af4abe7d58
This commit fixes the inconsistency between the payment state as reported by routerrpc.SendPayment/routerrpc.TrackPayment and the main rpc ListPayments call. In addition to that, payment state changes are now sent out for every state change. This opens the door to user interfaces giving more feedback to the user about the payment process. This is especially interesting for multi-part payments.
15106 lines
472 KiB
Go
15106 lines
472 KiB
Go
// +build rpctest
|
|
|
|
package itest
|
|
|
|
import (
|
|
"bytes"
|
|
"context"
|
|
"crypto/rand"
|
|
"crypto/sha256"
|
|
"encoding/hex"
|
|
"fmt"
|
|
"io"
|
|
"io/ioutil"
|
|
"math"
|
|
"os"
|
|
"path/filepath"
|
|
"reflect"
|
|
"strings"
|
|
"sync"
|
|
"sync/atomic"
|
|
"testing"
|
|
"time"
|
|
|
|
"github.com/btcsuite/btcd/btcjson"
|
|
"github.com/btcsuite/btcd/chaincfg"
|
|
"github.com/btcsuite/btcd/chaincfg/chainhash"
|
|
"github.com/btcsuite/btcd/integration/rpctest"
|
|
"github.com/btcsuite/btcd/rpcclient"
|
|
"github.com/btcsuite/btcd/wire"
|
|
"github.com/btcsuite/btcutil"
|
|
"github.com/davecgh/go-spew/spew"
|
|
"github.com/go-errors/errors"
|
|
"github.com/lightningnetwork/lnd"
|
|
"github.com/lightningnetwork/lnd/chanbackup"
|
|
"github.com/lightningnetwork/lnd/channeldb"
|
|
"github.com/lightningnetwork/lnd/input"
|
|
"github.com/lightningnetwork/lnd/lnrpc"
|
|
"github.com/lightningnetwork/lnd/lnrpc/invoicesrpc"
|
|
"github.com/lightningnetwork/lnd/lnrpc/routerrpc"
|
|
"github.com/lightningnetwork/lnd/lnrpc/signrpc"
|
|
"github.com/lightningnetwork/lnd/lnrpc/watchtowerrpc"
|
|
"github.com/lightningnetwork/lnd/lnrpc/wtclientrpc"
|
|
"github.com/lightningnetwork/lnd/lntest"
|
|
"github.com/lightningnetwork/lnd/lntest/wait"
|
|
"github.com/lightningnetwork/lnd/lntypes"
|
|
"github.com/lightningnetwork/lnd/lnwallet/chainfee"
|
|
"github.com/lightningnetwork/lnd/lnwire"
|
|
"github.com/lightningnetwork/lnd/routing"
|
|
)
|
|
|
|
var (
|
|
harnessNetParams = &chaincfg.RegressionNetParams
|
|
)
|
|
|
|
const (
|
|
testFeeBase = 1e+6
|
|
defaultCSV = lntest.DefaultCSV
|
|
defaultTimeout = lntest.DefaultTimeout
|
|
minerMempoolTimeout = lntest.MinerMempoolTimeout
|
|
channelOpenTimeout = lntest.ChannelOpenTimeout
|
|
channelCloseTimeout = lntest.ChannelCloseTimeout
|
|
itestLndBinary = "../../lnd-itest"
|
|
anchorSize = 330
|
|
)
|
|
|
|
// harnessTest wraps a regular testing.T providing enhanced error detection
|
|
// and propagation. All error will be augmented with a full stack-trace in
|
|
// order to aid in debugging. Additionally, any panics caused by active
|
|
// test cases will also be handled and represented as fatals.
|
|
type harnessTest struct {
|
|
t *testing.T
|
|
|
|
// testCase is populated during test execution and represents the
|
|
// current test case.
|
|
testCase *testCase
|
|
|
|
// lndHarness is a reference to the current network harness. Will be
|
|
// nil if not yet set up.
|
|
lndHarness *lntest.NetworkHarness
|
|
}
|
|
|
|
// newHarnessTest creates a new instance of a harnessTest from a regular
|
|
// testing.T instance.
|
|
func newHarnessTest(t *testing.T, net *lntest.NetworkHarness) *harnessTest {
|
|
return &harnessTest{t, nil, net}
|
|
}
|
|
|
|
// Skipf calls the underlying testing.T's Skip method, causing the current test
|
|
// to be skipped.
|
|
func (h *harnessTest) Skipf(format string, args ...interface{}) {
|
|
h.t.Skipf(format, args...)
|
|
}
|
|
|
|
// Fatalf causes the current active test case to fail with a fatal error. All
|
|
// integration tests should mark test failures solely with this method due to
|
|
// the error stack traces it produces.
|
|
func (h *harnessTest) Fatalf(format string, a ...interface{}) {
|
|
if h.lndHarness != nil {
|
|
h.lndHarness.SaveProfilesPages()
|
|
}
|
|
|
|
stacktrace := errors.Wrap(fmt.Sprintf(format, a...), 1).ErrorStack()
|
|
|
|
if h.testCase != nil {
|
|
h.t.Fatalf("Failed: (%v): exited with error: \n"+
|
|
"%v", h.testCase.name, stacktrace)
|
|
} else {
|
|
h.t.Fatalf("Error outside of test: %v", stacktrace)
|
|
}
|
|
}
|
|
|
|
// RunTestCase executes a harness test case. Any errors or panics will be
|
|
// represented as fatal.
|
|
func (h *harnessTest) RunTestCase(testCase *testCase) {
|
|
h.testCase = testCase
|
|
defer func() {
|
|
h.testCase = nil
|
|
}()
|
|
|
|
defer func() {
|
|
if err := recover(); err != nil {
|
|
description := errors.Wrap(err, 2).ErrorStack()
|
|
h.t.Fatalf("Failed: (%v) panicked with: \n%v",
|
|
h.testCase.name, description)
|
|
}
|
|
}()
|
|
|
|
testCase.test(h.lndHarness, h)
|
|
|
|
return
|
|
}
|
|
|
|
func (h *harnessTest) Logf(format string, args ...interface{}) {
|
|
h.t.Logf(format, args...)
|
|
}
|
|
|
|
func (h *harnessTest) Log(args ...interface{}) {
|
|
h.t.Log(args...)
|
|
}
|
|
|
|
func assertTxInBlock(t *harnessTest, block *wire.MsgBlock, txid *chainhash.Hash) {
|
|
for _, tx := range block.Transactions {
|
|
sha := tx.TxHash()
|
|
if bytes.Equal(txid[:], sha[:]) {
|
|
return
|
|
}
|
|
}
|
|
|
|
t.Fatalf("tx was not included in block")
|
|
}
|
|
|
|
func rpcPointToWirePoint(t *harnessTest, chanPoint *lnrpc.ChannelPoint) wire.OutPoint {
|
|
txid, err := lnd.GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
|
|
return wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: chanPoint.OutputIndex,
|
|
}
|
|
}
|
|
|
|
// mineBlocks mine 'num' of blocks and check that blocks are present in
|
|
// node blockchain. numTxs should be set to the number of transactions
|
|
// (excluding the coinbase) we expect to be included in the first mined block.
|
|
func mineBlocks(t *harnessTest, net *lntest.NetworkHarness,
|
|
num uint32, numTxs int) []*wire.MsgBlock {
|
|
|
|
// If we expect transactions to be included in the blocks we'll mine,
|
|
// we wait here until they are seen in the miner's mempool.
|
|
var txids []*chainhash.Hash
|
|
var err error
|
|
if numTxs > 0 {
|
|
txids, err = waitForNTxsInMempool(
|
|
net.Miner.Node, numTxs, minerMempoolTimeout,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to find txns in mempool: %v", err)
|
|
}
|
|
}
|
|
|
|
blocks := make([]*wire.MsgBlock, num)
|
|
|
|
blockHashes, err := net.Miner.Node.Generate(num)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate blocks: %v", err)
|
|
}
|
|
|
|
for i, blockHash := range blockHashes {
|
|
block, err := net.Miner.Node.GetBlock(blockHash)
|
|
if err != nil {
|
|
t.Fatalf("unable to get block: %v", err)
|
|
}
|
|
|
|
blocks[i] = block
|
|
}
|
|
|
|
// Finally, assert that all the transactions were included in the first
|
|
// block.
|
|
for _, txid := range txids {
|
|
assertTxInBlock(t, blocks[0], txid)
|
|
}
|
|
|
|
return blocks
|
|
}
|
|
|
|
// openChannelStream blocks until an OpenChannel request for a channel funding
|
|
// by alice succeeds. If it does, a stream client is returned to receive events
|
|
// about the opening channel.
|
|
func openChannelStream(ctx context.Context, t *harnessTest,
|
|
net *lntest.NetworkHarness, alice, bob *lntest.HarnessNode,
|
|
p lntest.OpenChannelParams) lnrpc.Lightning_OpenChannelClient {
|
|
|
|
t.t.Helper()
|
|
|
|
// Wait until we are able to fund a channel successfully. This wait
|
|
// prevents us from erroring out when trying to create a channel while
|
|
// the node is starting up.
|
|
var chanOpenUpdate lnrpc.Lightning_OpenChannelClient
|
|
err := wait.NoError(func() error {
|
|
var err error
|
|
chanOpenUpdate, err = net.OpenChannel(ctx, alice, bob, p)
|
|
return err
|
|
}, defaultTimeout)
|
|
if err != nil {
|
|
t.Fatalf("unable to open channel: %v", err)
|
|
}
|
|
|
|
return chanOpenUpdate
|
|
}
|
|
|
|
// openChannelAndAssert attempts to open a channel with the specified
|
|
// parameters extended from Alice to Bob. Additionally, two items are asserted
|
|
// after the channel is considered open: the funding transaction should be
|
|
// found within a block, and that Alice can report the status of the new
|
|
// channel.
|
|
func openChannelAndAssert(ctx context.Context, t *harnessTest,
|
|
net *lntest.NetworkHarness, alice, bob *lntest.HarnessNode,
|
|
p lntest.OpenChannelParams) *lnrpc.ChannelPoint {
|
|
|
|
t.t.Helper()
|
|
|
|
chanOpenUpdate := openChannelStream(ctx, t, net, alice, bob, p)
|
|
|
|
// Mine 6 blocks, then wait for Alice's node to notify us that the
|
|
// channel has been opened. The funding transaction should be found
|
|
// within the first newly mined block. We mine 6 blocks so that in the
|
|
// case that the channel is public, it is announced to the network.
|
|
block := mineBlocks(t, net, 6, 1)[0]
|
|
|
|
fundingChanPoint, err := net.WaitForChannelOpen(ctx, chanOpenUpdate)
|
|
if err != nil {
|
|
t.Fatalf("error while waiting for channel open: %v", err)
|
|
}
|
|
fundingTxID, err := lnd.GetChanPointFundingTxid(fundingChanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
assertTxInBlock(t, block, fundingTxID)
|
|
|
|
// The channel should be listed in the peer information returned by
|
|
// both peers.
|
|
chanPoint := wire.OutPoint{
|
|
Hash: *fundingTxID,
|
|
Index: fundingChanPoint.OutputIndex,
|
|
}
|
|
if err := net.AssertChannelExists(ctx, alice, &chanPoint); err != nil {
|
|
t.Fatalf("unable to assert channel existence: %v", err)
|
|
}
|
|
if err := net.AssertChannelExists(ctx, bob, &chanPoint); err != nil {
|
|
t.Fatalf("unable to assert channel existence: %v", err)
|
|
}
|
|
|
|
return fundingChanPoint
|
|
}
|
|
|
|
// closeChannelAndAssert attempts to close a channel identified by the passed
|
|
// channel point owned by the passed Lightning node. A fully blocking channel
|
|
// closure is attempted, therefore the passed context should be a child derived
|
|
// via timeout from a base parent. Additionally, once the channel has been
|
|
// detected as closed, an assertion checks that the transaction is found within
|
|
// a block. Finally, this assertion verifies that the node always sends out a
|
|
// disable update when closing the channel if the channel was previously enabled.
|
|
//
|
|
// NOTE: This method assumes that the provided funding point is confirmed
|
|
// on-chain AND that the edge exists in the node's channel graph. If the funding
|
|
// transactions was reorged out at some point, use closeReorgedChannelAndAssert.
|
|
func closeChannelAndAssert(ctx context.Context, t *harnessTest,
|
|
net *lntest.NetworkHarness, node *lntest.HarnessNode,
|
|
fundingChanPoint *lnrpc.ChannelPoint, force bool) *chainhash.Hash {
|
|
|
|
return closeChannelAndAssertType(ctx, t, net, node, fundingChanPoint, false, force)
|
|
}
|
|
|
|
func closeChannelAndAssertType(ctx context.Context, t *harnessTest,
|
|
net *lntest.NetworkHarness, node *lntest.HarnessNode,
|
|
fundingChanPoint *lnrpc.ChannelPoint, anchors, force bool) *chainhash.Hash {
|
|
|
|
// Fetch the current channel policy. If the channel is currently
|
|
// enabled, we will register for graph notifications before closing to
|
|
// assert that the node sends out a disabling update as a result of the
|
|
// channel being closed.
|
|
curPolicy := getChannelPolicies(t, node, node.PubKeyStr, fundingChanPoint)[0]
|
|
expectDisable := !curPolicy.Disabled
|
|
|
|
// If the current channel policy is enabled, begin subscribing the graph
|
|
// updates before initiating the channel closure.
|
|
var graphSub *graphSubscription
|
|
if expectDisable {
|
|
sub := subscribeGraphNotifications(t, ctx, node)
|
|
graphSub = &sub
|
|
defer close(graphSub.quit)
|
|
}
|
|
|
|
closeUpdates, _, err := net.CloseChannel(ctx, node, fundingChanPoint, force)
|
|
if err != nil {
|
|
t.Fatalf("unable to close channel: %v", err)
|
|
}
|
|
|
|
// If the channel policy was enabled prior to the closure, wait until we
|
|
// received the disabled update.
|
|
if expectDisable {
|
|
curPolicy.Disabled = true
|
|
waitForChannelUpdate(
|
|
t, *graphSub,
|
|
[]expectedChanUpdate{
|
|
{node.PubKeyStr, curPolicy, fundingChanPoint},
|
|
},
|
|
)
|
|
}
|
|
|
|
return assertChannelClosed(
|
|
ctx, t, net, node, fundingChanPoint, anchors, closeUpdates,
|
|
)
|
|
}
|
|
|
|
// closeReorgedChannelAndAssert attempts to close a channel identified by the
|
|
// passed channel point owned by the passed Lightning node. A fully blocking
|
|
// channel closure is attempted, therefore the passed context should be a child
|
|
// derived via timeout from a base parent. Additionally, once the channel has
|
|
// been detected as closed, an assertion checks that the transaction is found
|
|
// within a block.
|
|
//
|
|
// NOTE: This method does not verify that the node sends a disable update for
|
|
// the closed channel.
|
|
func closeReorgedChannelAndAssert(ctx context.Context, t *harnessTest,
|
|
net *lntest.NetworkHarness, node *lntest.HarnessNode,
|
|
fundingChanPoint *lnrpc.ChannelPoint, force bool) *chainhash.Hash {
|
|
|
|
closeUpdates, _, err := net.CloseChannel(ctx, node, fundingChanPoint, force)
|
|
if err != nil {
|
|
t.Fatalf("unable to close channel: %v", err)
|
|
}
|
|
|
|
return assertChannelClosed(
|
|
ctx, t, net, node, fundingChanPoint, false, closeUpdates,
|
|
)
|
|
}
|
|
|
|
// assertChannelClosed asserts that the channel is properly cleaned up after
|
|
// initiating a cooperative or local close.
|
|
func assertChannelClosed(ctx context.Context, t *harnessTest,
|
|
net *lntest.NetworkHarness, node *lntest.HarnessNode,
|
|
fundingChanPoint *lnrpc.ChannelPoint, anchors bool,
|
|
closeUpdates lnrpc.Lightning_CloseChannelClient) *chainhash.Hash {
|
|
|
|
txid, err := lnd.GetChanPointFundingTxid(fundingChanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
chanPointStr := fmt.Sprintf("%v:%v", txid, fundingChanPoint.OutputIndex)
|
|
|
|
// If the channel appears in list channels, ensure that its state
|
|
// contains ChanStatusCoopBroadcasted.
|
|
ctxt, _ := context.WithTimeout(ctx, defaultTimeout)
|
|
listChansRequest := &lnrpc.ListChannelsRequest{}
|
|
listChansResp, err := node.ListChannels(ctxt, listChansRequest)
|
|
if err != nil {
|
|
t.Fatalf("unable to query for list channels: %v", err)
|
|
}
|
|
for _, channel := range listChansResp.Channels {
|
|
// Skip other channels.
|
|
if channel.ChannelPoint != chanPointStr {
|
|
continue
|
|
}
|
|
|
|
// Assert that the channel is in coop broadcasted.
|
|
if !strings.Contains(channel.ChanStatusFlags,
|
|
channeldb.ChanStatusCoopBroadcasted.String()) {
|
|
t.Fatalf("channel not coop broadcasted, "+
|
|
"got: %v", channel.ChanStatusFlags)
|
|
}
|
|
}
|
|
|
|
// At this point, the channel should now be marked as being in the
|
|
// state of "waiting close".
|
|
ctxt, _ = context.WithTimeout(ctx, defaultTimeout)
|
|
pendingChansRequest := &lnrpc.PendingChannelsRequest{}
|
|
pendingChanResp, err := node.PendingChannels(ctxt, pendingChansRequest)
|
|
if err != nil {
|
|
t.Fatalf("unable to query for pending channels: %v", err)
|
|
}
|
|
var found bool
|
|
for _, pendingClose := range pendingChanResp.WaitingCloseChannels {
|
|
if pendingClose.Channel.ChannelPoint == chanPointStr {
|
|
found = true
|
|
break
|
|
}
|
|
}
|
|
if !found {
|
|
t.Fatalf("channel not marked as waiting close")
|
|
}
|
|
|
|
// We'll now, generate a single block, wait for the final close status
|
|
// update, then ensure that the closing transaction was included in the
|
|
// block. If there are anchors, we also expect an anchor sweep.
|
|
expectedTxes := 1
|
|
if anchors {
|
|
expectedTxes = 2
|
|
}
|
|
|
|
block := mineBlocks(t, net, 1, expectedTxes)[0]
|
|
|
|
closingTxid, err := net.WaitForChannelClose(ctx, closeUpdates)
|
|
if err != nil {
|
|
t.Fatalf("error while waiting for channel close: %v", err)
|
|
}
|
|
|
|
assertTxInBlock(t, block, closingTxid)
|
|
|
|
// Finally, the transaction should no longer be in the waiting close
|
|
// state as we've just mined a block that should include the closing
|
|
// transaction.
|
|
err = wait.Predicate(func() bool {
|
|
pendingChansRequest := &lnrpc.PendingChannelsRequest{}
|
|
pendingChanResp, err := node.PendingChannels(
|
|
ctx, pendingChansRequest,
|
|
)
|
|
if err != nil {
|
|
return false
|
|
}
|
|
|
|
for _, pendingClose := range pendingChanResp.WaitingCloseChannels {
|
|
if pendingClose.Channel.ChannelPoint == chanPointStr {
|
|
return false
|
|
}
|
|
}
|
|
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("closing transaction not marked as fully closed")
|
|
}
|
|
|
|
return closingTxid
|
|
}
|
|
|
|
// waitForChannelPendingForceClose waits for the node to report that the
|
|
// channel is pending force close, and that the UTXO nursery is aware of it.
|
|
func waitForChannelPendingForceClose(ctx context.Context,
|
|
node *lntest.HarnessNode, fundingChanPoint *lnrpc.ChannelPoint) error {
|
|
|
|
txid, err := lnd.GetChanPointFundingTxid(fundingChanPoint)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
op := wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: fundingChanPoint.OutputIndex,
|
|
}
|
|
|
|
var predErr error
|
|
err = wait.Predicate(func() bool {
|
|
pendingChansRequest := &lnrpc.PendingChannelsRequest{}
|
|
pendingChanResp, err := node.PendingChannels(
|
|
ctx, pendingChansRequest,
|
|
)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to get pending "+
|
|
"channels: %v", err)
|
|
return false
|
|
}
|
|
|
|
forceClose, err := findForceClosedChannel(pendingChanResp, &op)
|
|
if err != nil {
|
|
predErr = err
|
|
return false
|
|
}
|
|
|
|
// We must wait until the UTXO nursery has received the channel
|
|
// and is aware of its maturity height.
|
|
if forceClose.MaturityHeight == 0 {
|
|
predErr = fmt.Errorf("channel had maturity height of 0")
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
return predErr
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// cleanupForceClose mines a force close commitment found in the mempool and
|
|
// the following sweep transaction from the force closing node.
|
|
func cleanupForceClose(t *harnessTest, net *lntest.NetworkHarness,
|
|
node *lntest.HarnessNode, chanPoint *lnrpc.ChannelPoint) {
|
|
ctxb := context.Background()
|
|
|
|
// Wait for the channel to be marked pending force close.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
err := waitForChannelPendingForceClose(ctxt, node, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("channel not pending force close: %v", err)
|
|
}
|
|
|
|
// Mine enough blocks for the node to sweep its funds from the force
|
|
// closed channel.
|
|
_, err = net.Miner.Node.Generate(defaultCSV)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate blocks: %v", err)
|
|
}
|
|
|
|
// The node should now sweep the funds, clean up by mining the sweeping
|
|
// tx.
|
|
mineBlocks(t, net, 1, 1)
|
|
}
|
|
|
|
// numOpenChannelsPending sends an RPC request to a node to get a count of the
|
|
// node's channels that are currently in a pending state (with a broadcast, but
|
|
// not confirmed funding transaction).
|
|
func numOpenChannelsPending(ctxt context.Context, node *lntest.HarnessNode) (int, error) {
|
|
pendingChansRequest := &lnrpc.PendingChannelsRequest{}
|
|
resp, err := node.PendingChannels(ctxt, pendingChansRequest)
|
|
if err != nil {
|
|
return 0, err
|
|
}
|
|
return len(resp.PendingOpenChannels), nil
|
|
}
|
|
|
|
// assertNumOpenChannelsPending asserts that a pair of nodes have the expected
|
|
// number of pending channels between them.
|
|
func assertNumOpenChannelsPending(ctxt context.Context, t *harnessTest,
|
|
alice, bob *lntest.HarnessNode, expected int) {
|
|
|
|
err := wait.NoError(func() error {
|
|
aliceNumChans, err := numOpenChannelsPending(ctxt, alice)
|
|
if err != nil {
|
|
return fmt.Errorf("error fetching alice's node (%v) "+
|
|
"pending channels %v", alice.NodeID, err)
|
|
}
|
|
bobNumChans, err := numOpenChannelsPending(ctxt, bob)
|
|
if err != nil {
|
|
return fmt.Errorf("error fetching bob's node (%v) "+
|
|
"pending channels %v", bob.NodeID, err)
|
|
}
|
|
|
|
aliceStateCorrect := aliceNumChans == expected
|
|
if !aliceStateCorrect {
|
|
return fmt.Errorf("number of pending channels for "+
|
|
"alice incorrect. expected %v, got %v",
|
|
expected, aliceNumChans)
|
|
}
|
|
|
|
bobStateCorrect := bobNumChans == expected
|
|
if !bobStateCorrect {
|
|
return fmt.Errorf("number of pending channels for bob "+
|
|
"incorrect. expected %v, got %v", expected,
|
|
bobNumChans)
|
|
}
|
|
|
|
return nil
|
|
}, 15*time.Second)
|
|
if err != nil {
|
|
t.Fatalf(err.Error())
|
|
}
|
|
}
|
|
|
|
// assertNumConnections asserts number current connections between two peers.
|
|
func assertNumConnections(t *harnessTest, alice, bob *lntest.HarnessNode,
|
|
expected int) {
|
|
ctxb := context.Background()
|
|
|
|
const nPolls = 10
|
|
|
|
tick := time.NewTicker(300 * time.Millisecond)
|
|
defer tick.Stop()
|
|
|
|
for i := nPolls - 1; i >= 0; i-- {
|
|
select {
|
|
case <-tick.C:
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
aNumPeers, err := alice.ListPeers(ctxt, &lnrpc.ListPeersRequest{})
|
|
if err != nil {
|
|
t.Fatalf("unable to fetch alice's node (%v) list peers %v",
|
|
alice.NodeID, err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
bNumPeers, err := bob.ListPeers(ctxt, &lnrpc.ListPeersRequest{})
|
|
if err != nil {
|
|
t.Fatalf("unable to fetch bob's node (%v) list peers %v",
|
|
bob.NodeID, err)
|
|
}
|
|
if len(aNumPeers.Peers) != expected {
|
|
// Continue polling if this is not the final
|
|
// loop.
|
|
if i > 0 {
|
|
continue
|
|
}
|
|
t.Fatalf("number of peers connected to alice is incorrect: "+
|
|
"expected %v, got %v", expected, len(aNumPeers.Peers))
|
|
}
|
|
if len(bNumPeers.Peers) != expected {
|
|
// Continue polling if this is not the final
|
|
// loop.
|
|
if i > 0 {
|
|
continue
|
|
}
|
|
t.Fatalf("number of peers connected to bob is incorrect: "+
|
|
"expected %v, got %v", expected, len(bNumPeers.Peers))
|
|
}
|
|
|
|
// Alice and Bob both have the required number of
|
|
// peers, stop polling and return to caller.
|
|
return
|
|
}
|
|
}
|
|
}
|
|
|
|
// shutdownAndAssert shuts down the given node and asserts that no errors
|
|
// occur.
|
|
func shutdownAndAssert(net *lntest.NetworkHarness, t *harnessTest,
|
|
node *lntest.HarnessNode) {
|
|
if err := net.ShutdownNode(node); err != nil {
|
|
t.Fatalf("unable to shutdown %v: %v", node.Name(), err)
|
|
}
|
|
}
|
|
|
|
// completePaymentRequests sends payments from a lightning node to complete all
|
|
// payment requests. If the awaitResponse parameter is true, this function
|
|
// does not return until all payments successfully complete without errors.
|
|
func completePaymentRequests(ctx context.Context, client lnrpc.LightningClient,
|
|
paymentRequests []string, awaitResponse bool) error {
|
|
|
|
// We start by getting the current state of the client's channels. This
|
|
// is needed to ensure the payments actually have been committed before
|
|
// we return.
|
|
ctxt, _ := context.WithTimeout(ctx, defaultTimeout)
|
|
req := &lnrpc.ListChannelsRequest{}
|
|
listResp, err := client.ListChannels(ctxt, req)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
ctxc, cancel := context.WithCancel(ctx)
|
|
defer cancel()
|
|
|
|
payStream, err := client.SendPayment(ctxc)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
for _, payReq := range paymentRequests {
|
|
sendReq := &lnrpc.SendRequest{
|
|
PaymentRequest: payReq,
|
|
}
|
|
err := payStream.Send(sendReq)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
}
|
|
|
|
if awaitResponse {
|
|
for range paymentRequests {
|
|
resp, err := payStream.Recv()
|
|
if err != nil {
|
|
return err
|
|
}
|
|
if resp.PaymentError != "" {
|
|
return fmt.Errorf("received payment error: %v",
|
|
resp.PaymentError)
|
|
}
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// We are not waiting for feedback in the form of a response, but we
|
|
// should still wait long enough for the server to receive and handle
|
|
// the send before cancelling the request. We wait for the number of
|
|
// updates to one of our channels has increased before we return.
|
|
err = wait.Predicate(func() bool {
|
|
ctxt, _ = context.WithTimeout(ctx, defaultTimeout)
|
|
newListResp, err := client.ListChannels(ctxt, req)
|
|
if err != nil {
|
|
return false
|
|
}
|
|
|
|
for _, c1 := range listResp.Channels {
|
|
for _, c2 := range newListResp.Channels {
|
|
if c1.ChannelPoint != c2.ChannelPoint {
|
|
continue
|
|
}
|
|
|
|
// If this channel has an increased numbr of
|
|
// updates, we assume the payments are
|
|
// committed, and we can return.
|
|
if c2.NumUpdates > c1.NumUpdates {
|
|
return true
|
|
}
|
|
}
|
|
}
|
|
|
|
return false
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// makeFakePayHash creates random pre image hash
|
|
func makeFakePayHash(t *harnessTest) []byte {
|
|
randBuf := make([]byte, 32)
|
|
|
|
if _, err := rand.Read(randBuf); err != nil {
|
|
t.Fatalf("internal error, cannot generate random string: %v", err)
|
|
}
|
|
|
|
return randBuf
|
|
}
|
|
|
|
// createPayReqs is a helper method that will create a slice of payment
|
|
// requests for the given node.
|
|
func createPayReqs(node *lntest.HarnessNode, paymentAmt btcutil.Amount,
|
|
numInvoices int) ([]string, [][]byte, []*lnrpc.Invoice, error) {
|
|
|
|
payReqs := make([]string, numInvoices)
|
|
rHashes := make([][]byte, numInvoices)
|
|
invoices := make([]*lnrpc.Invoice, numInvoices)
|
|
for i := 0; i < numInvoices; i++ {
|
|
preimage := make([]byte, 32)
|
|
_, err := rand.Read(preimage)
|
|
if err != nil {
|
|
return nil, nil, nil, fmt.Errorf("unable to generate "+
|
|
"preimage: %v", err)
|
|
}
|
|
invoice := &lnrpc.Invoice{
|
|
Memo: "testing",
|
|
RPreimage: preimage,
|
|
Value: int64(paymentAmt),
|
|
}
|
|
ctxt, _ := context.WithTimeout(
|
|
context.Background(), defaultTimeout,
|
|
)
|
|
resp, err := node.AddInvoice(ctxt, invoice)
|
|
if err != nil {
|
|
return nil, nil, nil, fmt.Errorf("unable to add "+
|
|
"invoice: %v", err)
|
|
}
|
|
|
|
payReqs[i] = resp.PaymentRequest
|
|
rHashes[i] = resp.RHash
|
|
invoices[i] = invoice
|
|
}
|
|
return payReqs, rHashes, invoices, nil
|
|
}
|
|
|
|
// getChanInfo is a helper method for getting channel info for a node's sole
|
|
// channel.
|
|
func getChanInfo(ctx context.Context, node *lntest.HarnessNode) (
|
|
*lnrpc.Channel, error) {
|
|
|
|
req := &lnrpc.ListChannelsRequest{}
|
|
channelInfo, err := node.ListChannels(ctx, req)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
if len(channelInfo.Channels) != 1 {
|
|
return nil, fmt.Errorf("node should only have a single "+
|
|
"channel, instead it has %v", len(channelInfo.Channels))
|
|
}
|
|
|
|
return channelInfo.Channels[0], nil
|
|
}
|
|
|
|
const (
|
|
AddrTypeWitnessPubkeyHash = lnrpc.AddressType_WITNESS_PUBKEY_HASH
|
|
AddrTypeNestedPubkeyHash = lnrpc.AddressType_NESTED_PUBKEY_HASH
|
|
)
|
|
|
|
// testOnchainFundRecovery checks lnd's ability to rescan for onchain outputs
|
|
// when providing a valid aezeed that owns outputs on the chain. This test
|
|
// performs multiple restorations using the same seed and various recovery
|
|
// windows to ensure we detect funds properly.
|
|
func testOnchainFundRecovery(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
// First, create a new node with strong passphrase and grab the mnemonic
|
|
// used for key derivation. This will bring up Carol with an empty
|
|
// wallet, and such that she is synced up.
|
|
password := []byte("The Magic Words are Squeamish Ossifrage")
|
|
carol, mnemonic, err := net.NewNodeWithSeed("Carol", nil, password)
|
|
if err != nil {
|
|
t.Fatalf("unable to create node with seed; %v", err)
|
|
}
|
|
shutdownAndAssert(net, t, carol)
|
|
|
|
// Create a closure for testing the recovery of Carol's wallet. This
|
|
// method takes the expected value of Carol's balance when using the
|
|
// given recovery window. Additionally, the caller can specify an action
|
|
// to perform on the restored node before the node is shutdown.
|
|
restoreCheckBalance := func(expAmount int64, expectedNumUTXOs uint32,
|
|
recoveryWindow int32, fn func(*lntest.HarnessNode)) {
|
|
|
|
// Restore Carol, passing in the password, mnemonic, and
|
|
// desired recovery window.
|
|
node, err := net.RestoreNodeWithSeed(
|
|
"Carol", nil, password, mnemonic, recoveryWindow, nil,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to restore node: %v", err)
|
|
}
|
|
|
|
// Query carol for her current wallet balance, and also that we
|
|
// gain the expected number of UTXOs.
|
|
var (
|
|
currBalance int64
|
|
currNumUTXOs uint32
|
|
)
|
|
err = wait.Predicate(func() bool {
|
|
req := &lnrpc.WalletBalanceRequest{}
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := node.WalletBalance(ctxt, req)
|
|
if err != nil {
|
|
t.Fatalf("unable to query wallet balance: %v",
|
|
err)
|
|
}
|
|
currBalance = resp.ConfirmedBalance
|
|
|
|
utxoReq := &lnrpc.ListUnspentRequest{
|
|
MaxConfs: math.MaxInt32,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
utxoResp, err := node.ListUnspent(ctxt, utxoReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to query utxos: %v", err)
|
|
}
|
|
currNumUTXOs = uint32(len(utxoResp.Utxos))
|
|
|
|
// Verify that Carol's balance and number of UTXOs
|
|
// matches what's expected.
|
|
if expAmount != currBalance {
|
|
return false
|
|
}
|
|
if currNumUTXOs != expectedNumUTXOs {
|
|
return false
|
|
}
|
|
|
|
return true
|
|
}, 15*time.Second)
|
|
if err != nil {
|
|
t.Fatalf("expected restored node to have %d satoshis, "+
|
|
"instead has %d satoshis, expected %d utxos "+
|
|
"instead has %d", expAmount, currBalance,
|
|
expectedNumUTXOs, currNumUTXOs)
|
|
}
|
|
|
|
// If the user provided a callback, execute the commands against
|
|
// the restored Carol.
|
|
if fn != nil {
|
|
fn(node)
|
|
}
|
|
|
|
// Lastly, shutdown this Carol so we can move on to the next
|
|
// restoration.
|
|
shutdownAndAssert(net, t, node)
|
|
}
|
|
|
|
// Create a closure-factory for building closures that can generate and
|
|
// skip a configurable number of addresses, before finally sending coins
|
|
// to a next generated address. The returned closure will apply the same
|
|
// behavior to both default P2WKH and NP2WKH scopes.
|
|
skipAndSend := func(nskip int) func(*lntest.HarnessNode) {
|
|
return func(node *lntest.HarnessNode) {
|
|
newP2WKHAddrReq := &lnrpc.NewAddressRequest{
|
|
Type: AddrTypeWitnessPubkeyHash,
|
|
}
|
|
|
|
newNP2WKHAddrReq := &lnrpc.NewAddressRequest{
|
|
Type: AddrTypeNestedPubkeyHash,
|
|
}
|
|
|
|
// Generate and skip the number of addresses requested.
|
|
for i := 0; i < nskip; i++ {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
_, err = node.NewAddress(ctxt, newP2WKHAddrReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate new "+
|
|
"p2wkh address: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
_, err = node.NewAddress(ctxt, newNP2WKHAddrReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate new "+
|
|
"np2wkh address: %v", err)
|
|
}
|
|
}
|
|
|
|
// Send one BTC to the next P2WKH address.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(
|
|
ctxt, btcutil.SatoshiPerBitcoin, node,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to node: %v",
|
|
err)
|
|
}
|
|
|
|
// And another to the next NP2WKH address.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoinsNP2WKH(
|
|
ctxt, btcutil.SatoshiPerBitcoin, node,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to node: %v",
|
|
err)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Restore Carol with a recovery window of 0. Since no coins have been
|
|
// sent, her balance should be zero.
|
|
//
|
|
// After, one BTC is sent to both her first external P2WKH and NP2WKH
|
|
// addresses.
|
|
restoreCheckBalance(0, 0, 0, skipAndSend(0))
|
|
|
|
// Check that restoring without a look-ahead results in having no funds
|
|
// in the wallet, even though they exist on-chain.
|
|
restoreCheckBalance(0, 0, 0, nil)
|
|
|
|
// Now, check that using a look-ahead of 1 recovers the balance from
|
|
// the two transactions above. We should also now have 2 UTXOs in the
|
|
// wallet at the end of the recovery attempt.
|
|
//
|
|
// After, we will generate and skip 9 P2WKH and NP2WKH addresses, and
|
|
// send another BTC to the subsequent 10th address in each derivation
|
|
// path.
|
|
restoreCheckBalance(2*btcutil.SatoshiPerBitcoin, 2, 1, skipAndSend(9))
|
|
|
|
// Check that using a recovery window of 9 does not find the two most
|
|
// recent txns.
|
|
restoreCheckBalance(2*btcutil.SatoshiPerBitcoin, 2, 9, nil)
|
|
|
|
// Extending our recovery window to 10 should find the most recent
|
|
// transactions, leaving the wallet with 4 BTC total. We should also
|
|
// learn of the two additional UTXOs created above.
|
|
//
|
|
// After, we will skip 19 more addrs, sending to the 20th address past
|
|
// our last found address, and repeat the same checks.
|
|
restoreCheckBalance(4*btcutil.SatoshiPerBitcoin, 4, 10, skipAndSend(19))
|
|
|
|
// Check that recovering with a recovery window of 19 fails to find the
|
|
// most recent transactions.
|
|
restoreCheckBalance(4*btcutil.SatoshiPerBitcoin, 4, 19, nil)
|
|
|
|
// Ensure that using a recovery window of 20 succeeds with all UTXOs
|
|
// found and the final balance reflected.
|
|
|
|
// After these checks are done, we'll want to make sure we can also
|
|
// recover change address outputs. This is mainly motivated by a now
|
|
// fixed bug in the wallet in which change addresses could at times be
|
|
// created outside of the default key scopes. Recovery only used to be
|
|
// performed on the default key scopes, so ideally this test case
|
|
// would've caught the bug earlier. Carol has received 6 BTC so far from
|
|
// the miner, we'll send 5 back to ensure all of her UTXOs get spent to
|
|
// avoid fee discrepancies and a change output is formed.
|
|
const minerAmt = 5 * btcutil.SatoshiPerBitcoin
|
|
const finalBalance = 6 * btcutil.SatoshiPerBitcoin
|
|
promptChangeAddr := func(node *lntest.HarnessNode) {
|
|
minerAddr, err := net.Miner.NewAddress()
|
|
if err != nil {
|
|
t.Fatalf("unable to create new miner address: %v", err)
|
|
}
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := node.SendCoins(ctxt, &lnrpc.SendCoinsRequest{
|
|
Addr: minerAddr.String(),
|
|
Amount: minerAmt,
|
|
})
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to miner: %v", err)
|
|
}
|
|
txid, err := waitForTxInMempool(
|
|
net.Miner.Node, minerMempoolTimeout,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("transaction not found in mempool: %v", err)
|
|
}
|
|
if resp.Txid != txid.String() {
|
|
t.Fatalf("txid mismatch: %v vs %v", resp.Txid,
|
|
txid.String())
|
|
}
|
|
block := mineBlocks(t, net, 1, 1)[0]
|
|
assertTxInBlock(t, block, txid)
|
|
}
|
|
restoreCheckBalance(finalBalance, 6, 20, promptChangeAddr)
|
|
|
|
// We should expect a static fee of 27750 satoshis for spending 6 inputs
|
|
// (3 P2WPKH, 3 NP2WPKH) to two P2WPKH outputs. Carol should therefore
|
|
// only have one UTXO present (the change output) of 6 - 5 - fee BTC.
|
|
const fee = 27750
|
|
restoreCheckBalance(finalBalance-minerAmt-fee, 1, 21, nil)
|
|
}
|
|
|
|
// commitType is a simple enum used to run though the basic funding flow with
|
|
// different commitment formats.
|
|
type commitType byte
|
|
|
|
const (
|
|
// commitTypeLegacy is the old school commitment type.
|
|
commitTypeLegacy commitType = iota
|
|
|
|
// commiTypeTweakless is the commitment type where the remote key is
|
|
// static (non-tweaked).
|
|
commitTypeTweakless
|
|
|
|
// commitTypeAnchors is the kind of commitment that has extra outputs
|
|
// used for anchoring down to commitment using CPFP.
|
|
commitTypeAnchors
|
|
)
|
|
|
|
// String returns that name of the commitment type.
|
|
func (c commitType) String() string {
|
|
switch c {
|
|
case commitTypeLegacy:
|
|
return "legacy"
|
|
case commitTypeTweakless:
|
|
return "tweakless"
|
|
case commitTypeAnchors:
|
|
return "anchors"
|
|
default:
|
|
return "invalid"
|
|
}
|
|
}
|
|
|
|
// Args returns the command line flag to supply to enable this commitment type.
|
|
func (c commitType) Args() []string {
|
|
switch c {
|
|
case commitTypeLegacy:
|
|
return []string{"--protocol.committweak"}
|
|
case commitTypeTweakless:
|
|
return []string{}
|
|
case commitTypeAnchors:
|
|
return []string{"--protocol.anchors"}
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// calcStaticFee calculates appropriate fees for commitment transactions. This
|
|
// function provides a simple way to allow test balance assertions to take fee
|
|
// calculations into account.
|
|
func (c commitType) calcStaticFee(numHTLCs int) btcutil.Amount {
|
|
const htlcWeight = input.HTLCWeight
|
|
var (
|
|
feePerKw = chainfee.SatPerKVByte(50000).FeePerKWeight()
|
|
commitWeight = input.CommitWeight
|
|
anchors = btcutil.Amount(0)
|
|
)
|
|
|
|
// The anchor commitment type is slightly heavier, and we must also add
|
|
// the value of the two anchors to the resulting fee the initiator
|
|
// pays.
|
|
if c == commitTypeAnchors {
|
|
commitWeight = input.AnchorCommitWeight
|
|
anchors = 2 * anchorSize
|
|
}
|
|
|
|
return feePerKw.FeeForWeight(int64(commitWeight+htlcWeight*numHTLCs)) +
|
|
anchors
|
|
}
|
|
|
|
// channelCommitType retrieves the active channel commitment type for the given
|
|
// chan point.
|
|
func channelCommitType(node *lntest.HarnessNode,
|
|
chanPoint *lnrpc.ChannelPoint) (commitType, error) {
|
|
|
|
ctxb := context.Background()
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
|
|
req := &lnrpc.ListChannelsRequest{}
|
|
channels, err := node.ListChannels(ctxt, req)
|
|
if err != nil {
|
|
return 0, fmt.Errorf("listchannels failed: %v", err)
|
|
}
|
|
|
|
for _, c := range channels.Channels {
|
|
if c.ChannelPoint == txStr(chanPoint) {
|
|
switch c.CommitmentType {
|
|
|
|
// If the anchor output size is non-zero, we are
|
|
// dealing with the anchor type.
|
|
case lnrpc.CommitmentType_ANCHORS:
|
|
return commitTypeAnchors, nil
|
|
|
|
// StaticRemoteKey means it is tweakless,
|
|
case lnrpc.CommitmentType_STATIC_REMOTE_KEY:
|
|
return commitTypeTweakless, nil
|
|
|
|
// Otherwise legacy.
|
|
default:
|
|
return commitTypeLegacy, nil
|
|
}
|
|
}
|
|
}
|
|
|
|
return 0, fmt.Errorf("channel point %v not found", chanPoint)
|
|
}
|
|
|
|
// basicChannelFundingTest is a sub-test of the main testBasicChannelFunding
|
|
// test. Given two nodes: Alice and Bob, it'll assert proper channel creation,
|
|
// then return a function closure that should be called to assert proper
|
|
// channel closure.
|
|
func basicChannelFundingTest(t *harnessTest, net *lntest.NetworkHarness,
|
|
alice *lntest.HarnessNode, bob *lntest.HarnessNode,
|
|
fundingShim *lnrpc.FundingShim) (*lnrpc.Channel, *lnrpc.Channel, func(), error) {
|
|
|
|
chanAmt := lnd.MaxBtcFundingAmount
|
|
pushAmt := btcutil.Amount(100000)
|
|
|
|
// First establish a channel with a capacity of 0.5 BTC between Alice
|
|
// and Bob with Alice pushing 100k satoshis to Bob's side during
|
|
// funding. This function will block until the channel itself is fully
|
|
// open or an error occurs in the funding process. A series of
|
|
// assertions will be executed to ensure the funding process completed
|
|
// successfully.
|
|
ctxb := context.Background()
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, alice, bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
FundingShim: fundingShim,
|
|
},
|
|
)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err := alice.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
return nil, nil, nil, fmt.Errorf("alice didn't report "+
|
|
"channel: %v", err)
|
|
}
|
|
err = bob.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
return nil, nil, nil, fmt.Errorf("bob didn't report "+
|
|
"channel: %v", err)
|
|
}
|
|
|
|
cType, err := channelCommitType(alice, chanPoint)
|
|
if err != nil {
|
|
return nil, nil, nil, fmt.Errorf("unable to get channel "+
|
|
"type: %v", err)
|
|
}
|
|
|
|
// With the channel open, ensure that the amount specified above has
|
|
// properly been pushed to Bob.
|
|
balReq := &lnrpc.ChannelBalanceRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
aliceBal, err := alice.ChannelBalance(ctxt, balReq)
|
|
if err != nil {
|
|
return nil, nil, nil, fmt.Errorf("unable to get alice's "+
|
|
"balance: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
bobBal, err := bob.ChannelBalance(ctxt, balReq)
|
|
if err != nil {
|
|
return nil, nil, nil, fmt.Errorf("unable to get bobs's "+
|
|
"balance: %v", err)
|
|
}
|
|
|
|
expBalanceAlice := chanAmt - pushAmt - cType.calcStaticFee(0)
|
|
aliceBalance := btcutil.Amount(aliceBal.Balance)
|
|
if aliceBalance != expBalanceAlice {
|
|
return nil, nil, nil, fmt.Errorf("alice's balance is "+
|
|
"incorrect: expected %v got %v",
|
|
expBalanceAlice, aliceBalance)
|
|
}
|
|
|
|
bobBalance := btcutil.Amount(bobBal.Balance)
|
|
if bobBalance != pushAmt {
|
|
return nil, nil, nil, fmt.Errorf("bob's balance is incorrect: "+
|
|
"expected %v got %v", pushAmt, bobBalance)
|
|
}
|
|
|
|
req := &lnrpc.ListChannelsRequest{}
|
|
aliceChannel, err := alice.ListChannels(context.Background(), req)
|
|
if err != nil {
|
|
return nil, nil, nil, fmt.Errorf("unable to obtain chan: %v", err)
|
|
}
|
|
|
|
bobChannel, err := bob.ListChannels(context.Background(), req)
|
|
if err != nil {
|
|
return nil, nil, nil, fmt.Errorf("unable to obtain chan: %v", err)
|
|
}
|
|
|
|
closeChan := func() {
|
|
// Finally, immediately close the channel. This function will
|
|
// also block until the channel is closed and will additionally
|
|
// assert the relevant channel closing post conditions.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, alice, chanPoint, false)
|
|
}
|
|
|
|
return aliceChannel.Channels[0], bobChannel.Channels[0], closeChan, nil
|
|
}
|
|
|
|
// testBasicChannelFunding performs a test exercising expected behavior from a
|
|
// basic funding workflow. The test creates a new channel between Alice and
|
|
// Bob, then immediately closes the channel after asserting some expected post
|
|
// conditions. Finally, the chain itself is checked to ensure the closing
|
|
// transaction was mined.
|
|
func testBasicChannelFunding(net *lntest.NetworkHarness, t *harnessTest) {
|
|
|
|
ctxb := context.Background()
|
|
|
|
// Run through the test with combinations of all the different
|
|
// commitment types.
|
|
allTypes := []commitType{
|
|
commitTypeLegacy,
|
|
commitTypeTweakless,
|
|
commitTypeAnchors,
|
|
}
|
|
|
|
test:
|
|
// We'll test all possible combinations of the feature bit presence
|
|
// that both nodes can signal for this new channel type. We'll make a
|
|
// new Carol+Dave for each test instance as well.
|
|
for _, carolCommitType := range allTypes {
|
|
for _, daveCommitType := range allTypes {
|
|
// Based on the current tweak variable for Carol, we'll
|
|
// preferentially signal the legacy commitment format.
|
|
// We do the same for Dave shortly below.
|
|
carolArgs := carolCommitType.Args()
|
|
carol, err := net.NewNode("Carol", carolArgs)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new node: %v", err)
|
|
}
|
|
|
|
// Each time, we'll send Carol a new set of coins in
|
|
// order to fund the channel.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
|
|
daveArgs := daveCommitType.Args()
|
|
dave, err := net.NewNode("Dave", daveArgs)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new node: %v", err)
|
|
}
|
|
|
|
// Before we start the test, we'll ensure both sides
|
|
// are connected to the funding flow can properly be
|
|
// executed.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.EnsureConnected(ctxt, carol, dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to connect peers: %v", err)
|
|
}
|
|
|
|
testName := fmt.Sprintf("carol_commit=%v,dave_commit=%v",
|
|
carolCommitType, daveCommitType)
|
|
|
|
ht := t
|
|
success := t.t.Run(testName, func(t *testing.T) {
|
|
carolChannel, daveChannel, closeChan, err := basicChannelFundingTest(
|
|
ht, net, carol, dave, nil,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("failed funding flow: %v", err)
|
|
}
|
|
|
|
// Both nodes should report the same commitment
|
|
// type.
|
|
chansCommitType := carolChannel.CommitmentType
|
|
if daveChannel.CommitmentType != chansCommitType {
|
|
t.Fatalf("commit types don't match, "+
|
|
"carol got %v, dave got %v",
|
|
carolChannel.CommitmentType,
|
|
daveChannel.CommitmentType,
|
|
)
|
|
}
|
|
|
|
// Now check that the commitment type reported
|
|
// by both nodes is what we expect. It will be
|
|
// the minimum of the two nodes' preference, in
|
|
// the order Legacy, Tweakless, Anchors.
|
|
expType := carolCommitType
|
|
|
|
switch daveCommitType {
|
|
|
|
// Dave supports anchors, type will be what
|
|
// Carol supports.
|
|
case commitTypeAnchors:
|
|
|
|
// Dave only supports tweakless, channel will
|
|
// be downgraded to this type if Carol supports
|
|
// anchors.
|
|
case commitTypeTweakless:
|
|
if expType == commitTypeAnchors {
|
|
expType = commitTypeTweakless
|
|
}
|
|
|
|
// Dave only supoprts legacy type, channel will
|
|
// be downgraded to this type.
|
|
case commitTypeLegacy:
|
|
expType = commitTypeLegacy
|
|
|
|
default:
|
|
t.Fatalf("invalid commit type %v",
|
|
daveCommitType)
|
|
}
|
|
|
|
// Check that the signalled type matches what we
|
|
// expect.
|
|
switch {
|
|
case expType == commitTypeAnchors && chansCommitType == lnrpc.CommitmentType_ANCHORS:
|
|
case expType == commitTypeTweakless && chansCommitType == lnrpc.CommitmentType_STATIC_REMOTE_KEY:
|
|
case expType == commitTypeLegacy && chansCommitType == lnrpc.CommitmentType_LEGACY:
|
|
|
|
default:
|
|
t.Fatalf("expected nodes to signal "+
|
|
"commit type %v, instead got "+
|
|
"%v", expType, chansCommitType)
|
|
}
|
|
|
|
// As we've concluded this sub-test case we'll
|
|
// now close out the channel for both sides.
|
|
closeChan()
|
|
})
|
|
if !success {
|
|
break test
|
|
}
|
|
|
|
shutdownAndAssert(net, t, carol)
|
|
shutdownAndAssert(net, t, dave)
|
|
}
|
|
}
|
|
}
|
|
|
|
// testUnconfirmedChannelFunding tests that our unconfirmed change outputs can
|
|
// be used to fund channels.
|
|
func testUnconfirmedChannelFunding(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const (
|
|
chanAmt = lnd.MaxBtcFundingAmount
|
|
pushAmt = btcutil.Amount(100000)
|
|
)
|
|
|
|
// We'll start off by creating a node for Carol.
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create carol's node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
// We'll send her some confirmed funds.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, 2*chanAmt, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
|
|
// Now let Carol send some funds to herself, making a unconfirmed
|
|
// change output.
|
|
addrReq := &lnrpc.NewAddressRequest{
|
|
Type: lnrpc.AddressType_WITNESS_PUBKEY_HASH,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := carol.NewAddress(ctxt, addrReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get new address: %v", err)
|
|
}
|
|
|
|
sendReq := &lnrpc.SendCoinsRequest{
|
|
Addr: resp.Address,
|
|
Amount: int64(chanAmt) / 5,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
_, err = carol.SendCoins(ctxt, sendReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins: %v", err)
|
|
}
|
|
|
|
// Make sure the unconfirmed tx is seen in the mempool.
|
|
_, err = waitForTxInMempool(net.Miner.Node, minerMempoolTimeout)
|
|
if err != nil {
|
|
t.Fatalf("failed to find tx in miner mempool: %v", err)
|
|
}
|
|
|
|
// Now, we'll connect her to Alice so that they can open a channel
|
|
// together. The funding flow should select Carol's unconfirmed output
|
|
// as she doesn't have any other funds since it's a new node.
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, net.Alice); err != nil {
|
|
t.Fatalf("unable to connect dave to alice: %v", err)
|
|
}
|
|
|
|
chanOpenUpdate := openChannelStream(
|
|
ctxt, t, net, carol, net.Alice,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
SpendUnconfirmed: true,
|
|
},
|
|
)
|
|
|
|
// Confirm the channel and wait for it to be recognized by both
|
|
// parties. Two transactions should be mined, the unconfirmed spend and
|
|
// the funding tx.
|
|
mineBlocks(t, net, 6, 2)
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
chanPoint, err := net.WaitForChannelOpen(ctxt, chanOpenUpdate)
|
|
if err != nil {
|
|
t.Fatalf("error while waiting for channel open: %v", err)
|
|
}
|
|
|
|
cType, err := channelCommitType(net.Alice, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get channel type: %v", err)
|
|
}
|
|
|
|
// With the channel open, we'll check the balances on each side of the
|
|
// channel as a sanity check to ensure things worked out as intended.
|
|
balReq := &lnrpc.ChannelBalanceRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolBal, err := carol.ChannelBalance(ctxt, balReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol's balance: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
aliceBal, err := net.Alice.ChannelBalance(ctxt, balReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get alice's balance: %v", err)
|
|
}
|
|
if carolBal.Balance != int64(chanAmt-pushAmt-cType.calcStaticFee(0)) {
|
|
t.Fatalf("carol's balance is incorrect: expected %v got %v",
|
|
chanAmt-pushAmt-cType.calcStaticFee(0), carolBal)
|
|
}
|
|
if aliceBal.Balance != int64(pushAmt) {
|
|
t.Fatalf("alice's balance is incorrect: expected %v got %v",
|
|
pushAmt, aliceBal.Balance)
|
|
}
|
|
|
|
// Now that we're done with the test, the channel can be closed.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPoint, false)
|
|
}
|
|
|
|
// testPaymentFollowingChannelOpen tests that the channel transition from
|
|
// 'pending' to 'open' state does not cause any inconsistencies within other
|
|
// subsystems trying to udpate the channel state in the db. We follow this
|
|
// transition with a payment that updates the commitment state and verify that
|
|
// the pending state is up to date.
|
|
func testPaymentFollowingChannelOpen(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const paymentAmt = btcutil.Amount(100)
|
|
channelCapacity := btcutil.Amount(paymentAmt * 1000)
|
|
|
|
// We first establish a channel between Alice and Bob.
|
|
ctxt, cancel := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
defer cancel()
|
|
pendingUpdate, err := net.OpenPendingChannel(
|
|
ctxt, net.Alice, net.Bob, channelCapacity, 0,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to open channel: %v", err)
|
|
}
|
|
|
|
// At this point, the channel's funding transaction will have been
|
|
// broadcast, but not confirmed. Alice and Bob's nodes
|
|
// should reflect this when queried via RPC.
|
|
ctxt, cancel = context.WithTimeout(ctxb, defaultTimeout)
|
|
defer cancel()
|
|
assertNumOpenChannelsPending(ctxt, t, net.Alice, net.Bob, 1)
|
|
|
|
// We are restarting Bob's node to let the link be created for the
|
|
// pending channel.
|
|
if err := net.RestartNode(net.Bob, nil); err != nil {
|
|
t.Fatalf("Bob restart failed: %v", err)
|
|
}
|
|
|
|
// We ensure that Bob reconnets to Alice.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.EnsureConnected(ctxt, net.Bob, net.Alice); err != nil {
|
|
t.Fatalf("peers unable to reconnect after restart: %v", err)
|
|
}
|
|
|
|
// We mine one block for the channel to be confirmed.
|
|
_ = mineBlocks(t, net, 6, 1)[0]
|
|
|
|
// We verify that the chanel is open from both nodes point of view.
|
|
ctxt, cancel = context.WithTimeout(ctxb, defaultTimeout)
|
|
defer cancel()
|
|
assertNumOpenChannelsPending(ctxt, t, net.Alice, net.Bob, 0)
|
|
|
|
// With the channel open, we'll create invoices for Bob that Alice will
|
|
// pay to in order to advance the state of the channel.
|
|
bobPayReqs, _, _, err := createPayReqs(
|
|
net.Bob, paymentAmt, 1,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// Send payment to Bob so that a channel update to disk will be
|
|
// executed.
|
|
ctxt, cancel = context.WithTimeout(ctxb, defaultTimeout)
|
|
defer cancel()
|
|
_, err = net.Alice.SendPaymentSync(
|
|
ctxt,
|
|
&lnrpc.SendRequest{
|
|
PaymentRequest: bobPayReqs[0],
|
|
},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create payment stream for alice: %v", err)
|
|
}
|
|
|
|
// At this point we want to make sure the channel is opened and not
|
|
// pending.
|
|
ctxt, cancel = context.WithTimeout(ctxb, defaultTimeout)
|
|
defer cancel()
|
|
res, err := net.Bob.ListChannels(ctxt, &lnrpc.ListChannelsRequest{})
|
|
if err != nil {
|
|
t.Fatalf("unable to list bob channels: %v", err)
|
|
}
|
|
if len(res.Channels) == 0 {
|
|
t.Fatalf("bob list of channels is empty")
|
|
}
|
|
|
|
// Finally, immediately close the channel. This function will also
|
|
// block until the channel is closed and will additionally assert the
|
|
// relevant channel closing post conditions.
|
|
chanPoint := &lnrpc.ChannelPoint{
|
|
FundingTxid: &lnrpc.ChannelPoint_FundingTxidBytes{
|
|
FundingTxidBytes: pendingUpdate.Txid,
|
|
},
|
|
OutputIndex: pendingUpdate.OutputIndex,
|
|
}
|
|
ctxt, cancel = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
defer cancel()
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPoint, false)
|
|
}
|
|
|
|
// txStr returns the string representation of the channel's funding transaction.
|
|
func txStr(chanPoint *lnrpc.ChannelPoint) string {
|
|
fundingTxID, err := lnd.GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
return ""
|
|
}
|
|
cp := wire.OutPoint{
|
|
Hash: *fundingTxID,
|
|
Index: chanPoint.OutputIndex,
|
|
}
|
|
return cp.String()
|
|
}
|
|
|
|
// expectedChanUpdate houses params we expect a ChannelUpdate to advertise.
|
|
type expectedChanUpdate struct {
|
|
advertisingNode string
|
|
expectedPolicy *lnrpc.RoutingPolicy
|
|
chanPoint *lnrpc.ChannelPoint
|
|
}
|
|
|
|
// calculateMaxHtlc re-implements the RequiredRemoteChannelReserve of the
|
|
// funding manager's config, which corresponds to the maximum MaxHTLC value we
|
|
// allow users to set when updating a channel policy.
|
|
func calculateMaxHtlc(chanCap btcutil.Amount) uint64 {
|
|
reserve := lnwire.NewMSatFromSatoshis(chanCap / 100)
|
|
max := lnwire.NewMSatFromSatoshis(chanCap) - reserve
|
|
return uint64(max)
|
|
}
|
|
|
|
// waitForChannelUpdate waits for a node to receive the expected channel
|
|
// updates.
|
|
func waitForChannelUpdate(t *harnessTest, subscription graphSubscription,
|
|
expUpdates []expectedChanUpdate) {
|
|
|
|
// Create an array indicating which expected channel updates we have
|
|
// received.
|
|
found := make([]bool, len(expUpdates))
|
|
out:
|
|
for {
|
|
select {
|
|
case graphUpdate := <-subscription.updateChan:
|
|
for _, update := range graphUpdate.ChannelUpdates {
|
|
// For each expected update, check if it matches
|
|
// the update we just received.
|
|
for i, exp := range expUpdates {
|
|
fundingTxStr := txStr(update.ChanPoint)
|
|
if fundingTxStr != txStr(exp.chanPoint) {
|
|
continue
|
|
}
|
|
|
|
if update.AdvertisingNode !=
|
|
exp.advertisingNode {
|
|
continue
|
|
}
|
|
|
|
err := checkChannelPolicy(
|
|
update.RoutingPolicy,
|
|
exp.expectedPolicy,
|
|
)
|
|
if err != nil {
|
|
continue
|
|
}
|
|
|
|
// We got a policy update that matched
|
|
// the values and channel point of what
|
|
// we expected, mark it as found.
|
|
found[i] = true
|
|
|
|
// If we have no more channel updates
|
|
// we are waiting for, break out of the
|
|
// loop.
|
|
rem := 0
|
|
for _, f := range found {
|
|
if !f {
|
|
rem++
|
|
}
|
|
}
|
|
|
|
if rem == 0 {
|
|
break out
|
|
}
|
|
|
|
// Since we found a match among the
|
|
// expected updates, break out of the
|
|
// inner loop.
|
|
break
|
|
}
|
|
}
|
|
case err := <-subscription.errChan:
|
|
t.Fatalf("unable to recv graph update: %v", err)
|
|
case <-time.After(20 * time.Second):
|
|
t.Fatalf("did not receive channel update")
|
|
}
|
|
}
|
|
}
|
|
|
|
// assertNoChannelUpdates ensures that no ChannelUpdates are sent via the
|
|
// graphSubscription. This method will block for the provided duration before
|
|
// returning to the caller if successful.
|
|
func assertNoChannelUpdates(t *harnessTest, subscription graphSubscription,
|
|
duration time.Duration) {
|
|
|
|
timeout := time.After(duration)
|
|
for {
|
|
select {
|
|
case graphUpdate := <-subscription.updateChan:
|
|
if len(graphUpdate.ChannelUpdates) > 0 {
|
|
t.Fatalf("received %d channel updates when "+
|
|
"none were expected",
|
|
len(graphUpdate.ChannelUpdates))
|
|
}
|
|
|
|
case err := <-subscription.errChan:
|
|
t.Fatalf("graph subscription failure: %v", err)
|
|
|
|
case <-timeout:
|
|
// No updates received, success.
|
|
return
|
|
}
|
|
}
|
|
}
|
|
|
|
// getChannelPolicies queries the channel graph and retrieves the current edge
|
|
// policies for the provided channel points.
|
|
func getChannelPolicies(t *harnessTest, node *lntest.HarnessNode,
|
|
advertisingNode string,
|
|
chanPoints ...*lnrpc.ChannelPoint) []*lnrpc.RoutingPolicy {
|
|
|
|
ctxb := context.Background()
|
|
|
|
descReq := &lnrpc.ChannelGraphRequest{
|
|
IncludeUnannounced: true,
|
|
}
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
chanGraph, err := node.DescribeGraph(ctxt, descReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to query for alice's graph: %v", err)
|
|
}
|
|
|
|
var policies []*lnrpc.RoutingPolicy
|
|
out:
|
|
for _, chanPoint := range chanPoints {
|
|
for _, e := range chanGraph.Edges {
|
|
if e.ChanPoint != txStr(chanPoint) {
|
|
continue
|
|
}
|
|
|
|
if e.Node1Pub == advertisingNode {
|
|
policies = append(policies, e.Node1Policy)
|
|
} else {
|
|
policies = append(policies, e.Node2Policy)
|
|
}
|
|
|
|
continue out
|
|
}
|
|
|
|
// If we've iterated over all the known edges and we weren't
|
|
// able to find this specific one, then we'll fail.
|
|
t.Fatalf("did not find edge %v", txStr(chanPoint))
|
|
}
|
|
|
|
return policies
|
|
}
|
|
|
|
// assertChannelPolicy asserts that the passed node's known channel policy for
|
|
// the passed chanPoint is consistent with the expected policy values.
|
|
func assertChannelPolicy(t *harnessTest, node *lntest.HarnessNode,
|
|
advertisingNode string, expectedPolicy *lnrpc.RoutingPolicy,
|
|
chanPoints ...*lnrpc.ChannelPoint) {
|
|
|
|
policies := getChannelPolicies(t, node, advertisingNode, chanPoints...)
|
|
for _, policy := range policies {
|
|
err := checkChannelPolicy(policy, expectedPolicy)
|
|
if err != nil {
|
|
t.Fatalf(err.Error())
|
|
}
|
|
}
|
|
}
|
|
|
|
// checkChannelPolicy checks that the policy matches the expected one.
|
|
func checkChannelPolicy(policy, expectedPolicy *lnrpc.RoutingPolicy) error {
|
|
if policy.FeeBaseMsat != expectedPolicy.FeeBaseMsat {
|
|
return fmt.Errorf("expected base fee %v, got %v",
|
|
expectedPolicy.FeeBaseMsat, policy.FeeBaseMsat)
|
|
}
|
|
if policy.FeeRateMilliMsat != expectedPolicy.FeeRateMilliMsat {
|
|
return fmt.Errorf("expected fee rate %v, got %v",
|
|
expectedPolicy.FeeRateMilliMsat,
|
|
policy.FeeRateMilliMsat)
|
|
}
|
|
if policy.TimeLockDelta != expectedPolicy.TimeLockDelta {
|
|
return fmt.Errorf("expected time lock delta %v, got %v",
|
|
expectedPolicy.TimeLockDelta,
|
|
policy.TimeLockDelta)
|
|
}
|
|
if policy.MinHtlc != expectedPolicy.MinHtlc {
|
|
return fmt.Errorf("expected min htlc %v, got %v",
|
|
expectedPolicy.MinHtlc, policy.MinHtlc)
|
|
}
|
|
if policy.MaxHtlcMsat != expectedPolicy.MaxHtlcMsat {
|
|
return fmt.Errorf("expected max htlc %v, got %v",
|
|
expectedPolicy.MaxHtlcMsat, policy.MaxHtlcMsat)
|
|
}
|
|
if policy.Disabled != expectedPolicy.Disabled {
|
|
return errors.New("edge should be disabled but isn't")
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// testUpdateChannelPolicy tests that policy updates made to a channel
|
|
// gets propagated to other nodes in the network.
|
|
func testUpdateChannelPolicy(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const (
|
|
defaultFeeBase = 1000
|
|
defaultFeeRate = 1
|
|
defaultTimeLockDelta = lnd.DefaultBitcoinTimeLockDelta
|
|
defaultMinHtlc = 1000
|
|
)
|
|
defaultMaxHtlc := calculateMaxHtlc(lnd.MaxBtcFundingAmount)
|
|
|
|
// Launch notification clients for all nodes, such that we can
|
|
// get notified when they discover new channels and updates in the
|
|
// graph.
|
|
aliceSub := subscribeGraphNotifications(t, ctxb, net.Alice)
|
|
defer close(aliceSub.quit)
|
|
bobSub := subscribeGraphNotifications(t, ctxb, net.Bob)
|
|
defer close(bobSub.quit)
|
|
|
|
chanAmt := lnd.MaxBtcFundingAmount
|
|
pushAmt := chanAmt / 2
|
|
|
|
// Create a channel Alice->Bob.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
|
|
// We add all the nodes' update channels to a slice, such that we can
|
|
// make sure they all receive the expected updates.
|
|
graphSubs := []graphSubscription{aliceSub, bobSub}
|
|
nodes := []*lntest.HarnessNode{net.Alice, net.Bob}
|
|
|
|
// Alice and Bob should see each other's ChannelUpdates, advertising the
|
|
// default routing policies.
|
|
expectedPolicy := &lnrpc.RoutingPolicy{
|
|
FeeBaseMsat: defaultFeeBase,
|
|
FeeRateMilliMsat: defaultFeeRate,
|
|
TimeLockDelta: defaultTimeLockDelta,
|
|
MinHtlc: defaultMinHtlc,
|
|
MaxHtlcMsat: defaultMaxHtlc,
|
|
}
|
|
|
|
for _, graphSub := range graphSubs {
|
|
waitForChannelUpdate(
|
|
t, graphSub,
|
|
[]expectedChanUpdate{
|
|
{net.Alice.PubKeyStr, expectedPolicy, chanPoint},
|
|
{net.Bob.PubKeyStr, expectedPolicy, chanPoint},
|
|
},
|
|
)
|
|
}
|
|
|
|
// They should now know about the default policies.
|
|
for _, node := range nodes {
|
|
assertChannelPolicy(
|
|
t, node, net.Alice.PubKeyStr, expectedPolicy, chanPoint,
|
|
)
|
|
assertChannelPolicy(
|
|
t, node, net.Bob.PubKeyStr, expectedPolicy, chanPoint,
|
|
)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err := net.Alice.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't report channel: %v", err)
|
|
}
|
|
err = net.Bob.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("bob didn't report channel: %v", err)
|
|
}
|
|
|
|
// Create Carol and a new channel Bob->Carol.
