47235e2f98
In this commit, we add `msats` to the return value of `DecodePayReq` to ensure we always show full value information as we're moving to do generally for all RPC calls that deal with off-chain amounts.
5499 lines
167 KiB
Go
5499 lines
167 KiB
Go
package lnd
|
|
|
|
import (
|
|
"bytes"
|
|
"context"
|
|
"crypto/tls"
|
|
"encoding/hex"
|
|
"errors"
|
|
"fmt"
|
|
"io"
|
|
"math"
|
|
"net"
|
|
"net/http"
|
|
"sort"
|
|
"strings"
|
|
"sync"
|
|
"sync/atomic"
|
|
"time"
|
|
|
|
"github.com/btcsuite/btcd/blockchain"
|
|
"github.com/btcsuite/btcd/btcec"
|
|
"github.com/btcsuite/btcd/chaincfg/chainhash"
|
|
"github.com/btcsuite/btcd/txscript"
|
|
"github.com/btcsuite/btcd/wire"
|
|
"github.com/btcsuite/btcutil"
|
|
"github.com/btcsuite/btcwallet/wallet/txauthor"
|
|
"github.com/coreos/bbolt"
|
|
"github.com/davecgh/go-spew/spew"
|
|
grpc_middleware "github.com/grpc-ecosystem/go-grpc-middleware"
|
|
proxy "github.com/grpc-ecosystem/grpc-gateway/runtime"
|
|
"github.com/lightningnetwork/lnd/autopilot"
|
|
"github.com/lightningnetwork/lnd/build"
|
|
"github.com/lightningnetwork/lnd/chanacceptor"
|
|
"github.com/lightningnetwork/lnd/chanbackup"
|
|
"github.com/lightningnetwork/lnd/channeldb"
|
|
"github.com/lightningnetwork/lnd/channelnotifier"
|
|
"github.com/lightningnetwork/lnd/contractcourt"
|
|
"github.com/lightningnetwork/lnd/discovery"
|
|
"github.com/lightningnetwork/lnd/htlcswitch"
|
|
"github.com/lightningnetwork/lnd/input"
|
|
"github.com/lightningnetwork/lnd/invoices"
|
|
"github.com/lightningnetwork/lnd/lncfg"
|
|
"github.com/lightningnetwork/lnd/lnrpc"
|
|
"github.com/lightningnetwork/lnd/lnrpc/invoicesrpc"
|
|
"github.com/lightningnetwork/lnd/lnrpc/routerrpc"
|
|
"github.com/lightningnetwork/lnd/lntypes"
|
|
"github.com/lightningnetwork/lnd/lnwallet"
|
|
"github.com/lightningnetwork/lnd/lnwallet/chainfee"
|
|
"github.com/lightningnetwork/lnd/lnwire"
|
|
"github.com/lightningnetwork/lnd/macaroons"
|
|
"github.com/lightningnetwork/lnd/monitoring"
|
|
"github.com/lightningnetwork/lnd/routing"
|
|
"github.com/lightningnetwork/lnd/routing/route"
|
|
"github.com/lightningnetwork/lnd/signal"
|
|
"github.com/lightningnetwork/lnd/sweep"
|
|
"github.com/lightningnetwork/lnd/tlv"
|
|
"github.com/lightningnetwork/lnd/watchtower"
|
|
"github.com/lightningnetwork/lnd/zpay32"
|
|
"github.com/tv42/zbase32"
|
|
"google.golang.org/grpc"
|
|
"gopkg.in/macaroon-bakery.v2/bakery"
|
|
)
|
|
|
|
const (
|
|
// maxBtcPaymentMSat is the maximum allowed Bitcoin payment currently
|
|
// permitted as defined in BOLT-0002.
|
|
maxBtcPaymentMSat = lnwire.MilliSatoshi(math.MaxUint32)
|
|
|
|
// maxLtcPaymentMSat is the maximum allowed Litecoin payment currently
|
|
// permitted.
|
|
maxLtcPaymentMSat = lnwire.MilliSatoshi(math.MaxUint32) *
|
|
btcToLtcConversionRate
|
|
)
|
|
|
|
var (
|
|
// MaxPaymentMSat is the maximum allowed payment currently permitted as
|
|
// defined in BOLT-002. This value depends on which chain is active.
|
|
// It is set to the value under the Bitcoin chain as default.
|
|
MaxPaymentMSat = maxBtcPaymentMSat
|
|
|
|
// defaultAcceptorTimeout is the time after which an RPCAcceptor will time
|
|
// out and return false if it hasn't yet received a response.
|
|
//
|
|
// TODO: Make this configurable
|
|
defaultAcceptorTimeout = 15 * time.Second
|
|
|
|
// readPermissions is a slice of all entities that allow read
|
|
// permissions for authorization purposes, all lowercase.
|
|
readPermissions = []bakery.Op{
|
|
{
|
|
Entity: "onchain",
|
|
Action: "read",
|
|
},
|
|
{
|
|
Entity: "offchain",
|
|
Action: "read",
|
|
},
|
|
{
|
|
Entity: "address",
|
|
Action: "read",
|
|
},
|
|
{
|
|
Entity: "message",
|
|
Action: "read",
|
|
},
|
|
{
|
|
Entity: "peers",
|
|
Action: "read",
|
|
},
|
|
{
|
|
Entity: "info",
|
|
Action: "read",
|
|
},
|
|
{
|
|
Entity: "invoices",
|
|
Action: "read",
|
|
},
|
|
}
|
|
|
|
// writePermissions is a slice of all entities that allow write
|
|
// permissions for authorization purposes, all lowercase.
|
|
writePermissions = []bakery.Op{
|
|
{
|
|
Entity: "onchain",
|
|
Action: "write",
|
|
},
|
|
{
|
|
Entity: "offchain",
|
|
Action: "write",
|
|
},
|
|
{
|
|
Entity: "address",
|
|
Action: "write",
|
|
},
|
|
{
|
|
Entity: "message",
|
|
Action: "write",
|
|
},
|
|
{
|
|
Entity: "peers",
|
|
Action: "write",
|
|
},
|
|
{
|
|
Entity: "info",
|
|
Action: "write",
|
|
},
|
|
{
|
|
Entity: "invoices",
|
|
Action: "write",
|
|
},
|
|
{
|
|
Entity: "signer",
|
|
Action: "generate",
|
|
},
|
|
{
|
|
Entity: "macaroon",
|
|
Action: "generate",
|
|
},
|
|
}
|
|
|
|
// invoicePermissions is a slice of all the entities that allows a user
|
|
// to only access calls that are related to invoices, so: streaming
|
|
// RPCs, generating, and listening invoices.
|
|
invoicePermissions = []bakery.Op{
|
|
{
|
|
Entity: "invoices",
|
|
Action: "read",
|
|
},
|
|
{
|
|
Entity: "invoices",
|
|
Action: "write",
|
|
},
|
|
{
|
|
Entity: "address",
|
|
Action: "read",
|
|
},
|
|
{
|
|
Entity: "address",
|
|
Action: "write",
|
|
},
|
|
{
|
|
Entity: "onchain",
|
|
Action: "read",
|
|
},
|
|
}
|
|
|
|
// TODO(guggero): Refactor into constants that are used for all
|
|
// permissions in this file. Also expose the list of possible
|
|
// permissions in an RPC when per RPC permissions are
|
|
// implemented.
|
|
validActions = []string{"read", "write", "generate"}
|
|
validEntities = []string{
|
|
"onchain", "offchain", "address", "message",
|
|
"peers", "info", "invoices", "signer", "macaroon",
|
|
"address",
|
|
}
|
|
)
|
|
|
|
// stringInSlice returns true if a string is contained in the given slice.
|
|
func stringInSlice(a string, slice []string) bool {
|
|
for _, b := range slice {
|
|
if b == a {
|
|
return true
|
|
}
|
|
}
|
|
return false
|
|
}
|
|
|
|
// mainRPCServerPermissions returns a mapping of the main RPC server calls to
|
|
// the permissions they require.
|
|
func mainRPCServerPermissions() map[string][]bakery.Op {
|
|
return map[string][]bakery.Op{
|
|
"/lnrpc.Lightning/SendCoins": {{
|
|
Entity: "onchain",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/ListUnspent": {{
|
|
Entity: "onchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/SendMany": {{
|
|
Entity: "onchain",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/NewAddress": {{
|
|
Entity: "address",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/SignMessage": {{
|
|
Entity: "message",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/VerifyMessage": {{
|
|
Entity: "message",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/ConnectPeer": {{
|
|
Entity: "peers",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/DisconnectPeer": {{
|
|
Entity: "peers",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/OpenChannel": {{
|
|
Entity: "onchain",
|
|
Action: "write",
|
|
}, {
|
|
Entity: "offchain",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/OpenChannelSync": {{
|
|
Entity: "onchain",
|
|
Action: "write",
|
|
}, {
|
|
Entity: "offchain",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/CloseChannel": {{
|
|
Entity: "onchain",
|
|
Action: "write",
|
|
}, {
|
|
Entity: "offchain",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/AbandonChannel": {{
|
|
Entity: "offchain",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/GetInfo": {{
|
|
Entity: "info",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/ListPeers": {{
|
|
Entity: "peers",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/WalletBalance": {{
|
|
Entity: "onchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/EstimateFee": {{
|
|
Entity: "onchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/ChannelBalance": {{
|
|
Entity: "offchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/PendingChannels": {{
|
|
Entity: "offchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/ListChannels": {{
|
|
Entity: "offchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/SubscribeChannelEvents": {{
|
|
Entity: "offchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/ClosedChannels": {{
|
|
Entity: "offchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/SendPayment": {{
|
|
Entity: "offchain",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/SendPaymentSync": {{
|
|
Entity: "offchain",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/SendToRoute": {{
|
|
Entity: "offchain",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/SendToRouteSync": {{
|
|
Entity: "offchain",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/AddInvoice": {{
|
|
Entity: "invoices",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/LookupInvoice": {{
|
|
Entity: "invoices",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/ListInvoices": {{
|
|
Entity: "invoices",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/SubscribeInvoices": {{
|
|
Entity: "invoices",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/SubscribeTransactions": {{
|
|
Entity: "onchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/GetTransactions": {{
|
|
Entity: "onchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/DescribeGraph": {{
|
|
Entity: "info",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/GetChanInfo": {{
|
|
Entity: "info",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/GetNodeInfo": {{
|
|
Entity: "info",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/QueryRoutes": {{
|
|
Entity: "info",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/GetNetworkInfo": {{
|
|
Entity: "info",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/StopDaemon": {{
|
|
Entity: "info",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/SubscribeChannelGraph": {{
|
|
Entity: "info",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/ListPayments": {{
|
|
Entity: "offchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/DeleteAllPayments": {{
|
|
Entity: "offchain",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/DebugLevel": {{
|
|
Entity: "info",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/DecodePayReq": {{
|
|
Entity: "offchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/FeeReport": {{
|
|
Entity: "offchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/UpdateChannelPolicy": {{
|
|
Entity: "offchain",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/ForwardingHistory": {{
|
|
Entity: "offchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/RestoreChannelBackups": {{
|
|
Entity: "offchain",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/ExportChannelBackup": {{
|
|
Entity: "offchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/VerifyChanBackup": {{
|
|
Entity: "offchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/ExportAllChannelBackups": {{
|
|
Entity: "offchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/SubscribeChannelBackups": {{
|
|
Entity: "offchain",
|
|
Action: "read",
|
|
}},
|
|
"/lnrpc.Lightning/ChannelAcceptor": {{
|
|
Entity: "onchain",
|
|
Action: "write",
|
|
}, {
|
|
Entity: "offchain",
|
|
Action: "write",
|
|
}},
|
|
"/lnrpc.Lightning/BakeMacaroon": {{
|
|
Entity: "macaroon",
|
|
Action: "generate",
|
|
}},
|
|
}
|
|
}
|
|
|
|
// rpcServer is a gRPC, RPC front end to the lnd daemon.
|
|
// TODO(roasbeef): pagination support for the list-style calls
|
|
type rpcServer struct {
|
|
started int32 // To be used atomically.
|
|
shutdown int32 // To be used atomically.
|
|
|
|
server *server
|
|
|
|
// subServers are a set of sub-RPC servers that use the same gRPC and
|
|
// listening sockets as the main RPC server, but which maintain their
|
|
// own independent service. This allows us to expose a set of
|
|
// micro-service like abstractions to the outside world for users to
|
|
// consume.
|
|
subServers []lnrpc.SubServer
|
|
|
|
// grpcServer is the main gRPC server that this RPC server, and all the
|
|
// sub-servers will use to register themselves and accept client
|
|
// requests from.
|
|
grpcServer *grpc.Server
|
|
|
|
// listeners is a list of listeners to use when starting the grpc
|
|
// server. We make it configurable such that the grpc server can listen
|
|
// on custom interfaces.
|
|
listeners []net.Listener
|
|
|
|
// listenerCleanUp are a set of closures functions that will allow this
|
|
// main RPC server to clean up all the listening socket created for the
|
|
// server.
|
|
listenerCleanUp []func()
|
|
|
|
// restDialOpts are a set of gRPC dial options that the REST server
|
|
// proxy will use to connect to the main gRPC server.
|
|
restDialOpts []grpc.DialOption
|
|
|
|
// restProxyDest is the address to forward REST requests to.
|
|
restProxyDest string
|
|
|
|
// tlsCfg is the TLS config that allows the REST server proxy to
|
|
// connect to the main gRPC server to proxy all incoming requests.
|
|
tlsCfg *tls.Config
|
|
|
|
// routerBackend contains the backend implementation of the router
|
|
// rpc sub server.
|
|
routerBackend *routerrpc.RouterBackend
|
|
|
|
// chanPredicate is used in the bidirectional ChannelAcceptor streaming
|
|
// method.
|
|
chanPredicate *chanacceptor.ChainedAcceptor
|
|
|
|
quit chan struct{}
|
|
|
|
// macService is the macaroon service that we need to mint new
|
|
// macaroons.
|
|
macService *macaroons.Service
|
|
|
|
// selfNode is our own pubkey.
|
|
selfNode route.Vertex
|
|
}
|
|
|
|
// A compile time check to ensure that rpcServer fully implements the
|
|
// LightningServer gRPC service.
|
|
var _ lnrpc.LightningServer = (*rpcServer)(nil)
|
|
|
|
// newRPCServer creates and returns a new instance of the rpcServer. The
|
|
// rpcServer will handle creating all listening sockets needed by it, and any
|
|
// of the sub-servers that it maintains. The set of serverOpts should be the
|
|
// base level options passed to the grPC server. This typically includes things
|
|
// like requiring TLS, etc.
|
|
func newRPCServer(s *server, macService *macaroons.Service,
|
|
subServerCgs *subRPCServerConfigs, restDialOpts []grpc.DialOption,
|
|
restProxyDest string, atpl *autopilot.Manager,
|
|
invoiceRegistry *invoices.InvoiceRegistry, tower *watchtower.Standalone,
|
|
tlsCfg *tls.Config, getListeners rpcListeners,
|
|
chanPredicate *chanacceptor.ChainedAcceptor) (*rpcServer, error) {
|
|
|
|
// Set up router rpc backend.
|
|
channelGraph := s.chanDB.ChannelGraph()
|
|
selfNode, err := channelGraph.SourceNode()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
graph := s.chanDB.ChannelGraph()
|
|
routerBackend := &routerrpc.RouterBackend{
|
|
MaxPaymentMSat: MaxPaymentMSat,
|
|
SelfNode: selfNode.PubKeyBytes,
|
|
FetchChannelCapacity: func(chanID uint64) (btcutil.Amount,
|
|
error) {
|
|
|
|
info, _, _, err := graph.FetchChannelEdgesByID(chanID)
|
|
if err != nil {
|
|
return 0, err
|
|
}
|
|
return info.Capacity, nil
|
|
},
|
|
FetchChannelEndpoints: func(chanID uint64) (route.Vertex,
|
|
route.Vertex, error) {
|
|
|
|
info, _, _, err := graph.FetchChannelEdgesByID(
|
|
chanID,
|
|
)
|
|
if err != nil {
|
|
return route.Vertex{}, route.Vertex{},
|
|
fmt.Errorf("unable to fetch channel "+
|
|
"edges by channel ID %d: %v",
|
|
chanID, err)
|
|
}
|
|
|
|
return info.NodeKey1Bytes, info.NodeKey2Bytes, nil
|
|
},
|
|
FindRoute: s.chanRouter.FindRoute,
|
|
MissionControl: s.missionControl,
|
|
ActiveNetParams: activeNetParams.Params,
|
|
Tower: s.controlTower,
|
|
MaxTotalTimelock: cfg.MaxOutgoingCltvExpiry,
|
|
}
|
|
|
|
var (
|
|
subServers []lnrpc.SubServer
|
|
subServerPerms []lnrpc.MacaroonPerms
|
|
)
|
|
|
|
// Before we create any of the sub-servers, we need to ensure that all
|
|
// the dependencies they need are properly populated within each sub
|
|
// server configuration struct.
|
|
err = subServerCgs.PopulateDependencies(
|
|
s.cc, networkDir, macService, atpl, invoiceRegistry,
|
|
s.htlcSwitch, activeNetParams.Params, s.chanRouter,
|
|
routerBackend, s.nodeSigner, s.chanDB, s.sweeper, tower,
|
|
s.towerClient, cfg.net.ResolveTCPAddr,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Now that the sub-servers have all their dependencies in place, we
|
|
// can create each sub-server!
|
|
registeredSubServers := lnrpc.RegisteredSubServers()
|
|
for _, subServer := range registeredSubServers {
|
|
subServerInstance, macPerms, err := subServer.New(subServerCgs)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// We'll collect the sub-server, and also the set of
|
|
// permissions it needs for macaroons so we can apply the
|
|
// interceptors below.
|
|
subServers = append(subServers, subServerInstance)
|
|
subServerPerms = append(subServerPerms, macPerms)
|
|
}
|
|
|
|
// Next, we need to merge the set of sub server macaroon permissions
|
|
// with the main RPC server permissions so we can unite them under a
|
|
// single set of interceptors.
|
|
permissions := mainRPCServerPermissions()
|
|
for _, subServerPerm := range subServerPerms {
|
|
for method, ops := range subServerPerm {
|
|
// For each new method:ops combo, we also ensure that
|
|
// non of the sub-servers try to override each other.
|
|
if _, ok := permissions[method]; ok {
|
|
return nil, fmt.Errorf("detected duplicate "+
|
|
"macaroon constraints for path: %v",
|
|
method)
|
|
}
|
|
|
|
permissions[method] = ops
|
|
}
|
|
}
|
|
|
|
// If macaroons aren't disabled (a non-nil service), then we'll set up
|
|
// our set of interceptors which will allow us to handle the macaroon
|
|
// authentication in a single location.
|
|
macUnaryInterceptors := []grpc.UnaryServerInterceptor{}
|
|
macStrmInterceptors := []grpc.StreamServerInterceptor{}
|
|
if macService != nil {
|
|
unaryInterceptor := macService.UnaryServerInterceptor(permissions)
|
|
macUnaryInterceptors = append(macUnaryInterceptors, unaryInterceptor)
|
|
|
|
strmInterceptor := macService.StreamServerInterceptor(permissions)
|
|
macStrmInterceptors = append(macStrmInterceptors, strmInterceptor)
|
|
}
|
|
|
|
// Get interceptors for Prometheus to gather gRPC performance metrics.
|
|
// If monitoring is disabled, GetPromInterceptors() will return empty
|
|
// slices.
|
|
promUnaryInterceptors, promStrmInterceptors := monitoring.GetPromInterceptors()
|
|
|
|
// Concatenate the slices of unary and stream interceptors respectively.
|
|
unaryInterceptors := append(macUnaryInterceptors, promUnaryInterceptors...)
|
|
strmInterceptors := append(macStrmInterceptors, promStrmInterceptors...)
|
|
|
|
// We'll also add our logging interceptors as well, so we can
|
|
// automatically log all errors that happen during RPC calls.
|
|
unaryInterceptors = append(
|
|
unaryInterceptors, errorLogUnaryServerInterceptor(rpcsLog),
|
|
)
|
|
strmInterceptors = append(
|
|
strmInterceptors, errorLogStreamServerInterceptor(rpcsLog),
|
|
)
|
|
|
|
// Get the listeners and server options to use for this rpc server.
|
|
listeners, cleanup, serverOpts, err := getListeners()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// If any interceptors have been set up, add them to the server options.
|
|
if len(unaryInterceptors) != 0 && len(strmInterceptors) != 0 {
|
|
chainedUnary := grpc_middleware.WithUnaryServerChain(
|
|
unaryInterceptors...,
|
|
)
|
|
chainedStream := grpc_middleware.WithStreamServerChain(
|
|
strmInterceptors...,
|
|
)
|
|
serverOpts = append(serverOpts, chainedUnary, chainedStream)
|
|
}
|
|
|
|
// Finally, with all the pre-set up complete, we can create the main
|
|
// gRPC server, and register the main lnrpc server along side.
|
|
grpcServer := grpc.NewServer(serverOpts...)
|
|
rootRPCServer := &rpcServer{
|
|
restDialOpts: restDialOpts,
|
|
listeners: listeners,
|
|
listenerCleanUp: []func(){cleanup},
|
|
restProxyDest: restProxyDest,
|
|
subServers: subServers,
|
|
tlsCfg: tlsCfg,
|
|
grpcServer: grpcServer,
|
|
server: s,
|
|
routerBackend: routerBackend,
|
|
chanPredicate: chanPredicate,
|
|
quit: make(chan struct{}, 1),
|
|
macService: macService,
|
|
selfNode: selfNode.PubKeyBytes,
|
|
}
|
|
lnrpc.RegisterLightningServer(grpcServer, rootRPCServer)
|
|
|
|
// Now the main RPC server has been registered, we'll iterate through
|
|
// all the sub-RPC servers and register them to ensure that requests
|
|
// are properly routed towards them.
|
|
for _, subServer := range subServers {
|
|
err := subServer.RegisterWithRootServer(grpcServer)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to register "+
|
|
"sub-server %v with root: %v",
|
|
subServer.Name(), err)
|
|
}
|
|
}
|
|
|
|
return rootRPCServer, nil
|
|
}
|
|
|
|
// Start launches any helper goroutines required for the rpcServer to function.
|
|
func (r *rpcServer) Start() error {
|
|
if atomic.AddInt32(&r.started, 1) != 1 {
|
|
return nil
|
|
}
|
|
|
|
// First, we'll start all the sub-servers to ensure that they're ready
|
|
// to take new requests in.
|
|
//
|
|
// TODO(roasbeef): some may require that the entire daemon be started
|
|
// at that point
|
|
for _, subServer := range r.subServers {
|
|
rpcsLog.Debugf("Starting sub RPC server: %v", subServer.Name())
|
|
|
|
if err := subServer.Start(); err != nil {
|
|
return err
|
|
}
|
|
}
|
|
|
|
// With all the sub-servers started, we'll spin up the listeners for
|
|
// the main RPC server itself.
|
|
for _, lis := range r.listeners {
|
|
go func(lis net.Listener) {
|
|
rpcsLog.Infof("RPC server listening on %s", lis.Addr())
|
|
r.grpcServer.Serve(lis)
|
|
}(lis)
|
|
}
|
|
|
|
// If Prometheus monitoring is enabled, start the Prometheus exporter.
|
|
if cfg.Prometheus.Enabled() {
|
|
err := monitoring.ExportPrometheusMetrics(
|
|
r.grpcServer, cfg.Prometheus,
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
}
|
|
|
|
// Finally, start the REST proxy for our gRPC server above. We'll ensure
|
|
// we direct LND to connect to its loopback address rather than a
|
|
// wildcard to prevent certificate issues when accessing the proxy
|
|
// externally.
|
|
//
|
|
// TODO(roasbeef): eventually also allow the sub-servers to themselves
|
|
// have a REST proxy.
