diff --git a/beacon/engine/ed_codec.go b/beacon/engine/ed_codec.go index 02a1fd3805..83603616a7 100644 --- a/beacon/engine/ed_codec.go +++ b/beacon/engine/ed_codec.go @@ -10,7 +10,6 @@ import ( "github.com/ethereum/go-ethereum/common" "github.com/ethereum/go-ethereum/common/hexutil" "github.com/ethereum/go-ethereum/core/types" - "github.com/ethereum/go-ethereum/core/types/bal" ) var _ = (*executableDataMarshaling)(nil) @@ -36,7 +35,7 @@ func (e ExecutableData) MarshalJSON() ([]byte, error) { BlobGasUsed *hexutil.Uint64 `json:"blobGasUsed"` ExcessBlobGas *hexutil.Uint64 `json:"excessBlobGas"` SlotNumber *hexutil.Uint64 `json:"slotNumber,omitempty"` - BlockAccessList *bal.BlockAccessList `json:"blockAccessList,omitempty"` + BlockAccessList hexutil.Bytes `json:"blockAccessList,omitempty"` } var enc ExecutableData enc.ParentHash = e.ParentHash @@ -87,7 +86,7 @@ func (e *ExecutableData) UnmarshalJSON(input []byte) error { BlobGasUsed *hexutil.Uint64 `json:"blobGasUsed"` ExcessBlobGas *hexutil.Uint64 `json:"excessBlobGas"` SlotNumber *hexutil.Uint64 `json:"slotNumber,omitempty"` - BlockAccessList *bal.BlockAccessList `json:"blockAccessList,omitempty"` + BlockAccessList *hexutil.Bytes `json:"blockAccessList,omitempty"` } var dec ExecutableData if err := json.Unmarshal(input, &dec); err != nil { @@ -165,7 +164,7 @@ func (e *ExecutableData) UnmarshalJSON(input []byte) error { e.SlotNumber = (*uint64)(dec.SlotNumber) } if dec.BlockAccessList != nil { - e.BlockAccessList = dec.BlockAccessList + e.BlockAccessList = *dec.BlockAccessList } return nil } diff --git a/beacon/engine/types.go b/beacon/engine/types.go index 980e8f9771..c06c4b44ab 100644 --- a/beacon/engine/types.go +++ b/beacon/engine/types.go @@ -17,15 +17,18 @@ package engine import ( + "bytes" "fmt" + "github.com/ethereum/go-ethereum/core/types/bal" "math/big" "slices" "github.com/ethereum/go-ethereum/common" "github.com/ethereum/go-ethereum/common/hexutil" "github.com/ethereum/go-ethereum/core/types" - "github.com/ethereum/go-ethereum/core/types/bal" + "github.com/ethereum/go-ethereum/crypto" "github.com/ethereum/go-ethereum/params" + "github.com/ethereum/go-ethereum/rlp" "github.com/ethereum/go-ethereum/trie" ) @@ -83,25 +86,25 @@ type payloadAttributesMarshaling struct { // ExecutableData is the data necessary to execute an EL payload. type ExecutableData struct { - ParentHash common.Hash `json:"parentHash" gencodec:"required"` - FeeRecipient common.Address `json:"feeRecipient" gencodec:"required"` - StateRoot common.Hash `json:"stateRoot" gencodec:"required"` - ReceiptsRoot common.Hash `json:"receiptsRoot" gencodec:"required"` - LogsBloom []byte `json:"logsBloom" gencodec:"required"` - Random common.Hash `json:"prevRandao" gencodec:"required"` - Number uint64 `json:"blockNumber" gencodec:"required"` - GasLimit uint64 `json:"gasLimit" gencodec:"required"` - GasUsed uint64 `json:"gasUsed" gencodec:"required"` - Timestamp uint64 `json:"timestamp" gencodec:"required"` - ExtraData []byte `json:"extraData" gencodec:"required"` - BaseFeePerGas *big.Int `json:"baseFeePerGas" gencodec:"required"` - BlockHash common.Hash `json:"blockHash" gencodec:"required"` - Transactions [][]byte `json:"transactions" gencodec:"required"` - Withdrawals []*types.Withdrawal `json:"withdrawals"` - BlobGasUsed *uint64 `json:"blobGasUsed"` - ExcessBlobGas *uint64 `json:"excessBlobGas"` - SlotNumber *uint64 `json:"slotNumber,omitempty"` - BlockAccessList *bal.BlockAccessList `json:"blockAccessList,omitempty"` + ParentHash common.Hash `json:"parentHash" gencodec:"required"` + FeeRecipient common.Address `json:"feeRecipient" gencodec:"required"` + StateRoot common.Hash `json:"stateRoot" gencodec:"required"` + ReceiptsRoot common.Hash `json:"receiptsRoot" gencodec:"required"` + LogsBloom []byte `json:"logsBloom" gencodec:"required"` + Random common.Hash `json:"prevRandao" gencodec:"required"` + Number uint64 `json:"blockNumber" gencodec:"required"` + GasLimit uint64 `json:"gasLimit" gencodec:"required"` + GasUsed uint64 `json:"gasUsed" gencodec:"required"` + Timestamp uint64 `json:"timestamp" gencodec:"required"` + ExtraData []byte `json:"extraData" gencodec:"required"` + BaseFeePerGas *big.Int `json:"baseFeePerGas" gencodec:"required"` + BlockHash common.Hash `json:"blockHash" gencodec:"required"` + Transactions [][]byte `json:"transactions" gencodec:"required"` + Withdrawals []*types.Withdrawal `json:"withdrawals"` + BlobGasUsed *uint64 `json:"blobGasUsed"` + ExcessBlobGas *uint64 `json:"excessBlobGas"` + SlotNumber *uint64 `json:"slotNumber,omitempty"` + BlockAccessList hexutil.Bytes `json:"blockAccessList,omitempty"` } // JSON type overrides for executableData. @@ -315,12 +318,9 @@ func ExecutableDataToBlockNoHash(data ExecutableData, versionedHashes []common.H // If Amsterdam is enabled, data.BlockAccessList is always non-nil, // even for empty blocks with no state transitions. - // - // If Amsterdam is not enabled yet, blockAccessListHash is expected - // to be nil. var blockAccessListHash *common.Hash if data.BlockAccessList != nil { - hash := data.BlockAccessList.Hash() + hash := crypto.Keccak256Hash(data.BlockAccessList) blockAccessListHash = &hash } header := &types.Header{ @@ -347,32 +347,48 @@ func ExecutableDataToBlockNoHash(data ExecutableData, versionedHashes []common.H SlotNumber: data.SlotNumber, BlockAccessListHash: blockAccessListHash, } - return types.NewBlockWithHeader(header).WithBody(types.Body{Transactions: txs, Uncles: nil, Withdrawals: data.Withdrawals}), nil + body := types.Body{Transactions: txs, Uncles: nil, Withdrawals: data.Withdrawals} + if data.BlockAccessList != nil { + balHash := crypto.Keccak256Hash(data.BlockAccessList) + header.BlockAccessListHash = &balHash + var accessList bal.BlockAccessList + if err := rlp.DecodeBytes(data.BlockAccessList, &accessList); err != nil { + return nil, fmt.Errorf("failed to decode BAL: %w\n", err) + } + block := types.NewBlockWithHeader(header).WithBody(body).WithAccessList(&accessList) + return block, nil + } + return types.NewBlockWithHeader(header).WithBody(body), nil } // BlockToExecutableData constructs the ExecutableData structure by filling the // fields from the given block. It assumes the given block is post-merge block. func BlockToExecutableData(block *types.Block, fees *big.Int, sidecars []*types.BlobTxSidecar, requests [][]byte) *ExecutionPayloadEnvelope { data := &ExecutableData{ - BlockHash: block.Hash(), - ParentHash: block.ParentHash(), - FeeRecipient: block.Coinbase(), - StateRoot: block.Root(), - Number: block.NumberU64(), - GasLimit: block.GasLimit(), - GasUsed: block.GasUsed(), - BaseFeePerGas: block.BaseFee(), - Timestamp: block.Time(), - ReceiptsRoot: block.ReceiptHash(), - LogsBloom: block.Bloom().Bytes(), - Transactions: encodeTransactions(block.Transactions()), - Random: block.MixDigest(), - ExtraData: block.Extra(), - Withdrawals: block.Withdrawals(), - BlobGasUsed: block.BlobGasUsed(), - ExcessBlobGas: block.ExcessBlobGas(), - SlotNumber: block.SlotNumber(), - BlockAccessList: block.AccessList(), + BlockHash: block.Hash(), + ParentHash: block.ParentHash(), + FeeRecipient: block.Coinbase(), + StateRoot: block.Root(), + Number: block.NumberU64(), + GasLimit: block.GasLimit(), + GasUsed: block.GasUsed(), + BaseFeePerGas: block.BaseFee(), + Timestamp: block.Time(), + ReceiptsRoot: block.ReceiptHash(), + LogsBloom: block.Bloom().Bytes(), + Transactions: encodeTransactions(block.Transactions()), + Random: block.MixDigest(), + ExtraData: block.Extra(), + Withdrawals: block.Withdrawals(), + BlobGasUsed: block.BlobGasUsed(), + ExcessBlobGas: block.ExcessBlobGas(), + SlotNumber: block.SlotNumber(), + } + if al := block.AccessList(); al != nil { + var buf bytes.Buffer + if err := rlp.Encode(&buf, al); err == nil { + data.BlockAccessList = buf.Bytes() + } } // Add blobs. diff --git a/cmd/evm/blockrunner.go b/cmd/evm/blockrunner.go index b175f05dc7..d5d957cc40 100644 --- a/cmd/evm/blockrunner.go +++ b/cmd/evm/blockrunner.go @@ -117,7 +117,7 @@ func runBlockTest(ctx *cli.Context, fname string) ([]testResult, error) { test := tests[name] result := &testResult{Name: name, Pass: true} var finalRoot *common.Hash - if err := test.Run(false, rawdb.PathScheme, ctx.Bool(WitnessCrossCheckFlag.Name), tracer, func(res error, chain *core.BlockChain) { + if err := test.Run(false, rawdb.PathScheme, ctx.Bool(WitnessCrossCheckFlag.Name), true, tracer, func(res error, chain *core.BlockChain) { if ctx.Bool(DumpFlag.Name) { if s, _ := chain.State(); s != nil { result.State = dump(s) diff --git a/core/bal_test.go b/core/bal_test.go index 64c2714b16..ec9e6ccff9 100644 --- a/core/bal_test.go +++ b/core/bal_test.go @@ -26,6 +26,7 @@ import ( "github.com/ethereum/go-ethereum/common" "github.com/ethereum/go-ethereum/consensus/beacon" "github.com/ethereum/go-ethereum/consensus/ethash" + "github.com/ethereum/go-ethereum/core/rawdb" "github.com/ethereum/go-ethereum/core/types" "github.com/ethereum/go-ethereum/core/types/bal" "github.com/ethereum/go-ethereum/crypto" @@ -1453,3 +1454,48 @@ func TestBALAuthCodeOverwrittenFinalRecorded(t *testing.T) { t.Fatalf("expected final nonce 2, got %+v", aa.NonceChanges) } } + +// ============================== Preimages ============================== + +// TestBALPreimages tests that preimage tracking works when executing a block +// with an access list. +func TestBALPreimages(t *testing.T) { + // Runtime code: store 0x42 at memory[0], then KECCAK256 over memory[0:1]. + // PUSH1 0x42 ; PUSH1 0x00 ; MSTORE8 ; PUSH1 0x01 ; PUSH1 0x00 ; KECCAK256 ; POP ; STOP + contract := common.HexToAddress("0xca11ee") + runtime := []byte{0x60, 0x42, 0x60, 0x00, 0x53, 0x60, 0x01, 0x60, 0x00, 0x20, 0x50, 0x00} + + env := newBALTestEnv(types.GenesisAlloc{ + contract: {Code: runtime, Balance: common.Big0}, + }) + + engine := beacon.New(ethash.NewFaker()) + _, blocks, _ := GenerateChainWithGenesis(env.gspec, engine, 1, func(_ int, g *BlockGen) { + // Run the EIP-4788 beacon-root system call during generation so the + // generated BAL matches what the processor recomputes on import. + g.SetParentBeaconRoot(common.Hash{}) + g.AddTx(env.tx(0, &contract, big.NewInt(0), 200_000, 0, nil)) + }) + // Import the block through the parallel BAL processor with preimage + // recording enabled, so the KECCAK256 above is captured. + db := rawdb.NewMemoryDatabase() + cfg := DefaultConfig() + cfg.VmConfig.EnablePreimageRecording = true + chain, err := NewBlockChain(db, env.gspec, engine, cfg) + if err != nil { + t.Fatalf("new blockchain: %v", err) + } + defer chain.Stop() + + if _, err := chain.InsertChain(blocks); err != nil { + t.Fatalf("insert chain: %v", err) + } + + // The transaction hashed the single byte 0x42; the preimage must have been + // accumulated by the state transition and written to disk on commit. + want := []byte{0x42} + hash := crypto.Keccak256Hash(want) + if got := rawdb.ReadPreimage(db, hash); !bytes.Equal(got, want) { + t.Fatalf("preimage for %x: got %x, want %x", hash, got, want) + } +} diff --git a/core/block_validator.go b/core/block_validator.go index 962fffb82a..0b45cb020f 