package contractcourt import ( "encoding/binary" "fmt" "io" "sync" "github.com/btcsuite/btcd/wire" "github.com/btcsuite/btcutil" "github.com/davecgh/go-spew/spew" "github.com/lightningnetwork/lnd/chainntnfs" "github.com/lightningnetwork/lnd/channeldb" "github.com/lightningnetwork/lnd/input" "github.com/lightningnetwork/lnd/lntypes" "github.com/lightningnetwork/lnd/lnwallet" "github.com/lightningnetwork/lnd/lnwire" "github.com/lightningnetwork/lnd/sweep" ) // htlcTimeoutResolver is a ContractResolver that's capable of resolving an // outgoing HTLC. The HTLC may be on our commitment transaction, or on the // commitment transaction of the remote party. An output on our commitment // transaction is considered fully resolved once the second-level transaction // has been confirmed (and reached a sufficient depth). An output on the // commitment transaction of the remote party is resolved once we detect a // spend of the direct HTLC output using the timeout clause. type htlcTimeoutResolver struct { // htlcResolution contains all the information required to properly // resolve this outgoing HTLC. htlcResolution lnwallet.OutgoingHtlcResolution // outputIncubating returns true if we've sent the output to the output // incubator (utxo nursery). outputIncubating bool // resolved reflects if the contract has been fully resolved or not. resolved bool // broadcastHeight is the height that the original contract was // broadcast to the main-chain at. We'll use this value to bound any // historical queries to the chain for spends/confirmations. // // TODO(roasbeef): wrap above into definite resolution embedding? broadcastHeight uint32 // htlc contains information on the htlc that we are resolving on-chain. htlc channeldb.HTLC // currentReport stores the current state of the resolver for reporting // over the rpc interface. This should only be reported in case we have // a non-nil SignDetails on the htlcResolution, otherwise the nursery // will produce reports. currentReport ContractReport // reportLock prevents concurrent access to the resolver report. reportLock sync.Mutex contractResolverKit } // newTimeoutResolver instantiates a new timeout htlc resolver. func newTimeoutResolver(res lnwallet.OutgoingHtlcResolution, broadcastHeight uint32, htlc channeldb.HTLC, resCfg ResolverConfig) *htlcTimeoutResolver { h := &htlcTimeoutResolver{ contractResolverKit: *newContractResolverKit(resCfg), htlcResolution: res, broadcastHeight: broadcastHeight, htlc: htlc, } h.initReport() return h } // ResolverKey returns an identifier which should be globally unique for this // particular resolver within the chain the original contract resides within. // // NOTE: Part of the ContractResolver interface. func (h *htlcTimeoutResolver) ResolverKey() []byte { // The primary key for this resolver will be the outpoint of the HTLC // on the commitment transaction itself. If this is our commitment, // then the output can be found within the signed timeout tx, // otherwise, it's just the ClaimOutpoint. var op wire.OutPoint if h.htlcResolution.SignedTimeoutTx != nil { op = h.htlcResolution.SignedTimeoutTx.TxIn[0].PreviousOutPoint } else { op = h.htlcResolution.ClaimOutpoint } key := newResolverID(op) return key[:] } const ( // expectedRemoteWitnessSuccessSize is the expected size of the witness // on the remote commitment transaction for an outgoing HTLC that is // swept on-chain by them with pre-image. expectedRemoteWitnessSuccessSize = 5 // remotePreimageIndex index within the witness on the remote // commitment transaction that will hold they pre-image if they go to // sweep it on chain. remotePreimageIndex = 3 // localPreimageIndex is the index within the witness on the local // commitment transaction for an outgoing HTLC that will hold the // pre-image if the remote party sweeps it. localPreimageIndex = 1 ) // claimCleanUp is a helper method that's called once the HTLC output is spent // by the remote party. It'll extract the preimage, add it to the global cache, // and finally send the appropriate clean up message. func (h *htlcTimeoutResolver) claimCleanUp( commitSpend *chainntnfs.SpendDetail) (ContractResolver, error) { // Depending on if this is our commitment or not, then we'll be looking // for a different witness pattern. spenderIndex := commitSpend.SpenderInputIndex spendingInput := commitSpend.SpendingTx.TxIn[spenderIndex] log.Infof("%T(%v): extracting preimage! remote party spent "+ "HTLC with tx=%v", h, h.htlcResolution.ClaimOutpoint, spew.Sdump(commitSpend.SpendingTx)) // If this is the remote party's commitment, then we'll be looking for // them to spend using the second-level success transaction. var preimageBytes []byte if h.htlcResolution.SignedTimeoutTx == nil { // The witness stack when the remote party sweeps the output to // them looks like: // // * <0> preimageBytes = spendingInput.Witness[remotePreimageIndex] } else { // Otherwise, they'll be spending directly from our commitment // output. In which case the witness stack looks like: // // * preimageBytes = spendingInput.Witness[localPreimageIndex] } preimage, err := lntypes.MakePreimage(preimageBytes) if err != nil { return nil, fmt.Errorf("unable to create pre-image from "+ "witness: %v", err) } log.Infof("%T(%v): extracting preimage=%v from on-chain "+ "spend!", h, h.htlcResolution.ClaimOutpoint, preimage) // With the preimage obtained, we can now add it to the global cache. if err := h.PreimageDB.AddPreimages(preimage); err != nil { log.Errorf("%T(%v): unable to add witness to cache", h, h.htlcResolution.ClaimOutpoint) } var pre [32]byte copy(pre[:], preimage[:]) // Finally, we'll send the clean up message, mark ourselves as // resolved, then exit. if err := h.DeliverResolutionMsg(ResolutionMsg{ SourceChan: h.ShortChanID, HtlcIndex: h.htlc.HtlcIndex, PreImage: &pre, }); err != nil { return nil, err } h.resolved = true // Checkpoint our resolver with a report which reflects the preimage // claim by the remote party. amt := btcutil.Amount(h.htlcResolution.SweepSignDesc.Output.Value) report := &channeldb.ResolverReport{ OutPoint: h.htlcResolution.ClaimOutpoint, Amount: amt, ResolverType: channeldb.ResolverTypeOutgoingHtlc, ResolverOutcome: channeldb.ResolverOutcomeClaimed, SpendTxID: commitSpend.SpenderTxHash, } return nil, h.Checkpoint(h, report) } // chainDetailsToWatch returns the output and script which we use to watch for // spends from the direct HTLC output on the commitment transaction. // // TODO(joostjager): output already set properly in // lnwallet.newOutgoingHtlcResolution? And script too? func (h *htlcTimeoutResolver) chainDetailsToWatch() (*wire.OutPoint, []byte, error) { // If there's no timeout transaction, then the claim output is the // output directly on the commitment transaction, so we'll just use // that. if h.htlcResolution.SignedTimeoutTx == nil { outPointToWatch := h.htlcResolution.ClaimOutpoint scriptToWatch := h.htlcResolution.SweepSignDesc.Output.PkScript return &outPointToWatch, scriptToWatch, nil } // If this is the remote party's commitment, then we'll need to grab // watch the output that our timeout transaction points to. We can // directly grab the outpoint, then also extract the witness script // (the last element of the witness stack) to re-construct the pkScript // we need to watch. outPointToWatch := h.htlcResolution.SignedTimeoutTx.TxIn[0].PreviousOutPoint witness := h.htlcResolution.SignedTimeoutTx.TxIn[0].Witness scriptToWatch, err := input.WitnessScriptHash(witness[len(witness)-1]) if err != nil { return nil, nil, err } return &outPointToWatch, scriptToWatch, nil } // isSuccessSpend returns true if the passed spend on the specified commitment // is a success spend that reveals the pre-image or not. func isSuccessSpend(spend *chainntnfs.SpendDetail, localCommit bool) bool { // Based on the spending input index and transaction, obtain the // witness that tells us what type of spend this is. spenderIndex := spend.SpenderInputIndex spendingInput := spend.SpendingTx.TxIn[spenderIndex] spendingWitness := spendingInput.Witness // If this is the remote commitment then the only possible spends for // outgoing HTLCs are: // // RECVR: <0> (2nd level success spend) // REVOK: // SENDR: 0 // // In this case, if 5 witness elements are present (factoring the // witness script), and the 3rd element is the size of the pre-image, // then this is a remote spend. If not, then we swept it ourselves, or // revoked their output. if !localCommit { return len(spendingWitness) == expectedRemoteWitnessSuccessSize && len(spendingWitness[remotePreimageIndex]) == lntypes.HashSize } // Otherwise, for our commitment, the only possible spends for an // outgoing HTLC are: // // SENDR: <0> <0> (2nd level timeout) // RECVR: // REVOK: // // So the only success case has the pre-image as the 2nd (index 1) // element in the witness. return