forks/go-ethereum
1
0
Fork 1
mirror of https://github.com/ethereum/go-ethereum.git synced 2026-08-20 10:52:25 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Delete trie directory

Browse source
This commit is contained in:
Isabel Schöps Thiel 2024-01-01 00:30:59 +01:00 committed by GitHub
parent 4b27ecf9a7
commit ff12c4b147
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
50 changed files with 0 additions and 16802 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

182
trie/committer.go
View file

@ -1,182 +0,0 @@
// Copyright 2020 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 trie
import (
"fmt"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/trie/trienode"
)
// committer is the tool used for the trie Commit operation. The committer will
// capture all dirty nodes during the commit process and keep them cached in
// insertion order.
type committer struct {
nodes *trienode.NodeSet
tracer *tracer
collectLeaf bool
}
// newCommitter creates a new committer or picks one from the pool.
func newCommitter(nodeset *trienode.NodeSet, tracer *tracer, collectLeaf bool) *committer {
return &committer{
nodes: nodeset,
tracer: tracer,
collectLeaf: collectLeaf,
}
}
// Commit collapses a node down into a hash node.
func (c *committer) Commit(n node) hashNode {
return c.commit(nil, n).(hashNode)
}
// commit collapses a node down into a hash node and returns it.
func (c *committer) commit(path []byte, n node) node {
// if this path is clean, use available cached data
hash, dirty := n.cache()
if hash != nil && !dirty {
return hash
}
// Commit children, then parent, and remove the dirty flag.
switch cn := n.(type) {
case *shortNode:
// Commit child
collapsed := cn.copy()
// If the child is fullNode, recursively commit,
// otherwise it can only be hashNode or valueNode.
if _, ok := cn.Val.(*fullNode); ok {
collapsed.Val = c.commit(append(path, cn.Key...), cn.Val)
}
// The key needs to be copied, since we're adding it to the
// modified nodeset.
collapsed.Key = hexToCompact(cn.Key)
hashedNode := c.store(path, collapsed)
if hn, ok := hashedNode.(hashNode); ok {
return hn
}
return collapsed
case *fullNode:
hashedKids := c.commitChildren(path, cn)
collapsed := cn.copy()
collapsed.Children = hashedKids
hashedNode := c.store(path, collapsed)
if hn, ok := hashedNode.(hashNode); ok {
return hn
}
return collapsed
case hashNode:
return cn
default:
// nil, valuenode shouldn't be committed
panic(fmt.Sprintf("%T: invalid node: %v", n, n))
}
}
// commitChildren commits the children of the given fullnode
func (c *committer) commitChildren(path []byte, n *fullNode) [17]node {
var children [17]node
for i := 0; i < 16; i++ {
child := n.Children[i]
if child == nil {
continue
}
// If it's the hashed child, save the hash value directly.
// Note: it's impossible that the child in range [0, 15]
// is a valueNode.
if hn, ok := child.(hashNode); ok {
children[i] = hn
continue
}
// Commit the child recursively and store the "hashed" value.
// Note the returned node can be some embedded nodes, so it's
// possible the type is not hashNode.
children[i] = c.commit(append(path, byte(i)), child)
}
// For the 17th child, it's possible the type is valuenode.
if n.Children[16] != nil {
children[16] = n.Children[16]
}
return children
}
// store hashes the node n and adds it to the modified nodeset. If leaf collection
// is enabled, leaf nodes will be tracked in the modified nodeset as well.
func (c *committer) store(path []byte, n node) node {
// Larger nodes are replaced by their hash and stored in the database.
var hash, _ = n.cache()
// This was not generated - must be a small node stored in the parent.
// In theory, we should check if the node is leaf here (embedded node
// usually is leaf node). But small value (less than 32bytes) is not
// our target (leaves in account trie only).
if hash == nil {
// The node is embedded in its parent, in other words, this node
// will not be stored in the database independently, mark it as
// deleted only if the node was existent in database before.
_, ok := c.tracer.accessList[string(path)]
if ok {
c.nodes.AddNode(path, trienode.NewDeleted())
}
return n
}
// Collect the dirty node to nodeset for return.
nhash := common.BytesToHash(hash)
c.nodes.AddNode(path, trienode.New(nhash, nodeToBytes(n)))
// Collect the corresponding leaf node if it's required. We don't check
// full node since it's impossible to store value in fullNode. The key
// length of leaves should be exactly same.
if c.collectLeaf {
if sn, ok := n.(*shortNode); ok {
if val, ok := sn.Val.(valueNode); ok {
c.nodes.AddLeaf(nhash, val)
}
}
}
return hash
}
// mptResolver the children resolver in merkle-patricia-tree.
type mptResolver struct{}
// ForEach implements childResolver, decodes the provided node and
// traverses the children inside.
func (resolver mptResolver) ForEach(node []byte, onChild func(common.Hash)) {
forGatherChildren(mustDecodeNodeUnsafe(nil, node), onChild)
}
// forGatherChildren traverses the node hierarchy and invokes the callback
// for all the hashnode children.
func forGatherChildren(n node, onChild func(hash common.Hash)) {
switch n := n.(type) {
case *shortNode:
forGatherChildren(n.Val, onChild)
case *fullNode:
for i := 0; i < 16; i++ {
forGatherChildren(n.Children[i], onChild)
}
case hashNode:
onChild(common.BytesToHash(n))
case valueNode, nil:
default:
panic(fmt.Sprintf("unknown node type: %T", n))
}
}

315
trie/database.go
View file

@ -1,315 +0,0 @@
// 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 trie
import (
"errors"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/trie/triedb/hashdb"
"github.com/ethereum/go-ethereum/trie/triedb/pathdb"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
)
// Config defines all necessary options for database.
type Config struct {
Preimages bool // Flag whether the preimage of node key is recorded
IsVerkle bool // Flag whether the db is holding a verkle tree
HashDB *hashdb.Config // Configs for hash-based scheme
PathDB *pathdb.Config // Configs for experimental path-based scheme
}
// HashDefaults represents a config for using hash-based scheme with
// default settings.
var HashDefaults = &Config{
Preimages: false,
HashDB: hashdb.Defaults,
}
// backend defines the methods needed to access/update trie nodes in different
// state scheme.
type backend interface {
// Scheme returns the identifier of used storage scheme.
Scheme() string
// Initialized returns an indicator if the state data is already initialized
// according to the state scheme.
Initialized(genesisRoot common.Hash) bool
// Size returns the current storage size of the diff layers on top of the
// disk layer and the storage size of the nodes cached in the disk layer.
//
// For hash scheme, there is no differentiation between diff layer nodes
// and dirty disk layer nodes, so both are merged into the second return.
Size() (common.StorageSize, common.StorageSize)
// Update performs a state transition by committing dirty nodes contained
// in the given set in order to update state from the specified parent to
// the specified root.
//
// The passed in maps(nodes, states) will be retained to avoid copying
// everything. Therefore, these maps must not be changed afterwards.
Update(root common.Hash, parent common.Hash, block uint64, nodes *trienode.MergedNodeSet, states *triestate.Set) error
// Commit writes all relevant trie nodes belonging to the specified state
// to disk. Report specifies whether logs will be displayed in info level.
Commit(root common.Hash, report bool) error
// Close closes the trie database backend and releases all held resources.
Close() error
}
// Database is the wrapper of the underlying backend which is shared by different
// types of node backend as an entrypoint. It's responsible for all interactions
// relevant with trie nodes and node preimages.
type Database struct {
config *Config // Configuration for trie database
diskdb ethdb.Database // Persistent database to store the snapshot
preimages *preimageStore // The store for caching preimages
backend backend // The backend for managing trie nodes
}
// NewDatabase initializes the trie database with default settings, note
// the legacy hash-based scheme is used by default.
func NewDatabase(diskdb ethdb.Database, config *Config) *Database {
// Sanitize the config and use the default one if it's not specified.
if config == nil {
config = HashDefaults
}
var preimages *preimageStore
if config.Preimages {
preimages = newPreimageStore(diskdb)
}
db := &Database{
config: config,
diskdb: diskdb,
preimages: preimages,
}
if config.HashDB != nil && config.PathDB != nil {
log.Crit("Both 'hash' and 'path' mode are configured")
}
if config.PathDB != nil {
db.backend = pathdb.New(diskdb, config.PathDB)
} else {
db.backend = hashdb.New(diskdb, config.HashDB, mptResolver{})
}
return db
}
// Reader returns a reader for accessing all trie nodes with provided state root.
// An error will be returned if the requested state is not available.
func (db *Database) Reader(blockRoot common.Hash) (Reader, error) {
switch b := db.backend.(type) {
case *hashdb.Database:
return b.Reader(blockRoot)
case *pathdb.Database:
return b.Reader(blockRoot)
}
return nil, errors.New("unknown backend")
}
// Update performs a state transition by committing dirty nodes contained in the
// given set in order to update state from the specified parent to the specified
// root. The held pre-images accumulated up to this point will be flushed in case
// the size exceeds the threshold.
//
// The passed in maps(nodes, states) will be retained to avoid copying everything.
// Therefore, these maps must not be changed afterwards.
func (db *Database) Update(root common.Hash, parent common.Hash, block uint64, nodes *trienode.MergedNodeSet, states *triestate.Set) error {
if db.preimages != nil {
db.preimages.commit(false)
}
return db.backend.Update(root, parent, block, nodes, states)
}
// Commit iterates over all the children of a particular node, writes them out
// to disk. As a side effect, all pre-images accumulated up to this point are
// also written.
func (db *Database) Commit(root common.Hash, report bool) error {
if db.preimages != nil {
db.preimages.commit(true)
}
return db.backend.Commit(root, report)
}
// Size returns the storage size of diff layer nodes above the persistent disk
// layer, the dirty nodes buffered within the disk layer, and the size of cached
// preimages.
func (db *Database) Size() (common.StorageSize, common.StorageSize, common.StorageSize) {
var (
diffs, nodes common.StorageSize
preimages common.StorageSize
)
diffs, nodes = db.backend.Size()
if db.preimages != nil {
preimages = db.preimages.size()
}
return diffs, nodes, preimages
}
// Initialized returns an indicator if the state data is already initialized
// according to the state scheme.
func (db *Database) Initialized(genesisRoot common.Hash) bool {
return db.backend.Initialized(genesisRoot)
}
// Scheme returns the node scheme used in the database.
func (db *Database) Scheme() string {
return db.backend.Scheme()
}
// Close flushes the dangling preimages to disk and closes the trie database.
// It is meant to be called when closing the blockchain object, so that all
// resources held can be released correctly.
func (db *Database) Close() error {
db.WritePreimages()
return db.backend.Close()
}
// WritePreimages flushes all accumulated preimages to disk forcibly.
func (db *Database) WritePreimages() {
if db.preimages != nil {
db.preimages.commit(true)
}
}
// Preimage retrieves a cached trie node pre-image from memory. If it cannot be
// found cached, the method queries the persistent database for the content.
func (db *Database) Preimage(hash common.Hash) []byte {
if db.preimages == nil {
return nil
}
return db.preimages.preimage(hash)
}
// Cap iteratively flushes old but still referenced trie nodes until the total
// memory usage goes below the given threshold. The held pre-images accumulated
// up to this point will be flushed in case the size exceeds the threshold.
//
// It's only supported by hash-based database and will return an error for others.
func (db *Database) Cap(limit common.StorageSize) error {
hdb, ok := db.backend.(*hashdb.Database)
if !ok {
return errors.New("not supported")
}
if db.preimages != nil {
db.preimages.commit(false)
}
return hdb.Cap(limit)
}
// Reference adds a new reference from a parent node to a child node. This function
// is used to add reference between internal trie node and external node(e.g. storage
// trie root), all internal trie nodes are referenced together by database itself.
//
// It's only supported by hash-based database and will return an error for others.
func (db *Database) Reference(root common.Hash, parent common.Hash) error {
hdb, ok := db.backend.(*hashdb.Database)
if !ok {
return errors.New("not supported")
}
hdb.Reference(root, parent)
return nil
}
// Dereference removes an existing reference from a root node. It's only
// supported by hash-based database and will return an error for others.
func (db *Database) Dereference(root common.Hash) error {
hdb, ok := db.backend.(*hashdb.Database)
if !ok {
return errors.New("not supported")
}
hdb.Dereference(root)
return nil
}
// Recover rollbacks the database to a specified historical point. The state is
// supported as the rollback destination only if it's canonical state and the
// corresponding trie histories are existent. It's only supported by path-based
// database and will return an error for others.
func (db *Database) Recover(target common.Hash) error {
pdb, ok := db.backend.(*pathdb.Database)
if !ok {
return errors.New("not supported")
}
return pdb.Recover(target, &trieLoader{db: db})
}
// Recoverable returns the indicator if the specified state is enabled to be
// recovered. It's only supported by path-based database and will return an
// error for others.
func (db *Database) Recoverable(root common.Hash) (bool, error) {
pdb, ok := db.backend.(*pathdb.Database)
if !ok {
return false, errors.New("not supported")
}
return pdb.Recoverable(root), nil
}
// Disable deactivates the database and invalidates all available state layers
// as stale to prevent access to the persistent state, which is in the syncing
// stage.
//
// It's only supported by path-based database and will return an error for others.
func (db *Database) Disable() error {
pdb, ok := db.backend.(*pathdb.Database)
if !ok {
return errors.New("not supported")
}
return pdb.Disable()
}
// Enable activates database and resets the state tree with the provided persistent
// state root once the state sync is finished.
func (db *Database) Enable(root common.Hash) error {
pdb, ok := db.backend.(*pathdb.Database)
if !ok {
return errors.New("not supported")
}
return pdb.Enable(root)
}
// Journal commits an entire diff hierarchy to disk into a single journal entry.
// This is meant to be used during shutdown to persist the snapshot without
// flattening everything down (bad for reorgs). It's only supported by path-based
// database and will return an error for others.
func (db *Database) Journal(root common.Hash) error {
pdb, ok := db.backend.(*pathdb.Database)
if !ok {
return errors.New("not supported")
}
return pdb.Journal(root)
}
// SetBufferSize sets the node buffer size to the provided value(in bytes).
// It's only supported by path-based database and will return an error for
// others.
func (db *Database) SetBufferSize(size int) error {
pdb, ok := db.backend.(*pathdb.Database)
if !ok {
return errors.New("not supported")
}
return pdb.SetBufferSize(size)
}
// IsVerkle returns the indicator if the database is holding a verkle tree.
func (db *Database) IsVerkle() bool {
return db.config.IsVerkle
}

40
trie/database_test.go
View file

@ -1,40 +0,0 @@
// Copyright 2019 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 trie
import (
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/trie/triedb/hashdb"
"github.com/ethereum/go-ethereum/trie/triedb/pathdb"
)
// newTestDatabase initializes the trie database with specified scheme.
func newTestDatabase(diskdb ethdb.Database, scheme string) *Database {
config := &Config{Preimages: false}
if scheme == rawdb.HashScheme {
config.HashDB = &hashdb.Config{
CleanCacheSize: 0,
} // disable clean cache
} else {
config.PathDB = &pathdb.Config{
CleanCacheSize: 0,
DirtyCacheSize: 0,
} // disable clean/dirty cache
}
return NewDatabase(diskdb, config)
}

144
trie/encoding.go
View file

@ -1,144 +0,0 @@
// Copyright 2014 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 trie
// Trie keys are dealt with in three distinct encodings:
//
// KEYBYTES encoding contains the actual key and nothing else. This encoding is the
// input to most API functions.
//
// HEX encoding contains one byte for each nibble of the key and an optional trailing
// 'terminator' byte of value 0x10 which indicates whether or not the node at the key
// contains a value. Hex key encoding is used for nodes loaded in memory because it's
// convenient to access.
//
// COMPACT encoding is defined by the Ethereum Yellow Paper (it's called "hex prefix
// encoding" there) and contains the bytes of the key and a flag. The high nibble of the
// first byte contains the flag; the lowest bit encoding the oddness of the length and
// the second-lowest encoding whether the node at the key is a value node. The low nibble
// of the first byte is zero in the case of an even number of nibbles and the first nibble
// in the case of an odd number. All remaining nibbles (now an even number) fit properly
// into the remaining bytes. Compact encoding is used for nodes stored on disk.
func hexToCompact(hex []byte) []byte {
terminator := byte(0)
if hasTerm(hex) {
terminator = 1
hex = hex[:len(hex)-1]
}
buf := make([]byte, len(hex)/2+1)
buf[0] = terminator << 5 // the flag byte
if len(hex)&1 == 1 {
buf[0] |= 1 << 4 // odd flag
buf[0] |= hex[0] // first nibble is contained in the first byte
hex = hex[1:]
}
decodeNibbles(hex, buf[1:])
return buf
}
// hexToCompactInPlace places the compact key in input buffer, returning the compacted key.
func hexToCompactInPlace(hex []byte) []byte {
var (
hexLen = len(hex) // length of the hex input
firstByte = byte(0)
)
// Check if we have a terminator there
if hexLen > 0 && hex[hexLen-1] == 16 {
firstByte = 1 << 5
hexLen-- // last part was the terminator, ignore that
}
var (
binLen = hexLen/2 + 1
ni = 0 // index in hex
bi = 1 // index in bin (compact)
)
if hexLen&1 == 1 {
firstByte |= 1 << 4 // odd flag
firstByte |= hex[0] // first nibble is contained in the first byte
ni++
}
for ; ni < hexLen; bi, ni = bi+1, ni+2 {
hex[bi] = hex[ni]<<4 | hex[ni+1]
}
hex[0] = firstByte
return hex[:binLen]
}
func compactToHex(compact []byte) []byte {
if len(compact) == 0 {
return compact
}
base := keybytesToHex(compact)
// delete terminator flag
if base[0] < 2 {
base = base[:len(base)-1]
}
// apply odd flag
chop := 2 - base[0]&1
return base[chop:]
}
func keybytesToHex(str []byte) []byte {
l := len(str)*2 + 1
var nibbles = make([]byte, l)
for i, b := range str {
nibbles[i*2] = b / 16
nibbles[i*2+1] = b % 16
}
nibbles[l-1] = 16
return nibbles
}
// hexToKeybytes turns hex nibbles into key bytes.
// This can only be used for keys of even length.
func hexToKeybytes(hex []byte) []byte {
if hasTerm(hex) {
hex = hex[:len(hex)-1]
}
if len(hex)&1 != 0 {
panic("can't convert hex key of odd length")
}
key := make([]byte, len(hex)/2)
decodeNibbles(hex, key)
return key
}
func decodeNibbles(nibbles []byte, bytes []byte) {
for bi, ni := 0, 0; ni < len(nibbles); bi, ni = bi+1, ni+2 {
bytes[bi] = nibbles[ni]<<4 | nibbles[ni+1]
}
}
// prefixLen returns the length of the common prefix of a and b.
func prefixLen(a, b []byte) int {
var i, length = 0, len(a)
if len(b) < length {
length = len(b)
}
for ; i < length; i++ {
if a[i] != b[i] {
break
}
}
return i
}
// hasTerm returns whether a hex key has the terminator flag.
func hasTerm(s []byte) bool {
return len(s) > 0 && s[len(s)-1] == 16
}

146
trie/encoding_test.go
View file

@ -1,146 +0,0 @@
// Copyright 2014 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 trie
import (
"bytes"
crand "crypto/rand"
"encoding/hex"
"math/rand"
"testing"
)
func TestHexCompact(t *testing.T) {
tests := []struct{ hex, compact []byte }{
// empty keys, with and without terminator.
{hex: []byte{}, compact: []byte{0x00}},
{hex: []byte{16}, compact: []byte{0x20}},
// odd length, no terminator
{hex: []byte{1, 2, 3, 4, 5}, compact: []byte{0x11, 0x23, 0x45}},
// even length, no terminator
{hex: []byte{0, 1, 2, 3, 4, 5}, compact: []byte{0x00, 0x01, 0x23, 0x45}},
// odd length, terminator
{hex: []byte{15, 1, 12, 11, 8, 16 /*term*/}, compact: []byte{0x3f, 0x1c, 0xb8}},
// even length, terminator
{hex: []byte{0, 15, 1, 12, 11, 8, 16 /*term*/}, compact: []byte{0x20, 0x0f, 0x1c, 0xb8}},
}
for _, test := range tests {
if c := hexToCompact(test.hex); !bytes.Equal(c, test.compact) {
t.Errorf("hexToCompact(%x) -> %x, want %x", test.hex, c, test.compact)
}
if h := compactToHex(test.compact); !bytes.Equal(h, test.hex) {
t.Errorf("compactToHex(%x) -> %x, want %x", test.compact, h, test.hex)
}
}
}
func TestHexKeybytes(t *testing.T) {
tests := []struct{ key, hexIn, hexOut []byte }{
{key: []byte{}, hexIn: []byte{16}, hexOut: []byte{16}},
{key: []byte{}, hexIn: []byte{}, hexOut: []byte{16}},
{
key: []byte{0x12, 0x34, 0x56},
hexIn: []byte{1, 2, 3, 4, 5, 6, 16},
hexOut: []byte{1, 2, 3, 4, 5, 6, 16},
},
{
key: []byte{0x12, 0x34, 0x5},
hexIn: []byte{1, 2, 3, 4, 0, 5, 16},
hexOut: []byte{1, 2, 3, 4, 0, 5, 16},
},
{
key: []byte{0x12, 0x34, 0x56},
hexIn: []byte{1, 2, 3, 4, 5, 6},
hexOut: []byte{1, 2, 3, 4, 5, 6, 16},
},
}
for _, test := range tests {
if h := keybytesToHex(test.key); !bytes.Equal(h, test.hexOut) {
t.Errorf("keybytesToHex(%x) -> %x, want %x", test.key, h, test.hexOut)
}
if k := hexToKeybytes(test.hexIn); !bytes.Equal(k, test.key) {
t.Errorf("hexToKeybytes(%x) -> %x, want %x", test.hexIn, k, test.key)
}
}
}
func TestHexToCompactInPlace(t *testing.T) {
for i, key := range []string{
"00",
"060a040c0f000a090b040803010801010900080d090a0a0d0903000b10",
"10",
} {
hexBytes, _ := hex.DecodeString(key)
exp := hexToCompact(hexBytes)
got := hexToCompactInPlace(hexBytes)
if !bytes.Equal(exp, got) {
t.Fatalf("test %d: encoding err\ninp %v\ngot %x\nexp %x\n", i, key, got, exp)
}
}
}
func TestHexToCompactInPlaceRandom(t *testing.T) {
for i := 0; i < 10000; i++ {
l := rand.Intn(128)
key := make([]byte, l)
crand.Read(key)
hexBytes := keybytesToHex(key)
hexOrig := []byte(string(hexBytes))
exp := hexToCompact(hexBytes)
got := hexToCompactInPlace(hexBytes)
if !bytes.Equal(exp, got) {
t.Fatalf("encoding err \ncpt %x\nhex %x\ngot %x\nexp %x\n",
key, hexOrig, got, exp)
}
}
}
func BenchmarkHexToCompact(b *testing.B) {
testBytes := []byte{0, 15, 1, 12, 11, 8, 16 /*term*/}
for i := 0; i < b.N; i++ {
hexToCompact(testBytes)
}
}
func BenchmarkHexToCompactInPlace(b *testing.B) {
testBytes := []byte{0, 15, 1, 12, 11, 8, 16 /*term*/}
for i := 0; i < b.N; i++ {
hexToCompactInPlace(testBytes)
}
}
func BenchmarkCompactToHex(b *testing.B) {
testBytes := []byte{0, 15, 1, 12, 11, 8, 16 /*term*/}
for i := 0; i < b.N; i++ {
compactToHex(testBytes)
}
}
func BenchmarkKeybytesToHex(b *testing.B) {
testBytes := []byte{7, 6, 6, 5, 7, 2, 6, 2, 16}
for i := 0; i < b.N; i++ {
keybytesToHex(testBytes)
}
}
func BenchmarkHexToKeybytes(b *testing.B) {
testBytes := []byte{7, 6, 6, 5, 7, 2, 6, 2, 16}
for i := 0; i < b.N; i++ {
hexToKeybytes(testBytes)
}
}

52
trie/errors.go
View file

@ -1,52 +0,0 @@
// Copyright 2015 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 trie
import (
"errors"
"fmt"
"github.com/ethereum/go-ethereum/common"
)
// ErrCommitted is returned when a already committed trie is requested for usage.
// The potential usages can be `Get`, `Update`, `Delete`, `NodeIterator`, `Prove`
// and so on.
var ErrCommitted = errors.New("trie is already committed")
// MissingNodeError is returned by the trie functions (Get, Update, Delete)
// in the case where a trie node is not present in the local database. It contains
// information necessary for retrieving the missing node.
type MissingNodeError struct {
Owner common.Hash // owner of the trie if it's 2-layered trie
NodeHash common.Hash // hash of the missing node
Path []byte // hex-encoded path to the missing node
err error // concrete error for missing trie node
}
// Unwrap returns the concrete error for missing trie node which
// allows us for further analysis outside.
func (err *MissingNodeError) Unwrap() error {
return err.err
}
func (err *MissingNodeError) Error() string {
if err.Owner == (common.Hash{}) {
return fmt.Sprintf("missing trie node %x (path %x) %v", err.NodeHash, err.Path, err.err)
}
return fmt.Sprintf("missing trie node %x (owner %x) (path %x) %v", err.NodeHash, err.Owner, err.Path, err.err)
}

208
trie/hasher.go
View file

@ -1,208 +0,0 @@
// Copyright 2016 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 trie
import (
"sync"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/rlp"
"golang.org/x/crypto/sha3"
)
// hasher is a type used for the trie Hash operation. A hasher has some
// internal preallocated temp space
type hasher struct {
sha crypto.KeccakState
tmp []byte
encbuf rlp.EncoderBuffer
parallel bool // Whether to use parallel threads when hashing
}
// hasherPool holds pureHashers
var hasherPool = sync.Pool{
New: func() interface{} {
return &hasher{
tmp: make([]byte, 0, 550), // cap is as large as a full fullNode.
sha: sha3.NewLegacyKeccak256().(crypto.KeccakState),
encbuf: rlp.NewEncoderBuffer(nil),
}
},
}
func newHasher(parallel bool) *hasher {
h := hasherPool.Get().(*hasher)
h.parallel = parallel
return h
}
func returnHasherToPool(h *hasher) {
hasherPool.Put(h)
}
// hash collapses a node down into a hash node, also returning a copy of the
// original node initialized with the computed hash to replace the original one.
func (h *hasher) hash(n node, force bool) (hashed node, cached node) {
// Return the cached hash if it's available
if hash, _ := n.cache(); hash != nil {
return hash, n
}
// Trie not processed yet, walk the children
switch n := n.(type) {
case *shortNode:
collapsed, cached := h.hashShortNodeChildren(n)
hashed := h.shortnodeToHash(collapsed, force)
// We need to retain the possibly _not_ hashed node, in case it was too
// small to be hashed
if hn, ok := hashed.(hashNode); ok {
cached.flags.hash = hn
} else {
cached.flags.hash = nil
}
return hashed, cached
case *fullNode:
collapsed, cached := h.hashFullNodeChildren(n)
hashed = h.fullnodeToHash(collapsed, force)
if hn, ok := hashed.(hashNode); ok {
cached.flags.hash = hn
} else {
cached.flags.hash = nil
}
return hashed, cached
default:
// Value and hash nodes don't have children, so they're left as were
return n, n
}
}
// hashShortNodeChildren collapses the short node. The returned collapsed node
// holds a live reference to the Key, and must not be modified.
func (h *hasher) hashShortNodeChildren(n *shortNode) (collapsed, cached *shortNode) {
// Hash the short node's child, caching the newly hashed subtree
collapsed, cached = n.copy(), n.copy()
// Previously, we did copy this one. We don't seem to need to actually
// do that, since we don't overwrite/reuse keys
// cached.Key = common.CopyBytes(n.Key)
collapsed.Key = hexToCompact(n.Key)
// Unless the child is a valuenode or hashnode, hash it
switch n.Val.(type) {
case *fullNode, *shortNode:
collapsed.Val, cached.Val = h.hash(n.Val, false)
}
return collapsed, cached
}
func (h *hasher) hashFullNodeChildren(n *fullNode) (collapsed *fullNode, cached *fullNode) {
// Hash the full node's children, caching the newly hashed subtrees
cached = n.copy()
collapsed = n.copy()
if h.parallel {
var wg sync.WaitGroup
wg.Add(16)
for i := 0; i < 16; i++ {
go func(i int) {
hasher := newHasher(false)
if child := n.Children[i]; child != nil {
collapsed.Children[i], cached.Children[i] = hasher.hash(child, false)
} else {
collapsed.Children[i] = nilValueNode
}
returnHasherToPool(hasher)
wg.Done()
}(i)
}
wg.Wait()
} else {
for i := 0; i < 16; i++ {
if child := n.Children[i]; child != nil {
collapsed.Children[i], cached.Children[i] = h.hash(child, false)
} else {
collapsed.Children[i] = nilValueNode
}
}
}
return collapsed, cached
}
// shortnodeToHash creates a hashNode from a shortNode. The supplied shortnode
// should have hex-type Key, which will be converted (without modification)
// into compact form for RLP encoding.
// If the rlp data is smaller than 32 bytes, `nil` is returned.
func (h *hasher) shortnodeToHash(n *shortNode, force bool) node {
n.encode(h.encbuf)
enc := h.encodedBytes()
if len(enc) < 32 && !force {
return n // Nodes smaller than 32 bytes are stored inside their parent
}
return h.hashData(enc)
}
// fullnodeToHash is used to create a hashNode from a fullNode, (which
// may contain nil values)
func (h *hasher) fullnodeToHash(n *fullNode, force bool) node {
n.encode(h.encbuf)
enc := h.encodedBytes()
if len(enc) < 32 && !force {
return n // Nodes smaller than 32 bytes are stored inside their parent
}
return h.hashData(enc)
}
// encodedBytes returns the result of the last encoding operation on h.encbuf.
// This also resets the encoder buffer.
//
// All node encoding must be done like this:
//
// node.encode(h.encbuf)
// enc := h.encodedBytes()
//
// This convention exists because node.encode can only be inlined/escape-analyzed when
// called on a concrete receiver type.
func (h *hasher) encodedBytes() []byte {
h.tmp = h.encbuf.AppendToBytes(h.tmp[:0])
h.encbuf.Reset(nil)
return h.tmp
}
// hashData hashes the provided data
func (h *hasher) hashData(data []byte) hashNode {
n := make(hashNode, 32)
h.sha.Reset()
h.sha.Write(data)
h.sha.Read(n)
return n
}
// proofHash is used to construct trie proofs, and returns the 'collapsed'
// node (for later RLP encoding) as well as the hashed node -- unless the
// node is smaller than 32 bytes, in which case it will be returned as is.
// This method does not do anything on value- or hash-nodes.
func (h *hasher) proofHash(original node) (collapsed, hashed node) {
switch n := original.(type) {
case *shortNode:
sn, _ := h.hashShortNodeChildren(n)
return sn, h.shortnodeToHash(sn, false)
case *fullNode:
fn, _ := h.hashFullNodeChildren(n)
return fn, h.fullnodeToHash(fn, false)
default:
// Value and hash nodes don't have children, so they're left as were
return n, n
}
}

791
trie/iterator.go
View file

@ -1,791 +0,0 @@
// Copyright 2014 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 trie
import (
"bytes"
"container/heap"
"errors"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/types"
)
// NodeResolver is used for looking up trie nodes before reaching into the real
// persistent layer. This is not mandatory, rather is an optimization for cases
// where trie nodes can be recovered from some external mechanism without reading
// from disk. In those cases, this resolver allows short circuiting accesses and
// returning them from memory.
type NodeResolver func(owner common.Hash, path []byte, hash common.Hash) []byte
// Iterator is a key-value trie iterator that traverses a Trie.
type Iterator struct {
nodeIt NodeIterator
Key []byte // Current data key on which the iterator is positioned on
Value []byte // Current data value on which the iterator is positioned on
Err error
}
// NewIterator creates a new key-value iterator from a node iterator.
// Note that the value returned by the iterator is raw. If the content is encoded
// (e.g. storage value is RLP-encoded), it's caller's duty to decode it.
func NewIterator(it NodeIterator) *Iterator {
return &Iterator{
nodeIt: it,
}
}
// Next moves the iterator forward one key-value entry.
func (it *Iterator) Next() bool {
for it.nodeIt.Next(true) {
if it.nodeIt.Leaf() {
it.Key = it.nodeIt.LeafKey()
it.Value = it.nodeIt.LeafBlob()
return true
}
}
it.Key = nil
it.Value = nil
it.Err = it.nodeIt.Error()
return false
}
// Prove generates the Merkle proof for the leaf node the iterator is currently
// positioned on.
func (it *Iterator) Prove() [][]byte {
return it.nodeIt.LeafProof()
}
// NodeIterator is an iterator to traverse the trie pre-order.
type NodeIterator interface {
// Next moves the iterator to the next node. If the parameter is false, any child
// nodes will be skipped.
Next(bool) bool
// Error returns the error status of the iterator.
Error() error
// Hash returns the hash of the current node.
Hash() common.Hash
// Parent returns the hash of the parent of the current node. The hash may be the one
// grandparent if the immediate parent is an internal node with no hash.
Parent() common.Hash
// Path returns the hex-encoded path to the current node.
// Callers must not retain references to the return value after calling Next.
// For leaf nodes, the last element of the path is the 'terminator symbol' 0x10.
Path() []byte
// NodeBlob returns the rlp-encoded value of the current iterated node.
// If the node is an embedded node in its parent, nil is returned then.
NodeBlob() []byte
// Leaf returns true iff the current node is a leaf node.
Leaf() bool
// LeafKey returns the key of the leaf. The method panics if the iterator is not
// positioned at a leaf. Callers must not retain references to the value after
// calling Next.
LeafKey() []byte
// LeafBlob returns the content of the leaf. The method panics if the iterator
// is not positioned at a leaf. Callers must not retain references to the value
// after calling Next.
LeafBlob() []byte
// LeafProof returns the Merkle proof of the leaf. The method panics if the
// iterator is not positioned at a leaf. Callers must not retain references
// to the value after calling Next.
LeafProof() [][]byte
// AddResolver sets a node resolver to use for looking up trie nodes before
// reaching into the real persistent layer.
//
// This is not required for normal operation, rather is an optimization for
// cases where trie nodes can be recovered from some external mechanism without
// reading from disk. In those cases, this resolver allows short circuiting
// accesses and returning them from memory.
//
// Before adding a similar mechanism to any other place in Geth, consider
// making trie.Database an interface and wrapping at that level. It's a huge
// refactor, but it could be worth it if another occurrence arises.
AddResolver(NodeResolver)
}
// nodeIteratorState represents the iteration state at one particular node of the
// trie, which can be resumed at a later invocation.
type nodeIteratorState struct {
hash common.Hash // Hash of the node being iterated (nil if not standalone)
node node // Trie node being iterated
parent common.Hash // Hash of the first full ancestor node (nil if current is the root)
index int // Child to be processed next
pathlen int // Length of the path to this node
}
type nodeIterator struct {
trie *Trie // Trie being iterated
stack []*nodeIteratorState // Hierarchy of trie nodes persisting the iteration state
path []byte // Path to the current node
err error // Failure set in case of an internal error in the iterator
resolver NodeResolver // optional node resolver for avoiding disk hits
pool []*nodeIteratorState // local pool for iteratorstates
}
// errIteratorEnd is stored in nodeIterator.err when iteration is done.
var errIteratorEnd = errors.New("end of iteration")
// seekError is stored in nodeIterator.err if the initial seek has failed.
type seekError struct {
key []byte
err error
}
func (e seekError) Error() string {
return "seek error: " + e.err.Error()
}
func newNodeIterator(trie *Trie, start []byte) NodeIterator {
if trie.Hash() == types.EmptyRootHash {
return &nodeIterator{
trie: trie,
err: errIteratorEnd,
}
}
it := &nodeIterator{trie: trie}
it.err = it.seek(start)
return it
}
func (it *nodeIterator) putInPool(item *nodeIteratorState) {
if len(it.pool) < 40 {
item.node = nil
it.pool = append(it.pool, item)
}
}
func (it *nodeIterator) getFromPool() *nodeIteratorState {
idx := len(it.pool) - 1
if idx < 0 {
return new(nodeIteratorState)
}
el := it.pool[idx]
it.pool[idx] = nil
it.pool = it.pool[:idx]
return el
}
func (it *nodeIterator) AddResolver(resolver NodeResolver) {
it.resolver = resolver
}
func (it *nodeIterator) Hash() common.Hash {
if len(it.stack) == 0 {
return common.Hash{}
}
return it.stack[len(it.stack)-1].hash
}
func (it *nodeIterator) Parent() common.Hash {
if len(it.stack) == 0 {
return common.Hash{}
}
return it.stack[len(it.stack)-1].parent
}
func (it *nodeIterator) Leaf() bool {
return hasTerm(it.path)
}
func (it *nodeIterator) LeafKey() []byte {
if len(it.stack) > 0 {
if _, ok := it.stack[len(it.stack)-1].node.(valueNode); ok {
return hexToKeybytes(it.path)
}
}
panic("not at leaf")
}
func (it *nodeIterator) LeafBlob() []byte {
if len(it.stack) > 0 {
if node, ok := it.stack[len(it.stack)-1].node.(valueNode); ok {
return node
}
}
panic("not at leaf")
}
func (it *nodeIterator) LeafProof() [][]byte {
if len(it.stack) > 0 {
if _, ok := it.stack[len(it.stack)-1].node.(valueNode); ok {
hasher := newHasher(false)
defer returnHasherToPool(hasher)
proofs := make([][]byte, 0, len(it.stack))
for i, item := range it.stack[:len(it.stack)-1] {
// Gather nodes that end up as hash nodes (or the root)
node, hashed := hasher.proofHash(item.node)
if _, ok := hashed.(hashNode); ok || i == 0 {
proofs = append(proofs, nodeToBytes(node))
}
}
return proofs
}
}
panic("not at leaf")
}
func (it *nodeIterator) Path() []byte {
return it.path
}
func (it *nodeIterator) NodeBlob() []byte {
if it.Hash() == (common.Hash{}) {
return nil // skip the non-standalone node
}
blob, err := it.resolveBlob(it.Hash().Bytes(), it.Path())
if err != nil {
it.err = err
return nil
}
return blob
}
func (it *nodeIterator) Error() error {
if it.err == errIteratorEnd {
return nil
}
if seek, ok := it.err.(seekError); ok {
return seek.err
}
return it.err
}
// Next moves the iterator to the next node, returning whether there are any
// further nodes. In case of an internal error this method returns false and
// sets the Error field to the encountered failure. If `descend` is false,
// skips iterating over any subnodes of the current node.
func (it *nodeIterator) Next(descend bool) bool {
if it.err == errIteratorEnd {
return false
}
if seek, ok := it.err.(seekError); ok {
if it.err = it.seek(seek.key); it.err != nil {
return false
}
}
// Otherwise step forward with the iterator and report any errors.
state, parentIndex, path, err := it.peek(descend)
it.err = err
if it.err != nil {
return false
}
it.push(state, parentIndex, path)
return true
}
func (it *nodeIterator) seek(prefix []byte) error {
// The path we're looking for is the hex encoded key without terminator.
key := keybytesToHex(prefix)
key = key[:len(key)-1]
// Move forward until we're just before the closest match to key.
for {
state, parentIndex, path, err := it.peekSeek(key)
if err == errIteratorEnd {
return errIteratorEnd
} else if err != nil {
return seekError{prefix, err}
} else if bytes.Compare(path, key) >= 0 {
return nil
}
it.push(state, parentIndex, path)
}
}
// init initializes the iterator.
func (it *nodeIterator) init() (*nodeIteratorState, error) {
root := it.trie.Hash()
state := &nodeIteratorState{node: it.trie.root, index: -1}
if root != types.EmptyRootHash {
state.hash = root
}
return state, state.resolve(it, nil)
}
// peek creates the next state of the iterator.
func (it *nodeIterator) peek(descend bool) (*nodeIteratorState, *int, []byte, error) {
// Initialize the iterator if we've just started.
if len(it.stack) == 0 {
state, err := it.init()
return state, nil, nil, err
}
if !descend {
// If we're skipping children, pop the current node first
it.pop()
}
// Continue iteration to the next child
for len(it.stack) > 0 {
parent := it.stack[len(it.stack)-1]
ancestor := parent.hash
if (ancestor == common.Hash{}) {
ancestor = parent.parent
}
state, path, ok := it.nextChild(parent, ancestor)
if ok {
if err := state.resolve(it, path); err != nil {
return parent, &parent.index, path, err
}
return state, &parent.index, path, nil
}
// No more child nodes, move back up.
it.pop()
}
return nil, nil, nil, errIteratorEnd
}
// peekSeek is like peek, but it also tries to skip resolving hashes by skipping
// over the siblings that do not lead towards the desired seek position.
func (it *nodeIterator) peekSeek(seekKey []byte) (*nodeIteratorState, *int, []byte, error) {
// Initialize the iterator if we've just started.
if len(it.stack) == 0 {
state, err := it.init()
return state, nil, nil, err
}
if !bytes.HasPrefix(seekKey, it.path) {
// If we're skipping children, pop the current node first
it.pop()
}
// Continue iteration to the next child
for len(it.stack) > 0 {
parent := it.stack[len(it.stack)-1]
ancestor := parent.hash
if (ancestor == common.Hash{}) {
ancestor = parent.parent
}
state, path, ok := it.nextChildAt(parent, ancestor, seekKey)
if ok {
if err := state.resolve(it, path); err != nil {
return parent, &parent.index, path, err
}
return state, &parent.index, path, nil
}
// No more child nodes, move back up.
it.pop()
}
return nil, nil, nil, errIteratorEnd
}
func (it *nodeIterator) resolveHash(hash hashNode, path []byte) (node, error) {
if it.resolver != nil {
if blob := it.resolver(it.trie.owner, path, common.BytesToHash(hash)); len(blob) > 0 {
if resolved, err := decodeNode(hash, blob); err == nil {
return resolved, nil
}
}
}
// Retrieve the specified node from the underlying node reader.
// it.trie.resolveAndTrack is not used since in that function the
// loaded blob will be tracked, while it's not required here since
// all loaded nodes won't be linked to trie at all and track nodes
// may lead to out-of-memory issue.
blob, err := it.trie.reader.node(path, common.BytesToHash(hash))
if err != nil {
return nil, err
}
// The raw-blob format nodes are loaded either from the
// clean cache or the database, they are all in their own
// copy and safe to use unsafe decoder.
return mustDecodeNodeUnsafe(hash, blob), nil
}
func (it *nodeIterator) resolveBlob(hash hashNode, path []byte) ([]byte, error) {
if it.resolver != nil {
if blob := it.resolver(it.trie.owner, path, common.BytesToHash(hash)); len(blob) > 0 {
return blob, nil
}
}
// Retrieve the specified node from the underlying node reader.
// it.trie.resolveAndTrack is not used since in that function the
// loaded blob will be tracked, while it's not required here since
// all loaded nodes won't be linked to trie at all and track nodes
// may lead to out-of-memory issue.
return it.trie.reader.node(path, common.BytesToHash(hash))
}
func (st *nodeIteratorState) resolve(it *nodeIterator, path []byte) error {
if hash, ok := st.node.(hashNode); ok {
resolved, err := it.resolveHash(hash, path)
if err != nil {
return err
}
st.node = resolved
st.hash = common.BytesToHash(hash)
}
return nil
}
func (it *nodeIterator) findChild(n *fullNode, index int, ancestor common.Hash) (node, *nodeIteratorState, []byte, int) {
var (
path = it.path
child node
state *nodeIteratorState
childPath []byte
)
for ; index < len(n.Children); index++ {
if n.Children[index] != nil {
child = n.Children[index]
hash, _ := child.cache()
state = it.getFromPool()
state.hash = common.BytesToHash(hash)
state.node = child
state.parent = ancestor
state.index = -1
state.pathlen = len(path)
childPath = append(childPath, path...)
childPath = append(childPath, byte(index))
return child, state, childPath, index
}
}
return nil, nil, nil, 0
}
func (it *nodeIterator) nextChild(parent *nodeIteratorState, ancestor common.Hash) (*nodeIteratorState, []byte, bool) {
switch node := parent.node.(type) {
case *fullNode:
// Full node, move to the first non-nil child.
if child, state, path, index := it.findChild(node, parent.index+1, ancestor); child != nil {
parent.index = index - 1
return state, path, true
}
case *shortNode:
// Short node, return the pointer singleton child
if parent.index < 0 {
hash, _ := node.Val.cache()
state := it.getFromPool()
state.hash = common.BytesToHash(hash)
state.node = node.Val
state.parent = ancestor
state.index = -1
state.pathlen = len(it.path)
path := append(it.path, node.Key...)
return state, path, true
}
}
return parent, it.path, false
}
// nextChildAt is similar to nextChild, except that it targets a child as close to the
// target key as possible, thus skipping siblings.
func (it *nodeIterator) nextChildAt(parent *nodeIteratorState, ancestor common.Hash, key []byte) (*nodeIteratorState, []byte, bool) {
switch n := parent.node.(type) {
case *fullNode:
// Full node, move to the first non-nil child before the desired key position
child, state, path, index := it.findChild(n, parent.index+1, ancestor)
if child == nil {
// No more children in this fullnode
return parent, it.path, false
}
// If the child we found is already past the seek position, just return it.
if bytes.Compare(path, key) >= 0 {
parent.index = index - 1
return state, path, true
}
// The child is before the seek position. Try advancing
for {
nextChild, nextState, nextPath, nextIndex := it.findChild(n, index+1, ancestor)
// If we run out of children, or skipped past the target, return the
// previous one
if nextChild == nil || bytes.Compare(nextPath, key) >= 0 {
parent.index = index - 1
return state, path, true
}
// We found a better child closer to the target
state, path, index = nextState, nextPath, nextIndex
}
case *shortNode:
// Short node, return the pointer singleton child
if parent.index < 0 {
hash, _ := n.Val.cache()
state := it.getFromPool()
state.hash = common.BytesToHash(hash)
state.node = n.Val
state.parent = ancestor
state.index = -1
state.pathlen = len(it.path)
path := append(it.path, n.Key...)
return state, path, true
}
}
return parent, it.path, false
}
func (it *nodeIterator) push(state *nodeIteratorState, parentIndex *int, path []byte) {
it.path = path
it.stack = append(it.stack, state)
if parentIndex != nil {
*parentIndex++
}
}
func (it *nodeIterator) pop() {
last := it.stack[len(it.stack)-1]
it.path = it.path[:last.pathlen]
it.stack[len(it.stack)-1] = nil
it.stack = it.stack[:len(it.stack)-1]
// last is now unused
it.putInPool(last)
}
func compareNodes(a, b NodeIterator) int {
if cmp := bytes.Compare(a.Path(), b.Path()); cmp != 0 {
return cmp
}
if a.Leaf() && !b.Leaf() {
return -1
} else if b.Leaf() && !a.Leaf() {
return 1
}
if cmp := bytes.Compare(a.Hash().Bytes(), b.Hash().Bytes()); cmp != 0 {
return cmp
}
if a.Leaf() && b.Leaf() {
return bytes.Compare(a.LeafBlob(), b.LeafBlob())
}
return 0
}
type differenceIterator struct {
a, b NodeIterator // Nodes returned are those in b - a.
eof bool // Indicates a has run out of elements
count int // Number of nodes scanned on either trie
}
// NewDifferenceIterator constructs a NodeIterator that iterates over elements in b that
// are not in a. Returns the iterator, and a pointer to an integer recording the number
// of nodes seen.
func NewDifferenceIterator(a, b NodeIterator) (NodeIterator, *int) {
a.Next(true)
it := &differenceIterator{
a: a,
b: b,
}
return it, &it.count
}
func (it *differenceIterator) Hash() common.Hash {
return it.b.Hash()
}
func (it *differenceIterator) Parent() common.Hash {
return it.b.Parent()
}
func (it *differenceIterator) Leaf() bool {
return it.b.Leaf()
}
func (it *differenceIterator) LeafKey() []byte {
return it.b.LeafKey()
}
func (it *differenceIterator) LeafBlob() []byte {
return it.b.LeafBlob()
}
func (it *differenceIterator) LeafProof() [][]byte {
return it.b.LeafProof()
}
func (it *differenceIterator) Path() []byte {
return it.b.Path()
}
func (it *differenceIterator) NodeBlob() []byte {
return it.b.NodeBlob()
}
func (it *differenceIterator) AddResolver(resolver NodeResolver) {
panic("not implemented")
}
func (it *differenceIterator) Next(bool) bool {
// Invariants:
// - We always advance at least one element in b.
// - At the start of this function, a's path is lexically greater than b's.
if !it.b.Next(true) {
return false
}
it.count++
if it.eof {
// a has reached eof, so we just return all elements from b
return true
}
for {
switch compareNodes(it.a, it.b) {
case -1:
// b jumped past a; advance a
if !it.a.Next(true) {
it.eof = true
return true
}
it.count++
case 1:
// b is before a
return true
case 0:
// a and b are identical; skip this whole subtree if the nodes have hashes
hasHash := it.a.Hash() == common.Hash{}
if !it.b.Next(hasHash) {
return false
}
it.count++
if !it.a.Next(hasHash) {
it.eof = true
return true
}
it.count++
}
}
}
func (it *differenceIterator) Error() error {
if err := it.a.Error(); err != nil {
return err
}
return it.b.Error()
}
type nodeIteratorHeap []NodeIterator
func (h nodeIteratorHeap) Len() int { return len(h) }
func (h nodeIteratorHeap) Less(i, j int) bool { return compareNodes(h[i], h[j]) < 0 }
func (h nodeIteratorHeap) Swap(i, j int) { h[i], h[j] = h[j], h[i] }
func (h *nodeIteratorHeap) Push(x interface{}) { *h = append(*h, x.(NodeIterator)) }
func (h *nodeIteratorHeap) Pop() interface{} {
n := len(*h)
x := (*h)[n-1]
*h = (*h)[0 : n-1]
return x
}
type unionIterator struct {
items *nodeIteratorHeap // Nodes returned are the union of the ones in these iterators
count int // Number of nodes scanned across all tries
}
// NewUnionIterator constructs a NodeIterator that iterates over elements in the union
// of the provided NodeIterators. Returns the iterator, and a pointer to an integer
// recording the number of nodes visited.
func NewUnionIterator(iters []NodeIterator) (NodeIterator, *int) {
h := make(nodeIteratorHeap, len(iters))
copy(h, iters)
heap.Init(&h)
ui := &unionIterator{items: &h}
return ui, &ui.count
}
func (it *unionIterator) Hash() common.Hash {
return (*it.items)[0].Hash()
}
func (it *unionIterator) Parent() common.Hash {
return (*it.items)[0].Parent()
}
func (it *unionIterator) Leaf() bool {
return (*it.items)[0].Leaf()
}
func (it *unionIterator) LeafKey() []byte {
return (*it.items)[0].LeafKey()
}
func (it *unionIterator) LeafBlob() []byte {
return (*it.items)[0].LeafBlob()
}
func (it *unionIterator) LeafProof() [][]byte {
return (*it.items)[0].LeafProof()
}
func (it *unionIterator) Path() []byte {
return (*it.items)[0].Path()
}
func (it *unionIterator) NodeBlob() []byte {
return (*it.items)[0].NodeBlob()
}
func (it *unionIterator) AddResolver(resolver NodeResolver) {
panic("not implemented")
}
// Next returns the next node in the union of tries being iterated over.
//
// It does this by maintaining a heap of iterators, sorted by the iteration
// order of their next elements, with one entry for each source trie. Each
// time Next() is called, it takes the least element from the heap to return,
// advancing any other iterators that also point to that same element. These
// iterators are called with descend=false, since we know that any nodes under
// these nodes will also be duplicates, found in the currently selected iterator.
// Whenever an iterator is advanced, it is pushed back into the heap if it still
// has elements remaining.
//
// In the case that descend=false - eg, we're asked to ignore all subnodes of the
// current node - we also advance any iterators in the heap that have the current
// path as a prefix.
func (it *unionIterator) Next(descend bool) bool {
if len(*it.items) == 0 {
return false
}
// Get the next key from the union
least := heap.Pop(it.items).(NodeIterator)
// Skip over other nodes as long as they're identical, or, if we're not descending, as
// long as they have the same prefix as the current node.
for len(*it.items) > 0 && ((!descend && bytes.HasPrefix((*it.items)[0].Path(), least.Path())) || compareNodes(least, (*it.items)[0]) == 0) {
skipped := heap.Pop(it.items).(NodeIterator)
// Skip the whole subtree if the nodes have hashes; otherwise just skip this node
if skipped.Next(skipped.Hash() == common.Hash{}) {
it.count++
// If there are more elements, push the iterator back on the heap
heap.Push(it.items, skipped)
}
}
if least.Next(descend) {
it.count++
heap.Push(it.items, least)
}
return len(*it.items) > 0
}
func (it *unionIterator) Error() error {
for i := 0; i < len(*it.items); i++ {
if err := (*it.items)[i].Error(); err != nil {
return err
}
}
return nil
}

