package consensus import ( "math/big" "sync/atomic" "time" "github.com/harmony-one/harmony/core/types" "github.com/harmony-one/harmony/crypto/bls" "github.com/ethereum/go-ethereum/common" protobuf "github.com/golang/protobuf/proto" bls_core "github.com/harmony-one/bls/ffi/go/bls" msg_pb "github.com/harmony-one/harmony/api/proto/message" consensus_engine "github.com/harmony-one/harmony/consensus/engine" "github.com/harmony-one/harmony/consensus/quorum" "github.com/harmony-one/harmony/consensus/signature" bls_cosi "github.com/harmony-one/harmony/crypto/bls" "github.com/harmony-one/harmony/crypto/hash" "github.com/harmony-one/harmony/internal/chain" "github.com/harmony-one/harmony/internal/utils" "github.com/harmony-one/harmony/shard" "github.com/harmony-one/harmony/shard/committee" "github.com/pkg/errors" "github.com/rs/zerolog" ) // WaitForNewRandomness listens to the RndChannel to receive new VDF randomness. func (consensus *Consensus) WaitForNewRandomness() { go func() { for { vdfOutput := <-consensus.RndChannel consensus.pendingRnds = append(consensus.pendingRnds, vdfOutput) } }() } // GetNextRnd returns the oldest available randomness along with the hash of the block there randomness preimage is committed. func (consensus *Consensus) GetNextRnd() ([vdFAndProofSize]byte, [32]byte, error) { if len(consensus.pendingRnds) == 0 { return [vdFAndProofSize]byte{}, [32]byte{}, errors.New("No available randomness") } vdfOutput := consensus.pendingRnds[0] vdfBytes := [vdFAndProofSize]byte{} seed := [32]byte{} copy(vdfBytes[:], vdfOutput[:vdFAndProofSize]) copy(seed[:], vdfOutput[vdFAndProofSize:]) //pop the first vdfOutput from the list consensus.pendingRnds = consensus.pendingRnds[1:] return vdfBytes, seed, nil } var ( empty = []byte{} ) // Signs the consensus message and returns the marshaled message. func (consensus *Consensus) signAndMarshalConsensusMessage(message *msg_pb.Message, priKey *bls_core.SecretKey) ([]byte, error) { if err := consensus.signConsensusMessage(message, priKey); err != nil { return empty, err } marshaledMessage, err := protobuf.Marshal(message) if err != nil { return empty, err } return marshaledMessage, nil } // UpdatePublicKeys updates the PublicKeys for // quorum on current subcommittee, protected by a mutex func (consensus *Consensus) UpdatePublicKeys(pubKeys, allowlist []bls_cosi.PublicKeyWrapper) int64 { consensus.Decider.UpdateParticipants(pubKeys, allowlist) consensus.getLogger().Info().Msg("My Committee updated") for i := range pubKeys { consensus.getLogger().Info(). Int("index", i). Str("BLSPubKey", pubKeys[i].Bytes.Hex()). Msg("Member") } allKeys := consensus.Decider.Participants() if len(allKeys) != 0 { consensus.LeaderPubKey = &allKeys[0] consensus.getLogger().Info(). Str("info", consensus.LeaderPubKey.Bytes.Hex()).Msg("My Leader") } else { consensus.getLogger().Error(). Msg("[UpdatePublicKeys] Participants is empty") } // reset states after update public keys // TODO: incorporate bitmaps in the decider, so their state can't be inconsistent. consensus.UpdateBitmaps() consensus.ResetState() // do not reset view change state if it is in view changing mode if !consensus.isViewChangingMode() { consensus.ResetViewChangeState() } return consensus.Decider.ParticipantsCount() } // Sign on the hash of the message func (consensus *Consensus) signMessage(message []byte, priKey *bls_core.SecretKey) []byte { hash := hash.Keccak256(message) signature := priKey.SignHash(hash[:]) return signature.Serialize() } // Sign on the consensus message signature field. func (consensus *Consensus) signConsensusMessage(message *msg_pb.Message, priKey *bls_core.SecretKey) error { message.Signature = nil marshaledMessage, err := protobuf.Marshal(message) if err != nil { return err } // 64 byte of signature on previous data signature := consensus.signMessage(marshaledMessage, priKey) message.Signature = signature return nil } // UpdateBitmaps update the bitmaps for prepare and commit phase func (consensus *Consensus) UpdateBitmaps() { consensus.getLogger().Debug(). Str("MessageType", consensus.phase.String()). Msg("[UpdateBitmaps] Updating consensus bitmaps") members := consensus.Decider.Participants() prepareBitmap, _ := bls_cosi.NewMask(members, nil) commitBitmap, _ := bls_cosi.NewMask(members, nil) multiSigBitmap, _ := bls_cosi.NewMask(members, nil) consensus.prepareBitmap = prepareBitmap consensus.commitBitmap = commitBitmap consensus.multiSigBitmap = multiSigBitmap } // ResetState resets the state of the consensus func (consensus *Consensus) ResetState() { consensus.switchPhase("ResetState", FBFTAnnounce) consensus.blockHash = [32]byte{} consensus.block = []byte{} consensus.Decider.ResetPrepareAndCommitVotes() if consensus.prepareBitmap != nil { consensus.prepareBitmap.Clear() } if consensus.commitBitmap != nil { consensus.commitBitmap.Clear() } consensus.aggregatedPrepareSig = nil consensus.aggregatedCommitSig = nil } // IsValidatorInCommittee returns whether the given validator BLS address is part of my committee func (consensus *Consensus) IsValidatorInCommittee(pubKey bls.SerializedPublicKey) bool { return consensus.Decider.IndexOf(pubKey) != -1 } // SetMode sets the mode of consensus func (consensus *Consensus) SetMode(m Mode) { if m == Normal && consensus.isBackup { m = NormalBackup } consensus.getLogger().Debug(). Str("Mode", m.String()). Msg("[SetMode]") consensus.current.SetMode(m) } // SetIsBackup sets the mode of consensus func (consensus *Consensus) SetIsBackup(isBackup bool) { consensus.getLogger().Debug(). Bool("IsBackup", isBackup). Msg("[SetIsBackup]") consensus.isBackup = isBackup consensus.current.SetIsBackup(isBackup) } // Mode returns the mode of consensus func (consensus *Consensus) Mode() Mode { return consensus.current.Mode() } // RegisterPRndChannel registers the channel for receiving randomness preimage from DRG protocol func (consensus *Consensus) RegisterPRndChannel(pRndChannel chan []byte) { consensus.PRndChannel = pRndChannel } // RegisterRndChannel registers the channel for receiving final randomness from DRG protocol func (consensus *Consensus) RegisterRndChannel(rndChannel chan [548]byte) { consensus.RndChannel = rndChannel } // Check viewID, caller's responsibility to hold lock when change ignoreViewIDCheck func (consensus *Consensus) checkViewID(msg *FBFTMessage) error { // just ignore consensus check for the first time when node join if consensus.IgnoreViewIDCheck.IsSet() { //in syncing mode, node accepts incoming messages without viewID/leaderKey checking //so only set mode to normal when new node enters consensus and need checking viewID consensus.SetMode(Normal) consensus.SetViewIDs(msg.ViewID) if !msg.HasSingleSender() { return errors.New("Leader message can not have multiple sender keys") } consensus.LeaderPubKey = msg.SenderPubkeys[0] consensus.IgnoreViewIDCheck.UnSet() consensus.consensusTimeout[timeoutConsensus].Start() consensus.getLogger().Info(). Str("leaderKey", consensus.LeaderPubKey.Bytes.Hex()). Msg("[checkViewID] Start consensus timer") return nil } else if msg.ViewID > consensus.GetCurBlockViewID() { return consensus_engine.ErrViewIDNotMatch } else if msg.ViewID < consensus.GetCurBlockViewID() { return errors.New("view ID belongs to the past") } return nil } // SetBlockNum sets the blockNum in consensus object, called at node bootstrap func (consensus *Consensus) SetBlockNum(blockNum uint64) { atomic.StoreUint64(&consensus.blockNum, blockNum) } // ReadSignatureBitmapPayload read the payload for signature and bitmap; offset is the beginning position of reading func (consensus *Consensus) ReadSignatureBitmapPayload( recvPayload []byte, offset int, ) (*bls_core.Sign, *bls_cosi.Mask, error) { if offset+bls.BLSSignatureSizeInBytes > len(recvPayload) { return nil, nil, errors.New("payload not have enough length") } sigAndBitmapPayload := recvPayload[offset:] // TODO(audit): keep a Mask in the Decider so it won't be reconstructed on the fly. members := consensus.Decider.Participants() return chain.ReadSignatureBitmapByPublicKeys( sigAndBitmapPayload, members, ) } // UpdateConsensusInformation will update shard information (epoch, publicKeys, blockNum, viewID) // based on the local blockchain. It is called in two cases for now: // 1. consensus object initialization. because of current dependency where chainreader is only available // after node is initialized; node is only available after consensus is initialized // we need call this function separately after create consensus object // 2. after state syncing is finished // It will return the mode: // (a) node not in committed: Listening mode // (b) node in committed but has any err during processing: Syncing mode // (c) node in committed and everything looks good: Normal mode func (consensus *Consensus) UpdateConsensusInformation() Mode { curHeader := consensus.Blockchain().CurrentHeader() curEpoch := curHeader.Epoch() nextEpoch := new(big.Int).Add(curHeader.Epoch(), common.Big1) // Overwrite nextEpoch if the shard state has a epoch number if curHeader.IsLastBlockInEpoch() { nextShardState, err := curHeader.GetShardState() if err != nil { consensus.getLogger().Error(). Err(err). Uint32("shard", consensus.ShardID). Msg("[UpdateConsensusInformation] Error retrieving current shard state in the first block") return Syncing } if nextShardState.Epoch != nil { nextEpoch = nextShardState.Epoch } } consensus.BlockPeriod = 5 * time.Second // Enable 2s block time at the twoSecondsEpoch if consensus.Blockchain().Config().IsTwoSeconds(nextEpoch) { consensus.BlockPeriod = 2 * time.Second } isFirstTimeStaking := consensus.Blockchain().Config().IsStaking(nextEpoch) && curHeader.IsLastBlockInEpoch() && !consensus.Blockchain().Config().IsStaking(curEpoch) haventUpdatedDecider := consensus.Blockchain().Config().IsStaking(curEpoch) && consensus.Decider.Policy() != quorum.SuperMajorityStake // Only happens once, the flip-over to a new Decider policy if isFirstTimeStaking || haventUpdatedDecider { decider := quorum.NewDecider(quorum.SuperMajorityStake, consensus.ShardID) consensus.Decider = decider } var committeeToSet *shard.Committee epochToSet := curEpoch hasError := false curShardState, err := committee.WithStakingEnabled.ReadFromDB( curEpoch, consensus.Blockchain(), ) if err != nil { consensus.getLogger().Error(). Err(err). Uint32("shard", consensus.ShardID). Msg("[UpdateConsensusInformation] Error retrieving current shard state") return Syncing } consensus.getLogger().Info().Msg("[UpdateConsensusInformation] Updating.....") // genesis block is a special case that will have shard state and needs to skip processing isNotGenesisBlock := curHeader.Number().Cmp(big.NewInt(0)) > 0 if curHeader.IsLastBlockInEpoch() && isNotGenesisBlock { nextShardState, err := committee.WithStakingEnabled.ReadFromDB( nextEpoch, consensus.Blockchain(), ) if err != nil { consensus.getLogger().Error(). Err(err). Uint32("shard", consensus.ShardID). Msg("Error retrieving nextEpoch shard state") return Syncing } subComm, err := nextShardState.FindCommitteeByID(curHeader.ShardID()) if err != nil { consensus.getLogger().Error(). Err(err). Uint32("shard", consensus.ShardID). Msg("Error retrieving nextEpoch shard state") return Syncing } committeeToSet = subComm epochToSet = nextEpoch } else { subComm, err := curShardState.FindCommitteeByID(curHeader.ShardID()) if err != nil { consensus.getLogger().Error(). Err(err). Uint32("shard", consensus.ShardID). Msg("Error retrieving current shard state") return Syncing } committeeToSet = subComm } if len(committeeToSet.Slots) == 0 { consensus.getLogger().Warn(). Msg("[UpdateConsensusInformation] No