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
|
|
// Clean up carol's node when the test finishes.
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
carolSub := subscribeGraphNotifications(t, ctxb, carol)
|
|
defer close(carolSub.quit)
|
|
|
|
graphSubs = append(graphSubs, carolSub)
|
|
nodes = append(nodes, carol)
|
|
|
|
// Send some coins to Carol that can be used for channel funding.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
|
|
if err := net.ConnectNodes(ctxb, carol, net.Bob); err != nil {
|
|
t.Fatalf("unable to connect dave to alice: %v", err)
|
|
}
|
|
|
|
// Open the channel Carol->Bob with a custom min_htlc value set. Since
|
|
// Carol is opening the channel, she will require Bob to not forward
|
|
// HTLCs smaller than this value, and hence he should advertise it as
|
|
// part of his ChannelUpdate.
|
|
const customMinHtlc = 5000
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint2 := openChannelAndAssert(
|
|
ctxt, t, net, carol, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
MinHtlc: customMinHtlc,
|
|
},
|
|
)
|
|
|
|
expectedPolicyBob := &lnrpc.RoutingPolicy{
|
|
FeeBaseMsat: defaultFeeBase,
|
|
FeeRateMilliMsat: defaultFeeRate,
|
|
TimeLockDelta: defaultTimeLockDelta,
|
|
MinHtlc: customMinHtlc,
|
|
MaxHtlcMsat: defaultMaxHtlc,
|
|
}
|
|
|
|
expectedPolicyCarol := &lnrpc.RoutingPolicy{
|
|
FeeBaseMsat: defaultFeeBase,
|
|
FeeRateMilliMsat: defaultFeeRate,
|
|
TimeLockDelta: defaultTimeLockDelta,
|
|
MinHtlc: defaultMinHtlc,
|
|
MaxHtlcMsat: defaultMaxHtlc,
|
|
}
|
|
|
|
for _, graphSub := range graphSubs {
|
|
waitForChannelUpdate(
|
|
t, graphSub,
|
|
[]expectedChanUpdate{
|
|
{net.Bob.PubKeyStr, expectedPolicyBob, chanPoint2},
|
|
{carol.PubKeyStr, expectedPolicyCarol, chanPoint2},
|
|
},
|
|
)
|
|
}
|
|
|
|
// Check that all nodes now know about the updated policies.
|
|
for _, node := range nodes {
|
|
assertChannelPolicy(
|
|
t, node, net.Bob.PubKeyStr, expectedPolicyBob,
|
|
chanPoint2,
|
|
)
|
|
assertChannelPolicy(
|
|
t, node, carol.PubKeyStr, expectedPolicyCarol,
|
|
chanPoint2,
|
|
)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Alice.WaitForNetworkChannelOpen(ctxt, chanPoint2)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't report channel: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Bob.WaitForNetworkChannelOpen(ctxt, chanPoint2)
|
|
if err != nil {
|
|
t.Fatalf("bob didn't report channel: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPoint2)
|
|
if err != nil {
|
|
t.Fatalf("carol didn't report channel: %v", err)
|
|
}
|
|
|
|
// First we'll try to send a payment from Alice to Carol with an amount
|
|
// less than the min_htlc value required by Carol. This payment should
|
|
// fail, as the channel Bob->Carol cannot carry HTLCs this small.
|
|
payAmt := btcutil.Amount(4)
|
|
invoice := &lnrpc.Invoice{
|
|
Memo: "testing",
|
|
Value: int64(payAmt),
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := carol.AddInvoice(ctxt, invoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(
|
|
ctxt, net.Alice, []string{resp.PaymentRequest}, true,
|
|
)
|
|
|
|
// Alice knows about the channel policy of Carol and should therefore
|
|
// not be able to find a path during routing.
|
|
expErr := channeldb.FailureReasonNoRoute.Error()
|
|
if err == nil ||
|
|
!strings.Contains(err.Error(), expErr) {
|
|
t.Fatalf("expected payment to fail, instead got %v", err)
|
|
}
|
|
|
|
// Now we try to send a payment over the channel with a value too low
|
|
// to be accepted. First we query for a route to route a payment of
|
|
// 5000 mSAT, as this is accepted.
|
|
payAmt = btcutil.Amount(5)
|
|
routesReq := &lnrpc.QueryRoutesRequest{
|
|
PubKey: carol.PubKeyStr,
|
|
Amt: int64(payAmt),
|
|
FinalCltvDelta: defaultTimeLockDelta,
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
routes, err := net.Alice.QueryRoutes(ctxt, routesReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get route: %v", err)
|
|
}
|
|
|
|
if len(routes.Routes) != 1 {
|
|
t.Fatalf("expected to find 1 route, got %v", len(routes.Routes))
|
|
}
|
|
|
|
// We change the route to carry a payment of 4000 mSAT instead of 5000
|
|
// mSAT.
|
|
payAmt = btcutil.Amount(4)
|
|
amtSat := int64(payAmt)
|
|
amtMSat := int64(lnwire.NewMSatFromSatoshis(payAmt))
|
|
routes.Routes[0].Hops[0].AmtToForward = amtSat
|
|
routes.Routes[0].Hops[0].AmtToForwardMsat = amtMSat
|
|
routes.Routes[0].Hops[1].AmtToForward = amtSat
|
|
routes.Routes[0].Hops[1].AmtToForwardMsat = amtMSat
|
|
|
|
// Send the payment with the modified value.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
alicePayStream, err := net.Alice.SendToRoute(ctxt)
|
|
if err != nil {
|
|
t.Fatalf("unable to create payment stream for alice: %v", err)
|
|
}
|
|
sendReq := &lnrpc.SendToRouteRequest{
|
|
PaymentHash: resp.RHash,
|
|
Route: routes.Routes[0],
|
|
}
|
|
|
|
err = alicePayStream.Send(sendReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
|
|
// We expect this payment to fail, and that the min_htlc value is
|
|
// communicated back to us, since the attempted HTLC value was too low.
|
|
sendResp, err := alicePayStream.Recv()
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
|
|
// Expected as part of the error message.
|
|
substrs := []string{
|
|
"AmountBelowMinimum",
|
|
"HtlcMinimumMsat: (lnwire.MilliSatoshi) 5000 mSAT",
|
|
}
|
|
for _, s := range substrs {
|
|
if !strings.Contains(sendResp.PaymentError, s) {
|
|
t.Fatalf("expected error to contain \"%v\", instead "+
|
|
"got %v", s, sendResp.PaymentError)
|
|
}
|
|
}
|
|
|
|
// Make sure sending using the original value succeeds.
|
|
payAmt = btcutil.Amount(5)
|
|
amtSat = int64(payAmt)
|
|
amtMSat = int64(lnwire.NewMSatFromSatoshis(payAmt))
|
|
routes.Routes[0].Hops[0].AmtToForward = amtSat
|
|
routes.Routes[0].Hops[0].AmtToForwardMsat = amtMSat
|
|
routes.Routes[0].Hops[1].AmtToForward = amtSat
|
|
routes.Routes[0].Hops[1].AmtToForwardMsat = amtMSat
|
|
|
|
sendReq = &lnrpc.SendToRouteRequest{
|
|
PaymentHash: resp.RHash,
|
|
Route: routes.Routes[0],
|
|
}
|
|
|
|
err = alicePayStream.Send(sendReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
|
|
sendResp, err = alicePayStream.Recv()
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
|
|
if sendResp.PaymentError != "" {
|
|
t.Fatalf("expected payment to succeed, instead got %v",
|
|
sendResp.PaymentError)
|
|
}
|
|
|
|
// With our little cluster set up, we'll update the fees and the max htlc
|
|
// size for the Bob side of the Alice->Bob channel, and make sure
|
|
// all nodes learn about it.
|
|
baseFee := int64(1500)
|
|
feeRate := int64(12)
|
|
timeLockDelta := uint32(66)
|
|
maxHtlc := uint64(500000)
|
|
|
|
expectedPolicy = &lnrpc.RoutingPolicy{
|
|
FeeBaseMsat: baseFee,
|
|
FeeRateMilliMsat: testFeeBase * feeRate,
|
|
TimeLockDelta: timeLockDelta,
|
|
MinHtlc: defaultMinHtlc,
|
|
MaxHtlcMsat: maxHtlc,
|
|
}
|
|
|
|
req := &lnrpc.PolicyUpdateRequest{
|
|
BaseFeeMsat: baseFee,
|
|
FeeRate: float64(feeRate),
|
|
TimeLockDelta: timeLockDelta,
|
|
MaxHtlcMsat: maxHtlc,
|
|
Scope: &lnrpc.PolicyUpdateRequest_ChanPoint{
|
|
ChanPoint: chanPoint,
|
|
},
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if _, err := net.Bob.UpdateChannelPolicy(ctxt, req); err != nil {
|
|
t.Fatalf("unable to get alice's balance: %v", err)
|
|
}
|
|
|
|
// Wait for all nodes to have seen the policy update done by Bob.
|
|
for _, graphSub := range graphSubs {
|
|
waitForChannelUpdate(
|
|
t, graphSub,
|
|
[]expectedChanUpdate{
|
|
{net.Bob.PubKeyStr, expectedPolicy, chanPoint},
|
|
},
|
|
)
|
|
}
|
|
|
|
// Check that all nodes now know about Bob's updated policy.
|
|
for _, node := range nodes {
|
|
assertChannelPolicy(
|
|
t, node, net.Bob.PubKeyStr, expectedPolicy, chanPoint,
|
|
)
|
|
}
|
|
|
|
// Now that all nodes have received the new channel update, we'll try
|
|
// to send a payment from Alice to Carol to ensure that Alice has
|
|
// internalized this fee update. This shouldn't affect the route that
|
|
// Alice takes though: we updated the Alice -> Bob channel and she
|
|
// doesn't pay for transit over that channel as it's direct.
|
|
// Note that the payment amount is >= the min_htlc value for the
|
|
// channel Bob->Carol, so it should successfully be forwarded.
|
|
payAmt = btcutil.Amount(5)
|
|
invoice = &lnrpc.Invoice{
|
|
Memo: "testing",
|
|
Value: int64(payAmt),
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err = carol.AddInvoice(ctxt, invoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(
|
|
ctxt, net.Alice, []string{resp.PaymentRequest}, true,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
|
|
// We'll now open a channel from Alice directly to Carol.
|
|
if err := net.ConnectNodes(ctxb, net.Alice, carol); err != nil {
|
|
t.Fatalf("unable to connect dave to alice: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint3 := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Alice.WaitForNetworkChannelOpen(ctxt, chanPoint3)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't report channel: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPoint3)
|
|
if err != nil {
|
|
t.Fatalf("bob didn't report channel: %v", err)
|
|
}
|
|
|
|
// Make a global update, and check that both channels' new policies get
|
|
// propagated.
|
|
baseFee = int64(800)
|
|
feeRate = int64(123)
|
|
timeLockDelta = uint32(22)
|
|
maxHtlc = maxHtlc * 2
|
|
|
|
expectedPolicy.FeeBaseMsat = baseFee
|
|
expectedPolicy.FeeRateMilliMsat = testFeeBase * feeRate
|
|
expectedPolicy.TimeLockDelta = timeLockDelta
|
|
expectedPolicy.MaxHtlcMsat = maxHtlc
|
|
|
|
req = &lnrpc.PolicyUpdateRequest{
|
|
BaseFeeMsat: baseFee,
|
|
FeeRate: float64(feeRate),
|
|
TimeLockDelta: timeLockDelta,
|
|
MaxHtlcMsat: maxHtlc,
|
|
}
|
|
req.Scope = &lnrpc.PolicyUpdateRequest_Global{}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
_, err = net.Alice.UpdateChannelPolicy(ctxt, req)
|
|
if err != nil {
|
|
t.Fatalf("unable to update alice's channel policy: %v", err)
|
|
}
|
|
|
|
// Wait for all nodes to have seen the policy updates for both of
|
|
// Alice's channels.
|
|
for _, graphSub := range graphSubs {
|
|
waitForChannelUpdate(
|
|
t, graphSub,
|
|
[]expectedChanUpdate{
|
|
{net.Alice.PubKeyStr, expectedPolicy, chanPoint},
|
|
{net.Alice.PubKeyStr, expectedPolicy, chanPoint3},
|
|
},
|
|
)
|
|
}
|
|
|
|
// And finally check that all nodes remembers the policy update they
|
|
// received.
|
|
for _, node := range nodes {
|
|
assertChannelPolicy(
|
|
t, node, net.Alice.PubKeyStr, expectedPolicy,
|
|
chanPoint, chanPoint3,
|
|
)
|
|
}
|
|
|
|
// Close the channels.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPoint, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Bob, chanPoint2, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPoint3, false)
|
|
}
|
|
|
|
// waitForNodeBlockHeight queries the node for its current block height until
|
|
// it reaches the passed height.
|
|
func waitForNodeBlockHeight(ctx context.Context, node *lntest.HarnessNode,
|
|
height int32) error {
|
|
var predErr error
|
|
err := wait.Predicate(func() bool {
|
|
ctxt, _ := context.WithTimeout(ctx, 10*time.Second)
|
|
info, err := node.GetInfo(ctxt, &lnrpc.GetInfoRequest{})
|
|
if err != nil {
|
|
predErr = err
|
|
return false
|
|
}
|
|
|
|
if int32(info.BlockHeight) != height {
|
|
predErr = fmt.Errorf("expected block height to "+
|
|
"be %v, was %v", height, info.BlockHeight)
|
|
return false
|
|
}
|
|
return true
|
|
}, 15*time.Second)
|
|
if err != nil {
|
|
return predErr
|
|
}
|
|
return nil
|
|
}
|
|
|
|
// assertMinerBlockHeightDelta ensures that tempMiner is 'delta' blocks ahead
|
|
// of miner.
|
|
func assertMinerBlockHeightDelta(t *harnessTest,
|
|
miner, tempMiner *rpctest.Harness, delta int32) {
|
|
|
|
// Ensure the chain lengths are what we expect.
|
|
var predErr error
|
|
err := wait.Predicate(func() bool {
|
|
_, tempMinerHeight, err := tempMiner.Node.GetBestBlock()
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to get current "+
|
|
"blockheight %v", err)
|
|
return false
|
|
}
|
|
|
|
_, minerHeight, err := miner.Node.GetBestBlock()
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to get current "+
|
|
"blockheight %v", err)
|
|
return false
|
|
}
|
|
|
|
if tempMinerHeight != minerHeight+delta {
|
|
predErr = fmt.Errorf("expected new miner(%d) to be %d "+
|
|
"blocks ahead of original miner(%d)",
|
|
tempMinerHeight, delta, minerHeight)
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf(predErr.Error())
|
|
}
|
|
}
|
|
|
|
// testOpenChannelAfterReorg tests that in the case where we have an open
|
|
// channel where the funding tx gets reorged out, the channel will no
|
|
// longer be present in the node's routing table.
|
|
func testOpenChannelAfterReorg(net *lntest.NetworkHarness, t *harnessTest) {
|
|
// Skip test for neutrino, as we cannot disconnect the miner at will.
|
|
// TODO(halseth): remove when either can disconnect at will, or restart
|
|
// node with connection to new miner.
|
|
if net.BackendCfg.Name() == "neutrino" {
|
|
t.Skipf("skipping reorg test for neutrino backend")
|
|
}
|
|
|
|
var (
|
|
ctxb = context.Background()
|
|
temp = "temp"
|
|
)
|
|
|
|
// Set up a new miner that we can use to cause a reorg.
|
|
args := []string{"--rejectnonstd", "--txindex"}
|
|
tempMiner, err := rpctest.New(harnessNetParams,
|
|
&rpcclient.NotificationHandlers{}, args)
|
|
if err != nil {
|
|
t.Fatalf("unable to create mining node: %v", err)
|
|
}
|
|
if err := tempMiner.SetUp(false, 0); err != nil {
|
|
t.Fatalf("unable to set up mining node: %v", err)
|
|
}
|
|
defer tempMiner.TearDown()
|
|
|
|
// We start by connecting the new miner to our original miner,
|
|
// such that it will sync to our original chain.
|
|
err = net.Miner.Node.Node(
|
|
btcjson.NConnect, tempMiner.P2PAddress(), &temp,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to remove node: %v", err)
|
|
}
|
|
nodeSlice := []*rpctest.Harness{net.Miner, tempMiner}
|
|
if err := rpctest.JoinNodes(nodeSlice, rpctest.Blocks); err != nil {
|
|
t.Fatalf("unable to join node on blocks: %v", err)
|
|
}
|
|
|
|
// The two miners should be on the same blockheight.
|
|
assertMinerBlockHeightDelta(t, net.Miner, tempMiner, 0)
|
|
|
|
// We disconnect the two miners, such that we can mine two different
|
|
// chains and can cause a reorg later.
|
|
err = net.Miner.Node.Node(
|
|
btcjson.NDisconnect, tempMiner.P2PAddress(), &temp,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to remove node: %v", err)
|
|
}
|
|
|
|
// Create a new channel that requires 1 confs before it's considered
|
|
// open, then broadcast the funding transaction
|
|
chanAmt := lnd.MaxBtcFundingAmount
|
|
pushAmt := btcutil.Amount(0)
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
pendingUpdate, err := net.OpenPendingChannel(ctxt, net.Alice, net.Bob,
|
|
chanAmt, pushAmt)
|
|
if err != nil {
|
|
t.Fatalf("unable to open channel: %v", err)
|
|
}
|
|
|
|
// Wait for miner to have seen the funding tx. The temporary miner is
|
|
// disconnected, and won't see the transaction.
|
|
_, err = waitForTxInMempool(net.Miner.Node, minerMempoolTimeout)
|
|
if err != nil {
|
|
t.Fatalf("failed to find funding tx in mempool: %v", err)
|
|
}
|
|
|
|
// At this point, the channel's funding transaction will have been
|
|
// broadcast, but not confirmed, and the channel should be pending.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
assertNumOpenChannelsPending(ctxt, t, net.Alice, net.Bob, 1)
|
|
|
|
fundingTxID, err := chainhash.NewHash(pendingUpdate.Txid)
|
|
if err != nil {
|
|
t.Fatalf("unable to convert funding txid into chainhash.Hash:"+
|
|
" %v", err)
|
|
}
|
|
|
|
// We now cause a fork, by letting our original miner mine 10 blocks,
|
|
// and our new miner mine 15. This will also confirm our pending
|
|
// channel on the original miner's chain, which should be considered
|
|
// open.
|
|
block := mineBlocks(t, net, 10, 1)[0]
|
|
assertTxInBlock(t, block, fundingTxID)
|
|
if _, err := tempMiner.Node.Generate(15); err != nil {
|
|
t.Fatalf("unable to generate blocks: %v", err)
|
|
}
|
|
|
|
// Ensure the chain lengths are what we expect, with the temp miner
|
|
// being 5 blocks ahead.
|
|
assertMinerBlockHeightDelta(t, net.Miner, tempMiner, 5)
|
|
|
|
// Wait for Alice to sync to the original miner's chain.
|
|
_, minerHeight, err := net.Miner.Node.GetBestBlock()
|
|
if err != nil {
|
|
t.Fatalf("unable to get current blockheight %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = waitForNodeBlockHeight(ctxt, net.Alice, minerHeight)
|
|
if err != nil {
|
|
t.Fatalf("unable to sync to chain: %v", err)
|
|
}
|
|
|
|
chanPoint := &lnrpc.ChannelPoint{
|
|
FundingTxid: &lnrpc.ChannelPoint_FundingTxidBytes{
|
|
FundingTxidBytes: pendingUpdate.Txid,
|
|
},
|
|
OutputIndex: pendingUpdate.OutputIndex,
|
|
}
|
|
|
|
// Ensure channel is no longer pending.
|
|
assertNumOpenChannelsPending(ctxt, t, net.Alice, net.Bob, 0)
|
|
|
|
// Wait for Alice and Bob to recognize and advertise the new channel
|
|
// generated above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Alice.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't advertise channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Bob.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("bob didn't advertise channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// Alice should now have 1 edge in her graph.
|
|
req := &lnrpc.ChannelGraphRequest{
|
|
IncludeUnannounced: true,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
chanGraph, err := net.Alice.DescribeGraph(ctxt, req)
|
|
if err != nil {
|
|
t.Fatalf("unable to query for alice's routing table: %v", err)
|
|
}
|
|
|
|
numEdges := len(chanGraph.Edges)
|
|
if numEdges != 1 {
|
|
t.Fatalf("expected to find one edge in the graph, found %d",
|
|
numEdges)
|
|
}
|
|
|
|
// Now we disconnect Alice's chain backend from the original miner, and
|
|
// connect the two miners together. Since the temporary miner knows
|
|
// about a longer chain, both miners should sync to that chain.
|
|
err = net.BackendCfg.DisconnectMiner()
|
|
if err != nil {
|
|
t.Fatalf("unable to remove node: %v", err)
|
|
}
|
|
|
|
// Connecting to the temporary miner should now cause our original
|
|
// chain to be re-orged out.
|
|
err = net.Miner.Node.Node(
|
|
btcjson.NConnect, tempMiner.P2PAddress(), &temp,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to remove node: %v", err)
|
|
}
|
|
|
|
nodes := []*rpctest.Harness{tempMiner, net.Miner}
|
|
if err := rpctest.JoinNodes(nodes, rpctest.Blocks); err != nil {
|
|
t.Fatalf("unable to join node on blocks: %v", err)
|
|
}
|
|
|
|
// Once again they should be on the same chain.
|
|
assertMinerBlockHeightDelta(t, net.Miner, tempMiner, 0)
|
|
|
|
// Now we disconnect the two miners, and connect our original miner to
|
|
// our chain backend once again.
|
|
err = net.Miner.Node.Node(
|
|
btcjson.NDisconnect, tempMiner.P2PAddress(), &temp,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to remove node: %v", err)
|
|
}
|
|
|
|
err = net.BackendCfg.ConnectMiner()
|
|
if err != nil {
|
|
t.Fatalf("unable to remove node: %v", err)
|
|
}
|
|
|
|
// This should have caused a reorg, and Alice should sync to the longer
|
|
// chain, where the funding transaction is not confirmed.
|
|
_, tempMinerHeight, err := tempMiner.Node.GetBestBlock()
|
|
if err != nil {
|
|
t.Fatalf("unable to get current blockheight %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = waitForNodeBlockHeight(ctxt, net.Alice, tempMinerHeight)
|
|
if err != nil {
|
|
t.Fatalf("unable to sync to chain: %v", err)
|
|
}
|
|
|
|
// Since the fundingtx was reorged out, Alice should now have no edges
|
|
// in her graph.
|
|
req = &lnrpc.ChannelGraphRequest{
|
|
IncludeUnannounced: true,
|
|
}
|
|
|
|
var predErr error
|
|
err = wait.Predicate(func() bool {
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
chanGraph, err = net.Alice.DescribeGraph(ctxt, req)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to query for alice's routing table: %v", err)
|
|
return false
|
|
}
|
|
|
|
numEdges = len(chanGraph.Edges)
|
|
if numEdges != 0 {
|
|
predErr = fmt.Errorf("expected to find no edge in the graph, found %d",
|
|
numEdges)
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf(predErr.Error())
|
|
}
|
|
|
|
// Cleanup by mining the funding tx again, then closing the channel.
|
|
block = mineBlocks(t, net, 1, 1)[0]
|
|
assertTxInBlock(t, block, fundingTxID)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeReorgedChannelAndAssert(ctxt, t, net, net.Alice, chanPoint, false)
|
|
}
|
|
|
|
// testDisconnectingTargetPeer performs a test which disconnects Alice-peer from
|
|
// Bob-peer and then re-connects them again. We expect Alice to be able to
|
|
// disconnect at any point.
|
|
func testDisconnectingTargetPeer(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
// We'll start both nodes with a high backoff so that they don't
|
|
// reconnect automatically during our test.
|
|
args := []string{
|
|
"--minbackoff=1m",
|
|
"--maxbackoff=1m",
|
|
}
|
|
|
|
alice, err := net.NewNode("Alice", args)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, alice)
|
|
|
|
bob, err := net.NewNode("Bob", args)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, bob)
|
|
|
|
// Start by connecting Alice and Bob with no channels.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, alice, bob); err != nil {
|
|
t.Fatalf("unable to connect Alice's peer to Bob's: err %v", err)
|
|
}
|
|
|
|
// Check existing connection.
|
|
assertNumConnections(t, alice, bob, 1)
|
|
|
|
// Give Alice some coins so she can fund a channel.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, alice)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
|
|
chanAmt := lnd.MaxBtcFundingAmount
|
|
pushAmt := btcutil.Amount(0)
|
|
|
|
// Create a new channel that requires 1 confs before it's considered
|
|
// open, then broadcast the funding transaction
|
|
const numConfs = 1
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
pendingUpdate, err := net.OpenPendingChannel(
|
|
ctxt, alice, bob, chanAmt, pushAmt,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to open channel: %v", err)
|
|
}
|
|
|
|
// At this point, the channel's funding transaction will have been
|
|
// broadcast, but not confirmed. Alice and Bob's nodes should reflect
|
|
// this when queried via RPC.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
assertNumOpenChannelsPending(ctxt, t, alice, bob, 1)
|
|
|
|
// Disconnect Alice-peer from Bob-peer and get error causes by one
|
|
// pending channel with detach node is existing.
|
|
if err := net.DisconnectNodes(ctxt, alice, bob); err != nil {
|
|
t.Fatalf("Bob's peer was disconnected from Alice's"+
|
|
" while one pending channel is existing: err %v", err)
|
|
}
|
|
|
|
time.Sleep(time.Millisecond * 300)
|
|
|
|
// Assert that the connection was torn down.
|
|
assertNumConnections(t, alice, bob, 0)
|
|
|
|
fundingTxID, err := chainhash.NewHash(pendingUpdate.Txid)
|
|
if err != nil {
|
|
t.Fatalf("unable to convert funding txid into chainhash.Hash:"+
|
|
" %v", err)
|
|
}
|
|
|
|
// Mine a block, then wait for Alice's node to notify us that the
|
|
// channel has been opened. The funding transaction should be found
|
|
// within the newly mined block.
|
|
block := mineBlocks(t, net, numConfs, 1)[0]
|
|
assertTxInBlock(t, block, fundingTxID)
|
|
|
|
// At this point, the channel should be fully opened and there should be
|
|
// no pending channels remaining for either node.
|
|
time.Sleep(time.Millisecond * 300)
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
|
|
assertNumOpenChannelsPending(ctxt, t, alice, bob, 0)
|
|
|
|
// Reconnect the nodes so that the channel can become active.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, alice, bob); err != nil {
|
|
t.Fatalf("unable to connect Alice's peer to Bob's: err %v", err)
|
|
}
|
|
|
|
// The channel should be listed in the peer information returned by both
|
|
// peers.
|
|
outPoint := wire.OutPoint{
|
|
Hash: *fundingTxID,
|
|
Index: pendingUpdate.OutputIndex,
|
|
}
|
|
|
|
// Check both nodes to ensure that the channel is ready for operation.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.AssertChannelExists(ctxt, alice, &outPoint); err != nil {
|
|
t.Fatalf("unable to assert channel existence: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.AssertChannelExists(ctxt, bob, &outPoint); err != nil {
|
|
t.Fatalf("unable to assert channel existence: %v", err)
|
|
}
|
|
|
|
// Disconnect Alice-peer from Bob-peer and get error causes by one
|
|
// active channel with detach node is existing.
|
|
if err := net.DisconnectNodes(ctxt, alice, bob); err != nil {
|
|
t.Fatalf("Bob's peer was disconnected from Alice's"+
|
|
" while one active channel is existing: err %v", err)
|
|
}
|
|
|
|
// Check existing connection.
|
|
assertNumConnections(t, alice, bob, 0)
|
|
|
|
// Reconnect both nodes before force closing the channel.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, alice, bob); err != nil {
|
|
t.Fatalf("unable to connect Alice's peer to Bob's: err %v", err)
|
|
}
|
|
|
|
// Finally, immediately close the channel. This function will also block
|
|
// until the channel is closed and will additionally assert the relevant
|
|
// channel closing post conditions.
|
|
chanPoint := &lnrpc.ChannelPoint{
|
|
FundingTxid: &lnrpc.ChannelPoint_FundingTxidBytes{
|
|
FundingTxidBytes: pendingUpdate.Txid,
|
|
},
|
|
OutputIndex: pendingUpdate.OutputIndex,
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, alice, chanPoint, true)
|
|
|
|
// Disconnect Alice-peer from Bob-peer without getting error about
|
|
// existing channels.
|
|
if err := net.DisconnectNodes(ctxt, alice, bob); err != nil {
|
|
t.Fatalf("unable to disconnect Bob's peer from Alice's: err %v",
|
|
err)
|
|
}
|
|
|
|
// Check zero peer connections.
|
|
assertNumConnections(t, alice, bob, 0)
|
|
|
|
// Finally, re-connect both nodes.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, alice, bob); err != nil {
|
|
t.Fatalf("unable to connect Alice's peer to Bob's: err %v", err)
|
|
}
|
|
|
|
// Check existing connection.
|
|
assertNumConnections(t, alice, net.Bob, 1)
|
|
|
|
// Cleanup by mining the force close and sweep transaction.
|
|
cleanupForceClose(t, net, alice, chanPoint)
|
|
}
|
|
|
|
// testFundingPersistence is intended to ensure that the Funding Manager
|
|
// persists the state of new channels prior to broadcasting the channel's
|
|
// funding transaction. This ensures that the daemon maintains an up-to-date
|
|
// representation of channels if the system is restarted or disconnected.
|
|
// testFundingPersistence mirrors testBasicChannelFunding, but adds restarts
|
|
// and checks for the state of channels with unconfirmed funding transactions.
|
|
func testChannelFundingPersistence(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
chanAmt := lnd.MaxBtcFundingAmount
|
|
pushAmt := btcutil.Amount(0)
|
|
|
|
// As we need to create a channel that requires more than 1
|
|
// confirmation before it's open, with the current set of defaults,
|
|
// we'll need to create a new node instance.
|
|
const numConfs = 5
|
|
carolArgs := []string{fmt.Sprintf("--bitcoin.defaultchanconfs=%v", numConfs)}
|
|
carol, err := net.NewNode("Carol", carolArgs)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new node: %v", err)
|
|
}
|
|
|
|
// Clean up carol's node when the test finishes.
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, net.Alice, carol); err != nil {
|
|
t.Fatalf("unable to connect alice to carol: %v", err)
|
|
}
|
|
|
|
// Create a new channel that requires 5 confs before it's considered
|
|
// open, then broadcast the funding transaction
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
pendingUpdate, err := net.OpenPendingChannel(ctxt, net.Alice, carol,
|
|
chanAmt, pushAmt)
|
|
if err != nil {
|
|
t.Fatalf("unable to open channel: %v", err)
|
|
}
|
|
|
|
// At this point, the channel's funding transaction will have been
|
|
// broadcast, but not confirmed. Alice and Bob's nodes should reflect
|
|
// this when queried via RPC.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
assertNumOpenChannelsPending(ctxt, t, net.Alice, carol, 1)
|
|
|
|
// Restart both nodes to test that the appropriate state has been
|
|
// persisted and that both nodes recover gracefully.
|
|
if err := net.RestartNode(net.Alice, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
if err := net.RestartNode(carol, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
fundingTxID, err := chainhash.NewHash(pendingUpdate.Txid)
|
|
if err != nil {
|
|
t.Fatalf("unable to convert funding txid into chainhash.Hash:"+
|
|
" %v", err)
|
|
}
|
|
|
|
// Mine a block, then wait for Alice's node to notify us that the
|
|
// channel has been opened. The funding transaction should be found
|
|
// within the newly mined block.
|
|
block := mineBlocks(t, net, 1, 1)[0]
|
|
assertTxInBlock(t, block, fundingTxID)
|
|
|
|
// Restart both nodes to test that the appropriate state has been
|
|
// persisted and that both nodes recover gracefully.
|
|
if err := net.RestartNode(net.Alice, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
if err := net.RestartNode(carol, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// The following block ensures that after both nodes have restarted,
|
|
// they have reconnected before the execution of the next test.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.EnsureConnected(ctxt, net.Alice, carol); err != nil {
|
|
t.Fatalf("peers unable to reconnect after restart: %v", err)
|
|
}
|
|
|
|
// Next, mine enough blocks s.t the channel will open with a single
|
|
// additional block mined.
|
|
if _, err := net.Miner.Node.Generate(3); err != nil {
|
|
t.Fatalf("unable to mine blocks: %v", err)
|
|
}
|
|
|
|
// Both nodes should still show a single channel as pending.
|
|
time.Sleep(time.Second * 1)
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
assertNumOpenChannelsPending(ctxt, t, net.Alice, carol, 1)
|
|
|
|
// Finally, mine the last block which should mark the channel as open.
|
|
if _, err := net.Miner.Node.Generate(1); err != nil {
|
|
t.Fatalf("unable to mine blocks: %v", err)
|
|
}
|
|
|
|
// At this point, the channel should be fully opened and there should
|
|
// be no pending channels remaining for either node.
|
|
time.Sleep(time.Second * 1)
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
assertNumOpenChannelsPending(ctxt, t, net.Alice, carol, 0)
|
|
|
|
// The channel should be listed in the peer information returned by
|
|
// both peers.
|
|
outPoint := wire.OutPoint{
|
|
Hash: *fundingTxID,
|
|
Index: pendingUpdate.OutputIndex,
|
|
}
|
|
|
|
// Check both nodes to ensure that the channel is ready for operation.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.AssertChannelExists(ctxt, net.Alice, &outPoint); err != nil {
|
|
t.Fatalf("unable to assert channel existence: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.AssertChannelExists(ctxt, carol, &outPoint); err != nil {
|
|
t.Fatalf("unable to assert channel existence: %v", err)
|
|
}
|
|
|
|
// Finally, immediately close the channel. This function will also
|
|
// block until the channel is closed and will additionally assert the
|
|
// relevant channel closing post conditions.
|
|
chanPoint := &lnrpc.ChannelPoint{
|
|
FundingTxid: &lnrpc.ChannelPoint_FundingTxidBytes{
|
|
FundingTxidBytes: pendingUpdate.Txid,
|
|
},
|
|
OutputIndex: pendingUpdate.OutputIndex,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPoint, false)
|
|
}
|
|
|
|
// testChannelBalance creates a new channel between Alice and Bob, then
|
|
// checks channel balance to be equal amount specified while creation of channel.
|
|
func testChannelBalance(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
// Open a channel with 0.16 BTC between Alice and Bob, ensuring the
|
|
// channel has been opened properly.
|
|
amount := lnd.MaxBtcFundingAmount
|
|
|
|
// Creates a helper closure to be used below which asserts the proper
|
|
// response to a channel balance RPC.
|
|
checkChannelBalance := func(node lnrpc.LightningClient,
|
|
amount btcutil.Amount) {
|
|
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
response, err := node.ChannelBalance(ctxt, &lnrpc.ChannelBalanceRequest{})
|
|
if err != nil {
|
|
t.Fatalf("unable to get channel balance: %v", err)
|
|
}
|
|
|
|
balance := btcutil.Amount(response.Balance)
|
|
if balance != amount {
|
|
t.Fatalf("channel balance wrong: %v != %v", balance,
|
|
amount)
|
|
}
|
|
}
|
|
|
|
// Before beginning, make sure alice and bob are connected.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.EnsureConnected(ctxt, net.Alice, net.Bob); err != nil {
|
|
t.Fatalf("unable to connect alice and bob: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: amount,
|
|
},
|
|
)
|
|
|
|
// Wait for both Alice and Bob to recognize this new channel.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err := net.Alice.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't advertise channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Bob.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("bob didn't advertise channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
cType, err := channelCommitType(net.Alice, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get channel type: %v", err)
|
|
}
|
|
|
|
// As this is a single funder channel, Alice's balance should be
|
|
// exactly 0.5 BTC since now state transitions have taken place yet.
|
|
checkChannelBalance(net.Alice, amount-cType.calcStaticFee(0))
|
|
|
|
// Ensure Bob currently has no available balance within the channel.
|
|
checkChannelBalance(net.Bob, 0)
|
|
|
|
// Finally close the channel between Alice and Bob, asserting that the
|
|
// channel has been properly closed on-chain.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPoint, false)
|
|
}
|
|
|
|
// testChannelUnsettledBalance will test that the UnsettledBalance field
|
|
// is updated according to the number of Pending Htlcs.
|
|
// Alice will send Htlcs to Carol while she is in hodl mode. This will result
|
|
// in a build of pending Htlcs. We expect the channels unsettled balance to
|
|
// equal the sum of all the Pending Htlcs.
|
|
func testChannelUnsettledBalance(net *lntest.NetworkHarness, t *harnessTest) {
|
|
const chanAmt = btcutil.Amount(1000000)
|
|
ctxb := context.Background()
|
|
|
|
// Create carol in hodl mode.
|
|
carol, err := net.NewNode("Carol", []string{"--hodl.exit-settle"})
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
// Connect Alice to Carol.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxb, net.Alice, carol); err != nil {
|
|
t.Fatalf("unable to connect alice to carol: %v", err)
|
|
}
|
|
|
|
// Open a channel between Alice and Carol.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAlice := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Wait for Alice and Carol to receive the channel edge from the
|
|
// funding manager.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Alice.WaitForNetworkChannelOpen(ctxt, chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't see the alice->carol channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't see the alice->carol channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// Channel should be ready for payments.
|
|
const (
|
|
payAmt = 100
|
|
numInvoices = 6
|
|
)
|
|
|
|
// Create a paystream from Alice to Carol to enable Alice to make
|
|
// a series of payments.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
alicePayStream, err := net.Alice.SendPayment(ctxt)
|
|
if err != nil {
|
|
t.Fatalf("unable to create payment stream for alice: %v", err)
|
|
}
|
|
|
|
// Send payments from Alice to Carol a number of numInvoices
|
|
// times.
|
|
carolPubKey := carol.PubKey[:]
|
|
for i := 0; i < numInvoices; i++ {
|
|
err = alicePayStream.Send(&lnrpc.SendRequest{
|
|
Dest: carolPubKey,
|
|
Amt: int64(payAmt),
|
|
PaymentHash: makeFakePayHash(t),
|
|
FinalCltvDelta: lnd.DefaultBitcoinTimeLockDelta,
|
|
})
|
|
if err != nil {
|
|
t.Fatalf("unable to send alice htlc: %v", err)
|
|
}
|
|
}
|
|
|
|
// Test that the UnsettledBalance for both Alice and Carol
|
|
// is equal to the amount of invoices * payAmt.
|
|
var unsettledErr error
|
|
nodes := []*lntest.HarnessNode{net.Alice, carol}
|
|
err = wait.Predicate(func() bool {
|
|
// There should be a number of PendingHtlcs equal
|
|
// to the amount of Invoices sent.
|
|
unsettledErr = assertNumActiveHtlcs(nodes, numInvoices)
|
|
if unsettledErr != nil {
|
|
return false
|
|
}
|
|
|
|
// Set the amount expected for the Unsettled Balance for
|
|
// this channel.
|
|
expectedBalance := numInvoices * payAmt
|
|
|
|
// Check each nodes UnsettledBalance field.
|
|
for _, node := range nodes {
|
|
// Get channel info for the node.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
chanInfo, err := getChanInfo(ctxt, node)
|
|
if err != nil {
|
|
unsettledErr = err
|
|
return false
|
|
}
|
|
|
|
// Check that UnsettledBalance is what we expect.
|
|
if int(chanInfo.UnsettledBalance) != expectedBalance {
|
|
unsettledErr = fmt.Errorf("unsettled balance failed "+
|
|
"expected: %v, received: %v", expectedBalance,
|
|
chanInfo.UnsettledBalance)
|
|
return false
|
|
}
|
|
}
|
|
|
|
return true
|
|
}, defaultTimeout)
|
|
if err != nil {
|
|
t.Fatalf("unsettled balace error: %v", unsettledErr)
|
|
}
|
|
|
|
// Force and assert the channel closure.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointAlice, true)
|
|
|
|
// Cleanup by mining the force close and sweep transaction.
|
|
cleanupForceClose(t, net, net.Alice, chanPointAlice)
|
|
}
|
|
|
|
// findForceClosedChannel searches a pending channel response for a particular
|
|
// channel, returning the force closed channel upon success.
|
|
func findForceClosedChannel(pendingChanResp *lnrpc.PendingChannelsResponse,
|
|
op *wire.OutPoint) (*lnrpc.PendingChannelsResponse_ForceClosedChannel, error) {
|
|
|
|
for _, forceClose := range pendingChanResp.PendingForceClosingChannels {
|
|
if forceClose.Channel.ChannelPoint == op.String() {
|
|
return forceClose, nil
|
|
}
|
|
}
|
|
|
|
return nil, errors.New("channel not marked as force closed")
|
|
}
|
|
|
|
// findWaitingCloseChannel searches a pending channel response for a particular
|
|
// channel, returning the waiting close channel upon success.
|
|
func findWaitingCloseChannel(pendingChanResp *lnrpc.PendingChannelsResponse,
|
|
op *wire.OutPoint) (*lnrpc.PendingChannelsResponse_WaitingCloseChannel, error) {
|
|
|
|
for _, waitingClose := range pendingChanResp.WaitingCloseChannels {
|
|
if waitingClose.Channel.ChannelPoint == op.String() {
|
|
return waitingClose, nil
|
|
}
|
|
}
|
|
|
|
return nil, errors.New("channel not marked as waiting close")
|
|
}
|
|
|
|
func checkCommitmentMaturity(
|
|
forceClose *lnrpc.PendingChannelsResponse_ForceClosedChannel,
|
|
maturityHeight uint32, blocksTilMaturity int32) error {
|
|
|
|
if forceClose.MaturityHeight != maturityHeight {
|
|
return fmt.Errorf("expected commitment maturity height to be "+
|
|
"%d, found %d instead", maturityHeight,
|
|
forceClose.MaturityHeight)
|
|
}
|
|
if forceClose.BlocksTilMaturity != blocksTilMaturity {
|
|
return fmt.Errorf("expected commitment blocks til maturity to "+
|
|
"be %d, found %d instead", blocksTilMaturity,
|
|
forceClose.BlocksTilMaturity)
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// checkForceClosedChannelNumHtlcs verifies that a force closed channel has the
|
|
// proper number of htlcs.
|
|
func checkPendingChannelNumHtlcs(
|
|
forceClose *lnrpc.PendingChannelsResponse_ForceClosedChannel,
|
|
expectedNumHtlcs int) error {
|
|
|
|
if len(forceClose.PendingHtlcs) != expectedNumHtlcs {
|
|
return fmt.Errorf("expected force closed channel to have %d "+
|
|
"pending htlcs, found %d instead", expectedNumHtlcs,
|
|
len(forceClose.PendingHtlcs))
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// checkNumForceClosedChannels checks that a pending channel response has the
|
|
// expected number of force closed channels.
|
|
func checkNumForceClosedChannels(pendingChanResp *lnrpc.PendingChannelsResponse,
|
|
expectedNumChans int) error {
|
|
|
|
if len(pendingChanResp.PendingForceClosingChannels) != expectedNumChans {
|
|
return fmt.Errorf("expected to find %d force closed channels, "+
|
|
"got %d", expectedNumChans,
|
|
len(pendingChanResp.PendingForceClosingChannels))
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// checkNumWaitingCloseChannels checks that a pending channel response has the
|
|
// expected number of channels waiting for closing tx to confirm.
|
|
func checkNumWaitingCloseChannels(pendingChanResp *lnrpc.PendingChannelsResponse,
|
|
expectedNumChans int) error {
|
|
|
|
if len(pendingChanResp.WaitingCloseChannels) != expectedNumChans {
|
|
return fmt.Errorf("expected to find %d channels waiting "+
|
|
"closure, got %d", expectedNumChans,
|
|
len(pendingChanResp.WaitingCloseChannels))
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// checkPendingHtlcStageAndMaturity uniformly tests all pending htlc's belonging
|
|
// to a force closed channel, testing for the expected stage number, blocks till
|
|
// maturity, and the maturity height.
|
|
func checkPendingHtlcStageAndMaturity(
|
|
forceClose *lnrpc.PendingChannelsResponse_ForceClosedChannel,
|
|
stage, maturityHeight uint32, blocksTillMaturity int32) error {
|
|
|
|
for _, pendingHtlc := range forceClose.PendingHtlcs {
|
|
if pendingHtlc.Stage != stage {
|
|
return fmt.Errorf("expected pending htlc to be stage "+
|
|
"%d, found %d", stage, pendingHtlc.Stage)
|
|
}
|
|
if pendingHtlc.MaturityHeight != maturityHeight {
|
|
return fmt.Errorf("expected pending htlc maturity "+
|
|
"height to be %d, instead has %d",
|
|
maturityHeight, pendingHtlc.MaturityHeight)
|
|
}
|
|
if pendingHtlc.BlocksTilMaturity != blocksTillMaturity {
|
|
return fmt.Errorf("expected pending htlc blocks til "+
|
|
"maturity to be %d, instead has %d",
|
|
blocksTillMaturity,
|
|
pendingHtlc.BlocksTilMaturity)
|
|
}
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// padCLTV is a small helper function that pads a cltv value with a block
|
|
// padding.
|
|
func padCLTV(cltv uint32) uint32 {
|
|
return cltv + uint32(routing.BlockPadding)
|
|
}
|
|
|
|
// testChannelForceClosure performs a test to exercise the behavior of "force"
|
|
// closing a channel or unilaterally broadcasting the latest local commitment
|
|
// state on-chain. The test creates a new channel between Alice and Carol, then
|
|
// force closes the channel after some cursory assertions. Within the test, a
|
|
// total of 3 + n transactions will be broadcast, representing the commitment
|
|
// transaction, a transaction sweeping the local CSV delayed output, a
|
|
// transaction sweeping the CSV delayed 2nd-layer htlcs outputs, and n
|
|
// htlc timeout transactions, where n is the number of payments Alice attempted
|
|
// to send to Carol. This test includes several restarts to ensure that the
|
|
// transaction output states are persisted throughout the forced closure
|
|
// process.
|
|
//
|
|
// TODO(roasbeef): also add an unsettled HTLC before force closing.
|
|
func testChannelForceClosure(net *lntest.NetworkHarness, t *harnessTest) {
|
|
// We'll test the scenario for some of the commitment types, to ensure
|
|
// outputs can be swept.
|
|
commitTypes := []commitType{
|
|
commitTypeLegacy,
|
|
commitTypeAnchors,
|
|
}
|
|
|
|
for _, channelType := range commitTypes {
|
|
testName := fmt.Sprintf("committype=%v", channelType)
|
|
|
|
success := t.t.Run(testName, func(t *testing.T) {
|
|
ht := newHarnessTest(t, net)
|
|
|
|
args := channelType.Args()
|
|
alice, err := net.NewNode("Alice", args)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, ht, alice)
|
|
|
|
// Since we'd like to test failure scenarios with
|
|
// outstanding htlcs, we'll introduce another node into
|
|
// our test network: Carol.
|
|
carolArgs := []string{"--hodl.exit-settle"}
|
|
carolArgs = append(carolArgs, args...)
|
|
carol, err := net.NewNode("Carol", carolArgs)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, ht, carol)
|
|
|
|
// Each time, we'll send Alice new set of coins in
|
|
// order to fund the channel.
|
|
ctxt, _ := context.WithTimeout(
|
|
context.Background(), defaultTimeout,
|
|
)
|
|
err = net.SendCoins(
|
|
ctxt, btcutil.SatoshiPerBitcoin, alice,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to Alice: %v",
|
|
err)
|
|
}
|
|
|
|
// Also give Carol some coins to allow her to sweep her
|
|
// anchor.
|
|
err = net.SendCoins(
|
|
ctxt, btcutil.SatoshiPerBitcoin, carol,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to Alice: %v",
|
|
err)
|
|
}
|
|
|
|
channelForceClosureTest(
|
|
net, ht, alice, carol, channelType,
|
|
)
|
|
})
|
|
if !success {
|
|
return
|
|
}
|
|
}
|
|
}
|
|
|
|
func channelForceClosureTest(net *lntest.NetworkHarness, t *harnessTest,
|
|
alice, carol *lntest.HarnessNode, channelType commitType) {
|
|
|
|
ctxb := context.Background()
|
|
|
|
const (
|
|
chanAmt = btcutil.Amount(10e6)
|
|
pushAmt = btcutil.Amount(5e6)
|
|
paymentAmt = 100000
|
|
numInvoices = 6
|
|
)
|
|
|
|
// TODO(roasbeef): should check default value in config here
|
|
// instead, or make delay a param
|
|
defaultCLTV := uint32(lnd.DefaultBitcoinTimeLockDelta)
|
|
|
|
// We must let Alice have an open channel before she can send a node
|
|
// announcement, so we open a channel with Carol,
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, alice, carol); err != nil {
|
|
t.Fatalf("unable to connect alice to carol: %v", err)
|
|
}
|
|
|
|
// Before we start, obtain Carol's current wallet balance, we'll check
|
|
// to ensure that at the end of the force closure by Alice, Carol
|
|
// recognizes his new on-chain output.
|
|
carolBalReq := &lnrpc.WalletBalanceRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolBalResp, err := carol.WalletBalance(ctxt, carolBalReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol's balance: %v", err)
|
|
}
|
|
|
|
carolStartingBalance := carolBalResp.ConfirmedBalance
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, alice, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
|
|
// Wait for Alice and Carol to receive the channel edge from the
|
|
// funding manager.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = alice.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't see the alice->carol channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't see the alice->carol channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// Send payments from Alice to Carol, since Carol is htlchodl mode, the
|
|
// htlc outputs should be left unsettled, and should be swept by the
|
|
// utxo nursery.
|
|
ctx, cancel := context.WithCancel(ctxb)
|
|
defer cancel()
|
|
|
|
alicePayStream, err := alice.SendPayment(ctx)
|
|
if err != nil {
|
|
t.Fatalf("unable to create payment stream for alice: %v", err)
|
|
}
|
|
|
|
carolPubKey := carol.PubKey[:]
|
|
for i := 0; i < numInvoices; i++ {
|
|
err = alicePayStream.Send(&lnrpc.SendRequest{
|
|
Dest: carolPubKey,
|
|
Amt: int64(paymentAmt),
|
|
PaymentHash: makeFakePayHash(t),
|
|
FinalCltvDelta: lnd.DefaultBitcoinTimeLockDelta,
|
|
})
|
|
if err != nil {
|
|
t.Fatalf("unable to send alice htlc: %v", err)
|
|
}
|
|
}
|
|
|
|
// Once the HTLC has cleared, all the nodes n our mini network should
|
|
// show that the HTLC has been locked in.
|
|
nodes := []*lntest.HarnessNode{alice, carol}
|
|
var predErr error
|
|
err = wait.Predicate(func() bool {
|
|
predErr = assertNumActiveHtlcs(nodes, numInvoices)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("htlc mismatch: %v", predErr)
|
|
}
|
|
|
|
// Fetch starting height of this test so we can compute the block
|
|
// heights we expect certain events to take place.
|
|
_, curHeight, err := net.Miner.Node.GetBestBlock()
|
|
if err != nil {
|
|
t.Fatalf("unable to get best block height")
|
|
}
|
|
|
|
// Using the current height of the chain, derive the relevant heights
|
|
// for incubating two-stage htlcs.
|
|
var (
|
|
startHeight = uint32(curHeight)
|
|
commCsvMaturityHeight = startHeight + 1 + defaultCSV
|
|
htlcExpiryHeight = padCLTV(startHeight + defaultCLTV)
|
|
htlcCsvMaturityHeight = padCLTV(startHeight + defaultCLTV + 1 + defaultCSV)
|
|
)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
aliceChan, err := getChanInfo(ctxt, alice)
|
|
if err != nil {
|
|
t.Fatalf("unable to get alice's channel info: %v", err)
|
|
}
|
|
if aliceChan.NumUpdates == 0 {
|
|
t.Fatalf("alice should see at least one update to her channel")
|
|
}
|
|
|
|
// Now that the channel is open and we have unsettled htlcs, immediately
|
|
// execute a force closure of the channel. This will also assert that
|
|
// the commitment transaction was immediately broadcast in order to
|
|
// fulfill the force closure request.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
_, closingTxID, err := net.CloseChannel(ctxt, alice, chanPoint, true)
|
|
if err != nil {
|
|
t.Fatalf("unable to execute force channel closure: %v", err)
|
|
}
|
|
|
|
// Now that the channel has been force closed, it should show up in the
|
|
// PendingChannels RPC under the waiting close section.
|
|
pendingChansRequest := &lnrpc.PendingChannelsRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
pendingChanResp, err := alice.PendingChannels(ctxt, pendingChansRequest)
|
|
if err != nil {
|
|
t.Fatalf("unable to query for pending channels: %v", err)
|
|
}
|
|
err = checkNumWaitingCloseChannels(pendingChanResp, 1)
|
|
if err != nil {
|
|
t.Fatalf(err.Error())
|
|
}
|
|
|
|
// Compute the outpoint of the channel, which we will use repeatedly to
|
|
// locate the pending channel information in the rpc responses.
|
|
txid, err := lnd.GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
op := wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: chanPoint.OutputIndex,
|
|
}
|
|
|
|
waitingClose, err := findWaitingCloseChannel(pendingChanResp, &op)
|
|
if err != nil {
|
|
t.Fatalf(err.Error())
|
|
}
|
|
|
|
// Immediately after force closing, all of the funds should be in limbo.
|
|
if waitingClose.LimboBalance == 0 {
|
|
t.Fatalf("all funds should still be in limbo")
|
|
}
|
|
|
|
// The several restarts in this test are intended to ensure that when a
|
|
// channel is force-closed, the UTXO nursery has persisted the state of
|
|
// the channel in the closure process and will recover the correct state
|
|
// when the system comes back on line. This restart tests state
|
|
// persistence at the beginning of the process, when the commitment
|
|
// transaction has been broadcast but not yet confirmed in a block.
|
|
if err := net.RestartNode(alice, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// Mine a block which should confirm the commitment transaction
|
|
// broadcast as a result of the force closure. If there are anchors, we
|
|
// also expect the anchor sweep tx to be in the mempool.
|
|
expectedTxes := 1
|
|
if channelType == commitTypeAnchors {
|
|
expectedTxes = 2
|
|
}
|
|
|
|
_, err = waitForNTxsInMempool(
|
|
net.Miner.Node, expectedTxes, minerMempoolTimeout,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("failed to find commitment in miner mempool: %v", err)
|
|
}
|
|
|
|
if _, err := net.Miner.Node.Generate(1); err != nil {
|
|
t.Fatalf("unable to generate block: %v", err)
|
|
}
|
|
|
|
// Now that the commitment has been confirmed, the channel should be
|
|
// marked as force closed.
|
|
err = wait.NoError(func() error {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
pendingChanResp, err := alice.PendingChannels(
|
|
ctxt, pendingChansRequest,
|
|
)
|
|
if err != nil {
|
|
return fmt.Errorf("unable to query for pending "+
|
|
"channels: %v", err)
|
|
}
|
|
|
|
err = checkNumForceClosedChannels(pendingChanResp, 1)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
forceClose, err := findForceClosedChannel(pendingChanResp, &op)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
// Now that the channel has been force closed, it should now
|
|
// have the height and number of blocks to confirm populated.
|
|
err = checkCommitmentMaturity(
|
|
forceClose, commCsvMaturityHeight, int32(defaultCSV),
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
// None of our outputs have been swept, so they should all be in
|
|
// limbo. For anchors, we expect the anchor amount to be
|
|
// recovered.
|
|
if forceClose.LimboBalance == 0 {
|
|
return errors.New("all funds should still be in " +
|
|
"limbo")
|
|
}
|
|
expectedRecoveredBalance := int64(0)
|
|
if channelType == commitTypeAnchors {
|
|
expectedRecoveredBalance = anchorSize
|
|
}
|
|
if forceClose.RecoveredBalance != expectedRecoveredBalance {
|
|
return errors.New("no funds should yet be shown " +
|
|
"as recovered")
|
|
}
|
|
|
|
return nil
|
|
}, 15*time.Second)
|
|
if err != nil {
|
|
t.Fatalf(predErr.Error())
|
|
}
|
|
|
|
// The following restart is intended to ensure that outputs from the
|
|
// force close commitment transaction have been persisted once the
|
|
// transaction has been confirmed, but before the outputs are spendable
|
|
// (the "kindergarten" bucket.)
|
|
if err := net.RestartNode(alice, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// Carol's sweep tx should be in the mempool already, as her output is
|
|
// not timelocked. If there are anchors, we also expect Carol's anchor
|
|
// sweep now.
|
|
_, err = waitForNTxsInMempool(
|
|
net.Miner.Node, expectedTxes, minerMempoolTimeout,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("failed to find Carol's sweep in miner mempool: %v",
|
|
err)
|
|
}
|
|
|
|
// Currently within the codebase, the default CSV is 4 relative blocks.
|
|
// For the persistence test, we generate two blocks, then trigger
|
|
// a restart and then generate the final block that should trigger
|
|
// the creation of the sweep transaction.
|
|
if _, err := net.Miner.Node.Generate(defaultCSV - 2); err != nil {
|
|
t.Fatalf("unable to mine blocks: %v", err)
|
|
}
|
|
|
|
// The following restart checks to ensure that outputs in the
|
|
// kindergarten bucket are persisted while waiting for the required
|
|
// number of confirmations to be reported.
|
|
if err := net.RestartNode(alice, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// Alice should see the channel in her set of pending force closed
|
|
// channels with her funds still in limbo.
|
|
err = wait.NoError(func() error {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
pendingChanResp, err := alice.PendingChannels(
|
|
ctxt, pendingChansRequest,
|
|
)
|
|
if err != nil {
|
|
return fmt.Errorf("unable to query for pending "+
|
|
"channels: %v", err)
|
|
}
|
|
|
|
err = checkNumForceClosedChannels(pendingChanResp, 1)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
forceClose, err := findForceClosedChannel(
|
|
pendingChanResp, &op,
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
// At this point, the nursery should show that the commitment
|
|
// output has 2 block left before its CSV delay expires. In
|
|
// total, we have mined exactly defaultCSV blocks, so the htlc
|
|
// outputs should also reflect that this many blocks have
|
|
// passed.
|
|
err = checkCommitmentMaturity(
|
|
forceClose, commCsvMaturityHeight, 2,
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
// All funds should still be shown in limbo.