630
trie/iterator_test.go
View file

@ -1,630 +0,0 @@
// Copyright 2014 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 trie
import (
"bytes"
"fmt"
"math/rand"
"testing"
"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/trie/trienode"
)
func TestEmptyIterator(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
iter := trie.MustNodeIterator(nil)
seen := make(map[string]struct{})
for iter.Next(true) {
seen[string(iter.Path())] = struct{}{}
}
if len(seen) != 0 {
t.Fatal("Unexpected trie node iterated")
}
}
func TestIterator(t *testing.T) {
db := NewDatabase(rawdb.NewMemoryDatabase(), nil)
trie := NewEmpty(db)
vals := []struct{ k, v string }{
{"do", "verb"},
{"ether", "wookiedoo"},
{"horse", "stallion"},
{"shaman", "horse"},
{"doge", "coin"},
{"dog", "puppy"},
{"somethingveryoddindeedthis is", "myothernodedata"},
}
all := make(map[string]string)
for _, val := range vals {
all[val.k] = val.v
trie.MustUpdate([]byte(val.k), []byte(val.v))
}
root, nodes, _ := trie.Commit(false)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
trie, _ = New(TrieID(root), db)
found := make(map[string]string)
it := NewIterator(trie.MustNodeIterator(nil))
for it.Next() {
found[string(it.Key)] = string(it.Value)
}
for k, v := range all {
if found[k] != v {
t.Errorf("iterator value mismatch for %s: got %q want %q", k, found[k], v)
}
}
}
type kv struct {
k, v []byte
t bool
}
func (k *kv) cmp(other *kv) int {
return bytes.Compare(k.k, other.k)
}
func TestIteratorLargeData(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
vals := make(map[string]*kv)
for i := byte(0); i < 255; i++ {
value := &kv{common.LeftPadBytes([]byte{i}, 32), []byte{i}, false}
value2 := &kv{common.LeftPadBytes([]byte{10, i}, 32), []byte{i}, false}
trie.MustUpdate(value.k, value.v)
trie.MustUpdate(value2.k, value2.v)
vals[string(value.k)] = value
vals[string(value2.k)] = value2
}
it := NewIterator(trie.MustNodeIterator(nil))
for it.Next() {
vals[string(it.Key)].t = true
}
var untouched []*kv
for _, value := range vals {
if !value.t {
untouched = append(untouched, value)
}
}
if len(untouched) > 0 {
t.Errorf("Missed %d nodes", len(untouched))
for _, value := range untouched {
t.Error(value)
}
}
}
type iterationElement struct {
hash common.Hash
path []byte
blob []byte
}
// Tests that the node iterator indeed walks over the entire database contents.
func TestNodeIteratorCoverage(t *testing.T) {
testNodeIteratorCoverage(t, rawdb.HashScheme)
testNodeIteratorCoverage(t, rawdb.PathScheme)
}
func testNodeIteratorCoverage(t *testing.T, scheme string) {
// Create some arbitrary test trie to iterate
db, nodeDb, trie, _ := makeTestTrie(scheme)
// Gather all the node hashes found by the iterator
var elements = make(map[common.Hash]iterationElement)
for it := trie.MustNodeIterator(nil); it.Next(true); {
if it.Hash() != (common.Hash{}) {
elements[it.Hash()] = iterationElement{
hash: it.Hash(),
path: common.CopyBytes(it.Path()),
blob: common.CopyBytes(it.NodeBlob()),
}
}
}
// Cross check the hashes and the database itself
reader, err := nodeDb.Reader(trie.Hash())
if err != nil {
t.Fatalf("state is not available %x", trie.Hash())
}
for _, element := range elements {
if blob, err := reader.Node(common.Hash{}, element.path, element.hash); err != nil {
t.Errorf("failed to retrieve reported node %x: %v", element.hash, err)
} else if !bytes.Equal(blob, element.blob) {
t.Errorf("node blob is different, want %v got %v", element.blob, blob)
}
}
var (
count int
it = db.NewIterator(nil, nil)
)
for it.Next() {
res, _, _ := isTrieNode(nodeDb.Scheme(), it.Key(), it.Value())
if !res {
continue
}
count += 1
if elem, ok := elements[crypto.Keccak256Hash(it.Value())]; !ok {
t.Error("state entry not reported")
} else if !bytes.Equal(it.Value(), elem.blob) {
t.Errorf("node blob is different, want %v got %v", elem.blob, it.Value())
}
}
it.Release()
if count != len(elements) {
t.Errorf("state entry is mismatched %d %d", count, len(elements))
}
}
type kvs struct{ k, v string }
var testdata1 = []kvs{
{"barb", "ba"},
{"bard", "bc"},
{"bars", "bb"},
{"bar", "b"},
{"fab", "z"},
{"food", "ab"},
{"foos", "aa"},
{"foo", "a"},
}
var testdata2 = []kvs{
{"aardvark", "c"},
{"bar", "b"},
{"barb", "bd"},
{"bars", "be"},
{"fab", "z"},
{"foo", "a"},
{"foos", "aa"},
{"food", "ab"},
{"jars", "d"},
}
func TestIteratorSeek(t *testing.T) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
for _, val := range testdata1 {
trie.MustUpdate([]byte(val.k), []byte(val.v))
}
// Seek to the middle.
it := NewIterator(trie.MustNodeIterator([]byte("fab")))
if err := checkIteratorOrder(testdata1[4:], it); err != nil {
t.Fatal(err)
}
// Seek to a non-existent key.
it = NewIterator(trie.MustNodeIterator([]byte("barc")))
if err := checkIteratorOrder(testdata1[1:], it); err != nil {
t.Fatal(err)
}
// Seek beyond the end.
it = NewIterator(trie.MustNodeIterator([]byte("z")))
if err := checkIteratorOrder(nil, it); err != nil {
t.Fatal(err)
}
}
func checkIteratorOrder(want []kvs, it *Iterator) error {
for it.Next() {
if len(want) == 0 {
return fmt.Errorf("didn't expect any more values, got key %q", it.Key)
}
if !bytes.Equal(it.Key, []byte(want[0].k)) {
return fmt.Errorf("wrong key: got %q, want %q", it.Key, want[0].k)
}
want = want[1:]
}
if len(want) > 0 {
return fmt.Errorf("iterator ended early, want key %q", want[0])
}
return nil
}
func TestDifferenceIterator(t *testing.T) {
dba := NewDatabase(rawdb.NewMemoryDatabase(), nil)
triea := NewEmpty(dba)
for _, val := range testdata1 {
triea.MustUpdate([]byte(val.k), []byte(val.v))
}
rootA, nodesA, _ := triea.Commit(false)
dba.Update(rootA, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodesA), nil)
triea, _ = New(TrieID(rootA), dba)
dbb := NewDatabase(rawdb.NewMemoryDatabase(), nil)
trieb := NewEmpty(dbb)
for _, val := range testdata2 {
trieb.MustUpdate([]byte(val.k), []byte(val.v))
}
rootB, nodesB, _ := trieb.Commit(false)
dbb.Update(rootB, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodesB), nil)
trieb, _ = New(TrieID(rootB), dbb)
found := make(map[string]string)
di, _ := NewDifferenceIterator(triea.MustNodeIterator(nil), trieb.MustNodeIterator(nil))
it := NewIterator(di)
for it.Next() {
found[string(it.Key)] = string(it.Value)
}
all := []struct{ k, v string }{
{"aardvark", "c"},
{"barb", "bd"},
{"bars", "be"},
{"jars", "d"},
}
for _, item := range all {
if found[item.k] != item.v {
t.Errorf("iterator value mismatch for %s: got %v want %v", item.k, found[item.k], item.v)
}
}
if len(found) != len(all) {
t.Errorf("iterator count mismatch: got %d values, want %d", len(found), len(all))
}
}
func TestUnionIterator(t *testing.T) {
dba := NewDatabase(rawdb.NewMemoryDatabase(), nil)
triea := NewEmpty(dba)
for _, val := range testdata1 {
triea.MustUpdate([]byte(val.k), []byte(val.v))
}
rootA, nodesA, _ := triea.Commit(false)
dba.Update(rootA, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodesA), nil)
triea, _ = New(TrieID(rootA), dba)
dbb := NewDatabase(rawdb.NewMemoryDatabase(), nil)
trieb := NewEmpty(dbb)
for _, val := range testdata2 {
trieb.MustUpdate([]byte(val.k), []byte(val.v))
}
rootB, nodesB, _ := trieb.Commit(false)
dbb.Update(rootB, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodesB), nil)
trieb, _ = New(TrieID(rootB), dbb)
di, _ := NewUnionIterator([]NodeIterator{triea.MustNodeIterator(nil), trieb.MustNodeIterator(nil)})
it := NewIterator(di)
all := []struct{ k, v string }{
{"aardvark", "c"},
{"barb", "ba"},
{"barb", "bd"},
{"bard", "bc"},
{"bars", "bb"},
{"bars", "be"},
{"bar", "b"},
{"fab", "z"},
{"food", "ab"},
{"foos", "aa"},
{"foo", "a"},
{"jars", "d"},
}
for i, kv := range all {
if !it.Next() {
t.Errorf("Iterator ends prematurely at element %d", i)
}
if kv.k != string(it.Key) {
t.Errorf("iterator value mismatch for element %d: got key %s want %s", i, it.Key, kv.k)
}
if kv.v != string(it.Value) {
t.Errorf("iterator value mismatch for element %d: got value %s want %s", i, it.Value, kv.v)
}
}
if it.Next() {
t.Errorf("Iterator returned extra values.")
}
}
func TestIteratorNoDups(t *testing.T) {
tr := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
for _, val := range testdata1 {
tr.MustUpdate([]byte(val.k), []byte(val.v))
}
checkIteratorNoDups(t, tr.MustNodeIterator(nil), nil)
}
// This test checks that nodeIterator.Next can be retried after inserting missing trie nodes.
func TestIteratorContinueAfterError(t *testing.T) {
testIteratorContinueAfterError(t, false, rawdb.HashScheme)
testIteratorContinueAfterError(t, true, rawdb.HashScheme)
testIteratorContinueAfterError(t, false, rawdb.PathScheme)
testIteratorContinueAfterError(t, true, rawdb.PathScheme)
}
func testIteratorContinueAfterError(t *testing.T, memonly bool, scheme string) {
diskdb := rawdb.NewMemoryDatabase()
tdb := newTestDatabase(diskdb, scheme)
tr := NewEmpty(tdb)
for _, val := range testdata1 {
tr.MustUpdate([]byte(val.k), []byte(val.v))
}
root, nodes, _ := tr.Commit(false)
tdb.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
if !memonly {
tdb.Commit(root, false)
}
tr, _ = New(TrieID(root), tdb)
wantNodeCount := checkIteratorNoDups(t, tr.MustNodeIterator(nil), nil)
var (
paths [][]byte
hashes []common.Hash
)
if memonly {
for path, n := range nodes.Nodes {
paths = append(paths, []byte(path))
hashes = append(hashes, n.Hash)
}
} else {
it := diskdb.NewIterator(nil, nil)
for it.Next() {
ok, path, hash := isTrieNode(tdb.Scheme(), it.Key(), it.Value())
if !ok {
continue
}
paths = append(paths, path)
hashes = append(hashes, hash)
}
it.Release()
}
for i := 0; i < 20; i++ {
// Create trie that will load all nodes from DB.
tr, _ := New(TrieID(tr.Hash()), tdb)
// Remove a random node from the database. It can't be the root node
// because that one is already loaded.
var (
rval []byte
rpath []byte
rhash common.Hash
)
for {
if memonly {
rpath = paths[rand.Intn(len(paths))]
n := nodes.Nodes[string(rpath)]
if n == nil {
continue
}
rhash = n.Hash
} else {
index := rand.Intn(len(paths))
rpath = paths[index]
rhash = hashes[index]
}
if rhash != tr.Hash() {
break
}
}
if memonly {
tr.reader.banned = map[string]struct{}{string(rpath): {}}
} else {
rval = rawdb.ReadTrieNode(diskdb, common.Hash{}, rpath, rhash, tdb.Scheme())
rawdb.DeleteTrieNode(diskdb, common.Hash{}, rpath, rhash, tdb.Scheme())
}
// Iterate until the error is hit.
seen := make(map[string]bool)
it := tr.MustNodeIterator(nil)
checkIteratorNoDups(t, it, seen)
missing, ok := it.Error().(*MissingNodeError)
if !ok || missing.NodeHash != rhash {
t.Fatal("didn't hit missing node, got", it.Error())
}
// Add the node back and continue iteration.
if memonly {
delete(tr.reader.banned, string(rpath))
} else {
rawdb.WriteTrieNode(diskdb, common.Hash{}, rpath, rhash, rval, tdb.Scheme())
}
checkIteratorNoDups(t, it, seen)
if it.Error() != nil {
t.Fatal("unexpected error", it.Error())
}
if len(seen) != wantNodeCount {
t.Fatal("wrong node iteration count, got", len(seen), "want", wantNodeCount)
}
}
}
// Similar to the test above, this one checks that failure to create nodeIterator at a
// certain key prefix behaves correctly when Next is called. The expectation is that Next
// should retry seeking before returning true for the first time.
func TestIteratorContinueAfterSeekError(t *testing.T) {
testIteratorContinueAfterSeekError(t, false, rawdb.HashScheme)
testIteratorContinueAfterSeekError(t, true, rawdb.HashScheme)
testIteratorContinueAfterSeekError(t, false, rawdb.PathScheme)
testIteratorContinueAfterSeekError(t, true, rawdb.PathScheme)
}
func testIteratorContinueAfterSeekError(t *testing.T, memonly bool, scheme string) {
// Commit test trie to db, then remove the node containing "bars".
var (
barNodePath []byte
barNodeHash = common.HexToHash("05041990364eb72fcb1127652ce40d8bab765f2bfe53225b1170d276cc101c2e")
)
diskdb := rawdb.NewMemoryDatabase()
triedb := newTestDatabase(diskdb, scheme)
ctr := NewEmpty(triedb)
for _, val := range testdata1 {
ctr.MustUpdate([]byte(val.k), []byte(val.v))
}
root, nodes, _ := ctr.Commit(false)
for path, n := range nodes.Nodes {
if n.Hash == barNodeHash {
barNodePath = []byte(path)
break
}
}
triedb.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
if !memonly {
triedb.Commit(root, false)
}
var (
barNodeBlob []byte
)
tr, _ := New(TrieID(root), triedb)
if memonly {
tr.reader.banned = map[string]struct{}{string(barNodePath): {}}
} else {
barNodeBlob = rawdb.ReadTrieNode(diskdb, common.Hash{}, barNodePath, barNodeHash, triedb.Scheme())
rawdb.DeleteTrieNode(diskdb, common.Hash{}, barNodePath, barNodeHash, triedb.Scheme())
}
// Create a new iterator that seeks to "bars". Seeking can't proceed because
// the node is missing.
it := tr.MustNodeIterator([]byte("bars"))
missing, ok := it.Error().(*MissingNodeError)
if !ok {
t.Fatal("want MissingNodeError, got", it.Error())
} else if missing.NodeHash != barNodeHash {
t.Fatal("wrong node missing")
}
// Reinsert the missing node.
if memonly {
delete(tr.reader.banned, string(barNodePath))
} else {
rawdb.WriteTrieNode(diskdb, common.Hash{}, barNodePath, barNodeHash, barNodeBlob, triedb.Scheme())
}
// Check that iteration produces the right set of values.
if err := checkIteratorOrder(testdata1[2:], NewIterator(it)); err != nil {
t.Fatal(err)
}
}
func checkIteratorNoDups(t *testing.T, it NodeIterator, seen map[string]bool) int {
if seen == nil {
seen = make(map[string]bool)
}
for it.Next(true) {
if seen[string(it.Path())] {
t.Fatalf("iterator visited node path %x twice", it.Path())
}
seen[string(it.Path())] = true
}
return len(seen)
}
func TestIteratorNodeBlob(t *testing.T) {
testIteratorNodeBlob(t, rawdb.HashScheme)
testIteratorNodeBlob(t, rawdb.PathScheme)
}
func testIteratorNodeBlob(t *testing.T, scheme string) {
var (
db = rawdb.NewMemoryDatabase()
triedb = newTestDatabase(db, scheme)
trie = NewEmpty(triedb)
)
vals := []struct{ k, v string }{
{"do", "verb"},
{"ether", "wookiedoo"},
{"horse", "stallion"},
{"shaman", "horse"},
{"doge", "coin"},
{"dog", "puppy"},
{"somethingveryoddindeedthis is", "myothernodedata"},
}
all := make(map[string]string)
for _, val := range vals {
all[val.k] = val.v
trie.MustUpdate([]byte(val.k), []byte(val.v))
}
root, nodes, _ := trie.Commit(false)
triedb.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
triedb.Commit(root, false)
var found = make(map[common.Hash][]byte)
trie, _ = New(TrieID(root), triedb)
it := trie.MustNodeIterator(nil)
for it.Next(true) {
if it.Hash() == (common.Hash{}) {
continue
}
found[it.Hash()] = it.NodeBlob()
}
dbIter := db.NewIterator(nil, nil)
defer dbIter.Release()
var count int
for dbIter.Next() {
ok, _, _ := isTrieNode(triedb.Scheme(), dbIter.Key(), dbIter.Value())
if !ok {
continue
}
got, present := found[crypto.Keccak256Hash(dbIter.Value())]
if !present {
t.Fatal("Miss trie node")
}
if !bytes.Equal(got, dbIter.Value()) {
t.Fatalf("Unexpected trie node want %v got %v", dbIter.Value(), got)
}
count += 1
}
if count != len(found) {
t.Fatal("Find extra trie node via iterator")
}
}
// isTrieNode is a helper function which reports if the provided
// database entry belongs to a trie node or not. Note in tests
// only single layer trie is used, namely storage trie is not
// considered at all.
func isTrieNode(scheme string, key, val []byte) (bool, []byte, common.Hash) {
var (
path []byte
hash common.Hash
)
if scheme == rawdb.HashScheme {
ok := rawdb.IsLegacyTrieNode(key, val)
if !ok {
return false, nil, common.Hash{}
}
hash = common.BytesToHash(key)
} else {
ok, remain := rawdb.ResolveAccountTrieNodeKey(key)
if !ok {
return false, nil, common.Hash{}
}
path = common.CopyBytes(remain)
hash = crypto.Keccak256Hash(val)
}
return true, path, hash
}
func BenchmarkIterator(b *testing.B) {
diskDb, srcDb, tr, _ := makeTestTrie(rawdb.HashScheme)
root := tr.Hash()
b.ReportAllocs()
b.ResetTimer()
for i := 0; i < b.N; i++ {
if err := checkTrieConsistency(diskDb, srcDb.Scheme(), root, false); err != nil {
b.Fatal(err)
}
}
}

254
trie/node.go
View file

@ -1,254 +0,0 @@
// Copyright 2014 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 trie
import (
"fmt"
"io"
"strings"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/rlp"
)
var indices = []string{"0", "1", "2", "3", "4", "5", "6", "7", "8", "9", "a", "b", "c", "d", "e", "f", "[17]"}
type node interface {
cache() (hashNode, bool)
encode(w rlp.EncoderBuffer)
fstring(string) string
}
type (
fullNode struct {
Children [17]node // Actual trie node data to encode/decode (needs custom encoder)
flags nodeFlag
}
shortNode struct {
Key []byte
Val node
flags nodeFlag
}
hashNode []byte
valueNode []byte
)
// nilValueNode is used when collapsing internal trie nodes for hashing, since
// unset children need to serialize correctly.
var nilValueNode = valueNode(nil)
// EncodeRLP encodes a full node into the consensus RLP format.
func (n *fullNode) EncodeRLP(w io.Writer) error {
eb := rlp.NewEncoderBuffer(w)
n.encode(eb)
return eb.Flush()
}
func (n *fullNode) copy() *fullNode { copy := *n; return &copy }
func (n *shortNode) copy() *shortNode { copy := *n; return &copy }
// nodeFlag contains caching-related metadata about a node.
type nodeFlag struct {
hash hashNode // cached hash of the node (may be nil)
dirty bool // whether the node has changes that must be written to the database
}
func (n *fullNode) cache() (hashNode, bool) { return n.flags.hash, n.flags.dirty }
func (n *shortNode) cache() (hashNode, bool) { return n.flags.hash, n.flags.dirty }
func (n hashNode) cache() (hashNode, bool) { return nil, true }
func (n valueNode) cache() (hashNode, bool) { return nil, true }
// Pretty printing.
func (n *fullNode) String() string { return n.fstring("") }
func (n *shortNode) String() string { return n.fstring("") }
func (n hashNode) String() string { return n.fstring("") }
func (n valueNode) String() string { return n.fstring("") }
func (n *fullNode) fstring(ind string) string {
resp := fmt.Sprintf("[\n%s ", ind)
for i, node := range &n.Children {
if node == nil {
resp += fmt.Sprintf("%s: <nil> ", indices[i])
} else {
resp += fmt.Sprintf("%s: %v", indices[i], node.fstring(ind+" "))
}
}
return resp + fmt.Sprintf("\n%s] ", ind)
}
func (n *shortNode) fstring(ind string) string {
return fmt.Sprintf("{%x: %v} ", n.Key, n.Val.fstring(ind+" "))
}
func (n hashNode) fstring(ind string) string {
return fmt.Sprintf("<%x> ", []byte(n))
}
func (n valueNode) fstring(ind string) string {
return fmt.Sprintf("%x ", []byte(n))
}
// rawNode is a simple binary blob used to differentiate between collapsed trie
// nodes and already encoded RLP binary blobs (while at the same time store them
// in the same cache fields).
type rawNode []byte
func (n rawNode) cache() (hashNode, bool) { panic("this should never end up in a live trie") }
func (n rawNode) fstring(ind string) string { panic("this should never end up in a live trie") }
func (n rawNode) EncodeRLP(w io.Writer) error {
_, err := w.Write(n)
return err
}
// mustDecodeNode is a wrapper of decodeNode and panic if any error is encountered.
func mustDecodeNode(hash, buf []byte) node {
n, err := decodeNode(hash, buf)
if err != nil {
panic(fmt.Sprintf("node %x: %v", hash, err))
}
return n
}
// mustDecodeNodeUnsafe is a wrapper of decodeNodeUnsafe and panic if any error is
// encountered.
func mustDecodeNodeUnsafe(hash, buf []byte) node {
n, err := decodeNodeUnsafe(hash, buf)
if err != nil {
panic(fmt.Sprintf("node %x: %v", hash, err))
}
return n
}
// decodeNode parses the RLP encoding of a trie node. It will deep-copy the passed
// byte slice for decoding, so it's safe to modify the byte slice afterwards. The-
// decode performance of this function is not optimal, but it is suitable for most
// scenarios with low performance requirements and hard to determine whether the
// byte slice be modified or not.
func decodeNode(hash, buf []byte) (node, error) {
return decodeNodeUnsafe(hash, common.CopyBytes(buf))
}
// decodeNodeUnsafe parses the RLP encoding of a trie node. The passed byte slice
// will be directly referenced by node without bytes deep copy, so the input MUST
// not be changed after.
func decodeNodeUnsafe(hash, buf []byte) (node, error) {
if len(buf) == 0 {
return nil, io.ErrUnexpectedEOF
}
elems, _, err := rlp.SplitList(buf)
if err != nil {
return nil, fmt.Errorf("decode error: %v", err)
}
switch c, _ := rlp.CountValues(elems); c {
case 2:
n, err := decodeShort(hash, elems)
return n, wrapError(err, "short")
case 17:
n, err := decodeFull(hash, elems)
return n, wrapError(err, "full")
default:
return nil, fmt.Errorf("invalid number of list elements: %v", c)
}
}
func decodeShort(hash, elems []byte) (node, error) {
kbuf, rest, err := rlp.SplitString(elems)
if err != nil {
return nil, err
}
flag := nodeFlag{hash: hash}
key := compactToHex(kbuf)
if hasTerm(key) {
// value node
val, _, err := rlp.SplitString(rest)
if err != nil {
return nil, fmt.Errorf("invalid value node: %v", err)
}
return &shortNode{key, valueNode(val), flag}, nil
}
r, _, err := decodeRef(rest)
if err != nil {
return nil, wrapError(err, "val")
}
return &shortNode{key, r, flag}, nil
}
func decodeFull(hash, elems []byte) (*fullNode, error) {
n := &fullNode{flags: nodeFlag{hash: hash}}
for i := 0; i < 16; i++ {
cld, rest, err := decodeRef(elems)
if err != nil {
return n, wrapError(err, fmt.Sprintf("[%d]", i))
}
n.Children[i], elems = cld, rest
}
val, _, err := rlp.SplitString(elems)
if err != nil {
return n, err
}
if len(val) > 0 {
n.Children[16] = valueNode(val)
}
return n, nil
}
const hashLen = len(common.Hash{})
func decodeRef(buf []byte) (node, []byte, error) {
kind, val, rest, err := rlp.Split(buf)
if err != nil {
return nil, buf, err
}
switch {
case kind == rlp.List:
// 'embedded' node reference. The encoding must be smaller
// than a hash in order to be valid.
if size := len(buf) - len(rest); size > hashLen {
err := fmt.Errorf("oversized embedded node (size is %d bytes, want size < %d)", size, hashLen)
return nil, buf, err
}
n, err := decodeNode(nil, buf)
return n, rest, err
case kind == rlp.String && len(val) == 0:
// empty node
return nil, rest, nil
case kind == rlp.String && len(val) == 32:
return hashNode(val), rest, nil
default:
return nil, nil, fmt.Errorf("invalid RLP string size %d (want 0 or 32)", len(val))
}
}
// wraps a decoding error with information about the path to the
// invalid child node (for debugging encoding issues).
type decodeError struct {
what error
stack []string
}
func wrapError(err error, ctx string) error {
if err == nil {
return nil
}
if decErr, ok := err.(*decodeError); ok {
decErr.stack = append(decErr.stack, ctx)
return decErr
}
return &decodeError{err, []string{ctx}}
}
func (err *decodeError) Error() string {
return fmt.Sprintf("%v (decode path: %s)", err.what, strings.Join(err.stack, "<-"))
}

64
trie/node_enc.go
View file

@ -1,64 +0,0 @@
// 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 trie
import (
"github.com/ethereum/go-ethereum/rlp"
)
func nodeToBytes(n node) []byte {
w := rlp.NewEncoderBuffer(nil)
n.encode(w)
result := w.ToBytes()
w.Flush()
return result
}
func (n *fullNode) encode(w rlp.EncoderBuffer) {
offset := w.List()
for _, c := range n.Children {
if c != nil {
c.encode(w)
} else {
w.Write(rlp.EmptyString)
}
}
w.ListEnd(offset)
}
func (n *shortNode) encode(w rlp.EncoderBuffer) {
offset := w.List()
w.WriteBytes(n.Key)
if n.Val != nil {
n.Val.encode(w)
} else {
w.Write(rlp.EmptyString)
}
w.ListEnd(offset)
}
func (n hashNode) encode(w rlp.EncoderBuffer) {
w.WriteBytes(n)
}
func (n valueNode) encode(w rlp.EncoderBuffer) {
w.WriteBytes(n)
}
func (n rawNode) encode(w rlp.EncoderBuffer) {
w.Write(n)
}

215
trie/node_test.go
View file

@ -1,215 +0,0 @@
// Copyright 2016 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 trie
import (
"bytes"
"testing"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/rlp"
)
func newTestFullNode(v []byte) []interface{} {
fullNodeData := []interface{}{}
for i := 0; i < 16; i++ {
k := bytes.Repeat([]byte{byte(i + 1)}, 32)
fullNodeData = append(fullNodeData, k)
}
fullNodeData = append(fullNodeData, v)
return fullNodeData
}
func TestDecodeNestedNode(t *testing.T) {
fullNodeData := newTestFullNode([]byte("fullnode"))
data := [][]byte{}
for i := 0; i < 16; i++ {
data = append(data, nil)
}
data = append(data, []byte("subnode"))
fullNodeData[15] = data
buf := bytes.NewBuffer([]byte{})
rlp.Encode(buf, fullNodeData)
if _, err := decodeNode([]byte("testdecode"), buf.Bytes()); err != nil {
t.Fatalf("decode nested full node err: %v", err)
}
}
func TestDecodeFullNodeWrongSizeChild(t *testing.T) {
fullNodeData := newTestFullNode([]byte("wrongsizechild"))
fullNodeData[0] = []byte("00")
buf := bytes.NewBuffer([]byte{})
rlp.Encode(buf, fullNodeData)
_, err := decodeNode([]byte("testdecode"), buf.Bytes())
if _, ok := err.(*decodeError); !ok {
t.Fatalf("decodeNode returned wrong err: %v", err)
}
}
func TestDecodeFullNodeWrongNestedFullNode(t *testing.T) {
fullNodeData := newTestFullNode([]byte("fullnode"))
data := [][]byte{}
for i := 0; i < 16; i++ {
data = append(data, []byte("123456"))
}
data = append(data, []byte("subnode"))
fullNodeData[15] = data
buf := bytes.NewBuffer([]byte{})
rlp.Encode(buf, fullNodeData)
_, err := decodeNode([]byte("testdecode"), buf.Bytes())
if _, ok := err.(*decodeError); !ok {
t.Fatalf("decodeNode returned wrong err: %v", err)
}
}
func TestDecodeFullNode(t *testing.T) {
fullNodeData := newTestFullNode([]byte("decodefullnode"))
buf := bytes.NewBuffer([]byte{})
rlp.Encode(buf, fullNodeData)
_, err := decodeNode([]byte("testdecode"), buf.Bytes())
if err != nil {
t.Fatalf("decode full node err: %v", err)
}
}
// goos: darwin
// goarch: arm64
// pkg: github.com/ethereum/go-ethereum/trie
// BenchmarkEncodeShortNode
// BenchmarkEncodeShortNode-8 16878850 70.81 ns/op 48 B/op 1 allocs/op
func BenchmarkEncodeShortNode(b *testing.B) {
node := &shortNode{
Key: []byte{0x1, 0x2},
Val: hashNode(randBytes(32)),
}
b.ResetTimer()
b.ReportAllocs()
for i := 0; i < b.N; i++ {
nodeToBytes(node)
}
}
// goos: darwin
// goarch: arm64
// pkg: github.com/ethereum/go-ethereum/trie
// BenchmarkEncodeFullNode
// BenchmarkEncodeFullNode-8 4323273 284.4 ns/op 576 B/op 1 allocs/op
func BenchmarkEncodeFullNode(b *testing.B) {
node := &fullNode{}
for i := 0; i < 16; i++ {
node.Children[i] = hashNode(randBytes(32))
}
b.ResetTimer()
b.ReportAllocs()
for i := 0; i < b.N; i++ {
nodeToBytes(node)
}
}
// goos: darwin
// goarch: arm64
// pkg: github.com/ethereum/go-ethereum/trie
// BenchmarkDecodeShortNode
// BenchmarkDecodeShortNode-8 7925638 151.0 ns/op 157 B/op 4 allocs/op
func BenchmarkDecodeShortNode(b *testing.B) {
node := &shortNode{
Key: []byte{0x1, 0x2},
Val: hashNode(randBytes(32)),
}
blob := nodeToBytes(node)
hash := crypto.Keccak256(blob)
b.ResetTimer()
b.ReportAllocs()
for i := 0; i < b.N; i++ {
mustDecodeNode(hash, blob)
}
}
// goos: darwin
// goarch: arm64
// pkg: github.com/ethereum/go-ethereum/trie
// BenchmarkDecodeShortNodeUnsafe
// BenchmarkDecodeShortNodeUnsafe-8 9027476 128.6 ns/op 109 B/op 3 allocs/op
func BenchmarkDecodeShortNodeUnsafe(b *testing.B) {
node := &shortNode{
Key: []byte{0x1, 0x2},
Val: hashNode(randBytes(32)),
}
blob := nodeToBytes(node)
hash := crypto.Keccak256(blob)
b.ResetTimer()
b.ReportAllocs()
for i := 0; i < b.N; i++ {
mustDecodeNodeUnsafe(hash, blob)
}
}
// goos: darwin
// goarch: arm64
// pkg: github.com/ethereum/go-ethereum/trie
// BenchmarkDecodeFullNode
// BenchmarkDecodeFullNode-8 1597462 761.9 ns/op 1280 B/op 18 allocs/op
func BenchmarkDecodeFullNode(b *testing.B) {
node := &fullNode{}
for i := 0; i < 16; i++ {
node.Children[i] = hashNode(randBytes(32))
}
blob := nodeToBytes(node)
hash := crypto.Keccak256(blob)
b.ResetTimer()
b.ReportAllocs()
for i := 0; i < b.N; i++ {
mustDecodeNode(hash, blob)
}
}
// goos: darwin
// goarch: arm64
// pkg: github.com/ethereum/go-ethereum/trie
// BenchmarkDecodeFullNodeUnsafe
// BenchmarkDecodeFullNodeUnsafe-8 1789070 687.1 ns/op 704 B/op 17 allocs/op
func BenchmarkDecodeFullNodeUnsafe(b *testing.B) {
node := &fullNode{}
for i := 0; i < 16; i++ {
node.Children[i] = hashNode(randBytes(32))
}
blob := nodeToBytes(node)
hash := crypto.Keccak256(blob)
b.ResetTimer()
b.ReportAllocs()
for i := 0; i < b.N; i++ {
mustDecodeNodeUnsafe(hash, blob)
}
}

95
trie/preimages.go
View file

@ -1,95 +0,0 @@
// 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 trie
import (
"sync"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/ethdb"
)
// preimageStore is the store for caching preimages of node key.
type preimageStore struct {
lock sync.RWMutex
disk ethdb.KeyValueStore
preimages map[common.Hash][]byte // Preimages of nodes from the secure trie
preimagesSize common.StorageSize // Storage size of the preimages cache
}
// newPreimageStore initializes the store for caching preimages.
func newPreimageStore(disk ethdb.KeyValueStore) *preimageStore {
return &preimageStore{
disk: disk,
preimages: make(map[common.Hash][]byte),
}
}
// insertPreimage writes a new trie node pre-image to the memory database if it's
// yet unknown. The method will NOT make a copy of the slice, only use if the
// preimage will NOT be changed later on.
func (store *preimageStore) insertPreimage(preimages map[common.Hash][]byte) {
store.lock.Lock()
defer store.lock.Unlock()
for hash, preimage := range preimages {
if _, ok := store.preimages[hash]; ok {
continue
}
store.preimages[hash] = preimage
store.preimagesSize += common.StorageSize(common.HashLength + len(preimage))
}
}
// preimage retrieves a cached trie node pre-image from memory. If it cannot be
// found cached, the method queries the persistent database for the content.
func (store *preimageStore) preimage(hash common.Hash) []byte {
store.lock.RLock()
preimage := store.preimages[hash]
store.lock.RUnlock()
if preimage != nil {
return preimage
}
return rawdb.ReadPreimage(store.disk, hash)
}
// commit flushes the cached preimages into the disk.
func (store *preimageStore) commit(force bool) error {
store.lock.Lock()
defer store.lock.Unlock()
if store.preimagesSize <= 4*1024*1024 && !force {
return nil
}
batch := store.disk.NewBatch()
rawdb.WritePreimages(batch, store.preimages)
if err := batch.Write(); err != nil {
return err
}
store.preimages, store.preimagesSize = make(map[common.Hash][]byte), 0
return nil
}
// size returns the current storage size of accumulated preimages.
func (store *preimageStore) size() common.StorageSize {
store.lock.RLock()
defer store.lock.RUnlock()
return store.preimagesSize
}

616
trie/proof.go
View file

@ -1,616 +0,0 @@
// Copyright 2015 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 trie
import (
"bytes"
"errors"
"fmt"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/log"
)
// Prove constructs a merkle proof for key. The result contains all encoded nodes
// on the path to the value at key. The value itself is also included in the last
// node and can be retrieved by verifying the proof.
//
// If the trie does not contain a value for key, the returned proof contains all
// nodes of the longest existing prefix of the key (at least the root node), ending
// with the node that proves the absence of the key.
func (t *Trie) Prove(key []byte, proofDb ethdb.KeyValueWriter) error {
// Short circuit if the trie is already committed and not usable.
if t.committed {
return ErrCommitted
}
// Collect all nodes on the path to key.
var (
prefix []byte
nodes []node
tn = t.root
)
key = keybytesToHex(key)
for len(key) > 0 && tn != nil {
switch n := tn.(type) {
case *shortNode:
if len(key) < len(n.Key) || !bytes.Equal(n.Key, key[:len(n.Key)]) {
// The trie doesn't contain the key.
tn = nil
} else {
tn = n.Val
prefix = append(prefix, n.Key...)
key = key[len(n.Key):]
}
nodes = append(nodes, n)
case *fullNode:
tn = n.Children[key[0]]
prefix = append(prefix, key[0])
key = key[1:]
nodes = append(nodes, n)
case hashNode:
// Retrieve the specified node from the underlying node reader.
// trie.resolveAndTrack is not used since in that function the
// loaded blob will be tracked, while it's not required here since
// all loaded nodes won't be linked to trie at all and track nodes
// may lead to out-of-memory issue.
blob, err := t.reader.node(prefix, common.BytesToHash(n))
if err != nil {
log.Error("Unhandled trie error in Trie.Prove", "err", err)
return err
}
// The raw-blob format nodes are loaded either from the
// clean cache or the database, they are all in their own
// copy and safe to use unsafe decoder.
tn = mustDecodeNodeUnsafe(n, blob)
default:
panic(fmt.Sprintf("%T: invalid node: %v", tn, tn))
}
}
hasher := newHasher(false)
defer returnHasherToPool(hasher)
for i, n := range nodes {
var hn node
n, hn = hasher.proofHash(n)
if hash, ok := hn.(hashNode); ok || i == 0 {
// If the node's database encoding is a hash (or is the
// root node), it becomes a proof element.
enc := nodeToBytes(n)
if !ok {
hash = hasher.hashData(enc)
}
proofDb.Put(hash, enc)
}
}
return nil
}
// Prove constructs a merkle proof for key. The result contains all encoded nodes
// on the path to the value at key. The value itself is also included in the last
// node and can be retrieved by verifying the proof.
//
// If the trie does not contain a value for key, the returned proof contains all
// nodes of the longest existing prefix of the key (at least the root node), ending
// with the node that proves the absence of the key.
func (t *StateTrie) Prove(key []byte, proofDb ethdb.KeyValueWriter) error {
return t.trie.Prove(key, proofDb)
}
// VerifyProof checks merkle proofs. The given proof must contain the value for
// key in a trie with the given root hash. VerifyProof returns an error if the
// proof contains invalid trie nodes or the wrong value.
func VerifyProof(rootHash common.Hash, key []byte, proofDb ethdb.KeyValueReader) (value []byte, err error) {
key = keybytesToHex(key)
wantHash := rootHash
for i := 0; ; i++ {
buf, _ := proofDb.Get(wantHash[:])
if buf == nil {
return nil, fmt.Errorf("proof node %d (hash %064x) missing", i, wantHash)
}
n, err := decodeNode(wantHash[:], buf)
if err != nil {
return nil, fmt.Errorf("bad proof node %d: %v", i, err)
}
keyrest, cld := get(n, key, true)
switch cld := cld.(type) {
case nil:
// The trie doesn't contain the key.
return nil, nil
case hashNode:
key = keyrest
copy(wantHash[:], cld)
case valueNode:
return cld, nil
}
}
}
// proofToPath converts a merkle proof to trie node path. The main purpose of
// this function is recovering a node path from the merkle proof stream. All
// necessary nodes will be resolved and leave the remaining as hashnode.
//
// The given edge proof is allowed to be an existent or non-existent proof.
func proofToPath(rootHash common.Hash, root node, key []byte, proofDb ethdb.KeyValueReader, allowNonExistent bool) (node, []byte, error) {
// resolveNode retrieves and resolves trie node from merkle proof stream
resolveNode := func(hash common.Hash) (node, error) {
buf, _ := proofDb.Get(hash[:])
if buf == nil {
return nil, fmt.Errorf("proof node (hash %064x) missing", hash)
}
n, err := decodeNode(hash[:], buf)
if err != nil {
return nil, fmt.Errorf("bad proof node %v", err)
}
return n, err
}
// If the root node is empty, resolve it first.
// Root node must be included in the proof.
if root == nil {
n, err := resolveNode(rootHash)
if err != nil {
return nil, nil, err
}
root = n
}
var (
err error
child, parent node
keyrest []byte
valnode []byte
)
key, parent = keybytesToHex(key), root
for {
keyrest, child = get(parent, key, false)
switch cld := child.(type) {
case nil:
// The trie doesn't contain the key. It's possible
// the proof is a non-existing proof, but at least
// we can prove all resolved nodes are correct, it's
// enough for us to prove range.
if allowNonExistent {
return root, nil, nil
}
return nil, nil, errors.New("the node is not contained in trie")
case *shortNode:
key, parent = keyrest, child // Already resolved
continue
case *fullNode:
key, parent = keyrest, child // Already resolved
continue
case hashNode:
child, err = resolveNode(common.BytesToHash(cld))
if err != nil {
return nil, nil, err
}
case valueNode:
valnode = cld
}
// Link the parent and child.
switch pnode := parent.(type) {
case *shortNode:
pnode.Val = child
case *fullNode:
pnode.Children[key[0]] = child
default:
panic(fmt.Sprintf("%T: invalid node: %v", pnode, pnode))
}
if len(valnode) > 0 {
return root, valnode, nil // The whole path is resolved
}
key, parent = keyrest, child
}
}
// unsetInternal removes all internal node references(hashnode, embedded node).
// It should be called after a trie is constructed with two edge paths. Also
// the given boundary keys must be the one used to construct the edge paths.
//
// It's the key step for range proof. All visited nodes should be marked dirty
// since the node content might be modified. Besides it can happen that some
// fullnodes only have one child which is disallowed. But if the proof is valid,
// the missing children will be filled, otherwise it will be thrown anyway.
//
// Note we have the assumption here the given boundary keys are different
// and right is larger than left.
func unsetInternal(n node, left []byte, right []byte) (bool, error) {
left, right = keybytesToHex(left), keybytesToHex(right)
// Step down to the fork point. There are two scenarios can happen:
// - the fork point is a shortnode: either the key of left proof or
// right proof doesn't match with shortnode's key.
// - the fork point is a fullnode: both two edge proofs are allowed
// to point to a non-existent key.
var (
pos = 0
parent node
// fork indicator, 0 means no fork, -1 means proof is less, 1 means proof is greater
shortForkLeft, shortForkRight int
)
findFork:
for {
switch rn := (n).(type) {
case *shortNode:
rn.flags = nodeFlag{dirty: true}
// If either the key of left proof or right proof doesn't match with
// shortnode, stop here and the forkpoint is the shortnode.
if len(left)-pos < len(rn.Key) {
shortForkLeft = bytes.Compare(left[pos:], rn.Key)
} else {
shortForkLeft = bytes.Compare(left[pos:pos+len(rn.Key)], rn.Key)
}
if len(right)-pos < len(rn.Key) {
shortForkRight = bytes.Compare(right[pos:], rn.Key)
} else {
shortForkRight = bytes.Compare(right[pos:pos+len(rn.Key)], rn.Key)
}
if shortForkLeft != 0 || shortForkRight != 0 {
break findFork
}
parent = n
n, pos = rn.Val, pos+len(rn.Key)
case *fullNode:
rn.flags = nodeFlag{dirty: true}
// If either the node pointed by left proof or right proof is nil,
// stop here and the forkpoint is the fullnode.
leftnode, rightnode := rn.Children[left[pos]], rn.Children[right[pos]]
if leftnode == nil || rightnode == nil || leftnode != rightnode {
break findFork
}
parent = n
n, pos = rn.Children[left[pos]], pos+1
default:
panic(fmt.Sprintf("%T: invalid node: %v", n, n))
}
}
switch rn := n.(type) {
case *shortNode:
// There can have these five scenarios:
// - both proofs are less than the trie path => no valid range
// - both proofs are greater than the trie path => no valid range
// - left proof is less and right proof is greater => valid range, unset the shortnode entirely
// - left proof points to the shortnode, but right proof is greater
// - right proof points to the shortnode, but left proof is less
if shortForkLeft == -1 && shortForkRight == -1 {
return false, errors.New("empty range")
}
if shortForkLeft == 1 && shortForkRight == 1 {
return false, errors.New("empty range")
}
if shortForkLeft != 0 && shortForkRight != 0 {
// The fork point is root node, unset the entire trie
if parent == nil {
return true, nil
}
parent.(*fullNode).Children[left[pos-1]] = nil
return false, nil
}
// Only one proof points to non-existent key.
if shortForkRight != 0 {
if _, ok := rn.Val.(valueNode); ok {
// The fork point is root node, unset the entire trie
if parent == nil {
return true, nil
}
parent.(*fullNode).Children[left[pos-1]] = nil
return false, nil
}
return false, unset(rn, rn.Val, left[pos:], len(rn.Key), false)
}
if shortForkLeft != 0 {
if _, ok := rn.Val.(valueNode); ok {
// The fork point is root node, unset the entire trie
if parent == nil {
return true, nil
}
parent.(*fullNode).Children[right[pos-1]] = nil
return false, nil
}
return false, unset(rn, rn.Val, right[pos:], len(rn.Key), true)
}
return false, nil
case *fullNode:
// unset all internal nodes in the forkpoint
for i := left[pos] + 1; i < right[pos]; i++ {
rn.Children[i] = nil
}
if err := unset(rn, rn.Children[left[pos]], left[pos:], 1, false); err != nil {
return false, err
}
if err := unset(rn, rn.Children[right[pos]], right[pos:], 1, true); err != nil {
return false, err
}
return false, nil
default:
panic(fmt.Sprintf("%T: invalid node: %v", n, n))
}
}
// unset removes all internal node references either the left most or right most.
// It can meet these scenarios:
//
// - The given path is existent in the trie, unset the associated nodes with the
// specific direction
// - The given path is non-existent in the trie
// - the fork point is a fullnode, the corresponding child pointed by path
// is nil, return
// - the fork point is a shortnode, the shortnode is included in the range,
// keep the entire branch and return.
// - the fork point is a shortnode, the shortnode is excluded in the range,
// unset the entire branch.
func unset(parent node, child node, key []byte, pos int, removeLeft bool) error {
switch cld := child.(type) {
case *fullNode:
if removeLeft {
for i := 0; i < int(key[pos]); i++ {
cld.Children[i] = nil
}
cld.flags = nodeFlag{dirty: true}
} else {
for i := key[pos] + 1; i < 16; i++ {
cld.Children[i] = nil
}
cld.flags = nodeFlag{dirty: true}
}
return unset(cld, cld.Children[key[pos]], key, pos+1, removeLeft)
case *shortNode:
if len(key[pos:]) < len(cld.Key) || !bytes.Equal(cld.Key, key[pos:pos+len(cld.Key)]) {
// Find the fork point, it's an non-existent branch.
if removeLeft {
if bytes.Compare(cld.Key, key[pos:]) < 0 {
// The key of fork shortnode is less than the path
// (it belongs to the range), unset the entire
// branch. The parent must be a fullnode.
fn := parent.(*fullNode)
fn.Children[key[pos-1]] = nil
}
//else {
// The key of fork shortnode is greater than the
// path(it doesn't belong to the range), keep
// it with the cached hash available.
//}
} else {
if bytes.Compare(cld.Key, key[pos:]) > 0 {
// The key of fork shortnode is greater than the
// path(it belongs to the range), unset the entrie
// branch. The parent must be a fullnode.
fn := parent.(*fullNode)
fn.Children[key[pos-1]] = nil
}
//else {
// The key of fork shortnode is less than the
// path(it doesn't belong to the range), keep
// it with the cached hash available.
//}
}
return nil
}
if _, ok := cld.Val.(valueNode); ok {
fn := parent.(*fullNode)
fn.Children[key[pos-1]] = nil
return nil
}
cld.flags = nodeFlag{dirty: true}
return unset(cld, cld.Val, key, pos+len(cld.Key), removeLeft)
case nil:
// If the node is nil, then it's a child of the fork point
// fullnode(it's a non-existent branch).
return nil
default:
panic("it shouldn't happen") // hashNode, valueNode
}
}
// hasRightElement returns the indicator whether there exists more elements
// on the right side of the given path. The given path can point to an existent
// key or a non-existent one. This function has the assumption that the whole
// path should already be resolved.
func hasRightElement(node node, key []byte) bool {
pos, key := 0, keybytesToHex(key)
for node != nil {
switch rn := node.(type) {
case *fullNode:
for i := key[pos] + 1; i < 16; i++ {
if rn.Children[i] != nil {
return true
}
}
node, pos = rn.Children[key[pos]], pos+1
case *shortNode:
if len(key)-pos < len(rn.Key) || !bytes.Equal(rn.Key, key[pos:pos+len(rn.Key)]) {
return bytes.Compare(rn.Key, key[pos:]) > 0
}
node, pos = rn.Val, pos+len(rn.Key)
case valueNode:
return false // We have resolved the whole path
default:
panic(fmt.Sprintf("%T: invalid node: %v", node, node)) // hashnode
}
}
return false
}
// VerifyRangeProof checks whether the given leaf nodes and edge proof
// can prove the given trie leaves range is matched with the specific root.
// Besides, the range should be consecutive (no gap inside) and monotonic
// increasing.
//
// Note the given proof actually contains two edge proofs. Both of them can
// be non-existent proofs. For example the first proof is for a non-existent
// key 0x03, the last proof is for a non-existent key 0x10. The given batch
// leaves are [0x04, 0x05, .. 0x09]. It's still feasible to prove the given
// batch is valid.
//
// The firstKey is paired with firstProof, not necessarily the same as keys[0]
// (unless firstProof is an existent proof). Similarly, lastKey and lastProof
// are paired.
//
// Expect the normal case, this function can also be used to verify the following
// range proofs:
//
// - All elements proof. In this case the proof can be nil, but the range should
// be all the leaves in the trie.
//
// - One element proof. In this case no matter the edge proof is a non-existent
// proof or not, we can always verify the correctness of the proof.
//
// - Zero element proof. In this case a single non-existent proof is enough to prove.
// Besides, if there are still some other leaves available on the right side, then
// an error will be returned.
//
// Except returning the error to indicate the proof is valid or not, the function will
// also return a flag to indicate whether there exists more accounts/slots in the trie.
//
// Note: This method does not verify that the proof is of minimal form. If the input
// proofs are 'bloated' with neighbour leaves or random data, aside from the 'useful'
// data, then the proof will still be accepted.
func VerifyRangeProof(rootHash common.Hash, firstKey []byte, keys [][]byte, values [][]byte, proof ethdb.KeyValueReader) (bool, error) {
if len(keys) != len(values) {
return false, fmt.Errorf("inconsistent proof data, keys: %d, values: %d", len(keys), len(values))
}
// Ensure the received batch is monotonic increasing and contains no deletions
for i := 0; i < len(keys)-1; i++ {
if bytes.Compare(keys[i], keys[i+1]) >= 0 {
return false, errors.New("range is not monotonically increasing")
}
}
for _, value := range values {
if len(value) == 0 {
return false, errors.New("range contains deletion")
}
}
// Special case, there is no edge proof at all. The given range is expected
// to be the whole leaf-set in the trie.
if proof == nil {
tr := NewStackTrie(nil)
for index, key := range keys {
tr.Update(key, values[index])
}
if have, want := tr.Hash(), rootHash; have != want {
return false, fmt.Errorf("invalid proof, want hash %x, got %x", want, have)
}
return false, nil // No more elements
}
// Special case, there is a provided edge proof but zero key/value
// pairs, ensure there are no more accounts / slots in the trie.
if len(keys) == 0 {
root, val, err := proofToPath(rootHash, nil, firstKey, proof, true)
if err != nil {
return false, err
}
if val != nil || hasRightElement(root, firstKey) {
return false, errors.New("more entries available")
}
return false, nil
}
var lastKey = keys[len(keys)-1]
// Special case, there is only one element and two edge keys are same.
// In this case, we can't construct two edge paths. So handle it here.
if len(keys) == 1 && bytes.Equal(firstKey, lastKey) {
root, val, err := proofToPath(rootHash, nil, firstKey, proof, false)
if err != nil {
return false, err
}
if !bytes.Equal(firstKey, keys[0]) {
return false, errors.New("correct proof but invalid key")
}
if !bytes.Equal(val, values[0]) {
return false, errors.New("correct proof but invalid data")
}
return hasRightElement(root, firstKey), nil
}
// Ok, in all other cases, we require two edge paths available.
// First check the validity of edge keys.
if bytes.Compare(firstKey, lastKey) >= 0 {
return false, errors.New("invalid edge keys")
}
// todo(rjl493456442) different length edge keys should be supported
if len(firstKey) != len(lastKey) {
return false, errors.New("inconsistent edge keys")
}
// Convert the edge proofs to edge trie paths. Then we can
// have the same tree architecture with the original one.
// For the first edge proof, non-existent proof is allowed.
root, _, err := proofToPath(rootHash, nil, firstKey, proof, true)
if err != nil {
return false, err
}
// Pass the root node here, the second path will be merged
// with the first one. For the last edge proof, non-existent
// proof is also allowed.
root, _, err = proofToPath(rootHash, root, lastKey, proof, true)
if err != nil {
return false, err
}
// Remove all internal references. All the removed parts should
// be re-filled(or re-constructed) by the given leaves range.
empty, err := unsetInternal(root, firstKey, lastKey)
if err != nil {
return false, err
}
// Rebuild the trie with the leaf stream, the shape of trie
// should be same with the original one.
tr := &Trie{root: root, reader: newEmptyReader(), tracer: newTracer()}
if empty {
tr.root = nil
}
for index, key := range keys {
tr.Update(key, values[index])
}
if tr.Hash() != rootHash {
return false, fmt.Errorf("invalid proof, want hash %x, got %x", rootHash, tr.Hash())
}
return hasRightElement(tr.root, keys[len(keys)-1]), nil
}
// get returns the child of the given node. Return nil if the
// node with specified key doesn't exist at all.
//
// There is an additional flag `skipResolved`. If it's set then
// all resolved nodes won't be returned.
func get(tn node, key []byte, skipResolved bool) ([]byte, node) {
for {
switch n := tn.(type) {
case *shortNode:
if len(key) < len(n.Key) || !bytes.Equal(n.Key, key[:len(n.Key)]) {
return nil, nil
}
tn = n.Val
key = key[len(n.Key):]
if !skipResolved {
return key, tn
}
case *fullNode:
tn = n.Children[key[0]]
key = key[1:]
if !skipResolved {
return key, tn
}
case hashNode:
return key, n
case nil:
return key, nil
case valueNode:
return nil, n
default:
panic(fmt.Sprintf("%T: invalid node: %v", tn, tn))
}
}
}

