2
0
mirror of https://github.com/lightninglabs/loop synced 2024-11-04 06:00:21 +00:00
loop/loopin.go
Andras Banki-Horvath 5294b4ff07
loop: clean up server cost calculation for slightly better UX
Previously we'd calculate the server costs (swap fees) by accounting for
both the on-chain HTLC and the off-chain payment which was confusing as
server cost fluctuated by the amount of the swap itself. Now we'll only
add the actual cost when the swap happened, so the server cost will go
form zero to the actual fee value paid.
2024-03-04 16:15:21 +01:00

1191 lines
33 KiB
Go

package loop
import (
"context"
"crypto/rand"
"crypto/sha256"
"errors"
"fmt"
"sync"
"time"
"github.com/btcsuite/btcd/btcec/v2"
"github.com/btcsuite/btcd/btcutil"
"github.com/btcsuite/btcd/chaincfg/chainhash"
"github.com/btcsuite/btcd/mempool"
"github.com/btcsuite/btcd/wire"
"github.com/lightninglabs/lndclient"
"github.com/lightninglabs/loop/labels"
"github.com/lightninglabs/loop/loopdb"
"github.com/lightninglabs/loop/swap"
"github.com/lightninglabs/loop/utils"
"github.com/lightningnetwork/lnd/chainntnfs"
invpkg "github.com/lightningnetwork/lnd/invoices"
"github.com/lightningnetwork/lnd/keychain"
"github.com/lightningnetwork/lnd/lnrpc/invoicesrpc"
"github.com/lightningnetwork/lnd/lnrpc/walletrpc"
"github.com/lightningnetwork/lnd/lntypes"
"github.com/lightningnetwork/lnd/lnwallet/chainfee"
"github.com/lightningnetwork/lnd/lnwire"
"github.com/lightningnetwork/lnd/routing/route"
)
var (
// MaxLoopInAcceptDelta configures the maximum acceptable number of
// remaining blocks until the on-chain htlc expires. This value is used
// to decide whether we want to continue with the swap parameters as
// proposed by the server. It is a protection to prevent the server from
// getting us to lock up our funds to an arbitrary point in the future.
MaxLoopInAcceptDelta = int32(1500)
// MinLoopInPublishDelta defines the minimum number of remaining blocks
// until on-chain htlc expiry required to proceed to publishing the htlc
// tx. This value isn't critical, as we could even safely publish the
// htlc after expiry. The reason we do implement this check is to
// prevent us from publishing an htlc that the server surely wouldn't
// follow up to.
MinLoopInPublishDelta = int32(10)
// TimeoutTxConfTarget defines the confirmation target for the loop in
// timeout tx.
TimeoutTxConfTarget = int32(2)
// ErrSwapFinalized is returned when a to be executed swap is already in
// a final state.
ErrSwapFinalized = errors.New("swap is in a final state")
)
// loopInSwap contains all the in-memory state related to a pending loop in
// swap.
type loopInSwap struct {
swapKit
executeConfig
loopdb.LoopInContract
htlc *swap.Htlc
htlcP2WSH *swap.Htlc
htlcP2TR *swap.Htlc
// htlcTxHash is the confirmed htlc tx id.
htlcTxHash *chainhash.Hash
timeoutAddr btcutil.Address
abandonChan chan struct{}
wg sync.WaitGroup
}
// loopInInitResult contains information about a just-initiated loop in swap.
type loopInInitResult struct {
swap *loopInSwap
serverMessage string
}
// newLoopInSwap initiates a new loop in swap.
func newLoopInSwap(globalCtx context.Context, cfg *swapConfig,
currentHeight int32, request *LoopInRequest) (*loopInInitResult,
error) {
var err error
// Private and route hints are mutually exclusive as setting private
// means we retrieve our own route hints from the connected node.
if len(request.RouteHints) != 0 && request.Private {
return nil, fmt.Errorf("private and route_hints both set")
}
// If Private is set, we generate route hints.
if request.Private {
// If last_hop is set, we'll only add channels with peers set to
// the last_hop parameter.
includeNodes := make(map[route.Vertex]struct{})
if request.LastHop != nil {
includeNodes[*request.LastHop] = struct{}{}
}
// Because the Private flag is set, we'll generate our own set
// of hop hints.
request.RouteHints, err = SelectHopHints(
globalCtx, cfg.lnd, request.Amount, DefaultMaxHopHints,
includeNodes,
)
if err != nil {
return nil, err
}
}
// Request current server loop in terms and use these to calculate the
// swap fee that we should subtract from the swap amount in the payment
// request that we send to the server. We pass nil as optional route
// hints as hop hint selection when generating invoices with private
// channels is an LND side black box feature. Advanced users will quote
// directly anyway and there they have the option to add specific route
// hints.
