Merge pull request #8045 from gammazero/feat/cid-to-piece-idx

Add indexer pubsub message authentication and rate limiting

chain/sub/incoming.go

@@ -1,19 +1,24 @@
 package sub
 
 import (
+	"bytes"
 	"context"
-	"fmt"
+	"encoding/binary"
+	"sync"
 	"time"
 
 	address "github.com/filecoin-project/go-address"
+	"github.com/filecoin-project/go-legs/dtsync"
 	"github.com/filecoin-project/lotus/build"
 	"github.com/filecoin-project/lotus/chain"
 	"github.com/filecoin-project/lotus/chain/consensus"
 	"github.com/filecoin-project/lotus/chain/messagepool"
 	"github.com/filecoin-project/lotus/chain/store"
+	"github.com/filecoin-project/lotus/chain/sub/ratelimit"
 	"github.com/filecoin-project/lotus/chain/types"
 	"github.com/filecoin-project/lotus/metrics"
 	"github.com/filecoin-project/lotus/node/impl/client"
+	"github.com/filecoin-project/lotus/node/impl/full"
 	lru "github.com/hashicorp/golang-lru"
 	blocks "github.com/ipfs/go-block-format"
 	bserv "github.com/ipfs/go-blockservice"
@@ -168,12 +173,12 @@ func fetchCids(
 	cidIndex := make(map[cid.Cid]int)
 	for i, c := range cids {
 		if c.Prefix() != msgCidPrefix {
-			return fmt.Errorf("invalid msg CID: %s", c)
+			return xerrors.Errorf("invalid msg CID: %s", c)
 		}
 		cidIndex[c] = i
 	}
 	if len(cids) != len(cidIndex) {
-		return fmt.Errorf("duplicate CIDs in fetchCids input")
+		return xerrors.Errorf("duplicate CIDs in fetchCids input")
 	}
 
 	for block := range bserv.GetBlocks(ctx, cids) {
@@ -196,7 +201,7 @@ func fetchCids(
 	if len(cidIndex) > 0 {
 		err := ctx.Err()
 		if err == nil {
-			err = fmt.Errorf("failed to fetch %d messages for unknown reasons", len(cidIndex))
+			err = xerrors.Errorf("failed to fetch %d messages for unknown reasons", len(cidIndex))
 		}
 		return err
 	}
@@ -445,23 +450,166 @@ func recordFailure(ctx context.Context, metric *stats.Int64Measure, failureType
 	stats.Record(ctx, metric.M(1))
 }
 
-type IndexerMessageValidator struct {
-	self peer.ID
+type peerMsgInfo struct {
+	peerID    peer.ID
+	lastCid   cid.Cid
+	lastSeqno uint64
+	rateLimit *ratelimit.Window
+	mutex     sync.Mutex
 }
 
-func NewIndexerMessageValidator(self peer.ID) *IndexerMessageValidator {
-	return &IndexerMessageValidator{self: self}
+type IndexerMessageValidator struct {
+	self peer.ID
+
+	peerCache *lru.TwoQueueCache
+	chainApi  full.ChainModuleAPI
+	stateApi  full.StateModuleAPI
+}
+
+func NewIndexerMessageValidator(self peer.ID, chainApi full.ChainModuleAPI, stateApi full.StateModuleAPI) *IndexerMessageValidator {
+	peerCache, _ := lru.New2Q(8192)
+
+	return &IndexerMessageValidator{
+		self:      self,
+		peerCache: peerCache,
+		chainApi:  chainApi,
+		stateApi:  stateApi,
+	}
 }
 