1002
trie/proof_test.go
View file

File diff suppressed because it is too large Load diff

290
trie/secure_trie.go
View file

@ -1,290 +0,0 @@
// Copyright 2015 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 trie
import (
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/rlp"
"github.com/ethereum/go-ethereum/trie/trienode"
)
// SecureTrie is the old name of StateTrie.
// Deprecated: use StateTrie.
type SecureTrie = StateTrie
// NewSecure creates a new StateTrie.
// Deprecated: use NewStateTrie.
func NewSecure(stateRoot common.Hash, owner common.Hash, root common.Hash, db *Database) (*SecureTrie, error) {
id := &ID{
StateRoot: stateRoot,
Owner: owner,
Root: root,
}
return NewStateTrie(id, db)
}
// StateTrie wraps a trie with key hashing. In a stateTrie trie, all
// access operations hash the key using keccak256. This prevents
// calling code from creating long chains of nodes that
// increase the access time.
//
// Contrary to a regular trie, a StateTrie can only be created with
// New and must have an attached database. The database also stores
// the preimage of each key if preimage recording is enabled.
//
// StateTrie is not safe for concurrent use.
type StateTrie struct {
trie Trie
preimages *preimageStore
hashKeyBuf [common.HashLength]byte
secKeyCache map[string][]byte
secKeyCacheOwner *StateTrie // Pointer to self, replace the key cache on mismatch
}
// NewStateTrie creates a trie with an existing root node from a backing database.
//
// If root is the zero hash or the sha3 hash of an empty string, the
// trie is initially empty. Otherwise, New will panic if db is nil
// and returns MissingNodeError if the root node cannot be found.
func NewStateTrie(id *ID, db *Database) (*StateTrie, error) {
if db == nil {
panic("trie.NewStateTrie called without a database")
}
trie, err := New(id, db)
if err != nil {
return nil, err
}
return &StateTrie{trie: *trie, preimages: db.preimages}, nil
}
// MustGet returns the value for key stored in the trie.
// The value bytes must not be modified by the caller.
//
// This function will omit any encountered error but just
// print out an error message.
func (t *StateTrie) MustGet(key []byte) []byte {
return t.trie.MustGet(t.hashKey(key))
}
// GetStorage attempts to retrieve a storage slot with provided account address
// and slot key. The value bytes must not be modified by the caller.
// If the specified storage slot is not in the trie, nil will be returned.
// If a trie node is not found in the database, a MissingNodeError is returned.
func (t *StateTrie) GetStorage(_ common.Address, key []byte) ([]byte, error) {
enc, err := t.trie.Get(t.hashKey(key))
if err != nil || len(enc) == 0 {
return nil, err
}
_, content, _, err := rlp.Split(enc)
return content, err
}
// GetAccount attempts to retrieve an account with provided account address.
// If the specified account is not in the trie, nil will be returned.
// If a trie node is not found in the database, a MissingNodeError is returned.
func (t *StateTrie) GetAccount(address common.Address) (*types.StateAccount, error) {
res, err := t.trie.Get(t.hashKey(address.Bytes()))
if res == nil || err != nil {
return nil, err
}
ret := new(types.StateAccount)
err = rlp.DecodeBytes(res, ret)
return ret, err
}
// GetAccountByHash does the same thing as GetAccount, however it expects an
// account hash that is the hash of address. This constitutes an abstraction
// leak, since the client code needs to know the key format.
func (t *StateTrie) GetAccountByHash(addrHash common.Hash) (*types.StateAccount, error) {
res, err := t.trie.Get(addrHash.Bytes())
if res == nil || err != nil {
return nil, err
}
ret := new(types.StateAccount)
err = rlp.DecodeBytes(res, ret)
return ret, err
}
// GetNode 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.
// If the specified trie node is not in the trie, nil will be returned.
// If a trie node is not found in the database, a MissingNodeError is returned.
func (t *StateTrie) GetNode(path []byte) ([]byte, int, error) {
return t.trie.GetNode(path)
}
// MustUpdate 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.
//
// The value bytes must not be modified by the caller while they are
// stored in the trie.
//
// This function will omit any encountered error but just print out an
// error message.
func (t *StateTrie) MustUpdate(key, value []byte) {
hk := t.hashKey(key)
t.trie.MustUpdate(hk, value)
t.getSecKeyCache()[string(hk)] = common.CopyBytes(key)
}
// UpdateStorage 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.
//
// The value bytes must not be modified by the caller while they are
// stored in the trie.
//
// If a node is not found in the database, a MissingNodeError is returned.
func (t *StateTrie) UpdateStorage(_ common.Address, key, value []byte) error {
hk := t.hashKey(key)
v, _ := rlp.EncodeToBytes(value)
err := t.trie.Update(hk, v)
if err != nil {
return err
}
t.getSecKeyCache()[string(hk)] = common.CopyBytes(key)
return nil
}
// UpdateAccount will abstract the write of an account to the secure trie.
func (t *StateTrie) UpdateAccount(address common.Address, acc *types.StateAccount) error {
hk := t.hashKey(address.Bytes())
data, err := rlp.EncodeToBytes(acc)
if err != nil {
return err
}
if err := t.trie.Update(hk, data); err != nil {
return err
}
t.getSecKeyCache()[string(hk)] = address.Bytes()
return nil
}
func (t *StateTrie) UpdateContractCode(_ common.Address, _ common.Hash, _ []byte) error {
return nil
}
// MustDelete removes any existing value for key from the trie. This function
// will omit any encountered error but just print out an error message.
func (t *StateTrie) MustDelete(key []byte) {
hk := t.hashKey(key)
delete(t.getSecKeyCache(), string(hk))
t.trie.MustDelete(hk)
}
// DeleteStorage removes any existing storage slot from the trie.
// If the specified trie node is not in the trie, nothing will be changed.
// If a node is not found in the database, a MissingNodeError is returned.
func (t *StateTrie) DeleteStorage(_ common.Address, key []byte) error {
hk := t.hashKey(key)
delete(t.getSecKeyCache(), string(hk))
return t.trie.Delete(hk)
}
// DeleteAccount abstracts an account deletion from the trie.
func (t *StateTrie) DeleteAccount(address common.Address) error {
hk := t.hashKey(address.Bytes())
delete(t.getSecKeyCache(), string(hk))
return t.trie.Delete(hk)
}
// GetKey returns the sha3 preimage of a hashed key that was
// previously used to store a value.
func (t *StateTrie) GetKey(shaKey []byte) []byte {
if key, ok := t.getSecKeyCache()[string(shaKey)]; ok {
return key
}
if t.preimages == nil {
return nil
}
return t.preimages.preimage(common.BytesToHash(shaKey))
}
// Commit collects all dirty nodes in the trie and replaces them with the
// corresponding node hash. All collected nodes (including dirty leaves if
// collectLeaf is true) will be encapsulated into a nodeset for return.
// The returned nodeset can be nil if the trie is clean (nothing to commit).
// All cached preimages will be also flushed if preimages recording is enabled.
// Once the trie is committed, it's not usable anymore. A new trie must
// be created with new root and updated trie database for following usage
func (t *StateTrie) Commit(collectLeaf bool) (common.Hash, *trienode.NodeSet, error) {
// Write all the pre-images to the actual disk database
if len(t.getSecKeyCache()) > 0 {
if t.preimages != nil {
preimages := make(map[common.Hash][]byte)
for hk, key := range t.secKeyCache {
preimages[common.BytesToHash([]byte(hk))] = key
}
t.preimages.insertPreimage(preimages)
}
t.secKeyCache = make(map[string][]byte)
}
// Commit the trie and return its modified nodeset.
return t.trie.Commit(collectLeaf)
}
// Hash returns the root hash of StateTrie. It does not write to the
// database and can be used even if the trie doesn't have one.
func (t *StateTrie) Hash() common.Hash {
return t.trie.Hash()
}
// Copy returns a copy of StateTrie.
func (t *StateTrie) Copy() *StateTrie {
return &StateTrie{
trie: *t.trie.Copy(),
preimages: t.preimages,
secKeyCache: t.secKeyCache,
}
}
// NodeIterator returns an iterator that returns nodes of the underlying trie.
// Iteration starts at the key after the given start key.
func (t *StateTrie) NodeIterator(start []byte) (NodeIterator, error) {
return t.trie.NodeIterator(start)
}
// MustNodeIterator is a wrapper of NodeIterator and will omit any encountered
// error but just print out an error message.
func (t *StateTrie) MustNodeIterator(start []byte) NodeIterator {
return t.trie.MustNodeIterator(start)
}
// hashKey returns the hash of key as an ephemeral buffer.
// The caller must not hold onto the return value because it will become
// invalid on the next call to hashKey or secKey.
func (t *StateTrie) hashKey(key []byte) []byte {
h := newHasher(false)
h.sha.Reset()
h.sha.Write(key)
h.sha.Read(t.hashKeyBuf[:])
returnHasherToPool(h)
return t.hashKeyBuf[:]
}
// getSecKeyCache returns the current secure key cache, creating a new one if
// ownership changed (i.e. the current secure trie is a copy of another owning
// the actual cache).
func (t *StateTrie) getSecKeyCache() map[string][]byte {
if t != t.secKeyCacheOwner {
t.secKeyCacheOwner = t
t.secKeyCache = make(map[string][]byte)
}
return t.secKeyCache
}

149
trie/secure_trie_test.go
View file

@ -1,149 +0,0 @@
// Copyright 2015 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 trie
import (
"bytes"
"fmt"
"runtime"
"sync"
"testing"
"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/trie/trienode"
)
func newEmptySecure() *StateTrie {
trie, _ := NewStateTrie(TrieID(types.EmptyRootHash), NewDatabase(rawdb.NewMemoryDatabase(), nil))
return trie
}
// makeTestStateTrie creates a large enough secure trie for testing.
func makeTestStateTrie() (*Database, *StateTrie, map[string][]byte) {
// Create an empty trie
triedb := NewDatabase(rawdb.NewMemoryDatabase(), nil)
trie, _ := NewStateTrie(TrieID(types.EmptyRootHash), triedb)
// Fill it with some arbitrary data
content := make(map[string][]byte)
for i := byte(0); i < 255; i++ {
// Map the same data under multiple keys
key, val := common.LeftPadBytes([]byte{1, i}, 32), []byte{i}
content[string(key)] = val
trie.MustUpdate(key, val)
key, val = common.LeftPadBytes([]byte{2, i}, 32), []byte{i}
content[string(key)] = val
trie.MustUpdate(key, val)
// Add some other data to inflate the trie
for j := byte(3); j < 13; j++ {
key, val = common.LeftPadBytes([]byte{j, i}, 32), []byte{j, i}
content[string(key)] = val
trie.MustUpdate(key, val)
}
}
root, nodes, _ := trie.Commit(false)
if err := triedb.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil); err != nil {
panic(fmt.Errorf("failed to commit db %v", err))
}
// Re-create the trie based on the new state
trie, _ = NewStateTrie(TrieID(root), triedb)
return triedb, trie, content
}
func TestSecureDelete(t *testing.T) {
trie := newEmptySecure()
vals := []struct{ k, v string }{
{"do", "verb"},
{"ether", "wookiedoo"},
{"horse", "stallion"},
{"shaman", "horse"},
{"doge", "coin"},
{"ether", ""},
{"dog", "puppy"},
{"shaman", ""},
}
for _, val := range vals {
if val.v != "" {
trie.MustUpdate([]byte(val.k), []byte(val.v))
} else {
trie.MustDelete([]byte(val.k))
}
}
hash := trie.Hash()
exp := common.HexToHash("29b235a58c3c25ab83010c327d5932bcf05324b7d6b1185e650798034783ca9d")
if hash != exp {
t.Errorf("expected %x got %x", exp, hash)
}
}
func TestSecureGetKey(t *testing.T) {
trie := newEmptySecure()
trie.MustUpdate([]byte("foo"), []byte("bar"))
key := []byte("foo")
value := []byte("bar")
seckey := crypto.Keccak256(key)
if !bytes.Equal(trie.MustGet(key), value) {
t.Errorf("Get did not return bar")
}
if k := trie.GetKey(seckey); !bytes.Equal(k, key) {
t.Errorf("GetKey returned %q, want %q", k, key)
}
}
func TestStateTrieConcurrency(t *testing.T) {
// Create an initial trie and copy if for concurrent access
_, trie, _ := makeTestStateTrie()
threads := runtime.NumCPU()
tries := make([]*StateTrie, threads)
for i := 0; i < threads; i++ {
tries[i] = trie.Copy()
}
// Start a batch of goroutines interacting with the trie
pend := new(sync.WaitGroup)
pend.Add(threads)
for i := 0; i < threads; i++ {
go func(index int) {
defer pend.Done()
for j := byte(0); j < 255; j++ {
// Map the same data under multiple keys
key, val := common.LeftPadBytes([]byte{byte(index), 1, j}, 32), []byte{j}
tries[index].MustUpdate(key, val)
key, val = common.LeftPadBytes([]byte{byte(index), 2, j}, 32), []byte{j}
tries[index].MustUpdate(key, val)
// Add some other data to inflate the trie
for k := byte(3); k < 13; k++ {
key, val = common.LeftPadBytes([]byte{byte(index), k, j}, 32), []byte{k, j}
tries[index].MustUpdate(key, val)
}
}
tries[index].Commit(false)
}(i)
}
// Wait for all threads to finish
pend.Wait()
}

479
trie/stacktrie.go
View file

@ -1,479 +0,0 @@
// Copyright 2020 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 trie
import (
"bytes"
"errors"
"sync"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/metrics"
)
var (
stPool = sync.Pool{New: func() any { return new(stNode) }}
_ = types.TrieHasher((*StackTrie)(nil))
)
// StackTrieOptions contains the configured options for manipulating the stackTrie.
type StackTrieOptions struct {
Writer func(path []byte, hash common.Hash, blob []byte) // The function to commit the dirty nodes
Cleaner func(path []byte) // The function to clean up dangling nodes
SkipLeftBoundary bool // Flag whether the nodes on the left boundary are skipped for committing
SkipRightBoundary bool // Flag whether the nodes on the right boundary are skipped for committing
boundaryGauge metrics.Gauge // Gauge to track how many boundary nodes are met
}
// NewStackTrieOptions initializes an empty options for stackTrie.
func NewStackTrieOptions() *StackTrieOptions { return &StackTrieOptions{} }
// WithWriter configures trie node writer within the options.
func (o *StackTrieOptions) WithWriter(writer func(path []byte, hash common.Hash, blob []byte)) *StackTrieOptions {
o.Writer = writer
return o
}
// WithCleaner configures the cleaner in the option for removing dangling nodes.
func (o *StackTrieOptions) WithCleaner(cleaner func(path []byte)) *StackTrieOptions {
o.Cleaner = cleaner
return o
}
// WithSkipBoundary configures whether the left and right boundary nodes are
// filtered for committing, along with a gauge metrics to track how many
// boundary nodes are met.
func (o *StackTrieOptions) WithSkipBoundary(skipLeft, skipRight bool, gauge metrics.Gauge) *StackTrieOptions {
o.SkipLeftBoundary = skipLeft
o.SkipRightBoundary = skipRight
o.boundaryGauge = gauge
return o
}
// StackTrie is a trie implementation that expects keys to be inserted
// in order. Once it determines that a subtree will no longer be inserted
// into, it will hash it and free up the memory it uses.
type StackTrie struct {
options *StackTrieOptions
root *stNode
h *hasher
first []byte // The (hex-encoded without terminator) key of first inserted entry, tracked as left boundary.
last []byte // The (hex-encoded without terminator) key of last inserted entry, tracked as right boundary.
}
// NewStackTrie allocates and initializes an empty trie.
func NewStackTrie(options *StackTrieOptions) *StackTrie {
if options == nil {
options = NewStackTrieOptions()
}
return &StackTrie{
options: options,
root: stPool.Get().(*stNode),
h: newHasher(false),
}
}
// Update inserts a (key, value) pair into the stack trie.
func (t *StackTrie) Update(key, value []byte) error {
if len(value) == 0 {
return errors.New("trying to insert empty (deletion)")
}
k := keybytesToHex(key)
k = k[:len(k)-1] // chop the termination flag
if bytes.Compare(t.last, k) >= 0 {
return errors.New("non-ascending key order")
}
// track the first and last inserted entries.
if t.first == nil {
t.first = append([]byte{}, k...)
}
if t.last == nil {
t.last = append([]byte{}, k...) // allocate key slice
} else {
t.last = append(t.last[:0], k...) // reuse key slice
}
t.insert(t.root, k, value, nil)
return nil
}
// MustUpdate is a wrapper of Update and will omit any encountered error but
// just print out an error message.
func (t *StackTrie) MustUpdate(key, value []byte) {
if err := t.Update(key, value); err != nil {
log.Error("Unhandled trie error in StackTrie.Update", "err", err)
}
}
// Reset resets the stack trie object to empty state.
func (t *StackTrie) Reset() {
t.options = NewStackTrieOptions()
t.root = stPool.Get().(*stNode)
t.first = nil
t.last = nil
}
// stNode represents a node within a StackTrie
type stNode struct {
typ uint8 // node type (as in branch, ext, leaf)
key []byte // key chunk covered by this (leaf|ext) node
val []byte // value contained by this node if it's a leaf
children [16]*stNode // list of children (for branch and exts)
}
// newLeaf constructs a leaf node with provided node key and value. The key
// will be deep-copied in the function and safe to modify afterwards, but
// value is not.
func newLeaf(key, val []byte) *stNode {
st := stPool.Get().(*stNode)
st.typ = leafNode
st.key = append(st.key, key...)
st.val = val
return st
}
// newExt constructs an extension node with provided node key and child. The
// key will be deep-copied in the function and safe to modify afterwards.
func newExt(key []byte, child *stNode) *stNode {
st := stPool.Get().(*stNode)
st.typ = extNode
st.key = append(st.key, key...)
st.children[0] = child
return st
}
// List all values that stNode#nodeType can hold
const (
emptyNode = iota
branchNode
extNode
leafNode
hashedNode
)
func (n *stNode) reset() *stNode {
n.key = n.key[:0]
n.val = nil
for i := range n.children {
n.children[i] = nil
}
n.typ = emptyNode
return n
}
// Helper function that, given a full key, determines the index
// at which the chunk pointed by st.keyOffset is different from
// the same chunk in the full key.
func (n *stNode) getDiffIndex(key []byte) int {
for idx, nibble := range n.key {
if nibble != key[idx] {
return idx
}
}
return len(n.key)
}
// Helper function to that inserts a (key, value) pair into
// the trie.
func (t *StackTrie) insert(st *stNode, key, value []byte, path []byte) {
switch st.typ {
case branchNode: /* Branch */
idx := int(key[0])
// Unresolve elder siblings
for i := idx - 1; i >= 0; i-- {
if st.children[i] != nil {
if st.children[i].typ != hashedNode {
t.hash(st.children[i], append(path, byte(i)))
}
break
}
}
// Add new child
if st.children[idx] == nil {
st.children[idx] = newLeaf(key[1:], value)
} else {
t.insert(st.children[idx], key[1:], value, append(path, key[0]))
}
case extNode: /* Ext */
// Compare both key chunks and see where they differ
diffidx := st.getDiffIndex(key)
// Check if chunks are identical. If so, recurse into
// the child node. Otherwise, the key has to be split
// into 1) an optional common prefix, 2) the fullnode
// representing the two differing path, and 3) a leaf
// for each of the differentiated subtrees.
if diffidx == len(st.key) {
// Ext key and key segment are identical, recurse into
// the child node.
t.insert(st.children[0], key[diffidx:], value, append(path, key[:diffidx]...))
return
}
// Save the original part. Depending if the break is
// at the extension's last byte or not, create an
// intermediate extension or use the extension's child
// node directly.
var n *stNode
if diffidx < len(st.key)-1 {
// Break on the non-last byte, insert an intermediate
// extension. The path prefix of the newly-inserted
// extension should also contain the different byte.
n = newExt(st.key[diffidx+1:], st.children[0])
t.hash(n, append(path, st.key[:diffidx+1]...))
} else {
// Break on the last byte, no need to insert
// an extension node: reuse the current node.
// The path prefix of the original part should
// still be same.
n = st.children[0]
t.hash(n, append(path, st.key...))
}
var p *stNode
if diffidx == 0 {
// the break is on the first byte, so
// the current node is converted into
// a branch node.
st.children[0] = nil
p = st
st.typ = branchNode
} else {
// the common prefix is at least one byte
// long, insert a new intermediate branch
// node.
st.children[0] = stPool.Get().(*stNode)
st.children[0].typ = branchNode
p = st.children[0]
}
// Create a leaf for the inserted part
o := newLeaf(key[diffidx+1:], value)
// Insert both child leaves where they belong:
origIdx := st.key[diffidx]
newIdx := key[diffidx]
p.children[origIdx] = n
p.children[newIdx] = o
st.key = st.key[:diffidx]
case leafNode: /* Leaf */
// Compare both key chunks and see where they differ
diffidx := st.getDiffIndex(key)
// Overwriting a key isn't supported, which means that
// the current leaf is expected to be split into 1) an
// optional extension for the common prefix of these 2
// keys, 2) a fullnode selecting the path on which the
// keys differ, and 3) one leaf for the differentiated
// component of each key.
if diffidx >= len(st.key) {
panic("Trying to insert into existing key")
}
// Check if the split occurs at the first nibble of the
// chunk. In that case, no prefix extnode is necessary.
// Otherwise, create that
var p *stNode
if diffidx == 0 {
// Convert current leaf into a branch
st.typ = branchNode
p = st
st.children[0] = nil
} else {
// Convert current node into an ext,
// and insert a child branch node.
st.typ = extNode
st.children[0] = stPool.Get().(*stNode)
st.children[0].typ = branchNode
p = st.children[0]
}
// Create the two child leaves: one containing the original
// value and another containing the new value. The child leaf
// is hashed directly in order to free up some memory.
origIdx := st.key[diffidx]
p.children[origIdx] = newLeaf(st.key[diffidx+1:], st.val)
t.hash(p.children[origIdx], append(path, st.key[:diffidx+1]...))
newIdx := key[diffidx]
p.children[newIdx] = newLeaf(key[diffidx+1:], value)
// Finally, cut off the key part that has been passed
// over to the children.
st.key = st.key[:diffidx]
st.val = nil
case emptyNode: /* Empty */
st.typ = leafNode
st.key = key
st.val = value
case hashedNode:
panic("trying to insert into hash")
default:
panic("invalid type")
}
}
// hash converts st into a 'hashedNode', if possible. Possible outcomes:
//
// 1. The rlp-encoded value was >= 32 bytes:
// - Then the 32-byte `hash` will be accessible in `st.val`.
// - And the 'st.type' will be 'hashedNode'
//
// 2. The rlp-encoded value was < 32 bytes
// - Then the <32 byte rlp-encoded value will be accessible in 'st.val'.
// - And the 'st.type' will be 'hashedNode' AGAIN
//
// This method also sets 'st.type' to hashedNode, and clears 'st.key'.
func (t *StackTrie) hash(st *stNode, path []byte) {
var (
blob []byte // RLP-encoded node blob
internal [][]byte // List of node paths covered by the extension node
)
switch st.typ {
case hashedNode:
return
case emptyNode:
st.val = types.EmptyRootHash.Bytes()
st.key = st.key[:0]
st.typ = hashedNode
return
case branchNode:
var nodes fullNode
for i, child := range st.children {
if child == nil {
nodes.Children[i] = nilValueNode
continue
}
t.hash(child, append(path, byte(i)))
if len(child.val) < 32 {
nodes.Children[i] = rawNode(child.val)
} else {
nodes.Children[i] = hashNode(child.val)
}
st.children[i] = nil
stPool.Put(child.reset()) // Release child back to pool.
}
nodes.encode(t.h.encbuf)
blob = t.h.encodedBytes()
case extNode:
// recursively hash and commit child as the first step
t.hash(st.children[0], append(path, st.key...))
// Collect the path of internal nodes between shortNode and its **in disk**
// child. This is essential in the case of path mode scheme to avoid leaving
// danging nodes within the range of this internal path on disk, which would
// break the guarantee for state healing.
if len(st.children[0].val) >= 32 && t.options.Cleaner != nil {
for i := 1; i < len(st.key); i++ {
internal = append(internal, append(path, st.key[:i]...))
}
}
// encode the extension node
n := shortNode{Key: hexToCompactInPlace(st.key)}
if len(st.children[0].val) < 32 {
n.Val = rawNode(st.children[0].val)
} else {
n.Val = hashNode(st.children[0].val)
}
n.encode(t.h.encbuf)
blob = t.h.encodedBytes()
stPool.Put(st.children[0].reset()) // Release child back to pool.
st.children[0] = nil
case leafNode:
st.key = append(st.key, byte(16))
n := shortNode{Key: hexToCompactInPlace(st.key), Val: valueNode(st.val)}
n.encode(t.h.encbuf)
blob = t.h.encodedBytes()
default:
panic("invalid node type")
}
st.typ = hashedNode
st.key = st.key[:0]
// Skip committing the non-root node if the size is smaller than 32 bytes.
if len(blob) < 32 && len(path) > 0 {
st.val = common.CopyBytes(blob)
return
}
// Write the hash to the 'val'. We allocate a new val here to not mutate
// input values.
st.val = t.h.hashData(blob)
// Short circuit if the stack trie is not configured for writing.
if t.options.Writer == nil {
return
}
// Skip committing if the node is on the left boundary and stackTrie is
// configured to filter the boundary.
if t.options.SkipLeftBoundary && bytes.HasPrefix(t.first, path) {
if t.options.boundaryGauge != nil {
t.options.boundaryGauge.Inc(1)
}
return
}
// Skip committing if the node is on the right boundary and stackTrie is
// configured to filter the boundary.
if t.options.SkipRightBoundary && bytes.HasPrefix(t.last, path) {
if t.options.boundaryGauge != nil {
t.options.boundaryGauge.Inc(1)
}
return
}
// Clean up the internal dangling nodes covered by the extension node.
// This should be done before writing the node to adhere to the committing
// order from bottom to top.
for _, path := range internal {
t.options.Cleaner(path)
}
t.options.Writer(path, common.BytesToHash(st.val), blob)
}
// Hash will firstly hash the entire trie if it's still not hashed and then commit
// all nodes to the associated database. Actually most of the trie nodes have been
// committed already. The main purpose here is to commit the nodes on right boundary.
//
// For stack trie, Hash and Commit are functionally identical.
func (t *StackTrie) Hash() common.Hash {
n := t.root
t.hash(n, nil)
return common.BytesToHash(n.val)
}
// Commit will firstly hash the entire trie if it's still not hashed and then commit
// all nodes to the associated database. Actually most of the trie nodes have been
// committed already. The main purpose here is to commit the nodes on right boundary.
//
// For stack trie, Hash and Commit are functionally identical.
func (t *StackTrie) Commit() common.Hash {
return t.Hash()
}

155
trie/stacktrie_fuzzer_test.go
View file

@ -1,155 +0,0 @@
// Copyright 2020 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 trie
import (
"bytes"
"encoding/binary"
"fmt"
"testing"
"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/trie/trienode"
"golang.org/x/crypto/sha3"
"golang.org/x/exp/slices"
)
func FuzzStackTrie(f *testing.F) {
f.Fuzz(func(t *testing.T, data []byte) {
fuzz(data, false)
})
}
func fuzz(data []byte, debugging bool) {
// This spongeDb is used to check the sequence of disk-db-writes
var (
input = bytes.NewReader(data)
spongeA = &spongeDb{sponge: sha3.NewLegacyKeccak256()}
dbA = NewDatabase(rawdb.NewDatabase(spongeA), nil)
trieA = NewEmpty(dbA)
spongeB = &spongeDb{sponge: sha3.NewLegacyKeccak256()}
dbB = NewDatabase(rawdb.NewDatabase(spongeB), nil)
options = NewStackTrieOptions().WithWriter(func(path []byte, hash common.Hash, blob []byte) {
rawdb.WriteTrieNode(spongeB, common.Hash{}, path, hash, blob, dbB.Scheme())
})
trieB = NewStackTrie(options)
vals []*kv
maxElements = 10000
// operate on unique keys only
keys = make(map[string]struct{})
)
// Fill the trie with elements
for i := 0; input.Len() > 0 && i < maxElements; i++ {
k := make([]byte, 32)
input.Read(k)
var a uint16
binary.Read(input, binary.LittleEndian, &a)
a = 1 + a%100
v := make([]byte, a)
input.Read(v)
if input.Len() == 0 {
// If it was exhausted while reading, the value may be all zeroes,
// thus 'deletion' which is not supported on stacktrie
break
}
if _, present := keys[string(k)]; present {
// This key is a duplicate, ignore it
continue
}
keys[string(k)] = struct{}{}
vals = append(vals, &kv{k: k, v: v})
trieA.MustUpdate(k, v)
}
if len(vals) == 0 {
return
}
// Flush trie -> database
rootA, nodes, err := trieA.Commit(false)
if err != nil {
panic(err)
}
if nodes != nil {
dbA.Update(rootA, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
}
// Flush memdb -> disk (sponge)
dbA.Commit(rootA, false)
// Stacktrie requires sorted insertion
slices.SortFunc(vals, (*kv).cmp)
for _, kv := range vals {
if debugging {
fmt.Printf("{\"%#x\" , \"%#x\"} // stacktrie.Update\n", kv.k, kv.v)
}
trieB.MustUpdate(kv.k, kv.v)
}
rootB := trieB.Hash()
trieB.Commit()
if rootA != rootB {
panic(fmt.Sprintf("roots differ: (trie) %x != %x (stacktrie)", rootA, rootB))
}
sumA := spongeA.sponge.Sum(nil)
sumB := spongeB.sponge.Sum(nil)
if !bytes.Equal(sumA, sumB) {
panic(fmt.Sprintf("sequence differ: (trie) %x != %x (stacktrie)", sumA, sumB))
}
// Ensure all the nodes are persisted correctly
var (
nodeset = make(map[string][]byte) // path -> blob
optionsC = NewStackTrieOptions().WithWriter(func(path []byte, hash common.Hash, blob []byte) {
if crypto.Keccak256Hash(blob) != hash {
panic("invalid node blob")
}
nodeset[string(path)] = common.CopyBytes(blob)
})
trieC = NewStackTrie(optionsC)
checked int
)
for _, kv := range vals {
trieC.MustUpdate(kv.k, kv.v)
}
rootC := trieC.Commit()
if rootA != rootC {
panic(fmt.Sprintf("roots differ: (trie) %x != %x (stacktrie)", rootA, rootC))
}
trieA, _ = New(TrieID(rootA), dbA)
iterA := trieA.MustNodeIterator(nil)
for iterA.Next(true) {
if iterA.Hash() == (common.Hash{}) {
if _, present := nodeset[string(iterA.Path())]; present {
panic("unexpected tiny node")
}
continue
}
nodeBlob, present := nodeset[string(iterA.Path())]
if !present {
panic("missing node")
}
if !bytes.Equal(nodeBlob, iterA.NodeBlob()) {
panic("node blob is not matched")
}
checked += 1
}
if checked != len(nodeset) {
panic("node number is not matched")
}
}

487
trie/stacktrie_test.go
View file

@ -1,487 +0,0 @@
// Copyright 2020 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 trie
import (
"bytes"
"math/big"
"math/rand"
"testing"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/trie/testutil"
"github.com/stretchr/testify/assert"
"golang.org/x/exp/slices"
)
func TestStackTrieInsertAndHash(t *testing.T) {
type KeyValueHash struct {
K string // Hex string for key.
V string // Value, directly converted to bytes.
H string // Expected root hash after insert of (K, V) to an existing trie.
}
tests := [][]KeyValueHash{
{ // {0:0, 7:0, f:0}
{"00", "v_______________________0___0", "5cb26357b95bb9af08475be00243ceb68ade0b66b5cd816b0c18a18c612d2d21"},
{"70", "v_______________________0___1", "8ff64309574f7a437a7ad1628e690eb7663cfde10676f8a904a8c8291dbc1603"},
{"f0", "v_______________________0___2", "9e3a01bd8d43efb8e9d4b5506648150b8e3ed1caea596f84ee28e01a72635470"},
},
{ // {1:0cc, e:{1:fc, e:fc}}
{"10cc", "v_______________________1___0", "233e9b257843f3dfdb1cce6676cdaf9e595ac96ee1b55031434d852bc7ac9185"},
{"e1fc", "v_______________________1___1", "39c5e908ae83d0c78520c7c7bda0b3782daf594700e44546e93def8f049cca95"},
{"eefc", "v_______________________1___2", "d789567559fd76fe5b7d9cc42f3750f942502ac1c7f2a466e2f690ec4b6c2a7c"},
},
{ // {b:{a:ac, b:ac}, d:acc}
{"baac", "v_______________________2___0", "8be1c86ba7ec4c61e14c1a9b75055e0464c2633ae66a055a24e75450156a5d42"},
{"bbac", "v_______________________2___1", "8495159b9895a7d88d973171d737c0aace6fe6ac02a4769fff1bc43bcccce4cc"},
{"dacc", "v_______________________2___2", "9bcfc5b220a27328deb9dc6ee2e3d46c9ebc9c69e78acda1fa2c7040602c63ca"},
},
{ // {0:0cccc, 2:456{0:0, 2:2}
{"00cccc", "v_______________________3___0", "e57dc2785b99ce9205080cb41b32ebea7ac3e158952b44c87d186e6d190a6530"},
{"245600", "v_______________________3___1", "0335354adbd360a45c1871a842452287721b64b4234dfe08760b243523c998db"},
{"245622", "v_______________________3___2", "9e6832db0dca2b5cf81c0e0727bfde6afc39d5de33e5720bccacc183c162104e"},
},
{ // {1:4567{1:1c, 3:3c}, 3:0cccccc}
{"1456711c", "v_______________________4___0", "f2389e78d98fed99f3e63d6d1623c1d4d9e8c91cb1d585de81fbc7c0e60d3529"},
{"1456733c", "v_______________________4___1", "101189b3fab852be97a0120c03d95eefcf984d3ed639f2328527de6def55a9c0"},
{"30cccccc", "v_______________________4___2", "3780ce111f98d15751dfde1eb21080efc7d3914b429e5c84c64db637c55405b3"},
},
{ // 8800{1:f, 2:e, 3:d}
{"88001f", "v_______________________5___0", "e817db50d84f341d443c6f6593cafda093fc85e773a762421d47daa6ac993bd5"},
{"88002e", "v_______________________5___1", "d6e3e6047bdc110edd296a4d63c030aec451bee9d8075bc5a198eee8cda34f68"},
{"88003d", "v_______________________5___2", "b6bdf8298c703342188e5f7f84921a402042d0e5fb059969dd53a6b6b1fb989e"},
},
{ // 0{1:fc, 2:ec, 4:dc}
{"01fc", "v_______________________6___0", "693268f2ca80d32b015f61cd2c4dba5a47a6b52a14c34f8e6945fad684e7a0d5"},
{"02ec", "v_______________________6___1", "e24ddd44469310c2b785a2044618874bf486d2f7822603a9b8dce58d6524d5de"},
{"04dc", "v_______________________6___2", "33fc259629187bbe54b92f82f0cd8083b91a12e41a9456b84fc155321e334db7"},
},
{ // f{0:fccc, f:ff{0:f, f:f}}
{"f0fccc", "v_______________________7___0", "b0966b5aa469a3e292bc5fcfa6c396ae7a657255eef552ea7e12f996de795b90"},
{"ffff0f", "v_______________________7___1", "3b1ca154ec2a3d96d8d77bddef0abfe40a53a64eb03cecf78da9ec43799fa3d0"},
{"ffffff", "v_______________________7___2", "e75463041f1be8252781be0ace579a44ea4387bf5b2739f4607af676f7719678"},
},
{ // ff{0:f{0:f, f:f}, f:fcc}
{"ff0f0f", "v_______________________8___0", "0928af9b14718ec8262ab89df430f1e5fbf66fac0fed037aff2b6767ae8c8684"},
{"ff0fff", "v_______________________8___1", "d870f4d3ce26b0bf86912810a1960693630c20a48ba56be0ad04bc3e9ddb01e6"},
{"ffffcc", "v_______________________8___2", "4239f10dd9d9915ecf2e047d6a576bdc1733ed77a30830f1bf29deaf7d8e966f"},
},
{
{"123d", "x___________________________0", "fc453d88b6f128a77c448669710497380fa4588abbea9f78f4c20c80daa797d0"},
{"123e", "x___________________________1", "5af48f2d8a9a015c1ff7fa8b8c7f6b676233bd320e8fb57fd7933622badd2cec"},
{"123f", "x___________________________2", "1164d7299964e74ac40d761f9189b2a3987fae959800d0f7e29d3aaf3eae9e15"},
},
{
{"123d", "x___________________________0", "fc453d88b6f128a77c448669710497380fa4588abbea9f78f4c20c80daa797d0"},
{"123e", "x___________________________1", "5af48f2d8a9a015c1ff7fa8b8c7f6b676233bd320e8fb57fd7933622badd2cec"},
{"124a", "x___________________________2", "661a96a669869d76b7231380da0649d013301425fbea9d5c5fae6405aa31cfce"},
},
{
{"123d", "x___________________________0", "fc453d88b6f128a77c448669710497380fa4588abbea9f78f4c20c80daa797d0"},
{"123e", "x___________________________1", "5af48f2d8a9a015c1ff7fa8b8c7f6b676233bd320e8fb57fd7933622badd2cec"},
{"13aa", "x___________________________2", "6590120e1fd3ffd1a90e8de5bb10750b61079bb0776cca4414dd79a24e4d4356"},
},
{
{"123d", "x___________________________0", "fc453d88b6f128a77c448669710497380fa4588abbea9f78f4c20c80daa797d0"},
{"123e", "x___________________________1", "5af48f2d8a9a015c1ff7fa8b8c7f6b676233bd320e8fb57fd7933622badd2cec"},
{"2aaa", "x___________________________2", "f869b40e0c55eace1918332ef91563616fbf0755e2b946119679f7ef8e44b514"},
},
{
{"1234da", "x___________________________0", "1c4b4462e9f56a80ca0f5d77c0d632c41b0102290930343cf1791e971a045a79"},
{"1234ea", "x___________________________1", "2f502917f3ba7d328c21c8b45ee0f160652e68450332c166d4ad02d1afe31862"},
{"1234fa", "x___________________________2", "4f4e368ab367090d5bc3dbf25f7729f8bd60df84de309b4633a6b69ab66142c0"},
},
{
{"1234da", "x___________________________0", "1c4b4462e9f56a80ca0f5d77c0d632c41b0102290930343cf1791e971a045a79"},
{"1234ea", "x___________________________1", "2f502917f3ba7d328c21c8b45ee0f160652e68450332c166d4ad02d1afe31862"},
{"1235aa", "x___________________________2", "21840121d11a91ac8bbad9a5d06af902a5c8d56a47b85600ba813814b7bfcb9b"},
},
{
{"1234da", "x___________________________0", "1c4b4462e9f56a80ca0f5d77c0d632c41b0102290930343cf1791e971a045a79"},
{"1234ea", "x___________________________1", "2f502917f3ba7d328c21c8b45ee0f160652e68450332c166d4ad02d1afe31862"},
{"124aaa", "x___________________________2", "ea4040ddf6ae3fbd1524bdec19c0ab1581015996262006632027fa5cf21e441e"},
},
{
{"1234da", "x___________________________0", "1c4b4462e9f56a80ca0f5d77c0d632c41b0102290930343cf1791e971a045a79"},
{"1234ea", "x___________________________1", "2f502917f3ba7d328c21c8b45ee0f160652e68450332c166d4ad02d1afe31862"},
{"13aaaa", "x___________________________2", "e4beb66c67e44f2dd8ba36036e45a44ff68f8d52942472b1911a45f886a34507"},
},
{
{"1234da", "x___________________________0", "1c4b4462e9f56a80ca0f5d77c0d632c41b0102290930343cf1791e971a045a79"},
{"1234ea", "x___________________________1", "2f502917f3ba7d328c21c8b45ee0f160652e68450332c166d4ad02d1afe31862"},
{"2aaaaa", "x___________________________2", "5f5989b820ff5d76b7d49e77bb64f26602294f6c42a1a3becc669cd9e0dc8ec9"},
},
{
{"000000", "x___________________________0", "3b32b7af0bddc7940e7364ee18b5a59702c1825e469452c8483b9c4e0218b55a"},
{"1234da", "x___________________________1", "3ab152a1285dca31945566f872c1cc2f17a770440eda32aeee46a5e91033dde2"},
{"1234ea", "x___________________________2", "0cccc87f96ddef55563c1b3be3c64fff6a644333c3d9cd99852cb53b6412b9b8"},
{"1234fa", "x___________________________3", "65bb3aafea8121111d693ffe34881c14d27b128fd113fa120961f251fe28428d"},
},
{
{"000000", "x___________________________0", "3b32b7af0bddc7940e7364ee18b5a59702c1825e469452c8483b9c4e0218b55a"},
{"1234da", "x___________________________1", "3ab152a1285dca31945566f872c1cc2f17a770440eda32aeee46a5e91033dde2"},
{"1234ea", "x___________________________2", "0cccc87f96ddef55563c1b3be3c64fff6a644333c3d9cd99852cb53b6412b9b8"},
{"1235aa", "x___________________________3", "f670e4d2547c533c5f21e0045442e2ecb733f347ad6d29ef36e0f5ba31bb11a8"},
},
{
{"000000", "x___________________________0", "3b32b7af0bddc7940e7364ee18b5a59702c1825e469452c8483b9c4e0218b55a"},
{"1234da", "x___________________________1", "3ab152a1285dca31945566f872c1cc2f17a770440eda32aeee46a5e91033dde2"},
{"1234ea", "x___________________________2", "0cccc87f96ddef55563c1b3be3c64fff6a644333c3d9cd99852cb53b6412b9b8"},
{"124aaa", "x___________________________3", "c17464123050a9a6f29b5574bb2f92f6d305c1794976b475b7fb0316b6335598"},
},
{
{"000000", "x___________________________0", "3b32b7af0bddc7940e7364ee18b5a59702c1825e469452c8483b9c4e0218b55a"},
{"1234da", "x___________________________1", "3ab152a1285dca31945566f872c1cc2f17a770440eda32aeee46a5e91033dde2"},
{"1234ea", "x___________________________2", "0cccc87f96ddef55563c1b3be3c64fff6a644333c3d9cd99852cb53b6412b9b8"},
{"13aaaa", "x___________________________3", "aa8301be8cb52ea5cd249f5feb79fb4315ee8de2140c604033f4b3fff78f0105"},
},
{
{"0000", "x___________________________0", "cb8c09ad07ae882136f602b3f21f8733a9f5a78f1d2525a8d24d1c13258000b2"},
{"123d", "x___________________________1", "8f09663deb02f08958136410dc48565e077f76bb6c9d8c84d35fc8913a657d31"},
{"123e", "x___________________________2", "0d230561e398c579e09a9f7b69ceaf7d3970f5a436fdb28b68b7a37c5bdd6b80"},
{"123f", "x___________________________3", "80f7bad1893ca57e3443bb3305a517723a74d3ba831bcaca22a170645eb7aafb"},
},
{
{"0000", "x___________________________0", "cb8c09ad07ae882136f602b3f21f8733a9f5a78f1d2525a8d24d1c13258000b2"},
{"123d", "x___________________________1", "8f09663deb02f08958136410dc48565e077f76bb6c9d8c84d35fc8913a657d31"},
{"123e", "x___________________________2", "0d230561e398c579e09a9f7b69ceaf7d3970f5a436fdb28b68b7a37c5bdd6b80"},
{"124a", "x___________________________3", "383bc1bb4f019e6bc4da3751509ea709b58dd1ac46081670834bae072f3e9557"},
},
{
{"0000", "x___________________________0", "cb8c09ad07ae882136f602b3f21f8733a9f5a78f1d2525a8d24d1c13258000b2"},
{"123d", "x___________________________1", "8f09663deb02f08958136410dc48565e077f76bb6c9d8c84d35fc8913a657d31"},
{"123e", "x___________________________2", "0d230561e398c579e09a9f7b69ceaf7d3970f5a436fdb28b68b7a37c5bdd6b80"},
{"13aa", "x___________________________3", "ff0dc70ce2e5db90ee42a4c2ad12139596b890e90eb4e16526ab38fa465b35cf"},
},
{ // branch node with short values
{"01", "a", "b48605025f5f4b129d40a420e721aa7d504487f015fce85b96e52126365ef7dc"},
{"80", "b", "2dc6b680daf74db067cb7aeaad73265ded93d96fce190fcbf64f498d475672ab"},
{"ee", "c", "017dc705a54ac5328dd263fa1bae68d655310fb3e3f7b7bc57e9a43ddf99c4bf"},
{"ff", "d", "bd5a3584d271d459bd4eb95247b2fc88656b3671b60c1125ffe7bc0b689470d0"},
},
{ // ext node with short branch node, then becoming long
{"a0", "a", "a83e028cb1e4365935661a9fd36a5c65c30b9ab416eaa877424146ca2a69d088"},
{"a1", "b", "f586a4639b07b01798ca65e05c253b75d51135ebfbf6f8d6e87c0435089e65f0"},
{"a2", "c", "63e297c295c008e09a8d531e18d57f270b6bc403e23179b915429db948cd62e3"},
{"a3", "d", "94a7b721535578e9381f1f4e4b6ec29f8bdc5f0458a30320684c562f5d47b4b5"},
{"a4", "e", "4b7e66d1c81965cdbe8fab8295ef56bc57fefdc5733d4782d2f8baf630f083c6"},
{"a5", "f", "2997e7b502198ce1783b5277faacf52b25844fb55a99b63e88bdbbafac573106"},
{"a6", "g", "bee629dd27a40772b2e1a67ec6db270d26acdf8d3b674dfae27866ad6ae1f48b"},
},
{ // branch node with short values, then long ones
{"a001", "v1", "b9cc982d995392b51e6787f1915f0b88efd4ad8b30f138da0a3e2242f2323e35"},
{"b002", "v2", "a7b474bc77ef5097096fa0ee6298fdae8928c0bc3724e7311cd0fa9ed1942fc7"},
{"c003", "v___________________________3", "dceb5bb7c92b0e348df988a8d9fc36b101397e38ebd405df55ba6ee5f14a264a"},
{"d004", "v___________________________4", "36e60ecb86b9626165e1c6543c42ecbe4d83bca58e8e1124746961511fce362a"},
},
{ // ext node to branch node with short values, then long ones
{"8002", "v1", "3258fcb3e9e7d7234ecd3b8d4743999e4ab3a21592565e0a5ca64c141e8620d9"},
{"8004", "v2", "b6cb95b7024a83c17624a3c9bed09b4b5e8ed426f49f54b8ad13c39028b1e75a"},
{"8008", "v___________________________3", "c769d82963abe6f0900bf69754738eeb2f84559777cfa87a44f54e1aab417871"},
{"800d", "v___________________________4", "1cad1fdaab1a6fa95d7b780fd680030e423eb76669971368ba04797a8d9cdfc9"},
},
{ // ext node with a child of size 31 (Y) and branch node with a child of size 31 (X)
{"000001", "ZZZZZZZZZ", "cef154b87c03c563408520ff9b26923c360cbc3ddb590c079bedeeb25a8c9c77"},
{"000002", "Y", "2130735e600f612f6e657a32bd7be64ddcaec6512c5694844b19de713922895d"},
{"000003", "XXXXXXXXXXXXXXXXXXXXXXXXXXXX", "962c0fffdeef7612a4f7bff1950d67e3e81c878e48b9ae45b3b374253b050bd8"},
},
}
for i, test := range tests {
// The StackTrie does not allow Insert(), Hash(), Insert(), ...
// so we will create new trie for every sequence length of inserts.
for l := 1; l <= len(test); l++ {
st := NewStackTrie(nil)
for j := 0; j < l; j++ {
kv := &test[j]
if err := st.Update(common.FromHex(kv.K), []byte(kv.V)); err != nil {
t.Fatal(err)
}
}
expected := common.HexToHash(test[l-1].H)
if h := st.Hash(); h != expected {
t.Errorf("%d(%d): root hash mismatch: %x, expected %x", i, l, h, expected)
}
}
}
}
func TestSizeBug(t *testing.T) {
st := NewStackTrie(nil)
nt := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
leaf := common.FromHex("290decd9548b62a8d60345a988386fc84ba6bc95484008f6362f93160ef3e563")
value := common.FromHex("94cf40d0d2b44f2b66e07cace1372ca42b73cf21a3")
nt.Update(leaf, value)
st.Update(leaf, value)
if nt.Hash() != st.Hash() {
t.Fatalf("error %x != %x", st.Hash(), nt.Hash())
}
}
func TestEmptyBug(t *testing.T) {
st := NewStackTrie(nil)
nt := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
//leaf := common.FromHex("290decd9548b62a8d60345a988386fc84ba6bc95484008f6362f93160ef3e563")
//value := common.FromHex("94cf40d0d2b44f2b66e07cace1372ca42b73cf21a3")
kvs := []struct {
K string
V string
}{
{K: "405787fa12a823e0f2b7631cc41b3ba8828b3321ca811111fa75cd3aa3bb5ace", V: "9496f4ec2bf9dab484cac6be589e8417d84781be08"},
{K: "40edb63a35fcf86c08022722aa3287cdd36440d671b4918131b2514795fefa9c", V: "01"},
{K: "b10e2d527612073b26eecdfd717e6a320cf44b4afac2b0732d9fcbe2b7fa0cf6", V: "947a30f7736e48d6599356464ba4c150d8da0302ff"},
{K: "c2575a0e9e593c00f959f8c92f12db2869c3395a3b0502d05e2516446f71f85b", V: "02"},
}
for _, kv := range kvs {
nt.Update(common.FromHex(kv.K), common.FromHex(kv.V))
st.Update(common.FromHex(kv.K), common.FromHex(kv.V))
}
if nt.Hash() != st.Hash() {
t.Fatalf("error %x != %x", st.Hash(), nt.Hash())
}
}
func TestValLength56(t *testing.T) {
st := NewStackTrie(nil)
nt := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
//leaf := common.FromHex("290decd9548b62a8d60345a988386fc84ba6bc95484008f6362f93160ef3e563")
//value := common.FromHex("94cf40d0d2b44f2b66e07cace1372ca42b73cf21a3")
kvs := []struct {
K string
V string
}{
{K: "405787fa12a823e0f2b7631cc41b3ba8828b3321ca811111fa75cd3aa3bb5ace", V: "1111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111111"},
}
for _, kv := range kvs {
nt.Update(common.FromHex(kv.K), common.FromHex(kv.V))
st.Update(common.FromHex(kv.K), common.FromHex(kv.V))
}
if nt.Hash() != st.Hash() {
t.Fatalf("error %x != %x", st.Hash(), nt.Hash())
}
}
// TestUpdateSmallNodes tests a case where the leaves are small (both key and value),
// which causes a lot of node-within-node. This case was found via fuzzing.
func TestUpdateSmallNodes(t *testing.T) {
st := NewStackTrie(nil)
nt := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
kvs := []struct {
K string
V string
}{
{"63303030", "3041"}, // stacktrie.Update
{"65", "3000"}, // stacktrie.Update
}
for _, kv := range kvs {
nt.Update(common.FromHex(kv.K), common.FromHex(kv.V))
st.Update(common.FromHex(kv.K), common.FromHex(kv.V))
}
if nt.Hash() != st.Hash() {
t.Fatalf("error %x != %x", st.Hash(), nt.Hash())
}
}
// TestUpdateVariableKeys contains a case which stacktrie fails: when keys of different
// sizes are used, and the second one has the same prefix as the first, then the
// stacktrie fails, since it's unable to 'expand' on an already added leaf.
// For all practical purposes, this is fine, since keys are fixed-size length
// in account and storage tries.
//
// The test is marked as 'skipped', and exists just to have the behaviour documented.
// This case was found via fuzzing.
func TestUpdateVariableKeys(t *testing.T) {
t.SkipNow()
st := NewStackTrie(nil)
nt := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
kvs := []struct {
K string
V string
}{
{"0x33303534636532393561313031676174", "303030"},
{"0x3330353463653239356131303167617430", "313131"},
}
for _, kv := range kvs {
nt.Update(common.FromHex(kv.K), common.FromHex(kv.V))
st.Update(common.FromHex(kv.K), common.FromHex(kv.V))
}
if nt.Hash() != st.Hash() {
t.Fatalf("error %x != %x", st.Hash(), nt.Hash())
}
}
// TestStacktrieNotModifyValues checks that inserting blobs of data into the
// stacktrie does not mutate the blobs
func TestStacktrieNotModifyValues(t *testing.T) {
st := NewStackTrie(nil)
{ // Test a very small trie
// Give it the value as a slice with large backing alloc,
// so if the stacktrie tries to append, it won't have to realloc
value := make([]byte, 1, 100)
value[0] = 0x2
want := common.CopyBytes(value)
st.Update([]byte{0x01}, value)
st.Hash()
if have := value; !bytes.Equal(have, want) {
t.Fatalf("tiny trie: have %#x want %#x", have, want)
}
st = NewStackTrie(nil)
}
// Test with a larger trie
keyB := big.NewInt(1)
keyDelta := big.NewInt(1)
var vals [][]byte
getValue := func(i int) []byte {
if i%2 == 0 { // large
return crypto.Keccak256(big.NewInt(int64(i)).Bytes())
} else { //small
return big.NewInt(int64(i)).Bytes()
}
}
for i := 0; i < 1000; i++ {
key := common.BigToHash(keyB)
value := getValue(i)
st.Update(key.Bytes(), value)
vals = append(vals, value)
keyB = keyB.Add(keyB, keyDelta)
keyDelta.Add(keyDelta, common.Big1)
}
st.Hash()
for i := 0; i < 1000; i++ {
want := getValue(i)
have := vals[i]
if !bytes.Equal(have, want) {
t.Fatalf("item %d, have %#x want %#x", i, have, want)
}
}
}
func buildPartialTree(entries []*kv, t *testing.T) map[string]common.Hash {
var (
options = NewStackTrieOptions()
nodes = make(map[string]common.Hash)
)
var (
first int
last = len(entries) - 1
noLeft bool
noRight bool
)
// Enter split mode if there are at least two elements
if rand.Intn(5) != 0 {
for {
first = rand.Intn(len(entries))
last = rand.Intn(len(entries))
if first <= last {
break
}
}
if first != 0 {
noLeft = true
}
if last != len(entries)-1 {
noRight = true
}
}
options = options.WithSkipBoundary(noLeft, noRight, nil)
options = options.WithWriter(func(path []byte, hash common.Hash, blob []byte) {
nodes[string(path)] = hash
})
tr := NewStackTrie(options)
for i := first; i <= last; i++ {
tr.MustUpdate(entries[i].k, entries[i].v)
}
tr.Commit()
return nodes
}
func TestPartialStackTrie(t *testing.T) {
for round := 0; round < 100; round++ {
var (
n = rand.Intn(100) + 1
entries []*kv
)
for i := 0; i < n; i++ {
var val []byte
if rand.Intn(3) == 0 {
val = testutil.RandBytes(3)
} else {
val = testutil.RandBytes(32)
}
entries = append(entries, &kv{
k: testutil.RandBytes(32),
v: val,
})
}
slices.SortFunc(entries, (*kv).cmp)
var (
nodes = make(map[string]common.Hash)
options = NewStackTrieOptions().WithWriter(func(path []byte, hash common.Hash, blob []byte) {
nodes[string(path)] = hash
})
)
tr := NewStackTrie(options)
for i := 0; i < len(entries); i++ {
tr.MustUpdate(entries[i].k, entries[i].v)
}
tr.Commit()
for j := 0; j < 100; j++ {
for path, hash := range buildPartialTree(entries, t) {
if nodes[path] != hash {
t.Errorf("%v, want %x, got %x", []byte(path), nodes[path], hash)
}
}
}
}
}
func TestStackTrieErrors(t *testing.T) {
s := NewStackTrie(nil)
// Deletion
if err := s.Update(nil, nil); err == nil {
t.Fatal("expected error")
}
if err := s.Update(nil, []byte{}); err == nil {
t.Fatal("expected error")
}
if err := s.Update([]byte{0xa}, []byte{}); err == nil {
t.Fatal("expected error")
}
// Non-ascending keys (going backwards or repeating)
assert.Nil(t, s.Update([]byte{0xaa}, []byte{0xa}))
assert.NotNil(t, s.Update([]byte{0xaa}, []byte{0xa}), "repeat insert same key")
assert.NotNil(t, s.Update([]byte{0xaa}, []byte{0xb}), "repeat insert same key")
assert.Nil(t, s.Update([]byte{0xab}, []byte{0xa}))
assert.NotNil(t, s.Update([]byte{0x10}, []byte{0xb}), "out of order insert")
assert.NotNil(t, s.Update([]byte{0xaa}, []byte{0xb}), "repeat insert same key")
}

