core, eth, trie: prepare trie sync for path based operation

core/state/sync_test.go

@@ -26,6 +26,7 @@ import (
 	"github.com/ethereum/go-ethereum/crypto"
 	"github.com/ethereum/go-ethereum/ethdb"
 	"github.com/ethereum/go-ethereum/ethdb/memorydb"
+	"github.com/ethereum/go-ethereum/rlp"
 	"github.com/ethereum/go-ethereum/trie"
 )
 
@@ -44,7 +45,7 @@ func makeTestState() (Database, common.Hash, []*testAccount) {
 	state, _ := New(common.Hash{}, db, nil)
 
 	// Fill it with some arbitrary data
-	accounts := []*testAccount{}
+	var accounts []*testAccount
 	for i := byte(0); i < 96; i++ {
 		obj := state.GetOrNewStateObject(common.BytesToAddress([]byte{i}))
 		acc := &testAccount{address: common.BytesToAddress([]byte{i})}
@@ -59,6 +60,11 @@ func makeTestState() (Database, common.Hash, []*testAccount) {
 			obj.SetCode(crypto.Keccak256Hash([]byte{i, i, i, i, i}), []byte{i, i, i, i, i})
 			acc.code = []byte{i, i, i, i, i}
 		}
+		if i%5 == 0 {
+			for j := byte(0); j < 5; j++ {
+				obj.SetState(db, crypto.Keccak256Hash([]byte{i, i, i, i, i, j, j}), crypto.Keccak256Hash([]byte{i, i, i, i, i, j, j}))
+			}
+		}
 		state.updateStateObject(obj)
 		accounts = append(accounts, acc)
 	}
@@ -126,44 +132,94 @@ func checkStateConsistency(db ethdb.Database, root common.Hash) error {
 // Tests that an empty state is not scheduled for syncing.
 func TestEmptyStateSync(t *testing.T) {
 	empty := common.HexToHash("56e81f171bcc55a6ff8345e692c0f86e5b48e01b996cadc001622fb5e363b421")
-	if req := NewStateSync(empty, rawdb.NewMemoryDatabase(), trie.NewSyncBloom(1, memorydb.New())).Missing(1); len(req) != 0 {
-		t.Errorf("content requested for empty state: %v", req)
+	sync := NewStateSync(empty, rawdb.NewMemoryDatabase(), trie.NewSyncBloom(1, memorydb.New()))
+	if nodes, paths, codes := sync.Missing(1); len(nodes) != 0 || len(paths) != 0 || len(codes) != 0 {
+		t.Errorf(" content requested for empty state: %v, %v, %v", nodes, paths, codes)
 	}
 }
 
 // Tests that given a root hash, a state can sync iteratively on a single thread,
 // requesting retrieval tasks and returning all of them in one go.
-func TestIterativeStateSyncIndividual(t *testing.T)         { testIterativeStateSync(t, 1, false) }
-func TestIterativeStateSyncBatched(t *testing.T)            { testIterativeStateSync(t, 100, false) }
-func TestIterativeStateSyncIndividualFromDisk(t *testing.T) { testIterativeStateSync(t, 1, true) }
-func TestIterativeStateSyncBatchedFromDisk(t *testing.T)    { testIterativeStateSync(t, 100, true) }
+func TestIterativeStateSyncIndividual(t *testing.T) {
+	testIterativeStateSync(t, 1, false, false)
+}
+func TestIterativeStateSyncBatched(t *testing.T) {
+	testIterativeStateSync(t, 100, false, false)
+}
+func TestIterativeStateSyncIndividualFromDisk(t *testing.T) {
+	testIterativeStateSync(t, 1, true, false)
+}
+func TestIterativeStateSyncBatchedFromDisk(t *testing.T) {
+	testIterativeStateSync(t, 100, true, false)
+}
+func TestIterativeStateSyncIndividualByPath(t *testing.T) {
+	testIterativeStateSync(t, 1, false, true)
+}
+func TestIterativeStateSyncBatchedByPath(t *testing.T) {
+	testIterativeStateSync(t, 100, false, true)
+}
 
