables batch as the address
mappings. Each forward emits a prerouting DNAT rule keyed on
(iifname fips0, nfproto ipv6, l4proto, tcp/udp dport) that rewrites
destination address and port via Nat::with_ip_register +
with_port_register. When any forwards are configured, a single
LAN-side masquerade is installed on (iifname fips0, oifname
lan_interface, nfproto ipv6) so the LAN host sees the gateway as
source and replies flow back through conntrack. This rule is
distinct from the existing oifname fips0 masquerade that serves the
outbound pool.
Binary: fips-gateway validates port_forwards at startup and calls
set_port_forwards after NatManager construction; startup failure
cleans up the nftables table before exiting.
Test: extend testing/static/scripts/gateway-test.sh with Phase 7
that runs a marker HTTP server on the LAN-side client (fd02::20:8080)
and, from the mesh peer, curls the gateway's fips0 address on port
18080 to exercise the full DNAT + LAN masquerade path. The
LAN-side HTTP server is started with 'docker exec -d' plus a
bind-ready poll on ss; 'docker exec bash -c "cmd &"' does not
keep the child alive past the exec session even with nohup.
Test-infra: ci-local.sh now builds/clippies/tests with
--features gateway, matching GitHub CI. Without this the release
fips-gateway binary silently stays stale across runs, since
cargo build --release alone does not compile the gateway bin.
Verified locally: cargo test --features gateway --lib (991 pass),
clippy + fmt clean, full testing/ci-local.sh green (21/21 suites
in 8m36s, including the new gateway Phase 7).
macOS platform:
- Platform-native TUN interface management with shutdown pipe
- Raw Ethernet transport with macOS socket backend (socket_macos.rs)
- EthernetTransport and TransportHandle::Ethernet ungated from Linux-only
- macOS .pkg packaging (build-pkg.sh, launchd plist, uninstall script)
- CI: macOS build and unit test jobs; x86_64 cross-compiled from
macos-latest via rustup target add x86_64-apple-darwin
Gateway feature flag:
- New opt-in `gateway` Cargo feature activates optional `rustables` dep
- `pub mod gateway` and `Config.gateway` gated behind the feature so
macOS builds never pull in Linux-only nftables bindings
- `fips-gateway` bin has `required-features = ["gateway"]`
- All Linux/OpenWrt/AUR packaging passes `--features gateway`
CI / packaging:
- package-linux, package-macos, package-openwrt now trigger on push to
master/maint/next and on pull requests; release uploads remain tag-gated
- Bloom filter routing fix: fall through to tree routing when no candidate
is strictly closer
- MMP intervals: raise MIN to 1s / MAX to 5s with 5-sample cold-start phase
Add fips-gateway binary: a separate daemon that allows unmodified LAN
hosts to reach FIPS mesh destinations via DNS-allocated virtual IPs
and kernel nftables NAT.
Gateway DNS resolver: forwarding proxy on [::]:53 that intercepts
.fips queries, forwards to daemon resolver (localhost:5354), allocates
virtual IPs from pool, returns AAAA records. Always sends AAAA upstream
regardless of client query type, returns proper NODATA for non-AAAA.
Virtual IP pool: fd01::/112 pool with state machine lifecycle
(Allocated → Active → Draining → Free), TTL-based reclamation,
conntrack integration for session tracking.
NAT manager: nftables DNAT/SNAT rules via rustables netlink API,
per-mapping rule lifecycle, fips0 masquerade for LAN client source
address rewriting.
Network setup: local pool route, proxy NDP for virtual IPs on LAN
interface, IPv6 forwarding validation.
Control socket at /run/fips/gateway.sock with show_gateway and
show_mappings queries. fipstop Gateway tab with pool summary gauge
and mappings table.
Gateway config section in fips.yaml with pool CIDR, LAN interface,
DNS upstream, TTL, and grace period settings.
Design doc at docs/design/fips-gateway.md.
Integration test (testing/static/scripts/gateway-test.sh): three
containers verifying DNS resolution, end-to-end HTTP, NAT state,
TTL expiration, SERVFAIL fallback, and clean shutdown.
