package contractcourt import ( "errors" "fmt" "io/ioutil" "os" "path/filepath" "reflect" "sync" "testing" "time" "github.com/btcsuite/btcd/chaincfg/chainhash" "github.com/btcsuite/btcd/wire" "github.com/btcsuite/btcutil" "github.com/lightningnetwork/lnd/chainntnfs" "github.com/lightningnetwork/lnd/channeldb" "github.com/lightningnetwork/lnd/channeldb/kvdb" "github.com/lightningnetwork/lnd/clock" "github.com/lightningnetwork/lnd/input" "github.com/lightningnetwork/lnd/lnwallet" "github.com/lightningnetwork/lnd/lnwire" ) const ( defaultTimeout = time.Second * 5 // stateTimeout is the timeout we allow when waiting for state // transitions. stateTimeout = time.Second * 15 ) type mockArbitratorLog struct { state ArbitratorState newStates chan ArbitratorState failLog bool failFetch error failCommit bool failCommitState ArbitratorState resolutions *ContractResolutions resolvers map[ContractResolver]struct{} commitSet *CommitSet sync.Mutex } // A compile time check to ensure mockArbitratorLog meets the ArbitratorLog // interface. var _ ArbitratorLog = (*mockArbitratorLog)(nil) func (b *mockArbitratorLog) CurrentState() (ArbitratorState, error) { return b.state, nil } func (b *mockArbitratorLog) CommitState(s ArbitratorState) error { if b.failCommit && s == b.failCommitState { return fmt.Errorf("intentional commit error at state %v", b.failCommitState) } b.state = s b.newStates <- s return nil } func (b *mockArbitratorLog) FetchUnresolvedContracts() ([]ContractResolver, error) { b.Lock() v := make([]ContractResolver, len(b.resolvers)) idx := 0 for resolver := range b.resolvers { v[idx] = resolver idx++ } b.Unlock() return v, nil } func (b *mockArbitratorLog) InsertUnresolvedContracts(_ []*channeldb.ResolverReport, resolvers ...ContractResolver) error { b.Lock() for _, resolver := range resolvers { resKey := resolver.ResolverKey() if resKey == nil { continue } b.resolvers[resolver] = struct{}{} } b.Unlock() return nil } func (b *mockArbitratorLog) SwapContract(oldContract, newContract ContractResolver) error { b.Lock() delete(b.resolvers, oldContract) b.resolvers[newContract] = struct{}{} b.Unlock() return nil } func (b *mockArbitratorLog) ResolveContract(res ContractResolver) error { b.Lock() delete(b.resolvers, res) b.Unlock() return nil } func (b *mockArbitratorLog) LogContractResolutions(c *ContractResolutions) error { if b.failLog { return fmt.Errorf("intentional log failure") } b.resolutions = c return nil } func (b *mockArbitratorLog) FetchContractResolutions() (*ContractResolutions, error) { if b.failFetch != nil { return nil, b.failFetch } return b.resolutions, nil } func (b *mockArbitratorLog) FetchChainActions() (ChainActionMap, error) { return nil, nil } func (b *mockArbitratorLog) InsertConfirmedCommitSet(c *CommitSet) error { b.commitSet = c return nil } func (b *mockArbitratorLog) FetchConfirmedCommitSet() (*CommitSet, error) { return b.commitSet, nil } func (b *mockArbitratorLog) WipeHistory() error { return nil } // testArbLog is a wrapper around an existing (ideally fully concrete // ArbitratorLog) that lets us intercept certain calls like transitioning to a // new state. type testArbLog struct { ArbitratorLog newStates chan ArbitratorState } func (t *testArbLog) CommitState(s ArbitratorState) error { if err := t.ArbitratorLog.CommitState(s); err != nil { return err } t.newStates <- s return nil } type mockChainIO struct{} var _ lnwallet.BlockChainIO = (*mockChainIO)(nil) func (*mockChainIO) GetBestBlock() (*chainhash.Hash, int32, error) { return nil, 0, nil } func (*mockChainIO) GetUtxo(op *wire.OutPoint, _ []byte, heightHint uint32, _ <-chan struct{}) (*wire.TxOut, error) { return nil, nil } func (*mockChainIO) GetBlockHash(blockHeight int64) (*chainhash.Hash, error) { return nil, nil } func (*mockChainIO) GetBlock(blockHash *chainhash.Hash) (*wire.MsgBlock, error) { return nil, nil } type chanArbTestCtx struct { t *testing.T chanArb *ChannelArbitrator cleanUp func() resolvedChan chan struct{} blockEpochs chan *chainntnfs.BlockEpoch incubationRequests chan struct{} resolutions chan []ResolutionMsg log ArbitratorLog sweeper *mockSweeper } func (c *chanArbTestCtx) CleanUp() { if err := c.chanArb.Stop(); err != nil { c.t.Fatalf("unable to stop chan arb: %v", err) } if c.cleanUp != nil { c.cleanUp() } } // AssertStateTransitions asserts that the state machine steps through the // passed states in order. func (c *chanArbTestCtx) AssertStateTransitions(expectedStates ...ArbitratorState) { c.t.Helper() var newStatesChan chan ArbitratorState switch log := c.log.(type) { case *mockArbitratorLog: newStatesChan = log.newStates case *testArbLog: newStatesChan = log.newStates default: c.t.Fatalf("unable to assert state transitions with %T", log) } for _, exp := range expectedStates { var state ArbitratorState select { case state = <-newStatesChan: case <-time.After(defaultTimeout): c.t.Fatalf("new state not received") } if state != exp { c.t.Fatalf("expected new state %v, got %v", exp, state) } } } // AssertState checks that the ChannelArbitrator is in the state we expect it // to be. func (c *chanArbTestCtx) AssertState(expected ArbitratorState) { if c.chanArb.state != expected { c.t.Fatalf("expected state %v, was %v", expected, c.chanArb.state) } } // Restart simulates a clean restart of the channel arbitrator, forcing it to // walk through it's recovery logic. If this function returns nil, then a // restart was successful. Note that the restart process keeps the log in // place, in order to simulate proper persistence of the log. The caller can // optionally provide a restart closure which will be executed before the // resolver is started again, but after it is created. func (c *chanArbTestCtx) Restart(restartClosure func(*chanArbTestCtx)) (*chanArbTestCtx, error) { if err := c.chanArb.Stop(); err != nil { return nil, err } newCtx, err := createTestChannelArbitrator(c.t, c.log) if err != nil { return nil, err } if restartClosure != nil { restartClosure(newCtx) } if err := newCtx.chanArb.Start(); err != nil { return nil, err } return newCtx, nil } // testChanArbOption applies custom settings to a channel arbitrator config for // testing purposes. type testChanArbOption func(cfg *ChannelArbitratorConfig) // remoteInitiatorOption sets the MarkChannelClosed function in the // Channel Arbitrator's config. func withMarkClosed(markClosed func(*channeldb.ChannelCloseSummary, ...channeldb.ChannelStatus) error) testChanArbOption { return func(cfg *ChannelArbitratorConfig) { cfg.MarkChannelClosed = markClosed } } // createTestChannelArbitrator returns a channel arbitrator test context which // contains a channel arbitrator with default values. These values can be // changed by providing options which overwrite the default config. func createTestChannelArbitrator(t *testing.T, log ArbitratorLog, opts ...testChanArbOption) (*chanArbTestCtx, error) { blockEpochs := make(chan *chainntnfs.BlockEpoch) blockEpoch := &chainntnfs.BlockEpochEvent{ Epochs: blockEpochs, Cancel: func() {}, } chanPoint := wire.OutPoint{} shortChanID := lnwire.ShortChannelID{} chanEvents := &ChainEventSubscription{ RemoteUnilateralClosure: make(chan *RemoteUnilateralCloseInfo, 1), LocalUnilateralClosure: make(chan *LocalUnilateralCloseInfo, 1), CooperativeClosure: make(chan *CooperativeCloseInfo, 1), ContractBreach: make(chan *lnwallet.BreachRetribution, 1), } resolutionChan := make(chan []ResolutionMsg, 1) incubateChan := make(chan struct{}) chainIO := &mockChainIO{} mockSweeper := newMockSweeper() chainArbCfg := ChainArbitratorConfig{ ChainIO: chainIO, PublishTx: func(*wire.MsgTx, string) error { return nil }, DeliverResolutionMsg: func(msgs ...ResolutionMsg) error { resolutionChan <- msgs