|
|
if forceClose.LimboBalance == 0 {
|
|
return errors.New("all funds should still be in " +
|
|
"limbo")
|
|
}
|
|
expectedRecoveredBalance := int64(0)
|
|
if channelType == commitTypeAnchors {
|
|
expectedRecoveredBalance = anchorSize
|
|
}
|
|
if forceClose.RecoveredBalance != expectedRecoveredBalance {
|
|
return errors.New("no funds should yet be shown " +
|
|
"as recovered")
|
|
}
|
|
|
|
return nil
|
|
}, 15*time.Second)
|
|
if err != nil {
|
|
t.Fatalf(err.Error())
|
|
}
|
|
|
|
// Generate an additional block, which should cause the CSV delayed
|
|
// output from the commitment txn to expire.
|
|
if _, err := net.Miner.Node.Generate(1); err != nil {
|
|
t.Fatalf("unable to mine blocks: %v", err)
|
|
}
|
|
|
|
// At this point, the CSV will expire in the next block, meaning that
|
|
// the sweeping transaction should now be broadcast. So we fetch the
|
|
// node's mempool to ensure it has been properly broadcast.
|
|
sweepingTXID, err := waitForTxInMempool(net.Miner.Node, minerMempoolTimeout)
|
|
if err != nil {
|
|
t.Fatalf("failed to get sweep tx from mempool: %v", err)
|
|
}
|
|
|
|
// Fetch the sweep transaction, all input it's spending should be from
|
|
// the commitment transaction which was broadcast on-chain.
|
|
sweepTx, err := net.Miner.Node.GetRawTransaction(sweepingTXID)
|
|
if err != nil {
|
|
t.Fatalf("unable to fetch sweep tx: %v", err)
|
|
}
|
|
for _, txIn := range sweepTx.MsgTx().TxIn {
|
|
if !closingTxID.IsEqual(&txIn.PreviousOutPoint.Hash) {
|
|
t.Fatalf("sweep transaction not spending from commit "+
|
|
"tx %v, instead spending %v",
|
|
closingTxID, txIn.PreviousOutPoint)
|
|
}
|
|
}
|
|
|
|
// Restart Alice to ensure that she resumes watching the finalized
|
|
// commitment sweep txid.
|
|
if err := net.RestartNode(alice, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// Next, we mine an additional block which should include the sweep
|
|
// transaction as the input scripts and the sequence locks on the
|
|
// inputs should be properly met.
|
|
blockHash, err := net.Miner.Node.Generate(1)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate block: %v", err)
|
|
}
|
|
block, err := net.Miner.Node.GetBlock(blockHash[0])
|
|
if err != nil {
|
|
t.Fatalf("unable to get block: %v", err)
|
|
}
|
|
|
|
assertTxInBlock(t, block, sweepTx.Hash())
|
|
|
|
// Update current height
|
|
_, curHeight, err = net.Miner.Node.GetBestBlock()
|
|
if err != nil {
|
|
t.Fatalf("unable to get best block height")
|
|
}
|
|
|
|
err = wait.Predicate(func() bool {
|
|
// Now that the commit output has been fully swept, check to see
|
|
// that the channel remains open for the pending htlc outputs.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
pendingChanResp, err := alice.PendingChannels(
|
|
ctxt, pendingChansRequest,
|
|
)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to query for pending "+
|
|
"channels: %v", err)
|
|
return false
|
|
}
|
|
|
|
err = checkNumForceClosedChannels(pendingChanResp, 1)
|
|
if err != nil {
|
|
predErr = err
|
|
return false
|
|
}
|
|
|
|
// The commitment funds will have been recovered after the
|
|
// commit txn was included in the last block. The htlc funds
|
|
// will be shown in limbo.
|
|
forceClose, err := findForceClosedChannel(pendingChanResp, &op)
|
|
if err != nil {
|
|
predErr = err
|
|
return false
|
|
}
|
|
predErr = checkPendingChannelNumHtlcs(forceClose, numInvoices)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
predErr = checkPendingHtlcStageAndMaturity(
|
|
forceClose, 1, htlcExpiryHeight,
|
|
int32(htlcExpiryHeight)-curHeight,
|
|
)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
if forceClose.LimboBalance == 0 {
|
|
predErr = fmt.Errorf("expected funds in limbo, found 0")
|
|
return false
|
|
}
|
|
|
|
return true
|
|
}, 15*time.Second)
|
|
if err != nil {
|
|
t.Fatalf(predErr.Error())
|
|
}
|
|
|
|
// Compute the height preceding that which will cause the htlc CLTV
|
|
// timeouts will expire. The outputs entered at the same height as the
|
|
// output spending from the commitment txn, so we must deduct the number
|
|
// of blocks we have generated since adding it to the nursery, and take
|
|
// an additional block off so that we end up one block shy of the expiry
|
|
// height, and add the block padding.
|
|
cltvHeightDelta := padCLTV(defaultCLTV - defaultCSV - 1 - 1)
|
|
|
|
// Advance the blockchain until just before the CLTV expires, nothing
|
|
// exciting should have happened during this time.
|
|
blockHash, err = net.Miner.Node.Generate(cltvHeightDelta)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate block: %v", err)
|
|
}
|
|
|
|
// We now restart Alice, to ensure that she will broadcast the presigned
|
|
// htlc timeout txns after the delay expires after experiencing a while
|
|
// waiting for the htlc outputs to incubate.
|
|
if err := net.RestartNode(alice, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// Alice should now see the channel in her set of pending force closed
|
|
// channels with one pending HTLC.
|
|
err = wait.NoError(func() error {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
pendingChanResp, err := alice.PendingChannels(
|
|
ctxt, pendingChansRequest,
|
|
)
|
|
if err != nil {
|
|
return fmt.Errorf("unable to query for pending "+
|
|
"channels: %v", err)
|
|
}
|
|
|
|
err = checkNumForceClosedChannels(pendingChanResp, 1)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
forceClose, err := findForceClosedChannel(
|
|
pendingChanResp, &op,
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
// We should now be at the block just before the utxo nursery
|
|
// will attempt to broadcast the htlc timeout transactions.
|
|
err = checkPendingChannelNumHtlcs(forceClose, numInvoices)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
err = checkPendingHtlcStageAndMaturity(
|
|
forceClose, 1, htlcExpiryHeight, 1,
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
// Now that our commitment confirmation depth has been
|
|
// surpassed, we should now see a non-zero recovered balance.
|
|
// All htlc outputs are still left in limbo, so it should be
|
|
// non-zero as well.
|
|
if forceClose.LimboBalance == 0 {
|
|
return errors.New("htlc funds should still be in " +
|
|
"limbo")
|
|
}
|
|
|
|
return nil
|
|
}, 15*time.Second)
|
|
if err != nil {
|
|
t.Fatalf(err.Error())
|
|
}
|
|
|
|
// Now, generate the block which will cause Alice to broadcast the
|
|
// presigned htlc timeout txns.
|
|
blockHash, err = net.Miner.Node.Generate(1)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate block: %v", err)
|
|
}
|
|
|
|
// Since Alice had numInvoices (6) htlcs extended to Carol before force
|
|
// closing, we expect Alice to broadcast an htlc timeout txn for each
|
|
// one. Wait for them all to show up in the mempool.
|
|
htlcTxIDs, err := waitForNTxsInMempool(net.Miner.Node, numInvoices,
|
|
minerMempoolTimeout)
|
|
if err != nil {
|
|
t.Fatalf("unable to find htlc timeout txns in mempool: %v", err)
|
|
}
|
|
|
|
// Retrieve each htlc timeout txn from the mempool, and ensure it is
|
|
// well-formed. This entails verifying that each only spends from
|
|
// output, and that that output is from the commitment txn.
|
|
for _, htlcTxID := range htlcTxIDs {
|
|
// Fetch the sweep transaction, all input it's spending should
|
|
// be from the commitment transaction which was broadcast
|
|
// on-chain.
|
|
htlcTx, err := net.Miner.Node.GetRawTransaction(htlcTxID)
|
|
if err != nil {
|
|
t.Fatalf("unable to fetch sweep tx: %v", err)
|
|
}
|
|
// Ensure the htlc transaction only has one input.
|
|
if len(htlcTx.MsgTx().TxIn) != 1 {
|
|
t.Fatalf("htlc transaction should only have one txin, "+
|
|
"has %d", len(htlcTx.MsgTx().TxIn))
|
|
}
|
|
// Ensure the htlc transaction is spending from the commitment
|
|
// transaction.
|
|
txIn := htlcTx.MsgTx().TxIn[0]
|
|
if !closingTxID.IsEqual(&txIn.PreviousOutPoint.Hash) {
|
|
t.Fatalf("htlc transaction not spending from commit "+
|
|
"tx %v, instead spending %v",
|
|
closingTxID, txIn.PreviousOutPoint)
|
|
}
|
|
}
|
|
|
|
// With the htlc timeout txns still in the mempool, we restart Alice to
|
|
// verify that she can resume watching the htlc txns she broadcasted
|
|
// before crashing.
|
|
if err := net.RestartNode(alice, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// Generate a block that mines the htlc timeout txns. Doing so now
|
|
// activates the 2nd-stage CSV delayed outputs.
|
|
blockHash, err = net.Miner.Node.Generate(1)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate block: %v", err)
|
|
}
|
|
|
|
// Alice is restarted here to ensure that she promptly moved the crib
|
|
// outputs to the kindergarten bucket after the htlc timeout txns were
|
|
// confirmed.
|
|
if err := net.RestartNode(alice, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// Advance the chain until just before the 2nd-layer CSV delays expire.
|
|
blockHash, err = net.Miner.Node.Generate(defaultCSV - 1)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate block: %v", err)
|
|
}
|
|
|
|
// Restart Alice to ensure that she can recover from a failure before
|
|
// having graduated the htlc outputs in the kindergarten bucket.
|
|
if err := net.RestartNode(alice, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// Now that the channel has been fully swept, it should no longer show
|
|
// incubated, check to see that Alice's node still reports the channel
|
|
// as pending force closed.
|
|
err = wait.Predicate(func() bool {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
pendingChanResp, err = alice.PendingChannels(
|
|
ctxt, pendingChansRequest,
|
|
)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to query for pending "+
|
|
"channels: %v", err)
|
|
return false
|
|
}
|
|
err = checkNumForceClosedChannels(pendingChanResp, 1)
|
|
if err != nil {
|
|
predErr = err
|
|
return false
|
|
}
|
|
|
|
forceClose, err := findForceClosedChannel(pendingChanResp, &op)
|
|
if err != nil {
|
|
predErr = err
|
|
return false
|
|
}
|
|
|
|
if forceClose.LimboBalance == 0 {
|
|
predErr = fmt.Errorf("htlc funds should still be in limbo")
|
|
return false
|
|
}
|
|
|
|
predErr = checkPendingChannelNumHtlcs(forceClose, numInvoices)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
|
|
return true
|
|
}, 15*time.Second)
|
|
if err != nil {
|
|
t.Fatalf(predErr.Error())
|
|
}
|
|
|
|
// Generate a block that causes Alice to sweep the htlc outputs in the
|
|
// kindergarten bucket.
|
|
blockHash, err = net.Miner.Node.Generate(1)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate block: %v", err)
|
|
}
|
|
|
|
// Wait for the single sweep txn to appear in the mempool.
|
|
htlcSweepTxID, err := waitForTxInMempool(
|
|
net.Miner.Node, minerMempoolTimeout,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("failed to get sweep tx from mempool: %v", err)
|
|
}
|
|
|
|
// Construct a map of the already confirmed htlc timeout txids, that
|
|
// will count the number of times each is spent by the sweep txn. We
|
|
// prepopulate it in this way so that we can later detect if we are
|
|
// spending from an output that was not a confirmed htlc timeout txn.
|
|
var htlcTxIDSet = make(map[chainhash.Hash]int)
|
|
for _, htlcTxID := range htlcTxIDs {
|
|
htlcTxIDSet[*htlcTxID] = 0
|
|
}
|
|
|
|
// Fetch the htlc sweep transaction from the mempool.
|
|
htlcSweepTx, err := net.Miner.Node.GetRawTransaction(htlcSweepTxID)
|
|
if err != nil {
|
|
t.Fatalf("unable to fetch sweep tx: %v", err)
|
|
}
|
|
// Ensure the htlc sweep transaction only has one input for each htlc
|
|
// Alice extended before force closing.
|
|
if len(htlcSweepTx.MsgTx().TxIn) != numInvoices {
|
|
t.Fatalf("htlc transaction should have %d txin, "+
|
|
"has %d", numInvoices, len(htlcSweepTx.MsgTx().TxIn))
|
|
}
|
|
// Ensure that each output spends from exactly one htlc timeout txn.
|
|
for _, txIn := range htlcSweepTx.MsgTx().TxIn {
|
|
outpoint := txIn.PreviousOutPoint.Hash
|
|
// Check that the input is a confirmed htlc timeout txn.
|
|
if _, ok := htlcTxIDSet[outpoint]; !ok {
|
|
t.Fatalf("htlc sweep output not spending from htlc "+
|
|
"tx, instead spending output %v", outpoint)
|
|
}
|
|
// Increment our count for how many times this output was spent.
|
|
htlcTxIDSet[outpoint]++
|
|
|
|
// Check that each is only spent once.
|
|
if htlcTxIDSet[outpoint] > 1 {
|
|
t.Fatalf("htlc sweep tx has multiple spends from "+
|
|
"outpoint %v", outpoint)
|
|
}
|
|
}
|
|
|
|
// The following restart checks to ensure that the nursery store is
|
|
// storing the txid of the previously broadcast htlc sweep txn, and that
|
|
// it begins watching that txid after restarting.
|
|
if err := net.RestartNode(alice, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// Now that the channel has been fully swept, it should no longer show
|
|
// incubated, check to see that Alice's node still reports the channel
|
|
// as pending force closed.
|
|
err = wait.Predicate(func() bool {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
pendingChanResp, err := alice.PendingChannels(
|
|
ctxt, pendingChansRequest,
|
|
)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to query for pending "+
|
|
"channels: %v", err)
|
|
return false
|
|
}
|
|
err = checkNumForceClosedChannels(pendingChanResp, 1)
|
|
if err != nil {
|
|
predErr = err
|
|
return false
|
|
}
|
|
|
|
// All htlcs should show zero blocks until maturity, as
|
|
// evidenced by having checked the sweep transaction in the
|
|
// mempool.
|
|
forceClose, err := findForceClosedChannel(pendingChanResp, &op)
|
|
if err != nil {
|
|
predErr = err
|
|
return false
|
|
}
|
|
predErr = checkPendingChannelNumHtlcs(forceClose, numInvoices)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
err = checkPendingHtlcStageAndMaturity(
|
|
forceClose, 2, htlcCsvMaturityHeight, 0,
|
|
)
|
|
if err != nil {
|
|
predErr = err
|
|
return false
|
|
}
|
|
|
|
return true
|
|
}, 15*time.Second)
|
|
if err != nil {
|
|
t.Fatalf(predErr.Error())
|
|
}
|
|
|
|
// Generate the final block that sweeps all htlc funds into the user's
|
|
// wallet, and make sure the sweep is in this block.
|
|
block = mineBlocks(t, net, 1, 1)[0]
|
|
assertTxInBlock(t, block, htlcSweepTxID)
|
|
|
|
// Now that the channel has been fully swept, it should no longer show
|
|
// up within the pending channels RPC.
|
|
err = wait.Predicate(func() bool {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
pendingChanResp, err := alice.PendingChannels(
|
|
ctxt, pendingChansRequest,
|
|
)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to query for pending "+
|
|
"channels: %v", err)
|
|
return false
|
|
}
|
|
|
|
predErr = checkNumForceClosedChannels(pendingChanResp, 0)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
|
|
// In addition to there being no pending channels, we verify
|
|
// that pending channels does not report any money still in
|
|
// limbo.
|
|
if pendingChanResp.TotalLimboBalance != 0 {
|
|
predErr = errors.New("no user funds should be left " +
|
|
"in limbo after incubation")
|
|
return false
|
|
}
|
|
|
|
return true
|
|
}, 15*time.Second)
|
|
if err != nil {
|
|
t.Fatalf(predErr.Error())
|
|
}
|
|
|
|
// At this point, Bob should now be aware of his new immediately
|
|
// spendable on-chain balance, as it was Alice who broadcast the
|
|
// commitment transaction.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolBalResp, err = net.Bob.WalletBalance(ctxt, carolBalReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol's balance: %v", err)
|
|
}
|
|
carolExpectedBalance := btcutil.Amount(carolStartingBalance) + pushAmt
|
|
if btcutil.Amount(carolBalResp.ConfirmedBalance) < carolExpectedBalance {
|
|
t.Fatalf("carol's balance is incorrect: expected %v got %v",
|
|
carolExpectedBalance,
|
|
carolBalResp.ConfirmedBalance)
|
|
}
|
|
}
|
|
|
|
// assertAmountSent generates a closure which queries listchannels for sndr and
|
|
// rcvr, and asserts that sndr sent amt satoshis, and that rcvr received amt
|
|
// satoshis.
|
|
//
|
|
// NOTE: This method assumes that each node only has one channel, and it is the
|
|
// channel used to send the payment.
|
|
func assertAmountSent(amt btcutil.Amount, sndr, rcvr *lntest.HarnessNode) func() error {
|
|
return func() error {
|
|
// Both channels should also have properly accounted from the
|
|
// amount that has been sent/received over the channel.
|
|
listReq := &lnrpc.ListChannelsRequest{}
|
|
ctxb := context.Background()
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
sndrListChannels, err := sndr.ListChannels(ctxt, listReq)
|
|
if err != nil {
|
|
return fmt.Errorf("unable to query for %s's channel "+
|
|
"list: %v", sndr.Name(), err)
|
|
}
|
|
sndrSatoshisSent := sndrListChannels.Channels[0].TotalSatoshisSent
|
|
if sndrSatoshisSent != int64(amt) {
|
|
return fmt.Errorf("%s's satoshis sent is incorrect "+
|
|
"got %v, expected %v", sndr.Name(),
|
|
sndrSatoshisSent, amt)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
rcvrListChannels, err := rcvr.ListChannels(ctxt, listReq)
|
|
if err != nil {
|
|
return fmt.Errorf("unable to query for %s's channel "+
|
|
"list: %v", rcvr.Name(), err)
|
|
}
|
|
rcvrSatoshisReceived := rcvrListChannels.Channels[0].TotalSatoshisReceived
|
|
if rcvrSatoshisReceived != int64(amt) {
|
|
return fmt.Errorf("%s's satoshis received is "+
|
|
"incorrect got %v, expected %v", rcvr.Name(),
|
|
rcvrSatoshisReceived, amt)
|
|
}
|
|
|
|
return nil
|
|
}
|
|
}
|
|
|
|
// assertLastHTLCError checks that the last sent HTLC of the last payment sent
|
|
// by the given node failed with the expected failure code.
|
|
func assertLastHTLCError(t *harnessTest, node *lntest.HarnessNode,
|
|
code lnrpc.Failure_FailureCode) {
|
|
|
|
req := &lnrpc.ListPaymentsRequest{
|
|
IncludeIncomplete: true,
|
|
}
|
|
ctxt, _ := context.WithTimeout(context.Background(), defaultTimeout)
|
|
paymentsResp, err := node.ListPayments(ctxt, req)
|
|
if err != nil {
|
|
t.Fatalf("error when obtaining payments: %v", err)
|
|
}
|
|
|
|
payments := paymentsResp.Payments
|
|
if len(payments) == 0 {
|
|
t.Fatalf("no payments found")
|
|
}
|
|
|
|
payment := payments[len(payments)-1]
|
|
htlcs := payment.Htlcs
|
|
if len(htlcs) == 0 {
|
|
t.Fatalf("no htlcs")
|
|
}
|
|
|
|
htlc := htlcs[len(htlcs)-1]
|
|
if htlc.Failure == nil {
|
|
t.Fatalf("expected failure")
|
|
}
|
|
|
|
if htlc.Failure.Code != code {
|
|
t.Fatalf("expected failure %v, got %v", code, htlc.Failure.Code)
|
|
}
|
|
}
|
|
|
|
// testSphinxReplayPersistence verifies that replayed onion packets are rejected
|
|
// by a remote peer after a restart. We use a combination of unsafe
|
|
// configuration arguments to force Carol to replay the same sphinx packet after
|
|
// reconnecting to Dave, and compare the returned failure message with what we
|
|
// expect for replayed onion packets.
|
|
func testSphinxReplayPersistence(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
// Open a channel with 100k satoshis between Carol and Dave with Carol being
|
|
// the sole funder of the channel.
|
|
chanAmt := btcutil.Amount(100000)
|
|
|
|
// First, we'll create Dave, the receiver, and start him in hodl mode.
|
|
dave, err := net.NewNode("Dave", []string{"--hodl.exit-settle"})
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
|
|
// We must remember to shutdown the nodes we created for the duration
|
|
// of the tests, only leaving the two seed nodes (Alice and Bob) within
|
|
// our test network.
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
// Next, we'll create Carol and establish a channel to from her to
|
|
// Dave. Carol is started in both unsafe-replay which will cause her to
|
|
// replay any pending Adds held in memory upon reconnection.
|
|
carol, err := net.NewNode("Carol", []string{"--unsafe-replay"})
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, dave); err != nil {
|
|
t.Fatalf("unable to connect carol to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, carol, dave,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Now that the channel is open, create an invoice for Dave which
|
|
// expects a payment of 1000 satoshis from Carol paid via a particular
|
|
// preimage.
|
|
const paymentAmt = 1000
|
|
preimage := bytes.Repeat([]byte("A"), 32)
|
|
invoice := &lnrpc.Invoice{
|
|
Memo: "testing",
|
|
RPreimage: preimage,
|
|
Value: paymentAmt,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
invoiceResp, err := dave.AddInvoice(ctxt, invoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
|
|
// Wait for Carol to recognize and advertise the new channel generated
|
|
// above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't advertise channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
err = dave.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("bob didn't advertise channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// With the invoice for Dave added, send a payment from Carol paying
|
|
// to the above generated invoice.
|
|
ctx, cancel := context.WithCancel(ctxb)
|
|
defer cancel()
|
|
|
|
payStream, err := carol.SendPayment(ctx)
|
|
if err != nil {
|
|
t.Fatalf("unable to open payment stream: %v", err)
|
|
}
|
|
|
|
sendReq := &lnrpc.SendRequest{PaymentRequest: invoiceResp.PaymentRequest}
|
|
err = payStream.Send(sendReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
|
|
time.Sleep(200 * time.Millisecond)
|
|
|
|
// Dave's invoice should not be marked as settled.
|
|
payHash := &lnrpc.PaymentHash{
|
|
RHash: invoiceResp.RHash,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
dbInvoice, err := dave.LookupInvoice(ctxt, payHash)
|
|
if err != nil {
|
|
t.Fatalf("unable to lookup invoice: %v", err)
|
|
}
|
|
if dbInvoice.Settled {
|
|
t.Fatalf("dave's invoice should not be marked as settled: %v",
|
|
spew.Sdump(dbInvoice))
|
|
}
|
|
|
|
// With the payment sent but hedl, all balance related stats should not
|
|
// have changed.
|
|
err = wait.InvariantNoError(
|
|
assertAmountSent(0, carol, dave), 3*time.Second,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf(err.Error())
|
|
}
|
|
|
|
// With the first payment sent, restart dave to make sure he is
|
|
// persisting the information required to detect replayed sphinx
|
|
// packets.
|
|
if err := net.RestartNode(dave, nil); err != nil {
|
|
t.Fatalf("unable to restart dave: %v", err)
|
|
}
|
|
|
|
// Carol should retransmit the Add hedl in her mailbox on startup. Dave
|
|
// should not accept the replayed Add, and actually fail back the
|
|
// pending payment. Even though he still holds the original settle, if
|
|
// he does fail, it is almost certainly caused by the sphinx replay
|
|
// protection, as it is the only validation we do in hodl mode.
|
|
resp, err := payStream.Recv()
|
|
if err != nil {
|
|
t.Fatalf("unable to receive payment response: %v", err)
|
|
}
|
|
|
|
// Construct the response we expect after sending a duplicate packet
|
|
// that fails due to sphinx replay detection.
|
|
if resp.PaymentError == "" {
|
|
t.Fatalf("expected payment error")
|
|
}
|
|
assertLastHTLCError(t, carol, lnrpc.Failure_INVALID_ONION_KEY)
|
|
|
|
// Since the payment failed, the balance should still be left
|
|
// unaltered.
|
|
err = wait.InvariantNoError(
|
|
assertAmountSent(0, carol, dave), 3*time.Second,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf(err.Error())
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPoint, true)
|
|
|
|
// Cleanup by mining the force close and sweep transaction.
|
|
cleanupForceClose(t, net, carol, chanPoint)
|
|
}
|
|
|
|
func testListPayments(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
// First start by deleting all payments that Alice knows of. This will
|
|
// allow us to execute the test with a clean state for Alice.
|
|
delPaymentsReq := &lnrpc.DeleteAllPaymentsRequest{}
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if _, err := net.Alice.DeleteAllPayments(ctxt, delPaymentsReq); err != nil {
|
|
t.Fatalf("unable to delete payments: %v", err)
|
|
}
|
|
|
|
// Check that there are no payments before test.
|
|
reqInit := &lnrpc.ListPaymentsRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxt, defaultTimeout)
|
|
paymentsRespInit, err := net.Alice.ListPayments(ctxt, reqInit)
|
|
if err != nil {
|
|
t.Fatalf("error when obtaining Alice payments: %v", err)
|
|
}
|
|
if len(paymentsRespInit.Payments) != 0 {
|
|
t.Fatalf("incorrect number of payments, got %v, want %v",
|
|
len(paymentsRespInit.Payments), 0)
|
|
}
|
|
|
|
// Open a channel with 100k satoshis between Alice and Bob with Alice
|
|
// being the sole funder of the channel.
|
|
chanAmt := btcutil.Amount(100000)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Now that the channel is open, create an invoice for Bob which
|
|
// expects a payment of 1000 satoshis from Alice paid via a particular
|
|
// preimage.
|
|
const paymentAmt = 1000
|
|
preimage := bytes.Repeat([]byte("B"), 32)
|
|
invoice := &lnrpc.Invoice{
|
|
Memo: "testing",
|
|
RPreimage: preimage,
|
|
Value: paymentAmt,
|
|
}
|
|
addInvoiceCtxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
invoiceResp, err := net.Bob.AddInvoice(addInvoiceCtxt, invoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
|
|
// Wait for Alice to recognize and advertise the new channel generated
|
|
// above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err = net.Alice.WaitForNetworkChannelOpen(ctxt, chanPoint); err != nil {
|
|
t.Fatalf("alice didn't advertise channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
if err = net.Bob.WaitForNetworkChannelOpen(ctxt, chanPoint); err != nil {
|
|
t.Fatalf("bob didn't advertise channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// With the invoice for Bob added, send a payment towards Alice paying
|
|
// to the above generated invoice.
|
|
sendReq := &lnrpc.SendRequest{
|
|
PaymentRequest: invoiceResp.PaymentRequest,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := net.Alice.SendPaymentSync(ctxt, sendReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
if resp.PaymentError != "" {
|
|
t.Fatalf("error when attempting recv: %v", resp.PaymentError)
|
|
}
|
|
|
|
// Grab Alice's list of payments, she should show the existence of
|
|
// exactly one payment.
|
|
req := &lnrpc.ListPaymentsRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxt, defaultTimeout)
|
|
paymentsResp, err := net.Alice.ListPayments(ctxt, req)
|
|
if err != nil {
|
|
t.Fatalf("error when obtaining Alice payments: %v", err)
|
|
}
|
|
if len(paymentsResp.Payments) != 1 {
|
|
t.Fatalf("incorrect number of payments, got %v, want %v",
|
|
len(paymentsResp.Payments), 1)
|
|
}
|
|
p := paymentsResp.Payments[0]
|
|
|
|
// Ensure that the stored path shows a direct payment to Bob with no
|
|
// other nodes in-between.
|
|
expectedPath := []string{
|
|
net.Bob.PubKeyStr,
|
|
}
|
|
if !reflect.DeepEqual(p.Path, expectedPath) {
|
|
t.Fatalf("incorrect path, got %v, want %v",
|
|
p.Path, expectedPath)
|
|
}
|
|
|
|
// The payment amount should also match our previous payment directly.
|
|
if p.Value != paymentAmt {
|
|
t.Fatalf("incorrect amount, got %v, want %v",
|
|
p.Value, paymentAmt)
|
|
}
|
|
|
|
// The payment hash (or r-hash) should have been stored correctly.
|
|
correctRHash := hex.EncodeToString(invoiceResp.RHash)
|
|
if !reflect.DeepEqual(p.PaymentHash, correctRHash) {
|
|
t.Fatalf("incorrect RHash, got %v, want %v",
|
|
p.PaymentHash, correctRHash)
|
|
}
|
|
|
|
// As we made a single-hop direct payment, there should have been no fee
|
|
// applied.
|
|
if p.Fee != 0 {
|
|
t.Fatalf("incorrect Fee, got %v, want %v", p.Fee, 0)
|
|
}
|
|
|
|
// Finally, verify that the payment request returned by the rpc matches
|
|
// the invoice that we paid.
|
|
if p.PaymentRequest != invoiceResp.PaymentRequest {
|
|
t.Fatalf("incorrect payreq, got: %v, want: %v",
|
|
p.PaymentRequest, invoiceResp.PaymentRequest)
|
|
}
|
|
|
|
// Delete all payments from Alice. DB should have no payments.
|
|
delReq := &lnrpc.DeleteAllPaymentsRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxt, defaultTimeout)
|
|
_, err = net.Alice.DeleteAllPayments(ctxt, delReq)
|
|
if err != nil {
|
|
t.Fatalf("Can't delete payments at the end: %v", err)
|
|
}
|
|
|
|
// Check that there are no payments after test.
|
|
listReq := &lnrpc.ListPaymentsRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxt, defaultTimeout)
|
|
paymentsResp, err = net.Alice.ListPayments(ctxt, listReq)
|
|
if err != nil {
|
|
t.Fatalf("error when obtaining Alice payments: %v", err)
|
|
}
|
|
if len(paymentsResp.Payments) != 0 {
|
|
t.Fatalf("incorrect number of payments, got %v, want %v",
|
|
len(paymentsRespInit.Payments), 0)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPoint, false)
|
|
}
|
|
|
|
// assertAmountPaid checks that the ListChannels command of the provided
|
|
// node list the total amount sent and received as expected for the
|
|
// provided channel.
|
|
func assertAmountPaid(t *harnessTest, channelName string,
|
|
node *lntest.HarnessNode, chanPoint wire.OutPoint, amountSent,
|
|
amountReceived int64) {
|
|
ctxb := context.Background()
|
|
|
|
checkAmountPaid := func() error {
|
|
listReq := &lnrpc.ListChannelsRequest{}
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := node.ListChannels(ctxt, listReq)
|
|
if err != nil {
|
|
return fmt.Errorf("unable to for node's "+
|
|
"channels: %v", err)
|
|
}
|
|
for _, channel := range resp.Channels {
|
|
if channel.ChannelPoint != chanPoint.String() {
|
|
continue
|
|
}
|
|
|
|
if channel.TotalSatoshisSent != amountSent {
|
|
return fmt.Errorf("%v: incorrect amount"+
|
|
" sent: %v != %v", channelName,
|
|
channel.TotalSatoshisSent,
|
|
amountSent)
|
|
}
|
|
if channel.TotalSatoshisReceived !=
|
|
amountReceived {
|
|
return fmt.Errorf("%v: incorrect amount"+
|
|
" received: %v != %v",
|
|
channelName,
|
|
channel.TotalSatoshisReceived,
|
|
amountReceived)
|
|
}
|
|
|
|
return nil
|
|
}
|
|
return fmt.Errorf("channel not found")
|
|
}
|
|
|
|
// As far as HTLC inclusion in commitment transaction might be
|
|
// postponed we will try to check the balance couple of times,
|
|
// and then if after some period of time we receive wrong
|
|
// balance return the error.
|
|
// TODO(roasbeef): remove sleep after invoice notification hooks
|
|
// are in place
|
|
var timeover uint32
|
|
go func() {
|
|
<-time.After(time.Second * 20)
|
|
atomic.StoreUint32(&timeover, 1)
|
|
}()
|
|
|
|
for {
|
|
isTimeover := atomic.LoadUint32(&timeover) == 1
|
|
if err := checkAmountPaid(); err != nil {
|
|
if isTimeover {
|
|
t.Fatalf("Check amount Paid failed: %v", err)
|
|
}
|
|
} else {
|
|
break
|
|
}
|
|
}
|
|
}
|
|
|
|
// updateChannelPolicy updates the channel policy of node to the
|
|
// given fees and timelock delta. This function blocks until
|
|
// listenerNode has received the policy update.
|
|
func updateChannelPolicy(t *harnessTest, node *lntest.HarnessNode,
|
|
chanPoint *lnrpc.ChannelPoint, baseFee int64, feeRate int64,
|
|
timeLockDelta uint32, maxHtlc uint64, listenerNode *lntest.HarnessNode) {
|
|
|
|
ctxb := context.Background()
|
|
|
|
expectedPolicy := &lnrpc.RoutingPolicy{
|
|
FeeBaseMsat: baseFee,
|
|
FeeRateMilliMsat: feeRate,
|
|
TimeLockDelta: timeLockDelta,
|
|
MinHtlc: 1000, // default value
|
|
MaxHtlcMsat: maxHtlc,
|
|
}
|
|
|
|
updateFeeReq := &lnrpc.PolicyUpdateRequest{
|
|
BaseFeeMsat: baseFee,
|
|
FeeRate: float64(feeRate) / testFeeBase,
|
|
TimeLockDelta: timeLockDelta,
|
|
Scope: &lnrpc.PolicyUpdateRequest_ChanPoint{
|
|
ChanPoint: chanPoint,
|
|
},
|
|
MaxHtlcMsat: maxHtlc,
|
|
}
|
|
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if _, err := node.UpdateChannelPolicy(ctxt, updateFeeReq); err != nil {
|
|
t.Fatalf("unable to update chan policy: %v", err)
|
|
}
|
|
|
|
// Wait for listener node to receive the channel update from node.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
graphSub := subscribeGraphNotifications(t, ctxt, listenerNode)
|
|
defer close(graphSub.quit)
|
|
|
|
waitForChannelUpdate(
|
|
t, graphSub,
|
|
[]expectedChanUpdate{
|
|
{node.PubKeyStr, expectedPolicy, chanPoint},
|
|
},
|
|
)
|
|
}
|
|
|
|
func testMultiHopPayments(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const chanAmt = btcutil.Amount(100000)
|
|
var networkChans []*lnrpc.ChannelPoint
|
|
|
|
// Open a channel with 100k satoshis between Alice and Bob with Alice
|
|
// being the sole funder of the channel.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAlice := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointAlice)
|
|
|
|
aliceChanTXID, err := lnd.GetChanPointFundingTxid(chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
aliceFundPoint := wire.OutPoint{
|
|
Hash: *aliceChanTXID,
|
|
Index: chanPointAlice.OutputIndex,
|
|
}
|
|
|
|
// As preliminary setup, we'll create two new nodes: Carol and Dave,
|
|
// such that we now have a 4 node, 3 channel topology. Dave will make a
|
|
// channel with Alice, and Carol with Dave. After this setup, the
|
|
// network topology should now look like:
|
|
// Carol -> Dave -> Alice -> Bob
|
|
//
|
|
// First, we'll create Dave and establish a channel to Alice. Dave will
|
|
// be running an older node that requires the legacy onion payload.
|
|
daveArgs := []string{"--protocol.legacyonion"}
|
|
dave, err := net.NewNode("Dave", daveArgs)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, dave, net.Alice); err != nil {
|
|
t.Fatalf("unable to connect dave to alice: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointDave := openChannelAndAssert(
|
|
ctxt, t, net, dave, net.Alice,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointDave)
|
|
daveChanTXID, err := lnd.GetChanPointFundingTxid(chanPointDave)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
daveFundPoint := wire.OutPoint{
|
|
Hash: *daveChanTXID,
|
|
Index: chanPointDave.OutputIndex,
|
|
}
|
|
|
|
// Next, we'll create Carol and establish a channel to from her to
|
|
// Dave.
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, dave); err != nil {
|
|
t.Fatalf("unable to connect carol to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointCarol := openChannelAndAssert(
|
|
ctxt, t, net, carol, dave,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointCarol)
|
|
|
|
carolChanTXID, err := lnd.GetChanPointFundingTxid(chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
carolFundPoint := wire.OutPoint{
|
|
Hash: *carolChanTXID,
|
|
Index: chanPointCarol.OutputIndex,
|
|
}
|
|
|
|
// Wait for all nodes to have seen all channels.
|
|
nodes := []*lntest.HarnessNode{net.Alice, net.Bob, carol, dave}
|
|
nodeNames := []string{"Alice", "Bob", "Carol", "Dave"}
|
|
for _, chanPoint := range networkChans {
|
|
for i, node := range nodes {
|
|
txid, err := lnd.GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
point := wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: chanPoint.OutputIndex,
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = node.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("%s(%d): timeout waiting for "+
|
|
"channel(%s) open: %v", nodeNames[i],
|
|
node.NodeID, point, err)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Create 5 invoices for Bob, which expect a payment from Carol for 1k
|
|
// satoshis with a different preimage each time.
|
|
const numPayments = 5
|
|
const paymentAmt = 1000
|
|
payReqs, _, _, err := createPayReqs(
|
|
net.Bob, paymentAmt, numPayments,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// We'll wait for all parties to recognize the new channels within the
|
|
// network.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = dave.WaitForNetworkChannelOpen(ctxt, chanPointDave)
|
|
if err != nil {
|
|
t.Fatalf("dave didn't advertise his channel: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("carol didn't advertise her channel in time: %v",
|
|
err)
|
|
}
|
|
|
|
time.Sleep(time.Millisecond * 50)
|
|
|
|
// Set the fee policies of the Alice -> Bob and the Dave -> Alice
|
|
// channel edges to relatively large non default values. This makes it
|
|
// possible to pick up more subtle fee calculation errors.
|
|
maxHtlc := uint64(calculateMaxHtlc(chanAmt))
|
|
updateChannelPolicy(
|
|
t, net.Alice, chanPointAlice, 1000, 100000,
|
|
lnd.DefaultBitcoinTimeLockDelta, maxHtlc, carol,
|
|
)
|
|
|
|
updateChannelPolicy(
|
|
t, dave, chanPointDave, 5000, 150000,
|
|
lnd.DefaultBitcoinTimeLockDelta, maxHtlc, carol,
|
|
)
|
|
|
|
// Using Carol as the source, pay to the 5 invoices from Bob created
|
|
// above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, carol, payReqs, true)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
// When asserting the amount of satoshis moved, we'll factor in the
|
|
// default base fee, as we didn't modify the fee structure when
|
|
// creating the seed nodes in the network.
|
|
const baseFee = 1
|
|
|
|
// At this point all the channels within our proto network should be
|
|
// shifted by 5k satoshis in the direction of Bob, the sink within the
|
|
// payment flow generated above. The order of asserts corresponds to
|
|
// increasing of time is needed to embed the HTLC in commitment
|
|
// transaction, in channel Carol->David->Alice->Bob, order is Bob,
|
|
// Alice, David, Carol.
|
|
|
|
// The final node bob expects to get paid five times 1000 sat.
|
|
expectedAmountPaidAtoB := int64(5 * 1000)
|
|
|
|
assertAmountPaid(t, "Alice(local) => Bob(remote)", net.Bob,
|
|
aliceFundPoint, int64(0), expectedAmountPaidAtoB)
|
|
assertAmountPaid(t, "Alice(local) => Bob(remote)", net.Alice,
|
|
aliceFundPoint, expectedAmountPaidAtoB, int64(0))
|
|
|
|
// To forward a payment of 1000 sat, Alice is charging a fee of
|
|
// 1 sat + 10% = 101 sat.
|
|
const expectedFeeAlice = 5 * 101
|
|
|
|
// Dave needs to pay what Alice pays plus Alice's fee.
|
|
expectedAmountPaidDtoA := expectedAmountPaidAtoB + expectedFeeAlice
|
|
|
|
assertAmountPaid(t, "Dave(local) => Alice(remote)", net.Alice,
|
|
daveFundPoint, int64(0), expectedAmountPaidDtoA)
|
|
assertAmountPaid(t, "Dave(local) => Alice(remote)", dave,
|
|
daveFundPoint, expectedAmountPaidDtoA, int64(0))
|
|
|
|
// To forward a payment of 1101 sat, Dave is charging a fee of
|
|
// 5 sat + 15% = 170.15 sat. This is rounded down in rpcserver to 170.
|
|
const expectedFeeDave = 5 * 170
|
|
|
|
// Carol needs to pay what Dave pays plus Dave's fee.
|
|
expectedAmountPaidCtoD := expectedAmountPaidDtoA + expectedFeeDave
|
|
|
|
assertAmountPaid(t, "Carol(local) => Dave(remote)", dave,
|
|
carolFundPoint, int64(0), expectedAmountPaidCtoD)
|
|
assertAmountPaid(t, "Carol(local) => Dave(remote)", carol,
|
|
carolFundPoint, expectedAmountPaidCtoD, int64(0))
|
|
|
|
// Now that we know all the balances have been settled out properly,
|
|
// we'll ensure that our internal record keeping for completed circuits
|
|
// was properly updated.
|
|
|
|
// First, check that the FeeReport response shows the proper fees
|
|
// accrued over each time range. Dave should've earned 170 satoshi for
|
|
// each of the forwarded payments.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
feeReport, err := dave.FeeReport(ctxt, &lnrpc.FeeReportRequest{})
|
|
if err != nil {
|
|
t.Fatalf("unable to query for fee report: %v", err)
|
|
}
|
|
|
|
if feeReport.DayFeeSum != uint64(expectedFeeDave) {
|
|
t.Fatalf("fee mismatch: expected %v, got %v", expectedFeeDave,
|
|
feeReport.DayFeeSum)
|
|
}
|
|
if feeReport.WeekFeeSum != uint64(expectedFeeDave) {
|
|
t.Fatalf("fee mismatch: expected %v, got %v", expectedFeeDave,
|
|
feeReport.WeekFeeSum)
|
|
}
|
|
if feeReport.MonthFeeSum != uint64(expectedFeeDave) {
|
|
t.Fatalf("fee mismatch: expected %v, got %v", expectedFeeDave,
|
|
feeReport.MonthFeeSum)
|
|
}
|
|
|
|
// Next, ensure that if we issue the vanilla query for the forwarding
|
|
// history, it returns 5 values, and each entry is formatted properly.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
fwdingHistory, err := dave.ForwardingHistory(
|
|
ctxt, &lnrpc.ForwardingHistoryRequest{},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to query for fee report: %v", err)
|
|
}
|
|
if len(fwdingHistory.ForwardingEvents) != 5 {
|
|
t.Fatalf("wrong number of forwarding event: expected %v, "+
|
|
"got %v", 5, len(fwdingHistory.ForwardingEvents))
|
|
}
|
|
expectedForwardingFee := uint64(expectedFeeDave / numPayments)
|
|
for _, event := range fwdingHistory.ForwardingEvents {
|
|
// Each event should show a fee of 170 satoshi.
|
|
if event.Fee != expectedForwardingFee {
|
|
t.Fatalf("fee mismatch: expected %v, got %v",
|
|
expectedForwardingFee, event.Fee)
|
|
}
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointAlice, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, dave, chanPointDave, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPointCarol, false)
|
|
}
|
|
|
|
type singleHopSendToRouteCase struct {
|
|
name string
|
|
|
|
// streaming tests streaming SendToRoute if true, otherwise tests
|
|
// synchronous SenToRoute.
|
|
streaming bool
|
|
|
|
// routerrpc submits the request to the routerrpc subserver if true,
|
|
// otherwise submits to the main rpc server.
|
|
routerrpc bool
|
|
|
|
// mpp sets the MPP fields on the request if true, otherwise submits a
|
|
// regular payment.
|
|
mpp bool
|
|
}
|
|
|
|
var singleHopSendToRouteCases = []singleHopSendToRouteCase{
|
|
{
|
|
name: "regular main sync",
|
|
},
|
|
{
|
|
name: "regular main stream",
|
|
streaming: true,
|
|
},
|
|
{
|
|
name: "regular routerrpc sync",
|
|
routerrpc: true,
|
|
},
|
|
{
|
|
name: "mpp main sync",
|
|
mpp: true,
|
|
},
|
|
{
|
|
name: "mpp main stream",
|
|
streaming: true,
|
|
mpp: true,
|
|
},
|
|
{
|
|
name: "mpp routerrpc sync",
|
|
routerrpc: true,
|
|
mpp: true,
|
|
},
|
|
}
|
|
|
|
// testSingleHopSendToRoute tests that payments are properly processed through a
|
|
// provided route with a single hop. We'll create the following network
|
|
// topology:
|
|
// Carol --100k--> Dave
|
|
// We'll query the daemon for routes from Carol to Dave and then send payments
|
|
// by feeding the route back into the various SendToRoute RPC methods. Here we
|
|
// test all three SendToRoute endpoints, forcing each to perform both a regular
|
|
// payment and an MPP payment.
|
|
func testSingleHopSendToRoute(net *lntest.NetworkHarness, t *harnessTest) {
|
|
for _, test := range singleHopSendToRouteCases {
|
|
test := test
|
|
|
|
t.t.Run(test.name, func(t1 *testing.T) {
|
|
ht := newHarnessTest(t1, t.lndHarness)
|
|
ht.RunTestCase(&testCase{
|
|
name: test.name,
|
|
test: func(_ *lntest.NetworkHarness, tt *harnessTest) {
|
|
testSingleHopSendToRouteCase(net, tt, test)
|
|
},
|
|
})
|
|
})
|
|
}
|
|
}
|
|
|
|
func testSingleHopSendToRouteCase(net *lntest.NetworkHarness, t *harnessTest,
|
|
test singleHopSendToRouteCase) {
|
|
|
|
const chanAmt = btcutil.Amount(100000)
|
|
const paymentAmtSat = 1000
|
|
const numPayments = 5
|
|
const amountPaid = int64(numPayments * paymentAmtSat)
|
|
|
|
ctxb := context.Background()
|
|
var networkChans []*lnrpc.ChannelPoint
|
|
|
|
// Create Carol and Dave, then establish a channel between them. Carol
|
|
// is the sole funder of the channel with 100k satoshis. The network
|
|
// topology should look like:
|
|
// Carol -> 100k -> Dave
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
dave, err := net.NewNode("Dave", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, dave); err != nil {
|
|
t.Fatalf("unable to connect carol to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
|
|
// Open a channel with 100k satoshis between Carol and Dave with Carol
|
|
// being the sole funder of the channel.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointCarol := openChannelAndAssert(
|
|
ctxt, t, net, carol, dave,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointCarol)
|
|
|
|
carolChanTXID, err := lnd.GetChanPointFundingTxid(chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
carolFundPoint := wire.OutPoint{
|
|
Hash: *carolChanTXID,
|
|
Index: chanPointCarol.OutputIndex,
|
|
}
|
|
|
|
// Wait for all nodes to have seen all channels.
|
|
nodes := []*lntest.HarnessNode{carol, dave}
|
|
for _, chanPoint := range networkChans {
|
|
for _, node := range nodes {
|
|
txid, err := lnd.GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
point := wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: chanPoint.OutputIndex,
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = node.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("%s(%d): timeout waiting for "+
|
|
"channel(%s) open: %v", node.Name(),
|
|
node.NodeID, point, err)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Create invoices for Dave, which expect a payment from Carol.
|
|
payReqs, rHashes, _, err := createPayReqs(
|
|
dave, paymentAmtSat, numPayments,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// Reconstruct payment addresses.
|
|
var payAddrs [][]byte
|
|
for _, payReq := range payReqs {
|
|
ctx, _ := context.WithTimeout(
|
|
context.Background(), defaultTimeout,
|
|
)
|
|
resp, err := dave.DecodePayReq(
|
|
ctx,
|
|
&lnrpc.PayReqString{PayReq: payReq},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("decode pay req: %v", err)
|
|
}
|
|
payAddrs = append(payAddrs, resp.PaymentAddr)
|
|
}
|
|
|
|
// Query for routes to pay from Carol to Dave.
|
|
// We set FinalCltvDelta to 40 since by default QueryRoutes returns
|
|
// the last hop with a final cltv delta of 9 where as the default in
|
|
// htlcswitch is 40.
|
|
routesReq := &lnrpc.QueryRoutesRequest{
|
|
PubKey: dave.PubKeyStr,
|
|
Amt: paymentAmtSat,
|
|
FinalCltvDelta: lnd.DefaultBitcoinTimeLockDelta,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
routes, err := carol.QueryRoutes(ctxt, routesReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get route from %s: %v",
|
|
carol.Name(), err)
|
|
}
|
|
|
|
// There should only be one route to try, so take the first item.
|
|
r := routes.Routes[0]
|
|
|
|
// Construct a closure that will set MPP fields on the route, which
|
|
// allows us to test MPP payments.
|
|
setMPPFields := func(i int) {
|
|
hop := r.Hops[len(r.Hops)-1]
|
|
hop.TlvPayload = true
|
|
hop.MppRecord = &lnrpc.MPPRecord{
|
|
PaymentAddr: payAddrs[i],
|
|
TotalAmtMsat: paymentAmtSat * 1000,
|
|
}
|
|
}
|
|
|
|
// Construct closures for each of the payment types covered:
|
|
// - main rpc server sync
|
|
// - main rpc server streaming
|
|
// - routerrpc server sync
|
|
sendToRouteSync := func() {
|
|
for i, rHash := range rHashes {
|
|
// Populate the MPP fields for the final hop if we are
|
|
// testing MPP payments.
|
|
if test.mpp {
|
|
setMPPFields(i)
|
|
}
|
|
|
|
sendReq := &lnrpc.SendToRouteRequest{
|
|
PaymentHash: rHash,
|
|
Route: r,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := carol.SendToRouteSync(
|
|
ctxt, sendReq,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to send to route for "+
|
|
"%s: %v", carol.Name(), err)
|
|
}
|
|
if resp.PaymentError != "" {
|
|
t.Fatalf("received payment error from %s: %v",
|
|
carol.Name(), resp.PaymentError)
|
|
}
|
|
}
|
|
}
|
|
sendToRouteStream := func() {
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
alicePayStream, err := carol.SendToRoute(ctxt)
|
|
if err != nil {
|
|
t.Fatalf("unable to create payment stream for "+
|
|
"carol: %v", err)
|
|
}
|
|
|
|
for i, rHash := range rHashes {
|
|
// Populate the MPP fields for the final hop if we are
|
|
// testing MPP payments.
|
|
if test.mpp {
|
|
setMPPFields(i)
|
|
}
|
|
|
|
sendReq := &lnrpc.SendToRouteRequest{
|
|
PaymentHash: rHash,
|
|
Route: routes.Routes[0],
|
|
}
|
|
err := alicePayStream.Send(sendReq)
|
|
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
|
|
resp, err := alicePayStream.Recv()
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
if resp.PaymentError != "" {
|
|
t.Fatalf("received payment error: %v",
|
|
resp.PaymentError)
|
|
}
|
|
}
|
|
}
|
|
sendToRouteRouterRPC := func() {
|
|
for i, rHash := range rHashes {
|
|
// Populate the MPP fields for the final hop if we are
|
|
// testing MPP payments.
|
|
if test.mpp {
|
|
setMPPFields(i)
|
|
}
|
|
|
|
sendReq := &routerrpc.SendToRouteRequest{
|
|
PaymentHash: rHash,
|
|
Route: r,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := carol.RouterClient.SendToRoute(
|
|
ctxt, sendReq,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to send to route for "+
|
|
"%s: %v", carol.Name(), err)
|
|
}
|
|
if resp.Failure != nil {
|
|
t.Fatalf("received payment error from %s: %v",
|
|
carol.Name(), resp.Failure)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Using Carol as the node as the source, send the payments
|
|
// synchronously via the the routerrpc's SendToRoute, or via the main RPC
|
|
// server's SendToRoute streaming or sync calls.
|
|
switch {
|
|
case !test.routerrpc && test.streaming:
|
|
sendToRouteStream()
|
|
case !test.routerrpc && !test.streaming:
|
|
sendToRouteSync()
|
|
case test.routerrpc && !test.streaming:
|
|
sendToRouteRouterRPC()
|
|
default:
|
|
t.Fatalf("routerrpc does not support streaming send_to_route")
|
|
}
|
|
|
|
// Verify that the payment's from Carol's PoV have the correct payment
|
|
// hash and amount.
|
|
ctxt, _ = context.WithTimeout(ctxt, defaultTimeout)
|
|
paymentsResp, err := carol.ListPayments(
|
|
ctxt, &lnrpc.ListPaymentsRequest{},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("error when obtaining %s payments: %v",
|
|
carol.Name(), err)
|
|
}
|
|
if len(paymentsResp.Payments) != numPayments {
|
|
t.Fatalf("incorrect number of payments, got %v, want %v",
|
|
len(paymentsResp.Payments), numPayments)
|
|
}
|
|
|
|
for i, p := range paymentsResp.Payments {
|
|
// Assert that the payment hashes for each payment match up.
|
|
rHashHex := hex.EncodeToString(rHashes[i])
|
|
if p.PaymentHash != rHashHex {
|
|
t.Fatalf("incorrect payment hash for payment %d, "+
|
|
"want: %s got: %s",
|
|
i, rHashHex, p.PaymentHash)
|
|
}
|
|
|
|
// Assert that each payment has no invoice since the payment was
|
|
// completed using SendToRoute.
|
|
if p.PaymentRequest != "" {
|
|
t.Fatalf("incorrect payment request for payment: %d, "+
|
|
"want: \"\", got: %s",
|
|
i, p.PaymentRequest)
|
|
}
|
|
|
|
// Assert the payment ammount is correct.
|
|
if p.ValueSat != paymentAmtSat {
|
|
t.Fatalf("incorrect payment amt for payment %d, "+
|
|
"want: %d, got: %d",
|
|
i, paymentAmtSat, p.ValueSat)
|
|
}
|
|
|
|
// Assert exactly one htlc was made.
|
|
if len(p.Htlcs) != 1 {
|
|
t.Fatalf("expected 1 htlc for payment %d, got: %d",
|
|
i, len(p.Htlcs))
|
|
}
|
|
|
|
// Assert the htlc's route is populated.
|
|
htlc := p.Htlcs[0]
|
|
if htlc.Route == nil {
|
|
t.Fatalf("expected route for payment %d", i)
|
|
}
|
|
|
|
// Assert the hop has exactly one hop.
|
|
if len(htlc.Route.Hops) != 1 {
|
|
t.Fatalf("expected 1 hop for payment %d, got: %d",
|
|
i, len(htlc.Route.Hops))
|
|
}
|
|
|
|
// If this is an MPP test, assert the MPP record's fields are
|
|
// properly populated. Otherwise the hop should not have an MPP
|
|
// record.
|
|
hop := htlc.Route.Hops[0]
|
|
if test.mpp {
|
|
if hop.MppRecord == nil {
|
|
t.Fatalf("expected mpp record for mpp payment")
|
|
}
|
|
|
|
if hop.MppRecord.TotalAmtMsat != paymentAmtSat*1000 {
|
|
t.Fatalf("incorrect mpp total msat for payment %d "+
|
|
"want: %d, got: %d",
|
|
i, paymentAmtSat*1000,
|
|
hop.MppRecord.TotalAmtMsat)
|
|
}
|
|
|
|
expAddr := payAddrs[i]
|
|
if !bytes.Equal(hop.MppRecord.PaymentAddr, expAddr) {
|
|
t.Fatalf("incorrect mpp payment addr for payment %d "+
|
|
"want: %x, got: %x",
|
|
i, expAddr, hop.MppRecord.PaymentAddr)
|
|
}
|
|
} else if hop.MppRecord != nil {
|
|
t.Fatalf("unexpected mpp record for non-mpp payment")
|
|
}
|
|
}
|
|
|
|
// Verify that the invoices's from Dave's PoV have the correct payment
|
|
// hash and amount.
|
|
ctxt, _ = context.WithTimeout(ctxt, defaultTimeout)
|
|
invoicesResp, err := dave.ListInvoices(
|
|
ctxt, &lnrpc.ListInvoiceRequest{},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("error when obtaining %s payments: %v",
|
|
dave.Name(), err)
|
|
}
|
|
if len(invoicesResp.Invoices) != numPayments {
|
|
t.Fatalf("incorrect number of invoices, got %v, want %v",
|
|
len(invoicesResp.Invoices), numPayments)
|
|
}
|
|
|
|
for i, inv := range invoicesResp.Invoices {
|
|
// Assert that the payment hashes match up.
|
|
if !bytes.Equal(inv.RHash, rHashes[i]) {
|
|
t.Fatalf("incorrect payment hash for invoice %d, "+
|
|
"want: %x got: %x",
|
|
i, rHashes[i], inv.RHash)
|
|
}
|
|
|
|
// Assert that the amount paid to the invoice is correct.
|
|
if inv.AmtPaidSat != paymentAmtSat {
|
|
t.Fatalf("incorrect payment amt for invoice %d, "+
|
|
"want: %d, got %d",
|
|
i, paymentAmtSat, inv.AmtPaidSat)
|
|
}
|
|
}
|
|
|
|
// At this point all the channels within our proto network should be
|
|
// shifted by 5k satoshis in the direction of Dave, the sink within the
|
|
// payment flow generated above. The order of asserts corresponds to
|
|
// increasing of time is needed to embed the HTLC in commitment
|
|
// transaction, in channel Carol->Dave, order is Dave and then Carol.
|
|
assertAmountPaid(t, "Carol(local) => Dave(remote)", dave,
|
|
carolFundPoint, int64(0), amountPaid)
|
|
assertAmountPaid(t, "Carol(local) => Dave(remote)", carol,
|
|
carolFundPoint, amountPaid, int64(0))
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPointCarol, false)
|
|
}
|
|
|
|
// testMultiHopSendToRoute tests that payments are properly processed
|
|
// through a provided route. We'll create the following network topology:
|
|
// Alice --100k--> Bob --100k--> Carol
|
|
// We'll query the daemon for routes from Alice to Carol and then
|
|
// send payments through the routes.
|
|
func testMultiHopSendToRoute(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const chanAmt = btcutil.Amount(100000)
|
|
var networkChans []*lnrpc.ChannelPoint
|
|
|
|
// Open a channel with 100k satoshis between Alice and Bob with Alice
|
|
// being the sole funder of the channel.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAlice := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointAlice)
|
|
|
|
aliceChanTXID, err := lnd.GetChanPointFundingTxid(chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
aliceFundPoint := wire.OutPoint{
|
|
Hash: *aliceChanTXID,
|
|
Index: chanPointAlice.OutputIndex,
|
|
}
|
|
|
|
// Create Carol and establish a channel from Bob. Bob is the sole funder
|
|
// of the channel with 100k satoshis. The network topology should look like:
|
|
// Alice -> Bob -> Carol
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, net.Bob); err != nil {
|
|
t.Fatalf("unable to connect carol to alice: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, net.Bob)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to bob: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointBob := openChannelAndAssert(
|
|
ctxt, t, net, net.Bob, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointBob)
|
|
bobChanTXID, err := lnd.GetChanPointFundingTxid(chanPointBob)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
bobFundPoint := wire.OutPoint{
|
|
Hash: *bobChanTXID,
|
|
Index: chanPointBob.OutputIndex,
|
|
}
|
|
|
|
// Wait for all nodes to have seen all channels.
|
|
nodes := []*lntest.HarnessNode{net.Alice, net.Bob, carol}
|
|
nodeNames := []string{"Alice", "Bob", "Carol"}
|
|
for _, chanPoint := range networkChans {
|
|
for i, node := range nodes {
|
|
txid, err := lnd.GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
point := wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: chanPoint.OutputIndex,
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = node.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("%s(%d): timeout waiting for "+
|
|
"channel(%s) open: %v", nodeNames[i],
|
|
node.NodeID, point, err)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Query for routes to pay from Alice to Carol.
|
|
// We set FinalCltvDelta to 40 since by default QueryRoutes returns
|
|
// the last hop with a final cltv delta of 9 where as the default in
|
|
// htlcswitch is 40.
|
|
const paymentAmt = 1000
|
|
routesReq := &lnrpc.QueryRoutesRequest{
|
|
PubKey: carol.PubKeyStr,
|
|
Amt: paymentAmt,
|
|
FinalCltvDelta: lnd.DefaultBitcoinTimeLockDelta,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
routes, err := net.Alice.QueryRoutes(ctxt, routesReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get route: %v", err)
|
|
}
|
|
|
|
// Create 5 invoices for Carol, which expect a payment from Alice for 1k
|
|
// satoshis with a different preimage each time.
|
|
const numPayments = 5
|
|
_, rHashes, _, err := createPayReqs(
|
|
carol, paymentAmt, numPayments,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// We'll wait for all parties to recognize the new channels within the
|
|
// network.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPointBob)
|
|
if err != nil {
|
|
t.Fatalf("bob didn't advertise his channel in time: %v", err)
|
|
}
|
|
|
|
time.Sleep(time.Millisecond * 50)
|
|
|
|
// Using Alice as the source, pay to the 5 invoices from Carol created
|
|
// above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
alicePayStream, err := net.Alice.SendToRoute(ctxt)
|
|
if err != nil {
|
|
t.Fatalf("unable to create payment stream for alice: %v", err)
|
|
}
|
|
|
|
for _, rHash := range rHashes {
|
|
sendReq := &lnrpc.SendToRouteRequest{
|
|
PaymentHash: rHash,
|
|
Route: routes.Routes[0],
|
|
}
|
|
err := alicePayStream.Send(sendReq)
|
|
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
}
|
|
|
|
for range rHashes {
|
|
resp, err := alicePayStream.Recv()
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
if resp.PaymentError != "" {
|
|
t.Fatalf("received payment error: %v", resp.PaymentError)
|
|
}
|
|
}
|
|
|
|
// When asserting the amount of satoshis moved, we'll factor in the
|
|
// default base fee, as we didn't modify the fee structure when
|
|
// creating the seed nodes in the network.
|
|
const baseFee = 1
|
|
|
|
// At this point all the channels within our proto network should be
|
|
// shifted by 5k satoshis in the direction of Carol, the sink within the
|
|
// payment flow generated above. The order of asserts corresponds to
|
|
// increasing of time is needed to embed the HTLC in commitment
|
|
// transaction, in channel Alice->Bob->Carol, order is Carol, Bob,
|
|
// Alice.
|
|
const amountPaid = int64(5000)
|
|
assertAmountPaid(t, "Bob(local) => Carol(remote)", carol,
|
|
bobFundPoint, int64(0), amountPaid)
|
|
assertAmountPaid(t, "Bob(local) => Carol(remote)", net.Bob,
|
|
bobFundPoint, amountPaid, int64(0))
|
|
assertAmountPaid(t, "Alice(local) => Bob(remote)", net.Bob,
|
|
aliceFundPoint, int64(0), amountPaid+(baseFee*numPayments))
|
|
assertAmountPaid(t, "Alice(local) => Bob(remote)", net.Alice,
|
|
aliceFundPoint, amountPaid+(baseFee*numPayments), int64(0))
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointAlice, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPointBob, false)
|
|
}
|
|
|
|
// testSendToRouteErrorPropagation tests propagation of errors that occur
|
|
// while processing a multi-hop payment through an unknown route.
|
|
func testSendToRouteErrorPropagation(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const chanAmt = btcutil.Amount(100000)
|
|
|
|
// Open a channel with 100k satoshis between Alice and Bob with Alice
|
|
// being the sole funder of the channel.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAlice := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err := net.Alice.WaitForNetworkChannelOpen(ctxt, chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't advertise her channel: %v", err)
|
|
}
|
|
|
|
// Create a new nodes (Carol and Charlie), load her with some funds,
|
|
// then establish a connection between Carol and Charlie with a channel
|
|
// that has identical capacity to the one created above.Then we will
|
|
// get route via queryroutes call which will be fake route for Alice ->
|
|
// Bob graph.
|
|
//
|
|
// The network topology should now look like: Alice -> Bob; Carol -> Charlie.
