49d8f04197
This commit moves all localized instances of mock implementations of the Signer interface to the lntest/mock package. This allows us to remove a lot of code and have it housed under a single interface in many cases.
394 lines
10 KiB
Go
394 lines
10 KiB
Go
package sweep
|
|
|
|
import (
|
|
"bytes"
|
|
"fmt"
|
|
"testing"
|
|
|
|
"github.com/btcsuite/btcd/chaincfg"
|
|
"github.com/btcsuite/btcd/txscript"
|
|
"github.com/btcsuite/btcd/wire"
|
|
"github.com/btcsuite/btcutil"
|
|
"github.com/lightningnetwork/lnd/lntest/mock"
|
|
"github.com/lightningnetwork/lnd/lnwallet"
|
|
"github.com/lightningnetwork/lnd/lnwallet/chainfee"
|
|
)
|
|
|
|
// TestDetermineFeePerKw tests that given a fee preference, the
|
|
// DetermineFeePerKw will properly map it to a concrete fee in sat/kw.
|
|
func TestDetermineFeePerKw(t *testing.T) {
|
|
t.Parallel()
|
|
|
|
defaultFee := chainfee.SatPerKWeight(999)
|
|
relayFee := chainfee.SatPerKWeight(300)
|
|
|
|
feeEstimator := newMockFeeEstimator(defaultFee, relayFee)
|
|
|
|
// We'll populate two items in the internal map which is used to query
|
|
// a fee based on a confirmation target: the default conf target, and
|
|
// an arbitrary conf target. We'll ensure below that both of these are
|
|
// properly
|
|
feeEstimator.blocksToFee[50] = 300
|
|
feeEstimator.blocksToFee[defaultNumBlocksEstimate] = 1000
|
|
|
|
testCases := []struct {
|
|
// feePref is the target fee preference for this case.
|
|
feePref FeePreference
|
|
|
|
// fee is the value the DetermineFeePerKw should return given
|
|
// the FeePreference above
|
|
fee chainfee.SatPerKWeight
|
|
|
|
// fail determines if this test case should fail or not.
|
|
fail bool
|
|
}{
|
|
// A fee rate below the fee rate floor should output the floor.
|
|
{
|
|
feePref: FeePreference{
|
|
FeeRate: chainfee.SatPerKWeight(99),
|
|
},
|
|
fee: chainfee.FeePerKwFloor,
|
|
},
|
|
|
|
// A fee rate above the floor, should pass through and return
|
|
// the target fee rate.
|
|
{
|
|
feePref: FeePreference{
|
|
FeeRate: 900,
|
|
},
|
|
fee: 900,
|
|
},
|
|
|
|
// A specified confirmation target should cause the function to
|
|
// query the estimator which will return our value specified
|
|
// above.
|
|
{
|
|
feePref: FeePreference{
|
|
ConfTarget: 50,
|
|
},
|
|
fee: 300,
|
|
},
|
|
|
|
// If the caller doesn't specify any values at all, then we
|
|
// should query for the default conf target.
|
|
{
|
|
feePref: FeePreference{},
|
|
fee: 1000,
|
|
},
|
|
|
|
// Both conf target and fee rate are set, we should return with
|
|
// an error.