100644 --- a/core/block_validator.go +++ b/core/block_validator.go @@ -20,8 +20,9 @@ import ( "errors" "fmt" + "github.com/ethereum/go-ethereum/common" + "github.com/ethereum/go-ethereum/consensus" - "github.com/ethereum/go-ethereum/core/state" "github.com/ethereum/go-ethereum/core/types" "github.com/ethereum/go-ethereum/params" "github.com/ethereum/go-ethereum/trie" @@ -143,9 +144,15 @@ func (v *BlockValidator) ValidateBody(block *types.Block) error { return nil } +// StateRootSource computes a new state root during block execution. +type StateRootSource interface { + IntermediateRoot(deleteEmptyObjects bool) common.Hash + Error() error +} + // ValidateState validates the various changes that happen after a state transition, // such as amount of used gas, the receipt roots and the state root itself. -func (v *BlockValidator) ValidateState(block *types.Block, statedb *state.StateDB, res *ProcessResult, stateless bool) error { +func (v *BlockValidator) ValidateState(block *types.Block, state StateRootSource, res *ProcessResult, stateless bool) error { if res == nil { return errors.New("nil ProcessResult value") } @@ -201,8 +208,8 @@ func (v *BlockValidator) ValidateState(block *types.Block, statedb *state.StateD } // Validate the state root against the received state root and throw // an error if they don't match. - if root := statedb.IntermediateRoot(v.config.IsEIP158(header.Number)); header.Root != root { - return fmt.Errorf("invalid merkle root (remote: %x local: %x) dberr: %w", header.Root, root, statedb.Error()) + if root := state.IntermediateRoot(v.config.IsEIP158(header.Number)); header.Root != root { + return fmt.Errorf("invalid merkle root (remote: %x local: %x) dberr: %w", header.Root, root, state.Error()) } return nil } diff --git a/core/blockchain.go b/core/blockchain.go index 798d3f1ae6..fd50098277 100644 --- a/core/blockchain.go +++ b/core/blockchain.go @@ -31,6 +31,8 @@ import ( "sync/atomic" "time" + "github.com/ethereum/go-ethereum/core/types/bal" + "github.com/ethereum/go-ethereum/common" "github.com/ethereum/go-ethereum/common/lru" "github.com/ethereum/go-ethereum/common/mclock" @@ -365,12 +367,13 @@ type BlockChain struct { stopping atomic.Bool // false if chain is running, true when stopped procInterrupt atomic.Bool // interrupt signaler for block processing - engine consensus.Engine - validator Validator // Block and state validator interface - prefetcher Prefetcher - processor Processor // Block transaction processor interface - logger *tracing.Hooks - stateSizer *state.SizeTracker // State size tracking + engine consensus.Engine + validator Validator // Block and state validator interface + prefetcher Prefetcher + processor Processor // Block transaction processor interface + parallelProcessor ParallelStateProcessor // block processor for use with access lists + logger *tracing.Hooks + stateSizer *state.SizeTracker // State size tracking lastForkReadyAlert time.Time // Last time there was a fork readiness print out slowBlockThreshold time.Duration // Block execution time threshold beyond which detailed statistics will be logged @@ -433,6 +436,7 @@ func NewBlockChain(db ethdb.Database, genesis *Genesis, engine consensus.Engine, bc.validator = NewBlockValidator(chainConfig, bc) bc.prefetcher = newStatePrefetcher(chainConfig, bc.hc) bc.processor = NewStateProcessor(bc.hc) + bc.parallelProcessor = *NewParallelStateProcessor(bc.hc, bc.GetVMConfig()) genesisHeader := bc.GetHeaderByNumber(0) if genesisHeader == nil { @@ -1656,7 +1660,7 @@ func (bc *BlockChain) writeKnownBlock(block *types.Block) error { // writeBlockWithState writes block, metadata and corresponding state data to the // database. -func (bc *BlockChain) writeBlockWithState(block *types.Block, receipts []*types.Receipt, statedb *state.StateDB) error { +func (bc *BlockChain) writeBlockWithState(block *types.Block, receipts []*types.Receipt, statedb state.Committer) error { if !bc.HasHeader(block.ParentHash(), block.NumberU64()-1) { return consensus.ErrUnknownAncestor } @@ -1770,7 +1774,7 @@ func (bc *BlockChain) writeBlockWithState(block *types.Block, receipts []*types. // writeBlockAndSetHead is the internal implementation of WriteBlockAndSetHead. // This function expects the chain mutex to be held. -func (bc *BlockChain) writeBlockAndSetHead(block *types.Block, receipts []*types.Receipt, logs []*types.Log, state *state.StateDB, emitHeadEvent bool) (status WriteStatus, err error) { +func (bc *BlockChain) writeBlockAndSetHead(block *types.Block, receipts []*types.Receipt, logs []*types.Log, state state.Committer, emitHeadEvent bool) (status WriteStatus, err error) { if err := bc.writeBlockWithState(block, receipts, state); err != nil { return NonStatTy, err } @@ -2125,16 +2129,129 @@ type ExecuteConfig struct { EnableWitnessStats bool } +func (bc *BlockChain) processBlockWithAccessList(parentRoot common.Hash, block *types.Block, setHead bool) (procRes *blockProcessingResult, blockEndErr error) { + var ( + startTime = time.Now() + procTime time.Duration + statedb *state.StateDB + ) + + sdb := state.NewMPTDatabase(bc.triedb, bc.codedb).WithSnapshot(bc.snaps) + + al := block.AccessList() + reader := bal.NewAccessListReader(*al) + prefetchReader, err := sdb.ReaderWithPrefetch(parentRoot, reader.StorageKeys()) + if err != nil { + return nil, err + } + + stateTransition, err := state.NewBALStateTransition(block, prefetchReader, sdb, parentRoot, reader) + if err != nil { + return nil, err + } + statedb, err = state.NewWithReader(parentRoot, sdb, prefetchReader) + if err != nil { + return nil, err + } + + if bc.logger != nil && bc.logger.OnBlockStart != nil { + bc.logger.OnBlockStart(tracing.BlockEvent{ + Block: block, + Finalized: bc.CurrentFinalBlock(), + Safe: bc.CurrentSafeBlock(), + }) + } + if bc.logger != nil && bc.logger.OnBlockEnd != nil { + defer func() { + bc.logger.OnBlockEnd(blockEndErr) + }() + } + + res, err := bc.parallelProcessor.Process(block, stateTransition, statedb, bc.cfg.VmConfig) + if err != nil { + return nil, err + } + + if err := bc.validator.ValidateState(block, stateTransition, res.ProcessResult, false); err != nil { + return nil, err + } + + procTime = time.Since(startTime) + writeStart := time.Now() + // Write the block to the chain and get the status. + var ( + //wstart = time.Now() + status WriteStatus + ) + if !setHead { + // Don't set the head, only insert the block + err = bc.writeBlockWithState(block, res.ProcessResult.Receipts, stateTransition) + } else { + status, err = bc.writeBlockAndSetHead(block, res.ProcessResult.Receipts, res.ProcessResult.Logs, stateTransition, false) + } + if err != nil { + return nil, err + } + writeTime := time.Since(writeStart) + var stats ExecuteStats + + wc := stateTransition.WrittenCounts() + d := stateTransition.Deletions() + //codeLoaded, codeLoadBytes := prefetchReader.(state.CodeLoadTracker).CodeLoads() + //stats.AccountLoaded = al.UniqueAccountCount() + stats.AccountUpdated = wc.Accounts - d.Accounts + stats.AccountDeleted = d.Accounts + //stats.StorageLoaded = al.UniqueStorageSlotCount() + stats.StorageUpdated = wc.StorageSlots - d.Storage + stats.StorageDeleted = d.Storage + //stats.CodeLoaded = codeLoaded + //stats.CodeLoadBytes = codeLoadBytes + stats.CodeUpdated = wc.Codes + stats.CodeUpdateBytes = wc.CodeBytes + + //stats.ExecWall = res.ExecTime + //stats.PostProcess = res.PostProcessTime + + if m := res.StateTransitionMetrics; m != nil { + stats.AccountHashes = m.AccountUpdate + m.StateUpdate + m.StateHash + //stats.Prefetch = m.StatePrefetch + } + //stats.Prefetch = prefetchReader.(state.PrefetcherMetricer).Metrics().Elapsed + + stats.StateReadCacheStats = prefetchReader.(state.ReaderStater).GetStats() + + elapsed := time.Since(startTime) + 1 // prevent zero division + stats.TotalTime = elapsed + stats.MgasPerSecond = float64(res.ProcessResult.GasUsed) * 1000 / float64(elapsed) + stats.BlockWrite = writeTime + + // TODO: reinstate + //stats.balTransitionStats = res.StateTransitionMetrics + + return &blockProcessingResult{ + usedGas: res.ProcessResult.GasUsed, + procTime: procTime, + status: status, + witness: nil, + stats: &stats, + }, nil +} + // ProcessBlock executes and validates the given block. If there was no error // it writes the block and associated state to database. func (bc *BlockChain) ProcessBlock(ctx context.Context, parentRoot common.Hash, block *types.Block, config ExecuteConfig) (result *blockProcessingResult, blockEndErr error) { var ( - err error - startTime = time.Now() - statedb *state.StateDB - interrupt atomic.Bool - sdb state.Database + err error + startTime = time.Now() + statedb *state.StateDB + interrupt atomic.Bool + sdb state.Database + blockHasAccessList = block.AccessList() != nil ) + + if blockHasAccessList { + return bc.processBlockWithAccessList(parentRoot, block, config.WriteHead) + } defer interrupt.Store(true) // terminate the prefetch at the end if bc.chainConfig.IsUBT(block.Number(), block.Time()) { diff --git a/core/parallel_state_processor.go b/core/parallel_state_processor.go new file mode 100644 index 0000000000..0f021bdff0 --- /dev/null +++ b/core/parallel_state_processor.go @@ -0,0 +1,347 @@ +package core + +import ( + "cmp" + "context" + "fmt" + "runtime" + "slices" + "time" + + "github.com/ethereum/go-ethereum/common" + "github.com/ethereum/go-ethereum/core/state" + "github.com/ethereum/go-ethereum/core/types" + "github.com/ethereum/go-ethereum/core/types/bal" + "github.com/ethereum/go-ethereum/core/vm" + "golang.org/x/sync/errgroup" +) + +// ProcessResultWithMetrics wraps ProcessResult with timing breakdown for BAL block processing. +type ProcessResultWithMetrics struct { + ProcessResult *ProcessResult + PreProcessTime time.Duration + StateTransitionMetrics *state.BALStateTransitionMetrics + ExecTime time.Duration + PostProcessTime time.Duration + + // Preimages holds the SHA3 preimages recorded during the execution of + // the block. + Preimages map[common.Hash][]byte +} + +// errResult wraps an error into a new ProcessResultWithMetrics instance +func errResult(err error) *ProcessResultWithMetrics { + return &ProcessResultWithMetrics{ProcessResult: &ProcessResult{Error: err}} +} + +// ParallelStateProcessor is used to execute and verify blocks containing +// access lists. +type ParallelStateProcessor struct { + *StateProcessor + chain *HeaderChain + vmCfg *vm.Config +} + +// NewParallelStateProcessor returns a new ParallelStateProcessor instance. +func NewParallelStateProcessor(chain *HeaderChain, vmConfig *vm.Config) *ParallelStateProcessor { + return &ParallelStateProcessor{ + StateProcessor: NewStateProcessor(chain), + chain: chain, + vmCfg: vmConfig, + } +} + +// execVMConfig returns the subset of the configured VM options that is safe to +// reuse across the parallel per-transaction and post-transaction executions. +// Only the fields explicitly copied here are propagated (mirroring the original +// per-tx behaviour); notably the full caller-supplied config is used only for +// pre-execution in processBlockPreTx. +func (p *ParallelStateProcessor) execVMConfig() vm.Config { + return vm.Config{ + NoBaseFee: p.vmCfg.NoBaseFee, + EnablePreimageRecording: p.vmCfg.EnablePreimageRecording, + ExtraEips: slices.Clone(p.vmCfg.ExtraEips), + } +} + +// called by resultHandler when all transactions have successfully executed. +// performs post-tx state transition (system contracts and withdrawals) +// and calculates the ProcessResult, returning it to be sent on resCh +// by resultHandler +func (p *ParallelStateProcessor) prepareExecResult(block *types.Block, tExecStart time.Time, preTxBAL *bal.ConstructionBlockAccessList, accessList *bal.AccessListReader, statedb *state.StateDB, results []txExecResult) *ProcessResultWithMetrics { + tExec := time.Since(tExecStart) + tPostprocessStart := time.Now() + header := block.Header() + + // The post-execution changes are recorded at the BAL index immediately + // following the last transaction. + lastBALIdx := len(block.Transactions()) + 1 + postTxState := statedb.WithReader(state.NewReaderWithAccessList(statedb.Reader(), accessList, lastBALIdx)) + + evm := vm.NewEVM(NewEVMBlockContext(header, p.chain, nil), postTxState, p.chainConfig(), p.execVMConfig()) + + // 1. order the receipts by tx index + // 2. correctly calculate the cumulative gas used per receipt, returning bad block error if it goes over the allowed + slices.SortFunc(results, func(a, b txExecResult) int { + return cmp.Compare(a.receipt.TransactionIndex, b.receipt.TransactionIndex) + }) + + var ( + // Per-dimension cumulative sums for 2D block gas (EIP-8037). + sumRegular uint64 + sumState uint64 + cumulativeReceipt uint64 // cumulative receipt gas (what users pay) + + allLogs []*types.Log + allReceipts []*types.Receipt + ) + for _, result := range results { + sumRegular += result.txRegular + sumState += result.txState + + cumulativeReceipt += result.execGas + result.receipt.CumulativeGasUsed = cumulativeReceipt + allLogs = append(allLogs, result.receipt.Logs...) + allReceipts = append(allReceipts, result.receipt) + } + // Block gas = max(sum_regular, sum_state) per EIP-8037. + blockGasUsed := max(sumRegular, sumState) + if blockGasUsed > header.GasLimit { + return errResult(fmt.Errorf("gas limit exceeded")) + } + + requests, postBAL, err := PostExecution(context.Background(), p.chainConfig(), block.Number(), block.Time(), allLogs, evm, uint32(lastBALIdx)) + if err != nil { + return errResult(err) + } + + p.chain.Engine().Finalize(p.chain, block.Header(), evm.StateDB, block.Body(), uint32(lastBALIdx), postBAL) + + // Gather preimages from block execution. postTxState contains + // preimages from pre/post tx system contract execution. + preimages := make(map[common.Hash][]byte) + for hash, preimage := range postTxState.Preimages() { + preimages[hash] = preimage + } + for i := range results { + for hash, preimage := range results[i].preimages { + if _, ok := preimages[hash]; !ok { + preimages[hash] = preimage + } + } + } + + blockAccessList := bal.NewConstructionBlockAccessList() + blockAccessList.Merge(preTxBAL) + blockAccessList.Merge(postBAL) + for _, res := range results { + blockAccessList.Merge(res.blockAccessList) + } + + // TODO: do we move validation to ValidateState? + if block.AccessList().Hash() != blockAccessList.ToEncodingObj().Hash() { + // TODO: expose json string method on encoding block access list and log it here + return errResult(fmt.Errorf("invalid block access list: mismatch between local and remote block access list")) + } + + tPostprocess := time.Since(tPostprocessStart) + + return &ProcessResultWithMetrics{ + ProcessResult: &ProcessResult{ + Receipts: allReceipts, + Requests: requests, + Logs: allLogs, + GasUsed: blockGasUsed, + Bal: blockAccessList, + }, + PostProcessTime: tPostprocess, + ExecTime: tExec, + Preimages: preimages, + } +} + +type txExecResult struct { + receipt *types.Receipt + err error // non-EVM error which would render the block invalid + execGas uint64 // gas reported on the receipt (what the user pays) + + // Per-tx dimensional gas for Amsterdam 2D gas accounting (EIP-8037). + txRegular uint64 + txState uint64 + + blockAccessList *bal.ConstructionBlockAccessList + + // holds preimages recorded during execution of the tx + preimages map[common.Hash][]byte +} + +// resultHandler polls until all transactions have finished executing and the +// state root calculation is complete. The result is emitted on resCh. +func (p *ParallelStateProcessor) resultHandler(block *types.Block, preTxBAL *bal.ConstructionBlockAccessList, prepared *bal.AccessListReader, statedb *state.StateDB, tExecStart time.Time, txResCh <-chan txExecResult, stateRootCalcResCh <-chan stateRootCalculationResult, resCh chan *ProcessResultWithMetrics) { + // 1. if the block has transactions, receive the execution results from all of them and return an error on resCh if any txs err'd + // 2. once all txs are executed, compute the post-tx state transition and produce the ProcessResult sending it on resCh (or an error if the post-tx state didn't match what is reported in the BAL) + var ( + results []txExecResult + cumulativeStateGas uint64 + cumulativeRegularGas uint64 + execErr error + ) + + if numTx := len(block.Transactions()); numTx > 0 { + for completed := 0; completed < numTx; completed++ { + res := <-txResCh + if execErr != nil { + // A block-invalidating result was already seen; keep draining so + // the worker goroutines don't block on their sends. + continue + } + switch { + case res.err != nil: + execErr = res.err + default: + bottleneck := max(cumulativeRegularGas+res.txRegular, cumulativeStateGas+res.txState) + if bottleneck > block.GasLimit() { + execErr = fmt.Errorf("block used too much gas in bottleneck dimension: %d. block gas limit is %d", bottleneck, block.GasLimit()) + continue + } + cumulativeRegularGas += res.txRegular + cumulativeStateGas += res.txState + results = append(results, res) + } + } + + if execErr != nil { + // Drain stateRootCalcResCh so the calcAndVerifyRoot goroutine can exit. + <-stateRootCalcResCh + resCh <- errResult(execErr) + return + } + } + + execResults := p.prepareExecResult(block, tExecStart, preTxBAL, prepared, statedb, results) + rootCalcRes := <-stateRootCalcResCh + + switch { + case execResults.ProcessResult.Error != nil: + resCh <- execResults + case rootCalcRes.err != nil: + resCh <- errResult(rootCalcRes.err) + default: + execResults.StateTransitionMetrics = rootCalcRes.metrics + resCh <- execResults + } +} + +type stateRootCalculationResult struct { + err error + metrics *state.BALStateTransitionMetrics +} + +// calcAndVerifyRoot performs the post-state root hash calculation, verifying +// it against what is reported by the block and returning a result on resCh. +func (p *ParallelStateProcessor) calcAndVerifyRoot(block *types.Block, stateTransition *state.BALStateTransition, resCh chan stateRootCalculationResult) { + root := stateTransition.IntermediateRoot(false) + + res := stateRootCalculationResult{ + metrics: stateTransition.Metrics(), + } + if root != block.Root() { + res.err = fmt.Errorf("state root mismatch. local: %x. remote: %x", root, block.Root()) + } + resCh <- res +} + +// execTx executes a single transaction returning a result which includes state accessed/modified. +func (p *ParallelStateProcessor) execTx(block *types.Block, tx *types.Transaction, balIdx int, db *state.StateDB, signer types.Signer) *txExecResult { + header := block.Header() + evmContext := NewEVMBlockContext(header, p.chain, nil) + evm := vm.NewEVM(evmContext, db, p.chainConfig(), p.execVMConfig()) + + msg, err := TransactionToMessage(tx, signer, header.BaseFee) + if err != nil { + return &txExecResult{err: fmt.Errorf("could not apply tx %d [%v]: %w", balIdx, tx.Hash().Hex(), err)} + } + sender, err := signer.Sender(tx) + if err != nil { + return &txExecResult{err: fmt.Errorf("could not recover sender for tx at bal idx %d: %w", balIdx, err)} + } + + gp := NewGasPool(block.GasLimit()) + // TODO: make precompiled addresses be resolvable from chain config + block + db.Prepare(evm.GetRules(), sender, block.Coinbase(), tx.To(), vm.PrecompiledAddressesCancun, tx.AccessList()) + db.SetTxContext(tx.Hash(), balIdx-1, uint32(balIdx)) + + receipt, txBAL, err := ApplyTransactionWithEVM(msg, gp, db, block.Number(), block.Hash(), evmContext.Time, tx, evm) + if err != nil { + return &txExecResult{err: fmt.Errorf("could not apply tx %d [%v]: %w", balIdx, tx.Hash().Hex(), err)} + } + + return &txExecResult{ + receipt: receipt, + execGas: receipt.GasUsed, + txRegular: gp.cumulativeRegular, + txState: gp.cumulativeState, + preimages: db.Preimages(), + blockAccessList: txBAL, + } +} + +func (p *ParallelStateProcessor) processBlockPreTx(block *types.Block, parent *types.Header, statedb *state.StateDB, cfg vm.Config) *bal.ConstructionBlockAccessList { + header := block.Header() + evm := vm.NewEVM(NewEVMBlockContext(header, p.chain, nil), statedb, p.chainConfig(), cfg) + return PreExecution(context.Background(), block.BeaconRoot(), parent, p.chainConfig(), evm, block.Number(), block.Time()) +} + +// Process performs EVM execution and state root computation for a block which is known +// to contain an access list. +func (p *ParallelStateProcessor) Process(block *types.Block, stateTransition *state.BALStateTransition, statedb *state.StateDB, cfg vm.Config) (*ProcessResultWithMetrics, error) { + header := block.Header() + signer := types.MakeSigner(p.chainConfig(), header.Number, header.Time) + + var ( + resCh = make(chan *ProcessResultWithMetrics) + rootCalcResultCh = make(chan stateRootCalculationResult) + txResCh = make(chan txExecResult) + ) + + // Pre-transaction processing: system-contract updates and the pre-tx BAL. + pStart := time.Now() + startingState := statedb.Copy() + prepared := stateTransition.PreparedAccessList() + preTxBAL := p.processBlockPreTx(block, p.chain.GetHeader(block.ParentHash(), block.NumberU64()), statedb, cfg) + tPreprocess := time.Since(pStart) + + // Execute transactions and the state-root calculation in parallel. + tExecStart := time.Now() + go p.resultHandler(block, preTxBAL, prepared, statedb, tExecStart, txResCh, rootCalcResultCh, resCh) + + // Workers execute transactions concurrently against per-tx state copies. + // Each worker reports completion (and any block-invalidating error) on + // txResCh, which resultHandler drains. Worker errors therefore flow through + // the channel rather than the errgroup, so the group is used purely to bound + // concurrency and Wait() is intentionally not called. + var workers errgroup.Group + workers.SetLimit(runtime.NumCPU()) + for i, tx := range block.Transactions() { + balIdx := i + 1 + prestate := startingState.Copy() + workers.Go(func() error { + prestate = prestate.WithReader(state.NewReaderWithAccessList(statedb.Reader(), prepared, balIdx)) + res := p.execTx(block, tx, balIdx, prestate, signer) + txResCh <- *res + return nil + }) + } + + go p.calcAndVerifyRoot(block, stateTransition, rootCalcResultCh) + + res := <-resCh + if res.ProcessResult.Error != nil { + return nil, res.ProcessResult.Error + } + + stateTransition.SetPreimages(res.Preimages) + // TODO: remove preprocess metric ? + res.PreProcessTime = tPreprocess + return res, nil +} diff --git a/core/state/bal_state_transition.go b/core/state/bal_state_transition.go new file mode 100644 index 0000000000..bafaee671c --- /dev/null +++ b/core/state/bal_state_transition.go @@ -0,0 +1,531 @@ +package state + +import ( + "slices" + "sync" + "sync/atomic" + "time" + + "github.com/ethereum/go-ethereum/common" + "github.com/ethereum/go-ethereum/core/types" + "github.com/ethereum/go-ethereum/core/types/bal" + "github.com/ethereum/go-ethereum/crypto" + "github.com/ethereum/go-ethereum/trie/trienode" + "github.com/holiman/uint256" + "golang.org/x/sync/errgroup" +) + +// BALStateTransition computes the state root update and commits a single block +// which contains an access list. +type BALStateTransition struct { + accessList *bal.AccessListReader + written bal.WrittenCounts + db Database + reader Reader + stateTrie Trie + parentRoot common.Hash + + // the computed state root of the block + rootHash common.Hash + // the state modifications performed by the block + diffs bal.StateMutations + + // a map of common.Address -> *types.StateAccount containing the block + // prestate of all accounts that will be modified + prestates sync.Map + + postStates map[common.Address]*types.StateAccount + // a map of common.Address -> Trie containing the account tries for all + // accounts with mutated storage + tries sync.Map //map[common.Address]Trie + deletions map[common.Address]struct{} + + // Deletion counters; not derivable from the BAL alone (selfdestruct vs + // balance drain is indistinguishable without prestate). + accountDeleted int + storageDeleted atomic.Int64 + + stateUpdate *StateUpdate + + // preimages holds the SHA3 preimages seen by the VM across the whole block: + // the pre/post-transaction system executions and every individual + // transaction execution. It is populated once via SetPreimages. + preimages map[common.Hash][]byte + + metrics BALStateTransitionMetrics + maxBALIdx int + + err error +} + +// Metrics returns metrics if Commit has previously been called. +func (s *BALStateTransition) Metrics() *BALStateTransitionMetrics { + return &s.metrics +} + +// DeletionCounts holds per-block deletion counters for accounts/storage +type DeletionCounts struct { + Accounts int + Storage int +} + +func (s *BALStateTransition) Deletions() DeletionCounts { + return DeletionCounts{ + Accounts: s.accountDeleted, + Storage: int(s.storageDeleted.Load()), + } +} + +// BALStateTransitionMetrics contains metrics from a committed block. +type BALStateTransitionMetrics struct { + // trie hashing metrics + AccountUpdate time.Duration + StatePrefetch time.Duration + StateUpdate time.Duration + StateHash time.Duration +} + +// NewBALStateTransition creates a BALStateTransition instance +func NewBALStateTransition(block *types.Block, prefetchReader Reader, db Database, parentRoot common.Hash, prepared *bal.AccessListReader) (*BALStateTransition, error) { + stateTrie, err := db.OpenTrie(parentRoot) + if err != nil { + return nil, err + } + + return &BALStateTransition{ + accessList: prepared, + written: block.AccessList().WrittenCounts(), + db: db, + reader: prefetchReader, + stateTrie: stateTrie, + parentRoot: parentRoot, + rootHash: common.Hash{}, + diffs: make(bal.StateMutations), + prestates: sync.Map{}, + postStates: make(map[common.Address]*types.StateAccount), + tries: sync.Map{}, + deletions: make(map[common.Address]struct{}), + stateUpdate: nil, + preimages: make(map[common.Hash][]byte), + maxBALIdx: len(block.Transactions()) + 1, + }, nil +} + +// WrittenCounts returns the cached BAL write counts (computed once per block). +func (s *BALStateTransition) WrittenCounts() bal.WrittenCounts { + return s.written +} + +// PreparedAccessList returns the shared, read-only preprocessed access list for +// the block. It is built once per block and reused by the parallel execution +// readers so the preprocessing is not repeated per transaction. +func (s *BALStateTransition) PreparedAccessList() *bal.AccessListReader { + return s.accessList +} + +// Error returns an error if IntermediateRoot or Commit failed. +func (s *BALStateTransition) Error() error { + return s.err +} + +func (s *BALStateTransition) setError(err error) { + if s.err == nil { + s.err = err + } +} + +// isAccountDeleted checks whether the state account was deleted in this block. Post selfdestruct-removal, +// deletions can only occur if an account which has a balance becomes the target of a CREATE2 initcode +// which calls SENDALL, clearing the account and marking it for deletion. +func isAccountDeleted(prestate *types.StateAccount, mutations bal.AccountMutations) bool { + if mutations.Code != nil && len(mutations.Code) != 0 { + return false + } + if mutations.Nonce != nil && *mutations.Nonce != 0 { + return false + } + if mutations.StorageWrites != nil && len(mutations.StorageWrites) > 0 { + return false + } + if mutations.Balance != nil { + if mutations.Balance.IsZero() { + if prestate.Nonce != 0 || prestate.Balance.IsZero() || common.BytesToHash(prestate.CodeHash) != types.EmptyCodeHash { + return false + } + // consider an empty account with storage to be deleted, so we don't check root here + return true + } + } + return false +} + +// updateAccount applies the block state mutations to a given account returning +// the updated state account and new code (if the account code changed) +func (s *BALStateTransition) updateAccount(addr common.Address) (*types.StateAccount, []byte) { + a, _ := s.prestates.Load(addr) + acct := a.(*types.StateAccount) + + acct, diff := acct.Copy(), s.diffs[addr] + code := diff.Code + + if diff.Nonce != nil { + acct.Nonce = *diff.Nonce + } + if diff.Balance != nil { + acct.Balance = new(uint256.Int).Set(diff.Balance) + } + if tr, ok := s.tries.Load(addr); ok { + acct.Root = tr.(Trie).Hash() + } + return acct, code +} + +func (s *BALStateTransition) commitAccount(addr common.Address) (*AccountUpdate, *trienode.NodeSet, error) { + op := &AccountUpdate{ + Address: addr, + Data: s.postStates[addr], + } + var prestate *types.StateAccount + if ps, exist := s.prestates.Load(addr); exist { + op.Origin = ps.(*types.StateAccount) + } + + if s.diffs[addr].Code != nil { + code := ContractCode{ + Hash: crypto.Keccak256Hash(s.diffs[addr].Code), + Blob: s.diffs[addr].Code, + } + if prestate == nil { + code.OriginHash = types.EmptyCodeHash + } else { + code.OriginHash = common.BytesToHash(prestate.CodeHash) + } + op.Code = &code + } + + if len(s.diffs[addr].StorageWrites) == 0 { + return op, nil, nil + } + + op.Storages = make(map[common.Hash]common.Hash) + op.StoragesOriginByHash = make(map[common.Hash]common.Hash) + op.StoragesOriginByKey = make(map[common.Hash]common.Hash) + + for key, value := range s.diffs[addr].StorageWrites { + hash := crypto.Keccak256Hash(key[:]) + op.Storages[hash] = value + origin, err := s.reader.Storage(addr, key) + if err != nil { + return nil, nil, err + } + op.StoragesOriginByHash[hash] = origin + op.StoragesOriginByKey[key] = origin + } + tr, _ := s.tries.Load(addr) + root, nodes := tr.(Trie).Commit(false) + s.postStates[addr].Root = root + return op, nodes, nil +} + +// CommitWithUpdate flushes mutated trie nodes and state accounts to disk. +func (s *BALStateTransition) CommitWithUpdate(block uint64, deleteEmptyObjects bool, noStorageWiping bool) (common.Hash, *StateUpdate, error) { + // 1) create a stateUpdate object + // Commit objects to the trie, measuring the elapsed time + var ( + //commitStart = time.Now() + accountTrieNodesUpdated int + accountTrieNodesDeleted int + storageTrieNodesUpdated int + storageTrieNodesDeleted int + + lock sync.Mutex // protect two maps below + nodes = trienode.NewMergedNodeSet() // aggregated trie nodes + updates = make(map[common.Hash]*AccountUpdate, len(s.diffs)) // aggregated account updates + + // merge aggregates the dirty trie nodes into the global set. + // + // Given that some accounts may be destroyed and then recreated within + // the same block, it's possible that a node set with the same owner + // may already exist. In such cases, these two sets are combined, with + // the later one overwriting the previous one if any nodes are modified + // or deleted in both sets. + // + // merge run concurrently across all the state objects and account trie. + merge = func(set *trienode.NodeSet) error { + if set == nil { + return nil + } + lock.Lock() + defer lock.Unlock() + + updates, deletes := set.Size() + if set.Owner == (common.Hash{}) { + accountTrieNodesUpdated += updates + accountTrieNodesDeleted += deletes + } else { + storageTrieNodesUpdated += updates + storageTrieNodesDeleted += deletes + } + return nodes.Merge(set) + } + ) + + destructedPrestates := make(map[common.Address]*types.StateAccount) + s.prestates.Range(func(key, value any) bool { + addr := key.(common.Address) + acct := value.(*types.StateAccount) + destructedPrestates[addr] = acct + return true + }) + + deletes, delNodes, err := handleDestruction(s.db, s.stateTrie, s.parentRoot, noStorageWiping, slices.Values(s.accessList.AllDestructions()), destructedPrestates) + if err != nil { + return common.Hash{}, nil, err + } + for _, set := range delNodes { + if err := merge(set); err != nil { + return common.Hash{}, nil, err + } + } + + // Handle all state updates afterwards, concurrently to one another to shave + // off some milliseconds from the commit operation. Also accumulate the code + // writes to run in parallel with the computations. + var ( + root common.Hash + workers errgroup.Group + ) + // Schedule the account trie first since that will be the biggest, so give + // it the most time to crunch. + // + // TODO(karalabe): This account trie commit is *very* heavy. 