core, trie: prevent genesis and last snapshot from being pruned

core/blockchain.go

@@ -95,9 +95,11 @@ type BlockChain struct {
 	chainConfig *params.ChainConfig // Chain & network configuration
 	cacheConfig *CacheConfig        // Cache configuration for pruning
 
-	db     ethdb.Database // Low level persistent database to store final content in
+	db ethdb.Database // Low level persistent database to store final content in
-	triegc *prque.Prque   // Priority queue mapping block numbers to tries to gc
+
-	gcproc time.Duration  // Accumulates canonical block processing for trie dumping
+	gcqueue *prque.Prque  // Priority queue mapping block numbers to tries to gc
+	gcsave  common.Hash   // Root hash of the last trie committed to disk
+	gcproc  time.Duration // Accumulates canonical block processing for trie dumping
 
 	hc            *HeaderChain
 	rmLogsFeed    event.Feed
@@ -159,7 +161,7 @@ func NewBlockChain(db ethdb.Database, cacheConfig *CacheConfig, chainConfig *par
 		chainConfig:    chainConfig,
 		cacheConfig:    cacheConfig,
 		db:             db,
-		triegc:         prque.New(nil),
+		gcqueue:        prque.New(nil),
 		stateCache:     state.NewDatabaseWithCache(db, cacheConfig.TrieCleanLimit),
 		quit:           make(chan struct{}),
 		shouldPreserve: shouldPreserve,
@@ -200,6 +202,12 @@ func NewBlockChain(db ethdb.Database, cacheConfig *CacheConfig, chainConfig *par
 			}
 		}
 	}
+	// Forbid the genesis state (forever) and latest state (temporarilly) from being pruned
+	bc.stateCache.TrieDB().ForbidPrune(bc.genesisBlock.Root())
+	if head := bc.CurrentBlock(); head.NumberU64() > 0 {
+		bc.gcsave = head.Root()
+		bc.stateCache.TrieDB().ForbidPrune(bc.gcsave)
+	}
 	// Take ownership of this particular state
 	go bc.update()
 	return bc, nil
@@ -718,8 +726,8 @@ func (bc *BlockChain) Stop() {
 				}
 			}
 		}
-		for !bc.triegc.Empty() {
+		for !bc.gcqueue.Empty() {
-			triedb.Dereference(bc.triegc.PopItem().(common.Hash), false)
+			triedb.Dereference(bc.gcqueue.PopItem().(common.Hash), false)
 		}
 		if size, _ := triedb.Size(); size != 0 {
 			log.Error("Dangling trie nodes after full cleanup", "size", size)
@@ -967,7 +975,7 @@ func (bc *BlockChain) writeBlockWithState(block *types.Block, receipts []*types.
 	} else {
 		// Full but not archive node, do proper garbage collection
 		triedb.Reference(common.Hash{}, root, common.Hash{}) // metadata reference to keep trie alive
-		bc.triegc.Push(root, -int64(block.NumberU64()))
+		bc.gcqueue.Push(root, -int64(block.NumberU64()))
 
 		if current := block.NumberU64(); current > triesInMemory {
 			// If we exceeded our memory allowance, flush matured singleton nodes to disk
@@ -998,13 +1006,21 @@ func (bc *BlockChain) writeBlockWithState(block *types.Block, receipts []*types.
 					triedb.Commit(header.Root, true)
 					lastWrite = chosen
 					bc.gcproc = 0
+
+					// A new snapshot was flushed to disk, swap the prune allowance
+					triedb.ForbidPrune(header.Root)
+					if bc.gcsave != (common.Hash{}) {
+						triedb.PermitPrune(bc.gcsave)
+						triedb.Dereference(bc.gcsave, true)
+					}
+					bc.gcsave = header.Root
 				}
 			}
 			// Garbage collect anything below our required write retention
-			for !bc.triegc.Empty() {
+			for !bc.gcqueue.Empty() {
-				root, number := bc.triegc.Pop()
+				root, number := bc.gcqueue.Pop()
 				if uint64(-number) > chosen {
-					bc.triegc.Push(root, number)
+					bc.gcqueue.Push(root, number)
 					break
 				}
 				triedb.Dereference(root.(common.Hash), true)

