les: fix checkpoint sync (#20120)

les/checkpointoracle.go

@@ -35,11 +35,8 @@ type checkpointOracle struct {
 	config   *params.CheckpointOracleConfig
 	contract *checkpointoracle.CheckpointOracle
 
-	// Whether the contract backend is set.
-	running int32
-
+	running  int32                                 // Flag whether the contract backend is set or not
 	getLocal func(uint64) params.TrustedCheckpoint // Function used to retrieve local checkpoint
-	syncDoneHook func()                                // Function used to notify that light syncing has completed.
 }
 
 // newCheckpointOracle returns a checkpoint registrar handler.

les/client_handler.go

@@ -42,10 +42,12 @@ type clientHandler struct {
 
 	closeCh  chan struct{}
 	wg       sync.WaitGroup // WaitGroup used to track all connected peers.
+	syncDone func()         // Test hooks when syncing is done.
 }
 
 func newClientHandler(ulcServers []string, ulcFraction int, checkpoint *params.TrustedCheckpoint, backend *LightEthereum) *clientHandler {
 	handler := &clientHandler{
+		checkpoint: checkpoint,
 		backend:    backend,
 		closeCh:    make(chan struct{}),
 	}

les/sync.go

@@ -135,21 +135,24 @@ func (h *clientHandler) synchronise(peer *peer) {
 		mode = legacyCheckpointSync
 		log.Debug("Disable checkpoint syncing", "reason", "checkpoint is hardcoded")
 	case h.backend.oracle == nil || !h.backend.oracle.isRunning():
+		if h.checkpoint == nil {
+			mode = lightSync // Downgrade to light sync unfortunately.
+		} else {
+			checkpoint = h.checkpoint
 			mode = legacyCheckpointSync
+		}
 		log.Debug("Disable checkpoint syncing", "reason", "checkpoint syncing is not activated")
 	}
 	// Notify testing framework if syncing has completed(for testing purpose).
 	defer func() {
-		if h.backend.oracle != nil && h.backend.oracle.syncDoneHook != nil {
-			h.backend.oracle.syncDoneHook()
+		if h.syncDone != nil {
+			h.syncDone()
 		}
 	}()
 	start := time.Now()
 	if mode == checkpointSync || mode == legacyCheckpointSync {
 		// Validate the advertised checkpoint
-		if mode == legacyCheckpointSync {
-			checkpoint = h.checkpoint
-		} else if mode == checkpointSync {
+		if mode == checkpointSync {
 			if err := h.validateCheckpoint(peer); err != nil {
 				log.Debug("Failed to validate checkpoint", "reason", err)
 				h.removePeer(peer.id)

les/sync_test.go

@@ -102,7 +102,7 @@ func testCheckpointSyncing(t *testing.T, protocol int, syncMode int) {
 	}
 
 	done := make(chan error)
-	client.handler.backend.oracle.syncDoneHook = func() {
+	client.handler.syncDone = func() {
 		header := client.handler.backend.blockchain.CurrentHeader()
 		if header.Number.Uint64() == expected {
 			done <- nil
@@ -131,3 +131,102 @@ func testCheckpointSyncing(t *testing.T, protocol int, syncMode int) {
 		t.Error("checkpoint syncing timeout")
 	}
 }
+
+func TestMissOracleBackend(t *testing.T)             { testMissOracleBackend(t, true) }
+func TestMissOracleBackendNoCheckpoint(t *testing.T) { testMissOracleBackend(t, false) }
+
+func testMissOracleBackend(t *testing.T, hasCheckpoint bool) {
+	config := light.TestServerIndexerConfig
+
+	waitIndexers := func(cIndexer, bIndexer, btIndexer *core.ChainIndexer) {
+		for {
+			cs, _, _ := cIndexer.Sections()
+			bts, _, _ := btIndexer.Sections()
+			if cs >= 1 && bts >= 1 {
+				break
+			}
+			time.Sleep(10 * time.Millisecond)
+		}
+	}
+	// Generate 512+4 blocks (totally 1 CHT sections)
+	server, client, tearDown := newClientServerEnv(t, int(config.ChtSize+config.ChtConfirms), 3, waitIndexers, nil, 0, false, false)
+	defer tearDown()
+
+	expected := config.ChtSize + config.ChtConfirms
+
+	s, _, head := server.chtIndexer.Sections()
+	cp := &params.TrustedCheckpoint{
+		SectionIndex: 0,
+		SectionHead:  head,
+		CHTRoot:      light.GetChtRoot(server.db, s-1, head),
+		BloomRoot:    light.GetBloomTrieRoot(server.db, s-1, head),
+	}
+	// Register the assembled checkpoint into oracle.
+	header := server.backend.Blockchain().CurrentHeader()
+
+	data := append([]byte{0x19, 0x00}, append(registrarAddr.Bytes(), append([]byte{0, 0, 0, 0, 0, 0, 0, 0}, cp.Hash().Bytes()...)...)...)
+	sig, _ := crypto.Sign(crypto.Keccak256(data), signerKey)
+	sig[64] += 27 // Transform V from 0/1 to 27/28 according to the yellow paper
+	if _, err := server.handler.server.oracle.contract.RegisterCheckpoint(bind.NewKeyedTransactor(signerKey), cp.SectionIndex, cp.Hash().Bytes(), new(big.Int).Sub(header.Number, big.NewInt(1)), header.ParentHash, [][]byte{sig}); err != nil {
+		t.Error("register checkpoint failed", err)
+	}
+	server.backend.Commit()
+
+	// Wait for the checkpoint registration
+	for {
+		_, hash, _, err := server.handler.server.oracle.contract.Contract().GetLatestCheckpoint(nil)
+		if err != nil || hash == [32]byte{} {
+			time.Sleep(100 * time.Millisecond)
+			continue
+		}
+		break
+	}
+	expected += 1
+
+	// Explicitly set the oracle as nil. In normal use case it can happen
+	// that user wants to unlock something which blocks the oracle backend
+	// initialisation. But at the same time syncing starts.
+	//
+	// See https://github.com/ethereum/go-ethereum/issues/20097 for more detail.
+	//
+	// In this case, client should run light sync or legacy checkpoint sync
+	// if hardcoded checkpoint is configured.
+	client.handler.backend.oracle = nil
+
+	// For some private networks it can happen checkpoint syncing is enabled
+	// but there is no hardcoded checkpoint configured.
+	if hasCheckpoint {
+		client.handler.checkpoint = cp
+		client.handler.backend.blockchain.AddTrustedCheckpoint(cp)
+	}
+
+	done := make(chan error)
+	client.handler.syncDone = func() {
+		header := client.handler.backend.blockchain.CurrentHeader()
+		if header.Number.Uint64() == expected {
+			done <- nil
+		} else {
+			done <- fmt.Errorf("blockchain length mismatch, want %d, got %d", expected, header.Number)
+		}
+	}
+
+	// Create connected peer pair.
+	_, err1, _, err2 := newTestPeerPair("peer", 2, server.handler, client.handler)
+	select {
+	case <-time.After(time.Millisecond * 100):
+	case err := <-err1:
+		t.Fatalf("peer 1 handshake error: %v", err)
+	case err := <-err2:
+		t.Fatalf("peer 2 handshake error: %v", err)
+	}
+
+	select {
+	case err := <-done:
+		if err != nil {
+			t.Error("sync failed", err)
+		}
+		return
+	case <-time.NewTimer(10 * time.Second).C:
+		t.Error("checkpoint syncing timeout")
+	}
+}