5-6ms at chain + // heads, which seems excessive given that it doesn't do hashing, it just + // shuffles some data. For comparison, the *hashing* at chain head is 2-3ms. + // We need to investigate what's happening as it seems something's wonky. + // Obviously it's not an end of the world issue, just something the original + // code didn't anticipate for. + workers.Go(func() error { + // Write the account trie changes, measuring the amount of wasted time + newroot, set := s.stateTrie.Commit(true) + root = newroot + + if err := merge(set); err != nil { + return err + } + return nil + }) + + // Schedule each of the storage tries that need to be updated, so they can + // run concurrently to one another. + // + // TODO(karalabe): Experimentally, the account commit takes approximately the + // same time as all the storage commits combined, so we could maybe only have + // 2 threads in total. But that kind of depends on the account commit being + // more expensive than it should be, so let's fix that and revisit this todo. + for addr, _ := range s.diffs { + if _, isDeleted := s.deletions[addr]; isDeleted { + continue + } + + address := addr + // Run the storage updates concurrently to one another + workers.Go(func() error { + // Write any storage changes in the state object to its storage trie + update, set, err := s.commitAccount(address) + if err != nil { + return err + } + + if err := merge(set); err != nil { + return err + } + lock.Lock() + updates[crypto.Keccak256Hash(address[:])] = update + lock.Unlock() + return nil + }) + } + // Wait for everything to finish and update the metrics + if err := workers.Wait(); err != nil { + return common.Hash{}, nil, err + } + + storageDeleted := s.storageDeleted.Load() + accountUpdatedMeter.Mark(int64(s.written.Accounts - s.accountDeleted)) + storageUpdatedMeter.Mark(int64(s.written.StorageSlots) - storageDeleted) + accountDeletedMeter.Mark(int64(s.accountDeleted)) + storageDeletedMeter.Mark(storageDeleted) + accountTrieUpdatedMeter.Mark(int64(accountTrieNodesUpdated)) + accountTrieDeletedMeter.Mark(int64(accountTrieNodesDeleted)) + storageTriesUpdatedMeter.Mark(int64(storageTrieNodesUpdated)) + storageTriesDeletedMeter.Mark(int64(storageTrieNodesDeleted)) + + storageKeyType := StorageKeyHashed + if noStorageWiping { + storageKeyType = StorageKeyPlain + } + update := NewStateUpdate(storageKeyType, s.parentRoot, root, block, deletes, updates, nodes) + + if err := s.db.Commit(update); err != nil { + return common.Hash{}, nil, err + } + return root, update, nil +} + +func (s *BALStateTransition) Commit(block uint64, deleteEmptyObjects bool, noStorageWiping bool) (common.Hash, error) { + hash, _, err := s.CommitWithUpdate(block, deleteEmptyObjects, noStorageWiping) + return hash, err +} + +// IntermediateRoot applies block state mutations and computes the updated state +// trie root. +func (s *BALStateTransition) IntermediateRoot(_ bool) common.Hash { + if s.rootHash != (common.Hash{}) { + return s.rootHash + } + + // State root calculation proceeds as follows: + + // 1 (a): load the origin storage values for all slots which were modified during the block (this is needed for computing the stateUpdate) + // 1 (b): update each mutated account, producing the post-block state object by applying the state mutations to the prestate (retrieved in 1a). + // 1 (c): prefetch the intermediate trie nodes of the mutated state set from the account trie. + // + // 2: compute the post-state root of the account trie + // + // Steps 1/2 are performed sequentially, with steps 1a-d performed in parallel + + start := time.Now() + + var wg sync.WaitGroup + + s.diffs = *s.accessList.Mutations(s.maxBALIdx + 1) + + for addr, d := range s.diffs { + wg.Add(1) + address := addr + diff := d + go func() { + defer wg.Done() + + // 1 (b): update each mutated account, producing the post-block state object by applying the state mutations to the prestate (retrieved in 1a). + acct, err := s.reader.Account(address) + if err != nil { + s.setError(err) + return + } + + if acct == nil { + acct = types.NewEmptyStateAccount() + } + s.prestates.Store(address, acct) + + if len(diff.StorageWrites) > 0 { + tr, err := s.db.OpenStorageTrie(s.parentRoot, address, acct.Root, s.stateTrie) + if err != nil { + s.setError(err) + return + } + s.tries.Store(address, tr) + + var ( + updateKeys, updateValues [][]byte + deleteKeys [][]byte + ) + for key, val := range diff.StorageWrites { + if val != (common.Hash{}) { + updateKeys = append(updateKeys, key[:]) + updateValues = append(updateValues, common.TrimLeftZeroes(val[:])) + } else { + deleteKeys = append(deleteKeys, key[:]) + } + } + + if err := tr.UpdateStorageBatch(address, updateKeys, updateValues); err != nil { + s.setError(err) + return + } + + for _, key := range deleteKeys { + if err := tr.DeleteStorage(address, key); err != nil { + s.setError(err) + return + } + } + + hashStart := time.Now() + tr.Hash() + s.metrics.StateHash = time.Since(hashStart) + } + }() + } + + wg.Add(1) + // 1 (c): prefetch the intermediate trie nodes of the mutated state set from the account trie. + go func() { + defer wg.Done() + prefetchStart := time.Now() + var prefetchAddrs []common.Address + for addr, _ := range s.diffs { + prefetchAddrs = append(prefetchAddrs, addr) + } + if err := s.stateTrie.PrefetchAccount(prefetchAddrs); err != nil { + s.setError(err) + return + } + s.metrics.StatePrefetch = time.Since(prefetchStart) + }() + + wg.Wait() + s.metrics.AccountUpdate = time.Since(start) + + // 2: compute the post-state root of the account trie + stateUpdateStart := time.Now() + for mutatedAddr, _ := range s.diffs { + p, _ := s.prestates.Load(mutatedAddr) + prestate := p.(*types.StateAccount) + + isDeleted := isAccountDeleted(prestate, s.diffs[mutatedAddr]) + if isDeleted { + if err := s.stateTrie.DeleteAccount(mutatedAddr); err != nil { + s.setError(err) + return common.Hash{} + } + s.deletions[mutatedAddr] = struct{}{} + s.accountDeleted++ + } else { + acct, code := s.updateAccount(mutatedAddr) + + if code != nil { + codeHash := crypto.Keccak256Hash(code) + acct.CodeHash = codeHash.Bytes() + if err := s.stateTrie.UpdateContractCode(mutatedAddr, codeHash, code); err != nil { + s.setError(err) + return common.Hash{} + } + } + if err := s.stateTrie.UpdateAccount(mutatedAddr, acct, len(code)); err != nil { + s.setError(err) + return common.Hash{} + } + s.postStates[mutatedAddr] = acct + } + } + + s.metrics.StateUpdate = time.Since(stateUpdateStart) + + stateTrieHashStart := time.Now() + s.rootHash = s.stateTrie.Hash() + s.metrics.StateHash = time.Since(stateTrieHashStart) + return s.rootHash +} + +// SetPreimages sets the preimages. +func (s *BALStateTransition) SetPreimages(preimages map[common.Hash][]byte) { + s.preimages = preimages +} + +// Preimages returns the SHA3 preimages seen by the VM over the course of the +// block. +func (s *BALStateTransition) Preimages() map[common.Hash][]byte { + return s.preimages +} diff --git a/core/state/database.go b/core/state/database.go index 1001a4b4cf..277ff60d3a 100644 --- a/core/state/database.go +++ b/core/state/database.go @@ -54,6 +54,10 @@ type Database interface { // Reader returns a state reader associated with the specified state root. Reader(root common.Hash) (Reader, error) + // ReaderWithPrefetch returns a reader which asynchronously fetches block + // access list state in the background. + ReaderWithPrefetch(stateRoot common.Hash, accessList map[common.Address][]common.Hash) (Reader, error) + // Iteratee returns a state iteratee associated with the specified state root, // through which the account iterator and storage iterator can be created. Iteratee(root common.Hash) (Iteratee, error) diff --git a/core/state/database_history.go b/core/state/database_history.go index fbf4ab5f9c..d7d051874b 100644 --- a/core/state/database_history.go +++ b/core/state/database_history.go @@ -223,6 +223,10 @@ type HistoricDB struct { codedb *CodeDB } +func (db *HistoricDB) ReaderWithPrefetch(stateRoot common.Hash, accessList map[common.Address][]common.Hash) (Reader, error) { + panic("not implemented") +} + // Type returns the trie type of the underlying database. func (db *HistoricDB) Type() DatabaseType { // TODO(rjl493456442) support UBT in the future diff --git a/core/state/database_mpt.go b/core/state/database_mpt.go index 42c5f2e5ef..228b7eb61b 100644 --- a/core/state/database_mpt.go +++ b/core/state/database_mpt.go @@ -24,6 +24,7 @@ import ( "github.com/ethereum/go-ethereum/log" "github.com/ethereum/go-ethereum/trie" "github.com/ethereum/go-ethereum/triedb" + "runtime" ) // MPTDatabase is an implementation of Database interface for Merkle Patricia Tries. @@ -185,3 +186,17 @@ func (db *MPTDatabase) Commit(update *StateUpdate) error { func (db *MPTDatabase) Iteratee(root common.Hash) (Iteratee, error) { return newStateIteratee(true, root, db.triedb, db.snap) } + +func (db *MPTDatabase) ReaderWithPrefetch(stateRoot common.Hash, accessList map[common.Address][]common.Hash) (Reader, error) { + base, err := db.StateReader(stateRoot) + if err != nil { + return nil, err + } + // Construct the state reader with native cache and associated statistics + r := newStateReaderWithStats(newStateReaderWithCache(base)) + + // Construct the state reader with background prefetching + pr := newPrefetchStateReader(r, accessList, runtime.NumCPU()) + + return newReaderWithPrefetch(db.codedb.Reader(), pr, pr), nil +} diff --git a/core/state/database_ubt.go b/core/state/database_ubt.go index d9b2f07a77..b088927dea 100644 --- a/core/state/database_ubt.go +++ b/core/state/database_ubt.go @@ -96,6 +96,10 @@ func (db *UBTDatabase) Reader(stateRoot common.Hash) (Reader, error) { return newReader(db.codedb.Reader(), sr), nil } +func (db *UBTDatabase) ReaderWithPrefetch(stateRoot common.Hash, accessList map[common.Address][]common.Hash) (Reader, error) { + panic("not implemented") +} + // ReadersWithCacheStats creates a pair of state readers that share the same // underlying state reader and internal state cache, while maintaining separate // statistics respectively. diff --git a/core/state/reader.go b/core/state/reader.go index be07cec0f9..c485af96ed 100644 --- a/core/state/reader.go +++ b/core/state/reader.go @@ -560,6 +560,7 @@ func (r *stateReaderWithStats) GetStateStats() StateReaderStats { type reader struct { ContractCodeReader StateReader + PrefetcherMetricer } // newReader constructs a reader with the supplied code reader and state reader. @@ -570,6 +571,14 @@ func newReader(codeReader ContractCodeReader, stateReader StateReader) *reader { } } +func newReaderWithPrefetch(codeReader ContractCodeReader, stateReader StateReader, metricer PrefetcherMetricer) *reader { + return &reader{ + ContractCodeReader: codeReader, + StateReader: stateReader, + PrefetcherMetricer: metricer, + } +} + // GetCodeStats returns the statistics of code access. func (r *reader) GetCodeStats() ContractCodeReaderStats { if stater, ok := r.ContractCodeReader.