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
|
|
charlie, err := net.NewNode("Charlie", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, charlie)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, charlie)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to charlie: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, charlie); err != nil {
|
|
t.Fatalf("unable to connect carol to alice: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointCarol := openChannelAndAssert(
|
|
ctxt, t, net, carol, charlie,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("carol didn't advertise her channel: %v", err)
|
|
}
|
|
|
|
// Query routes from Carol to Charlie which will be an invalid route
|
|
// for Alice -> Bob.
|
|
fakeReq := &lnrpc.QueryRoutesRequest{
|
|
PubKey: charlie.PubKeyStr,
|
|
Amt: int64(1),
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
fakeRoute, err := carol.QueryRoutes(ctxt, fakeReq)
|
|
if err != nil {
|
|
t.Fatalf("unable get fake route: %v", err)
|
|
}
|
|
|
|
// Create 1 invoices for Bob, which expect a payment from Alice for 1k
|
|
// satoshis
|
|
const paymentAmt = 1000
|
|
|
|
invoice := &lnrpc.Invoice{
|
|
Memo: "testing",
|
|
Value: paymentAmt,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := net.Bob.AddInvoice(ctxt, invoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
|
|
rHash := resp.RHash
|
|
|
|
// Using Alice as the source, pay to the 5 invoices from Bob created above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
alicePayStream, err := net.Alice.SendToRoute(ctxt)
|
|
if err != nil {
|
|
t.Fatalf("unable to create payment stream for alice: %v", err)
|
|
}
|
|
|
|
sendReq := &lnrpc.SendToRouteRequest{
|
|
PaymentHash: rHash,
|
|
Route: fakeRoute.Routes[0],
|
|
}
|
|
|
|
if err := alicePayStream.Send(sendReq); err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
|
|
// At this place we should get an rpc error with notification
|
|
// that edge is not found on hop(0)
|
|
if _, err := alicePayStream.Recv(); err != nil && strings.Contains(err.Error(),
|
|
"edge not found") {
|
|
|
|
} else if err != nil {
|
|
t.Fatalf("payment stream has been closed but fake route has consumed: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointAlice, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPointCarol, false)
|
|
}
|
|
|
|
// testUnannouncedChannels checks unannounced channels are not returned by
|
|
// describeGraph RPC request unless explicitly asked for.
|
|
func testUnannouncedChannels(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
amount := lnd.MaxBtcFundingAmount
|
|
|
|
// Open a channel between Alice and Bob, ensuring the
|
|
// channel has been opened properly.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanOpenUpdate := openChannelStream(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: amount,
|
|
},
|
|
)
|
|
|
|
// Mine 2 blocks, and check that the channel is opened but not yet
|
|
// announced to the network.
|
|
mineBlocks(t, net, 2, 1)
|
|
|
|
// One block is enough to make the channel ready for use, since the
|
|
// nodes have defaultNumConfs=1 set.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
fundingChanPoint, err := net.WaitForChannelOpen(ctxt, chanOpenUpdate)
|
|
if err != nil {
|
|
t.Fatalf("error while waiting for channel open: %v", err)
|
|
}
|
|
|
|
// Alice should have 1 edge in her graph.
|
|
req := &lnrpc.ChannelGraphRequest{
|
|
IncludeUnannounced: true,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
chanGraph, err := net.Alice.DescribeGraph(ctxt, req)
|
|
if err != nil {
|
|
t.Fatalf("unable to query alice's graph: %v", err)
|
|
}
|
|
|
|
numEdges := len(chanGraph.Edges)
|
|
if numEdges != 1 {
|
|
t.Fatalf("expected to find 1 edge in the graph, found %d", numEdges)
|
|
}
|
|
|
|
// Channels should not be announced yet, hence Alice should have no
|
|
// announced edges in her graph.
|
|
req.IncludeUnannounced = false
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
chanGraph, err = net.Alice.DescribeGraph(ctxt, req)
|
|
if err != nil {
|
|
t.Fatalf("unable to query alice's graph: %v", err)
|
|
}
|
|
|
|
numEdges = len(chanGraph.Edges)
|
|
if numEdges != 0 {
|
|
t.Fatalf("expected to find 0 announced edges in the graph, found %d",
|
|
numEdges)
|
|
}
|
|
|
|
// Mine 4 more blocks, and check that the channel is now announced.
|
|
mineBlocks(t, net, 4, 0)
|
|
|
|
// Give the network a chance to learn that auth proof is confirmed.
|
|
var predErr error
|
|
err = wait.Predicate(func() bool {
|
|
// The channel should now be announced. Check that Alice has 1
|
|
// announced edge.
|
|
req.IncludeUnannounced = false
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
chanGraph, err = net.Alice.DescribeGraph(ctxt, req)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to query alice's graph: %v", err)
|
|
return false
|
|
}
|
|
|
|
numEdges = len(chanGraph.Edges)
|
|
if numEdges != 1 {
|
|
predErr = fmt.Errorf("expected to find 1 announced edge in "+
|
|
"the graph, found %d", numEdges)
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("%v", predErr)
|
|
}
|
|
|
|
// The channel should now be announced. Check that Alice has 1 announced
|
|
// edge.
|
|
req.IncludeUnannounced = false
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
chanGraph, err = net.Alice.DescribeGraph(ctxt, req)
|
|
if err != nil {
|
|
t.Fatalf("unable to query alice's graph: %v", err)
|
|
}
|
|
|
|
numEdges = len(chanGraph.Edges)
|
|
if numEdges != 1 {
|
|
t.Fatalf("expected to find 1 announced edge in the graph, found %d",
|
|
numEdges)
|
|
}
|
|
|
|
// Close the channel used during the test.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, fundingChanPoint, false)
|
|
}
|
|
|
|
// testPrivateChannels tests that a private channel can be used for
|
|
// routing by the two endpoints of the channel, but is not known by
|
|
// the rest of the nodes in the graph.
|
|
func testPrivateChannels(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const chanAmt = btcutil.Amount(100000)
|
|
var networkChans []*lnrpc.ChannelPoint
|
|
|
|
// We create the following topology:
|
|
//
|
|
// Dave --100k--> Alice --200k--> Bob
|
|
// ^ ^
|
|
// | |
|
|
// 100k 100k
|
|
// | |
|
|
// +---- Carol ----+
|
|
//
|
|
// where the 100k channel between Carol and Alice is private.
|
|
|
|
// Open a channel with 200k satoshis between Alice and Bob.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAlice := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt * 2,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointAlice)
|
|
|
|
aliceChanTXID, err := lnd.GetChanPointFundingTxid(chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
aliceFundPoint := wire.OutPoint{
|
|
Hash: *aliceChanTXID,
|
|
Index: chanPointAlice.OutputIndex,
|
|
}
|
|
|
|
// Create Dave, and a channel to Alice of 100k.
|
|
dave, err := net.NewNode("Dave", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, dave, net.Alice); err != nil {
|
|
t.Fatalf("unable to connect dave to alice: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointDave := openChannelAndAssert(
|
|
ctxt, t, net, dave, net.Alice,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointDave)
|
|
daveChanTXID, err := lnd.GetChanPointFundingTxid(chanPointDave)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
daveFundPoint := wire.OutPoint{
|
|
Hash: *daveChanTXID,
|
|
Index: chanPointDave.OutputIndex,
|
|
}
|
|
|
|
// Next, we'll create Carol and establish a channel from her to
|
|
// Dave of 100k.
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, dave); err != nil {
|
|
t.Fatalf("unable to connect carol to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointCarol := openChannelAndAssert(
|
|
ctxt, t, net, carol, dave,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointCarol)
|
|
|
|
carolChanTXID, err := lnd.GetChanPointFundingTxid(chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
carolFundPoint := wire.OutPoint{
|
|
Hash: *carolChanTXID,
|
|
Index: chanPointCarol.OutputIndex,
|
|
}
|
|
|
|
// Wait for all nodes to have seen all these channels, as they
|
|
// are all public.
|
|
nodes := []*lntest.HarnessNode{net.Alice, net.Bob, carol, dave}
|
|
nodeNames := []string{"Alice", "Bob", "Carol", "Dave"}
|
|
for _, chanPoint := range networkChans {
|
|
for i, node := range nodes {
|
|
txid, err := lnd.GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
point := wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: chanPoint.OutputIndex,
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = node.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("%s(%d): timeout waiting for "+
|
|
"channel(%s) open: %v", nodeNames[i],
|
|
node.NodeID, point, err)
|
|
}
|
|
}
|
|
}
|
|
// Now create a _private_ channel directly between Carol and
|
|
// Alice of 100k.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, net.Alice); err != nil {
|
|
t.Fatalf("unable to connect dave to alice: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanOpenUpdate := openChannelStream(
|
|
ctxt, t, net, carol, net.Alice,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
Private: true,
|
|
},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to open channel: %v", err)
|
|
}
|
|
|
|
// One block is enough to make the channel ready for use, since the
|
|
// nodes have defaultNumConfs=1 set.
|
|
block := mineBlocks(t, net, 1, 1)[0]
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
chanPointPrivate, err := net.WaitForChannelOpen(ctxt, chanOpenUpdate)
|
|
if err != nil {
|
|
t.Fatalf("error while waiting for channel open: %v", err)
|
|
}
|
|
fundingTxID, err := lnd.GetChanPointFundingTxid(chanPointPrivate)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
assertTxInBlock(t, block, fundingTxID)
|
|
|
|
// The channel should be listed in the peer information returned by
|
|
// both peers.
|
|
privateFundPoint := wire.OutPoint{
|
|
Hash: *fundingTxID,
|
|
Index: chanPointPrivate.OutputIndex,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.AssertChannelExists(ctxt, carol, &privateFundPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to assert channel existence: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.AssertChannelExists(ctxt, net.Alice, &privateFundPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to assert channel existence: %v", err)
|
|
}
|
|
|
|
// The channel should be available for payments between Carol and Alice.
|
|
// We check this by sending payments from Carol to Bob, that
|
|
// collectively would deplete at least one of Carol's channels.
|
|
|
|
// Create 2 invoices for Bob, each of 70k satoshis. Since each of
|
|
// Carol's channels is of size 100k, these payments cannot succeed
|
|
// by only using one of the channels.
|
|
const numPayments = 2
|
|
const paymentAmt = 70000
|
|
payReqs, _, _, err := createPayReqs(
|
|
net.Bob, paymentAmt, numPayments,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
time.Sleep(time.Millisecond * 50)
|
|
|
|
// Let Carol pay the invoices.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, carol, payReqs, true)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
// When asserting the amount of satoshis moved, we'll factor in the
|
|
// default base fee, as we didn't modify the fee structure when
|
|
// creating the seed nodes in the network.
|
|
const baseFee = 1
|
|
|
|
// Bob should have received 140k satoshis from Alice.
|
|
assertAmountPaid(t, "Alice(local) => Bob(remote)", net.Bob,
|
|
aliceFundPoint, int64(0), 2*paymentAmt)
|
|
|
|
// Alice sent 140k to Bob.
|
|
assertAmountPaid(t, "Alice(local) => Bob(remote)", net.Alice,
|
|
aliceFundPoint, 2*paymentAmt, int64(0))
|
|
|
|
// Alice received 70k + fee from Dave.
|
|
assertAmountPaid(t, "Dave(local) => Alice(remote)", net.Alice,
|
|
daveFundPoint, int64(0), paymentAmt+baseFee)
|
|
|
|
// Dave sent 70k+fee to Alice.
|
|
assertAmountPaid(t, "Dave(local) => Alice(remote)", dave,
|
|
daveFundPoint, paymentAmt+baseFee, int64(0))
|
|
|
|
// Dave received 70k+fee of two hops from Carol.
|
|
assertAmountPaid(t, "Carol(local) => Dave(remote)", dave,
|
|
carolFundPoint, int64(0), paymentAmt+baseFee*2)
|
|
|
|
// Carol sent 70k+fee of two hops to Dave.
|
|
assertAmountPaid(t, "Carol(local) => Dave(remote)", carol,
|
|
carolFundPoint, paymentAmt+baseFee*2, int64(0))
|
|
|
|
// Alice received 70k+fee from Carol.
|
|
assertAmountPaid(t, "Carol(local) [private=>] Alice(remote)",
|
|
net.Alice, privateFundPoint, int64(0), paymentAmt+baseFee)
|
|
|
|
// Carol sent 70k+fee to Alice.
|
|
assertAmountPaid(t, "Carol(local) [private=>] Alice(remote)",
|
|
carol, privateFundPoint, paymentAmt+baseFee, int64(0))
|
|
|
|
// Alice should also be able to route payments using this channel,
|
|
// so send two payments of 60k back to Carol.
|
|
const paymentAmt60k = 60000
|
|
payReqs, _, _, err = createPayReqs(
|
|
carol, paymentAmt60k, numPayments,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
time.Sleep(time.Millisecond * 50)
|
|
|
|
// Let Bob pay the invoices.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, net.Alice, payReqs, true)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
// Finally, we make sure Dave and Bob does not know about the
|
|
// private channel between Carol and Alice. We first mine
|
|
// plenty of blocks, such that the channel would have been
|
|
// announced in case it was public.
|
|
mineBlocks(t, net, 10, 0)
|
|
|
|
// We create a helper method to check how many edges each of the
|
|
// nodes know about. Carol and Alice should know about 4, while
|
|
// Bob and Dave should only know about 3, since one channel is
|
|
// private.
|
|
numChannels := func(node *lntest.HarnessNode, includeUnannounced bool) int {
|
|
req := &lnrpc.ChannelGraphRequest{
|
|
IncludeUnannounced: includeUnannounced,
|
|
}
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
chanGraph, err := node.DescribeGraph(ctxt, req)
|
|
if err != nil {
|
|
t.Fatalf("unable go describegraph: %v", err)
|
|
}
|
|
return len(chanGraph.Edges)
|
|
}
|
|
|
|
var predErr error
|
|
err = wait.Predicate(func() bool {
|
|
aliceChans := numChannels(net.Alice, true)
|
|
if aliceChans != 4 {
|
|
predErr = fmt.Errorf("expected Alice to know 4 edges, "+
|
|
"had %v", aliceChans)
|
|
return false
|
|
}
|
|
alicePubChans := numChannels(net.Alice, false)
|
|
if alicePubChans != 3 {
|
|
predErr = fmt.Errorf("expected Alice to know 3 public edges, "+
|
|
"had %v", alicePubChans)
|
|
return false
|
|
}
|
|
bobChans := numChannels(net.Bob, true)
|
|
if bobChans != 3 {
|
|
predErr = fmt.Errorf("expected Bob to know 3 edges, "+
|
|
"had %v", bobChans)
|
|
return false
|
|
}
|
|
carolChans := numChannels(carol, true)
|
|
if carolChans != 4 {
|
|
predErr = fmt.Errorf("expected Carol to know 4 edges, "+
|
|
"had %v", carolChans)
|
|
return false
|
|
}
|
|
carolPubChans := numChannels(carol, false)
|
|
if carolPubChans != 3 {
|
|
predErr = fmt.Errorf("expected Carol to know 3 public edges, "+
|
|
"had %v", carolPubChans)
|
|
return false
|
|
}
|
|
daveChans := numChannels(dave, true)
|
|
if daveChans != 3 {
|
|
predErr = fmt.Errorf("expected Dave to know 3 edges, "+
|
|
"had %v", daveChans)
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("%v", predErr)
|
|
}
|
|
|
|
// Close all channels.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointAlice, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, dave, chanPointDave, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPointCarol, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPointPrivate, false)
|
|
}
|
|
|
|
// testInvoiceRoutingHints tests that the routing hints for an invoice are
|
|
// created properly.
|
|
func testInvoiceRoutingHints(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const chanAmt = btcutil.Amount(100000)
|
|
|
|
// Throughout this test, we'll be opening a channel between Alice and
|
|
// several other parties.
|
|
//
|
|
// First, we'll create a private channel between Alice and Bob. This
|
|
// will be the only channel that will be considered as a routing hint
|
|
// throughout this test. We'll include a push amount since we currently
|
|
// require channels to have enough remote balance to cover the invoice's
|
|
// payment.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointBob := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: chanAmt / 2,
|
|
Private: true,
|
|
},
|
|
)
|
|
|
|
// Then, we'll create Carol's node and open a public channel between her
|
|
// and Alice. This channel will not be considered as a routing hint due
|
|
// to it being public.
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create carol's node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, net.Alice, carol); err != nil {
|
|
t.Fatalf("unable to connect alice to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointCarol := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: chanAmt / 2,
|
|
},
|
|
)
|
|
|
|
// We'll also create a public channel between Bob and Carol to ensure
|
|
// that Bob gets selected as the only routing hint. We do this as
|
|
// we should only include routing hints for nodes that are publicly
|
|
// advertised, otherwise we'd end up leaking information about nodes
|
|
// that wish to stay unadvertised.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, net.Bob, carol); err != nil {
|
|
t.Fatalf("unable to connect alice to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointBobCarol := openChannelAndAssert(
|
|
ctxt, t, net, net.Bob, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: chanAmt / 2,
|
|
},
|
|
)
|
|
|
|
// Then, we'll create Dave's node and open a private channel between him
|
|
// and Alice. We will not include a push amount in order to not consider
|
|
// this channel as a routing hint as it will not have enough remote
|
|
// balance for the invoice's amount.
|
|
dave, err := net.NewNode("Dave", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create dave's node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, net.Alice, dave); err != nil {
|
|
t.Fatalf("unable to connect alice to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointDave := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, dave,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
Private: true,
|
|
},
|
|
)
|
|
|
|
// Finally, we'll create Eve's node and open a private channel between
|
|
// her and Alice. This time though, we'll take Eve's node down after the
|
|
// channel has been created to avoid populating routing hints for
|
|
// inactive channels.
|
|
eve, err := net.NewNode("Eve", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create eve's node: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, net.Alice, eve); err != nil {
|
|
t.Fatalf("unable to connect alice to eve: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointEve := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, eve,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: chanAmt / 2,
|
|
Private: true,
|
|
},
|
|
)
|
|
|
|
// Make sure all the channels have been opened.
|
|
nodeNames := []string{"bob", "carol", "dave", "eve"}
|
|
aliceChans := []*lnrpc.ChannelPoint{
|
|
chanPointBob, chanPointCarol, chanPointBobCarol, chanPointDave,
|
|
chanPointEve,
|
|
}
|
|
for i, chanPoint := range aliceChans {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Alice.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("timed out waiting for channel open between "+
|
|
"alice and %s: %v", nodeNames[i], err)
|
|
}
|
|
}
|
|
|
|
// Now that the channels are open, we'll take down Eve's node.
|
|
shutdownAndAssert(net, t, eve)
|
|
|
|
// Create an invoice for Alice that will populate the routing hints.
|
|
invoice := &lnrpc.Invoice{
|
|
Memo: "routing hints",
|
|
Value: int64(chanAmt / 4),
|
|
Private: true,
|
|
}
|
|
|
|
// Due to the way the channels were set up above, the channel between
|
|
// Alice and Bob should be the only channel used as a routing hint.
|
|
var predErr error
|
|
var decoded *lnrpc.PayReq
|
|
err = wait.Predicate(func() bool {
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := net.Alice.AddInvoice(ctxt, invoice)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to add invoice: %v", err)
|
|
return false
|
|
}
|
|
|
|
// We'll decode the invoice's payment request to determine which
|
|
// channels were used as routing hints.
|
|
payReq := &lnrpc.PayReqString{
|
|
PayReq: resp.PaymentRequest,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
decoded, err = net.Alice.DecodePayReq(ctxt, payReq)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to decode payment "+
|
|
"request: %v", err)
|
|
return false
|
|
}
|
|
|
|
if len(decoded.RouteHints) != 1 {
|
|
predErr = fmt.Errorf("expected one route hint, got %d",
|
|
len(decoded.RouteHints))
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf(predErr.Error())
|
|
}
|
|
|
|
hops := decoded.RouteHints[0].HopHints
|
|
if len(hops) != 1 {
|
|
t.Fatalf("expected one hop in route hint, got %d", len(hops))
|
|
}
|
|
chanID := hops[0].ChanId
|
|
|
|
// We'll need the short channel ID of the channel between Alice and Bob
|
|
// to make sure the routing hint is for this channel.
|
|
listReq := &lnrpc.ListChannelsRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
listResp, err := net.Alice.ListChannels(ctxt, listReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to retrieve alice's channels: %v", err)
|
|
}
|
|
|
|
var aliceBobChanID uint64
|
|
for _, channel := range listResp.Channels {
|
|
if channel.RemotePubkey == net.Bob.PubKeyStr {
|
|
aliceBobChanID = channel.ChanId
|
|
}
|
|
}
|
|
|
|
if aliceBobChanID == 0 {
|
|
t.Fatalf("channel between alice and bob not found")
|
|
}
|
|
|
|
if chanID != aliceBobChanID {
|
|
t.Fatalf("expected channel ID %d, got %d", aliceBobChanID,
|
|
chanID)
|
|
}
|
|
|
|
// Now that we've confirmed the routing hints were added correctly, we
|
|
// can close all the channels and shut down all the nodes created.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointBob, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointCarol, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Bob, chanPointBobCarol, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointDave, false)
|
|
|
|
// The channel between Alice and Eve should be force closed since Eve
|
|
// is offline.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointEve, true)
|
|
|
|
// Cleanup by mining the force close and sweep transaction.
|
|
cleanupForceClose(t, net, net.Alice, chanPointEve)
|
|
}
|
|
|
|
// testMultiHopOverPrivateChannels tests that private channels can be used as
|
|
// intermediate hops in a route for payments.
|
|
func testMultiHopOverPrivateChannels(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
// We'll test that multi-hop payments over private channels work as
|
|
// intended. To do so, we'll create the following topology:
|
|
// private public private
|
|
// Alice <--100k--> Bob <--100k--> Carol <--100k--> Dave
|
|
const chanAmt = btcutil.Amount(100000)
|
|
|
|
// First, we'll open a private channel between Alice and Bob with Alice
|
|
// being the funder.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAlice := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
Private: true,
|
|
},
|
|
)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err := net.Alice.WaitForNetworkChannelOpen(ctxt, chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't see the channel alice <-> bob before "+
|
|
"timeout: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Bob.WaitForNetworkChannelOpen(ctxt, chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("bob didn't see the channel alice <-> bob before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// Retrieve Alice's funding outpoint.
|
|
aliceChanTXID, err := lnd.GetChanPointFundingTxid(chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
aliceFundPoint := wire.OutPoint{
|
|
Hash: *aliceChanTXID,
|
|
Index: chanPointAlice.OutputIndex,
|
|
}
|
|
|
|
// Next, we'll create Carol's node and open a public channel between
|
|
// her and Bob with Bob being the funder.
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create carol's node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, net.Bob, carol); err != nil {
|
|
t.Fatalf("unable to connect bob to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointBob := openChannelAndAssert(
|
|
ctxt, t, net, net.Bob, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Bob.WaitForNetworkChannelOpen(ctxt, chanPointBob)
|
|
if err != nil {
|
|
t.Fatalf("bob didn't see the channel bob <-> carol before "+
|
|
"timeout: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPointBob)
|
|
if err != nil {
|
|
t.Fatalf("carol didn't see the channel bob <-> carol before "+
|
|
"timeout: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Alice.WaitForNetworkChannelOpen(ctxt, chanPointBob)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't see the channel bob <-> carol before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// Retrieve Bob's funding outpoint.
|
|
bobChanTXID, err := lnd.GetChanPointFundingTxid(chanPointBob)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
bobFundPoint := wire.OutPoint{
|
|
Hash: *bobChanTXID,
|
|
Index: chanPointBob.OutputIndex,
|
|
}
|
|
|
|
// Next, we'll create Dave's node and open a private channel between him
|
|
// and Carol with Carol being the funder.
|
|
dave, err := net.NewNode("Dave", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create dave's node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, dave); err != nil {
|
|
t.Fatalf("unable to connect carol to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointCarol := openChannelAndAssert(
|
|
ctxt, t, net, carol, dave,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
Private: true,
|
|
},
|
|
)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("carol didn't see the channel carol <-> dave before "+
|
|
"timeout: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = dave.WaitForNetworkChannelOpen(ctxt, chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("dave didn't see the channel carol <-> dave before "+
|
|
"timeout: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = dave.WaitForNetworkChannelOpen(ctxt, chanPointBob)
|
|
if err != nil {
|
|
t.Fatalf("dave didn't see the channel bob <-> carol before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// Retrieve Carol's funding point.
|
|
carolChanTXID, err := lnd.GetChanPointFundingTxid(chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
carolFundPoint := wire.OutPoint{
|
|
Hash: *carolChanTXID,
|
|
Index: chanPointCarol.OutputIndex,
|
|
}
|
|
|
|
// Now that all the channels are set up according to the topology from
|
|
// above, we can proceed to test payments. We'll create an invoice for
|
|
// Dave of 20k satoshis and pay it with Alice. Since there is no public
|
|
// route from Alice to Dave, we'll need to use the private channel
|
|
// between Carol and Dave as a routing hint encoded in the invoice.
|
|
const paymentAmt = 20000
|
|
|
|
// Create the invoice for Dave.
|
|
invoice := &lnrpc.Invoice{
|
|
Memo: "two hopz!",
|
|
Value: paymentAmt,
|
|
Private: true,
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := dave.AddInvoice(ctxt, invoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice for dave: %v", err)
|
|
}
|
|
|
|
// Let Alice pay the invoice.
|
|
payReqs := []string{resp.PaymentRequest}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, net.Alice, payReqs, true)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments from alice to dave: %v", err)
|
|
}
|
|
|
|
// When asserting the amount of satoshis moved, we'll factor in the
|
|
// default base fee, as we didn't modify the fee structure when opening
|
|
// the channels.
|
|
const baseFee = 1
|
|
|
|
// Dave should have received 20k satoshis from Carol.
|
|
assertAmountPaid(t, "Carol(local) [private=>] Dave(remote)",
|
|
dave, carolFundPoint, 0, paymentAmt)
|
|
|
|
// Carol should have sent 20k satoshis to Dave.
|
|
assertAmountPaid(t, "Carol(local) [private=>] Dave(remote)",
|
|
carol, carolFundPoint, paymentAmt, 0)
|
|
|
|
// Carol should have received 20k satoshis + fee for one hop from Bob.
|
|
assertAmountPaid(t, "Bob(local) => Carol(remote)",
|
|
carol, bobFundPoint, 0, paymentAmt+baseFee)
|
|
|
|
// Bob should have sent 20k satoshis + fee for one hop to Carol.
|
|
assertAmountPaid(t, "Bob(local) => Carol(remote)",
|
|
net.Bob, bobFundPoint, paymentAmt+baseFee, 0)
|
|
|
|
// Bob should have received 20k satoshis + fee for two hops from Alice.
|
|
assertAmountPaid(t, "Alice(local) [private=>] Bob(remote)", net.Bob,
|
|
aliceFundPoint, 0, paymentAmt+baseFee*2)
|
|
|
|
// Alice should have sent 20k satoshis + fee for two hops to Bob.
|
|
assertAmountPaid(t, "Alice(local) [private=>] Bob(remote)", net.Alice,
|
|
aliceFundPoint, paymentAmt+baseFee*2, 0)
|
|
|
|
// At this point, the payment was successful. We can now close all the
|
|
// channels and shutdown the nodes created throughout this test.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointAlice, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Bob, chanPointBob, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPointCarol, false)
|
|
}
|
|
|
|
func testInvoiceSubscriptions(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const chanAmt = btcutil.Amount(500000)
|
|
|
|
// Open a channel with 500k satoshis between Alice and Bob with Alice
|
|
// being the sole funder of the channel.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Next create a new invoice for Bob requesting 1k satoshis.
|
|
// TODO(roasbeef): make global list of invoices for each node to re-use
|
|
// and avoid collisions
|
|
const paymentAmt = 1000
|
|
invoice := &lnrpc.Invoice{
|
|
Memo: "testing",
|
|
RPreimage: makeFakePayHash(t),
|
|
Value: paymentAmt,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
invoiceResp, err := net.Bob.AddInvoice(ctxt, invoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
lastAddIndex := invoiceResp.AddIndex
|
|
|
|
// Create a new invoice subscription client for Bob, the notification
|
|
// should be dispatched shortly below.
|
|
req := &lnrpc.InvoiceSubscription{}
|
|
ctx, cancelInvoiceSubscription := context.WithCancel(ctxb)
|
|
bobInvoiceSubscription, err := net.Bob.SubscribeInvoices(ctx, req)
|
|
if err != nil {
|
|
t.Fatalf("unable to subscribe to bob's invoice updates: %v", err)
|
|
}
|
|
|
|
var settleIndex uint64
|
|
quit := make(chan struct{})
|
|
updateSent := make(chan struct{})
|
|
go func() {
|
|
invoiceUpdate, err := bobInvoiceSubscription.Recv()
|
|
select {
|
|
case <-quit:
|
|
// Received cancellation
|
|
return
|
|
default:
|
|
}
|
|
|
|
if err != nil {
|
|
t.Fatalf("unable to recv invoice update: %v", err)
|
|
}
|
|
|
|
// The invoice update should exactly match the invoice created
|
|
// above, but should now be settled and have SettleDate
|
|
if !invoiceUpdate.Settled {
|
|
t.Fatalf("invoice not settled but should be")
|
|
}
|
|
if invoiceUpdate.SettleDate == 0 {
|
|
t.Fatalf("invoice should have non zero settle date, but doesn't")
|
|
}
|
|
|
|
if !bytes.Equal(invoiceUpdate.RPreimage, invoice.RPreimage) {
|
|
t.Fatalf("payment preimages don't match: expected %v, got %v",
|
|
invoice.RPreimage, invoiceUpdate.RPreimage)
|
|
}
|
|
|
|
if invoiceUpdate.SettleIndex == 0 {
|
|
t.Fatalf("invoice should have settle index")
|
|
}
|
|
|
|
settleIndex = invoiceUpdate.SettleIndex
|
|
|
|
close(updateSent)
|
|
}()
|
|
|
|
// Wait for the channel to be recognized by both Alice and Bob before
|
|
// continuing the rest of the test.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Alice.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
// TODO(roasbeef): will need to make num blocks to advertise a
|
|
// node param
|
|
close(quit)
|
|
t.Fatalf("channel not seen by alice before timeout: %v", err)
|
|
}
|
|
|
|
// With the assertion above set up, send a payment from Alice to Bob
|
|
// which should finalize and settle the invoice.
|
|
sendReq := &lnrpc.SendRequest{
|
|
PaymentRequest: invoiceResp.PaymentRequest,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := net.Alice.SendPaymentSync(ctxt, sendReq)
|
|
if err != nil {
|
|
close(quit)
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
if resp.PaymentError != "" {
|
|
close(quit)
|
|
t.Fatalf("error when attempting recv: %v", resp.PaymentError)
|
|
}
|
|
|
|
select {
|
|
case <-time.After(time.Second * 10):
|
|
close(quit)
|
|
t.Fatalf("update not sent after 10 seconds")
|
|
case <-updateSent: // Fall through on success
|
|
}
|
|
|
|
// With the base case working, we'll now cancel Bob's current
|
|
// subscription in order to exercise the backlog fill behavior.
|
|
cancelInvoiceSubscription()
|
|
|
|
// We'll now add 3 more invoices to Bob's invoice registry.
|
|
const numInvoices = 3
|
|
payReqs, _, newInvoices, err := createPayReqs(
|
|
net.Bob, paymentAmt, numInvoices,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// Now that the set of invoices has been added, we'll re-register for
|
|
// streaming invoice notifications for Bob, this time specifying the
|
|
// add invoice of the last prior invoice.
|
|
req = &lnrpc.InvoiceSubscription{
|
|
AddIndex: lastAddIndex,
|
|
}
|
|
ctx, cancelInvoiceSubscription = context.WithCancel(ctxb)
|
|
bobInvoiceSubscription, err = net.Bob.SubscribeInvoices(ctx, req)
|
|
if err != nil {
|
|
t.Fatalf("unable to subscribe to bob's invoice updates: %v", err)
|
|
}
|
|
|
|
// Since we specified a value of the prior add index above, we should
|
|
// now immediately get the invoices we just added as we should get the
|
|
// backlog of notifications.
|
|
for i := 0; i < numInvoices; i++ {
|
|
invoiceUpdate, err := bobInvoiceSubscription.Recv()
|
|
if err != nil {
|
|
t.Fatalf("unable to receive subscription")
|
|
}
|
|
|
|
// We should now get the ith invoice we added, as they should
|
|
// be returned in order.
|
|
if invoiceUpdate.Settled {
|
|
t.Fatalf("should have only received add events")
|
|
}
|
|
originalInvoice := newInvoices[i]
|
|
rHash := sha256.Sum256(originalInvoice.RPreimage[:])
|
|
if !bytes.Equal(invoiceUpdate.RHash, rHash[:]) {
|
|
t.Fatalf("invoices have mismatched payment hashes: "+
|
|
"expected %x, got %x", rHash[:],
|
|
invoiceUpdate.RHash)
|
|
}
|
|
}
|
|
|
|
cancelInvoiceSubscription()
|
|
|
|
// We'll now have Bob settle out the remainder of these invoices so we
|
|
// can test that all settled invoices are properly notified.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(
|
|
ctxt, net.Alice, payReqs, true,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
|
|
// With the set of invoices paid, we'll now cancel the old
|
|
// subscription, and create a new one for Bob, this time using the
|
|
// settle index to obtain the backlog of settled invoices.
|
|
req = &lnrpc.InvoiceSubscription{
|
|
SettleIndex: settleIndex,
|
|
}
|
|
ctx, cancelInvoiceSubscription = context.WithCancel(ctxb)
|
|
bobInvoiceSubscription, err = net.Bob.SubscribeInvoices(ctx, req)
|
|
if err != nil {
|
|
t.Fatalf("unable to subscribe to bob's invoice updates: %v", err)
|
|
}
|
|
|
|
defer cancelInvoiceSubscription()
|
|
|
|
// As we specified the index of the past settle index, we should now
|
|
// receive notifications for the three HTLCs that we just settled. As
|
|
// the order that the HTLCs will be settled in is partially randomized,
|
|
// we'll use a map to assert that the proper set has been settled.
|
|
settledInvoices := make(map[[32]byte]struct{})
|
|
for _, invoice := range newInvoices {
|
|
rHash := sha256.Sum256(invoice.RPreimage[:])
|
|
settledInvoices[rHash] = struct{}{}
|
|
}
|
|
for i := 0; i < numInvoices; i++ {
|
|
invoiceUpdate, err := bobInvoiceSubscription.Recv()
|
|
if err != nil {
|
|
t.Fatalf("unable to receive subscription")
|
|
}
|
|
|
|
// We should now get the ith invoice we added, as they should
|
|
// be returned in order.
|
|
if !invoiceUpdate.Settled {
|
|
t.Fatalf("should have only received settle events")
|
|
}
|
|
|
|
var rHash [32]byte
|
|
copy(rHash[:], invoiceUpdate.RHash)
|
|
if _, ok := settledInvoices[rHash]; !ok {
|
|
t.Fatalf("unknown invoice settled: %x", rHash)
|
|
}
|
|
|
|
delete(settledInvoices, rHash)
|
|
}
|
|
|
|
// At this point, all the invoices should be fully settled.
|
|
if len(settledInvoices) != 0 {
|
|
t.Fatalf("not all invoices settled")
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPoint, false)
|
|
}
|
|
|
|
// channelSubscription houses the proxied update and error chans for a node's
|
|
// channel subscriptions.
|
|
type channelSubscription struct {
|
|
updateChan chan *lnrpc.ChannelEventUpdate
|
|
errChan chan error
|
|
quit chan struct{}
|
|
}
|
|
|
|
// subscribeChannelNotifications subscribes to channel updates and launches a
|
|
// goroutine that forwards these to the returned channel.
|
|
func subscribeChannelNotifications(ctxb context.Context, t *harnessTest,
|
|
node *lntest.HarnessNode) channelSubscription {
|
|
|
|
// We'll first start by establishing a notification client which will
|
|
// send us notifications upon channels becoming active, inactive or
|
|
// closed.
|
|
req := &lnrpc.ChannelEventSubscription{}
|
|
ctx, cancelFunc := context.WithCancel(ctxb)
|
|
|
|
chanUpdateClient, err := node.SubscribeChannelEvents(ctx, req)
|
|
if err != nil {
|
|
t.Fatalf("unable to create channel update client: %v", err)
|
|
}
|
|
|
|
// We'll launch a goroutine that will be responsible for proxying all
|
|
// notifications recv'd from the client into the channel below.
|
|
errChan := make(chan error, 1)
|
|
quit := make(chan struct{})
|
|
chanUpdates := make(chan *lnrpc.ChannelEventUpdate, 20)
|
|
go func() {
|
|
defer cancelFunc()
|
|
for {
|
|
select {
|
|
case <-quit:
|
|
return
|
|
default:
|
|
chanUpdate, err := chanUpdateClient.Recv()
|
|
select {
|
|
case <-quit:
|
|
return
|
|
default:
|
|
}
|
|
|
|
if err == io.EOF {
|
|
return
|
|
} else if err != nil {
|
|
select {
|
|
case errChan <- err:
|
|
case <-quit:
|
|
}
|
|
return
|
|
}
|
|
|
|
select {
|
|
case chanUpdates <- chanUpdate:
|
|
case <-quit:
|
|
return
|
|
}
|
|
}
|
|
}
|
|
}()
|
|
|
|
return channelSubscription{
|
|
updateChan: chanUpdates,
|
|
errChan: errChan,
|
|
quit: quit,
|
|
}
|
|
}
|
|
|
|
// verifyCloseUpdate is used to verify that a closed channel update is of the
|
|
// expected type.
|
|
func verifyCloseUpdate(chanUpdate *lnrpc.ChannelEventUpdate,
|
|
closeType lnrpc.ChannelCloseSummary_ClosureType,
|
|
closeInitiator lnrpc.Initiator) error {
|
|
|
|
// We should receive one inactive and one closed notification
|
|
// for each channel.
|
|
switch update := chanUpdate.Channel.(type) {
|
|
case *lnrpc.ChannelEventUpdate_InactiveChannel:
|
|
if chanUpdate.Type != lnrpc.ChannelEventUpdate_INACTIVE_CHANNEL {
|
|
return fmt.Errorf("update type mismatch: expected %v, got %v",
|
|
lnrpc.ChannelEventUpdate_INACTIVE_CHANNEL,
|
|
chanUpdate.Type)
|
|
}
|
|
case *lnrpc.ChannelEventUpdate_ClosedChannel:
|
|
if chanUpdate.Type !=
|
|
lnrpc.ChannelEventUpdate_CLOSED_CHANNEL {
|
|
return fmt.Errorf("update type mismatch: expected %v, got %v",
|
|
lnrpc.ChannelEventUpdate_CLOSED_CHANNEL,
|
|
chanUpdate.Type)
|
|
}
|
|
|
|
if update.ClosedChannel.CloseType != closeType {
|
|
return fmt.Errorf("channel closure type "+
|
|
"mismatch: expected %v, got %v",
|
|
closeType,
|
|
update.ClosedChannel.CloseType)
|
|
}
|
|
|
|
if update.ClosedChannel.CloseInitiator != closeInitiator {
|
|
return fmt.Errorf("expected close intiator: %v, got: %v",
|
|
closeInitiator,
|
|
update.ClosedChannel.CloseInitiator)
|
|
}
|
|
|
|
default:
|
|
return fmt.Errorf("channel update channel of wrong type, "+
|
|
"expected closed channel, got %T",
|
|
update)
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// testBasicChannelCreationAndUpdates tests multiple channel opening and closing,
|
|
// and ensures that if a node is subscribed to channel updates they will be
|
|
// received correctly for both cooperative and force closed channels.
|
|
func testBasicChannelCreationAndUpdates(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
const (
|
|
numChannels = 2
|
|
amount = lnd.MaxBtcFundingAmount
|
|
)
|
|
|
|
// Subscribe Bob and Alice to channel event notifications.
|
|
bobChanSub := subscribeChannelNotifications(ctxb, t, net.Bob)
|
|
defer close(bobChanSub.quit)
|
|
|
|
aliceChanSub := subscribeChannelNotifications(ctxb, t, net.Alice)
|
|
defer close(aliceChanSub.quit)
|
|
|
|
// Open the channel between Alice and Bob, asserting that the
|
|
// channel has been properly open on-chain.
|
|
chanPoints := make([]*lnrpc.ChannelPoint, numChannels)
|
|
for i := 0; i < numChannels; i++ {
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoints[i] = openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: amount,
|
|
},
|
|
)
|
|
}
|
|
|
|
// Since each of the channels just became open, Bob and Alice should
|
|
// each receive an open and an active notification for each channel.
|
|
var numChannelUpds int
|
|
const totalNtfns = 3 * numChannels
|
|
verifyOpenUpdatesReceived := func(sub channelSubscription) error {
|
|
numChannelUpds = 0
|
|
for numChannelUpds < totalNtfns {
|
|
select {
|
|
case update := <-sub.updateChan:
|
|
switch update.Type {
|
|
case lnrpc.ChannelEventUpdate_PENDING_OPEN_CHANNEL:
|
|
if numChannelUpds%3 != 0 {
|
|
return fmt.Errorf("expected " +
|
|
"open or active" +
|
|
"channel ntfn, got pending open " +
|
|
"channel ntfn instead")
|
|
}
|
|
case lnrpc.ChannelEventUpdate_OPEN_CHANNEL:
|
|
if numChannelUpds%3 != 1 {
|
|
return fmt.Errorf("expected " +
|
|
"pending open or active" +
|
|
"channel ntfn, got open" +
|
|
"channel ntfn instead")
|
|
}
|
|
case lnrpc.ChannelEventUpdate_ACTIVE_CHANNEL:
|
|
if numChannelUpds%3 != 2 {
|
|
return fmt.Errorf("expected " +
|
|
"pending open or open" +
|
|
"channel ntfn, got active " +
|
|
"channel ntfn instead")
|
|
}
|
|
default:
|
|
return fmt.Errorf("update type mismatch: "+
|
|
"expected open or active channel "+
|
|
"notification, got: %v",
|
|
update.Type)
|
|
}
|
|
numChannelUpds++
|
|
case <-time.After(time.Second * 10):
|
|
return fmt.Errorf("timeout waiting for channel "+
|
|
"notifications, only received %d/%d "+
|
|
"chanupds", numChannelUpds,
|
|
totalNtfns)
|
|
}
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
if err := verifyOpenUpdatesReceived(bobChanSub); err != nil {
|
|
t.Fatalf("error verifying open updates: %v", err)
|
|
}
|
|
if err := verifyOpenUpdatesReceived(aliceChanSub); err != nil {
|
|
t.Fatalf("error verifying open updates: %v", err)
|
|
}
|
|
|
|
// Close the channel between Alice and Bob, asserting that the channel
|
|
// has been properly closed on-chain.
|
|
for i, chanPoint := range chanPoints {
|
|
ctx, _ := context.WithTimeout(context.Background(), defaultTimeout)
|
|
|
|
// Force close half of the channels.
|
|
force := i%2 == 0
|
|
closeChannelAndAssert(ctx, t, net, net.Alice, chanPoint, force)
|
|
if force {
|
|
cleanupForceClose(t, net, net.Alice, chanPoint)
|
|
}
|
|
}
|
|
|
|
// verifyCloseUpdatesReceived is used to verify that Alice and Bob
|
|
// receive the correct channel updates in order.
|
|
verifyCloseUpdatesReceived := func(sub channelSubscription,
|
|
forceType lnrpc.ChannelCloseSummary_ClosureType,
|
|
closeInitiator lnrpc.Initiator) error {
|
|
|
|
// Ensure one inactive and one closed notification is received for each
|
|
// closed channel.
|
|
numChannelUpds := 0
|
|
for numChannelUpds < 2*numChannels {
|
|
expectedCloseType := lnrpc.ChannelCloseSummary_COOPERATIVE_CLOSE
|
|
|
|
// Every other channel should be force closed. If this
|
|
// channel was force closed, set the expected close type
|
|
// the the type passed in.
|
|
force := (numChannelUpds/2)%2 == 0
|
|
if force {
|
|
expectedCloseType = forceType
|
|
}
|
|
|
|
select {
|
|
case chanUpdate := <-sub.updateChan:
|
|
err := verifyCloseUpdate(
|
|
chanUpdate, expectedCloseType,
|
|
closeInitiator,
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
numChannelUpds++
|
|
case err := <-sub.errChan:
|
|
return err
|
|
case <-time.After(time.Second * 10):
|
|
return fmt.Errorf("timeout waiting "+
|
|
"for channel notifications, only "+
|
|
"received %d/%d chanupds",
|
|
numChannelUpds, 2*numChannels)
|
|
}
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// Verify Bob receives all closed channel notifications. He should
|
|
// receive a remote force close notification for force closed channels.
|
|
// All channels (cooperatively and force closed) should have a remote
|
|
// close initiator because Alice closed the channels.
|
|
if err := verifyCloseUpdatesReceived(bobChanSub,
|
|
lnrpc.ChannelCloseSummary_REMOTE_FORCE_CLOSE,
|
|
lnrpc.Initiator_INITIATOR_REMOTE); err != nil {
|
|
t.Fatalf("errored verifying close updates: %v", err)
|
|
}
|
|
|
|
// Verify Alice receives all closed channel notifications. She should
|
|
// receive a remote force close notification for force closed channels.
|
|
// All channels (cooperatively and force closed) should have a local
|
|
// close initiator because Alice closed the channels.
|
|
if err := verifyCloseUpdatesReceived(aliceChanSub,
|
|
lnrpc.ChannelCloseSummary_LOCAL_FORCE_CLOSE,
|
|
lnrpc.Initiator_INITIATOR_LOCAL); err != nil {
|
|
t.Fatalf("errored verifying close updates: %v", err)
|
|
}
|
|
}
|
|
|
|
// testMaxPendingChannels checks that error is returned from remote peer if
|
|
// max pending channel number was exceeded and that '--maxpendingchannels' flag
|
|
// exists and works properly.
|
|
func testMaxPendingChannels(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
maxPendingChannels := lnd.DefaultMaxPendingChannels + 1
|
|
amount := lnd.MaxBtcFundingAmount
|
|
|
|
// Create a new node (Carol) with greater number of max pending
|
|
// channels.
|
|
args := []string{
|
|
fmt.Sprintf("--maxpendingchannels=%v", maxPendingChannels),
|
|
}
|
|
carol, err := net.NewNode("Carol", args)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, net.Alice, carol); err != nil {
|
|
t.Fatalf("unable to connect carol to alice: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolBalance := btcutil.Amount(maxPendingChannels) * amount
|
|
if err := net.SendCoins(ctxt, carolBalance, carol); err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
|
|
// Send open channel requests without generating new blocks thereby
|
|
// increasing pool of pending channels. Then check that we can't open
|
|
// the channel if the number of pending channels exceed max value.
|
|
openStreams := make([]lnrpc.Lightning_OpenChannelClient, maxPendingChannels)
|
|
for i := 0; i < maxPendingChannels; i++ {
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
stream := openChannelStream(
|
|
ctxt, t, net, net.Alice, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: amount,
|
|
},
|
|
)
|
|
openStreams[i] = stream
|
|
}
|
|
|
|
// Carol exhausted available amount of pending channels, next open
|
|
// channel request should cause ErrorGeneric to be sent back to Alice.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
_, err = net.OpenChannel(
|
|
ctxt, net.Alice, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: amount,
|
|
},
|
|
)
|
|
|
|
if err == nil {
|
|
t.Fatalf("error wasn't received")
|
|
} else if !strings.Contains(
|
|
err.Error(), lnwire.ErrMaxPendingChannels.Error(),
|
|
) {
|
|
t.Fatalf("not expected error was received: %v", err)
|
|
}
|
|
|
|
// For now our channels are in pending state, in order to not interfere
|
|
// with other tests we should clean up - complete opening of the
|
|
// channel and then close it.
|
|
|
|
// Mine 6 blocks, then wait for node's to notify us that the channel has
|
|
// been opened. The funding transactions should be found within the
|
|
// first newly mined block. 6 blocks make sure the funding transaction
|
|
// has enough confirmations to be announced publicly.
|
|
block := mineBlocks(t, net, 6, maxPendingChannels)[0]
|
|
|
|
chanPoints := make([]*lnrpc.ChannelPoint, maxPendingChannels)
|
|
for i, stream := range openStreams {
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
fundingChanPoint, err := net.WaitForChannelOpen(ctxt, stream)
|
|
if err != nil {
|
|
t.Fatalf("error while waiting for channel open: %v", err)
|
|
}
|
|
|
|
fundingTxID, err := lnd.GetChanPointFundingTxid(fundingChanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
|
|
// Ensure that the funding transaction enters a block, and is
|
|
// properly advertised by Alice.
|
|
assertTxInBlock(t, block, fundingTxID)
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Alice.WaitForNetworkChannelOpen(ctxt, fundingChanPoint)
|
|
if err != nil {
|
|
t.Fatalf("channel not seen on network before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// The channel should be listed in the peer information
|
|
// returned by both peers.
|
|
chanPoint := wire.OutPoint{
|
|
Hash: *fundingTxID,
|
|
Index: fundingChanPoint.OutputIndex,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.AssertChannelExists(ctxt, net.Alice, &chanPoint); err != nil {
|
|
t.Fatalf("unable to assert channel existence: %v", err)
|
|
}
|
|
|
|
chanPoints[i] = fundingChanPoint
|
|
}
|
|
|
|
// Next, close the channel between Alice and Carol, asserting that the
|
|
// channel has been properly closed on-chain.
|
|
for _, chanPoint := range chanPoints {
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPoint, false)
|
|
}
|
|
}
|
|
|
|
// waitForTxInMempool polls until finding one transaction in the provided
|
|
// miner's mempool. An error is returned if *one* transaction isn't found within
|
|
// the given timeout.
|
|
func waitForTxInMempool(miner *rpcclient.Client,
|
|
timeout time.Duration) (*chainhash.Hash, error) {
|
|
|
|
txs, err := waitForNTxsInMempool(miner, 1, timeout)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
return txs[0], err
|
|
}
|
|
|
|
// waitForNTxsInMempool polls until finding the desired number of transactions
|
|
// in the provided miner's mempool. An error is returned if this number is not
|
|
// met after the given timeout.
|
|
func waitForNTxsInMempool(miner *rpcclient.Client, n int,
|
|
timeout time.Duration) ([]*chainhash.Hash, error) {
|
|
|
|
breakTimeout := time.After(timeout)
|
|
ticker := time.NewTicker(50 * time.Millisecond)
|
|
defer ticker.Stop()
|
|
|
|
var err error
|
|
var mempool []*chainhash.Hash
|
|
for {
|
|
select {
|
|
case <-breakTimeout:
|
|
return nil, fmt.Errorf("wanted %v, found %v txs "+
|
|
"in mempool: %v", n, len(mempool), mempool)
|
|
case <-ticker.C:
|
|
mempool, err = miner.GetRawMempool()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
if len(mempool) == n {
|
|
return mempool, nil
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// getNTxsFromMempool polls until finding the desired number of transactions in
|
|
// the provided miner's mempool and returns the full transactions to the caller.
|
|
func getNTxsFromMempool(miner *rpcclient.Client, n int,
|
|
timeout time.Duration) ([]*wire.MsgTx, error) {
|
|
|
|
txids, err := waitForNTxsInMempool(miner, n, timeout)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
var txes []*wire.MsgTx
|
|
for _, txid := range txids {
|
|
tx, err := miner.GetRawTransaction(txid)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
txes = append(txes, tx.MsgTx())
|
|
}
|
|
return txes, nil
|
|
}
|
|
|
|
// testFailingChannel tests that we will fail the channel by force closing ii
|
|
// in the case where a counterparty tries to settle an HTLC with the wrong
|
|
// preimage.
|
|
func testFailingChannel(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const (
|
|
paymentAmt = 10000
|
|
)
|
|
|
|
chanAmt := lnd.MaxFundingAmount
|
|
|
|
// We'll introduce Carol, which will settle any incoming invoice with a
|
|
// totally unrelated preimage.
|
|
carol, err := net.NewNode("Carol", []string{"--hodl.bogus-settle"})
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
// Let Alice connect and open a channel to Carol,
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, net.Alice, carol); err != nil {
|
|
t.Fatalf("unable to connect alice to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// With the channel open, we'll create a invoice for Carol that Alice
|
|
// will attempt to pay.
|
|
preimage := bytes.Repeat([]byte{byte(192)}, 32)
|
|
invoice := &lnrpc.Invoice{
|
|
Memo: "testing",
|
|
RPreimage: preimage,
|
|
Value: paymentAmt,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := carol.AddInvoice(ctxt, invoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
carolPayReqs := []string{resp.PaymentRequest}
|
|
|
|
// Wait for Alice to receive the channel edge from the funding manager.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Alice.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't see the alice->carol channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// Send the payment from Alice to Carol. We expect Carol to attempt to
|
|
// settle this payment with the wrong preimage.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, net.Alice, carolPayReqs, false)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
// Since Alice detects that Carol is trying to trick her by providing a
|
|
// fake preimage, she should fail and force close the channel.
|
|
var predErr error
|
|
err = wait.Predicate(func() bool {
|
|
pendingChansRequest := &lnrpc.PendingChannelsRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
pendingChanResp, err := net.Alice.PendingChannels(ctxt,
|
|
pendingChansRequest)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to query for pending "+
|
|
"channels: %v", err)
|
|
return false
|
|
}
|
|
n := len(pendingChanResp.WaitingCloseChannels)
|
|
if n != 1 {
|
|
predErr = fmt.Errorf("Expected to find %d channels "+
|
|
"waiting close, found %d", 1, n)
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("%v", predErr)
|
|
}
|
|
|
|
// Mine a block to confirm the broadcasted commitment.
|
|
block := mineBlocks(t, net, 1, 1)[0]
|
|
if len(block.Transactions) != 2 {
|
|
t.Fatalf("transaction wasn't mined")
|
|
}
|
|
|
|
// The channel should now show up as force closed both for Alice and
|
|
// Carol.
|
|
err = wait.Predicate(func() bool {
|
|
pendingChansRequest := &lnrpc.PendingChannelsRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
pendingChanResp, err := net.Alice.PendingChannels(ctxt,
|
|
pendingChansRequest)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to query for pending "+
|
|
"channels: %v", err)
|
|
return false
|
|
}
|
|
n := len(pendingChanResp.WaitingCloseChannels)
|
|
if n != 0 {
|
|
predErr = fmt.Errorf("Expected to find %d channels "+
|
|
"waiting close, found %d", 0, n)
|
|
return false
|
|
}
|
|
n = len(pendingChanResp.PendingForceClosingChannels)
|
|
if n != 1 {
|
|
predErr = fmt.Errorf("expected to find %d channel "+
|
|
"pending force close, found %d", 1, n)
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("%v", predErr)
|
|
}
|
|
|
|
err = wait.Predicate(func() bool {
|
|
pendingChansRequest := &lnrpc.PendingChannelsRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
pendingChanResp, err := carol.PendingChannels(ctxt,
|
|
pendingChansRequest)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to query for pending "+
|
|
"channels: %v", err)
|
|
return false
|
|
}
|
|
n := len(pendingChanResp.PendingForceClosingChannels)
|
|
if n != 1 {
|
|
predErr = fmt.Errorf("expected to find %d channel "+
|
|
"pending force close, found %d", 1, n)
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("%v", predErr)
|
|
}
|
|
|
|
// Carol will use the correct preimage to resolve the HTLC on-chain.
|
|
_, err = waitForTxInMempool(net.Miner.Node, minerMempoolTimeout)
|
|
if err != nil {
|
|
t.Fatalf("unable to find Carol's resolve tx in mempool: %v", err)
|
|
}
|
|
|
|
// Mine enough blocks for Alice to sweep her funds from the force
|
|
// closed channel.
|
|
_, err = net.Miner.Node.Generate(defaultCSV)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate blocks: %v", err)
|
|
}
|
|
|
|
// Wait for the sweeping tx to be broadcast.
|
|
_, err = waitForTxInMempool(net.Miner.Node, minerMempoolTimeout)
|
|
if err != nil {
|
|
t.Fatalf("unable to find Alice's sweep tx in mempool: %v", err)
|
|
}
|
|
|
|
// Mine the sweep.
|
|
_, err = net.Miner.Node.Generate(1)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate blocks: %v", err)
|
|
}
|
|
|
|
// No pending channels should be left.
|
|
err = wait.Predicate(func() bool {
|
|
pendingChansRequest := &lnrpc.PendingChannelsRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
pendingChanResp, err := net.Alice.PendingChannels(ctxt,
|
|
pendingChansRequest)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to query for pending "+
|
|
"channels: %v", err)
|
|
return false
|
|
}
|
|
n := len(pendingChanResp.PendingForceClosingChannels)
|
|
if n != 0 {
|
|
predErr = fmt.Errorf("expected to find %d channel "+
|
|
"pending force close, found %d", 0, n)
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("%v", predErr)
|
|
}
|
|
}
|
|
|
|
// testGarbageCollectLinkNodes tests that we properly garbase collect link nodes
|
|
// from the database and the set of persistent connections within the server.
|
|
func testGarbageCollectLinkNodes(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const (
|
|
chanAmt = 1000000
|
|
)
|
|
|
|
// Open a channel between Alice and Bob which will later be
|
|
// cooperatively closed.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
coopChanPoint := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Create Carol's node and connect Alice to her.