quote, err := cfg.server.GetLoopInQuote(
globalCtx, request.Amount, cfg.lnd.NodePubkey, request.LastHop,
request.RouteHints, request.Initiator,
)
if err != nil {
return nil, wrapGrpcError("loop in terms", err)
}
swapFee := quote.SwapFee
if swapFee > request.MaxSwapFee {
log.Warnf("Swap fee %v exceeding maximum of %v",
swapFee, request.MaxSwapFee)
return nil, ErrSwapFeeTooHigh
}
// Calculate the swap invoice amount. The prepay is added which
// effectively forces the server to pay us back our prepayment on a
// successful swap.
swapInvoiceAmt := request.Amount - swapFee
// Generate random preimage.
var swapPreimage lntypes.Preimage
if _, err := rand.Read(swapPreimage[:]); err != nil {
log.Error("Cannot generate preimage")
}
swapHash := lntypes.Hash(sha256.Sum256(swapPreimage[:]))
// Derive a sender key for this swap.
keyDesc, err := cfg.lnd.WalletKit.DeriveNextKey(
globalCtx, swap.KeyFamily,
)
if err != nil {
return nil, err
}
var senderKey [33]byte
copy(senderKey[:], keyDesc.PubKey.SerializeCompressed())
// Create the swap invoice in lnd.
_, swapInvoice, err := cfg.lnd.Client.AddInvoice(
globalCtx, &invoicesrpc.AddInvoiceData{
Preimage: &swapPreimage,
Value: lnwire.NewMSatFromSatoshis(swapInvoiceAmt),
Memo: "swap",
Expiry: 3600 * 24 * 365,
RouteHints: request.RouteHints,
},
)
if err != nil {
return nil, err
}
// Create the probe invoice in lnd. Derive the payment hash
// deterministically from the swap hash in such a way that the server
// can be sure that we don't know the preimage.
probeHash := lntypes.Hash(sha256.Sum256(swapHash[:]))
probeHash[0] ^= 1
log.Infof("Creating probe invoice %v", probeHash)
probeInvoice, err := cfg.lnd.Invoices.AddHoldInvoice(
globalCtx, &invoicesrpc.AddInvoiceData{
Hash: &probeHash,
Value: lnwire.NewMSatFromSatoshis(swapInvoiceAmt),
Memo: "loop in probe",
Expiry: 3600,
RouteHints: request.RouteHints,
},
)
if err != nil {
return nil, err
}
// Default the HTLC internal key to our sender key.
senderInternalPubKey := senderKey
// If this is a MuSig2 swap then we'll generate a brand new key pair and
// will use that as the internal key for the HTLC.
if loopdb.CurrentProtocolVersion() >= loopdb.ProtocolVersionMuSig2 {
secret, err := sharedSecretFromHash(
globalCtx, cfg.lnd.Signer, swapHash,
)
if err != nil {
return nil, err
}
_, pubKey := btcec.PrivKeyFromBytes(secret[:])
copy(senderInternalPubKey[:], pubKey.SerializeCompressed())
}
// Create a cancellable context that is used for monitoring the probe.
probeWaitCtx, probeWaitCancel := context.WithCancel(globalCtx)
// Launch a goroutine to monitor the probe.
probeResult, err := awaitProbe(probeWaitCtx, *cfg.lnd, probeHash)
if err != nil {
probeWaitCancel()
return nil, fmt.Errorf("probe failed: %v", err)
}
// Post the swap parameters to the swap server. The response contains
// the server success key and the expiry height of the on-chain swap
// htlc.
log.Infof("Initiating swap request at height %v", currentHeight)
swapResp, err := cfg.server.NewLoopInSwap(globalCtx, swapHash,
request.Amount, senderKey, senderInternalPubKey, swapInvoice,
probeInvoice, request.LastHop, request.Initiator,
)
probeWaitCancel()
if err != nil {
return nil, wrapGrpcError("cannot initiate swap", err)
}
// Because the context is cancelled, it is guaranteed that we will be
// able to read from the probeResult channel.
err = <-probeResult
if err != nil {
return nil, fmt.Errorf("probe error: %v", err)
}
// Validate if the response parameters are outside our allowed range
// preventing us from continuing with a swap.
err = validateLoopInContract(currentHeight, swapResp)
if err != nil {
return nil, err
}
// Instantiate a struct that contains all required data to start the
// swap.