(ContractCodeReaderStater); ok { diff --git a/core/state/reader_eip_7928.go b/core/state/reader_eip_7928.go index ff315ac5eb..0b1bfb0874 100644 --- a/core/state/reader_eip_7928.go +++ b/core/state/reader_eip_7928.go @@ -16,14 +16,6 @@ package state -import ( - "sync" - - "github.com/ethereum/go-ethereum/common" - "github.com/ethereum/go-ethereum/core/types" - "github.com/ethereum/go-ethereum/core/types/bal" -) - // The EIP27928 reader utilizes a hierarchical architecture to optimize state // access during block execution: // @@ -39,15 +31,13 @@ import ( // This layer provides a "unified view" by merging the pre-transition state // with mutated states from preceding transactions in the block. // -// - Tracking Layer: Finally, the readerTracker wraps the execution reader to -// capture all state reads made during a specific transaction. These individual -// reads are subsequently merged to construct a comprehensive access list -// for the entire block. -// // The architecture can be illustrated by the diagram below: -// + +// [ Block Level Access List ] <────────────────┐ +// ▲ │ (Merge) +// │ │ // ┌──────────────┴──────────────┐ ┌──────────────┴──────────────┐ -// │ ReaderWithBlockLevelAL │ │ ReaderWithBlockLevelAL │ +// │ ReaderWithBlockLevelAL │ │ ReaderWithBlockLevelAL │ (Unified View) // │ (Pre-state + Mutations) │ │ (Pre-state + Mutations) │ // └──────────────┬──────────────┘ └──────────────┬──────────────┘ // │ │ @@ -63,11 +53,16 @@ import ( // │ (State & Contract Code) │ // └─────────────────────────────┘ -// Note: The block producer, which is responsible for generating the block -// along with the block-level access list, does not maintain the internal -// hierarchy (e.g., PrefetchStateReader or ReaderWithBlockLevelAL). -// Instead, it directly utilizes the readerTracker, wrapped around the -// base reader, to construct the access list. +import ( + "sync" + "time" + + "github.com/ethereum/go-ethereum/crypto" + + "github.com/ethereum/go-ethereum/common" + "github.com/ethereum/go-ethereum/core/types" + "github.com/ethereum/go-ethereum/core/types/bal" +) type fetchTask struct { addr common.Address @@ -78,16 +73,27 @@ func (t *fetchTask) weight() int { return 1 + len(t.slots) } type prefetchStateReader struct { StateReader - tasks []*fetchTask nThreads int done chan struct{} term chan struct{} closeOnce sync.Once + start time.Time + metrics PrefetchMetrics } -// nolint:unused -func newPrefetchStateReader(reader StateReader, accessList map[common.Address][]common.Hash, nThreads int) *prefetchStateReader { +type PrefetchMetrics struct { + // the total amount of time it took to complete the scheduled workload + Elapsed time.Duration +} + +// PrefetcherMetricer is an object that can expose metrics related to the state +// prefetching. +type PrefetcherMetricer interface { + Metrics() PrefetchMetrics +} + +func newPrefetchStateReader(reader StateReader, accessList bal.StorageKeys, nThreads int) *prefetchStateReader { tasks := make([]*fetchTask, 0, len(accessList)) for addr, slots := range accessList { tasks = append(tasks, &fetchTask{ @@ -105,11 +111,16 @@ func newPrefetchStateReaderInternal(reader StateReader, tasks []*fetchTask, nThr nThreads: nThreads, done: make(chan struct{}), term: make(chan struct{}), + start: time.Now(), } go r.prefetch() return r } +func (r *prefetchStateReader) Metrics() PrefetchMetrics { + return r.metrics +} + func (r *prefetchStateReader) Close() { r.closeOnce.Do(func() { close(r.term) @@ -127,7 +138,10 @@ func (r *prefetchStateReader) Wait() error { } func (r *prefetchStateReader) prefetch() { - defer close(r.done) + defer func() { + r.metrics = PrefetchMetrics{time.Since(r.start)} + close(r.done) + }() if len(r.tasks) == 0 { return @@ -196,52 +210,108 @@ func (r *prefetchStateReader) process(start, limit int) { // ReaderWithBlockLevelAccessList provides state access that reflects the // pre-transition state combined with the mutations made by transactions // prior to TxIndex. +// +// It is a cheap, per-transaction view over a shared, read-only +// bal.AccessListReader: constructing one is O(1) and every lookup is an +// allocation-free binary search. type ReaderWithBlockLevelAccessList struct { Reader - AccessList *bal.ConstructionBlockAccessList - TxIndex int + prepared *bal.AccessListReader + TxIndex int } -// NewReaderWithBlockLevelAccessList constructs a reader for accessing states +// NewReaderWithAccessList constructs a reader for accessing states // with the mutations made by transactions prior to txIndex. // // The txIndex refers to the call frame as such: // - 0 for pre‑execution system contract calls. // - 1 … n for transactions (in block order). // - n + 1 for post‑execution system contract calls. -func NewReaderWithBlockLevelAccessList(base Reader, accessList *bal.ConstructionBlockAccessList, txIndex int) *ReaderWithBlockLevelAccessList { +func NewReaderWithAccessList(base Reader, prepared *bal.AccessListReader, txIndex int) *ReaderWithBlockLevelAccessList { return &ReaderWithBlockLevelAccessList{ - Reader: base, - AccessList: accessList, - TxIndex: txIndex, + Reader: base, + prepared: prepared, + TxIndex: txIndex, } } +// NewReaderWithBlockLevelAccessList wraps a base reader with a raw access list, +// preprocessing it on the spot. Prefer NewReaderWithAccessList when the +// prepared list can be built once and shared across multiple readers. +func NewReaderWithBlockLevelAccessList(base Reader, accessList bal.BlockAccessList, txIndex int) *ReaderWithBlockLevelAccessList { + return NewReaderWithAccessList(base, bal.NewAccessListReader(accessList), txIndex) +} + // Account implements Reader, returning the account with the specific address. -func (r *ReaderWithBlockLevelAccessList) Account(addr common.Address) (*types.StateAccount, error) { - panic("implement me") +func (r *ReaderWithBlockLevelAccessList) Account(addr common.Address) (acct *types.StateAccount, err error) { + acct, err = r.Reader.Account(addr) + if err != nil { + return nil, err + } + + balance := r.prepared.Balance(addr, r.TxIndex) + code := r.prepared.Code(addr, r.TxIndex) + nonce, hasNonce := r.prepared.Nonce(addr, r.TxIndex) + if balance == nil && code == nil && !hasNonce { + return acct, nil + } + + if acct == nil { + acct = types.NewEmptyStateAccount() + } else { + // the account returned by the underlying reader is a reference + // copy it to avoid mutating the reader's instance + acct = acct.Copy() + } + + // balance and code alias the shared access list; this is safe because the + // EVM never mutates them in place (it replaces the pointer/slice wholesale, + // and the journal clones before stashing). + if balance != nil { + acct.Balance = balance + } + if code != nil { + codeHash := crypto.Keccak256Hash(code) + acct.CodeHash = codeHash[:] + } + if hasNonce { + acct.Nonce = nonce + } + return } // Storage implements Reader, returning the storage slot with the specific // address and slot key. func (r *ReaderWithBlockLevelAccessList) Storage(addr common.Address, slot common.Hash) (common.Hash, error) { - panic("implement me") + if val, ok := r.prepared.StorageAt(addr, slot, r.TxIndex); ok { + return val, nil + } + return r.Reader.Storage(addr, slot) } // Has implements Reader, returning the flag indicating whether the contract // code with specified address and hash exists or not. func (r *ReaderWithBlockLevelAccessList) Has(addr common.Address, codeHash common.Hash) bool { - panic("implement me") + if code := r.prepared.Code(addr, r.TxIndex); code != nil { + return crypto.Keccak256Hash(code) == codeHash + } + return r.Reader.Has(addr, codeHash) } // Code implements Reader, returning the contract code with specified address // and hash. -func (r *ReaderWithBlockLevelAccessList) Code(addr common.Address, codeHash common.Hash) ([]byte, error) { - panic("implement me") +func (r *ReaderWithBlockLevelAccessList) Code(addr common.Address, codeHash common.Hash) []byte { + if code := r.prepared.Code(addr, r.TxIndex); code != nil && crypto.Keccak256Hash(code) == codeHash { + return code + } + return r.Reader.Code(addr, codeHash) } // CodeSize implements Reader, returning the contract code size with specified // address and hash. -func (r *ReaderWithBlockLevelAccessList) CodeSize(addr common.Address, codeHash common.Hash) (int, error) { - panic("implement me") +func (r *ReaderWithBlockLevelAccessList) CodeSize(addr common.Address, codeHash common.Hash) int { + if code := r.prepared.Code(addr, r.TxIndex); code != nil && crypto.Keccak256Hash(code) == codeHash { + return len(code) + } + return r.Reader.CodeSize(addr, codeHash) } diff --git a/core/state/statedb.go b/core/state/statedb.go index 09a896a89a..ef67315507 100644 --- a/core/state/statedb.go +++ b/core/state/statedb.go @@ -21,6 +21,7 @@ import ( "bytes" "errors" "fmt" + "iter" "maps" "slices" "sort" @@ -182,6 +183,13 @@ func New(root common.Hash, db Database) (*StateDB, error) { return NewWithReader(root, db, reader) } +// WithReader returns a copy of the statedb instance with the specified reader. +func (s *StateDB) WithReader(reader Reader) *StateDB { + cpy := s.Copy() + cpy.reader = reader + return cpy +} + // NewWithReader creates a new state for the specified state root. Unlike New, // this function accepts an additional Reader which is bound to the given root. func NewWithReader(root common.Hash, db Database, reader Reader) (*StateDB, error) { @@ -1133,12 +1141,15 @@ func (s *StateDB) clearJournalAndRefund() { // deleteStorage is designed to delete the storage trie of a designated account. func (s *StateDB) deleteStorage(addrHash common.Hash, root common.Hash) (map[common.Hash]common.Hash, map[common.Hash]common.Hash, *trienode.NodeSet, error) { + return deleteStorage(s.db, s.originalRoot, addrHash, root) +} +func deleteStorage(db Database, originalRoot common.Hash, addrHash common.Hash, root common.Hash) (map[common.Hash]common.Hash, map[common.Hash]common.Hash, *trienode.NodeSet, error) { var ( nodes = trienode.NewNodeSet(addrHash) // the set for trie node mutations (value is nil) storages = make(map[common.Hash]common.Hash) // the set for storage mutations (value is nil) storageOrigins = make(map[common.Hash]common.Hash) // the set for tracking the original value of slot ) - iteratee, err := s.db.Iteratee(s.originalRoot) + iteratee, err := db.Iteratee(originalRoot) if err != nil { return nil, nil, nil, err } @@ -1567,3 +1578,72 @@ func (s *StateDB) Witness() *stateless.Witness { func (s *StateDB) AccessEvents() *AccessEvents { return s.accessEvents } + +// handleDestruction processes all destruction markers and deletes the account +// and associated storage slots if necessary. There are four potential scenarios +// as following: +// +// (a) the account was not existent and be marked as destructed +// (b) the account was not existent and be marked as destructed, +// however, it's resurrected later in the same block. +// (c) the account was existent and be marked as destructed +// (d) the account was existent and be marked as destructed, +// however it's resurrected later in the same block. +// +// In case (a), nothing needs be deleted, nil to nil transition can be ignored. +// In case (b), nothing needs be deleted, nil is used as the original value for +// newly created account and storages +// In case (c), **original** account along with its storages should be deleted, +// with their values be tracked as original value. +// In case (d), **original** account along with its storages should be deleted, +// with their values be tracked as original value. +func handleDestruction(db Database, trie Trie, root common.Hash, noStorageWiping bool, destructions iter.Seq[common.Address], prestates map[common.Address]*types.StateAccount) (map[common.Hash]*AccountDelete, []*trienode.NodeSet, error) { + var ( + nodes []*trienode.NodeSet + deletes = make(map[common.Hash]*AccountDelete) + ) + for addr := range destructions { + prestate := prestates[addr] + // The account was non-existent, and it's marked as destructed in the scope + // of block. It can be either case (a) or (b) and will be interpreted as + // null->null state transition. + // - for (a), skip it without doing anything + // - for (b), the resurrected account with nil as original will be handled afterwards + if prestate == nil { + continue + } + // The account was existent, it can be either case (c) or (d). + addrHash := crypto.Keccak256Hash(addr.Bytes()) + op := &AccountDelete{ + Address: addr, + Origin: prestate, + } + deletes[addrHash] = op + + // Short