 func (v *IndexerMessageValidator) Validate(ctx context.Context, pid peer.ID, msg *pubsub.Message) pubsub.ValidationResult {
 	// This chain-node should not be publishing its own messages.  These are
-	// relayed from miner-nodes or index publishers.  If a node appears to be
-	// local, reject it.
+	// relayed from market-nodes.  If a node appears to be local, reject it.
 	if pid == v.self {
-		log.Warnf("refusing to relay indexer message from self")
+		log.Debug("ignoring indexer message from self")
 		stats.Record(ctx, metrics.IndexerMessageValidationFailure.M(1))
+		return pubsub.ValidationIgnore
+	}
+	originPeer := msg.GetFrom()
+	if originPeer == v.self {
+		log.Debug("ignoring indexer message originating from self")
+		stats.Record(ctx, metrics.IndexerMessageValidationFailure.M(1))
+		return pubsub.ValidationIgnore
+	}
+
+	idxrMsg := dtsync.Message{}
+	err := idxrMsg.UnmarshalCBOR(bytes.NewBuffer(msg.Data))
+	if err != nil {
+		log.Errorw("Could not decode indexer pubsub message", "err", err)
 		return pubsub.ValidationReject
 	}
+	if len(idxrMsg.ExtraData) == 0 {
+		log.Debugw("ignoring messsage missing miner id", "peer", originPeer)
+		return pubsub.ValidationIgnore
+	}
+
+	minerID := string(idxrMsg.ExtraData)
+	msgCid := idxrMsg.Cid
+
+	var msgInfo *peerMsgInfo
+	val, ok := v.peerCache.Get(minerID)
+	if !ok {
+		msgInfo = &peerMsgInfo{}
+	} else {
+		msgInfo = val.(*peerMsgInfo)
+	}
+
+	// Lock this peer's message info.
+	msgInfo.mutex.Lock()
+	defer msgInfo.mutex.Unlock()
+
+	if ok {
+		// Reject replayed messages.
+		seqno := binary.BigEndian.Uint64(msg.Message.GetSeqno())
+		if seqno <= msgInfo.lastSeqno {
+			log.Debugf("ignoring replayed indexer message")
+			return pubsub.ValidationIgnore
+		}
+		msgInfo.lastSeqno = seqno
+	}
+
+	if !ok || originPeer != msgInfo.peerID {
+		// Check that the miner ID maps to the peer that sent the message.
+		err = v.authenticateMessage(ctx, minerID, originPeer)
+		if err != nil {
+			log.Warnw("cannot authenticate messsage", "err", err, "peer", originPeer, "minerID", minerID)
+			stats.Record(ctx, metrics.IndexerMessageValidationFailure.M(1))
+			return pubsub.ValidationReject
+		}
+		msgInfo.peerID = originPeer
+		if !ok {
+			// Add msgInfo to cache only after being authenticated.  If two
+			// messages from the same peer are handled concurrently, there is a
+			// small chance that one msgInfo could replace the other here when
+			// the info is first cached.  This is OK, so no need to prevent it.
+			v.peerCache.Add(minerID, msgInfo)
+		}
+	}
+
+	// See if message needs to be ignored due to rate limiting.
+	if v.rateLimitPeer(msgInfo, msgCid) {
+		return pubsub.ValidationIgnore
+	}
+
 	stats.Record(ctx, metrics.IndexerMessageValidationSuccess.M(1))
 	return pubsub.ValidationAccept
 }
+
+func (v *IndexerMessageValidator) rateLimitPeer(msgInfo *peerMsgInfo, msgCid cid.Cid) bool {
+	const (
+		msgLimit        = 5
+		msgTimeLimit    = 10 * time.Second
+		repeatTimeLimit = 2 * time.Hour
+	)
+
+	timeWindow := msgInfo.rateLimit
+
+	// Check overall message rate.
+	if timeWindow == nil {
+		timeWindow = ratelimit.NewWindow(msgLimit, msgTimeLimit)
+		msgInfo.rateLimit = timeWindow
+	} else if msgInfo.lastCid == msgCid {
+		// Check if this is a repeat of the previous message data.
+		if time.Since(timeWindow.Newest()) < repeatTimeLimit {
+			log.Warnw("ignoring repeated indexer message", "sender", msgInfo.peerID)
+			return true
+		}
+	}
+
+	err := timeWindow.Add()
+	if err != nil {
+		log.Warnw("ignoring indexer message", "sender", msgInfo.peerID, "err", err)
+		return true
+	}
+
+	msgInfo.lastCid = msgCid
+
+	return false
+}
+
+func (v *IndexerMessageValidator) authenticateMessage(ctx context.Context, minerID string, peerID peer.ID) error {
+	// Get miner info from lotus
+	minerAddress, err := address.NewFromString(minerID)
+	if err != nil {
+		return xerrors.Errorf("invalid miner id: %w", err)
+	}
+
+	ts, err := v.chainApi.ChainHead(ctx)
+	if err != nil {
+		return err
+	}
+
+	minerInfo, err := v.stateApi.StateMinerInfo(ctx, minerAddress, ts.Key())
+	if err != nil {
+		return err
+	}
+
+	if minerInfo.PeerId == nil {
+		return xerrors.New("no peer id for miner")
+	}
+	if *minerInfo.PeerId != peerID {
+		return xerrors.New("miner id does not map to peer that sent message")
+	}
+
+	return nil
+}