initiationTime := time.Now()
contract := loopdb.LoopInContract{
HtlcConfTarget: request.HtlcConfTarget,
LastHop: request.LastHop,
ExternalHtlc: request.ExternalHtlc,
SwapContract: loopdb.SwapContract{
InitiationHeight: currentHeight,
InitiationTime: initiationTime,
HtlcKeys: loopdb.HtlcKeys{
SenderScriptKey: senderKey,
SenderInternalPubKey: senderKey,
ReceiverScriptKey: swapResp.receiverKey,
ReceiverInternalPubKey: swapResp.receiverKey,
ClientScriptKeyLocator: keyDesc.KeyLocator,
},
Preimage: swapPreimage,
AmountRequested: request.Amount,
CltvExpiry: swapResp.expiry,
MaxMinerFee: request.MaxMinerFee,
MaxSwapFee: request.MaxSwapFee,
Label: request.Label,
ProtocolVersion: loopdb.CurrentProtocolVersion(),
},
}
// For MuSig2 swaps we store the proper internal keys that we generated
// and received from the server.
if loopdb.CurrentProtocolVersion() >= loopdb.ProtocolVersionMuSig2 {
contract.HtlcKeys.SenderInternalPubKey = senderInternalPubKey
contract.HtlcKeys.ReceiverInternalPubKey = swapResp.receiverInternalKey
}
swapKit := newSwapKit(
swapHash, swap.TypeIn,
cfg, &contract.SwapContract,
)
swapKit.lastUpdateTime = initiationTime
swap := &loopInSwap{
LoopInContract: contract,
swapKit: *swapKit,
}
if err := swap.initHtlcs(); err != nil {
return nil, err
}
// Persist the data before exiting this function, so that the caller can
// trust that this swap will be resumed on restart.
err = cfg.store.CreateLoopIn(globalCtx, swapHash, &swap.LoopInContract)
if err != nil {
return nil, fmt.Errorf("cannot store swap: %v", err)
}
if swapResp.serverMessage != "" {
swap.log.Infof("Server message: %v", swapResp.serverMessage)
}
swap.abandonChan = make(chan struct{}, 1)
return &loopInInitResult{
swap: swap,
serverMessage: swapResp.serverMessage,
}, nil
}
// awaitProbe waits for a probe payment to arrive and cancels it. This is a
// workaround for the current lack of multi-path probing.
func awaitProbe(ctx context.Context, lnd lndclient.LndServices,
probeHash lntypes.Hash) (chan error, error) {
// Subscribe to the probe invoice.
updateChan, errChan, err := lnd.Invoices.SubscribeSingleInvoice(
ctx, probeHash,
)
if err != nil {
return nil, err
}
// Wait in the background for the probe to arrive.
probeResult := make(chan error, 1)
go func() {
for {
select {
case update := <-updateChan:
switch update.State {
case invpkg.ContractAccepted:
log.Infof("Server probe successful")
probeResult <- nil
// Cancel probe invoice so that the
// server will know that its probe was
// successful.
err := lnd.Invoices.CancelInvoice(
ctx, probeHash,
)
if err != nil {
log.Errorf("Cancel probe "+
"invoice: %v", err)
}
return
case invpkg.ContractCanceled:
probeResult <- errors.New(
"probe invoice expired")
return
case invpkg.ContractSettled:
probeResult <- errors.New(
"impossible that probe " +
"invoice was settled")
return
}
case err := <-errChan:
probeResult <- err
return
case <-ctx.Done():
probeResult <- ctx.Err()
return
}
}
}()
return probeResult, nil
}
// resumeLoopInSwap returns a swap object representing a pending swap that has
// been restored from the database.
func resumeLoopInSwap(_ context.Context, cfg *swapConfig,
pend *loopdb.LoopIn) (*loopInSwap, error) {
hash := lntypes.Hash(sha256.Sum256(pend.Contract.Preimage[:]))
log.Infof("Resuming loop in swap %v", hash)
swapKit := newSwapKit(
hash, swap.TypeIn, cfg,
&pend.Contract.SwapContract,
)
swap := &loopInSwap{
LoopInContract: *pend.Contract,
swapKit: *swapKit,
}
if err := swap.initHtlcs(); err != nil {
return nil, err
}
lastUpdate := pend.LastUpdate()
if lastUpdate == nil {
swap.lastUpdateTime = pend.Contract.InitiationTime
} else {
swap.state = lastUpdate.State
swap.lastUpdateTime = lastUpdate.Time
swap.htlcTxHash = lastUpdate.HtlcTxHash
swap.cost = lastUpdate.Cost
}
// Upon restoring the swap we also need to assign a new abandon channel
// that the client can use to signal that the swap should be abandoned.