return nil }, OutgoingBroadcastDelta: 5, IncomingBroadcastDelta: 5, Notifier: &mockNotifier{ epochChan: make(chan *chainntnfs.BlockEpoch), spendChan: make(chan *chainntnfs.SpendDetail), confChan: make(chan *chainntnfs.TxConfirmation), }, IncubateOutputs: func(wire.OutPoint, *lnwallet.OutgoingHtlcResolution, *lnwallet.IncomingHtlcResolution, uint32) error { incubateChan <- struct{}{} return nil }, OnionProcessor: &mockOnionProcessor{}, IsForwardedHTLC: func(chanID lnwire.ShortChannelID, htlcIndex uint64) bool { return true }, Clock: clock.NewDefaultClock(), Sweeper: mockSweeper, } // We'll use the resolvedChan to synchronize on call to // MarkChannelResolved. resolvedChan := make(chan struct{}, 1) // Next we'll create the matching configuration struct that contains // all interfaces and methods the arbitrator needs to do its job. arbCfg := &ChannelArbitratorConfig{ ChanPoint: chanPoint, ShortChanID: shortChanID, BlockEpochs: blockEpoch, MarkChannelResolved: func() error { resolvedChan <- struct{}{} return nil }, Channel: &mockChannel{}, MarkCommitmentBroadcasted: func(_ *wire.MsgTx, _ bool) error { return nil }, MarkChannelClosed: func(*channeldb.ChannelCloseSummary, ...channeldb.ChannelStatus) error { return nil }, IsPendingClose: false, ChainArbitratorConfig: chainArbCfg, ChainEvents: chanEvents, PutResolverReport: func(_ kvdb.RwTx, _ *channeldb.ResolverReport) error { return nil }, } // Apply all custom options to the config struct. for _, option := range opts { option(arbCfg) } var cleanUp func() if log == nil { dbDir, err := ioutil.TempDir("", "chanArb") if err != nil { return nil, err } dbPath := filepath.Join(dbDir, "testdb") db, err := kvdb.Create(kvdb.BoltBackendName, dbPath, true) if err != nil { return nil, err } backingLog, err := newBoltArbitratorLog( db, *arbCfg, chainhash.Hash{}, chanPoint, ) if err != nil { return nil, err } cleanUp = func() { db.Close() os.RemoveAll(dbDir) } log = &testArbLog{ ArbitratorLog: backingLog, newStates: make(chan ArbitratorState), } } htlcSets := make(map[HtlcSetKey]htlcSet) chanArb := NewChannelArbitrator(*arbCfg, htlcSets, log) return &chanArbTestCtx{ t: t, chanArb: chanArb, cleanUp: cleanUp, resolvedChan: resolvedChan, resolutions: resolutionChan, blockEpochs: blockEpochs, log: log, incubationRequests: incubateChan, sweeper: mockSweeper, }, nil } // TestChannelArbitratorCooperativeClose tests that the ChannelArbitertor // correctly marks the channel resolved in case a cooperative close is // confirmed. func TestChannelArbitratorCooperativeClose(t *testing.T) { log := &mockArbitratorLog{ state: StateDefault, newStates: make(chan ArbitratorState, 5), } chanArbCtx, err := createTestChannelArbitrator(t, log) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } if err := chanArbCtx.chanArb.Start(); err != nil { t.Fatalf("unable to start ChannelArbitrator: %v", err) } defer func() { if err := chanArbCtx.chanArb.Stop(); err != nil { t.Fatalf("unable to stop chan arb: %v", err) } }() // It should start out in the default state. chanArbCtx.AssertState(StateDefault) // We set up a channel to detect when MarkChannelClosed is called. closeInfos := make(chan *channeldb.ChannelCloseSummary) chanArbCtx.chanArb.cfg.MarkChannelClosed = func( closeInfo *channeldb.ChannelCloseSummary, statuses ...channeldb.ChannelStatus) error { closeInfos <- closeInfo return nil } // Cooperative close should do trigger a MarkChannelClosed + // MarkChannelResolved. closeInfo := &CooperativeCloseInfo{ &channeldb.ChannelCloseSummary{}, } chanArbCtx.chanArb.cfg.ChainEvents.CooperativeClosure <- closeInfo select { case c := <-closeInfos: if c.CloseType != channeldb.CooperativeClose { t.Fatalf("expected cooperative close, got %v", c.CloseType) } case <-time.After(defaultTimeout): t.Fatalf("timeout waiting for channel close") } // It should mark the channel as resolved. select { case <-chanArbCtx.resolvedChan: // Expected. case <-time.After(defaultTimeout): t.Fatalf("contract was not resolved") } } // TestChannelArbitratorRemoteForceClose checks that the ChannelArbitrator goes // through the expected states if a remote force close is observed in the // chain. func TestChannelArbitratorRemoteForceClose(t *testing.T) { log := &mockArbitratorLog{ state: StateDefault, newStates: make(chan ArbitratorState, 5), } chanArbCtx, err := createTestChannelArbitrator(t, log) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } chanArb := chanArbCtx.chanArb if err := chanArb.Start(); err != nil { t.Fatalf("unable to start ChannelArbitrator: %v", err) } defer chanArb.Stop() // It should start out in the default state. chanArbCtx.AssertState(StateDefault) // Send a remote force close event. commitSpend := &chainntnfs.SpendDetail{ SpenderTxHash: &chainhash.Hash{}, } uniClose := &lnwallet.UnilateralCloseSummary{ SpendDetail: commitSpend, HtlcResolutions: &lnwallet.HtlcResolutions{}, } chanArb.cfg.ChainEvents.RemoteUnilateralClosure <- &RemoteUnilateralCloseInfo{ UnilateralCloseSummary: uniClose, CommitSet: CommitSet{ ConfCommitKey: &RemoteHtlcSet, HtlcSets: make(map[HtlcSetKey][]channeldb.HTLC), }, } // It should transition StateDefault -> StateContractClosed -> // StateFullyResolved. chanArbCtx.AssertStateTransitions( StateContractClosed, StateFullyResolved, ) // It should also mark the channel as resolved. select { case <-chanArbCtx.resolvedChan: // Expected. case <-time.After(defaultTimeout): t.Fatalf("contract was not resolved") } } // TestChannelArbitratorLocalForceClose tests that the ChannelArbitrator goes // through the expected states in case we request it to force close the channel, // and the local force close event is observed in chain. func TestChannelArbitratorLocalForceClose(t *testing.T) { log := &mockArbitratorLog{ state: StateDefault, newStates: make(chan ArbitratorState, 5), } chanArbCtx, err := createTestChannelArbitrator(t, log) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } chanArb := chanArbCtx.chanArb if err := chanArb.Start(); err != nil { t.Fatalf("unable to start ChannelArbitrator: %v", err) } defer chanArb.Stop() // It should start out in the default state. chanArbCtx.AssertState(StateDefault) // We create a channel we can use to pause the ChannelArbitrator at the // point where it broadcasts the close tx, and check its state. stateChan := make(chan ArbitratorState) chanArb.cfg.PublishTx = func(*wire.MsgTx, string) error { // When the force close tx is being broadcasted, check that the // state is correct at that point. select { case stateChan <- chanArb.state: case <-chanArb.quit: return fmt.Errorf("exiting") } return nil } errChan := make(chan error, 1) respChan := make(chan *wire.MsgTx, 1) // With the channel found, and the request crafted, we'll send over a // force close request to the arbitrator that watches this channel. chanArb.forceCloseReqs <- &forceCloseReq{ errResp: errChan, closeTx: respChan, } // It should transition to StateBroadcastCommit. chanArbCtx.AssertStateTransitions(StateBroadcastCommit) // When it is broadcasting the force close, its state should be // StateBroadcastCommit. select { case state := <-stateChan: if state != StateBroadcastCommit { t.Fatalf("state during PublishTx was %v", state) } case <-time.After(stateTimeout): t.Fatalf("did not get state update") } // After broadcasting, transition should be to // StateCommitmentBroadcasted. chanArbCtx.AssertStateTransitions(StateCommitmentBroadcasted) select { case <-respChan: case <-time.After(defaultTimeout): t.Fatalf("no response received") } select { case err := <-errChan: if err != nil { t.Fatalf("error force closing channel: %v", err) } case <-time.After(defaultTimeout): t.Fatalf("no response received") } // After broadcasting the close tx, it should be