# Nimbus # Copyright (c) 2022-2024 Status Research & Development GmbH # Licensed and distributed under either of # * MIT license (license terms in the root directory or at https://opensource.org/licenses/MIT). # * Apache v2 license (license terms in the root directory or at https://www.apache.org/licenses/LICENSE-2.0). # at your option. This file may not be copied, modified, or distributed except according to those terms. {.push raises: [].} import eth/rlp, eth/common/eth_types_rlp, ssz_serialization, ssz_serialization/[proofs, merkleization], ../../common/common_types, ./history_content export ssz_serialization, merkleization, proofs, eth_types_rlp # Header Accumulator, as per specification: # https://github.com/ethereum/portal-network-specs/blob/master/history-network.md#the-header-accumulator # But with the adjustment to finish the accumulator at merge point. const epochSize* = 8192 # blocks # Allow this to be adjusted at compile time for testing. If more constants # need to be adjusted we can add some presets file. mergeBlockNumber* {.intdefine.}: uint64 = 15537394 # Note: This is like a ceil(mergeBlockNumber / epochSize) # Could use ceilDiv(mergeBlockNumber, epochSize) in future versions preMergeEpochs* = (mergeBlockNumber + epochSize - 1) div epochSize # TODO: # Currently disabled, because issue when testing with other # `mergeBlockNumber`, but it could be used as value to double check on at # merge block. # TODO: Could also be used as value to actual finish the accumulator, instead # of `mergeBlockNumber`, but: # - Still need to store the actual `mergeBlockNumber` and run-time somewhere # as it allows for each pre vs post merge block header checking. # - Can't limit `historicalEpochs` SSZ list at `preMergeEpochs` value. # - Should probably be stated in the portal network specs. # TERMINAL_TOTAL_DIFFICULTY = u256"58750000000000000000000" type HeaderRecord* = object blockHash*: BlockHash totalDifficulty*: UInt256 EpochAccumulator* = List[HeaderRecord, epochSize] # In the core code of Fluffy the `EpochAccumulator` type is solely used, as # `hash_tree_root` is done either once or never on this object after # serialization. # However for the generation of the proofs for all the headers in an epoch, it # needs to be run many times and the cached version of the SSZ list is # obviously much faster, so this second type is added for this usage. EpochAccumulatorCached* = HashList[HeaderRecord, epochSize] Accumulator* = object historicalEpochs*: List[Bytes32, int(preMergeEpochs)] currentEpoch*: EpochAccumulator FinishedAccumulator* = object historicalEpochs*: List[Bytes32, int(preMergeEpochs)] BlockEpochData* = object epochHash*: Bytes32 blockRelativeIndex*: uint64 func init*(T: type Accumulator): T = Accumulator( historicalEpochs: List[Bytes32, int(preMergeEpochs)].init(@[]), currentEpoch: EpochAccumulator.init(@[]), ) func getEpochAccumulatorRoot*(headerRecords: openArray[HeaderRecord]): Digest = let epochAccumulator = EpochAccumulator.init(@headerRecords) hash_tree_root(epochAccumulator) func updateAccumulator*(a: var Accumulator, header: BlockHeader) = doAssert( header.blockNumber.truncate(uint64) < mergeBlockNumber, "No post merge blocks for header accumulator", ) let lastTotalDifficulty = if a.currentEpoch.len() == 0: 0.stuint(256) else: a.currentEpoch[^1].totalDifficulty # TODO: It is a bit annoying to require an extra header + update call to # finish an epoch. However, if we were to move this after adding the # `HeaderRecord`, there would be no way to get the current total difficulty, # unless another field is introduced in the `Accumulator` object. if a.currentEpoch.len() == epochSize: let epochHash = hash_tree_root(a.currentEpoch) doAssert(a.historicalEpochs.add(epochHash.data)) a.currentEpoch = EpochAccumulator.init(@[]) let headerRecord = HeaderRecord( blockHash: header.blockHash(), totalDifficulty: lastTotalDifficulty + header.difficulty, ) let res = a.currentEpoch.add(headerRecord) doAssert(res, "Can't fail because of currentEpoch length check") func finishAccumulator*(a: var