swap.abandonChan = make(chan struct{}, 1)
return swap, nil
}
// validateLoopInContract validates the contract parameters against our request.
func validateLoopInContract(height int32, response *newLoopInResponse) error {
// Verify that we are not forced to publish a htlc that locks up our
// funds for too long in case the server doesn't follow through.
if response.expiry-height > MaxLoopInAcceptDelta {
return ErrExpiryTooFar
}
return nil
}
// initHtlcs creates and updates the native and nested segwit htlcs of the
// loopInSwap.
func (s *loopInSwap) initHtlcs() error {
htlc, err := utils.GetHtlc(
s.hash, &s.SwapContract, s.swapKit.lnd.ChainParams,
)
if err != nil {
return err
}
switch htlc.OutputType {
case swap.HtlcP2WSH:
s.htlcP2WSH = htlc
case swap.HtlcP2TR:
s.htlcP2TR = htlc
default:
return fmt.Errorf("invalid output type")
}
s.swapKit.log.Infof("Htlc address (%s): %v", htlc.OutputType,
htlc.Address)
return nil
}
// sendUpdate reports an update to the swap state.
func (s *loopInSwap) sendUpdate(ctx context.Context) error {
info := s.swapInfo()
s.log.Infof("Loop in swap state: %v", info.State)
if IsTaprootSwap(&s.SwapContract) {
info.HtlcAddressP2TR = s.htlcP2TR.Address
} else {
info.HtlcAddressP2WSH = s.htlcP2WSH.Address
}
info.ExternalHtlc = s.ExternalHtlc
// In order to avoid potentially dangerous ownership sharing we copy the
// last hop vertex.
if s.LastHop != nil {
lastHop := &route.Vertex{}
copy(lastHop[:], s.LastHop[:])
info.LastHop = lastHop
}
select {
case s.statusChan <- *info:
case <-ctx.Done():
return ctx.Err()
}
return nil
}
// execute starts/resumes the swap. It is a thin wrapper around executeSwap to
// conveniently handle the error case.
func (s *loopInSwap) execute(mainCtx context.Context,
cfg *executeConfig, height int32) error {
defer s.wg.Wait()
s.executeConfig = *cfg
s.height = height
// Create context for our state subscription which we will cancel once
// swap execution has completed, ensuring that we kill the subscribe
// goroutine.
subCtx, cancel := context.WithCancel(mainCtx)
defer cancel()
s.wg.Add(1)
go func() {
defer s.wg.Done()
subscribeAndLogUpdates(
subCtx, s.hash, s.log, s.server.SubscribeLoopInUpdates,
)
}()
// Announce swap by sending out an initial update.
err := s.sendUpdate(mainCtx)
if err != nil {
return err
}
// Execute the swap until it either reaches a final state or a temporary
// error occurs.
err = s.executeSwap(mainCtx)
// If there are insufficient confirmed funds to publish the swap, we
// finalize its state so a new swap will be published if funds become
// available.
if errors.Is(err, ErrInsufficientBalance) {
return err
}
// Stop the execution if the swap has been abandoned.
if err != nil && s.state == loopdb.StateFailAbandoned {
return err
}
// Sanity check. If there is no error, the swap must be in a final
// state.
if err == nil && s.state.Type() == loopdb.StateTypePending {
err = fmt.Errorf("swap in non-final state %v", s.state)
}
// If an unexpected error happened, report a temporary failure but don't
// persist the error. Otherwise, for example a connection error could
// lead to abandoning the swap permanently and losing funds.
if err != nil {
s.log.Errorf("Swap error: %v", err)
s.setState(loopdb.StateFailTemporary)
// If we cannot send out this update, there is nothing we can
// do.
_ = s.sendUpdate(mainCtx)
return err
}
s.log.Infof("Loop in swap completed: %v "+
"(final cost: server %v, onchain %v, offchain %v)",
s.state,
s.cost.Server,
s.cost.Onchain,
s.cost.Offchain,
)
return nil
}
// executeSwap executes the swap.
func (s *loopInSwap) executeSwap(globalCtx context.Context) error {
var err error
// If the swap is already in a final state, we can return immediately.
if s.state.IsFinal() {
return ErrSwapFinalized
}
// For loop in, the client takes the first step by publishing the
// on-chain htlc. Only do this if we haven't already done so in a
// previous run.
if s.state == loopdb.StateInitiated {
if s.ExternalHtlc {
// If an external htlc was indicated, we can move to the
// HtlcPublished state directly and wait for
// confirmation.
