go-ethereum/triedb/pathdb/journal.go

// Copyright 2022 The go-ethereum Authors
// This file is part of the go-ethereum library.
//
// The go-ethereum library is free software: you can redistribute it and/or modify
// it under the terms of the GNU Lesser General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
//
// The go-ethereum library is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU Lesser General Public License for more details.
//
// You should have received a copy of the GNU Lesser General Public License
// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>.

package pathdb

import (
	"bytes"
	"errors"
	"fmt"
	"io"
	"time"

	"github.com/ethereum/go-ethereum/common"
	"github.com/ethereum/go-ethereum/core/rawdb"
	"github.com/ethereum/go-ethereum/core/types"
	"github.com/ethereum/go-ethereum/crypto"
	"github.com/ethereum/go-ethereum/log"
	"github.com/ethereum/go-ethereum/rlp"
)

var (
	errMissJournal       = errors.New("journal not found")
	errMissVersion       = errors.New("version not found")
	errUnexpectedVersion = errors.New("unexpected journal version")
	errMissDiskRoot      = errors.New("disk layer root not found")
	errUnmatchedJournal  = errors.New("unmatched journal")
)

// journalVersion ensures that an incompatible journal is detected and discarded.
//
// Changelog:
//
// - Version 0: initial version
// - Version 1: storage.Incomplete field is removed
const journalVersion uint64 = 1

// loadJournal tries to parse the layer journal from the disk.
func (db *Database) loadJournal(diskRoot common.Hash) (layer, error) {
	journal := rawdb.ReadTrieJournal(db.diskdb)
	if len(journal) == 0 {
		return nil, errMissJournal
	}
	r := rlp.NewStream(bytes.NewReader(journal), 0)

	// Firstly, resolve the first element as the journal version
	version, err := r.Uint64()
	if err != nil {
		return nil, errMissVersion
	}
	if version != journalVersion {
		return nil, fmt.Errorf("%w want %d got %d", errUnexpectedVersion, journalVersion, version)
	}
	// Secondly, resolve the disk layer root, ensure it's continuous
	// with disk layer. Note now we can ensure it's the layer journal
	// correct version, so we expect everything can be resolved properly.
	var root common.Hash
	if err := r.Decode(&root); err != nil {
		return nil, errMissDiskRoot
	}
	// The journal is not matched with persistent state, discard them.
	// It can happen that geth crashes without persisting the journal.
	if !bytes.Equal(root.Bytes(), diskRoot.Bytes()) {
		return nil, fmt.Errorf("%w want %x got %x", errUnmatchedJournal, root, diskRoot)
	}
	// Load the disk layer from the journal
	base, err := db.loadDiskLayer(r)
	if err != nil {
		return nil, err
	}
	// Load all the diff layers from the journal
	head, err := db.loadDiffLayer(base, r)
	if err != nil {
		return nil, err
	}
	log.Debug("Loaded layer journal", "diskroot", diskRoot, "diffhead", head.rootHash())
	return head, nil
}

// loadLayers loads a pre-existing state layer backed by a key-value store.
func (db *Database) loadLayers() layer {
	// Retrieve the root node of persistent state.
	var root = types.EmptyRootHash
	if blob := rawdb.ReadAccountTrieNode(db.diskdb, nil); len(blob) > 0 {
		root = crypto.Keccak256Hash(blob)
	}
	// Load the layers by resolving the journal
	head, err := db.loadJournal(root)
	if err == nil {
		return head
	}
	// journal is not matched(or missing) with the persistent state, discard
	// it. Display log for discarding journal, but try to avoid showing
	// useless information when the db is created from scratch.
	if !(root == types.EmptyRootHash && errors.Is(err, errMissJournal)) {
		log.Info("Failed to load journal, discard it", "err", err)
	}
	// Return single layer with persistent state.
	return newDiskLayer(root, rawdb.ReadPersistentStateID(db.diskdb), db, nil, newBuffer(db.config.WriteBufferSize, nil, 0))
}

// loadDiskLayer reads the binary blob from the layer journal, reconstructing
// a new disk layer on it.
func (db *Database) loadDiskLayer(r *rlp.Stream) (layer, error) {
	// Resolve disk layer root
	var root common.Hash
	if err := r.Decode(&root); err != nil {
		return nil, fmt.Errorf("load disk root: %v", err)
	}
	// Resolve the state id of disk layer, it can be different
	// with the persistent id tracked in disk, the id distance
	// is the number of transitions aggregated in disk layer.
	var id uint64
	if err := r.Decode(&id); err != nil {
		return nil, fmt.Errorf("load state id: %v", err)
	}
	stored := rawdb.ReadPersistentStateID(db.diskdb)
	if stored > id {
		return nil, fmt.Errorf("invalid state id: stored %d resolved %d", stored, id)
	}
	// Resolve nodes cached in aggregated buffer
	var nodes nodeSet
	if err := nodes.decode(r); err != nil {
		return nil, err
	}
	return newDiskLayer(root, id, db, nil, newBuffer(db.config.WriteBufferSize, &nodes, id-stored)), nil
}

