114 lines
3.8 KiB
Nim
114 lines
3.8 KiB
Nim
# Nimbus - Portal Network
|
|
# Copyright (c) 2021-2024 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
|
|
std/net,
|
|
eth/[common, keys, rlp],
|
|
eth/p2p/discoveryv5/[enr, node, routing_table],
|
|
eth/p2p/discoveryv5/protocol as discv5_protocol,
|
|
../network/history/[accumulator, history_content],
|
|
../database/content_db
|
|
|
|
proc localAddress*(port: int): Address {.raises: [ValueError].} =
|
|
Address(ip: parseIpAddress("127.0.0.1"), port: Port(port))
|
|
|
|
proc initDiscoveryNode*(
|
|
rng: ref HmacDrbgContext,
|
|
privKey: PrivateKey,
|
|
address: Address,
|
|
bootstrapRecords: openArray[Record] = [],
|
|
localEnrFields: openArray[(string, seq[byte])] = [],
|
|
previousRecord = none[enr.Record](),
|
|
): discv5_protocol.Protocol {.raises: [CatchableError].} =
|
|
# set bucketIpLimit to allow bucket split
|
|
let config = DiscoveryConfig.init(1000, 24, 5)
|
|
|
|
result = newProtocol(
|
|
privKey,
|
|
some(address.ip),
|
|
some(address.port),
|
|
some(address.port),
|
|
bindPort = address.port,
|
|
bootstrapRecords = bootstrapRecords,
|
|
localEnrFields = localEnrFields,
|
|
previousRecord = previousRecord,
|
|
config = config,
|
|
rng = rng,
|
|
)
|
|
|
|
result.open()
|
|
|
|
proc genByteSeq*(length: int): seq[byte] =
|
|
var i = 0
|
|
var resultSeq = newSeq[byte](length)
|
|
while i < length:
|
|
resultSeq[i] = byte(i)
|
|
inc i
|
|
return resultSeq
|
|
|
|
func buildAccumulator*(headers: seq[BlockHeader]): Result[FinishedAccumulator, string] =
|
|
var accumulator: Accumulator
|
|
for header in headers:
|
|
updateAccumulator(accumulator, header)
|
|
|
|
if header.number == mergeBlockNumber - 1:
|
|
return ok(finishAccumulator(accumulator))
|
|
|
|
err("Not enough headers provided to finish the accumulator")
|
|
|
|
func buildAccumulatorData*(
|
|
headers: seq[BlockHeader]
|
|
): Result[(FinishedAccumulator, seq[EpochAccumulator]), string] =
|
|
var accumulator: Accumulator
|
|
var epochAccumulators: seq[EpochAccumulator]
|
|
for header in headers:
|
|
updateAccumulator(accumulator, header)
|
|
|
|
if accumulator.currentEpoch.len() == epochSize:
|
|
epochAccumulators.add(accumulator.currentEpoch)
|
|
|
|
if header.number == mergeBlockNumber - 1:
|
|
epochAccumulators.add(accumulator.currentEpoch)
|
|
|
|
return ok((finishAccumulator(accumulator), epochAccumulators))
|
|
|
|
err("Not enough headers provided to finish the accumulator")
|
|
|
|
func buildProof*(
|
|
header: BlockHeader, epochAccumulators: seq[EpochAccumulator]
|
|
): Result[AccumulatorProof, string] =
|
|
let epochIndex = getEpochIndex(header)
|
|
doAssert(epochIndex < uint64(epochAccumulators.len()))
|
|
let epochAccumulator = epochAccumulators[epochIndex]
|
|
|
|
buildProof(header, epochAccumulator)
|
|
|
|
func buildHeaderWithProof*(
|
|
header: BlockHeader, epochAccumulators: seq[EpochAccumulator]
|
|
): Result[BlockHeaderWithProof, string] =
|
|
## Construct the accumulator proof for a specific header.
|
|
## Returns the block header with the proof
|
|
if header.isPreMerge():
|
|
let epochIndex = getEpochIndex(header)
|
|
doAssert(epochIndex < uint64(epochAccumulators.len()))
|
|
let epochAccumulator = epochAccumulators[epochIndex]
|
|
|
|
buildHeaderWithProof(header, epochAccumulator)
|
|
else:
|
|
err("Cannot build accumulator proof for post merge header")
|
|
|
|
func buildHeadersWithProof*(
|
|
headers: seq[BlockHeader], epochAccumulators: seq[EpochAccumulator]
|
|
): Result[seq[BlockHeaderWithProof], string] =
|
|
var headersWithProof: seq[BlockHeaderWithProof]
|
|
for header in headers:
|
|
headersWithProof.add(?buildHeaderWithProof(header, epochAccumulators))
|
|
|
|
ok(headersWithProof)
|