s.setState(loopdb.StateHtlcPublished)
err = s.persistAndAnnounceState(globalCtx)
if err != nil {
return err
}
} else {
published, err := s.publishOnChainHtlc(globalCtx)
if err != nil {
return err
}
if !published {
return nil
}
}
}
// Wait for the htlc to confirm. After a restart, this will pick up a
// previously published tx.
conf, err := s.waitForHtlcConf(globalCtx)
if err != nil {
return err
}
// Determine the htlc outpoint by inspecting the htlc tx.
htlcOutpoint, htlcValue, err := swap.GetScriptOutput(
conf.Tx, s.htlc.PkScript,
)
if err != nil {
return err
}
// Verify that the confirmed (external) htlc value matches the swap
// amount. If the amounts mismatch we update the swap state to indicate
// this, but end processing the swap. Instead, we continue to wait for
// the htlc to expire and publish a timeout tx to reclaim the funds. We
// skip this part if the swap was recovered from this state.
if s.state != loopdb.StateFailIncorrectHtlcAmt &&
htlcValue != s.LoopInContract.AmountRequested {
s.setState(loopdb.StateFailIncorrectHtlcAmt)
err = s.persistAndAnnounceState(globalCtx)
if err != nil {
log.Errorf("Error persisting state: %v", err)
}
}
// The server is expected to see the htlc on-chain and know that it can
// sweep that htlc with the preimage, it should pay our swap invoice,
// receive the preimage and sweep the htlc. We are waiting for this to
// happen and simultaneously watch the htlc expiry height. When the htlc
// expires, we will publish a timeout tx to reclaim the funds.
err = s.waitForSwapComplete(globalCtx, htlcOutpoint, htlcValue)
if err != nil {
return err
}
// Persist swap outcome.
if err := s.persistAndAnnounceState(globalCtx); err != nil {
return err
}
return nil
}
// waitForHtlcConf watches the chain until the htlc confirms.
func (s *loopInSwap) waitForHtlcConf(globalCtx context.Context) (
*chainntnfs.TxConfirmation, error) {
// Register for confirmation of the htlc. It is essential to specify not
// just the pk script, because an attacker may publish the same htlc
// with a lower value, and we don't want to follow through with that tx.
// In the unlikely event that our call to SendOutputs crashes, and we
// restart, htlcTxHash will be nil at this point. Then only register
// with PkScript and accept the risk that the call triggers on a
// different htlc outpoint.
s.log.Infof("Register for htlc conf (hh=%v, txid=%v)",
s.InitiationHeight, s.htlcTxHash)
if s.htlcTxHash == nil {
s.log.Warnf("No htlc tx hash available, registering with " +
"just the pkscript")
}
ctx, cancel := context.WithCancel(globalCtx)
defer cancel()
notifier := s.lnd.ChainNotifier
notifyConfirmation := func(htlc *swap.Htlc) (
chan *chainntnfs.TxConfirmation, chan error, error) {
if htlc == nil {
return nil, nil, nil
}
return notifier.RegisterConfirmationsNtfn(
ctx, s.htlcTxHash, htlc.PkScript, 1, s.InitiationHeight,
)
}
confChanP2WSH, confErrP2WSH, err := notifyConfirmation(s.htlcP2WSH)
if err != nil {
return nil, err
}
confChanP2TR, confErrP2TR, err := notifyConfirmation(s.htlcP2TR)
if err != nil {
return nil, err
}
var conf *chainntnfs.TxConfirmation
for conf == nil {
select {
// P2WSH htlc confirmed.
case conf = <-confChanP2WSH:
s.htlc = s.htlcP2WSH
s.log.Infof("P2WSH htlc confirmed")
// P2TR htlc confirmed.
case conf = <-confChanP2TR:
s.htlc = s.htlcP2TR
s.log.Infof("P2TR htlc confirmed")
// Conf ntfn error.
case err := <-confErrP2WSH:
return nil, err
// Conf ntfn error.
case err := <-confErrP2TR:
return nil, err
// Keep up with block height.
case notification := <-s.blockEpochChan:
s.height = notification.(int32)
// If the client requested the swap to be abandoned, we override
// the status in the database.
case <-s.abandonChan:
return nil, s.setStateAbandoned(ctx)
// Cancel.
case <-globalCtx.Done():
return nil, globalCtx.Err()
}
}
// Store htlc tx hash for accounting purposes. Usually this call is a
// no-op because the htlc tx hash was already known. Exceptions are:
//
// - Old pending swaps that were initiated before we persisted the htlc
// tx hash directly after publish.
