core/state: accumulate writes and only update tries when must

core/state/state_object.go

@@ -79,9 +79,10 @@ type stateObject struct {
 	trie Trie // storage trie, which becomes non-nil on first access
 	code Code // contract bytecode, which gets set when code is loaded
 
-	originStorage Storage // Storage cache of original entries to dedup rewrites
-	dirtyStorage  Storage // Storage entries that need to be flushed to disk
-	fakeStorage   Storage // Fake storage which constructed by caller for debugging purpose.
+	originStorage  Storage // Storage cache of original entries to dedup rewrites, reset for every transaction
+	pendingStorage Storage // Storage entries that need to be flushed to disk, at the end of an entire block
+	dirtyStorage   Storage // Storage entries that have been modified in the current transaction execution
+	fakeStorage    Storage // Fake storage which constructed by caller for debugging purpose.
 
 	// Cache flags.
 	// When an object is marked suicided it will be delete from the trie
@@ -113,13 +114,17 @@ func newObject(db *StateDB, address common.Address, data Account) *stateObject {
 	if data.CodeHash == nil {
 		data.CodeHash = emptyCodeHash
 	}
+	if data.Root == (common.Hash{}) {
+		data.Root = emptyRoot
+	}
 	return &stateObject{
-		db:            db,
-		address:       address,
-		addrHash:      crypto.Keccak256Hash(address[:]),
-		data:          data,
-		originStorage: make(Storage),
-		dirtyStorage:  make(Storage),
+		db:             db,
+		address:        address,
+		addrHash:       crypto.Keccak256Hash(address[:]),
+		data:           data,
+		originStorage:  make(Storage),
+		pendingStorage: make(Storage),
+		dirtyStorage:   make(Storage),
 	}
 }
 
@@ -183,9 +188,11 @@ func (s *stateObject) GetCommittedState(db Database, key common.Hash) common.Has
 	if s.fakeStorage != nil {
 		return s.fakeStorage[key]
 	}
-	// If we have the original value cached, return that
-	value, cached := s.originStorage[key]
-	if cached {
+	// If we have a pending write or clean cached, return that
+	if value, pending := s.pendingStorage[key]; pending {
+		return value
+	}
+	if value, cached := s.originStorage[key]; cached {
 		return value
 	}
 	// Track the amount of time wasted on reading the storage trie
@@ -198,6 +205,7 @@ func (s *stateObject) GetCommittedState(db Database, key common.Hash) common.Has
 		s.setError(err)
 		return common.Hash{}
 	}
+	var value common.Hash
 	if len(enc) > 0 {
 		_, content, _, err := rlp.Split(enc)
 		if err != nil {
@@ -252,17 +260,29 @@ func (s *stateObject) setState(key, value common.Hash) {
 	s.dirtyStorage[key] = value
 }
 
+// finalise moves all dirty storage slots into the pending area to be hashed or
+// committed later. It is invoked at the end of every transaction.
+func (s *stateObject) finalise() {
+	for key, value := range s.dirtyStorage {
+		s.pendingStorage[key] = value
+	}
+	if len(s.dirtyStorage) > 0 {
+		s.dirtyStorage = make(Storage)
+	}
+}
+
 // updateTrie writes cached storage modifications into the object's storage trie.
 func (s *stateObject) updateTrie(db Database) Trie {
+	// Make sure all dirty slots are finalized into the pending storage area
+	s.finalise()
+
 	// Track the amount of time wasted on updating the storge trie
 	if metrics.EnabledExpensive {
 		defer func(start time.Time) { s.db.StorageUpdates += time.Since(start) }(time.Now())
 	}
-	// Update all the dirty slots in the trie
+	// Insert all the pending updates into the trie
 	tr := s.getTrie(db)
-	for key, value := range s.dirtyStorage {
-		delete(s.dirtyStorage, key)
-
+	for key, value := range s.pendingStorage {
 		// Skip noop changes, persist actual changes
 		if value == s.originStorage[key] {
 			continue
@@ -277,6 +297,9 @@ func (s *stateObject) updateTrie(db Database) Trie {
 		v, _ := rlp.EncodeToBytes(common.TrimLeftZeroes(value[:]))
 		s.setError(tr.TryUpdate(key[:], v))
 	}
+	if len(s.pendingStorage) > 0 {
+		s.pendingStorage = make(Storage)
+	}
 	return tr
 }
 

core/state/statedb_test.go

@@ -449,3 +449,38 @@ func TestCopyOfCopy(t *testing.T) {
 		t.Fatalf("2nd copy fail, expected 42, got %v", got)
 	}
 }
+
+// TestDeleteCreateRevert tests a weird state transition corner case that we hit
+// while changing the internals of statedb. The workflow is that a contract is
+// self destructed, then in a followup transaction (but same block) it's created
+// again and the transaction reverted.
+//
+// The original statedb implementation flushed dirty objects to the tries after
+// each transaction, so this works ok. The rework accumulated writes in memory
+// first, but the journal wiped the entire state object on create-revert.
+func TestDeleteCreateRevert(t *testing.T) {
+	// Create an initial state with a single contract
+	state, _ := New(common.Hash{}, NewDatabase(rawdb.NewMemoryDatabase()))
+
+	addr := toAddr([]byte("so"))
+	state.SetBalance(addr, big.NewInt(1))
+
+	root, _ := state.Commit(false)
+	state.Reset(root)
+
+	// Simulate self-destructing in one transaction, then create-reverting in another
+	state.Suicide(addr)
+	state.Finalise(true)
+
+	id := state.Snapshot()
+	state.SetBalance(addr, big.NewInt(2))
+	state.RevertToSnapshot(id)
+
+	// Commit the entire state and make sure we don't crash and have the correct state
+	root, _ = state.Commit(true)
+	state.Reset(root)
+
+	if state.getStateObject(addr) != nil {
+		t.Fatalf("self-destructed contract came alive")
+	}
+}