|
|
mux := proxy.NewServeMux()
|
|
|
|
err := lnrpc.RegisterLightningHandlerFromEndpoint(
|
|
context.Background(), mux, r.restProxyDest,
|
|
r.restDialOpts,
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
for _, restEndpoint := range cfg.RESTListeners {
|
|
lis, err := lncfg.TLSListenOnAddress(restEndpoint, r.tlsCfg)
|
|
if err != nil {
|
|
ltndLog.Errorf(
|
|
"gRPC proxy unable to listen on %s",
|
|
restEndpoint,
|
|
)
|
|
return err
|
|
}
|
|
|
|
r.listenerCleanUp = append(r.listenerCleanUp, func() {
|
|
lis.Close()
|
|
})
|
|
|
|
go func() {
|
|
rpcsLog.Infof("gRPC proxy started at %s", lis.Addr())
|
|
http.Serve(lis, mux)
|
|
}()
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// Stop signals any active goroutines for a graceful closure.
|
|
func (r *rpcServer) Stop() error {
|
|
if atomic.AddInt32(&r.shutdown, 1) != 1 {
|
|
return nil
|
|
}
|
|
|
|
rpcsLog.Infof("Stopping RPC Server")
|
|
|
|
close(r.quit)
|
|
|
|
// After we've signalled all of our active goroutines to exit, we'll
|
|
// then do the same to signal a graceful shutdown of all the sub
|
|
// servers.
|
|
for _, subServer := range r.subServers {
|
|
rpcsLog.Infof("Stopping %v Sub-RPC Server",
|
|
subServer.Name())
|
|
|
|
if err := subServer.Stop(); err != nil {
|
|
rpcsLog.Errorf("unable to stop sub-server %v: %v",
|
|
subServer.Name(), err)
|
|
continue
|
|
}
|
|
}
|
|
|
|
// Finally, we can clean up all the listening sockets to ensure that we
|
|
// give the file descriptors back to the OS.
|
|
for _, cleanUp := range r.listenerCleanUp {
|
|
cleanUp()
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// addrPairsToOutputs converts a map describing a set of outputs to be created,
|
|
// the outputs themselves. The passed map pairs up an address, to a desired
|
|
// output value amount. Each address is converted to its corresponding pkScript
|
|
// to be used within the constructed output(s).
|
|
func addrPairsToOutputs(addrPairs map[string]int64) ([]*wire.TxOut, error) {
|
|
outputs := make([]*wire.TxOut, 0, len(addrPairs))
|
|
for addr, amt := range addrPairs {
|
|
addr, err := btcutil.DecodeAddress(addr, activeNetParams.Params)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
pkscript, err := txscript.PayToAddrScript(addr)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
outputs = append(outputs, wire.NewTxOut(amt, pkscript))
|
|
}
|
|
|
|
return outputs, nil
|
|
}
|
|
|
|
// sendCoinsOnChain makes an on-chain transaction in or to send coins to one or
|
|
// more addresses specified in the passed payment map. The payment map maps an
|
|
// address to a specified output value to be sent to that address.
|
|
func (r *rpcServer) sendCoinsOnChain(paymentMap map[string]int64,
|
|
feeRate chainfee.SatPerKWeight) (*chainhash.Hash, error) {
|
|
|
|
outputs, err := addrPairsToOutputs(paymentMap)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
tx, err := r.server.cc.wallet.SendOutputs(outputs, feeRate)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
txHash := tx.TxHash()
|
|
return &txHash, nil
|
|
}
|
|
|
|
// ListUnspent returns useful information about each unspent output owned by
|
|
// the wallet, as reported by the underlying `ListUnspentWitness`; the
|
|
// information returned is: outpoint, amount in satoshis, address, address
|
|
// type, scriptPubKey in hex and number of confirmations. The result is
|
|
// filtered to contain outputs whose number of confirmations is between a
|
|
// minimum and maximum number of confirmations specified by the user, with 0
|
|
// meaning unconfirmed.
|
|
func (r *rpcServer) ListUnspent(ctx context.Context,
|
|
in *lnrpc.ListUnspentRequest) (*lnrpc.ListUnspentResponse, error) {
|
|
|
|
minConfs := in.MinConfs
|
|
maxConfs := in.MaxConfs
|
|
|
|
switch {
|
|
// Ensure that the user didn't attempt to specify a negative number of
|
|
// confirmations, as that isn't possible.
|
|
case minConfs < 0:
|
|
return nil, fmt.Errorf("min confirmations must be >= 0")
|
|
|
|
// We'll also ensure that the min number of confs is strictly less than
|
|
// or equal to the max number of confs for sanity.
|
|
case minConfs > maxConfs:
|
|
return nil, fmt.Errorf("max confirmations must be >= min " +
|
|
"confirmations")
|
|
}
|
|
|
|
// With our arguments validated, we'll query the internal wallet for
|
|
// the set of UTXOs that match our query.
|
|
utxos, err := r.server.cc.wallet.ListUnspentWitness(minConfs, maxConfs)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
resp := &lnrpc.ListUnspentResponse{
|
|
Utxos: make([]*lnrpc.Utxo, 0, len(utxos)),
|
|
}
|
|
|
|
for _, utxo := range utxos {
|
|
// Translate lnwallet address type to the proper gRPC proto
|
|
// address type.
|
|
var addrType lnrpc.AddressType
|
|
switch utxo.AddressType {
|
|
|
|
case lnwallet.WitnessPubKey:
|
|
addrType = lnrpc.AddressType_WITNESS_PUBKEY_HASH
|
|
|
|
case lnwallet.NestedWitnessPubKey:
|
|
addrType = lnrpc.AddressType_NESTED_PUBKEY_HASH
|
|
|
|
case lnwallet.UnknownAddressType:
|
|
rpcsLog.Warnf("[listunspent] utxo with address of "+
|
|
"unknown type ignored: %v",
|
|
utxo.OutPoint.String())
|
|
continue
|
|
|
|
default:
|
|
return nil, fmt.Errorf("invalid utxo address type")
|
|
}
|
|
|
|
// Now that we know we have a proper mapping to an address,
|
|
// we'll convert the regular outpoint to an lnrpc variant.
|
|
outpoint := &lnrpc.OutPoint{
|
|
TxidBytes: utxo.OutPoint.Hash[:],
|
|
TxidStr: utxo.OutPoint.Hash.String(),
|
|
OutputIndex: utxo.OutPoint.Index,
|
|
}
|
|
|
|
utxoResp := lnrpc.Utxo{
|
|
Type: addrType,
|
|
AmountSat: int64(utxo.Value),
|
|
PkScript: hex.EncodeToString(utxo.PkScript),
|
|
Outpoint: outpoint,
|
|
Confirmations: utxo.Confirmations,
|
|
}
|
|
|
|
// Finally, we'll attempt to extract the raw address from the
|
|
// script so we can display a human friendly address to the end
|
|
// user.
|
|
_, outAddresses, _, err := txscript.ExtractPkScriptAddrs(
|
|
utxo.PkScript, activeNetParams.Params,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// If we can't properly locate a single address, then this was
|
|
// an error in our mapping, and we'll return an error back to
|
|
// the user.
|
|
if len(outAddresses) != 1 {
|
|
return nil, fmt.Errorf("an output was unexpectedly " +
|
|
"multisig")
|
|
}
|
|
|
|
utxoResp.Address = outAddresses[0].String()
|
|
|
|
resp.Utxos = append(resp.Utxos, &utxoResp)
|
|
}
|
|
|
|
maxStr := ""
|
|
if maxConfs != math.MaxInt32 {
|
|
maxStr = " max=" + fmt.Sprintf("%d", maxConfs)
|
|
}
|
|
|
|
rpcsLog.Debugf("[listunspent] min=%v%v, generated utxos: %v", minConfs,
|
|
maxStr, utxos)
|
|
|
|
return resp, nil
|
|
}
|
|
|
|
// EstimateFee handles a request for estimating the fee for sending a
|
|
// transaction spending to multiple specified outputs in parallel.
|
|
func (r *rpcServer) EstimateFee(ctx context.Context,
|
|
in *lnrpc.EstimateFeeRequest) (*lnrpc.EstimateFeeResponse, error) {
|
|
|
|
// Create the list of outputs we are spending to.
|
|
outputs, err := addrPairsToOutputs(in.AddrToAmount)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Query the fee estimator for the fee rate for the given confirmation
|
|
// target.
|
|
target := in.TargetConf
|
|
feePerKw, err := sweep.DetermineFeePerKw(
|
|
r.server.cc.feeEstimator, sweep.FeePreference{
|
|
ConfTarget: uint32(target),
|
|
},
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// We will ask the wallet to create a tx using this fee rate. We set
|
|
// dryRun=true to avoid inflating the change addresses in the db.
|
|
var tx *txauthor.AuthoredTx
|
|
wallet := r.server.cc.wallet
|
|
err = wallet.WithCoinSelectLock(func() error {
|
|
tx, err = wallet.CreateSimpleTx(outputs, feePerKw, true)
|
|
return err
|
|
})
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Use the created tx to calculate the total fee.
|
|
totalOutput := int64(0)
|
|
for _, out := range tx.Tx.TxOut {
|
|
totalOutput += out.Value
|
|
}
|
|
totalFee := int64(tx.TotalInput) - totalOutput
|
|
|
|
resp := &lnrpc.EstimateFeeResponse{
|
|
FeeSat: totalFee,
|
|
FeerateSatPerByte: int64(feePerKw.FeePerKVByte() / 1000),
|
|
}
|
|
|
|
rpcsLog.Debugf("[estimatefee] fee estimate for conf target %d: %v",
|
|
target, resp)
|
|
|
|
return resp, nil
|
|
}
|
|
|
|
// SendCoins executes a request to send coins to a particular address. Unlike
|
|
// SendMany, this RPC call only allows creating a single output at a time.
|
|
func (r *rpcServer) SendCoins(ctx context.Context,
|
|
in *lnrpc.SendCoinsRequest) (*lnrpc.SendCoinsResponse, error) {
|
|
|
|
// Based on the passed fee related parameters, we'll determine an
|
|
// appropriate fee rate for this transaction.
|
|
satPerKw := chainfee.SatPerKVByte(in.SatPerByte * 1000).FeePerKWeight()
|
|
feePerKw, err := sweep.DetermineFeePerKw(
|
|
r.server.cc.feeEstimator, sweep.FeePreference{
|
|
ConfTarget: uint32(in.TargetConf),
|
|
FeeRate: satPerKw,
|
|
},
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
rpcsLog.Infof("[sendcoins] addr=%v, amt=%v, sat/kw=%v, sweep_all=%v",
|
|
in.Addr, btcutil.Amount(in.Amount), int64(feePerKw),
|
|
in.SendAll)
|
|
|
|
// Decode the address receiving the coins, we need to check whether the
|
|
// address is valid for this network.
|
|
targetAddr, err := btcutil.DecodeAddress(in.Addr, activeNetParams.Params)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Make the check on the decoded address according to the active network.
|
|
if !targetAddr.IsForNet(activeNetParams.Params) {
|
|
return nil, fmt.Errorf("address: %v is not valid for this "+
|
|
"network: %v", targetAddr.String(),
|
|
activeNetParams.Params.Name)
|
|
}
|
|
|
|
// If the destination address parses to a valid pubkey, we assume the user
|
|
// accidentally tried to send funds to a bare pubkey address. This check is
|
|
// here to prevent unintended transfers.
|
|
decodedAddr, _ := hex.DecodeString(in.Addr)
|
|
_, err = btcec.ParsePubKey(decodedAddr, btcec.S256())
|
|
if err == nil {
|
|
return nil, fmt.Errorf("cannot send coins to pubkeys")
|
|
}
|
|
|
|
var txid *chainhash.Hash
|
|
|
|
wallet := r.server.cc.wallet
|
|
|
|
// If the send all flag is active, then we'll attempt to sweep all the
|
|
// coins in the wallet in a single transaction (if possible),
|
|
// otherwise, we'll respect the amount, and attempt a regular 2-output
|
|
// send.
|
|
if in.SendAll {
|
|
// At this point, the amount shouldn't be set since we've been
|
|
// instructed to sweep all the coins from the wallet.
|
|
if in.Amount != 0 {
|
|
return nil, fmt.Errorf("amount set while SendAll is " +
|
|
"active")
|
|
}
|
|
|
|
_, bestHeight, err := r.server.cc.chainIO.GetBestBlock()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// With the sweeper instance created, we can now generate a
|
|
// transaction that will sweep ALL outputs from the wallet in a
|
|
// single transaction. This will be generated in a concurrent
|
|
// safe manner, so no need to worry about locking.
|
|
sweepTxPkg, err := sweep.CraftSweepAllTx(
|
|
feePerKw, uint32(bestHeight), targetAddr, wallet,
|
|
wallet.WalletController, wallet.WalletController,
|
|
r.server.cc.feeEstimator, r.server.cc.signer,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
rpcsLog.Debugf("Sweeping all coins from wallet to addr=%v, "+
|
|
"with tx=%v", in.Addr, spew.Sdump(sweepTxPkg.SweepTx))
|
|
|
|
// As our sweep transaction was created, successfully, we'll
|
|
// now attempt to publish it, cancelling the sweep pkg to
|
|
// return all outputs if it fails.
|
|
err = wallet.PublishTransaction(sweepTxPkg.SweepTx)
|
|
if err != nil {
|
|
sweepTxPkg.CancelSweepAttempt()
|
|
|
|
return nil, fmt.Errorf("unable to broadcast sweep "+
|
|
"transaction: %v", err)
|
|
}
|
|
|
|
sweepTXID := sweepTxPkg.SweepTx.TxHash()
|
|
txid = &sweepTXID
|
|
} else {
|
|
|
|
// We'll now construct out payment map, and use the wallet's
|
|
// coin selection synchronization method to ensure that no coin
|
|
// selection (funding, sweep alls, other sends) can proceed
|
|
// while we instruct the wallet to send this transaction.
|
|
paymentMap := map[string]int64{targetAddr.String(): in.Amount}
|
|
err := wallet.WithCoinSelectLock(func() error {
|
|
newTXID, err := r.sendCoinsOnChain(paymentMap, feePerKw)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
txid = newTXID
|
|
|
|
return nil
|
|
})
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
}
|
|
|
|
rpcsLog.Infof("[sendcoins] spend generated txid: %v", txid.String())
|
|
|
|
return &lnrpc.SendCoinsResponse{Txid: txid.String()}, nil
|
|
}
|
|
|
|
// SendMany handles a request for a transaction create multiple specified
|
|
// outputs in parallel.
|
|
func (r *rpcServer) SendMany(ctx context.Context,
|
|
in *lnrpc.SendManyRequest) (*lnrpc.SendManyResponse, error) {
|
|
|
|
// Based on the passed fee related parameters, we'll determine an
|
|
// appropriate fee rate for this transaction.
|
|
satPerKw := chainfee.SatPerKVByte(in.SatPerByte * 1000).FeePerKWeight()
|
|
feePerKw, err := sweep.DetermineFeePerKw(
|
|
r.server.cc.feeEstimator, sweep.FeePreference{
|
|
ConfTarget: uint32(in.TargetConf),
|
|
FeeRate: satPerKw,
|
|
},
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
rpcsLog.Infof("[sendmany] outputs=%v, sat/kw=%v",
|
|
spew.Sdump(in.AddrToAmount), int64(feePerKw))
|
|
|
|
var txid *chainhash.Hash
|
|
|
|
// We'll attempt to send to the target set of outputs, ensuring that we
|
|
// synchronize with any other ongoing coin selection attempts which
|
|
// happen to also be concurrently executing.
|
|
wallet := r.server.cc.wallet
|
|
err = wallet.WithCoinSelectLock(func() error {
|
|
sendManyTXID, err := r.sendCoinsOnChain(
|
|
in.AddrToAmount, feePerKw,
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
txid = sendManyTXID
|
|
|
|
return nil
|
|
})
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
rpcsLog.Infof("[sendmany] spend generated txid: %v", txid.String())
|
|
|
|
return &lnrpc.SendManyResponse{Txid: txid.String()}, nil
|
|
}
|
|
|
|
// NewAddress creates a new address under control of the local wallet.
|
|
func (r *rpcServer) NewAddress(ctx context.Context,
|
|
in *lnrpc.NewAddressRequest) (*lnrpc.NewAddressResponse, error) {
|
|
|
|
// Translate the gRPC proto address type to the wallet controller's
|
|
// available address types.
|
|
var (
|
|
addr btcutil.Address
|
|
err error
|
|
)
|
|
switch in.Type {
|
|
case lnrpc.AddressType_WITNESS_PUBKEY_HASH:
|
|
addr, err = r.server.cc.wallet.NewAddress(
|
|
lnwallet.WitnessPubKey, false,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
case lnrpc.AddressType_NESTED_PUBKEY_HASH:
|
|
addr, err = r.server.cc.wallet.NewAddress(
|
|
lnwallet.NestedWitnessPubKey, false,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
case lnrpc.AddressType_UNUSED_WITNESS_PUBKEY_HASH:
|
|
addr, err = r.server.cc.wallet.LastUnusedAddress(
|
|
lnwallet.WitnessPubKey,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
case lnrpc.AddressType_UNUSED_NESTED_PUBKEY_HASH:
|
|
addr, err = r.server.cc.wallet.LastUnusedAddress(
|
|
lnwallet.NestedWitnessPubKey,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
}
|
|
|
|
rpcsLog.Debugf("[newaddress] type=%v addr=%v", in.Type, addr.String())
|
|
return &lnrpc.NewAddressResponse{Address: addr.String()}, nil
|
|
}
|
|
|
|
var (
|
|
// signedMsgPrefix is a special prefix that we'll prepend to any
|
|
// messages we sign/verify. We do this to ensure that we don't
|
|
// accidentally sign a sighash, or other sensitive material. By
|
|
// prepending this fragment, we mind message signing to our particular
|
|
// context.
|
|
signedMsgPrefix = []byte("Lightning Signed Message:")
|
|
)
|
|
|
|
// SignMessage signs a message with the resident node's private key. The
|
|
// returned signature string is zbase32 encoded and pubkey recoverable, meaning
|
|
// that only the message digest and signature are needed for verification.
|
|
func (r *rpcServer) SignMessage(ctx context.Context,
|
|
in *lnrpc.SignMessageRequest) (*lnrpc.SignMessageResponse, error) {
|
|
|
|
if in.Msg == nil {
|
|
return nil, fmt.Errorf("need a message to sign")
|
|
}
|
|
|
|
in.Msg = append(signedMsgPrefix, in.Msg...)
|
|
sigBytes, err := r.server.nodeSigner.SignCompact(in.Msg)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
sig := zbase32.EncodeToString(sigBytes)
|
|
return &lnrpc.SignMessageResponse{Signature: sig}, nil
|
|
}
|
|
|
|
// VerifyMessage verifies a signature over a msg. The signature must be zbase32
|
|
// encoded and signed by an active node in the resident node's channel
|
|
// database. In addition to returning the validity of the signature,
|
|
// VerifyMessage also returns the recovered pubkey from the signature.
|
|
func (r *rpcServer) VerifyMessage(ctx context.Context,
|
|
in *lnrpc.VerifyMessageRequest) (*lnrpc.VerifyMessageResponse, error) {
|
|
|
|
if in.Msg == nil {
|
|
return nil, fmt.Errorf("need a message to verify")
|
|
}
|
|
|
|
// The signature should be zbase32 encoded
|
|
sig, err := zbase32.DecodeString(in.Signature)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("failed to decode signature: %v", err)
|
|
}
|
|
|
|
// The signature is over the double-sha256 hash of the message.
|
|
in.Msg = append(signedMsgPrefix, in.Msg...)
|
|
digest := chainhash.DoubleHashB(in.Msg)
|
|
|
|
// RecoverCompact both recovers the pubkey and validates the signature.
|
|
pubKey, _, err := btcec.RecoverCompact(btcec.S256(), sig, digest)
|
|
if err != nil {
|
|
return &lnrpc.VerifyMessageResponse{Valid: false}, nil
|
|
}
|
|
pubKeyHex := hex.EncodeToString(pubKey.SerializeCompressed())
|
|
|
|
var pub [33]byte
|
|
copy(pub[:], pubKey.SerializeCompressed())
|
|
|
|
// Query the channel graph to ensure a node in the network with active
|
|
// channels signed the message.
|
|
//
|
|
// TODO(phlip9): Require valid nodes to have capital in active channels.
|
|
graph := r.server.chanDB.ChannelGraph()
|
|
_, active, err := graph.HasLightningNode(pub)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("failed to query graph: %v", err)
|
|
}
|
|
|
|
return &lnrpc.VerifyMessageResponse{
|
|
Valid: active,
|
|
Pubkey: pubKeyHex,
|
|
}, nil
|
|
}
|
|
|
|
// ConnectPeer attempts to establish a connection to a remote peer.
|
|
func (r *rpcServer) ConnectPeer(ctx context.Context,
|
|
in *lnrpc.ConnectPeerRequest) (*lnrpc.ConnectPeerResponse, error) {
|
|
|
|
// The server hasn't yet started, so it won't be able to service any of
|
|
// our requests, so we'll bail early here.
|
|
if !r.server.Started() {
|
|
return nil, ErrServerNotActive
|
|
}
|
|
|
|
if in.Addr == nil {
|
|
return nil, fmt.Errorf("need: lnc pubkeyhash@hostname")
|
|
}
|
|
|
|
pubkeyHex, err := hex.DecodeString(in.Addr.Pubkey)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
pubKey, err := btcec.ParsePubKey(pubkeyHex, btcec.S256())
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Connections to ourselves are disallowed for obvious reasons.
|
|
if pubKey.IsEqual(r.server.identityPriv.PubKey()) {
|
|
return nil, fmt.Errorf("cannot make connection to self")
|
|
}
|
|
|
|
addr, err := parseAddr(in.Addr.Host)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
peerAddr := &lnwire.NetAddress{
|
|
IdentityKey: pubKey,
|
|
Address: addr,
|
|
ChainNet: activeNetParams.Net,
|
|
}
|
|
|
|
rpcsLog.Debugf("[connectpeer] requested connection to %x@%s",
|
|
peerAddr.IdentityKey.SerializeCompressed(), peerAddr.Address)
|
|
|
|
if err := r.server.ConnectToPeer(peerAddr, in.Perm); err != nil {
|
|
rpcsLog.Errorf("[connectpeer]: error connecting to peer: %v", err)
|
|
return nil, err
|
|
}
|
|
|
|
rpcsLog.Debugf("Connected to peer: %v", peerAddr.String())
|
|
return &lnrpc.ConnectPeerResponse{}, nil
|
|
}
|
|
|
|
// DisconnectPeer attempts to disconnect one peer from another identified by a
|
|
// given pubKey. In the case that we currently have a pending or active channel
|
|
// with the target peer, this action will be disallowed.
|
|
func (r *rpcServer) DisconnectPeer(ctx context.Context,
|
|
in *lnrpc.DisconnectPeerRequest) (*lnrpc.DisconnectPeerResponse, error) {
|
|
|
|
rpcsLog.Debugf("[disconnectpeer] from peer(%s)", in.PubKey)
|
|
|
|
if !r.server.Started() {
|
|
return nil, ErrServerNotActive
|
|
}
|
|
|
|
// First we'll validate the string passed in within the request to
|
|
// ensure that it's a valid hex-string, and also a valid compressed
|
|
// public key.
|
|
pubKeyBytes, err := hex.DecodeString(in.PubKey)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to decode pubkey bytes: %v", err)
|
|
}
|
|
peerPubKey, err := btcec.ParsePubKey(pubKeyBytes, btcec.S256())
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to parse pubkey: %v", err)
|
|
}
|
|
|
|
// Next, we'll fetch the pending/active channels we have with a
|
|
// particular peer.
|
|
nodeChannels, err := r.server.chanDB.FetchOpenChannels(peerPubKey)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to fetch channels for peer: %v", err)
|
|
}
|
|
|
|
// In order to avoid erroneously disconnecting from a peer that we have
|
|
// an active channel with, if we have any channels active with this
|
|
// peer, then we'll disallow disconnecting from them.
|
|
if len(nodeChannels) > 0 && !cfg.UnsafeDisconnect {
|
|
return nil, fmt.Errorf("cannot disconnect from peer(%x), "+
|
|
"all active channels with the peer need to be closed "+
|
|
"first", pubKeyBytes)
|
|
}
|
|
|
|
// With all initial validation complete, we'll now request that the
|
|
// server disconnects from the peer.