//
// - Swaps that experienced a crash during their call to SendOutputs. In
// that case, we weren't able to record the tx hash.
txHash := conf.Tx.TxHash()
s.htlcTxHash = &txHash
return conf, nil
}
// publishOnChainHtlc checks whether there are still enough blocks left and if
// so, it publishes the htlc and advances the swap state.
func (s *loopInSwap) publishOnChainHtlc(ctx context.Context) (bool, error) {
var err error
blocksRemaining := s.CltvExpiry - s.height
s.log.Infof("Blocks left until on-chain expiry: %v", blocksRemaining)
// Verify whether it still makes sense to publish the htlc.
if blocksRemaining < MinLoopInPublishDelta {
s.setState(loopdb.StateFailTimeout)
return false, s.persistAndAnnounceState(ctx)
}
// Get fee estimate from lnd.
feeRate, err := s.lnd.WalletKit.EstimateFeeRate(
ctx, s.LoopInContract.HtlcConfTarget,
)
if err != nil {
return false, fmt.Errorf("estimate fee: %v", err)
}
// Transition to state HtlcPublished before calling SendOutputs to
// prevent us from ever paying multiple times after a crash.
s.setState(loopdb.StateHtlcPublished)
err = s.persistAndAnnounceState(ctx)
if err != nil {
return false, err
}
s.log.Infof("Publishing on chain HTLC with fee rate %v", feeRate)
var pkScript []byte
if IsTaprootSwap(&s.SwapContract) {
pkScript = s.htlcP2TR.PkScript
} else {
pkScript = s.htlcP2WSH.PkScript
}
tx, err := s.lnd.WalletKit.SendOutputs(
ctx, []*wire.TxOut{{
PkScript: pkScript,
Value: int64(s.LoopInContract.AmountRequested),
}}, feeRate, labels.LoopInHtlcLabel(swap.ShortHash(&s.hash)),
)
if err != nil {
s.log.Errorf("send outputs: %v", err)
s.setState(loopdb.StateFailInsufficientConfirmedBalance)
// If we cannot send out this update, there is nothing we can
// do.
_ = s.persistAndAnnounceState(ctx)
return false, ErrInsufficientBalance
}
txHash := tx.TxHash()
fee := getTxFee(tx, feeRate.FeePerKVByte())
s.log.Infof("Published on chain HTLC tx %v, fee: %v", txHash, fee)
// Persist the htlc hash so that after a restart we are still waiting
// for our own htlc. We don't need to announce to clients, because the
// state remains unchanged.
s.htlcTxHash = &txHash
// We do not expect any on-chain fees to be recorded yet, and we only
// publish our htlc once, so we set our total on-chain costs to equal
// the fee for publishing the htlc.
s.cost.Onchain = fee
s.lastUpdateTime = time.Now()
if err := s.persistState(ctx); err != nil {
return false, fmt.Errorf("persist htlc tx: %v", err)
}
return true, nil
}
// getTxFee calculates our fee for a transaction that we have broadcast. We use
// sat per kvbyte because this is what lnd uses, and we will run into rounding
// issues if we do not use the same fee rate as lnd.
func getTxFee(tx *wire.MsgTx, fee chainfee.SatPerKVByte) btcutil.Amount {
btcTx := btcutil.NewTx(tx)
vsize := mempool.GetTxVirtualSize(btcTx)
return fee.FeeForVSize(vsize)
}
// waitForSwapComplete waits until a spending tx of the htlc gets confirmed and
// the swap invoice is either settled or canceled. If the htlc times out, the
// timeout tx will be published.
func (s *loopInSwap) waitForSwapComplete(ctx context.Context,
htlcOutpoint *wire.OutPoint, htlcValue btcutil.Amount) error {
// Register the htlc spend notification.
rpcCtx, cancel := context.WithCancel(ctx)
defer cancel()
spendChan, spendErr, err := s.lnd.ChainNotifier.RegisterSpendNtfn(
rpcCtx, htlcOutpoint, s.htlc.PkScript, s.InitiationHeight,
)
if err != nil {
return fmt.Errorf("register spend ntfn: %v", err)
}
// Register for swap invoice updates.
rpcCtx, cancel = context.WithCancel(ctx)
defer cancel()
s.log.Infof("Subscribing to swap invoice %v", s.hash)
swapInvoiceChan, swapInvoiceErr, err := s.lnd.Invoices.SubscribeSingleInvoice(
rpcCtx, s.hash,
)
if err != nil {
return fmt.Errorf("subscribe to swap invoice: %v", err)
}
// publishTxOnTimeout publishes the timeout tx if the contract has
// expired.
publishTxOnTimeout := func() (btcutil.Amount, error) {
if s.height >= s.LoopInContract.CltvExpiry {
return s.publishTimeoutTx(ctx, htlcOutpoint, htlcValue)
}
return 0, nil
}
// Check timeout at current height. After a restart we may want to
// publish the tx immediately.
