# beacon_chain # Copyright (c) 2018-2021 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: [Defect].} import std/tables, stew/results, chronicles, chronos, metrics, ../spec/[helpers, forks], ../spec/datatypes/[altair, phase0], ../consensus_object_pools/[ block_clearance, blockchain_dag, exit_pool, attestation_pool, sync_committee_msg_pool], ../validators/validator_pool, ../beacon_clock, "."/[gossip_validation, block_processor, batch_validation] export results, block_clearance, blockchain_dag, exit_pool, attestation_pool, sync_committee_msg_pool, validator_pool, beacon_clock, gossip_validation, block_processor, batch_validation # Metrics for tracking attestation and beacon block loss declareCounter beacon_attestations_received, "Number of valid unaggregated attestations processed by this node" declareCounter beacon_attestations_dropped, "Number of invalid unaggregated attestations dropped by this node", labels = ["reason"] declareCounter beacon_aggregates_received, "Number of valid aggregated attestations processed by this node" declareCounter beacon_aggregates_dropped, "Number of invalid aggregated attestations dropped by this node", labels = ["reason"] declareCounter beacon_blocks_received, "Number of valid blocks processed by this node" declareCounter beacon_blocks_dropped, "Number of invalid blocks dropped by this node", labels = ["reason"] declareCounter beacon_attester_slashings_received, "Number of valid attester slashings processed by this node" declareCounter beacon_attester_slashings_dropped, "Number of invalid attester slashings dropped by this node", labels = ["reason"] declareCounter beacon_proposer_slashings_received, "Number of valid proposer slashings processed by this node" declareCounter beacon_proposer_slashings_dropped, "Number of invalid proposer slashings dropped by this node", labels = ["reason"] declareCounter beacon_voluntary_exits_received, "Number of valid voluntary exits processed by this node" declareCounter beacon_voluntary_exits_dropped, "Number of invalid voluntary exits dropped by this node", labels = ["reason"] declareCounter beacon_sync_committee_messages_received, "Number of valid sync committee messages processed by this node" declareCounter beacon_sync_committee_messages_dropped, "Number of invalid sync committee messages dropped by this node", labels = ["reason"] declareCounter beacon_sync_committee_contributions_received, "Number of valid sync committee contributions processed by this node" declareCounter beacon_sync_committee_contributions_dropped, "Number of invalid sync committee contributions dropped by this node", labels = ["reason"] const delayBuckets = [2.0, 4.0, 6.0, 8.0, 10.0, 12.0, 14.0, Inf] declareHistogram beacon_attestation_delay, "Time(s) between slot start and attestation reception", buckets = delayBuckets declareHistogram beacon_aggregate_delay, "Time(s) between slot start and aggregate reception", buckets = delayBuckets declareHistogram beacon_block_delay, "Time(s) between slot start and beacon block reception", buckets = delayBuckets type DoppelgangerProtection = object broadcastStartEpoch*: Epoch ##\ ## Set anew, each time gossip is re-enabled after syncing completes, so ## might reset multiple times per instance. This allows some safe level ## of gossip interleaving between nodes so long as they don't gossip at ## the same time. nodeLaunchSlot: Slot ##\ ## Set once, at node launch. This functions as a basic protection against ## false positives from attestations persisting within the gossip network ## across quick restarts. Eth2Processor* = object ## The Eth2Processor is the entry point for untrusted message processing - ## when we receive messages from various sources, we pass them to the ## processor for validation and routing - the messages are generally ## validated, and if valid, passed on to the various pools, monitors and ## managers to update the state of the application. ## ## Block processing is special in that part of it is done in the ## `BlockProcessor` instead, via a special block processing queue. ## ## Each validating function generally will do a sanity check on the message ## whose purpose is to quickly filter out spam, then will (usually) delegate ## full validation to the proper manager - finally, metrics and monitoring ## are updated. doppelGangerDetectionEnabled*: bool # Local sources of truth for validation # ---------------------------------------------------------------- dag*: ChainDAGRef attestationPool*: ref AttestationPool validatorPool: ref ValidatorPool syncCommitteeMsgPool: ref SyncCommitteeMsgPool doppelgangerDetection*: DoppelgangerProtection # Gossip validated -> enqueue for further verification # ---------------------------------------------------------------- blockProcessor: ref BlockProcessor # Validated with no further verification required # ---------------------------------------------------------------- exitPool: ref ExitPool # Almost validated, pending cryptographic signature check # ---------------------------------------------------------------- batchCrypto*: ref BatchCrypto # Missing information # ---------------------------------------------------------------- quarantine*: QuarantineRef # Application-provided current time provider (to facilitate testing) getCurrentBeaconTime*: GetBeaconTimeFn ValidationRes* = Result[void, ValidationError] # Initialization # ------------------------------------------------------------------------------ proc new*(T: type Eth2Processor, doppelGangerDetectionEnabled: bool, blockProcessor: ref BlockProcessor, dag: ChainDAGRef, attestationPool: ref AttestationPool, exitPool: ref ExitPool, validatorPool: ref ValidatorPool, syncCommitteeMsgPool: ref SyncCommitteeMsgPool, quarantine: QuarantineRef, rng: ref BrHmacDrbgContext, getBeaconTime: GetBeaconTimeFn, taskpool: batch_validation.TaskPoolPtr ): ref Eth2Processor = (ref Eth2Processor)( doppelGangerDetectionEnabled: doppelGangerDetectionEnabled, doppelgangerDetection: DoppelgangerProtection( nodeLaunchSlot: getBeaconTime().slotOrZero), blockProcessor: blockProcessor, dag: dag, attestationPool: attestationPool, exitPool: exitPool, validatorPool: validatorPool, syncCommitteeMsgPool: syncCommitteeMsgPool, quarantine: quarantine, getCurrentBeaconTime: getBeaconTime, batchCrypto: BatchCrypto.new( rng = rng, # Only run eager attestation signature verification if we're not # processing blocks in order to give priority to block processing eager = proc(): bool = not blockProcessor[].hasBlocks(), taskpool) ) # Each validator logs, validates then passes valid data to its destination # further down the line - in particular, validation should generally not have # any side effects until the message is fully validated, or invalid messages # could be used to push out valid messages. proc blockValidator*( self: var Eth2Processor, signedBlock: phase0.SignedBeaconBlock | altair.SignedBeaconBlock): ValidationRes = let wallTime = self.getCurrentBeaconTime() (afterGenesis, wallSlot) = wallTime.toSlot() logScope: blockRoot = shortLog(signedBlock.root) blck = shortLog(signedBlock.message) signature = shortLog(signedBlock.signature) wallSlot if not afterGenesis: notice "Block before genesis" return errIgnore("Block before genesis") # Potential under/overflows are fine; would just create odd metrics and logs let delay = wallTime - signedBlock.message.slot.toBeaconTime # Start of block processing - in reality, we have already gone through SSZ # decoding at this stage, which may be significant debug "Block received", delay let v = self.dag.validateBeaconBlock(self.quarantine, signedBlock, wallTime, {}) if v.isOk(): # Block passed validation - enqueue it for processing. The block processing # queue is effectively unbounded as we use a freestanding task to enqueue # the block - this is done so that when blocks arrive concurrently with # sync, we don't lose the gossip blocks, but also don't block the gossip # propagation of seemingly good blocks trace "Block validated" self.blockProcessor[].addBlock( ForkedSignedBeaconBlock.init(signedBlock), validationDur = self.getCurrentBeaconTime() - wallTime) # Validator monitor registration for blocks is done by the processor beacon_blocks_received.inc() beacon_block_delay.observe(delay.toFloatSeconds()) else: debug "Dropping block", error = v.error() self.blockProcessor[].dumpInvalidBlock(signedBlock) beacon_blocks_dropped.inc(1, [$v.error[0]]) v