|
|
if err := r.server.DisconnectPeer(peerPubKey); err != nil {
|
|
return nil, fmt.Errorf("unable to disconnect peer: %v", err)
|
|
}
|
|
|
|
return &lnrpc.DisconnectPeerResponse{}, nil
|
|
}
|
|
|
|
// extractOpenChannelMinConfs extracts the minimum number of confirmations from
|
|
// the OpenChannelRequest that each output used to fund the channel's funding
|
|
// transaction should satisfy.
|
|
func extractOpenChannelMinConfs(in *lnrpc.OpenChannelRequest) (int32, error) {
|
|
switch {
|
|
// Ensure that the MinConfs parameter is non-negative.
|
|
case in.MinConfs < 0:
|
|
return 0, errors.New("minimum number of confirmations must " +
|
|
"be a non-negative number")
|
|
|
|
// The funding transaction should not be funded with unconfirmed outputs
|
|
// unless explicitly specified by SpendUnconfirmed. We do this to
|
|
// provide sane defaults to the OpenChannel RPC, as otherwise, if the
|
|
// MinConfs field isn't explicitly set by the caller, we'll use
|
|
// unconfirmed outputs without the caller being aware.
|
|
case in.MinConfs == 0 && !in.SpendUnconfirmed:
|
|
return 1, nil
|
|
|
|
// In the event that the caller set MinConfs > 0 and SpendUnconfirmed to
|
|
// true, we'll return an error to indicate the conflict.
|
|
case in.MinConfs > 0 && in.SpendUnconfirmed:
|
|
return 0, errors.New("SpendUnconfirmed set to true with " +
|
|
"MinConfs > 0")
|
|
|
|
// The funding transaction of the new channel to be created can be
|
|
// funded with unconfirmed outputs.
|
|
case in.SpendUnconfirmed:
|
|
return 0, nil
|
|
|
|
// If none of the above cases matched, we'll return the value set
|
|
// explicitly by the caller.
|
|
default:
|
|
return in.MinConfs, nil
|
|
}
|
|
}
|
|
|
|
// OpenChannel attempts to open a singly funded channel specified in the
|
|
// request to a remote peer.
|
|
func (r *rpcServer) OpenChannel(in *lnrpc.OpenChannelRequest,
|
|
updateStream lnrpc.Lightning_OpenChannelServer) error {
|
|
|
|
rpcsLog.Tracef("[openchannel] request to NodeKey(%v) "+
|
|
"allocation(us=%v, them=%v)", in.NodePubkeyString,
|
|
in.LocalFundingAmount, in.PushSat)
|
|
|
|
if !r.server.Started() {
|
|
return ErrServerNotActive
|
|
}
|
|
|
|
localFundingAmt := btcutil.Amount(in.LocalFundingAmount)
|
|
remoteInitialBalance := btcutil.Amount(in.PushSat)
|
|
minHtlc := lnwire.MilliSatoshi(in.MinHtlcMsat)
|
|
remoteCsvDelay := uint16(in.RemoteCsvDelay)
|
|
|
|
// Ensure that the initial balance of the remote party (if pushing
|
|
// satoshis) does not exceed the amount the local party has requested
|
|
// for funding.
|
|
//
|
|
// TODO(roasbeef): incorporate base fee?
|
|
if remoteInitialBalance >= localFundingAmt {
|
|
return fmt.Errorf("amount pushed to remote peer for initial " +
|
|
"state must be below the local funding amount")
|
|
}
|
|
|
|
// Ensure that the user doesn't exceed the current soft-limit for
|
|
// channel size. If the funding amount is above the soft-limit, then
|
|
// we'll reject the request.
|
|
if localFundingAmt > MaxFundingAmount {
|
|
return fmt.Errorf("funding amount is too large, the max "+
|
|
"channel size is: %v", MaxFundingAmount)
|
|
}
|
|
|
|
// Restrict the size of the channel we'll actually open. At a later
|
|
// level, we'll ensure that the output we create after accounting for
|
|
// fees that a dust output isn't created.
|
|
if localFundingAmt < minChanFundingSize {
|
|
return fmt.Errorf("channel is too small, the minimum channel "+
|
|
"size is: %v SAT", int64(minChanFundingSize))
|
|
}
|
|
|
|
// Then, we'll extract the minimum number of confirmations that each
|
|
// output we use to fund the channel's funding transaction should
|
|
// satisfy.
|
|
minConfs, err := extractOpenChannelMinConfs(in)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
var (
|
|
nodePubKey *btcec.PublicKey
|
|
nodePubKeyBytes []byte
|
|
)
|
|
|
|
// TODO(roasbeef): also return channel ID?
|
|
|
|
// Ensure that the NodePubKey is set before attempting to use it
|
|
if len(in.NodePubkey) == 0 {
|
|
return fmt.Errorf("NodePubKey is not set")
|
|
}
|
|
|
|
// Parse the raw bytes of the node key into a pubkey object so we
|
|
// can easily manipulate it.
|
|
nodePubKey, err = btcec.ParsePubKey(in.NodePubkey, btcec.S256())
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
// Making a channel to ourselves wouldn't be of any use, so we
|
|
// explicitly disallow them.
|
|
if nodePubKey.IsEqual(r.server.identityPriv.PubKey()) {
|
|
return fmt.Errorf("cannot open channel to self")
|
|
}
|
|
|
|
nodePubKeyBytes = nodePubKey.SerializeCompressed()
|
|
|
|
// Based on the passed fee related parameters, we'll determine an
|
|
// appropriate fee rate for the funding transaction.
|
|
satPerKw := chainfee.SatPerKVByte(in.SatPerByte * 1000).FeePerKWeight()
|
|
feeRate, err := sweep.DetermineFeePerKw(
|
|
r.server.cc.feeEstimator, sweep.FeePreference{
|
|
ConfTarget: uint32(in.TargetConf),
|
|
FeeRate: satPerKw,
|
|
},
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
rpcsLog.Debugf("[openchannel]: using fee of %v sat/kw for funding tx",
|
|
int64(feeRate))
|
|
|
|
// Instruct the server to trigger the necessary events to attempt to
|
|
// open a new channel. A stream is returned in place, this stream will
|
|
// be used to consume updates of the state of the pending channel.
|
|
req := &openChanReq{
|
|
targetPubkey: nodePubKey,
|
|
chainHash: *activeNetParams.GenesisHash,
|
|
localFundingAmt: localFundingAmt,
|
|
pushAmt: lnwire.NewMSatFromSatoshis(remoteInitialBalance),
|
|
minHtlc: minHtlc,
|
|
fundingFeePerKw: feeRate,
|
|
private: in.Private,
|
|
remoteCsvDelay: remoteCsvDelay,
|
|
minConfs: minConfs,
|
|
}
|
|
|
|
updateChan, errChan := r.server.OpenChannel(req)
|
|
|
|
var outpoint wire.OutPoint
|
|
out:
|
|
for {
|
|
select {
|
|
case err := <-errChan:
|
|
rpcsLog.Errorf("unable to open channel to NodeKey(%x): %v",
|
|
nodePubKeyBytes, err)
|
|
return err
|
|
case fundingUpdate := <-updateChan:
|
|
rpcsLog.Tracef("[openchannel] sending update: %v",
|
|
fundingUpdate)
|
|
if err := updateStream.Send(fundingUpdate); err != nil {
|
|
return err
|
|
}
|
|
|
|
// If a final channel open update is being sent, then
|
|
// we can break out of our recv loop as we no longer
|
|
// need to process any further updates.
|
|
switch update := fundingUpdate.Update.(type) {
|
|
case *lnrpc.OpenStatusUpdate_ChanOpen:
|
|
chanPoint := update.ChanOpen.ChannelPoint
|
|
txid, err := GetChanPointFundingTxid(chanPoint)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
outpoint = wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: chanPoint.OutputIndex,
|
|
}
|
|
|
|
break out
|
|
}
|
|
case <-r.quit:
|
|
return nil
|
|
}
|
|
}
|
|
|
|
rpcsLog.Tracef("[openchannel] success NodeKey(%x), ChannelPoint(%v)",
|
|
nodePubKeyBytes, outpoint)
|
|
return nil
|
|
}
|
|
|
|
// OpenChannelSync is a synchronous version of the OpenChannel RPC call. This
|
|
// call is meant to be consumed by clients to the REST proxy. As with all other
|
|
// sync calls, all byte slices are instead to be populated as hex encoded
|
|
// strings.
|
|
func (r *rpcServer) OpenChannelSync(ctx context.Context,
|
|
in *lnrpc.OpenChannelRequest) (*lnrpc.ChannelPoint, error) {
|
|
|
|
rpcsLog.Tracef("[openchannel] request to NodeKey(%v) "+
|
|
"allocation(us=%v, them=%v)", in.NodePubkeyString,
|
|
in.LocalFundingAmount, in.PushSat)
|
|
|
|
// We don't allow new channels to be open while the server is still
|
|
// syncing, as otherwise we may not be able to obtain the relevant
|
|
// notifications.
|
|
if !r.server.Started() {
|
|
return nil, ErrServerNotActive
|
|
}
|
|
|
|
// Creation of channels before the wallet syncs up is currently
|
|
// disallowed.
|
|
isSynced, _, err := r.server.cc.wallet.IsSynced()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
if !isSynced {
|
|
return nil, errors.New("channels cannot be created before the " +
|
|
"wallet is fully synced")
|
|
}
|
|
|
|
// Decode the provided target node's public key, parsing it into a pub
|
|
// key object. For all sync call, byte slices are expected to be
|
|
// encoded as hex strings.
|
|
keyBytes, err := hex.DecodeString(in.NodePubkeyString)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
nodepubKey, err := btcec.ParsePubKey(keyBytes, btcec.S256())
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
localFundingAmt := btcutil.Amount(in.LocalFundingAmount)
|
|
remoteInitialBalance := btcutil.Amount(in.PushSat)
|
|
minHtlc := lnwire.MilliSatoshi(in.MinHtlcMsat)
|
|
remoteCsvDelay := uint16(in.RemoteCsvDelay)
|
|
|
|
// Ensure that the initial balance of the remote party (if pushing
|
|
// satoshis) does not exceed the amount the local party has requested
|
|
// for funding.
|
|
if remoteInitialBalance >= localFundingAmt {
|
|
return nil, fmt.Errorf("amount pushed to remote peer for " +
|
|
"initial state must be below the local funding amount")
|
|
}
|
|
|
|
// Restrict the size of the channel we'll actually open. At a later
|
|
// level, we'll ensure that the output we create after accounting for
|
|
// fees that a dust output isn't created.
|
|
if localFundingAmt < minChanFundingSize {
|
|
return nil, fmt.Errorf("channel is too small, the minimum channel "+
|
|
"size is: %v SAT", int64(minChanFundingSize))
|
|
}
|
|
|
|
// Then, we'll extract the minimum number of confirmations that each
|
|
// output we use to fund the channel's funding transaction should
|
|
// satisfy.
|
|
minConfs, err := extractOpenChannelMinConfs(in)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Based on the passed fee related parameters, we'll determine an
|
|
// appropriate fee rate for the funding transaction.
|
|
satPerKw := chainfee.SatPerKVByte(in.SatPerByte * 1000).FeePerKWeight()
|
|
feeRate, err := sweep.DetermineFeePerKw(
|
|
r.server.cc.feeEstimator, sweep.FeePreference{
|
|
ConfTarget: uint32(in.TargetConf),
|
|
FeeRate: satPerKw,
|
|
},
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
rpcsLog.Tracef("[openchannel] target sat/kw for funding tx: %v",
|
|
int64(feeRate))
|
|
|
|
req := &openChanReq{
|
|
targetPubkey: nodepubKey,
|
|
chainHash: *activeNetParams.GenesisHash,
|
|
localFundingAmt: localFundingAmt,
|
|
pushAmt: lnwire.NewMSatFromSatoshis(remoteInitialBalance),
|
|
minHtlc: minHtlc,
|
|
fundingFeePerKw: feeRate,
|
|
private: in.Private,
|
|
remoteCsvDelay: remoteCsvDelay,
|
|
minConfs: minConfs,
|
|
}
|
|
|
|
updateChan, errChan := r.server.OpenChannel(req)
|
|
select {
|
|
// If an error occurs them immediately return the error to the client.
|
|
case err := <-errChan:
|
|
rpcsLog.Errorf("unable to open channel to NodeKey(%x): %v",
|
|
nodepubKey, err)
|
|
return nil, err
|
|
|
|
// Otherwise, wait for the first channel update. The first update sent
|
|
// is when the funding transaction is broadcast to the network.
|
|
case fundingUpdate := <-updateChan:
|
|
rpcsLog.Tracef("[openchannel] sending update: %v",
|
|
fundingUpdate)
|
|
|
|
// Parse out the txid of the pending funding transaction. The
|
|
// sync client can use this to poll against the list of
|
|
// PendingChannels.
|
|
openUpdate := fundingUpdate.Update.(*lnrpc.OpenStatusUpdate_ChanPending)
|
|
chanUpdate := openUpdate.ChanPending
|
|
|
|
return &lnrpc.ChannelPoint{
|
|
FundingTxid: &lnrpc.ChannelPoint_FundingTxidBytes{
|
|
FundingTxidBytes: chanUpdate.Txid,
|
|
},
|
|
OutputIndex: chanUpdate.OutputIndex,
|
|
}, nil
|
|
case <-r.quit:
|
|
return nil, nil
|
|
}
|
|
}
|
|
|
|
// GetChanPointFundingTxid returns the given channel point's funding txid in
|
|
// raw bytes.
|
|
func GetChanPointFundingTxid(chanPoint *lnrpc.ChannelPoint) (*chainhash.Hash, error) {
|
|
var txid []byte
|
|
|
|
// A channel point's funding txid can be get/set as a byte slice or a
|
|
// string. In the case it is a string, decode it.
|
|
switch chanPoint.GetFundingTxid().(type) {
|
|
case *lnrpc.ChannelPoint_FundingTxidBytes:
|
|
txid = chanPoint.GetFundingTxidBytes()
|
|
case *lnrpc.ChannelPoint_FundingTxidStr:
|
|
s := chanPoint.GetFundingTxidStr()
|
|
h, err := chainhash.NewHashFromStr(s)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
txid = h[:]
|
|
}
|
|
|
|
return chainhash.NewHash(txid)
|
|
}
|
|
|
|
// CloseChannel attempts to close an active channel identified by its channel
|
|
// point. The actions of this method can additionally be augmented to attempt
|
|
// a force close after a timeout period in the case of an inactive peer.
|
|
func (r *rpcServer) CloseChannel(in *lnrpc.CloseChannelRequest,
|
|
updateStream lnrpc.Lightning_CloseChannelServer) error {
|
|
|
|
if !r.server.Started() {
|
|
return ErrServerNotActive
|
|
}
|
|
|
|
// If the user didn't specify a channel point, then we'll reject this
|
|
// request all together.
|
|
if in.GetChannelPoint() == nil {
|
|
return fmt.Errorf("must specify channel point in close channel")
|
|
}
|
|
|
|
// If force closing a channel, the fee set in the commitment transaction
|
|
// is used.
|
|
if in.Force && (in.SatPerByte != 0 || in.TargetConf != 0) {
|
|
return fmt.Errorf("force closing a channel uses a pre-defined fee")
|
|
}
|
|
|
|
force := in.Force
|
|
index := in.ChannelPoint.OutputIndex
|
|
txid, err := GetChanPointFundingTxid(in.GetChannelPoint())
|
|
if err != nil {
|
|
rpcsLog.Errorf("[closechannel] unable to get funding txid: %v", err)
|
|
return err
|
|
}
|
|
chanPoint := wire.NewOutPoint(txid, index)
|
|
|
|
rpcsLog.Tracef("[closechannel] request for ChannelPoint(%v), force=%v",
|
|
chanPoint, force)
|
|
|
|
var (
|
|
updateChan chan interface{}
|
|
errChan chan error
|
|
)
|
|
|
|
// TODO(roasbeef): if force and peer online then don't force?
|
|
|
|
// First, we'll fetch the channel as is, as we'll need to examine it
|
|
// regardless of if this is a force close or not.
|
|
channel, err := r.fetchActiveChannel(*chanPoint)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
// If a force closure was requested, then we'll handle all the details
|
|
// around the creation and broadcast of the unilateral closure
|
|
// transaction here rather than going to the switch as we don't require
|
|
// interaction from the peer.
|
|
if force {
|
|
_, bestHeight, err := r.server.cc.chainIO.GetBestBlock()
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
// As we're force closing this channel, as a precaution, we'll
|
|
// ensure that the switch doesn't continue to see this channel
|
|
// as eligible for forwarding HTLC's. If the peer is online,
|
|
// then we'll also purge all of its indexes.
|
|
remotePub := &channel.StateSnapshot().RemoteIdentity
|
|
if peer, err := r.server.FindPeer(remotePub); err == nil {
|
|
// TODO(roasbeef): actually get the active channel
|
|
// instead too?
|
|
// * so only need to grab from database
|
|
peer.WipeChannel(channel.ChannelPoint())
|
|
} else {
|
|
chanID := lnwire.NewChanIDFromOutPoint(channel.ChannelPoint())
|
|
r.server.htlcSwitch.RemoveLink(chanID)
|
|
}
|
|
|
|
// With the necessary indexes cleaned up, we'll now force close
|
|
// the channel.
|
|
chainArbitrator := r.server.chainArb
|
|
closingTx, err := chainArbitrator.ForceCloseContract(
|
|
*chanPoint,
|
|
)
|
|
if err != nil {
|
|
rpcsLog.Errorf("unable to force close transaction: %v", err)
|
|
return err
|
|
}
|
|
|
|
closingTxid := closingTx.TxHash()
|
|
|
|
// With the transaction broadcast, we send our first update to
|
|
// the client.
|
|
updateChan = make(chan interface{}, 2)
|
|
updateChan <- &pendingUpdate{
|
|
Txid: closingTxid[:],
|
|
}
|
|
|
|
errChan = make(chan error, 1)
|
|
notifier := r.server.cc.chainNotifier
|
|
go waitForChanToClose(uint32(bestHeight), notifier, errChan, chanPoint,
|
|
&closingTxid, closingTx.TxOut[0].PkScript, func() {
|
|
// Respond to the local subsystem which
|
|
// requested the channel closure.
|
|
updateChan <- &channelCloseUpdate{
|
|
ClosingTxid: closingTxid[:],
|
|
Success: true,
|
|
}
|
|
})
|
|
} else {
|
|
// If the link is not known by the switch, we cannot gracefully close
|
|
// the channel.
|
|
channelID := lnwire.NewChanIDFromOutPoint(chanPoint)
|
|
if _, err := r.server.htlcSwitch.GetLink(channelID); err != nil {
|
|
rpcsLog.Debugf("Trying to non-force close offline channel with "+
|
|
"chan_point=%v", chanPoint)
|
|
return fmt.Errorf("unable to gracefully close channel while peer "+
|
|
"is offline (try force closing it instead): %v", err)
|
|
}
|
|
|
|
// Based on the passed fee related parameters, we'll determine
|
|
// an appropriate fee rate for the cooperative closure
|
|
// transaction.
|
|
satPerKw := chainfee.SatPerKVByte(
|
|
in.SatPerByte * 1000,
|
|
).FeePerKWeight()
|
|
feeRate, err := sweep.DetermineFeePerKw(
|
|
r.server.cc.feeEstimator, sweep.FeePreference{
|
|
ConfTarget: uint32(in.TargetConf),
|
|
FeeRate: satPerKw,
|
|
},
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
rpcsLog.Debugf("Target sat/kw for closing transaction: %v",
|
|
int64(feeRate))
|
|
|
|
// Before we attempt the cooperative channel closure, we'll
|
|
// examine the channel to ensure that it doesn't have a
|
|
// lingering HTLC.
|
|
if len(channel.ActiveHtlcs()) != 0 {
|
|
return fmt.Errorf("cannot co-op close channel " +
|
|
"with active htlcs")
|
|
}
|
|
|
|
// Otherwise, the caller has requested a regular interactive
|
|
// cooperative channel closure. So we'll forward the request to
|
|
// the htlc switch which will handle the negotiation and
|
|
// broadcast details.
|
|
updateChan, errChan = r.server.htlcSwitch.CloseLink(
|
|
chanPoint, htlcswitch.CloseRegular, feeRate,
|
|
)
|
|
}
|
|
out:
|
|
for {
|
|
select {
|
|
case err := <-errChan:
|
|
rpcsLog.Errorf("[closechannel] unable to close "+
|
|
"ChannelPoint(%v): %v", chanPoint, err)
|
|
return err
|
|
case closingUpdate := <-updateChan:
|
|
rpcClosingUpdate, err := createRPCCloseUpdate(
|
|
closingUpdate,
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
rpcsLog.Tracef("[closechannel] sending update: %v",
|
|
rpcClosingUpdate)
|
|
|
|
if err := updateStream.Send(rpcClosingUpdate); err != nil {
|
|
return err
|
|
}
|
|
|
|
// If a final channel closing updates is being sent,
|
|
// then we can break out of our dispatch loop as we no
|
|
// longer need to process any further updates.
|
|
switch closeUpdate := closingUpdate.(type) {
|
|
case *channelCloseUpdate:
|
|
h, _ := chainhash.NewHash(closeUpdate.ClosingTxid)
|
|
rpcsLog.Infof("[closechannel] close completed: "+
|
|
"txid(%v)", h)
|
|
break out
|
|
}
|
|
case <-r.quit:
|
|
return nil
|
|
}
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
func createRPCCloseUpdate(update interface{}) (
|
|
*lnrpc.CloseStatusUpdate, error) {
|
|
|
|
switch u := update.(type) {
|
|
case *channelCloseUpdate:
|
|
return &lnrpc.CloseStatusUpdate{
|
|
Update: &lnrpc.CloseStatusUpdate_ChanClose{
|
|
ChanClose: &lnrpc.ChannelCloseUpdate{
|
|
ClosingTxid: u.ClosingTxid,
|
|
},
|
|
},
|
|
}, nil
|
|
case *pendingUpdate:
|
|
return &lnrpc.CloseStatusUpdate{
|
|
Update: &lnrpc.CloseStatusUpdate_ClosePending{
|
|
ClosePending: &lnrpc.PendingUpdate{
|
|
Txid: u.Txid,
|
|
OutputIndex: u.OutputIndex,
|
|
},
|
|
},
|
|
}, nil
|
|
}
|
|
|
|
return nil, errors.New("unknown close status update")
|
|
}
|
|
|
|
// abandonChanFromGraph attempts to remove a channel from the channel graph. If
|
|
// we can't find the chanID in the graph, then we assume it has already been
|
|
// removed, and will return a nop.
|
|
func abandonChanFromGraph(chanGraph *channeldb.ChannelGraph,
|
|
chanPoint *wire.OutPoint) error {
|
|
|
|
// First, we'll obtain the channel ID. If we can't locate this, then
|
|
// it's the case that the channel may have already been removed from
|
|
// the graph, so we'll return a nil error.
|
|
chanID, err := chanGraph.ChannelID(chanPoint)
|
|
switch {
|
|
case err == channeldb.ErrEdgeNotFound:
|
|
return nil
|
|
case err != nil:
|
|
return err
|
|
}
|
|
|
|
// If the channel ID is still in the graph, then that means the channel
|
|
// is still open, so we'll now move to purge it from the graph.
|
|
return chanGraph.DeleteChannelEdges(chanID)
|
|
}
|
|
|
|
// AbandonChannel removes all channel state from the database except for a
|
|
// close summary. This method can be used to get rid of permanently unusable
|
|
// channels due to bugs fixed in newer versions of lnd.
|
|
func (r *rpcServer) AbandonChannel(ctx context.Context,
|
|
in *lnrpc.AbandonChannelRequest) (*lnrpc.AbandonChannelResponse, error) {
|
|
|
|
// If this isn't the dev build, then we won't allow the RPC to be
|
|
// executed, as it's an advanced feature and won't be activated in
|
|
// regular production/release builds.