714
trie/sync.go
View file

@ -1,714 +0,0 @@
// Copyright 2015 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 trie
import (
"errors"
"fmt"
"sync"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/common/prque"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/metrics"
)
// ErrNotRequested is returned by the trie sync when it's requested to process a
// node it did not request.
var ErrNotRequested = errors.New("not requested")
// ErrAlreadyProcessed is returned by the trie sync when it's requested to process a
// node it already processed previously.
var ErrAlreadyProcessed = errors.New("already processed")
// maxFetchesPerDepth is the maximum number of pending trie nodes per depth. The
// role of this value is to limit the number of trie nodes that get expanded in
// memory if the node was configured with a significant number of peers.
const maxFetchesPerDepth = 16384
var (
// deletionGauge is the metric to track how many trie node deletions
// are performed in total during the sync process.
deletionGauge = metrics.NewRegisteredGauge("trie/sync/delete", nil)
// lookupGauge is the metric to track how many trie node lookups are
// performed to determine if node needs to be deleted.
lookupGauge = metrics.NewRegisteredGauge("trie/sync/lookup", nil)
// accountNodeSyncedGauge is the metric to track how many account trie
// node are written during the sync.
accountNodeSyncedGauge = metrics.NewRegisteredGauge("trie/sync/nodes/account", nil)
// storageNodeSyncedGauge is the metric to track how many account trie
// node are written during the sync.
storageNodeSyncedGauge = metrics.NewRegisteredGauge("trie/sync/nodes/storage", nil)
// codeSyncedGauge is the metric to track how many contract codes are
// written during the sync.
codeSyncedGauge = metrics.NewRegisteredGauge("trie/sync/codes", nil)
)
// 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 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:])}
}
// LeafCallback is a callback type invoked when a trie operation reaches a leaf
// node.
//
// The keys is a path tuple identifying a particular trie node either in a single
// trie (account) or a layered trie (account -> storage). Each key in the tuple
// is in the raw format(32 bytes).
//
// The path is a composite hexary path identifying the trie node. All the key
// bytes are converted to the hexary nibbles and composited with the parent path
// if the trie node is in a layered trie.
//
// It's used by state sync and commit to allow handling external references
// between account and storage tries. And also it's used in the state healing
// for extracting the raw states(leaf nodes) with corresponding paths.
type LeafCallback func(keys [][]byte, path []byte, leaf []byte, parent common.Hash, parentPath []byte) error
// nodeRequest represents a scheduled or already in-flight trie node retrieval request.
type nodeRequest struct {
hash common.Hash // Hash of the trie node to retrieve
path []byte // Merkle path leading to this node for prioritization
data []byte // Data content of the node, cached until all subtrees complete
parent *nodeRequest // Parent state node referencing this entry
deps int // Number of dependencies before allowed to commit this node
callback LeafCallback // Callback to invoke if a leaf node it reached on this branch
}
// codeRequest represents a scheduled or already in-flight bytecode retrieval request.
type codeRequest struct {
hash common.Hash // Hash of the contract bytecode to retrieve
path []byte // Merkle path leading to this node for prioritization
data []byte // Data content of the node, cached until all subtrees complete
parents []*nodeRequest // Parent state nodes referencing this entry (notify all upon completion)
}
// NodeSyncResult is a response with requested trie node along with its node path.
type NodeSyncResult struct {
Path string // Path of the originally unknown trie node
Data []byte // Data content of the retrieved trie node
}
// CodeSyncResult is a response with requested bytecode along with its hash.
type CodeSyncResult struct {
Hash common.Hash // Hash the originally unknown bytecode
Data []byte // Data content of the retrieved bytecode
}
// nodeOp represents an operation upon the trie node. It can either represent a
// deletion to the specific node or a node write for persisting retrieved node.
type nodeOp struct {
owner common.Hash // identifier of the trie (empty for account trie)
path []byte // path from the root to the specified node.
blob []byte // the content of the node (nil for deletion)
hash common.Hash // hash of the node content (empty for node deletion)
}
// isDelete indicates if the operation is a database deletion.
func (op *nodeOp) isDelete() bool {
return len(op.blob) == 0
}
// syncMemBatch is an in-memory buffer of successfully downloaded but not yet
// persisted data items.
type syncMemBatch struct {
scheme string // State scheme identifier
codes map[common.Hash][]byte // In-memory batch of recently completed codes
nodes []nodeOp // In-memory batch of recently completed/deleted nodes
size uint64 // Estimated batch-size of in-memory data.
}
// newSyncMemBatch allocates a new memory-buffer for not-yet persisted trie nodes.
func newSyncMemBatch(scheme string) *syncMemBatch {
return &syncMemBatch{
scheme: scheme,
codes: make(map[common.Hash][]byte),
}
}
// hasCode reports the contract code with specific hash is already cached.
func (batch *syncMemBatch) hasCode(hash common.Hash) bool {
_, ok := batch.codes[hash]
return ok
}
// addCode caches a contract code database write operation.
func (batch *syncMemBatch) addCode(hash common.Hash, code []byte) {
batch.codes[hash] = code
batch.size += common.HashLength + uint64(len(code))
}
// addNode caches a node database write operation.
func (batch *syncMemBatch) addNode(owner common.Hash, path []byte, blob []byte, hash common.Hash) {
if batch.scheme == rawdb.PathScheme {
if owner == (common.Hash{}) {
batch.size += uint64(len(path) + len(blob))
} else {
batch.size += common.HashLength + uint64(len(path)+len(blob))
}
} else {
batch.size += common.HashLength + uint64(len(blob))
}
batch.nodes = append(batch.nodes, nodeOp{
owner: owner,
path: path,
blob: blob,
hash: hash,
})
}
// delNode caches a node database delete operation.
func (batch *syncMemBatch) delNode(owner common.Hash, path []byte) {
if batch.scheme != rawdb.PathScheme {
log.Error("Unexpected node deletion", "owner", owner, "path", path, "scheme", batch.scheme)
return // deletion is not supported in hash mode.
}
if owner == (common.Hash{}) {
batch.size += uint64(len(path))
} else {
batch.size += common.HashLength + uint64(len(path))
}
batch.nodes = append(batch.nodes, nodeOp{
owner: owner,
path: path,
})
}
// Sync is the main state trie synchronisation scheduler, which provides yet
// unknown trie hashes to retrieve, accepts node data associated with said hashes
// and reconstructs the trie step by step until all is done.
type Sync struct {
scheme string // Node scheme descriptor used in database.
database ethdb.KeyValueReader // Persistent database to check for existing entries
membatch *syncMemBatch // Memory buffer to avoid frequent database writes
nodeReqs map[string]*nodeRequest // Pending requests pertaining to a trie node path
codeReqs map[common.Hash]*codeRequest // Pending requests pertaining to a code hash
queue *prque.Prque[int64, any] // Priority queue with the pending requests
fetches map[int]int // Number of active fetches per trie node depth
}
// NewSync creates a new trie data download scheduler.
func NewSync(root common.Hash, database ethdb.KeyValueReader, callback LeafCallback, scheme string) *Sync {
ts := &Sync{
scheme: scheme,
database: database,
membatch: newSyncMemBatch(scheme),
nodeReqs: make(map[string]*nodeRequest),
codeReqs: make(map[common.Hash]*codeRequest),
queue: prque.New[int64, any](nil), // Ugh, can contain both string and hash, whyyy
fetches: make(map[int]int),
}
ts.AddSubTrie(root, nil, common.Hash{}, nil, callback)
return ts
}
// AddSubTrie registers a new trie to the sync code, rooted at the designated
// parent for completion tracking. The given path is a unique node path in
// hex format and contain all the parent path if it's layered trie node.
func (s *Sync) AddSubTrie(root common.Hash, path []byte, parent common.Hash, parentPath []byte, callback LeafCallback) {
if root == types.EmptyRootHash {
return
}
owner, inner := ResolvePath(path)
exist, inconsistent := s.hasNode(owner, inner, root)
if exist {
// The entire subtrie is already present in the database.
return
} else if inconsistent {
// There is a pre-existing node with the wrong hash in DB, remove it.
s.membatch.delNode(owner, inner)
}
// Assemble the new sub-trie sync request
req := &nodeRequest{
hash: root,
path: path,
callback: callback,
}
// If this sub-trie has a designated parent, link them together
if parent != (common.Hash{}) {
ancestor := s.nodeReqs[string(parentPath)]
if ancestor == nil {
panic(fmt.Sprintf("sub-trie ancestor not found: %x", parent))
}
ancestor.deps++
req.parent = ancestor
}
s.scheduleNodeRequest(req)
}
// AddCodeEntry schedules the direct retrieval of a contract code that should not
// be interpreted as a trie node, but rather accepted and stored into the database
// as is.
func (s *Sync) AddCodeEntry(hash common.Hash, path []byte, parent common.Hash, parentPath []byte) {
// Short circuit if the entry is empty or already known
if hash == types.EmptyCodeHash {
return
}
if s.membatch.hasCode(hash) {
return
}
// If database says duplicate, the blob is present for sure.
// Note we only check the existence with new code scheme, snap
// sync is expected to run with a fresh new node. Even there
// exists the code with legacy format, fetch and store with
// new scheme anyway.
if rawdb.HasCodeWithPrefix(s.database, hash) {
return
}
// Assemble the new sub-trie sync request
req := &codeRequest{
path: path,
hash: hash,
}
// If this sub-trie has a designated parent, link them together
if parent != (common.Hash{}) {
ancestor := s.nodeReqs[string(parentPath)] // the parent of codereq can ONLY be nodereq
if ancestor == nil {
panic(fmt.Sprintf("raw-entry ancestor not found: %x", parent))
}
ancestor.deps++
req.parents = append(req.parents, ancestor)
}
s.scheduleCodeRequest(req)
}
// 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) ([]string, []common.Hash, []common.Hash) {
var (
nodePaths []string
nodeHashes []common.Hash
codeHashes []common.Hash
)
for !s.queue.Empty() && (max == 0 || len(nodeHashes)+len(codeHashes) < max) {
// Retrieve the next item in line
item, prio := s.queue.Peek()
// If we have too many already-pending tasks for this depth, throttle
depth := int(prio >> 56)
if s.fetches[depth] > maxFetchesPerDepth {
break
}
// Item is allowed to be scheduled, add it to the task list
s.queue.Pop()
s.fetches[depth]++
switch item := item.(type) {
case common.Hash:
codeHashes = append(codeHashes, item)
case string:
req, ok := s.nodeReqs[item]
if !ok {
log.Error("Missing node request", "path", item)
continue // System very wrong, shouldn't happen
}
nodePaths = append(nodePaths, item)
nodeHashes = append(nodeHashes, req.hash)
}
}
return nodePaths, nodeHashes, codeHashes
}
// ProcessCode injects the received data for requested item. Note it can
// happen that the single response commits two pending requests(e.g.
// there are two requests one for code and one for node but the hash
// is same). In this case the second response for the same hash will
// be treated as "non-requested" item or "already-processed" item but
// there is no downside.
func (s *Sync) ProcessCode(result CodeSyncResult) error {
// If the code was not requested or it's already processed, bail out
req := s.codeReqs[result.Hash]
if req == nil {
return ErrNotRequested
}
if req.data != nil {
return ErrAlreadyProcessed
}
req.data = result.Data
return s.commitCodeRequest(req)
}
// ProcessNode injects the received data for requested item. Note it can
// happen that the single response commits two pending requests(e.g.
// there are two requests one for code and one for node but the hash
// is same). In this case the second response for the same hash will
// be treated as "non-requested" item or "already-processed" item but
// there is no downside.
func (s *Sync) ProcessNode(result NodeSyncResult) error {
// If the trie node was not requested or it's already processed, bail out
req := s.nodeReqs[result.Path]
if req == nil {
return ErrNotRequested
}
if req.data != nil {
return ErrAlreadyProcessed
}
// Decode the node data content and update the request
node, err := decodeNode(req.hash.Bytes(), result.Data)
if err != nil {
return err
}
req.data = result.Data
// Create and schedule a request for all the children nodes
requests, err := s.children(req, node)
if err != nil {
return err
}
if len(requests) == 0 && req.deps == 0 {
s.commitNodeRequest(req)
} else {
req.deps += len(requests)
for _, child := range requests {
s.scheduleNodeRequest(child)
}
}
return nil
}
// Commit flushes the data stored in the internal membatch out to persistent
// storage, returning any occurred error. The whole data set will be flushed
// in an atomic database batch.
func (s *Sync) Commit(dbw ethdb.Batch) error {
// Flush the pending node writes into database batch.
var (
account int
storage int
)
for _, op := range s.membatch.nodes {
if op.isDelete() {
// node deletion is only supported in path mode.
if op.owner == (common.Hash{}) {
rawdb.DeleteAccountTrieNode(dbw, op.path)
} else {
rawdb.DeleteStorageTrieNode(dbw, op.owner, op.path)
}
deletionGauge.Inc(1)
} else {
if op.owner == (common.Hash{}) {
account += 1
} else {
storage += 1
}
rawdb.WriteTrieNode(dbw, op.owner, op.path, op.hash, op.blob, s.scheme)
}
}
accountNodeSyncedGauge.Inc(int64(account))
storageNodeSyncedGauge.Inc(int64(storage))
// Flush the pending code writes into database batch.
for hash, value := range s.membatch.codes {
rawdb.WriteCode(dbw, hash, value)
}
codeSyncedGauge.Inc(int64(len(s.membatch.codes)))
s.membatch = newSyncMemBatch(s.scheme) // reset the batch
return nil
}
// MemSize returns an estimated size (in bytes) of the data held in the membatch.
func (s *Sync) MemSize() uint64 {
return s.membatch.size
}
// Pending returns the number of state entries currently pending for download.
func (s *Sync) Pending() int {
return len(s.nodeReqs) + len(s.codeReqs)
}
// scheduleNodeRequest inserts a new state retrieval request into the fetch queue. If there
// is already a pending request for this node, the new request will be discarded
// and only a parent reference added to the old one.
func (s *Sync) scheduleNodeRequest(req *nodeRequest) {
s.nodeReqs[string(req.path)] = req
// Schedule the request for future retrieval. This queue is shared
// by both node requests and code requests.
prio := int64(len(req.path)) << 56 // depth >= 128 will never happen, storage leaves will be included in their parents
for i := 0; i < 14 && i < len(req.path); i++ {
prio |= int64(15-req.path[i]) << (52 - i*4) // 15-nibble => lexicographic order
}
s.queue.Push(string(req.path), prio)
}
// scheduleCodeRequest inserts a new state retrieval request into the fetch queue. If there
// is already a pending request for this node, the new request will be discarded
// and only a parent reference added to the old one.
func (s *Sync) scheduleCodeRequest(req *codeRequest) {
// If we're already requesting this node, add a new reference and stop
if old, ok := s.codeReqs[req.hash]; ok {
old.parents = append(old.parents, req.parents...)
return
}
s.codeReqs[req.hash] = req
// Schedule the request for future retrieval. This queue is shared
// by both node requests and code requests.
prio := int64(len(req.path)) << 56 // depth >= 128 will never happen, storage leaves will be included in their parents
for i := 0; i < 14 && i < len(req.path); i++ {
prio |= int64(15-req.path[i]) << (52 - i*4) // 15-nibble => lexicographic order
}
s.queue.Push(req.hash, prio)
}
// children retrieves all the missing children of a state trie entry for future
// retrieval scheduling.
func (s *Sync) children(req *nodeRequest, object node) ([]*nodeRequest, error) {
// Gather all the children of the node, irrelevant whether known or not
type childNode struct {
path []byte
node node
}
var children []childNode
switch node := (object).(type) {
case *shortNode:
key := node.Key
if hasTerm(key) {
key = key[:len(key)-1]
}
children = []childNode{{
node: node.Val,
path: append(append([]byte(nil), req.path...), key...),
}}
// Mark all internal nodes between shortNode and its **in disk**
// child as invalid. This is essential in the case of path mode
// scheme; otherwise, state healing might overwrite existing child
// nodes silently while leaving a dangling parent node within the
// range of this internal path on disk and the persistent state
// ends up with a very weird situation that nodes on the same path
// are not inconsistent while they all present in disk. This property
// would break the guarantee for state healing.
//
// While it's possible for this shortNode to overwrite a previously
// existing full node, the other branches of the fullNode can be
// retained as they are not accessible with the new shortNode, and
// also the whole sub-trie is still untouched and complete.
//
// This step is only necessary for path mode, as there is no deletion
// in hash mode at all.
if _, ok := node.Val.(hashNode); ok && s.scheme == rawdb.PathScheme {
owner, inner := ResolvePath(req.path)
for i := 1; i < len(key); i++ {
// While checking for a non-existent item in Pebble can be less efficient
// without a bloom filter, the relatively low frequency of lookups makes
// the performance impact negligible.
var exists bool
if owner == (common.Hash{}) {
exists = rawdb.ExistsAccountTrieNode(s.database, append(inner, key[:i]...))
} else {
exists = rawdb.ExistsStorageTrieNode(s.database, owner, append(inner, key[:i]...))
}
if exists {
s.membatch.delNode(owner, append(inner, key[:i]...))
log.Debug("Detected dangling node", "owner", owner, "path", append(inner, key[:i]...))
}
}
lookupGauge.Inc(int64(len(key) - 1))
}
case *fullNode:
for i := 0; i < 17; i++ {
if node.Children[i] != nil {
children = append(children, childNode{
node: node.Children[i],
path: append(append([]byte(nil), req.path...), byte(i)),
})
}
}
default:
panic(fmt.Sprintf("unknown node: %+v", node))
}
// Iterate over the children, and request all unknown ones
var (
missing = make(chan *nodeRequest, len(children))
pending sync.WaitGroup
batchMu sync.Mutex
)
for _, child := range children {
// Notify any external watcher of a new key/value node
if req.callback != nil {
if node, ok := (child.node).(valueNode); ok {
var paths [][]byte
if len(child.path) == 2*common.HashLength {
paths = append(paths, hexToKeybytes(child.path))
} else if len(child.path) == 4*common.HashLength {
paths = append(paths, hexToKeybytes(child.path[:2*common.HashLength]))
paths = append(paths, hexToKeybytes(child.path[2*common.HashLength:]))
}
if err := req.callback(paths, child.path, node, req.hash, req.path); err != nil {
return nil, err
}
}
}
// If the child references another node, resolve or schedule.
// We check all children concurrently.
if node, ok := (child.node).(hashNode); ok {
path := child.path
hash := common.BytesToHash(node)
pending.Add(1)
go func() {
defer pending.Done()
owner, inner := ResolvePath(path)
exist, inconsistent := s.hasNode(owner, inner, hash)
if exist {
return
} else if inconsistent {
// There is a pre-existing node with the wrong hash in DB, remove it.
batchMu.Lock()
s.membatch.delNode(owner, inner)
batchMu.Unlock()
}
// Locally unknown node, schedule for retrieval
missing <- &nodeRequest{
path: path,
hash: hash,
parent: req,
callback: req.callback,
}
}()
}
}
pending.Wait()
requests := make([]*nodeRequest, 0, len(children))
for done := false; !done; {
select {
case miss := <-missing:
requests = append(requests, miss)
default:
done = true
}
}
return requests, nil
}
// commitNodeRequest finalizes a retrieval request and stores it into the membatch. If any
// of the referencing parent requests complete due to this commit, they are also
// committed themselves.
func (s *Sync) commitNodeRequest(req *nodeRequest) error {
// Write the node content to the membatch
owner, path := ResolvePath(req.path)
s.membatch.addNode(owner, path, req.data, req.hash)
// Removed the completed node request
delete(s.nodeReqs, string(req.path))
s.fetches[len(req.path)]--
// Check parent for completion
if req.parent != nil {
req.parent.deps--
if req.parent.deps == 0 {
if err := s.commitNodeRequest(req.parent); err != nil {
return err
}
}
}
return nil
}
// commitCodeRequest finalizes a retrieval request and stores it into the membatch. If any
// of the referencing parent requests complete due to this commit, they are also
// committed themselves.
func (s *Sync) commitCodeRequest(req *codeRequest) error {
// Write the node content to the membatch
s.membatch.addCode(req.hash, req.data)
// Removed the completed code request
delete(s.codeReqs, req.hash)
s.fetches[len(req.path)]--
// Check all parents for completion
for _, parent := range req.parents {
parent.deps--
if parent.deps == 0 {
if err := s.commitNodeRequest(parent); err != nil {
return err
}
}
}
return nil
}
// hasNode reports whether the specified trie node is present in the database.
// 'exists' is true when the node exists in the database and matches the given root
// hash. The 'inconsistent' return value is true when the node exists but does not
// match the expected hash.
func (s *Sync) hasNode(owner common.Hash, path []byte, hash common.Hash) (exists bool, inconsistent bool) {
// If node is running with hash scheme, check the presence with node hash.
if s.scheme == rawdb.HashScheme {
return rawdb.HasLegacyTrieNode(s.database, hash), false
}
// If node is running with path scheme, check the presence with node path.
var blob []byte
var dbHash common.Hash
if owner == (common.Hash{}) {
blob, dbHash = rawdb.ReadAccountTrieNode(s.database, path)
} else {
blob, dbHash = rawdb.ReadStorageTrieNode(s.database, owner, path)
}
exists = hash == dbHash
inconsistent = !exists && len(blob) != 0
return exists, inconsistent
}
// ResolvePath resolves the provided composite node path by separating the
// path in account trie if it's existent.
func ResolvePath(path []byte) (common.Hash, []byte) {
var owner common.Hash
if len(path) >= 2*common.HashLength {
owner = common.BytesToHash(hexToKeybytes(path[:2*common.HashLength]))
path = path[2*common.HashLength:]
}
return owner, path
}

1015
trie/sync_test.go
View file

File diff suppressed because it is too large Load diff

61
trie/testutil/utils.go
View file

@ -1,61 +0,0 @@
// Copyright 2023 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 testutil
import (
crand "crypto/rand"
"encoding/binary"
mrand "math/rand"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/trie/trienode"
)
// Prng is a pseudo random number generator seeded by strong randomness.
// The randomness is printed on startup in order to make failures reproducible.
var prng = initRand()
func initRand() *mrand.Rand {
var seed [8]byte
crand.Read(seed[:])
rnd := mrand.New(mrand.NewSource(int64(binary.LittleEndian.Uint64(seed[:]))))
return rnd
}
// RandBytes generates a random byte slice with specified length.
func RandBytes(n int) []byte {
r := make([]byte, n)
prng.Read(r)
return r
}
// RandomHash generates a random blob of data and returns it as a hash.
func RandomHash() common.Hash {
return common.BytesToHash(RandBytes(common.HashLength))
}
// RandomAddress generates a random blob of data and returns it as an address.
func RandomAddress() common.Address {
return common.BytesToAddress(RandBytes(common.AddressLength))
}
// RandomNode generates a random node.
func RandomNode() *trienode.Node {
val := RandBytes(100)
return trienode.New(crypto.Keccak256Hash(val), val)
}

130
trie/tracer.go
View file

@ -1,130 +0,0 @@
// 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 trie
import (
"github.com/ethereum/go-ethereum/common"
)
// tracer tracks the changes of trie nodes. During the trie operations,
// some nodes can be deleted from the trie, while these deleted nodes
// won't be captured by trie.Hasher or trie.Committer. Thus, these deleted
// nodes won't be removed from the disk at all. Tracer is an auxiliary tool
// used to track all insert and delete operations of trie and capture all
// deleted nodes eventually.
//
// The changed nodes can be mainly divided into two categories: the leaf
// node and intermediate node. The former is inserted/deleted by callers
// while the latter is inserted/deleted in order to follow the rule of trie.
// This tool can track all of them no matter the node is embedded in its
// parent or not, but valueNode is never tracked.
//
// Besides, it's also used for recording the original value of the nodes
// when they are resolved from the disk. The pre-value of the nodes will
// be used to construct trie history in the future.
//
// Note tracer is not thread-safe, callers should be responsible for handling
// the concurrency issues by themselves.
type tracer struct {
inserts map[string]struct{}
deletes map[string]struct{}
accessList map[string][]byte
}
// newTracer initializes the tracer for capturing trie changes.
func newTracer() *tracer {
return &tracer{
inserts: make(map[string]struct{}),
deletes: make(map[string]struct{}),
accessList: make(map[string][]byte),
}
}
// onRead tracks the newly loaded trie node and caches the rlp-encoded
// blob internally. Don't change the value outside of function since
// it's not deep-copied.
func (t *tracer) onRead(path []byte, val []byte) {
t.accessList[string(path)] = val
}
// onInsert tracks the newly inserted trie node. If it's already
// in the deletion set (resurrected node), then just wipe it from
// the deletion set as it's "untouched".
func (t *tracer) onInsert(path []byte) {
if _, present := t.deletes[string(path)]; present {
delete(t.deletes, string(path))
return
}
t.inserts[string(path)] = struct{}{}
}
// onDelete tracks the newly deleted trie node. If it's already
// in the addition set, then just wipe it from the addition set
// as it's untouched.
func (t *tracer) onDelete(path []byte) {
if _, present := t.inserts[string(path)]; present {
delete(t.inserts, string(path))
return
}
t.deletes[string(path)] = struct{}{}
}
// reset clears the content tracked by tracer.
func (t *tracer) reset() {
t.inserts = make(map[string]struct{})
t.deletes = make(map[string]struct{})
t.accessList = make(map[string][]byte)
}
// copy returns a deep copied tracer instance.
func (t *tracer) copy() *tracer {
var (
inserts = make(map[string]struct{})
deletes = make(map[string]struct{})
accessList = make(map[string][]byte)
)
for path := range t.inserts {
inserts[path] = struct{}{}
}
for path := range t.deletes {
deletes[path] = struct{}{}
}
for path, blob := range t.accessList {
accessList[path] = common.CopyBytes(blob)
}
return &tracer{
inserts: inserts,
deletes: deletes,
accessList: accessList,
}
}
// deletedNodes returns a list of node paths which are deleted from the trie.
func (t *tracer) deletedNodes() []string {
var paths []string
for path := range t.deletes {
// It's possible a few deleted nodes were embedded
// in their parent before, the deletions can be no
// effect by deleting nothing, filter them out.
_, ok := t.accessList[path]
if !ok {
continue
}
paths = append(paths, path)
}
return paths
}

375
trie/tracer_test.go
View file

@ -1,375 +0,0 @@
// 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 trie
import (
"bytes"
"testing"
"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/trie/trienode"
)
var (
tiny = []struct{ k, v string }{
{"k1", "v1"},
{"k2", "v2"},
{"k3", "v3"},
}
nonAligned = []struct{ k, v string }{
{"do", "verb"},
{"ether", "wookiedoo"},
{"horse", "stallion"},
{"shaman", "horse"},
{"doge", "coin"},
{"dog", "puppy"},
{"somethingveryoddindeedthis is", "myothernodedata"},
}
standard = []struct{ k, v string }{
{string(randBytes(32)), "verb"},
{string(randBytes(32)), "wookiedoo"},
{string(randBytes(32)), "stallion"},
{string(randBytes(32)), "horse"},
{string(randBytes(32)), "coin"},
{string(randBytes(32)), "puppy"},
{string(randBytes(32)), "myothernodedata"},
}
)
func TestTrieTracer(t *testing.T) {
testTrieTracer(t, tiny)
testTrieTracer(t, nonAligned)
testTrieTracer(t, standard)
}
// Tests if the trie diffs are tracked correctly. Tracer should capture
// all non-leaf dirty nodes, no matter the node is embedded or not.
func testTrieTracer(t *testing.T, vals []struct{ k, v string }) {
db := NewDatabase(rawdb.NewMemoryDatabase(), nil)
trie := NewEmpty(db)
// Determine all new nodes are tracked
for _, val := range vals {
trie.MustUpdate([]byte(val.k), []byte(val.v))
}
insertSet := copySet(trie.tracer.inserts) // copy before commit
deleteSet := copySet(trie.tracer.deletes) // copy before commit
root, nodes, _ := trie.Commit(false)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
seen := setKeys(iterNodes(db, root))
if !compareSet(insertSet, seen) {
t.Fatal("Unexpected insertion set")
}
if !compareSet(deleteSet, nil) {
t.Fatal("Unexpected deletion set")
}
// Determine all deletions are tracked
trie, _ = New(TrieID(root), db)
for _, val := range vals {
trie.MustDelete([]byte(val.k))
}
insertSet, deleteSet = copySet(trie.tracer.inserts), copySet(trie.tracer.deletes)
if !compareSet(insertSet, nil) {
t.Fatal("Unexpected insertion set")
}
if !compareSet(deleteSet, seen) {
t.Fatal("Unexpected deletion set")
}
}
// Test that after inserting a new batch of nodes and deleting them immediately,
// the trie tracer should be cleared normally as no operation happened.
func TestTrieTracerNoop(t *testing.T) {
testTrieTracerNoop(t, tiny)
testTrieTracerNoop(t, nonAligned)
testTrieTracerNoop(t, standard)
}
func testTrieTracerNoop(t *testing.T, vals []struct{ k, v string }) {
trie := NewEmpty(NewDatabase(rawdb.NewMemoryDatabase(), nil))
for _, val := range vals {
trie.MustUpdate([]byte(val.k), []byte(val.v))
}
for _, val := range vals {
trie.MustDelete([]byte(val.k))
}
if len(trie.tracer.inserts) != 0 {
t.Fatal("Unexpected insertion set")
}
if len(trie.tracer.deletes) != 0 {
t.Fatal("Unexpected deletion set")
}
}
// Tests if the accessList is correctly tracked.
func TestAccessList(t *testing.T) {
testAccessList(t, tiny)
testAccessList(t, nonAligned)
testAccessList(t, standard)
}
func testAccessList(t *testing.T, vals []struct{ k, v string }) {
var (
db = NewDatabase(rawdb.NewMemoryDatabase(), nil)
trie = NewEmpty(db)
orig = trie.Copy()
)
// Create trie from scratch
for _, val := range vals {
trie.MustUpdate([]byte(val.k), []byte(val.v))
}
root, nodes, _ := trie.Commit(false)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
trie, _ = New(TrieID(root), db)
if err := verifyAccessList(orig, trie, nodes); err != nil {
t.Fatalf("Invalid accessList %v", err)
}
// Update trie
parent := root
trie, _ = New(TrieID(root), db)
orig = trie.Copy()
for _, val := range vals {
trie.MustUpdate([]byte(val.k), randBytes(32))
}
root, nodes, _ = trie.Commit(false)
db.Update(root, parent, 0, trienode.NewWithNodeSet(nodes), nil)
trie, _ = New(TrieID(root), db)
if err := verifyAccessList(orig, trie, nodes); err != nil {
t.Fatalf("Invalid accessList %v", err)
}
// Add more new nodes
parent = root
trie, _ = New(TrieID(root), db)
orig = trie.Copy()
var keys []string
for i := 0; i < 30; i++ {
key := randBytes(32)
keys = append(keys, string(key))
trie.MustUpdate(key, randBytes(32))
}
root, nodes, _ = trie.Commit(false)
db.Update(root, parent, 0, trienode.NewWithNodeSet(nodes), nil)
trie, _ = New(TrieID(root), db)
if err := verifyAccessList(orig, trie, nodes); err != nil {
t.Fatalf("Invalid accessList %v", err)
}
// Partial deletions
parent = root
trie, _ = New(TrieID(root), db)
orig = trie.Copy()
for _, key := range keys {
trie.MustUpdate([]byte(key), nil)
}
root, nodes, _ = trie.Commit(false)
db.Update(root, parent, 0, trienode.NewWithNodeSet(nodes), nil)
trie, _ = New(TrieID(root), db)
if err := verifyAccessList(orig, trie, nodes); err != nil {
t.Fatalf("Invalid accessList %v", err)
}
// Delete all
parent = root
trie, _ = New(TrieID(root), db)
orig = trie.Copy()
for _, val := range vals {
trie.MustUpdate([]byte(val.k), nil)
}
root, nodes, _ = trie.Commit(false)
db.Update(root, parent, 0, trienode.NewWithNodeSet(nodes), nil)
trie, _ = New(TrieID(root), db)
if err := verifyAccessList(orig, trie, nodes); err != nil {
t.Fatalf("Invalid accessList %v", err)
}
}
// Tests origin values won't be tracked in Iterator or Prover
func TestAccessListLeak(t *testing.T) {
var (
db = NewDatabase(rawdb.NewMemoryDatabase(), nil)
trie = NewEmpty(db)
)
// Create trie from scratch
for _, val := range standard {
trie.MustUpdate([]byte(val.k), []byte(val.v))
}
root, nodes, _ := trie.Commit(false)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(nodes), nil)
var cases = []struct {
op func(tr *Trie)
}{
{
func(tr *Trie) {
it := tr.MustNodeIterator(nil)
for it.Next(true) {
}
},
},
{
func(tr *Trie) {
it := NewIterator(tr.MustNodeIterator(nil))
for it.Next() {
}
},
},
{
func(tr *Trie) {
for _, val := range standard {
tr.Prove([]byte(val.k), rawdb.NewMemoryDatabase())
}
},
},
}
for _, c := range cases {
trie, _ = New(TrieID(root), db)
n1 := len(trie.tracer.accessList)
c.op(trie)
n2 := len(trie.tracer.accessList)
if n1 != n2 {
t.Fatalf("AccessList is leaked, prev %d after %d", n1, n2)
}
}
}
// Tests whether the original tree node is correctly deleted after being embedded
// in its parent due to the smaller size of the original tree node.
func TestTinyTree(t *testing.T) {
var (
db = NewDatabase(rawdb.NewMemoryDatabase(), nil)
trie = NewEmpty(db)
)
for _, val := range tiny {
trie.MustUpdate([]byte(val.k), randBytes(32))
}
root, set, _ := trie.Commit(false)
db.Update(root, types.EmptyRootHash, 0, trienode.NewWithNodeSet(set), nil)
parent := root
trie, _ = New(TrieID(root), db)
orig := trie.Copy()
for _, val := range tiny {
trie.MustUpdate([]byte(val.k), []byte(val.v))
}
root, set, _ = trie.Commit(false)
db.Update(root, parent, 0, trienode.NewWithNodeSet(set), nil)
trie, _ = New(TrieID(root), db)
if err := verifyAccessList(orig, trie, set); err != nil {
t.Fatalf("Invalid accessList %v", err)
}
}
func compareSet(setA, setB map[string]struct{}) bool {
if len(setA) != len(setB) {
return false
}
for key := range setA {
if _, ok := setB[key]; !ok {
return false
}
}
return true
}
func forNodes(tr *Trie) map[string][]byte {
var (
it = tr.MustNodeIterator(nil)
nodes = make(map[string][]byte)
)
for it.Next(true) {
if it.Leaf() {
continue
}
nodes[string(it.Path())] = common.CopyBytes(it.NodeBlob())
}
return nodes
}
func iterNodes(db *Database, root common.Hash) map[string][]byte {
tr, _ := New(TrieID(root), db)
return forNodes(tr)
}
func forHashedNodes(tr *Trie) map[string][]byte {
var (
it = tr.MustNodeIterator(nil)
nodes = make(map[string][]byte)
)
for it.Next(true) {
if it.Hash() == (common.Hash{}) {
continue
}
nodes[string(it.Path())] = common.CopyBytes(it.NodeBlob())
}
return nodes
}
func diffTries(trieA, trieB *Trie) (map[string][]byte, map[string][]byte, map[string][]byte) {
var (
nodesA = forHashedNodes(trieA)
nodesB = forHashedNodes(trieB)
inA = make(map[string][]byte) // hashed nodes in trie a but not b
inB = make(map[string][]byte) // hashed nodes in trie b but not a
both = make(map[string][]byte) // hashed nodes in both tries but different value
)
for path, blobA := range nodesA {
if blobB, ok := nodesB[path]; ok {
if bytes.Equal(blobA, blobB) {
continue
}
both[path] = blobA
continue
}
inA[path] = blobA
}
for path, blobB := range nodesB {
if _, ok := nodesA[path]; ok {
continue
}
inB[path] = blobB
}
return inA, inB, both
}
func setKeys(set map[string][]byte) map[string]struct{} {
keys := make(map[string]struct{})
for k := range set {
keys[k] = struct{}{}
}
return keys
}
func copySet(set map[string]struct{}) map[string]struct{} {
copied := make(map[string]struct{})
for k := range set {
copied[k] = struct{}{}
}
return copied
}