len(spendingWitness[localPreimageIndex]) == lntypes.HashSize } // Resolve kicks off full resolution of an outgoing HTLC output. If it's our // commitment, it isn't resolved until we see the second level HTLC txn // confirmed. If it's the remote party's commitment, we don't resolve until we // see a direct sweep via the timeout clause. // // NOTE: Part of the ContractResolver interface. func (h *htlcTimeoutResolver) Resolve() (ContractResolver, error) { // If we're already resolved, then we can exit early. if h.resolved { return nil, nil } // Start by spending the HTLC output, either by broadcasting the // second-level timeout transaction, or directly if this is the remote // commitment. commitSpend, err := h.spendHtlcOutput() if err != nil { return nil, err } // If the spend reveals the pre-image, then we'll enter the clean up // workflow to pass the pre-image back to the incoming link, add it to // the witness cache, and exit. if isSuccessSpend(commitSpend, h.htlcResolution.SignedTimeoutTx != nil) { log.Infof("%T(%v): HTLC has been swept with pre-image by "+ "remote party during timeout flow! Adding pre-image to "+ "witness cache", h.htlcResolution.ClaimOutpoint) return h.claimCleanUp(commitSpend) } log.Infof("%T(%v): resolving htlc with incoming fail msg, fully "+ "confirmed", h, h.htlcResolution.ClaimOutpoint) // At this point, the second-level transaction is sufficiently // confirmed, or a transaction directly spending the output is. // Therefore, we can now send back our clean up message, failing the // HTLC on the incoming link. failureMsg := &lnwire.FailPermanentChannelFailure{} if err := h.DeliverResolutionMsg(ResolutionMsg{ SourceChan: h.ShortChanID, HtlcIndex: h.htlc.HtlcIndex, Failure: failureMsg, }); err != nil { return nil, err } // Depending on whether this was a local or remote commit, we must // handle the spending transaction accordingly. return h.handleCommitSpend(commitSpend) } // spendHtlcOutput handles the initial spend of an HTLC output via the timeout // clause. If this is our local commitment, the second-level timeout TX will be // used to spend the output into the next stage. If this is the remote // commitment, the output will be swept directly without the timeout // transaction. func (h *htlcTimeoutResolver) spendHtlcOutput() (*chainntnfs.SpendDetail, error) { switch { // If we have non-nil SignDetails, this means that have a 2nd level // HTLC transaction that is signed using sighash SINGLE|ANYONECANPAY // (the case for anchor type channels). In this case we can re-sign it // and attach fees at will. We let the sweeper handle this job. case h.htlcResolution.SignDetails != nil && !h.outputIncubating: log.Infof("%T(%x): offering second-layer timeout tx to "+ "sweeper: %v", h, h.htlc.RHash[:], spew.Sdump(h.htlcResolution.SignedTimeoutTx)) inp := input.MakeHtlcSecondLevelTimeoutAnchorInput( h.htlcResolution.SignedTimeoutTx, h.htlcResolution.SignDetails, h.broadcastHeight, ) _, err := h.Sweeper.SweepInput( &inp, sweep.Params{ Fee: sweep.FeePreference{ ConfTarget: secondLevelConfTarget, }, }, ) if err != nil { return nil, err } // If we have no SignDetails, and we haven't already sent the output to // the utxo nursery, then we'll do so now. case h.htlcResolution.SignDetails == nil && !h.outputIncubating: log.Tracef("%T(%v): incubating htlc output", h, h.htlcResolution.ClaimOutpoint) err := h.IncubateOutputs( h.ChanPoint, &h.htlcResolution, nil, h.broadcastHeight, ) if err != nil { return nil, err } h.outputIncubating = true if err := h.Checkpoint(h); err != nil { log.Errorf("unable to Checkpoint: %v", err) return nil, err } } // Now that we've handed off the HTLC to the nursery or sweeper, we'll // watch for a spend of the output, and make our next move off of that. // Depending on if this is our commitment, or the remote party's // commitment, we'll be watching a different outpoint and script. outpointToWatch, scriptToWatch, err := h.chainDetailsToWatch() if err != nil { return nil, err } log.Infof("%T(%v): waiting for HTLC output %v to be spent"+ "fully confirmed", h, h.htlcResolution.ClaimOutpoint, outpointToWatch) // We'll block here until either we exit, or the HTLC output on the // commitment transaction has been spent. spend, err := waitForSpend( outpointToWatch, scriptToWatch, h.broadcastHeight, h.Notifier, h.quit, ) if err != nil { return nil, err } // If this was the second level transaction published by the sweeper, // we can checkpoint the resolver