|
|
if !build.IsDevBuild() {
|
|
return nil, fmt.Errorf("AbandonChannel RPC call only " +
|
|
"available in dev builds")
|
|
}
|
|
|
|
// We'll parse out the arguments to we can obtain the chanPoint of the
|
|
// target channel.
|
|
txid, err := GetChanPointFundingTxid(in.GetChannelPoint())
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
index := in.ChannelPoint.OutputIndex
|
|
chanPoint := wire.NewOutPoint(txid, index)
|
|
|
|
// When we remove the channel from the database, we need to set a close
|
|
// height, so we'll just use the current best known height.
|
|
_, bestHeight, err := r.server.cc.chainIO.GetBestBlock()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
dbChan, err := r.server.chanDB.FetchChannel(*chanPoint)
|
|
switch {
|
|
// If the channel isn't found in the set of open channels, then we can
|
|
// continue on as it can't be loaded into the link/peer.
|
|
case err == channeldb.ErrChannelNotFound:
|
|
break
|
|
|
|
// If the channel is still known to be open, then before we modify any
|
|
// on-disk state, we'll remove the channel from the switch and peer
|
|
// state if it's been loaded in.
|
|
case err == nil:
|
|
// We'll mark the channel as borked before we remove the state
|
|
// from the switch/peer so it won't be loaded back in if the
|
|
// peer reconnects.
|
|
if err := dbChan.MarkBorked(); err != nil {
|
|
return nil, err
|
|
}
|
|
remotePub := dbChan.IdentityPub
|
|
if peer, err := r.server.FindPeer(remotePub); err == nil {
|
|
if err := peer.WipeChannel(chanPoint); err != nil {
|
|
return nil, fmt.Errorf("unable to wipe "+
|
|
"channel state: %v", err)
|
|
}
|
|
}
|
|
|
|
default:
|
|
return nil, err
|
|
}
|
|
|
|
// Abandoning a channel is a three step process: remove from the open
|
|
// channel state, remove from the graph, remove from the contract
|
|
// court. Between any step it's possible that the users restarts the
|
|
// process all over again. As a result, each of the steps below are
|
|
// intended to be idempotent.
|
|
err = r.server.chanDB.AbandonChannel(chanPoint, uint32(bestHeight))
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
err = abandonChanFromGraph(
|
|
r.server.chanDB.ChannelGraph(), chanPoint,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
err = r.server.chainArb.ResolveContract(*chanPoint)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// If this channel was in the process of being closed, but didn't fully
|
|
// close, then it's possible that the nursery is hanging on to some
|
|
// state. To err on the side of caution, we'll now attempt to wipe any
|
|
// state for this channel from the nursery.
|
|
err = r.server.utxoNursery.cfg.Store.RemoveChannel(chanPoint)
|
|
if err != nil && err != ErrContractNotFound {
|
|
return nil, err
|
|
}
|
|
|
|
return &lnrpc.AbandonChannelResponse{}, nil
|
|
}
|
|
|
|
// fetchActiveChannel attempts to locate a channel identified by its channel
|
|
// point from the database's set of all currently opened channels and
|
|
// return it as a fully populated state machine
|
|
func (r *rpcServer) fetchActiveChannel(chanPoint wire.OutPoint) (
|
|
*lnwallet.LightningChannel, error) {
|
|
|
|
dbChan, err := r.server.chanDB.FetchChannel(chanPoint)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// If the channel is successfully fetched from the database,
|
|
// we create a fully populated channel state machine which
|
|
// uses the db channel as backing storage.
|
|
return lnwallet.NewLightningChannel(
|
|
r.server.cc.wallet.Cfg.Signer, dbChan, nil,
|
|
)
|
|
}
|
|
|
|
// GetInfo returns general information concerning the lightning node including
|
|
// its identity pubkey, alias, the chains it is connected to, and information
|
|
// concerning the number of open+pending channels.
|
|
func (r *rpcServer) GetInfo(ctx context.Context,
|
|
in *lnrpc.GetInfoRequest) (*lnrpc.GetInfoResponse, error) {
|
|
|
|
serverPeers := r.server.Peers()
|
|
|
|
openChannels, err := r.server.chanDB.FetchAllOpenChannels()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
var activeChannels uint32
|
|
for _, channel := range openChannels {
|
|
chanID := lnwire.NewChanIDFromOutPoint(&channel.FundingOutpoint)
|
|
if r.server.htlcSwitch.HasActiveLink(chanID) {
|
|
activeChannels++
|
|
}
|
|
}
|
|
|
|
inactiveChannels := uint32(len(openChannels)) - activeChannels
|
|
|
|
pendingChannels, err := r.server.chanDB.FetchPendingChannels()
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to get retrieve pending "+
|
|
"channels: %v", err)
|
|
}
|
|
nPendingChannels := uint32(len(pendingChannels))
|
|
|
|
idPub := r.server.identityPriv.PubKey().SerializeCompressed()
|
|
encodedIDPub := hex.EncodeToString(idPub)
|
|
|
|
bestHash, bestHeight, err := r.server.cc.chainIO.GetBestBlock()
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to get best block info: %v", err)
|
|
}
|
|
|
|
isSynced, bestHeaderTimestamp, err := r.server.cc.wallet.IsSynced()
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to sync PoV of the wallet "+
|
|
"with current best block in the main chain: %v", err)
|
|
}
|
|
|
|
network := normalizeNetwork(activeNetParams.Name)
|
|
activeChains := make([]*lnrpc.Chain, registeredChains.NumActiveChains())
|
|
for i, chain := range registeredChains.ActiveChains() {
|
|
activeChains[i] = &lnrpc.Chain{
|
|
Chain: chain.String(),
|
|
Network: network,
|
|
}
|
|
|
|
}
|
|
|
|
// Check if external IP addresses were provided to lnd and use them
|
|
// to set the URIs.
|
|
nodeAnn, err := r.server.genNodeAnnouncement(false)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to retrieve current fully signed "+
|
|
"node announcement: %v", err)
|
|
}
|
|
addrs := nodeAnn.Addresses
|
|
uris := make([]string, len(addrs))
|
|
for i, addr := range addrs {
|
|
uris[i] = fmt.Sprintf("%s@%s", encodedIDPub, addr.String())
|
|
}
|
|
|
|
isGraphSynced := r.server.authGossiper.SyncManager().IsGraphSynced()
|
|
|
|
// TODO(roasbeef): add synced height n stuff
|
|
return &lnrpc.GetInfoResponse{
|
|
IdentityPubkey: encodedIDPub,
|
|
NumPendingChannels: nPendingChannels,
|
|
NumActiveChannels: activeChannels,
|
|
NumInactiveChannels: inactiveChannels,
|
|
NumPeers: uint32(len(serverPeers)),
|
|
BlockHeight: uint32(bestHeight),
|
|
BlockHash: bestHash.String(),
|
|
SyncedToChain: isSynced,
|
|
Testnet: isTestnet(&activeNetParams),
|
|
Chains: activeChains,
|
|
Uris: uris,
|
|
Alias: nodeAnn.Alias.String(),
|
|
Color: routing.EncodeHexColor(nodeAnn.RGBColor),
|
|
BestHeaderTimestamp: int64(bestHeaderTimestamp),
|
|
Version: build.Version(),
|
|
SyncedToGraph: isGraphSynced,
|
|
}, nil
|
|
}
|
|
|
|
// ListPeers returns a verbose listing of all currently active peers.
|
|
func (r *rpcServer) ListPeers(ctx context.Context,
|
|
in *lnrpc.ListPeersRequest) (*lnrpc.ListPeersResponse, error) {
|
|
|
|
rpcsLog.Tracef("[listpeers] request")
|
|
|
|
serverPeers := r.server.Peers()
|
|
resp := &lnrpc.ListPeersResponse{
|
|
Peers: make([]*lnrpc.Peer, 0, len(serverPeers)),
|
|
}
|
|
|
|
for _, serverPeer := range serverPeers {
|
|
var (
|
|
satSent int64
|
|
satRecv int64
|
|
)
|
|
|
|
// In order to display the total number of satoshis of outbound
|
|
// (sent) and inbound (recv'd) satoshis that have been
|
|
// transported through this peer, we'll sum up the sent/recv'd
|
|
// values for each of the active channels we have with the
|
|
// peer.
|
|
chans := serverPeer.ChannelSnapshots()
|
|
for _, c := range chans {
|
|
satSent += int64(c.TotalMSatSent.ToSatoshis())
|
|
satRecv += int64(c.TotalMSatReceived.ToSatoshis())
|
|
}
|
|
|
|
nodePub := serverPeer.PubKey()
|
|
|
|
// Retrieve the peer's sync type. If we don't currently have a
|
|
// syncer for the peer, then we'll default to a passive sync.
|
|
// This can happen if the RPC is called while a peer is
|
|
// initializing.
|
|
syncer, ok := r.server.authGossiper.SyncManager().GossipSyncer(
|
|
nodePub,
|
|
)
|
|
|
|
var lnrpcSyncType lnrpc.Peer_SyncType
|
|
if !ok {
|
|
rpcsLog.Warnf("Gossip syncer for peer=%x not found",
|
|
nodePub)
|
|
lnrpcSyncType = lnrpc.Peer_UNKNOWN_SYNC
|
|
} else {
|
|
syncType := syncer.SyncType()
|
|
switch syncType {
|
|
case discovery.ActiveSync:
|
|
lnrpcSyncType = lnrpc.Peer_ACTIVE_SYNC
|
|
case discovery.PassiveSync:
|
|
lnrpcSyncType = lnrpc.Peer_PASSIVE_SYNC
|
|
default:
|
|
return nil, fmt.Errorf("unhandled sync type %v",
|
|
syncType)
|
|
}
|
|
}
|
|
|
|
peer := &lnrpc.Peer{
|
|
PubKey: hex.EncodeToString(nodePub[:]),
|
|
Address: serverPeer.conn.RemoteAddr().String(),
|
|
Inbound: serverPeer.inbound,
|
|
BytesRecv: atomic.LoadUint64(&serverPeer.bytesReceived),
|
|
BytesSent: atomic.LoadUint64(&serverPeer.bytesSent),
|
|
SatSent: satSent,
|
|
SatRecv: satRecv,
|
|
PingTime: serverPeer.PingTime(),
|
|
SyncType: lnrpcSyncType,
|
|
}
|
|
|
|
resp.Peers = append(resp.Peers, peer)
|
|
}
|
|
|
|
rpcsLog.Debugf("[listpeers] yielded %v peers", serverPeers)
|
|
|
|
return resp, nil
|
|
}
|
|
|
|
// WalletBalance returns total unspent outputs(confirmed and unconfirmed), all
|
|
// confirmed unspent outputs and all unconfirmed unspent outputs under control
|
|
// by the wallet. This method can be modified by having the request specify
|
|
// only witness outputs should be factored into the final output sum.
|
|
// TODO(roasbeef): add async hooks into wallet balance changes
|
|
func (r *rpcServer) WalletBalance(ctx context.Context,
|
|
in *lnrpc.WalletBalanceRequest) (*lnrpc.WalletBalanceResponse, error) {
|
|
|
|
// Get total balance, from txs that have >= 0 confirmations.
|
|
totalBal, err := r.server.cc.wallet.ConfirmedBalance(0)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Get confirmed balance, from txs that have >= 1 confirmations.
|
|
// TODO(halseth): get both unconfirmed and confirmed balance in one
|
|
// call, as this is racy.
|
|
confirmedBal, err := r.server.cc.wallet.ConfirmedBalance(1)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Get unconfirmed balance, from txs with 0 confirmations.
|
|
unconfirmedBal := totalBal - confirmedBal
|
|
|
|
rpcsLog.Debugf("[walletbalance] Total balance=%v (confirmed=%v, "+
|
|
"unconfirmed=%v)", totalBal, confirmedBal, unconfirmedBal)
|
|
|
|
return &lnrpc.WalletBalanceResponse{
|
|
TotalBalance: int64(totalBal),
|
|
ConfirmedBalance: int64(confirmedBal),
|
|
UnconfirmedBalance: int64(unconfirmedBal),
|
|
}, nil
|
|
}
|
|
|
|
// ChannelBalance returns the total available channel flow across all open
|
|
// channels in satoshis.
|
|
func (r *rpcServer) ChannelBalance(ctx context.Context,
|
|
in *lnrpc.ChannelBalanceRequest) (*lnrpc.ChannelBalanceResponse, error) {
|
|
|
|
openChannels, err := r.server.chanDB.FetchAllOpenChannels()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
var balance btcutil.Amount
|
|
for _, channel := range openChannels {
|
|
balance += channel.LocalCommitment.LocalBalance.ToSatoshis()
|
|
}
|
|
|
|
pendingChannels, err := r.server.chanDB.FetchPendingChannels()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
var pendingOpenBalance btcutil.Amount
|
|
for _, channel := range pendingChannels {
|
|
pendingOpenBalance += channel.LocalCommitment.LocalBalance.ToSatoshis()
|
|
}
|
|
|
|
rpcsLog.Debugf("[channelbalance] balance=%v pending-open=%v",
|
|
balance, pendingOpenBalance)
|
|
|
|
return &lnrpc.ChannelBalanceResponse{
|
|
Balance: int64(balance),
|
|
PendingOpenBalance: int64(pendingOpenBalance),
|
|
}, nil
|
|
}
|
|
|
|
// PendingChannels returns a list of all the channels that are currently
|
|
// considered "pending". A channel is pending if it has finished the funding
|
|
// workflow and is waiting for confirmations for the funding txn, or is in the
|
|
// process of closure, either initiated cooperatively or non-cooperatively.
|
|
func (r *rpcServer) PendingChannels(ctx context.Context,
|
|
in *lnrpc.PendingChannelsRequest) (*lnrpc.PendingChannelsResponse, error) {
|
|
|
|
rpcsLog.Debugf("[pendingchannels]")
|
|
|
|
resp := &lnrpc.PendingChannelsResponse{}
|
|
|
|
// First, we'll populate the response with all the channels that are
|
|
// soon to be opened. We can easily fetch this data from the database
|
|
// and map the db struct to the proto response.
|
|
pendingOpenChannels, err := r.server.chanDB.FetchPendingChannels()
|
|
if err != nil {
|
|
rpcsLog.Errorf("unable to fetch pending channels: %v", err)
|
|
return nil, err
|
|
}
|
|
resp.PendingOpenChannels = make([]*lnrpc.PendingChannelsResponse_PendingOpenChannel,
|
|
len(pendingOpenChannels))
|
|
for i, pendingChan := range pendingOpenChannels {
|
|
pub := pendingChan.IdentityPub.SerializeCompressed()
|
|
|
|
// As this is required for display purposes, we'll calculate
|
|
// the weight of the commitment transaction. We also add on the
|
|
// estimated weight of the witness to calculate the weight of
|
|
// the transaction if it were to be immediately unilaterally
|
|
// broadcast.
|
|
// TODO(roasbeef): query for funding tx from wallet, display
|
|
// that also?
|
|
localCommitment := pendingChan.LocalCommitment
|
|
utx := btcutil.NewTx(localCommitment.CommitTx)
|
|
commitBaseWeight := blockchain.GetTransactionWeight(utx)
|
|
commitWeight := commitBaseWeight + input.WitnessCommitmentTxWeight
|
|
|
|
resp.PendingOpenChannels[i] = &lnrpc.PendingChannelsResponse_PendingOpenChannel{
|
|
Channel: &lnrpc.PendingChannelsResponse_PendingChannel{
|
|
RemoteNodePub: hex.EncodeToString(pub),
|
|
ChannelPoint: pendingChan.FundingOutpoint.String(),
|
|
Capacity: int64(pendingChan.Capacity),
|
|
LocalBalance: int64(localCommitment.LocalBalance.ToSatoshis()),
|
|
RemoteBalance: int64(localCommitment.RemoteBalance.ToSatoshis()),
|
|
LocalChanReserveSat: int64(pendingChan.LocalChanCfg.ChanReserve),
|
|
RemoteChanReserveSat: int64(pendingChan.RemoteChanCfg.ChanReserve),
|
|
},
|
|
CommitWeight: commitWeight,
|
|
CommitFee: int64(localCommitment.CommitFee),
|
|
FeePerKw: int64(localCommitment.FeePerKw),
|
|
// TODO(roasbeef): need to track confirmation height
|
|
}
|
|
}
|
|
|
|
_, currentHeight, err := r.server.cc.chainIO.GetBestBlock()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Next, we'll examine the channels that are soon to be closed so we
|
|
// can populate these fields within the response.
|
|
pendingCloseChannels, err := r.server.chanDB.FetchClosedChannels(true)
|
|
if err != nil {
|
|
rpcsLog.Errorf("unable to fetch closed channels: %v", err)
|
|
return nil, err
|
|
}
|
|
for _, pendingClose := range pendingCloseChannels {
|
|
// First construct the channel struct itself, this will be
|
|
// needed regardless of how this channel was closed.
|
|
pub := pendingClose.RemotePub.SerializeCompressed()
|
|
chanPoint := pendingClose.ChanPoint
|
|
channel := &lnrpc.PendingChannelsResponse_PendingChannel{
|
|
RemoteNodePub: hex.EncodeToString(pub),
|
|
ChannelPoint: chanPoint.String(),
|
|
Capacity: int64(pendingClose.Capacity),
|
|
LocalBalance: int64(pendingClose.SettledBalance),
|
|
}
|
|
|
|
closeTXID := pendingClose.ClosingTXID.String()
|
|
|
|
switch pendingClose.CloseType {
|
|
|
|
// If the channel was closed cooperatively, then we'll only
|
|
// need to tack on the closing txid.
|
|
// TODO(halseth): remove. After recent changes, a coop closed
|
|
// channel should never be in the "pending close" state.
|
|
// Keeping for now to let someone that upgraded in the middle
|
|
// of a close let their closing tx confirm.
|
|
case channeldb.CooperativeClose:
|
|
resp.PendingClosingChannels = append(
|
|
resp.PendingClosingChannels,
|
|
&lnrpc.PendingChannelsResponse_ClosedChannel{
|
|
Channel: channel,
|
|
ClosingTxid: closeTXID,
|
|
},
|
|
)
|
|
|
|
resp.TotalLimboBalance += channel.LocalBalance
|
|
|
|
// If the channel was force closed, then we'll need to query
|
|
// the utxoNursery for additional information.
|
|
// TODO(halseth): distinguish remote and local case?
|
|
case channeldb.LocalForceClose, channeldb.RemoteForceClose:
|
|
forceClose := &lnrpc.PendingChannelsResponse_ForceClosedChannel{
|
|
Channel: channel,
|
|
ClosingTxid: closeTXID,
|
|
}
|
|
|
|
// Fetch reports from both nursery and resolvers. At the
|
|
// moment this is not an atomic snapshot. This is
|
|
// planned to be resolved when the nursery is removed
|
|
// and channel arbitrator will be the single source for
|
|
// these kind of reports.
|
|
err := r.nurseryPopulateForceCloseResp(
|
|
&chanPoint, currentHeight, forceClose,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
err = r.arbitratorPopulateForceCloseResp(
|
|
&chanPoint, currentHeight, forceClose,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
resp.TotalLimboBalance += int64(forceClose.LimboBalance)
|
|
|
|
resp.PendingForceClosingChannels = append(
|
|
resp.PendingForceClosingChannels,
|
|
forceClose,
|
|
)
|
|
}
|
|
}
|
|
|
|
// We'll also fetch all channels that are open, but have had their
|
|
// commitment broadcasted, meaning they are waiting for the closing
|
|
// transaction to confirm.
|
|
waitingCloseChans, err := r.server.chanDB.FetchWaitingCloseChannels()
|
|
if err != nil {
|
|
rpcsLog.Errorf("unable to fetch channels waiting close: %v",
|
|
err)
|
|
return nil, err
|
|
}
|
|
|
|
for _, waitingClose := range waitingCloseChans {
|
|
pub := waitingClose.IdentityPub.SerializeCompressed()
|
|
chanPoint := waitingClose.FundingOutpoint
|
|
channel := &lnrpc.PendingChannelsResponse_PendingChannel{
|
|
RemoteNodePub: hex.EncodeToString(pub),
|
|
ChannelPoint: chanPoint.String(),
|
|
Capacity: int64(waitingClose.Capacity),
|
|
LocalBalance: int64(waitingClose.LocalCommitment.LocalBalance.ToSatoshis()),
|
|
RemoteBalance: int64(waitingClose.LocalCommitment.RemoteBalance.ToSatoshis()),
|
|
LocalChanReserveSat: int64(waitingClose.LocalChanCfg.ChanReserve),
|
|
RemoteChanReserveSat: int64(waitingClose.RemoteChanCfg.ChanReserve),
|
|
}
|
|
|
|
// A close tx has been broadcasted, all our balance will be in
|
|
// limbo until it confirms.
|
|
resp.WaitingCloseChannels = append(
|
|
resp.WaitingCloseChannels,
|
|
&lnrpc.PendingChannelsResponse_WaitingCloseChannel{
|
|
Channel: channel,
|
|
LimboBalance: channel.LocalBalance,
|
|
},
|
|
)
|
|
|
|
resp.TotalLimboBalance += channel.LocalBalance
|
|
}
|
|
|
|
return resp, nil
|
|
}
|
|
|
|
// arbitratorPopulateForceCloseResp populates the pending channels response
|
|
// message with channel resolution information from the contract resolvers.
|
|
func (r *rpcServer) arbitratorPopulateForceCloseResp(chanPoint *wire.OutPoint,
|
|
currentHeight int32,
|
|
forceClose *lnrpc.PendingChannelsResponse_ForceClosedChannel) error {
|
|
|
|
// Query for contract resolvers state.
|
|
arbitrator, err := r.server.chainArb.GetChannelArbitrator(*chanPoint)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
reports := arbitrator.Report()
|
|
|
|
for _, report := range reports {
|
|
switch report.Type {
|
|
|
|
// For a direct output, populate/update the top level
|
|
// response properties.
|
|
case contractcourt.ReportOutputUnencumbered:
|
|
// Populate the maturity height fields for the direct
|
|
// commitment output to us.
|
|
forceClose.MaturityHeight = report.MaturityHeight
|
|
|
|
// If the transaction has been confirmed, then we can
|
|
// compute how many blocks it has left.
|
|
if forceClose.MaturityHeight != 0 {
|
|
forceClose.BlocksTilMaturity =
|
|
int32(forceClose.MaturityHeight) -
|
|
currentHeight
|
|
}
|
|
|
|
// Add htlcs to the PendingHtlcs response property.
|
|
case contractcourt.ReportOutputIncomingHtlc,
|
|
contractcourt.ReportOutputOutgoingHtlc:
|
|
|
|
incoming := report.Type == contractcourt.ReportOutputIncomingHtlc
|
|
htlc := &lnrpc.PendingHTLC{
|
|
Incoming: incoming,
|
|
Amount: int64(report.Amount),
|
|
Outpoint: report.Outpoint.String(),
|
|
MaturityHeight: report.MaturityHeight,
|
|
Stage: report.Stage,
|
|
}
|
|
|
|
if htlc.MaturityHeight != 0 {
|
|
htlc.BlocksTilMaturity =
|
|
int32(htlc.MaturityHeight) - currentHeight
|
|
}
|
|
|
|
forceClose.PendingHtlcs = append(forceClose.PendingHtlcs, htlc)
|
|
|
|
default:
|
|
return fmt.Errorf("unknown report output type: %v",
|
|
report.Type)
|
|
}
|
|
|
|
forceClose.LimboBalance += int64(report.LimboBalance)
|
|
forceClose.RecoveredBalance += int64(report.RecoveredBalance)
|
|
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// nurseryPopulateForceCloseResp populates the pending channels response
|
|
// message with contract resolution information from utxonursery.
|
|
func (r *rpcServer) nurseryPopulateForceCloseResp(chanPoint *wire.OutPoint,
|
|
currentHeight int32,
|
|
forceClose *lnrpc.PendingChannelsResponse_ForceClosedChannel) error {
|
|
|
|
// Query for the maturity state for this force closed channel. If we
|
|
// didn't have any time-locked outputs, then the nursery may not know of
|
|
// the contract.
