The default 30s post-failure backoff (300s cap, doubling per
consecutive failure) was set to bound traffic from chatty apps
looking up unreachable targets, but in practice it dominates
cold-start mesh convergence: a single timed-out lookup during
initial bloom-filter propagation suppresses any retry for 30s, and
the existing reset triggers (parent change, new peer, first RTT,
reconnection) don't fire on a stable post-handshake topology. The
suppression window winds up dictating the protocol's effective
time-to-converge instead of bounding repeat traffic.
Replaces the single-lookup-with-internal-retry model
(`timeout_secs`/`retry_interval_secs`/`max_attempts`) with a
per-attempt timeout sequence in `node.discovery.attempt_timeouts_secs`,
defaulting to `[1, 2, 4, 8]`. Each attempt sends a fresh LookupRequest
with a new random request_id so successive attempts can take different
forwarding paths as the bloom and tree state evolve. The destination
is declared unreachable only after the sequence is exhausted (15s
total at the default).
Disables post-failure suppression by default (`backoff_base_secs`/
`backoff_max_secs` now `0`/`0`). The `DiscoveryBackoff` machinery
stays in tree (inert at zero base/cap); operators with chatty apps
generating repeat lookups against unreachable destinations can opt
back in.
`PendingLookup` field shape unchanged so the control-socket
`show_routing` JSON (`pending_lookups[].attempt`/`initiated_ms`/
`last_sent_ms`) keeps the same schema for fipstop and external
consumers; `last_sent_ms` now means "current-attempt start" under
the new state machine.
A malformed FilterAnnounce whose fill ratio produces an implausibly
high false-positive rate is mostly useless for routing and, once
merged into our outgoing filter via bitwise OR, propagates the
saturated state to tree peers one hop per announce tick. A saturated
filter also made estimated_count() return f64::INFINITY, which
compute_mesh_size summed into its cached estimate.
handle_filter_announce now rejects inbound FilterAnnounce whose
derived FPR exceeds `node.bloom.max_inbound_fpr` (new config field,
default 0.05 ≈ fill 0.549 at k=5). Rejection is silent on the wire,
logs at WARN, and increments a new `bloom.fill_exceeded` counter. The
peer's prior stored filter and filter_sequence are left unchanged so
a single rejected announce does not wipe the peer's existing
contribution to aggregation.
After a successful outgoing FilterAnnounce send, a rate-limited WARN
fires if our own filter's FPR exceeds the same cap, surfacing
aggregation drift. Limited to once per 60 seconds via a new
Node.last_self_warn field.
BloomFilter::estimated_count() now takes max_fpr and returns
Option<f64>. Returns None for saturated filters (regardless of cap)
or when the filter's FPR exceeds max_fpr. Callers updated: debug
logs render None as "—", the Debug impl uses f64::INFINITY as "no
cap" and prints "saturated" instead of inf, control-socket JSON
emits null, and compute_mesh_size propagates None into the already-
Option<u64> estimated_mesh_size field.
Gate platform-specific code behind cfg attributes and add full Windows
support: TUN device via wintun, TCP control socket on localhost:21210,
Windows Service lifecycle (--install-service/--uninstall-service/--service),
CI build and test matrix, and packaging with ZIP builder and PowerShell
service management scripts.
Key changes:
- Cargo.toml: move tun/libc/rtnetlink behind cfg(unix); add wintun and
windows-service dependencies for Windows
- upper/tun.rs: wintun-based TUN implementation with netsh configuration
for IPv6 address, MTU, and fd00::/8 routing
- control/mod.rs: split into unix_impl/windows_impl; Windows uses TCP on
localhost:21210 with shared connection handler
- bin/fips.rs: refactor main() into run_daemon() accepting a shutdown
signal; add Windows Service support via windows-service crate
- transport/udp/socket.rs: platform-gated modules; Windows uses
tokio::net::UdpSocket (kernel drop count unavailable, returns 0)
- transport/ethernet: gate to cfg(unix); add Windows stub types
- config: platform-conditional default paths (socket, hosts) for Windows
- CI: add windows-latest to build matrix and test-windows job with
cargo-nextest
- packaging/windows: build-zip.ps1, install-service.ps1,
uninstall-service.ps1, and package-windows.yml workflow
- README/docs: Windows build instructions, service management, and
control socket platform differences
Linux and macOS behavior is unchanged.
Add log_level field to NodeConfig (case-insensitive, default: info).
The daemon now loads config before initializing tracing so the
configured level takes effect. RUST_LOG env var still overrides if
set.
Remove hardcoded Environment=RUST_LOG=info from systemd units and
OpenWrt procd init script — these prevented config-driven log levels
from working.