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create carol's node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, net.Alice, carol); err != nil {
|
|
t.Fatalf("unable to connect alice and carol: %v", err)
|
|
}
|
|
|
|
// Open a channel between Alice and Carol which will later be force
|
|
// closed.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
forceCloseChanPoint := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Now, create Dave's a node and also open a channel between Alice and
|
|
// him. This link will serve as the only persistent link throughout
|
|
// restarts in this test.
|
|
dave, err := net.NewNode("Dave", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create dave's node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
if err := net.ConnectNodes(ctxt, net.Alice, dave); err != nil {
|
|
t.Fatalf("unable to connect alice to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
persistentChanPoint := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, dave,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// isConnected is a helper closure that checks if a peer is connected to
|
|
// Alice.
|
|
isConnected := func(pubKey string) bool {
|
|
req := &lnrpc.ListPeersRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := net.Alice.ListPeers(ctxt, req)
|
|
if err != nil {
|
|
t.Fatalf("unable to retrieve alice's peers: %v", err)
|
|
}
|
|
|
|
for _, peer := range resp.Peers {
|
|
if peer.PubKey == pubKey {
|
|
return true
|
|
}
|
|
}
|
|
|
|
return false
|
|
}
|
|
|
|
// Restart both Bob and Carol to ensure Alice is able to reconnect to
|
|
// them.
|
|
if err := net.RestartNode(net.Bob, nil); err != nil {
|
|
t.Fatalf("unable to restart bob's node: %v", err)
|
|
}
|
|
if err := net.RestartNode(carol, nil); err != nil {
|
|
t.Fatalf("unable to restart carol's node: %v", err)
|
|
}
|
|
|
|
err = wait.Predicate(func() bool {
|
|
return isConnected(net.Bob.PubKeyStr)
|
|
}, 15*time.Second)
|
|
if err != nil {
|
|
t.Fatalf("alice did not reconnect to bob")
|
|
}
|
|
err = wait.Predicate(func() bool {
|
|
return isConnected(carol.PubKeyStr)
|
|
}, 15*time.Second)
|
|
if err != nil {
|
|
t.Fatalf("alice did not reconnect to carol")
|
|
}
|
|
|
|
// We'll also restart Alice to ensure she can reconnect to her peers
|
|
// with open channels.
|
|
if err := net.RestartNode(net.Alice, nil); err != nil {
|
|
t.Fatalf("unable to restart alice's node: %v", err)
|
|
}
|
|
|
|
err = wait.Predicate(func() bool {
|
|
return isConnected(net.Bob.PubKeyStr)
|
|
}, 15*time.Second)
|
|
if err != nil {
|
|
t.Fatalf("alice did not reconnect to bob")
|
|
}
|
|
err = wait.Predicate(func() bool {
|
|
return isConnected(carol.PubKeyStr)
|
|
}, 15*time.Second)
|
|
if err != nil {
|
|
t.Fatalf("alice did not reconnect to carol")
|
|
}
|
|
err = wait.Predicate(func() bool {
|
|
return isConnected(dave.PubKeyStr)
|
|
}, 15*time.Second)
|
|
if err != nil {
|
|
t.Fatalf("alice did not reconnect to dave")
|
|
}
|
|
|
|
// testReconnection is a helper closure that restarts the nodes at both
|
|
// ends of a channel to ensure they do not reconnect after restarting.
|
|
// When restarting Alice, we'll first need to ensure she has
|
|
// reestablished her connection with Dave, as they still have an open
|
|
// channel together.
|
|
testReconnection := func(node *lntest.HarnessNode) {
|
|
// Restart both nodes, to trigger the pruning logic.
|
|
if err := net.RestartNode(node, nil); err != nil {
|
|
t.Fatalf("unable to restart %v's node: %v",
|
|
node.Name(), err)
|
|
}
|
|
|
|
if err := net.RestartNode(net.Alice, nil); err != nil {
|
|
t.Fatalf("unable to restart alice's node: %v", err)
|
|
}
|
|
|
|
// Now restart both nodes and make sure they don't reconnect.
|
|
if err := net.RestartNode(node, nil); err != nil {
|
|
t.Fatalf("unable to restart %v's node: %v", node.Name(),
|
|
err)
|
|
}
|
|
err = wait.Invariant(func() bool {
|
|
return !isConnected(node.PubKeyStr)
|
|
}, 5*time.Second)
|
|
if err != nil {
|
|
t.Fatalf("alice reconnected to %v", node.Name())
|
|
}
|
|
|
|
if err := net.RestartNode(net.Alice, nil); err != nil {
|
|
t.Fatalf("unable to restart alice's node: %v", err)
|
|
}
|
|
err = wait.Predicate(func() bool {
|
|
return isConnected(dave.PubKeyStr)
|
|
}, 20*time.Second)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't reconnect to Dave")
|
|
}
|
|
|
|
err = wait.Invariant(func() bool {
|
|
return !isConnected(node.PubKeyStr)
|
|
}, 5*time.Second)
|
|
if err != nil {
|
|
t.Fatalf("alice reconnected to %v", node.Name())
|
|
}
|
|
}
|
|
|
|
// Now, we'll close the channel between Alice and Bob and ensure there
|
|
// is no reconnection logic between the both once the channel is fully
|
|
// closed.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, coopChanPoint, false)
|
|
|
|
testReconnection(net.Bob)
|
|
|
|
// We'll do the same with Alice and Carol, but this time we'll force
|
|
// close the channel instead.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, forceCloseChanPoint, true)
|
|
|
|
// Cleanup by mining the force close and sweep transaction.
|
|
cleanupForceClose(t, net, net.Alice, forceCloseChanPoint)
|
|
|
|
// We'll need to mine some blocks in order to mark the channel fully
|
|
// closed.
|
|
_, err = net.Miner.Node.Generate(lnd.DefaultBitcoinTimeLockDelta - defaultCSV)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate blocks: %v", err)
|
|
}
|
|
|
|
// Before we test reconnection, we'll ensure that the channel has been
|
|
// fully cleaned up for both Carol and Alice.
|
|
var predErr error
|
|
pendingChansRequest := &lnrpc.PendingChannelsRequest{}
|
|
err = wait.Predicate(func() bool {
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
pendingChanResp, err := net.Alice.PendingChannels(
|
|
ctxt, pendingChansRequest,
|
|
)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to query for pending "+
|
|
"channels: %v", err)
|
|
return false
|
|
}
|
|
|
|
predErr = checkNumForceClosedChannels(pendingChanResp, 0)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
pendingChanResp, err = carol.PendingChannels(
|
|
ctxt, pendingChansRequest,
|
|
)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to query for pending "+
|
|
"channels: %v", err)
|
|
return false
|
|
}
|
|
|
|
predErr = checkNumForceClosedChannels(pendingChanResp, 0)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("channels not marked as fully resolved: %v", predErr)
|
|
}
|
|
|
|
testReconnection(carol)
|
|
|
|
// Finally, we'll ensure that Bob and Carol no longer show in Alice's
|
|
// channel graph.
|
|
describeGraphReq := &lnrpc.ChannelGraphRequest{
|
|
IncludeUnannounced: true,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
channelGraph, err := net.Alice.DescribeGraph(ctxt, describeGraphReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to query for alice's channel graph: %v", err)
|
|
}
|
|
for _, node := range channelGraph.Nodes {
|
|
if node.PubKey == net.Bob.PubKeyStr {
|
|
t.Fatalf("did not expect to find bob in the channel " +
|
|
"graph, but did")
|
|
}
|
|
if node.PubKey == carol.PubKeyStr {
|
|
t.Fatalf("did not expect to find carol in the channel " +
|
|
"graph, but did")
|
|
}
|
|
}
|
|
|
|
// Now that the test is done, we can also close the persistent link.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, persistentChanPoint, false)
|
|
}
|
|
|
|
// testRevokedCloseRetribution tests that Carol is able carry out
|
|
// retribution in the event that she fails immediately after detecting Bob's
|
|
// breach txn in the mempool.
|
|
func testRevokedCloseRetribution(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const (
|
|
chanAmt = lnd.MaxBtcFundingAmount
|
|
paymentAmt = 10000
|
|
numInvoices = 6
|
|
)
|
|
|
|
// Carol will be the breached party. We set --nolisten to ensure Bob
|
|
// won't be able to connect to her and trigger the channel data
|
|
// protection logic automatically. We also can't have Carol
|
|
// automatically re-connect too early, otherwise DLP would be initiated
|
|
// instead of the breach we want to provoke.
|
|
carol, err := net.NewNode(
|
|
"Carol",
|
|
[]string{"--hodl.exit-settle", "--nolisten", "--minbackoff=1h"},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new carol node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
// We must let Bob communicate with Carol before they are able to open
|
|
// channel, so we connect Bob and Carol,
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, net.Bob); err != nil {
|
|
t.Fatalf("unable to connect dave to carol: %v", err)
|
|
}
|
|
|
|
// Before we make a channel, we'll load up Carol with some coins sent
|
|
// directly from the miner.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
|
|
// In order to test Carol's response to an uncooperative channel
|
|
// closure by Bob, we'll first open up a channel between them with a
|
|
// 0.5 BTC value.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, carol, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// With the channel open, we'll create a few invoices for Bob that
|
|
// Carol will pay to in order to advance the state of the channel.
|
|
bobPayReqs, _, _, err := createPayReqs(
|
|
net.Bob, paymentAmt, numInvoices,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// Wait for Carol to receive the channel edge from the funding manager.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("carol didn't see the carol->bob channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// Send payments from Carol to Bob using 3 of Bob's payment hashes
|
|
// generated above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, carol, bobPayReqs[:numInvoices/2],
|
|
true)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
// Next query for Bob's channel state, as we sent 3 payments of 10k
|
|
// satoshis each, Bob should now see his balance as being 30k satoshis.
|
|
var bobChan *lnrpc.Channel
|
|
var predErr error
|
|
err = wait.Predicate(func() bool {
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
bChan, err := getChanInfo(ctxt, net.Bob)
|
|
if err != nil {
|
|
t.Fatalf("unable to get bob's channel info: %v", err)
|
|
}
|
|
if bChan.LocalBalance != 30000 {
|
|
predErr = fmt.Errorf("bob's balance is incorrect, "+
|
|
"got %v, expected %v", bChan.LocalBalance,
|
|
30000)
|
|
return false
|
|
}
|
|
|
|
bobChan = bChan
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("%v", predErr)
|
|
}
|
|
|
|
// Grab Bob's current commitment height (update number), we'll later
|
|
// revert him to this state after additional updates to force him to
|
|
// broadcast this soon to be revoked state.
|
|
bobStateNumPreCopy := bobChan.NumUpdates
|
|
|
|
// Create a temporary file to house Bob's database state at this
|
|
// particular point in history.
|
|
bobTempDbPath, err := ioutil.TempDir("", "bob-past-state")
|
|
if err != nil {
|
|
t.Fatalf("unable to create temp db folder: %v", err)
|
|
}
|
|
bobTempDbFile := filepath.Join(bobTempDbPath, "channel.db")
|
|
defer os.Remove(bobTempDbPath)
|
|
|
|
// With the temporary file created, copy Bob's current state into the
|
|
// temporary file we created above. Later after more updates, we'll
|
|
// restore this state.
|
|
if err := lntest.CopyFile(bobTempDbFile, net.Bob.DBPath()); err != nil {
|
|
t.Fatalf("unable to copy database files: %v", err)
|
|
}
|
|
|
|
// Finally, send payments from Carol to Bob, consuming Bob's remaining
|
|
// payment hashes.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, carol, bobPayReqs[numInvoices/2:],
|
|
true)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
bobChan, err = getChanInfo(ctxt, net.Bob)
|
|
if err != nil {
|
|
t.Fatalf("unable to get bob chan info: %v", err)
|
|
}
|
|
|
|
// Now we shutdown Bob, copying over the his temporary database state
|
|
// which has the *prior* channel state over his current most up to date
|
|
// state. With this, we essentially force Bob to travel back in time
|
|
// within the channel's history.
|
|
if err = net.RestartNode(net.Bob, func() error {
|
|
return os.Rename(bobTempDbFile, net.Bob.DBPath())
|
|
}); err != nil {
|
|
t.Fatalf("unable to restart node: %v", err)
|
|
}
|
|
|
|
// Now query for Bob's channel state, it should show that he's at a
|
|
// state number in the past, not the *latest* state.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
bobChan, err = getChanInfo(ctxt, net.Bob)
|
|
if err != nil {
|
|
t.Fatalf("unable to get bob chan info: %v", err)
|
|
}
|
|
if bobChan.NumUpdates != bobStateNumPreCopy {
|
|
t.Fatalf("db copy failed: %v", bobChan.NumUpdates)
|
|
}
|
|
|
|
// Now force Bob to execute a *force* channel closure by unilaterally
|
|
// broadcasting his current channel state. This is actually the
|
|
// commitment transaction of a prior *revoked* state, so he'll soon
|
|
// feel the wrath of Carol's retribution.
|
|
var closeUpdates lnrpc.Lightning_CloseChannelClient
|
|
force := true
|
|
err = wait.Predicate(func() bool {
|
|
ctxt, _ := context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeUpdates, _, err = net.CloseChannel(ctxt, net.Bob, chanPoint, force)
|
|
if err != nil {
|
|
predErr = err
|
|
return false
|
|
}
|
|
|
|
return true
|
|
}, time.Second*10)
|
|
if err != nil {
|
|
t.Fatalf("unable to close channel: %v", predErr)
|
|
}
|
|
|
|
// Wait for Bob's breach transaction to show up in the mempool to ensure
|
|
// that Carol's node has started waiting for confirmations.
|
|
_, err = waitForTxInMempool(net.Miner.Node, minerMempoolTimeout)
|
|
if err != nil {
|
|
t.Fatalf("unable to find Bob's breach tx in mempool: %v", err)
|
|
}
|
|
|
|
// Here, Carol sees Bob's breach transaction in the mempool, but is waiting
|
|
// for it to confirm before continuing her retribution. We restart Carol to
|
|
// ensure that she is persisting her retribution state and continues
|
|
// watching for the breach transaction to confirm even after her node
|
|
// restarts.
|
|
if err := net.RestartNode(carol, nil); err != nil {
|
|
t.Fatalf("unable to restart Carol's node: %v", err)
|
|
}
|
|
|
|
// Finally, generate a single block, wait for the final close status
|
|
// update, then ensure that the closing transaction was included in the
|
|
// block.
|
|
block := mineBlocks(t, net, 1, 1)[0]
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
breachTXID, err := net.WaitForChannelClose(ctxt, closeUpdates)
|
|
if err != nil {
|
|
t.Fatalf("error while waiting for channel close: %v", err)
|
|
}
|
|
assertTxInBlock(t, block, breachTXID)
|
|
|
|
// Query the mempool for Carol's justice transaction, this should be
|
|
// broadcast as Bob's contract breaching transaction gets confirmed
|
|
// above.
|
|
justiceTXID, err := waitForTxInMempool(net.Miner.Node, minerMempoolTimeout)
|
|
if err != nil {
|
|
t.Fatalf("unable to find Carol's justice tx in mempool: %v", err)
|
|
}
|
|
time.Sleep(100 * time.Millisecond)
|
|
|
|
// Query for the mempool transaction found above. Then assert that all
|
|
// the inputs of this transaction are spending outputs generated by
|
|
// Bob's breach transaction above.
|
|
justiceTx, err := net.Miner.Node.GetRawTransaction(justiceTXID)
|
|
if err != nil {
|
|
t.Fatalf("unable to query for justice tx: %v", err)
|
|
}
|
|
for _, txIn := range justiceTx.MsgTx().TxIn {
|
|
if !bytes.Equal(txIn.PreviousOutPoint.Hash[:], breachTXID[:]) {
|
|
t.Fatalf("justice tx not spending commitment utxo "+
|
|
"instead is: %v", txIn.PreviousOutPoint)
|
|
}
|
|
}
|
|
|
|
// We restart Carol here to ensure that she persists her retribution state
|
|
// and successfully continues exacting retribution after restarting. At
|
|
// this point, Carol has broadcast the justice transaction, but it hasn't
|
|
// been confirmed yet; when Carol restarts, she should start waiting for
|
|
// the justice transaction to confirm again.
|
|
if err := net.RestartNode(carol, nil); err != nil {
|
|
t.Fatalf("unable to restart Carol's node: %v", err)
|
|
}
|
|
|
|
// Now mine a block, this transaction should include Carol's justice
|
|
// transaction which was just accepted into the mempool.
|
|
block = mineBlocks(t, net, 1, 1)[0]
|
|
|
|
// The block should have exactly *two* transactions, one of which is
|
|
// the justice transaction.
|
|
if len(block.Transactions) != 2 {
|
|
t.Fatalf("transaction wasn't mined")
|
|
}
|
|
justiceSha := block.Transactions[1].TxHash()
|
|
if !bytes.Equal(justiceTx.Hash()[:], justiceSha[:]) {
|
|
t.Fatalf("justice tx wasn't mined")
|
|
}
|
|
|
|
assertNodeNumChannels(t, carol, 0)
|
|
}
|
|
|
|
// testRevokedCloseRetributionZeroValueRemoteOutput tests that Dave is able
|
|
// carry out retribution in the event that she fails in state where the remote
|
|
// commitment output has zero-value.
|
|
func testRevokedCloseRetributionZeroValueRemoteOutput(net *lntest.NetworkHarness,
|
|
t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const (
|
|
chanAmt = lnd.MaxBtcFundingAmount
|
|
paymentAmt = 10000
|
|
numInvoices = 6
|
|
)
|
|
|
|
// Since we'd like to test some multi-hop failure scenarios, we'll
|
|
// introduce another node into our test network: Carol.
|
|
carol, err := net.NewNode("Carol", []string{"--hodl.exit-settle"})
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
// Dave will be the breached party. We set --nolisten to ensure Carol
|
|
// won't be able to connect to him and trigger the channel data
|
|
// protection logic automatically. We also can't have Dave automatically
|
|
// re-connect too early, otherwise DLP would be initiated instead of the
|
|
// breach we want to provoke.
|
|
dave, err := net.NewNode(
|
|
"Dave",
|
|
[]string{"--hodl.exit-settle", "--nolisten", "--minbackoff=1h"},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
// We must let Dave have an open channel before she can send a node
|
|
// announcement, so we open a channel with Carol,
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, dave, carol); err != nil {
|
|
t.Fatalf("unable to connect dave to carol: %v", err)
|
|
}
|
|
|
|
// Before we make a channel, we'll load up Dave with some coins sent
|
|
// directly from the miner.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to dave: %v", err)
|
|
}
|
|
|
|
// In order to test Dave's response to an uncooperative channel
|
|
// closure by Carol, we'll first open up a channel between them with a
|
|
// 0.5 BTC value.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, dave, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// With the channel open, we'll create a few invoices for Carol that
|
|
// Dave will pay to in order to advance the state of the channel.
|
|
carolPayReqs, _, _, err := createPayReqs(
|
|
carol, paymentAmt, numInvoices,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// Wait for Dave to receive the channel edge from the funding manager.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = dave.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("dave didn't see the dave->carol channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// Next query for Carol's channel state, as we sent 0 payments, Carol
|
|
// should now see her balance as being 0 satoshis.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolChan, err := getChanInfo(ctxt, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol's channel info: %v", err)
|
|
}
|
|
if carolChan.LocalBalance != 0 {
|
|
t.Fatalf("carol's balance is incorrect, got %v, expected %v",
|
|
carolChan.LocalBalance, 0)
|
|
}
|
|
|
|
// Grab Carol's current commitment height (update number), we'll later
|
|
// revert her to this state after additional updates to force him to
|
|
// broadcast this soon to be revoked state.
|
|
carolStateNumPreCopy := carolChan.NumUpdates
|
|
|
|
// Create a temporary file to house Carol's database state at this
|
|
// particular point in history.
|
|
carolTempDbPath, err := ioutil.TempDir("", "carol-past-state")
|
|
if err != nil {
|
|
t.Fatalf("unable to create temp db folder: %v", err)
|
|
}
|
|
carolTempDbFile := filepath.Join(carolTempDbPath, "channel.db")
|
|
defer os.Remove(carolTempDbPath)
|
|
|
|
// With the temporary file created, copy Carol's current state into the
|
|
// temporary file we created above. Later after more updates, we'll
|
|
// restore this state.
|
|
if err := lntest.CopyFile(carolTempDbFile, carol.DBPath()); err != nil {
|
|
t.Fatalf("unable to copy database files: %v", err)
|
|
}
|
|
|
|
// Finally, send payments from Dave to Carol, consuming Carol's remaining
|
|
// payment hashes.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, dave, carolPayReqs, false)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolChan, err = getChanInfo(ctxt, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol chan info: %v", err)
|
|
}
|
|
|
|
// Now we shutdown Carol, copying over the his temporary database state
|
|
// which has the *prior* channel state over his current most up to date
|
|
// state. With this, we essentially force Carol to travel back in time
|
|
// within the channel's history.
|
|
if err = net.RestartNode(carol, func() error {
|
|
return os.Rename(carolTempDbFile, carol.DBPath())
|
|
}); err != nil {
|
|
t.Fatalf("unable to restart node: %v", err)
|
|
}
|
|
|
|
// Now query for Carol's channel state, it should show that he's at a
|
|
// state number in the past, not the *latest* state.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolChan, err = getChanInfo(ctxt, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol chan info: %v", err)
|
|
}
|
|
if carolChan.NumUpdates != carolStateNumPreCopy {
|
|
t.Fatalf("db copy failed: %v", carolChan.NumUpdates)
|
|
}
|
|
|
|
// Now force Carol to execute a *force* channel closure by unilaterally
|
|
// broadcasting his current channel state. This is actually the
|
|
// commitment transaction of a prior *revoked* state, so he'll soon
|
|
// feel the wrath of Dave's retribution.
|
|
var (
|
|
closeUpdates lnrpc.Lightning_CloseChannelClient
|
|
closeTxId *chainhash.Hash
|
|
closeErr error
|
|
force bool = true
|
|
)
|
|
err = wait.Predicate(func() bool {
|
|
ctxt, _ := context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeUpdates, closeTxId, closeErr = net.CloseChannel(
|
|
ctxt, carol, chanPoint, force,
|
|
)
|
|
return closeErr == nil
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("unable to close channel: %v", closeErr)
|
|
}
|
|
|
|
// Query the mempool for the breaching closing transaction, this should
|
|
// be broadcast by Carol when she force closes the channel above.
|
|
txid, err := waitForTxInMempool(net.Miner.Node, minerMempoolTimeout)
|
|
if err != nil {
|
|
t.Fatalf("unable to find Carol's force close tx in mempool: %v",
|
|
err)
|
|
}
|
|
if *txid != *closeTxId {
|
|
t.Fatalf("expected closeTx(%v) in mempool, instead found %v",
|
|
closeTxId, txid)
|
|
}
|
|
|
|
// Finally, generate a single block, wait for the final close status
|
|
// update, then ensure that the closing transaction was included in the
|
|
// block.
|
|
block := mineBlocks(t, net, 1, 1)[0]
|
|
|
|
// Here, Dave receives a confirmation of Carol's breach transaction.
|
|
// We restart Dave to ensure that she is persisting her retribution
|
|
// state and continues exacting justice after her node restarts.
|
|
if err := net.RestartNode(dave, nil); err != nil {
|
|
t.Fatalf("unable to stop Dave's node: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
breachTXID, err := net.WaitForChannelClose(ctxt, closeUpdates)
|
|
if err != nil {
|
|
t.Fatalf("error while waiting for channel close: %v", err)
|
|
}
|
|
assertTxInBlock(t, block, breachTXID)
|
|
|
|
// Query the mempool for Dave's justice transaction, this should be
|
|
// broadcast as Carol's contract breaching transaction gets confirmed
|
|
// above.
|
|
justiceTXID, err := waitForTxInMempool(net.Miner.Node, minerMempoolTimeout)
|
|
if err != nil {
|
|
t.Fatalf("unable to find Dave's justice tx in mempool: %v",
|
|
err)
|
|
}
|
|
time.Sleep(100 * time.Millisecond)
|
|
|
|
// Query for the mempool transaction found above. Then assert that all
|
|
// the inputs of this transaction are spending outputs generated by
|
|
// Carol's breach transaction above.
|
|
justiceTx, err := net.Miner.Node.GetRawTransaction(justiceTXID)
|
|
if err != nil {
|
|
t.Fatalf("unable to query for justice tx: %v", err)
|
|
}
|
|
for _, txIn := range justiceTx.MsgTx().TxIn {
|
|
if !bytes.Equal(txIn.PreviousOutPoint.Hash[:], breachTXID[:]) {
|
|
t.Fatalf("justice tx not spending commitment utxo "+
|
|
"instead is: %v", txIn.PreviousOutPoint)
|
|
}
|
|
}
|
|
|
|
// We restart Dave here to ensure that he persists her retribution state
|
|
// and successfully continues exacting retribution after restarting. At
|
|
// this point, Dave has broadcast the justice transaction, but it hasn't
|
|
// been confirmed yet; when Dave restarts, she should start waiting for
|
|
// the justice transaction to confirm again.
|
|
if err := net.RestartNode(dave, nil); err != nil {
|
|
t.Fatalf("unable to restart Dave's node: %v", err)
|
|
}
|
|
|
|
// Now mine a block, this transaction should include Dave's justice
|
|
// transaction which was just accepted into the mempool.
|
|
block = mineBlocks(t, net, 1, 1)[0]
|
|
|
|
// The block should have exactly *two* transactions, one of which is
|
|
// the justice transaction.
|
|
if len(block.Transactions) != 2 {
|
|
t.Fatalf("transaction wasn't mined")
|
|
}
|
|
justiceSha := block.Transactions[1].TxHash()
|
|
if !bytes.Equal(justiceTx.Hash()[:], justiceSha[:]) {
|
|
t.Fatalf("justice tx wasn't mined")
|
|
}
|
|
|
|
assertNodeNumChannels(t, dave, 0)
|
|
}
|
|
|
|
// testRevokedCloseRetributionRemoteHodl tests that Dave properly responds to a
|
|
// channel breach made by the remote party, specifically in the case that the
|
|
// remote party breaches before settling extended HTLCs.
|
|
func testRevokedCloseRetributionRemoteHodl(net *lntest.NetworkHarness,
|
|
t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const (
|
|
chanAmt = lnd.MaxBtcFundingAmount
|
|
pushAmt = 200000
|
|
paymentAmt = 10000
|
|
numInvoices = 6
|
|
)
|
|
|
|
// Since this test will result in the counterparty being left in a
|
|
// weird state, we will introduce another node into our test network:
|
|
// Carol.
|
|
carol, err := net.NewNode("Carol", []string{"--hodl.exit-settle"})
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
// We'll also create a new node Dave, who will have a channel with
|
|
// Carol, and also use similar settings so we can broadcast a commit
|
|
// with active HTLCs. Dave will be the breached party. We set
|
|
// --nolisten to ensure Carol won't be able to connect to him and
|
|
// trigger the channel data protection logic automatically.
|
|
dave, err := net.NewNode(
|
|
"Dave",
|
|
[]string{"--hodl.exit-settle", "--nolisten"},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new dave node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
// We must let Dave communicate with Carol before they are able to open
|
|
// channel, so we connect Dave and Carol,
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, dave, carol); err != nil {
|
|
t.Fatalf("unable to connect dave to carol: %v", err)
|
|
}
|
|
|
|
// Before we make a channel, we'll load up Dave with some coins sent
|
|
// directly from the miner.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to dave: %v", err)
|
|
}
|
|
|
|
// In order to test Dave's response to an uncooperative channel closure
|
|
// by Carol, we'll first open up a channel between them with a
|
|
// lnd.MaxBtcFundingAmount (2^24) satoshis value.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, dave, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
|
|
// With the channel open, we'll create a few invoices for Carol that
|
|
// Dave will pay to in order to advance the state of the channel.
|
|
carolPayReqs, _, _, err := createPayReqs(
|
|
carol, paymentAmt, numInvoices,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// We'll introduce a closure to validate that Carol's current balance
|
|
// matches the given expected amount.
|
|
checkCarolBalance := func(expectedAmt int64) {
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolChan, err := getChanInfo(ctxt, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol's channel info: %v", err)
|
|
}
|
|
if carolChan.LocalBalance != expectedAmt {
|
|
t.Fatalf("carol's balance is incorrect, "+
|
|
"got %v, expected %v", carolChan.LocalBalance,
|
|
expectedAmt)
|
|
}
|
|
}
|
|
|
|
// We'll introduce another closure to validate that Carol's current
|
|
// number of updates is at least as large as the provided minimum
|
|
// number.
|
|
checkCarolNumUpdatesAtLeast := func(minimum uint64) {
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolChan, err := getChanInfo(ctxt, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol's channel info: %v", err)
|
|
}
|
|
if carolChan.NumUpdates < minimum {
|
|
t.Fatalf("carol's numupdates is incorrect, want %v "+
|
|
"to be at least %v", carolChan.NumUpdates,
|
|
minimum)
|
|
}
|
|
}
|
|
|
|
// Wait for Dave to receive the channel edge from the funding manager.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = dave.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("dave didn't see the dave->carol channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// Ensure that carol's balance starts with the amount we pushed to her.
|
|
checkCarolBalance(pushAmt)
|
|
|
|
// Send payments from Dave to Carol using 3 of Carol's payment hashes
|
|
// generated above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(
|
|
ctxt, dave, carolPayReqs[:numInvoices/2], false,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
// At this point, we'll also send over a set of HTLC's from Carol to
|
|
// Dave. This ensures that the final revoked transaction has HTLC's in
|
|
// both directions.
|
|
davePayReqs, _, _, err := createPayReqs(
|
|
dave, paymentAmt, numInvoices,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// Send payments from Carol to Dave using 3 of Dave's payment hashes
|
|
// generated above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(
|
|
ctxt, carol, davePayReqs[:numInvoices/2], false,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
// Next query for Carol's channel state, as we sent 3 payments of 10k
|
|
// satoshis each, however Carol should now see her balance as being
|
|
// equal to the push amount in satoshis since she has not settled.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolChan, err := getChanInfo(ctxt, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol's channel info: %v", err)
|
|
}
|
|
|
|
// Grab Carol's current commitment height (update number), we'll later
|
|
// revert her to this state after additional updates to force her to
|
|
// broadcast this soon to be revoked state.
|
|
carolStateNumPreCopy := carolChan.NumUpdates
|
|
|
|
// Ensure that carol's balance still reflects the original amount we
|
|
// pushed to her, minus the HTLCs she just sent to Dave.
|
|
checkCarolBalance(pushAmt - 3*paymentAmt)
|
|
|
|
// Since Carol has not settled, she should only see at least one update
|
|
// to her channel.
|
|
checkCarolNumUpdatesAtLeast(1)
|
|
|
|
// Create a temporary file to house Carol's database state at this
|
|
// particular point in history.
|
|
carolTempDbPath, err := ioutil.TempDir("", "carol-past-state")
|
|
if err != nil {
|
|
t.Fatalf("unable to create temp db folder: %v", err)
|
|
}
|
|
carolTempDbFile := filepath.Join(carolTempDbPath, "channel.db")
|
|
defer os.Remove(carolTempDbPath)
|
|
|
|
// With the temporary file created, copy Carol's current state into the
|
|
// temporary file we created above. Later after more updates, we'll
|
|
// restore this state.
|
|
if err := lntest.CopyFile(carolTempDbFile, carol.DBPath()); err != nil {
|
|
t.Fatalf("unable to copy database files: %v", err)
|
|
}
|
|
|
|
// Finally, send payments from Dave to Carol, consuming Carol's
|
|
// remaining payment hashes.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(
|
|
ctxt, dave, carolPayReqs[numInvoices/2:], false,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
// Ensure that carol's balance still shows the amount we originally
|
|
// pushed to her (minus the HTLCs she sent to Bob), and that at least
|
|
// one more update has occurred.
|
|
time.Sleep(500 * time.Millisecond)
|
|
checkCarolBalance(pushAmt - 3*paymentAmt)
|
|
checkCarolNumUpdatesAtLeast(carolStateNumPreCopy + 1)
|
|
|
|
// Suspend Dave, such that Carol won't reconnect at startup, triggering
|
|
// the data loss protection.
|
|
restartDave, err := net.SuspendNode(dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to suspend Dave: %v", err)
|
|
}
|
|
|
|
// Now we shutdown Carol, copying over the her temporary database state
|
|
// which has the *prior* channel state over her current most up to date
|
|
// state. With this, we essentially force Carol to travel back in time
|
|
// within the channel's history.
|
|
if err = net.RestartNode(carol, func() error {
|
|
return os.Rename(carolTempDbFile, carol.DBPath())
|
|
}); err != nil {
|
|
t.Fatalf("unable to restart node: %v", err)
|
|
}
|
|
|
|
time.Sleep(200 * time.Millisecond)
|
|
|
|
// Ensure that Carol's view of the channel is consistent with the state
|
|
// of the channel just before it was snapshotted.
|
|
checkCarolBalance(pushAmt - 3*paymentAmt)
|
|
checkCarolNumUpdatesAtLeast(1)
|
|
|
|
// Now query for Carol's channel state, it should show that she's at a
|
|
// state number in the past, *not* the latest state.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolChan, err = getChanInfo(ctxt, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol chan info: %v", err)
|
|
}
|
|
if carolChan.NumUpdates != carolStateNumPreCopy {
|
|
t.Fatalf("db copy failed: %v", carolChan.NumUpdates)
|
|
}
|
|
|
|
// Now force Carol to execute a *force* channel closure by unilaterally
|
|
// broadcasting her current channel state. This is actually the
|
|
// commitment transaction of a prior *revoked* state, so she'll soon
|
|
// feel the wrath of Dave's retribution.
|
|
force := true
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeUpdates, closeTxId, err := net.CloseChannel(ctxt, carol,
|
|
chanPoint, force)
|
|
if err != nil {
|
|
t.Fatalf("unable to close channel: %v", err)
|
|
}
|
|
|
|
// Query the mempool for the breaching closing transaction, this should
|
|
// be broadcast by Carol when she force closes the channel above.
|
|
txid, err := waitForTxInMempool(net.Miner.Node, minerMempoolTimeout)
|
|
if err != nil {
|
|
t.Fatalf("unable to find Carol's force close tx in mempool: %v",
|
|
err)
|
|
}
|
|
if *txid != *closeTxId {
|
|
t.Fatalf("expected closeTx(%v) in mempool, instead found %v",
|
|
closeTxId, txid)
|
|
}
|
|
|
|
// Generate a single block to mine the breach transaction.
|
|
block := mineBlocks(t, net, 1, 1)[0]
|
|
|
|
// We resurrect Dave to ensure he will be exacting justice after his
|
|
// node restarts.
|
|
if err := restartDave(); err != nil {
|
|
t.Fatalf("unable to stop Dave's node: %v", err)
|
|
}
|
|
|
|
// Finally, wait for the final close status update, then ensure that
|
|
// the closing transaction was included in the block.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
breachTXID, err := net.WaitForChannelClose(ctxt, closeUpdates)
|
|
if err != nil {
|
|
t.Fatalf("error while waiting for channel close: %v", err)
|
|
}
|
|
if *breachTXID != *closeTxId {
|
|
t.Fatalf("expected breach ID(%v) to be equal to close ID (%v)",
|
|
breachTXID, closeTxId)
|
|
}
|
|
assertTxInBlock(t, block, breachTXID)
|
|
|
|
// Query the mempool for Dave's justice transaction, this should be
|
|
// broadcast as Carol's contract breaching transaction gets confirmed
|
|
// above. Since Carol might have had the time to take some of the HTLC
|
|
// outputs to the second level before Dave broadcasts his justice tx,
|
|
// we'll search through the mempool for a tx that matches the number of
|
|
// expected inputs in the justice tx.
|
|
var predErr error
|
|
var justiceTxid *chainhash.Hash
|
|
errNotFound := errors.New("justice tx not found")
|
|
findJusticeTx := func() (*chainhash.Hash, error) {
|
|
mempool, err := net.Miner.Node.GetRawMempool()
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to get mempool from "+
|
|
"miner: %v", err)
|
|
}
|
|
|
|
for _, txid := range mempool {
|
|
// Check that the justice tx has the appropriate number
|
|
// of inputs.
|
|
tx, err := net.Miner.Node.GetRawTransaction(txid)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to query for "+
|
|
"txs: %v", err)
|
|
}
|
|
|
|
exNumInputs := 2 + numInvoices
|
|
if len(tx.MsgTx().TxIn) == exNumInputs {
|
|
return txid, nil
|
|
}
|
|
}
|
|
return nil, errNotFound
|
|
}
|
|
|
|
err = wait.Predicate(func() bool {
|
|
txid, err := findJusticeTx()
|
|
if err != nil {
|
|
predErr = err
|
|
return false
|
|
}
|
|
|
|
justiceTxid = txid
|
|
return true
|
|
}, time.Second*10)
|
|
if err != nil && predErr == errNotFound {
|
|
// If Dave is unable to broadcast his justice tx on first
|
|
// attempt because of the second layer transactions, he will
|
|
// wait until the next block epoch before trying again. Because
|
|
// of this, we'll mine a block if we cannot find the justice tx
|
|
// immediately. Since we cannot tell for sure how many
|
|
// transactions will be in the mempool at this point, we pass 0
|
|
// as the last argument, indicating we don't care what's in the
|
|
// mempool.
|
|
mineBlocks(t, net, 1, 0)
|
|
err = wait.Predicate(func() bool {
|
|
txid, err := findJusticeTx()
|
|
if err != nil {
|
|
predErr = err
|
|
return false
|
|
}
|
|
|
|
justiceTxid = txid
|
|
return true
|
|
}, time.Second*10)
|
|
}
|
|
if err != nil {
|
|
t.Fatalf(predErr.Error())
|
|
}
|
|
|
|
justiceTx, err := net.Miner.Node.GetRawTransaction(justiceTxid)
|
|
if err != nil {
|
|
t.Fatalf("unable to query for justice tx: %v", err)
|
|
}
|
|
|
|
// isSecondLevelSpend checks that the passed secondLevelTxid is a
|
|
// potentitial second level spend spending from the commit tx.
|
|
isSecondLevelSpend := func(commitTxid, secondLevelTxid *chainhash.Hash) bool {
|
|
secondLevel, err := net.Miner.Node.GetRawTransaction(
|
|
secondLevelTxid)
|
|
if err != nil {
|
|
t.Fatalf("unable to query for tx: %v", err)
|
|
}
|
|
|
|
// A second level spend should have only one input, and one
|
|
// output.
|
|
if len(secondLevel.MsgTx().TxIn) != 1 {
|
|
return false
|
|
}
|
|
if len(secondLevel.MsgTx().TxOut) != 1 {
|
|
return false
|
|
}
|
|
|
|
// The sole input should be spending from the commit tx.
|
|
txIn := secondLevel.MsgTx().TxIn[0]
|
|
if !bytes.Equal(txIn.PreviousOutPoint.Hash[:], commitTxid[:]) {
|
|
return false
|
|
}
|
|
|
|
return true
|
|
}
|
|
|
|
// Check that all the inputs of this transaction are spending outputs
|
|
// generated by Carol's breach transaction above.
|
|
for _, txIn := range justiceTx.MsgTx().TxIn {
|
|
if bytes.Equal(txIn.PreviousOutPoint.Hash[:], breachTXID[:]) {
|
|
continue
|
|
}
|
|
|
|
// If the justice tx is spending from an output that was not on
|
|
// the breach tx, Carol might have had the time to take an
|
|
// output to the second level. In that case, check that the
|
|
// justice tx is spending this second level output.
|
|
if isSecondLevelSpend(breachTXID, &txIn.PreviousOutPoint.Hash) {
|
|
continue
|
|
}
|
|
t.Fatalf("justice tx not spending commitment utxo "+
|
|
"instead is: %v", txIn.PreviousOutPoint)
|
|
}
|
|
time.Sleep(100 * time.Millisecond)
|
|
|
|
// We restart Dave here to ensure that he persists he retribution state
|
|
// and successfully continues exacting retribution after restarting. At
|
|
// this point, Dave has broadcast the justice transaction, but it
|
|
// hasn't been confirmed yet; when Dave restarts, he should start
|
|
// waiting for the justice transaction to confirm again.
|
|
if err := net.RestartNode(dave, nil); err != nil {
|
|
t.Fatalf("unable to restart Dave's node: %v", err)
|
|
}
|
|
|
|
// Now mine a block, this transaction should include Dave's justice
|
|
// transaction which was just accepted into the mempool.
|
|
block = mineBlocks(t, net, 1, 1)[0]
|
|
assertTxInBlock(t, block, justiceTxid)
|
|
|
|
// Dave should have no open channels.
|
|
assertNodeNumChannels(t, dave, 0)
|
|
}
|
|
|
|
// testRevokedCloseRetributionAltruistWatchtower establishes a channel between
|
|
// Carol and Dave, where Carol is using a third node Willy as her watchtower.
|
|
// After sending some payments, Dave reverts his state and force closes to
|
|
// trigger a breach. Carol is kept offline throughout the process and the test
|
|
// asserts that Willy responds by broadcasting the justice transaction on
|
|
// Carol's behalf sweeping her funds without a reward.
|
|
func testRevokedCloseRetributionAltruistWatchtower(net *lntest.NetworkHarness,
|
|
t *harnessTest) {
|
|
|
|
ctxb := context.Background()
|
|
const (
|
|
chanAmt = lnd.MaxBtcFundingAmount
|
|
paymentAmt = 10000
|
|
numInvoices = 6
|
|
externalIP = "1.2.3.4"
|
|
)
|
|
|
|
// Since we'd like to test some multi-hop failure scenarios, we'll
|
|
// introduce another node into our test network: Carol.
|
|
carol, err := net.NewNode("Carol", []string{"--hodl.exit-settle"})
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
// Willy the watchtower will protect Dave from Carol's breach. He will
|
|
// remain online in order to punish Carol on Dave's behalf, since the
|
|
// breach will happen while Dave is offline.
|
|
willy, err := net.NewNode("Willy", []string{
|
|
"--watchtower.active",
|
|
"--watchtower.externalip=" + externalIP,
|
|
})
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, willy)
|
|
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
willyInfo, err := willy.Watchtower.GetInfo(
|
|
ctxt, &watchtowerrpc.GetInfoRequest{},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to getinfo from willy: %v", err)
|
|
}
|
|
|
|
// Assert that Willy has one listener and it is 0.0.0.0:9911 or
|
|
// [::]:9911. Since no listener is explicitly specified, one of these
|
|
// should be the default depending on whether the host supports IPv6 or
|
|
// not.
|
|
if len(willyInfo.Listeners) != 1 {
|
|
t.Fatalf("Willy should have 1 listener, has %d",
|
|
len(willyInfo.Listeners))
|
|
}
|
|
listener := willyInfo.Listeners[0]
|
|
if listener != "0.0.0.0:9911" && listener != "[::]:9911" {
|
|
t.Fatalf("expected listener on 0.0.0.0:9911 or [::]:9911, "+
|
|
"got %v", listener)
|
|
}
|
|
|
|
// Assert the Willy's URIs properly display the chosen external IP.
|
|
if len(willyInfo.Uris) != 1 {
|
|
t.Fatalf("Willy should have 1 uri, has %d",
|
|
len(willyInfo.Uris))
|
|
}
|
|
if !strings.Contains(willyInfo.Uris[0], externalIP) {
|
|
t.Fatalf("expected uri with %v, got %v",
|
|
externalIP, willyInfo.Uris[0])
|
|
}
|
|
|
|
// Dave will be the breached party. We set --nolisten to ensure Carol
|
|
// won't be able to connect to him and trigger the channel data
|
|
// protection logic automatically.
|
|
dave, err := net.NewNode("Dave", []string{
|
|
"--nolisten",
|
|
"--wtclient.active",
|
|
})
|
|
if err != nil {
|
|
t.Fatalf("unable to create new node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
addTowerReq := &wtclientrpc.AddTowerRequest{
|
|
Pubkey: willyInfo.Pubkey,
|
|
Address: listener,
|
|
}
|
|
if _, err := dave.WatchtowerClient.AddTower(ctxt, addTowerReq); err != nil {
|
|
t.Fatalf("unable to add willy's watchtower: %v", err)
|
|
}
|
|
|
|
// We must let Dave have an open channel before she can send a node
|
|
// announcement, so we open a channel with Carol,
|
|
if err := net.ConnectNodes(ctxb, dave, carol); err != nil {
|
|
t.Fatalf("unable to connect dave to carol: %v", err)
|
|
}
|
|
|
|
// Before we make a channel, we'll load up Dave with some coins sent
|
|
// directly from the miner.
|
|
err = net.SendCoins(ctxb, btcutil.SatoshiPerBitcoin, dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to dave: %v", err)
|
|
}
|
|
|
|
// In order to test Dave's response to an uncooperative channel
|
|
// closure by Carol, we'll first open up a channel between them with a
|
|
// 0.5 BTC value.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, dave, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: 3 * (chanAmt / 4),
|
|
PushAmt: chanAmt / 4,
|
|
},
|
|
)
|
|
|
|
// With the channel open, we'll create a few invoices for Carol that
|
|
// Dave will pay to in order to advance the state of the channel.
|
|
carolPayReqs, _, _, err := createPayReqs(
|
|
carol, paymentAmt, numInvoices,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// Wait for Dave to receive the channel edge from the funding manager.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = dave.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("dave didn't see the dave->carol channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// Next query for Carol's channel state, as we sent 0 payments, Carol
|
|
// should still see her balance as the push amount, which is 1/4 of the
|
|
// capacity.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolChan, err := getChanInfo(ctxt, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol's channel info: %v", err)
|
|
}
|
|
if carolChan.LocalBalance != int64(chanAmt/4) {
|
|
t.Fatalf("carol's balance is incorrect, got %v, expected %v",
|
|
carolChan.LocalBalance, chanAmt/4)
|
|
}
|
|
|
|
// Grab Carol's current commitment height (update number), we'll later
|
|
// revert her to this state after additional updates to force him to
|
|
// broadcast this soon to be revoked state.
|
|
carolStateNumPreCopy := carolChan.NumUpdates
|
|
|
|
// Create a temporary file to house Carol's database state at this
|
|
// particular point in history.
|
|
carolTempDbPath, err := ioutil.TempDir("", "carol-past-state")
|
|
if err != nil {
|
|
t.Fatalf("unable to create temp db folder: %v", err)
|
|
}
|
|
carolTempDbFile := filepath.Join(carolTempDbPath, "channel.db")
|
|
defer os.Remove(carolTempDbPath)
|
|
|
|
// With the temporary file created, copy Carol's current state into the
|
|
// temporary file we created above. Later after more updates, we'll
|
|
// restore this state.
|
|
if err := lntest.CopyFile(carolTempDbFile, carol.DBPath()); err != nil {
|
|
t.Fatalf("unable to copy database files: %v", err)
|
|
}
|
|
|
|
// Finally, send payments from Dave to Carol, consuming Carol's remaining
|
|
// payment hashes.
|
|
err = completePaymentRequests(ctxb, dave, carolPayReqs, false)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
daveBalReq := &lnrpc.WalletBalanceRequest{}
|
|
daveBalResp, err := dave.WalletBalance(ctxt, daveBalReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get dave's balance: %v", err)
|
|
}
|
|
|
|
davePreSweepBalance := daveBalResp.ConfirmedBalance
|
|
|
|
// Shutdown Dave to simulate going offline for an extended period of
|
|
// time. Once he's not watching, Carol will try to breach the channel.
|
|
restart, err := net.SuspendNode(dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to suspend Dave: %v", err)
|
|
}
|
|
|
|
// Now we shutdown Carol, copying over the his temporary database state
|
|
// which has the *prior* channel state over his current most up to date
|
|
// state. With this, we essentially force Carol to travel back in time
|
|
// within the channel's history.
|
|
if err = net.RestartNode(carol, func() error {
|
|
return os.Rename(carolTempDbFile, carol.DBPath())
|
|
}); err != nil {
|
|
t.Fatalf("unable to restart node: %v", err)
|
|
}
|
|
|
|
// Now query for Carol's channel state, it should show that he's at a
|
|
// state number in the past, not the *latest* state.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolChan, err = getChanInfo(ctxt, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol chan info: %v", err)
|
|
}
|
|
if carolChan.NumUpdates != carolStateNumPreCopy {
|
|
t.Fatalf("db copy failed: %v", carolChan.NumUpdates)
|
|
}
|
|
|
|
// TODO(conner): add hook for backup completion
|
|
time.Sleep(3 * time.Second)
|
|
|
|
// Now force Carol to execute a *force* channel closure by unilaterally
|
|
// broadcasting his current channel state. This is actually the
|
|
// commitment transaction of a prior *revoked* state, so he'll soon
|
|
// feel the wrath of Dave's retribution.
|
|
closeUpdates, closeTxId, err := net.CloseChannel(
|
|
ctxb, carol, chanPoint, true,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to close channel: %v", err)
|
|
}
|
|
|
|
// Query the mempool for the breaching closing transaction, this should
|
|
// be broadcast by Carol when she force closes the channel above.
|
|
txid, err := waitForTxInMempool(net.Miner.Node, minerMempoolTimeout)
|
|
if err != nil {
|
|
t.Fatalf("unable to find Carol's force close tx in mempool: %v",
|
|
err)
|
|
}
|
|
if *txid != *closeTxId {
|
|
t.Fatalf("expected closeTx(%v) in mempool, instead found %v",
|
|
closeTxId, txid)
|
|
}
|
|
|
|
// Finally, generate a single block, wait for the final close status
|
|
// update, then ensure that the closing transaction was included in the
|
|
// block.
|
|
block := mineBlocks(t, net, 1, 1)[0]
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
breachTXID, err := net.WaitForChannelClose(ctxt, closeUpdates)
|
|
if err != nil {
|
|
t.Fatalf("error while waiting for channel close: %v", err)
|
|
}
|
|
assertTxInBlock(t, block, breachTXID)
|
|
|
|
// Query the mempool for Dave's justice transaction, this should be
|
|
// broadcast as Carol's contract breaching transaction gets confirmed
|
|
// above.
|
|
justiceTXID, err := waitForTxInMempool(net.Miner.Node, minerMempoolTimeout)
|
|
if err != nil {
|
|
t.Fatalf("unable to find Dave's justice tx in mempool: %v",
|
|
err)
|
|
}
|
|
time.Sleep(100 * time.Millisecond)
|
|
|
|
// Query for the mempool transaction found above. Then assert that all
|
|
// the inputs of this transaction are spending outputs generated by
|
|
// Carol's breach transaction above.
|
|
justiceTx, err := net.Miner.Node.GetRawTransaction(justiceTXID)
|
|
if err != nil {
|
|
t.Fatalf("unable to query for justice tx: %v", err)
|
|
}
|
|
for _, txIn := range justiceTx.MsgTx().TxIn {
|
|
if !bytes.Equal(txIn.PreviousOutPoint.Hash[:], breachTXID[:]) {
|
|
t.Fatalf("justice tx not spending commitment utxo "+
|
|
"instead is: %v", txIn.PreviousOutPoint)
|
|
}
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
willyBalReq := &lnrpc.WalletBalanceRequest{}
|
|
willyBalResp, err := willy.WalletBalance(ctxt, willyBalReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get willy's balance: %v", err)
|
|
}
|
|
|
|
if willyBalResp.ConfirmedBalance != 0 {
|
|
t.Fatalf("willy should have 0 balance before mining "+
|
|
"justice transaction, instead has %d",
|
|
willyBalResp.ConfirmedBalance)
|
|
}
|
|
|
|
// Now mine a block, this transaction should include Dave's justice
|
|
// transaction which was just accepted into the mempool.
|
|
block = mineBlocks(t, net, 1, 1)[0]
|
|
|
|
// The block should have exactly *two* transactions, one of which is
|
|
// the justice transaction.
|
|
if len(block.Transactions) != 2 {
|
|
t.Fatalf("transaction wasn't mined")
|
|
}
|
|
justiceSha := block.Transactions[1].TxHash()
|
|
if !bytes.Equal(justiceTx.Hash()[:], justiceSha[:]) {
|
|
t.Fatalf("justice tx wasn't mined")
|
|
}
|
|
|
|
// Ensure that Willy doesn't get any funds, as he is acting as an
|
|
// altruist watchtower.
|
|
var predErr error
|
|
err = wait.Invariant(func() bool {
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
willyBalReq := &lnrpc.WalletBalanceRequest{}
|
|
willyBalResp, err := willy.WalletBalance(ctxt, willyBalReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get willy's balance: %v", err)
|
|
}
|
|
|
|
if willyBalResp.ConfirmedBalance != 0 {
|
|
predErr = fmt.Errorf("Expected Willy to have no funds "+
|
|
"after justice transaction was mined, found %v",
|
|
willyBalResp)
|
|
return false
|
|
}
|
|
|
|
return true
|
|
}, time.Second*5)
|
|
if err != nil {
|
|
t.Fatalf("%v", predErr)
|
|
}
|
|
|
|
// Restart Dave, who will still think his channel with Carol is open.
|
|
// We should him to detect the breach, but realize that the funds have
|
|
// then been swept to his wallet by Willy.
|
|
err = restart()
|
|
if err != nil {
|
|
t.Fatalf("unable to restart dave: %v", err)
|
|
}
|
|
|
|
err = wait.Predicate(func() bool {
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
daveBalReq := &lnrpc.ChannelBalanceRequest{}
|
|
daveBalResp, err := dave.ChannelBalance(ctxt, daveBalReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get dave's balance: %v", err)
|
|
}
|
|
|
|
if daveBalResp.Balance != 0 {
|
|
predErr = fmt.Errorf("Dave should end up with zero "+
|
|
"channel balance, instead has %d",
|
|
daveBalResp.Balance)
|
|
return false
|
|
}
|
|
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("%v", predErr)
|
|
}
|
|
|
|
assertNumPendingChannels(t, dave, 0, 0)
|
|
|
|
err = wait.Predicate(func() bool {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
daveBalReq := &lnrpc.WalletBalanceRequest{}
|
|
daveBalResp, err := dave.WalletBalance(ctxt, daveBalReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get dave's balance: %v", err)
|
|
}
|
|
|
|
if daveBalResp.ConfirmedBalance <= davePreSweepBalance {
|
|
predErr = fmt.Errorf("Dave should have more than %d "+
|
|
"after sweep, instead has %d",
|
|
davePreSweepBalance,
|
|
daveBalResp.ConfirmedBalance)
|
|
return false
|
|
}
|
|
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("%v", predErr)
|
|
}
|
|
|
|
// Dave should have no open channels.
|
|
assertNodeNumChannels(t, dave, 0)
|
|
}
|
|
|
|
// assertNumPendingChannels checks that a PendingChannels response from the
|
|
// node reports the expected number of pending channels.
|
|
func assertNumPendingChannels(t *harnessTest, node *lntest.HarnessNode,
|
|
expWaitingClose, expPendingForceClose int) {
|
|
ctxb := context.Background()
|
|
|
|
var predErr error
|
|
err := wait.Predicate(func() bool {
|
|
pendingChansRequest := &lnrpc.PendingChannelsRequest{}
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
pendingChanResp, err := node.PendingChannels(ctxt,
|
|
pendingChansRequest)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to query for pending "+
|
|
"channels: %v", err)
|
|
return false
|
|
}
|
|
n := len(pendingChanResp.WaitingCloseChannels)
|
|
if n != expWaitingClose {
|
|
predErr = fmt.Errorf("Expected to find %d channels "+
|
|
"waiting close, found %d", expWaitingClose, n)
|
|
return false
|
|
}
|
|
n = len(pendingChanResp.PendingForceClosingChannels)
|
|
if n != expPendingForceClose {
|
|
predErr = fmt.Errorf("expected to find %d channel "+
|
|
"pending force close, found %d", expPendingForceClose, n)
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("%v", predErr)
|
|
}
|
|
}
|
|
|
|
// assertDLPExecuted asserts that Dave is a node that has recovered their state
|
|
// form scratch. Carol should then force close on chain, with Dave sweeping his
|
|
// funds immediately, and Carol sweeping her fund after her CSV delay is up. If
|
|
// the blankSlate value is true, then this means that Dave won't need to sweep
|
|
// on chain as he has no funds in the channel.
|
|
func assertDLPExecuted(net *lntest.NetworkHarness, t *harnessTest,
|
|
carol *lntest.HarnessNode, carolStartingBalance int64,
|
|
dave *lntest.HarnessNode, daveStartingBalance int64,
|
|
anchors bool) {
|
|
|
|
// We disabled auto-reconnect for some tests to avoid timing issues.
|
|
// To make sure the nodes are initiating DLP now, we have to manually
|
|
// re-connect them.
|
|
ctxb := context.Background()
|
|
err := net.ConnectNodes(ctxb, carol, dave)
|
|
if err != nil && !strings.Contains(err.Error(), "already connected") {
|
|
t.Fatalf("unable to connect Carol to Dave to initiate DLP: %v",
|
|
err)
|
|
}
|
|
|
|
// Upon reconnection, the nodes should detect that Dave is out of sync.
|
|
// Carol should force close the channel using her latest commitment.
|
|
expectedTxes := 1
|
|
if anchors {
|
|
expectedTxes = 2
|
|
}
|
|
_, err = waitForNTxsInMempool(
|
|
net.Miner.Node, expectedTxes, minerMempoolTimeout,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to find Carol's force close tx in mempool: %v",
|
|
err)
|
|
}
|
|
|
|
// Channel should be in the state "waiting close" for Carol since she
|
|
// broadcasted the force close tx.
|
|
assertNumPendingChannels(t, carol, 1, 0)
|
|
|
|
// Dave should also consider the channel "waiting close", as he noticed
|
|
// the channel was out of sync, and is now waiting for a force close to
|
|
// hit the chain.
|
|
assertNumPendingChannels(t, dave, 1, 0)
|
|
|
|
// Restart Dave to make sure he is able to sweep the funds after
|
|
// shutdown.
|
|
if err := net.RestartNode(dave, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// Generate a single block, which should confirm the closing tx.
|
|
block := mineBlocks(t, net, 1, expectedTxes)[0]
|
|
|
|
// Dave should sweep his funds immediately, as they are not timelocked.
|
|
// We also expect Dave to sweep his anchor, if present.
|
|
|
|
_, err = waitForNTxsInMempool(
|
|
net.Miner.Node, expectedTxes, minerMempoolTimeout,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to find Dave's sweep tx in mempool: %v", err)
|
|
}
|
|
|
|
// Dave should consider the channel pending force close (since he is
|
|
// waiting for his sweep to confirm).
|
|
assertNumPendingChannels(t, dave, 0, 1)
|
|
|
|
// Carol is considering it "pending force close", as we must wait
|
|
// before she can sweep her outputs.
|
|
assertNumPendingChannels(t, carol, 0, 1)
|
|
|
|
// Mine the sweep tx.
|
|
block = mineBlocks(t, net, 1, expectedTxes)[0]
|
|
|
|
// Now Dave should consider the channel fully closed.
|
|
assertNumPendingChannels(t, dave, 0, 0)
|
|
|
|
// We query Dave's balance to make sure it increased after the channel
|
|
// closed. This checks that he was able to sweep the funds he had in
|
|
// the channel.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
balReq := &lnrpc.WalletBalanceRequest{}
|
|
daveBalResp, err := dave.WalletBalance(ctxt, balReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get dave's balance: %v", err)
|
|
}
|
|
|
|
daveBalance := daveBalResp.ConfirmedBalance
|
|
if daveBalance <= daveStartingBalance {
|
|
t.Fatalf("expected dave to have balance above %d, "+
|
|
"instead had %v", daveStartingBalance, daveBalance)
|
|
}
|
|
|
|
// After the Carol's output matures, she should also reclaim her funds.
|
|
mineBlocks(t, net, defaultCSV-1, 0)
|
|
carolSweep, err := waitForTxInMempool(
|
|
net.Miner.Node, minerMempoolTimeout,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to find Carol's sweep tx in mempool: %v", err)
|
|
}
|
|
block = mineBlocks(t, net, 1, 1)[0]
|
|
assertTxInBlock(t, block, carolSweep)
|
|
|
|
// Now the channel should be fully closed also from Carol's POV.
|
|
assertNumPendingChannels(t, carol, 0, 0)
|
|
|
|
// Make sure Carol got her balance back.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolBalResp, err := carol.WalletBalance(ctxt, balReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol's balance: %v", err)
|
|
}
|
|
carolBalance := carolBalResp.ConfirmedBalance
|
|
if carolBalance <= carolStartingBalance {
|
|
t.Fatalf("expected carol to have balance above %d, "+
|
|
"instead had %v", carolStartingBalance,
|
|
carolBalance)
|
|
}
|
|
|
|
assertNodeNumChannels(t, dave, 0)
|
|
assertNodeNumChannels(t, carol, 0)
|
|
}
|
|
|
|
// testDataLossProtection tests that if one of the nodes in a channel
|
|
// relationship lost state, they will detect this during channel sync, and the
|
|
// up-to-date party will force close the channel, giving the outdated party the
|
|
// opportunity to sweep its output.
|
|
func testDataLossProtection(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
const (
|
|
chanAmt = lnd.MaxBtcFundingAmount
|
|
paymentAmt = 10000
|
|
numInvoices = 6
|
|
)
|
|
|
|
// Carol will be the up-to-date party. We set --nolisten to ensure Dave
|
|
// won't be able to connect to her and trigger the channel data
|
|
// protection logic automatically. We also can't have Carol
|
|
// automatically re-connect too early, otherwise DLP would be initiated
|
|
// at the wrong moment.
|
|
carol, err := net.NewNode(
|
|
"Carol", []string{"--nolisten", "--minbackoff=1h"},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new carol node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
// Dave will be the party losing his state.
|
|
dave, err := net.NewNode("Dave", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
// Before we make a channel, we'll load up Carol with some coins sent
|
|
// directly from the miner.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
|
|
// timeTravel is a method that will make Carol open a channel to the
|
|
// passed node, settle a series of payments, then reset the node back
|
|
// to the state before the payments happened. When this method returns
|
|
// the node will be unaware of the new state updates. The returned
|
|
// function can be used to restart the node in this state.
|
|
timeTravel := func(node *lntest.HarnessNode) (func() error,
|
|
*lnrpc.ChannelPoint, int64, error) {
|
|
|
|
// We must let the node communicate with Carol before they are
|
|
// able to open channel, so we connect them.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.EnsureConnected(ctxt, carol, node); err != nil {
|
|
t.Fatalf("unable to connect %v to carol: %v",
|
|
node.Name(), err)
|
|
}
|
|
|
|
// We'll first open up a channel between them with a 0.5 BTC
|
|
// value.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, carol, node,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// With the channel open, we'll create a few invoices for the
|
|
// node that Carol will pay to in order to advance the state of
|
|
// the channel.
|
|
// TODO(halseth): have dangling HTLCs on the commitment, able to
|
|
// retrive funds?
|
|
payReqs, _, _, err := createPayReqs(
|
|
node, paymentAmt, numInvoices,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// Wait for Carol to receive the channel edge from the funding
|
|
// manager.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("carol didn't see the carol->%s channel "+
|
|
"before timeout: %v", node.Name(), err)
|
|
}
|
|
|
|
// Send payments from Carol using 3 of the payment hashes
|
|
// generated above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, carol,
|
|
payReqs[:numInvoices/2], true)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
// Next query for the node's channel state, as we sent 3
|
|
// payments of 10k satoshis each, it should now see his balance
|
|
// as being 30k satoshis.
|
|
var nodeChan *lnrpc.Channel
|
|
var predErr error
|
|
err = wait.Predicate(func() bool {
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
bChan, err := getChanInfo(ctxt, node)
|
|
if err != nil {
|
|
t.Fatalf("unable to get channel info: %v", err)
|
|
}
|
|
if bChan.LocalBalance != 30000 {
|
|
predErr = fmt.Errorf("balance is incorrect, "+
|
|
"got %v, expected %v",
|
|
bChan.LocalBalance, 30000)
|
|
return false
|
|
}
|
|
|
|
nodeChan = bChan
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("%v", predErr)
|
|
}
|
|
|
|
// Grab the current commitment height (update number), we'll
|
|
// later revert him to this state after additional updates to
|
|
// revoke this state.
|
|
stateNumPreCopy := nodeChan.NumUpdates
|
|
|
|
// Create a temporary file to house the database state at this
|
|
// particular point in history.
|
|
tempDbPath, err := ioutil.TempDir("", node.Name()+"-past-state")
|
|
if err != nil {
|
|
t.Fatalf("unable to create temp db folder: %v", err)
|
|
}
|
|
tempDbFile := filepath.Join(tempDbPath, "channel.db")
|
|
defer os.Remove(tempDbPath)
|
|
|
|
// With the temporary file created, copy the current state into
|
|
// the temporary file we created above. Later after more
|
|
// updates, we'll restore this state.
|
|
if err := lntest.CopyFile(tempDbFile, node.DBPath()); err != nil {
|
|
t.Fatalf("unable to copy database files: %v", err)
|
|
}
|
|
|
|
// Finally, send more payments from , using the remaining
|
|
// payment hashes.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, carol,
|
|
payReqs[numInvoices/2:], true)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
nodeChan, err = getChanInfo(ctxt, node)
|
|
if err != nil {
|
|
t.Fatalf("unable to get dave chan info: %v", err)
|
|
}
|
|
|
|
// Now we shutdown the node, copying over the its temporary
|
|
// database state which has the *prior* channel state over his
|
|
// current most up to date state. With this, we essentially
|
|
// force the node to travel back in time within the channel's
|
|
// history.
|
|
if err = net.RestartNode(node, func() error {
|
|
return os.Rename(tempDbFile, node.DBPath())
|
|
}); err != nil {
|
|
t.Fatalf("unable to restart node: %v", err)
|
|
}
|
|
|
|
// Make sure the channel is still there from the PoV of the
|
|
// node.
|
|
assertNodeNumChannels(t, node, 1)
|
|
|
|
// Now query for the channel state, it should show that it's at
|
|
// a state number in the past, not the *latest* state.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
nodeChan, err = getChanInfo(ctxt, node)
|
|
if err != nil {
|
|
t.Fatalf("unable to get dave chan info: %v", err)
|
|
}
|
|
if nodeChan.NumUpdates != stateNumPreCopy {
|
|
t.Fatalf("db copy failed: %v", nodeChan.NumUpdates)
|
|
}
|
|
|
|
balReq := &lnrpc.WalletBalanceRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
balResp, err := node.WalletBalance(ctxt, balReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get dave's balance: %v", err)
|
|
}
|
|
|
|
restart, err := net.SuspendNode(node)
|
|
if err != nil {
|
|
t.Fatalf("unable to suspend node: %v", err)
|
|
}
|
|
|
|
return restart, chanPoint, balResp.ConfirmedBalance, nil
|
|
}
|
|
|
|
// Reset Dave to a state where he has an outdated channel state.
|
|
restartDave, _, daveStartingBalance, err := timeTravel(dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to time travel dave: %v", err)
|
|
}
|
|
|
|
// We make a note of the nodes' current on-chain balances, to make sure
|
|
// they are able to retrieve the channel funds eventually,
|
|
balReq := &lnrpc.WalletBalanceRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolBalResp, err := carol.WalletBalance(ctxt, balReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol's balance: %v", err)
|
|
}
|
|
carolStartingBalance := carolBalResp.ConfirmedBalance
|
|
|
|
// Restart Dave to trigger a channel resync.
|
|
if err := restartDave(); err != nil {
|
|
t.Fatalf("unable to restart dave: %v", err)
|
|
}
|
|
|
|
// Assert that once Dave comes up, they reconnect, Carol force closes
|
|
// on chain, and both of them properly carry out the DLP protocol.
|
|
assertDLPExecuted(
|
|
net, t, carol, carolStartingBalance, dave, daveStartingBalance,
|
|
false,
|
|
)
|
|
|
|
// As a second part of this test, we will test the scenario where a
|
|
// channel is closed while Dave is offline, loses his state and comes
|
|
// back online. In this case the node should attempt to resync the
|
|
// channel, and the peer should resend a channel sync message for the
|
|
// closed channel, such that Dave can retrieve his funds.
|
|
//
|
|
// We start by letting Dave time travel back to an outdated state.
|
|
restartDave, chanPoint2, daveStartingBalance, err := timeTravel(dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to time travel eve: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolBalResp, err = carol.WalletBalance(ctxt, balReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol's balance: %v", err)
|
|
}
|
|
carolStartingBalance = carolBalResp.ConfirmedBalance
|
|
|
|
// Now let Carol force close the channel while Dave is offline.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPoint2, true)
|
|
|
|
// Wait for the channel to be marked pending force close.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = waitForChannelPendingForceClose(ctxt, carol, chanPoint2)
|
|
if err != nil {
|
|
t.Fatalf("channel not pending force close: %v", err)
|
|
}
|
|
|
|
// Mine enough blocks for Carol to sweep her funds.
|
|
mineBlocks(t, net, defaultCSV, 0)
|
|
|
|
carolSweep, err := waitForTxInMempool(net.Miner.Node, minerMempoolTimeout)
|
|
if err != nil {
|
|
t.Fatalf("unable to find Carol's sweep tx in mempool: %v", err)
|
|
}
|
|
block := mineBlocks(t, net, 1, 1)[0]
|
|
assertTxInBlock(t, block, carolSweep)
|
|
|
|
// Now the channel should be fully closed also from Carol's POV.
|
|
assertNumPendingChannels(t, carol, 0, 0)
|
|
|
|
// Make sure Carol got her balance back.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolBalResp, err = carol.WalletBalance(ctxt, balReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol's balance: %v", err)
|
|
}
|
|
carolBalance := carolBalResp.ConfirmedBalance
|
|
if carolBalance <= carolStartingBalance {
|
|
t.Fatalf("expected carol to have balance above %d, "+
|
|
"instead had %v", carolStartingBalance,
|
|
carolBalance)
|
|
}
|
|
|
|
assertNodeNumChannels(t, carol, 0)
|
|
|
|
// When Dave comes online, he will reconnect to Carol, try to resync
|
|
// the channel, but it will already be closed. Carol should resend the
|
|
// information Dave needs to sweep his funds.
|
|
if err := restartDave(); err != nil {
|
|
t.Fatalf("unable to restart Eve: %v", err)
|
|
}
|
|
|
|
// Dave should sweep his funds.
|
|
_, err = waitForTxInMempool(net.Miner.Node, minerMempoolTimeout)
|
|
if err != nil {
|
|
t.Fatalf("unable to find Dave's sweep tx in mempool: %v", err)
|
|
}
|
|
|
|
// Mine a block to confirm the sweep, and make sure Dave got his
|
|
// balance back.
|
|
mineBlocks(t, net, 1, 1)
|
|
assertNodeNumChannels(t, dave, 0)
|
|
|
|
err = wait.NoError(func() error {
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
daveBalResp, err := dave.WalletBalance(ctxt, balReq)
|
|
if err != nil {
|
|
return fmt.Errorf("unable to get dave's balance: %v",
|
|
err)
|
|
}
|
|
|
|
daveBalance := daveBalResp.ConfirmedBalance
|
|
if daveBalance <= daveStartingBalance {
|
|
return fmt.Errorf("expected dave to have balance "+
|
|
"above %d, intead had %v", daveStartingBalance,
|
|
daveBalance)
|
|
}
|
|
|
|
return nil
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("%v", err)
|
|
}
|
|
}
|
|
|
|
// assertNodeNumChannels polls the provided node's list channels rpc until it
|
|
// reaches the desired number of total channels.
|
|
func assertNodeNumChannels(t *harnessTest, node *lntest.HarnessNode,
|
|
numChannels int) {
|
|
ctxb := context.Background()
|
|
|
|
// Poll node for its list of channels.
|
|
req := &lnrpc.ListChannelsRequest{}
|
|
|
|
var predErr error
|
|
pred := func() bool {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
chanInfo, err := node.ListChannels(ctxt, req)
|
|
if err != nil {
|
|
predErr = fmt.Errorf("unable to query for node's "+
|
|
"channels: %v", err)
|
|
return false
|
|
}
|
|
|
|
// Return true if the query returned the expected number of
|
|
// channels.
|
|
num := len(chanInfo.Channels)
|
|
if num != numChannels {
|
|
predErr = fmt.Errorf("expected %v channels, got %v",
|
|
numChannels, num)
|
|
return false
|
|
}
|
|
return true
|
|
}
|
|
|
|
if err := wait.Predicate(pred, time.Second*15); err != nil {
|
|
t.Fatalf("node has incorrect number of channels: %v", predErr)
|
|
}
|
|
}
|
|
|
|
func testHtlcErrorPropagation(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
// In this test we wish to exercise the daemon's correct parsing,
|
|
// handling, and propagation of errors that occur while processing a
|
|
// multi-hop payment.
|
|
const chanAmt = lnd.MaxBtcFundingAmount
|
|
|
|
// First establish a channel with a capacity of 0.5 BTC between Alice
|
|
// and Bob.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAlice := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.Alice.WaitForNetworkChannelOpen(ctxt, chanPointAlice); err != nil {
|
|
t.Fatalf("channel not seen by alice before timeout: %v", err)
|
|
}
|
|
|
|
cType, err := channelCommitType(net.Alice, chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("unable to get channel type: %v", err)
|
|
}
|
|
|
|
commitFee := cType.calcStaticFee(0)
|
|
assertBaseBalance := func() {
|
|
balReq := &lnrpc.ChannelBalanceRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
aliceBal, err := net.Alice.ChannelBalance(ctxt, balReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get channel balance: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
bobBal, err := net.Bob.ChannelBalance(ctxt, balReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get channel balance: %v", err)
|
|
}
|
|
if aliceBal.Balance != int64(chanAmt-commitFee) {
|
|
t.Fatalf("alice has an incorrect balance: expected %v got %v",
|
|
int64(chanAmt-commitFee), aliceBal)
|
|
}
|
|
if bobBal.Balance != int64(chanAmt-commitFee) {
|
|
t.Fatalf("bob has an incorrect balance: expected %v got %v",
|
|
int64(chanAmt-commitFee), bobBal)
|
|
}
|
|
}
|
|
|
|
// Since we'd like to test some multi-hop failure scenarios, we'll
|
|
// introduce another node into our test network: Carol.