in state // StateCommitmentBroadcasted. chanArbCtx.AssertState(StateCommitmentBroadcasted) // Now notify about the local force close getting confirmed. chanArb.cfg.ChainEvents.LocalUnilateralClosure <- &LocalUnilateralCloseInfo{ SpendDetail: &chainntnfs.SpendDetail{}, LocalForceCloseSummary: &lnwallet.LocalForceCloseSummary{ CloseTx: &wire.MsgTx{}, HtlcResolutions: &lnwallet.HtlcResolutions{}, }, ChannelCloseSummary: &channeldb.ChannelCloseSummary{}, } // It should transition StateContractClosed -> StateFullyResolved. chanArbCtx.AssertStateTransitions(StateContractClosed, StateFullyResolved) // It should also mark the channel as resolved. select { case <-chanArbCtx.resolvedChan: // Expected. case <-time.After(defaultTimeout): t.Fatalf("contract was not resolved") } } // TestChannelArbitratorBreachClose tests that the ChannelArbitrator goes // through the expected states in case we notice a breach in the chain, and // gracefully exits. func TestChannelArbitratorBreachClose(t *testing.T) { log := &mockArbitratorLog{ state: StateDefault, newStates: make(chan ArbitratorState, 5), } chanArbCtx, err := createTestChannelArbitrator(t, log) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } chanArb := chanArbCtx.chanArb if err := chanArb.Start(); err != nil { t.Fatalf("unable to start ChannelArbitrator: %v", err) } defer func() { if err := chanArb.Stop(); err != nil { t.Fatal(err) } }() // It should start out in the default state. chanArbCtx.AssertState(StateDefault) // Send a breach close event. chanArb.cfg.ChainEvents.ContractBreach <- &lnwallet.BreachRetribution{} // It should transition StateDefault -> StateFullyResolved. chanArbCtx.AssertStateTransitions( StateFullyResolved, ) // It should also mark the channel as resolved. select { case <-chanArbCtx.resolvedChan: // Expected. case <-time.After(defaultTimeout): t.Fatalf("contract was not resolved") } } // TestChannelArbitratorLocalForceClosePendingHtlc tests that the // ChannelArbitrator goes through the expected states in case we request it to // force close a channel that still has an HTLC pending. func TestChannelArbitratorLocalForceClosePendingHtlc(t *testing.T) { // We create a new test context for this channel arb, notice that we // pass in a nil ArbitratorLog which means that a default one backed by // a real DB will be created. We need this for our test as we want to // test proper restart recovery and resolver population. chanArbCtx, err := createTestChannelArbitrator(t, nil) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } chanArb := chanArbCtx.chanArb chanArb.cfg.PreimageDB = newMockWitnessBeacon() chanArb.cfg.Registry = &mockRegistry{} if err := chanArb.Start(); err != nil { t.Fatalf("unable to start ChannelArbitrator: %v", err) } defer chanArb.Stop() // Create htlcUpdates channel. htlcUpdates := make(chan *ContractUpdate) signals := &ContractSignals{ HtlcUpdates: htlcUpdates, ShortChanID: lnwire.ShortChannelID{}, } chanArb.UpdateContractSignals(signals) // Add HTLC to channel arbitrator. htlcAmt := 10000 htlc := channeldb.HTLC{ Incoming: false, Amt: lnwire.MilliSatoshi(htlcAmt), HtlcIndex: 99, } outgoingDustHtlc := channeldb.HTLC{ Incoming: false, Amt: 100, HtlcIndex: 100, OutputIndex: -1, } incomingDustHtlc := channeldb.HTLC{ Incoming: true, Amt: 105, HtlcIndex: 101, OutputIndex: -1, } htlcSet := []channeldb.HTLC{ htlc, outgoingDustHtlc, incomingDustHtlc, } htlcUpdates <- &ContractUpdate{ HtlcKey: LocalHtlcSet, Htlcs: htlcSet, } errChan := make(chan error, 1) respChan := make(chan *wire.MsgTx, 1) // With the channel found, and the request crafted, we'll send over a // force close request to the arbitrator that watches this channel. chanArb.forceCloseReqs <- &forceCloseReq{ errResp: errChan, closeTx: respChan, } // The force close request should trigger broadcast of the commitment // transaction. chanArbCtx.AssertStateTransitions( StateBroadcastCommit, StateCommitmentBroadcasted, ) select { case <-respChan: case <-time.After(defaultTimeout): t.Fatalf("no response received") } select { case err := <-errChan: if err != nil { t.Fatalf("error force closing channel: %v", err) } case <-time.After(defaultTimeout): t.Fatalf("no response received") } // Now notify about the local force close getting confirmed. closeTx := &wire.MsgTx{ TxIn: []*wire.TxIn{ { PreviousOutPoint: wire.OutPoint{}, Witness: [][]byte{ {0x1}, {0x2}, }, }, }, } htlcOp := wire.OutPoint{ Hash: closeTx.TxHash(), Index: 0, } // Set up the outgoing resolution. Populate SignedTimeoutTx because our // commitment transaction got confirmed. outgoingRes := lnwallet.OutgoingHtlcResolution{ Expiry: 10, SweepSignDesc: input.SignDescriptor{ Output: &wire.TxOut{}, }, SignedTimeoutTx: &wire.MsgTx{ TxIn: []*wire.TxIn{ { PreviousOutPoint: htlcOp, Witness: [][]byte{{}}, }, }, TxOut: []*wire.TxOut{ {}, }, }, } chanArb.cfg.ChainEvents.LocalUnilateralClosure <- &LocalUnilateralCloseInfo{ SpendDetail: &chainntnfs.SpendDetail{}, LocalForceCloseSummary: &lnwallet.LocalForceCloseSummary{ CloseTx: closeTx, HtlcResolutions: &lnwallet.HtlcResolutions{ OutgoingHTLCs: []lnwallet.OutgoingHtlcResolution{ outgoingRes, }, }, }, ChannelCloseSummary: &channeldb.ChannelCloseSummary{}, CommitSet: CommitSet{ ConfCommitKey: &LocalHtlcSet, HtlcSets: map[HtlcSetKey][]channeldb.HTLC{ LocalHtlcSet: htlcSet, }, }, } chanArbCtx.AssertStateTransitions( StateContractClosed, StateWaitingFullResolution, ) // We expect an immediate resolution message for the outgoing dust htlc. // It is not resolvable on-chain. select { case msgs := <-chanArbCtx.resolutions: if len(msgs) != 1 { t.Fatalf("expected 1 message, instead got %v", len(msgs)) } if msgs[0].HtlcIndex != outgoingDustHtlc.HtlcIndex { t.Fatalf("wrong htlc index: expected %v, got %v", outgoingDustHtlc.HtlcIndex, msgs[0].HtlcIndex) } case <-time.After(defaultTimeout): t.Fatalf("resolution msgs not sent") } // We'll grab the old notifier here as our resolvers are still holding // a reference to this instance, and a new one will be created when we // restart the channel arb below. oldNotifier := chanArb.cfg.Notifier.(*mockNotifier) // At this point, in order to simulate a restart, we'll re-create the // channel arbitrator. We do this to ensure that all information // required to properly resolve this HTLC are populated. if err := chanArb.Stop(); err != nil { t.Fatalf("unable to stop chan arb: %v", err) } // We'll no re-create the resolver, notice that we use the existing // arbLog so it carries over the same on-disk state. chanArbCtxNew, err := chanArbCtx.Restart(nil) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } chanArb = chanArbCtxNew.chanArb defer chanArbCtxNew.CleanUp() // Post restart, it should be the case that our resolver was properly // supplemented, and we only have a single resolver in the final set. if len(chanArb.activeResolvers) != 1 { t.Fatalf("expected single resolver, instead got: %v", len(chanArb.activeResolvers)) } // We'll now examine the in-memory state of the active resolvers to // ensure t hey were populated properly. resolver := chanArb.activeResolvers[0] outgoingResolver, ok := resolver.