|
|
{
|
|
feePref: FeePreference{
|
|
ConfTarget: 50,
|
|
FeeRate: 90000,
|
|
},
|
|
fee: 300,
|
|
fail: true,
|
|
},
|
|
}
|
|
for i, testCase := range testCases {
|
|
targetFee, err := DetermineFeePerKw(
|
|
feeEstimator, testCase.feePref,
|
|
)
|
|
switch {
|
|
case testCase.fail && err != nil:
|
|
continue
|
|
|
|
case testCase.fail && err == nil:
|
|
t.Fatalf("expected failure for #%v", i)
|
|
|
|
case !testCase.fail && err != nil:
|
|
t.Fatalf("unable to estimate fee; %v", err)
|
|
}
|
|
|
|
if targetFee != testCase.fee {
|
|
t.Fatalf("#%v: wrong fee: expected %v got %v", i,
|
|
testCase.fee, targetFee)
|
|
}
|
|
}
|
|
}
|
|
|
|
type mockUtxoSource struct {
|
|
outputs []*lnwallet.Utxo
|
|
}
|
|
|
|
func newMockUtxoSource(utxos []*lnwallet.Utxo) *mockUtxoSource {
|
|
return &mockUtxoSource{
|
|
outputs: utxos,
|
|
}
|
|
}
|
|
|
|
func (m *mockUtxoSource) ListUnspentWitness(minConfs int32,
|
|
maxConfs int32) ([]*lnwallet.Utxo, error) {
|
|
|
|
return m.outputs, nil
|
|
}
|
|
|
|
type mockCoinSelectionLocker struct {
|
|
fail bool
|
|
}
|
|
|
|
func (m *mockCoinSelectionLocker) WithCoinSelectLock(f func() error) error {
|
|
if err := f(); err != nil {
|
|
return err
|
|
}
|
|
|
|
if m.fail {
|
|
return fmt.Errorf("kek")
|
|
}
|
|
|
|
return nil
|
|
|
|
}
|
|
|
|
type mockOutpointLocker struct {
|
|
lockedOutpoints map[wire.OutPoint]struct{}
|
|
|
|
unlockedOutpoints map[wire.OutPoint]struct{}
|
|
}
|
|
|
|
func newMockOutpointLocker() *mockOutpointLocker {
|
|
return &mockOutpointLocker{
|
|
lockedOutpoints: make(map[wire.OutPoint]struct{}),
|
|
|
|
unlockedOutpoints: make(map[wire.OutPoint]struct{}),
|
|
}
|
|
}
|
|
|
|
func (m *mockOutpointLocker) LockOutpoint(o wire.OutPoint) {
|
|
m.lockedOutpoints[o] = struct{}{}
|
|
}
|
|
func (m *mockOutpointLocker) UnlockOutpoint(o wire.OutPoint) {
|
|
m.unlockedOutpoints[o] = struct{}{}
|
|
}
|
|
|
|
var sweepScript = []byte{
|
|
0x0, 0x14, 0x64, 0x3d, 0x8b, 0x15, 0x69, 0x4a, 0x54,
|
|
0x7d, 0x57, 0x33, 0x6e, 0x51, 0xdf, 0xfd, 0x38, 0xe3,
|
|
0xe, 0x6e, 0xf8, 0xef,
|
|
}
|
|
|
|
var deliveryAddr = func() btcutil.Address {
|
|
_, addrs, _, err := txscript.ExtractPkScriptAddrs(
|
|
sweepScript, &chaincfg.TestNet3Params,
|
|
)
|
|
if err != nil {
|
|
panic(err)
|
|
}
|
|
|
|
return addrs[0]
|
|
}()
|
|
|
|
var testUtxos = []*lnwallet.Utxo{
|
|
{
|
|
// A p2wkh output.
|
|
AddressType: lnwallet.WitnessPubKey,
|
|
PkScript: []byte{
|
|
0x0, 0x14, 0x64, 0x3d, 0x8b, 0x15, 0x69, 0x4a, 0x54,
|
|
0x7d, 0x57, 0x33, 0x6e, 0x51, 0xdf, 0xfd, 0x38, 0xe3,
|
|
0xe, 0x6e, 0xf7, 0xef,
|
|
},
|
|
Value: 1000,
|
|
OutPoint: wire.OutPoint{
|
|
Index: 1,
|
|
},
|
|
},
|
|
|
|
{
|
|
// A np2wkh output.
|
|
AddressType: lnwallet.NestedWitnessPubKey,
|
|
PkScript: []byte{
|
|
0xa9, 0x14, 0x97, 0x17, 0xf7, 0xd1, 0x5f, 0x6f, 0x8b,
|
|
0x7, 0xe3, 0x58, 0x43, 0x19, 0xb9, 0x7e, 0xa9, 0x20,
|
|
0x18, 0xc3, 0x17, 0xd7, 0x87,
|
|
},
|
|
Value: 2000,
|
|
OutPoint: wire.OutPoint{
|
|
Index: 2,
|
|
},
|
|
},
|
|
|
|
// A p2wsh output.