chain/sub/ratelimit/queue.go

@@ -0,0 +1,89 @@
+package ratelimit
+
+import "errors"
+
+var ErrRateLimitExceeded = errors.New("rate limit exceeded")
+
+type queue struct {
+	buf   []int64
+	count int
+	head  int
+	tail  int
+}
+
+// cap returns the queue capacity
+func (q *queue) cap() int {
+	return len(q.buf)
+}
+
+// len returns the number of items in the queue
+func (q *queue) len() int {
+	return q.count
+}
+
+// push adds an element to the end of the queue.
+func (q *queue) push(elem int64) error {
+	if q.count == len(q.buf) {
+		return ErrRateLimitExceeded
+	}
+
+	q.buf[q.tail] = elem
+	// Calculate new tail position.
+	q.tail = q.next(q.tail)
+	q.count++
+	return nil
+}
+
+// pop removes and returns the element from the front of the queue.
+func (q *queue) pop() int64 {
+	if q.count <= 0 {
+		panic("pop from empty queue")
+	}
+	ret := q.buf[q.head]
+
+	// Calculate new head position.
+	q.head = q.next(q.head)
+	q.count--
+
+	return ret
+}
+
+// front returns the element at the front of the queue.  This is the element
+// that would be returned by pop().  This call panics if the queue is empty.
+func (q *queue) front() int64 {
+	if q.count <= 0 {
+		panic("front() called when empty")
+	}
+	return q.buf[q.head]
+}
+
+// back returns the element at the back of the queue.  This call panics if the
+// queue is empty.
+func (q *queue) back() int64 {
+	if q.count <= 0 {
+		panic("back() called when empty")
+	}
+	return q.buf[q.prev(q.tail)]
+}
+
+// prev returns the previous buffer position wrapping around buffer.
+func (q *queue) prev(i int) int {
+	if i == 0 {
+		return len(q.buf) - 1
+	}
+	return (i - 1) % len(q.buf)
+}
+
+// next returns the next buffer position wrapping around buffer.
+func (q *queue) next(i int) int {
+	return (i + 1) % len(q.buf)
+}
+
+// truncate pops values that are less than or equal the specified threshold.
+func (q *queue) truncate(threshold int64) {
+	for q.count != 0 && q.buf[q.head] <= threshold {
+		// pop() without returning a value
+		q.head = q.next(q.head)
+		q.count--
+	}
+}

chain/sub/ratelimit/window.go

@@ -0,0 +1,70 @@
+package ratelimit
+
+import "time"
+
+// Window is a time windows for counting events within a span of time.  The
+// windows slides forward in time so that it spans from the most recent event
+// to size time in the past.
+type Window struct {
+	q    *queue
+	size int64
+}
+
+// NewWindow creates a new Window that limits the number of events to maximum
+// count of events within a duration of time.  The capacity sets the maximum
+// number of events, and size sets the span of time over which the events are
+// counted.
+func NewWindow(capacity int, size time.Duration) *Window {
+	return &Window{
+		q: &queue{
+			buf: make([]int64, capacity),
+		},
+		size: int64(size),
+	}
+}
+
+// Add attempts to append a new timestamp into the current window.  Previously
+// added values that are not not within `size` difference from the value being
+// added are first removed.  Add fails if adding the value would cause the
+// window to exceed capacity.
+func (w *Window) Add() error {
+	now := time.Now().UnixNano()
+	if w.Len() != 0 {
+		w.q.truncate(now - w.size)
+	}
+	return w.q.push(now)
+}
+
+// Cap returns the maximum number of items the window can hold.
+func (w *Window) Cap() int {
+	return w.q.cap()
+}
+
+// Len returns the number of elements currently in the window.
+func (w *Window) Len() int {
+	return w.q.len()
+}
+
+// Span returns the distance from the first to the last item in the window.
+func (w *Window) Span() time.Duration {
+	if w.q.len() < 2 {
+		return 0
+	}
+	return time.Duration(w.q.back() - w.q.front())
+}
+
+// Oldest returns the oldest timestamp in the window.
+func (w *Window) Oldest() time.Time {
+	if w.q.len() == 0 {
+		return time.Time{}
+	}
+	return time.Unix(0, w.q.front())
+}
+
+// Newest returns the newest timestamp in the window.
+func (w *Window) Newest() time.Time {
+	if w.q.len() == 0 {
+		return time.Time{}
+	}
+	return time.Unix(0, w.q.back())
+}

chain/sub/ratelimit/window_test.go

@@ -0,0 +1,61 @@
+package ratelimit
+
+import (
+	"testing"
+	"time"
+)
+
+func TestWindow(t *testing.T) {
+	const (
+		maxEvents = 3
+		timeLimit = 100 * time.Millisecond
+	)
+	w := NewWindow(maxEvents, timeLimit)
+	if w.Len() != 0 {
+		t.Fatal("q.Len() =", w.Len(), "expect 0")
+	}
+	if w.Cap() != maxEvents {
+		t.Fatal("q.Cap() =", w.Cap(), "expect 3")
+	}
+	if !w.Newest().IsZero() {
+		t.Fatal("expected newest to be zero time with empty window")
+	}
+	if !w.Oldest().IsZero() {
+		t.Fatal("expected oldest to be zero time with empty window")
+	}
+	if w.Span() != 0 {
+		t.Fatal("expected span to be zero time with empty window")
+	}
+
+	var err error
+	for i := 0; i < maxEvents; i++ {
+		err = w.Add()
+		if err != nil {
+			t.Fatalf("cannot add event %d", i)
+		}
+	}
+	if w.Len() != maxEvents {
+		t.Fatalf("q.Len() is %d, expected %d", w.Len(), maxEvents)
+	}
+	if err = w.Add(); err != ErrRateLimitExceeded {
+		t.Fatalf("add event %d within time limit should have failed with err: %s", maxEvents+1, ErrRateLimitExceeded)
+	}
+
+	time.Sleep(timeLimit)
+	if err = w.Add(); err != nil {
+		t.Fatalf("cannot add event after time limit: %s", err)
+	}
+
+	prev := w.Newest()
+	time.Sleep(timeLimit)
+	err = w.Add()
+	if err != nil {
+		t.Fatalf("cannot add event")
+	}
+	if w.Newest().Before(prev) {
+		t.Fatal("newest is before previous value")
+	}
+	if w.Oldest().Before(prev) {
+		t.Fatal("oldest is before previous value")
+	}
+}