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
|
|
// Next, we'll create a connection from Bob to Carol, and open a
|
|
// channel between them so we have the topology: Alice -> Bob -> Carol.
|
|
// The channel created will be of lower capacity that the one created
|
|
// above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, net.Bob, carol); err != nil {
|
|
t.Fatalf("unable to connect bob to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
const bobChanAmt = lnd.MaxBtcFundingAmount
|
|
chanPointBob := openChannelAndAssert(
|
|
ctxt, t, net, net.Bob, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Ensure that Alice has Carol in her routing table before proceeding.
|
|
nodeInfoReq := &lnrpc.NodeInfoRequest{
|
|
PubKey: carol.PubKeyStr,
|
|
}
|
|
checkTableTimeout := time.After(time.Second * 10)
|
|
checkTableTicker := time.NewTicker(100 * time.Millisecond)
|
|
defer checkTableTicker.Stop()
|
|
|
|
out:
|
|
// TODO(roasbeef): make into async hook for node announcements
|
|
for {
|
|
select {
|
|
case <-checkTableTicker.C:
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
_, err := net.Alice.GetNodeInfo(ctxt, nodeInfoReq)
|
|
if err != nil && strings.Contains(err.Error(),
|
|
"unable to find") {
|
|
|
|
continue
|
|
}
|
|
|
|
break out
|
|
case <-checkTableTimeout:
|
|
t.Fatalf("carol's node announcement didn't propagate within " +
|
|
"the timeout period")
|
|
}
|
|
}
|
|
|
|
// With the channels, open we can now start to test our multi-hop error
|
|
// scenarios. First, we'll generate an invoice from carol that we'll
|
|
// use to test some error cases.
|
|
const payAmt = 10000
|
|
invoiceReq := &lnrpc.Invoice{
|
|
Memo: "kek99",
|
|
Value: payAmt,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolInvoice, err := carol.AddInvoice(ctxt, invoiceReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate carol invoice: %v", err)
|
|
}
|
|
|
|
carolPayReq, err := carol.DecodePayReq(ctxb,
|
|
&lnrpc.PayReqString{
|
|
PayReq: carolInvoice.PaymentRequest,
|
|
})
|
|
if err != nil {
|
|
t.Fatalf("unable to decode generated payment request: %v", err)
|
|
}
|
|
|
|
// Before we send the payment, ensure that the announcement of the new
|
|
// channel has been processed by Alice.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.Alice.WaitForNetworkChannelOpen(ctxt, chanPointBob); err != nil {
|
|
t.Fatalf("channel not seen by alice before timeout: %v", err)
|
|
}
|
|
|
|
// For the first scenario, we'll test the cancellation of an HTLC with
|
|
// an unknown payment hash.
|
|
// TODO(roasbeef): return failure response rather than failing entire
|
|
// stream on payment error.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
sendReq := &lnrpc.SendRequest{
|
|
PaymentHashString: hex.EncodeToString(makeFakePayHash(t)),
|
|
DestString: hex.EncodeToString(carol.PubKey[:]),
|
|
Amt: payAmt,
|
|
FinalCltvDelta: int32(carolPayReq.CltvExpiry),
|
|
}
|
|
resp, err := net.Alice.SendPaymentSync(ctxt, sendReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
|
|
// The payment should have resulted in an error since we sent it with the
|
|
// wrong payment hash.
|
|
if resp.PaymentError == "" {
|
|
t.Fatalf("payment should have been rejected due to invalid " +
|
|
"payment hash")
|
|
}
|
|
|
|
assertLastHTLCError(
|
|
t, net.Alice,
|
|
lnrpc.Failure_INCORRECT_OR_UNKNOWN_PAYMENT_DETAILS,
|
|
)
|
|
|
|
// The balances of all parties should be the same as initially since
|
|
// the HTLC was canceled.
|
|
assertBaseBalance()
|
|
|
|
// Next, we'll test the case of a recognized payHash but, an incorrect
|
|
// value on the extended HTLC.
|
|
htlcAmt := lnwire.NewMSatFromSatoshis(1000)
|
|
sendReq = &lnrpc.SendRequest{
|
|
PaymentHashString: hex.EncodeToString(carolInvoice.RHash),
|
|
DestString: hex.EncodeToString(carol.PubKey[:]),
|
|
Amt: int64(htlcAmt.ToSatoshis()), // 10k satoshis are expected.
|
|
FinalCltvDelta: int32(carolPayReq.CltvExpiry),
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err = net.Alice.SendPaymentSync(ctxt, sendReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
|
|
// The payment should fail with an error since we sent 1k satoshis isn't of
|
|
// 10k as was requested.
|
|
if resp.PaymentError == "" {
|
|
t.Fatalf("payment should have been rejected due to wrong " +
|
|
"HTLC amount")
|
|
}
|
|
|
|
assertLastHTLCError(
|
|
t, net.Alice,
|
|
lnrpc.Failure_INCORRECT_OR_UNKNOWN_PAYMENT_DETAILS,
|
|
)
|
|
|
|
// The balances of all parties should be the same as initially since
|
|
// the HTLC was canceled.
|
|
assertBaseBalance()
|
|
|
|
// Next we'll test an error that occurs mid-route due to an outgoing
|
|
// link having insufficient capacity. In order to do so, we'll first
|
|
// need to unbalance the link connecting Bob<->Carol.
|
|
ctx, cancel := context.WithCancel(ctxb)
|
|
defer cancel()
|
|
|
|
bobPayStream, err := net.Bob.SendPayment(ctx)
|
|
if err != nil {
|
|
t.Fatalf("unable to create payment stream: %v", err)
|
|
}
|
|
|
|
// To do so, we'll push most of the funds in the channel over to
|
|
// Alice's side, leaving on 10k satoshis of available balance for bob.
|
|
// There's a max payment amount, so we'll have to do this
|
|
// incrementally.
|
|
chanReserve := int64(chanAmt / 100)
|
|
amtToSend := int64(chanAmt) - chanReserve - 20000
|
|
amtSent := int64(0)
|
|
for amtSent != amtToSend {
|
|
// We'll send in chunks of the max payment amount. If we're
|
|
// about to send too much, then we'll only send the amount
|
|
// remaining.
|
|
toSend := int64(lnd.MaxPaymentMSat.ToSatoshis())
|
|
if toSend+amtSent > amtToSend {
|
|
toSend = amtToSend - amtSent
|
|
}
|
|
|
|
invoiceReq = &lnrpc.Invoice{
|
|
Value: toSend,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolInvoice2, err := carol.AddInvoice(ctxt, invoiceReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate carol invoice: %v", err)
|
|
}
|
|
if err := bobPayStream.Send(&lnrpc.SendRequest{
|
|
PaymentRequest: carolInvoice2.PaymentRequest,
|
|
}); err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
|
|
if resp, err := bobPayStream.Recv(); err != nil {
|
|
t.Fatalf("payment stream has been closed: %v", err)
|
|
} else if resp.PaymentError != "" {
|
|
t.Fatalf("bob's payment failed: %v", resp.PaymentError)
|
|
}
|
|
|
|
amtSent += toSend
|
|
}
|
|
|
|
// At this point, Alice has 50mil satoshis on her side of the channel,
|
|
// but Bob only has 10k available on his side of the channel. So a
|
|
// payment from Alice to Carol worth 100k satoshis should fail.
|
|
invoiceReq = &lnrpc.Invoice{
|
|
Value: 100000,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolInvoice3, err := carol.AddInvoice(ctxt, invoiceReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate carol invoice: %v", err)
|
|
}
|
|
|
|
sendReq = &lnrpc.SendRequest{
|
|
PaymentRequest: carolInvoice3.PaymentRequest,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err = net.Alice.SendPaymentSync(ctxt, sendReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
if resp.PaymentError == "" {
|
|
t.Fatalf("payment should fail due to insufficient "+
|
|
"capacity: %v", err)
|
|
}
|
|
|
|
assertLastHTLCError(
|
|
t, net.Alice, lnrpc.Failure_TEMPORARY_CHANNEL_FAILURE,
|
|
)
|
|
|
|
// Generate new invoice to not pay same invoice twice.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolInvoice, err = carol.AddInvoice(ctxt, invoiceReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate carol invoice: %v", err)
|
|
}
|
|
|
|
// For our final test, we'll ensure that if a target link isn't
|
|
// available for what ever reason then the payment fails accordingly.
|
|
//
|
|
// We'll attempt to complete the original invoice we created with Carol
|
|
// above, but before we do so, Carol will go offline, resulting in a
|
|
// failed payment.
|
|
shutdownAndAssert(net, t, carol)
|
|
|
|
// Reset mission control to forget the temporary channel failure above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
_, err = net.Alice.RouterClient.ResetMissionControl(
|
|
ctxt, &routerrpc.ResetMissionControlRequest{},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to reset mission control: %v", err)
|
|
}
|
|
|
|
sendReq = &lnrpc.SendRequest{
|
|
PaymentRequest: carolInvoice.PaymentRequest,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err = net.Alice.SendPaymentSync(ctxt, sendReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
|
|
if resp.PaymentError == "" {
|
|
t.Fatalf("payment should have failed")
|
|
}
|
|
|
|
assertLastHTLCError(t, net.Alice, lnrpc.Failure_UNKNOWN_NEXT_PEER)
|
|
|
|
// Finally, immediately close the channel. This function will also
|
|
// block until the channel is closed and will additionally assert the
|
|
// relevant channel closing post conditions.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointAlice, false)
|
|
|
|
// Force close Bob's final channel.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Bob, chanPointBob, true)
|
|
|
|
// Cleanup by mining the force close and sweep transaction.
|
|
cleanupForceClose(t, net, net.Bob, chanPointBob)
|
|
}
|
|
|
|
// testRejectHTLC tests that a node can be created with the flag --rejecthtlc.
|
|
// This means that the node will reject all forwarded HTLCs but can still
|
|
// accept direct HTLCs as well as send HTLCs.
|
|
func testRejectHTLC(net *lntest.NetworkHarness, t *harnessTest) {
|
|
// RejectHTLC
|
|
// Alice ------> Carol ------> Bob
|
|
//
|
|
const chanAmt = btcutil.Amount(1000000)
|
|
ctxb := context.Background()
|
|
|
|
// Create Carol with reject htlc flag.
|
|
carol, err := net.NewNode("Carol", []string{"--rejecthtlc"})
|
|
if err != nil {
|
|
t.Fatalf("unable to create new node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
// Connect Alice to Carol.
|
|
if err := net.ConnectNodes(ctxb, net.Alice, carol); err != nil {
|
|
t.Fatalf("unable to connect alice to carol: %v", err)
|
|
}
|
|
|
|
// Connect Carol to Bob.
|
|
if err := net.ConnectNodes(ctxb, carol, net.Bob); err != nil {
|
|
t.Fatalf("unable to conenct carol to net.Bob: %v", err)
|
|
}
|
|
|
|
// Send coins to Carol.
|
|
err = net.SendCoins(ctxb, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
|
|
// Send coins to Alice.
|
|
err = net.SendCoins(ctxb, btcutil.SatoshiPerBitcent, net.Alice)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to alice: %v", err)
|
|
}
|
|
|
|
// Open a channel between Alice and Carol.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAlice := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Open a channel between Carol and Bob.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointCarol := openChannelAndAssert(
|
|
ctxt, t, net, carol, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Channel should be ready for payments.
|
|
const payAmt = 100
|
|
|
|
// Helper closure to generate a random pre image.
|
|
genPreImage := func() []byte {
|
|
preimage := make([]byte, 32)
|
|
|
|
_, err = rand.Read(preimage)
|
|
if err != nil {
|
|
t.Fatalf("unable to generate preimage: %v", err)
|
|
}
|
|
|
|
return preimage
|
|
}
|
|
|
|
// Create an invoice from Carol of 100 satoshis.
|
|
// We expect Alice to be able to pay this invoice.
|
|
preimage := genPreImage()
|
|
|
|
carolInvoice := &lnrpc.Invoice{
|
|
Memo: "testing - alice should pay carol",
|
|
RPreimage: preimage,
|
|
Value: payAmt,
|
|
}
|
|
|
|
// Carol adds the invoice to her database.
|
|
resp, err := carol.AddInvoice(ctxb, carolInvoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
|
|
// Alice pays Carols invoice.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(
|
|
ctxt, net.Alice, []string{resp.PaymentRequest}, true,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments from alice to carol: %v", err)
|
|
}
|
|
|
|
// Create an invoice from Bob of 100 satoshis.
|
|
// We expect Carol to be able to pay this invoice.
|
|
preimage = genPreImage()
|
|
|
|
bobInvoice := &lnrpc.Invoice{
|
|
Memo: "testing - carol should pay bob",
|
|
RPreimage: preimage,
|
|
Value: payAmt,
|
|
}
|
|
|
|
// Bob adds the invoice to his database.
|
|
resp, err = net.Bob.AddInvoice(ctxb, bobInvoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
|
|
// Carol pays Bobs invoice.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(
|
|
ctxt, carol, []string{resp.PaymentRequest}, true,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments from carol to bob: %v", err)
|
|
}
|
|
|
|
// Create an invoice from Bob of 100 satoshis.
|
|
// Alice attempts to pay Bob but this should fail, since we are
|
|
// using Carol as a hop and her node will reject onward HTLCs.
|
|
preimage = genPreImage()
|
|
|
|
bobInvoice = &lnrpc.Invoice{
|
|
Memo: "testing - alice tries to pay bob",
|
|
RPreimage: preimage,
|
|
Value: payAmt,
|
|
}
|
|
|
|
// Bob adds the invoice to his database.
|
|
resp, err = net.Bob.AddInvoice(ctxb, bobInvoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
|
|
// Alice attempts to pay Bobs invoice. This payment should be rejected since
|
|
// we are using Carol as an intermediary hop, Carol is running lnd with
|
|
// --rejecthtlc.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(
|
|
ctxt, net.Alice, []string{resp.PaymentRequest}, true,
|
|
)
|
|
if err == nil {
|
|
t.Fatalf(
|
|
"should have been rejected, carol will not accept forwarded htlcs",
|
|
)
|
|
}
|
|
|
|
assertLastHTLCError(t, net.Alice, lnrpc.Failure_CHANNEL_DISABLED)
|
|
|
|
// Close all channels.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointAlice, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPointCarol, false)
|
|
}
|
|
|
|
// graphSubscription houses the proxied update and error chans for a node's
|
|
// graph subscriptions.
|
|
type graphSubscription struct {
|
|
updateChan chan *lnrpc.GraphTopologyUpdate
|
|
errChan chan error
|
|
quit chan struct{}
|
|
}
|
|
|
|
// subscribeGraphNotifications subscribes to channel graph updates and launches
|
|
// a goroutine that forwards these to the returned channel.
|
|
func subscribeGraphNotifications(t *harnessTest, ctxb context.Context,
|
|
node *lntest.HarnessNode) graphSubscription {
|
|
|
|
// We'll first start by establishing a notification client which will
|
|
// send us notifications upon detected changes in the channel graph.
|
|
req := &lnrpc.GraphTopologySubscription{}
|
|
ctx, cancelFunc := context.WithCancel(ctxb)
|
|
topologyClient, err := node.SubscribeChannelGraph(ctx, req)
|
|
if err != nil {
|
|
t.Fatalf("unable to create topology client: %v", err)
|
|
}
|
|
|
|
// We'll launch a goroutine that will be responsible for proxying all
|
|
// notifications recv'd from the client into the channel below.
|
|
errChan := make(chan error, 1)
|
|
quit := make(chan struct{})
|
|
graphUpdates := make(chan *lnrpc.GraphTopologyUpdate, 20)
|
|
go func() {
|
|
for {
|
|
defer cancelFunc()
|
|
|
|
select {
|
|
case <-quit:
|
|
return
|
|
default:
|
|
graphUpdate, err := topologyClient.Recv()
|
|
select {
|
|
case <-quit:
|
|
return
|
|
default:
|
|
}
|
|
|
|
if err == io.EOF {
|
|
return
|
|
} else if err != nil {
|
|
select {
|
|
case errChan <- err:
|
|
case <-quit:
|
|
}
|
|
return
|
|
}
|
|
|
|
select {
|
|
case graphUpdates <- graphUpdate:
|
|
case <-quit:
|
|
return
|
|
}
|
|
}
|
|
}
|
|
}()
|
|
|
|
return graphSubscription{
|
|
updateChan: graphUpdates,
|
|
errChan: errChan,
|
|
quit: quit,
|
|
}
|
|
}
|
|
|
|
func testGraphTopologyNotifications(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const chanAmt = lnd.MaxBtcFundingAmount
|
|
|
|
// Let Alice subscribe to graph notifications.
|
|
graphSub := subscribeGraphNotifications(
|
|
t, ctxb, net.Alice,
|
|
)
|
|
defer close(graphSub.quit)
|
|
|
|
// Open a new channel between Alice and Bob.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// The channel opening above should have triggered a few notifications
|
|
// sent to the notification client. We'll expect two channel updates,
|
|
// and two node announcements.
|
|
var numChannelUpds int
|
|
var numNodeAnns int
|
|
for numChannelUpds < 2 && numNodeAnns < 2 {
|
|
select {
|
|
// Ensure that a new update for both created edges is properly
|
|
// dispatched to our registered client.
|
|
case graphUpdate := <-graphSub.updateChan:
|
|
// Process all channel updates prsented in this update
|
|
// message.
|
|
for _, chanUpdate := range graphUpdate.ChannelUpdates {
|
|
switch chanUpdate.AdvertisingNode {
|
|
case net.Alice.PubKeyStr:
|
|
case net.Bob.PubKeyStr:
|
|
default:
|
|
t.Fatalf("unknown advertising node: %v",
|
|
chanUpdate.AdvertisingNode)
|
|
}
|
|
switch chanUpdate.ConnectingNode {
|
|
case net.Alice.PubKeyStr:
|
|
case net.Bob.PubKeyStr:
|
|
default:
|
|
t.Fatalf("unknown connecting node: %v",
|
|
chanUpdate.ConnectingNode)
|
|
}
|
|
|
|
if chanUpdate.Capacity != int64(chanAmt) {
|
|
t.Fatalf("channel capacities mismatch:"+
|
|
" expected %v, got %v", chanAmt,
|
|
btcutil.Amount(chanUpdate.Capacity))
|
|
}
|
|
numChannelUpds++
|
|
}
|
|
|
|
for _, nodeUpdate := range graphUpdate.NodeUpdates {
|
|
switch nodeUpdate.IdentityKey {
|
|
case net.Alice.PubKeyStr:
|
|
case net.Bob.PubKeyStr:
|
|
default:
|
|
t.Fatalf("unknown node: %v",
|
|
nodeUpdate.IdentityKey)
|
|
}
|
|
numNodeAnns++
|
|
}
|
|
case err := <-graphSub.errChan:
|
|
t.Fatalf("unable to recv graph update: %v", err)
|
|
case <-time.After(time.Second * 10):
|
|
t.Fatalf("timeout waiting for graph notifications, "+
|
|
"only received %d/2 chanupds and %d/2 nodeanns",
|
|
numChannelUpds, numNodeAnns)
|
|
}
|
|
}
|
|
|
|
_, blockHeight, err := net.Miner.Node.GetBestBlock()
|
|
if err != nil {
|
|
t.Fatalf("unable to get current blockheight %v", err)
|
|
}
|
|
|
|
// Now we'll test that updates are properly sent after channels are closed
|
|
// within the network.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPoint, false)
|
|
|
|
// Now that the channel has been closed, we should receive a
|
|
// notification indicating so.
|
|
out:
|
|
for {
|
|
select {
|
|
case graphUpdate := <-graphSub.updateChan:
|
|
if len(graphUpdate.ClosedChans) != 1 {
|
|
continue
|
|
}
|
|
|
|
closedChan := graphUpdate.ClosedChans[0]
|
|
if closedChan.ClosedHeight != uint32(blockHeight+1) {
|
|
t.Fatalf("close heights of channel mismatch: "+
|
|
"expected %v, got %v", blockHeight+1,
|
|
closedChan.ClosedHeight)
|
|
}
|
|
chanPointTxid, err := lnd.GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
closedChanTxid, err := lnd.GetChanPointFundingTxid(
|
|
closedChan.ChanPoint,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
if !bytes.Equal(closedChanTxid[:], chanPointTxid[:]) {
|
|
t.Fatalf("channel point hash mismatch: "+
|
|
"expected %v, got %v", chanPointTxid,
|
|
closedChanTxid)
|
|
}
|
|
if closedChan.ChanPoint.OutputIndex != chanPoint.OutputIndex {
|
|
t.Fatalf("output index mismatch: expected %v, "+
|
|
"got %v", chanPoint.OutputIndex,
|
|
closedChan.ChanPoint)
|
|
}
|
|
|
|
break out
|
|
|
|
case err := <-graphSub.errChan:
|
|
t.Fatalf("unable to recv graph update: %v", err)
|
|
case <-time.After(time.Second * 10):
|
|
t.Fatalf("notification for channel closure not " +
|
|
"sent")
|
|
}
|
|
}
|
|
|
|
// For the final portion of the test, we'll ensure that once a new node
|
|
// appears in the network, the proper notification is dispatched. Note
|
|
// that a node that does not have any channels open is ignored, so first
|
|
// we disconnect Alice and Bob, open a channel between Bob and Carol,
|
|
// and finally connect Alice to Bob again.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.DisconnectNodes(ctxt, net.Alice, net.Bob); err != nil {
|
|
t.Fatalf("unable to disconnect alice and bob: %v", err)
|
|
}
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, net.Bob, carol); err != nil {
|
|
t.Fatalf("unable to connect bob to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint = openChannelAndAssert(
|
|
ctxt, t, net, net.Bob, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Reconnect Alice and Bob. This should result in the nodes syncing up
|
|
// their respective graph state, with the new addition being the
|
|
// existence of Carol in the graph, and also the channel between Bob
|
|
// and Carol. Note that we will also receive a node announcement from
|
|
// Bob, since a node will update its node announcement after a new
|
|
// channel is opened.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.EnsureConnected(ctxt, net.Alice, net.Bob); err != nil {
|
|
t.Fatalf("unable to connect alice to bob: %v", err)
|
|
}
|
|
|
|
// We should receive an update advertising the newly connected node,
|
|
// Bob's new node announcement, and the channel between Bob and Carol.
|
|
numNodeAnns = 0
|
|
numChannelUpds = 0
|
|
for numChannelUpds < 2 && numNodeAnns < 1 {
|
|
select {
|
|
case graphUpdate := <-graphSub.updateChan:
|
|
for _, nodeUpdate := range graphUpdate.NodeUpdates {
|
|
switch nodeUpdate.IdentityKey {
|
|
case carol.PubKeyStr:
|
|
case net.Bob.PubKeyStr:
|
|
default:
|
|
t.Fatalf("unknown node update pubey: %v",
|
|
nodeUpdate.IdentityKey)
|
|
}
|
|
numNodeAnns++
|
|
}
|
|
|
|
for _, chanUpdate := range graphUpdate.ChannelUpdates {
|
|
switch chanUpdate.AdvertisingNode {
|
|
case carol.PubKeyStr:
|
|
case net.Bob.PubKeyStr:
|
|
default:
|
|
t.Fatalf("unknown advertising node: %v",
|
|
chanUpdate.AdvertisingNode)
|
|
}
|
|
switch chanUpdate.ConnectingNode {
|
|
case carol.PubKeyStr:
|
|
case net.Bob.PubKeyStr:
|
|
default:
|
|
t.Fatalf("unknown connecting node: %v",
|
|
chanUpdate.ConnectingNode)
|
|
}
|
|
|
|
if chanUpdate.Capacity != int64(chanAmt) {
|
|
t.Fatalf("channel capacities mismatch:"+
|
|
" expected %v, got %v", chanAmt,
|
|
btcutil.Amount(chanUpdate.Capacity))
|
|
}
|
|
numChannelUpds++
|
|
}
|
|
case err := <-graphSub.errChan:
|
|
t.Fatalf("unable to recv graph update: %v", err)
|
|
case <-time.After(time.Second * 10):
|
|
t.Fatalf("timeout waiting for graph notifications, "+
|
|
"only received %d/2 chanupds and %d/2 nodeanns",
|
|
numChannelUpds, numNodeAnns)
|
|
}
|
|
}
|
|
|
|
// Close the channel between Bob and Carol.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Bob, chanPoint, false)
|
|
}
|
|
|
|
// testNodeAnnouncement ensures that when a node is started with one or more
|
|
// external IP addresses specified on the command line, that those addresses
|
|
// announced to the network and reported in the network graph.
|
|
func testNodeAnnouncement(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
aliceSub := subscribeGraphNotifications(t, ctxb, net.Alice)
|
|
defer close(aliceSub.quit)
|
|
|
|
advertisedAddrs := []string{
|
|
"192.168.1.1:8333",
|
|
"[2001:db8:85a3:8d3:1319:8a2e:370:7348]:8337",
|
|
"bkb6azqggsaiskzi.onion:9735",
|
|
"fomvuglh6h6vcag73xo5t5gv56ombih3zr2xvplkpbfd7wrog4swjwid.onion:1234",
|
|
}
|
|
|
|
var lndArgs []string
|
|
for _, addr := range advertisedAddrs {
|
|
lndArgs = append(lndArgs, "--externalip="+addr)
|
|
}
|
|
|
|
dave, err := net.NewNode("Dave", lndArgs)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
// We must let Dave have an open channel before he can send a node
|
|
// announcement, so we open a channel with Bob,
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, net.Bob, dave); err != nil {
|
|
t.Fatalf("unable to connect bob to carol: %v", err)
|
|
}
|
|
|
|
// Alice shouldn't receive any new updates yet since the channel has yet
|
|
// to be opened.
|
|
select {
|
|
case <-aliceSub.updateChan:
|
|
t.Fatalf("received unexpected update from dave")
|
|
case <-time.After(time.Second):
|
|
}
|
|
|
|
// We'll then go ahead and open a channel between Bob and Dave. This
|
|
// ensures that Alice receives the node announcement from Bob as part of
|
|
// the announcement broadcast.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, net.Bob, dave,
|
|
lntest.OpenChannelParams{
|
|
Amt: 1000000,
|
|
},
|
|
)
|
|
|
|
assertAddrs := func(addrsFound []string, targetAddrs ...string) {
|
|
addrs := make(map[string]struct{}, len(addrsFound))
|
|
for _, addr := range addrsFound {
|
|
addrs[addr] = struct{}{}
|
|
}
|
|
|
|
for _, addr := range targetAddrs {
|
|
if _, ok := addrs[addr]; !ok {
|
|
t.Fatalf("address %v not found in node "+
|
|
"announcement", addr)
|
|
}
|
|
}
|
|
}
|
|
|
|
waitForAddrsInUpdate := func(graphSub graphSubscription,
|
|
nodePubKey string, targetAddrs ...string) {
|
|
|
|
for {
|
|
select {
|
|
case graphUpdate := <-graphSub.updateChan:
|
|
for _, update := range graphUpdate.NodeUpdates {
|
|
if update.IdentityKey == nodePubKey {
|
|
assertAddrs(
|
|
update.Addresses,
|
|
targetAddrs...,
|
|
)
|
|
return
|
|
}
|
|
}
|
|
case err := <-graphSub.errChan:
|
|
t.Fatalf("unable to recv graph update: %v", err)
|
|
case <-time.After(20 * time.Second):
|
|
t.Fatalf("did not receive node ann update")
|
|
}
|
|
}
|
|
}
|
|
|
|
// We'll then wait for Alice to receive Dave's node announcement
|
|
// including the expected advertised addresses from Bob since they
|
|
// should already be connected.
|
|
waitForAddrsInUpdate(
|
|
aliceSub, dave.PubKeyStr, advertisedAddrs...,
|
|
)
|
|
|
|
// Close the channel between Bob and Dave.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Bob, chanPoint, false)
|
|
}
|
|
|
|
func testNodeSignVerify(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
chanAmt := lnd.MaxBtcFundingAmount
|
|
pushAmt := btcutil.Amount(100000)
|
|
|
|
// Create a channel between alice and bob.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
aliceBobCh := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
|
|
aliceMsg := []byte("alice msg")
|
|
|
|
// alice signs "alice msg" and sends her signature to bob.
|
|
sigReq := &lnrpc.SignMessageRequest{Msg: aliceMsg}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
sigResp, err := net.Alice.SignMessage(ctxt, sigReq)
|
|
if err != nil {
|
|
t.Fatalf("SignMessage rpc call failed: %v", err)
|
|
}
|
|
aliceSig := sigResp.Signature
|
|
|
|
// bob verifying alice's signature should succeed since alice and bob are
|
|
// connected.
|
|
verifyReq := &lnrpc.VerifyMessageRequest{Msg: aliceMsg, Signature: aliceSig}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
verifyResp, err := net.Bob.VerifyMessage(ctxt, verifyReq)
|
|
if err != nil {
|
|
t.Fatalf("VerifyMessage failed: %v", err)
|
|
}
|
|
if !verifyResp.Valid {
|
|
t.Fatalf("alice's signature didn't validate")
|
|
}
|
|
if verifyResp.Pubkey != net.Alice.PubKeyStr {
|
|
t.Fatalf("alice's signature doesn't contain alice's pubkey.")
|
|
}
|
|
|
|
// carol is a new node that is unconnected to alice or bob.
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
carolMsg := []byte("carol msg")
|
|
|
|
// carol signs "carol msg" and sends her signature to bob.
|
|
sigReq = &lnrpc.SignMessageRequest{Msg: carolMsg}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
sigResp, err = carol.SignMessage(ctxt, sigReq)
|
|
if err != nil {
|
|
t.Fatalf("SignMessage rpc call failed: %v", err)
|
|
}
|
|
carolSig := sigResp.Signature
|
|
|
|
// bob verifying carol's signature should fail since they are not connected.
|
|
verifyReq = &lnrpc.VerifyMessageRequest{Msg: carolMsg, Signature: carolSig}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
verifyResp, err = net.Bob.VerifyMessage(ctxt, verifyReq)
|
|
if err != nil {
|
|
t.Fatalf("VerifyMessage failed: %v", err)
|
|
}
|
|
if verifyResp.Valid {
|
|
t.Fatalf("carol's signature should not be valid")
|
|
}
|
|
if verifyResp.Pubkey != carol.PubKeyStr {
|
|
t.Fatalf("carol's signature doesn't contain her pubkey")
|
|
}
|
|
|
|
// Close the channel between alice and bob.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, aliceBobCh, false)
|
|
}
|
|
|
|
// testAsyncPayments tests the performance of the async payments.
|
|
func testAsyncPayments(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const (
|
|
paymentAmt = 100
|
|
)
|
|
|
|
// First establish a channel with a capacity equals to the overall
|
|
// amount of payments, between Alice and Bob, at the end of the test
|
|
// Alice should send all money from her side to Bob.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
channelCapacity := btcutil.Amount(paymentAmt * 2000)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: channelCapacity,
|
|
},
|
|
)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
info, err := getChanInfo(ctxt, net.Alice)
|
|
if err != nil {
|
|
t.Fatalf("unable to get alice channel info: %v", err)
|
|
}
|
|
|
|
// We'll create a number of invoices equal the max number of HTLCs that
|
|
// can be carried in one direction. The number on the commitment will
|
|
// likely be lower, but we can't guarantee that any more HTLCs will
|
|
// succeed due to the limited path diversity and inability of the router
|
|
// to retry via another path.
|
|
numInvoices := int(input.MaxHTLCNumber / 2)
|
|
|
|
bobAmt := int64(numInvoices * paymentAmt)
|
|
aliceAmt := info.LocalBalance - bobAmt
|
|
|
|
// With the channel open, we'll create invoices for Bob that Alice
|
|
// will pay to in order to advance the state of the channel.
|
|
bobPayReqs, _, _, err := createPayReqs(
|
|
net.Bob, paymentAmt, numInvoices,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// Wait for Alice to receive the channel edge from the funding manager.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Alice.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't see the alice->bob channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// Open up a payment stream to Alice that we'll use to send payment to
|
|
// Bob. We also create a small helper function to send payments to Bob,
|
|
// consuming the payment hashes we generated above.
|
|
ctxt, _ = context.WithTimeout(ctxb, lntest.AsyncBenchmarkTimeout)
|
|
alicePayStream, err := net.Alice.SendPayment(ctxt)
|
|
if err != nil {
|
|
t.Fatalf("unable to create payment stream for alice: %v", err)
|
|
}
|
|
|
|
// Send payments from Alice to Bob using of Bob's payment hashes
|
|
// generated above.
|
|
now := time.Now()
|
|
for i := 0; i < numInvoices; i++ {
|
|
sendReq := &lnrpc.SendRequest{
|
|
PaymentRequest: bobPayReqs[i],
|
|
}
|
|
|
|
if err := alicePayStream.Send(sendReq); err != nil {
|
|
t.Fatalf("unable to send payment: "+
|
|
"stream has been closed: %v", err)
|
|
}
|
|
}
|
|
|
|
// Wait until all the payments have settled.
|
|
for i := 0; i < numInvoices; i++ {
|
|
if _, err := alicePayStream.Recv(); err != nil {
|
|
t.Fatalf("payment stream have been closed: %v", err)
|
|
}
|
|
}
|
|
|
|
// All payments have been sent, mark the finish time.
|
|
timeTaken := time.Since(now)
|
|
|
|
// Next query for Bob's and Alice's channel states, in order to confirm
|
|
// that all payment have been successful transmitted.
|
|
|
|
// Wait for the revocation to be received so alice no longer has pending
|
|
// htlcs listed and has correct balances. This is needed due to the fact
|
|
// that we now pipeline the settles.
|
|
err = wait.Predicate(func() bool {
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
aliceChan, err := getChanInfo(ctxt, net.Alice)
|
|
if err != nil {
|
|
return false
|
|
}
|
|
if len(aliceChan.PendingHtlcs) != 0 {
|
|
return false
|
|
}
|
|
if aliceChan.RemoteBalance != bobAmt {
|
|
return false
|
|
}
|
|
if aliceChan.LocalBalance != aliceAmt {
|
|
return false
|
|
}
|
|
|
|
return true
|
|
}, time.Second*5)
|
|
if err != nil {
|
|
t.Fatalf("failed to assert alice's pending htlcs and/or remote/local balance")
|
|
}
|
|
|
|
// Wait for Bob to receive revocation from Alice.
|
|
time.Sleep(2 * time.Second)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
bobChan, err := getChanInfo(ctxt, net.Bob)
|
|
if err != nil {
|
|
t.Fatalf("unable to get bob's channel info: %v", err)
|
|
}
|
|
if len(bobChan.PendingHtlcs) != 0 {
|
|
t.Fatalf("bob's pending htlcs is incorrect, got %v, "+
|
|
"expected %v", len(bobChan.PendingHtlcs), 0)
|
|
}
|
|
if bobChan.LocalBalance != bobAmt {
|
|
t.Fatalf("bob's local balance is incorrect, got %v, expected"+
|
|
" %v", bobChan.LocalBalance, bobAmt)
|
|
}
|
|
if bobChan.RemoteBalance != aliceAmt {
|
|
t.Fatalf("bob's remote balance is incorrect, got %v, "+
|
|
"expected %v", bobChan.RemoteBalance, aliceAmt)
|
|
}
|
|
|
|
t.Log("\tBenchmark info: Elapsed time: ", timeTaken)
|
|
t.Log("\tBenchmark info: TPS: ", float64(numInvoices)/float64(timeTaken.Seconds()))
|
|
|
|
// Finally, immediately close the channel. This function will also
|
|
// block until the channel is closed and will additionally assert the
|
|
// relevant channel closing post conditions.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPoint, false)
|
|
}
|
|
|
|
// testBidirectionalAsyncPayments tests that nodes are able to send the
|
|
// payments to each other in async manner without blocking.
|
|
func testBidirectionalAsyncPayments(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const (
|
|
paymentAmt = 1000
|
|
)
|
|
|
|
// First establish a channel with a capacity equals to the overall
|
|
// amount of payments, between Alice and Bob, at the end of the test
|
|
// Alice should send all money from her side to Bob.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: paymentAmt * 2000,
|
|
PushAmt: paymentAmt * 1000,
|
|
},
|
|
)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
info, err := getChanInfo(ctxt, net.Alice)
|
|
if err != nil {
|
|
t.Fatalf("unable to get alice channel info: %v", err)
|
|
}
|
|
|
|
// We'll create a number of invoices equal the max number of HTLCs that
|
|
// can be carried in one direction. The number on the commitment will
|
|
// likely be lower, but we can't guarantee that any more HTLCs will
|
|
// succeed due to the limited path diversity and inability of the router
|
|
// to retry via another path.
|
|
numInvoices := int(input.MaxHTLCNumber / 2)
|
|
|
|
// Nodes should exchange the same amount of money and because of this
|
|
// at the end balances should remain the same.
|
|
aliceAmt := info.LocalBalance
|
|
bobAmt := info.RemoteBalance
|
|
|
|
// With the channel open, we'll create invoices for Bob that Alice
|
|
// will pay to in order to advance the state of the channel.
|
|
bobPayReqs, _, _, err := createPayReqs(
|
|
net.Bob, paymentAmt, numInvoices,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// With the channel open, we'll create invoices for Alice that Bob
|
|
// will pay to in order to advance the state of the channel.
|
|
alicePayReqs, _, _, err := createPayReqs(
|
|
net.Alice, paymentAmt, numInvoices,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// Wait for Alice to receive the channel edge from the funding manager.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err = net.Alice.WaitForNetworkChannelOpen(ctxt, chanPoint); err != nil {
|
|
t.Fatalf("alice didn't see the alice->bob channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
if err = net.Bob.WaitForNetworkChannelOpen(ctxt, chanPoint); err != nil {
|
|
t.Fatalf("bob didn't see the bob->alice channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// Reset mission control to prevent previous payment results from
|
|
// interfering with this test. A new channel has been opened, but
|
|
// mission control operates on node pairs.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
_, err = net.Alice.RouterClient.ResetMissionControl(
|
|
ctxt, &routerrpc.ResetMissionControlRequest{},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to reset mc for alice: %v", err)
|
|
}
|
|
|
|
// Open up a payment streams to Alice and to Bob, that we'll use to
|
|
// send payment between nodes.
|
|
ctx, cancel := context.WithTimeout(ctxb, lntest.AsyncBenchmarkTimeout)
|
|
defer cancel()
|
|
|
|
alicePayStream, err := net.Alice.SendPayment(ctx)
|
|
if err != nil {
|
|
t.Fatalf("unable to create payment stream for alice: %v", err)
|
|
}
|
|
|
|
ctx, cancel = context.WithTimeout(ctxb, lntest.AsyncBenchmarkTimeout)
|
|
defer cancel()
|
|
|
|
bobPayStream, err := net.Bob.SendPayment(ctx)
|
|
if err != nil {
|
|
t.Fatalf("unable to create payment stream for bob: %v", err)
|
|
}
|
|
|
|
// Send payments from Alice to Bob and from Bob to Alice in async
|
|
// manner.
|
|
for i := 0; i < numInvoices; i++ {
|
|
aliceSendReq := &lnrpc.SendRequest{
|
|
PaymentRequest: bobPayReqs[i],
|
|
}
|
|
|
|
bobSendReq := &lnrpc.SendRequest{
|
|
PaymentRequest: alicePayReqs[i],
|
|
}
|
|
|
|
if err := alicePayStream.Send(aliceSendReq); err != nil {
|
|
t.Fatalf("unable to send payment: "+
|
|
"%v", err)
|
|
}
|
|
|
|
if err := bobPayStream.Send(bobSendReq); err != nil {
|
|
t.Fatalf("unable to send payment: "+
|
|
"%v", err)
|
|
}
|
|
}
|
|
|
|
errChan := make(chan error)
|
|
go func() {
|
|
for i := 0; i < numInvoices; i++ {
|
|
if resp, err := alicePayStream.Recv(); err != nil {
|
|
errChan <- errors.Errorf("payment stream has"+
|
|
" been closed: %v", err)
|
|
return
|
|
} else if resp.PaymentError != "" {
|
|
errChan <- errors.Errorf("unable to send "+
|
|
"payment from alice to bob: %v",
|
|
resp.PaymentError)
|
|
return
|
|
}
|
|
}
|
|
errChan <- nil
|
|
}()
|
|
|
|
go func() {
|
|
for i := 0; i < numInvoices; i++ {
|
|
if resp, err := bobPayStream.Recv(); err != nil {
|
|
errChan <- errors.Errorf("payment stream has"+
|
|
" been closed: %v", err)
|
|
return
|
|
} else if resp.PaymentError != "" {
|
|
errChan <- errors.Errorf("unable to send "+
|
|
"payment from bob to alice: %v",
|
|
resp.PaymentError)
|
|
return
|
|
}
|
|
}
|
|
errChan <- nil
|
|
}()
|
|
|
|
// Wait for Alice and Bob receive their payments, and throw and error
|
|
// if something goes wrong.
|
|
for i := 0; i < 2; i++ {
|
|
select {
|
|
case err := <-errChan:
|
|
if err != nil {
|
|
t.Fatalf(err.Error())
|
|
}
|
|
case <-time.After(lntest.AsyncBenchmarkTimeout):
|
|
t.Fatalf("waiting for payments to finish too long "+
|
|
"(%v)", lntest.AsyncBenchmarkTimeout)
|
|
}
|
|
}
|
|
|
|
// Wait for Alice and Bob to receive revocations messages, and update
|
|
// states, i.e. balance info.
|
|
time.Sleep(1 * time.Second)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
aliceInfo, err := getChanInfo(ctxt, net.Alice)
|
|
if err != nil {
|
|
t.Fatalf("unable to get bob's channel info: %v", err)
|
|
}
|
|
if aliceInfo.RemoteBalance != bobAmt {
|
|
t.Fatalf("alice's remote balance is incorrect, got %v, "+
|
|
"expected %v", aliceInfo.RemoteBalance, bobAmt)
|
|
}
|
|
if aliceInfo.LocalBalance != aliceAmt {
|
|
t.Fatalf("alice's local balance is incorrect, got %v, "+
|
|
"expected %v", aliceInfo.LocalBalance, aliceAmt)
|
|
}
|
|
if len(aliceInfo.PendingHtlcs) != 0 {
|
|
t.Fatalf("alice's pending htlcs is incorrect, got %v, "+
|
|
"expected %v", len(aliceInfo.PendingHtlcs), 0)
|
|
}
|
|
|
|
// Next query for Bob's and Alice's channel states, in order to confirm
|
|
// that all payment have been successful transmitted.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
bobInfo, err := getChanInfo(ctxt, net.Bob)
|
|
if err != nil {
|
|
t.Fatalf("unable to get bob's channel info: %v", err)
|
|
}
|
|
|
|
if bobInfo.LocalBalance != bobAmt {
|
|
t.Fatalf("bob's local balance is incorrect, got %v, expected"+
|
|
" %v", bobInfo.LocalBalance, bobAmt)
|
|
}
|
|
if bobInfo.RemoteBalance != aliceAmt {
|
|
t.Fatalf("bob's remote balance is incorrect, got %v, "+
|
|
"expected %v", bobInfo.RemoteBalance, aliceAmt)
|
|
}
|
|
if len(bobInfo.PendingHtlcs) != 0 {
|
|
t.Fatalf("bob's pending htlcs is incorrect, got %v, "+
|
|
"expected %v", len(bobInfo.PendingHtlcs), 0)
|
|
}
|
|
|
|
// Finally, immediately close the channel. This function will also
|
|
// block until the channel is closed and will additionally assert the
|
|
// relevant channel closing post conditions.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPoint, false)
|
|
}
|
|
|
|
// assertActiveHtlcs makes sure all the passed nodes have the _exact_ HTLCs
|
|
// matching payHashes on _all_ their channels.
|
|
func assertActiveHtlcs(nodes []*lntest.HarnessNode, payHashes ...[]byte) error {
|
|
ctxb := context.Background()
|
|
|
|
req := &lnrpc.ListChannelsRequest{}
|
|
for _, node := range nodes {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
nodeChans, err := node.ListChannels(ctxt, req)
|
|
if err != nil {
|
|
return fmt.Errorf("unable to get node chans: %v", err)
|
|
}
|
|
|
|
for _, channel := range nodeChans.Channels {
|
|
// Record all payment hashes active for this channel.
|
|
htlcHashes := make(map[string]struct{})
|
|
for _, htlc := range channel.PendingHtlcs {
|
|
_, ok := htlcHashes[string(htlc.HashLock)]
|
|
if ok {
|
|
return fmt.Errorf("duplicate HashLock")
|
|
}
|
|
htlcHashes[string(htlc.HashLock)] = struct{}{}
|
|
}
|
|
|
|
// Channel should have exactly the payHashes active.
|
|
if len(payHashes) != len(htlcHashes) {
|
|
return fmt.Errorf("node %x had %v htlcs active, "+
|
|
"expected %v", node.PubKey[:],
|
|
len(htlcHashes), len(payHashes))
|
|
}
|
|
|
|
// Make sure all the payHashes are active.
|
|
for _, payHash := range payHashes {
|
|
if _, ok := htlcHashes[string(payHash)]; ok {
|
|
continue
|
|
}
|
|
return fmt.Errorf("node %x didn't have the "+
|
|
"payHash %v active", node.PubKey[:],
|
|
payHash)
|
|
}
|
|
}
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
func assertNumActiveHtlcsChanPoint(node *lntest.HarnessNode,
|
|
chanPoint wire.OutPoint, numHtlcs int) error {
|
|
ctxb := context.Background()
|
|
|
|
req := &lnrpc.ListChannelsRequest{}
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
nodeChans, err := node.ListChannels(ctxt, req)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
for _, channel := range nodeChans.Channels {
|
|
if channel.ChannelPoint != chanPoint.String() {
|
|
continue
|
|
}
|
|
|
|
if len(channel.PendingHtlcs) != numHtlcs {
|
|
return fmt.Errorf("expected %v active HTLCs, got %v",
|
|
numHtlcs, len(channel.PendingHtlcs))
|
|
}
|
|
return nil
|
|
}
|
|
|
|
return fmt.Errorf("channel point %v not found", chanPoint)
|
|
}
|
|
|
|
func assertNumActiveHtlcs(nodes []*lntest.HarnessNode, numHtlcs int) error {
|
|
ctxb := context.Background()
|
|
|
|
req := &lnrpc.ListChannelsRequest{}
|
|
for _, node := range nodes {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
nodeChans, err := node.ListChannels(ctxt, req)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
for _, channel := range nodeChans.Channels {
|
|
if len(channel.PendingHtlcs) != numHtlcs {
|
|
return fmt.Errorf("expected %v HTLCs, got %v",
|
|
numHtlcs, len(channel.PendingHtlcs))
|
|
}
|
|
}
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
func assertSpendingTxInMempool(t *harnessTest, miner *rpcclient.Client,
|
|
timeout time.Duration, chanPoint wire.OutPoint) chainhash.Hash {
|
|
|
|
tx := getSpendingTxInMempool(t, miner, timeout, chanPoint)
|
|
return tx.TxHash()
|
|
}
|
|
|
|
// getSpendingTxInMempool waits for a transaction spending the given outpoint to
|
|
// appear in the mempool and returns that tx in full.
|
|
func getSpendingTxInMempool(t *harnessTest, miner *rpcclient.Client,
|
|
timeout time.Duration, chanPoint wire.OutPoint) *wire.MsgTx {
|
|
|
|
breakTimeout := time.After(timeout)
|
|
ticker := time.NewTicker(50 * time.Millisecond)
|
|
defer ticker.Stop()
|
|
|
|
for {
|
|
select {
|
|
case <-breakTimeout:
|
|
t.Fatalf("didn't find tx in mempool")
|
|
case <-ticker.C:
|
|
mempool, err := miner.GetRawMempool()
|
|
if err != nil {
|
|
t.Fatalf("unable to get mempool: %v", err)
|
|
}
|
|
|
|
if len(mempool) == 0 {
|
|
continue
|
|
}
|
|
|
|
for _, txid := range mempool {
|
|
tx, err := miner.GetRawTransaction(txid)
|
|
if err != nil {
|
|
t.Fatalf("unable to fetch tx: %v", err)
|
|
}
|
|
|
|
msgTx := tx.MsgTx()
|
|
for _, txIn := range msgTx.TxIn {
|
|
if txIn.PreviousOutPoint == chanPoint {
|
|
return msgTx
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// testSwitchCircuitPersistence creates a multihop network to ensure the sender
|
|
// and intermediaries are persisting their open payment circuits. After
|
|
// forwarding a packet via an outgoing link, all are restarted, and expected to
|
|
// forward a response back from the receiver once back online.
|
|
//
|
|
// The general flow of this test:
|
|
// 1. Carol --> Dave --> Alice --> Bob forward payment
|
|
// 2. X X X Bob restart sender and intermediaries
|
|
// 3. Carol <-- Dave <-- Alice <-- Bob expect settle to propagate
|
|
func testSwitchCircuitPersistence(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const chanAmt = btcutil.Amount(1000000)
|
|
const pushAmt = btcutil.Amount(900000)
|
|
var networkChans []*lnrpc.ChannelPoint
|
|
|
|
// Open a channel with 100k satoshis between Alice and Bob with Alice
|
|
// being the sole funder of the channel.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAlice := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointAlice)
|
|
|
|
aliceChanTXID, err := lnd.GetChanPointFundingTxid(chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
aliceFundPoint := wire.OutPoint{
|
|
Hash: *aliceChanTXID,
|
|
Index: chanPointAlice.OutputIndex,
|
|
}
|
|
|
|
// As preliminary setup, we'll create two new nodes: Carol and Dave,
|
|
// such that we now have a 4 ndoe, 3 channel topology. Dave will make
|
|
// a channel with Alice, and Carol with Dave. After this setup, the
|
|
// network topology should now look like:
|
|
// Carol -> Dave -> Alice -> Bob
|
|
//
|
|
// First, we'll create Dave and establish a channel to Alice.
|
|
dave, err := net.NewNode("Dave", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, dave, net.Alice); err != nil {
|
|
t.Fatalf("unable to connect dave to alice: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointDave := openChannelAndAssert(
|
|
ctxt, t, net, dave, net.Alice,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointDave)
|
|
daveChanTXID, err := lnd.GetChanPointFundingTxid(chanPointDave)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
daveFundPoint := wire.OutPoint{
|
|
Hash: *daveChanTXID,
|
|
Index: chanPointDave.OutputIndex,
|
|
}
|
|
|
|
// Next, we'll create Carol and establish a channel to from her to
|
|
// Dave. Carol is started in htlchodl mode so that we can disconnect the
|
|
// intermediary hops before starting the settle.
|
|
carol, err := net.NewNode("Carol", []string{"--hodl.exit-settle"})
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, dave); err != nil {
|
|
t.Fatalf("unable to connect carol to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointCarol := openChannelAndAssert(
|
|
ctxt, t, net, carol, dave,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointCarol)
|
|
|
|
carolChanTXID, err := lnd.GetChanPointFundingTxid(chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
carolFundPoint := wire.OutPoint{
|
|
Hash: *carolChanTXID,
|
|
Index: chanPointCarol.OutputIndex,
|
|
}
|
|
|
|
// Wait for all nodes to have seen all channels.
|
|
nodes := []*lntest.HarnessNode{net.Alice, net.Bob, carol, dave}
|
|
nodeNames := []string{"Alice", "Bob", "Carol", "Dave"}
|
|
for _, chanPoint := range networkChans {
|
|
for i, node := range nodes {
|
|
txid, err := lnd.GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
point := wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: chanPoint.OutputIndex,
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = node.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("%s(%d): timeout waiting for "+
|
|
"channel(%s) open: %v", nodeNames[i],
|
|
node.NodeID, point, err)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Create 5 invoices for Carol, which expect a payment from Bob for 1k
|
|
// satoshis with a different preimage each time.
|
|
const numPayments = 5
|
|
const paymentAmt = 1000
|
|
payReqs, _, _, err := createPayReqs(
|
|
carol, paymentAmt, numPayments,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// We'll wait for all parties to recognize the new channels within the
|
|
// network.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = dave.WaitForNetworkChannelOpen(ctxt, chanPointDave)
|
|
if err != nil {
|
|
t.Fatalf("dave didn't advertise his channel: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("carol didn't advertise her channel in time: %v",
|
|
err)
|
|
}
|
|
|
|
time.Sleep(time.Millisecond * 50)
|
|
|
|
// Using Carol as the source, pay to the 5 invoices from Bob created
|
|
// above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, net.Bob, payReqs, false)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
// Wait until all nodes in the network have 5 outstanding htlcs.
|
|
var predErr error
|
|
err = wait.Predicate(func() bool {
|
|
predErr = assertNumActiveHtlcs(nodes, numPayments)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("htlc mismatch: %v", predErr)
|
|
}
|
|
|
|
// Restart the intermediaries and the sender.
|
|
if err := net.RestartNode(dave, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
if err := net.RestartNode(net.Alice, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
if err := net.RestartNode(net.Bob, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// Ensure all of the intermediate links are reconnected.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.EnsureConnected(ctxt, net.Alice, dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to reconnect alice and dave: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.EnsureConnected(ctxt, net.Bob, net.Alice)
|
|
if err != nil {
|
|
t.Fatalf("unable to reconnect bob and alice: %v", err)
|
|
}
|
|
|
|
// Ensure all nodes in the network still have 5 outstanding htlcs.
|
|
err = wait.Predicate(func() bool {
|
|
predErr = assertNumActiveHtlcs(nodes, numPayments)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
return true
|
|
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("htlc mismatch: %v", predErr)
|
|
}
|
|
|
|
// Now restart carol without hodl mode, to settle back the outstanding
|
|
// payments.
|
|
carol.SetExtraArgs(nil)
|
|
if err := net.RestartNode(carol, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.EnsureConnected(ctxt, dave, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to reconnect dave and carol: %v", err)
|
|
}
|
|
|
|
// After the payments settle, there should be no active htlcs on any of
|
|
// the nodes in the network.
|
|
err = wait.Predicate(func() bool {
|
|
predErr = assertNumActiveHtlcs(nodes, 0)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
return true
|
|
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("htlc mismatch: %v", predErr)
|
|
}
|
|
|
|
// When asserting the amount of satoshis moved, we'll factor in the
|
|
// default base fee, as we didn't modify the fee structure when
|
|
// creating the seed nodes in the network.
|
|
const baseFee = 1
|
|
|
|
// At this point all the channels within our proto network should be
|
|
// shifted by 5k satoshis in the direction of Carol, the sink within the
|
|
// payment flow generated above. The order of asserts corresponds to
|
|
// increasing of time is needed to embed the HTLC in commitment
|
|
// transaction, in channel Bob->Alice->David->Carol, order is Carol,
|
|
// David, Alice, Bob.
|
|
var amountPaid = int64(5000)
|
|
assertAmountPaid(t, "Dave(local) => Carol(remote)", carol,
|
|
carolFundPoint, int64(0), amountPaid)
|
|
assertAmountPaid(t, "Dave(local) => Carol(remote)", dave,
|
|
carolFundPoint, amountPaid, int64(0))
|
|
assertAmountPaid(t, "Alice(local) => Dave(remote)", dave,
|
|
daveFundPoint, int64(0), amountPaid+(baseFee*numPayments))
|
|
assertAmountPaid(t, "Alice(local) => Dave(remote)", net.Alice,
|
|
daveFundPoint, amountPaid+(baseFee*numPayments), int64(0))
|
|
assertAmountPaid(t, "Bob(local) => Alice(remote)", net.Alice,
|
|
aliceFundPoint, int64(0), amountPaid+((baseFee*numPayments)*2))
|
|
assertAmountPaid(t, "Bob(local) => Alice(remote)", net.Bob,
|
|
aliceFundPoint, amountPaid+(baseFee*numPayments)*2, int64(0))
|
|
|
|
// Lastly, we will send one more payment to ensure all channels are
|
|
// still functioning properly.
|
|
finalInvoice := &lnrpc.Invoice{
|
|
Memo: "testing",
|
|
Value: paymentAmt,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := carol.AddInvoice(ctxt, finalInvoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
|
|
payReqs = []string{resp.PaymentRequest}
|
|
|
|
// Using Carol as the source, pay to the 5 invoices from Bob created
|
|
// above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, net.Bob, payReqs, true)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
amountPaid = int64(6000)
|
|
assertAmountPaid(t, "Dave(local) => Carol(remote)", carol,
|
|
carolFundPoint, int64(0), amountPaid)
|
|
assertAmountPaid(t, "Dave(local) => Carol(remote)", dave,
|
|
carolFundPoint, amountPaid, int64(0))
|
|
assertAmountPaid(t, "Alice(local) => Dave(remote)", dave,
|
|
daveFundPoint, int64(0), amountPaid+(baseFee*(numPayments+1)))
|
|
assertAmountPaid(t, "Alice(local) => Dave(remote)", net.Alice,
|
|
daveFundPoint, amountPaid+(baseFee*(numPayments+1)), int64(0))
|
|
assertAmountPaid(t, "Bob(local) => Alice(remote)", net.Alice,
|
|
aliceFundPoint, int64(0), amountPaid+((baseFee*(numPayments+1))*2))
|
|
assertAmountPaid(t, "Bob(local) => Alice(remote)", net.Bob,
|
|
aliceFundPoint, amountPaid+(baseFee*(numPayments+1))*2, int64(0))
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointAlice, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, dave, chanPointDave, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPointCarol, false)
|
|
}
|
|
|
|
// testSwitchOfflineDelivery constructs a set of multihop payments, and tests
|
|
// that the returning payments are not lost if a peer on the backwards path is
|
|
// offline when the settle/fails are received. We expect the payments to be
|
|
// buffered in memory, and transmitted as soon as the disconnect link comes back
|
|
// online.
|
|
//
|
|
// The general flow of this test:
|
|
// 1. Carol --> Dave --> Alice --> Bob forward payment
|
|
// 2. Carol --- Dave X Alice --- Bob disconnect intermediaries
|
|
// 3. Carol --- Dave X Alice <-- Bob settle last hop
|
|
// 4. Carol <-- Dave <-- Alice --- Bob reconnect, expect settle to propagate
|
|
func testSwitchOfflineDelivery(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const chanAmt = btcutil.Amount(1000000)
|
|
const pushAmt = btcutil.Amount(900000)
|
|
var networkChans []*lnrpc.ChannelPoint
|
|
|
|
// Open a channel with 100k satoshis between Alice and Bob with Alice
|
|
// being the sole funder of the channel.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAlice := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointAlice)
|
|
|
|
aliceChanTXID, err := lnd.GetChanPointFundingTxid(chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
aliceFundPoint := wire.OutPoint{
|
|
Hash: *aliceChanTXID,
|
|
Index: chanPointAlice.OutputIndex,
|
|
}
|
|
|
|
// As preliminary setup, we'll create two new nodes: Carol and Dave,
|
|
// such that we now have a 4 ndoe, 3 channel topology. Dave will make
|
|
// a channel with Alice, and Carol with Dave. After this setup, the
|
|
// network topology should now look like:
|
|
// Carol -> Dave -> Alice -> Bob
|
|
//
|
|
// First, we'll create Dave and establish a channel to Alice.
|
|
dave, err := net.NewNode("Dave", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, dave, net.Alice); err != nil {
|
|
t.Fatalf("unable to connect dave to alice: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointDave := openChannelAndAssert(
|
|
ctxt, t, net, dave, net.Alice,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointDave)
|
|
daveChanTXID, err := lnd.GetChanPointFundingTxid(chanPointDave)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
daveFundPoint := wire.OutPoint{
|
|
Hash: *daveChanTXID,
|
|
Index: chanPointDave.OutputIndex,
|
|
}
|
|
|
|
// Next, we'll create Carol and establish a channel to from her to
|
|
// Dave. Carol is started in htlchodl mode so that we can disconnect the
|
|
// intermediary hops before starting the settle.
|
|
carol, err := net.NewNode("Carol", []string{"--hodl.exit-settle"})
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, dave); err != nil {
|
|
t.Fatalf("unable to connect carol to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointCarol := openChannelAndAssert(
|
|
ctxt, t, net, carol, dave,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointCarol)
|
|
|
|
carolChanTXID, err := lnd.GetChanPointFundingTxid(chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
carolFundPoint := wire.OutPoint{
|
|
Hash: *carolChanTXID,
|
|
Index: chanPointCarol.OutputIndex,
|
|
}
|
|
|
|
// Wait for all nodes to have seen all channels.