trie/database.go

@@ -101,7 +101,8 @@ func splitNodeKey(key string) (common.Hash, common.Hash) {
 // the disk database. The aim is to accumulate trie writes in-memory and only
 // periodically flush a couple tries to disk, garbage collecting the remainder.
 type Database struct {
-	diskdb ethdb.Database // Persistent storage for matured trie nodes
+	diskdb  ethdb.Database           // Persistent storage for matured trie nodes
+	noprune map[common.Hash]struct{} // Root hashes of the tries that aren't prunable
 
 	cleans  *bigcache.BigCache     // GC friendly memory cache of clean node RLPs
 	dirties map[string]*cachedNode // Data and references relationships of dirty nodes
@@ -363,12 +364,34 @@ func NewDatabaseWithCache(diskdb ethdb.Database, cache int) *Database {
 	}
 	return &Database{
 		diskdb:    diskdb,
+		noprune:   make(map[common.Hash]struct{}),
 		cleans:    cleans,
 		dirties:   map[string]*cachedNode{metaRoot: {}},
 		preimages: make(map[common.Hash][]byte),
 	}
 }
 
+// ForbidPrune adds a root hash to the list of tries that are disallowed from
+// being pruned. The only two ever to be used are the genesis trie and the last
+// committed trie (snapshot).
+func (db *Database) ForbidPrune(root common.Hash) {
+	db.lock.Lock()
+	defer db.lock.Unlock()
+
+	log.Debug("Forbidding pruner to delete trie", "root", root)
+	db.noprune[root] = struct{}{}
+}
+
+// PermitPrune allows a particular trie to be pruned from the disk database. To
+// actually run the pruner, please call a dereference on the root hash.
+func (db *Database) PermitPrune(root common.Hash) {
+	db.lock.Lock()
+	defer db.lock.Unlock()
+
+	log.Debug("Permitting pruner to delete trie", "root", root)
+	delete(db.noprune, root)
+}
+
 // DiskDB retrieves the persistent storage backing the trie database.
 func (db *Database) DiskDB() DatabaseReader {
 	return db.diskdb
@@ -459,7 +482,7 @@ func (db *Database) node(owner common.Hash, hash common.Hash, cachegen uint16) n
 		return nil
 	}
 	if db.cleans != nil {
-		db.cleans.Set(string(hash[:]), enc)
+		db.cleans.Set(key, enc)
 		memcacheCleanMissMeter.Mark(1)
 		memcacheCleanWriteMeter.Mark(int64(len(enc)))
 	}

trie/database_pruning.go

@@ -77,6 +77,16 @@ func (p *pruner) execute() {
 			},
 		})
 	}
+	// Beside all the tries kept in memory, keep anything forbidden from pruning
+	for hash := range p.db.noprune {
+		p.tries = append(p.tries, &traverser{
+			db: p.db,
+			state: &traverserState{
+				node: hashNode(common.CopyBytes(hash[:])), // Need closure, take care!!
+				hash: hash,
+			},
+		})
+	}
 	// Iterate over all the nodes marked for pruning and delete them
 	for _, mark := range p.marks {
 		p.prune(mark.owner, mark.hash, mark.path)
@@ -169,17 +179,13 @@ func (t *traverser) live(owner common.Hash, hash common.Hash, path []byte, unref
 	// Short circuit the liveness check if we already covered this prefix (if this
 	// prefix path was not yet seen in previous tries, no parent could have been
 	// seen either, so no point in checkin upwards further than the first hash).
-	state := t.state
+	for state := t.state; state != nil; state = state.parent {
-	for state != nil {
-		// If we've found a hash node, check if it's an already known result
 		if state.hash != (common.Hash{}) {
 			if unrefs[state.hash] {
 				return false
 			}
 			break
 		}
-		// Not a hash node, traverse further up
-		state = state.parent
 	}
 	// Traverse downward until the prefix matches the path completely
 	path = path[len(t.state.prefix):]
@@ -206,7 +212,7 @@ func (t *traverser) live(owner common.Hash, hash common.Hash, path []byte, unref
 			} else {
 				blob, err := t.db.diskdb.Get([]byte(key))
 				if blob == nil || err != nil {
-					log.Error("Missing referenced node", "owner", owner, "hash", t.state.hash, "path", fmt.Sprintf("%x%x", t.state.prefix, path))
+					log.Error("Missing referenced node", "owner", owner, "hash", t.state.hash.Hex(), "path", fmt.Sprintf("%x%x", t.state.prefix, path))
 					return false
 					//panic(fmt.Sprintf("missing referenced node %x (searching for %x:%x at %x%x)", key, owner, t.state.hash, t.state.prefix, path))
 				}