672
trie/trie.go
View file

@ -1,672 +0,0 @@
// Copyright 2014 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 trie implements Merkle Patricia Tries.
package trie
import (
"bytes"
"errors"
"fmt"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/trie/trienode"
)
// Trie is a Merkle Patricia Trie. Use New to create a trie that sits on
// top of a database. Whenever trie performs a commit operation, the generated
// nodes will be gathered and returned in a set. Once the trie is committed,
// it's not usable anymore. Callers have to re-create the trie with new root
// based on the updated trie database.
//
// Trie is not safe for concurrent use.
type Trie struct {
root node
owner common.Hash
// Flag whether the commit operation is already performed. If so the
// trie is not usable(latest states is invisible).
committed bool
// Keep track of the number leaves which have been inserted since the last
// hashing operation. This number will not directly map to the number of
// actually unhashed nodes.
unhashed int
// reader is the handler trie can retrieve nodes from.
reader *trieReader
// tracer is the tool to track the trie changes.
// It will be reset after each commit operation.
tracer *tracer
}
// newFlag returns the cache flag value for a newly created node.
func (t *Trie) newFlag() nodeFlag {
return nodeFlag{dirty: true}
}
// Copy returns a copy of Trie.
func (t *Trie) Copy() *Trie {
return &Trie{
root: t.root,
owner: t.owner,
committed: t.committed,
unhashed: t.unhashed,
reader: t.reader,
tracer: t.tracer.copy(),
}
}
// New creates the trie instance with provided trie id and the read-only
// database. The state specified by trie id must be available, otherwise
// an error will be returned. The trie root specified by trie id can be
// zero hash or the sha3 hash of an empty string, then trie is initially
// empty, otherwise, the root node must be present in database or returns
// a MissingNodeError if not.
func New(id *ID, db *Database) (*Trie, error) {
reader, err := newTrieReader(id.StateRoot, id.Owner, db)
if err != nil {
return nil, err
}
trie := &Trie{
owner: id.Owner,
reader: reader,
tracer: newTracer(),
}
if id.Root != (common.Hash{}) && id.Root != types.EmptyRootHash {
rootnode, err := trie.resolveAndTrack(id.Root[:], nil)
if err != nil {
return nil, err
}
trie.root = rootnode
}
return trie, nil
}
// NewEmpty is a shortcut to create empty tree. It's mostly used in tests.
func NewEmpty(db *Database) *Trie {
tr, _ := New(TrieID(types.EmptyRootHash), db)
return tr
}
// MustNodeIterator is a wrapper of NodeIterator and will omit any encountered
// error but just print out an error message.
func (t *Trie) MustNodeIterator(start []byte) NodeIterator {
it, err := t.NodeIterator(start)
if err != nil {
log.Error("Unhandled trie error in Trie.NodeIterator", "err", err)
}
return it
}
// NodeIterator returns an iterator that returns nodes of the trie. Iteration starts at
// the key after the given start key.
func (t *Trie) NodeIterator(start []byte) (NodeIterator, error) {
// Short circuit if the trie is already committed and not usable.
if t.committed {
return nil, ErrCommitted
}
return newNodeIterator(t, start), nil
}
// MustGet is a wrapper of Get and will omit any encountered error but just
// print out an error message.
func (t *Trie) MustGet(key []byte) []byte {
res, err := t.Get(key)
if err != nil {
log.Error("Unhandled trie error in Trie.Get", "err", err)
}
return res
}
// Get returns the value for key stored in the trie.
// The value bytes must not be modified by the caller.
//
// If the requested node is not present in trie, no error will be returned.
// If the trie is corrupted, a MissingNodeError is returned.
func (t *Trie) Get(key []byte) ([]byte, error) {
// Short circuit if the trie is already committed and not usable.
if t.committed {
return nil, ErrCommitted
}
value, newroot, didResolve, err := t.get(t.root, keybytesToHex(key), 0)
if err == nil && didResolve {
t.root = newroot
}
return value, err
}
func (t *Trie) get(origNode node, key []byte, pos int) (value []byte, newnode node, didResolve bool, err error) {
switch n := (origNode).(type) {
case nil:
return nil, nil, false, nil
case valueNode:
return n, n, false, nil
case *shortNode:
if len(key)-pos < len(n.Key) || !bytes.Equal(n.Key, key[pos:pos+len(n.Key)]) {
// key not found in trie
return nil, n, false, nil
}
value, newnode, didResolve, err = t.get(n.Val, key, pos+len(n.Key))
if err == nil && didResolve {
n = n.copy()
n.Val = newnode
}
return value, n, didResolve, err
case *fullNode:
value, newnode, didResolve, err = t.get(n.Children[key[pos]], key, pos+1)
if err == nil && didResolve {
n = n.copy()
n.Children[key[pos]] = newnode
}
return value, n, didResolve, err
case hashNode:
child, err := t.resolveAndTrack(n, key[:pos])
if err != nil {
return nil, n, true, err
}
value, newnode, _, err := t.get(child, key, pos)
return value, newnode, true, err
default:
panic(fmt.Sprintf("%T: invalid node: %v", origNode, origNode))
}
}
// MustGetNode is a wrapper of GetNode and will omit any encountered error but
// just print out an error message.
func (t *Trie) MustGetNode(path []byte) ([]byte, int) {
item, resolved, err := t.GetNode(path)
if err != nil {
log.Error("Unhandled trie error in Trie.GetNode", "err", err)
}
return item, resolved
}
// GetNode retrieves a trie node by compact-encoded path. It is not possible
// to use keybyte-encoding as the path might contain odd nibbles.
//
// If the requested node is not present in trie, no error will be returned.
// If the trie is corrupted, a MissingNodeError is returned.
func (t *Trie) GetNode(path []byte) ([]byte, int, error) {
// Short circuit if the trie is already committed and not usable.
if t.committed {
return nil, 0, ErrCommitted
}
item, newroot, resolved, err := t.getNode(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
}
return item, resolved, nil
}
func (t *Trie) getNode(origNode node, path []byte, pos int) (item []byte, newnode node, resolved int, err error) {
// If non-existent path requested, abort
if origNode == nil {
return nil, nil, 0, nil
}
// If we reached the requested path, return the current node
if pos >= len(path) {
// Although we most probably have the original node expanded, encoding
// that into consensus form can be nasty (needs to cascade down) and
// time consuming. Instead, just pull the hash up from disk directly.
var hash hashNode
if node, ok := origNode.(hashNode); ok {
hash = node
} else {
hash, _ = origNode.cache()
}
if hash == nil {
return nil, origNode, 0, errors.New("non-consensus node")
}
blob, err := t.reader.node(path, common.BytesToHash(hash))
return blob, origNode, 1, err
}
// Path still needs to be traversed, descend into children
switch n := (origNode).(type) {
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.getNode(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.getNode(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.resolveAndTrack(n, path[:pos])
if err != nil {
return nil, n, 1, err
}
item, newnode, resolved, err := t.getNode(child, path, pos)
return item, newnode, resolved + 1, err
default:
panic(fmt.Sprintf("%T: invalid node: %v", origNode, origNode))
}
}
// MustUpdate is a wrapper of Update and will omit any encountered error but
// just print out an error message.
func (t *Trie) MustUpdate(key, value []byte) {
if err := t.Update(key, value); err != nil {
log.Error("Unhandled trie error in Trie.Update", "err", err)
}
}
// 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.
//
// The value bytes must not be modified by the caller while they are
// stored in the trie.
//
// If the requested node is not present in trie, no error will be returned.
// If the trie is corrupted, a MissingNodeError is returned.
func (t *Trie) Update(key, value []byte) error {
// Short circuit if the trie is already committed and not usable.
if t.committed {
return ErrCommitted
}
return t.update(key, value)
}
func (t *Trie) update(key, value []byte) error {
t.unhashed++
k := keybytesToHex(key)
if len(value) != 0 {
_, n, err := t.insert(t.root, nil, k, valueNode(value))
if err != nil {
return err
}
t.root = n
} else {
_, n, err := t.delete(t.root, nil, k)
if err != nil {
return err
}
t.root = n
}
return nil
}
func (t *Trie) insert(n node, prefix, key []byte, value node) (bool, node, error) {
if len(key) == 0 {
if v, ok := n.(valueNode); ok {
return !bytes.Equal(v, value.(valueNode)), value, nil
}
return true, value, nil
}
switch n := n.(type) {
case *shortNode:
matchlen := prefixLen(key, n.Key)
// If the whole key matches, keep this short node as is
// and only update the value.
if matchlen == len(n.Key) {
dirty, nn, err := t.insert(n.Val, append(prefix, key[:matchlen]...), key[matchlen:], value)
if !dirty || err != nil {
return false, n, err
}
return true, &shortNode{n.Key, nn, t.newFlag()}, nil
}
// Otherwise branch out at the index where they differ.
branch := &fullNode{flags: t.newFlag()}
var err error
_, branch.Children[n.Key[matchlen]], err = t.insert(nil, append(prefix, n.Key[:matchlen+1]...), n.Key[matchlen+1:], n.Val)
if err != nil {
return false, nil, err
}
_, branch.Children[key[matchlen]], err = t.insert(nil, append(prefix, key[:matchlen+1]...), key[matchlen+1:], value)
if err != nil {
return false, nil, err
}
// Replace this shortNode with the branch if it occurs at index 0.
if matchlen == 0 {
return true, branch, nil
}
// New branch node is created as a child of the original short node.
// Track the newly inserted node in the tracer. The node identifier
// passed is the path from the root node.
t.tracer.onInsert(append(prefix, key[:matchlen]...))
// Replace it with a short node leading up to the branch.
return true, &shortNode{key[:matchlen], branch, t.newFlag()}, nil
case *fullNode:
dirty, nn, err := t.insert(n.Children[key[0]], append(prefix, key[0]), key[1:], value)
if !dirty || err != nil {
return false, n, err
}
n = n.copy()
n.flags = t.newFlag()
n.Children[key[0]] = nn
return true, n, nil
case nil:
// New short node is created and track it in the tracer. The node identifier
// passed is the path from the root node. Note the valueNode won't be tracked
// since it's always embedded in its parent.
t.tracer.onInsert(prefix)
return true, &shortNode{key, value, t.newFlag()}, nil
case hashNode:
// We've hit a part of the trie that isn't loaded yet. Load
// the node and insert into it. This leaves all child nodes on
// the path to the value in the trie.
rn, err := t.resolveAndTrack(n, prefix)
if err != nil {
return false, nil, err
}
dirty, nn, err := t.insert(rn, prefix, key, value)
if !dirty || err != nil {
return false, rn, err
}
return true, nn, nil
default:
panic(fmt.Sprintf("%T: invalid node: %v", n, n))
}
}
// MustDelete is a wrapper of Delete and will omit any encountered error but
// just print out an error message.
func (t *Trie) MustDelete(key []byte) {
if err := t.Delete(key); err != nil {
log.Error("Unhandled trie error in Trie.Delete", "err", err)
}
}
// Delete removes any existing value for key from the trie.
//
// If the requested node is not present in trie, no error will be returned.
// If the trie is corrupted, a MissingNodeError is returned.
func (t *Trie) Delete(key []byte) error {
// Short circuit if the trie is already committed and not usable.
if t.committed {
return ErrCommitted
}
t.unhashed++
k := keybytesToHex(key)
_, n, err := t.delete(t.root, nil, k)
if err != nil {
return err
}
t.root = n
return nil
}
// delete returns the new root of the trie with key deleted.
// It reduces the trie to minimal form by simplifying
// nodes on the way up after deleting recursively.
func (t *Trie) delete(n node, prefix, key []byte) (bool, node, error) {
switch n := n.(type) {
case *shortNode:
matchlen := prefixLen(key, n.Key)
if matchlen < len(n.Key) {
return false, n, nil // don't replace n on mismatch
}
if matchlen == len(key) {
// The matched short node is deleted entirely and track
// it in the deletion set. The same the valueNode doesn't
// need to be tracked at all since it's always embedded.
t.tracer.onDelete(prefix)
return true, nil, nil // remove n entirely for whole matches
}
// The key is longer than n.Key. Remove the remaining suffix
// from the subtrie. Child can never be nil here since the
// subtrie must contain at least two other values with keys
// longer than n.Key.
dirty, child, err := t.delete(n.Val, append(prefix, key[:len(n.Key)]...), key[len(n.Key):])
if !dirty || err != nil {
return false, n, err
}
switch child := child.(type) {
case *shortNode:
// The child shortNode is merged into its parent, track
// is deleted as well.
t.tracer.onDelete(append(prefix, n.Key...))
// Deleting from the subtrie reduced it to another
// short node. Merge the nodes to avoid creating a
// shortNode{..., shortNode{...}}. Use concat (which
// always creates a new slice) instead of append to
// avoid modifying n.Key since it might be shared with
// other nodes.
return true, &shortNode{concat(n.Key, child.Key...), child.Val, t.newFlag()}, nil
default:
return true, &shortNode{n.Key, child, t.newFlag()}, nil
}
case *fullNode:
dirty, nn, err := t.delete(n.Children[key[0]], append(prefix, key[0]), key[1:])
if !dirty || err != nil {
return false, n, err
}
n = n.copy()
n.flags = t.newFlag()
n.Children[key[0]] = nn
// Because n is a full node, it must've contained at least two children
// before the delete operation. If the new child value is non-nil, n still
// has at least two children after the deletion, and cannot be reduced to
// a short node.
if nn != nil {
return true, n, nil
}
// Reduction:
// Check how many non-nil entries are left after deleting and
// reduce the full node to a short node if only one entry is
// left. Since n must've contained at least two children
// before deletion (otherwise it would not be a full node) n
// can never be reduced to nil.
//
// When the loop is done, pos contains the index of the single
// value that is left in n or -2 if n contains at least two
// values.
pos := -1
for i, cld := range &n.Children {
if cld != nil {
if pos == -1 {
pos = i
} else {
pos = -2
break
}
}
}
if pos >= 0 {
if pos != 16 {
// If the remaining entry is a short node, it replaces
// n and its key gets the missing nibble tacked to the
// front. This avoids creating an invalid
// shortNode{..., shortNode{...}}. Since the entry
// might not be loaded yet, resolve it just for this
// check.
cnode, err := t.resolve(n.Children[pos], append(prefix, byte(pos)))
if err != nil {
return false, nil, err
}
if cnode, ok := cnode.(*shortNode); ok {
// Replace the entire full node with the short node.
// Mark the original short node as deleted since the
// value is embedded into the parent now.
t.tracer.onDelete(append(prefix, byte(pos)))
k := append([]byte{byte(pos)}, cnode.Key...)
return true, &shortNode{k, cnode.Val, t.newFlag()}, nil
}
}
// Otherwise, n is replaced by a one-nibble short node
// containing the child.
return true, &shortNode{[]byte{byte(pos)}, n.Children[pos], t.newFlag()}, nil
}
// n still contains at least two values and cannot be reduced.
return true, n, nil
case valueNode:
return true, nil, nil
case nil:
return false, nil, nil
case hashNode:
// We've hit a part of the trie that isn't loaded yet. Load
// the node and delete from it. This leaves all child nodes on
// the path to the value in the trie.
rn, err := t.resolveAndTrack(n, prefix)
if err != nil {
return false, nil, err
}
dirty, nn, err := t.delete(rn, prefix, key)
if !dirty || err != nil {
return false, rn, err
}
return true, nn, nil
default:
panic(fmt.Sprintf("%T: invalid node: %v (%v)", n, n, key))
}
}
func concat(s1 []byte, s2 ...byte) []byte {
r := make([]byte, len(s1)+len(s2))
copy(r, s1)
copy(r[len(s1):], s2)
return r
}
func (t *Trie) resolve(n node, prefix []byte) (node, error) {
if n, ok := n.(hashNode); ok {
return t.resolveAndTrack(n, prefix)
}
return n, nil
}
// resolveAndTrack loads node from the underlying store with the given node hash
// and path prefix and also tracks the loaded node blob in tracer treated as the
// node's original value. The rlp-encoded blob is preferred to be loaded from
// database because it's easy to decode node while complex to encode node to blob.
func (t *Trie) resolveAndTrack(n hashNode, prefix []byte) (node, error) {
blob, err := t.reader.node(prefix, common.BytesToHash(n))
if err != nil {
return nil, err
}
t.tracer.onRead(prefix, blob)
return mustDecodeNode(n, blob), nil
}
// Hash returns the root hash of the trie. It does not write to the
// database and can be used even if the trie doesn't have one.
func (t *Trie) Hash() common.Hash {
hash, cached := t.hashRoot()
t.root = cached
return common.BytesToHash(hash.(hashNode))
}
// Commit collects all dirty nodes in the trie and replaces them with the
// corresponding node hash. All collected nodes (including dirty leaves if
// collectLeaf is true) will be encapsulated into a nodeset for return.
// The returned nodeset can be nil if the trie is clean (nothing to commit).
// Once the trie is committed, it's not usable anymore. A new trie must
// be created with new root and updated trie database for following usage
func (t *Trie) Commit(collectLeaf bool) (common.Hash, *trienode.NodeSet, error) {
defer t.tracer.reset()
defer func() {
t.committed = true
}()
// Trie is empty and can be classified into two types of situations:
// (a) The trie was empty and no update happens => return nil
// (b) The trie was non-empty and all nodes are dropped => return
// the node set includes all deleted nodes
if t.root == nil {
paths := t.tracer.deletedNodes()
if len(paths) == 0 {
return types.EmptyRootHash, nil, nil // case (a)
}
nodes := trienode.NewNodeSet(t.owner)
for _, path := range paths {
nodes.AddNode([]byte(path), trienode.NewDeleted())
}
return types.EmptyRootHash, nodes, nil // case (b)
}
// Derive the hash for all dirty nodes first. We hold the assumption
// in the following procedure that all nodes are hashed.
rootHash := t.Hash()
// Do a quick check if we really need to commit. This can happen e.g.
// if we load a trie for reading storage values, but don't write to it.
if hashedNode, dirty := t.root.cache(); !dirty {
// Replace the root node with the origin hash in order to
// ensure all resolved nodes are dropped after the commit.
t.root = hashedNode
return rootHash, nil, nil
}
nodes := trienode.NewNodeSet(t.owner)
for _, path := range t.tracer.deletedNodes() {
nodes.AddNode([]byte(path), trienode.NewDeleted())
}
t.root = newCommitter(nodes, t.tracer, collectLeaf).Commit(t.root)
return rootHash, nodes, nil
}
// hashRoot calculates the root hash of the given trie
func (t *Trie) hashRoot() (node, node) {
if t.root == nil {
return hashNode(types.EmptyRootHash.Bytes()), nil
}
// If the number of changes is below 100, we let one thread handle it
h := newHasher(t.unhashed >= 100)
defer func() {
returnHasherToPool(h)
t.unhashed = 0
}()
hashed, cached := h.hash(t.root, true)
return hashed, cached
}
// Reset drops the referenced root node and cleans all internal state.
func (t *Trie) Reset() {
t.root = nil
t.owner = common.Hash{}
t.unhashed = 0
t.tracer.reset()
t.committed = false
}

55
trie/trie_id.go
View file

@ -1,55 +0,0 @@
// 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 trie
import "github.com/ethereum/go-ethereum/common"
// ID is the identifier for uniquely identifying a trie.
type ID struct {
StateRoot common.Hash // The root of the corresponding state(block.root)
Owner common.Hash // The contract address hash which the trie belongs to
Root common.Hash // The root hash of trie
}
// StateTrieID constructs an identifier for state trie with the provided state root.
func StateTrieID(root common.Hash) *ID {
return &ID{
StateRoot: root,
Owner: common.Hash{},
Root: root,
}
}
// StorageTrieID constructs an identifier for storage trie which belongs to a certain
// state and contract specified by the stateRoot and owner.
func StorageTrieID(stateRoot common.Hash, owner common.Hash, root common.Hash) *ID {
return &ID{
StateRoot: stateRoot,
Owner: owner,
Root: root,
}
}
// TrieID constructs an identifier for a standard trie(not a second-layer trie)
// with provided root. It's mostly used in tests and some other tries like CHT trie.
func TrieID(root common.Hash) *ID {
return &ID{
StateRoot: root,
Owner: common.Hash{},
Root: root,
}
}

101
trie/trie_reader.go
View file

@ -1,101 +0,0 @@
// 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 trie
import (
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/trie/triestate"
)
// Reader wraps the Node method of a backing trie store.
type Reader interface {
// Node retrieves the trie node blob with the provided trie identifier, node path and
// the corresponding node hash. No error will be returned if the node is not found.
//
// When looking up nodes in the account trie, 'owner' is the zero hash. For contract
// storage trie nodes, 'owner' is the hash of the account address that containing the
// storage.
//
// TODO(rjl493456442): remove the 'hash' parameter, it's redundant in PBSS.
Node(owner common.Hash, path []byte, hash common.Hash) ([]byte, error)
}
// trieReader is a wrapper of the underlying node reader. It's not safe
// for concurrent usage.
type trieReader struct {
owner common.Hash
reader Reader
banned map[string]struct{} // Marker to prevent node from being accessed, for tests
}
// newTrieReader initializes the trie reader with the given node reader.
func newTrieReader(stateRoot, owner common.Hash, db *Database) (*trieReader, error) {
if stateRoot == (common.Hash{}) || stateRoot == types.EmptyRootHash {
if stateRoot == (common.Hash{}) {
log.Error("Zero state root hash!")
}
return &trieReader{owner: owner}, nil
}
reader, err := db.Reader(stateRoot)
if err != nil {
return nil, &MissingNodeError{Owner: owner, NodeHash: stateRoot, err: err}
}
return &trieReader{owner: owner, reader: reader}, nil
}
// newEmptyReader initializes the pure in-memory reader. All read operations
// should be forbidden and returns the MissingNodeError.
func newEmptyReader() *trieReader {
return &trieReader{}
}
// node retrieves the rlp-encoded trie node with the provided trie node
// information. An MissingNodeError will be returned in case the node is
// not found or any error is encountered.
func (r *trieReader) node(path []byte, hash common.Hash) ([]byte, error) {
// Perform the logics in tests for preventing trie node access.
if r.banned != nil {
if _, ok := r.banned[string(path)]; ok {
return nil, &MissingNodeError{Owner: r.owner, NodeHash: hash, Path: path}
}
}
if r.reader == nil {
return nil, &MissingNodeError{Owner: r.owner, NodeHash: hash, Path: path}
}
blob, err := r.reader.Node(r.owner, path, hash)
if err != nil || len(blob) == 0 {
return nil, &MissingNodeError{Owner: r.owner, NodeHash: hash, Path: path, err: err}
}
return blob, nil
}
// trieLoader implements triestate.TrieLoader for constructing tries.
type trieLoader struct {
db *Database
}
// OpenTrie opens the main account trie.
func (l *trieLoader) OpenTrie(root common.Hash) (triestate.Trie, error) {
return New(TrieID(root), l.db)
}
// OpenStorageTrie opens the storage trie of an account.
func (l *trieLoader) OpenStorageTrie(stateRoot common.Hash, addrHash, root common.Hash) (triestate.Trie, error) {
return New(StorageTrieID(stateRoot, addrHash, root), l.db)
}

1224
trie/trie_test.go
View file

File diff suppressed because it is too large Load diff

651
trie/triedb/hashdb/database.go
View file

@ -1,651 +0,0 @@
// Copyright 2018 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 hashdb
import (
"errors"
"fmt"
"reflect"
"sync"
"time"
"github.com/VictoriaMetrics/fastcache"
"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/ethdb"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/metrics"
"github.com/ethereum/go-ethereum/rlp"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
)
var (
memcacheCleanHitMeter = metrics.NewRegisteredMeter("hashdb/memcache/clean/hit", nil)
memcacheCleanMissMeter = metrics.NewRegisteredMeter("hashdb/memcache/clean/miss", nil)
memcacheCleanReadMeter = metrics.NewRegisteredMeter("hashdb/memcache/clean/read", nil)
memcacheCleanWriteMeter = metrics.NewRegisteredMeter("hashdb/memcache/clean/write", nil)
memcacheDirtyHitMeter = metrics.NewRegisteredMeter("hashdb/memcache/dirty/hit", nil)
memcacheDirtyMissMeter = metrics.NewRegisteredMeter("hashdb/memcache/dirty/miss", nil)
memcacheDirtyReadMeter = metrics.NewRegisteredMeter("hashdb/memcache/dirty/read", nil)
memcacheDirtyWriteMeter = metrics.NewRegisteredMeter("hashdb/memcache/dirty/write", nil)
memcacheFlushTimeTimer = metrics.NewRegisteredResettingTimer("hashdb/memcache/flush/time", nil)
memcacheFlushNodesMeter = metrics.NewRegisteredMeter("hashdb/memcache/flush/nodes", nil)
memcacheFlushBytesMeter = metrics.NewRegisteredMeter("hashdb/memcache/flush/bytes", nil)
memcacheGCTimeTimer = metrics.NewRegisteredResettingTimer("hashdb/memcache/gc/time", nil)
memcacheGCNodesMeter = metrics.NewRegisteredMeter("hashdb/memcache/gc/nodes", nil)
memcacheGCBytesMeter = metrics.NewRegisteredMeter("hashdb/memcache/gc/bytes", nil)
memcacheCommitTimeTimer = metrics.NewRegisteredResettingTimer("hashdb/memcache/commit/time", nil)
memcacheCommitNodesMeter = metrics.NewRegisteredMeter("hashdb/memcache/commit/nodes", nil)
memcacheCommitBytesMeter = metrics.NewRegisteredMeter("hashdb/memcache/commit/bytes", nil)
)
// ChildResolver defines the required method to decode the provided
// trie node and iterate the children on top.
type ChildResolver interface {
ForEach(node []byte, onChild func(common.Hash))
}
// Config contains the settings for database.
type Config struct {
CleanCacheSize int // Maximum memory allowance (in bytes) for caching clean nodes
}
// Defaults is the default setting for database if it's not specified.
// Notably, clean cache is disabled explicitly,
var Defaults = &Config{
// Explicitly set clean cache size to 0 to avoid creating fastcache,
// otherwise database must be closed when it's no longer needed to
// prevent memory leak.
CleanCacheSize: 0,
}
// Database is an intermediate write layer between the trie data structures and
// 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
resolver ChildResolver // The handler to resolve children of nodes
cleans *fastcache.Cache // GC friendly memory cache of clean node RLPs
dirties map[common.Hash]*cachedNode // Data and references relationships of dirty trie nodes
oldest common.Hash // Oldest tracked node, flush-list head
newest common.Hash // Newest tracked node, flush-list tail
gctime time.Duration // Time spent on garbage collection since last commit
gcnodes uint64 // Nodes garbage collected since last commit
gcsize common.StorageSize // Data storage garbage collected since last commit
flushtime time.Duration // Time spent on data flushing since last commit
flushnodes uint64 // Nodes flushed since last commit
flushsize common.StorageSize // Data storage flushed since last commit
dirtiesSize common.StorageSize // Storage size of the dirty node cache (exc. metadata)
childrenSize common.StorageSize // Storage size of the external children tracking
lock sync.RWMutex
}
// cachedNode is all the information we know about a single cached trie node
// in the memory database write layer.
type cachedNode struct {
node []byte // Encoded node blob, immutable
parents uint32 // Number of live nodes referencing this one
external map[common.Hash]struct{} // The set of external children
flushPrev common.Hash // Previous node in the flush-list
flushNext common.Hash // Next node in the flush-list
}
// cachedNodeSize is the raw size of a cachedNode data structure without any
// node data included. It's an approximate size, but should be a lot better
// than not counting them.
var cachedNodeSize = int(reflect.TypeOf(cachedNode{}).Size())
// forChildren invokes the callback for all the tracked children of this node,
// both the implicit ones from inside the node as well as the explicit ones
// from outside the node.
func (n *cachedNode) forChildren(resolver ChildResolver, onChild func(hash common.Hash)) {
for child := range n.external {
onChild(child)
}
resolver.ForEach(n.node, onChild)
}
// New initializes the hash-based node database.
func New(diskdb ethdb.Database, config *Config, resolver ChildResolver) *Database {
if config == nil {
config = Defaults
}
var cleans *fastcache.Cache
if config.CleanCacheSize > 0 {
cleans = fastcache.New(config.CleanCacheSize)
}
return &Database{
diskdb: diskdb,
resolver: resolver,
cleans: cleans,
dirties: make(map[common.Hash]*cachedNode),
}
}
// insert inserts a trie node into the memory database. All nodes inserted by
// this function will be reference tracked. This function assumes the lock is
// already held.
func (db *Database) insert(hash common.Hash, node []byte) {
// If the node's already cached, skip
if _, ok := db.dirties[hash]; ok {
return
}
memcacheDirtyWriteMeter.Mark(int64(len(node)))
// Create the cached entry for this node
entry := &cachedNode{
node: node,
flushPrev: db.newest,
}
entry.forChildren(db.resolver, func(child common.Hash) {
if c := db.dirties[child]; c != nil {
c.parents++
}
})
db.dirties[hash] = entry
// Update the flush-list endpoints
if db.oldest == (common.Hash{}) {
db.oldest, db.newest = hash, hash
} else {
db.dirties[db.newest].flushNext, db.newest = hash, hash
}
db.dirtiesSize += common.StorageSize(common.HashLength + len(node))
}
// node retrieves an encoded cached trie node from memory. If it cannot be found
// cached, the method queries the persistent database for the content.
func (db *Database) node(hash common.Hash) ([]byte, error) {
// It doesn't make sense to retrieve the metaroot
if hash == (common.Hash{}) {
return nil, errors.New("not found")
}
// Retrieve the node from the clean cache if available
if db.cleans != nil {
if enc := db.cleans.Get(nil, hash[:]); enc != nil {
memcacheCleanHitMeter.Mark(1)
memcacheCleanReadMeter.Mark(int64(len(enc)))
return enc, nil
}
}
// Retrieve the node from the dirty cache if available.
db.lock.RLock()
dirty := db.dirties[hash]
db.lock.RUnlock()
// Return the cached node if it's found in the dirty set.
// The dirty.node field is immutable and safe to read it
// even without lock guard.
if dirty != nil {
memcacheDirtyHitMeter.Mark(1)
memcacheDirtyReadMeter.Mark(int64(len(dirty.node)))
return dirty.node, nil
}
memcacheDirtyMissMeter.Mark(1)
// Content unavailable in memory, attempt to retrieve from disk
enc := rawdb.ReadLegacyTrieNode(db.diskdb, hash)
if len(enc) != 0 {
if db.cleans != nil {
db.cleans.Set(hash[:], enc)
memcacheCleanMissMeter.Mark(1)
memcacheCleanWriteMeter.Mark(int64(len(enc)))
}
return enc, nil
}
return nil, errors.New("not found")
}
// Reference adds a new reference from a parent node to a child node.
// This function is used to add reference between internal trie node
// and external node(e.g. storage trie root), all internal trie nodes
// are referenced together by database itself.
func (db *Database) Reference(child common.Hash, parent common.Hash) {
db.lock.Lock()
defer db.lock.Unlock()
db.reference(child, parent)
}
// reference is the private locked version of Reference.
func (db *Database) reference(child common.Hash, parent common.Hash) {
// If the node does not exist, it's a node pulled from disk, skip
node, ok := db.dirties[child]
if !ok {
return
}
// The reference is for state root, increase the reference counter.
if parent == (common.Hash{}) {
node.parents += 1
return
}
// The reference is for external storage trie, don't duplicate if
// the reference is already existent.
if db.dirties[parent].external == nil {
db.dirties[parent].external = make(map[common.Hash]struct{})
}
if _, ok := db.dirties[parent].external[child]; ok {
return
}
node.parents++
db.dirties[parent].external[child] = struct{}{}
db.childrenSize += common.HashLength
}
// Dereference removes an existing reference from a root node.
func (db *Database) Dereference(root common.Hash) {
// Sanity check to ensure that the meta-root is not removed
if root == (common.Hash{}) {
log.Error("Attempted to dereference the trie cache meta root")
return
}
db.lock.Lock()
defer db.lock.Unlock()
nodes, storage, start := len(db.dirties), db.dirtiesSize, time.Now()
db.dereference(root)
db.gcnodes += uint64(nodes - len(db.dirties))
db.gcsize += storage - db.dirtiesSize
db.gctime += time.Since(start)
memcacheGCTimeTimer.Update(time.Since(start))
memcacheGCBytesMeter.Mark(int64(storage - db.dirtiesSize))
memcacheGCNodesMeter.Mark(int64(nodes - len(db.dirties)))
log.Debug("Dereferenced trie from memory database", "nodes", nodes-len(db.dirties), "size", storage-db.dirtiesSize, "time", time.Since(start),
"gcnodes", db.gcnodes, "gcsize", db.gcsize, "gctime", db.gctime, "livenodes", len(db.dirties), "livesize", db.dirtiesSize)
}
// dereference is the private locked version of Dereference.
func (db *Database) dereference(hash common.Hash) {
// If the node does not exist, it's a previously committed node.
node, ok := db.dirties[hash]
if !ok {
return
}
// If there are no more references to the node, delete it and cascade
if node.parents > 0 {
// This is a special cornercase where a node loaded from disk (i.e. not in the
// memcache any more) gets reinjected as a new node (short node split into full,
// then reverted into short), causing a cached node to have no parents. That is
// no problem in itself, but don't make maxint parents out of it.
node.parents--
}
if node.parents == 0 {
// Remove the node from the flush-list
switch hash {
case db.oldest:
db.oldest = node.flushNext
if node.flushNext != (common.Hash{}) {
db.dirties[node.flushNext].flushPrev = common.Hash{}
}
case db.newest:
db.newest = node.flushPrev
if node.flushPrev != (common.Hash{}) {
db.dirties[node.flushPrev].flushNext = common.Hash{}
}
default:
db.dirties[node.flushPrev].flushNext = node.flushNext
db.dirties[node.flushNext].flushPrev = node.flushPrev
}
// Dereference all children and delete the node
node.forChildren(db.resolver, func(child common.Hash) {
db.dereference(child)
})
delete(db.dirties, hash)
db.dirtiesSize -= common.StorageSize(common.HashLength + len(node.node))
if node.external != nil {
db.childrenSize -= common.StorageSize(len(node.external) * common.HashLength)
}
}
}
// Cap iteratively flushes old but still referenced trie nodes until the total
// memory usage goes below the given threshold.
func (db *Database) Cap(limit common.StorageSize) error {
db.lock.Lock()
defer db.lock.Unlock()
// Create a database batch to flush persistent data out. It is important that
// outside code doesn't see an inconsistent state (referenced data removed from
// memory cache during commit but not yet in persistent storage). This is ensured
// by only uncaching existing data when the database write finalizes.
batch := db.diskdb.NewBatch()
nodes, storage, start := len(db.dirties), db.dirtiesSize, time.Now()
// db.dirtiesSize only contains the useful data in the cache, but when reporting
// the total memory consumption, the maintenance metadata is also needed to be
// counted.
size := db.dirtiesSize + common.StorageSize(len(db.dirties)*cachedNodeSize)
size += db.childrenSize
// Keep committing nodes from the flush-list until we're below allowance
oldest := db.oldest
for size > limit && oldest != (common.Hash{}) {
// Fetch the oldest referenced node and push into the batch
node := db.dirties[oldest]
rawdb.WriteLegacyTrieNode(batch, oldest, node.node)
// If we exceeded the ideal batch size, commit and reset
if batch.ValueSize() >= ethdb.IdealBatchSize {
if err := batch.Write(); err != nil {
log.Error("Failed to write flush list to disk", "err", err)
return err
}
batch.Reset()
}
// Iterate to the next flush item, or abort if the size cap was achieved. Size
// is the total size, including the useful cached data (hash -> blob), the
// cache item metadata, as well as external children mappings.
size -= common.StorageSize(common.HashLength + len(node.node) + cachedNodeSize)
if node.external != nil {
size -= common.StorageSize(len(node.external) * common.HashLength)
}
oldest = node.flushNext
}
// Flush out any remainder data from the last batch
if err := batch.Write(); err != nil {
log.Error("Failed to write flush list to disk", "err", err)
return err
}
// Write successful, clear out the flushed data
for db.oldest != oldest {
node := db.dirties[db.oldest]
delete(db.dirties, db.oldest)
db.oldest = node.flushNext
db.dirtiesSize -= common.StorageSize(common.HashLength + len(node.node))
if node.external != nil {
db.childrenSize -= common.StorageSize(len(node.external) * common.HashLength)
}
}
if db.oldest != (common.Hash{}) {
db.dirties[db.oldest].flushPrev = common.Hash{}
}
db.flushnodes += uint64(nodes - len(db.dirties))
db.flushsize += storage - db.dirtiesSize
db.flushtime += time.Since(start)
memcacheFlushTimeTimer.Update(time.Since(start))
memcacheFlushBytesMeter.Mark(int64(storage - db.dirtiesSize))
memcacheFlushNodesMeter.Mark(int64(nodes - len(db.dirties)))
log.Debug("Persisted nodes from memory database", "nodes", nodes-len(db.dirties), "size", storage-db.dirtiesSize, "time", time.Since(start),
"flushnodes", db.flushnodes, "flushsize", db.flushsize, "flushtime", db.flushtime, "livenodes", len(db.dirties), "livesize", db.dirtiesSize)
return nil
}
// Commit iterates over all the children of a particular node, writes them out
// to disk, forcefully tearing down all references in both directions. As a side
// effect, all pre-images accumulated up to this point are also written.
func (db *Database) Commit(node common.Hash, report bool) error {
db.lock.Lock()
defer db.lock.Unlock()
// Create a database batch to flush persistent data out. It is important that
// outside code doesn't see an inconsistent state (referenced data removed from
// memory cache during commit but not yet in persistent storage). This is ensured
// by only uncaching existing data when the database write finalizes.
start := time.Now()
batch := db.diskdb.NewBatch()
// Move the trie itself into the batch, flushing if enough data is accumulated
nodes, storage := len(db.dirties), db.dirtiesSize
uncacher := &cleaner{db}
if err := db.commit(node, batch, uncacher); err != nil {
log.Error("Failed to commit trie from trie database", "err", err)
return err
}
// Trie mostly committed to disk, flush any batch leftovers
if err := batch.Write(); err != nil {
log.Error("Failed to write trie to disk", "err", err)
return err
}
// Uncache any leftovers in the last batch
if err := batch.Replay(uncacher); err != nil {
return err
}
batch.Reset()
// Reset the storage counters and bumped metrics
memcacheCommitTimeTimer.Update(time.Since(start))
memcacheCommitBytesMeter.Mark(int64(storage - db.dirtiesSize))
memcacheCommitNodesMeter.Mark(int64(nodes - len(db.dirties)))
logger := log.Info
if !report {
logger = log.Debug
}
logger("Persisted trie from memory database", "nodes", nodes-len(db.dirties)+int(db.flushnodes), "size", storage-db.dirtiesSize+db.flushsize, "time", time.Since(start)+db.flushtime,
"gcnodes", db.gcnodes, "gcsize", db.gcsize, "gctime", db.gctime, "livenodes", len(db.dirties), "livesize", db.dirtiesSize)
// Reset the garbage collection statistics
db.gcnodes, db.gcsize, db.gctime = 0, 0, 0
db.flushnodes, db.flushsize, db.flushtime = 0, 0, 0
return nil
}
// commit is the private locked version of Commit.
func (db *Database) commit(hash common.Hash, batch ethdb.Batch, uncacher *cleaner) error {
// If the node does not exist, it's a previously committed node
node, ok := db.dirties[hash]
if !ok {
return nil
}
var err error
// Dereference all children and delete the node
node.forChildren(db.resolver, func(child common.Hash) {
if err == nil {
err = db.commit(child, batch, uncacher)
}
})
if err != nil {
return err
}
// If we've reached an optimal batch size, commit and start over
rawdb.WriteLegacyTrieNode(batch, hash, node.node)
if batch.ValueSize() >= ethdb.IdealBatchSize {
if err := batch.Write(); err != nil {
return err
}
err := batch.Replay(uncacher)
if err != nil {
return err
}
batch.Reset()
}
return nil
}
// cleaner is a database batch replayer that takes a batch of write operations
// and cleans up the trie database from anything written to disk.
type cleaner struct {
db *Database
}
// Put reacts to database writes and implements dirty data uncaching. This is the
// post-processing step of a commit operation where the already persisted trie is
// removed from the dirty cache and moved into the clean cache. The reason behind
// the two-phase commit is to ensure data availability while moving from memory
// to disk.
func (c *cleaner) Put(key []byte, rlp []byte) error {
hash := common.BytesToHash(key)
// If the node does not exist, we're done on this path
node, ok := c.db.dirties[hash]
if !ok {
return nil
}
// Node still exists, remove it from the flush-list
switch hash {
case c.db.oldest:
c.db.oldest = node.flushNext
if node.flushNext != (common.Hash{}) {
c.db.dirties[node.flushNext].flushPrev = common.Hash{}
}
case c.db.newest:
c.db.newest = node.flushPrev
if node.flushPrev != (common.Hash{}) {
c.db.dirties[node.flushPrev].flushNext = common.Hash{}
}
default:
c.db.dirties[node.flushPrev].flushNext = node.flushNext
c.db.dirties[node.flushNext].flushPrev = node.flushPrev
}
// Remove the node from the dirty cache
delete(c.db.dirties, hash)
c.db.dirtiesSize -= common.StorageSize(common.HashLength + len(node.node))
if node.external != nil {
c.db.childrenSize -= common.StorageSize(len(node.external) * common.HashLength)
}
// Move the flushed node into the clean cache to prevent insta-reloads
if c.db.cleans != nil {
c.db.cleans.Set(hash[:], rlp)
memcacheCleanWriteMeter.Mark(int64(len(rlp)))
}
return nil
}
func (c *cleaner) Delete(key []byte) error {
panic("not implemented")
}
// Initialized returns an indicator if state data is already initialized
// in hash-based scheme by checking the presence of genesis state.
func (db *Database) Initialized(genesisRoot common.Hash) bool {
return rawdb.HasLegacyTrieNode(db.diskdb, genesisRoot)
}
// Update inserts the dirty nodes in provided nodeset into database and link the
// account trie with multiple storage tries if necessary.
func (db *Database) Update(root common.Hash, parent common.Hash, block uint64, nodes *trienode.MergedNodeSet, states *triestate.Set) error {
// Ensure the parent state is present and signal a warning if not.
if parent != types.EmptyRootHash {
if blob, _ := db.node(parent); len(blob) == 0 {
log.Error("parent state is not present")
}
}
db.lock.Lock()
defer db.lock.Unlock()
// Insert dirty nodes into the database. In the same tree, it must be
// ensured that children are inserted first, then parent so that children
// can be linked with their parent correctly.
//
// Note, the storage tries must be flushed before the account trie to
// retain the invariant that children go into the dirty cache first.
var order []common.Hash
for owner := range nodes.Sets {
if owner == (common.Hash{}) {
continue
}
order = append(order, owner)
}
if _, ok := nodes.Sets[common.Hash{}]; ok {
order = append(order, common.Hash{})
}
for _, owner := range order {
subset := nodes.Sets[owner]
subset.ForEachWithOrder(func(path string, n *trienode.Node) {
if n.IsDeleted() {
return // ignore deletion
}
db.insert(n.Hash, n.Blob)
})
}
// Link up the account trie and storage trie if the node points
// to an account trie leaf.
if set, present := nodes.Sets[common.Hash{}]; present {
for _, n := range set.Leaves {
var account types.StateAccount
if err := rlp.DecodeBytes(n.Blob, &account); err != nil {
return err
}
if account.Root != types.EmptyRootHash {
db.reference(account.Root, n.Parent)
}
}
}
return nil
}
// Size returns the current storage size of the memory cache in front of the
// persistent database layer.
//
// The first return will always be 0, representing the memory stored in unbounded
// diff layers above the dirty cache. This is only available in pathdb.
func (db *Database) Size() (common.StorageSize, common.StorageSize) {
db.lock.RLock()
defer db.lock.RUnlock()
// db.dirtiesSize only contains the useful data in the cache, but when reporting
// the total memory consumption, the maintenance metadata is also needed to be
// counted.
var metadataSize = common.StorageSize(len(db.dirties) * cachedNodeSize)
return 0, db.dirtiesSize + db.childrenSize + metadataSize
}
// Close closes the trie database and releases all held resources.
func (db *Database) Close() error {
if db.cleans != nil {
db.cleans.Reset()
db.cleans = nil
}
return nil
}
// Scheme returns the node scheme used in the database.
func (db *Database) Scheme() string {
return rawdb.HashScheme
}
// Reader retrieves a node reader belonging to the given state root.
// An error will be returned if the requested state is not available.
func (db *Database) Reader(root common.Hash) (*reader, error) {
if _, err := db.node(root); err != nil {
return nil, fmt.Errorf("state %#x is not available, %v", root, err)
}
return &reader{db: db}, nil
}
// reader is a state reader of Database which implements the Reader interface.
type reader struct {
db *Database
}
// Node retrieves the trie node with the given node hash. No error will be
// returned if the node is not found.
func (reader *reader) Node(owner common.Hash, path []byte, hash common.Hash) ([]byte, error) {
blob, _ := reader.db.node(hash)
return blob, nil
}