proc checkForPotentialDoppelganger( self: var Eth2Processor, attestation: Attestation, attesterIndices: openArray[ValidatorIndex]) = # Only check for attestations after node launch. There might be one slot of # overlap in quick intra-slot restarts so trade off a few true negatives in # the service of avoiding more likely false positives. if attestation.data.slot <= self.doppelgangerDetection.nodeLaunchSlot + 1: return if attestation.data.slot.epoch < self.doppelgangerDetection.broadcastStartEpoch: let tgtBlck = self.dag.getRef(attestation.data.target.root) doAssert not tgtBlck.isNil # because attestation is valid above let epochRef = self.dag.getEpochRef( tgtBlck, attestation.data.target.epoch) for validatorIndex in attesterIndices: let validatorPubkey = epochRef.validatorKey(validatorIndex).get().toPubKey() if self.doppelgangerDetectionEnabled and self.validatorPool[].getValidator(validatorPubkey) != default(AttachedValidator): warn "We believe you are currently running another instance of the same validator. We've disconnected you from the network as this presents a significant slashing risk. Possible next steps are (a) making sure you've disconnected your validator from your old machine before restarting the client; and (b) running the client again with the gossip-slashing-protection option disabled, only if you are absolutely sure this is the only instance of your validator running, and reporting the issue at https://github.com/status-im/nimbus-eth2/issues.", validatorIndex, validatorPubkey, attestation = shortLog(attestation) quit QuitFailure proc attestationValidator*( self: ref Eth2Processor, attestation: Attestation, subnet_id: SubnetId, checkSignature: bool = true): Future[ValidationRes] {.async.} = let wallTime = self.getCurrentBeaconTime() var (afterGenesis, wallSlot) = wallTime.toSlot() logScope: attestation = shortLog(attestation) subnet_id wallSlot if not afterGenesis: notice "Attestation before genesis" return errIgnore("Attestation before genesis") # Potential under/overflows are fine; would just create odd metrics and logs let delay = wallTime - attestation.data.slot.toBeaconTime debug "Attestation received", delay # Now proceed to validation let v = await self.attestationPool.validateAttestation( self.batchCrypto, attestation, wallTime, subnet_id, checkSignature) return if v.isOk(): # Due to async validation the wallSlot here might have changed wallSlot = self.getCurrentBeaconTime().slotOrZero() let (attester_index, sig) = v.get() self[].checkForPotentialDoppelganger(attestation, [attester_index]) trace "Attestation validated" self.attestationPool[].addAttestation( attestation, [attester_index], sig, wallSlot) beacon_attestations_received.inc() beacon_attestation_delay.observe(delay.toFloatSeconds()) ok() else: debug "Dropping attestation", validationError = v.error beacon_attestations_dropped.inc(1, [$v.error[0]]) err(v.error()) proc aggregateValidator*( self: ref Eth2Processor, signedAggregateAndProof: SignedAggregateAndProof): Future[ValidationRes] {.async.} = let wallTime = self.getCurrentBeaconTime() var (afterGenesis, wallSlot) = wallTime.toSlot() logScope: aggregate = shortLog(signedAggregateAndProof.message.aggregate) aggregator_index = signedAggregateAndProof.message.aggregator_index selection_proof = shortLog(signedAggregateAndProof.message.selection_proof) signature = shortLog(signedAggregateAndProof.signature) wallSlot if not afterGenesis: notice "Aggregate before genesis" return errIgnore("Aggreagte before genesis") # Potential under/overflows are fine; would just create odd logs let delay = wallTime - signedAggregateAndProof.message.aggregate.data.slot.toBeaconTime debug "Aggregate received", delay let v = await self.attestationPool.validateAggregate( self.batchCrypto, signedAggregateAndProof, wallTime) return if v.isOk(): # Due to async validation the wallSlot here might have changed wallSlot = self.getCurrentBeaconTime().slotOrZero() let (attesting_indices, sig) = v.get() self[].checkForPotentialDoppelganger( signedAggregateAndProof.message.aggregate, attesting_indices) trace "Aggregate