|
|
nodes := []*lntest.HarnessNode{net.Alice, net.Bob, carol, dave}
|
|
nodeNames := []string{"Alice", "Bob", "Carol", "Dave"}
|
|
for _, chanPoint := range networkChans {
|
|
for i, node := range nodes {
|
|
txid, err := lnd.GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
point := wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: chanPoint.OutputIndex,
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = node.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("%s(%d): timeout waiting for "+
|
|
"channel(%s) open: %v", nodeNames[i],
|
|
node.NodeID, point, err)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Create 5 invoices for Carol, which expect a payment from Bob for 1k
|
|
// satoshis with a different preimage each time.
|
|
const numPayments = 5
|
|
const paymentAmt = 1000
|
|
payReqs, _, _, err := createPayReqs(
|
|
carol, paymentAmt, numPayments,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// We'll wait for all parties to recognize the new channels within the
|
|
// network.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = dave.WaitForNetworkChannelOpen(ctxt, chanPointDave)
|
|
if err != nil {
|
|
t.Fatalf("dave didn't advertise his channel: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("carol didn't advertise her channel in time: %v",
|
|
err)
|
|
}
|
|
|
|
time.Sleep(time.Millisecond * 50)
|
|
|
|
// Using Carol as the source, pay to the 5 invoices from Bob created
|
|
// above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, net.Bob, payReqs, false)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
// Wait for all of the payments to reach Carol.
|
|
var predErr error
|
|
err = wait.Predicate(func() bool {
|
|
predErr = assertNumActiveHtlcs(nodes, numPayments)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("htlc mismatch: %v", predErr)
|
|
}
|
|
|
|
// First, disconnect Dave and Alice so that their link is broken.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.DisconnectNodes(ctxt, dave, net.Alice); err != nil {
|
|
t.Fatalf("unable to disconnect alice from dave: %v", err)
|
|
}
|
|
|
|
// Then, reconnect them to ensure Dave doesn't just fail back the htlc.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, dave, net.Alice); err != nil {
|
|
t.Fatalf("unable to reconnect alice to dave: %v", err)
|
|
}
|
|
|
|
// Wait to ensure that the payment remain are not failed back after
|
|
// reconnecting. All node should report the number payments initiated
|
|
// for the duration of the interval.
|
|
err = wait.Invariant(func() bool {
|
|
predErr = assertNumActiveHtlcs(nodes, numPayments)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*2)
|
|
if err != nil {
|
|
t.Fatalf("htlc change: %v", predErr)
|
|
}
|
|
|
|
// Now, disconnect Dave from Alice again before settling back the
|
|
// payment.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.DisconnectNodes(ctxt, dave, net.Alice); err != nil {
|
|
t.Fatalf("unable to disconnect alice from dave: %v", err)
|
|
}
|
|
|
|
// Now restart carol without hodl mode, to settle back the outstanding
|
|
// payments.
|
|
carol.SetExtraArgs(nil)
|
|
if err := net.RestartNode(carol, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// Wait for Carol to report no outstanding htlcs.
|
|
carolNode := []*lntest.HarnessNode{carol}
|
|
err = wait.Predicate(func() bool {
|
|
predErr = assertNumActiveHtlcs(carolNode, 0)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("htlc mismatch: %v", predErr)
|
|
}
|
|
|
|
// Now that the settles have reached Dave, reconnect him with Alice,
|
|
// allowing the settles to return to the sender.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.EnsureConnected(ctxt, dave, net.Alice); err != nil {
|
|
t.Fatalf("unable to reconnect alice to dave: %v", err)
|
|
}
|
|
|
|
// Wait until all outstanding htlcs in the network have been settled.
|
|
err = wait.Predicate(func() bool {
|
|
predErr = assertNumActiveHtlcs(nodes, 0)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("htlc mismatch: %v", predErr)
|
|
}
|
|
|
|
// When asserting the amount of satoshis moved, we'll factor in the
|
|
// default base fee, as we didn't modify the fee structure when
|
|
// creating the seed nodes in the network.
|
|
const baseFee = 1
|
|
|
|
// At this point all the channels within our proto network should be
|
|
// shifted by 5k satoshis in the direction of Carol, the sink within the
|
|
// payment flow generated above. The order of asserts corresponds to
|
|
// increasing of time is needed to embed the HTLC in commitment
|
|
// transaction, in channel Bob->Alice->David->Carol, order is Carol,
|
|
// David, Alice, Bob.
|
|
var amountPaid = int64(5000)
|
|
assertAmountPaid(t, "Dave(local) => Carol(remote)", carol,
|
|
carolFundPoint, int64(0), amountPaid)
|
|
assertAmountPaid(t, "Dave(local) => Carol(remote)", dave,
|
|
carolFundPoint, amountPaid, int64(0))
|
|
assertAmountPaid(t, "Alice(local) => Dave(remote)", dave,
|
|
daveFundPoint, int64(0), amountPaid+(baseFee*numPayments))
|
|
assertAmountPaid(t, "Alice(local) => Dave(remote)", net.Alice,
|
|
daveFundPoint, amountPaid+(baseFee*numPayments), int64(0))
|
|
assertAmountPaid(t, "Bob(local) => Alice(remote)", net.Alice,
|
|
aliceFundPoint, int64(0), amountPaid+((baseFee*numPayments)*2))
|
|
assertAmountPaid(t, "Bob(local) => Alice(remote)", net.Bob,
|
|
aliceFundPoint, amountPaid+(baseFee*numPayments)*2, int64(0))
|
|
|
|
// Lastly, we will send one more payment to ensure all channels are
|
|
// still functioning properly.
|
|
finalInvoice := &lnrpc.Invoice{
|
|
Memo: "testing",
|
|
Value: paymentAmt,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := carol.AddInvoice(ctxt, finalInvoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
|
|
payReqs = []string{resp.PaymentRequest}
|
|
|
|
// Using Carol as the source, pay to the 5 invoices from Bob created
|
|
// above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, net.Bob, payReqs, true)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
amountPaid = int64(6000)
|
|
assertAmountPaid(t, "Dave(local) => Carol(remote)", carol,
|
|
carolFundPoint, int64(0), amountPaid)
|
|
assertAmountPaid(t, "Dave(local) => Carol(remote)", dave,
|
|
carolFundPoint, amountPaid, int64(0))
|
|
assertAmountPaid(t, "Alice(local) => Dave(remote)", dave,
|
|
daveFundPoint, int64(0), amountPaid+(baseFee*(numPayments+1)))
|
|
assertAmountPaid(t, "Alice(local) => Dave(remote)", net.Alice,
|
|
daveFundPoint, amountPaid+(baseFee*(numPayments+1)), int64(0))
|
|
assertAmountPaid(t, "Bob(local) => Alice(remote)", net.Alice,
|
|
aliceFundPoint, int64(0), amountPaid+((baseFee*(numPayments+1))*2))
|
|
assertAmountPaid(t, "Bob(local) => Alice(remote)", net.Bob,
|
|
aliceFundPoint, amountPaid+(baseFee*(numPayments+1))*2, int64(0))
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointAlice, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, dave, chanPointDave, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPointCarol, false)
|
|
}
|
|
|
|
// testSwitchOfflineDeliveryPersistence constructs a set of multihop payments,
|
|
// and tests that the returning payments are not lost if a peer on the backwards
|
|
// path is offline when the settle/fails are received AND the peer buffering the
|
|
// responses is completely restarts. We expect the payments to be reloaded from
|
|
// disk, and transmitted as soon as the intermediaries are reconnected.
|
|
//
|
|
// The general flow of this test:
|
|
// 1. Carol --> Dave --> Alice --> Bob forward payment
|
|
// 2. Carol --- Dave X Alice --- Bob disconnect intermediaries
|
|
// 3. Carol --- Dave X Alice <-- Bob settle last hop
|
|
// 4. Carol --- Dave X X Bob restart Alice
|
|
// 5. Carol <-- Dave <-- Alice --- Bob expect settle to propagate
|
|
func testSwitchOfflineDeliveryPersistence(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const chanAmt = btcutil.Amount(1000000)
|
|
const pushAmt = btcutil.Amount(900000)
|
|
var networkChans []*lnrpc.ChannelPoint
|
|
|
|
// Open a channel with 100k satoshis between Alice and Bob with Alice
|
|
// being the sole funder of the channel.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAlice := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointAlice)
|
|
|
|
aliceChanTXID, err := lnd.GetChanPointFundingTxid(chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
aliceFundPoint := wire.OutPoint{
|
|
Hash: *aliceChanTXID,
|
|
Index: chanPointAlice.OutputIndex,
|
|
}
|
|
|
|
// As preliminary setup, we'll create two new nodes: Carol and Dave,
|
|
// such that we now have a 4 ndoe, 3 channel topology. Dave will make
|
|
// a channel with Alice, and Carol with Dave. After this setup, the
|
|
// network topology should now look like:
|
|
// Carol -> Dave -> Alice -> Bob
|
|
//
|
|
// First, we'll create Dave and establish a channel to Alice.
|
|
dave, err := net.NewNode("Dave", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, dave, net.Alice); err != nil {
|
|
t.Fatalf("unable to connect dave to alice: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointDave := openChannelAndAssert(
|
|
ctxt, t, net, dave, net.Alice,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
|
|
networkChans = append(networkChans, chanPointDave)
|
|
daveChanTXID, err := lnd.GetChanPointFundingTxid(chanPointDave)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
daveFundPoint := wire.OutPoint{
|
|
Hash: *daveChanTXID,
|
|
Index: chanPointDave.OutputIndex,
|
|
}
|
|
|
|
// Next, we'll create Carol and establish a channel to from her to
|
|
// Dave. Carol is started in htlchodl mode so that we can disconnect the
|
|
// intermediary hops before starting the settle.
|
|
carol, err := net.NewNode("Carol", []string{"--hodl.exit-settle"})
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, dave); err != nil {
|
|
t.Fatalf("unable to connect carol to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointCarol := openChannelAndAssert(
|
|
ctxt, t, net, carol, dave,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointCarol)
|
|
|
|
carolChanTXID, err := lnd.GetChanPointFundingTxid(chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
carolFundPoint := wire.OutPoint{
|
|
Hash: *carolChanTXID,
|
|
Index: chanPointCarol.OutputIndex,
|
|
}
|
|
|
|
// Wait for all nodes to have seen all channels.
|
|
nodes := []*lntest.HarnessNode{net.Alice, net.Bob, carol, dave}
|
|
nodeNames := []string{"Alice", "Bob", "Carol", "Dave"}
|
|
for _, chanPoint := range networkChans {
|
|
for i, node := range nodes {
|
|
txid, err := lnd.GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
point := wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: chanPoint.OutputIndex,
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = node.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("%s(%d): timeout waiting for "+
|
|
"channel(%s) open: %v", nodeNames[i],
|
|
node.NodeID, point, err)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Create 5 invoices for Carol, which expect a payment from Bob for 1k
|
|
// satoshis with a different preimage each time.
|
|
const numPayments = 5
|
|
const paymentAmt = 1000
|
|
payReqs, _, _, err := createPayReqs(
|
|
carol, paymentAmt, numPayments,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// We'll wait for all parties to recognize the new channels within the
|
|
// network.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = dave.WaitForNetworkChannelOpen(ctxt, chanPointDave)
|
|
if err != nil {
|
|
t.Fatalf("dave didn't advertise his channel: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("carol didn't advertise her channel in time: %v",
|
|
err)
|
|
}
|
|
|
|
// Using Carol as the source, pay to the 5 invoices from Bob created
|
|
// above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, net.Bob, payReqs, false)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
var predErr error
|
|
err = wait.Predicate(func() bool {
|
|
predErr = assertNumActiveHtlcs(nodes, numPayments)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
return true
|
|
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("htlc mismatch: %v", predErr)
|
|
}
|
|
|
|
// Disconnect the two intermediaries, Alice and Dave, by shutting down
|
|
// Alice.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.StopNode(net.Alice); err != nil {
|
|
t.Fatalf("unable to shutdown alice: %v", err)
|
|
}
|
|
|
|
// Now restart carol without hodl mode, to settle back the outstanding
|
|
// payments.
|
|
carol.SetExtraArgs(nil)
|
|
if err := net.RestartNode(carol, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// Make Carol and Dave are reconnected before waiting for the htlcs to
|
|
// clear.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.EnsureConnected(ctxt, dave, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to reconnect dave and carol: %v", err)
|
|
}
|
|
|
|
// Wait for Carol to report no outstanding htlcs, and also for Dav to
|
|
// receive all the settles from Carol.
|
|
carolNode := []*lntest.HarnessNode{carol}
|
|
err = wait.Predicate(func() bool {
|
|
predErr = assertNumActiveHtlcs(carolNode, 0)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
|
|
predErr = assertNumActiveHtlcsChanPoint(dave, carolFundPoint, 0)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("htlc mismatch: %v", predErr)
|
|
}
|
|
|
|
// Finally, restart dave who received the settles, but was unable to
|
|
// deliver them to Alice since they were disconnected.
|
|
if err := net.RestartNode(dave, nil); err != nil {
|
|
t.Fatalf("unable to restart dave: %v", err)
|
|
}
|
|
if err = net.RestartNode(net.Alice, nil); err != nil {
|
|
t.Fatalf("unable to restart alice: %v", err)
|
|
}
|
|
|
|
// Force Dave and Alice to reconnect before waiting for the htlcs to
|
|
// clear.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.EnsureConnected(ctxt, dave, net.Alice)
|
|
if err != nil {
|
|
t.Fatalf("unable to reconnect dave and carol: %v", err)
|
|
}
|
|
|
|
// After reconnection succeeds, the settles should be propagated all
|
|
// the way back to the sender. All nodes should report no active htlcs.
|
|
err = wait.Predicate(func() bool {
|
|
predErr = assertNumActiveHtlcs(nodes, 0)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("htlc mismatch: %v", predErr)
|
|
}
|
|
|
|
// When asserting the amount of satoshis moved, we'll factor in the
|
|
// default base fee, as we didn't modify the fee structure when
|
|
// creating the seed nodes in the network.
|
|
const baseFee = 1
|
|
|
|
// At this point all the channels within our proto network should be
|
|
// shifted by 5k satoshis in the direction of Carol, the sink within the
|
|
// payment flow generated above. The order of asserts corresponds to
|
|
// increasing of time is needed to embed the HTLC in commitment
|
|
// transaction, in channel Bob->Alice->David->Carol, order is Carol,
|
|
// David, Alice, Bob.
|
|
var amountPaid = int64(5000)
|
|
assertAmountPaid(t, "Dave(local) => Carol(remote)", carol,
|
|
carolFundPoint, int64(0), amountPaid)
|
|
assertAmountPaid(t, "Dave(local) => Carol(remote)", dave,
|
|
carolFundPoint, amountPaid, int64(0))
|
|
assertAmountPaid(t, "Alice(local) => Dave(remote)", dave,
|
|
daveFundPoint, int64(0), amountPaid+(baseFee*numPayments))
|
|
assertAmountPaid(t, "Alice(local) => Dave(remote)", net.Alice,
|
|
daveFundPoint, amountPaid+(baseFee*numPayments), int64(0))
|
|
assertAmountPaid(t, "Bob(local) => Alice(remote)", net.Alice,
|
|
aliceFundPoint, int64(0), amountPaid+((baseFee*numPayments)*2))
|
|
assertAmountPaid(t, "Bob(local) => Alice(remote)", net.Bob,
|
|
aliceFundPoint, amountPaid+(baseFee*numPayments)*2, int64(0))
|
|
|
|
// Lastly, we will send one more payment to ensure all channels are
|
|
// still functioning properly.
|
|
finalInvoice := &lnrpc.Invoice{
|
|
Memo: "testing",
|
|
Value: paymentAmt,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := carol.AddInvoice(ctxt, finalInvoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
|
|
payReqs = []string{resp.PaymentRequest}
|
|
|
|
// Before completing the final payment request, ensure that the
|
|
// connection between Dave and Carol has been healed.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.EnsureConnected(ctxt, dave, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to reconnect dave and carol: %v", err)
|
|
}
|
|
|
|
// Using Carol as the source, pay to the 5 invoices from Bob created
|
|
// above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, net.Bob, payReqs, true)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
amountPaid = int64(6000)
|
|
assertAmountPaid(t, "Dave(local) => Carol(remote)", carol,
|
|
carolFundPoint, int64(0), amountPaid)
|
|
assertAmountPaid(t, "Dave(local) => Carol(remote)", dave,
|
|
carolFundPoint, amountPaid, int64(0))
|
|
assertAmountPaid(t, "Alice(local) => Dave(remote)", dave,
|
|
daveFundPoint, int64(0), amountPaid+(baseFee*(numPayments+1)))
|
|
assertAmountPaid(t, "Alice(local) => Dave(remote)", net.Alice,
|
|
daveFundPoint, amountPaid+(baseFee*(numPayments+1)), int64(0))
|
|
assertAmountPaid(t, "Bob(local) => Alice(remote)", net.Alice,
|
|
aliceFundPoint, int64(0), amountPaid+((baseFee*(numPayments+1))*2))
|
|
assertAmountPaid(t, "Bob(local) => Alice(remote)", net.Bob,
|
|
aliceFundPoint, amountPaid+(baseFee*(numPayments+1))*2, int64(0))
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointAlice, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, dave, chanPointDave, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPointCarol, false)
|
|
}
|
|
|
|
// testSwitchOfflineDeliveryOutgoingOffline constructs a set of multihop payments,
|
|
// and tests that the returning payments are not lost if a peer on the backwards
|
|
// path is offline when the settle/fails are received AND the peer buffering the
|
|
// responses is completely restarts. We expect the payments to be reloaded from
|
|
// disk, and transmitted as soon as the intermediaries are reconnected.
|
|
//
|
|
// The general flow of this test:
|
|
// 1. Carol --> Dave --> Alice --> Bob forward payment
|
|
// 2. Carol --- Dave X Alice --- Bob disconnect intermediaries
|
|
// 3. Carol --- Dave X Alice <-- Bob settle last hop
|
|
// 4. Carol --- Dave X X shutdown Bob, restart Alice
|
|
// 5. Carol <-- Dave <-- Alice X expect settle to propagate
|
|
func testSwitchOfflineDeliveryOutgoingOffline(
|
|
net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const chanAmt = btcutil.Amount(1000000)
|
|
const pushAmt = btcutil.Amount(900000)
|
|
var networkChans []*lnrpc.ChannelPoint
|
|
|
|
// Open a channel with 100k satoshis between Alice and Bob with Alice
|
|
// being the sole funder of the channel.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAlice := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointAlice)
|
|
|
|
aliceChanTXID, err := lnd.GetChanPointFundingTxid(chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
aliceFundPoint := wire.OutPoint{
|
|
Hash: *aliceChanTXID,
|
|
Index: chanPointAlice.OutputIndex,
|
|
}
|
|
|
|
// As preliminary setup, we'll create two new nodes: Carol and Dave,
|
|
// such that we now have a 4 ndoe, 3 channel topology. Dave will make
|
|
// a channel with Alice, and Carol with Dave. After this setup, the
|
|
// network topology should now look like:
|
|
// Carol -> Dave -> Alice -> Bob
|
|
//
|
|
// First, we'll create Dave and establish a channel to Alice.
|
|
dave, err := net.NewNode("Dave", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, dave, net.Alice); err != nil {
|
|
t.Fatalf("unable to connect dave to alice: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointDave := openChannelAndAssert(
|
|
ctxt, t, net, dave, net.Alice,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointDave)
|
|
daveChanTXID, err := lnd.GetChanPointFundingTxid(chanPointDave)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
daveFundPoint := wire.OutPoint{
|
|
Hash: *daveChanTXID,
|
|
Index: chanPointDave.OutputIndex,
|
|
}
|
|
|
|
// Next, we'll create Carol and establish a channel to from her to
|
|
// Dave. Carol is started in htlchodl mode so that we can disconnect the
|
|
// intermediary hops before starting the settle.
|
|
carol, err := net.NewNode("Carol", []string{"--hodl.exit-settle"})
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, dave); err != nil {
|
|
t.Fatalf("unable to connect carol to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointCarol := openChannelAndAssert(
|
|
ctxt, t, net, carol, dave,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointCarol)
|
|
|
|
carolChanTXID, err := lnd.GetChanPointFundingTxid(chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
carolFundPoint := wire.OutPoint{
|
|
Hash: *carolChanTXID,
|
|
Index: chanPointCarol.OutputIndex,
|
|
}
|
|
|
|
// Wait for all nodes to have seen all channels.
|
|
nodes := []*lntest.HarnessNode{net.Alice, net.Bob, carol, dave}
|
|
nodeNames := []string{"Alice", "Bob", "Carol", "Dave"}
|
|
for _, chanPoint := range networkChans {
|
|
for i, node := range nodes {
|
|
txid, err := lnd.GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
point := wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: chanPoint.OutputIndex,
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = node.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("%s(%d): timeout waiting for "+
|
|
"channel(%s) open: %v", nodeNames[i],
|
|
node.NodeID, point, err)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Create 5 invoices for Carol, which expect a payment from Bob for 1k
|
|
// satoshis with a different preimage each time.
|
|
const numPayments = 5
|
|
const paymentAmt = 1000
|
|
payReqs, _, _, err := createPayReqs(
|
|
carol, paymentAmt, numPayments,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create pay reqs: %v", err)
|
|
}
|
|
|
|
// We'll wait for all parties to recognize the new channels within the
|
|
// network.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = dave.WaitForNetworkChannelOpen(ctxt, chanPointDave)
|
|
if err != nil {
|
|
t.Fatalf("dave didn't advertise his channel: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPointCarol)
|
|
if err != nil {
|
|
t.Fatalf("carol didn't advertise her channel in time: %v",
|
|
err)
|
|
}
|
|
|
|
// Using Carol as the source, pay to the 5 invoices from Bob created
|
|
// above.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(ctxt, net.Bob, payReqs, false)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payments: %v", err)
|
|
}
|
|
|
|
// Wait for all payments to reach Carol.
|
|
var predErr error
|
|
err = wait.Predicate(func() bool {
|
|
predErr = assertNumActiveHtlcs(nodes, numPayments)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("htlc mismatch: %v", predErr)
|
|
}
|
|
|
|
// Disconnect the two intermediaries, Alice and Dave, so that when carol
|
|
// restarts, the response will be held by Dave.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.StopNode(net.Alice); err != nil {
|
|
t.Fatalf("unable to shutdown alice: %v", err)
|
|
}
|
|
|
|
// Now restart carol without hodl mode, to settle back the outstanding
|
|
// payments.
|
|
carol.SetExtraArgs(nil)
|
|
if err := net.RestartNode(carol, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// Wait for Carol to report no outstanding htlcs.
|
|
carolNode := []*lntest.HarnessNode{carol}
|
|
err = wait.Predicate(func() bool {
|
|
predErr = assertNumActiveHtlcs(carolNode, 0)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
|
|
predErr = assertNumActiveHtlcsChanPoint(dave, carolFundPoint, 0)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("htlc mismatch: %v", predErr)
|
|
}
|
|
|
|
// Now check that the total amount was transferred from Dave to Carol.
|
|
// The amount transferred should be exactly equal to the invoice total
|
|
// payment amount, 5k satsohis.
|
|
const amountPaid = int64(5000)
|
|
assertAmountPaid(t, "Dave(local) => Carol(remote)", carol,
|
|
carolFundPoint, int64(0), amountPaid)
|
|
assertAmountPaid(t, "Dave(local) => Carol(remote)", dave,
|
|
carolFundPoint, amountPaid, int64(0))
|
|
|
|
// Shutdown carol and leave her offline for the rest of the test. This
|
|
// is critical, as we wish to see if Dave can propragate settles even if
|
|
// the outgoing link is never revived.
|
|
shutdownAndAssert(net, t, carol)
|
|
|
|
// Now restart Dave, ensuring he is both persisting the settles, and is
|
|
// able to reforward them to Alice after recovering from a restart.
|
|
if err := net.RestartNode(dave, nil); err != nil {
|
|
t.Fatalf("unable to restart dave: %v", err)
|
|
}
|
|
if err = net.RestartNode(net.Alice, nil); err != nil {
|
|
t.Fatalf("unable to restart alice: %v", err)
|
|
}
|
|
|
|
// Ensure that Dave is reconnected to Alice before waiting for the
|
|
// htlcs to clear.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.EnsureConnected(ctxt, dave, net.Alice)
|
|
if err != nil {
|
|
t.Fatalf("unable to reconnect alice and dave: %v", err)
|
|
}
|
|
|
|
// Since Carol has been shutdown permanently, we will wait until all
|
|
// other nodes in the network report no active htlcs.
|
|
nodesMinusCarol := []*lntest.HarnessNode{net.Bob, net.Alice, dave}
|
|
err = wait.Predicate(func() bool {
|
|
predErr = assertNumActiveHtlcs(nodesMinusCarol, 0)
|
|
if predErr != nil {
|
|
return false
|
|
}
|
|
return true
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("htlc mismatch: %v", predErr)
|
|
}
|
|
|
|
// When asserting the amount of satoshis moved, we'll factor in the
|
|
// default base fee, as we didn't modify the fee structure when
|
|
// creating the seed nodes in the network.
|
|
const baseFee = 1
|
|
|
|
// At this point, all channels (minus Carol, who is shutdown) should
|
|
// show a shift of 5k satoshis towards Carol. The order of asserts
|
|
// corresponds to increasing of time is needed to embed the HTLC in
|
|
// commitment transaction, in channel Bob->Alice->David, order is
|
|
// David, Alice, Bob.
|
|
assertAmountPaid(t, "Alice(local) => Dave(remote)", dave,
|
|
daveFundPoint, int64(0), amountPaid+(baseFee*numPayments))
|
|
assertAmountPaid(t, "Alice(local) => Dave(remote)", net.Alice,
|
|
daveFundPoint, amountPaid+(baseFee*numPayments), int64(0))
|
|
assertAmountPaid(t, "Bob(local) => Alice(remote)", net.Alice,
|
|
aliceFundPoint, int64(0), amountPaid+((baseFee*numPayments)*2))
|
|
assertAmountPaid(t, "Bob(local) => Alice(remote)", net.Bob,
|
|
aliceFundPoint, amountPaid+(baseFee*numPayments)*2, int64(0))
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointAlice, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, dave, chanPointDave, false)
|
|
}
|
|
|
|
// computeFee calculates the payment fee as specified in BOLT07
|
|
func computeFee(baseFee, feeRate, amt lnwire.MilliSatoshi) lnwire.MilliSatoshi {
|
|
return baseFee + amt*feeRate/1000000
|
|
}
|
|
|
|
// testQueryRoutes checks the response of queryroutes.
|
|
// We'll create the following network topology:
|
|
// Alice --> Bob --> Carol --> Dave
|
|
// and query the daemon for routes from Alice to Dave.
|
|
func testQueryRoutes(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const chanAmt = btcutil.Amount(100000)
|
|
var networkChans []*lnrpc.ChannelPoint
|
|
|
|
// Open a channel between Alice and Bob.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAlice := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointAlice)
|
|
|
|
// Create Carol and establish a channel from Bob.
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, net.Bob); err != nil {
|
|
t.Fatalf("unable to connect carol to bob: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, net.Bob)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to bob: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointBob := openChannelAndAssert(
|
|
ctxt, t, net, net.Bob, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointBob)
|
|
|
|
// Create Dave and establish a channel from Carol.
|
|
dave, err := net.NewNode("Dave", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, dave, carol); err != nil {
|
|
t.Fatalf("unable to connect dave to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointCarol := openChannelAndAssert(
|
|
ctxt, t, net, carol, dave,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
networkChans = append(networkChans, chanPointCarol)
|
|
|
|
// Wait for all nodes to have seen all channels.
|
|
nodes := []*lntest.HarnessNode{net.Alice, net.Bob, carol, dave}
|
|
nodeNames := []string{"Alice", "Bob", "Carol", "Dave"}
|
|
for _, chanPoint := range networkChans {
|
|
for i, node := range nodes {
|
|
txid, err := lnd.GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
point := wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: chanPoint.OutputIndex,
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = node.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("%s(%d): timeout waiting for "+
|
|
"channel(%s) open: %v", nodeNames[i],
|
|
node.NodeID, point, err)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Query for routes to pay from Alice to Dave.
|
|
const paymentAmt = 1000
|
|
routesReq := &lnrpc.QueryRoutesRequest{
|
|
PubKey: dave.PubKeyStr,
|
|
Amt: paymentAmt,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
routesRes, err := net.Alice.QueryRoutes(ctxt, routesReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get route: %v", err)
|
|
}
|
|
|
|
const mSat = 1000
|
|
feePerHopMSat := computeFee(1000, 1, paymentAmt*mSat)
|
|
|
|
for i, route := range routesRes.Routes {
|
|
expectedTotalFeesMSat :=
|
|
lnwire.MilliSatoshi(len(route.Hops)-1) * feePerHopMSat
|
|
expectedTotalAmtMSat := (paymentAmt * mSat) + expectedTotalFeesMSat
|
|
|
|
if route.TotalFees != route.TotalFeesMsat/mSat {
|
|
t.Fatalf("route %v: total fees %v (msat) does not "+
|
|
"round down to %v (sat)",
|
|
i, route.TotalFeesMsat, route.TotalFees)
|
|
}
|
|
if route.TotalFeesMsat != int64(expectedTotalFeesMSat) {
|
|
t.Fatalf("route %v: total fees in msat expected %v got %v",
|
|
i, expectedTotalFeesMSat, route.TotalFeesMsat)
|
|
}
|
|
|
|
if route.TotalAmt != route.TotalAmtMsat/mSat {
|
|
t.Fatalf("route %v: total amt %v (msat) does not "+
|
|
"round down to %v (sat)",
|
|
i, route.TotalAmtMsat, route.TotalAmt)
|
|
}
|
|
if route.TotalAmtMsat != int64(expectedTotalAmtMSat) {
|
|
t.Fatalf("route %v: total amt in msat expected %v got %v",
|
|
i, expectedTotalAmtMSat, route.TotalAmtMsat)
|
|
}
|
|
|
|
// For all hops except the last, we check that fee equals feePerHop
|
|
// and amount to forward deducts feePerHop on each hop.
|
|
expectedAmtToForwardMSat := expectedTotalAmtMSat
|
|
for j, hop := range route.Hops[:len(route.Hops)-1] {
|
|
expectedAmtToForwardMSat -= feePerHopMSat
|
|
|
|
if hop.Fee != hop.FeeMsat/mSat {
|
|
t.Fatalf("route %v hop %v: fee %v (msat) does not "+
|
|
"round down to %v (sat)",
|
|
i, j, hop.FeeMsat, hop.Fee)
|
|
}
|
|
if hop.FeeMsat != int64(feePerHopMSat) {
|
|
t.Fatalf("route %v hop %v: fee in msat expected %v got %v",
|
|
i, j, feePerHopMSat, hop.FeeMsat)
|
|
}
|
|
|
|
if hop.AmtToForward != hop.AmtToForwardMsat/mSat {
|
|
t.Fatalf("route %v hop %v: amt to forward %v (msat) does not "+
|
|
"round down to %v (sat)",
|
|
i, j, hop.AmtToForwardMsat, hop.AmtToForward)
|
|
}
|
|
if hop.AmtToForwardMsat != int64(expectedAmtToForwardMSat) {
|
|
t.Fatalf("route %v hop %v: amt to forward in msat "+
|
|
"expected %v got %v",
|
|
i, j, expectedAmtToForwardMSat, hop.AmtToForwardMsat)
|
|
}
|
|
}
|
|
// Last hop should have zero fee and amount to forward should equal
|
|
// payment amount.
|
|
hop := route.Hops[len(route.Hops)-1]
|
|
|
|
if hop.Fee != 0 || hop.FeeMsat != 0 {
|
|
t.Fatalf("route %v hop %v: fee expected 0 got %v (sat) %v (msat)",
|
|
i, len(route.Hops)-1, hop.Fee, hop.FeeMsat)
|
|
}
|
|
|
|
if hop.AmtToForward != hop.AmtToForwardMsat/mSat {
|
|
t.Fatalf("route %v hop %v: amt to forward %v (msat) does not "+
|
|
"round down to %v (sat)",
|
|
i, len(route.Hops)-1, hop.AmtToForwardMsat, hop.AmtToForward)
|
|
}
|
|
if hop.AmtToForwardMsat != paymentAmt*mSat {
|
|
t.Fatalf("route %v hop %v: amt to forward in msat "+
|
|
"expected %v got %v",
|
|
i, len(route.Hops)-1, paymentAmt*mSat, hop.AmtToForwardMsat)
|
|
}
|
|
}
|
|
|
|
// We clean up the test case by closing channels that were created for
|
|
// the duration of the tests.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointAlice, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Bob, chanPointBob, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPointCarol, false)
|
|
}
|
|
|
|
// testRouteFeeCutoff tests that we are able to prevent querying routes and
|
|
// sending payments that incur a fee higher than the fee limit.
|
|
func testRouteFeeCutoff(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
// For this test, we'll create the following topology:
|
|
//
|
|
// --- Bob ---
|
|
// / \
|
|
// Alice ---- ---- Dave
|
|
// \ /
|
|
// -- Carol --
|
|
//
|
|
// Alice will attempt to send payments to Dave that should not incur a
|
|
// fee greater than the fee limit expressed as a percentage of the
|
|
// amount and as a fixed amount of satoshis.
|
|
const chanAmt = btcutil.Amount(100000)
|
|
|
|
// Open a channel between Alice and Bob.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAliceBob := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Create Carol's node and open a channel between her and Alice with
|
|
// Alice being the funder.
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create carol's node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, net.Alice); err != nil {
|
|
t.Fatalf("unable to connect carol to alice: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAliceCarol := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Create Dave's node and open a channel between him and Bob with Bob
|
|
// being the funder.
|
|
dave, err := net.NewNode("Dave", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create dave's node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, dave, net.Bob); err != nil {
|
|
t.Fatalf("unable to connect dave to bob: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointBobDave := openChannelAndAssert(
|
|
ctxt, t, net, net.Bob, dave,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Open a channel between Carol and Dave.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, dave); err != nil {
|
|
t.Fatalf("unable to connect carol to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointCarolDave := openChannelAndAssert(
|
|
ctxt, t, net, carol, dave,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Now that all the channels were set up, we'll wait for all the nodes
|
|
// to have seen all the channels.
|
|
nodes := []*lntest.HarnessNode{net.Alice, net.Bob, carol, dave}
|
|
nodeNames := []string{"alice", "bob", "carol", "dave"}
|
|
networkChans := []*lnrpc.ChannelPoint{
|
|
chanPointAliceBob, chanPointAliceCarol, chanPointBobDave,
|
|
chanPointCarolDave,
|
|
}
|
|
for _, chanPoint := range networkChans {
|
|
for i, node := range nodes {
|
|
txid, err := lnd.GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
outpoint := wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: chanPoint.OutputIndex,
|
|
}
|
|
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
err = node.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("%s(%d) timed out waiting for "+
|
|
"channel(%s) open: %v", nodeNames[i],
|
|
node.NodeID, outpoint, err)
|
|
}
|
|
}
|
|
}
|
|
|
|
// The payments should only be successful across the route:
|
|
// Alice -> Bob -> Dave
|
|
// Therefore, we'll update the fee policy on Carol's side for the
|
|
// channel between her and Dave to invalidate the route:
|
|
// Alice -> Carol -> Dave
|
|
baseFee := int64(10000)
|
|
feeRate := int64(5)
|
|
timeLockDelta := uint32(lnd.DefaultBitcoinTimeLockDelta)
|
|
maxHtlc := calculateMaxHtlc(chanAmt)
|
|
|
|
expectedPolicy := &lnrpc.RoutingPolicy{
|
|
FeeBaseMsat: baseFee,
|
|
FeeRateMilliMsat: testFeeBase * feeRate,
|
|
TimeLockDelta: timeLockDelta,
|
|
MinHtlc: 1000, // default value
|
|
MaxHtlcMsat: maxHtlc,
|
|
}
|
|
|
|
updateFeeReq := &lnrpc.PolicyUpdateRequest{
|
|
BaseFeeMsat: baseFee,
|
|
FeeRate: float64(feeRate),
|
|
TimeLockDelta: timeLockDelta,
|
|
MaxHtlcMsat: maxHtlc,
|
|
Scope: &lnrpc.PolicyUpdateRequest_ChanPoint{
|
|
ChanPoint: chanPointCarolDave,
|
|
},
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if _, err := carol.UpdateChannelPolicy(ctxt, updateFeeReq); err != nil {
|
|
t.Fatalf("unable to update chan policy: %v", err)
|
|
}
|
|
|
|
// Wait for Alice to receive the channel update from Carol.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
aliceSub := subscribeGraphNotifications(t, ctxt, net.Alice)
|
|
defer close(aliceSub.quit)
|
|
|
|
waitForChannelUpdate(
|
|
t, aliceSub,
|
|
[]expectedChanUpdate{
|
|
{carol.PubKeyStr, expectedPolicy, chanPointCarolDave},
|
|
},
|
|
)
|
|
|
|
// We'll also need the channel IDs for Bob's channels in order to
|
|
// confirm the route of the payments.
|
|
listReq := &lnrpc.ListChannelsRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
listResp, err := net.Bob.ListChannels(ctxt, listReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to retrieve bob's channels: %v", err)
|
|
}
|
|
|
|
var aliceBobChanID, bobDaveChanID uint64
|
|
for _, channel := range listResp.Channels {
|
|
switch channel.RemotePubkey {
|
|
case net.Alice.PubKeyStr:
|
|
aliceBobChanID = channel.ChanId
|
|
case dave.PubKeyStr:
|
|
bobDaveChanID = channel.ChanId
|
|
}
|
|
}
|
|
|
|
if aliceBobChanID == 0 {
|
|
t.Fatalf("channel between alice and bob not found")
|
|
}
|
|
if bobDaveChanID == 0 {
|
|
t.Fatalf("channel between bob and dave not found")
|
|
}
|
|
hopChanIDs := []uint64{aliceBobChanID, bobDaveChanID}
|
|
|
|
// checkRoute is a helper closure to ensure the route contains the
|
|
// correct intermediate hops.
|
|
checkRoute := func(route *lnrpc.Route) {
|
|
if len(route.Hops) != 2 {
|
|
t.Fatalf("expected two hops, got %d", len(route.Hops))
|
|
}
|
|
|
|
for i, hop := range route.Hops {
|
|
if hop.ChanId != hopChanIDs[i] {
|
|
t.Fatalf("expected chan id %d, got %d",
|
|
hopChanIDs[i], hop.ChanId)
|
|
}
|
|
}
|
|
}
|
|
|
|
// We'll be attempting to send two payments from Alice to Dave. One will
|
|
// have a fee cutoff expressed as a percentage of the amount and the
|
|
// other will have it expressed as a fixed amount of satoshis.
|
|
const paymentAmt = 100
|
|
carolFee := computeFee(lnwire.MilliSatoshi(baseFee), 1, paymentAmt)
|
|
|
|
// testFeeCutoff is a helper closure that will ensure the different
|
|
// types of fee limits work as intended when querying routes and sending
|
|
// payments.
|
|
testFeeCutoff := func(feeLimit *lnrpc.FeeLimit) {
|
|
queryRoutesReq := &lnrpc.QueryRoutesRequest{
|
|
PubKey: dave.PubKeyStr,
|
|
Amt: paymentAmt,
|
|
FeeLimit: feeLimit,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
routesResp, err := net.Alice.QueryRoutes(ctxt, queryRoutesReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get routes: %v", err)
|
|
}
|
|
|
|
checkRoute(routesResp.Routes[0])
|
|
|
|
invoice := &lnrpc.Invoice{Value: paymentAmt}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
invoiceResp, err := dave.AddInvoice(ctxt, invoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to create invoice: %v", err)
|
|
}
|
|
|
|
sendReq := &lnrpc.SendRequest{
|
|
PaymentRequest: invoiceResp.PaymentRequest,
|
|
FeeLimit: feeLimit,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
paymentResp, err := net.Alice.SendPaymentSync(ctxt, sendReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to send payment: %v", err)
|
|
}
|
|
if paymentResp.PaymentError != "" {
|
|
t.Fatalf("unable to send payment: %v",
|
|
paymentResp.PaymentError)
|
|
}
|
|
|
|
checkRoute(paymentResp.PaymentRoute)
|
|
}
|
|
|
|
// We'll start off using percentages first. Since the fee along the
|
|
// route using Carol as an intermediate hop is 10% of the payment's
|
|
// amount, we'll use a lower percentage in order to invalid that route.
|
|
feeLimitPercent := &lnrpc.FeeLimit{
|
|
Limit: &lnrpc.FeeLimit_Percent{
|
|
Percent: baseFee/1000 - 1,
|
|
},
|
|
}
|
|
testFeeCutoff(feeLimitPercent)
|
|
|
|
// Now we'll test using fixed fee limit amounts. Since we computed the
|
|
// fee for the route using Carol as an intermediate hop earlier, we can
|
|
// use a smaller value in order to invalidate that route.
|
|
feeLimitFixed := &lnrpc.FeeLimit{
|
|
Limit: &lnrpc.FeeLimit_Fixed{
|
|
Fixed: int64(carolFee.ToSatoshis()) - 1,
|
|
},
|
|
}
|
|
testFeeCutoff(feeLimitFixed)
|
|
|
|
// Once we're done, close the channels and shut down the nodes created
|
|
// throughout this test.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointAliceBob, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Alice, chanPointAliceCarol, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Bob, chanPointBobDave, false)
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, carol, chanPointCarolDave, false)
|
|
}
|
|
|
|
// testSendUpdateDisableChannel ensures that a channel update with the disable
|
|
// flag set is sent once a channel has been either unilaterally or cooperatively
|
|
// closed.
|
|
func testSendUpdateDisableChannel(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const (
|
|
chanAmt = 100000
|
|
)
|
|
|
|
// Open a channel between Alice and Bob and Alice and Carol. These will
|
|
// be closed later on in order to trigger channel update messages
|
|
// marking the channels as disabled.
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAliceBob := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
carol, err := net.NewNode("Carol", []string{
|
|
"--minbackoff=10s",
|
|
"--chan-enable-timeout=1.5s",
|
|
"--chan-disable-timeout=3s",
|
|
"--chan-status-sample-interval=.5s",
|
|
})
|
|
if err != nil {
|
|
t.Fatalf("unable to create carol's node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, net.Alice, carol); err != nil {
|
|
t.Fatalf("unable to connect alice to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAliceCarol := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// We create a new node Eve that has an inactive channel timeout of
|
|
// just 2 seconds (down from the default 20m). It will be used to test
|
|
// channel updates for channels going inactive.
|
|
eve, err := net.NewNode("Eve", []string{
|
|
"--minbackoff=10s",
|
|
"--chan-enable-timeout=1.5s",
|
|
"--chan-disable-timeout=3s",
|
|
"--chan-status-sample-interval=.5s",
|
|
})
|
|
if err != nil {
|
|
t.Fatalf("unable to create eve's node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, eve)
|
|
|
|
// Give Eve some coins.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, eve)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to eve: %v", err)
|
|
}
|
|
|
|
// Connect Eve to Carol and Bob, and open a channel to carol.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, eve, carol); err != nil {
|
|
t.Fatalf("unable to connect alice to carol: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, eve, net.Bob); err != nil {
|
|
t.Fatalf("unable to connect eve to bob: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointEveCarol := openChannelAndAssert(
|
|
ctxt, t, net, eve, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Launch a node for Dave which will connect to Bob in order to receive
|
|
// graph updates from. This will ensure that the channel updates are
|
|
// propagated throughout the network.
|
|
dave, err := net.NewNode("Dave", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create dave's node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, net.Bob, dave); err != nil {
|
|
t.Fatalf("unable to connect bob to dave: %v", err)
|
|
}
|
|
|
|
daveSub := subscribeGraphNotifications(t, ctxb, dave)
|
|
defer close(daveSub.quit)
|
|
|
|
// We should expect to see a channel update with the default routing
|
|
// policy, except that it should indicate the channel is disabled.
|
|
expectedPolicy := &lnrpc.RoutingPolicy{
|
|
FeeBaseMsat: int64(lnd.DefaultBitcoinBaseFeeMSat),
|
|
FeeRateMilliMsat: int64(lnd.DefaultBitcoinFeeRate),
|
|
TimeLockDelta: lnd.DefaultBitcoinTimeLockDelta,
|
|
MinHtlc: 1000, // default value
|
|
MaxHtlcMsat: calculateMaxHtlc(chanAmt),
|
|
Disabled: true,
|
|
}
|
|
|
|
// Let Carol go offline. Since Eve has an inactive timeout of 2s, we
|
|
// expect her to send an update disabling the channel.
|
|
restartCarol, err := net.SuspendNode(carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to suspend carol: %v", err)
|
|
}
|
|
waitForChannelUpdate(
|
|
t, daveSub,
|
|
[]expectedChanUpdate{
|
|
{eve.PubKeyStr, expectedPolicy, chanPointEveCarol},
|
|
},
|
|
)
|
|
|
|
// We restart Carol. Since the channel now becomes active again, Eve
|
|
// should send a ChannelUpdate setting the channel no longer disabled.
|
|
if err := restartCarol(); err != nil {
|
|
t.Fatalf("unable to restart carol: %v", err)
|
|
}
|
|
|
|
expectedPolicy.Disabled = false
|
|
waitForChannelUpdate(
|
|
t, daveSub,
|
|
[]expectedChanUpdate{
|
|
{eve.PubKeyStr, expectedPolicy, chanPointEveCarol},
|
|
},
|
|
)
|
|
|
|
// Now we'll test a long disconnection. Disconnect Carol and Eve and
|
|
// ensure they both detect each other as disabled. Their min backoffs
|
|
// are high enough to not interfere with disabling logic.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.DisconnectNodes(ctxt, carol, eve); err != nil {
|
|
t.Fatalf("unable to disconnect Carol from Eve: %v", err)
|
|
}
|
|
|
|
// Wait for a disable from both Carol and Eve to come through.
|
|
expectedPolicy.Disabled = true
|
|
waitForChannelUpdate(
|
|
t, daveSub,
|
|
[]expectedChanUpdate{
|
|
{eve.PubKeyStr, expectedPolicy, chanPointEveCarol},
|
|
{carol.PubKeyStr, expectedPolicy, chanPointEveCarol},
|
|
},
|
|
)
|
|
|
|
// Reconnect Carol and Eve, this should cause them to reenable the
|
|
// channel from both ends after a short delay.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.EnsureConnected(ctxt, carol, eve); err != nil {
|
|
t.Fatalf("unable to reconnect Carol to Eve: %v", err)
|
|
}
|
|
|
|
expectedPolicy.Disabled = false
|
|
waitForChannelUpdate(
|
|
t, daveSub,
|
|
[]expectedChanUpdate{
|
|
{eve.PubKeyStr, expectedPolicy, chanPointEveCarol},
|
|
{carol.PubKeyStr, expectedPolicy, chanPointEveCarol},
|
|
},
|
|
)
|
|
|
|
// Now we'll test a short disconnection. Disconnect Carol and Eve, then
|
|
// reconnect them after one second so that their scheduled disables are
|
|
// aborted. One second is twice the status sample interval, so this
|
|
// should allow for the disconnect to be detected, but still leave time
|
|
// to cancel the announcement before the 3 second inactive timeout is
|
|
// hit.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.DisconnectNodes(ctxt, carol, eve); err != nil {
|
|
t.Fatalf("unable to disconnect Carol from Eve: %v", err)
|
|
}
|
|
time.Sleep(time.Second)
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.EnsureConnected(ctxt, eve, carol); err != nil {
|
|
t.Fatalf("unable to reconnect Carol to Eve: %v", err)
|
|
}
|
|
|
|
// Since the disable should have been canceled by both Carol and Eve, we
|
|
// expect no channel updates to appear on the network.
|
|
assertNoChannelUpdates(t, daveSub, 4*time.Second)
|
|
|
|
// Close Alice's channels with Bob and Carol cooperatively and
|
|
// unilaterally respectively.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
_, _, err = net.CloseChannel(ctxt, net.Alice, chanPointAliceBob, false)
|
|
if err != nil {
|
|
t.Fatalf("unable to close channel: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
_, _, err = net.CloseChannel(ctxt, net.Alice, chanPointAliceCarol, true)
|
|
if err != nil {
|
|
t.Fatalf("unable to close channel: %v", err)
|
|
}
|
|
|
|
// Now that the channel close processes have been started, we should
|
|
// receive an update marking each as disabled.
|
|
expectedPolicy.Disabled = true
|
|
waitForChannelUpdate(
|
|
t, daveSub,
|
|
[]expectedChanUpdate{
|
|
{net.Alice.PubKeyStr, expectedPolicy, chanPointAliceBob},
|
|
{net.Alice.PubKeyStr, expectedPolicy, chanPointAliceCarol},
|
|
},
|
|
)
|
|
|
|
// Finally, close the channels by mining the closing transactions.
|
|
mineBlocks(t, net, 1, 2)
|
|
|
|
// Also do this check for Eve's channel with Carol.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
_, _, err = net.CloseChannel(ctxt, eve, chanPointEveCarol, false)
|
|
if err != nil {
|
|
t.Fatalf("unable to close channel: %v", err)
|
|
}
|
|
|
|
waitForChannelUpdate(
|
|
t, daveSub,
|
|
[]expectedChanUpdate{
|
|
{eve.PubKeyStr, expectedPolicy, chanPointEveCarol},
|
|
},
|
|
)
|
|
mineBlocks(t, net, 1, 1)
|
|
|
|
// And finally, clean up the force closed channel by mining the
|
|
// sweeping transaction.
|
|
cleanupForceClose(t, net, net.Alice, chanPointAliceCarol)
|
|
}
|
|
|
|
// testAbandonChannel abandones a channel and asserts that it is no
|
|
// longer open and not in one of the pending closure states. It also
|
|
// verifies that the abandoned channel is reported as closed with close
|
|
// type 'abandoned'.
|
|
func testAbandonChannel(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
// First establish a channel between Alice and Bob.
|
|
channelParam := lntest.OpenChannelParams{
|
|
Amt: lnd.MaxBtcFundingAmount,
|
|
PushAmt: btcutil.Amount(100000),
|
|
}
|
|
|
|
ctxt, _ := context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, net.Bob, channelParam,
|
|
)
|
|
txid, err := lnd.GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("unable to get txid: %v", err)
|
|
}
|
|
chanPointStr := fmt.Sprintf("%v:%v", txid, chanPoint.OutputIndex)
|
|
|
|
// Wait for channel to be confirmed open.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Alice.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't report channel: %v", err)
|
|
}
|
|
err = net.Bob.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("bob didn't report channel: %v", err)
|
|
}
|
|
|
|
// Now that the channel is open, we'll obtain its channel ID real quick
|
|
// so we can use it to query the graph below.
|
|
listReq := &lnrpc.ListChannelsRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
aliceChannelList, err := net.Alice.ListChannels(ctxt, listReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to fetch alice's channels: %v", err)
|
|
}
|
|
var chanID uint64
|
|
for _, channel := range aliceChannelList.Channels {
|
|
if channel.ChannelPoint == chanPointStr {
|
|
chanID = channel.ChanId
|
|
}
|
|
}
|
|
|
|
if chanID == 0 {
|
|
t.Fatalf("unable to find channel")
|
|
}
|
|
|
|
// To make sure the channel is removed from the backup file as well when
|
|
// being abandoned, grab a backup snapshot so we can compare it with the
|
|
// later state.
|
|
bkupBefore, err := ioutil.ReadFile(net.Alice.ChanBackupPath())
|
|
if err != nil {
|
|
t.Fatalf("could not get channel backup before abandoning "+
|
|
"channel: %v", err)
|
|
}
|
|
|
|
// Send request to abandon channel.
|
|
abandonChannelRequest := &lnrpc.AbandonChannelRequest{
|
|
ChannelPoint: chanPoint,
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
_, err = net.Alice.AbandonChannel(ctxt, abandonChannelRequest)
|
|
if err != nil {
|
|
t.Fatalf("unable to abandon channel: %v", err)
|
|
}
|
|
|
|
// Assert that channel in no longer open.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
aliceChannelList, err = net.Alice.ListChannels(ctxt, listReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to list channels: %v", err)
|
|
}
|
|
if len(aliceChannelList.Channels) != 0 {
|
|
t.Fatalf("alice should only have no channels open, "+
|
|
"instead she has %v",
|
|
len(aliceChannelList.Channels))
|
|
}
|
|
|
|
// Assert that channel is not pending closure.
|
|
pendingReq := &lnrpc.PendingChannelsRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
alicePendingList, err := net.Alice.PendingChannels(ctxt, pendingReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to list pending channels: %v", err)
|
|
}
|
|
if len(alicePendingList.PendingClosingChannels) != 0 {
|
|
t.Fatalf("alice should only have no pending closing channels, "+
|
|
"instead she has %v",
|
|
len(alicePendingList.PendingClosingChannels))
|
|
}
|
|
if len(alicePendingList.PendingForceClosingChannels) != 0 {
|
|
t.Fatalf("alice should only have no pending force closing "+
|
|
"channels instead she has %v",
|
|
len(alicePendingList.PendingForceClosingChannels))
|
|
}
|
|
if len(alicePendingList.WaitingCloseChannels) != 0 {
|
|
t.Fatalf("alice should only have no waiting close "+
|
|
"channels instead she has %v",
|
|
len(alicePendingList.WaitingCloseChannels))
|
|
}
|
|
|
|
// Assert that channel is listed as abandoned.
|
|
closedReq := &lnrpc.ClosedChannelsRequest{
|
|
Abandoned: true,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
aliceClosedList, err := net.Alice.ClosedChannels(ctxt, closedReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to list closed channels: %v", err)
|
|
}
|
|
if len(aliceClosedList.Channels) != 1 {
|
|
t.Fatalf("alice should only have a single abandoned channel, "+
|
|
"instead she has %v",
|
|
len(aliceClosedList.Channels))
|
|
}
|
|
|
|
// Ensure that the channel can no longer be found in the channel graph.
|
|
_, err = net.Alice.GetChanInfo(ctxb, &lnrpc.ChanInfoRequest{
|
|
ChanId: chanID,
|
|
})
|
|
if !strings.Contains(err.Error(), "marked as zombie") {
|
|
t.Fatalf("channel shouldn't be found in the channel " +
|
|
"graph!")
|
|
}
|
|
|
|
// Make sure the channel is no longer in the channel backup list.
|
|
err = wait.Predicate(func() bool {
|
|
bkupAfter, err := ioutil.ReadFile(net.Alice.ChanBackupPath())
|
|
if err != nil {
|
|
t.Fatalf("could not get channel backup before "+
|
|
"abandoning channel: %v", err)
|
|
}
|
|
|
|
return len(bkupAfter) < len(bkupBefore)
|
|
}, defaultTimeout)
|
|
if err != nil {
|
|
t.Fatalf("channel wasn't removed from channel backup file")
|
|
}
|
|
|
|
// Calling AbandonChannel again, should result in no new errors, as the
|
|
// channel has already been removed.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
_, err = net.Alice.AbandonChannel(ctxt, abandonChannelRequest)
|
|
if err != nil {
|
|
t.Fatalf("unable to abandon channel a second time: %v", err)
|
|
}
|
|
|
|
// Now that we're done with the test, the channel can be closed. This
|
|
// is necessary to avoid unexpected outcomes of other tests that use
|
|
// Bob's lnd instance.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, net.Bob, chanPoint, true)
|
|
|
|
// Cleanup by mining the force close and sweep transaction.
|
|
cleanupForceClose(t, net, net.Bob, chanPoint)
|
|
}
|
|
|
|
// testSweepAllCoins tests that we're able to properly sweep all coins from the
|
|
// wallet into a single target address at the specified fee rate.
|
|
func testSweepAllCoins(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
// First, we'll make a new node, ainz who'll we'll use to test wallet
|
|
// sweeping.
|
|
ainz, err := net.NewNode("Ainz", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, ainz)
|
|
|
|
// Next, we'll give Ainz exactly 2 utxos of 1 BTC each, with one of
|
|
// them being p2wkh and the other being a n2wpkh address.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, ainz)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to eve: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoinsNP2WKH(ctxt, btcutil.SatoshiPerBitcoin, ainz)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to eve: %v", err)
|
|
}
|
|
|
|
// Ensure that we can't send coins to our own Pubkey.
|
|
info, err := ainz.GetInfo(ctxt, &lnrpc.GetInfoRequest{})
|
|
if err != nil {
|
|
t.Fatalf("unable to get node info: %v", err)
|
|
}
|
|
|
|
sweepReq := &lnrpc.SendCoinsRequest{
|
|
Addr: info.IdentityPubkey,
|
|
SendAll: true,
|
|
}
|
|
_, err = ainz.SendCoins(ctxt, sweepReq)
|
|
if err == nil {
|
|
t.Fatalf("expected SendCoins to users own pubkey to fail")
|
|
}
|
|
|
|
// Ensure that we can't send coins to another users Pubkey.
|
|
info, err = net.Alice.GetInfo(ctxt, &lnrpc.GetInfoRequest{})
|
|
if err != nil {
|
|
t.Fatalf("unable to get node info: %v", err)
|
|
}
|
|
|
|
sweepReq = &lnrpc.SendCoinsRequest{
|
|
Addr: info.IdentityPubkey,
|
|
SendAll: true,
|
|
}
|
|
_, err = ainz.SendCoins(ctxt, sweepReq)
|
|
if err == nil {
|
|
t.Fatalf("expected SendCoins to Alices pubkey to fail")
|
|
}
|
|
|
|
// With the two coins above mined, we'll now instruct ainz to sweep all
|
|
// the coins to an external address not under its control.
|
|
// We will first attempt to send the coins to addresses that are not
|
|
// compatible with the current network. This is to test that the wallet
|
|
// will prevent any onchain transactions to addresses that are not on the
|
|
// same network as the user.
|
|
|
|
// Send coins to a testnet3 address.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
sweepReq = &lnrpc.SendCoinsRequest{
|
|
Addr: "tb1qfc8fusa98jx8uvnhzavxccqlzvg749tvjw82tg",
|
|
SendAll: true,
|
|
}
|
|
_, err = ainz.SendCoins(ctxt, sweepReq)
|
|
if err == nil {
|
|
t.Fatalf("expected SendCoins to different network to fail")
|
|
}
|
|
|
|
// Send coins to a mainnet address.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
sweepReq = &lnrpc.SendCoinsRequest{
|
|
Addr: "1MPaXKp5HhsLNjVSqaL7fChE3TVyrTMRT3",
|
|
SendAll: true,
|
|
}
|
|
_, err = ainz.SendCoins(ctxt, sweepReq)
|
|
if err == nil {
|
|
t.Fatalf("expected SendCoins to different network to fail")
|
|
}
|
|
|
|
// Send coins to a compatible address.
|
|
minerAddr, err := net.Miner.NewAddress()
|
|
if err != nil {
|
|
t.Fatalf("unable to create new miner addr: %v", err)
|
|
}
|
|
|
|
sweepReq = &lnrpc.SendCoinsRequest{
|
|
Addr: minerAddr.String(),
|
|
SendAll: true,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
_, err = ainz.SendCoins(ctxt, sweepReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to sweep coins: %v", err)
|
|
}
|
|
|
|
// We'll mine a block which should include the sweep transaction we
|
|
// generated above.
|
|
block := mineBlocks(t, net, 1, 1)[0]
|
|
|
|
// The sweep transaction should have exactly two inputs as we only had
|
|
// two UTXOs in the wallet.
|
|
sweepTx := block.Transactions[1]
|
|
if len(sweepTx.TxIn) != 2 {
|
|
t.Fatalf("expected 2 inputs instead have %v", len(sweepTx.TxIn))
|
|
}
|
|
|
|
// Finally, Ainz should now have no coins at all within his wallet.
|
|
balReq := &lnrpc.WalletBalanceRequest{}
|
|
resp, err := ainz.WalletBalance(ctxt, balReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get ainz's balance: %v", err)
|
|
}
|
|
switch {
|
|
case resp.ConfirmedBalance != 0:
|
|
t.Fatalf("expected no confirmed balance, instead have %v",
|
|
resp.ConfirmedBalance)
|
|
|
|
case resp.UnconfirmedBalance != 0:
|
|
t.Fatalf("expected no unconfirmed balance, instead have %v",
|
|
resp.UnconfirmedBalance)
|
|
}
|
|
|
|
// If we try again, but this time specifying an amount, then the call
|
|
// should fail.
|
|
sweepReq.Amount = 10000
|
|
_, err = ainz.SendCoins(ctxt, sweepReq)
|
|
if err == nil {
|
|
t.Fatalf("sweep attempt should fail")
|
|
}
|
|
}
|
|
|
|
// testChannelBackupUpdates tests that both the streaming channel update RPC,
|
|
// and the on-disk channels.backup are updated each time a channel is
|
|
// opened/closed.
|
|
func testChannelBackupUpdates(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
// First, we'll make a temp directory that we'll use to store our
|
|
// backup file, so we can check in on it during the test easily.
|
|
backupDir, err := ioutil.TempDir("", "")
|
|
if err != nil {
|
|
t.Fatalf("unable to create backup dir: %v", err)
|
|
}
|
|
defer os.RemoveAll(backupDir)
|
|
|
|
// First, we'll create a new node, Carol. We'll also create a temporary
|
|
// file that Carol will use to store her channel backups.
|
|
backupFilePath := filepath.Join(
|
|
backupDir, chanbackup.DefaultBackupFileName,
|
|
)
|
|
carolArgs := fmt.Sprintf("--backupfilepath=%v", backupFilePath)
|
|
carol, err := net.NewNode("carol", []string{carolArgs})
|
|
if err != nil {
|
|
t.Fatalf("unable to create new node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
// Next, we'll register for streaming notifications for changes to the
|
|
// backup file.
|
|
backupStream, err := carol.SubscribeChannelBackups(
|
|
ctxb, &lnrpc.ChannelBackupSubscription{},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create backup stream: %v", err)
|
|
}
|
|
|
|
// We'll use this goroutine to proxy any updates to a channel we can
|
|
// easily use below.
|
|
var wg sync.WaitGroup
|
|
backupUpdates := make(chan *lnrpc.ChanBackupSnapshot)
|
|
streamErr := make(chan error)
|
|
streamQuit := make(chan struct{})
|
|
|
|
wg.Add(1)
|
|
go func() {
|
|
defer wg.Done()
|
|
for {
|
|
snapshot, err := backupStream.Recv()
|
|
if err != nil {
|
|
select {
|
|
case streamErr <- err:
|
|
case <-streamQuit:
|
|
return
|
|
}
|
|
}
|
|
|
|
select {
|
|
case backupUpdates <- snapshot:
|
|
case <-streamQuit:
|
|
return
|
|
}
|
|
}
|
|
}()
|
|
defer close(streamQuit)
|
|
|
|
// With Carol up, we'll now connect her to Alice, and open a channel
|
|
// between them.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, net.Alice); err != nil {
|
|
t.Fatalf("unable to connect carol to alice: %v", err)
|
|
}
|
|
|
|
// Next, we'll open two channels between Alice and Carol back to back.
|
|
var chanPoints []*lnrpc.ChannelPoint
|
|
numChans := 2
|
|
chanAmt := btcutil.Amount(1000000)
|
|
for i := 0; i < numChans; i++ {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
chanPoints = append(chanPoints, chanPoint)
|
|
}
|
|
|
|
// Using this helper function, we'll maintain a pointer to the latest
|
|
// channel backup so we can compare it to the on disk state.
|
|
var currentBackup *lnrpc.ChanBackupSnapshot
|
|
assertBackupNtfns := func(numNtfns int) {
|
|
for i := 0; i < numNtfns; i++ {
|
|
select {
|
|
case err := <-streamErr:
|
|
t.Fatalf("error with backup stream: %v", err)
|
|
|
|
case currentBackup = <-backupUpdates:
|
|
|
|
case <-time.After(time.Second * 5):
|
|
t.Fatalf("didn't receive channel backup "+
|
|
"notification %v", i+1)
|
|
}
|
|
}
|
|
}
|
|
|
|
// assertBackupFileState is a helper function that we'll use to compare
|
|
// the on disk back up file to our currentBackup pointer above.