485
trie/triedb/pathdb/database.go
View file

@ -1,485 +0,0 @@
// 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 (
"errors"
"fmt"
"io"
"sync"
"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/ethdb"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/params"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
)
const (
// maxDiffLayers is the maximum diff layers allowed in the layer tree.
maxDiffLayers = 128
// defaultCleanSize is the default memory allowance of clean cache.
defaultCleanSize = 16 * 1024 * 1024
// maxBufferSize is the maximum memory allowance of node buffer.
// Too large nodebuffer will cause the system to pause for a long
// time when write happens. Also, the largest batch that pebble can
// support is 4GB, node will panic if batch size exceeds this limit.
maxBufferSize = 256 * 1024 * 1024
// DefaultBufferSize is the default memory allowance of node buffer
// that aggregates the writes from above until it's flushed into the
// disk. It's meant to be used once the initial sync is finished.
// Do not increase the buffer size arbitrarily, otherwise the system
// pause time will increase when the database writes happen.
DefaultBufferSize = 64 * 1024 * 1024
)
// layer is the interface implemented by all state layers which includes some
// public methods and some additional methods for internal usage.
type layer interface {
// Node retrieves the trie node with the node info. An error will be returned
// if the read operation exits abnormally. For example, if the layer is already
// stale, or the associated state is regarded as corrupted. Notably, no error
// will be returned if the requested node is not found in database.
Node(owner common.Hash, path []byte, hash common.Hash) ([]byte, error)
// rootHash returns the root hash for which this layer was made.
rootHash() common.Hash
// stateID returns the associated state id of layer.
stateID() uint64
// parentLayer returns the subsequent layer of it, or nil if the disk was reached.
parentLayer() layer
// update creates a new layer on top of the existing layer diff tree with
// the provided dirty trie nodes along with the state change set.
//
// Note, the maps are retained by the method to avoid copying everything.
update(root common.Hash, id uint64, block uint64, nodes map[common.Hash]map[string]*trienode.Node, states *triestate.Set) *diffLayer
// 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).
journal(w io.Writer) error
}
// Config contains the settings for database.
type Config struct {
StateHistory uint64 // Number of recent blocks to maintain state history for
CleanCacheSize int // Maximum memory allowance (in bytes) for caching clean nodes
DirtyCacheSize int // Maximum memory allowance (in bytes) for caching dirty nodes
ReadOnly bool // Flag whether the database is opened in read only mode.
}
// sanitize checks the provided user configurations and changes anything that's
// unreasonable or unworkable.
func (c *Config) sanitize() *Config {
conf := *c
if conf.DirtyCacheSize > maxBufferSize {
log.Warn("Sanitizing invalid node buffer size", "provided", common.StorageSize(conf.DirtyCacheSize), "updated", common.StorageSize(maxBufferSize))
conf.DirtyCacheSize = maxBufferSize
}
return &conf
}
// Defaults contains default settings for Ethereum mainnet.
var Defaults = &Config{
StateHistory: params.FullImmutabilityThreshold,
CleanCacheSize: defaultCleanSize,
DirtyCacheSize: DefaultBufferSize,
}
// ReadOnly is the config in order to open database in read only mode.
var ReadOnly = &Config{ReadOnly: true}
// Database is a multiple-layered structure for maintaining in-memory trie nodes.
// It consists of one persistent base layer backed by a key-value store, on top
// of which arbitrarily many in-memory diff layers are stacked. The memory diffs
// can form a tree with branching, but the disk layer is singleton and common to
// all. If a reorg goes deeper than the disk layer, a batch of reverse diffs can
// be applied to rollback. The deepest reorg that can be handled depends on the
// amount of state histories tracked in the disk.
//
// At most one readable and writable database can be opened at the same time in
// the whole system which ensures that only one database writer can operate disk
// state. Unexpected open operations can cause the system to panic.
type Database struct {
// readOnly is the flag whether the mutation is allowed to be applied.
// It will be set automatically when the database is journaled during
// the shutdown to reject all following unexpected mutations.
readOnly bool // Flag if database is opened in read only mode
waitSync bool // Flag if database is deactivated due to initial state sync
bufferSize int // Memory allowance (in bytes) for caching dirty nodes
config *Config // Configuration for database
diskdb ethdb.Database // Persistent storage for matured trie nodes
tree *layerTree // The group for all known layers
freezer *rawdb.ResettableFreezer // Freezer for storing trie histories, nil possible in tests
lock sync.RWMutex // Lock to prevent mutations from happening at the same time
}
// New attempts to load an already existing layer from a persistent key-value
// store (with a number of memory layers from a journal). If the journal is not
// matched with the base persistent layer, all the recorded diff layers are discarded.
func New(diskdb ethdb.Database, config *Config) *Database {
if config == nil {
config = Defaults
}
config = config.sanitize()
db := &Database{
readOnly: config.ReadOnly,
bufferSize: config.DirtyCacheSize,
config: config,
diskdb: diskdb,
}
// Construct the layer tree by resolving the in-disk singleton state
// and in-memory layer journal.
db.tree = newLayerTree(db.loadLayers())
// Open the freezer for state history if the passed database contains an
// ancient store. Otherwise, all the relevant functionalities are disabled.
//
// Because the freezer can only be opened once at the same time, this
// mechanism also ensures that at most one **non-readOnly** database
// is opened at the same time to prevent accidental mutation.
if ancient, err := diskdb.AncientDatadir(); err == nil && ancient != "" && !db.readOnly {
freezer, err := rawdb.NewStateFreezer(ancient, false)
if err != nil {
log.Crit("Failed to open state history freezer", "err", err)
}
db.freezer = freezer
diskLayerID := db.tree.bottom().stateID()
if diskLayerID == 0 {
// Reset the entire state histories in case the trie database is
// not initialized yet, as these state histories are not expected.
frozen, err := db.freezer.Ancients()
if err != nil {
log.Crit("Failed to retrieve head of state history", "err", err)
}
if frozen != 0 {
err := db.freezer.Reset()
if err != nil {
log.Crit("Failed to reset state histories", "err", err)
}
log.Info("Truncated extraneous state history")
}
} else {
// Truncate the extra state histories above in freezer in case
// it's not aligned with the disk layer.
pruned, err := truncateFromHead(db.diskdb, freezer, diskLayerID)
if err != nil {
log.Crit("Failed to truncate extra state histories", "err", err)
}
if pruned != 0 {
log.Warn("Truncated extra state histories", "number", pruned)
}
}
}
// Disable database in case node is still in the initial state sync stage.
if rawdb.ReadSnapSyncStatusFlag(diskdb) == rawdb.StateSyncRunning && !db.readOnly {
if err := db.Disable(); err != nil {
log.Crit("Failed to disable database", "err", err) // impossible to happen
}
}
log.Warn("Path-based state scheme is an experimental feature")
return db
}
// Reader retrieves a layer belonging to the given state root.
func (db *Database) Reader(root common.Hash) (layer, error) {
l := db.tree.get(root)
if l == nil {
return nil, fmt.Errorf("state %#x is not available", root)
}
return l, nil
}
// Update adds a new layer into the tree, if that can be linked to an existing
// old parent. It is disallowed to insert a disk layer (the origin of all). Apart
// from that this function will flatten the extra diff layers at bottom into disk
// to only keep 128 diff layers in memory by default.
//
// The passed in maps(nodes, states) will be retained to avoid copying everything.
// Therefore, these maps must not be changed afterwards.
func (db *Database) Update(root common.Hash, parentRoot common.Hash, block uint64, nodes *trienode.MergedNodeSet, states *triestate.Set) error {
// Hold the lock to prevent concurrent mutations.
db.lock.Lock()
defer db.lock.Unlock()
// Short circuit if the mutation is not allowed.
if err := db.modifyAllowed(); err != nil {
return err
}
if err := db.tree.add(root, parentRoot, block, nodes, states); err != nil {
return err
}
// Keep 128 diff layers in the memory, persistent layer is 129th.
// - head layer is paired with HEAD state
// - head-1 layer is paired with HEAD-1 state
// - head-127 layer(bottom-most diff layer) is paired with HEAD-127 state
// - head-128 layer(disk layer) is paired with HEAD-128 state
return db.tree.cap(root, maxDiffLayers)
}
// Commit traverses downwards the layer tree from a specified layer with the
// provided state root and all the layers below are flattened downwards. It
// can be used alone and mostly for test purposes.
func (db *Database) Commit(root common.Hash, report bool) error {
// Hold the lock to prevent concurrent mutations.
db.lock.Lock()
defer db.lock.Unlock()
// Short circuit if the mutation is not allowed.
if err := db.modifyAllowed(); err != nil {
return err
}
return db.tree.cap(root, 0)
}
// Disable deactivates the database and invalidates all available state layers
// as stale to prevent access to the persistent state, which is in the syncing
// stage.
func (db *Database) Disable() error {
db.lock.Lock()
defer db.lock.Unlock()
// Short circuit if the database is in read only mode.
if db.readOnly {
return errDatabaseReadOnly
}
// Prevent duplicated disable operation.
if db.waitSync {
log.Error("Reject duplicated disable operation")
return nil
}
db.waitSync = true
// Mark the disk layer as stale to prevent access to persistent state.
db.tree.bottom().markStale()
// Write the initial sync flag to persist it across restarts.
rawdb.WriteSnapSyncStatusFlag(db.diskdb, rawdb.StateSyncRunning)
log.Info("Disabled trie database due to state sync")
return nil
}
// Enable activates database and resets the state tree with the provided persistent
// state root once the state sync is finished.
func (db *Database) Enable(root common.Hash) error {
db.lock.Lock()
defer db.lock.Unlock()
// Short circuit if the database is in read only mode.
if db.readOnly {
return errDatabaseReadOnly
}
// Ensure the provided state root matches the stored one.
root = types.TrieRootHash(root)
_, stored := rawdb.ReadAccountTrieNode(db.diskdb, nil)
if stored != root {
return fmt.Errorf("state root mismatch: stored %x, synced %x", stored, root)
}
// Drop the stale state journal in persistent database and
// reset the persistent state id back to zero.
batch := db.diskdb.NewBatch()
rawdb.DeleteTrieJournal(batch)
rawdb.WritePersistentStateID(batch, 0)
if err := batch.Write(); err != nil {
return err
}
// Clean up all state histories in freezer. Theoretically
// all root->id mappings should be removed as well. Since
// mappings can be huge and might take a while to clear
// them, just leave them in disk and wait for overwriting.
if db.freezer != nil {
if err := db.freezer.Reset(); err != nil {
return err
}
}
// Re-construct a new disk layer backed by persistent state
// with **empty clean cache and node buffer**.
db.tree.reset(newDiskLayer(root, 0, db, nil, newNodeBuffer(db.bufferSize, nil, 0)))
// Re-enable the database as the final step.
db.waitSync = false
rawdb.WriteSnapSyncStatusFlag(db.diskdb, rawdb.StateSyncFinished)
log.Info("Rebuilt trie database", "root", root)
return nil
}
// Recover rollbacks the database to a specified historical point.
// The state is supported as the rollback destination only if it's
// canonical state and the corresponding trie histories are existent.
func (db *Database) Recover(root common.Hash, loader triestate.TrieLoader) error {
db.lock.Lock()
defer db.lock.Unlock()
// Short circuit if rollback operation is not supported.
if err := db.modifyAllowed(); err != nil {
return err
}
if db.freezer == nil {
return errors.New("state rollback is non-supported")
}
// Short circuit if the target state is not recoverable.
root = types.TrieRootHash(root)
if !db.Recoverable(root) {
return errStateUnrecoverable
}
// Apply the state histories upon the disk layer in order.
var (
start = time.Now()
dl = db.tree.bottom()
)
for dl.rootHash() != root {
h, err := readHistory(db.freezer, dl.stateID())
if err != nil {
return err
}
dl, err = dl.revert(h, loader)
if err != nil {
return err
}
// reset layer with newly created disk layer. It must be
// done after each revert operation, otherwise the new
// disk layer won't be accessible from outside.
db.tree.reset(dl)
}
rawdb.DeleteTrieJournal(db.diskdb)
_, err := truncateFromHead(db.diskdb, db.freezer, dl.stateID())
if err != nil {
return err
}
log.Debug("Recovered state", "root", root, "elapsed", common.PrettyDuration(time.Since(start)))
return nil
}
// Recoverable returns the indicator if the specified state is recoverable.
func (db *Database) Recoverable(root common.Hash) bool {
// Ensure the requested state is a known state.
root = types.TrieRootHash(root)
id := rawdb.ReadStateID(db.diskdb, root)
if id == nil {
return false
}
// Recoverable state must below the disk layer. The recoverable
// state only refers the state that is currently not available,
// but can be restored by applying state history.
dl := db.tree.bottom()
if *id >= dl.stateID() {
return false
}
// Ensure the requested state is a canonical state and all state
// histories in range [id+1, disklayer.ID] are present and complete.
parent := root
return checkHistories(db.freezer, *id+1, dl.stateID()-*id, func(m *meta) error {
if m.parent != parent {
return errors.New("unexpected state history")
}
if len(m.incomplete) > 0 {
return errors.New("incomplete state history")
}
parent = m.root
return nil
}) == nil
}
// Close closes the trie database and the held freezer.
func (db *Database) Close() error {
db.lock.Lock()
defer db.lock.Unlock()
// Set the database to read-only mode to prevent all
// following mutations.
db.readOnly = true
// Release the memory held by clean cache.
db.tree.bottom().resetCache()
// Close the attached state history freezer.
if db.freezer == nil {
return nil
}
return db.freezer.Close()
}
// Size returns the current storage size of the memory cache in front of the
// persistent database layer.
func (db *Database) Size() (diffs common.StorageSize, nodes common.StorageSize) {
db.tree.forEach(func(layer layer) {
if diff, ok := layer.(*diffLayer); ok {
diffs += common.StorageSize(diff.memory)
}
if disk, ok := layer.(*diskLayer); ok {
nodes += disk.size()
}
})
return diffs, nodes
}
// Initialized returns an indicator if the state data is already
// initialized in path-based scheme.
func (db *Database) Initialized(genesisRoot common.Hash) bool {
var inited bool
db.tree.forEach(func(layer layer) {
if layer.rootHash() != types.EmptyRootHash {
inited = true
}
})
if !inited {
inited = rawdb.ReadSnapSyncStatusFlag(db.diskdb) != rawdb.StateSyncUnknown
}
return inited
}
// SetBufferSize sets the node buffer size to the provided value(in bytes).
func (db *Database) SetBufferSize(size int) error {
db.lock.Lock()
defer db.lock.Unlock()
if size > maxBufferSize {
log.Info("Capped node buffer size", "provided", common.StorageSize(size), "adjusted", common.StorageSize(maxBufferSize))
size = maxBufferSize
}
db.bufferSize = size
return db.tree.bottom().setBufferSize(db.bufferSize)
}
// Scheme returns the node scheme used in the database.
func (db *Database) Scheme() string {
return rawdb.PathScheme
}
// modifyAllowed returns the indicator if mutation is allowed. This function
// assumes the db.lock is already held.
func (db *Database) modifyAllowed() error {
if db.readOnly {
return errDatabaseReadOnly
}
if db.waitSync {
return errDatabaseWaitSync
}
return nil
}

608
trie/triedb/pathdb/database_test.go
View file

@ -1,608 +0,0 @@
// 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"
"math/big"
"math/rand"
"testing"
"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/rlp"
"github.com/ethereum/go-ethereum/trie/testutil"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
)
func updateTrie(addrHash common.Hash, root common.Hash, dirties, cleans map[common.Hash][]byte) (common.Hash, *trienode.NodeSet) {
h, err := newTestHasher(addrHash, root, cleans)
if err != nil {
panic(fmt.Errorf("failed to create hasher, err: %w", err))
}
for key, val := range dirties {
if len(val) == 0 {
h.Delete(key.Bytes())
} else {
h.Update(key.Bytes(), val)
}
}
root, nodes, _ := h.Commit(false)
return root, nodes
}
func generateAccount(storageRoot common.Hash) types.StateAccount {
return types.StateAccount{
Nonce: uint64(rand.Intn(100)),
Balance: big.NewInt(rand.Int63()),
CodeHash: testutil.RandBytes(32),
Root: storageRoot,
}
}
const (
createAccountOp int = iota
modifyAccountOp
deleteAccountOp
opLen
)
type genctx struct {
accounts map[common.Hash][]byte
storages map[common.Hash]map[common.Hash][]byte
accountOrigin map[common.Address][]byte
storageOrigin map[common.Address]map[common.Hash][]byte
nodes *trienode.MergedNodeSet
}
func newCtx() *genctx {
return &genctx{
accounts: make(map[common.Hash][]byte),
storages: make(map[common.Hash]map[common.Hash][]byte),
accountOrigin: make(map[common.Address][]byte),
storageOrigin: make(map[common.Address]map[common.Hash][]byte),
nodes: trienode.NewMergedNodeSet(),
}
}
type tester struct {
db *Database
roots []common.Hash
preimages map[common.Hash]common.Address
accounts map[common.Hash][]byte
storages map[common.Hash]map[common.Hash][]byte
// state snapshots
snapAccounts map[common.Hash]map[common.Hash][]byte
snapStorages map[common.Hash]map[common.Hash]map[common.Hash][]byte
}
func newTester(t *testing.T, historyLimit uint64) *tester {
var (
disk, _ = rawdb.NewDatabaseWithFreezer(rawdb.NewMemoryDatabase(), t.TempDir(), "", false)
db = New(disk, &Config{
StateHistory: historyLimit,
CleanCacheSize: 256 * 1024,
DirtyCacheSize: 256 * 1024,
})
obj = &tester{
db: db,
preimages: make(map[common.Hash]common.Address),
accounts: make(map[common.Hash][]byte),
storages: make(map[common.Hash]map[common.Hash][]byte),
snapAccounts: make(map[common.Hash]map[common.Hash][]byte),
snapStorages: make(map[common.Hash]map[common.Hash]map[common.Hash][]byte),
}
)
for i := 0; i < 2*128; i++ {
var parent = types.EmptyRootHash
if len(obj.roots) != 0 {
parent = obj.roots[len(obj.roots)-1]
}
root, nodes, states := obj.generate(parent)
if err := db.Update(root, parent, uint64(i), nodes, states); err != nil {
panic(fmt.Errorf("failed to update state changes, err: %w", err))
}
obj.roots = append(obj.roots, root)
}
return obj
}
func (t *tester) release() {
t.db.Close()
t.db.diskdb.Close()
}
func (t *tester) randAccount() (common.Address, []byte) {
for addrHash, account := range t.accounts {
return t.preimages[addrHash], account
}
return common.Address{}, nil
}
func (t *tester) generateStorage(ctx *genctx, addr common.Address) common.Hash {
var (
addrHash = crypto.Keccak256Hash(addr.Bytes())
storage = make(map[common.Hash][]byte)
origin = make(map[common.Hash][]byte)
)
for i := 0; i < 10; i++ {
v, _ := rlp.EncodeToBytes(common.TrimLeftZeroes(testutil.RandBytes(32)))
hash := testutil.RandomHash()
storage[hash] = v
origin[hash] = nil
}
root, set := updateTrie(addrHash, types.EmptyRootHash, storage, nil)
ctx.storages[addrHash] = storage
ctx.storageOrigin[addr] = origin
ctx.nodes.Merge(set)
return root
}
func (t *tester) mutateStorage(ctx *genctx, addr common.Address, root common.Hash) common.Hash {
var (
addrHash = crypto.Keccak256Hash(addr.Bytes())
storage = make(map[common.Hash][]byte)
origin = make(map[common.Hash][]byte)
)
for hash, val := range t.storages[addrHash] {
origin[hash] = val
storage[hash] = nil
if len(origin) == 3 {
break
}
}
for i := 0; i < 3; i++ {
v, _ := rlp.EncodeToBytes(common.TrimLeftZeroes(testutil.RandBytes(32)))
hash := testutil.RandomHash()
storage[hash] = v
origin[hash] = nil
}
root, set := updateTrie(crypto.Keccak256Hash(addr.Bytes()), root, storage, t.storages[addrHash])
ctx.storages[addrHash] = storage
ctx.storageOrigin[addr] = origin
ctx.nodes.Merge(set)
return root
}
func (t *tester) clearStorage(ctx *genctx, addr common.Address, root common.Hash) common.Hash {
var (
addrHash = crypto.Keccak256Hash(addr.Bytes())
storage = make(map[common.Hash][]byte)
origin = make(map[common.Hash][]byte)
)
for hash, val := range t.storages[addrHash] {
origin[hash] = val
storage[hash] = nil
}
root, set := updateTrie(addrHash, root, storage, t.storages[addrHash])
if root != types.EmptyRootHash {
panic("failed to clear storage trie")
}
ctx.storages[addrHash] = storage
ctx.storageOrigin[addr] = origin
ctx.nodes.Merge(set)
return root
}
func (t *tester) generate(parent common.Hash) (common.Hash, *trienode.MergedNodeSet, *triestate.Set) {
var (
ctx = newCtx()
dirties = make(map[common.Hash]struct{})
)
for i := 0; i < 20; i++ {
switch rand.Intn(opLen) {
case createAccountOp:
// account creation
addr := testutil.RandomAddress()
addrHash := crypto.Keccak256Hash(addr.Bytes())
if _, ok := t.accounts[addrHash]; ok {
continue
}
if _, ok := dirties[addrHash]; ok {
continue
}
dirties[addrHash] = struct{}{}
root := t.generateStorage(ctx, addr)
ctx.accounts[addrHash] = types.SlimAccountRLP(generateAccount(root))
ctx.accountOrigin[addr] = nil
t.preimages[addrHash] = addr
case modifyAccountOp:
// account mutation
addr, account := t.randAccount()
if addr == (common.Address{}) {
continue
}
addrHash := crypto.Keccak256Hash(addr.Bytes())
if _, ok := dirties[addrHash]; ok {
continue
}
dirties[addrHash] = struct{}{}
acct, _ := types.FullAccount(account)
stRoot := t.mutateStorage(ctx, addr, acct.Root)
newAccount := types.SlimAccountRLP(generateAccount(stRoot))
ctx.accounts[addrHash] = newAccount
ctx.accountOrigin[addr] = account
case deleteAccountOp:
// account deletion
addr, account := t.randAccount()
if addr == (common.Address{}) {
continue
}
addrHash := crypto.Keccak256Hash(addr.Bytes())
if _, ok := dirties[addrHash]; ok {
continue
}
dirties[addrHash] = struct{}{}
acct, _ := types.FullAccount(account)
if acct.Root != types.EmptyRootHash {
t.clearStorage(ctx, addr, acct.Root)
}
ctx.accounts[addrHash] = nil
ctx.accountOrigin[addr] = account
}
}
root, set := updateTrie(common.Hash{}, parent, ctx.accounts, t.accounts)
ctx.nodes.Merge(set)
// Save state snapshot before commit
t.snapAccounts[parent] = copyAccounts(t.accounts)
t.snapStorages[parent] = copyStorages(t.storages)
// Commit all changes to live state set
for addrHash, account := range ctx.accounts {
if len(account) == 0 {
delete(t.accounts, addrHash)
} else {
t.accounts[addrHash] = account
}
}
for addrHash, slots := range ctx.storages {
if _, ok := t.storages[addrHash]; !ok {
t.storages[addrHash] = make(map[common.Hash][]byte)
}
for sHash, slot := range slots {
if len(slot) == 0 {
delete(t.storages[addrHash], sHash)
} else {
t.storages[addrHash][sHash] = slot
}
}
}
return root, ctx.nodes, triestate.New(ctx.accountOrigin, ctx.storageOrigin, nil)
}
// lastRoot returns the latest root hash, or empty if nothing is cached.
func (t *tester) lastHash() common.Hash {
if len(t.roots) == 0 {
return common.Hash{}
}
return t.roots[len(t.roots)-1]
}
func (t *tester) verifyState(root common.Hash) error {
reader, err := t.db.Reader(root)
if err != nil {
return err
}
_, err = reader.Node(common.Hash{}, nil, root)
if err != nil {
return errors.New("root node is not available")
}
for addrHash, account := range t.snapAccounts[root] {
blob, err := reader.Node(common.Hash{}, addrHash.Bytes(), crypto.Keccak256Hash(account))
if err != nil || !bytes.Equal(blob, account) {
return fmt.Errorf("account is mismatched: %w", err)
}
}
for addrHash, slots := range t.snapStorages[root] {
for hash, slot := range slots {
blob, err := reader.Node(addrHash, hash.Bytes(), crypto.Keccak256Hash(slot))
if err != nil || !bytes.Equal(blob, slot) {
return fmt.Errorf("slot is mismatched: %w", err)
}
}
}
return nil
}
func (t *tester) verifyHistory() error {
bottom := t.bottomIndex()
for i, root := range t.roots {
// The state history related to the state above disk layer should not exist.
if i > bottom {
_, err := readHistory(t.db.freezer, uint64(i+1))
if err == nil {
return errors.New("unexpected state history")
}
continue
}
// The state history related to the state below or equal to the disk layer
// should exist.
obj, err := readHistory(t.db.freezer, uint64(i+1))
if err != nil {
return err
}
parent := types.EmptyRootHash
if i != 0 {
parent = t.roots[i-1]
}
if obj.meta.parent != parent {
return fmt.Errorf("unexpected parent, want: %x, got: %x", parent, obj.meta.parent)
}
if obj.meta.root != root {
return fmt.Errorf("unexpected root, want: %x, got: %x", root, obj.meta.root)
}
}
return nil
}
// bottomIndex returns the index of current disk layer.
func (t *tester) bottomIndex() int {
bottom := t.db.tree.bottom()
for i := 0; i < len(t.roots); i++ {
if t.roots[i] == bottom.rootHash() {
return i
}
}
return -1
}
func TestDatabaseRollback(t *testing.T) {
// Verify state histories
tester := newTester(t, 0)
defer tester.release()
if err := tester.verifyHistory(); err != nil {
t.Fatalf("Invalid state history, err: %v", err)
}
// Revert database from top to bottom
for i := tester.bottomIndex(); i >= 0; i-- {
root := tester.roots[i]
parent := types.EmptyRootHash
if i > 0 {
parent = tester.roots[i-1]
}
loader := newHashLoader(tester.snapAccounts[root], tester.snapStorages[root])
if err := tester.db.Recover(parent, loader); err != nil {
t.Fatalf("Failed to revert db, err: %v", err)
}
tester.verifyState(parent)
}
if tester.db.tree.len() != 1 {
t.Fatal("Only disk layer is expected")
}
}
func TestDatabaseRecoverable(t *testing.T) {
var (
tester = newTester(t, 0)
index = tester.bottomIndex()
)
defer tester.release()
var cases = []struct {
root common.Hash
expect bool
}{
// Unknown state should be unrecoverable
{common.Hash{0x1}, false},
// Initial state should be recoverable
{types.EmptyRootHash, true},
// Initial state should be recoverable
{common.Hash{}, true},
// Layers below current disk layer are recoverable
{tester.roots[index-1], true},
// Disklayer itself is not recoverable, since it's
// available for accessing.
{tester.roots[index], false},
// Layers above current disk layer are not recoverable
// since they are available for accessing.
{tester.roots[index+1], false},
}
for i, c := range cases {
result := tester.db.Recoverable(c.root)
if result != c.expect {
t.Fatalf("case: %d, unexpected result, want %t, got %t", i, c.expect, result)
}
}
}
func TestDisable(t *testing.T) {
tester := newTester(t, 0)
defer tester.release()
_, stored := rawdb.ReadAccountTrieNode(tester.db.diskdb, nil)
if err := tester.db.Disable(); err != nil {
t.Fatal("Failed to deactivate database")
}
if err := tester.db.Enable(types.EmptyRootHash); err == nil {
t.Fatalf("Invalid activation should be rejected")
}
if err := tester.db.Enable(stored); err != nil {
t.Fatal("Failed to activate database")
}
// Ensure journal is deleted from disk
if blob := rawdb.ReadTrieJournal(tester.db.diskdb); len(blob) != 0 {
t.Fatal("Failed to clean journal")
}
// Ensure all trie histories are removed
n, err := tester.db.freezer.Ancients()
if err != nil {
t.Fatal("Failed to clean state history")
}
if n != 0 {
t.Fatal("Failed to clean state history")
}
// Verify layer tree structure, single disk layer is expected
if tester.db.tree.len() != 1 {
t.Fatalf("Extra layer kept %d", tester.db.tree.len())
}
if tester.db.tree.bottom().rootHash() != stored {
t.Fatalf("Root hash is not matched exp %x got %x", stored, tester.db.tree.bottom().rootHash())
}
}
func TestCommit(t *testing.T) {
tester := newTester(t, 0)
defer tester.release()
if err := tester.db.Commit(tester.lastHash(), false); err != nil {
t.Fatalf("Failed to cap database, err: %v", err)
}
// Verify layer tree structure, single disk layer is expected
if tester.db.tree.len() != 1 {
t.Fatal("Layer tree structure is invalid")
}
if tester.db.tree.bottom().rootHash() != tester.lastHash() {
t.Fatal("Layer tree structure is invalid")
}
// Verify states
if err := tester.verifyState(tester.lastHash()); err != nil {
t.Fatalf("State is invalid, err: %v", err)
}
// Verify state histories
if err := tester.verifyHistory(); err != nil {
t.Fatalf("State history is invalid, err: %v", err)
}
}
func TestJournal(t *testing.T) {
tester := newTester(t, 0)
defer tester.release()
if err := tester.db.Journal(tester.lastHash()); err != nil {
t.Errorf("Failed to journal, err: %v", err)
}
tester.db.Close()
tester.db = New(tester.db.diskdb, nil)
// Verify states including disk layer and all diff on top.
for i := 0; i < len(tester.roots); i++ {
if i >= tester.bottomIndex() {
if err := tester.verifyState(tester.roots[i]); err != nil {
t.Fatalf("Invalid state, err: %v", err)
}
continue
}
if err := tester.verifyState(tester.roots[i]); err == nil {
t.Fatal("Unexpected state")
}
}
}
func TestCorruptedJournal(t *testing.T) {
tester := newTester(t, 0)
defer tester.release()
if err := tester.db.Journal(tester.lastHash()); err != nil {
t.Errorf("Failed to journal, err: %v", err)
}
tester.db.Close()
_, root := rawdb.ReadAccountTrieNode(tester.db.diskdb, nil)
// Mutate the journal in disk, it should be regarded as invalid
blob := rawdb.ReadTrieJournal(tester.db.diskdb)
blob[0] = 1
rawdb.WriteTrieJournal(tester.db.diskdb, blob)
// Verify states, all not-yet-written states should be discarded
tester.db = New(tester.db.diskdb, nil)
for i := 0; i < len(tester.roots); i++ {
if tester.roots[i] == root {
if err := tester.verifyState(root); err != nil {
t.Fatalf("Disk state is corrupted, err: %v", err)
}
continue
}
if err := tester.verifyState(tester.roots[i]); err == nil {
t.Fatal("Unexpected state")
}
}
}
// TestTailTruncateHistory function is designed to test a specific edge case where,
// when history objects are removed from the end, it should trigger a state flush
// if the ID of the new tail object is even higher than the persisted state ID.
//
// For example, let's say the ID of the persistent state is 10, and the current
// history objects range from ID(5) to ID(15). As we accumulate six more objects,
// the history will expand to cover ID(11) to ID(21). ID(11) then becomes the
// oldest history object, and its ID is even higher than the stored state.
//
// In this scenario, it is mandatory to update the persistent state before
// truncating the tail histories. This ensures that the ID of the persistent state
// always falls within the range of [oldest-history-id, latest-history-id].
func TestTailTruncateHistory(t *testing.T) {
tester := newTester(t, 10)
defer tester.release()
tester.db.Close()
tester.db = New(tester.db.diskdb, &Config{StateHistory: 10})
head, err := tester.db.freezer.Ancients()
if err != nil {
t.Fatalf("Failed to obtain freezer head")
}
stored := rawdb.ReadPersistentStateID(tester.db.diskdb)
if head != stored {
t.Fatalf("Failed to truncate excess history object above, stored: %d, head: %d", stored, head)
}
}
// copyAccounts returns a deep-copied account set of the provided one.
func copyAccounts(set map[common.Hash][]byte) map[common.Hash][]byte {
copied := make(map[common.Hash][]byte, len(set))
for key, val := range set {
copied[key] = common.CopyBytes(val)
}
return copied
}
// copyStorages returns a deep-copied storage set of the provided one.
func copyStorages(set map[common.Hash]map[common.Hash][]byte) map[common.Hash]map[common.Hash][]byte {
copied := make(map[common.Hash]map[common.Hash][]byte, len(set))
for addrHash, subset := range set {
copied[addrHash] = make(map[common.Hash][]byte, len(subset))
for key, val := range subset {
copied[addrHash][key] = common.CopyBytes(val)
}
}
return copied
}

174
trie/triedb/pathdb/difflayer.go
View file

@ -1,174 +0,0 @@
// 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 (
"fmt"
"sync"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
)
// diffLayer represents a collection of modifications made to the in-memory tries
// along with associated state changes after running a block on top.
//
// The goal of a diff layer is to act as a journal, tracking recent modifications
// made to the state, that have not yet graduated into a semi-immutable state.
type diffLayer struct {
// Immutables
root common.Hash // Root hash to which this layer diff belongs to
id uint64 // Corresponding state id
block uint64 // Associated block number
nodes map[common.Hash]map[string]*trienode.Node // Cached trie nodes indexed by owner and path
states *triestate.Set // Associated state change set for building history
memory uint64 // Approximate guess as to how much memory we use
parent layer // Parent layer modified by this one, never nil, **can be changed**
lock sync.RWMutex // Lock used to protect parent
}
// newDiffLayer creates a new diff layer on top of an existing layer.
func newDiffLayer(parent layer, root common.Hash, id uint64, block uint64, nodes map[common.Hash]map[string]*trienode.Node, states *triestate.Set) *diffLayer {
var (
size int64
count int
)
dl := &diffLayer{
root: root,
id: id,
block: block,
nodes: nodes,
states: states,
parent: parent,
}
for _, subset := range nodes {
for path, n := range subset {
dl.memory += uint64(n.Size() + len(path))
size += int64(len(n.Blob) + len(path))
}
count += len(subset)
}
if states != nil {
dl.memory += uint64(states.Size())
}
dirtyWriteMeter.Mark(size)
diffLayerNodesMeter.Mark(int64(count))
diffLayerBytesMeter.Mark(int64(dl.memory))
log.Debug("Created new diff layer", "id", id, "block", block, "nodes", count, "size", common.StorageSize(dl.memory))
return dl
}
// rootHash implements the layer interface, returning the root hash of
// corresponding state.
func (dl *diffLayer) rootHash() common.Hash {
return dl.root
}
// stateID implements the layer interface, returning the state id of the layer.
func (dl *diffLayer) stateID() uint64 {
return dl.id
}
// parentLayer implements the layer interface, returning the subsequent
// layer of the diff layer.
func (dl *diffLayer) parentLayer() layer {
dl.lock.RLock()
defer dl.lock.RUnlock()
return dl.parent
}
// node retrieves the node with provided node information. It's the internal
// version of Node function with additional accessed layer tracked. No error
// will be returned if node is not found.
func (dl *diffLayer) node(owner common.Hash, path []byte, hash common.Hash, depth int) ([]byte, error) {
// Hold the lock, ensure the parent won't be changed during the
// state accessing.
dl.lock.RLock()
defer dl.lock.RUnlock()
// If the trie node is known locally, return it
subset, ok := dl.nodes[owner]
if ok {
n, ok := subset[string(path)]
if ok {
// If the trie node is not hash matched, or marked as removed,
// bubble up an error here. It shouldn't happen at all.
if n.Hash != hash {
dirtyFalseMeter.Mark(1)
log.Error("Unexpected trie node in diff layer", "owner", owner, "path", path, "expect", hash, "got", n.Hash)
return nil, newUnexpectedNodeError("diff", hash, n.Hash, owner, path, n.Blob)
}
dirtyHitMeter.Mark(1)
dirtyNodeHitDepthHist.Update(int64(depth))
dirtyReadMeter.Mark(int64(len(n.Blob)))
return n.Blob, nil
}
}
// Trie node unknown to this layer, resolve from parent
if diff, ok := dl.parent.(*diffLayer); ok {
return diff.node(owner, path, hash, depth+1)
}
// Failed to resolve through diff layers, fallback to disk layer
return dl.parent.Node(owner, path, hash)
}
// Node implements the layer interface, retrieving the trie node blob with the
// provided node information. No error will be returned if the node is not found.
func (dl *diffLayer) Node(owner common.Hash, path []byte, hash common.Hash) ([]byte, error) {
return dl.node(owner, path, hash, 0)
}
// update implements the layer interface, creating a new layer on top of the
// existing layer tree with the specified data items.
func (dl *diffLayer) update(root common.Hash, id uint64, block uint64, nodes map[common.Hash]map[string]*trienode.Node, states *triestate.Set) *diffLayer {
return newDiffLayer(dl, root, id, block, nodes, states)
}
// persist flushes the diff layer and all its parent layers to disk layer.
func (dl *diffLayer) persist(force bool) (layer, error) {
if parent, ok := dl.parentLayer().(*diffLayer); ok {
// Hold the lock to prevent any read operation until the new
// parent is linked correctly.
dl.lock.Lock()
// The merging of diff layers starts at the bottom-most layer,
// therefore we recurse down here, flattening on the way up
// (diffToDisk).
result, err := parent.persist(force)
if err != nil {
dl.lock.Unlock()
return nil, err
}
dl.parent = result
dl.lock.Unlock()
}
return diffToDisk(dl, force)
}
// diffToDisk merges a bottom-most diff into the persistent disk layer underneath
// it. The method will panic if called onto a non-bottom-most diff layer.
func diffToDisk(layer *diffLayer, force bool) (layer, error) {
disk, ok := layer.parentLayer().(*diskLayer)
if !ok {
panic(fmt.Sprintf("unknown layer type: %T", layer.parentLayer()))
}
return disk.commit(layer, force)
}

170
trie/triedb/pathdb/difflayer_test.go
View file

@ -1,170 +0,0 @@
// Copyright 2019 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"
"testing"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/trie/testutil"
"github.com/ethereum/go-ethereum/trie/trienode"
)
func emptyLayer() *diskLayer {
return &diskLayer{
db: New(rawdb.NewMemoryDatabase(), nil),
buffer: newNodeBuffer(DefaultBufferSize, nil, 0),
}
}
// goos: darwin
// goarch: arm64
// pkg: github.com/ethereum/go-ethereum/trie
// BenchmarkSearch128Layers
// BenchmarkSearch128Layers-8 243826 4755 ns/op
func BenchmarkSearch128Layers(b *testing.B) { benchmarkSearch(b, 0, 128) }
// goos: darwin
// goarch: arm64
// pkg: github.com/ethereum/go-ethereum/trie
// BenchmarkSearch512Layers
// BenchmarkSearch512Layers-8 49686 24256 ns/op
func BenchmarkSearch512Layers(b *testing.B) { benchmarkSearch(b, 0, 512) }
// goos: darwin
// goarch: arm64
// pkg: github.com/ethereum/go-ethereum/trie
// BenchmarkSearch1Layer
// BenchmarkSearch1Layer-8 14062725 88.40 ns/op
func BenchmarkSearch1Layer(b *testing.B) { benchmarkSearch(b, 127, 128) }
func benchmarkSearch(b *testing.B, depth int, total int) {
var (
npath []byte
nhash common.Hash
nblob []byte
)
// First, we set up 128 diff layers, with 3K items each
fill := func(parent layer, index int) *diffLayer {
nodes := make(map[common.Hash]map[string]*trienode.Node)
nodes[common.Hash{}] = make(map[string]*trienode.Node)
for i := 0; i < 3000; i++ {
var (
path = testutil.RandBytes(32)
node = testutil.RandomNode()
)
nodes[common.Hash{}][string(path)] = trienode.New(node.Hash, node.Blob)
if npath == nil && depth == index {
npath = common.CopyBytes(path)
nblob = common.CopyBytes(node.Blob)
nhash = node.Hash
}
}
return newDiffLayer(parent, common.Hash{}, 0, 0, nodes, nil)
}
var layer layer
layer = emptyLayer()
for i := 0; i < total; i++ {
layer = fill(layer, i)
}
b.ResetTimer()
var (
have []byte
err error
)
for i := 0; i < b.N; i++ {
have, err = layer.Node(common.Hash{}, npath, nhash)
if err != nil {
b.Fatal(err)
}
}
if !bytes.Equal(have, nblob) {
b.Fatalf("have %x want %x", have, nblob)
}
}
// goos: darwin
// goarch: arm64
// pkg: github.com/ethereum/go-ethereum/trie
// BenchmarkPersist
// BenchmarkPersist-8 10 111252975 ns/op
func BenchmarkPersist(b *testing.B) {
// First, we set up 128 diff layers, with 3K items each
fill := func(parent layer) *diffLayer {
nodes := make(map[common.Hash]map[string]*trienode.Node)
nodes[common.Hash{}] = make(map[string]*trienode.Node)
for i := 0; i < 3000; i++ {
var (
path = testutil.RandBytes(32)
node = testutil.RandomNode()
)
nodes[common.Hash{}][string(path)] = trienode.New(node.Hash, node.Blob)
}
return newDiffLayer(parent, common.Hash{}, 0, 0, nodes, nil)
}
for i := 0; i < b.N; i++ {
b.StopTimer()
var layer layer
layer = emptyLayer()
for i := 1; i < 128; i++ {
layer = fill(layer)
}
b.StartTimer()
dl, ok := layer.(*diffLayer)
if !ok {
break
}
dl.persist(false)
}
}
// BenchmarkJournal benchmarks the performance for journaling the layers.
//
// BenchmarkJournal
// BenchmarkJournal-8 10 110969279 ns/op
func BenchmarkJournal(b *testing.B) {
b.SkipNow()
// First, we set up 128 diff layers, with 3K items each
fill := func(parent layer) *diffLayer {
nodes := make(map[common.Hash]map[string]*trienode.Node)
nodes[common.Hash{}] = make(map[string]*trienode.Node)
for i := 0; i < 3000; i++ {
var (
path = testutil.RandBytes(32)
node = testutil.RandomNode()
)
nodes[common.Hash{}][string(path)] = trienode.New(node.Hash, node.Blob)
}
// TODO(rjl493456442) a non-nil state set is expected.
return newDiffLayer(parent, common.Hash{}, 0, 0, nodes, nil)
}
var layer layer
layer = emptyLayer()
for i := 0; i < 128; i++ {
layer = fill(layer)
}
b.ResetTimer()
for i := 0; i < b.N; i++ {
layer.journal(new(bytes.Buffer))
}
}

338
trie/triedb/pathdb/disklayer.go
View file

@ -1,338 +0,0 @@
// 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 (
"errors"
"fmt"
"sync"
"github.com/VictoriaMetrics/fastcache"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
"golang.org/x/crypto/sha3"
)
// diskLayer is a low level persistent layer built on top of a key-value store.
type diskLayer struct {
root common.Hash // Immutable, root hash to which this layer was made for
id uint64 // Immutable, corresponding state id
db *Database // Path-based trie database
cleans *fastcache.Cache // GC friendly memory cache of clean node RLPs
buffer *nodebuffer // Node buffer to aggregate writes
stale bool // Signals that the layer became stale (state progressed)
lock sync.RWMutex // Lock used to protect stale flag
}
// newDiskLayer creates a new disk layer based on the passing arguments.
func newDiskLayer(root common.Hash, id uint64, db *Database, cleans *fastcache.Cache, buffer *nodebuffer) *diskLayer {
// Initialize a clean cache if the memory allowance is not zero
// or reuse the provided cache if it is not nil (inherited from
// the original disk layer).
if cleans == nil && db.config.CleanCacheSize != 0 {
cleans = fastcache.New(db.config.CleanCacheSize)
}
return &diskLayer{
root: root,
id: id,
db: db,
cleans: cleans,
buffer: buffer,
}
}
// root implements the layer interface, returning root hash of corresponding state.
func (dl *diskLayer) rootHash() common.Hash {
return dl.root
}
// stateID implements the layer interface, returning the state id of disk layer.
func (dl *diskLayer) stateID() uint64 {
return dl.id
}
// parent implements the layer interface, returning nil as there's no layer
// below the disk.
func (dl *diskLayer) parentLayer() layer {
return nil
}
// isStale return whether this layer has become stale (was flattened across) or if
// it's still live.
func (dl *diskLayer) isStale() bool {
dl.lock.RLock()
defer dl.lock.RUnlock()
return dl.stale
}
// markStale sets the stale flag as true.
func (dl *diskLayer) markStale() {
dl.lock.Lock()
defer dl.lock.Unlock()
if dl.stale {
panic("triedb disk layer is stale") // we've committed into the same base from two children, boom
}
dl.stale = true
}
// Node implements the layer interface, retrieving the trie node with the
// provided node info. No error will be returned if the node is not found.
func (dl *diskLayer) Node(owner common.Hash, path []byte, hash common.Hash) ([]byte, error) {
dl.lock.RLock()
defer dl.lock.RUnlock()
if dl.stale {
return nil, errSnapshotStale
}
// Try to retrieve the trie node from the not-yet-written
// node buffer first. Note the buffer is lock free since
// it's impossible to mutate the buffer before tagging the
// layer as stale.
n, err := dl.buffer.node(owner, path, hash)
if err != nil {
return nil, err
}
if n != nil {
dirtyHitMeter.Mark(1)
dirtyReadMeter.Mark(int64(len(n.Blob)))
return n.Blob, nil
}
dirtyMissMeter.Mark(1)
// Try to retrieve the trie node from the clean memory cache
key := cacheKey(owner, path)
if dl.cleans != nil {
if blob := dl.cleans.Get(nil, key); len(blob) > 0 {
h := newHasher()
defer h.release()
got := h.hash(blob)
if got == hash {
cleanHitMeter.Mark(1)
cleanReadMeter.Mark(int64(len(blob)))
return blob, nil
}
cleanFalseMeter.Mark(1)
log.Error("Unexpected trie node in clean cache", "owner", owner, "path", path, "expect", hash, "got", got)
}
cleanMissMeter.Mark(1)
}
// Try to retrieve the trie node from the disk.
var (
nBlob []byte
nHash common.Hash
)
if owner == (common.Hash{}) {
nBlob, nHash = rawdb.ReadAccountTrieNode(dl.db.diskdb, path)
} else {
nBlob, nHash = rawdb.ReadStorageTrieNode(dl.db.diskdb, owner, path)
}
if nHash != hash {
diskFalseMeter.Mark(1)
log.Error("Unexpected trie node in disk", "owner", owner, "path", path, "expect", hash, "got", nHash)
return nil, newUnexpectedNodeError("disk", hash, nHash, owner, path, nBlob)
}
if dl.cleans != nil && len(nBlob) > 0 {
dl.cleans.Set(key, nBlob)
cleanWriteMeter.Mark(int64(len(nBlob)))
}
return nBlob, nil
}
// update implements the layer interface, returning a new diff layer on top
// with the given state set.
func (dl *diskLayer) update(root common.Hash, id uint64, block uint64, nodes map[common.Hash]map[string]*trienode.Node, states *triestate.Set) *diffLayer {
return newDiffLayer(dl, root, id, block, nodes, states)
}
// commit merges the given bottom-most diff layer into the node buffer
// and returns a newly constructed disk layer. Note the current disk
// layer must be tagged as stale first to prevent re-access.
func (dl *diskLayer) commit(bottom *diffLayer, force bool) (*diskLayer, error) {
dl.lock.Lock()
defer dl.lock.Unlock()
// Construct and store the state history first. If crash happens after storing
// the state history but without flushing the corresponding states(journal),
// the stored state history will be truncated from head in the next restart.
var (
overflow bool
oldest uint64
)
if dl.db.freezer != nil {
err := writeHistory(dl.db.freezer, bottom)
if err != nil {
return nil, err
}
// Determine if the persisted history object has exceeded the configured
// limitation, set the overflow as true if so.
tail, err := dl.db.freezer.Tail()
if err != nil {
return nil, err
}
limit := dl.db.config.StateHistory
if limit != 0 && bottom.stateID()-tail > limit {
overflow = true
oldest = bottom.stateID() - limit + 1 // track the id of history **after truncation**
}
}
// Mark the diskLayer as stale before applying any mutations on top.
dl.stale = true
// Store the root->id lookup afterwards. All stored lookups are identified
// by the **unique** state root. It's impossible that in the same chain
// blocks are not adjacent but have the same root.
if dl.id == 0 {
rawdb.WriteStateID(dl.db.diskdb, dl.root, 0)
}
rawdb.WriteStateID(dl.db.diskdb, bottom.rootHash(), bottom.stateID())
// Construct a new disk layer by merging the nodes from the provided diff
// layer, and flush the content in disk layer if there are too many nodes
// cached. The clean cache is inherited from the original disk layer.
ndl := newDiskLayer(bottom.root, bottom.stateID(), dl.db, dl.cleans, dl.buffer.commit(bottom.nodes))
// In a unique scenario where the ID of the oldest history object (after tail
// truncation) surpasses the persisted state ID, we take the necessary action
// of forcibly committing the cached dirty nodes to ensure that the persisted
// state ID remains higher.
if !force && rawdb.ReadPersistentStateID(dl.db.diskdb) < oldest {
force = true
}
if err := ndl.buffer.flush(ndl.db.diskdb, ndl.cleans, ndl.id, force); err != nil {
return nil, err
}
// To remove outdated history objects from the end, we set the 'tail' parameter
// to 'oldest-1' due to the offset between the freezer index and the history ID.
if overflow {
pruned, err := truncateFromTail(ndl.db.diskdb, ndl.db.freezer, oldest-1)
if err != nil {
return nil, err
}
log.Debug("Pruned state history", "items", pruned, "tailid", oldest)
}
return ndl, nil
}
// revert applies the given state history and return a reverted disk layer.
func (dl *diskLayer) revert(h *history, loader triestate.TrieLoader) (*diskLayer, error) {
if h.meta.root != dl.rootHash() {
return nil, errUnexpectedHistory
}
// Reject if the provided state history is incomplete. It's due to
// a large construct SELF-DESTRUCT which can't be handled because
// of memory limitation.
if len(h.meta.incomplete) > 0 {
return nil, errors.New("incomplete state history")
}
if dl.id == 0 {
return nil, fmt.Errorf("%w: zero state id", errStateUnrecoverable)
}
// Apply the reverse state changes upon the current state. This must
// be done before holding the lock in order to access state in "this"
// layer.
nodes, err := triestate.Apply(h.meta.parent, h.meta.root, h.accounts, h.storages, loader)
if err != nil {
return nil, err
}
// Mark the diskLayer as stale before applying any mutations on top.
dl.lock.Lock()
defer dl.lock.Unlock()
dl.stale = true
// State change may be applied to node buffer, or the persistent
// state, depends on if node buffer is empty or not. If the node
// buffer is not empty, it means that the state transition that
// needs to be reverted is not yet flushed and cached in node
// buffer, otherwise, manipulate persistent state directly.
if !dl.buffer.empty() {
err := dl.buffer.revert(dl.db.diskdb, nodes)
if err != nil {
return nil, err
}
} else {
batch := dl.db.diskdb.NewBatch()
writeNodes(batch, nodes, dl.cleans)
rawdb.WritePersistentStateID(batch, dl.id-1)
if err := batch.Write(); err != nil {
log.Crit("Failed to write states", "err", err)
}
}
return newDiskLayer(h.meta.parent, dl.id-1, dl.db, dl.cleans, dl.buffer), nil
}
// setBufferSize sets the node buffer size to the provided value.
func (dl *diskLayer) setBufferSize(size int) error {
dl.lock.RLock()
defer dl.lock.RUnlock()
if dl.stale {
return errSnapshotStale
}
return dl.buffer.setSize(size, dl.db.diskdb, dl.cleans, dl.id)
}
// size returns the approximate size of cached nodes in the disk layer.
func (dl *diskLayer) size() common.StorageSize {
dl.lock.RLock()
defer dl.lock.RUnlock()
if dl.stale {
return 0
}
return common.StorageSize(dl.buffer.size)
}
// resetCache releases the memory held by clean cache to prevent memory leak.
func (dl *diskLayer) resetCache() {
dl.lock.RLock()
defer dl.lock.RUnlock()
// Stale disk layer loses the ownership of clean cache.
if dl.stale {
return
}
if dl.cleans != nil {
dl.cleans.Reset()
}
}
// hasher is used to compute the sha256 hash of the provided data.
type hasher struct{ sha crypto.KeccakState }
var hasherPool = sync.Pool{
New: func() interface{} { return &hasher{sha: sha3.NewLegacyKeccak256().(crypto.KeccakState)} },
}
func newHasher() *hasher {
return hasherPool.Get().(*hasher)
}
func (h *hasher) hash(data []byte) common.Hash {
return crypto.HashData(h.sha, data)
}
func (h *hasher) release() {
hasherPool.Put(h)
}

60
trie/triedb/pathdb/errors.go
View file

@ -1,60 +0,0 @@
// Copyright 2023 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 (
"errors"
"fmt"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/common/hexutil"
)
var (
// errDatabaseReadOnly is returned if the database is opened in read only mode
// to prevent any mutation.
errDatabaseReadOnly = errors.New("read only")
// errDatabaseWaitSync is returned if the initial state sync is not completed
// yet and database is disabled to prevent accessing state.
errDatabaseWaitSync = errors.New("waiting for sync")
// errSnapshotStale is returned from data accessors if the underlying layer
// layer had been invalidated due to the chain progressing forward far enough
// to not maintain the layer's original state.
errSnapshotStale = errors.New("layer stale")
// errUnexpectedHistory is returned if an unmatched state history is applied
// to the database for state rollback.
errUnexpectedHistory = errors.New("unexpected state history")
// errStateUnrecoverable is returned if state is required to be reverted to
// a destination without associated state history available.
errStateUnrecoverable = errors.New("state is unrecoverable")
// errUnexpectedNode is returned if the requested node with specified path is
// not hash matched with expectation.
errUnexpectedNode = errors.New("unexpected node")
)
func newUnexpectedNodeError(loc string, expHash common.Hash, gotHash common.Hash, owner common.Hash, path []byte, blob []byte) error {
blobHex := "nil"
if len(blob) > 0 {
blobHex = hexutil.Encode(blob)
}
return fmt.Errorf("%w, loc: %s, node: (%x %v), %x!=%x, blob: %s", errUnexpectedNode, loc, owner, path, expHash, gotHash, blobHex)
}