// loadDiffLayer reads the next sections of a layer journal, reconstructing a new
// diff and verifying that it can be linked to the requested parent.
func (db *Database) loadDiffLayer(parent layer, r *rlp.Stream) (layer, error) {
	// Read the next diff journal entry
	var root common.Hash
	if err := r.Decode(&root); err != nil {
		// The first read may fail with EOF, marking the end of the journal
		if err == io.EOF {
			return parent, nil
		}
		return nil, fmt.Errorf("load diff root: %v", err)
	}
	var block uint64
	if err := r.Decode(&block); err != nil {
		return nil, fmt.Errorf("load block number: %v", err)
	}
	// Read in-memory trie nodes from journal
	var nodes nodeSet
	if err := nodes.decode(r); err != nil {
		return nil, err
	}
	// Read flat states set (with original value attached) from journal
	var stateSet StateSetWithOrigin
	if err := stateSet.decode(r); err != nil {
		return nil, err
	}
	return db.loadDiffLayer(newDiffLayer(parent, root, parent.stateID()+1, block, &nodes, &stateSet), r)
}

// journal implements the layer interface, marshaling the un-flushed trie nodes
// along with layer meta data into provided byte buffer.
func (dl *diskLayer) journal(w io.Writer) error {
	dl.lock.RLock()
	defer dl.lock.RUnlock()

	// Ensure the layer didn't get stale
	if dl.stale {
		return errSnapshotStale
	}
	// Step one, write the disk root into the journal.
	if err := rlp.Encode(w, dl.root); err != nil {
		return err
	}
	// Step two, write the corresponding state id into the journal
	if err := rlp.Encode(w, dl.id); err != nil {
		return err
	}
	// Step three, write the accumulated trie nodes into the journal
	if err := dl.buffer.nodes.encode(w); err != nil {
		return err
	}
	log.Debug("Journaled pathdb disk layer", "root", dl.root)
	return nil
}

// journal implements the layer interface, writing the memory layer contents
// into a buffer to be stored in the database as the layer journal.
func (dl *diffLayer) journal(w io.Writer) error {
	dl.lock.RLock()
	defer dl.lock.RUnlock()

	// journal the parent first
	if err := dl.parent.journal(w); err != nil {
		return err
	}
	// Everything below was journaled, persist this layer too
	if err := rlp.Encode(w, dl.root); err != nil {
		return err
	}
	if err := rlp.Encode(w, dl.block); err != nil {
		return err
	}
	// Write the accumulated trie nodes into buffer
	if err := dl.nodes.encode(w); err != nil {
		return err
	}
	// Write the associated flat state set into buffer
	if err := dl.states.encode(w); err != nil {
		return err
	}
	log.Debug("Journaled pathdb diff layer", "root", dl.root, "parent", dl.parent.rootHash(), "id", dl.stateID(), "block", dl.block)
	return nil
}

// Journal commits an entire diff hierarchy to disk into a single journal entry.
// This is meant to be used during shutdown to persist the layer without
// flattening everything down (bad for reorgs). And this function will mark the
// database as read-only to prevent all following mutation to disk.
func (db *Database) Journal(root common.Hash) error {
	// Retrieve the head layer to journal from.
	l := db.tree.get(root)
	if l == nil {
		return fmt.Errorf("triedb layer [%#x] missing", root)
	}
	disk := db.tree.bottom()
	if l, ok := l.(*diffLayer); ok {
		log.Info("Persisting dirty state to disk", "head", l.block, "root", root, "layers", l.id-disk.id+disk.buffer.layers)
	} else { // disk layer only on noop runs (likely) or deep reorgs (unlikely)
		log.Info("Persisting dirty state to disk", "root", root, "layers", disk.buffer.layers)
	}
	start := time.Now()

	// Run the journaling
	db.lock.Lock()
	defer db.lock.Unlock()

	// Short circuit if the database is in read only mode.
	if db.readOnly {
		return errDatabaseReadOnly
	}
	// Firstly write out the metadata of journal
	journal := new(bytes.Buffer)
	if err := rlp.Encode(journal, journalVersion); err != nil {
		return err
	}
	// Secondly write out the state root in disk, ensure all layers
	// on top are continuous with disk.
	diskRoot := types.EmptyRootHash
	if blob := rawdb.ReadAccountTrieNode(db.diskdb, nil); len(blob) > 0 {
		diskRoot = crypto.Keccak256Hash(blob)
	}
	if err := rlp.Encode(journal, diskRoot); err != nil {
		return err
	}
	// Finally write out the journal of each layer in reverse order.
	if err := l.journal(journal); err != nil {
		return err
	}
	// Store the journal into the database and return
	rawdb.WriteTrieJournal(db.diskdb, journal.Bytes())

	// Set the db in read only mode to reject all following mutations
	db.readOnly = true
	log.Info("Persisted dirty state to disk", "size", common.StorageSize(journal.Len()), "elapsed", common.PrettyDuration(time.Since(start)))
	return nil
}