go.mod

@@ -39,6 +39,7 @@ require (
 	github.com/filecoin-project/go-fil-markets v1.19.1-0.20220210121001-44fb837759c1
 	github.com/filecoin-project/go-indexer-core v0.2.8
 	github.com/filecoin-project/go-jsonrpc v0.1.5
+	github.com/filecoin-project/go-legs v0.3.0
 	github.com/filecoin-project/go-padreader v0.0.1
 	github.com/filecoin-project/go-paramfetch v0.0.3-0.20220111000201-e42866db1a53
 	github.com/filecoin-project/go-state-types v0.1.3

node/modules/lp2p/pubsub.go

@@ -114,6 +114,22 @@ func GossipSub(in GossipIn) (service *pubsub.PubSub, err error) {
 		InvalidMessageDeliveriesDecay:  pubsub.ScoreParameterDecay(time.Hour),
 	}
 
+	ingestTopicParams := &pubsub.TopicScoreParams{
+		// expected ~0.5 confirmed deals / min. sampled
+		TopicWeight: 0.1,
+
+		TimeInMeshWeight:  0.00027, // ~1/3600
+		TimeInMeshQuantum: time.Second,
+		TimeInMeshCap:     1,
+
+		FirstMessageDeliveriesWeight: 0.5,
+		FirstMessageDeliveriesDecay:  pubsub.ScoreParameterDecay(time.Hour),
+		FirstMessageDeliveriesCap:    100, // allowing for burstiness
+
+		InvalidMessageDeliveriesWeight: -1000,
+		InvalidMessageDeliveriesDecay:  pubsub.ScoreParameterDecay(time.Hour),
+	}
+
 	topicParams := map[string]*pubsub.TopicScoreParams{
 		build.BlocksTopic(in.Nn): {
 			// expected 10 blocks/min
@@ -208,6 +224,9 @@ func GossipSub(in GossipIn) (service *pubsub.PubSub, err error) {
 		drandTopics = append(drandTopics, topic)
 	}
 
+	// Index ingestion whitelist
+	topicParams[build.IndexerIngestTopic(in.Nn)] = ingestTopicParams
+
 	// IP colocation whitelist
 	var ipcoloWhitelist []*net.IPNet
 	for _, cidr := range in.Cfg.IPColocationWhitelist {

node/modules/services.go

@@ -35,6 +35,7 @@ import (
 	"github.com/filecoin-project/lotus/lib/peermgr"
 	marketevents "github.com/filecoin-project/lotus/markets/loggers"
 	"github.com/filecoin-project/lotus/node/hello"
+	"github.com/filecoin-project/lotus/node/impl/full"
 	"github.com/filecoin-project/lotus/node/modules/dtypes"
 	"github.com/filecoin-project/lotus/node/modules/helpers"
 	"github.com/filecoin-project/lotus/node/repo"
@@ -198,10 +199,10 @@ func HandleIncomingMessages(mctx helpers.MetricsCtx, lc fx.Lifecycle, ps *pubsub
 	waitForSync(stmgr, pubsubMsgsSyncEpochs, subscribe)
 }
 
-func RelayIndexerMessages(lc fx.Lifecycle, ps *pubsub.PubSub, nn dtypes.NetworkName, h host.Host) error {
+func RelayIndexerMessages(lc fx.Lifecycle, ps *pubsub.PubSub, nn dtypes.NetworkName, h host.Host, chainModule full.ChainModuleAPI, stateModule full.StateModuleAPI) error {
 	topicName := build.IndexerIngestTopic(nn)
 
-	v := sub.NewIndexerMessageValidator(h.ID())
+	v := sub.NewIndexerMessageValidator(h.ID(), chainModule, stateModule)
 
 	if err := ps.RegisterTopicValidator(topicName, v.Validate); err != nil {
 		return xerrors.Errorf("failed to register validator for topic %s, err: %w", topicName, err)