Replace discovery flooding with bloom-filter-guided tree routing:
lookups sent only to tree peers (parent + children) whose bloom
filter contains the target. If no tree peer matches, fall back to
non-tree peers with bloom matches before dropping the request. This
produces single-path forwarding through the spanning tree (90%
traffic reduction) while recovering from dead ends caused by stale
bloom filters, tree restructuring, or transit node failures.
Remove visited bloom filter from LookupRequest wire format (-257
bytes per request). Tree routing is inherently loop-free; request_id
dedup handles edge cases during tree restructuring. Add response-
forwarded flag to prevent response routing loops from convergent
request paths.
Add originator-side exponential backoff (30s base, 300s cap) after
lookup timeouts and bloom misses. Backoff resets on topology changes
(parent switch, new peer, first RTT, reconnection).
Add transit-side per-target rate limiting (2s minimum interval) for
forwarded lookups as defense-in-depth.
Add discovery retry within the timeout window (default: send at T=0,
retry at T=5s, fail at T=10s) to compensate for single-path fragility.
Lookups with zero eligible tree peers fail immediately.
Improve discovery logging: promote key events to info (initiation,
success, timeout with failure count). Add debug logging for dedup,
pending packet retry, backoff suppression, forward rate limiting,
and backoff reset.
New config: discovery.backoff_base_secs, backoff_max_secs,
forward_min_interval_secs, retry_interval_secs, max_attempts.
New stats: req_backoff_suppressed, req_forward_rate_limited,
req_bloom_miss, req_no_tree_peer, req_fallback_forwarded.
Removed: req_already_visited, visited bloom filter.
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.
Implement hop-by-hop ECN congestion signaling through the FMP layer,
transport-level congestion detection via kernel drop counters, and
chaos harness integration for end-to-end validation.
FMP/session ECN plumbing:
- Thread ce_flag parsed at link layer through dispatch_link_message,
handle_session_datagram, handle_session_payload, and
handle_encrypted_session_msg to session delivery
- Replace hardcoded false in session-layer record_recv() with actual
ce_flag, activating ecn_ce_count tracking in session MMP
ECN congestion detection and CE relay:
- Add EcnConfig (node.ecn.*) with configurable loss_threshold (5%)
and etx_threshold (3.0) for transit congestion detection
- Add send_encrypted_link_message_with_ce() that ORs FLAG_CE into FMP
header flags; original method delegates with ce_flag=false
- Compute outgoing_ce = incoming_ce || local congestion on next-hop
link, enabling hop-by-hop CE relay through transit nodes
IPv6 ECN-CE marking:
- Mark ECN-CE (0b11) in IPv6 Traffic Class on received DataPackets
before TUN delivery when FMP CE flag is set
- Only marks ECN-capable packets (ECT(0)/ECT(1)); Not-ECT packets
unchanged per RFC 3168
Transport congestion abstraction and UDP kernel drop detection:
- Add TransportCongestion struct to transport layer for transport-
agnostic local congestion indicators
- Replace tokio::UdpSocket with AsyncFd<socket2::Socket> using
libc::recvmsg() with ancillary data parsing
- Enable SO_RXQ_OVFL for kernel receive buffer drop counter on every
packet, wiring up previously-stubbed UdpStats.kernel_drops
- Add TransportDropState for per-transport delta tracking with 1s
tick sampling via sample_transport_congestion()
- Extend detect_congestion() with transport kernel drop check
alongside MMP loss metrics
Congestion monitoring and control:
- Add CongestionStats (ce_forwarded, ce_received, congestion_detected,
kernel_drop_events) to NodeStats with snapshot serialization
- Wire counters into forwarding path, session handler, and transport
drop sampling with rate-limited warn logging (5s interval)
- Expose congestion data in show_routing control query and
ecn_ce_count in show_mmp peer entries
- Add congestion counters to fipstop routing tab in two-column layout
Chaos harness integration:
- Add query_routing(), query_transports(), snapshot_all_congestion()
to chaos control module
- Add congestion/kernel-drop log analysis in logs module
- Add congestion-stress scenario: 10-node tree, 1 Mbps bandwidth,
5-10% netem loss, heavy iperf3 traffic
- Add IngressConfig for tc ingress policing with per-peer policer
filters simulating upstream bandwidth bottlenecks
- Add iperf3 JSON result capture to traffic manager for throughput
measurement across scenarios
- Add ECN A/B test scenarios (ecn-ab-on/off.yaml) with ingress
policing and comparison script
- Enable TCP ECN negotiation (tcp_ecn=1 sysctl) in container
entrypoint for end-to-end CE propagation
Tests:
- 10 ECN unit/integration tests: mark_ipv6_ecn_ce variants, CE relay
chain (3-node propagation), EcnConfig serde roundtrip
- 3 transport drop congestion detection unit tests
Documentation:
- Update fips-mesh-layer.md: replace outdated CE Echo stub with full
ECN Congestion Signaling section covering detection logic, CE relay,
IPv6 marking, session tracking, and monitoring counters
- Update fips-configuration.md: add node.ecn.* parameter table and
ecn block in complete reference YAML
- Update fips-transport-layer.md: add Congestion Reporting section
with TransportCongestion struct, congestion() trait method, and
per-transport status; document AsyncFd/recvmsg/SO_RXQ_OVFL in UDP
- Update chaos README: add congestion/ECN scenario docs, ingress
traffic control, and iperf3 JSON capture sections
- Update README.md: add ECN to features list and "What works today";
update transport and tooling entries
Track parent switch frequency in a sliding window. When switches exceed
a configurable threshold (default 4 in 60s), impose an extended hold-down
period (default 120s) that prevents non-mandatory parent changes.