now that it's confirmed. if h.htlcResolution.SignDetails != nil && !h.outputIncubating { h.outputIncubating = true if err := h.Checkpoint(h); err != nil { log.Errorf("unable to Checkpoint: %v", err) return nil, err } } return spend, err } // handleCommitSpend handles the spend of the HTLC output on the commitment // transaction. If this was our local commitment, the spend will be he // confirmed second-level timeout transaction, and we'll sweep that into our // wallet. If the was a remote commitment, the resolver will resolve // immetiately. func (h *htlcTimeoutResolver) handleCommitSpend( commitSpend *chainntnfs.SpendDetail) (ContractResolver, error) { var ( // claimOutpoint will be the outpoint of the second level // transaction, or on the remote commitment directly. It will // start out as set in the resolution, but we'll update it if // the second-level goes through the sweeper and changes its // txid. claimOutpoint = h.htlcResolution.ClaimOutpoint // spendTxID will be the ultimate spend of the claimOutpoint. // We set it to the commit spend for now, as this is the // ultimate spend in case this is a remote commitment. If we go // through the second-level transaction, we'll update this // accordingly. spendTxID = commitSpend.SpenderTxHash reports []*channeldb.ResolverReport ) switch { // If the sweeper is handling the second level transaction, wait for // the CSV lock to expire, before sweeping the output on the // second-level. case h.htlcResolution.SignDetails != nil: waitHeight := uint32(commitSpend.SpendingHeight) + h.htlcResolution.CsvDelay - 1 h.reportLock.Lock() h.currentReport.Stage = 2 h.currentReport.MaturityHeight = waitHeight h.reportLock.Unlock() log.Infof("%T(%x): waiting for CSV lock to expire at height %v", h, h.htlc.RHash[:], waitHeight) err := waitForHeight(waitHeight, h.Notifier, h.quit) if err != nil { return nil, err } // We'll use this input index to determine the second-level // output index on the transaction, as the signatures requires // the indexes to be the same. We don't look for the // second-level output script directly, as there might be more // than one HTLC output to the same pkScript. op := &wire.OutPoint{ Hash: *commitSpend.SpenderTxHash, Index: commitSpend.SpenderInputIndex, } // Let the sweeper sweep the second-level output now that the // CSV delay has passed. log.Infof("%T(%x): CSV lock expired, offering second-layer "+ "output to sweeper: %v", h, h.htlc.RHash[:], op) inp := input.NewCsvInput( op, input.HtlcOfferedTimeoutSecondLevel, &h.htlcResolution.SweepSignDesc, h.broadcastHeight, h.htlcResolution.CsvDelay, ) _, err = h.Sweeper.SweepInput( inp, sweep.Params{ Fee: sweep.FeePreference{ ConfTarget: sweepConfTarget, }, }, ) if err != nil { return nil, err } // Update the claim outpoint to point to the second-level // transaction created by the sweeper. claimOutpoint = *op fallthrough // Finally, if this was an output on our commitment transaction, we'll // wait for the second-level HTLC output to be spent, and for that // transaction itself to confirm. case h.htlcResolution.SignedTimeoutTx != nil: log.Infof("%T(%v): waiting for nursery/sweeper to spend CSV "+ "delayed output", h, claimOutpoint) sweep, err := waitForSpend( &claimOutpoint, h.htlcResolution.SweepSignDesc.Output.PkScript, h.broadcastHeight, h.Notifier, h.quit, ) if err != nil { return nil, err } // Update the spend txid to the hash of the sweep transaction. spendTxID = sweep.SpenderTxHash // Once our sweep of the timeout tx has confirmed, we add a // resolution for our timeoutTx tx first stage transaction. timeoutTx := commitSpend.SpendingTx index := commitSpend.SpenderInputIndex spendHash := commitSpend.SpenderTxHash reports = append(reports, &channeldb.ResolverReport{ OutPoint: timeoutTx.TxIn[index].PreviousOutPoint, Amount: h.htlc.Amt.ToSatoshis(), ResolverType: channeldb.ResolverTypeOutgoingHtlc, ResolverOutcome: channeldb.ResolverOutcomeFirstStage, SpendTxID: spendHash, }) } // With the clean up message sent, we'll now mark the contract // resolved, update the recovered balance, record the timeout and the // sweep txid on disk, and wait. h.resolved = true h.reportLock.Lock() h.currentReport.RecoveredBalance = h.currentReport.LimboBalance h.currentReport.LimboBalance = 0 h.reportLock.Unlock() amt := btcutil.Amount(h.htlcResolution.SweepSignDesc.Output.Value) reports = append(reports, &channeldb.ResolverReport{ OutPoint: claimOutpoint, Amount: amt, ResolverType: channeldb.ResolverTypeOutgoingHtlc, ResolverOutcome: channeldb.ResolverOutcomeTimeout, SpendTxID: spendTxID, }) return nil, h.Checkpoint(h, reports...) } // Stop signals the resolver to cancel any current resolution processes, and // suspend. // // NOTE: Part of the ContractResolver interface. func (h *htlcTimeoutResolver) Stop() { close(h.quit) } // IsResolved returns true if the stored state in the resolve is fully // resolved. In this case the target output can be forgotten. // // NOTE: Part of the ContractResolver interface. func (h *htlcTimeoutResolver) IsResolved() bool { return h.resolved } // report returns a report on the resolution state of the contract. func (h *htlcTimeoutResolver) report() *ContractReport { // If the sign details are nil, the report will be created by handled // by the nursery. if h.htlcResolution.SignDetails == nil { return nil } h.reportLock.Lock() defer h.reportLock.Unlock() copy := h.currentReport return © } func (h *htlcTimeoutResolver) initReport() { // We create the initial report. This will only be reported for // resolvers not handled by the nursery. finalAmt := h.htlc.Amt.ToSatoshis() if h.htlcResolution.SignedTimeoutTx != nil { finalAmt = btcutil.Amount( h.htlcResolution.SignedTimeoutTx.TxOut[0].Value, ) } h.currentReport = ContractReport{ Outpoint: h.htlcResolution.ClaimOutpoint, Type: ReportOutputOutgoingHtlc, Amount: finalAmt, MaturityHeight: h.htlcResolution.Expiry, LimboBalance: finalAmt, Stage: 1, } } // Encode writes an encoded version of the ContractResolver into the passed // Writer. // // NOTE: Part of the ContractResolver interface. func (h *htlcTimeoutResolver) Encode(w io.Writer) error { // First, we'll write out the relevant fields of the // OutgoingHtlcResolution to the writer. if err := encodeOutgoingResolution(w, &h.htlcResolution); err != nil { return err } // With that portion written, we can now write out the fields specific // to the resolver itself. if err := binary.Write(w, endian, h.outputIncubating); err != nil { return err } if err := binary.Write(w, endian, h.resolved); err != nil { return err } if err := binary.Write(w, endian, h.broadcastHeight); err != nil { return err } if err := binary.Write(w, endian, h.htlc.HtlcIndex); err != nil { return err } // We encode the sign details last for backwards compatibility. err := encodeSignDetails(w, h.htlcResolution.SignDetails) if err != nil { return err } return nil } // newTimeoutResolverFromReader attempts to decode an encoded ContractResolver // from the passed Reader instance, returning an active ContractResolver // instance. func newTimeoutResolverFromReader(r io.Reader, resCfg ResolverConfig) ( *htlcTimeoutResolver, error) { h := &htlcTimeoutResolver{ contractResolverKit: *newContractResolverKit(resCfg), } // First, we'll read out all the mandatory fields of the // OutgoingHtlcResolution that we store. if err := decodeOutgoingResolution(r, &h.htlcResolution); err != nil { return nil, err } // With those fields read, we can now read back the fields that are // specific to the resolver itself. if err := binary.Read(r, endian, &h.outputIncubating); err != nil { return nil, err } if err := binary.Read(r, endian, &h.resolved); err != nil { return nil, err } if err := binary.Read(r, endian, &h.broadcastHeight); err != nil { return nil, err } if err := binary.Read(r, endian, &h.htlc.HtlcIndex); err != nil { return nil, err } // Sign details is a new field that was added to the htlc resolution, // so it is serialized last for backwards compatibility. We try to read // it, but don't error out if there are not bytes left. signDetails, err := decodeSignDetails(r) if err == nil { h.htlcResolution.SignDetails = signDetails } else if err != io.EOF && err != io.ErrUnexpectedEOF { return nil, err } h.initReport() return h, nil } // Supplement adds additional information to the resolver that is required // before Resolve() is called. // // NOTE: Part of the htlcContractResolver interface. func (h *htlcTimeoutResolver) Supplement(htlc channeldb.HTLC) { h.htlc = htlc } // HtlcPoint returns the htlc's outpoint on the commitment tx. // // NOTE: Part of the htlcContractResolver interface. func (h *htlcTimeoutResolver) HtlcPoint() wire.OutPoint { return h.htlcResolution.HtlcPoint() } // A compile time assertion to ensure htlcTimeoutResolver meets the // ContractResolver interface. var _ htlcContractResolver = (*htlcTimeoutResolver)(nil)