|
|
nurseryInfo, err := r.server.utxoNursery.NurseryReport(chanPoint)
|
|
if err == ErrContractNotFound {
|
|
return nil
|
|
}
|
|
if err != nil {
|
|
return fmt.Errorf("unable to obtain "+
|
|
"nursery report for ChannelPoint(%v): %v",
|
|
chanPoint, err)
|
|
}
|
|
|
|
// If the nursery knows of this channel, then we can populate
|
|
// information detailing exactly how much funds are time locked and also
|
|
// the height in which we can ultimately sweep the funds into the
|
|
// wallet.
|
|
forceClose.LimboBalance = int64(nurseryInfo.limboBalance)
|
|
forceClose.RecoveredBalance = int64(nurseryInfo.recoveredBalance)
|
|
|
|
for _, htlcReport := range nurseryInfo.htlcs {
|
|
// TODO(conner) set incoming flag appropriately after handling
|
|
// incoming incubation
|
|
htlc := &lnrpc.PendingHTLC{
|
|
Incoming: false,
|
|
Amount: int64(htlcReport.amount),
|
|
Outpoint: htlcReport.outpoint.String(),
|
|
MaturityHeight: htlcReport.maturityHeight,
|
|
Stage: htlcReport.stage,
|
|
}
|
|
|
|
if htlc.MaturityHeight != 0 {
|
|
htlc.BlocksTilMaturity =
|
|
int32(htlc.MaturityHeight) -
|
|
currentHeight
|
|
}
|
|
|
|
forceClose.PendingHtlcs = append(forceClose.PendingHtlcs,
|
|
htlc)
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// ClosedChannels returns a list of all the channels have been closed.
|
|
// This does not include channels that are still in the process of closing.
|
|
func (r *rpcServer) ClosedChannels(ctx context.Context,
|
|
in *lnrpc.ClosedChannelsRequest) (*lnrpc.ClosedChannelsResponse,
|
|
error) {
|
|
|
|
// Show all channels when no filter flags are set.
|
|
filterResults := in.Cooperative || in.LocalForce ||
|
|
in.RemoteForce || in.Breach || in.FundingCanceled ||
|
|
in.Abandoned
|
|
|
|
resp := &lnrpc.ClosedChannelsResponse{}
|
|
|
|
dbChannels, err := r.server.chanDB.FetchClosedChannels(false)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// In order to make the response easier to parse for clients, we'll
|
|
// sort the set of closed channels by their closing height before
|
|
// serializing the proto response.
|
|
sort.Slice(dbChannels, func(i, j int) bool {
|
|
return dbChannels[i].CloseHeight < dbChannels[j].CloseHeight
|
|
})
|
|
|
|
for _, dbChannel := range dbChannels {
|
|
if dbChannel.IsPending {
|
|
continue
|
|
}
|
|
|
|
switch dbChannel.CloseType {
|
|
case channeldb.CooperativeClose:
|
|
if filterResults && !in.Cooperative {
|
|
continue
|
|
}
|
|
case channeldb.LocalForceClose:
|
|
if filterResults && !in.LocalForce {
|
|
continue
|
|
}
|
|
case channeldb.RemoteForceClose:
|
|
if filterResults && !in.RemoteForce {
|
|
continue
|
|
}
|
|
case channeldb.BreachClose:
|
|
if filterResults && !in.Breach {
|
|
continue
|
|
}
|
|
case channeldb.FundingCanceled:
|
|
if filterResults && !in.FundingCanceled {
|
|
continue
|
|
}
|
|
case channeldb.Abandoned:
|
|
if filterResults && !in.Abandoned {
|
|
continue
|
|
}
|
|
}
|
|
|
|
channel := createRPCClosedChannel(dbChannel)
|
|
resp.Channels = append(resp.Channels, channel)
|
|
}
|
|
|
|
return resp, nil
|
|
}
|
|
|
|
// ListChannels returns a description of all the open channels that this node
|
|
// is a participant in.
|
|
func (r *rpcServer) ListChannels(ctx context.Context,
|
|
in *lnrpc.ListChannelsRequest) (*lnrpc.ListChannelsResponse, error) {
|
|
|
|
if in.ActiveOnly && in.InactiveOnly {
|
|
return nil, fmt.Errorf("either `active_only` or " +
|
|
"`inactive_only` can be set, but not both")
|
|
}
|
|
|
|
if in.PublicOnly && in.PrivateOnly {
|
|
return nil, fmt.Errorf("either `public_only` or " +
|
|
"`private_only` can be set, but not both")
|
|
}
|
|
|
|
resp := &lnrpc.ListChannelsResponse{}
|
|
|
|
graph := r.server.chanDB.ChannelGraph()
|
|
|
|
dbChannels, err := r.server.chanDB.FetchAllOpenChannels()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
rpcsLog.Debugf("[listchannels] fetched %v channels from DB",
|
|
len(dbChannels))
|
|
|
|
for _, dbChannel := range dbChannels {
|
|
nodePub := dbChannel.IdentityPub
|
|
chanPoint := dbChannel.FundingOutpoint
|
|
|
|
var peerOnline bool
|
|
if _, err := r.server.FindPeer(nodePub); err == nil {
|
|
peerOnline = true
|
|
}
|
|
|
|
channelID := lnwire.NewChanIDFromOutPoint(&chanPoint)
|
|
var linkActive bool
|
|
if link, err := r.server.htlcSwitch.GetLink(channelID); err == nil {
|
|
// A channel is only considered active if it is known
|
|
// by the switch *and* able to forward
|
|
// incoming/outgoing payments.
|
|
linkActive = link.EligibleToForward()
|
|
}
|
|
|
|
// Next, we'll determine whether we should add this channel to
|
|
// our list depending on the type of channels requested to us.
|
|
isActive := peerOnline && linkActive
|
|
channel := createRPCOpenChannel(r, graph, dbChannel, isActive)
|
|
|
|
// We'll only skip returning this channel if we were requested
|
|
// for a specific kind and this channel doesn't satisfy it.
|
|
switch {
|
|
case in.ActiveOnly && !isActive:
|
|
continue
|
|
case in.InactiveOnly && isActive:
|
|
continue
|
|
case in.PublicOnly && channel.Private:
|
|
continue
|
|
case in.PrivateOnly && !channel.Private:
|
|
continue
|
|
}
|
|
|
|
resp.Channels = append(resp.Channels, channel)
|
|
}
|
|
|
|
return resp, nil
|
|
}
|
|
|
|
// createRPCOpenChannel creates an *lnrpc.Channel from the *channeldb.Channel.
|
|
func createRPCOpenChannel(r *rpcServer, graph *channeldb.ChannelGraph,
|
|
dbChannel *channeldb.OpenChannel, isActive bool) *lnrpc.Channel {
|
|
|
|
nodePub := dbChannel.IdentityPub
|
|
nodeID := hex.EncodeToString(nodePub.SerializeCompressed())
|
|
chanPoint := dbChannel.FundingOutpoint
|
|
|
|
// Next, we'll determine whether the channel is public or not.
|
|
isPublic := dbChannel.ChannelFlags&lnwire.FFAnnounceChannel != 0
|
|
|
|
// As this is required for display purposes, we'll calculate
|
|
// the weight of the commitment transaction. We also add on the
|
|
// estimated weight of the witness to calculate the weight of
|
|
// the transaction if it were to be immediately unilaterally
|
|
// broadcast.
|
|
localCommit := dbChannel.LocalCommitment
|
|
utx := btcutil.NewTx(localCommit.CommitTx)
|
|
commitBaseWeight := blockchain.GetTransactionWeight(utx)
|
|
commitWeight := commitBaseWeight + input.WitnessCommitmentTxWeight
|
|
|
|
localBalance := localCommit.LocalBalance
|
|
remoteBalance := localCommit.RemoteBalance
|
|
|
|
// As an artifact of our usage of mSAT internally, either party
|
|
// may end up in a state where they're holding a fractional
|
|
// amount of satoshis which can't be expressed within the
|
|
// actual commitment output. Since we round down when going
|
|
// from mSAT -> SAT, we may at any point be adding an
|
|
// additional SAT to miners fees. As a result, we display a
|
|
// commitment fee that accounts for this externally.
|
|
var sumOutputs btcutil.Amount
|
|
for _, txOut := range localCommit.CommitTx.TxOut {
|
|
sumOutputs += btcutil.Amount(txOut.Value)
|
|
}
|
|
externalCommitFee := dbChannel.Capacity - sumOutputs
|
|
|
|
chanID := dbChannel.ShortChannelID.ToUint64()
|
|
|
|
var (
|
|
uptime time.Duration
|
|
lifespan time.Duration
|
|
)
|
|
|
|
// Get the lifespan observed by the channel event store.
|
|
startTime, endTime, err := r.server.chanEventStore.GetLifespan(chanID)
|
|
if err != nil {
|
|
// If the channel cannot be found, log an error and do not perform
|
|
// further calculations for uptime and lifespan.
|
|
rpcsLog.Warnf("GetLifespan %v error: %v", chanID, err)
|
|
} else {
|
|
// If endTime is zero, the channel is still open, progress endTime to
|
|
// the present so we can calculate lifespan.
|
|
if endTime.IsZero() {
|
|
endTime = time.Now()
|
|
}
|
|
lifespan = endTime.Sub(startTime)
|
|
|
|
uptime, err = r.server.chanEventStore.GetUptime(
|
|
chanID,
|
|
startTime,
|
|
endTime,
|
|
)
|
|
if err != nil {
|
|
rpcsLog.Warnf("GetUptime %v error: %v", chanID, err)
|
|
}
|
|
}
|
|
|
|
channel := &lnrpc.Channel{
|
|
Active: isActive,
|
|
Private: !isPublic,
|
|
RemotePubkey: nodeID,
|
|
ChannelPoint: chanPoint.String(),
|
|
ChanId: chanID,
|
|
Capacity: int64(dbChannel.Capacity),
|
|
LocalBalance: int64(localBalance.ToSatoshis()),
|
|
RemoteBalance: int64(remoteBalance.ToSatoshis()),
|
|
CommitFee: int64(externalCommitFee),
|
|
CommitWeight: commitWeight,
|
|
FeePerKw: int64(localCommit.FeePerKw),
|
|
TotalSatoshisSent: int64(dbChannel.TotalMSatSent.ToSatoshis()),
|
|
TotalSatoshisReceived: int64(dbChannel.TotalMSatReceived.ToSatoshis()),
|
|
NumUpdates: localCommit.CommitHeight,
|
|
PendingHtlcs: make([]*lnrpc.HTLC, len(localCommit.Htlcs)),
|
|
CsvDelay: uint32(dbChannel.LocalChanCfg.CsvDelay),
|
|
Initiator: dbChannel.IsInitiator,
|
|
ChanStatusFlags: dbChannel.ChanStatus().String(),
|
|
LocalChanReserveSat: int64(dbChannel.LocalChanCfg.ChanReserve),
|
|
RemoteChanReserveSat: int64(dbChannel.RemoteChanCfg.ChanReserve),
|
|
StaticRemoteKey: dbChannel.ChanType.IsTweakless(),
|
|
Lifetime: int64(lifespan.Seconds()),
|
|
Uptime: int64(uptime.Seconds()),
|
|
}
|
|
|
|
for i, htlc := range localCommit.Htlcs {
|
|
var rHash [32]byte
|
|
copy(rHash[:], htlc.RHash[:])
|
|
channel.PendingHtlcs[i] = &lnrpc.HTLC{
|
|
Incoming: htlc.Incoming,
|
|
Amount: int64(htlc.Amt.ToSatoshis()),
|
|
HashLock: rHash[:],
|
|
ExpirationHeight: htlc.RefundTimeout,
|
|
}
|
|
|
|
// Add the Pending Htlc Amount to UnsettledBalance field.
|
|
channel.UnsettledBalance += channel.PendingHtlcs[i].Amount
|
|
}
|
|
|
|
return channel
|
|
}
|
|
|
|
// createRPCClosedChannel creates an *lnrpc.ClosedChannelSummary from a
|
|
// *channeldb.ChannelCloseSummary.
|
|
func createRPCClosedChannel(
|
|
dbChannel *channeldb.ChannelCloseSummary) *lnrpc.ChannelCloseSummary {
|
|
|
|
nodePub := dbChannel.RemotePub
|
|
nodeID := hex.EncodeToString(nodePub.SerializeCompressed())
|
|
|
|
var closeType lnrpc.ChannelCloseSummary_ClosureType
|
|
switch dbChannel.CloseType {
|
|
case channeldb.CooperativeClose:
|
|
closeType = lnrpc.ChannelCloseSummary_COOPERATIVE_CLOSE
|
|
case channeldb.LocalForceClose:
|
|
closeType = lnrpc.ChannelCloseSummary_LOCAL_FORCE_CLOSE
|
|
case channeldb.RemoteForceClose:
|
|
closeType = lnrpc.ChannelCloseSummary_REMOTE_FORCE_CLOSE
|
|
case channeldb.BreachClose:
|
|
closeType = lnrpc.ChannelCloseSummary_BREACH_CLOSE
|
|
case channeldb.FundingCanceled:
|
|
closeType = lnrpc.ChannelCloseSummary_FUNDING_CANCELED
|
|
case channeldb.Abandoned:
|
|
closeType = lnrpc.ChannelCloseSummary_ABANDONED
|
|
}
|
|
|
|
return &lnrpc.ChannelCloseSummary{
|
|
Capacity: int64(dbChannel.Capacity),
|
|
RemotePubkey: nodeID,
|
|
CloseHeight: dbChannel.CloseHeight,
|
|
CloseType: closeType,
|
|
ChannelPoint: dbChannel.ChanPoint.String(),
|
|
ChanId: dbChannel.ShortChanID.ToUint64(),
|
|
SettledBalance: int64(dbChannel.SettledBalance),
|
|
TimeLockedBalance: int64(dbChannel.TimeLockedBalance),
|
|
ChainHash: dbChannel.ChainHash.String(),
|
|
ClosingTxHash: dbChannel.ClosingTXID.String(),
|
|
}
|
|
}
|
|
|
|
// SubscribeChannelEvents returns a uni-directional stream (server -> client)
|
|
// for notifying the client of newly active, inactive or closed channels.
|
|
func (r *rpcServer) SubscribeChannelEvents(req *lnrpc.ChannelEventSubscription,
|
|
updateStream lnrpc.Lightning_SubscribeChannelEventsServer) error {
|
|
|
|
channelEventSub, err := r.server.channelNotifier.SubscribeChannelEvents()
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
// Ensure that the resources for the client is cleaned up once either
|
|
// the server, or client exits.
|
|
defer channelEventSub.Cancel()
|
|
|
|
graph := r.server.chanDB.ChannelGraph()
|
|
|
|
for {
|
|
select {
|
|
// A new update has been sent by the channel router, we'll
|
|
// marshal it into the form expected by the gRPC client, then
|
|
// send it off to the client(s).
|
|
case e := <-channelEventSub.Updates():
|
|
var update *lnrpc.ChannelEventUpdate
|
|
switch event := e.(type) {
|
|
case channelnotifier.OpenChannelEvent:
|
|
channel := createRPCOpenChannel(r, graph,
|
|
event.Channel, true)
|
|
update = &lnrpc.ChannelEventUpdate{
|
|
Type: lnrpc.ChannelEventUpdate_OPEN_CHANNEL,
|
|
Channel: &lnrpc.ChannelEventUpdate_OpenChannel{
|
|
OpenChannel: channel,
|
|
},
|
|
}
|
|
case channelnotifier.ClosedChannelEvent:
|
|
closedChannel := createRPCClosedChannel(event.CloseSummary)
|
|
update = &lnrpc.ChannelEventUpdate{
|
|
Type: lnrpc.ChannelEventUpdate_CLOSED_CHANNEL,
|
|
Channel: &lnrpc.ChannelEventUpdate_ClosedChannel{
|
|
ClosedChannel: closedChannel,
|
|
},
|
|
}
|
|
case channelnotifier.ActiveChannelEvent:
|
|
update = &lnrpc.ChannelEventUpdate{
|
|
Type: lnrpc.ChannelEventUpdate_ACTIVE_CHANNEL,
|
|
Channel: &lnrpc.ChannelEventUpdate_ActiveChannel{
|
|
ActiveChannel: &lnrpc.ChannelPoint{
|
|
FundingTxid: &lnrpc.ChannelPoint_FundingTxidBytes{
|
|
FundingTxidBytes: event.ChannelPoint.Hash[:],
|
|
},
|
|
OutputIndex: event.ChannelPoint.Index,
|
|
},
|
|
},
|
|
}
|
|
case channelnotifier.InactiveChannelEvent:
|
|
update = &lnrpc.ChannelEventUpdate{
|
|
Type: lnrpc.ChannelEventUpdate_INACTIVE_CHANNEL,
|
|
Channel: &lnrpc.ChannelEventUpdate_InactiveChannel{
|
|
InactiveChannel: &lnrpc.ChannelPoint{
|
|
FundingTxid: &lnrpc.ChannelPoint_FundingTxidBytes{
|
|
FundingTxidBytes: event.ChannelPoint.Hash[:],
|
|
},
|
|
OutputIndex: event.ChannelPoint.Index,
|
|
},
|
|
},
|
|
}
|
|
default:
|
|
return fmt.Errorf("unexpected channel event update: %v", event)
|
|
}
|
|
|
|
if err := updateStream.Send(update); err != nil {
|
|
return err
|
|
}
|
|
case <-r.quit:
|
|
return nil
|
|
}
|
|
}
|
|
}
|
|
|
|
// paymentStream enables different types of payment streams, such as:
|
|
// lnrpc.Lightning_SendPaymentServer and lnrpc.Lightning_SendToRouteServer to
|
|
// execute sendPayment. We use this struct as a sort of bridge to enable code
|
|
// re-use between SendPayment and SendToRoute.
|
|
type paymentStream struct {
|
|
recv func() (*rpcPaymentRequest, error)
|
|
send func(*lnrpc.SendResponse) error
|
|
}
|
|
|
|
// rpcPaymentRequest wraps lnrpc.SendRequest so that routes from
|
|
// lnrpc.SendToRouteRequest can be passed to sendPayment.
|
|
type rpcPaymentRequest struct {
|
|
*lnrpc.SendRequest
|
|
route *route.Route
|
|
}
|
|
|
|
// SendPayment dispatches a bi-directional streaming RPC for sending payments
|
|
// through the Lightning Network. A single RPC invocation creates a persistent
|
|
// bi-directional stream allowing clients to rapidly send payments through the
|
|
// Lightning Network with a single persistent connection.
|
|
func (r *rpcServer) SendPayment(stream lnrpc.Lightning_SendPaymentServer) error {
|
|
var lock sync.Mutex
|
|
|
|
return r.sendPayment(&paymentStream{
|
|
recv: func() (*rpcPaymentRequest, error) {
|
|
req, err := stream.Recv()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
return &rpcPaymentRequest{
|
|
SendRequest: req,
|
|
}, nil
|
|
},
|
|
send: func(r *lnrpc.SendResponse) error {
|
|
// Calling stream.Send concurrently is not safe.
|
|
lock.Lock()
|
|
defer lock.Unlock()
|
|
return stream.Send(r)
|
|
},
|
|
})
|
|
}
|
|
|
|
// SendToRoute dispatches a bi-directional streaming RPC for sending payments
|
|
// through the Lightning Network via predefined routes passed in. A single RPC
|
|
// invocation creates a persistent bi-directional stream allowing clients to
|
|
// rapidly send payments through the Lightning Network with a single persistent
|
|
// connection.
|
|
func (r *rpcServer) SendToRoute(stream lnrpc.Lightning_SendToRouteServer) error {
|
|
var lock sync.Mutex
|
|
|
|
return r.sendPayment(&paymentStream{
|
|
recv: func() (*rpcPaymentRequest, error) {
|
|
req, err := stream.Recv()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
return r.unmarshallSendToRouteRequest(req)
|
|
},
|
|
send: func(r *lnrpc.SendResponse) error {
|
|
// Calling stream.Send concurrently is not safe.
|
|
lock.Lock()
|
|
defer lock.Unlock()
|
|
return stream.Send(r)
|
|
},
|
|
})
|
|
}
|
|
|
|
// unmarshallSendToRouteRequest unmarshalls an rpc sendtoroute request
|
|
func (r *rpcServer) unmarshallSendToRouteRequest(
|
|
req *lnrpc.SendToRouteRequest) (*rpcPaymentRequest, error) {
|
|
|
|
if req.Route == nil {
|
|
return nil, fmt.Errorf("unable to send, no route provided")
|
|
}
|
|
|
|
route, err := r.routerBackend.UnmarshallRoute(req.Route)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
return &rpcPaymentRequest{
|
|
SendRequest: &lnrpc.SendRequest{
|
|
PaymentHash: req.PaymentHash,
|
|
PaymentHashString: req.PaymentHashString,
|
|
},
|
|
route: route,
|
|
}, nil
|
|
}
|
|
|
|
// rpcPaymentIntent is a small wrapper struct around the of values we can
|
|
// receive from a client over RPC if they wish to send a payment. We'll either
|
|
// extract these fields from a payment request (which may include routing
|
|
// hints), or we'll get a fully populated route from the user that we'll pass
|
|
// directly to the channel router for dispatching.
|
|
type rpcPaymentIntent struct {
|
|
msat lnwire.MilliSatoshi
|
|
feeLimit lnwire.MilliSatoshi
|
|
cltvLimit uint32
|
|
dest route.Vertex
|
|
rHash [32]byte
|
|
cltvDelta uint16
|
|
routeHints [][]zpay32.HopHint
|
|
outgoingChannelID *uint64
|
|
lastHop *route.Vertex
|
|
payReq []byte
|
|
|
|
destTLV []tlv.Record
|
|
|
|
route *route.Route
|
|
}
|
|
|
|
// extractPaymentIntent attempts to parse the complete details required to
|
|
// dispatch a client from the information presented by an RPC client. There are
|
|
// three ways a client can specify their payment details: a payment request,
|
|
// via manual details, or via a complete route.
|
|
func (r *rpcServer) extractPaymentIntent(rpcPayReq *rpcPaymentRequest) (rpcPaymentIntent, error) {
|
|
payIntent := rpcPaymentIntent{}
|
|
|
|
// If a route was specified, then we can use that directly.
|
|
if rpcPayReq.route != nil {
|
|
// If the user is using the REST interface, then they'll be
|
|
// passing the payment hash as a hex encoded string.
|
|
if rpcPayReq.PaymentHashString != "" {
|
|
paymentHash, err := hex.DecodeString(
|
|
rpcPayReq.PaymentHashString,
|
|
)
|
|
if err != nil {
|
|
return payIntent, err
|
|
}
|
|
|
|
copy(payIntent.rHash[:], paymentHash)
|
|
} else {
|
|
copy(payIntent.rHash[:], rpcPayReq.PaymentHash)
|
|
}
|
|
|
|
payIntent.route = rpcPayReq.route
|
|
return payIntent, nil
|
|
}
|
|
|
|
// If there are no routes specified, pass along a outgoing channel
|
|
// restriction if specified.
|
|
if rpcPayReq.OutgoingChanId != 0 {
|
|
payIntent.outgoingChannelID = &rpcPayReq.OutgoingChanId
|
|
}
|
|
|
|
// Pass along a last hop restriction if specified.
|
|
if len(rpcPayReq.LastHopPubkey) > 0 {
|
|
lastHop, err := route.NewVertexFromBytes(
|
|
rpcPayReq.LastHopPubkey,
|
|
)
|
|
if err != nil {
|
|
return payIntent, err
|
|
}
|
|
payIntent.lastHop = &lastHop
|
|
}
|
|
|
|
// Take the CLTV limit from the request if set, otherwise use the max.