validated" self.attestationPool[].addAttestation( signedAggregateAndProof.message.aggregate, attesting_indices, sig, wallSlot) beacon_aggregates_received.inc() beacon_aggregate_delay.observe(delay.toFloatSeconds()) ok() else: debug "Dropping aggregate", error = v.error beacon_aggregates_dropped.inc(1, [$v.error[0]]) err(v.error()) proc attesterSlashingValidator*( self: var Eth2Processor, attesterSlashing: AttesterSlashing): ValidationRes = logScope: attesterSlashing = shortLog(attesterSlashing) debug "Attester slashing received" let v = self.exitPool[].validateAttesterSlashing(attesterSlashing) if v.isOk(): trace "Attester slashing validated" self.exitPool[].addMessage(attesterSlashing) beacon_attester_slashings_received.inc() else: debug "Dropping attester slashing", validationError = v.error beacon_attester_slashings_dropped.inc(1, [$v.error[0]]) v proc proposerSlashingValidator*( self: var Eth2Processor, proposerSlashing: ProposerSlashing): Result[void, ValidationError] = logScope: proposerSlashing = shortLog(proposerSlashing) debug "Proposer slashing received" let v = self.exitPool[].validateProposerSlashing(proposerSlashing) if v.isOk(): trace "Proposer slashing validated" beacon_proposer_slashings_received.inc() else: debug "Dropping proposer slashing", validationError = v.error self.exitPool[].addMessage(proposerSlashing) beacon_proposer_slashings_dropped.inc(1, [$v.error[0]]) v proc voluntaryExitValidator*( self: var Eth2Processor, signedVoluntaryExit: SignedVoluntaryExit): Result[void, ValidationError] = logScope: signedVoluntaryExit = shortLog(signedVoluntaryExit) debug "Voluntary exit received" let v = self.exitPool[].validateVoluntaryExit(signedVoluntaryExit) if v.isOk(): trace "Voluntary exit validated" self.exitPool[].addMessage(signedVoluntaryExit) beacon_voluntary_exits_received.inc() else: debug "Dropping voluntary exit", error = v.error beacon_voluntary_exits_dropped.inc(1, [$v.error[0]]) v proc syncCommitteeMessageValidator*( self: ref Eth2Processor, syncCommitteeMsg: SyncCommitteeMessage, subcommitteeIdx: SyncSubcommitteeIndex, checkSignature: bool = true): Result[void, ValidationError] = let wallTime = self.getCurrentBeaconTime() wallSlot = wallTime.slotOrZero() logScope: syncCommitteeMsg = shortLog(syncCommitteeMsg) subcommitteeIdx wallSlot # Potential under/overflows are fine; would just create odd metrics and logs let delay = wallTime - syncCommitteeMsg.slot.toBeaconTime debug "Sync committee message received", delay # Now proceed to validation let v = validateSyncCommitteeMessage(self.dag, self.syncCommitteeMsgPool[], syncCommitteeMsg, subcommitteeIdx, wallTime, checkSignature) return if v.isOk(): trace "Sync committee message validated" let (positions, cookedSig) = v.get() self.syncCommitteeMsgPool[].addSyncCommitteeMessage( syncCommitteeMsg.slot, syncCommitteeMsg.beacon_block_root, syncCommitteeMsg.validator_index, cookedSig, subcommitteeIdx, positions) beacon_sync_committee_messages_received.inc() ok() else: debug "Dropping sync committee message", error = v.error beacon_sync_committee_messages_dropped.inc(1, [$v.error[0]]) err(v.error()) proc contributionValidator*( self: ref Eth2Processor, contributionAndProof: SignedContributionAndProof, checkSignature: bool = true): Result[void, ValidationError] = let wallTime = self.getCurrentBeaconTime() wallSlot = wallTime.slotOrZero() logScope: contribution = shortLog(contributionAndProof.message.contribution) signature = shortLog(contributionAndProof.signature) aggregator_index = contributionAndProof.message.aggregator_index selection_proof = contributionAndProof.message.selection_proof wallSlot # Potential under/overflows are fine; would just create odd metrics and logs let delay = wallTime - contributionAndProof.message.contribution.slot.toBeaconTime debug "Contribution received", delay # Now proceed to validation let v = validateContribution( self.dag, self.syncCommitteeMsgPool[], contributionAndProof, wallTime, checkSignature) return if v.isOk(): trace "Contribution validated" self.syncCommitteeMsgPool[].addContribution(contributionAndProof, v.get) beacon_sync_committee_contributions_received.inc() ok() else: debug "Dropping contribution", error = v.error beacon_sync_committee_contributions_dropped.inc(1, [$v.error[0]]) err(v.error())