|
|
assertBackupFileState := func() {
|
|
err := wait.NoError(func() error {
|
|
packedBackup, err := ioutil.ReadFile(backupFilePath)
|
|
if err != nil {
|
|
return fmt.Errorf("unable to read backup "+
|
|
"file: %v", err)
|
|
}
|
|
|
|
// As each back up file will be encrypted with a fresh
|
|
// nonce, we can't compare them directly, so instead
|
|
// we'll compare the length which is a proxy for the
|
|
// number of channels that the multi-backup contains.
|
|
rawBackup := currentBackup.MultiChanBackup.MultiChanBackup
|
|
if len(rawBackup) != len(packedBackup) {
|
|
return fmt.Errorf("backup files don't match: "+
|
|
"expected %x got %x", rawBackup, packedBackup)
|
|
}
|
|
|
|
// Additionally, we'll assert that both backups up
|
|
// returned are valid.
|
|
for i, backup := range [][]byte{rawBackup, packedBackup} {
|
|
snapshot := &lnrpc.ChanBackupSnapshot{
|
|
MultiChanBackup: &lnrpc.MultiChanBackup{
|
|
MultiChanBackup: backup,
|
|
},
|
|
}
|
|
_, err := carol.VerifyChanBackup(ctxb, snapshot)
|
|
if err != nil {
|
|
return fmt.Errorf("unable to verify "+
|
|
"backup #%d: %v", i, err)
|
|
}
|
|
}
|
|
|
|
return nil
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("backup state invalid: %v", err)
|
|
}
|
|
}
|
|
|
|
// As these two channels were just opened, we should've got two times
|
|
// the pending and open notifications for channel backups.
|
|
assertBackupNtfns(2 * 2)
|
|
|
|
// The on disk file should also exactly match the latest backup that we
|
|
// have.
|
|
assertBackupFileState()
|
|
|
|
// Next, we'll close the channels one by one. After each channel
|
|
// closure, we should get a notification, and the on-disk state should
|
|
// match this state as well.
|
|
for i := 0; i < numChans; i++ {
|
|
// To ensure force closes also trigger an update, we'll force
|
|
// close half of the channels.
|
|
forceClose := i%2 == 0
|
|
|
|
chanPoint := chanPoints[i]
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(
|
|
ctxt, t, net, net.Alice, chanPoint, forceClose,
|
|
)
|
|
|
|
// We should get a single notification after closing, and the
|
|
// on-disk state should match this latest notifications.
|
|
assertBackupNtfns(1)
|
|
assertBackupFileState()
|
|
|
|
// If we force closed the channel, then we'll mine enough
|
|
// blocks to ensure all outputs have been swept.
|
|
if forceClose {
|
|
cleanupForceClose(t, net, net.Alice, chanPoint)
|
|
}
|
|
}
|
|
}
|
|
|
|
// testExportChannelBackup tests that we're able to properly export either a
|
|
// targeted channel's backup, or export backups of all the currents open
|
|
// channels.
|
|
func testExportChannelBackup(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
// First, we'll create our primary test node: Carol. We'll use Carol to
|
|
// open channels and also export backups that we'll examine throughout
|
|
// the test.
|
|
carol, err := net.NewNode("carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
// With Carol up, we'll now connect her to Alice, and open a channel
|
|
// between them.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxt, carol, net.Alice); err != nil {
|
|
t.Fatalf("unable to connect carol to alice: %v", err)
|
|
}
|
|
|
|
// Next, we'll open two channels between Alice and Carol back to back.
|
|
var chanPoints []*lnrpc.ChannelPoint
|
|
numChans := 2
|
|
chanAmt := btcutil.Amount(1000000)
|
|
for i := 0; i < numChans; i++ {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
chanPoints = append(chanPoints, chanPoint)
|
|
}
|
|
|
|
// Now that the channels are open, we should be able to fetch the
|
|
// backups of each of the channels.
|
|
for _, chanPoint := range chanPoints {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
req := &lnrpc.ExportChannelBackupRequest{
|
|
ChanPoint: chanPoint,
|
|
}
|
|
chanBackup, err := carol.ExportChannelBackup(ctxt, req)
|
|
if err != nil {
|
|
t.Fatalf("unable to fetch backup for channel %v: %v",
|
|
chanPoint, err)
|
|
}
|
|
|
|
// The returned backup should be full populated. Since it's
|
|
// encrypted, we can't assert any more than that atm.
|
|
if len(chanBackup.ChanBackup) == 0 {
|
|
t.Fatalf("obtained empty backup for channel: %v", chanPoint)
|
|
}
|
|
|
|
// The specified chanPoint in the response should match our
|
|
// requested chanPoint.
|
|
if chanBackup.ChanPoint.String() != chanPoint.String() {
|
|
t.Fatalf("chanPoint mismatched: expected %v, got %v",
|
|
chanPoint.String(),
|
|
chanBackup.ChanPoint.String())
|
|
}
|
|
}
|
|
|
|
// Before we proceed, we'll make two utility methods we'll use below
|
|
// for our primary assertions.
|
|
assertNumSingleBackups := func(numSingles int) {
|
|
err := wait.NoError(func() error {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
req := &lnrpc.ChanBackupExportRequest{}
|
|
chanSnapshot, err := carol.ExportAllChannelBackups(
|
|
ctxt, req,
|
|
)
|
|
if err != nil {
|
|
return fmt.Errorf("unable to export channel "+
|
|
"backup: %v", err)
|
|
}
|
|
|
|
if chanSnapshot.SingleChanBackups == nil {
|
|
return fmt.Errorf("single chan backups not " +
|
|
"populated")
|
|
}
|
|
|
|
backups := chanSnapshot.SingleChanBackups.ChanBackups
|
|
if len(backups) != numSingles {
|
|
return fmt.Errorf("expected %v singles, "+
|
|
"got %v", len(backups), numSingles)
|
|
}
|
|
|
|
return nil
|
|
}, defaultTimeout)
|
|
if err != nil {
|
|
t.Fatalf(err.Error())
|
|
}
|
|
}
|
|
assertMultiBackupFound := func() func(bool, map[wire.OutPoint]struct{}) {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
req := &lnrpc.ChanBackupExportRequest{}
|
|
chanSnapshot, err := carol.ExportAllChannelBackups(ctxt, req)
|
|
if err != nil {
|
|
t.Fatalf("unable to export channel backup: %v", err)
|
|
}
|
|
|
|
return func(found bool, chanPoints map[wire.OutPoint]struct{}) {
|
|
switch {
|
|
case found && chanSnapshot.MultiChanBackup == nil:
|
|
t.Fatalf("multi-backup not present")
|
|
|
|
case !found && chanSnapshot.MultiChanBackup != nil &&
|
|
(len(chanSnapshot.MultiChanBackup.MultiChanBackup) !=
|
|
chanbackup.NilMultiSizePacked):
|
|
|
|
t.Fatalf("found multi-backup when non should " +
|
|
"be found")
|
|
}
|
|
|
|
if !found {
|
|
return
|
|
}
|
|
|
|
backedUpChans := chanSnapshot.MultiChanBackup.ChanPoints
|
|
if len(chanPoints) != len(backedUpChans) {
|
|
t.Fatalf("expected %v chans got %v", len(chanPoints),
|
|
len(backedUpChans))
|
|
}
|
|
|
|
for _, chanPoint := range backedUpChans {
|
|
wirePoint := rpcPointToWirePoint(t, chanPoint)
|
|
if _, ok := chanPoints[wirePoint]; !ok {
|
|
t.Fatalf("unexpected backup: %v", wirePoint)
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
chans := make(map[wire.OutPoint]struct{})
|
|
for _, chanPoint := range chanPoints {
|
|
chans[rpcPointToWirePoint(t, chanPoint)] = struct{}{}
|
|
}
|
|
|
|
// We should have exactly two single channel backups contained, and we
|
|
// should also have a multi-channel backup.
|
|
assertNumSingleBackups(2)
|
|
assertMultiBackupFound()(true, chans)
|
|
|
|
// We'll now close each channel on by one. After we close a channel, we
|
|
// shouldn't be able to find that channel as a backup still. We should
|
|
// also have one less single written to disk.
|
|
for i, chanPoint := range chanPoints {
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(
|
|
ctxt, t, net, net.Alice, chanPoint, false,
|
|
)
|
|
|
|
assertNumSingleBackups(len(chanPoints) - i - 1)
|
|
|
|
delete(chans, rpcPointToWirePoint(t, chanPoint))
|
|
assertMultiBackupFound()(true, chans)
|
|
}
|
|
|
|
// At this point we shouldn't have any single or multi-chan backups at
|
|
// all.
|
|
assertNumSingleBackups(0)
|
|
assertMultiBackupFound()(false, nil)
|
|
}
|
|
|
|
// nodeRestorer is a function closure that allows each chanRestoreTestCase to
|
|
// control exactly *how* the prior node is restored. This might be using an
|
|
// backup obtained over RPC, or the file system, etc.
|
|
type nodeRestorer func() (*lntest.HarnessNode, error)
|
|
|
|
// chanRestoreTestCase describes a test case for an end to end SCB restoration
|
|
// work flow. One node will start from scratch using an existing SCB. At the
|
|
// end of the est, both nodes should be made whole via the DLP protocol.
|
|
type chanRestoreTestCase struct {
|
|
// name is the name of the target test case.
|
|
name string
|
|
|
|
// channelsUpdated is false then this means that no updates
|
|
// have taken place within the channel before restore.
|
|
// Otherwise, HTLCs will be settled between the two parties
|
|
// before restoration modifying the balance beyond the initial
|
|
// allocation.
|
|
channelsUpdated bool
|
|
|
|
// initiator signals if Dave should be the one that opens the
|
|
// channel to Alice, or if it should be the other way around.
|
|
initiator bool
|
|
|
|
// private signals if the channel from Dave to Carol should be
|
|
// private or not.
|
|
private bool
|
|
|
|
// unconfirmed signals if the channel from Dave to Carol should be
|
|
// confirmed or not.
|
|
unconfirmed bool
|
|
|
|
// anchorCommit is true, then the new anchor commitment type will be
|
|
// used for the channels created in the test.
|
|
anchorCommit bool
|
|
|
|
// restoreMethod takes an old node, then returns a function
|
|
// closure that'll return the same node, but with its state
|
|
// restored via a custom method. We use this to abstract away
|
|
// _how_ a node is restored from our assertions once the node
|
|
// has been fully restored itself.
|
|
restoreMethod func(oldNode *lntest.HarnessNode,
|
|
backupFilePath string,
|
|
mnemonic []string) (nodeRestorer, error)
|
|
}
|
|
|
|
// testChanRestoreScenario executes a chanRestoreTestCase from end to end,
|
|
// ensuring that after Dave restores his channel state according to the
|
|
// testCase, the DLP protocol is executed properly and both nodes are made
|
|
// whole.
|
|
func testChanRestoreScenario(t *harnessTest, net *lntest.NetworkHarness,
|
|
testCase *chanRestoreTestCase, password []byte) {
|
|
|
|
const (
|
|
chanAmt = btcutil.Amount(10000000)
|
|
pushAmt = btcutil.Amount(5000000)
|
|
)
|
|
|
|
ctxb := context.Background()
|
|
|
|
var nodeArgs []string
|
|
if testCase.anchorCommit {
|
|
nodeArgs = commitTypeAnchors.Args()
|
|
}
|
|
|
|
// First, we'll create a brand new node we'll use within the test. If
|
|
// we have a custom backup file specified, then we'll also create that
|
|
// for use.
|
|
dave, mnemonic, err := net.NewNodeWithSeed(
|
|
"dave", nodeArgs, password,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new node: %v", err)
|
|
}
|
|
// Defer to a closure instead of to shutdownAndAssert due to the value
|
|
// of 'dave' changing throughout the test.
|
|
defer func() {
|
|
shutdownAndAssert(net, t, dave)
|
|
}()
|
|
carol, err := net.NewNode("carol", nodeArgs)
|
|
if err != nil {
|
|
t.Fatalf("unable to make new node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
// Now that our new nodes are created, we'll give them some coins for
|
|
// channel opening and anchor sweeping.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to dave: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to dave: %v", err)
|
|
}
|
|
|
|
var from, to *lntest.HarnessNode
|
|
if testCase.initiator {
|
|
from, to = dave, carol
|
|
} else {
|
|
from, to = carol, dave
|
|
}
|
|
|
|
// Next, we'll connect Dave to Carol, and open a new channel to her
|
|
// with a portion pushed.
|
|
if err := net.ConnectNodes(ctxt, dave, carol); err != nil {
|
|
t.Fatalf("unable to connect dave to carol: %v", err)
|
|
}
|
|
|
|
// We will either open a confirmed or unconfirmed channel, depending on
|
|
// the requirements of the test case.
|
|
switch {
|
|
case testCase.unconfirmed:
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
_, err := net.OpenPendingChannel(
|
|
ctxt, from, to, chanAmt, pushAmt,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("couldn't open pending channel: %v", err)
|
|
}
|
|
|
|
// Give the pubsub some time to update the channel backup.
|
|
err = wait.NoError(func() error {
|
|
fi, err := os.Stat(dave.ChanBackupPath())
|
|
if err != nil {
|
|
return err
|
|
}
|
|
if fi.Size() <= chanbackup.NilMultiSizePacked {
|
|
return fmt.Errorf("backup file empty")
|
|
}
|
|
return nil
|
|
}, defaultTimeout)
|
|
if err != nil {
|
|
t.Fatalf("channel backup not updated in time: %v", err)
|
|
}
|
|
|
|
default:
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPoint := openChannelAndAssert(
|
|
ctxt, t, net, from, to,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
PushAmt: pushAmt,
|
|
Private: testCase.private,
|
|
},
|
|
)
|
|
|
|
// Wait for both sides to see the opened channel.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = dave.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("dave didn't report channel: %v", err)
|
|
}
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPoint)
|
|
if err != nil {
|
|
t.Fatalf("carol didn't report channel: %v", err)
|
|
}
|
|
}
|
|
|
|
// If both parties should start with existing channel updates, then
|
|
// we'll send+settle an HTLC between 'from' and 'to' now.
|
|
if testCase.channelsUpdated {
|
|
invoice := &lnrpc.Invoice{
|
|
Memo: "testing",
|
|
Value: 10000,
|
|
}
|
|
invoiceResp, err := to.AddInvoice(ctxt, invoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(
|
|
ctxt, from, []string{invoiceResp.PaymentRequest},
|
|
true,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to complete payments: %v", err)
|
|
}
|
|
}
|
|
|
|
// Before we start the recovery, we'll record the balances of both
|
|
// Carol and Dave to ensure they both sweep their coins at the end.
|
|
balReq := &lnrpc.WalletBalanceRequest{}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
carolBalResp, err := carol.WalletBalance(ctxt, balReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol's balance: %v", err)
|
|
}
|
|
carolStartingBalance := carolBalResp.ConfirmedBalance
|
|
|
|
daveBalance, err := dave.WalletBalance(ctxt, balReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol's balance: %v", err)
|
|
}
|
|
daveStartingBalance := daveBalance.ConfirmedBalance
|
|
|
|
// At this point, we'll now execute the restore method to give us the
|
|
// new node we should attempt our assertions against.
|
|
backupFilePath := dave.ChanBackupPath()
|
|
restoredNodeFunc, err := testCase.restoreMethod(
|
|
dave, backupFilePath, mnemonic,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to prep node restoration: %v", err)
|
|
}
|
|
|
|
// TODO(roasbeef): assert recovery state in channel
|
|
|
|
// Now that we're able to make our restored now, we'll shutdown the old
|
|
// Dave node as we'll be storing it shortly below.
|
|
shutdownAndAssert(net, t, dave)
|
|
|
|
// Next, we'll make a new Dave and start the bulk of our recovery
|
|
// workflow.
|
|
dave, err = restoredNodeFunc()
|
|
if err != nil {
|
|
t.Fatalf("unable to restore node: %v", err)
|
|
}
|
|
|
|
// First ensure that the on-chain balance is restored.
|
|
err = wait.NoError(func() error {
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
balReq := &lnrpc.WalletBalanceRequest{}
|
|
daveBalResp, err := dave.WalletBalance(ctxt, balReq)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
daveBal := daveBalResp.ConfirmedBalance
|
|
if daveBal <= 0 {
|
|
return fmt.Errorf("expected positive balance, had %v",
|
|
daveBal)
|
|
}
|
|
|
|
return nil
|
|
}, defaultTimeout)
|
|
if err != nil {
|
|
t.Fatalf("On-chain balance not restored: %v", err)
|
|
}
|
|
|
|
// Now that we have our new node up, we expect that it'll re-connect to
|
|
// Carol automatically based on the restored backup.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.EnsureConnected(ctxt, dave, carol)
|
|
if err != nil {
|
|
t.Fatalf("node didn't connect after recovery: %v", err)
|
|
}
|
|
|
|
// TODO(roasbeef): move dave restarts?
|
|
|
|
// Now we'll assert that both sides properly execute the DLP protocol.
|
|
// We grab their balances now to ensure that they're made whole at the
|
|
// end of the protocol.
|
|
assertDLPExecuted(
|
|
net, t, carol, carolStartingBalance, dave, daveStartingBalance,
|
|
testCase.anchorCommit,
|
|
)
|
|
}
|
|
|
|
// chanRestoreViaRPC is a helper test method that returns a nodeRestorer
|
|
// instance which will restore the target node from a password+seed, then
|
|
// trigger a SCB restore using the RPC interface.
|
|
func chanRestoreViaRPC(net *lntest.NetworkHarness,
|
|
password []byte, mnemonic []string,
|
|
multi []byte) (nodeRestorer, error) {
|
|
|
|
backup := &lnrpc.RestoreChanBackupRequest_MultiChanBackup{
|
|
MultiChanBackup: multi,
|
|
}
|
|
|
|
ctxb := context.Background()
|
|
|
|
return func() (*lntest.HarnessNode, error) {
|
|
newNode, err := net.RestoreNodeWithSeed(
|
|
"dave", nil, password, mnemonic, 1000, nil,
|
|
)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to "+
|
|
"restore node: %v", err)
|
|
}
|
|
|
|
_, err = newNode.RestoreChannelBackups(
|
|
ctxb, &lnrpc.RestoreChanBackupRequest{
|
|
Backup: backup,
|
|
},
|
|
)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable "+
|
|
"to restore backups: %v", err)
|
|
}
|
|
|
|
return newNode, nil
|
|
}, nil
|
|
}
|
|
|
|
// testChannelBackupRestore tests that we're able to recover from, and initiate
|
|
// the DLP protocol via: the RPC restore command, restoring on unlock, and
|
|
// restoring from initial wallet creation. We'll also alternate between
|
|
// restoring form the on disk file, and restoring from the exported RPC command
|
|
// as well.
|
|
func testChannelBackupRestore(net *lntest.NetworkHarness, t *harnessTest) {
|
|
password := []byte("El Psy Kongroo")
|
|
|
|
ctxb := context.Background()
|
|
|
|
var testCases = []chanRestoreTestCase{
|
|
// Restore from backups obtained via the RPC interface. Dave
|
|
// was the initiator, of the non-advertised channel.
|
|
{
|
|
name: "restore from RPC backup",
|
|
channelsUpdated: false,
|
|
initiator: true,
|
|
private: false,
|
|
restoreMethod: func(oldNode *lntest.HarnessNode,
|
|
backupFilePath string,
|
|
mnemonic []string) (nodeRestorer, error) {
|
|
|
|
// For this restoration method, we'll grab the
|
|
// current multi-channel backup from the old
|
|
// node, and use it to restore a new node
|
|
// within the closure.
|
|
req := &lnrpc.ChanBackupExportRequest{}
|
|
chanBackup, err := oldNode.ExportAllChannelBackups(
|
|
ctxb, req,
|
|
)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to obtain "+
|
|
"channel backup: %v", err)
|
|
}
|
|
|
|
multi := chanBackup.MultiChanBackup.MultiChanBackup
|
|
|
|
// In our nodeRestorer function, we'll restore
|
|
// the node from seed, then manually recover
|
|
// the channel backup.
|
|
return chanRestoreViaRPC(
|
|
net, password, mnemonic, multi,
|
|
)
|
|
},
|
|
},
|
|
|
|
// Restore the backup from the on-disk file, using the RPC
|
|
// interface.
|
|
{
|
|
name: "restore from backup file",
|
|
initiator: true,
|
|
private: false,
|
|
restoreMethod: func(oldNode *lntest.HarnessNode,
|
|
backupFilePath string,
|
|
mnemonic []string) (nodeRestorer, error) {
|
|
|
|
// Read the entire Multi backup stored within
|
|
// this node's channels.backup file.
|
|
multi, err := ioutil.ReadFile(backupFilePath)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Now that we have Dave's backup file, we'll
|
|
// create a new nodeRestorer that will restore
|
|
// using the on-disk channels.backup.
|
|
return chanRestoreViaRPC(
|
|
net, password, mnemonic, multi,
|
|
)
|
|
},
|
|
},
|
|
|
|
// Restore the backup as part of node initialization with the
|
|
// prior mnemonic and new backup seed.
|
|
{
|
|
name: "restore during creation",
|
|
initiator: true,
|
|
private: false,
|
|
restoreMethod: func(oldNode *lntest.HarnessNode,
|
|
backupFilePath string,
|
|
mnemonic []string) (nodeRestorer, error) {
|
|
|
|
// First, fetch the current backup state as is,
|
|
// to obtain our latest Multi.
|
|
chanBackup, err := oldNode.ExportAllChannelBackups(
|
|
ctxb, &lnrpc.ChanBackupExportRequest{},
|
|
)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to obtain "+
|
|
"channel backup: %v", err)
|
|
}
|
|
backupSnapshot := &lnrpc.ChanBackupSnapshot{
|
|
MultiChanBackup: chanBackup.MultiChanBackup,
|
|
}
|
|
|
|
// Create a new nodeRestorer that will restore
|
|
// the node using the Multi backup we just
|
|
// obtained above.
|
|
return func() (*lntest.HarnessNode, error) {
|
|
return net.RestoreNodeWithSeed(
|
|
"dave", nil, password,
|
|
mnemonic, 1000, backupSnapshot,
|
|
)
|
|
}, nil
|
|
},
|
|
},
|
|
|
|
// Restore the backup once the node has already been
|
|
// re-created, using the Unlock call.
|
|
{
|
|
name: "restore during unlock",
|
|
initiator: true,
|
|
private: false,
|
|
restoreMethod: func(oldNode *lntest.HarnessNode,
|
|
backupFilePath string,
|
|
mnemonic []string) (nodeRestorer, error) {
|
|
|
|
// First, fetch the current backup state as is,
|
|
// to obtain our latest Multi.
|
|
chanBackup, err := oldNode.ExportAllChannelBackups(
|
|
ctxb, &lnrpc.ChanBackupExportRequest{},
|
|
)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to obtain "+
|
|
"channel backup: %v", err)
|
|
}
|
|
backupSnapshot := &lnrpc.ChanBackupSnapshot{
|
|
MultiChanBackup: chanBackup.MultiChanBackup,
|
|
}
|
|
|
|
// Create a new nodeRestorer that will restore
|
|
// the node with its seed, but no channel
|
|
// backup, shutdown this initialized node, then
|
|
// restart it again using Unlock.
|
|
return func() (*lntest.HarnessNode, error) {
|
|
newNode, err := net.RestoreNodeWithSeed(
|
|
"dave", nil, password,
|
|
mnemonic, 1000, nil,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
err = net.RestartNode(
|
|
newNode, nil, backupSnapshot,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
return newNode, nil
|
|
}, nil
|
|
},
|
|
},
|
|
|
|
// Restore the backup from the on-disk file a second time to
|
|
// make sure imports can be canceled and later resumed.
|
|
{
|
|
name: "restore from backup file twice",
|
|
initiator: true,
|
|
private: false,
|
|
restoreMethod: func(oldNode *lntest.HarnessNode,
|
|
backupFilePath string,
|
|
mnemonic []string) (nodeRestorer, error) {
|
|
|
|
// Read the entire Multi backup stored within
|
|
// this node's channels.backup file.
|
|
multi, err := ioutil.ReadFile(backupFilePath)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Now that we have Dave's backup file, we'll
|
|
// create a new nodeRestorer that will restore
|
|
// using the on-disk channels.backup.
|
|
backup := &lnrpc.RestoreChanBackupRequest_MultiChanBackup{
|
|
MultiChanBackup: multi,
|
|
}
|
|
|
|
ctxb := context.Background()
|
|
|
|
return func() (*lntest.HarnessNode, error) {
|
|
newNode, err := net.RestoreNodeWithSeed(
|
|
"dave", nil, password, mnemonic,
|
|
1000, nil,
|
|
)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to "+
|
|
"restore node: %v", err)
|
|
}
|
|
|
|
_, err = newNode.RestoreChannelBackups(
|
|
ctxb,
|
|
&lnrpc.RestoreChanBackupRequest{
|
|
Backup: backup,
|
|
},
|
|
)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable "+
|
|
"to restore backups: %v",
|
|
err)
|
|
}
|
|
|
|
_, err = newNode.RestoreChannelBackups(
|
|
ctxb,
|
|
&lnrpc.RestoreChanBackupRequest{
|
|
Backup: backup,
|
|
},
|
|
)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable "+
|
|
"to restore backups the"+
|
|
"second time: %v",
|
|
err)
|
|
}
|
|
|
|
return newNode, nil
|
|
}, nil
|
|
},
|
|
},
|
|
|
|
// Use the channel backup file that contains an unconfirmed
|
|
// channel and make sure recovery works as well.
|
|
{
|
|
name: "restore unconfirmed channel file",
|
|
channelsUpdated: false,
|
|
initiator: true,
|
|
private: false,
|
|
unconfirmed: true,
|
|
restoreMethod: func(oldNode *lntest.HarnessNode,
|
|
backupFilePath string,
|
|
mnemonic []string) (nodeRestorer, error) {
|
|
|
|
// Read the entire Multi backup stored within
|
|
// this node's channels.backup file.
|
|
multi, err := ioutil.ReadFile(backupFilePath)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Let's assume time passes, the channel
|
|
// confirms in the meantime but for some reason
|
|
// the backup we made while it was still
|
|
// unconfirmed is the only backup we have. We
|
|
// should still be able to restore it. To
|
|
// simulate time passing, we mine some blocks
|
|
// to get the channel confirmed _after_ we saved
|
|
// the backup.
|
|
mineBlocks(t, net, 6, 1)
|
|
|
|
// In our nodeRestorer function, we'll restore
|
|
// the node from seed, then manually recover
|
|
// the channel backup.
|
|
return chanRestoreViaRPC(
|
|
net, password, mnemonic, multi,
|
|
)
|
|
},
|
|
},
|
|
|
|
// Create a backup using RPC that contains an unconfirmed
|
|
// channel and make sure recovery works as well.
|
|
{
|
|
name: "restore unconfirmed channel RPC",
|
|
channelsUpdated: false,
|
|
initiator: true,
|
|
private: false,
|
|
unconfirmed: true,
|
|
restoreMethod: func(oldNode *lntest.HarnessNode,
|
|
backupFilePath string,
|
|
mnemonic []string) (nodeRestorer, error) {
|
|
|
|
// For this restoration method, we'll grab the
|
|
// current multi-channel backup from the old
|
|
// node. The channel should be included, even if
|
|
// it is not confirmed yet.
|
|
req := &lnrpc.ChanBackupExportRequest{}
|
|
chanBackup, err := oldNode.ExportAllChannelBackups(
|
|
ctxb, req,
|
|
)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to obtain "+
|
|
"channel backup: %v", err)
|
|
}
|
|
chanPoints := chanBackup.MultiChanBackup.ChanPoints
|
|
if len(chanPoints) == 0 {
|
|
return nil, fmt.Errorf("unconfirmed " +
|
|
"channel not included in backup")
|
|
}
|
|
|
|
// Let's assume time passes, the channel
|
|
// confirms in the meantime but for some reason
|
|
// the backup we made while it was still
|
|
// unconfirmed is the only backup we have. We
|
|
// should still be able to restore it. To
|
|
// simulate time passing, we mine some blocks
|
|
// to get the channel confirmed _after_ we saved
|
|
// the backup.
|
|
mineBlocks(t, net, 6, 1)
|
|
|
|
// In our nodeRestorer function, we'll restore
|
|
// the node from seed, then manually recover
|
|
// the channel backup.
|
|
multi := chanBackup.MultiChanBackup.MultiChanBackup
|
|
return chanRestoreViaRPC(
|
|
net, password, mnemonic, multi,
|
|
)
|
|
},
|
|
},
|
|
|
|
// Restore the backup from the on-disk file, using the RPC
|
|
// interface, for anchor commitment channels.
|
|
{
|
|
name: "restore from backup file anchors",
|
|
initiator: true,
|
|
private: false,
|
|
anchorCommit: true,
|
|
restoreMethod: func(oldNode *lntest.HarnessNode,
|
|
backupFilePath string,
|
|
mnemonic []string) (nodeRestorer, error) {
|
|
|
|
// Read the entire Multi backup stored within
|
|
// this node's channels.backup file.
|
|
multi, err := ioutil.ReadFile(backupFilePath)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Now that we have Dave's backup file, we'll
|
|
// create a new nodeRestorer that will restore
|
|
// using the on-disk channels.backup.
|
|
return chanRestoreViaRPC(
|
|
net, password, mnemonic, multi,
|
|
)
|
|
},
|
|
},
|
|
}
|
|
|
|
// TODO(roasbeef): online vs offline close?
|
|
|
|
// TODO(roasbeef): need to re-trigger the on-disk file once the node
|
|
// ann is updated?
|
|
|
|
for _, testCase := range testCases {
|
|
success := t.t.Run(testCase.name, func(t *testing.T) {
|
|
h := newHarnessTest(t, net)
|
|
testChanRestoreScenario(h, net, &testCase, password)
|
|
})
|
|
if !success {
|
|
break
|
|
}
|
|
}
|
|
}
|
|
|
|
// testHoldInvoicePersistence tests that a sender to a hold-invoice, can be
|
|
// restarted before the payment gets settled, and still be able to receive the
|
|
// preimage.
|
|
func testHoldInvoicePersistence(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
const (
|
|
chanAmt = btcutil.Amount(1000000)
|
|
numPayments = 10
|
|
)
|
|
|
|
// Create carol, and clean up when the test finishes.
|
|
carol, err := net.NewNode("Carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to create new nodes: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
// Connect Alice to Carol.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
if err := net.ConnectNodes(ctxb, net.Alice, carol); err != nil {
|
|
t.Fatalf("unable to connect alice to carol: %v", err)
|
|
}
|
|
|
|
// Open a channel between Alice and Carol.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelOpenTimeout)
|
|
chanPointAlice := openChannelAndAssert(
|
|
ctxt, t, net, net.Alice, carol,
|
|
lntest.OpenChannelParams{
|
|
Amt: chanAmt,
|
|
},
|
|
)
|
|
|
|
// Wait for Alice and Carol to receive the channel edge from the
|
|
// funding manager.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.Alice.WaitForNetworkChannelOpen(ctxt, chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't see the alice->carol channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = carol.WaitForNetworkChannelOpen(ctxt, chanPointAlice)
|
|
if err != nil {
|
|
t.Fatalf("alice didn't see the alice->carol channel before "+
|
|
"timeout: %v", err)
|
|
}
|
|
|
|
// Create preimages for all payments we are going to initiate.
|
|
var preimages []lntypes.Preimage
|
|
for i := 0; i < numPayments; i++ {
|
|
var preimage lntypes.Preimage
|
|
_, err = rand.Read(preimage[:])
|
|
if err != nil {
|
|
t.Fatalf("unable to generate preimage: %v", err)
|
|
}
|
|
|
|
preimages = append(preimages, preimage)
|
|
}
|
|
|
|
// Let Carol create hold-invoices for all the payments.
|
|
var (
|
|
payAmt = btcutil.Amount(4)
|
|
payReqs []string
|
|
invoiceStreams []invoicesrpc.Invoices_SubscribeSingleInvoiceClient
|
|
)
|
|
|
|
for _, preimage := range preimages {
|
|
payHash := preimage.Hash()
|
|
invoiceReq := &invoicesrpc.AddHoldInvoiceRequest{
|
|
Memo: "testing",
|
|
Value: int64(payAmt),
|
|
Hash: payHash[:],
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := carol.AddHoldInvoice(ctxt, invoiceReq)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
|
|
ctx, cancel := context.WithCancel(ctxb)
|
|
defer cancel()
|
|
|
|
stream, err := carol.SubscribeSingleInvoice(
|
|
ctx,
|
|
&invoicesrpc.SubscribeSingleInvoiceRequest{
|
|
RHash: payHash[:],
|
|
},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to subscribe to invoice: %v", err)
|
|
}
|
|
|
|
invoiceStreams = append(invoiceStreams, stream)
|
|
payReqs = append(payReqs, resp.PaymentRequest)
|
|
}
|
|
|
|
// Wait for all the invoices to reach the OPEN state.
|
|
for _, stream := range invoiceStreams {
|
|
invoice, err := stream.Recv()
|
|
if err != nil {
|
|
t.Fatalf("err: %v", err)
|
|
}
|
|
|
|
if invoice.State != lnrpc.Invoice_OPEN {
|
|
t.Fatalf("expected OPEN, got state: %v", invoice.State)
|
|
}
|
|
}
|
|
|
|
// Let Alice initiate payments for all the created invoices.
|
|
var paymentStreams []routerrpc.Router_SendPaymentClient
|
|
for _, payReq := range payReqs {
|
|
ctx, cancel := context.WithCancel(ctxb)
|
|
defer cancel()
|
|
|
|
payStream, err := net.Alice.RouterClient.SendPayment(
|
|
ctx, &routerrpc.SendPaymentRequest{
|
|
PaymentRequest: payReq,
|
|
TimeoutSeconds: 60,
|
|
FeeLimitSat: 1000000,
|
|
},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to send alice htlc: %v", err)
|
|
}
|
|
|
|
paymentStreams = append(paymentStreams, payStream)
|
|
}
|
|
|
|
// Wait for inlight status update.
|
|
for _, payStream := range paymentStreams {
|
|
payment, err := payStream.Recv()
|
|
if err != nil {
|
|
t.Fatalf("Failed receiving status update: %v", err)
|
|
}
|
|
|
|
if payment.Status != lnrpc.Payment_IN_FLIGHT {
|
|
t.Fatalf("state not in flight: %v", payment.Status)
|
|
}
|
|
}
|
|
|
|
// The payments should now show up in Alice's ListInvoices, with a zero
|
|
// preimage, indicating they are not yet settled.
|
|
err = wait.NoError(func() error {
|
|
req := &lnrpc.ListPaymentsRequest{
|
|
IncludeIncomplete: true,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxt, defaultTimeout)
|
|
paymentsResp, err := net.Alice.ListPayments(ctxt, req)
|
|
if err != nil {
|
|
return fmt.Errorf("error when obtaining payments: %v",
|
|
err)
|
|
}
|
|
|
|
// Gather the payment hashes we are looking for in the
|
|
// response.
|
|
payHashes := make(map[string]struct{})
|
|
for _, preimg := range preimages {
|
|
payHashes[preimg.Hash().String()] = struct{}{}
|
|
}
|
|
|
|
var zeroPreimg lntypes.Preimage
|
|
for _, payment := range paymentsResp.Payments {
|
|
_, ok := payHashes[payment.PaymentHash]
|
|
if !ok {
|
|
continue
|
|
}
|
|
|
|
// The preimage should NEVER be non-zero at this point.
|
|
if payment.PaymentPreimage != zeroPreimg.String() {
|
|
t.Fatalf("expected zero preimage, got %v",
|
|
payment.PaymentPreimage)
|
|
}
|
|
|
|
// We wait for the payment attempt to have been
|
|
// properly recorded in the DB.
|
|
if len(payment.Path) == 0 {
|
|
return fmt.Errorf("path is empty")
|
|
}
|
|
|
|
delete(payHashes, payment.PaymentHash)
|
|
}
|
|
|
|
if len(payHashes) != 0 {
|
|
return fmt.Errorf("payhash not found in response")
|
|
}
|
|
|
|
return nil
|
|
}, time.Second*15)
|
|
if err != nil {
|
|
t.Fatalf("predicate not satisfied: %v", err)
|
|
}
|
|
|
|
// Wait for all invoices to be accepted.
|
|
for _, stream := range invoiceStreams {
|
|
invoice, err := stream.Recv()
|
|
if err != nil {
|
|
t.Fatalf("err: %v", err)
|
|
}
|
|
|
|
if invoice.State != lnrpc.Invoice_ACCEPTED {
|
|
t.Fatalf("expected ACCEPTED, got state: %v",
|
|
invoice.State)
|
|
}
|
|
}
|
|
|
|
// Restart alice. This to ensure she will still be able to handle
|
|
// settling the invoices after a restart.
|
|
if err := net.RestartNode(net.Alice, nil); err != nil {
|
|
t.Fatalf("Node restart failed: %v", err)
|
|
}
|
|
|
|
// Now after a restart, we must re-track the payments. We set up a
|
|
// goroutine for each to track thir status updates.
|
|
var (
|
|
statusUpdates []chan *lnrpc.Payment
|
|
wg sync.WaitGroup
|
|
quit = make(chan struct{})
|
|
)
|
|
|
|
defer close(quit)
|
|
for _, preimg := range preimages {
|
|
hash := preimg.Hash()
|
|
|
|
ctx, cancel := context.WithCancel(ctxb)
|
|
defer cancel()
|
|
|
|
payStream, err := net.Alice.RouterClient.TrackPayment(
|
|
ctx, &routerrpc.TrackPaymentRequest{
|
|
PaymentHash: hash[:],
|
|
},
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to send track payment: %v", err)
|
|
}
|
|
|
|
// We set up a channel where we'll forward any status update.
|
|
upd := make(chan *lnrpc.Payment)
|
|
wg.Add(1)
|
|
go func() {
|
|
defer wg.Done()
|
|
|
|
for {
|
|
payment, err := payStream.Recv()
|
|
if err != nil {
|
|
close(upd)
|
|
return
|
|
}
|
|
|
|
select {
|
|
case upd <- payment:
|
|
case <-quit:
|
|
return
|
|
}
|
|
}
|
|
}()
|
|
|
|
statusUpdates = append(statusUpdates, upd)
|
|
}
|
|
|
|
// Wait for the in-flight status update.
|
|
for _, upd := range statusUpdates {
|
|
select {
|
|
case payment, ok := <-upd:
|
|
if !ok {
|
|
t.Fatalf("failed getting payment update")
|
|
}
|
|
|
|
if payment.Status != lnrpc.Payment_IN_FLIGHT {
|
|
t.Fatalf("state not in in flight: %v",
|
|
payment.Status)
|
|
}
|
|
case <-time.After(5 * time.Second):
|
|
t.Fatalf("in flight status not recevied")
|
|
}
|
|
}
|
|
|
|
// Settle invoices half the invoices, cancel the rest.
|
|
for i, preimage := range preimages {
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
if i%2 == 0 {
|
|
settle := &invoicesrpc.SettleInvoiceMsg{
|
|
Preimage: preimage[:],
|
|
}
|
|
_, err = carol.SettleInvoice(ctxt, settle)
|
|
} else {
|
|
hash := preimage.Hash()
|
|
settle := &invoicesrpc.CancelInvoiceMsg{
|
|
PaymentHash: hash[:],
|
|
}
|
|
_, err = carol.CancelInvoice(ctxt, settle)
|
|
}
|
|
if err != nil {
|
|
t.Fatalf("unable to cancel/settle invoice: %v", err)
|
|
}
|
|
}
|
|
|
|
// Make sure we get the expected status update.
|
|
for i, upd := range statusUpdates {
|
|
// Read until the payment is in a terminal state.
|
|
var payment *lnrpc.Payment
|
|
for payment == nil {
|
|
select {
|
|
case p, ok := <-upd:
|
|
if !ok {
|
|
t.Fatalf("failed getting payment update")
|
|
}
|
|
|
|
if p.Status == lnrpc.Payment_IN_FLIGHT {
|
|
continue
|
|
}
|
|
|
|
payment = p
|
|
case <-time.After(5 * time.Second):
|
|
t.Fatalf("in flight status not recevied")
|
|
}
|
|
}
|
|
|
|
// Assert terminal payment state.
|
|
if i%2 == 0 {
|
|
if payment.Status != lnrpc.Payment_SUCCEEDED {
|
|
t.Fatalf("state not suceeded : %v",
|
|
payment.Status)
|
|
}
|
|
} else {
|
|
if payment.FailureReason !=
|
|
lnrpc.PaymentFailureReason_FAILURE_REASON_INCORRECT_PAYMENT_DETAILS {
|
|
|
|
t.Fatalf("state not failed: %v",
|
|
payment.FailureReason)
|
|
}
|
|
}
|
|
}
|
|
|
|
// Check that Alice's invoices to be shown as settled and failed
|
|
// accordingly, and preimages matching up.
|
|
req := &lnrpc.ListPaymentsRequest{
|
|
IncludeIncomplete: true,
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxt, defaultTimeout)
|
|
paymentsResp, err := net.Alice.ListPayments(ctxt, req)
|
|
if err != nil {
|
|
t.Fatalf("error when obtaining Alice payments: %v", err)
|
|
}
|
|
for i, preimage := range preimages {
|
|
paymentHash := preimage.Hash()
|
|
var p string
|
|
for _, resp := range paymentsResp.Payments {
|
|
if resp.PaymentHash == paymentHash.String() {
|
|
p = resp.PaymentPreimage
|
|
break
|
|
}
|
|
}
|
|
if p == "" {
|
|
t.Fatalf("payment not found")
|
|
}
|
|
|
|
if i%2 == 0 {
|
|
if p != preimage.String() {
|
|
t.Fatalf("preimage doesn't match: %v vs %v",
|
|
p, preimage.String())
|
|
}
|
|
} else {
|
|
if p != lntypes.ZeroHash.String() {
|
|
t.Fatalf("preimage not zero: %v", p)
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
// testExternalFundingChanPoint tests that we're able to carry out a normal
|
|
// channel funding workflow given a channel point that was constructed outside
|
|
// the main daemon.
|
|
func testExternalFundingChanPoint(net *lntest.NetworkHarness, t *harnessTest) {
|
|
ctxb := context.Background()
|
|
|
|
// First, we'll create two new nodes that we'll use to open channel
|
|
// between for this test.
|
|
carol, err := net.NewNode("carol", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to start new node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, carol)
|
|
|
|
dave, err := net.NewNode("dave", nil)
|
|
if err != nil {
|
|
t.Fatalf("unable to start new node: %v", err)
|
|
}
|
|
defer shutdownAndAssert(net, t, dave)
|
|
|
|
// Carol will be funding the channel, so we'll send some coins over to
|
|
// her and ensure they have enough confirmations before we proceed.
|
|
ctxt, _ := context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.SendCoins(ctxt, btcutil.SatoshiPerBitcoin, carol)
|
|
if err != nil {
|
|
t.Fatalf("unable to send coins to carol: %v", err)
|
|
}
|
|
|
|
// Before we start the test, we'll ensure both sides are connected to
|
|
// the funding flow can properly be executed.
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = net.EnsureConnected(ctxt, carol, dave)
|
|
if err != nil {
|
|
t.Fatalf("unable to connect peers: %v", err)
|
|
}
|
|
|
|
// At this point, we're ready to simulate our external channle funding
|
|
// flow. To start with, we'll get to new keys from both sides which
|
|
// will be used to create the multi-sig output for the external funding
|
|
// transaction.
|
|
keyLoc := &signrpc.KeyLocator{
|
|
KeyFamily: 9999,
|
|
KeyIndex: 1,
|
|
}
|
|
carolFundingKey, err := carol.WalletKitClient.DeriveKey(ctxb, keyLoc)
|
|
if err != nil {
|
|
t.Fatalf("unable to get carol funding key: %v", err)
|
|
}
|
|
daveFundingKey, err := dave.WalletKitClient.DeriveKey(ctxb, keyLoc)
|
|
if err != nil {
|
|
t.Fatalf("unable to get dave funding key: %v", err)
|
|
}
|
|
|
|
// Now that we have the multi-sig keys for each party, we can manually
|
|
// construct the funding transaction. We'll instruct the backend to
|
|
// immediately create and broadcast a transaction paying out an exact
|
|
// amount. Normally this would reside in the mempool, but we just
|
|
// confirm it now for simplicity.
|
|
const chanSize = lnd.MaxBtcFundingAmount
|
|
_, fundingOutput, err := input.GenFundingPkScript(
|
|
carolFundingKey.RawKeyBytes, daveFundingKey.RawKeyBytes,
|
|
int64(chanSize),
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create funding script: %v", err)
|
|
}
|
|
txid, err := net.Miner.SendOutputsWithoutChange(
|
|
[]*wire.TxOut{fundingOutput}, 5,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to create funding output: %v", err)
|
|
}
|
|
|
|
// At this point, we can being our external channel funding workflow.
|
|
// We'll start by generating a pending channel ID externally that will
|
|
// be used to track this new funding type.
|
|
var pendingChanID [32]byte
|
|
if _, err := rand.Read(pendingChanID[:]); err != nil {
|
|
t.Fatalf("unable to gen pending chan ID: %v", err)
|
|
}
|
|
|
|
_, currentHeight, err := net.Miner.Node.GetBestBlock()
|
|
if err != nil {
|
|
t.Fatalf("unable to get current blockheight %v", err)
|
|
}
|
|
|
|
// Now that we have the pending channel ID, Dave (our responder) will
|
|
// register the intent to receive a new channel funding workflow using
|
|
// the pending channel ID.
|
|
chanPoint := &lnrpc.ChannelPoint{
|
|
FundingTxid: &lnrpc.ChannelPoint_FundingTxidBytes{
|
|
FundingTxidBytes: txid[:],
|
|
},
|
|
}
|
|
thawHeight := uint32(currentHeight + 10)
|
|
chanPointShim := &lnrpc.ChanPointShim{
|
|
Amt: int64(chanSize),
|
|
ChanPoint: chanPoint,
|
|
LocalKey: &lnrpc.KeyDescriptor{
|
|
RawKeyBytes: daveFundingKey.RawKeyBytes,
|
|
KeyLoc: &lnrpc.KeyLocator{
|
|
KeyFamily: daveFundingKey.KeyLoc.KeyFamily,
|
|
KeyIndex: daveFundingKey.KeyLoc.KeyIndex,
|
|
},
|
|
},
|
|
RemoteKey: carolFundingKey.RawKeyBytes,
|
|
PendingChanId: pendingChanID[:],
|
|
ThawHeight: thawHeight,
|
|
}
|
|
fundingShim := &lnrpc.FundingShim{
|
|
Shim: &lnrpc.FundingShim_ChanPointShim{
|
|
ChanPointShim: chanPointShim,
|
|
},
|
|
}
|
|
_, err = dave.FundingStateStep(ctxb, &lnrpc.FundingTransitionMsg{
|
|
Trigger: &lnrpc.FundingTransitionMsg_ShimRegister{
|
|
ShimRegister: fundingShim,
|
|
},
|
|
})
|
|
if err != nil {
|
|
t.Fatalf("unable to walk funding state forward: %v", err)
|
|
}
|
|
|
|
// If we attempt to register the same shim (has the same pending chan
|
|
// ID), then we should get an error.
|
|
_, err = dave.FundingStateStep(ctxb, &lnrpc.FundingTransitionMsg{
|
|
Trigger: &lnrpc.FundingTransitionMsg_ShimRegister{
|
|
ShimRegister: fundingShim,
|
|
},
|
|
})
|
|
if err == nil {
|
|
t.Fatalf("duplicate pending channel ID funding shim " +
|
|
"registration should trigger an error")
|
|
}
|
|
|
|
// We'll take the chan point shim we just registered for Dave (the
|
|
// responder), and swap the local/remote keys before we feed it in as
|
|
// Carol's funding shim as the initiator.
|
|
fundingShim.GetChanPointShim().LocalKey = &lnrpc.KeyDescriptor{
|
|
RawKeyBytes: carolFundingKey.RawKeyBytes,
|
|
KeyLoc: &lnrpc.KeyLocator{
|
|
KeyFamily: carolFundingKey.KeyLoc.KeyFamily,
|
|
KeyIndex: carolFundingKey.KeyLoc.KeyIndex,
|
|
},
|
|
}
|
|
fundingShim.GetChanPointShim().RemoteKey = daveFundingKey.RawKeyBytes
|
|
|
|
// At this point, we'll now carry out the normal basic channel funding
|
|
// test as everything should now proceed as normal (a regular channel
|
|
// funding flow).
|
|
carolChan, daveChan, _, err := basicChannelFundingTest(
|
|
t, net, carol, dave, fundingShim,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to open channels: %v", err)
|
|
}
|
|
|
|
// Both channels should be marked as frozen with the proper thaw
|
|
// height.
|
|
if carolChan.ThawHeight != thawHeight {
|
|
t.Fatalf("expected thaw height of %v, got %v",
|
|
carolChan.ThawHeight, thawHeight)
|
|
}
|
|
if daveChan.ThawHeight != thawHeight {
|
|
t.Fatalf("expected thaw height of %v, got %v",
|
|
daveChan.ThawHeight, thawHeight)
|
|
}
|
|
|
|
// Next, to make sure the channel functions as normal, we'll make some
|
|
// payments within the channel.
|
|
payAmt := btcutil.Amount(100000)
|
|
invoice := &lnrpc.Invoice{
|
|
Memo: "new chans",
|
|
Value: int64(payAmt),
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
resp, err := dave.AddInvoice(ctxt, invoice)
|
|
if err != nil {
|
|
t.Fatalf("unable to add invoice: %v", err)
|
|
}
|
|
ctxt, _ = context.WithTimeout(ctxb, defaultTimeout)
|
|
err = completePaymentRequests(
|
|
ctxt, carol, []string{resp.PaymentRequest}, true,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to make payments between Carol and Dave")
|
|
}
|
|
|
|
// Now that the channels are open, and we've confirmed that they're
|
|
// operational, we'll now ensure that the channels are frozen as
|
|
// intended (if requested).
|
|
//
|
|
// First, we'll try to close the channel as Carol, the initiator. This
|
|
// should fail as a frozen channel only allows the responder to
|
|
// initiate a channel close.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
_, _, err = net.CloseChannel(ctxt, carol, chanPoint, false)
|
|
if err == nil {
|
|
t.Fatalf("carol wasn't denied a co-op close attempt for a " +
|
|
"frozen channel")
|
|
}
|
|
|
|
// Next we'll try but this time with Dave (the responder) as the
|
|
// initiator. This time the channel should be closed as normal.
|
|
ctxt, _ = context.WithTimeout(ctxb, channelCloseTimeout)
|
|
closeChannelAndAssert(ctxt, t, net, dave, chanPoint, false)
|
|
}
|
|
|
|
type testCase struct {
|
|
name string
|
|
test func(net *lntest.NetworkHarness, t *harnessTest)
|
|
}
|
|
|
|
var testsCases = []*testCase{
|
|
{
|
|
name: "sweep coins",
|
|
test: testSweepAllCoins,
|
|
},
|
|
{
|
|
name: "onchain fund recovery",
|
|
test: testOnchainFundRecovery,
|
|
},
|
|
{
|
|
name: "basic funding flow",
|
|
test: testBasicChannelFunding,
|
|
},
|
|
{
|
|
name: "unconfirmed channel funding",
|
|
test: testUnconfirmedChannelFunding,
|
|
},
|
|
{
|
|
name: "update channel policy",
|
|
test: testUpdateChannelPolicy,
|
|
},
|
|
{
|
|
name: "open channel reorg test",
|
|
test: testOpenChannelAfterReorg,
|
|
},
|
|
{
|
|
name: "disconnecting target peer",
|
|
test: testDisconnectingTargetPeer,
|
|
},
|
|
{
|
|
name: "graph topology notifications",
|
|
test: testGraphTopologyNotifications,
|
|
},
|
|
{
|
|
name: "funding flow persistence",
|
|
test: testChannelFundingPersistence,
|
|
},
|
|
{
|
|
name: "channel force closure",
|
|
test: testChannelForceClosure,
|
|
},
|
|
{
|
|
name: "channel balance",
|
|
test: testChannelBalance,
|
|
},
|
|
{
|
|
name: "channel unsettled balance",
|
|
test: testChannelUnsettledBalance,
|
|
},
|
|
{
|
|
name: "single hop invoice",
|
|
test: testSingleHopInvoice,
|
|
},
|
|
{
|
|
name: "sphinx replay persistence",
|
|
test: testSphinxReplayPersistence,
|
|
},
|
|
{
|
|
name: "list outgoing payments",
|
|
test: testListPayments,
|
|
},
|
|
{
|
|
name: "max pending channel",
|
|
test: testMaxPendingChannels,
|
|
},
|
|
{
|
|
name: "multi-hop payments",
|
|
test: testMultiHopPayments,
|
|
},
|
|
{
|
|
name: "single-hop send to route",
|
|
test: testSingleHopSendToRoute,
|
|
},
|
|
{
|
|
name: "multi-hop send to route",
|
|
test: testMultiHopSendToRoute,
|
|
},
|
|
{
|
|
name: "send to route error propagation",
|
|
test: testSendToRouteErrorPropagation,
|
|
},
|
|
{
|
|
name: "unannounced channels",
|
|
test: testUnannouncedChannels,
|
|
},
|
|
{
|
|
name: "private channels",
|
|
test: testPrivateChannels,
|
|
},
|
|
{
|
|
name: "invoice routing hints",
|
|
test: testInvoiceRoutingHints,
|
|
},
|
|
{
|
|
name: "multi-hop payments over private channels",
|
|
test: testMultiHopOverPrivateChannels,
|
|
},
|
|
{
|
|
name: "multiple channel creation and update subscription",
|
|
test: testBasicChannelCreationAndUpdates,
|
|
},
|
|
{
|
|
name: "invoice update subscription",
|
|
test: testInvoiceSubscriptions,
|
|
},
|
|
{
|
|
name: "multi-hop htlc error propagation",
|
|
test: testHtlcErrorPropagation,
|
|
},
|
|
{
|
|
name: "reject onward htlc",
|
|
test: testRejectHTLC,
|
|
},
|
|
// TODO(roasbeef): multi-path integration test
|
|
{
|
|
name: "node announcement",
|
|
test: testNodeAnnouncement,
|
|
},
|
|
{
|
|
name: "node sign verify",
|
|
test: testNodeSignVerify,
|
|
},
|
|
{
|
|
name: "async payments benchmark",
|
|
test: testAsyncPayments,
|
|
},
|
|
{
|
|
name: "async bidirectional payments",
|
|
test: testBidirectionalAsyncPayments,
|
|
},
|
|
{
|
|
name: "test multi-hop htlc",
|
|
test: testMultiHopHtlcClaims,
|
|
},
|
|
{
|
|
name: "switch circuit persistence",
|
|
test: testSwitchCircuitPersistence,
|
|
},
|
|
{
|
|
name: "switch offline delivery",
|
|
test: testSwitchOfflineDelivery,
|
|
},
|
|
{
|
|
name: "switch offline delivery persistence",
|
|
test: testSwitchOfflineDeliveryPersistence,
|
|
},
|
|
{
|
|
name: "switch offline delivery outgoing offline",
|
|
test: testSwitchOfflineDeliveryOutgoingOffline,
|
|
},
|
|
{
|
|
// TODO(roasbeef): test always needs to be last as Bob's state
|
|
// is borked since we trick him into attempting to cheat Alice?
|
|
name: "revoked uncooperative close retribution",
|
|
test: testRevokedCloseRetribution,
|
|
},
|
|
{
|
|
name: "failing link",
|
|
test: testFailingChannel,
|
|
},
|
|
{
|
|
name: "garbage collect link nodes",
|
|
test: testGarbageCollectLinkNodes,
|
|
},
|
|
{
|
|
name: "abandonchannel",
|
|
test: testAbandonChannel,
|
|
},
|
|
{
|
|
name: "revoked uncooperative close retribution zero value remote output",
|
|
test: testRevokedCloseRetributionZeroValueRemoteOutput,
|
|
},
|
|
{
|
|
name: "revoked uncooperative close retribution remote hodl",
|
|
test: testRevokedCloseRetributionRemoteHodl,
|
|
},
|
|
{
|
|
name: "revoked uncooperative close retribution altruist watchtower",
|
|
test: testRevokedCloseRetributionAltruistWatchtower,
|
|
},
|
|
{
|
|
name: "data loss protection",
|
|
test: testDataLossProtection,
|
|
},
|
|
{
|
|
name: "query routes",
|
|
test: testQueryRoutes,
|
|
},
|
|
{
|
|
name: "route fee cutoff",
|
|
test: testRouteFeeCutoff,
|
|
},
|
|
{
|
|
name: "send update disable channel",
|
|
test: testSendUpdateDisableChannel,
|
|
},
|
|
{
|
|
name: "streaming channel backup update",
|
|
test: testChannelBackupUpdates,
|
|
},
|
|
{
|
|
name: "export channel backup",
|
|
test: testExportChannelBackup,
|
|
},
|
|
{
|
|
name: "channel backup restore",
|
|
test: testChannelBackupRestore,
|
|
},
|
|
{
|
|
name: "hold invoice sender persistence",
|
|
test: testHoldInvoicePersistence,
|
|
},
|
|
{
|
|
name: "cpfp",
|
|
test: testCPFP,
|
|
},
|
|
{
|
|
name: "macaroon authentication",
|
|
test: testMacaroonAuthentication,
|
|
},
|
|
{
|
|
name: "immediate payment after channel opened",
|
|
test: testPaymentFollowingChannelOpen,
|
|
},
|
|
{
|
|
name: "external channel funding",
|
|
test: testExternalFundingChanPoint,
|
|
},
|
|
{
|
|
name: "psbt channel funding",
|
|
test: testPsbtChanFunding,
|
|
},
|
|
{
|
|
name: "sendtoroute multi path payment",
|
|
test: testSendToRouteMultiPath,
|
|
},
|
|
}
|
|
|
|
// TestLightningNetworkDaemon performs a series of integration tests amongst a
|
|
// programmatically driven network of lnd nodes.
|
|
func TestLightningNetworkDaemon(t *testing.T) {
|
|
ht := newHarnessTest(t, nil)
|
|
|
|
// Declare the network harness here to gain access to its
|
|
// 'OnTxAccepted' call back.
|
|
var lndHarness *lntest.NetworkHarness
|
|
|
|
// Create an instance of the btcd's rpctest.Harness that will act as
|
|
// the miner for all tests. This will be used to fund the wallets of
|
|
// the nodes within the test network and to drive blockchain related
|
|
// events within the network. Revert the default setting of accepting
|
|
// non-standard transactions on simnet to reject them. Transactions on
|
|
// the lightning network should always be standard to get better
|
|
// guarantees of getting included in to blocks.
|
|
//
|
|
// We will also connect it to our chain backend.
|
|
minerLogDir := "./.minerlogs"
|
|
args := []string{
|
|
"--rejectnonstd",
|
|
"--txindex",
|
|
"--debuglevel=debug",
|
|
"--logdir=" + minerLogDir,
|
|
"--trickleinterval=100ms",
|
|
}
|
|
handlers := &rpcclient.NotificationHandlers{
|
|
OnTxAccepted: func(hash *chainhash.Hash, amt btcutil.Amount) {
|
|
lndHarness.OnTxAccepted(hash)
|
|
},
|
|
}
|
|
|
|
miner, err := rpctest.New(harnessNetParams, handlers, args)
|
|
if err != nil {
|
|
ht.Fatalf("unable to create mining node: %v", err)
|
|
}
|
|
defer func() {
|
|
miner.TearDown()
|
|
|
|
// After shutting down the miner, we'll make a copy of the log
|
|
// file before deleting the temporary log dir.
|
|
logFile := fmt.Sprintf(
|
|
"%s/%s/btcd.log", minerLogDir, harnessNetParams.Name,
|
|
)
|
|
err := lntest.CopyFile("./output_btcd_miner.log", logFile)
|
|
if err != nil {
|
|
fmt.Printf("unable to copy file: %v\n", err)
|
|
}
|
|
if err = os.RemoveAll(minerLogDir); err != nil {
|
|
fmt.Printf("Cannot remove dir %s: %v\n",
|
|
minerLogDir, err)
|
|
}
|
|
}()
|
|
|
|
// Start a chain backend.
|
|
chainBackend, cleanUp, err := lntest.NewBackend(
|
|
miner.P2PAddress(), harnessNetParams,
|
|
)
|
|
if err != nil {
|
|
ht.Fatalf("unable to start backend: %v", err)
|
|
}
|
|
defer cleanUp()
|
|
|
|
if err := miner.SetUp(true, 50); err != nil {
|
|
ht.Fatalf("unable to set up mining node: %v", err)
|
|
}
|
|
if err := miner.Node.NotifyNewTransactions(false); err != nil {
|
|
ht.Fatalf("unable to request transaction notifications: %v", err)
|
|
}
|
|
|
|
// Now we can set up our test harness (LND instance), with the chain
|
|
// backend we just created.
|
|
lndHarness, err = lntest.NewNetworkHarness(
|
|
miner, chainBackend, itestLndBinary,
|
|
)
|
|
if err != nil {
|
|
ht.Fatalf("unable to create lightning network harness: %v", err)
|
|
}
|
|
defer lndHarness.TearDownAll()
|
|
|
|
// Spawn a new goroutine to watch for any fatal errors that any of the
|
|
// running lnd processes encounter. If an error occurs, then the test
|
|
// case should naturally as a result and we log the server error here to
|
|
// help debug.
|
|
go func() {
|
|
for {
|
|
select {
|
|
case err, more := <-lndHarness.ProcessErrors():
|
|
if !more {
|
|
return
|
|
}
|
|
ht.Logf("lnd finished with error (stderr):\n%v", err)
|
|
}
|
|
}
|
|
}()
|
|
|
|
// Next mine enough blocks in order for segwit and the CSV package
|
|
// soft-fork to activate on SimNet.
|
|
numBlocks := harnessNetParams.MinerConfirmationWindow * 2
|
|
if _, err := miner.Node.Generate(numBlocks); err != nil {
|
|
ht.Fatalf("unable to generate blocks: %v", err)
|
|
}
|
|
|
|
// With the btcd harness created, we can now complete the
|
|
// initialization of the network. args - list of lnd arguments,
|
|
// example: "--debuglevel=debug"
|
|
// TODO(roasbeef): create master balanced channel with all the monies?
|
|
if err = lndHarness.SetUp(nil); err != nil {
|
|
ht.Fatalf("unable to set up test lightning network: %v", err)
|
|
}
|
|
|
|
t.Logf("Running %v integration tests", len(testsCases))
|
|
for _, testCase := range testsCases {
|
|
logLine := fmt.Sprintf("STARTING ============ %v ============\n",
|
|
testCase.name)
|
|
|
|
err := lndHarness.EnsureConnected(
|
|
context.Background(), lndHarness.Alice, lndHarness.Bob,
|
|
)
|
|
if err != nil {
|
|
t.Fatalf("unable to connect alice to bob: %v", err)
|
|
}
|
|
|
|
if err := lndHarness.Alice.AddToLog(logLine); err != nil {
|
|
t.Fatalf("unable to add to log: %v", err)
|
|
}
|
|
if err := lndHarness.Bob.AddToLog(logLine); err != nil {
|
|
t.Fatalf("unable to add to log: %v", err)
|
|
}
|
|
|
|
success := t.Run(testCase.name, func(t1 *testing.T) {
|
|
ht := newHarnessTest(t1, lndHarness)
|
|
ht.RunTestCase(testCase)
|
|
})
|
|
|
|
// Stop at the first failure. Mimic behavior of original test
|
|
// framework.
|
|
if !success {
|
|
break
|
|
}
|
|
}
|
|
}
|