|
|
{
|
|
AddressType: lnwallet.UnknownAddressType,
|
|
PkScript: []byte{
|
|
0x0, 0x20, 0x70, 0x1a, 0x8d, 0x40, 0x1c, 0x84, 0xfb, 0x13,
|
|
0xe6, 0xba, 0xf1, 0x69, 0xd5, 0x96, 0x84, 0xe2, 0x7a, 0xbd,
|
|
0x9f, 0xa2, 0x16, 0xc8, 0xbc, 0x5b, 0x9f, 0xc6, 0x3d, 0x62,
|
|
0x2f, 0xf8, 0xc5, 0x8c,
|
|
},
|
|
Value: 3000,
|
|
OutPoint: wire.OutPoint{
|
|
Index: 3,
|
|
},
|
|
},
|
|
}
|
|
|
|
func assertUtxosLocked(t *testing.T, utxoLocker *mockOutpointLocker,
|
|
utxos []*lnwallet.Utxo) {
|
|
|
|
t.Helper()
|
|
|
|
for _, utxo := range utxos {
|
|
if _, ok := utxoLocker.lockedOutpoints[utxo.OutPoint]; !ok {
|
|
t.Fatalf("utxo %v was never locked", utxo.OutPoint)
|
|
}
|
|
}
|
|
|
|
}
|
|
|
|
func assertNoUtxosUnlocked(t *testing.T, utxoLocker *mockOutpointLocker,
|
|
utxos []*lnwallet.Utxo) {
|
|
|
|
t.Helper()
|
|
|
|
if len(utxoLocker.unlockedOutpoints) != 0 {
|
|
t.Fatalf("outputs have been locked, but shouldn't have been")
|
|
}
|
|
}
|
|
|
|
func assertUtxosUnlocked(t *testing.T, utxoLocker *mockOutpointLocker,
|
|
utxos []*lnwallet.Utxo) {
|
|
|
|
t.Helper()
|
|
|
|
for _, utxo := range utxos {
|
|
if _, ok := utxoLocker.unlockedOutpoints[utxo.OutPoint]; !ok {
|
|
t.Fatalf("utxo %v was never unlocked", utxo.OutPoint)
|
|
}
|
|
}
|
|
}
|
|
|
|
func assertUtxosLockedAndUnlocked(t *testing.T, utxoLocker *mockOutpointLocker,
|
|
utxos []*lnwallet.Utxo) {
|
|
|
|
t.Helper()
|
|
|
|
for _, utxo := range utxos {
|
|
if _, ok := utxoLocker.lockedOutpoints[utxo.OutPoint]; !ok {
|
|
t.Fatalf("utxo %v was never locked", utxo.OutPoint)
|
|
}
|
|
|
|
if _, ok := utxoLocker.unlockedOutpoints[utxo.OutPoint]; !ok {
|
|
t.Fatalf("utxo %v was never unlocked", utxo.OutPoint)
|
|
}
|
|
}
|
|
}
|
|
|
|
// TestCraftSweepAllTxCoinSelectFail tests that if coin selection fails, then
|
|
// we unlock any outputs we may have locked in the passed closure.
|
|
func TestCraftSweepAllTxCoinSelectFail(t *testing.T) {
|
|
t.Parallel()
|
|
|
|
utxoSource := newMockUtxoSource(testUtxos)
|
|
coinSelectLocker := &mockCoinSelectionLocker{
|
|
fail: true,
|
|
}
|
|
utxoLocker := newMockOutpointLocker()
|
|
|
|
_, err := CraftSweepAllTx(
|
|
0, 100, nil, coinSelectLocker, utxoSource, utxoLocker, nil, nil,
|
|
)
|
|
|
|
// Since we instructed the coin select locker to fail above, we should
|
|
// get an error.
|
|
if err == nil {
|
|
t.Fatalf("sweep tx should have failed: %v", err)
|
|
}
|
|
|
|
// At this point, we'll now verify that all outputs were initially
|
|
// locked, and then also unlocked due to the failure.
|
|
assertUtxosLockedAndUnlocked(t, utxoLocker, testUtxos)
|
|
}
|
|
|
|
// TestCraftSweepAllTxUnknownWitnessType tests that if one of the inputs we
|
|
// encounter is of an unknown witness type, then we fail and unlock any prior
|
|
// locked outputs.