(*htlcOutgoingContestResolver) if !ok { t.Fatalf("expected outgoing contest resolver, got %vT", resolver) } // The resolver should have its htlc amt field populated as it. if int64(outgoingResolver.htlc.Amt) != int64(htlcAmt) { t.Fatalf("wrong htlc amount: expected %v, got %v,", htlcAmt, int64(outgoingResolver.htlc.Amt)) } // htlcOutgoingContestResolver is now active and waiting for the HTLC to // expire. It should not yet have passed it on for incubation. select { case <-chanArbCtx.incubationRequests: t.Fatalf("contract should not be incubated yet") default: } // Send a notification that the expiry height has been reached. oldNotifier.epochChan <- &chainntnfs.BlockEpoch{Height: 10} // htlcOutgoingContestResolver is now transforming into a // htlcTimeoutResolver and should send the contract off for incubation. select { case <-chanArbCtx.incubationRequests: case <-time.After(defaultTimeout): t.Fatalf("no response received") } // Notify resolver that the HTLC output of the commitment has been // spent. oldNotifier.spendChan <- &chainntnfs.SpendDetail{SpendingTx: closeTx} // Finally, we should also receive a resolution message instructing the // switch to cancel back the HTLC. select { case msgs := <-chanArbCtx.resolutions: if len(msgs) != 1 { t.Fatalf("expected 1 message, instead got %v", len(msgs)) } if msgs[0].HtlcIndex != htlc.HtlcIndex { t.Fatalf("wrong htlc index: expected %v, got %v", htlc.HtlcIndex, msgs[0].HtlcIndex) } case <-time.After(defaultTimeout): t.Fatalf("resolution msgs not sent") } // As this is our own commitment transaction, the HTLC will go through // to the second level. Channel arbitrator should still not be marked // as resolved. select { case <-chanArbCtxNew.resolvedChan: t.Fatalf("channel resolved prematurely") default: } // Notify resolver that the second level transaction is spent. oldNotifier.spendChan <- &chainntnfs.SpendDetail{SpendingTx: closeTx} // At this point channel should be marked as resolved. chanArbCtxNew.AssertStateTransitions(StateFullyResolved) select { case <-chanArbCtxNew.resolvedChan: case <-time.After(defaultTimeout): t.Fatalf("contract was not resolved") } } // TestChannelArbitratorLocalForceCloseRemoteConfiremd tests that the // ChannelArbitrator behaves as expected in the case where we request a local // force close, but a remote commitment ends up being confirmed in chain. func TestChannelArbitratorLocalForceCloseRemoteConfirmed(t *testing.T) { log := &mockArbitratorLog{ state: StateDefault, newStates: make(chan ArbitratorState, 5), } chanArbCtx, err := createTestChannelArbitrator(t, log) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } chanArb := chanArbCtx.chanArb if err := chanArb.Start(); err != nil { t.Fatalf("unable to start ChannelArbitrator: %v", err) } defer chanArb.Stop() // It should start out in the default state. chanArbCtx.AssertState(StateDefault) // Create a channel we can use to assert the state when it publishes // the close tx. stateChan := make(chan ArbitratorState) chanArb.cfg.PublishTx = func(*wire.MsgTx, string) error { // When the force close tx is being broadcasted, check that the // state is correct at that point. select { case stateChan <- chanArb.state: case <-chanArb.quit: return fmt.Errorf("exiting") } return nil } errChan := make(chan error, 1) respChan := make(chan *wire.MsgTx, 1) // With the channel found, and the request crafted, we'll send over a // force close request to the arbitrator that watches this channel. chanArb.forceCloseReqs <- &forceCloseReq{ errResp: errChan, closeTx: respChan, } // It should transition to StateBroadcastCommit. chanArbCtx.AssertStateTransitions(StateBroadcastCommit) // We expect it to be in state StateBroadcastCommit when publishing // the force close. select { case state := <-stateChan: if state != StateBroadcastCommit { t.Fatalf("state during PublishTx was %v", state) } case <-time.After(stateTimeout): t.Fatalf("no state update received") } // After broadcasting, transition should be to // StateCommitmentBroadcasted. chanArbCtx.AssertStateTransitions(StateCommitmentBroadcasted) // Wait for a response to the force close. select { case <-respChan: case <-time.After(defaultTimeout): t.Fatalf("no response received") } select { case err := <-errChan: if err != nil { t.Fatalf("error force closing channel: %v", err) } case <-time.After(defaultTimeout): t.Fatalf("no response received") } // The state should be StateCommitmentBroadcasted. chanArbCtx.AssertState(StateCommitmentBroadcasted) // Now notify about the _REMOTE_ commitment getting confirmed. commitSpend := &chainntnfs.SpendDetail{ SpenderTxHash: &chainhash.Hash{}, } uniClose := &lnwallet.UnilateralCloseSummary{ SpendDetail: commitSpend, HtlcResolutions: &lnwallet.HtlcResolutions{}, } chanArb.cfg.ChainEvents.RemoteUnilateralClosure <- &RemoteUnilateralCloseInfo{ UnilateralCloseSummary: uniClose, } // It should transition StateContractClosed -> StateFullyResolved. chanArbCtx.AssertStateTransitions(StateContractClosed, StateFullyResolved) // It should resolve. select { case <-chanArbCtx.resolvedChan: // Expected. case <-time.After(stateTimeout): t.Fatalf("contract was not resolved") } } // TestChannelArbitratorLocalForceCloseDoubleSpend tests that the // ChannelArbitrator behaves as expected in the case where we request a local // force close, but we fail broadcasting our commitment because a remote // commitment has already been published. func TestChannelArbitratorLocalForceDoubleSpend(t *testing.T) { log := &mockArbitratorLog{ state: StateDefault, newStates: make(chan ArbitratorState, 5), } chanArbCtx, err := createTestChannelArbitrator(t, log) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } chanArb := chanArbCtx.chanArb if err := chanArb.Start(); err != nil { t.Fatalf("unable to start ChannelArbitrator: %v", err) } defer chanArb.Stop() // It should start out in the default state. chanArbCtx.AssertState(StateDefault) // Return ErrDoubleSpend when attempting to publish the tx. stateChan := make(chan ArbitratorState) chanArb.cfg.PublishTx = func(*wire.MsgTx, string) error { // When the force close tx is being broadcasted, check that the // state is correct at that point. select { case stateChan <- chanArb.state: case <-chanArb.quit: return fmt.Errorf("exiting") } return lnwallet.ErrDoubleSpend } errChan := make(chan error, 1) respChan := make(chan *wire.MsgTx, 1) // With the channel found, and the request crafted, we'll send over a // force close request to the arbitrator that watches this channel. chanArb.forceCloseReqs <- &forceCloseReq{ errResp: errChan, closeTx: respChan, } // It should transition to StateBroadcastCommit. chanArbCtx.AssertStateTransitions(StateBroadcastCommit) // We expect it to be in state StateBroadcastCommit when publishing // the force close. select { case state := <-stateChan: if state != StateBroadcastCommit { t.Fatalf("state during PublishTx was %v", state) } case <-time.After(stateTimeout): t.Fatalf("no state update received") } // After broadcasting, transition should be to // StateCommitmentBroadcasted. chanArbCtx.AssertStateTransitions(StateCommitmentBroadcasted) // Wait for a response to the force close. select { case <-respChan: case <-time.After(defaultTimeout): t.Fatalf("no response received") } select { case err := <-errChan: if err != nil { t.Fatalf("error force closing channel: %v", err) } case <-time.After(defaultTimeout): t.Fatalf("no response received") } // The state should be StateCommitmentBroadcasted. chanArbCtx.AssertState(StateCommitmentBroadcasted) // Now notify about the _REMOTE_ commitment getting confirmed. commitSpend := &chainntnfs.SpendDetail{ SpenderTxHash: &chainhash.Hash{}, } uniClose := &lnwallet.UnilateralCloseSummary{ SpendDetail: commitSpend, HtlcResolutions: &lnwallet.HtlcResolutions{}, } chanArb.cfg.ChainEvents.RemoteUnilateralClosure <- &RemoteUnilateralCloseInfo{ UnilateralCloseSummary: uniClose, } // It should transition StateContractClosed -> StateFullyResolved. chanArbCtx.AssertStateTransitions(StateContractClosed, StateFullyResolved) // It should resolve. select { case <-chanArbCtx.resolvedChan: // Expected. case <-time.After(stateTimeout): t.Fatalf("contract was not resolved") } } // TestChannelArbitratorPersistence tests that the ChannelArbitrator is able to // keep advancing the state machine from various states