Files
oc-discovery/daemons/node/indexer/behavior.go
2026-03-11 16:28:15 +01:00

255 lines
6.6 KiB
Go

package indexer
import (
"errors"
"sync"
"time"
"oc-discovery/conf"
pp "github.com/libp2p/go-libp2p/core/peer"
)
// ── defaults ──────────────────────────────────────────────────────────────────
const (
defaultMaxConnPerWindow = 20
defaultConnWindowSecs = 30
defaultMaxHBPerMinute = 5
defaultMaxPublishPerMin = 10
defaultMaxGetPerMin = 50
strikeThreshold = 3
banDuration = 10 * time.Minute
behaviorWindowDur = 60 * time.Second
)
func cfgOr(v, def int) int {
if v > 0 {
return v
}
return def
}
// ── ConnectionRateGuard ───────────────────────────────────────────────────────
// ConnectionRateGuard limits the number of NEW incoming connections accepted
// within a sliding time window. It protects public indexers against coordinated
// registration floods (Sybil bursts).
type ConnectionRateGuard struct {
mu sync.Mutex
window []time.Time
maxInWindow int
windowDur time.Duration
}
func newConnectionRateGuard() *ConnectionRateGuard {
cfg := conf.GetConfig()
return &ConnectionRateGuard{
maxInWindow: cfgOr(cfg.MaxConnPerWindow, defaultMaxConnPerWindow),
windowDur: time.Duration(cfgOr(cfg.ConnWindowSecs, defaultConnWindowSecs)) * time.Second,
}
}
// Allow returns true if a new connection may be accepted.
// The internal window is pruned on each call so memory stays bounded.
func (g *ConnectionRateGuard) Allow() bool {
g.mu.Lock()
defer g.mu.Unlock()
now := time.Now()
cutoff := now.Add(-g.windowDur)
i := 0
for i < len(g.window) && g.window[i].Before(cutoff) {
i++
}
g.window = g.window[i:]
if len(g.window) >= g.maxInWindow {
return false
}
g.window = append(g.window, now)
return true
}
// ── per-node state ────────────────────────────────────────────────────────────
type nodeBehavior struct {
mu sync.Mutex
knownDID string
hbTimes []time.Time
pubTimes []time.Time
getTimes []time.Time
strikes int
bannedUntil time.Time
}
func (nb *nodeBehavior) isBanned() bool {
return time.Now().Before(nb.bannedUntil)
}
func (nb *nodeBehavior) strike(n int) {
nb.strikes += n
if nb.strikes >= strikeThreshold {
nb.bannedUntil = time.Now().Add(banDuration)
}
}
func pruneWindow(ts []time.Time, dur time.Duration) []time.Time {
cutoff := time.Now().Add(-dur)
i := 0
for i < len(ts) && ts[i].Before(cutoff) {
i++
}
return ts[i:]
}
// recordInWindow appends now to the window slice and returns false (+ adds a
// strike) when the count exceeds max.
func (nb *nodeBehavior) recordInWindow(ts *[]time.Time, max int) bool {
*ts = pruneWindow(*ts, behaviorWindowDur)
if len(*ts) >= max {
nb.strike(1)
return false
}
*ts = append(*ts, time.Now())
return true
}
// ── NodeBehaviorTracker ───────────────────────────────────────────────────────
// NodeBehaviorTracker is the indexer-side per-node compliance monitor.
// It is entirely local: no state is shared with other indexers.
type NodeBehaviorTracker struct {
mu sync.RWMutex
nodes map[pp.ID]*nodeBehavior
maxHB int
maxPub int
maxGet int
}
func newNodeBehaviorTracker() *NodeBehaviorTracker {
cfg := conf.GetConfig()
return &NodeBehaviorTracker{
nodes: make(map[pp.ID]*nodeBehavior),
maxHB: cfgOr(cfg.MaxHBPerMinute, defaultMaxHBPerMinute),
maxPub: cfgOr(cfg.MaxPublishPerMinute, defaultMaxPublishPerMin),
maxGet: cfgOr(cfg.MaxGetPerMinute, defaultMaxGetPerMin),
}
}
func (t *NodeBehaviorTracker) get(pid pp.ID) *nodeBehavior {
t.mu.RLock()
nb := t.nodes[pid]
t.mu.RUnlock()
if nb != nil {
return nb
}
t.mu.Lock()
defer t.mu.Unlock()
if nb = t.nodes[pid]; nb == nil {
nb = &nodeBehavior{}
t.nodes[pid] = nb
}
return nb
}
// IsBanned returns true when the peer is in an active ban period.
func (t *NodeBehaviorTracker) IsBanned(pid pp.ID) bool {
nb := t.get(pid)
nb.mu.Lock()
defer nb.mu.Unlock()
return nb.isBanned()
}
// RecordHeartbeat checks heartbeat cadence. Returns an error if the peer is
// flooding (too many heartbeats in the sliding window).
func (t *NodeBehaviorTracker) RecordHeartbeat(pid pp.ID) error {
nb := t.get(pid)
nb.mu.Lock()
defer nb.mu.Unlock()
if nb.isBanned() {
return errors.New("peer is banned")
}
if !nb.recordInWindow(&nb.hbTimes, t.maxHB) {
return errors.New("heartbeat flood detected")
}
return nil
}
// CheckIdentity verifies that the DID associated with a PeerID never changes.
// A DID change is a strong signal of identity spoofing.
func (t *NodeBehaviorTracker) CheckIdentity(pid pp.ID, did string) error {
if did == "" {
return nil
}
nb := t.get(pid)
nb.mu.Lock()
defer nb.mu.Unlock()
if nb.knownDID == "" {
nb.knownDID = did
return nil
}
if nb.knownDID != did {
nb.strike(2) // identity change is severe
return errors.New("DID mismatch for peer " + pid.String())
}
return nil
}
// RecordBadSignature registers a cryptographic verification failure.
// A single bad signature is worth 2 strikes (near-immediate ban).
func (t *NodeBehaviorTracker) RecordBadSignature(pid pp.ID) {
nb := t.get(pid)
nb.mu.Lock()
defer nb.mu.Unlock()
nb.strike(2)
}
// RecordPublish checks publish volume. Returns an error if the peer is
// sending too many publish requests.
func (t *NodeBehaviorTracker) RecordPublish(pid pp.ID) error {
nb := t.get(pid)
nb.mu.Lock()
defer nb.mu.Unlock()
if nb.isBanned() {
return errors.New("peer is banned")
}
if !nb.recordInWindow(&nb.pubTimes, t.maxPub) {
return errors.New("publish volume exceeded")
}
return nil
}
// RecordGet checks get volume. Returns an error if the peer is enumerating
// the DHT at an abnormal rate.
func (t *NodeBehaviorTracker) RecordGet(pid pp.ID) error {
nb := t.get(pid)
nb.mu.Lock()
defer nb.mu.Unlock()
if nb.isBanned() {
return errors.New("peer is banned")
}
if !nb.recordInWindow(&nb.getTimes, t.maxGet) {
return errors.New("get volume exceeded")
}
return nil
}
// Cleanup removes the behavior entry for a peer if it is not currently banned.
// Called when the peer is evicted from StreamRecords by the GC.
func (t *NodeBehaviorTracker) Cleanup(pid pp.ID) {
t.mu.RLock()
nb := t.nodes[pid]
t.mu.RUnlock()
if nb == nil {
return
}
nb.mu.Lock()
banned := nb.isBanned()
nb.mu.Unlock()
if !banned {
t.mu.Lock()
delete(t.nodes, pid)
t.mu.Unlock()
}
}