-func testIterativeStateSync(t *testing.T, count int, commit bool) {
+func testIterativeStateSync(t *testing.T, count int, commit bool, bypath bool) {
 	// Create a random state to copy
 	srcDb, srcRoot, srcAccounts := makeTestState()
 	if commit {
 		srcDb.TrieDB().Commit(srcRoot, false, nil)
 	}
+	srcTrie, _ := trie.New(srcRoot, srcDb.TrieDB())
+
 	// Create a destination state and sync with the scheduler
 	dstDb := rawdb.NewMemoryDatabase()
 	sched := NewStateSync(srcRoot, dstDb, trie.NewSyncBloom(1, dstDb))
 
-	queue := append([]common.Hash{}, sched.Missing(count)...)
-	for len(queue) > 0 {
-		results := make([]trie.SyncResult, len(queue))
-		for i, hash := range queue {
+	nodes, paths, codes := sched.Missing(count)
+	var (
+		hashQueue []common.Hash
+		pathQueue []trie.SyncPath
+	)
+	if !bypath {
+		hashQueue = append(append(hashQueue[:0], nodes...), codes...)
+	} else {
+		hashQueue = append(hashQueue[:0], codes...)
+		pathQueue = append(pathQueue[:0], paths...)
+	}
+	for len(hashQueue)+len(pathQueue) > 0 {
+		results := make([]trie.SyncResult, len(hashQueue)+len(pathQueue))
+		for i, hash := range hashQueue {
 			data, err := srcDb.TrieDB().Node(hash)
 			if err != nil {
 				data, err = srcDb.ContractCode(common.Hash{}, hash)
 			}
 			if err != nil {
-				t.Fatalf("failed to retrieve node data for %x", hash)
+				t.Fatalf("failed to retrieve node data for hash %x", hash)
 			}
 			results[i] = trie.SyncResult{Hash: hash, Data: data}
 		}
+		for i, path := range pathQueue {
+			if len(path) == 1 {
+				data, _, err := srcTrie.TryGetNode(path[0])
+				if err != nil {
+					t.Fatalf("failed to retrieve node data for path %x: %v", path, err)
+				}
+				results[len(hashQueue)+i] = trie.SyncResult{Hash: crypto.Keccak256Hash(data), Data: data}
+			} else {
+				var acc Account
+				if err := rlp.DecodeBytes(srcTrie.Get(path[0]), &acc); err != nil {
+					t.Fatalf("failed to decode account on path %x: %v", path, err)
+				}
+				stTrie, err := trie.New(acc.Root, srcDb.TrieDB())
+				if err != nil {
+					t.Fatalf("failed to retriev storage trie for path %x: %v", path, err)
+				}
+				data, _, err := stTrie.TryGetNode(path[1])
+				if err != nil {
+					t.Fatalf("failed to retrieve node data for path %x: %v", path, err)
+				}
+				results[len(hashQueue)+i] = trie.SyncResult{Hash: crypto.Keccak256Hash(data), Data: data}
+			}
+		}
 		for _, result := range results {
 			if err := sched.Process(result); err != nil {
-				t.Fatalf("failed to process result %v", err)
+				t.Errorf("failed to process result %v", err)
 			}
 		}
 		batch := dstDb.NewBatch()
@@ -171,7 +227,14 @@ func testIterativeStateSync(t *testing.T, count int, commit bool) {
 			t.Fatalf("failed to commit data: %v", err)
 		}
 		batch.Write()
-		queue = append(queue[:0], sched.Missing(count)...)
+
+		nodes, paths, codes = sched.Missing(count)
+		if !bypath {
+			hashQueue = append(append(hashQueue[:0], nodes...), codes...)
+		} else {
+			hashQueue = append(hashQueue[:0], codes...)
+			pathQueue = append(pathQueue[:0], paths...)
+		}
 	}
 	// Cross check that the two states are in sync
 	checkStateAccounts(t, dstDb, srcRoot, srcAccounts)
@@ -187,7 +250,9 @@ func TestIterativeDelayedStateSync(t *testing.T) {
 	dstDb := rawdb.NewMemoryDatabase()
 	sched := NewStateSync(srcRoot, dstDb, trie.NewSyncBloom(1, dstDb))
 