649
trie/triedb/pathdb/history.go
View file

@ -1,649 +0,0 @@
// 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"
"encoding/binary"
"errors"
"fmt"
"time"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/trie/triestate"
"golang.org/x/exp/slices"
)
// State history records the state changes involved in executing a block. The
// state can be reverted to the previous version by applying the associated
// history object (state reverse diff). State history objects are kept to
// guarantee that the system can perform state rollbacks in case of deep reorg.
//
// Each state transition will generate a state history object. Note that not
// every block has a corresponding state history object. If a block performs
// no state changes whatsoever, no state is created for it. Each state history
// will have a sequentially increasing number acting as its unique identifier.
//
// The state history is written to disk (ancient store) when the corresponding
// diff layer is merged into the disk layer. At the same time, system can prune
// the oldest histories according to config.
//
// Disk State
// ^
// |
// +------------+ +---------+ +---------+ +---------+
// | Init State |---->| State 1 |---->| ... |---->| State n |
// +------------+ +---------+ +---------+ +---------+
//
// +-----------+ +------+ +-----------+
// | History 1 |----> | ... |---->| History n |
// +-----------+ +------+ +-----------+
//
// # Rollback
//
// If the system wants to roll back to a previous state n, it needs to ensure
// all history objects from n+1 up to the current disk layer are existent. The
// history objects are applied to the state in reverse order, starting from the
// current disk layer.
const (
accountIndexSize = common.AddressLength + 13 // The length of encoded account index
slotIndexSize = common.HashLength + 5 // The length of encoded slot index
historyMetaSize = 9 + 2*common.HashLength // The length of fixed size part of meta object
stateHistoryVersion = uint8(0) // initial version of state history structure.
)
// Each state history entry is consisted of five elements:
//
// # metadata
// This object contains a few meta fields, such as the associated state root,
// block number, version tag and so on. This object may contain an extra
// accountHash list which means the storage changes belong to these accounts
// are not complete due to large contract destruction. The incomplete history
// can not be used for rollback and serving archive state request.
//
// # account index
// This object contains some index information of account. For example, offset
// and length indicate the location of the data belonging to the account. Besides,
// storageOffset and storageSlots indicate the storage modification location
// belonging to the account.
//
// The size of each account index is *fixed*, and all indexes are sorted
// lexicographically. Thus binary search can be performed to quickly locate a
// specific account.
//
// # account data
// Account data is a concatenated byte stream composed of all account data.
// The account data can be solved by the offset and length info indicated
// by corresponding account index.
//
// fixed size
// ^ ^
// / \
// +-----------------+-----------------+----------------+-----------------+
// | Account index 1 | Account index 2 | ... | Account index N |
// +-----------------+-----------------+----------------+-----------------+
// |
// | length
// offset |----------------+
// v v
// +----------------+----------------+----------------+----------------+
// | Account data 1 | Account data 2 | ... | Account data N |
// +----------------+----------------+----------------+----------------+
//
// # storage index
// This object is similar with account index. It's also fixed size and contains
// the location info of storage slot data.
//
// # storage data
// Storage data is a concatenated byte stream composed of all storage slot data.
// The storage slot data can be solved by the location info indicated by
// corresponding account index and storage slot index.
//
// fixed size
// ^ ^
// / \
// +-----------------+-----------------+----------------+-----------------+
// | Account index 1 | Account index 2 | ... | Account index N |
// +-----------------+-----------------+----------------+-----------------+
// |
// | storage slots
// storage offset |-----------------------------------------------------+
// v v
// +-----------------+-----------------+-----------------+
// | storage index 1 | storage index 2 | storage index 3 |
// +-----------------+-----------------+-----------------+
// | length
// offset |-------------+
// v v
// +-------------+
// | slot data 1 |
// +-------------+
// accountIndex describes the metadata belonging to an account.
type accountIndex struct {
address common.Address // The address of account
length uint8 // The length of account data, size limited by 255
offset uint32 // The offset of item in account data table
storageOffset uint32 // The offset of storage index in storage index table
storageSlots uint32 // The number of mutated storage slots belonging to the account
}
// encode packs account index into byte stream.
func (i *accountIndex) encode() []byte {
var buf [accountIndexSize]byte
copy(buf[:], i.address.Bytes())
buf[common.AddressLength] = i.length
binary.BigEndian.PutUint32(buf[common.AddressLength+1:], i.offset)
binary.BigEndian.PutUint32(buf[common.AddressLength+5:], i.storageOffset)
binary.BigEndian.PutUint32(buf[common.AddressLength+9:], i.storageSlots)
return buf[:]
}
// decode unpacks account index from byte stream.
func (i *accountIndex) decode(blob []byte) {
i.address = common.BytesToAddress(blob[:common.AddressLength])
i.length = blob[common.AddressLength]
i.offset = binary.BigEndian.Uint32(blob[common.AddressLength+1:])
i.storageOffset = binary.BigEndian.Uint32(blob[common.AddressLength+5:])
i.storageSlots = binary.BigEndian.Uint32(blob[common.AddressLength+9:])
}
// slotIndex describes the metadata belonging to a storage slot.
type slotIndex struct {
hash common.Hash // The hash of slot key
length uint8 // The length of storage slot, up to 32 bytes defined in protocol
offset uint32 // The offset of item in storage slot data table
}
// encode packs slot index into byte stream.
func (i *slotIndex) encode() []byte {
var buf [slotIndexSize]byte
copy(buf[:common.HashLength], i.hash.Bytes())
buf[common.HashLength] = i.length
binary.BigEndian.PutUint32(buf[common.HashLength+1:], i.offset)
return buf[:]
}
// decode unpack slot index from the byte stream.
func (i *slotIndex) decode(blob []byte) {
i.hash = common.BytesToHash(blob[:common.HashLength])
i.length = blob[common.HashLength]
i.offset = binary.BigEndian.Uint32(blob[common.HashLength+1:])
}
// meta describes the meta data of state history object.
type meta struct {
version uint8 // version tag of history object
parent common.Hash // prev-state root before the state transition
root common.Hash // post-state root after the state transition
block uint64 // associated block number
incomplete []common.Address // list of address whose storage set is incomplete
}
// encode packs the meta object into byte stream.
func (m *meta) encode() []byte {
buf := make([]byte, historyMetaSize+len(m.incomplete)*common.AddressLength)
buf[0] = m.version
copy(buf[1:1+common.HashLength], m.parent.Bytes())
copy(buf[1+common.HashLength:1+2*common.HashLength], m.root.Bytes())
binary.BigEndian.PutUint64(buf[1+2*common.HashLength:historyMetaSize], m.block)
for i, h := range m.incomplete {
copy(buf[i*common.AddressLength+historyMetaSize:], h.Bytes())
}
return buf[:]
}
// decode unpacks the meta object from byte stream.
func (m *meta) decode(blob []byte) error {
if len(blob) < 1 {
return fmt.Errorf("no version tag")
}
switch blob[0] {
case stateHistoryVersion:
if len(blob) < historyMetaSize {
return fmt.Errorf("invalid state history meta, len: %d", len(blob))
}
if (len(blob)-historyMetaSize)%common.AddressLength != 0 {
return fmt.Errorf("corrupted state history meta, len: %d", len(blob))
}
m.version = blob[0]
m.parent = common.BytesToHash(blob[1 : 1+common.HashLength])
m.root = common.BytesToHash(blob[1+common.HashLength : 1+2*common.HashLength])
m.block = binary.BigEndian.Uint64(blob[1+2*common.HashLength : historyMetaSize])
for pos := historyMetaSize; pos < len(blob); {
m.incomplete = append(m.incomplete, common.BytesToAddress(blob[pos:pos+common.AddressLength]))
pos += common.AddressLength
}
return nil
default:
return fmt.Errorf("unknown version %d", blob[0])
}
}
// history represents a set of state changes belong to a block along with
// the metadata including the state roots involved in the state transition.
// State history objects in disk are linked with each other by a unique id
// (8-bytes integer), the oldest state history object can be pruned on demand
// in order to control the storage size.
type history struct {
meta *meta // Meta data of history
accounts map[common.Address][]byte // Account data keyed by its address hash
accountList []common.Address // Sorted account hash list
storages map[common.Address]map[common.Hash][]byte // Storage data keyed by its address hash and slot hash
storageList map[common.Address][]common.Hash // Sorted slot hash list
}
// newHistory constructs the state history object with provided state change set.
func newHistory(root common.Hash, parent common.Hash, block uint64, states *triestate.Set) *history {
var (
accountList []common.Address
storageList = make(map[common.Address][]common.Hash)
incomplete []common.Address
)
for addr := range states.Accounts {
accountList = append(accountList, addr)
}
slices.SortFunc(accountList, common.Address.Cmp)
for addr, slots := range states.Storages {
slist := make([]common.Hash, 0, len(slots))
for slotHash := range slots {
slist = append(slist, slotHash)
}
slices.SortFunc(slist, common.Hash.Cmp)
storageList[addr] = slist
}
for addr := range states.Incomplete {
incomplete = append(incomplete, addr)
}
slices.SortFunc(incomplete, common.Address.Cmp)
return &history{
meta: &meta{
version: stateHistoryVersion,
parent: parent,
root: root,
block: block,
incomplete: incomplete,
},
accounts: states.Accounts,
accountList: accountList,
storages: states.Storages,
storageList: storageList,
}
}
// encode serializes the state history and returns four byte streams represent
// concatenated account/storage data, account/storage indexes respectively.
func (h *history) encode() ([]byte, []byte, []byte, []byte) {
var (
slotNumber uint32 // the number of processed slots
accountData []byte // the buffer for concatenated account data
storageData []byte // the buffer for concatenated storage data
accountIndexes []byte // the buffer for concatenated account index
storageIndexes []byte // the buffer for concatenated storage index
)
for _, addr := range h.accountList {
accIndex := accountIndex{
address: addr,
length: uint8(len(h.accounts[addr])),
offset: uint32(len(accountData)),
}
slots, exist := h.storages[addr]
if exist {
// Encode storage slots in order
for _, slotHash := range h.storageList[addr] {
sIndex := slotIndex{
hash: slotHash,
length: uint8(len(slots[slotHash])),
offset: uint32(len(storageData)),
}
storageData = append(storageData, slots[slotHash]...)
storageIndexes = append(storageIndexes, sIndex.encode()...)
}
// Fill up the storage meta in account index
accIndex.storageOffset = slotNumber
accIndex.storageSlots = uint32(len(slots))
slotNumber += uint32(len(slots))
}
accountData = append(accountData, h.accounts[addr]...)
accountIndexes = append(accountIndexes, accIndex.encode()...)
}
return accountData, storageData, accountIndexes, storageIndexes
}
// decoder wraps the byte streams for decoding with extra meta fields.
type decoder struct {
accountData []byte // the buffer for concatenated account data
storageData []byte // the buffer for concatenated storage data
accountIndexes []byte // the buffer for concatenated account index
storageIndexes []byte // the buffer for concatenated storage index
lastAccount *common.Address // the address of last resolved account
lastAccountRead uint32 // the read-cursor position of account data
lastSlotIndexRead uint32 // the read-cursor position of storage slot index
lastSlotDataRead uint32 // the read-cursor position of storage slot data
}
// verify validates the provided byte streams for decoding state history. A few
// checks will be performed to quickly detect data corruption. The byte stream
// is regarded as corrupted if:
//
// - account indexes buffer is empty(empty state set is invalid)
// - account indexes/storage indexer buffer is not aligned
//
// note, these situations are allowed:
//
// - empty account data: all accounts were not present
// - empty storage set: no slots are modified
func (r *decoder) verify() error {
if len(r.accountIndexes)%accountIndexSize != 0 || len(r.accountIndexes) == 0 {
return fmt.Errorf("invalid account index, len: %d", len(r.accountIndexes))
}
if len(r.storageIndexes)%slotIndexSize != 0 {
return fmt.Errorf("invalid storage index, len: %d", len(r.storageIndexes))
}
return nil
}
// readAccount parses the account from the byte stream with specified position.
func (r *decoder) readAccount(pos int) (accountIndex, []byte, error) {
// Decode account index from the index byte stream.
var index accountIndex
if (pos+1)*accountIndexSize > len(r.accountIndexes) {
return accountIndex{}, nil, errors.New("account data buffer is corrupted")
}
index.decode(r.accountIndexes[pos*accountIndexSize : (pos+1)*accountIndexSize])
// Perform validation before parsing account data, ensure
// - account is sorted in order in byte stream
// - account data is strictly encoded with no gap inside
// - account data is not out-of-slice
if r.lastAccount != nil { // zero address is possible
if bytes.Compare(r.lastAccount.Bytes(), index.address.Bytes()) >= 0 {
return accountIndex{}, nil, errors.New("account is not in order")
}
}
if index.offset != r.lastAccountRead {
return accountIndex{}, nil, errors.New("account data buffer is gaped")
}
last := index.offset + uint32(index.length)
if uint32(len(r.accountData)) < last {
return accountIndex{}, nil, errors.New("account data buffer is corrupted")
}
data := r.accountData[index.offset:last]
r.lastAccount = &index.address
r.lastAccountRead = last
return index, data, nil
}
// readStorage parses the storage slots from the byte stream with specified account.
func (r *decoder) readStorage(accIndex accountIndex) ([]common.Hash, map[common.Hash][]byte, error) {
var (
last common.Hash
list []common.Hash
storage = make(map[common.Hash][]byte)
)
for j := 0; j < int(accIndex.storageSlots); j++ {
var (
index slotIndex
start = (accIndex.storageOffset + uint32(j)) * uint32(slotIndexSize)
end = (accIndex.storageOffset + uint32(j+1)) * uint32(slotIndexSize)
)
// Perform validation before parsing storage slot data, ensure
// - slot index is not out-of-slice
// - slot data is not out-of-slice
// - slot is sorted in order in byte stream
// - slot indexes is strictly encoded with no gap inside
// - slot data is strictly encoded with no gap inside
if start != r.lastSlotIndexRead {
return nil, nil, errors.New("storage index buffer is gapped")
}
if uint32(len(r.storageIndexes)) < end {
return nil, nil, errors.New("storage index buffer is corrupted")
}
index.decode(r.storageIndexes[start:end])
if bytes.Compare(last.Bytes(), index.hash.Bytes()) >= 0 {
return nil, nil, errors.New("storage slot is not in order")
}
if index.offset != r.lastSlotDataRead {
return nil, nil, errors.New("storage data buffer is gapped")
}
sEnd := index.offset + uint32(index.length)
if uint32(len(r.storageData)) < sEnd {
return nil, nil, errors.New("storage data buffer is corrupted")
}
storage[index.hash] = r.storageData[r.lastSlotDataRead:sEnd]
list = append(list, index.hash)
last = index.hash
r.lastSlotIndexRead = end
r.lastSlotDataRead = sEnd
}
return list, storage, nil
}
// decode deserializes the account and storage data from the provided byte stream.
func (h *history) decode(accountData, storageData, accountIndexes, storageIndexes []byte) error {
var (
accounts = make(map[common.Address][]byte)
storages = make(map[common.Address]map[common.Hash][]byte)
accountList []common.Address
storageList = make(map[common.Address][]common.Hash)
r = &decoder{
accountData: accountData,
storageData: storageData,
accountIndexes: accountIndexes,
storageIndexes: storageIndexes,
}
)
if err := r.verify(); err != nil {
return err
}
for i := 0; i < len(accountIndexes)/accountIndexSize; i++ {
// Resolve account first
accIndex, accData, err := r.readAccount(i)
if err != nil {
return err
}
accounts[accIndex.address] = accData
accountList = append(accountList, accIndex.address)
// Resolve storage slots
slotList, slotData, err := r.readStorage(accIndex)
if err != nil {
return err
}
if len(slotList) > 0 {
storageList[accIndex.address] = slotList
storages[accIndex.address] = slotData
}
}
h.accounts = accounts
h.accountList = accountList
h.storages = storages
h.storageList = storageList
return nil
}
// readHistory reads and decodes the state history object by the given id.
func readHistory(freezer *rawdb.ResettableFreezer, id uint64) (*history, error) {
blob := rawdb.ReadStateHistoryMeta(freezer, id)
if len(blob) == 0 {
return nil, fmt.Errorf("state history not found %d", id)
}
var m meta
if err := m.decode(blob); err != nil {
return nil, err
}
var (
dec = history{meta: &m}
accountData = rawdb.ReadStateAccountHistory(freezer, id)
storageData = rawdb.ReadStateStorageHistory(freezer, id)
accountIndexes = rawdb.ReadStateAccountIndex(freezer, id)
storageIndexes = rawdb.ReadStateStorageIndex(freezer, id)
)
if err := dec.decode(accountData, storageData, accountIndexes, storageIndexes); err != nil {
return nil, err
}
return &dec, nil
}
// writeHistory persists the state history with the provided state set.
func writeHistory(freezer *rawdb.ResettableFreezer, dl *diffLayer) error {
// Short circuit if state set is not available.
if dl.states == nil {
return errors.New("state change set is not available")
}
var (
start = time.Now()
history = newHistory(dl.rootHash(), dl.parentLayer().rootHash(), dl.block, dl.states)
)
accountData, storageData, accountIndex, storageIndex := history.encode()
dataSize := common.StorageSize(len(accountData) + len(storageData))
indexSize := common.StorageSize(len(accountIndex) + len(storageIndex))
// Write history data into five freezer table respectively.
rawdb.WriteStateHistory(freezer, dl.stateID(), history.meta.encode(), accountIndex, storageIndex, accountData, storageData)
historyDataBytesMeter.Mark(int64(dataSize))
historyIndexBytesMeter.Mark(int64(indexSize))
historyBuildTimeMeter.UpdateSince(start)
log.Debug("Stored state history", "id", dl.stateID(), "block", dl.block, "data", dataSize, "index", indexSize, "elapsed", common.PrettyDuration(time.Since(start)))
return nil
}
// checkHistories retrieves a batch of meta objects with the specified range
// and performs the callback on each item.
func checkHistories(freezer *rawdb.ResettableFreezer, start, count uint64, check func(*meta) error) error {
for count > 0 {
number := count
if number > 10000 {
number = 10000 // split the big read into small chunks
}
blobs, err := rawdb.ReadStateHistoryMetaList(freezer, start, number)
if err != nil {
return err
}
for _, blob := range blobs {
var dec meta
if err := dec.decode(blob); err != nil {
return err
}
if err := check(&dec); err != nil {
return err
}
}
count -= uint64(len(blobs))
start += uint64(len(blobs))
}
return nil
}
// truncateFromHead removes the extra state histories from the head with the given
// parameters. It returns the number of items removed from the head.
func truncateFromHead(db ethdb.Batcher, freezer *rawdb.ResettableFreezer, nhead uint64) (int, error) {
ohead, err := freezer.Ancients()
if err != nil {
return 0, err
}
otail, err := freezer.Tail()
if err != nil {
return 0, err
}
// Ensure that the truncation target falls within the specified range.
if ohead < nhead || nhead < otail {
return 0, fmt.Errorf("out of range, tail: %d, head: %d, target: %d", otail, ohead, nhead)
}
// Short circuit if nothing to truncate.
if ohead == nhead {
return 0, nil
}
// Load the meta objects in range [nhead+1, ohead]
blobs, err := rawdb.ReadStateHistoryMetaList(freezer, nhead+1, ohead-nhead)
if err != nil {
return 0, err
}
batch := db.NewBatch()
for _, blob := range blobs {
var m meta
if err := m.decode(blob); err != nil {
return 0, err
}
rawdb.DeleteStateID(batch, m.root)
}
if err := batch.Write(); err != nil {
return 0, err
}
ohead, err = freezer.TruncateHead(nhead)
if err != nil {
return 0, err
}
return int(ohead - nhead), nil
}
// truncateFromTail removes the extra state histories from the tail with the given
// parameters. It returns the number of items removed from the tail.
func truncateFromTail(db ethdb.Batcher, freezer *rawdb.ResettableFreezer, ntail uint64) (int, error) {
ohead, err := freezer.Ancients()
if err != nil {
return 0, err
}
otail, err := freezer.Tail()
if err != nil {
return 0, err
}
// Ensure that the truncation target falls within the specified range.
if otail > ntail || ntail > ohead {
return 0, fmt.Errorf("out of range, tail: %d, head: %d, target: %d", otail, ohead, ntail)
}
// Short circuit if nothing to truncate.
if otail == ntail {
return 0, nil
}
// Load the meta objects in range [otail+1, ntail]
blobs, err := rawdb.ReadStateHistoryMetaList(freezer, otail+1, ntail-otail)
if err != nil {
return 0, err
}
batch := db.NewBatch()
for _, blob := range blobs {
var m meta
if err := m.decode(blob); err != nil {
return 0, err
}
rawdb.DeleteStateID(batch, m.root)
}
if err := batch.Write(); err != nil {
return 0, err
}
otail, err = freezer.TruncateTail(ntail)
if err != nil {
return 0, err
}
return int(ntail - otail), nil
}

334
trie/triedb/pathdb/history_test.go
View file

@ -1,334 +0,0 @@
// 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"
"fmt"
"reflect"
"testing"
"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/ethdb"
"github.com/ethereum/go-ethereum/rlp"
"github.com/ethereum/go-ethereum/trie/testutil"
"github.com/ethereum/go-ethereum/trie/triestate"
)
// randomStateSet generates a random state change set.
func randomStateSet(n int) *triestate.Set {
var (
accounts = make(map[common.Address][]byte)
storages = make(map[common.Address]map[common.Hash][]byte)
)
for i := 0; i < n; i++ {
addr := testutil.RandomAddress()
storages[addr] = make(map[common.Hash][]byte)
for j := 0; j < 3; j++ {
v, _ := rlp.EncodeToBytes(common.TrimLeftZeroes(testutil.RandBytes(32)))
storages[addr][testutil.RandomHash()] = v
}
account := generateAccount(types.EmptyRootHash)
accounts[addr] = types.SlimAccountRLP(account)
}
return triestate.New(accounts, storages, nil)
}
func makeHistory() *history {
return newHistory(testutil.RandomHash(), types.EmptyRootHash, 0, randomStateSet(3))
}
func makeHistories(n int) []*history {
var (
parent = types.EmptyRootHash
result []*history
)
for i := 0; i < n; i++ {
root := testutil.RandomHash()
h := newHistory(root, parent, uint64(i), randomStateSet(3))
parent = root
result = append(result, h)
}
return result
}
func TestEncodeDecodeHistory(t *testing.T) {
var (
m meta
dec history
obj = makeHistory()
)
// check if meta data can be correctly encode/decode
blob := obj.meta.encode()
if err := m.decode(blob); err != nil {
t.Fatalf("Failed to decode %v", err)
}
if !reflect.DeepEqual(&m, obj.meta) {
t.Fatal("meta is mismatched")
}
// check if account/storage data can be correctly encode/decode
accountData, storageData, accountIndexes, storageIndexes := obj.encode()
if err := dec.decode(accountData, storageData, accountIndexes, storageIndexes); err != nil {
t.Fatalf("Failed to decode, err: %v", err)
}
if !compareSet(dec.accounts, obj.accounts) {
t.Fatal("account data is mismatched")
}
if !compareStorages(dec.storages, obj.storages) {
t.Fatal("storage data is mismatched")
}
if !compareList(dec.accountList, obj.accountList) {
t.Fatal("account list is mismatched")
}
if !compareStorageList(dec.storageList, obj.storageList) {
t.Fatal("storage list is mismatched")
}
}
func checkHistory(t *testing.T, db ethdb.KeyValueReader, freezer *rawdb.ResettableFreezer, id uint64, root common.Hash, exist bool) {
blob := rawdb.ReadStateHistoryMeta(freezer, id)
if exist && len(blob) == 0 {
t.Fatalf("Failed to load trie history, %d", id)
}
if !exist && len(blob) != 0 {
t.Fatalf("Unexpected trie history, %d", id)
}
if exist && rawdb.ReadStateID(db, root) == nil {
t.Fatalf("Root->ID mapping is not found, %d", id)
}
if !exist && rawdb.ReadStateID(db, root) != nil {
t.Fatalf("Unexpected root->ID mapping, %d", id)
}
}
func checkHistoriesInRange(t *testing.T, db ethdb.KeyValueReader, freezer *rawdb.ResettableFreezer, from, to uint64, roots []common.Hash, exist bool) {
for i, j := from, 0; i <= to; i, j = i+1, j+1 {
checkHistory(t, db, freezer, i, roots[j], exist)
}
}
func TestTruncateHeadHistory(t *testing.T) {
var (
roots []common.Hash
hs = makeHistories(10)
db = rawdb.NewMemoryDatabase()
freezer, _ = openFreezer(t.TempDir(), false)
)
defer freezer.Close()
for i := 0; i < len(hs); i++ {
accountData, storageData, accountIndex, storageIndex := hs[i].encode()
rawdb.WriteStateHistory(freezer, uint64(i+1), hs[i].meta.encode(), accountIndex, storageIndex, accountData, storageData)
rawdb.WriteStateID(db, hs[i].meta.root, uint64(i+1))
roots = append(roots, hs[i].meta.root)
}
for size := len(hs); size > 0; size-- {
pruned, err := truncateFromHead(db, freezer, uint64(size-1))
if err != nil {
t.Fatalf("Failed to truncate from head %v", err)
}
if pruned != 1 {
t.Error("Unexpected pruned items", "want", 1, "got", pruned)
}
checkHistoriesInRange(t, db, freezer, uint64(size), uint64(10), roots[size-1:], false)
checkHistoriesInRange(t, db, freezer, uint64(1), uint64(size-1), roots[:size-1], true)
}
}
func TestTruncateTailHistory(t *testing.T) {
var (
roots []common.Hash
hs = makeHistories(10)
db = rawdb.NewMemoryDatabase()
freezer, _ = openFreezer(t.TempDir(), false)
)
defer freezer.Close()
for i := 0; i < len(hs); i++ {
accountData, storageData, accountIndex, storageIndex := hs[i].encode()
rawdb.WriteStateHistory(freezer, uint64(i+1), hs[i].meta.encode(), accountIndex, storageIndex, accountData, storageData)
rawdb.WriteStateID(db, hs[i].meta.root, uint64(i+1))
roots = append(roots, hs[i].meta.root)
}
for newTail := 1; newTail < len(hs); newTail++ {
pruned, _ := truncateFromTail(db, freezer, uint64(newTail))
if pruned != 1 {
t.Error("Unexpected pruned items", "want", 1, "got", pruned)
}
checkHistoriesInRange(t, db, freezer, uint64(1), uint64(newTail), roots[:newTail], false)
checkHistoriesInRange(t, db, freezer, uint64(newTail+1), uint64(10), roots[newTail:], true)
}
}
func TestTruncateTailHistories(t *testing.T) {
var cases = []struct {
limit uint64
expPruned int
maxPruned uint64
minUnpruned uint64
empty bool
}{
{
1, 9, 9, 10, false,
},
{
0, 10, 10, 0 /* no meaning */, true,
},
{
10, 0, 0, 1, false,
},
}
for i, c := range cases {
var (
roots []common.Hash
hs = makeHistories(10)
db = rawdb.NewMemoryDatabase()
freezer, _ = openFreezer(t.TempDir()+fmt.Sprintf("%d", i), false)
)
defer freezer.Close()
for i := 0; i < len(hs); i++ {
accountData, storageData, accountIndex, storageIndex := hs[i].encode()
rawdb.WriteStateHistory(freezer, uint64(i+1), hs[i].meta.encode(), accountIndex, storageIndex, accountData, storageData)
rawdb.WriteStateID(db, hs[i].meta.root, uint64(i+1))
roots = append(roots, hs[i].meta.root)
}
pruned, _ := truncateFromTail(db, freezer, uint64(10)-c.limit)
if pruned != c.expPruned {
t.Error("Unexpected pruned items", "want", c.expPruned, "got", pruned)
}
if c.empty {
checkHistoriesInRange(t, db, freezer, uint64(1), uint64(10), roots, false)
} else {
tail := 10 - int(c.limit)
checkHistoriesInRange(t, db, freezer, uint64(1), c.maxPruned, roots[:tail], false)
checkHistoriesInRange(t, db, freezer, c.minUnpruned, uint64(10), roots[tail:], true)
}
}
}
func TestTruncateOutOfRange(t *testing.T) {
var (
hs = makeHistories(10)
db = rawdb.NewMemoryDatabase()
freezer, _ = openFreezer(t.TempDir(), false)
)
defer freezer.Close()
for i := 0; i < len(hs); i++ {
accountData, storageData, accountIndex, storageIndex := hs[i].encode()
rawdb.WriteStateHistory(freezer, uint64(i+1), hs[i].meta.encode(), accountIndex, storageIndex, accountData, storageData)
rawdb.WriteStateID(db, hs[i].meta.root, uint64(i+1))
}
truncateFromTail(db, freezer, uint64(len(hs)/2))
// Ensure of-out-range truncations are rejected correctly.
head, _ := freezer.Ancients()
tail, _ := freezer.Tail()
cases := []struct {
mode int
target uint64
expErr error
}{
{0, head, nil}, // nothing to delete
{0, head + 1, fmt.Errorf("out of range, tail: %d, head: %d, target: %d", tail, head, head+1)},
{0, tail - 1, fmt.Errorf("out of range, tail: %d, head: %d, target: %d", tail, head, tail-1)},
{1, tail, nil}, // nothing to delete
{1, head + 1, fmt.Errorf("out of range, tail: %d, head: %d, target: %d", tail, head, head+1)},
{1, tail - 1, fmt.Errorf("out of range, tail: %d, head: %d, target: %d", tail, head, tail-1)},
}
for _, c := range cases {
var gotErr error
if c.mode == 0 {
_, gotErr = truncateFromHead(db, freezer, c.target)
} else {
_, gotErr = truncateFromTail(db, freezer, c.target)
}
if !reflect.DeepEqual(gotErr, c.expErr) {
t.Errorf("Unexpected error, want: %v, got: %v", c.expErr, gotErr)
}
}
}
// openFreezer initializes the freezer instance for storing state histories.
func openFreezer(datadir string, readOnly bool) (*rawdb.ResettableFreezer, error) {
return rawdb.NewStateFreezer(datadir, readOnly)
}
func compareSet[k comparable](a, b map[k][]byte) bool {
if len(a) != len(b) {
return false
}
for key, valA := range a {
valB, ok := b[key]
if !ok {
return false
}
if !bytes.Equal(valA, valB) {
return false
}
}
return true
}
func compareList[k comparable](a, b []k) bool {
if len(a) != len(b) {
return false
}
for i := 0; i < len(a); i++ {
if a[i] != b[i] {
return false
}
}
return true
}
func compareStorages(a, b map[common.Address]map[common.Hash][]byte) bool {
if len(a) != len(b) {
return false
}
for h, subA := range a {
subB, ok := b[h]
if !ok {
return false
}
if !compareSet(subA, subB) {
return false
}
}
return true
}
func compareStorageList(a, b map[common.Address][]common.Hash) bool {
if len(a) != len(b) {
return false
}
for h, la := range a {
lb, ok := b[h]
if !ok {
return false
}
if !compareList(la, lb) {
return false
}
}
return true
}

387
trie/triedb/pathdb/journal.go
View file

@ -1,387 +0,0 @@
// 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"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
)
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")
)
const journalVersion uint64 = 0
// journalNode represents a trie node persisted in the journal.
type journalNode struct {
Path []byte // Path of the node in the trie
Blob []byte // RLP-encoded trie node blob, nil means the node is deleted
}
// journalNodes represents a list trie nodes belong to a single account
// or the main account trie.
type journalNodes struct {
Owner common.Hash
Nodes []journalNode
}
// journalAccounts represents a list accounts belong to the layer.
type journalAccounts struct {
Addresses []common.Address
Accounts [][]byte
}
// journalStorage represents a list of storage slots belong to an account.
type journalStorage struct {
Incomplete bool
Account common.Address
Hashes []common.Hash
Slots [][]byte
}
// 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.
_, root := rawdb.ReadAccountTrieNode(db.diskdb, nil)
root = types.TrieRootHash(root)
// 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, newNodeBuffer(db.bufferSize, 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 node buffer
var encoded []journalNodes
if err := r.Decode(&encoded); err != nil {
return nil, fmt.Errorf("load disk nodes: %v", err)
}
nodes := make(map[common.Hash]map[string]*trienode.Node)
for _, entry := range encoded {
subset := make(map[string]*trienode.Node)
for _, n := range entry.Nodes {
if len(n.Blob) > 0 {
subset[string(n.Path)] = trienode.New(crypto.Keccak256Hash(n.Blob), n.Blob)
} else {
subset[string(n.Path)] = trienode.NewDeleted()
}
}
nodes[entry.Owner] = subset
}
// Calculate the internal state transitions by id difference.
base := newDiskLayer(root, id, db, nil, newNodeBuffer(db.bufferSize, nodes, id-stored))
return base, 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 encoded []journalNodes
if err := r.Decode(&encoded); err != nil {
return nil, fmt.Errorf("load diff nodes: %v", err)
}
nodes := make(map[common.Hash]map[string]*trienode.Node)
for _, entry := range encoded {
subset := make(map[string]*trienode.Node)
for _, n := range entry.Nodes {
if len(n.Blob) > 0 {
subset[string(n.Path)] = trienode.New(crypto.Keccak256Hash(n.Blob), n.Blob)
} else {
subset[string(n.Path)] = trienode.NewDeleted()
}
}
nodes[entry.Owner] = subset
}
// Read state changes from journal
var (
jaccounts journalAccounts
jstorages []journalStorage
accounts = make(map[common.Address][]byte)
storages = make(map[common.Address]map[common.Hash][]byte)
incomplete = make(map[common.Address]struct{})
)
if err := r.Decode(&jaccounts); err != nil {
return nil, fmt.Errorf("load diff accounts: %v", err)
}
for i, addr := range jaccounts.Addresses {
accounts[addr] = jaccounts.Accounts[i]
}
if err := r.Decode(&jstorages); err != nil {
return nil, fmt.Errorf("load diff storages: %v", err)
}
for _, entry := range jstorages {
set := make(map[common.Hash][]byte)
for i, h := range entry.Hashes {
if len(entry.Slots[i]) > 0 {
set[h] = entry.Slots[i]
} else {
set[h] = nil
}
}
if entry.Incomplete {
incomplete[entry.Account] = struct{}{}
}
storages[entry.Account] = set
}
return db.loadDiffLayer(newDiffLayer(parent, root, parent.stateID()+1, block, nodes, triestate.New(accounts, storages, incomplete)), 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 all unwritten nodes into the journal
nodes := make([]journalNodes, 0, len(dl.buffer.nodes))
for owner, subset := range dl.buffer.nodes {
entry := journalNodes{Owner: owner}
for path, node := range subset {
entry.Nodes = append(entry.Nodes, journalNode{Path: []byte(path), Blob: node.Blob})
}
nodes = append(nodes, entry)
}
if err := rlp.Encode(w, nodes); err != nil {
return err
}
log.Debug("Journaled pathdb disk layer", "root", dl.root, "nodes", len(dl.buffer.nodes))
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
nodes := make([]journalNodes, 0, len(dl.nodes))
for owner, subset := range dl.nodes {
entry := journalNodes{Owner: owner}
for path, node := range subset {
entry.Nodes = append(entry.Nodes, journalNode{Path: []byte(path), Blob: node.Blob})
}
nodes = append(nodes, entry)
}
if err := rlp.Encode(w, nodes); err != nil {
return err
}
// Write the accumulated state changes into buffer
var jacct journalAccounts
for addr, account := range dl.states.Accounts {
jacct.Addresses = append(jacct.Addresses, addr)
jacct.Accounts = append(jacct.Accounts, account)
}
if err := rlp.Encode(w, jacct); err != nil {
return err
}
storage := make([]journalStorage, 0, len(dl.states.Storages))
for addr, slots := range dl.states.Storages {
entry := journalStorage{Account: addr}
if _, ok := dl.states.Incomplete[addr]; ok {
entry.Incomplete = true
}
for slotHash, slot := range slots {
entry.Hashes = append(entry.Hashes, slotHash)
entry.Slots = append(entry.Slots, slot)
}
storage = append(storage, entry)
}
if err := rlp.Encode(w, storage); err != nil {
return err
}
log.Debug("Journaled pathdb diff layer", "root", dl.root, "parent", dl.parent.rootHash(), "id", dl.stateID(), "block", dl.block, "nodes", len(dl.nodes))
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
}
// The stored state in disk might be empty, convert the
// root to emptyRoot in this case.
_, diskroot := rawdb.ReadAccountTrieNode(db.diskdb, nil)
diskroot = types.TrieRootHash(diskroot)
// Secondly write out the state root in disk, ensure all layers
// on top are continuous with disk.
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
}

214
trie/triedb/pathdb/layertree.go
View file

@ -1,214 +0,0 @@
// 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 (
"errors"
"fmt"
"sync"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
)
// layerTree is a group of state layers identified by the state root.
// This structure defines a few basic operations for manipulating
// state layers linked with each other in a tree structure. It's
// thread-safe to use. However, callers need to ensure the thread-safety
// of the referenced layer by themselves.
type layerTree struct {
lock sync.RWMutex
layers map[common.Hash]layer
}
// newLayerTree constructs the layerTree with the given head layer.
func newLayerTree(head layer) *layerTree {
tree := new(layerTree)
tree.reset(head)
return tree
}
// reset initializes the layerTree by the given head layer.
// All the ancestors will be iterated out and linked in the tree.
func (tree *layerTree) reset(head layer) {
tree.lock.Lock()
defer tree.lock.Unlock()
var layers = make(map[common.Hash]layer)
for head != nil {
layers[head.rootHash()] = head
head = head.parentLayer()
}
tree.layers = layers
}
// get retrieves a layer belonging to the given state root.
func (tree *layerTree) get(root common.Hash) layer {
tree.lock.RLock()
defer tree.lock.RUnlock()
return tree.layers[types.TrieRootHash(root)]
}
// forEach iterates the stored layers inside and applies the
// given callback on them.
func (tree *layerTree) forEach(onLayer func(layer)) {
tree.lock.RLock()
defer tree.lock.RUnlock()
for _, layer := range tree.layers {
onLayer(layer)
}
}
// len returns the number of layers cached.
func (tree *layerTree) len() int {
tree.lock.RLock()
defer tree.lock.RUnlock()
return len(tree.layers)
}
// add inserts a new layer into the tree if it can be linked to an existing old parent.
func (tree *layerTree) add(root common.Hash, parentRoot common.Hash, block uint64, nodes *trienode.MergedNodeSet, states *triestate.Set) error {
// Reject noop updates to avoid self-loops. This is a special case that can
// happen for clique networks and proof-of-stake networks where empty blocks
// don't modify the state (0 block subsidy).
//
// Although we could silently ignore this internally, it should be the caller's
// responsibility to avoid even attempting to insert such a layer.
root, parentRoot = types.TrieRootHash(root), types.TrieRootHash(parentRoot)
if root == parentRoot {
return errors.New("layer cycle")
}
parent := tree.get(parentRoot)
if parent == nil {
return fmt.Errorf("triedb parent [%#x] layer missing", parentRoot)
}
l := parent.update(root, parent.stateID()+1, block, nodes.Flatten(), states)
tree.lock.Lock()
tree.layers[l.rootHash()] = l
tree.lock.Unlock()
return nil
}
// cap traverses downwards the diff tree until the number of allowed diff layers
// are crossed. All diffs beyond the permitted number are flattened downwards.
func (tree *layerTree) cap(root common.Hash, layers int) error {
// Retrieve the head layer to cap from
root = types.TrieRootHash(root)
l := tree.get(root)
if l == nil {
return fmt.Errorf("triedb layer [%#x] missing", root)
}
diff, ok := l.(*diffLayer)
if !ok {
return fmt.Errorf("triedb layer [%#x] is disk layer", root)
}
tree.lock.Lock()
defer tree.lock.Unlock()
// If full commit was requested, flatten the diffs and merge onto disk
if layers == 0 {
base, err := diff.persist(true)
if err != nil {
return err
}
// Replace the entire layer tree with the flat base
tree.layers = map[common.Hash]layer{base.rootHash(): base}
return nil
}
// Dive until we run out of layers or reach the persistent database
for i := 0; i < layers-1; i++ {
// If we still have diff layers below, continue down
if parent, ok := diff.parentLayer().(*diffLayer); ok {
diff = parent
} else {
// Diff stack too shallow, return without modifications
return nil
}
}
// We're out of layers, flatten anything below, stopping if it's the disk or if
// the memory limit is not yet exceeded.
switch parent := diff.parentLayer().(type) {
case *diskLayer:
return nil
case *diffLayer:
// Hold the lock to prevent any read operations until the new
// parent is linked correctly.
diff.lock.Lock()
base, err := parent.persist(false)
if err != nil {
diff.lock.Unlock()
return err
}
tree.layers[base.rootHash()] = base
diff.parent = base
diff.lock.Unlock()
default:
panic(fmt.Sprintf("unknown data layer in triedb: %T", parent))
}
// Remove any layer that is stale or links into a stale layer
children := make(map[common.Hash][]common.Hash)
for root, layer := range tree.layers {
if dl, ok := layer.(*diffLayer); ok {
parent := dl.parentLayer().rootHash()
children[parent] = append(children[parent], root)
}
}
var remove func(root common.Hash)
remove = func(root common.Hash) {
delete(tree.layers, root)
for _, child := range children[root] {
remove(child)
}
delete(children, root)
}
for root, layer := range tree.layers {
if dl, ok := layer.(*diskLayer); ok && dl.isStale() {
remove(root)
}
}
return nil
}
// bottom returns the bottom-most disk layer in this tree.
func (tree *layerTree) bottom() *diskLayer {
tree.lock.RLock()
defer tree.lock.RUnlock()
if len(tree.layers) == 0 {
return nil // Shouldn't happen, empty tree
}
// pick a random one as the entry point
var current layer
for _, layer := range tree.layers {
current = layer
break
}
for current.parentLayer() != nil {
current = current.parentLayer()
}
return current.(*diskLayer)
}

50
trie/triedb/pathdb/metrics.go
View file

@ -1,50 +0,0 @@
// 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 "github.com/ethereum/go-ethereum/metrics"
var (
cleanHitMeter = metrics.NewRegisteredMeter("pathdb/clean/hit", nil)
cleanMissMeter = metrics.NewRegisteredMeter("pathdb/clean/miss", nil)
cleanReadMeter = metrics.NewRegisteredMeter("pathdb/clean/read", nil)
cleanWriteMeter = metrics.NewRegisteredMeter("pathdb/clean/write", nil)
dirtyHitMeter = metrics.NewRegisteredMeter("pathdb/dirty/hit", nil)
dirtyMissMeter = metrics.NewRegisteredMeter("pathdb/dirty/miss", nil)
dirtyReadMeter = metrics.NewRegisteredMeter("pathdb/dirty/read", nil)
dirtyWriteMeter = metrics.NewRegisteredMeter("pathdb/dirty/write", nil)
dirtyNodeHitDepthHist = metrics.NewRegisteredHistogram("pathdb/dirty/depth", nil, metrics.NewExpDecaySample(1028, 0.015))
cleanFalseMeter = metrics.NewRegisteredMeter("pathdb/clean/false", nil)
dirtyFalseMeter = metrics.NewRegisteredMeter("pathdb/dirty/false", nil)
diskFalseMeter = metrics.NewRegisteredMeter("pathdb/disk/false", nil)
commitTimeTimer = metrics.NewRegisteredTimer("pathdb/commit/time", nil)
commitNodesMeter = metrics.NewRegisteredMeter("pathdb/commit/nodes", nil)
commitBytesMeter = metrics.NewRegisteredMeter("pathdb/commit/bytes", nil)
gcNodesMeter = metrics.NewRegisteredMeter("pathdb/gc/nodes", nil)
gcBytesMeter = metrics.NewRegisteredMeter("pathdb/gc/bytes", nil)
diffLayerBytesMeter = metrics.NewRegisteredMeter("pathdb/diff/bytes", nil)
diffLayerNodesMeter = metrics.NewRegisteredMeter("pathdb/diff/nodes", nil)
historyBuildTimeMeter = metrics.NewRegisteredTimer("pathdb/history/time", nil)
historyDataBytesMeter = metrics.NewRegisteredMeter("pathdb/history/bytes/data", nil)
historyIndexBytesMeter = metrics.NewRegisteredMeter("pathdb/history/bytes/index", nil)
)

275
trie/triedb/pathdb/nodebuffer.go
View file

@ -1,275 +0,0 @@
// 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 (
"fmt"
"time"
"github.com/VictoriaMetrics/fastcache"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/rawdb"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/trie/trienode"
)
// nodebuffer is a collection of modified trie nodes to aggregate the disk
// write. The content of the nodebuffer must be checked before diving into
// disk (since it basically is not-yet-written data).
type nodebuffer struct {
layers uint64 // The number of diff layers aggregated inside
size uint64 // The size of aggregated writes
limit uint64 // The maximum memory allowance in bytes
nodes map[common.Hash]map[string]*trienode.Node // The dirty node set, mapped by owner and path
}
// newNodeBuffer initializes the node buffer with the provided nodes.
func newNodeBuffer(limit int, nodes map[common.Hash]map[string]*trienode.Node, layers uint64) *nodebuffer {
if nodes == nil {
nodes = make(map[common.Hash]map[string]*trienode.Node)
}
var size uint64
for _, subset := range nodes {
for path, n := range subset {
size += uint64(len(n.Blob) + len(path))
}
}
return &nodebuffer{
layers: layers,
nodes: nodes,
size: size,
limit: uint64(limit),
}
}
// node retrieves the trie node with given node info.
func (b *nodebuffer) node(owner common.Hash, path []byte, hash common.Hash) (*trienode.Node, error) {
subset, ok := b.nodes[owner]
if !ok {
return nil, nil
}
n, ok := subset[string(path)]
if !ok {
return nil, nil
}
if n.Hash != hash {
dirtyFalseMeter.Mark(1)
log.Error("Unexpected trie node in node buffer", "owner", owner, "path", path, "expect", hash, "got", n.Hash)
return nil, newUnexpectedNodeError("dirty", hash, n.Hash, owner, path, n.Blob)
}
return n, nil
}
// commit merges the dirty nodes into the nodebuffer. This operation won't take
// the ownership of the nodes map which belongs to the bottom-most diff layer.
// It will just hold the node references from the given map which are safe to
// copy.
func (b *nodebuffer) commit(nodes map[common.Hash]map[string]*trienode.Node) *nodebuffer {
var (
delta int64
overwrite int64
overwriteSize int64
)
for owner, subset := range nodes {
current, exist := b.nodes[owner]
if !exist {
// Allocate a new map for the subset instead of claiming it directly
// from the passed map to avoid potential concurrent map read/write.
// The nodes belong to original diff layer are still accessible even
// after merging, thus the ownership of nodes map should still belong
// to original layer and any mutation on it should be prevented.
current = make(map[string]*trienode.Node)
for path, n := range subset {
current[path] = n
delta += int64(len(n.Blob) + len(path))
}
b.nodes[owner] = current
continue
}
for path, n := range subset {
if orig, exist := current[path]; !exist {
delta += int64(len(n.Blob) + len(path))
} else {
delta += int64(len(n.Blob) - len(orig.Blob))
overwrite++
overwriteSize += int64(len(orig.Blob) + len(path))
}
current[path] = n
}
b.nodes[owner] = current
}
b.updateSize(delta)
b.layers++
gcNodesMeter.Mark(overwrite)
gcBytesMeter.Mark(overwriteSize)
return b
}
// revert is the reverse operation of commit. It also merges the provided nodes
// into the nodebuffer, the difference is that the provided node set should
// revert the changes made by the last state transition.
func (b *nodebuffer) revert(db ethdb.KeyValueReader, nodes map[common.Hash]map[string]*trienode.Node) error {
// Short circuit if no embedded state transition to revert.
if b.layers == 0 {
return errStateUnrecoverable
}
b.layers--
// Reset the entire buffer if only a single transition left.
if b.layers == 0 {
b.reset()
return nil
}
var delta int64
for owner, subset := range nodes {
current, ok := b.nodes[owner]
if !ok {
panic(fmt.Sprintf("non-existent subset (%x)", owner))
}
for path, n := range subset {
orig, ok := current[path]
if !ok {
// There is a special case in MPT that one child is removed from
// a fullNode which only has two children, and then a new child
// with different position is immediately inserted into the fullNode.
// In this case, the clean child of the fullNode will also be
// marked as dirty because of node collapse and expansion.
//
// In case of database rollback, don't panic if this "clean"
// node occurs which is not present in buffer.
var nhash common.Hash
if owner == (common.Hash{}) {
_, nhash = rawdb.ReadAccountTrieNode(db, []byte(path))
} else {
_, nhash = rawdb.ReadStorageTrieNode(db, owner, []byte(path))
}
// Ignore the clean node in the case described above.
if nhash == n.Hash {
continue
}
panic(fmt.Sprintf("non-existent node (%x %v) blob: %v", owner, path, crypto.Keccak256Hash(n.Blob).Hex()))
}
current[path] = n
delta += int64(len(n.Blob)) - int64(len(orig.Blob))
}
}
b.updateSize(delta)
return nil
}
// updateSize updates the total cache size by the given delta.
func (b *nodebuffer) updateSize(delta int64) {
size := int64(b.size) + delta
if size >= 0 {
b.size = uint64(size)
return
}
s := b.size
b.size = 0
log.Error("Invalid pathdb buffer size", "prev", common.StorageSize(s), "delta", common.StorageSize(delta))
}
// reset cleans up the disk cache.
func (b *nodebuffer) reset() {
b.layers = 0
b.size = 0
b.nodes = make(map[common.Hash]map[string]*trienode.Node)
}
// empty returns an indicator if nodebuffer contains any state transition inside.
func (b *nodebuffer) empty() bool {
return b.layers == 0
}
// setSize sets the buffer size to the provided number, and invokes a flush
// operation if the current memory usage exceeds the new limit.
func (b *nodebuffer) setSize(size int, db ethdb.KeyValueStore, clean *fastcache.Cache, id uint64) error {
b.limit = uint64(size)
return b.flush(db, clean, id, false)
}
// flush persists the in-memory dirty trie node into the disk if the configured
// memory threshold is reached. Note, all data must be written atomically.
func (b *nodebuffer) flush(db ethdb.KeyValueStore, clean *fastcache.Cache, id uint64, force bool) error {
if b.size <= b.limit && !force {
return nil
}
// Ensure the target state id is aligned with the internal counter.
head := rawdb.ReadPersistentStateID(db)
if head+b.layers != id {
return fmt.Errorf("buffer layers (%d) cannot be applied on top of persisted state id (%d) to reach requested state id (%d)", b.layers, head, id)
}
var (
start = time.Now()
batch = db.NewBatchWithSize(int(b.size))
)
nodes := writeNodes(batch, b.nodes, clean)
rawdb.WritePersistentStateID(batch, id)
// Flush all mutations in a single batch
size := batch.ValueSize()
if err := batch.Write(); err != nil {
return err
}
commitBytesMeter.Mark(int64(size))
commitNodesMeter.Mark(int64(nodes))
commitTimeTimer.UpdateSince(start)
log.Debug("Persisted pathdb nodes", "nodes", len(b.nodes), "bytes", common.StorageSize(size), "elapsed", common.PrettyDuration(time.Since(start)))
b.reset()
return nil
}
// writeNodes writes the trie nodes into the provided database batch.
// Note this function will also inject all the newly written nodes
// into clean cache.
func writeNodes(batch ethdb.Batch, nodes map[common.Hash]map[string]*trienode.Node, clean *fastcache.Cache) (total int) {
for owner, subset := range nodes {
for path, n := range subset {
if n.IsDeleted() {
if owner == (common.Hash{}) {
rawdb.DeleteAccountTrieNode(batch, []byte(path))
} else {
rawdb.DeleteStorageTrieNode(batch, owner, []byte(path))
}
if clean != nil {
clean.Del(cacheKey(owner, []byte(path)))
}
} else {
if owner == (common.Hash{}) {
rawdb.WriteAccountTrieNode(batch, []byte(path), n.Blob)
} else {
rawdb.WriteStorageTrieNode(batch, owner, []byte(path), n.Blob)
}
if clean != nil {
clean.Set(cacheKey(owner, []byte(path)), n.Blob)
}
}
}
total += len(subset)
}
return total
}
// cacheKey constructs the unique key of clean cache.
func cacheKey(owner common.Hash, path []byte) []byte {
if owner == (common.Hash{}) {
return path
}
return append(owner.Bytes(), path...)
}