after restart. func TestChannelArbitratorPersistence(t *testing.T) { // Start out with a log that will fail writing the set of resolutions. log := &mockArbitratorLog{ state: StateDefault, newStates: make(chan ArbitratorState, 5), failLog: true, } chanArbCtx, err := createTestChannelArbitrator(t, log) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } chanArb := chanArbCtx.chanArb if err := chanArb.Start(); err != nil { t.Fatalf("unable to start ChannelArbitrator: %v", err) } // It should start in StateDefault. chanArbCtx.AssertState(StateDefault) // Send a remote force close event. commitSpend := &chainntnfs.SpendDetail{ SpenderTxHash: &chainhash.Hash{}, } uniClose := &lnwallet.UnilateralCloseSummary{ SpendDetail: commitSpend, HtlcResolutions: &lnwallet.HtlcResolutions{}, } chanArb.cfg.ChainEvents.RemoteUnilateralClosure <- &RemoteUnilateralCloseInfo{ UnilateralCloseSummary: uniClose, } // Since writing the resolutions fail, the arbitrator should not // advance to the next state. time.Sleep(100 * time.Millisecond) if log.state != StateDefault { t.Fatalf("expected to stay in StateDefault") } // Restart the channel arb, this'll use the same long and prior // context. chanArbCtx, err = chanArbCtx.Restart(nil) if err != nil { t.Fatalf("unable to restart channel arb: %v", err) } chanArb = chanArbCtx.chanArb // Again, it should start up in the default state. chanArbCtx.AssertState(StateDefault) // Now we make the log succeed writing the resolutions, but fail when // attempting to close the channel. log.failLog = false chanArb.cfg.MarkChannelClosed = func(*channeldb.ChannelCloseSummary, ...channeldb.ChannelStatus) error { return fmt.Errorf("intentional close error") } // Send a new remote force close event. chanArb.cfg.ChainEvents.RemoteUnilateralClosure <- &RemoteUnilateralCloseInfo{ UnilateralCloseSummary: uniClose, } // Since closing the channel failed, the arbitrator should stay in the // default state. time.Sleep(100 * time.Millisecond) if log.state != StateDefault { t.Fatalf("expected to stay in StateDefault") } // Restart once again to simulate yet another restart. chanArbCtx, err = chanArbCtx.Restart(nil) if err != nil { t.Fatalf("unable to restart channel arb: %v", err) } chanArb = chanArbCtx.chanArb // Starts out in StateDefault. chanArbCtx.AssertState(StateDefault) // Now make fetching the resolutions fail. log.failFetch = fmt.Errorf("intentional fetch failure") chanArb.cfg.ChainEvents.RemoteUnilateralClosure <- &RemoteUnilateralCloseInfo{ UnilateralCloseSummary: uniClose, } // Since logging the resolutions and closing the channel now succeeds, // it should advance to StateContractClosed. chanArbCtx.AssertStateTransitions(StateContractClosed) // It should not advance further, however, as fetching resolutions // failed. time.Sleep(100 * time.Millisecond) if log.state != StateContractClosed { t.Fatalf("expected to stay in StateContractClosed") } chanArb.Stop() // Create a new arbitrator, and now make fetching resolutions succeed. log.failFetch = nil chanArbCtx, err = chanArbCtx.Restart(nil) if err != nil { t.Fatalf("unable to restart channel arb: %v", err) } defer chanArbCtx.CleanUp() // Finally it should advance to StateFullyResolved. chanArbCtx.AssertStateTransitions(StateFullyResolved) // It should also mark the channel as resolved. select { case <-chanArbCtx.resolvedChan: // Expected. case <-time.After(defaultTimeout): t.Fatalf("contract was not resolved") } } // TestChannelArbitratorForceCloseBreachedChannel tests that the channel // arbitrator is able to handle a channel in the process of being force closed // is breached by the remote node. In these cases we expect the // ChannelArbitrator to gracefully exit, as the breach is handled by other // subsystems. func TestChannelArbitratorForceCloseBreachedChannel(t *testing.T) { log := &mockArbitratorLog{ state: StateDefault, newStates: make(chan ArbitratorState, 5), } chanArbCtx, err := createTestChannelArbitrator(t, log) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } chanArb := chanArbCtx.chanArb if err := chanArb.Start(); err != nil { t.Fatalf("unable to start ChannelArbitrator: %v", err) } // It should start in StateDefault. chanArbCtx.AssertState(StateDefault) // We start by attempting a local force close. We'll return an // unexpected publication error, causing the state machine to halt. expErr := errors.New("intentional publication error") stateChan := make(chan ArbitratorState) chanArb.cfg.PublishTx = func(*wire.MsgTx, string) error { // When the force close tx is being broadcasted, check that the // state is correct at that point. select { case stateChan <- chanArb.state: case <-chanArb.quit: return fmt.Errorf("exiting") } return expErr } errChan := make(chan error, 1) respChan := make(chan *wire.MsgTx, 1) // With the channel found, and the request crafted, we'll send over a // force close request to the arbitrator that watches this channel. chanArb.forceCloseReqs <- &forceCloseReq{ errResp: errChan, closeTx: respChan, } // It should transition to StateBroadcastCommit. chanArbCtx.AssertStateTransitions(StateBroadcastCommit) // We expect it to be in state StateBroadcastCommit when attempting // the force close. select { case state := <-stateChan: if state != StateBroadcastCommit { t.Fatalf("state during PublishTx was %v", state) } case <-time.After(stateTimeout): t.Fatalf("no state update received") } // Make sure we get the expected error. select { case err := <-errChan: if err != expErr { t.Fatalf("unexpected error force closing channel: %v", err) } case <-time.After(defaultTimeout): t.Fatalf("no response received") } // We mimic that the channel is breached while the channel arbitrator // is down. This means that on restart it will be started with a // pending close channel, of type BreachClose. chanArbCtx, err = chanArbCtx.Restart(func(c *chanArbTestCtx) { c.chanArb.cfg.IsPendingClose = true c.chanArb.cfg.ClosingHeight = 100 c.chanArb.cfg.CloseType = channeldb.BreachClose }) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } defer chanArbCtx.CleanUp() // Finally it should advance to StateFullyResolved. chanArbCtx.AssertStateTransitions(StateFullyResolved) // It should also mark the channel as resolved. select { case <-chanArbCtx.resolvedChan: // Expected. case <-time.After(defaultTimeout): t.Fatalf("contract was not resolved") } } // TestChannelArbitratorCommitFailure tests that the channel arbitrator is able // to recover from a failed CommitState call at restart. func TestChannelArbitratorCommitFailure(t *testing.T) { testCases := []struct { // closeType is the type of channel close we want ot test. closeType channeldb.ClosureType // sendEvent is a function that will send the event // corresponding to this test's closeType to the passed // ChannelArbitrator. sendEvent func(chanArb *ChannelArbitrator) // expectedStates is the states we expect the state machine to // go through after a restart and successful log commit. expectedStates []ArbitratorState }{ { closeType: channeldb.CooperativeClose, sendEvent: func(chanArb *ChannelArbitrator) { closeInfo := &CooperativeCloseInfo{ &channeldb.ChannelCloseSummary{}, } chanArb.cfg.ChainEvents.CooperativeClosure <- closeInfo }, expectedStates: []ArbitratorState{StateFullyResolved}, }, { closeType: channeldb.RemoteForceClose, sendEvent: func(chanArb *ChannelArbitrator) { commitSpend := &chainntnfs.SpendDetail{ SpenderTxHash: &chainhash.Hash{}, } uniClose := &lnwallet.UnilateralCloseSummary{ SpendDetail: commitSpend, HtlcResolutions: &lnwallet.HtlcResolutions{}, } chanArb.cfg.ChainEvents.RemoteUnilateralClosure <- &RemoteUnilateralCloseInfo{ UnilateralCloseSummary: uniClose, } }, expectedStates: []ArbitratorState{StateContractClosed, StateFullyResolved}, }, { closeType: channeldb.LocalForceClose, sendEvent: func(chanArb *ChannelArbitrator) { chanArb.cfg.ChainEvents.LocalUnilateralClosure <- &LocalUnilateralCloseInfo{ SpendDetail: &chainntnfs.SpendDetail{}, LocalForceCloseSummary: &lnwallet.LocalForceCloseSummary{ CloseTx: &wire.MsgTx{}, HtlcResolutions: &lnwallet.HtlcResolutions{}, }, ChannelCloseSummary: &channeldb.ChannelCloseSummary{}, } }, expectedStates: []ArbitratorState{StateContractClosed, StateFullyResolved}, }, } for _, test := range testCases { test := test log := &mockArbitratorLog{ state: StateDefault, newStates: make(chan ArbitratorState, 5), failCommit: true, // Set the log to fail on the first expected state // after state machine progress for this test case. failCommitState: test.expectedStates[0], } chanArbCtx, err := createTestChannelArbitrator(t, log) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } chanArb := chanArbCtx.chanArb if err := chanArb.Start(); err != nil { t.Fatalf("unable to start ChannelArbitrator: %v", err) } // It should start in StateDefault. chanArbCtx.AssertState(StateDefault) closed := make(chan struct{}) chanArb.cfg.MarkChannelClosed = func( *channeldb.ChannelCloseSummary, ...channeldb.ChannelStatus) error { close(closed) return nil } // Send the test event to trigger the state machine. test.sendEvent(chanArb) select { case <-closed: case <-time.After(defaultTimeout): t.Fatalf("channel was not marked closed") } // Since the channel was marked closed in the database, but the // commit to the next state failed, the state should still be // StateDefault. time.Sleep(100 * time.Millisecond) if log.state != StateDefault { t.Fatalf("expected to stay in StateDefault, instead "+ "has %v", log.state) } chanArb.Stop() // Start the arbitrator again, with IsPendingClose reporting // the channel closed in the database. log.failCommit = false chanArbCtx, err = chanArbCtx.Restart(func(c *chanArbTestCtx) { c.chanArb.cfg.IsPendingClose = true c.chanArb.cfg.ClosingHeight = 100 c.chanArb.cfg.CloseType = test.closeType }) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } // Since the channel is marked closed in the database, it // should advance to the expected states. chanArbCtx.AssertStateTransitions(test.expectedStates...) // It should also mark the channel as resolved. select { case <-chanArbCtx.resolvedChan: // Expected. case <-time.After(defaultTimeout): t.Fatalf("contract was not resolved") } } } // TestChannelArbitratorEmptyResolutions makes sure that a channel that is // pending close in the database, but haven't had any resolutions logged will // not be marked resolved. This situation must be handled to avoid closing // channels from earlier versions of the ChannelArbitrator, which didn't have a // proper handoff from the ChainWatcher, and we could risk ending up in a state // where the channel was closed in the DB, but the resolutions weren't properly // written. func TestChannelArbitratorEmptyResolutions(t *testing.T) { // Start out with a log that will fail writing the set of resolutions. log := &mockArbitratorLog{ state: StateDefault, newStates: make(chan ArbitratorState, 5), failFetch: errNoResolutions, } chanArbCtx, err := createTestChannelArbitrator(t, log) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } chanArb := chanArbCtx.chanArb chanArb.cfg.IsPendingClose = true chanArb.cfg.ClosingHeight = 100 chanArb.cfg.CloseType = channeldb.RemoteForceClose if err := chanArb.Start(); err != nil { t.Fatalf("unable to start ChannelArbitrator: %v", err) } // It should not advance its state beyond StateContractClosed, since // fetching resolutions fails. chanArbCtx.AssertStateTransitions(StateContractClosed) // It should not advance further, however, as fetching resolutions // failed. time.Sleep(100 * time.Millisecond) if log.state != StateContractClosed { t.Fatalf("expected to stay in StateContractClosed") } chanArb.Stop() } // TestChannelArbitratorAlreadyForceClosed ensures that we cannot force close a // channel that is already in the process of doing so. func TestChannelArbitratorAlreadyForceClosed(t *testing.T) { t.Parallel() // We'll create the arbitrator and its backing log to signal that it's // already in the process of being force closed. log := &mockArbitratorLog{ state: StateCommitmentBroadcasted, } chanArbCtx, err := createTestChannelArbitrator(t, log) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } chanArb := chanArbCtx.chanArb if err := chanArb.Start(); err != nil { t.Fatalf("unable to start ChannelArbitrator: %v", err) } defer chanArb.Stop() // Then, we'll create a request to signal a force close request to the // channel arbitrator. errChan := make(chan error, 1) respChan := make(chan *wire.MsgTx, 1) select { case chanArb.forceCloseReqs <- &forceCloseReq{ closeTx: respChan, errResp: errChan, }: case <-chanArb.quit: } // Finally, we should ensure that we are not able to do so by seeing // the expected errAlreadyForceClosed error. select { case err = <-errChan: if err != errAlreadyForceClosed { t.Fatalf("expected errAlreadyForceClosed, got %v", err) } case <-time.After(time.Second): t.Fatal("expected to receive error response") } } // TestChannelArbitratorDanglingCommitForceClose tests that if there're HTLCs // on the remote party's commitment, but not ours, and they're about to time // out, then we'll go on chain so we can cancel back the HTLCs on the incoming // commitment. func TestChannelArbitratorDanglingCommitForceClose(t *testing.T) { t.Parallel() type testCase struct { htlcExpired bool remotePendingHTLC bool confCommit HtlcSetKey } var testCases []testCase testOptions := []bool{true, false} confOptions := []HtlcSetKey{ LocalHtlcSet, RemoteHtlcSet, RemotePendingHtlcSet, } for _, htlcExpired := range testOptions { for _, remotePendingHTLC := range testOptions { for _, commitConf := range confOptions { switch { // If the HTLC is on the remote commitment, and // that one confirms, then there's no special // behavior, we should play all the HTLCs on // that remote commitment as normal. case !remotePendingHTLC && commitConf == RemoteHtlcSet: fallthrough // If the HTLC is on the remote pending, and // that confirms, then we don't have any // special actions. case remotePendingHTLC && commitConf == RemotePendingHtlcSet: continue } testCases = append(testCases, testCase{ htlcExpired: htlcExpired, remotePendingHTLC: remotePendingHTLC, confCommit: commitConf, }) } } } for _, testCase := range testCases { testCase := testCase testName := fmt.Sprintf("testCase: htlcExpired=%v,"+ "remotePendingHTLC=%v,remotePendingCommitConf=%v", testCase.htlcExpired, testCase.remotePendingHTLC, testCase.confCommit) t.Run(testName, func(t *testing.T) { t.Parallel() arbLog := &mockArbitratorLog{ state: StateDefault, newStates: make(chan ArbitratorState, 5), resolvers: make(map[ContractResolver]struct{}), } chanArbCtx, err := createTestChannelArbitrator( t, arbLog, ) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } chanArb := chanArbCtx.chanArb if err := chanArb.Start(); err != nil { t.Fatalf("unable to start ChannelArbitrator: %v", err) } defer chanArb.Stop() // Now that our channel arb has started, we'll set up // its contract signals channel so we can send it // various HTLC updates for this test. htlcUpdates := make(chan *ContractUpdate) signals := &ContractSignals{ HtlcUpdates: htlcUpdates, ShortChanID: lnwire.ShortChannelID{}, } chanArb.UpdateContractSignals(signals) htlcKey := RemoteHtlcSet if testCase.remotePendingHTLC { htlcKey = RemotePendingHtlcSet } // Next, we'll send it a new HTLC that is set to expire // in 10 blocks, this HTLC will only appear on the // commitment transaction of the _remote_ party. htlcIndex := uint64(99) htlcExpiry := uint32(10) danglingHTLC := channeldb.HTLC{ Incoming: false, Amt: 10000, HtlcIndex: htlcIndex, RefundTimeout: htlcExpiry, } htlcUpdates <- &ContractUpdate{ HtlcKey: htlcKey, Htlcs: []channeldb.HTLC{danglingHTLC}, } // At this point, we now have a split commitment state // from the PoV of the channel arb. There's now an HTLC // that only exists on the commitment transaction of // the remote party. errChan := make(chan error, 1) respChan := make(chan *wire.MsgTx, 1) switch { // If we want an HTLC expiration trigger, then We'll // now mine a block (height 5), which is 5 blocks away // (our grace delta) from the expiry of that HTLC. case testCase.htlcExpired: chanArbCtx.blockEpochs <- &chainntnfs.BlockEpoch{Height: 5} // Otherwise, we'll just trigger a regular force close // request. case !testCase.htlcExpired: chanArb.forceCloseReqs <- &forceCloseReq{ errResp: errChan, closeTx: respChan, } } // At this point, the resolver should now have // determined that it needs to go to chain in order to // block off the redemption path so it can cancel the // incoming HTLC. chanArbCtx.AssertStateTransitions( StateBroadcastCommit, StateCommitmentBroadcasted, ) // Next we'll craft a fake commitment transaction to // send to signal that the channel has closed out on // chain. closeTx := &wire.MsgTx{ TxIn: []*wire.TxIn{ { PreviousOutPoint: wire.OutPoint{}, Witness: [][]byte{ {0x9}, }, }, }, } // We'll now signal to the channel arb that the HTLC // has fully closed on chain. Our local commit set // shows now HTLC on our commitment, but one on the // remote commitment. This should result in the HTLC // being canalled back. Also note that there're no HTLC // resolutions sent since we have none on our // commitment transaction. uniCloseInfo := &LocalUnilateralCloseInfo{ SpendDetail: &chainntnfs.SpendDetail{}, LocalForceCloseSummary: &lnwallet.LocalForceCloseSummary{ CloseTx: closeTx, HtlcResolutions: &lnwallet.HtlcResolutions{}, }, ChannelCloseSummary: &channeldb.ChannelCloseSummary{}, CommitSet: CommitSet{ ConfCommitKey: &testCase.confCommit, HtlcSets: make(map[HtlcSetKey][]channeldb.HTLC), }, } // If the HTLC was meant to expire, then we'll mark the // closing transaction at the proper expiry height // since our comparison "need to timeout" comparison is // based on the confirmation height. if testCase.htlcExpired { uniCloseInfo.SpendDetail.SpendingHeight = 5 } // Depending on if we're testing the remote pending // commitment or not, we'll populate either a fake // dangling remote commitment, or a regular locked in // one. htlcs := []channeldb.HTLC{danglingHTLC} if testCase.remotePendingHTLC { uniCloseInfo.CommitSet.HtlcSets[RemotePendingHtlcSet] = htlcs } else { uniCloseInfo.CommitSet.HtlcSets[RemoteHtlcSet] = htlcs } chanArb.cfg.ChainEvents.LocalUnilateralClosure <- uniCloseInfo // The channel arb should now transition to waiting // until the HTLCs have been fully resolved. chanArbCtx.AssertStateTransitions( StateContractClosed, StateWaitingFullResolution, ) // Now that we've sent this signal, we should have that // HTLC be canceled back immediately. select { case msgs := <-chanArbCtx.resolutions: if len(msgs) != 1 { t.Fatalf("expected 1 message, "+ "instead got %v", len(msgs)) } if msgs[0].HtlcIndex != htlcIndex { t.Fatalf("wrong htlc index: expected %v, got %v", htlcIndex, msgs[0].HtlcIndex) } case <-time.After(defaultTimeout): t.Fatalf("resolution msgs not sent") } // There's no contract to send a fully resolve message, // so instead, we'll mine another block which'll cause // it to re-examine its state and realize there're no // more HTLCs. chanArbCtx.blockEpochs <- &chainntnfs.BlockEpoch{Height: 6} chanArbCtx.AssertStateTransitions(StateFullyResolved) }) } } // TestChannelArbitratorPendingExpiredHTLC tests that if we have pending htlc // that is expired we will only go to chain if we are running at least the // time defined in PaymentsExpirationGracePeriod. // During this time the remote party is expected to send his updates and cancel // The htlc. func TestChannelArbitratorPendingExpiredHTLC(t *testing.T) { t.Parallel() // We'll create the arbitrator and its backing log in a default state. log := &mockArbitratorLog{ state: StateDefault, newStates: make(chan ArbitratorState, 5), resolvers: make(map[ContractResolver]struct{}), } chanArbCtx, err := createTestChannelArbitrator(t, log) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } chanArb := chanArbCtx.chanArb // We'll inject a test clock implementation so we can control the uptime. startTime := time.Date(2020, time.February, 3, 13, 0, 0, 0, time.UTC) testClock := clock.NewTestClock(startTime) chanArb.cfg.Clock = testClock // We also configure the grace period and the IsForwardedHTLC to identify // the htlc as our initiated payment. chanArb.cfg.PaymentsExpirationGracePeriod = time.Second * 15 chanArb.cfg.IsForwardedHTLC = func(chanID lnwire.ShortChannelID, htlcIndex uint64) bool { return false } if err := chanArb.Start(); err != nil { t.Fatalf("unable to start ChannelArbitrator: %v", err) } defer func() { if err := chanArb.Stop(); err != nil { t.Fatalf("unable to stop chan arb: %v", err) } }() // Now that our channel arb has started, we'll set up // its contract signals channel so we can send it // various HTLC updates for this test. htlcUpdates := make(chan *ContractUpdate) signals := &ContractSignals{ HtlcUpdates: htlcUpdates, ShortChanID: lnwire.ShortChannelID{}, } chanArb.UpdateContractSignals(signals) // Next, we'll send it a new HTLC that is set to expire // in 10 blocks. htlcIndex := uint64(99) htlcExpiry := uint32(10) pendingHTLC := channeldb.HTLC{ Incoming: false, Amt: 10000, HtlcIndex: htlcIndex, RefundTimeout: htlcExpiry, } htlcUpdates <- &ContractUpdate{ HtlcKey: RemoteHtlcSet, Htlcs: []channeldb.HTLC{pendingHTLC}, } // We will advance the uptime to 10 seconds which should be still within // the grace period and should not trigger going to chain. testClock.SetTime(startTime.Add(time.Second * 10)) chanArbCtx.blockEpochs <- &chainntnfs.BlockEpoch{Height: 5} chanArbCtx.AssertState(StateDefault) // We will advance the uptime to 16 seconds which should trigger going // to chain. testClock.SetTime(startTime.Add(time.Second * 16)) chanArbCtx.blockEpochs <- &chainntnfs.BlockEpoch{Height: 6} chanArbCtx.AssertStateTransitions( StateBroadcastCommit, StateCommitmentBroadcasted, ) } // TestRemoteCloseInitiator tests the setting of close initiator statuses // for remote force closes and breaches. func TestRemoteCloseInitiator(t *testing.T) { // getCloseSummary returns a unilateral close summary for the channel // provided. getCloseSummary := func(channel *channeldb.OpenChannel) *RemoteUnilateralCloseInfo { return &RemoteUnilateralCloseInfo{ UnilateralCloseSummary: &lnwallet.UnilateralCloseSummary{ SpendDetail: &chainntnfs.SpendDetail{ SpenderTxHash: &chainhash.Hash{}, SpendingTx: &wire.MsgTx{ TxIn: []*wire.TxIn{}, TxOut: []*wire.TxOut{}, }, }, ChannelCloseSummary: channeldb.ChannelCloseSummary{ ChanPoint: channel.FundingOutpoint, RemotePub: channel.IdentityPub, SettledBalance: btcutil.Amount(500), TimeLockedBalance: btcutil.Amount(10000), IsPending: false, }, HtlcResolutions: &lnwallet.HtlcResolutions{}, }, } } tests := []struct { name string // notifyClose sends the appropriate chain event to indicate // that the channel has closed. The event subscription channel // is expected to be buffered, as is the default for test // channel arbitrators. notifyClose func(sub *ChainEventSubscription, channel *channeldb.OpenChannel) // expectedStates is the set of states we expect the arbitrator // to progress through. expectedStates []ArbitratorState }{ { name: "force close", notifyClose: func(sub *ChainEventSubscription, channel *channeldb.OpenChannel) { s := getCloseSummary(channel) sub.RemoteUnilateralClosure <- s }, expectedStates: []ArbitratorState{ StateContractClosed, StateFullyResolved, }, }, } for _, test := range tests { test := test t.Run(test.name, func(t *testing.T) { t.Parallel() // First, create alice's channel. alice, _, cleanUp, err := lnwallet.CreateTestChannels( channeldb.SingleFunderTweaklessBit, ) if err != nil { t.Fatalf("unable to create test channels: %v", err) } defer cleanUp() // Create a mock log which will not block the test's // expected number of transitions transitions, and has // no commit resolutions so that the channel will // resolve immediately. log := &mockArbitratorLog{ state: StateDefault, newStates: make(chan ArbitratorState, len(test.expectedStates)), resolutions: &ContractResolutions{ CommitHash: chainhash.Hash{}, CommitResolution: nil, }, } // Mock marking the channel as closed, we only care // about setting of channel status. mockMarkClosed := func(_ *channeldb.ChannelCloseSummary, statuses ...channeldb.ChannelStatus) error { for _, status := range statuses { err := alice.State().ApplyChanStatus(status) if err != nil { return err } } return nil } chanArbCtx, err := createTestChannelArbitrator( t, log, withMarkClosed(mockMarkClosed), ) if err != nil { t.Fatalf("unable to create "+ "ChannelArbitrator: %v", err) } chanArb := chanArbCtx.chanArb if err := chanArb.Start(); err != nil { t.Fatalf("unable to start "+ "ChannelArbitrator: %v", err) } defer func() { if err := chanArb.Stop(); err != nil { t.Fatal(err) } }() // It should start out in the default state. chanArbCtx.AssertState(StateDefault) // Notify the close event. test.notifyClose(chanArb.cfg.ChainEvents, alice.State()) // Check that the channel transitions as expected. chanArbCtx.AssertStateTransitions( test.expectedStates..., ) // It should also mark the channel as resolved. select { case <-chanArbCtx.resolvedChan: // Expected. case <-time.After(defaultTimeout): t.Fatalf("contract was not resolved") } // Check that alice has the status we expect. if !alice.State().HasChanStatus( channeldb.ChanStatusRemoteCloseInitiator, ) { t.Fatalf("expected remote close initiator, "+ "got: %v", alice.State().ChanStatus()) } }) } } // TestChannelArbitratorAnchors asserts that the commitment tx anchor is swept. func TestChannelArbitratorAnchors(t *testing.T) { log := &mockArbitratorLog{ state: StateDefault, newStates: make(chan ArbitratorState, 5), } chanArbCtx, err := createTestChannelArbitrator(t, log) if err != nil { t.Fatalf("unable to create ChannelArbitrator: %v", err) } // Replace our mocked put report function with one which will push // reports into a channel for us to consume. We update this function // because our resolver will be created from the existing chanArb cfg. reports := make(chan *channeldb.ResolverReport) chanArbCtx.chanArb.cfg.PutResolverReport = putResolverReportInChannel( reports, ) chanArb := chanArbCtx.chanArb chanArb.cfg.PreimageDB = newMockWitnessBeacon() chanArb.cfg.Registry = &mockRegistry{} // Setup two pre-confirmation anchor resolutions on the mock channel. chanArb.cfg.Channel.(*mockChannel).anchorResolutions = []*lnwallet.AnchorResolution{ {}, {}, } if err := chanArb.Start(); err != nil { t.Fatalf("unable to start ChannelArbitrator: %v", err) } defer func() { if err := chanArb.Stop(); err != nil { t.Fatal(err) } }() // Create htlcUpdates channel. htlcUpdates := make(chan *ContractUpdate) signals := &ContractSignals{ HtlcUpdates: htlcUpdates, ShortChanID: lnwire.ShortChannelID{}, } chanArb.UpdateContractSignals(signals) errChan := make(chan error, 1) respChan := make(chan *wire.MsgTx, 1) // With the channel found, and the request crafted, we'll send over a // force close request to the arbitrator that watches this channel. chanArb.forceCloseReqs <- &forceCloseReq{ errResp: errChan, closeTx: respChan, } // The force close request should trigger broadcast of the commitment // transaction. chanArbCtx.AssertStateTransitions( StateBroadcastCommit, StateCommitmentBroadcasted, ) // With the commitment tx still unconfirmed, we expect sweep attempts // for all three versions of the commitment transaction. <-chanArbCtx.sweeper.sweptInputs <-chanArbCtx.sweeper.sweptInputs select { case <-respChan: case <-time.After(5 * time.Second): t.Fatalf("no response received") } select { case err := <-errChan: if err != nil { t.Fatalf("error force closing channel: %v", err) } case <-time.After(5 * time.Second): t.Fatalf("no response received") } // Now notify about the local force close getting confirmed. closeTx := &wire.MsgTx{ TxIn: []*wire.TxIn{ { PreviousOutPoint: wire.OutPoint{}, Witness: [][]byte{ {0x1}, {0x2}, }, }, }, } anchorResolution := &lnwallet.AnchorResolution{ AnchorSignDescriptor: input.SignDescriptor{ Output: &wire.TxOut{ Value: 1, }, }, } chanArb.cfg.ChainEvents.LocalUnilateralClosure <- &LocalUnilateralCloseInfo{ SpendDetail: &chainntnfs.SpendDetail{}, LocalForceCloseSummary: &lnwallet.LocalForceCloseSummary{ CloseTx: closeTx, HtlcResolutions: &lnwallet.HtlcResolutions{}, AnchorResolution: anchorResolution, }, ChannelCloseSummary: &channeldb.ChannelCloseSummary{}, CommitSet: CommitSet{ ConfCommitKey: &LocalHtlcSet, HtlcSets: map[HtlcSetKey][]channeldb.HTLC{}, }, } chanArbCtx.AssertStateTransitions( StateContractClosed, StateWaitingFullResolution, ) // We expect to only have the anchor resolver active. if len(chanArb.activeResolvers) != 1 { t.Fatalf("expected single resolver, instead got: %v", len(chanArb.activeResolvers)) } resolver := chanArb.activeResolvers[0] _, ok := resolver.(*anchorResolver) if !ok { t.Fatalf("expected anchor resolver, got %T", resolver) } // The anchor resolver is expected to offer the anchor input to the // sweeper. <-chanArbCtx.sweeper.updatedInputs // The mock sweeper immediately signals success for that input. This // should transition the channel to the resolved state. chanArbCtx.AssertStateTransitions(StateFullyResolved) select { case <-chanArbCtx.resolvedChan: case <-time.After(5 * time.Second): t.Fatalf("contract was not resolved") } anchorAmt := btcutil.Amount( anchorResolution.AnchorSignDescriptor.Output.Value, ) spendTx := chanArbCtx.sweeper.sweepTx.TxHash() expectedReport := &channeldb.ResolverReport{ OutPoint: anchorResolution.CommitAnchor, Amount: anchorAmt, ResolverType: channeldb.ResolverTypeAnchor, ResolverOutcome: channeldb.ResolverOutcomeClaimed, SpendTxID: &spendTx, } assertResolverReport(t, reports, expectedReport) } // putResolverReportInChannel returns a put report function which will pipe // reports into the channel provided. func putResolverReportInChannel(reports chan *channeldb.ResolverReport) func( _ kvdb.RwTx, report *channeldb.ResolverReport) error { return func(_ kvdb.RwTx, report *channeldb.ResolverReport) error { reports <- report return nil } } // assertResolverReport checks that a set of reports only contains a single // report, and that it is equal to the expected report passed in. func assertResolverReport(t *testing.T, reports chan *channeldb.ResolverReport, expected *channeldb.ResolverReport) { select { case report := <-reports: if !reflect.DeepEqual(report, expected) { t.Fatalf("expected: %v, got: %v", expected, report) } case <-time.After(defaultTimeout): t.Fatalf("no reports present") } } type mockChannel struct { anchorResolutions []*lnwallet.AnchorResolution } func (m *mockChannel) NewAnchorResolutions() ([]*lnwallet.AnchorResolution, error) { return m.anchorResolutions, nil } func (m *mockChannel) ForceCloseChan() (*lnwallet.LocalForceCloseSummary, error) { summary := &lnwallet.LocalForceCloseSummary{ CloseTx: &wire.MsgTx{}, HtlcResolutions: &lnwallet.HtlcResolutions{}, } return summary, nil }