var sweepFee btcutil.Amount
sweepFee, err = publishTxOnTimeout()
if err != nil {
return err
}
htlcSpend := false
invoiceFinalized := false
htlcKeyRevealed := false
for !htlcSpend || !invoiceFinalized {
select {
// If the client requested the swap to be abandoned, we override
// the status in the database.
case <-s.abandonChan:
return s.setStateAbandoned(ctx)
// Spend notification error.
case err := <-spendErr:
return err
// Receive block epochs and start publishing the timeout tx
// whenever possible.
case notification := <-s.blockEpochChan:
s.height = notification.(int32)
sweepFee, err = publishTxOnTimeout()
if err != nil {
return err
}
if invoiceFinalized && !htlcKeyRevealed {
htlcKeyRevealed = s.tryPushHtlcKey(ctx)
}
// The htlc spend is confirmed. Inspect the spending tx to
// determine the final swap state.
case spendDetails := <-spendChan:
s.log.Infof("Htlc spend by tx: %v",
spendDetails.SpenderTxHash)
err := s.processHtlcSpend(ctx, spendDetails, sweepFee)
if err != nil {
return err
}
htlcSpend = true
// Swap invoice ntfn error.
case err, ok := <-swapInvoiceErr:
// If the channel has been closed, the server has
// finished sending updates, so we set the channel to
// nil because we don't want to constantly select this
// case.
if !ok {
swapInvoiceErr = nil
continue
}
return err
// An update to the swap invoice occurred. Check the new state
// and update the swap state accordingly.
case update, ok := <-swapInvoiceChan:
// If the channel has been closed, the server has
// finished sending updates, so we set the channel to
// nil because we don't want to constantly select this
// case.
if !ok {
swapInvoiceChan = nil
continue
}
s.log.Infof("Received swap invoice update: %v",
update.State)
switch update.State {
// Swap invoice was paid, so update server cost balance.
case invpkg.ContractSettled:
// If invoice settlement and htlc spend happen
// in the expected order, move the swap to an
// intermediate state that indicates that the
// swap is complete from the user point of view,
// but still incomplete with regards to
// accounting data.
if s.state == loopdb.StateHtlcPublished {
s.setState(loopdb.StateInvoiceSettled)
err := s.persistAndAnnounceState(ctx)
if err != nil {
return err
}
}
invoiceFinalized = true
htlcKeyRevealed = s.tryPushHtlcKey(ctx)
s.cost.Server = s.AmountRequested -
update.AmtPaid
// Canceled invoice has no effect on server cost
// balance.
case invpkg.ContractCanceled:
invoiceFinalized = true
}
case <-ctx.Done():
return ctx.Err()
}
}
return nil
}
// tryPushHtlcKey attempts to push the htlc key to the server. If the server
// returns an error of any kind we'll log it as a warning but won't act as the
// swap execution can just go on without the server gaining knowledge of our
// internal key.
func (s *loopInSwap) tryPushHtlcKey(ctx context.Context) bool {
if s.ProtocolVersion < loopdb.ProtocolVersionMuSig2 {
return false
}
log.Infof("Attempting to reveal internal HTLC key to the server")
internalPrivKey, err := sharedSecretFromHash(
ctx, s.swapConfig.lnd.Signer, s.hash,
)
if err != nil {
s.log.Warnf("Unable to derive HTLC internal private key: %v",
err)
return false
}
err = s.server.PushKey(ctx, s.ProtocolVersion, s.hash, internalPrivKey)
if err != nil {
s.log.Warnf("Internal HTLC key reveal failed: %v", err)
return false
}
return true
}
func (s *loopInSwap) processHtlcSpend(ctx context.Context,
spend *chainntnfs.SpendDetail, sweepFee btcutil.Amount) error {
// Determine the htlc input of the spending tx and inspect the witness
// to find out whether a success or a timeout tx spent the htlc.
htlcInput := spend.SpendingTx.TxIn[spend.SpenderInputIndex]
if s.htlc.IsSuccessWitness(htlcInput.Witness) {
s.setState(loopdb.StateSuccess)
} else {
// We needed another on chain tx to sweep the timeout clause,
// which we now include in our costs.
s.cost.Onchain += sweepFee
// If the swap is in state StateFailIncorrectHtlcAmt we know
// that the deposited htlc amount wasn't equal to the contract
// amount. We can finalize the swap by setting an appropriate
// state.
if s.state == loopdb.StateFailIncorrectHtlcAmt {
s.setState(loopdb.StateFailIncorrectHtlcAmtSwept)
} else {
s.setState(loopdb.StateFailTimeout)
}
// Now that the timeout tx confirmed, we can safely cancel the
// swap invoice. We still need to query the final invoice state.