156
trie/triedb/pathdb/testutils.go
View file

@ -1,156 +0,0 @@
// Copyright 2023 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"
"fmt"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/triestate"
"golang.org/x/exp/slices"
)
// testHasher is a test utility for computing root hash of a batch of state
// elements. The hash algorithm is to sort all the elements in lexicographical
// order, concat the key and value in turn, and perform hash calculation on
// the concatenated bytes. Except the root hash, a nodeset will be returned
// once Commit is called, which contains all the changes made to hasher.
type testHasher struct {
owner common.Hash // owner identifier
root common.Hash // original root
dirties map[common.Hash][]byte // dirty states
cleans map[common.Hash][]byte // clean states
}
// newTestHasher constructs a hasher object with provided states.
func newTestHasher(owner common.Hash, root common.Hash, cleans map[common.Hash][]byte) (*testHasher, error) {
if cleans == nil {
cleans = make(map[common.Hash][]byte)
}
if got, _ := hash(cleans); got != root {
return nil, fmt.Errorf("state root mismatched, want: %x, got: %x", root, got)
}
return &testHasher{
owner: owner,
root: root,
dirties: make(map[common.Hash][]byte),
cleans: cleans,
}, nil
}
// Get returns the value for key stored in the trie.
func (h *testHasher) Get(key []byte) ([]byte, error) {
hash := common.BytesToHash(key)
val, ok := h.dirties[hash]
if ok {
return val, nil
}
return h.cleans[hash], nil
}
// Update associates key with value in the trie.
func (h *testHasher) Update(key, value []byte) error {
h.dirties[common.BytesToHash(key)] = common.CopyBytes(value)
return nil
}
// Delete removes any existing value for key from the trie.
func (h *testHasher) Delete(key []byte) error {
h.dirties[common.BytesToHash(key)] = nil
return nil
}
// Commit computes the new hash of the states and returns the set with all
// state changes.
func (h *testHasher) Commit(collectLeaf bool) (common.Hash, *trienode.NodeSet, error) {
var (
nodes = make(map[common.Hash][]byte)
set = trienode.NewNodeSet(h.owner)
)
for hash, val := range h.cleans {
nodes[hash] = val
}
for hash, val := range h.dirties {
nodes[hash] = val
if bytes.Equal(val, h.cleans[hash]) {
continue
}
if len(val) == 0 {
set.AddNode(hash.Bytes(), trienode.NewDeleted())
} else {
set.AddNode(hash.Bytes(), trienode.New(crypto.Keccak256Hash(val), val))
}
}
root, blob := hash(nodes)
// Include the dirty root node as well.
if root != types.EmptyRootHash && root != h.root {
set.AddNode(nil, trienode.New(root, blob))
}
if root == types.EmptyRootHash && h.root != types.EmptyRootHash {
set.AddNode(nil, trienode.NewDeleted())
}
return root, set, nil
}
// hash performs the hash computation upon the provided states.
func hash(states map[common.Hash][]byte) (common.Hash, []byte) {
var hs []common.Hash
for hash := range states {
hs = append(hs, hash)
}
slices.SortFunc(hs, common.Hash.Cmp)
var input []byte
for _, hash := range hs {
if len(states[hash]) == 0 {
continue
}
input = append(input, hash.Bytes()...)
input = append(input, states[hash]...)
}
if len(input) == 0 {
return types.EmptyRootHash, nil
}
return crypto.Keccak256Hash(input), input
}
type hashLoader struct {
accounts map[common.Hash][]byte
storages map[common.Hash]map[common.Hash][]byte
}
func newHashLoader(accounts map[common.Hash][]byte, storages map[common.Hash]map[common.Hash][]byte) *hashLoader {
return &hashLoader{
accounts: accounts,
storages: storages,
}
}
// OpenTrie opens the main account trie.
func (l *hashLoader) OpenTrie(root common.Hash) (triestate.Trie, error) {
return newTestHasher(common.Hash{}, root, l.accounts)
}
// OpenStorageTrie opens the storage trie of an account.
func (l *hashLoader) OpenStorageTrie(stateRoot common.Hash, addrHash, root common.Hash) (triestate.Trie, error) {
return newTestHasher(addrHash, root, l.storages[addrHash])
}

199
trie/trienode/node.go
View file

@ -1,199 +0,0 @@
// Copyright 2023 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 trienode
import (
"fmt"
"sort"
"strings"
"github.com/ethereum/go-ethereum/common"
)
// Node is a wrapper which contains the encoded blob of the trie node and its
// node hash. It is general enough that can be used to represent trie node
// corresponding to different trie implementations.
type Node struct {
Hash common.Hash // Node hash, empty for deleted node
Blob []byte // Encoded node blob, nil for the deleted node
}
// Size returns the total memory size used by this node.
func (n *Node) Size() int {
return len(n.Blob) + common.HashLength
}
// IsDeleted returns the indicator if the node is marked as deleted.
func (n *Node) IsDeleted() bool {
return len(n.Blob) == 0
}
// New constructs a node with provided node information.
func New(hash common.Hash, blob []byte) *Node {
return &Node{Hash: hash, Blob: blob}
}
// NewDeleted constructs a node which is deleted.
func NewDeleted() *Node { return New(common.Hash{}, nil) }
// leaf represents a trie leaf node
type leaf struct {
Blob []byte // raw blob of leaf
Parent common.Hash // the hash of parent node
}
// NodeSet contains a set of nodes collected during the commit operation.
// Each node is keyed by path. It's not thread-safe to use.
type NodeSet struct {
Owner common.Hash
Leaves []*leaf
Nodes map[string]*Node
updates int // the count of updated and inserted nodes
deletes int // the count of deleted nodes
}
// NewNodeSet initializes a node set. The owner is zero for the account trie and
// the owning account address hash for storage tries.
func NewNodeSet(owner common.Hash) *NodeSet {
return &NodeSet{
Owner: owner,
Nodes: make(map[string]*Node),
}
}
// ForEachWithOrder iterates the nodes with the order from bottom to top,
// right to left, nodes with the longest path will be iterated first.
func (set *NodeSet) ForEachWithOrder(callback func(path string, n *Node)) {
var paths []string
for path := range set.Nodes {
paths = append(paths, path)
}
// Bottom-up, the longest path first
sort.Sort(sort.Reverse(sort.StringSlice(paths)))
for _, path := range paths {
callback(path, set.Nodes[path])
}
}
// AddNode adds the provided node into set.
func (set *NodeSet) AddNode(path []byte, n *Node) {
if n.IsDeleted() {
set.deletes += 1
} else {
set.updates += 1
}
set.Nodes[string(path)] = n
}
// Merge adds a set of nodes into the set.
func (set *NodeSet) Merge(owner common.Hash, nodes map[string]*Node) error {
if set.Owner != owner {
return fmt.Errorf("nodesets belong to different owner are not mergeable %x-%x", set.Owner, owner)
}
for path, node := range nodes {
prev, ok := set.Nodes[path]
if ok {
// overwrite happens, revoke the counter
if prev.IsDeleted() {
set.deletes -= 1
} else {
set.updates -= 1
}
}
set.AddNode([]byte(path), node)
}
return nil
}
// AddLeaf adds the provided leaf node into set. TODO(rjl493456442) how can
// we get rid of it?
func (set *NodeSet) AddLeaf(parent common.Hash, blob []byte) {
set.Leaves = append(set.Leaves, &leaf{Blob: blob, Parent: parent})
}
// Size returns the number of dirty nodes in set.
func (set *NodeSet) Size() (int, int) {
return set.updates, set.deletes
}
// Hashes returns the hashes of all updated nodes. TODO(rjl493456442) how can
// we get rid of it?
func (set *NodeSet) Hashes() []common.Hash {
var ret []common.Hash
for _, node := range set.Nodes {
ret = append(ret, node.Hash)
}
return ret
}
// Summary returns a string-representation of the NodeSet.
func (set *NodeSet) Summary() string {
var out = new(strings.Builder)
fmt.Fprintf(out, "nodeset owner: %v\n", set.Owner)
if set.Nodes != nil {
for path, n := range set.Nodes {
// Deletion
if n.IsDeleted() {
fmt.Fprintf(out, " [-]: %x\n", path)
continue
}
// Insertion or update
fmt.Fprintf(out, " [+/*]: %x -> %v \n", path, n.Hash)
}
}
for _, n := range set.Leaves {
fmt.Fprintf(out, "[leaf]: %v\n", n)
}
return out.String()
}
// MergedNodeSet represents a merged node set for a group of tries.
type MergedNodeSet struct {
Sets map[common.Hash]*NodeSet
}
// NewMergedNodeSet initializes an empty merged set.
func NewMergedNodeSet() *MergedNodeSet {
return &MergedNodeSet{Sets: make(map[common.Hash]*NodeSet)}
}
// NewWithNodeSet constructs a merged nodeset with the provided single set.
func NewWithNodeSet(set *NodeSet) *MergedNodeSet {
merged := NewMergedNodeSet()
merged.Merge(set)
return merged
}
// Merge merges the provided dirty nodes of a trie into the set. The assumption
// is held that no duplicated set belonging to the same trie will be merged twice.
func (set *MergedNodeSet) Merge(other *NodeSet) error {
subset, present := set.Sets[other.Owner]
if present {
return subset.Merge(other.Owner, other.Nodes)
}
set.Sets[other.Owner] = other
return nil
}
// Flatten returns a two-dimensional map for internal nodes.
func (set *MergedNodeSet) Flatten() map[common.Hash]map[string]*Node {
nodes := make(map[common.Hash]map[string]*Node)
for owner, set := range set.Sets {
nodes[owner] = set.Nodes
}
return nodes
}

162
trie/trienode/proof.go
View file

@ -1,162 +0,0 @@
// Copyright 2017 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 trienode
import (
"errors"
"sync"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/rlp"
)
// ProofSet stores a set of trie nodes. It implements trie.Database and can also
// act as a cache for another trie.Database.
type ProofSet struct {
nodes map[string][]byte
order []string
dataSize int
lock sync.RWMutex
}
// NewProofSet creates an empty node set
func NewProofSet() *ProofSet {
return &ProofSet{
nodes: make(map[string][]byte),
}
}
// Put stores a new node in the set
func (db *ProofSet) Put(key []byte, value []byte) error {
db.lock.Lock()
defer db.lock.Unlock()
if _, ok := db.nodes[string(key)]; ok {
return nil
}
keystr := string(key)
db.nodes[keystr] = common.CopyBytes(value)
db.order = append(db.order, keystr)
db.dataSize += len(value)
return nil
}
// Delete removes a node from the set
func (db *ProofSet) Delete(key []byte) error {
db.lock.Lock()
defer db.lock.Unlock()
delete(db.nodes, string(key))
return nil
}
// Get returns a stored node
func (db *ProofSet) Get(key []byte) ([]byte, error) {
db.lock.RLock()
defer db.lock.RUnlock()
if entry, ok := db.nodes[string(key)]; ok {
return entry, nil
}
return nil, errors.New("not found")
}
// Has returns true if the node set contains the given key
func (db *ProofSet) Has(key []byte) (bool, error) {
_, err := db.Get(key)
return err == nil, nil
}
// KeyCount returns the number of nodes in the set
func (db *ProofSet) KeyCount() int {
db.lock.RLock()
defer db.lock.RUnlock()
return len(db.nodes)
}
// DataSize returns the aggregated data size of nodes in the set
func (db *ProofSet) DataSize() int {
db.lock.RLock()
defer db.lock.RUnlock()
return db.dataSize
}
// List converts the node set to a ProofList
func (db *ProofSet) List() ProofList {
db.lock.RLock()
defer db.lock.RUnlock()
var values ProofList
for _, key := range db.order {
values = append(values, db.nodes[key])
}
return values
}
// Store writes the contents of the set to the given database
func (db *ProofSet) Store(target ethdb.KeyValueWriter) {
db.lock.RLock()
defer db.lock.RUnlock()
for key, value := range db.nodes {
target.Put([]byte(key), value)
}
}
// ProofList stores an ordered list of trie nodes. It implements ethdb.KeyValueWriter.
type ProofList []rlp.RawValue
// Store writes the contents of the list to the given database
func (n ProofList) Store(db ethdb.KeyValueWriter) {
for _, node := range n {
db.Put(crypto.Keccak256(node), node)
}
}
// Set converts the node list to a ProofSet
func (n ProofList) Set() *ProofSet {
db := NewProofSet()
n.Store(db)
return db
}
// Put stores a new node at the end of the list
func (n *ProofList) Put(key []byte, value []byte) error {
*n = append(*n, value)
return nil
}
// Delete panics as there's no reason to remove a node from the list.
func (n *ProofList) Delete(key []byte) error {
panic("not supported")
}
// DataSize returns the aggregated data size of nodes in the list
func (n ProofList) DataSize() int {
var size int
for _, node := range n {
size += len(node)
}
return size
}

276
trie/triestate/state.go
View file

@ -1,276 +0,0 @@
// Copyright 2023 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 triestate
import (
"errors"
"fmt"
"sync"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/crypto"
"github.com/ethereum/go-ethereum/rlp"
"github.com/ethereum/go-ethereum/trie/trienode"
"golang.org/x/crypto/sha3"
)
// Trie is an Ethereum state trie, can be implemented by Ethereum Merkle Patricia
// tree or Verkle tree.
type Trie interface {
// Get returns the value for key stored in the trie.
Get(key []byte) ([]byte, error)
// Update associates key with value in the trie.
Update(key, value []byte) error
// Delete removes any existing value for key from the trie.
Delete(key []byte) error
// Commit the trie and returns a set of dirty nodes generated along with
// the new root hash.
Commit(collectLeaf bool) (common.Hash, *trienode.NodeSet, error)
}
// TrieLoader wraps functions to load tries.
type TrieLoader interface {
// OpenTrie opens the main account trie.
OpenTrie(root common.Hash) (Trie, error)
// OpenStorageTrie opens the storage trie of an account.
OpenStorageTrie(stateRoot common.Hash, addrHash, root common.Hash) (Trie, error)
}
// Set represents a collection of mutated states during a state transition.
// The value refers to the original content of state before the transition
// is made. Nil means that the state was not present previously.
type Set struct {
Accounts map[common.Address][]byte // Mutated account set, nil means the account was not present
Storages map[common.Address]map[common.Hash][]byte // Mutated storage set, nil means the slot was not present
Incomplete map[common.Address]struct{} // Indicator whether the storage is incomplete due to large deletion
size common.StorageSize // Approximate size of set
}
// New constructs the state set with provided data.
func New(accounts map[common.Address][]byte, storages map[common.Address]map[common.Hash][]byte, incomplete map[common.Address]struct{}) *Set {
return &Set{
Accounts: accounts,
Storages: storages,
Incomplete: incomplete,
}
}
// Size returns the approximate memory size occupied by the set.
func (s *Set) Size() common.StorageSize {
if s.size != 0 {
return s.size
}
for _, account := range s.Accounts {
s.size += common.StorageSize(common.AddressLength + len(account))
}
for _, slots := range s.Storages {
for _, val := range slots {
s.size += common.StorageSize(common.HashLength + len(val))
}
s.size += common.StorageSize(common.AddressLength)
}
s.size += common.StorageSize(common.AddressLength * len(s.Incomplete))
return s.size
}
// context wraps all fields for executing state diffs.
type context struct {
prevRoot common.Hash
postRoot common.Hash
accounts map[common.Address][]byte
storages map[common.Address]map[common.Hash][]byte
accountTrie Trie
nodes *trienode.MergedNodeSet
}
// Apply traverses the provided state diffs, apply them in the associated
// post-state and return the generated dirty trie nodes. The state can be
// loaded via the provided trie loader.
func Apply(prevRoot common.Hash, postRoot common.Hash, accounts map[common.Address][]byte, storages map[common.Address]map[common.Hash][]byte, loader TrieLoader) (map[common.Hash]map[string]*trienode.Node, error) {
tr, err := loader.OpenTrie(postRoot)
if err != nil {
return nil, err
}
ctx := &context{
prevRoot: prevRoot,
postRoot: postRoot,
accounts: accounts,
storages: storages,
accountTrie: tr,
nodes: trienode.NewMergedNodeSet(),
}
for addr, account := range accounts {
var err error
if len(account) == 0 {
err = deleteAccount(ctx, loader, addr)
} else {
err = updateAccount(ctx, loader, addr)
}
if err != nil {
return nil, fmt.Errorf("failed to revert state, err: %w", err)
}
}
root, result, err := tr.Commit(false)
if err != nil {
return nil, err
}
if root != prevRoot {
return nil, fmt.Errorf("failed to revert state, want %#x, got %#x", prevRoot, root)
}
if err := ctx.nodes.Merge(result); err != nil {
return nil, err
}
return ctx.nodes.Flatten(), nil
}
// updateAccount the account was present in prev-state, and may or may not
// existent in post-state. Apply the reverse diff and verify if the storage
// root matches the one in prev-state account.
func updateAccount(ctx *context, loader TrieLoader, addr common.Address) error {
// The account was present in prev-state, decode it from the
// 'slim-rlp' format bytes.
h := newHasher()
defer h.release()
addrHash := h.hash(addr.Bytes())
prev, err := types.FullAccount(ctx.accounts[addr])
if err != nil {
return err
}
// The account may or may not existent in post-state, try to
// load it and decode if it's found.
blob, err := ctx.accountTrie.Get(addrHash.Bytes())
if err != nil {
return err
}
post := types.NewEmptyStateAccount()
if len(blob) != 0 {
if err := rlp.DecodeBytes(blob, &post); err != nil {
return err
}
}
// Apply all storage changes into the post-state storage trie.
st, err := loader.OpenStorageTrie(ctx.postRoot, addrHash, post.Root)
if err != nil {
return err
}
for key, val := range ctx.storages[addr] {
var err error
if len(val) == 0 {
err = st.Delete(key.Bytes())
} else {
err = st.Update(key.Bytes(), val)
}
if err != nil {
return err
}
}
root, result, err := st.Commit(false)
if err != nil {
return err
}
if root != prev.Root {
return errors.New("failed to reset storage trie")
}
// The returned set can be nil if storage trie is not changed
// at all.
if result != nil {
if err := ctx.nodes.Merge(result); err != nil {
return err
}
}
// Write the prev-state account into the main trie
full, err := rlp.EncodeToBytes(prev)
if err != nil {
return err
}
return ctx.accountTrie.Update(addrHash.Bytes(), full)
}
// deleteAccount the account was not present in prev-state, and is expected
// to be existent in post-state. Apply the reverse diff and verify if the
// account and storage is wiped out correctly.
func deleteAccount(ctx *context, loader TrieLoader, addr common.Address) error {
// The account must be existent in post-state, load the account.
h := newHasher()
defer h.release()
addrHash := h.hash(addr.Bytes())
blob, err := ctx.accountTrie.Get(addrHash.Bytes())
if err != nil {
return err
}
if len(blob) == 0 {
return fmt.Errorf("account is non-existent %#x", addrHash)
}
var post types.StateAccount
if err := rlp.DecodeBytes(blob, &post); err != nil {
return err
}
st, err := loader.OpenStorageTrie(ctx.postRoot, addrHash, post.Root)
if err != nil {
return err
}
for key, val := range ctx.storages[addr] {
if len(val) != 0 {
return errors.New("expect storage deletion")
}
if err := st.Delete(key.Bytes()); err != nil {
return err
}
}
root, result, err := st.Commit(false)
if err != nil {
return err
}
if root != types.EmptyRootHash {
return errors.New("failed to clear storage trie")
}
// The returned set can be nil if storage trie is not changed
// at all.
if result != nil {
if err := ctx.nodes.Merge(result); err != nil {
return err
}
}
// Delete the post-state account from the main trie.
return ctx.accountTrie.Delete(addrHash.Bytes())
}
// hasher is used to compute the sha256 hash of the provided data.
type hasher struct{ sha crypto.KeccakState }
var hasherPool = sync.Pool{
New: func() interface{} { return &hasher{sha: sha3.NewLegacyKeccak256().(crypto.KeccakState)} },
}
func newHasher() *hasher {
return hasherPool.Get().(*hasher)
}
func (h *hasher) hash(data []byte) common.Hash {
return crypto.HashData(h.sha, data)
}
func (h *hasher) release() {
hasherPool.Put(h)
}

342
trie/utils/verkle.go
View file

@ -1,342 +0,0 @@
// Copyright 2023 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 utils
import (
"encoding/binary"
"sync"
"github.com/crate-crypto/go-ipa/bandersnatch/fr"
"github.com/ethereum/go-ethereum/common/lru"
"github.com/ethereum/go-ethereum/metrics"
"github.com/gballet/go-verkle"
"github.com/holiman/uint256"
)
const (
// The spec of verkle key encoding can be found here.
// https://notes.ethereum.org/@vbuterin/verkle_tree_eip#Tree-embedding
VersionLeafKey = 0
BalanceLeafKey = 1
NonceLeafKey = 2
CodeKeccakLeafKey = 3
CodeSizeLeafKey = 4
)
var (
zero = uint256.NewInt(0)
verkleNodeWidthLog2 = 8
headerStorageOffset = uint256.NewInt(64)
mainStorageOffsetLshVerkleNodeWidth = new(uint256.Int).Lsh(uint256.NewInt(256), 31-uint(verkleNodeWidthLog2))
codeOffset = uint256.NewInt(128)
verkleNodeWidth = uint256.NewInt(256)
codeStorageDelta = uint256.NewInt(0).Sub(codeOffset, headerStorageOffset)
index0Point *verkle.Point // pre-computed commitment of polynomial [2+256*64]
// cacheHitGauge is the metric to track how many cache hit occurred.
cacheHitGauge = metrics.NewRegisteredGauge("trie/verkle/cache/hit", nil)
// cacheMissGauge is the metric to track how many cache miss occurred.
cacheMissGauge = metrics.NewRegisteredGauge("trie/verkle/cache/miss", nil)
)
func init() {
// The byte array is the Marshalled output of the point computed as such:
//
// var (
// config = verkle.GetConfig()
// fr verkle.Fr
// )
// verkle.FromLEBytes(&fr, []byte{2, 64})
// point := config.CommitToPoly([]verkle.Fr{fr}, 1)
index0Point = new(verkle.Point)
err := index0Point.SetBytes([]byte{34, 25, 109, 242, 193, 5, 144, 224, 76, 52, 189, 92, 197, 126, 9, 145, 27, 152, 199, 130, 165, 3, 210, 27, 193, 131, 142, 28, 110, 26, 16, 191})
if err != nil {
panic(err)
}
}
// PointCache is the LRU cache for storing evaluated address commitment.
type PointCache struct {
lru lru.BasicLRU[string, *verkle.Point]
lock sync.RWMutex
}
// NewPointCache returns the cache with specified size.
func NewPointCache(maxItems int) *PointCache {
return &PointCache{
lru: lru.NewBasicLRU[string, *verkle.Point](maxItems),
}
}
// Get returns the cached commitment for the specified address, or computing
// it on the flight.
func (c *PointCache) Get(addr []byte) *verkle.Point {
c.lock.Lock()
defer c.lock.Unlock()
p, ok := c.lru.Get(string(addr))
if ok {
cacheHitGauge.Inc(1)
return p
}
cacheMissGauge.Inc(1)
p = evaluateAddressPoint(addr)
c.lru.Add(string(addr), p)
return p
}
// GetStem returns the first 31 bytes of the tree key as the tree stem. It only
// works for the account metadata whose treeIndex is 0.
func (c *PointCache) GetStem(addr []byte) []byte {
p := c.Get(addr)
return pointToHash(p, 0)[:31]
}
// GetTreeKey performs both the work of the spec's get_tree_key function, and that
// of pedersen_hash: it builds the polynomial in pedersen_hash without having to
// create a mostly zero-filled buffer and "type cast" it to a 128-long 16-byte
// array. Since at most the first 5 coefficients of the polynomial will be non-zero,
// these 5 coefficients are created directly.
func GetTreeKey(address []byte, treeIndex *uint256.Int, subIndex byte) []byte {
if len(address) < 32 {
var aligned [32]byte
address = append(aligned[:32-len(address)], address...)
}
// poly = [2+256*64, address_le_low, address_le_high, tree_index_le_low, tree_index_le_high]
var poly [5]fr.Element
// 32-byte address, interpreted as two little endian
// 16-byte numbers.
verkle.FromLEBytes(&poly[1], address[:16])
verkle.FromLEBytes(&poly[2], address[16:])
// treeIndex must be interpreted as a 32-byte aligned little-endian integer.
// e.g: if treeIndex is 0xAABBCC, we need the byte representation to be 0xCCBBAA00...00.
// poly[3] = LE({CC,BB,AA,00...0}) (16 bytes), poly[4]=LE({00,00,...}) (16 bytes).
//
// To avoid unnecessary endianness conversions for go-ipa, we do some trick:
// - poly[3]'s byte representation is the same as the *top* 16 bytes (trieIndexBytes[16:]) of
// 32-byte aligned big-endian representation (BE({00,...,AA,BB,CC})).
// - poly[4]'s byte representation is the same as the *low* 16 bytes (trieIndexBytes[:16]) of
// the 32-byte aligned big-endian representation (BE({00,00,...}).
trieIndexBytes := treeIndex.Bytes32()
verkle.FromBytes(&poly[3], trieIndexBytes[16:])
verkle.FromBytes(&poly[4], trieIndexBytes[:16])
cfg := verkle.GetConfig()
ret := cfg.CommitToPoly(poly[:], 0)
// add a constant point corresponding to poly[0]=[2+256*64].
ret.Add(ret, index0Point)
return pointToHash(ret, subIndex)
}
// GetTreeKeyWithEvaluatedAddress is basically identical to GetTreeKey, the only
// difference is a part of polynomial is already evaluated.
//
// Specifically, poly = [2+256*64, address_le_low, address_le_high] is already
// evaluated.
func GetTreeKeyWithEvaluatedAddress(evaluated *verkle.Point, treeIndex *uint256.Int, subIndex byte) []byte {
var poly [5]fr.Element
poly[0].SetZero()
poly[1].SetZero()
poly[2].SetZero()
// little-endian, 32-byte aligned treeIndex
var index [32]byte
for i := 0; i < len(treeIndex); i++ {
binary.LittleEndian.PutUint64(index[i*8:(i+1)*8], treeIndex[i])
}
verkle.FromLEBytes(&poly[3], index[:16])
verkle.FromLEBytes(&poly[4], index[16:])
cfg := verkle.GetConfig()
ret := cfg.CommitToPoly(poly[:], 0)
// add the pre-evaluated address
ret.Add(ret, evaluated)
return pointToHash(ret, subIndex)
}
// VersionKey returns the verkle tree key of the version field for the specified account.
func VersionKey(address []byte) []byte {
return GetTreeKey(address, zero, VersionLeafKey)
}
// BalanceKey returns the verkle tree key of the balance field for the specified account.
func BalanceKey(address []byte) []byte {
return GetTreeKey(address, zero, BalanceLeafKey)
}
// NonceKey returns the verkle tree key of the nonce field for the specified account.
func NonceKey(address []byte) []byte {
return GetTreeKey(address, zero, NonceLeafKey)
}
// CodeKeccakKey returns the verkle tree key of the code keccak field for
// the specified account.
func CodeKeccakKey(address []byte) []byte {
return GetTreeKey(address, zero, CodeKeccakLeafKey)
}
// CodeSizeKey returns the verkle tree key of the code size field for the
// specified account.
func CodeSizeKey(address []byte) []byte {
return GetTreeKey(address, zero, CodeSizeLeafKey)
}
func codeChunkIndex(chunk *uint256.Int) (*uint256.Int, byte) {
var (
chunkOffset = new(uint256.Int).Add(codeOffset, chunk)
treeIndex = new(uint256.Int).Div(chunkOffset, verkleNodeWidth)
subIndexMod = new(uint256.Int).Mod(chunkOffset, verkleNodeWidth)
)
var subIndex byte
if len(subIndexMod) != 0 {
subIndex = byte(subIndexMod[0])
}
return treeIndex, subIndex
}
// CodeChunkKey returns the verkle tree key of the code chunk for the
// specified account.
func CodeChunkKey(address []byte, chunk *uint256.Int) []byte {
treeIndex, subIndex := codeChunkIndex(chunk)
return GetTreeKey(address, treeIndex, subIndex)
}
func storageIndex(bytes []byte) (*uint256.Int, byte) {
// If the storage slot is in the header, we need to add the header offset.
var key uint256.Int
key.SetBytes(bytes)
if key.Cmp(codeStorageDelta) < 0 {
// This addition is always safe; it can't ever overflow since pos<codeStorageDelta.
key.Add(headerStorageOffset, &key)
// In this branch, the tree-index is zero since we're in the account header,
// and the sub-index is the LSB of the modified storage key.
return zero, byte(key[0] & 0xFF)
}
// We first divide by VerkleNodeWidth to create room to avoid an overflow next.
key.Rsh(&key, uint(verkleNodeWidthLog2))
// We add mainStorageOffset/VerkleNodeWidth which can't overflow.
key.Add(&key, mainStorageOffsetLshVerkleNodeWidth)
// The sub-index is the LSB of the original storage key, since mainStorageOffset
// doesn't affect this byte, so we can avoid masks or shifts.
return &key, byte(key[0] & 0xFF)
}
// StorageSlotKey returns the verkle tree key of the storage slot for the
// specified account.
func StorageSlotKey(address []byte, storageKey []byte) []byte {
treeIndex, subIndex := storageIndex(storageKey)
return GetTreeKey(address, treeIndex, subIndex)
}
// VersionKeyWithEvaluatedAddress returns the verkle tree key of the version
// field for the specified account. The difference between VersionKey is the
// address evaluation is already computed to minimize the computational overhead.
func VersionKeyWithEvaluatedAddress(evaluated *verkle.Point) []byte {
return GetTreeKeyWithEvaluatedAddress(evaluated, zero, VersionLeafKey)
}
// BalanceKeyWithEvaluatedAddress returns the verkle tree key of the balance
// field for the specified account. The difference between BalanceKey is the
// address evaluation is already computed to minimize the computational overhead.
func BalanceKeyWithEvaluatedAddress(evaluated *verkle.Point) []byte {
return GetTreeKeyWithEvaluatedAddress(evaluated, zero, BalanceLeafKey)
}
// NonceKeyWithEvaluatedAddress returns the verkle tree key of the nonce
// field for the specified account. The difference between NonceKey is the
// address evaluation is already computed to minimize the computational overhead.
func NonceKeyWithEvaluatedAddress(evaluated *verkle.Point) []byte {
return GetTreeKeyWithEvaluatedAddress(evaluated, zero, NonceLeafKey)
}
// CodeKeccakKeyWithEvaluatedAddress returns the verkle tree key of the code
// keccak for the specified account. The difference between CodeKeccakKey is the
// address evaluation is already computed to minimize the computational overhead.
func CodeKeccakKeyWithEvaluatedAddress(evaluated *verkle.Point) []byte {
return GetTreeKeyWithEvaluatedAddress(evaluated, zero, CodeKeccakLeafKey)
}
// CodeSizeKeyWithEvaluatedAddress returns the verkle tree key of the code
// size for the specified account. The difference between CodeSizeKey is the
// address evaluation is already computed to minimize the computational overhead.
func CodeSizeKeyWithEvaluatedAddress(evaluated *verkle.Point) []byte {
return GetTreeKeyWithEvaluatedAddress(evaluated, zero, CodeSizeLeafKey)
}
// CodeChunkKeyWithEvaluatedAddress returns the verkle tree key of the code
// chunk for the specified account. The difference between CodeChunkKey is the
// address evaluation is already computed to minimize the computational overhead.
func CodeChunkKeyWithEvaluatedAddress(addressPoint *verkle.Point, chunk *uint256.Int) []byte {
treeIndex, subIndex := codeChunkIndex(chunk)
return GetTreeKeyWithEvaluatedAddress(addressPoint, treeIndex, subIndex)
}
// StorageSlotKeyWithEvaluatedAddress returns the verkle tree key of the storage
// slot for the specified account. The difference between StorageSlotKey is the
// address evaluation is already computed to minimize the computational overhead.
func StorageSlotKeyWithEvaluatedAddress(evaluated *verkle.Point, storageKey []byte) []byte {
treeIndex, subIndex := storageIndex(storageKey)
return GetTreeKeyWithEvaluatedAddress(evaluated, treeIndex, subIndex)
}
func pointToHash(evaluated *verkle.Point, suffix byte) []byte {
// The output of Byte() is big endian for banderwagon. This
// introduces an imbalance in the tree, because hashes are
// elements of a 253-bit field. This means more than half the
// tree would be empty. To avoid this problem, use a little
// endian commitment and chop the MSB.
bytes := evaluated.Bytes()
for i := 0; i < 16; i++ {
bytes[31-i], bytes[i] = bytes[i], bytes[31-i]
}
bytes[31] = suffix
return bytes[:]
}
func evaluateAddressPoint(address []byte) *verkle.Point {
if len(address) < 32 {
var aligned [32]byte
address = append(aligned[:32-len(address)], address...)
}
var poly [3]fr.Element
poly[0].SetZero()
// 32-byte address, interpreted as two little endian
// 16-byte numbers.
verkle.FromLEBytes(&poly[1], address[:16])
verkle.FromLEBytes(&poly[2], address[16:])
cfg := verkle.GetConfig()
ret := cfg.CommitToPoly(poly[:], 0)
// add a constant point
ret.Add(ret, index0Point)
return ret
}

139
trie/utils/verkle_test.go
View file

@ -1,139 +0,0 @@
// Copyright 2023 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 utils
import (
"bytes"
"testing"
"github.com/gballet/go-verkle"
"github.com/holiman/uint256"
)
func TestTreeKey(t *testing.T) {
var (
address = []byte{0x01}
addressEval = evaluateAddressPoint(address)
smallIndex = uint256.NewInt(1)
largeIndex = uint256.NewInt(10000)
smallStorage = []byte{0x1}
largeStorage = bytes.Repeat([]byte{0xff}, 16)
)
if !bytes.Equal(VersionKey(address), VersionKeyWithEvaluatedAddress(addressEval)) {
t.Fatal("Unmatched version key")
}
if !bytes.Equal(BalanceKey(address), BalanceKeyWithEvaluatedAddress(addressEval)) {
t.Fatal("Unmatched balance key")
}
if !bytes.Equal(NonceKey(address), NonceKeyWithEvaluatedAddress(addressEval)) {
t.Fatal("Unmatched nonce key")
}
if !bytes.Equal(CodeKeccakKey(address), CodeKeccakKeyWithEvaluatedAddress(addressEval)) {
t.Fatal("Unmatched code keccak key")
}
if !bytes.Equal(CodeSizeKey(address), CodeSizeKeyWithEvaluatedAddress(addressEval)) {
t.Fatal("Unmatched code size key")
}
if !bytes.Equal(CodeChunkKey(address, smallIndex), CodeChunkKeyWithEvaluatedAddress(addressEval, smallIndex)) {
t.Fatal("Unmatched code chunk key")
}
if !bytes.Equal(CodeChunkKey(address, largeIndex), CodeChunkKeyWithEvaluatedAddress(addressEval, largeIndex)) {
t.Fatal("Unmatched code chunk key")
}
if !bytes.Equal(StorageSlotKey(address, smallStorage), StorageSlotKeyWithEvaluatedAddress(addressEval, smallStorage)) {
t.Fatal("Unmatched storage slot key")
}
if !bytes.Equal(StorageSlotKey(address, largeStorage), StorageSlotKeyWithEvaluatedAddress(addressEval, largeStorage)) {
t.Fatal("Unmatched storage slot key")
}
}
// goos: darwin
// goarch: amd64
// pkg: github.com/ethereum/go-ethereum/trie/utils
// cpu: VirtualApple @ 2.50GHz
// BenchmarkTreeKey
// BenchmarkTreeKey-8 398731 2961 ns/op 32 B/op 1 allocs/op
func BenchmarkTreeKey(b *testing.B) {
// Initialize the IPA settings which can be pretty expensive.
verkle.GetConfig()
b.ReportAllocs()
b.ResetTimer()
for i := 0; i < b.N; i++ {
BalanceKey([]byte{0x01})
}
}
// goos: darwin
// goarch: amd64
// pkg: github.com/ethereum/go-ethereum/trie/utils
// cpu: VirtualApple @ 2.50GHz
// BenchmarkTreeKeyWithEvaluation
// BenchmarkTreeKeyWithEvaluation-8 513855 2324 ns/op 32 B/op 1 allocs/op
func BenchmarkTreeKeyWithEvaluation(b *testing.B) {
// Initialize the IPA settings which can be pretty expensive.
verkle.GetConfig()
addr := []byte{0x01}
eval := evaluateAddressPoint(addr)
b.ReportAllocs()
b.ResetTimer()
for i := 0; i < b.N; i++ {
BalanceKeyWithEvaluatedAddress(eval)
}
}
// goos: darwin
// goarch: amd64
// pkg: github.com/ethereum/go-ethereum/trie/utils
// cpu: VirtualApple @ 2.50GHz
// BenchmarkStorageKey
// BenchmarkStorageKey-8 230516 4584 ns/op 96 B/op 3 allocs/op
func BenchmarkStorageKey(b *testing.B) {
// Initialize the IPA settings which can be pretty expensive.
verkle.GetConfig()
b.ReportAllocs()
b.ResetTimer()
for i := 0; i < b.N; i++ {
StorageSlotKey([]byte{0x01}, bytes.Repeat([]byte{0xff}, 32))
}
}
// goos: darwin
// goarch: amd64
// pkg: github.com/ethereum/go-ethereum/trie/utils
// cpu: VirtualApple @ 2.50GHz
// BenchmarkStorageKeyWithEvaluation
// BenchmarkStorageKeyWithEvaluation-8 320125 3753 ns/op 96 B/op 3 allocs/op
func BenchmarkStorageKeyWithEvaluation(b *testing.B) {
// Initialize the IPA settings which can be pretty expensive.
verkle.GetConfig()
addr := []byte{0x01}
eval := evaluateAddressPoint(addr)
b.ReportAllocs()
b.ResetTimer()
for i := 0; i < b.N; i++ {
StorageSlotKeyWithEvaluatedAddress(eval, bytes.Repeat([]byte{0xff}, 32))
}
}

375
trie/verkle.go
View file

@ -1,375 +0,0 @@
// Copyright 2023 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 trie
import (
"encoding/binary"
"errors"
"fmt"
"math/big"
"github.com/ethereum/go-ethereum/common"
"github.com/ethereum/go-ethereum/core/types"
"github.com/ethereum/go-ethereum/ethdb"
"github.com/ethereum/go-ethereum/trie/trienode"
"github.com/ethereum/go-ethereum/trie/utils"
"github.com/gballet/go-verkle"
"github.com/holiman/uint256"
)
var (
zero [32]byte
errInvalidRootType = errors.New("invalid node type for root")
)
// VerkleTrie is a wrapper around VerkleNode that implements the trie.Trie
// interface so that Verkle trees can be reused verbatim.
type VerkleTrie struct {
root verkle.VerkleNode
db *Database
cache *utils.PointCache
reader *trieReader
}
// NewVerkleTrie constructs a verkle tree based on the specified root hash.
func NewVerkleTrie(root common.Hash, db *Database, cache *utils.PointCache) (*VerkleTrie, error) {
reader, err := newTrieReader(root, common.Hash{}, db)
if err != nil {
return nil, err
}
// Parse the root verkle node if it's not empty.
node := verkle.New()
if root != types.EmptyVerkleHash && root != types.EmptyRootHash {
blob, err := reader.node(nil, common.Hash{})
if err != nil {
return nil, err
}
node, err = verkle.ParseNode(blob, 0)
if err != nil {
return nil, err
}
}
return &VerkleTrie{
root: node,
db: db,
cache: cache,
reader: reader,
}, nil
}
// GetKey returns the sha3 preimage of a hashed key that was previously used
// to store a value.
func (t *VerkleTrie) GetKey(key []byte) []byte {
return key
}
// GetAccount implements state.Trie, retrieving the account with the specified
// account address. If the specified account is not in the verkle tree, nil will
// be returned. If the tree is corrupted, an error will be returned.
func (t *VerkleTrie) GetAccount(addr common.Address) (*types.StateAccount, error) {
var (
acc = &types.StateAccount{}
values [][]byte
err error
)
switch n := t.root.(type) {
case *verkle.InternalNode:
values, err = n.GetValuesAtStem(t.cache.GetStem(addr[:]), t.nodeResolver)
if err != nil {
return nil, fmt.Errorf("GetAccount (%x) error: %v", addr, err)
}
default:
return nil, errInvalidRootType
}
if values == nil {
return nil, nil
}
// Decode nonce in little-endian
if len(values[utils.NonceLeafKey]) > 0 {
acc.Nonce = binary.LittleEndian.Uint64(values[utils.NonceLeafKey])
}
// Decode balance in little-endian
var balance [32]byte
copy(balance[:], values[utils.BalanceLeafKey])
for i := 0; i < len(balance)/2; i++ {
balance[len(balance)-i-1], balance[i] = balance[i], balance[len(balance)-i-1]
}
acc.Balance = new(big.Int).SetBytes(balance[:])
// Decode codehash
acc.CodeHash = values[utils.CodeKeccakLeafKey]
// TODO account.Root is leave as empty. How should we handle the legacy account?
return acc, nil
}
// GetStorage implements state.Trie, retrieving the storage slot with the specified
// account address and storage key. If the specified slot is not in the verkle tree,
// nil will be returned. If the tree is corrupted, an error will be returned.
func (t *VerkleTrie) GetStorage(addr common.Address, key []byte) ([]byte, error) {
k := utils.StorageSlotKeyWithEvaluatedAddress(t.cache.Get(addr.Bytes()), key)
val, err := t.root.Get(k, t.nodeResolver)
if err != nil {
return nil, err
}
return common.TrimLeftZeroes(val), nil
}
// UpdateAccount implements state.Trie, writing the provided account into the tree.
// If the tree is corrupted, an error will be returned.
func (t *VerkleTrie) UpdateAccount(addr common.Address, acc *types.StateAccount) error {
var (
err error
nonce, balance [32]byte
values = make([][]byte, verkle.NodeWidth)
)
values[utils.VersionLeafKey] = zero[:]
values[utils.CodeKeccakLeafKey] = acc.CodeHash[:]
// Encode nonce in little-endian
binary.LittleEndian.PutUint64(nonce[:], acc.Nonce)
values[utils.NonceLeafKey] = nonce[:]
// Encode balance in little-endian
bytes := acc.Balance.Bytes()
if len(bytes) > 0 {
for i, b := range bytes {
balance[len(bytes)-i-1] = b
}
}
values[utils.BalanceLeafKey] = balance[:]
switch n := t.root.(type) {
case *verkle.InternalNode:
err = n.InsertValuesAtStem(t.cache.GetStem(addr[:]), values, t.nodeResolver)
if err != nil {
return fmt.Errorf("UpdateAccount (%x) error: %v", addr, err)
}
default:
return errInvalidRootType
}
// TODO figure out if the code size needs to be updated, too
return nil
}
// UpdateStorage implements state.Trie, writing the provided storage slot into
// the tree. If the tree is corrupted, an error will be returned.
func (t *VerkleTrie) UpdateStorage(address common.Address, key, value []byte) error {
// Left padding the slot value to 32 bytes.
var v [32]byte
if len(value) >= 32 {
copy(v[:], value[:32])
} else {
copy(v[32-len(value):], value[:])
}
k := utils.StorageSlotKeyWithEvaluatedAddress(t.cache.Get(address.Bytes()), key)
return t.root.Insert(k, v[:], t.nodeResolver)
}
// DeleteAccount implements state.Trie, deleting the specified account from the
// trie. If the account was not existent in the trie, no error will be returned.
// If the trie is corrupted, an error will be returned.
func (t *VerkleTrie) DeleteAccount(addr common.Address) error {
var (
err error
values = make([][]byte, verkle.NodeWidth)
)
for i := 0; i < verkle.NodeWidth; i++ {
values[i] = zero[:]
}
switch n := t.root.(type) {
case *verkle.InternalNode:
err = n.InsertValuesAtStem(t.cache.GetStem(addr.Bytes()), values, t.nodeResolver)
if err != nil {
return fmt.Errorf("DeleteAccount (%x) error: %v", addr, err)
}
default:
return errInvalidRootType
}
return nil
}
// DeleteStorage implements state.Trie, deleting the specified storage slot from
// the trie. If the storage slot was not existent in the trie, no error will be
// returned. If the trie is corrupted, an error will be returned.
func (t *VerkleTrie) DeleteStorage(addr common.Address, key []byte) error {
var zero [32]byte
k := utils.StorageSlotKeyWithEvaluatedAddress(t.cache.Get(addr.Bytes()), key)
return t.root.Insert(k, zero[:], t.nodeResolver)
}
// Hash returns the root hash of the tree. It does not write to the database and
// can be used even if the tree doesn't have one.
func (t *VerkleTrie) Hash() common.Hash {
return t.root.Commit().Bytes()
}
// Commit writes all nodes to the tree's memory database.
func (t *VerkleTrie) Commit(_ bool) (common.Hash, *trienode.NodeSet, error) {
root, ok := t.root.(*verkle.InternalNode)
if !ok {
return common.Hash{}, nil, errors.New("unexpected root node type")
}
nodes, err := root.BatchSerialize()
if err != nil {
return common.Hash{}, nil, fmt.Errorf("serializing tree nodes: %s", err)
}
nodeset := trienode.NewNodeSet(common.Hash{})
for _, node := range nodes {
// hash parameter is not used in pathdb
nodeset.AddNode(node.Path, trienode.New(common.Hash{}, node.SerializedBytes))
}
// Serialize root commitment form
return t.Hash(), nodeset, nil
}
// NodeIterator implements state.Trie, returning an iterator that returns
// nodes of the trie. Iteration starts at the key after the given start key.
//
// TODO(gballet, rjl493456442) implement it.
func (t *VerkleTrie) NodeIterator(startKey []byte) (NodeIterator, error) {
panic("not implemented")
}
// Prove implements state.Trie, constructing a Merkle proof for key. The result
// contains all encoded nodes on the path to the value at key. The value itself
// is also included in the last node and can be retrieved by verifying the proof.
//
// If the trie does not contain a value for key, the returned proof contains all
// nodes of the longest existing prefix of the key (at least the root), ending
// with the node that proves the absence of the key.
//
// TODO(gballet, rjl493456442) implement it.
func (t *VerkleTrie) Prove(key []byte, proofDb ethdb.KeyValueWriter) error {
panic("not implemented")
}
// Copy returns a deep-copied verkle tree.
func (t *VerkleTrie) Copy() *VerkleTrie {
return &VerkleTrie{
root: t.root.Copy(),
db: t.db,
cache: t.cache,
reader: t.reader,
}
}
// IsVerkle indicates if the trie is a Verkle trie.
func (t *VerkleTrie) IsVerkle() bool {
return true
}
// ChunkedCode represents a sequence of 32-bytes chunks of code (31 bytes of which
// are actual code, and 1 byte is the pushdata offset).
type ChunkedCode []byte
// Copy the values here so as to avoid an import cycle
const (
PUSH1 = byte(0x60)
PUSH32 = byte(0x7f)
)
// ChunkifyCode generates the chunked version of an array representing EVM bytecode
func ChunkifyCode(code []byte) ChunkedCode {
var (
chunkOffset = 0 // offset in the chunk
chunkCount = len(code) / 31
codeOffset = 0 // offset in the code
)
if len(code)%31 != 0 {
chunkCount++
}
chunks := make([]byte, chunkCount*32)
for i := 0; i < chunkCount; i++ {
// number of bytes to copy, 31 unless the end of the code has been reached.
end := 31 * (i + 1)
if len(code) < end {
end = len(code)
}
copy(chunks[i*32+1:], code[31*i:end]) // copy the code itself
// chunk offset = taken from the last chunk.
if chunkOffset > 31 {
// skip offset calculation if push data covers the whole chunk
chunks[i*32] = 31
chunkOffset = 1
continue
}
chunks[32*i] = byte(chunkOffset)
chunkOffset = 0
// Check each instruction and update the offset it should be 0 unless
// a PUSH-N overflows.
for ; codeOffset < end; codeOffset++ {
if code[codeOffset] >= PUSH1 && code[codeOffset] <= PUSH32 {
codeOffset += int(code[codeOffset] - PUSH1 + 1)
if codeOffset+1 >= 31*(i+1) {
codeOffset++
chunkOffset = codeOffset - 31*(i+1)
break
}
}
}
}
return chunks
}
// UpdateContractCode implements state.Trie, writing the provided contract code
// into the trie.
func (t *VerkleTrie) UpdateContractCode(addr common.Address, codeHash common.Hash, code []byte) error {
var (
chunks = ChunkifyCode(code)
values [][]byte
key []byte
err error
)
for i, chunknr := 0, uint64(0); i < len(chunks); i, chunknr = i+32, chunknr+1 {
groupOffset := (chunknr + 128) % 256
if groupOffset == 0 /* start of new group */ || chunknr == 0 /* first chunk in header group */ {
values = make([][]byte, verkle.NodeWidth)
key = utils.CodeChunkKeyWithEvaluatedAddress(t.cache.Get(addr.Bytes()), uint256.NewInt(chunknr))
}
values[groupOffset] = chunks[i : i+32]
// Reuse the calculated key to also update the code size.
if i == 0 {
cs := make([]byte, 32)
binary.LittleEndian.PutUint64(cs, uint64(len(code)))
values[utils.CodeSizeLeafKey] = cs
}
if groupOffset == 255 || len(chunks)-i <= 32 {
switch root := t.root.(type) {
case *verkle.InternalNode:
err = root.InsertValuesAtStem(key[:31], values, t.nodeResolver)
if err != nil {
return fmt.Errorf("UpdateContractCode (addr=%x) error: %w", addr[:], err)
}
default:
return errInvalidRootType
}
}
}
return nil
}
func (t *VerkleTrie) ToDot() string {
return verkle.ToDot(t.root)
}
func (t *VerkleTrie) nodeResolver(path []byte) ([]byte, error) {
return t.reader.node(path, common.Hash{})
}

97
trie/verkle_test.go
View file

@ -1,97 +0,0 @@
// Copyright 2023 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 trie
import (
"bytes"
"math/big"
"reflect"
"testing"
"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/trie/triedb/pathdb"
"github.com/ethereum/go-ethereum/trie/utils"
)
var (
accounts = map[common.Address]*types.StateAccount{
{1}: {
Nonce: 100,
Balance: big.NewInt(100),
CodeHash: common.Hash{0x1}.Bytes(),
},
{2}: {
Nonce: 200,
Balance: big.NewInt(200),
CodeHash: common.Hash{0x2}.Bytes(),
},
}
storages = map[common.Address]map[common.Hash][]byte{
{1}: {
common.Hash{10}: []byte{10},
common.Hash{11}: []byte{11},
common.MaxHash: []byte{0xff},
},
{2}: {
common.Hash{20}: []byte{20},
common.Hash{21}: []byte{21},
common.MaxHash: []byte{0xff},
},
}
)
func TestVerkleTreeReadWrite(t *testing.T) {
db := NewDatabase(rawdb.NewMemoryDatabase(), &Config{
IsVerkle: true,
PathDB: pathdb.Defaults,
})
defer db.Close()
tr, _ := NewVerkleTrie(types.EmptyVerkleHash, db, utils.NewPointCache(100))
for addr, acct := range accounts {
if err := tr.UpdateAccount(addr, acct); err != nil {
t.Fatalf("Failed to update account, %v", err)
}
for key, val := range storages[addr] {
if err := tr.UpdateStorage(addr, key.Bytes(), val); err != nil {
t.Fatalf("Failed to update account, %v", err)
}
}
}
for addr, acct := range accounts {
stored, err := tr.GetAccount(addr)
if err != nil {
t.Fatalf("Failed to get account, %v", err)
}
if !reflect.DeepEqual(stored, acct) {
t.Fatal("account is not matched")
}
for key, val := range storages[addr] {
stored, err := tr.GetStorage(addr, key.Bytes())
if err != nil {
t.Fatalf("Failed to get storage, %v", err)
}
if !bytes.Equal(stored, val) {
t.Fatal("storage is not matched")
}
}
}
}