Mandatory switches (parent loss, root change, shouldn't-be-root) bypass
dampening. The flap counter resets when the window expires naturally.
Implements TASK-2026-0030 / IDEA-0013.
Cost-based parent selection:
- Replace depth-only parent selection with effective_depth = depth + link_cost
- link_cost computed from locally measured MMP metrics: etx * (1.0 + srtt_ms / 100.0)
- Prevents bottleneck subtrees in heterogeneous networks where a LoRa link
at depth 1 would otherwise always beat fiber at depth 2
- Configurable hysteresis (default 0.2) prevents marginal parent switches
- Configurable hold-down timer (default 30s) suppresses re-evaluation
after parent switch
- Mandatory switches (parent lost, root change) bypass both safeguards
- Link costs passed as HashMap parameter to keep TreeState pure
Periodic re-evaluation:
- evaluate_parent() was only called on TreeAnnounce receipt or parent loss;
after tree stabilization, link degradation went undetected
- Added timer-based re-evaluation (reeval_interval_secs, default 60s) that
calls evaluate_parent() from the tick handler with current MMP link costs
- Respects existing hold-down and hysteresis safeguards
- Short-circuits when disabled or <2 peers
Design documentation:
- Update 7 design docs to reflect cost-based parent selection
- Replace depth-only algorithm descriptions with effective_depth model
- Replace rejected cumulative path cost spec with local-only design rationale
- Rewrite Example 2 (heterogeneous links) for local-only cost model
- Update config docs: parent_switch_threshold replaced by parent_hysteresis,
hold_down_secs, reeval_interval_secs
Chaos simulation enhancements:
- fips_overrides with deep merge for per-scenario FIPS config customization
- Explicit topology algorithm for deterministic test graphs
- Control socket querying via fipsctl for tree/MMP snapshot collection
- Edge existence validation in netem manager
- Per-link netem policy overrides
- 9 new chaos scenarios covering cost avoidance, depth-vs-cost tradeoffs,
stability, mixed topologies, periodic re-evaluation, and bottleneck parent
12 new unit tests, 667 total passing, clippy clean.
Add a Unix domain socket interface for querying node state at runtime.
A spawned tokio task accepts connections and communicates with the main
event loop via mpsc/oneshot channels, keeping all Node access
single-threaded.
Includes:
- src/control/ module with socket lifecycle, JSON protocol, and 11
query handlers (status, peers, links, tree, sessions, bloom, mmp,
cache, connections, transports, routing)
- Separate fipsctl binary for CLI queries (fipsctl show <command>)
- ControlConfig in node configuration (enabled, socket_path)
- Integration into the main select! event loop
Sends a 1-byte encrypted heartbeat (0x51) to each peer every 10s.
If no frame is received from a peer within 30s, the peer is removed
via remove_active_peer(), triggering tree reconvergence, coord cache
flush, and bloom filter recomputation.
This fixes the critical bug where UDP peers that silently died
(e.g., container stopped) were never detected or removed, leaving
the spanning tree permanently stale.
Both intervals are configurable via node.heartbeat_interval_secs
and node.link_dead_timeout_secs.
Add message-level retry for Noise IK handshake within the 30s timeout
window. Previously, a lost msg1 or msg2 required the full timeout to
expire before cleanup and retry. Under 10% bidirectional loss (~19%
per attempt), this made connection establishment unreliable.
Initiator resends stored msg1 bytes with exponential backoff (1s, 2s,
4s, 8s, 16s — 5 resends). Responder stores msg2 and resends on
duplicate msg1 receipt. Duplicate msg2 at initiator drops silently
via existing pending_outbound cleanup.
Config: handshake_resend_interval_ms (1000), handshake_resend_backoff
(2.0), handshake_max_resends (5) on node.rate_limit.
P(all 6 attempts fail) under 19% loss = 0.19^6 ≈ 0.005%.