|
|
cltvLimit, err := routerrpc.ValidateCLTVLimit(
|
|
rpcPayReq.CltvLimit, cfg.MaxOutgoingCltvExpiry,
|
|
)
|
|
if err != nil {
|
|
return payIntent, err
|
|
}
|
|
payIntent.cltvLimit = cltvLimit
|
|
|
|
payIntent.destTLV, err = routerrpc.UnmarshallCustomRecords(
|
|
rpcPayReq.DestCustomRecords,
|
|
)
|
|
if err != nil {
|
|
return payIntent, err
|
|
}
|
|
|
|
validateDest := func(dest route.Vertex) error {
|
|
if rpcPayReq.AllowSelfPayment {
|
|
return nil
|
|
}
|
|
|
|
if dest == r.selfNode {
|
|
return errors.New("self-payments not allowed")
|
|
}
|
|
|
|
return nil
|
|
}
|
|
|
|
// If the payment request field isn't blank, then the details of the
|
|
// invoice are encoded entirely within the encoded payReq. So we'll
|
|
// attempt to decode it, populating the payment accordingly.
|
|
if rpcPayReq.PaymentRequest != "" {
|
|
payReq, err := zpay32.Decode(
|
|
rpcPayReq.PaymentRequest, activeNetParams.Params,
|
|
)
|
|
if err != nil {
|
|
return payIntent, err
|
|
}
|
|
|
|
// Next, we'll ensure that this payreq hasn't already expired.
|
|
err = routerrpc.ValidatePayReqExpiry(payReq)
|
|
if err != nil {
|
|
return payIntent, err
|
|
}
|
|
|
|
// If the amount was not included in the invoice, then we let
|
|
// the payee specify the amount of satoshis they wish to send.
|
|
// We override the amount to pay with the amount provided from
|
|
// the payment request.
|
|
if payReq.MilliSat == nil {
|
|
amt, err := lnrpc.UnmarshallAmt(
|
|
rpcPayReq.Amt, rpcPayReq.AmtMsat,
|
|
)
|
|
if err != nil {
|
|
return payIntent, err
|
|
}
|
|
if amt == 0 {
|
|
return payIntent, errors.New("amount must be " +
|
|
"specified when paying a zero amount " +
|
|
"invoice")
|
|
}
|
|
|
|
payIntent.msat = amt
|
|
} else {
|
|
payIntent.msat = *payReq.MilliSat
|
|
}
|
|
|
|
// Calculate the fee limit that should be used for this payment.
|
|
payIntent.feeLimit = lnrpc.CalculateFeeLimit(
|
|
rpcPayReq.FeeLimit, payIntent.msat,
|
|
)
|
|
|
|
copy(payIntent.rHash[:], payReq.PaymentHash[:])
|
|
destKey := payReq.Destination.SerializeCompressed()
|
|
copy(payIntent.dest[:], destKey)
|
|
payIntent.cltvDelta = uint16(payReq.MinFinalCLTVExpiry())
|
|
payIntent.routeHints = payReq.RouteHints
|
|
payIntent.payReq = []byte(rpcPayReq.PaymentRequest)
|
|
|
|
if err := validateDest(payIntent.dest); err != nil {
|
|
return payIntent, err
|
|
}
|
|
|
|
return payIntent, nil
|
|
}
|
|
|
|
// At this point, a destination MUST be specified, so we'll convert it
|
|
// into the proper representation now. The destination will either be
|
|
// encoded as raw bytes, or via a hex string.
|
|
var pubBytes []byte
|
|
if len(rpcPayReq.Dest) != 0 {
|
|
pubBytes = rpcPayReq.Dest
|
|
} else {
|
|
var err error
|
|
pubBytes, err = hex.DecodeString(rpcPayReq.DestString)
|
|
if err != nil {
|
|
return payIntent, err
|
|
}
|
|
}
|
|
if len(pubBytes) != 33 {
|
|
return payIntent, errors.New("invalid key length")
|
|
}
|
|
copy(payIntent.dest[:], pubBytes)
|
|
|
|
if err := validateDest(payIntent.dest); err != nil {
|
|
return payIntent, err
|
|
}
|
|
|
|
// Otherwise, If the payment request field was not specified
|
|
// (and a custom route wasn't specified), construct the payment
|
|
// from the other fields.
|
|
payIntent.msat, err = lnrpc.UnmarshallAmt(
|
|
rpcPayReq.Amt, rpcPayReq.AmtMsat,
|
|
)
|
|
if err != nil {
|
|
return payIntent, err
|
|
}
|
|
|
|
// Calculate the fee limit that should be used for this payment.
|
|
payIntent.feeLimit = lnrpc.CalculateFeeLimit(
|
|
rpcPayReq.FeeLimit, payIntent.msat,
|
|
)
|
|
|
|
if rpcPayReq.FinalCltvDelta != 0 {
|
|
payIntent.cltvDelta = uint16(rpcPayReq.FinalCltvDelta)
|
|
} else {
|
|
payIntent.cltvDelta = zpay32.DefaultFinalCLTVDelta
|
|
}
|
|
|
|
// If the user is manually specifying payment details, then the payment
|
|
// hash may be encoded as a string.
|
|
switch {
|
|
case rpcPayReq.PaymentHashString != "":
|
|
paymentHash, err := hex.DecodeString(
|
|
rpcPayReq.PaymentHashString,
|
|
)
|
|
if err != nil {
|
|
return payIntent, err
|
|
}
|
|
|
|
copy(payIntent.rHash[:], paymentHash)
|
|
|
|
default:
|
|
copy(payIntent.rHash[:], rpcPayReq.PaymentHash)
|
|
}
|
|
|
|
// Currently, within the bootstrap phase of the network, we limit the
|
|
// largest payment size allotted to (2^32) - 1 mSAT or 4.29 million
|
|
// satoshis.
|
|
if payIntent.msat > MaxPaymentMSat {
|
|
// In this case, we'll send an error to the caller, but
|
|
// continue our loop for the next payment.
|
|
return payIntent, fmt.Errorf("payment of %v is too large, "+
|
|
"max payment allowed is %v", payIntent.msat,
|
|
MaxPaymentMSat)
|
|
|
|
}
|
|
|
|
return payIntent, nil
|
|
}
|
|
|
|
type paymentIntentResponse struct {
|
|
Route *route.Route
|
|
Preimage [32]byte
|
|
Err error
|
|
}
|
|
|
|
// dispatchPaymentIntent attempts to fully dispatch an RPC payment intent.
|
|
// We'll either pass the payment as a whole to the channel router, or give it a
|
|
// pre-built route. The first error this method returns denotes if we were
|
|
// unable to save the payment. The second error returned denotes if the payment
|
|
// didn't succeed.
|
|
func (r *rpcServer) dispatchPaymentIntent(
|
|
payIntent *rpcPaymentIntent) (*paymentIntentResponse, error) {
|
|
|
|
// Construct a payment request to send to the channel router. If the
|
|
// payment is successful, the route chosen will be returned. Otherwise,
|
|
// we'll get a non-nil error.
|
|
var (
|
|
preImage [32]byte
|
|
route *route.Route
|
|
routerErr error
|
|
)
|
|
|
|
// If a route was specified, then we'll pass the route directly to the
|
|
// router, otherwise we'll create a payment session to execute it.
|
|
if payIntent.route == nil {
|
|
payment := &routing.LightningPayment{
|
|
Target: payIntent.dest,
|
|
Amount: payIntent.msat,
|
|
FinalCLTVDelta: payIntent.cltvDelta,
|
|
FeeLimit: payIntent.feeLimit,
|
|
CltvLimit: payIntent.cltvLimit,
|
|
PaymentHash: payIntent.rHash,
|
|
RouteHints: payIntent.routeHints,
|
|
OutgoingChannelID: payIntent.outgoingChannelID,
|
|
LastHop: payIntent.lastHop,
|
|
PaymentRequest: payIntent.payReq,
|
|
PayAttemptTimeout: routing.DefaultPayAttemptTimeout,
|
|
FinalDestRecords: payIntent.destTLV,
|
|
}
|
|
|
|
preImage, route, routerErr = r.server.chanRouter.SendPayment(
|
|
payment,
|
|
)
|
|
} else {
|
|
preImage, routerErr = r.server.chanRouter.SendToRoute(
|
|
payIntent.rHash, payIntent.route,
|
|
)
|
|
|
|
route = payIntent.route
|
|
}
|
|
|
|
// If the route failed, then we'll return a nil save err, but a non-nil
|
|
// routing err.
|
|
if routerErr != nil {
|
|
rpcsLog.Warnf("Unable to send payment: %v", routerErr)
|
|
|
|
return &paymentIntentResponse{
|
|
Err: routerErr,
|
|
}, nil
|
|
}
|
|
|
|
return &paymentIntentResponse{
|
|
Route: route,
|
|
Preimage: preImage,
|
|
}, nil
|
|
}
|
|
|
|
// sendPayment takes a paymentStream (a source of pre-built routes or payment
|
|
// requests) and continually attempt to dispatch payment requests written to
|
|
// the write end of the stream. Responses will also be streamed back to the
|
|
// client via the write end of the stream. This method is by both SendToRoute
|
|
// and SendPayment as the logic is virtually identical.
|
|
func (r *rpcServer) sendPayment(stream *paymentStream) error {
|
|
payChan := make(chan *rpcPaymentIntent)
|
|
errChan := make(chan error, 1)
|
|
|
|
// We don't allow payments to be sent while the daemon itself is still
|
|
// syncing as we may be trying to sent a payment over a "stale"
|
|
// channel.
|
|
if !r.server.Started() {
|
|
return ErrServerNotActive
|
|
}
|
|
|
|
// TODO(roasbeef): check payment filter to see if already used?
|
|
|
|
// In order to limit the level of concurrency and prevent a client from
|
|
// attempting to OOM the server, we'll set up a semaphore to create an
|
|
// upper ceiling on the number of outstanding payments.
|
|
const numOutstandingPayments = 2000
|
|
htlcSema := make(chan struct{}, numOutstandingPayments)
|
|
for i := 0; i < numOutstandingPayments; i++ {
|
|
htlcSema <- struct{}{}
|
|
}
|
|
|
|
// Launch a new goroutine to handle reading new payment requests from
|
|
// the client. This way we can handle errors independently of blocking
|
|
// and waiting for the next payment request to come through.
|
|
reqQuit := make(chan struct{})
|
|
defer func() {
|
|
close(reqQuit)
|
|
}()
|
|
|
|
// TODO(joostjager): Callers expect result to come in in the same order
|
|
// as the request were sent, but this is far from guarantueed in the
|
|
// code below.
|
|
go func() {
|
|
for {
|
|
select {
|
|
case <-reqQuit:
|
|
return
|
|
case <-r.quit:
|
|
errChan <- nil
|
|
return
|
|
default:
|
|
// Receive the next pending payment within the
|
|
// stream sent by the client. If we read the
|
|
// EOF sentinel, then the client has closed the
|
|
// stream, and we can exit normally.
|
|
nextPayment, err := stream.recv()
|
|
if err == io.EOF {
|
|
errChan <- nil
|
|
return
|
|
} else if err != nil {
|
|
select {
|
|
case errChan <- err:
|
|
case <-reqQuit:
|
|
return
|
|
}
|
|
return
|
|
}
|
|
|
|
// Populate the next payment, either from the
|
|
// payment request, or from the explicitly set
|
|
// fields. If the payment proto wasn't well
|
|
// formed, then we'll send an error reply and
|
|
// wait for the next payment.
|
|
payIntent, err := r.extractPaymentIntent(
|
|
nextPayment,
|
|
)
|
|
if err != nil {
|
|
if err := stream.send(&lnrpc.SendResponse{
|
|
PaymentError: err.Error(),
|
|
PaymentHash: payIntent.rHash[:],
|
|
}); err != nil {
|
|
select {
|
|
case errChan <- err:
|
|
case <-reqQuit:
|
|
return
|
|
}
|
|
}
|
|
continue
|
|
}
|
|
|
|
// If the payment was well formed, then we'll
|
|
// send to the dispatch goroutine, or exit,
|
|
// which ever comes first
|
|
select {
|
|
case payChan <- &payIntent:
|
|
case <-reqQuit:
|
|
return
|
|
}
|
|
}
|
|
}
|
|
}()
|
|
|
|
for {
|
|
select {
|
|
case err := <-errChan:
|
|
return err
|
|
|
|
case payIntent := <-payChan:
|
|
// We launch a new goroutine to execute the current
|
|
// payment so we can continue to serve requests while
|
|
// this payment is being dispatched.
|
|
go func() {
|
|
// Attempt to grab a free semaphore slot, using
|
|
// a defer to eventually release the slot
|
|
// regardless of payment success.
|
|
<-htlcSema
|
|
defer func() {
|
|
htlcSema <- struct{}{}
|
|
}()
|
|
|
|
resp, saveErr := r.dispatchPaymentIntent(
|
|
payIntent,
|
|
)
|
|
|
|
switch {
|
|
// If we were unable to save the state of the
|
|
// payment, then we'll return the error to the
|
|
// user, and terminate.
|
|
case saveErr != nil:
|
|
errChan <- saveErr
|
|
return
|
|
|
|
// If we receive payment error than, instead of
|
|
// terminating the stream, send error response
|
|
// to the user.
|
|
case resp.Err != nil:
|
|
err := stream.send(&lnrpc.SendResponse{
|
|
PaymentError: resp.Err.Error(),
|
|
PaymentHash: payIntent.rHash[:],
|
|
})
|
|
if err != nil {
|
|
errChan <- err
|
|
}
|
|
return
|
|
}
|
|
|
|
backend := r.routerBackend
|
|
marshalledRouted, err := backend.MarshallRoute(
|
|
resp.Route,
|
|
)
|
|
if err != nil {
|
|
errChan <- err
|
|
return
|
|
}
|
|
|
|
err = stream.send(&lnrpc.SendResponse{
|
|
PaymentHash: payIntent.rHash[:],
|
|
PaymentPreimage: resp.Preimage[:],
|
|
PaymentRoute: marshalledRouted,
|
|
})
|
|
if err != nil {
|
|
errChan <- err
|
|
return
|
|
}
|
|
}()
|
|
}
|
|
}
|
|
}
|
|
|
|
// SendPaymentSync is the synchronous non-streaming version of SendPayment.
|
|
// This RPC is intended to be consumed by clients of the REST proxy.
|
|
// Additionally, this RPC expects the destination's public key and the payment
|
|
// hash (if any) to be encoded as hex strings.
|
|
func (r *rpcServer) SendPaymentSync(ctx context.Context,
|
|
nextPayment *lnrpc.SendRequest) (*lnrpc.SendResponse, error) {
|
|
|
|
return r.sendPaymentSync(ctx, &rpcPaymentRequest{
|
|
SendRequest: nextPayment,
|
|
})
|
|
}
|
|
|
|
// SendToRouteSync is the synchronous non-streaming version of SendToRoute.
|
|
// This RPC is intended to be consumed by clients of the REST proxy.
|
|
// Additionally, this RPC expects the payment hash (if any) to be encoded as
|
|
// hex strings.
|
|
func (r *rpcServer) SendToRouteSync(ctx context.Context,
|
|
req *lnrpc.SendToRouteRequest) (*lnrpc.SendResponse, error) {
|
|
|
|
if req.Route == nil {
|
|
return nil, fmt.Errorf("unable to send, no routes provided")
|
|
}
|
|
|
|
paymentRequest, err := r.unmarshallSendToRouteRequest(req)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
return r.sendPaymentSync(ctx, paymentRequest)
|
|
}
|
|
|
|
// sendPaymentSync is the synchronous variant of sendPayment. It will block and
|
|
// wait until the payment has been fully completed.
|
|
func (r *rpcServer) sendPaymentSync(ctx context.Context,
|
|
nextPayment *rpcPaymentRequest) (*lnrpc.SendResponse, error) {
|
|
|
|
// We don't allow payments to be sent while the daemon itself is still
|
|
// syncing as we may be trying to sent a payment over a "stale"
|
|
// channel.
|
|
if !r.server.Started() {
|
|
return nil, ErrServerNotActive
|
|
}
|
|
|
|
// First we'll attempt to map the proto describing the next payment to
|
|
// an intent that we can pass to local sub-systems.
|
|
payIntent, err := r.extractPaymentIntent(nextPayment)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// With the payment validated, we'll now attempt to dispatch the
|
|
// payment.
|
|
resp, saveErr := r.dispatchPaymentIntent(&payIntent)
|
|
switch {
|
|
case saveErr != nil:
|
|
return nil, saveErr
|
|
|
|
case resp.Err != nil:
|
|
return &lnrpc.SendResponse{
|
|
PaymentError: resp.Err.Error(),
|
|
PaymentHash: payIntent.rHash[:],
|
|
}, nil
|
|
}
|
|
|
|
rpcRoute, err := r.routerBackend.MarshallRoute(resp.Route)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
return &lnrpc.SendResponse{
|
|
PaymentHash: payIntent.rHash[:],
|
|
PaymentPreimage: resp.Preimage[:],
|
|
PaymentRoute: rpcRoute,
|
|
}, nil
|
|
}
|
|
|
|
// AddInvoice attempts to add a new invoice to the invoice database. Any
|
|
// duplicated invoices are rejected, therefore all invoices *must* have a
|
|
// unique payment preimage.
|
|
func (r *rpcServer) AddInvoice(ctx context.Context,
|
|
invoice *lnrpc.Invoice) (*lnrpc.AddInvoiceResponse, error) {
|
|
|
|
defaultDelta := cfg.Bitcoin.TimeLockDelta
|
|
if registeredChains.PrimaryChain() == litecoinChain {
|
|
defaultDelta = cfg.Litecoin.TimeLockDelta
|
|
}
|
|
|
|
addInvoiceCfg := &invoicesrpc.AddInvoiceConfig{
|
|
AddInvoice: r.server.invoices.AddInvoice,
|
|
IsChannelActive: r.server.htlcSwitch.HasActiveLink,
|
|
ChainParams: activeNetParams.Params,
|
|
NodeSigner: r.server.nodeSigner,
|
|
MaxPaymentMSat: MaxPaymentMSat,
|
|
DefaultCLTVExpiry: defaultDelta,
|
|
ChanDB: r.server.chanDB,
|
|
}
|
|
|
|
value, err := lnrpc.UnmarshallAmt(invoice.Value, invoice.ValueMsat)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
addInvoiceData := &invoicesrpc.AddInvoiceData{
|
|
Memo: invoice.Memo,
|
|
Value: value,
|
|
DescriptionHash: invoice.DescriptionHash,
|
|
Expiry: invoice.Expiry,
|
|
FallbackAddr: invoice.FallbackAddr,
|
|
CltvExpiry: invoice.CltvExpiry,
|
|
Private: invoice.Private,
|
|
}
|
|
|
|
if invoice.RPreimage != nil {
|
|
preimage, err := lntypes.MakePreimage(invoice.RPreimage)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
addInvoiceData.Preimage = &preimage
|
|
}
|
|
|
|
hash, dbInvoice, err := invoicesrpc.AddInvoice(
|
|
ctx, addInvoiceCfg, addInvoiceData,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
return &lnrpc.AddInvoiceResponse{
|
|
AddIndex: dbInvoice.AddIndex,
|
|
PaymentRequest: string(dbInvoice.PaymentRequest),
|
|
RHash: hash[:],
|
|
}, nil
|
|
}
|
|
|
|
// LookupInvoice attempts to look up an invoice according to its payment hash.
|
|
// The passed payment hash *must* be exactly 32 bytes, if not an error is
|
|
// returned.
|
|
func (r *rpcServer) LookupInvoice(ctx context.Context,
|
|
req *lnrpc.PaymentHash) (*lnrpc.Invoice, error) {
|
|
|
|
var (
|
|
payHash [32]byte
|
|
rHash []byte
|
|
err error
|
|
)
|
|
|
|
// If the RHash as a raw string was provided, then decode that and use
|
|
// that directly. Otherwise, we use the raw bytes provided.
|
|
if req.RHashStr != "" {
|
|
rHash, err = hex.DecodeString(req.RHashStr)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
} else {
|
|
rHash = req.RHash
|
|
}
|
|
|
|
// Ensure that the payment hash is *exactly* 32-bytes.
|
|
if len(rHash) != 0 && len(rHash) != 32 {
|
|
return nil, fmt.Errorf("payment hash must be exactly "+
|
|
"32 bytes, is instead %v", len(rHash))
|
|
}
|
|
copy(payHash[:], rHash)
|
|
|
|
rpcsLog.Tracef("[lookupinvoice] searching for invoice %x", payHash[:])
|
|
|
|
invoice, err := r.server.invoices.LookupInvoice(payHash)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
rpcsLog.Tracef("[lookupinvoice] located invoice %v",
|
|
newLogClosure(func() string {
|
|
return spew.Sdump(invoice)
|
|
}))
|
|
|
|
rpcInvoice, err := invoicesrpc.CreateRPCInvoice(
|
|
&invoice, activeNetParams.Params,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
return rpcInvoice, nil
|
|
}
|
|
|
|
// ListInvoices returns a list of all the invoices currently stored within the
|
|
// database. Any active debug invoices are ignored.
|
|
func (r *rpcServer) ListInvoices(ctx context.Context,
|
|
req *lnrpc.ListInvoiceRequest) (*lnrpc.ListInvoiceResponse, error) {
|
|
|
|
// If the number of invoices was not specified, then we'll default to
|
|
// returning the latest 100 invoices.
|
|
if req.NumMaxInvoices == 0 {
|
|
req.NumMaxInvoices = 100
|
|
}
|
|
|
|
// Next, we'll map the proto request into a format that is understood by
|
|
// the database.
|
|
q := channeldb.InvoiceQuery{
|
|
IndexOffset: req.IndexOffset,
|
|
NumMaxInvoices: req.NumMaxInvoices,
|
|
PendingOnly: req.PendingOnly,
|
|
Reversed: req.Reversed,
|
|
}
|
|
invoiceSlice, err := r.server.chanDB.QueryInvoices(q)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to query invoices: %v", err)
|
|
}
|
|
|
|
// Before returning the response, we'll need to convert each invoice
|
|
// into it's proto representation.
|
|
resp := &lnrpc.ListInvoiceResponse{
|
|
Invoices: make([]*lnrpc.Invoice, len(invoiceSlice.Invoices)),
|
|
FirstIndexOffset: invoiceSlice.FirstIndexOffset,
|
|
LastIndexOffset: invoiceSlice.LastIndexOffset,
|
|
}
|
|
for i, invoice := range invoiceSlice.Invoices {
|
|
resp.Invoices[i], err = invoicesrpc.CreateRPCInvoice(
|
|
&invoice, activeNetParams.Params,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
}
|
|
|
|
return resp, nil
|
|
}
|
|
|
|
// SubscribeInvoices returns a uni-directional stream (server -> client) for
|
|
// notifying the client of newly added/settled invoices.
|
|
func (r *rpcServer) SubscribeInvoices(req *lnrpc.InvoiceSubscription,
|
|
updateStream lnrpc.Lightning_SubscribeInvoicesServer) error {
|
|
|
|
invoiceClient := r.server.invoices.SubscribeNotifications(
|
|
req.AddIndex, req.SettleIndex,
|
|
)
|
|
defer invoiceClient.Cancel()
|
|
|
|
for {
|
|
select {
|
|
case newInvoice := <-invoiceClient.NewInvoices:
|
|
rpcInvoice, err := invoicesrpc.CreateRPCInvoice(
|
|
newInvoice, activeNetParams.Params,
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
if err := updateStream.Send(rpcInvoice); err != nil {
|
|
return err
|
|
}
|
|
|
|
case settledInvoice := <-invoiceClient.SettledInvoices:
|
|
rpcInvoice, err := invoicesrpc.CreateRPCInvoice(
|
|
settledInvoice, activeNetParams.Params,
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
if err := updateStream.Send(rpcInvoice); err != nil {
|
|
return err
|
|
}
|
|
|
|
case <-r.quit:
|
|
return nil
|
|
}
|
|
}
|
|
}
|
|
|
|
// SubscribeTransactions creates a uni-directional stream (server -> client) in
|
|
// which any newly discovered transactions relevant to the wallet are sent
|
|
// over.