-	queue := append([]common.Hash{}, sched.Missing(0)...)
+	nodes, _, codes := sched.Missing(0)
+	queue := append(append([]common.Hash{}, nodes...), codes...)
+
 	for len(queue) > 0 {
 		// Sync only half of the scheduled nodes
 		results := make([]trie.SyncResult, len(queue)/2+1)
@@ -211,7 +276,9 @@ func TestIterativeDelayedStateSync(t *testing.T) {
 			t.Fatalf("failed to commit data: %v", err)
 		}
 		batch.Write()
-		queue = append(queue[len(results):], sched.Missing(0)...)
+
+		nodes, _, codes = sched.Missing(0)
+		queue = append(append(queue[len(results):], nodes...), codes...)
 	}
 	// Cross check that the two states are in sync
 	checkStateAccounts(t, dstDb, srcRoot, srcAccounts)
@@ -232,7 +299,8 @@ func testIterativeRandomStateSync(t *testing.T, count int) {
 	sched := NewStateSync(srcRoot, dstDb, trie.NewSyncBloom(1, dstDb))
 
 	queue := make(map[common.Hash]struct{})
-	for _, hash := range sched.Missing(count) {
+	nodes, _, codes := sched.Missing(count)
+	for _, hash := range append(nodes, codes...) {
 		queue[hash] = struct{}{}
 	}
 	for len(queue) > 0 {
@@ -259,8 +327,10 @@ func testIterativeRandomStateSync(t *testing.T, count int) {
 			t.Fatalf("failed to commit data: %v", err)
 		}
 		batch.Write()
+
 		queue = make(map[common.Hash]struct{})
-		for _, hash := range sched.Missing(count) {
+		nodes, _, codes = sched.Missing(count)
+		for _, hash := range append(nodes, codes...) {
 			queue[hash] = struct{}{}
 		}
 	}
@@ -279,7 +349,8 @@ func TestIterativeRandomDelayedStateSync(t *testing.T) {
 	sched := NewStateSync(srcRoot, dstDb, trie.NewSyncBloom(1, dstDb))
 
 	queue := make(map[common.Hash]struct{})
-	for _, hash := range sched.Missing(0) {
+	nodes, _, codes := sched.Missing(0)
+	for _, hash := range append(nodes, codes...) {
 		queue[hash] = struct{}{}
 	}
 	for len(queue) > 0 {
@@ -312,7 +383,11 @@ func TestIterativeRandomDelayedStateSync(t *testing.T) {
 			t.Fatalf("failed to commit data: %v", err)
 		}
 		batch.Write()
-		for _, hash := range sched.Missing(0) {
+		for _, result := range results {
+			delete(queue, result.Hash)
+		}
+		nodes, _, codes = sched.Missing(0)
+		for _, hash := range append(nodes, codes...) {
 			queue[hash] = struct{}{}
 		}
 	}
@@ -341,8 +416,11 @@ func TestIncompleteStateSync(t *testing.T) {
 	dstDb := rawdb.NewMemoryDatabase()
 	sched := NewStateSync(srcRoot, dstDb, trie.NewSyncBloom(1, dstDb))
 
-	added := []common.Hash{}
-	queue := append([]common.Hash{}, sched.Missing(1)...)
+	var added []common.Hash
+
+	nodes, _, codes := sched.Missing(1)
+	queue := append(append([]common.Hash{}, nodes...), codes...)
+
 	for len(queue) > 0 {
 		// Fetch a batch of state nodes
 		results := make([]trie.SyncResult, len(queue))
@@ -382,7 +460,8 @@ func TestIncompleteStateSync(t *testing.T) {
 			}
 		}
 		// Fetch the next batch to retrieve
-		queue = append(queue[:0], sched.Missing(1)...)
+		nodes, _, codes = sched.Missing(1)
+		queue = append(append(queue[:0], nodes...), codes...)
 	}
 	// Sanity check that removing any node from the database is detected
 	for _, node := range added[1:] {

trie/secure_trie.go

@@ -79,6 +79,12 @@ func (t *SecureTrie) TryGet(key []byte) ([]byte, error) {
 	return t.trie.TryGet(t.hashKey(key))
 }
 
+// TryGetNode attempts to retrieve a trie node by compact-encoded path. It is not
+// possible to use keybyte-encoding as the path might contain odd nibbles.
+func (t *SecureTrie) TryGetNode(path []byte) ([]byte, int, error) {
+	return t.trie.TryGetNode(path)
+}
+
 // Update associates key with value in the trie. Subsequent calls to
 // Get will return value. If value has length zero, any existing value
 // is deleted from the trie and calls to Get will return nil.

trie/sync.go

@@ -52,6 +52,39 @@ type request struct {
 	callback LeafCallback // Callback to invoke if a leaf node it reached on this branch
 }
 