Wait for all nodes to reach their expected peer counts instead of
only checking a single node. This prevents false failures on slower
CI runners where remote nodes (especially node E in mesh/chain
topologies) take longer to establish all links.
Wait for all nodes to reach their expected peer counts instead of
only checking a single node. This prevents false failures on slower
CI runners where remote nodes (especially node E in mesh/chain
topologies) take longer to establish all links.
Phase 1 pre-rekey baseline was failing intermittently on CI runners
because 5s wasn't enough for multi-hop discovery (B→D requires
B→C→D). The mesh always converged by Phase 3, confirming this was
purely a timing issue.
The rekey test greps container logs for initiator cutover messages that
were demoted to debug level. Override RUST_LOG in the rekey profile to
enable debug for fips::node::handlers::rekey so the test assertions
can find them.
The FSP XK rekey handshake has a race condition where the initiator
can cut over (K-bit flip) and send data encrypted with the new session
before msg3 reaches the responder. The responder has no pending session
yet, so K-bit detection fails and packets are dropped.
Defer FSP initiator cutover by 2 seconds after handshake completion
(FSP_CUTOVER_DELAY_MS) to give msg3 time to traverse the mesh. FMP
(IK, 2 messages) is unaffected since the responder completes during
msg1 processing.
Also fix MMP metric corruption after rekey cutover: the new session
starts with counter 0 but MMP state carries highest_counter and
GapTracker.expected_next from the old session, producing false
reorder counts, jitter spikes, and invalid OWD trends. Add
reset_for_rekey() methods that clear counter-dependent state while
preserving RTT estimates.
Additional fixes:
- Remove stale peers_by_index entry on abandon_rekey error path
- Replace redundant peers_by_index inserts with debug assertions
verifying the pre-registration invariant
- Tighten rekey integration test to zero tolerance (was 4 failures)
Update 9 documentation files to match current implementation:
- Add missing rekey config section (node.rekey.*) and host mapping
section to fips-configuration.md
- Update Ethernet frame format from [type:1][payload] to
[type:1][length:2 LE][payload] in wire-formats and transport docs
- Fix Ethernet effective MTU from interface-1 to interface-3
- Mark rekey as Implemented in mesh-layer and session-layer status tables
- Change TCP default port examples from 443 to 8443
- Add rekey, persistent identity, host mapping, mesh size estimation to
README features and status sections
- Update chaos scenario count from 16 to 20, add rekey topology to
static test docs
Phase 3 (post-first-rekey connectivity) can see brief disruptions while
links are mid-cutover. Add a 5s settle time after the rekey wait and
allow up to 4 transient pair failures in Phase 3. Phase 5 remains
strict, ensuring full recovery after the second rekey cycle.
Update the external pub node (vps-chi at 217.77.8.91) to use its
real persistent npub instead of a fabricated test key. Remove the
unnecessary nsec from the external node definition.
Implement periodic full rekey at both protocol layers using fresh DH
key exchanges. Uses the existing K-bit flag (FLAG_KEY_EPOCH /
FSP_FLAG_K) to coordinate cutover between peers.
FMP layer (IK pattern):
- ActivePeer gains rekey state: pending/previous sessions, K-bit epoch
tracking, drain window, dampening timer
- Handshake state stored on ActivePeer with msg1 sent on existing link
- Encrypted frame handler detects K-bit flips, promotes pending
sessions, falls back to previous session during drain
- Handshake handlers distinguish rekey from new connections using
addr_to_link lookup with identity-based fallback
- Free all session indices (current, rekey, pending, previous) on
peer removal
FSP layer (XK pattern):
- SessionEntry gains parallel rekey fields with XK-specific state
for the 3-message handshake
- Route availability check before FSP rekey initiation
- Encrypted session handler adds K-bit flip detection and dual-session
decrypt fallback
- SessionSetup/Ack/Msg3 handlers extended for rekey paths
Defense-in-depth:
- Consecutive decryption failure detector (threshold=20) triggers
forced peer removal instead of waiting for link-dead timeout
- Identity-based rekey detection as fallback when addr_to_link
doesn't match (e.g., TCP ephemeral ports)
Configuration: RekeyConfig with enabled flag, after_secs (default 120),
and after_messages (default 65536) thresholds.