|
|
func (r *rpcServer) SubscribeTransactions(req *lnrpc.GetTransactionsRequest,
|
|
updateStream lnrpc.Lightning_SubscribeTransactionsServer) error {
|
|
|
|
txClient, err := r.server.cc.wallet.SubscribeTransactions()
|
|
if err != nil {
|
|
return err
|
|
}
|
|
defer txClient.Cancel()
|
|
|
|
for {
|
|
select {
|
|
case tx := <-txClient.ConfirmedTransactions():
|
|
destAddresses := make([]string, 0, len(tx.DestAddresses))
|
|
for _, destAddress := range tx.DestAddresses {
|
|
destAddresses = append(destAddresses, destAddress.EncodeAddress())
|
|
}
|
|
detail := &lnrpc.Transaction{
|
|
TxHash: tx.Hash.String(),
|
|
Amount: int64(tx.Value),
|
|
NumConfirmations: tx.NumConfirmations,
|
|
BlockHash: tx.BlockHash.String(),
|
|
BlockHeight: tx.BlockHeight,
|
|
TimeStamp: tx.Timestamp,
|
|
TotalFees: tx.TotalFees,
|
|
DestAddresses: destAddresses,
|
|
RawTxHex: hex.EncodeToString(tx.RawTx),
|
|
}
|
|
if err := updateStream.Send(detail); err != nil {
|
|
return err
|
|
}
|
|
|
|
case tx := <-txClient.UnconfirmedTransactions():
|
|
var destAddresses []string
|
|
for _, destAddress := range tx.DestAddresses {
|
|
destAddresses = append(destAddresses, destAddress.EncodeAddress())
|
|
}
|
|
detail := &lnrpc.Transaction{
|
|
TxHash: tx.Hash.String(),
|
|
Amount: int64(tx.Value),
|
|
TimeStamp: tx.Timestamp,
|
|
TotalFees: tx.TotalFees,
|
|
DestAddresses: destAddresses,
|
|
RawTxHex: hex.EncodeToString(tx.RawTx),
|
|
}
|
|
if err := updateStream.Send(detail); err != nil {
|
|
return err
|
|
}
|
|
|
|
case <-r.quit:
|
|
return nil
|
|
}
|
|
}
|
|
}
|
|
|
|
// GetTransactions returns a list of describing all the known transactions
|
|
// relevant to the wallet.
|
|
func (r *rpcServer) GetTransactions(ctx context.Context,
|
|
_ *lnrpc.GetTransactionsRequest) (*lnrpc.TransactionDetails, error) {
|
|
|
|
// TODO(roasbeef): add pagination support
|
|
transactions, err := r.server.cc.wallet.ListTransactionDetails()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
txDetails := &lnrpc.TransactionDetails{
|
|
Transactions: make([]*lnrpc.Transaction, len(transactions)),
|
|
}
|
|
for i, tx := range transactions {
|
|
var destAddresses []string
|
|
for _, destAddress := range tx.DestAddresses {
|
|
destAddresses = append(destAddresses, destAddress.EncodeAddress())
|
|
}
|
|
|
|
// We also get unconfirmed transactions, so BlockHash can be
|
|
// nil.
|
|
blockHash := ""
|
|
if tx.BlockHash != nil {
|
|
blockHash = tx.BlockHash.String()
|
|
}
|
|
|
|
txDetails.Transactions[i] = &lnrpc.Transaction{
|
|
TxHash: tx.Hash.String(),
|
|
Amount: int64(tx.Value),
|
|
NumConfirmations: tx.NumConfirmations,
|
|
BlockHash: blockHash,
|
|
BlockHeight: tx.BlockHeight,
|
|
TimeStamp: tx.Timestamp,
|
|
TotalFees: tx.TotalFees,
|
|
DestAddresses: destAddresses,
|
|
RawTxHex: hex.EncodeToString(tx.RawTx),
|
|
}
|
|
}
|
|
|
|
return txDetails, nil
|
|
}
|
|
|
|
// DescribeGraph returns a description of the latest graph state from the PoV
|
|
// of the node. The graph information is partitioned into two components: all
|
|
// the nodes/vertexes, and all the edges that connect the vertexes themselves.
|
|
// As this is a directed graph, the edges also contain the node directional
|
|
// specific routing policy which includes: the time lock delta, fee
|
|
// information, etc.
|
|
func (r *rpcServer) DescribeGraph(ctx context.Context,
|
|
req *lnrpc.ChannelGraphRequest) (*lnrpc.ChannelGraph, error) {
|
|
|
|
resp := &lnrpc.ChannelGraph{}
|
|
includeUnannounced := req.IncludeUnannounced
|
|
|
|
// Obtain the pointer to the global singleton channel graph, this will
|
|
// provide a consistent view of the graph due to bolt db's
|
|
// transactional model.
|
|
graph := r.server.chanDB.ChannelGraph()
|
|
|
|
// First iterate through all the known nodes (connected or unconnected
|
|
// within the graph), collating their current state into the RPC
|
|
// response.
|
|
err := graph.ForEachNode(nil, func(_ *bbolt.Tx, node *channeldb.LightningNode) error {
|
|
nodeAddrs := make([]*lnrpc.NodeAddress, 0)
|
|
for _, addr := range node.Addresses {
|
|
nodeAddr := &lnrpc.NodeAddress{
|
|
Network: addr.Network(),
|
|
Addr: addr.String(),
|
|
}
|
|
nodeAddrs = append(nodeAddrs, nodeAddr)
|
|
}
|
|
|
|
resp.Nodes = append(resp.Nodes, &lnrpc.LightningNode{
|
|
LastUpdate: uint32(node.LastUpdate.Unix()),
|
|
PubKey: hex.EncodeToString(node.PubKeyBytes[:]),
|
|
Addresses: nodeAddrs,
|
|
Alias: node.Alias,
|
|
Color: routing.EncodeHexColor(node.Color),
|
|
})
|
|
|
|
return nil
|
|
})
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Next, for each active channel we know of within the graph, create a
|
|
// similar response which details both the edge information as well as
|
|
// the routing policies of th nodes connecting the two edges.
|
|
err = graph.ForEachChannel(func(edgeInfo *channeldb.ChannelEdgeInfo,
|
|
c1, c2 *channeldb.ChannelEdgePolicy) error {
|
|
|
|
// Do not include unannounced channels unless specifically
|
|
// requested. Unannounced channels include both private channels as
|
|
// well as public channels whose authentication proof were not
|
|
// confirmed yet, hence were not announced.
|
|
if !includeUnannounced && edgeInfo.AuthProof == nil {
|
|
return nil
|
|
}
|
|
|
|
edge := marshalDbEdge(edgeInfo, c1, c2)
|
|
resp.Edges = append(resp.Edges, edge)
|
|
|
|
return nil
|
|
})
|
|
if err != nil && err != channeldb.ErrGraphNoEdgesFound {
|
|
return nil, err
|
|
}
|
|
|
|
return resp, nil
|
|
}
|
|
|
|
func marshalDbEdge(edgeInfo *channeldb.ChannelEdgeInfo,
|
|
c1, c2 *channeldb.ChannelEdgePolicy) *lnrpc.ChannelEdge {
|
|
|
|
// Order the edges by increasing pubkey.
|
|
if bytes.Compare(edgeInfo.NodeKey2Bytes[:],
|
|
edgeInfo.NodeKey1Bytes[:]) < 0 {
|
|
|
|
c2, c1 = c1, c2
|
|
}
|
|
|
|
var lastUpdate int64
|
|
if c1 != nil {
|
|
lastUpdate = c1.LastUpdate.Unix()
|
|
}
|
|
if c2 != nil && c2.LastUpdate.Unix() > lastUpdate {
|
|
lastUpdate = c2.LastUpdate.Unix()
|
|
}
|
|
|
|
edge := &lnrpc.ChannelEdge{
|
|
ChannelId: edgeInfo.ChannelID,
|
|
ChanPoint: edgeInfo.ChannelPoint.String(),
|
|
// TODO(roasbeef): update should be on edge info itself
|
|
LastUpdate: uint32(lastUpdate),
|
|
Node1Pub: hex.EncodeToString(edgeInfo.NodeKey1Bytes[:]),
|
|
Node2Pub: hex.EncodeToString(edgeInfo.NodeKey2Bytes[:]),
|
|
Capacity: int64(edgeInfo.Capacity),
|
|
}
|
|
|
|
if c1 != nil {
|
|
edge.Node1Policy = &lnrpc.RoutingPolicy{
|
|
TimeLockDelta: uint32(c1.TimeLockDelta),
|
|
MinHtlc: int64(c1.MinHTLC),
|
|
MaxHtlcMsat: uint64(c1.MaxHTLC),
|
|
FeeBaseMsat: int64(c1.FeeBaseMSat),
|
|
FeeRateMilliMsat: int64(c1.FeeProportionalMillionths),
|
|
Disabled: c1.ChannelFlags&lnwire.ChanUpdateDisabled != 0,
|
|
LastUpdate: uint32(c1.LastUpdate.Unix()),
|
|
}
|
|
}
|
|
|
|
if c2 != nil {
|
|
edge.Node2Policy = &lnrpc.RoutingPolicy{
|
|
TimeLockDelta: uint32(c2.TimeLockDelta),
|
|
MinHtlc: int64(c2.MinHTLC),
|
|
MaxHtlcMsat: uint64(c2.MaxHTLC),
|
|
FeeBaseMsat: int64(c2.FeeBaseMSat),
|
|
FeeRateMilliMsat: int64(c2.FeeProportionalMillionths),
|
|
Disabled: c2.ChannelFlags&lnwire.ChanUpdateDisabled != 0,
|
|
LastUpdate: uint32(c2.LastUpdate.Unix()),
|
|
}
|
|
}
|
|
|
|
return edge
|
|
}
|
|
|
|
// GetChanInfo returns the latest authenticated network announcement for the
|
|
// given channel identified by its channel ID: an 8-byte integer which uniquely
|
|
// identifies the location of transaction's funding output within the block
|
|
// chain.
|
|
func (r *rpcServer) GetChanInfo(ctx context.Context,
|
|
in *lnrpc.ChanInfoRequest) (*lnrpc.ChannelEdge, error) {
|
|
|
|
graph := r.server.chanDB.ChannelGraph()
|
|
|
|
edgeInfo, edge1, edge2, err := graph.FetchChannelEdgesByID(in.ChanId)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Convert the database's edge format into the network/RPC edge format
|
|
// which couples the edge itself along with the directional node
|
|
// routing policies of each node involved within the channel.
|
|
channelEdge := marshalDbEdge(edgeInfo, edge1, edge2)
|
|
|
|
return channelEdge, nil
|
|
}
|
|
|
|
// GetNodeInfo returns the latest advertised and aggregate authenticated
|
|
// channel information for the specified node identified by its public key.
|
|
func (r *rpcServer) GetNodeInfo(ctx context.Context,
|
|
in *lnrpc.NodeInfoRequest) (*lnrpc.NodeInfo, error) {
|
|
|
|
graph := r.server.chanDB.ChannelGraph()
|
|
|
|
// First, parse the hex-encoded public key into a full in-memory public
|
|
// key object we can work with for querying.
|
|
pubKeyBytes, err := hex.DecodeString(in.PubKey)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
pubKey, err := btcec.ParsePubKey(pubKeyBytes, btcec.S256())
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// With the public key decoded, attempt to fetch the node corresponding
|
|
// to this public key. If the node cannot be found, then an error will
|
|
// be returned.
|
|
node, err := graph.FetchLightningNode(pubKey)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// With the node obtained, we'll now iterate through all its out going
|
|
// edges to gather some basic statistics about its out going channels.
|
|
var (
|
|
numChannels uint32
|
|
totalCapacity btcutil.Amount
|
|
channels []*lnrpc.ChannelEdge
|
|
)
|
|
|
|
if err := node.ForEachChannel(nil, func(_ *bbolt.Tx,
|
|
edge *channeldb.ChannelEdgeInfo,
|
|
c1, c2 *channeldb.ChannelEdgePolicy) error {
|
|
|
|
numChannels++
|
|
totalCapacity += edge.Capacity
|
|
|
|
// Only populate the node's channels if the user requested them.
|
|
if in.IncludeChannels {
|
|
// Do not include unannounced channels - private
|
|
// channels or public channels whose authentication
|
|
// proof were not confirmed yet.
|
|
if edge.AuthProof == nil {
|
|
return nil
|
|
}
|
|
|
|
// Convert the database's edge format into the
|
|
// network/RPC edge format.
|
|
channelEdge := marshalDbEdge(edge, c1, c2)
|
|
channels = append(channels, channelEdge)
|
|
}
|
|
|
|
return nil
|
|
}); err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
nodeAddrs := make([]*lnrpc.NodeAddress, 0)
|
|
for _, addr := range node.Addresses {
|
|
nodeAddr := &lnrpc.NodeAddress{
|
|
Network: addr.Network(),
|
|
Addr: addr.String(),
|
|
}
|
|
nodeAddrs = append(nodeAddrs, nodeAddr)
|
|
}
|
|
|
|
return &lnrpc.NodeInfo{
|
|
Node: &lnrpc.LightningNode{
|
|
LastUpdate: uint32(node.LastUpdate.Unix()),
|
|
PubKey: in.PubKey,
|
|
Addresses: nodeAddrs,
|
|
Alias: node.Alias,
|
|
Color: routing.EncodeHexColor(node.Color),
|
|
},
|
|
NumChannels: numChannels,
|
|
TotalCapacity: int64(totalCapacity),
|
|
Channels: channels,
|
|
}, nil
|
|
}
|
|
|
|
// QueryRoutes attempts to query the daemons' Channel Router for a possible
|
|
// route to a target destination capable of carrying a specific amount of
|
|
// satoshis within the route's flow. The retuned route contains the full
|
|
// details required to craft and send an HTLC, also including the necessary
|
|
// information that should be present within the Sphinx packet encapsulated
|
|
// within the HTLC.
|
|
//
|
|
// TODO(roasbeef): should return a slice of routes in reality
|
|
// * create separate PR to send based on well formatted route
|
|
func (r *rpcServer) QueryRoutes(ctx context.Context,
|
|
in *lnrpc.QueryRoutesRequest) (*lnrpc.QueryRoutesResponse, error) {
|
|
|
|
return r.routerBackend.QueryRoutes(ctx, in)
|
|
}
|
|
|
|
// GetNetworkInfo returns some basic stats about the known channel graph from
|
|
// the PoV of the node.
|
|
func (r *rpcServer) GetNetworkInfo(ctx context.Context,
|
|
_ *lnrpc.NetworkInfoRequest) (*lnrpc.NetworkInfo, error) {
|
|
|
|
graph := r.server.chanDB.ChannelGraph()
|
|
|
|
var (
|
|
numNodes uint32
|
|
numChannels uint32
|
|
maxChanOut uint32
|
|
totalNetworkCapacity btcutil.Amount
|
|
minChannelSize btcutil.Amount = math.MaxInt64
|
|
maxChannelSize btcutil.Amount
|
|
medianChanSize btcutil.Amount
|
|
)
|
|
|
|
// We'll use this map to de-duplicate channels during our traversal.
|
|
// This is needed since channels are directional, so there will be two
|
|
// edges for each channel within the graph.
|
|
seenChans := make(map[uint64]struct{})
|
|
|
|
// We also keep a list of all encountered capacities, in order to
|
|
// calculate the median channel size.
|
|
var allChans []btcutil.Amount
|
|
|
|
// We'll run through all the known nodes in the within our view of the
|
|
// network, tallying up the total number of nodes, and also gathering
|
|
// each node so we can measure the graph diameter and degree stats
|
|
// below.
|
|
if err := graph.ForEachNode(nil, func(tx *bbolt.Tx, node *channeldb.LightningNode) error {
|
|
// Increment the total number of nodes with each iteration.
|
|
numNodes++
|
|
|
|
// For each channel we'll compute the out degree of each node,
|
|
// and also update our running tallies of the min/max channel
|
|
// capacity, as well as the total channel capacity. We pass
|
|
// through the db transaction from the outer view so we can
|
|
// re-use it within this inner view.
|
|
var outDegree uint32
|
|
if err := node.ForEachChannel(tx, func(_ *bbolt.Tx,
|
|
edge *channeldb.ChannelEdgeInfo, _, _ *channeldb.ChannelEdgePolicy) error {
|
|
|
|
// Bump up the out degree for this node for each
|
|
// channel encountered.
|
|
outDegree++
|
|
|
|
// If we've already seen this channel, then we'll
|
|
// return early to ensure that we don't double-count
|
|
// stats.
|
|
if _, ok := seenChans[edge.ChannelID]; ok {
|
|
return nil
|
|
}
|
|
|
|
// Compare the capacity of this channel against the
|
|
// running min/max to see if we should update the
|
|
// extrema.
|
|
chanCapacity := edge.Capacity
|
|
if chanCapacity < minChannelSize {
|
|
minChannelSize = chanCapacity
|
|
}
|
|
if chanCapacity > maxChannelSize {
|
|
maxChannelSize = chanCapacity
|
|
}
|
|
|
|
// Accumulate the total capacity of this channel to the
|
|
// network wide-capacity.
|
|
totalNetworkCapacity += chanCapacity
|
|
|
|
numChannels++
|
|
|
|
seenChans[edge.ChannelID] = struct{}{}
|
|
allChans = append(allChans, edge.Capacity)
|
|
return nil
|
|
}); err != nil {
|
|
return err
|
|
}
|
|
|
|
// Finally, if the out degree of this node is greater than what
|
|
// we've seen so far, update the maxChanOut variable.
|
|
if outDegree > maxChanOut {
|
|
maxChanOut = outDegree
|
|
}
|
|
|
|
return nil
|
|
}); err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Query the graph for the current number of zombie channels.
|
|
numZombies, err := graph.NumZombies()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Find the median.
|
|
medianChanSize = autopilot.Median(allChans)
|
|
|
|
// If we don't have any channels, then reset the minChannelSize to zero
|
|
// to avoid outputting NaN in encoded JSON.
|
|
if numChannels == 0 {
|
|
minChannelSize = 0
|
|
}
|
|
|
|
// TODO(roasbeef): graph diameter
|
|
|
|
// TODO(roasbeef): also add oldest channel?
|
|
netInfo := &lnrpc.NetworkInfo{
|
|
MaxOutDegree: maxChanOut,
|
|
AvgOutDegree: float64(2*numChannels) / float64(numNodes),
|
|
NumNodes: numNodes,
|
|
NumChannels: numChannels,
|
|
TotalNetworkCapacity: int64(totalNetworkCapacity),
|
|
AvgChannelSize: float64(totalNetworkCapacity) / float64(numChannels),
|
|
|
|
MinChannelSize: int64(minChannelSize),
|
|
MaxChannelSize: int64(maxChannelSize),
|
|
MedianChannelSizeSat: int64(medianChanSize),
|
|
NumZombieChans: numZombies,
|
|
}
|
|
|
|
// Similarly, if we don't have any channels, then we'll also set the
|
|
// average channel size to zero in order to avoid weird JSON encoding
|
|
// outputs.
|
|
if numChannels == 0 {
|
|
netInfo.AvgChannelSize = 0
|
|
}
|
|
|
|
return netInfo, nil
|
|
}
|
|
|
|
// StopDaemon will send a shutdown request to the interrupt handler, triggering
|
|
// a graceful shutdown of the daemon.
|
|
func (r *rpcServer) StopDaemon(ctx context.Context,
|
|
_ *lnrpc.StopRequest) (*lnrpc.StopResponse, error) {
|
|
|
|
signal.RequestShutdown()
|
|
return &lnrpc.StopResponse{}, nil
|
|
}
|
|
|
|
// SubscribeChannelGraph launches a streaming RPC that allows the caller to
|
|
// receive notifications upon any changes the channel graph topology from the
|
|
// review of the responding node. Events notified include: new nodes coming
|
|
// online, nodes updating their authenticated attributes, new channels being
|
|
// advertised, updates in the routing policy for a directional channel edge,
|
|
// and finally when prior channels are closed on-chain.
|
|
func (r *rpcServer) SubscribeChannelGraph(req *lnrpc.GraphTopologySubscription,
|
|
updateStream lnrpc.Lightning_SubscribeChannelGraphServer) error {
|
|
|
|
// First, we start by subscribing to a new intent to receive
|
|
// notifications from the channel router.
|
|
client, err := r.server.chanRouter.SubscribeTopology()
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
// Ensure that the resources for the topology update client is cleaned
|
|
// up once either the server, or client exists.
|
|
defer client.Cancel()
|
|
|
|
for {
|
|
select {
|
|
|
|
// A new update has been sent by the channel router, we'll
|
|
// marshal it into the form expected by the gRPC client, then
|
|
// send it off.
|
|
case topChange, ok := <-client.TopologyChanges:
|
|
// If the second value from the channel read is nil,
|
|
// then this means that the channel router is exiting
|
|
// or the notification client was canceled. So we'll
|
|
// exit early.
|
|
if !ok {
|
|
return errors.New("server shutting down")
|
|
}
|
|
|
|
// Convert the struct from the channel router into the
|
|
// form expected by the gRPC service then send it off
|
|
// to the client.
|
|
graphUpdate := marshallTopologyChange(topChange)
|
|
if err := updateStream.Send(graphUpdate); err != nil {
|
|
return err
|
|
}
|
|
|
|
// The server is quitting, so we'll exit immediately. Returning
|
|
// nil will close the clients read end of the stream.
|
|
case <-r.quit:
|
|
return nil
|
|
}
|
|
}
|
|
}
|
|
|
|
// marshallTopologyChange performs a mapping from the topology change struct
|
|
// returned by the router to the form of notifications expected by the current
|
|
// gRPC service.
|
|
func marshallTopologyChange(topChange *routing.TopologyChange) *lnrpc.GraphTopologyUpdate {
|
|
|
|
// encodeKey is a simple helper function that converts a live public
|
|
// key into a hex-encoded version of the compressed serialization for
|
|
// the public key.
|
|
encodeKey := func(k *btcec.PublicKey) string {
|
|
return hex.EncodeToString(k.SerializeCompressed())
|
|
}
|
|
|
|
nodeUpdates := make([]*lnrpc.NodeUpdate, len(topChange.NodeUpdates))
|
|
for i, nodeUpdate := range topChange.NodeUpdates {
|
|
addrs := make([]string, len(nodeUpdate.Addresses))
|
|
for i, addr := range nodeUpdate.Addresses {
|
|
addrs[i] = addr.String()
|
|
}
|
|
|
|
nodeUpdates[i] = &lnrpc.NodeUpdate{
|
|
Addresses: addrs,
|
|
IdentityKey: encodeKey(nodeUpdate.IdentityKey),
|
|
GlobalFeatures: nodeUpdate.GlobalFeatures,
|
|
Alias: nodeUpdate.Alias,
|
|
Color: nodeUpdate.Color,
|
|
}
|
|
}
|
|
|
|
channelUpdates := make([]*lnrpc.ChannelEdgeUpdate, len(topChange.ChannelEdgeUpdates))
|
|
for i, channelUpdate := range topChange.ChannelEdgeUpdates {
|
|
channelUpdates[i] = &lnrpc.ChannelEdgeUpdate{
|
|
ChanId: channelUpdate.ChanID,
|
|
ChanPoint: &lnrpc.ChannelPoint{
|
|
FundingTxid: &lnrpc.ChannelPoint_FundingTxidBytes{
|
|
FundingTxidBytes: channelUpdate.ChanPoint.Hash[:],
|
|
},
|
|
OutputIndex: channelUpdate.ChanPoint.Index,
|
|
},
|
|
Capacity: int64(channelUpdate.Capacity),
|
|
RoutingPolicy: &lnrpc.RoutingPolicy{
|
|
TimeLockDelta: uint32(channelUpdate.TimeLockDelta),
|
|
MinHtlc: int64(channelUpdate.MinHTLC),
|
|
MaxHtlcMsat: uint64(channelUpdate.MaxHTLC),
|
|
FeeBaseMsat: int64(channelUpdate.BaseFee),
|
|
FeeRateMilliMsat: int64(channelUpdate.FeeRate),
|
|
Disabled: channelUpdate.Disabled,
|
|
},
|
|
AdvertisingNode: encodeKey(channelUpdate.AdvertisingNode),
|
|
ConnectingNode: encodeKey(channelUpdate.ConnectingNode),
|
|
}
|
|
}
|
|
|
|
closedChans := make([]*lnrpc.ClosedChannelUpdate, len(topChange.ClosedChannels))
|
|
for i, closedChan := range topChange.ClosedChannels {
|
|
closedChans[i] = &lnrpc.ClosedChannelUpdate{
|
|
ChanId: closedChan.ChanID,
|
|
Capacity: int64(closedChan.Capacity),
|
|
ClosedHeight: closedChan.ClosedHeight,
|
|
ChanPoint: &lnrpc.ChannelPoint{
|
|
FundingTxid: &lnrpc.ChannelPoint_FundingTxidBytes{
|
|
FundingTxidBytes: closedChan.ChanPoint.Hash[:],
|
|
},
|
|
OutputIndex: closedChan.ChanPoint.Index,
|
|
},
|
|
}
|
|
}
|
|
|
|
return &lnrpc.GraphTopologyUpdate{
|
|
NodeUpdates: nodeUpdates,
|
|
ChannelUpdates: channelUpdates,
|
|
ClosedChans: closedChans,
|
|
}
|
|
}
|
|
|
|
// ListPayments returns a list of all outgoing payments.