Implement hybrid coordinate cache warming strategy: piggyback coords
via CP flag when they fit within transport MTU, send standalone
CoordsWarmup (0x14) message when they don't. On CoordsRequired or
PathBroken receipt, send CoordsWarmup immediately with source-side
rate limiting (default 2s per destination, configurable).
- Add CoordsWarmup = 0x14 session message type (empty body, CP flag)
- Add send_coords_warmup() following send_session_msg() pattern
- Restructure send_session_data() to send standalone warmup before
data packet when piggybacked coords exceed MTU
- Add immediate CoordsWarmup response in handle_coords_required()
and handle_path_broken() with per-destination rate limiting
- Add coords_response_interval_ms config (node.session)
- Add RoutingErrorRateLimiter::with_interval() constructor
- Zero transit-path changes: existing try_warm_coord_cache() handles
CoordsWarmup identically to CP-flagged data packets
- Update design docs (session layer, wire formats, mesh operation,
configuration)
MMP reports (SenderReport, ReceiverReport, PathMtuNotification) were
resetting the idle timer on both RX and TX paths, preventing sessions
from ever timing out when MMP traffic kept flowing. Changed touch() to
only be called for DataPacket send/receive and session establishment,
so sessions with no application data tear down after the idle timeout
even with active MMP measurement traffic.
- PathBroken handler: convert to async, trigger re-discovery via
maybe_initiate_lookup(), reset COORDS_PRESENT warmup counter
(was a stub that only invalidated coord_cache)
- CoordsRequired recovery timing: reset warmup counter in
handle_lookup_response() when discovery completes for an
established session, so COORDS_PRESENT packets fire after
fresh coords are available (not just on CoordsRequired receipt)
- Routing error rate limiting: add RoutingErrorRateLimiter
(100ms per-destination, matching ICMP PTB pattern) to gate
send_routing_error() at transit nodes
- Remove root refresh dead code: the 1800s periodic root
re-announcement in check_tree_state() only propagated to
depth 1 (sequence-only changes don't cascade). Root loss
detection relies on link failure propagation which works
correctly.
Include source and destination coordinates on the first N DataPackets
of each session (default 5, configurable via
node.session.coords_warmup_packets). This warms transit node
coord_caches so multi-hop forwarding and error signal routing work
after SessionSetup cache entries expire.
On CoordsRequired receipt, reset the counter to re-enable coordinate
inclusion for the next N packets, handling mid-session cache expiry
and path changes.
Changes:
- SessionConfig: add coords_warmup_packets field (default 5)
- SessionEntry: add coords_warmup_remaining counter, initialized on
Established transition (both initiator and responder paths)
- send_session_data(): attach coords via DataPacket::with_coords()
while counter > 0, decrement per send
- handle_coords_required(): reset counter for affected session
- 4 new unit tests for counter lifecycle and config default
Identity cache: remove TTL-based expiry (60s TTL broke active sessions
after expiry since handle_tun_outbound checks identity_cache before
session table). Replace with LRU-only eviction bounded by configurable
identity_size (default 10K). Lookup now touches timestamp for LRU
freshness.
Cache merge: unify coord_cache and route_cache into single coordinate
cache. Both stored NodeAddr→TreeCoordinate; the layer distinction was
conceptual, not functional. Discovery-sourced entries now get the same
TTL+refresh treatment as session-sourced entries. Simplifies
find_next_hop() to single cache lookup.
Parent-change flush: clear coord_cache after recompute_coords() in both
parent-switch paths of handle_tree_announce(). Stale coordinates after
tree reconvergence cause dead-end routing that's more expensive than
re-discovery.
Tested: 493 unit tests passed, clippy clean, Docker mesh 20/20,
Docker chain 6/6.
Sessions in the Established state that have no activity for 90 seconds
are now automatically removed. This ensures idle sessions are torn down
before transit node coord_cache entries expire (300s TTL), so that when
traffic resumes a fresh SessionSetup re-warms transit node caches with
current coordinates.
The identity cache now stores registration timestamps and expires
entries after 60 seconds via lazy expiry on lookup. This prevents
unbounded growth while allowing natural repopulation through DNS
resolution on next use.
Timer ordering: identity (60s) < session (90s) < coord_cache (300s).
Both timeouts are configurable: node.session.idle_timeout_secs and
node.cache.identity_ttl_secs. Setting idle_timeout_secs to 0 disables
session idle purging.
Changes:
- Add idle_timeout_secs (default 90) to SessionConfig
- Add identity_ttl_secs (default 60) to CacheConfig
- Add timestamp to identity_cache entries, lazy expiry on lookup
- Add purge_idle_sessions() called from tick loop
- Remove #[cfg(test)] from SessionEntry::last_activity()
- 7 new tests covering timeout behavior and edge cases
Johnathan Corgan
OceanSlimandGitHub