|
|
func TestCraftSweepAllTxUnknownWitnessType(t *testing.T) {
|
|
t.Parallel()
|
|
|
|
utxoSource := newMockUtxoSource(testUtxos)
|
|
coinSelectLocker := &mockCoinSelectionLocker{}
|
|
utxoLocker := newMockOutpointLocker()
|
|
|
|
_, err := CraftSweepAllTx(
|
|
0, 100, nil, coinSelectLocker, utxoSource, utxoLocker, nil, nil,
|
|
)
|
|
|
|
// Since passed in a p2wsh output, which is unknown, we should fail to
|
|
// map the output to a witness type.
|
|
if err == nil {
|
|
t.Fatalf("sweep tx should have failed: %v", err)
|
|
}
|
|
|
|
// At this point, we'll now verify that all outputs were initially
|
|
// locked, and then also unlocked since we weren't able to find a
|
|
// witness type for the last output.
|
|
assertUtxosLockedAndUnlocked(t, utxoLocker, testUtxos)
|
|
}
|
|
|
|
// TestCraftSweepAllTx tests that we'll properly lock all available outputs
|
|
// within the wallet, and craft a single sweep transaction that pays to the
|
|
// target output.
|
|
func TestCraftSweepAllTx(t *testing.T) {
|
|
t.Parallel()
|
|
|
|
// First, we'll make a mock signer along with a fee estimator, We'll
|
|
// use zero fees to we can assert a precise output value.
|
|
signer := &mock.DummySigner{}
|
|
feeEstimator := newMockFeeEstimator(0, 0)
|
|
|
|
// For our UTXO source, we'll pass in all the UTXOs that we know of,
|
|
// other than the final one which is of an unknown witness type.
|
|
targetUTXOs := testUtxos[:2]
|
|
utxoSource := newMockUtxoSource(targetUTXOs)
|
|
coinSelectLocker := &mockCoinSelectionLocker{}
|
|
utxoLocker := newMockOutpointLocker()
|
|
|
|
sweepPkg, err := CraftSweepAllTx(
|
|
0, 100, deliveryAddr, coinSelectLocker, utxoSource, utxoLocker,
|
|
feeEstimator, signer,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to make sweep tx: %v", err)
|
|
}
|
|
|
|
// At this point, all of the UTXOs that we made above should be locked
|
|
// and none of them unlocked.
|
|
assertUtxosLocked(t, utxoLocker, testUtxos[:2])
|
|
assertNoUtxosUnlocked(t, utxoLocker, testUtxos[:2])
|
|
|
|
// Now that we have our sweep transaction, we should find that we have
|
|
// a UTXO for each input, and also that our final output value is the
|
|
// sum of all our inputs.
|
|
sweepTx := sweepPkg.SweepTx
|
|
if len(sweepTx.TxIn) != len(targetUTXOs) {
|
|
t.Fatalf("expected %v utxo, got %v", len(targetUTXOs),
|
|
len(sweepTx.TxIn))
|
|
}
|
|
|
|
// We should have a single output that pays to our sweep script
|
|
// generated above.
|
|
expectedSweepValue := int64(3000)
|
|
if len(sweepTx.TxOut) != 1 {
|
|
t.Fatalf("should have %v outputs, instead have %v", 1,
|
|
len(sweepTx.TxOut))
|
|
}
|
|
output := sweepTx.TxOut[0]
|
|
switch {
|
|
case output.Value != expectedSweepValue:
|
|
t.Fatalf("expected %v sweep value, instead got %v",
|
|
expectedSweepValue, output.Value)
|
|
|
|
case !bytes.Equal(sweepScript, output.PkScript):
|
|
t.Fatalf("expected %x sweep script, instead got %x", sweepScript,
|
|
output.PkScript)
|
|
}
|
|
|
|
// If we cancel the sweep attempt, then we should find that all the
|
|
// UTXOs within the sweep transaction are now unlocked.
|
|
sweepPkg.CancelSweepAttempt()
|
|
assertUtxosUnlocked(t, utxoLocker, testUtxos[:2])
|
|
}
|