core: only store pruned sidechains until they beat canonical

consensus/errors.go

@@ -23,6 +23,10 @@ var (
 	// that is unknown.
 	ErrUnknownAncestor = errors.New("unknown ancestor")
 
+	// ErrPrunedAncestor is returned when validating a block requires an ancestor
+	// that is known, but the state of which is not available.
+	ErrPrunedAncestor = errors.New("pruned ancestor")
+
 	// ErrFutureBlock is returned when a block's timestamp is in the future according
 	// to the current node.
 	ErrFutureBlock = errors.New("block in the future")

core/block_validator.go

@@ -54,7 +54,10 @@ func (v *BlockValidator) ValidateBody(block *types.Block) error {
 		return ErrKnownBlock
 	}
 	if !v.bc.HasBlockAndState(block.ParentHash(), block.NumberU64()-1) {
-		return consensus.ErrUnknownAncestor
+		if !v.bc.HasBlock(block.ParentHash(), block.NumberU64()-1) {
+			return consensus.ErrUnknownAncestor
+		}
+		return consensus.ErrPrunedAncestor
 	}
 	// Header validity is known at this point, check the uncles and transactions
 	header := block.Header()

core/blockchain.go

@@ -552,6 +552,12 @@ func (bc *BlockChain) HasBlock(hash common.Hash, number uint64) bool {
 	return ok
 }
 
+// HasState checks if state trie is fully present in the database or not.
+func (bc *BlockChain) HasState(hash common.Hash) bool {
+	_, err := bc.stateCache.OpenTrie(hash)
+	return err == nil
+}
+
 // HasBlockAndState checks if a block and associated state trie is fully present
 // in the database or not, caching it if present.
 func (bc *BlockChain) HasBlockAndState(hash common.Hash, number uint64) bool {
@@ -560,9 +566,7 @@ func (bc *BlockChain) HasBlockAndState(hash common.Hash, number uint64) bool {
 	if block == nil {
 		return false
 	}
-	// Ensure the associated state is also present
-	_, err := bc.stateCache.OpenTrie(block.Root())
-	return err == nil
+	return bc.HasState(block.Root())
 }
 
 // GetBlock retrieves a block from the database by hash and number,
@@ -843,8 +847,24 @@ func (bc *BlockChain) InsertReceiptChain(blockChain types.Blocks, receiptChain [
 
 var lastWrite uint64
 
-// WriteBlock writes the block to the chain.
-func (bc *BlockChain) WriteBlockAndState(block *types.Block, receipts []*types.Receipt, state *state.StateDB) (status WriteStatus, err error) {
+// WriteBlockWithoutState writes only the block and its metadata to the database,
+// but does not write any state. This is used to construct competing side forks
+// up to teh point where they exceed the canonical total difficulty.
+func (bc *BlockChain) WriteBlockWithoutState(block *types.Block, td *big.Int) (err error) {
+	bc.wg.Add(1)
+	defer bc.wg.Done()
+
+	if err := bc.hc.WriteTd(block.Hash(), block.NumberU64(), td); err != nil {
+		return err
+	}
+	if err := WriteBlock(bc.diskdb, block); err != nil {
+		return err
+	}
+	return nil
+}
+
+// WriteBlockWithState writes the block and all assocaited state to the database.
+func (bc *BlockChain) WriteBlockWithState(block *types.Block, receipts []*types.Receipt, state *state.StateDB) (status WriteStatus, err error) {
 	bc.wg.Add(1)
 	defer bc.wg.Done()
 
@@ -1017,31 +1037,60 @@ func (bc *BlockChain) insertChain(chain types.Blocks) (int, []interface{}, []*ty
 		if err == nil {
 			err = bc.Validator().ValidateBody(block)
 		}
-		if err != nil {
-			if err == ErrKnownBlock {
-				stats.ignored++
-				continue
-			}
+		switch {
+		case err == ErrKnownBlock:
+			stats.ignored++
+			continue
 
-			if err == consensus.ErrFutureBlock {
-				// Allow up to MaxFuture second in the future blocks. If this limit
-				// is exceeded the chain is discarded and processed at a later time
-				// if given.
-				max := big.NewInt(time.Now().Unix() + maxTimeFutureBlocks)
-				if block.Time().Cmp(max) > 0 {
-					return i, events, coalescedLogs, fmt.Errorf("future block: %v > %v", block.Time(), max)
+		case err == consensus.ErrFutureBlock:
+			// Allow up to MaxFuture second in the future blocks. If this limit is exceeded
+			// the chain is discarded and processed at a later time if given.
+			max := big.NewInt(time.Now().Unix() + maxTimeFutureBlocks)
+			if block.Time().Cmp(max) > 0 {
+				return i, events, coalescedLogs, fmt.Errorf("future block: %v > %v", block.Time(), max)
+			}
+			bc.futureBlocks.Add(block.Hash(), block)
+			stats.queued++
+			continue
+
+		case err == consensus.ErrUnknownAncestor && bc.futureBlocks.Contains(block.ParentHash()):
+			bc.futureBlocks.Add(block.Hash(), block)
+			stats.queued++
+			continue
+
+		case err == consensus.ErrPrunedAncestor:
+			// Block competing with the canonical chain, store in the db, but don't process
+			// until the competitor TD goes above the canonical TD
+			localTd := bc.GetTd(bc.currentBlock.Hash(), bc.currentBlock.NumberU64())
+			externTd := new(big.Int).Add(bc.GetTd(block.ParentHash(), block.NumberU64()-1), block.Difficulty())
+			if localTd.Cmp(externTd) > 0 {
+				if err = bc.WriteBlockWithoutState(block, externTd); err != nil {
+					return i, events, coalescedLogs, err
 				}
-				bc.futureBlocks.Add(block.Hash(), block)
-				stats.queued++
 				continue
 			}
+			// Competitor chain beat canonical, gather all blocks from the common ancestor
+			var winner []*types.Block
 