+// SyncPath is a path tuple identifying a particular trie node either in a single
+// trie (account) or a layered trie (account -> storage).
+//
+// Content wise the tuple either has 1 element if it addresses a node in a single
+// trie or 2 elements if it addresses a node in a stacked trie.
+//
+// To support aiming arbitrary trie nodes, the path needs to support odd nibble
+// lengths. To avoid transferring expanded hex form over the network, the last
+// part of the tuple (which needs to index into the middle of a trie) is compact
+// encoded. In case of a 2-tuple, the first item is always 32 bytes so that is
+// simple binary encoded.
+//
+// Examples:
+//   - Path 0x9  -> {0x19}
+//   - Path 0x99 -> {0x0099}
+//   - Path 0x01234567890123456789012345678901012345678901234567890123456789019  -> {0x0123456789012345678901234567890101234567890123456789012345678901, 0x19}
+//   - Path 0x012345678901234567890123456789010123456789012345678901234567890199 -> {0x0123456789012345678901234567890101234567890123456789012345678901, 0x0099}
+type SyncPath [][]byte
+
+// newSyncPath converts an expanded trie path from nibble form into a compact
+// version that can be sent over the network.
+func newSyncPath(path []byte) SyncPath {
+	// If the hash is from the account trie, append a single item, if it
+	// is from the a storage trie, append a tuple. Note, the length 64 is
+	// clashing between account leaf and storage root. It's fine though
+	// because having a trie node at 64 depth means a hash collision was
+	// found and we're long dead.
+	if len(path) < 64 {
+		return SyncPath{hexToCompact(path)}
+	}
+	return SyncPath{hexToKeybytes(path[:64]), hexToCompact(path[64:])}
+}
+
 // SyncResult is a response with requested data along with it's hash.
 type SyncResult struct {
 	Hash common.Hash // Hash of the originally unknown trie node
@@ -193,10 +226,16 @@ func (s *Sync) AddCodeEntry(hash common.Hash, path []byte, parent common.Hash) {
 	s.schedule(req)
 }
 
-// Missing retrieves the known missing nodes from the trie for retrieval.
-func (s *Sync) Missing(max int) []common.Hash {
-	var requests []common.Hash
-	for !s.queue.Empty() && (max == 0 || len(requests) < max) {
+// Missing retrieves the known missing nodes from the trie for retrieval. To aid
+// both eth/6x style fast sync and snap/1x style state sync, the paths of trie
+// nodes are returned too, as well as separate hash list for codes.
+func (s *Sync) Missing(max int) (nodes []common.Hash, paths []SyncPath, codes []common.Hash) {
+	var (
+		nodeHashes []common.Hash
+		nodePaths  []SyncPath
+		codeHashes []common.Hash
+	)
+	for !s.queue.Empty() && (max == 0 || len(nodeHashes)+len(codeHashes) < max) {
 		// Retrieve th enext item in line
 		item, prio := s.queue.Peek()
 
@@ -208,9 +247,16 @@ func (s *Sync) Missing(max int) []common.Hash {
 		// Item is allowed to be scheduled, add it to the task list
 		s.queue.Pop()
 		s.fetches[depth]++
-		requests = append(requests, item.(common.Hash))
+
+		hash := item.(common.Hash)
+		if req, ok := s.nodeReqs[hash]; ok {
+			nodeHashes = append(nodeHashes, hash)
+			nodePaths = append(nodePaths, newSyncPath(req.path))
+		} else {
+			codeHashes = append(codeHashes, hash)
+		}
 	}
-	return requests
+	return nodeHashes, nodePaths, codeHashes
 }
 
 // Process injects the received data for requested item. Note it can
@@ -322,9 +368,13 @@ func (s *Sync) children(req *request, object node) ([]*request, error) {
 
 	switch node := (object).(type) {
 	case *shortNode:
+		key := node.Key
+		if hasTerm(key) {
+			key = key[:len(key)-1]
+		}
 		children = []child{{
 			node: node.Val,
-			path: append(append([]byte(nil), req.path...), node.Key...),
+			path: append(append([]byte(nil), req.path...), key...),
 		}}
 	case *fullNode:
 		for i := 0; i < 17; i++ {
@@ -344,7 +394,7 @@ func (s *Sync) children(req *request, object node) ([]*request, error) {
 		// Notify any external watcher of a new key/value node
 		if req.callback != nil {
 			if node, ok := (child.node).(valueNode); ok {
-				if err := req.callback(req.path, node, req.hash); err != nil {
+				if err := req.callback(child.path, node, req.hash); err != nil {
 					return nil, err
 				}
 			}