members in the committee to update") hasError = true } // update public keys in the committee oldLeader := consensus.LeaderPubKey pubKeys, _ := committeeToSet.BLSPublicKeys() consensus.getLogger().Info(). Int("numPubKeys", len(pubKeys)). Msg("[UpdateConsensusInformation] Successfully updated public keys") consensus.UpdatePublicKeys(pubKeys, shard.Schedule.InstanceForEpoch(nextEpoch).ExternalAllowlist()) // Update voters in the committee if _, err := consensus.Decider.SetVoters( committeeToSet, epochToSet, ); err != nil { consensus.getLogger().Error(). Err(err). Uint32("shard", consensus.ShardID). Msg("Error when updating voters") return Syncing } // take care of possible leader change during the epoch // TODO: in a very rare case, when a M1 view change happened, the block contains coinbase for last leader // but the new leader is actually recognized by most of the nodes. At this time, if a node sync to this // exact block and set its leader, it will set with the failed leader as in the coinbase of the block. // This is a very rare case scenario and not likely to cause any issue in mainnet. But we need to think about // a solution to take care of this case because the coinbase of the latest block doesn't really represent the // the real current leader in case of M1 view change. if !curHeader.IsLastBlockInEpoch() && curHeader.Number().Uint64() != 0 { leaderPubKey, err := chain.GetLeaderPubKeyFromCoinbase(consensus.Blockchain(), curHeader) if err != nil || leaderPubKey == nil { consensus.getLogger().Error().Err(err). Msg("[UpdateConsensusInformation] Unable to get leaderPubKey from coinbase") consensus.IgnoreViewIDCheck.Set() hasError = true } else { consensus.getLogger().Info(). Str("leaderPubKey", leaderPubKey.Bytes.Hex()). Msg("[UpdateConsensusInformation] Most Recent LeaderPubKey Updated Based on BlockChain") consensus.LeaderPubKey = leaderPubKey } } for _, key := range pubKeys { // in committee myPubKeys := consensus.GetPublicKeys() if myPubKeys.Contains(key.Object) { if hasError { consensus.getLogger().Error(). Str("myKey", myPubKeys.SerializeToHexStr()). Msg("[UpdateConsensusInformation] hasError") return Syncing } // If the leader changed and I myself become the leader if (oldLeader != nil && consensus.LeaderPubKey != nil && !consensus.LeaderPubKey.Object.IsEqual(oldLeader.Object)) && consensus.IsLeader() { go func() { consensus.getLogger().Info(). Str("myKey", myPubKeys.SerializeToHexStr()). Msg("[UpdateConsensusInformation] I am the New Leader") consensus.ReadySignal <- SyncProposal }() } return Normal } } consensus.getLogger().Info(). Msg("[UpdateConsensusInformation] not in committee, Listening") // not in committee return Listening } // IsLeader check if the node is a leader or not by comparing the public key of // the node with the leader public key func (consensus *Consensus) IsLeader() bool { obj := consensus.LeaderPubKey.Object for _, key := range consensus.priKey { if key.Pub.Object.IsEqual(obj) { return true } } return false } // isLeader check if the node is a leader or not by comparing the public key of // the node with the leader public key. This function assume it runs under lock. func (consensus *Consensus) isLeader() bool { obj := consensus.LeaderPubKey.Object for _, key := range consensus.priKey { if key.Pub.Object.IsEqual(obj) { return true } } return false } // SetViewIDs set both current view ID and view changing ID to the height // of the blockchain. It is used during client startup to recover the state func (consensus *Consensus) SetViewIDs(height uint64) { fmt.Println("SetViewIDs", height) consensus.SetCurBlockViewID(height) consensus.SetViewChangingID(height) } // SetCurBlockViewID set the current view ID func (consensus *Consensus) SetCurBlockViewID(viewID uint64) uint64 { return consensus.current.SetCurBlockViewID(viewID) } // SetViewChangingID set the current view change ID func (consensus *Consensus) SetViewChangingID(viewID uint64) { consensus.current.SetViewChangingID(viewID) } // StartFinalityCount set the finality counter to current time func (consensus *Consensus) StartFinalityCount() { consensus.finalityCounter.Store(time.Now().UnixNano()) } // FinishFinalityCount calculate the current finality func (consensus *Consensus) FinishFinalityCount() { d := time.Now().UnixNano() consensus.finality = (d - consensus.finalityCounter.Load().