-			if err == consensus.ErrUnknownAncestor && bc.futureBlocks.Contains(block.ParentHash()) {
-				bc.futureBlocks.Add(block.Hash(), block)
-				stats.queued++
-				continue
+			parent := bc.GetBlock(block.ParentHash(), block.NumberU64()-1)
+			for !bc.HasState(parent.Root()) {
+				winner = append(winner, parent)
+				parent = bc.GetBlock(parent.ParentHash(), parent.NumberU64()-1)
+			}
+			for j := 0; j < len(winner)/2; j++ {
+				winner[j], winner[len(winner)-1-j] = winner[len(winner)-1-j], winner[j]
+			}
+			// Import all the pruned blocks to make the state available
+			bc.chainmu.Unlock()
+			_, evs, logs, err := bc.insertChain(winner)
+			bc.chainmu.Lock()
+			events, coalescedLogs = evs, logs
+
+			if err != nil {
+				return i, events, coalescedLogs, err
 			}
 
+		case err != nil:
 			bc.reportBlock(block, nil, err)
 			return i, events, coalescedLogs, err
 		}
@@ -1072,7 +1121,7 @@ func (bc *BlockChain) insertChain(chain types.Blocks) (int, []interface{}, []*ty
 		bc.procTime += time.Since(bstart)
 
 		// Write the block to the chain and get the status.
-		status, err := bc.WriteBlockAndState(block, receipts, state)
+		status, err := bc.WriteBlockWithState(block, receipts, state)
 		if err != nil {
 			return i, events, coalescedLogs, err
 		}

core/blockchain_test.go

@@ -1291,3 +1291,56 @@ func TestTrieForkGC(t *testing.T) {
 		t.Fatalf("stale tries still alive after garbase collection")
 	}
 }
+
+// Tests that doing large reorgs works even if the state associated with the
+// forking point is not available any more.
+func TestLargeReorgTrieGC(t *testing.T) {
+	// Generate the original common chain segment and the two competing forks
+	engine := ethash.NewFaker()
+
+	db, _ := ethdb.NewMemDatabase()
+	genesis := new(Genesis).MustCommit(db)
+
+	shared, _ := GenerateChain(params.TestChainConfig, genesis, engine, db, 64, func(i int, b *BlockGen) { b.SetCoinbase(common.Address{1}) })
+	original, _ := GenerateChain(params.TestChainConfig, shared[len(shared)-1], engine, db, 2*triesInMemory, func(i int, b *BlockGen) { b.SetCoinbase(common.Address{2}) })
+	competitor, _ := GenerateChain(params.TestChainConfig, shared[len(shared)-1], engine, db, 2*triesInMemory+1, func(i int, b *BlockGen) { b.SetCoinbase(common.Address{3}) })
+
+	// Import the shared chain and the original canonical one
+	diskdb, _ := ethdb.NewMemDatabase()
+	new(Genesis).MustCommit(diskdb)
+
+	chain, err := NewBlockChain(diskdb, nil, params.TestChainConfig, engine, vm.Config{})
+	if err != nil {
+		t.Fatalf("failed to create tester chain: %v", err)
+	}
+	if _, err := chain.InsertChain(shared); err != nil {
+		t.Fatalf("failed to insert shared chain: %v", err)
+	}
+	if _, err := chain.InsertChain(original); err != nil {
+		t.Fatalf("failed to insert shared chain: %v", err)
+	}
+	// Ensure that the state associated with the forking point is pruned away
+	if node, _ := chain.triedb.Node(shared[len(shared)-1].Root()); node != nil {
+		t.Fatalf("common-but-old ancestor still cache")
+	}
+	// Import the competitor chain without exceeding the canonical's TD and ensure
+	// we have not processed any of the blocks (protection against malicious blocks)
+	if _, err := chain.InsertChain(competitor[:len(competitor)-2]); err != nil {
+		t.Fatalf("failed to insert competitor chain: %v", err)
+	}
+	for i, block := range competitor[:len(competitor)-2] {
+		if node, _ := chain.triedb.Node(block.Root()); node != nil {
+			t.Fatalf("competitor %d: low TD chain became processed", i)
+		}
+	}
+	// Import the head of the competitor chain, triggering the reorg and ensure we
+	// successfully reprocess all the stashed away blocks.
+	if _, err := chain.InsertChain(competitor[len(competitor)-2:]); err != nil {
+		t.Fatalf("failed to finalize competitor chain: %v", err)
+	}
+	for i, block := range competitor[:len(competitor)-triesInMemory] {
+		if node, _ := chain.triedb.Node(block.Root()); node != nil {
+			t.Fatalf("competitor %d: competing chain state missing", i)
+		}
+	}
+}

miner/worker.go

@@ -309,7 +309,7 @@ func (self *worker) wait() {
 			for _, log := range work.state.Logs() {
 				log.BlockHash = block.Hash()
 			}
-			stat, err := self.chain.WriteBlockAndState(block, work.receipts, work.state)
+			stat, err := self.chain.WriteBlockWithState(block, work.receipts, work.state)
 			if err != nil {
 				log.Error("Failed writing block to chain", "err", err)
 				continue