Accumulator): FinishedAccumulator = # doAssert(a.currentEpoch[^2].totalDifficulty < TERMINAL_TOTAL_DIFFICULTY) # doAssert(a.currentEpoch[^1].totalDifficulty >= TERMINAL_TOTAL_DIFFICULTY) let epochHash = hash_tree_root(a.currentEpoch) doAssert(a.historicalEpochs.add(epochHash.data)) FinishedAccumulator(historicalEpochs: a.historicalEpochs) ## Calls and helper calls for building header proofs and verifying headers ## against the Accumulator and the header proofs. func getEpochIndex*(blockNumber: uint64): uint64 = blockNumber div epochSize func getEpochIndex*(header: BlockHeader): uint64 = let blockNumber = header.blockNumber.truncate(uint64) ## Get the index for the historical epochs getEpochIndex(blockNumber) func getHeaderRecordIndex*(blockNumber: uint64, epochIndex: uint64): uint64 = ## Get the relative header index for the epoch accumulator uint64(blockNumber - epochIndex * epochSize) func getHeaderRecordIndex*(header: BlockHeader, epochIndex: uint64): uint64 = ## Get the relative header index for the epoch accumulator getHeaderRecordIndex(header.blockNumber.truncate(uint64), epochIndex) func isPreMerge*(blockNumber: uint64): bool = blockNumber < mergeBlockNumber func isPreMerge*(header: BlockHeader): bool = isPreMerge(header.blockNumber.truncate(uint64)) func verifyProof( a: FinishedAccumulator, header: BlockHeader, proof: openArray[Digest] ): bool = let epochIndex = getEpochIndex(header) epochAccumulatorHash = Digest(data: a.historicalEpochs[epochIndex]) leave = hash_tree_root(header.blockHash()) headerRecordIndex = getHeaderRecordIndex(header, epochIndex) # TODO: Implement more generalized `get_generalized_index` gIndex = GeneralizedIndex(epochSize * 2 * 2 + (headerRecordIndex * 2)) verify_merkle_multiproof(@[leave], proof, @[gIndex], epochAccumulatorHash) func verifyAccumulatorProof*( a: FinishedAccumulator, header: BlockHeader, proof: AccumulatorProof ): Result[void, string] = if header.isPreMerge(): # Note: The proof is typed with correct depth, so no check on this is # required here. if a.verifyProof(header, proof): ok() else: err("Proof verification failed") else: err("Cannot verify post merge header with accumulator proof") func verifyHeader*( a: FinishedAccumulator, header: BlockHeader, proof: BlockHeaderProof ): Result[void, string] = case proof.proofType of BlockHeaderProofType.accumulatorProof: a.verifyAccumulatorProof(header, proof.accumulatorProof) of BlockHeaderProofType.none: if header.isPreMerge(): err("Pre merge header requires AccumulatorProof") else: # TODO: # Currently there is no proof solution for verifying headers post-merge. # Skipping canonical verification will allow for nodes to push block data # that is not part of the canonical chain. # For now we accept this flaw as the focus lies on testing data # availability up to the head of the chain. ok() func buildProof*( header: BlockHeader, epochAccumulator: EpochAccumulator | EpochAccumulatorCached ): Result[AccumulatorProof, string] = doAssert(header.isPreMerge(), "Must be pre merge header") let epochIndex = getEpochIndex(header) headerRecordIndex = getHeaderRecordIndex(header, epochIndex) # TODO: Implement more generalized `get_generalized_index` gIndex = GeneralizedIndex(epochSize * 2 * 2 + (headerRecordIndex * 2)) var proof: AccumulatorProof ?epochAccumulator.build_proof(gIndex, proof) ok(proof) func buildHeaderWithProof*( header: BlockHeader, epochAccumulator: EpochAccumulator | EpochAccumulatorCached ): Result[BlockHeaderWithProof, string] = let proof = ?buildProof(header, epochAccumulator) ok( BlockHeaderWithProof( header: ByteList.init(rlp.encode(header)), proof: BlockHeaderProof.init(proof) ) ) func getBlockEpochDataForBlockNumber*( a: FinishedAccumulator, bn: UInt256 ): Result[BlockEpochData, string] = let blockNumber = bn.truncate(uint64) if blockNumber.isPreMerge: let epochIndex = getEpochIndex(blockNumber) ok( BlockEpochData( epochHash: a.historicalEpochs[epochIndex], blockRelativeIndex: getHeaderRecordIndex(blockNumber, epochIndex), ) ) else: err("Block number is post merge: " & $blockNumber)