trie/trie.go

@@ -25,6 +25,7 @@ import (
 	"github.com/ethereum/go-ethereum/common"
 	"github.com/ethereum/go-ethereum/crypto"
 	"github.com/ethereum/go-ethereum/log"
+	"github.com/ethereum/go-ethereum/rlp"
 )
 
 var (
@@ -102,8 +103,7 @@ func (t *Trie) Get(key []byte) []byte {
 // The value bytes must not be modified by the caller.
 // If a node was not found in the database, a MissingNodeError is returned.
 func (t *Trie) TryGet(key []byte) ([]byte, error) {
-	key = keybytesToHex(key)
-	value, newroot, didResolve, err := t.tryGet(t.root, key, 0)
+	value, newroot, didResolve, err := t.tryGet(t.root, keybytesToHex(key), 0)
 	if err == nil && didResolve {
 		t.root = newroot
 	}
@@ -146,6 +146,86 @@ func (t *Trie) tryGet(origNode node, key []byte, pos int) (value []byte, newnode
 	}
 }
 
+// TryGetNode attempts to retrieve a trie node by compact-encoded path. It is not
+// possible to use keybyte-encoding as the path might contain odd nibbles.
+func (t *Trie) TryGetNode(path []byte) ([]byte, int, error) {
+	item, newroot, resolved, err := t.tryGetNode(t.root, compactToHex(path), 0)
+	if err != nil {
+		return nil, resolved, err
+	}
+	if resolved > 0 {
+		t.root = newroot
+	}
+	if item == nil {
+		return nil, resolved, nil
+	}
+	enc, err := rlp.EncodeToBytes(item)
+	if err != nil {
+		log.Error("Encoding existing trie node failed", "err", err)
+		return nil, resolved, err
+	}
+	return enc, resolved, err
+}
+
+func (t *Trie) tryGetNode(origNode node, path []byte, pos int) (item node, newnode node, resolved int, err error) {
+	// If we reached the requested path, return the current node
+	if pos >= len(path) {
+		// Don't return collapsed hash nodes though
+		if _, ok := origNode.(hashNode); !ok {
+			// Short nodes have expanded keys, compact them before returning
+			item := origNode
+			if sn, ok := item.(*shortNode); ok {
+				item = &shortNode{
+					Key: hexToCompact(sn.Key),
+					Val: sn.Val,
+				}
+			}
+			return item, origNode, 0, nil
+		}
+	}
+	// Path still needs to be traversed, descend into children
+	switch n := (origNode).(type) {
+	case nil:
+		// Non-existent path requested, abort
+		return nil, nil, 0, nil
+
+	case valueNode:
+		// Path prematurely ended, abort
+		return nil, nil, 0, nil
+
+	case *shortNode:
+		if len(path)-pos < len(n.Key) || !bytes.Equal(n.Key, path[pos:pos+len(n.Key)]) {
+			// Path branches off from short node
+			return nil, n, 0, nil
+		}
+		item, newnode, resolved, err = t.tryGetNode(n.Val, path, pos+len(n.Key))
+		if err == nil && resolved > 0 {
+			n = n.copy()
+			n.Val = newnode
+		}
+		return item, n, resolved, err
+
+	case *fullNode:
+		item, newnode, resolved, err = t.tryGetNode(n.Children[path[pos]], path, pos+1)
+		if err == nil && resolved > 0 {
+			n = n.copy()
+			n.Children[path[pos]] = newnode
+		}
+		return item, n, resolved, err
+
+	case hashNode:
+		child, err := t.resolveHash(n, path[:pos])
+		if err != nil {
+			return nil, n, 1, err
+		}
+		item, newnode, resolved, err := t.tryGetNode(child, path, pos)
+		return item, newnode, resolved + 1, err
+
+	default:
+		panic(fmt.Sprintf("%T: invalid node: %v", origNode, origNode))
+	}
+}
+
 // Update associates key with value in the trie. Subsequent calls to
 // Get will return value. If value has length zero, any existing value
 // is deleted from the trie and calls to Get will return nil.