Logging: info for successful K-bit cutover completions, warn for
failures, debug for intermediate handshake steps, trace for routine
operations (resends, drain cleanup).
Rekey lifecycle:
1. Timer/counter fires -> initiator starts new handshake
2. Old session continues handling traffic during handshake
3. Handshake completes -> initiator cuts over, flips K-bit
4. Responder sees flipped K-bit -> promotes new session
5. Both keep old session for 10s drain window
6. After drain, old session discarded
Integration test: Docker-based multi-phase test exercising both FMP
and FSP rekey with aggressive timers (35s). Verifies connectivity
across all 20 directed pairs survives two consecutive rekey cycles.
Includes rekey topology, docker-compose profile, and CI matrix entry.
Increase ping test convergence wait from 3s to 5s for CI reliability.
Chaos harness enhancements:
- transport_mix config: weighted random transport assignment for random
topologies (erdos_renyi, random_geometric) with UDP/Ethernet/TCP
- Replace LoRa with Bluetooth L2CAP in cost-based/mixed-tech scenarios
using realistic netem values (15-40ms delay, 5-15ms jitter, 2-8% loss)
- New churn-20-mixed scenario: 20-node Erdos-Renyi with 60% UDP,
20% Ethernet, 20% TCP, full netem/link-flap/churn/bandwidth config
- Expanded chaos README with full scenario catalog in four categories
Static harness:
- Updated README with topology table and scenario count
Implement TCP transport for FIPS enabling firewall traversal and serving
as the foundation for future Tor transport. This is the first
connection-oriented transport in the system.
Key design decisions:
- FMP header-based framing: reuses existing 4-byte FMP common prefix for
packet boundary recovery with zero framing overhead
- Session survives TCP reconnection: Noise/MMP/FSP state bound to npub,
not TCP connection; MMP liveness is sole authority for peer death
- Connect-on-send: fresh connection on first send, transparent reconnect
- close_connection() trait method for cross-connection deduplication cleanup
New transport files:
- src/transport/tcp/mod.rs: TcpTransport, connection pool, accept loop
- src/transport/tcp/stream.rs: FMP-aware stream reader (shared with Tor)
Modified: transport trait (close_connection), TcpConfig, TransportHandle
match arms, create_transports(), initiate_connection() for connection-
oriented links, cross-connection tie-breaker cleanup, design docs.
Tree announce loop and TCP stability fixes:
- Preserve tree announce rate-limit state across reconnection: carry
forward last_tree_announce_sent_ms when a peer reconnects so the
rate-limit window isn't reset to zero
- Drop oversize TCP packets at sender: pre-send MTU check returns
MtuExceeded instead of writing to the stream, preventing receiver-side
connection teardown and reset-reconnect cycles
Chaos harness:
- TCP transport support: tcp_edges/has_tcp/tcp_peers in SimTopology,
transport-aware config_gen with per-edge transport type, TCP port 443,
pure-TCP node support
- Include all non-Ethernet edges in directed_outbound()
- Fix netem/links log messages to say "IP-based" instead of "UDP"
- Add tcp-chain, tcp-only, and tcp-mesh scenario files
Static harness:
- Transport-aware config generation (get_default_transport, transport_port)
- TCP transport injection via Python post-processing
- Add tcp-chain topology and docker-compose profile
Update the default UDP bind port from 4000 to 2121 (decimal) and the
default Ethernet EtherType from 0x88B5 to 0x2121 across all source
code, documentation, configuration templates, test fixtures, and
scripts. Remove references to "IEEE 802 experimental range" since
0x2121 is not in that range.
Johnathan Corgan
Martti MalmiandJohnathan Corgan
Arjen
Origami74
joandJohnathan Corgan