circuit if the origin storage was empty. + if prestate.Root == types.EmptyRootHash || db.TrieDB().IsUBT() { + continue + } + if noStorageWiping { + return nil, nil, fmt.Errorf("unexpected storage wiping, %x", addr) + } + // Remove storage slots belonging to the account. + storages, storagesOrigin, set, err := deleteStorage(db, prestate.Root, addrHash, root) + if err != nil { + return nil, nil, fmt.Errorf("failed to delete storage, err: %w", err) + } + op.Storages = storages + op.StoragesOrigin = storagesOrigin + + // Aggregate the associated trie node changes. + nodes = append(nodes, set) + } + return deletes, nodes, nil +} + +// TODO: find better location for this +type Committer interface { + Commit(block uint64, deleteEmptyObjects bool, noStorageWiping bool) (common.Hash, error) + CommitWithUpdate(block uint64, deleteEmptyObjects bool, noStorageWiping bool) (common.Hash, *StateUpdate, error) + Preimages() map[common.Hash][]byte +} diff --git a/core/types.go b/core/types.go index f6f15101e0..98035262b1 100644 --- a/core/types.go +++ b/core/types.go @@ -34,7 +34,7 @@ type Validator interface { ValidateBody(block *types.Block) error // ValidateState validates the given statedb and optionally the process result. - ValidateState(block *types.Block, state *state.StateDB, res *ProcessResult, stateless bool) error + ValidateState(block *types.Block, state StateRootSource, res *ProcessResult, stateless bool) error } // Prefetcher is an interface for pre-caching transaction signatures and state. @@ -63,4 +63,6 @@ type ProcessResult struct { // BAL is only meaningful for post-Amsterdam blocks. Please ensure // fork validation is performed before accessing it. Bal *bal.ConstructionBlockAccessList + + Error error } diff --git a/core/types/bal/bal.go b/core/types/bal/bal.go index 2eb5fe93cd..9845916765 100644 --- a/core/types/bal/bal.go +++ b/core/types/bal/bal.go @@ -18,6 +18,7 @@ package bal import ( "bytes" + "encoding/json" "maps" "github.com/ethereum/go-ethereum/common" @@ -223,3 +224,137 @@ func (b *ConstructionBlockAccessList) Copy() *ConstructionBlockAccessList { } return res } + +type StorageMutations map[common.Hash]common.Hash + +// AccountMutations contains mutations that were made to an account across +// one or more access list indices. +type AccountMutations struct { + Balance *uint256.Int `json:"Balance,omitempty"` + Nonce *uint64 `json:"Nonce,omitempty"` + Code []byte `json:"Code,omitempty"` + StorageWrites StorageMutations `json:"StorageWrites,omitempty"` +} + +// String returns a human-readable JSON representation of the account mutations. +func (a *AccountMutations) String() string { + var res bytes.Buffer + enc := json.NewEncoder(&res) + enc.SetIndent("", " ") + enc.Encode(a) + return res.String() +} + +// Copy returns a deep-copy of the instance. +func (a *AccountMutations) Copy() *AccountMutations { + res := &AccountMutations{ + nil, + nil, + nil, + nil, + } + if a.Nonce != nil { + res.Nonce = new(uint64) + *res.Nonce = *a.Nonce + } + if a.Code != nil { + res.Code = bytes.Clone(a.Code) + } + if a.Balance != nil { + res.Balance = new(uint256.Int).Set(a.Balance) + } + if a.StorageWrites != nil { + res.StorageWrites = maps.Clone(a.StorageWrites) + } + return res +} + +// Eq returns whether the calling instance is equal to the provided one. +func (a *AccountMutations) Eq(other *AccountMutations) bool { + if a.Balance != nil || other.Balance != nil { + if a.Balance == nil || other.Balance == nil { + return false + } + + if !a.Balance.Eq(other.Balance) { + return false + } + } + + if (len(a.Code) != 0 || len(other.Code) != 0) && !bytes.Equal(a.Code, other.Code) { + return false + } + + if a.Nonce != nil || other.Nonce != nil { + if a.Nonce == nil || other.Nonce == nil { + return false + } + + if *a.Nonce != *other.Nonce { + return false + } + } + + if a.StorageWrites != nil || other.StorageWrites != nil { + if !maps.Equal(a.StorageWrites, other.StorageWrites) { + return false + } + } + return true +} + +type BALExecutionMode int + +const ( + BALExecutionOptimized BALExecutionMode = iota + BALExecutionNoBatchIO + BALExecutionSequential +) + +// WrittenCounts groups per-block aggregate write counts derived from the BAL. +type WrittenCounts struct { + Accounts int + StorageSlots int + Codes int + CodeBytes int +} + +// WrittenCounts walks the BAL once and returns the aggregate write counts. +func (e BlockAccessList) WrittenCounts() WrittenCounts { + var w WrittenCounts + for i := range e { + a := &e[i] + if len(a.StorageChanges) > 0 || len(a.BalanceChanges) > 0 || + len(a.NonceChanges) > 0 || len(a.CodeChanges) > 0 { + w.Accounts++ + } + w.StorageSlots += len(a.StorageChanges) + if n := len(a.CodeChanges); n > 0 { + w.Codes++ + w.CodeBytes += len(a.CodeChanges[n-1].NewCode) + } + } + return w +} + +type StateMutations map[common.Address]AccountMutations + +type StorageKeySet map[common.Hash]struct{} +type StateAccesses map[common.Address]StorageKeySet + +func (s StateAccesses) Eq(other StateAccesses) bool { + if len(s) != len(other) { + return false + } + + for addr, set := range s { + otherSet, ok := other[addr] + if !ok { + return false + } + if !maps.Equal(set, otherSet) { + return false + } + } + return true +} diff --git a/core/types/bal/bal_encoding.go b/core/types/bal/bal_encoding.go index db891b92d0..4822652332 100644 --- a/core/types/bal/bal_encoding.go +++ b/core/types/bal/bal_encoding.go @@ -400,7 +400,7 @@ func (a *ConstructionAccountAccess) toEncodingObj(addr common.Address) AccountAc obj.SlotChanges = make([]encodingStorageWrite, 0, len(slotWrites)) indices := slices.Collect(maps.Keys(slotWrites)) - slices.SortFunc(indices, cmp.Compare) + slices.Sort(indices) for _, index := range indices { val := slotWrites[index] obj.SlotChanges = append(obj.SlotChanges, encodingStorageWrite{ @@ -420,7 +420,7 @@ func (a *ConstructionAccountAccess) toEncodingObj(addr common.Address) AccountAc // Convert balance changes balanceIndices := slices.Collect(maps.Keys(a.BalanceChanges)) - slices.SortFunc(balanceIndices, cmp.Compare) + slices.Sort(balanceIndices) for _, idx := range balanceIndices { res.BalanceChanges = append(res.BalanceChanges, encodingBalanceChange{ BlockAccessIndex: idx, @@ -430,7 +430,7 @@ func (a *ConstructionAccountAccess) toEncodingObj(addr common.Address) AccountAc // Convert nonce changes nonceIndices := slices.Collect(maps.Keys(a.NonceChanges)) - slices.SortFunc(nonceIndices, cmp.Compare) + slices.Sort(nonceIndices) for _, idx := range nonceIndices { res.NonceChanges = append(res.NonceChanges, encodingAccountNonce{ BlockAccessIndex: idx, @@ -440,7 +440,7 @@ func (a *ConstructionAccountAccess) toEncodingObj(addr common.Address) AccountAc // Convert code change codeIndices := slices.Collect(maps.Keys(a.CodeChange)) - slices.SortFunc(codeIndices, cmp.Compare) + slices.Sort(codeIndices) for _, idx := range codeIndices { res.CodeChanges = append(res.CodeChanges, encodingCodeChange{ BlockAccessIndex: idx, diff --git a/core/types/bal/bal_reader.go b/core/types/bal/bal_reader.go new file mode 100644 index 0000000000..a08df17797 --- /dev/null +++ b/core/types/bal/bal_reader.go @@ -0,0 +1,193 @@ +package bal + +import ( + "sort" + + "github.com/ethereum/go-ethereum/common" + "github.com/holiman/uint256" +) + +// AccessListReader enables efficient state diff lookups from a block access +// list during block execution. +type AccessListReader struct { + accounts map[common.Address]*preparedAccount +} + +type preparedAccount struct { + storage map[common.Hash]preparedSlot + AccountAccess +} + +type preparedSlot struct { + changes []encodingStorageWrite // borrowed, sorted asc by BlockAccessIndex +} + +// NewAccessListReader instantiates an access list reader. +func NewAccessListReader(list BlockAccessList) *AccessListReader { + accounts := make(map[common.Address]*preparedAccount, len(list)) + for i := range list { + a := list[i] // index; do not range-copy the AccountAccess + pa := &preparedAccount{ + AccountAccess: a, + } + if len(a.StorageChanges) > 0 { + pa.storage = make(map[common.Hash]preparedSlot, len(a.StorageChanges)) + for j := range a.StorageChanges { + sc := &a.StorageChanges[j] + pa.storage[sc.Slot.Bytes32()] = preparedSlot{changes: sc.SlotChanges} + } + } + accounts[a.Address] = pa + } + return &AccessListReader{accounts: accounts} +} + +// lastBefore returns the position of the last element in a slice of n elements +// sorted ascending by BlockAccessIndex whose key is strictly less than idx, or +// -1 if no such element exists. keyAt returns the BlockAccessIndex at position k. +func lastBefore(n int, idx uint32, keyAt func(k int) uint32) int { + // sort.Search returns the smallest position whose key is >= idx; everything + // before it is strictly less than idx, so the answer is that position - 1. + return sort.Search(n, func(k int) bool { return keyAt(k) >= idx }) - 1 +} + +// Balance returns the post-balance in effect immediately before the given block +// access index, or nil if the account's balance was not changed before idx. +// The returned pointer aliases the access list and must not be mutated. +func (p *AccessListReader) Balance(addr common.Address, idx int) *uint256.Int { + a := p.accounts[addr] + if a == nil { + return nil + } + k := lastBefore(len(a.BalanceChanges), uint32(idx), func(i int) uint32 { return a.BalanceChanges[i].BlockAccessIndex }) + if k < 0 { + return nil + } + return a.BalanceChanges[k].PostBalance +} + +// Nonce returns the post-nonce in effect immediately before the given block +// access index. The boolean is false if the nonce was not changed before idx. +func (p *AccessListReader) Nonce(addr common.Address, idx int) (uint64, bool) { + a := p.accounts[addr] + if a == nil { + return 0, false + } + k := lastBefore(len(a.NonceChanges), uint32(idx), func(i int) uint32 { return a.NonceChanges[i].BlockAccessIndex }) + if k < 0 { + return 0, false + } + return a.NonceChanges[k].PostNonce, true +} + +// Code returns the contract code in effect immediately before the given block +// access index, or nil if the code was not changed before idx. The returned +// slice aliases the access list and must not be mutated. +func (p *AccessListReader) Code(addr common.Address, idx int) []byte { + a := p.accounts[addr] + if a == nil { + return nil + } + k := lastBefore(len(a.CodeChanges), uint32(idx), func(i int) uint32 { return a.CodeChanges[i].BlockAccessIndex }) + if k < 0 { + return nil + } + return a.CodeChanges[k].NewCode +} + +// StorageAt returns the post-value of a storage slot immediately before the +// given block access index. The boolean is false if the slot was not written +// before idx. +func (p *AccessListReader) StorageAt(addr common.Address, slot common.Hash, idx int) (common.Hash, bool) { + a := p.accounts[addr] + if a == nil { + return common.Hash{}, false + } + s, ok := a.storage[slot] + if !ok { + return common.Hash{}, false + } + k := lastBefore(len(s.changes), uint32(idx), func(i int) uint32 { return s.changes[i].BlockAccessIndex }) + if k < 0 { + return common.Hash{}, false + } + return s.changes[k].PostValue.Bytes32(), true +} + +// AccountMutations returns the aggregate mutation for an account up until (and +// not including) the given block access list index, or nil if the account was +// not mutated before idx. +func (p *AccessListReader) AccountMutations(addr common.Address, idx int) *AccountMutations { + a := p.accounts[addr] + if a == nil { + return nil + } + res := &AccountMutations{} + if bal := p.Balance(addr, idx); bal != nil { + res.Balance = bal.Clone() + } + if code := p.Code(addr, idx); code != nil { + res.Code = code + } + if nonce, ok := p.Nonce(addr, idx); ok { + res.Nonce = new(uint64) + *res.Nonce = nonce + } + for slot, s := range a.storage { + k := lastBefore(len(s.changes), uint32(idx), func(i int) uint32 { return s.changes[i].BlockAccessIndex }) + if k < 0 { + continue + } + if res.StorageWrites == nil { + res.StorageWrites = make(map[common.Hash]common.Hash) + } + res.StorageWrites[slot] = s.changes[k].PostValue.Bytes32() + } + if res.Code == nil && res.Nonce == nil && len(res.StorageWrites) == 0 && res.Balance == nil { + return nil + } + return res +} + +type StorageKeys map[common.Address][]common.Hash + +// StorageKeys returns the set of accounts and storage keys mutated in the access +// list. If reads is set, the un-mutated accounts/keys are included in the result. +func (p *AccessListReader) StorageKeys() (keys StorageKeys) { + keys = make(StorageKeys) + for addr, a := range p.accounts { + for _, storageChange := range a.StorageChanges { + keys[addr] = append(keys[addr], storageChange.Slot.Bytes32()) + } + for _, storageRead := range a.StorageReads { + keys[addr] = append(keys[addr], storageRead.Bytes32()) + } + } + return +} + +// Mutations returns the aggregate state mutations from bal indices [0, idx). +func (p *AccessListReader) Mutations(idx int) *StateMutations { + res := make(StateMutations) + for addr := range p.accounts { + if mut := p.AccountMutations(addr, idx); mut != nil { + res[addr] = *mut + } + } + return &res +} + +// AllDestructions returns all accounts that experienced a destruction, regardless +// of whether they were later resurrected and exist after the block. It excludes +// ephemeral contracts from the result. +func (p *AccessListReader) AllDestructions() (res []common.Address) { + for addr, a := range p.accounts { + for _, nonce := range a.NonceChanges { + if nonce.PostNonce == 0 { + res = append(res, addr) + break + } + } + } + return res +} diff --git a/core/vm/evm.go b/core/vm/evm.go index 15609a0205..0c596489aa 100644 --- a/core/vm/evm.go +++ b/core/vm/evm.go @@ -261,6 +261,12 @@ func (evm *EVM) Call(caller common.Address, addr common.Address, input []byte, g } syscall := isSystemCall(caller) + // EIP-7928: per the Amsterdam spec, delegation resolution happens before + // the value-transfer check, so the delegated-to must appear in the BAL + // even when the call later reverts with ErrInsufficientBalance. Touch the + // target's code here (a no-op for non-delegated accounts) to record it. + evm.resolveCode(addr) + // Fail if we're trying to transfer more than the available balance. if !syscall && !value.IsZero() && !evm.Context.CanTransfer(evm.StateDB, caller, value) { return nil, gas, ErrInsufficientBalance @@ -346,6 +352,12 @@ func (evm *EVM) CallCode(caller common.Address, addr common.Address, input []byt if evm.depth > int(params.CallCreateDepth) { return nil, gas, ErrDepth } + + // EIP-7928: per the Amsterdam spec, delegation resolution happens before + // the value-transfer check, so the delegated-to must appear in the BAL + // even when the call later reverts with ErrInsufficientBalance. + evm.resolveCode(addr) + // Fail if we're trying to transfer more than the available balance if !evm.Context.CanTransfer(evm.StateDB, caller, value) { return nil, gas, ErrInsufficientBalance diff --git a/eth/catalyst/api.go b/eth/catalyst/api.go index 4b0c3f6ea3..54e5e45c7a 100644 --- a/eth/catalyst/api.go +++ b/eth/catalyst/api.go @@ -804,10 +804,12 @@ func (api *ConsensusAPI) NewPayloadV5(ctx context.Context, params engine.Executa return invalidStatus, paramsErr("nil beaconRoot post-cancun") case executionRequests == nil: return invalidStatus, paramsErr("nil executionRequests post-prague") - case params.SlotNumber == nil: - return invalidStatus, paramsErr("nil slotnumber post-amsterdam") case !api.checkFork(params.Timestamp, forks.Amsterdam): return invalidStatus, unsupportedForkErr("newPayloadV5 must only be called for amsterdam payloads") + case params.SlotNumber == nil: + return invalidStatus, paramsErr("nil slotnumber post-amsterdam") + case params.BlockAccessList == nil: + return invalidStatus, paramsErr("nil block access list post-amsterdam") } requests := convertRequests(executionRequests) if err := validateRequests(requests); err != nil { diff --git a/tests/block_test.go b/tests/block_test.go index 8c1a7ace0f..dd85b4b3e3 100644 --- a/tests/block_test.go +++ b/tests/block_test.go @@ -82,8 +82,17 @@ func TestBlockchain(t *testing.T) { // TestExecutionSpecBlocktests runs the test fixtures from execution-spec-tests. func TestExecutionSpecBlocktests(t *testing.T) { - if !common.FileExist(executionSpecBlockchainTestDir) { - t.Skipf("directory %s does not exist", executionSpecBlockchainTestDir) + testExecutionSpecBlocktests(t, executionSpecBlockchainTestDir) +} + +// TestExecutionSpecBlocktestsBAL runs the BAL release test fixtures from execution-spec-tests. +func TestExecutionSpecBlocktestsBAL(t *testing.T) { + testExecutionSpecBlocktests(t, executionSpecBALBlockchainTestDir) +} + +func testExecutionSpecBlocktests(t *testing.T, testDir string) { + if !common.FileExist(testDir) { + t.Skipf("directory %s does not exist", testDir) } bt := new(testMatcher) @@ -96,7 +105,7 @@ func TestExecutionSpecBlocktests(t *testing.T) { bt.skipLoad(`.*eip7610_create_collision/revert_in_create/.*`) bt.skipLoad(`.*stRandom2/random_statetest642/.*`) - bt.walk(t, executionSpecBlockchainTestDir, func(t *testing.T, name string, test *BlockTest) { + bt.walk(t, testDir, func(t *testing.T, name string, test *BlockTest) { execBlockTest(t, bt, test) }) } @@ -117,7 +126,7 @@ func execBlockTest(t *testing.T, bt *testMatcher, test *BlockTest) { } for _, snapshot := range snapshotConf { for _, dbscheme := range dbschemeConf { - if err := bt.checkFailure(t, test.Run(snapshot, dbscheme, true, nil, nil)); err != nil { + if err := bt.checkFailure(t, test.Run(snapshot, dbscheme, true, true, nil, nil)); err != nil { t.Errorf("test with config {snapshotter:%v, scheme:%v} failed: %v", snapshot, dbscheme, err) return } diff --git a/tests/block_test_util.go b/tests/block_test_util.go index bece8ae610..a302504821 100644 --- a/tests/block_test_util.go +++ b/tests/block_test_util.go @@ -113,27 +113,20 @@ type btHeaderMarshaling struct { SlotNumber *math.HexOrDecimal64 } -func (t *BlockTest) Run(snapshotter bool, scheme string, witness bool, tracer *tracing.Hooks, postCheck func(error, *core.BlockChain)) (result error) { - config, ok := Forks[t.json.Network] - if !ok { - return UnsupportedForkError{t.json.Network} - } - +func (t *BlockTest) createTestBlockChain(config *params.ChainConfig, snapshotter bool, scheme string, witness, createAndVerifyBAL bool, tracer *tracing.Hooks) (*core.BlockChain, error) { // import pre accounts & construct test genesis block & state root - // Commit genesis state var ( - gspec = t.genesis(config) db = rawdb.NewMemoryDatabase() tconf = &triedb.Config{ Preimages: true, - IsUBT: gspec.Config.UBTTime != nil && *gspec.Config.UBTTime <= gspec.Timestamp, } ) - if scheme == rawdb.PathScheme || tconf.IsUBT { + if scheme == rawdb.PathScheme { tconf.PathDB = pathdb.Defaults } else { tconf.HashDB = hashdb.Defaults } + gspec := t.genesis(config) // if ttd is not specified, set an arbitrary huge value if gspec.Config.TerminalTotalDifficulty == nil { @@ -142,15 +135,15 @@ func (t *BlockTest) Run(snapshotter bool, scheme string, witness bool, tracer *t triedb := triedb.NewDatabase(db, tconf) gblock, err := gspec.Commit(db, triedb, nil) if err != nil { - return err + return nil, err } triedb.Close() // close the db to prevent memory leak if gblock.Hash() != t.json.Genesis.Hash { - return fmt.Errorf("genesis block hash doesn't match test: computed=%x, test=%x", gblock.Hash().Bytes()[:6], t.json.Genesis.Hash[:6]) + return nil, fmt.Errorf("genesis block hash doesn't match test: computed=%x, test=%x", gblock.Hash().Bytes()[:6], t.json.Genesis.Hash[:6]) } if gblock.Root() != t.json.Genesis.StateRoot { - return fmt.Errorf("genesis block state root does not match test: computed=%x, test=%x", gblock.Root().Bytes()[:6], t.json.Genesis.StateRoot[:6]) + return nil, fmt.Errorf("genesis block state root does not match test: computed=%x, test=%x", gblock.Root().Bytes()[:6], t.json.Genesis.StateRoot[:6]) } // Wrap the original engine within the beacon-engine engine := beacon.New(ethash.NewFaker()) @@ -164,12 +157,26 @@ func (t *BlockTest) Run(snapshotter bool, scheme string, witness bool, tracer *t Tracer: tracer, }, StatelessSelfValidation: witness, + NoPrefetch: true, } if snapshotter { options.SnapshotLimit = 1 options.SnapshotWait = true } chain, err := core.NewBlockChain(db, gspec, engine, options) + if err != nil { + return nil, err + } + return chain, nil +} + +func (t *BlockTest) Run(snapshotter bool, scheme string, witness, createAndVerifyBAL bool, tracer *tracing.Hooks, postCheck func(error, *core.BlockChain)) (result error) { + config, ok := Forks[t.json.Network] + if !ok { + return UnsupportedForkError{t.json.Network} + } + + chain, err := t.createTestBlockChain(config, snapshotter, scheme, witness, createAndVerifyBAL, tracer) if err != nil { return err } @@ -203,7 +210,50 @@ func (t *BlockTest) Run(snapshotter bool, scheme string, witness bool, tracer *t } } } - return t.validateImportedHeaders(chain, validBlocks) + err = t.validateImportedHeaders(chain, validBlocks) + if err != nil { + return err + } + + if createAndVerifyBAL { + newChain, _ := t.createTestBlockChain(config, snapshotter, scheme, witness, createAndVerifyBAL, tracer) + defer newChain.Stop() + + var blocksWithBAL types.Blocks + for i := uint64(1); i <= chain.CurrentBlock().Number.Uint64(); i++ { + block := chain.GetBlockByNumber(i) + if chain.Config().IsAmsterdam(block.Number(), block.Time()) && block.AccessList() == nil { + return fmt.Errorf("block %d missing BAL", block.NumberU64()) + } + blocksWithBAL = append(blocksWithBAL, block) + } + + amt, err := newChain.InsertChain(blocksWithBAL) + if err != nil { + return err + } + _ = amt + newDB, err := newChain.State() + if err != nil { + return err + } + if err = t.validatePostState(newDB); err != nil { + return fmt.Errorf("post state validation failed: %v", err) + } + // Cross-check the snapshot-to-hash against the trie hash + if snapshotter { + if newChain.Snapshots() != nil { + if err := chain.Snapshots().Verify(chain.CurrentBlock().Root); err != nil { + return err + } + } + } + err = t.validateImportedHeaders(newChain, validBlocks) + if err != nil { + return err + } + } + return nil } // Network returns the network/fork name for this test.