// This is not a hodl invoice, so it may be that the invoice was
// already settled. This means that the server didn't succeed in
// sweeping the htlc after paying the invoice.
err := s.lnd.Invoices.CancelInvoice(ctx, s.hash)
if err != nil && err != invpkg.ErrInvoiceAlreadySettled {
return err
}
}
return nil
}
// publishTimeoutTx publishes a timeout tx after the on-chain htlc has expired,
// returning the fee that is paid by the sweep tx. We cannot update our swap
// costs in this function because it is called multiple times. The swap failed
// and we are reclaiming our funds.
func (s *loopInSwap) publishTimeoutTx(ctx context.Context,
htlcOutpoint *wire.OutPoint, htlcValue btcutil.Amount) (btcutil.Amount,
error) {
if s.timeoutAddr == nil {
var err error
s.timeoutAddr, err = s.lnd.WalletKit.NextAddr(
ctx, "", walletrpc.AddressType_WITNESS_PUBKEY_HASH,
false,
)
if err != nil {
return 0, err
}
}
// Calculate sweep tx fee.
fee, err := s.sweeper.GetSweepFee(
ctx, s.htlc.AddTimeoutToEstimator, s.timeoutAddr,
TimeoutTxConfTarget,
)
if err != nil {
return 0, err
}
// Create a function that will assemble our timeout witness.
witnessFunc := func(sig []byte) (wire.TxWitness, error) {
return s.htlc.GenTimeoutWitness(sig)
}
// Retrieve the full script required to unlock the output.
redeemScript := s.htlc.TimeoutScript()
sequence := uint32(0)
timeoutTx, err := s.sweeper.CreateSweepTx(
ctx, s.height, sequence, s.htlc, *htlcOutpoint,
s.HtlcKeys.SenderScriptKey, redeemScript, witnessFunc,
htlcValue, fee, s.timeoutAddr,
)
if err != nil {
return 0, err
}
timeoutTxHash := timeoutTx.TxHash()
s.log.Infof("Publishing timeout tx %v with fee %v to addr %v",
timeoutTxHash, fee, s.timeoutAddr)
err = s.lnd.WalletKit.PublishTransaction(
ctx, timeoutTx,
labels.LoopInSweepTimeout(swap.ShortHash(&s.hash)),
)
if err != nil {
s.log.Warnf("publish timeout: %v", err)
}
return fee, nil
}
// setStateAbandoned stores the abandoned state and announces it. It also
// cancels the swap invoice so the server can't settle it.
func (s *loopInSwap) setStateAbandoned(ctx context.Context) error {
s.log.Infof("Abandoning swap %v...", s.hash)
if !s.state.IsPending() {
return fmt.Errorf("cannot abandon swap in state %v", s.state)
}
s.setState(loopdb.StateFailAbandoned)
err := s.persistAndAnnounceState(ctx)
if err != nil {
return err
}
// If the invoice is already settled or canceled, this is a nop.
_ = s.lnd.Invoices.CancelInvoice(ctx, s.hash)
return fmt.Errorf("swap hash "+
"abandoned by client, "+
"swap ID: %v, %v",
s.hash, err)
}
// persistAndAnnounceState updates the swap state on disk and sends out an
// update notification.
func (s *loopInSwap) persistAndAnnounceState(ctx context.Context) error {
// Update state in store.
if err := s.persistState(ctx); err != nil {
return err
}
// Send out swap update
return s.sendUpdate(ctx)
}
// persistState updates the swap state on disk.
func (s *loopInSwap) persistState(ctx context.Context) error {
return s.store.UpdateLoopIn(
ctx, s.hash, s.lastUpdateTime,
loopdb.SwapStateData{
State: s.state,
Cost: s.cost,
HtlcTxHash: s.htlcTxHash,
},
)
}
// setState updates the swap state and last update timestamp.
func (s *loopInSwap) setState(state loopdb.SwapState) {
s.lastUpdateTime = time.Now()
s.state = state
}
// sharedSecretFromHash derives the shared secret from the swap hash using the
// swap.KeyFamily family and zero as index.
func sharedSecretFromHash(ctx context.Context, signer lndclient.SignerClient,
hash lntypes.Hash) ([32]byte, error) {
_, hashPubKey := btcec.PrivKeyFromBytes(hash[:])
return signer.DeriveSharedKey(
ctx, hashPubKey, &keychain.KeyLocator{
Family: keychain.KeyFamily(swap.KeyFamily),
Index: 0,
},
)
}