(int64)) / 1000000 consensusFinalityHistogram.Observe(float64(consensus.finality)) } // GetFinality returns the finality time in milliseconds of previous consensus func (consensus *Consensus) GetFinality() int64 { return consensus.finality } // switchPhase will switch FBFTPhase to nextPhase if the desirePhase equals the nextPhase func (consensus *Consensus) switchPhase(subject string, desired FBFTPhase) { consensus.getLogger().Info(). Str("from:", consensus.phase.String()). Str("to:", desired.String()). Str("switchPhase:", subject) consensus.phase.Set(desired) return } var ( errGetPreparedBlock = errors.New("failed to get prepared block for self commit") errReadBitmapPayload = errors.New("failed to read signature bitmap payload") ) // selfCommit will create a commit message and commit it locally // it is used by the new leadder of the view change routine // when view change is succeeded and the new leader // received prepared payload from other validators or from local func (consensus *Consensus) selfCommit(payload []byte) error { var blockHash [32]byte copy(blockHash[:], payload[:32]) // Leader sign and add commit message block := consensus.FBFTLog.GetBlockByHash(blockHash) if block == nil { return errGetPreparedBlock } aggSig, mask, err := consensus.ReadSignatureBitmapPayload(payload, 32) if err != nil { return errReadBitmapPayload } // Have to keep the block hash so the leader can finish the commit phase of prepared block consensus.ResetState() copy(consensus.blockHash[:], blockHash[:]) consensus.switchPhase("selfCommit", FBFTCommit) consensus.aggregatedPrepareSig = aggSig consensus.prepareBitmap = mask commitPayload := signature.ConstructCommitPayload(consensus.Blockchain(), block.Epoch(), block.Hash(), block.NumberU64(), block.Header().ViewID().Uint64()) for i, key := range consensus.priKey { if err := consensus.commitBitmap.SetKey(key.Pub.Bytes, true); err != nil { consensus.getLogger().Error(). Err(err). Int("Index", i). Str("Key", key.Pub.Bytes.Hex()). Msg("[selfCommit] New Leader commit bitmap set failed") continue } if _, err := consensus.Decider.AddNewVote( quorum.Commit, []*bls_cosi.PublicKeyWrapper{key.Pub}, key.Pri.SignHash(commitPayload), common.BytesToHash(consensus.blockHash[:]), block.NumberU64(), block.Header().ViewID().Uint64(), ); err != nil { consensus.getLogger().Warn(). Err(err). Int("Index", i). Str("Key", key.Pub.Bytes.Hex()). Msg("[selfCommit] submit vote on viewchange commit failed") } } return nil } // NumSignaturesIncludedInBlock returns the number of signatures included in the block func (consensus *Consensus) NumSignaturesIncludedInBlock(block *types.Block) uint32 { count := uint32(0) members := consensus.Decider.Participants() // TODO(audit): do not reconstruct the Mask mask, err := bls.NewMask(members, nil) if err != nil { return count } err = mask.SetMask(block.Header().LastCommitBitmap()) if err != nil { return count } for _, key := range consensus.GetPublicKeys() { if ok, err := mask.KeyEnabled(key.Bytes); err == nil && ok { count++ } } return count } // getLogger returns logger for consensus contexts added func (consensus *Consensus) getLogger() *zerolog.Logger { logger := utils.Logger().With(). Uint64("myBlock", consensus.BlockNum()). Uint64("myViewID", consensus.GetCurBlockViewID()). Str("phase", consensus.phase.String()). Str("mode", consensus.current.Mode().String()). Logger() return &logger }