|
|
func (r *rpcServer) ListPayments(ctx context.Context,
|
|
req *lnrpc.ListPaymentsRequest) (*lnrpc.ListPaymentsResponse, error) {
|
|
|
|
rpcsLog.Debugf("[ListPayments]")
|
|
|
|
payments, err := r.server.chanDB.FetchPayments()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
paymentsResp := &lnrpc.ListPaymentsResponse{}
|
|
for _, payment := range payments {
|
|
// To keep compatibility with the old API, we only return
|
|
// non-suceeded payments if requested.
|
|
if payment.Status != channeldb.StatusSucceeded &&
|
|
!req.IncludeIncomplete {
|
|
continue
|
|
}
|
|
|
|
// Fetch the payment's route and preimage. If no HTLC was
|
|
// successful, an empty route and preimage will be used.
|
|
var (
|
|
route route.Route
|
|
preimage lntypes.Preimage
|
|
)
|
|
for _, htlc := range payment.HTLCs {
|
|
// Display the last route attempted.
|
|
route = htlc.Route
|
|
|
|
// If any of the htlcs have settled, extract a valid
|
|
// preimage.
|
|
if htlc.Settle != nil {
|
|
preimage = htlc.Settle.Preimage
|
|
}
|
|
}
|
|
|
|
// Encode the hops from the successful route, if any.
|
|
path := make([]string, len(route.Hops))
|
|
for i, hop := range route.Hops {
|
|
path[i] = hex.EncodeToString(hop.PubKeyBytes[:])
|
|
}
|
|
|
|
msatValue := int64(payment.Info.Value)
|
|
satValue := int64(payment.Info.Value.ToSatoshis())
|
|
|
|
status, err := convertPaymentStatus(payment.Status)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
htlcs := make([]*lnrpc.HTLCAttempt, 0, len(payment.HTLCs))
|
|
for _, dbHTLC := range payment.HTLCs {
|
|
htlc, err := r.routerBackend.MarshalHTLCAttempt(dbHTLC)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
htlcs = append(htlcs, htlc)
|
|
}
|
|
|
|
paymentHash := payment.Info.PaymentHash
|
|
creationTimeNS := routerrpc.MarshalTimeNano(payment.Info.CreationTime)
|
|
paymentsResp.Payments = append(paymentsResp.Payments, &lnrpc.Payment{
|
|
PaymentHash: hex.EncodeToString(paymentHash[:]),
|
|
Value: satValue,
|
|
ValueMsat: msatValue,
|
|
ValueSat: satValue,
|
|
CreationDate: payment.Info.CreationTime.Unix(),
|
|
CreationTimeNs: creationTimeNS,
|
|
Path: path,
|
|
Fee: int64(route.TotalFees().ToSatoshis()),
|
|
FeeSat: int64(route.TotalFees().ToSatoshis()),
|
|
FeeMsat: int64(route.TotalFees()),
|
|
PaymentPreimage: hex.EncodeToString(preimage[:]),
|
|
PaymentRequest: string(payment.Info.PaymentRequest),
|
|
Status: status,
|
|
Htlcs: htlcs,
|
|
})
|
|
}
|
|
|
|
return paymentsResp, nil
|
|
}
|
|
|
|
// convertPaymentStatus converts a channeldb.PaymentStatus to the type expected
|
|
// by the RPC.
|
|
func convertPaymentStatus(dbStatus channeldb.PaymentStatus) (
|
|
lnrpc.Payment_PaymentStatus, error) {
|
|
|
|
switch dbStatus {
|
|
case channeldb.StatusUnknown:
|
|
return lnrpc.Payment_UNKNOWN, nil
|
|
|
|
case channeldb.StatusInFlight:
|
|
return lnrpc.Payment_IN_FLIGHT, nil
|
|
|
|
case channeldb.StatusSucceeded:
|
|
return lnrpc.Payment_SUCCEEDED, nil
|
|
|
|
case channeldb.StatusFailed:
|
|
return lnrpc.Payment_FAILED, nil
|
|
|
|
default:
|
|
return 0, fmt.Errorf("unhandled payment status %v", dbStatus)
|
|
}
|
|
}
|
|
|
|
// DeleteAllPayments deletes all outgoing payments from DB.
|
|
func (r *rpcServer) DeleteAllPayments(ctx context.Context,
|
|
_ *lnrpc.DeleteAllPaymentsRequest) (*lnrpc.DeleteAllPaymentsResponse, error) {
|
|
|
|
rpcsLog.Debugf("[DeleteAllPayments]")
|
|
|
|
if err := r.server.chanDB.DeletePayments(); err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
return &lnrpc.DeleteAllPaymentsResponse{}, nil
|
|
}
|
|
|
|
// DebugLevel allows a caller to programmatically set the logging verbosity of
|
|
// lnd. The logging can be targeted according to a coarse daemon-wide logging
|
|
// level, or in a granular fashion to specify the logging for a target
|
|
// sub-system.
|
|
func (r *rpcServer) DebugLevel(ctx context.Context,
|
|
req *lnrpc.DebugLevelRequest) (*lnrpc.DebugLevelResponse, error) {
|
|
|
|
// If show is set, then we simply print out the list of available
|
|
// sub-systems.
|
|
if req.Show {
|
|
return &lnrpc.DebugLevelResponse{
|
|
SubSystems: strings.Join(
|
|
logWriter.SupportedSubsystems(), " ",
|
|
),
|
|
}, nil
|
|
}
|
|
|
|
rpcsLog.Infof("[debuglevel] changing debug level to: %v", req.LevelSpec)
|
|
|
|
// Otherwise, we'll attempt to set the logging level using the
|
|
// specified level spec.
|
|
err := build.ParseAndSetDebugLevels(req.LevelSpec, logWriter)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
return &lnrpc.DebugLevelResponse{}, nil
|
|
}
|
|
|
|
// DecodePayReq takes an encoded payment request string and attempts to decode
|
|
// it, returning a full description of the conditions encoded within the
|
|
// payment request.
|
|
func (r *rpcServer) DecodePayReq(ctx context.Context,
|
|
req *lnrpc.PayReqString) (*lnrpc.PayReq, error) {
|
|
|
|
rpcsLog.Tracef("[decodepayreq] decoding: %v", req.PayReq)
|
|
|
|
// Fist we'll attempt to decode the payment request string, if the
|
|
// request is invalid or the checksum doesn't match, then we'll exit
|
|
// here with an error.
|
|
payReq, err := zpay32.Decode(req.PayReq, activeNetParams.Params)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// Let the fields default to empty strings.
|
|
desc := ""
|
|
if payReq.Description != nil {
|
|
desc = *payReq.Description
|
|
}
|
|
|
|
descHash := []byte("")
|
|
if payReq.DescriptionHash != nil {
|
|
descHash = payReq.DescriptionHash[:]
|
|
}
|
|
|
|
fallbackAddr := ""
|
|
if payReq.FallbackAddr != nil {
|
|
fallbackAddr = payReq.FallbackAddr.String()
|
|
}
|
|
|
|
// Expiry time will default to 3600 seconds if not specified
|
|
// explicitly.
|
|
expiry := int64(payReq.Expiry().Seconds())
|
|
|
|
// Convert between the `lnrpc` and `routing` types.
|
|
routeHints := invoicesrpc.CreateRPCRouteHints(payReq.RouteHints)
|
|
|
|
var amtSat, amtMsat int64
|
|
if payReq.MilliSat != nil {
|
|
amtSat = int64(payReq.MilliSat.ToSatoshis())
|
|
amtMsat = int64(*payReq.MilliSat)
|
|
}
|
|
|
|
// Extract the payment address from the payment request, if present.
|
|
var paymentAddr []byte
|
|
if payReq.PaymentAddr != nil {
|
|
paymentAddr = payReq.PaymentAddr[:]
|
|
}
|
|
|
|
dest := payReq.Destination.SerializeCompressed()
|
|
return &lnrpc.PayReq{
|
|
Destination: hex.EncodeToString(dest),
|
|
PaymentHash: hex.EncodeToString(payReq.PaymentHash[:]),
|
|
NumSatoshis: amtSat,
|
|
NumMsat: amtMsat,
|
|
Timestamp: payReq.Timestamp.Unix(),
|
|
Description: desc,
|
|
DescriptionHash: hex.EncodeToString(descHash[:]),
|
|
FallbackAddr: fallbackAddr,
|
|
Expiry: expiry,
|
|
CltvExpiry: int64(payReq.MinFinalCLTVExpiry()),
|
|
RouteHints: routeHints,
|
|
PaymentAddr: paymentAddr,
|
|
}, nil
|
|
}
|
|
|
|
// feeBase is the fixed point that fee rate computation are performed over.
|
|
// Nodes on the network advertise their fee rate using this point as a base.
|
|
// This means that the minimal possible fee rate if 1e-6, or 0.000001, or
|
|
// 0.0001%.
|
|
const feeBase = 1000000
|
|
|
|
// FeeReport allows the caller to obtain a report detailing the current fee
|
|
// schedule enforced by the node globally for each channel.
|
|
func (r *rpcServer) FeeReport(ctx context.Context,
|
|
_ *lnrpc.FeeReportRequest) (*lnrpc.FeeReportResponse, error) {
|
|
|
|
// TODO(roasbeef): use UnaryInterceptor to add automated logging
|
|
|
|
rpcsLog.Debugf("[feereport]")
|
|
|
|
channelGraph := r.server.chanDB.ChannelGraph()
|
|
selfNode, err := channelGraph.SourceNode()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
var feeReports []*lnrpc.ChannelFeeReport
|
|
err = selfNode.ForEachChannel(nil, func(_ *bbolt.Tx, chanInfo *channeldb.ChannelEdgeInfo,
|
|
edgePolicy, _ *channeldb.ChannelEdgePolicy) error {
|
|
|
|
// Self node should always have policies for its channels.
|
|
if edgePolicy == nil {
|
|
return fmt.Errorf("no policy for outgoing channel %v ",
|
|
chanInfo.ChannelID)
|
|
}
|
|
|
|
// We'll compute the effective fee rate by converting from a
|
|
// fixed point fee rate to a floating point fee rate. The fee
|
|
// rate field in the database the amount of mSAT charged per
|
|
// 1mil mSAT sent, so will divide by this to get the proper fee
|
|
// rate.
|
|
feeRateFixedPoint := edgePolicy.FeeProportionalMillionths
|
|
feeRate := float64(feeRateFixedPoint) / float64(feeBase)
|
|
|
|
// TODO(roasbeef): also add stats for revenue for each channel
|
|
feeReports = append(feeReports, &lnrpc.ChannelFeeReport{
|
|
ChanPoint: chanInfo.ChannelPoint.String(),
|
|
BaseFeeMsat: int64(edgePolicy.FeeBaseMSat),
|
|
FeePerMil: int64(feeRateFixedPoint),
|
|
FeeRate: feeRate,
|
|
})
|
|
|
|
return nil
|
|
})
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
fwdEventLog := r.server.chanDB.ForwardingLog()
|
|
|
|
// computeFeeSum is a helper function that computes the total fees for
|
|
// a particular time slice described by a forwarding event query.
|
|
computeFeeSum := func(query channeldb.ForwardingEventQuery) (lnwire.MilliSatoshi, error) {
|
|
|
|
var totalFees lnwire.MilliSatoshi
|
|
|
|
// We'll continue to fetch the next query and accumulate the
|
|
// fees until the next query returns no events.
|
|
for {
|
|
timeSlice, err := fwdEventLog.Query(query)
|
|
if err != nil {
|
|
return 0, err
|
|
}
|
|
|
|
// If the timeslice is empty, then we'll return as
|
|
// we've retrieved all the entries in this range.
|
|
if len(timeSlice.ForwardingEvents) == 0 {
|
|
break
|
|
}
|
|
|
|
// Otherwise, we'll tally up an accumulate the total
|
|
// fees for this time slice.
|
|
for _, event := range timeSlice.ForwardingEvents {
|
|
fee := event.AmtIn - event.AmtOut
|
|
totalFees += fee
|
|
}
|
|
|
|
// We'll now take the last offset index returned as
|
|
// part of this response, and modify our query to start
|
|
// at this index. This has a pagination effect in the
|
|
// case that our query bounds has more than 100k
|
|
// entries.
|
|
query.IndexOffset = timeSlice.LastIndexOffset
|
|
}
|
|
|
|
return totalFees, nil
|
|
}
|
|
|
|
now := time.Now()
|
|
|
|
// Before we perform the queries below, we'll instruct the switch to
|
|
// flush any pending events to disk. This ensure we get a complete
|
|
// snapshot at this particular time.
|
|
if err := r.server.htlcSwitch.FlushForwardingEvents(); err != nil {
|
|
return nil, fmt.Errorf("unable to flush forwarding "+
|
|
"events: %v", err)
|
|
}
|
|
|
|
// In addition to returning the current fee schedule for each channel.
|
|
// We'll also perform a series of queries to obtain the total fees
|
|
// earned over the past day, week, and month.
|
|
dayQuery := channeldb.ForwardingEventQuery{
|
|
StartTime: now.Add(-time.Hour * 24),
|
|
EndTime: now,
|
|
NumMaxEvents: 1000,
|
|
}
|
|
dayFees, err := computeFeeSum(dayQuery)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to retrieve day fees: %v", err)
|
|
}
|
|
|
|
weekQuery := channeldb.ForwardingEventQuery{
|
|
StartTime: now.Add(-time.Hour * 24 * 7),
|
|
EndTime: now,
|
|
NumMaxEvents: 1000,
|
|
}
|
|
weekFees, err := computeFeeSum(weekQuery)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to retrieve day fees: %v", err)
|
|
}
|
|
|
|
monthQuery := channeldb.ForwardingEventQuery{
|
|
StartTime: now.Add(-time.Hour * 24 * 30),
|
|
EndTime: now,
|
|
NumMaxEvents: 1000,
|
|
}
|
|
monthFees, err := computeFeeSum(monthQuery)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to retrieve day fees: %v", err)
|
|
}
|
|
|
|
return &lnrpc.FeeReportResponse{
|
|
ChannelFees: feeReports,
|
|
DayFeeSum: uint64(dayFees.ToSatoshis()),
|
|
WeekFeeSum: uint64(weekFees.ToSatoshis()),
|
|
MonthFeeSum: uint64(monthFees.ToSatoshis()),
|
|
}, nil
|
|
}
|
|
|
|
// minFeeRate is the smallest permitted fee rate within the network. This is
|
|
// derived by the fact that fee rates are computed using a fixed point of
|
|
// 1,000,000. As a result, the smallest representable fee rate is 1e-6, or
|
|
// 0.000001, or 0.0001%.
|
|
const minFeeRate = 1e-6
|
|
|
|
// UpdateChannelPolicy allows the caller to update the channel forwarding policy
|
|
// for all channels globally, or a particular channel.
|
|
func (r *rpcServer) UpdateChannelPolicy(ctx context.Context,
|
|
req *lnrpc.PolicyUpdateRequest) (*lnrpc.PolicyUpdateResponse, error) {
|
|
|
|
var targetChans []wire.OutPoint
|
|
switch scope := req.Scope.(type) {
|
|
// If the request is targeting all active channels, then we don't need
|
|
// target any channels by their channel point.
|
|
case *lnrpc.PolicyUpdateRequest_Global:
|
|
|
|
// Otherwise, we're targeting an individual channel by its channel
|
|
// point.
|
|
case *lnrpc.PolicyUpdateRequest_ChanPoint:
|
|
txid, err := GetChanPointFundingTxid(scope.ChanPoint)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
targetChans = append(targetChans, wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: scope.ChanPoint.OutputIndex,
|
|
})
|
|
default:
|
|
return nil, fmt.Errorf("unknown scope: %v", scope)
|
|
}
|
|
|
|
switch {
|
|
// As a sanity check, if the fee isn't zero, we'll ensure that the
|
|
// passed fee rate is below 1e-6, or the lowest allowed non-zero fee
|
|
// rate expressible within the protocol.
|
|
case req.FeeRate != 0 && req.FeeRate < minFeeRate:
|
|
return nil, fmt.Errorf("fee rate of %v is too small, min fee "+
|
|
"rate is %v", req.FeeRate, minFeeRate)
|
|
|
|
// We'll also ensure that the user isn't setting a CLTV delta that
|
|
// won't give outgoing HTLCs enough time to fully resolve if needed.
|
|
case req.TimeLockDelta < minTimeLockDelta:
|
|
return nil, fmt.Errorf("time lock delta of %v is too small, "+
|
|
"minimum supported is %v", req.TimeLockDelta,
|
|
minTimeLockDelta)
|
|
}
|
|
|
|
// We'll also need to convert the floating point fee rate we accept
|
|
// over RPC to the fixed point rate that we use within the protocol. We
|
|
// do this by multiplying the passed fee rate by the fee base. This
|
|
// gives us the fixed point, scaled by 1 million that's used within the
|
|
// protocol.
|
|
feeRateFixed := uint32(req.FeeRate * feeBase)
|
|
baseFeeMsat := lnwire.MilliSatoshi(req.BaseFeeMsat)
|
|
feeSchema := routing.FeeSchema{
|
|
BaseFee: baseFeeMsat,
|
|
FeeRate: feeRateFixed,
|
|
}
|
|
|
|
chanPolicy := routing.ChannelPolicy{
|
|
FeeSchema: feeSchema,
|
|
TimeLockDelta: req.TimeLockDelta,
|
|
MaxHTLC: lnwire.MilliSatoshi(req.MaxHtlcMsat),
|
|
}
|
|
|
|
rpcsLog.Debugf("[updatechanpolicy] updating channel policy base_fee=%v, "+
|
|
"rate_float=%v, rate_fixed=%v, time_lock_delta: %v, targets=%v",
|
|
req.BaseFeeMsat, req.FeeRate, feeRateFixed, req.TimeLockDelta,
|
|
spew.Sdump(targetChans))
|
|
|
|
// With the scope resolved, we'll now send this to the local channel
|
|
// manager so it can propagate the new policy for our target channel(s).
|
|
err := r.server.localChanMgr.UpdatePolicy(chanPolicy, targetChans...)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
return &lnrpc.PolicyUpdateResponse{}, nil
|
|
}
|
|
|
|
// ForwardingHistory allows the caller to query the htlcswitch for a record of
|
|
// all HTLC's forwarded within the target time range, and integer offset within
|
|
// that time range. If no time-range is specified, then the first chunk of the
|
|
// past 24 hrs of forwarding history are returned.
|
|
|
|
// A list of forwarding events are returned. The size of each forwarding event
|
|
// is 40 bytes, and the max message size able to be returned in gRPC is 4 MiB.
|
|
// In order to safely stay under this max limit, we'll return 50k events per
|
|
// response. Each response has the index offset of the last entry. The index
|
|
// offset can be provided to the request to allow the caller to skip a series
|
|
// of records.
|
|
func (r *rpcServer) ForwardingHistory(ctx context.Context,
|
|
req *lnrpc.ForwardingHistoryRequest) (*lnrpc.ForwardingHistoryResponse, error) {
|
|
|
|
rpcsLog.Debugf("[forwardinghistory]")
|
|
|
|
// Before we perform the queries below, we'll instruct the switch to
|
|
// flush any pending events to disk. This ensure we get a complete
|
|
// snapshot at this particular time.
|
|
if err := r.server.htlcSwitch.FlushForwardingEvents(); err != nil {
|
|
return nil, fmt.Errorf("unable to flush forwarding "+
|
|
"events: %v", err)
|
|
}
|
|
|
|
var (
|
|
startTime, endTime time.Time
|
|
|
|
numEvents uint32
|
|
)
|
|
|
|
// startTime defaults to the Unix epoch (0 unixtime, or midnight 01-01-1970).
|
|
startTime = time.Unix(int64(req.StartTime), 0)
|
|
|
|
// If the end time wasn't specified, assume a default end time of now.
|
|
if req.EndTime == 0 {
|
|
now := time.Now()
|
|
endTime = now
|
|
} else {
|
|
endTime = time.Unix(int64(req.EndTime), 0)
|
|
}
|
|
|
|
// If the number of events wasn't specified, then we'll default to
|
|
// returning the last 100 events.
|
|
numEvents = req.NumMaxEvents
|
|
if numEvents == 0 {
|
|
numEvents = 100
|
|
}
|
|
|
|
// Next, we'll map the proto request into a format that is understood by
|
|
// the forwarding log.
|
|
eventQuery := channeldb.ForwardingEventQuery{
|
|
StartTime: startTime,
|
|
EndTime: endTime,
|
|
IndexOffset: req.IndexOffset,
|
|
NumMaxEvents: numEvents,
|
|
}
|
|
timeSlice, err := r.server.chanDB.ForwardingLog().Query(eventQuery)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to query forwarding log: %v", err)
|
|
}
|
|
|
|
// TODO(roasbeef): add settlement latency?
|
|
// * use FPE on all records?
|
|
|
|
// With the events retrieved, we'll now map them into the proper proto
|
|
// response.
|
|
//
|
|
// TODO(roasbeef): show in ns for the outside?
|
|
resp := &lnrpc.ForwardingHistoryResponse{
|
|
ForwardingEvents: make([]*lnrpc.ForwardingEvent, len(timeSlice.ForwardingEvents)),
|
|
LastOffsetIndex: timeSlice.LastIndexOffset,
|
|
}
|
|
for i, event := range timeSlice.ForwardingEvents {
|
|
amtInMsat := event.AmtIn
|
|
amtOutMsat := event.AmtOut
|
|
feeMsat := event.AmtIn - event.AmtOut
|
|
|
|
resp.ForwardingEvents[i] = &lnrpc.ForwardingEvent{
|
|
Timestamp: uint64(event.Timestamp.Unix()),
|
|
ChanIdIn: event.IncomingChanID.ToUint64(),
|
|
ChanIdOut: event.OutgoingChanID.ToUint64(),
|
|
AmtIn: uint64(amtInMsat.ToSatoshis()),
|
|
AmtOut: uint64(amtOutMsat.ToSatoshis()),
|
|
Fee: uint64(feeMsat.ToSatoshis()),
|
|
FeeMsat: uint64(feeMsat),
|
|
AmtInMsat: uint64(amtInMsat),
|
|
AmtOutMsat: uint64(amtOutMsat),
|
|
}
|
|
}
|
|
|
|
return resp, nil
|
|
}
|
|
|
|
// ExportChannelBackup attempts to return an encrypted static channel backup
|
|
// for the target channel identified by it channel point. The backup is
|
|
// encrypted with a key generated from the aezeed seed of the user. The
|
|
// returned backup can either be restored using the RestoreChannelBackup method
|
|
// once lnd is running, or via the InitWallet and UnlockWallet methods from the
|
|
// WalletUnlocker service.
|
|
func (r *rpcServer) ExportChannelBackup(ctx context.Context,
|
|
in *lnrpc.ExportChannelBackupRequest) (*lnrpc.ChannelBackup, error) {
|
|
|
|
// First, we'll convert the lnrpc channel point into a wire.OutPoint
|
|
// that we can manipulate.
|
|
txid, err := GetChanPointFundingTxid(in.ChanPoint)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
chanPoint := wire.OutPoint{
|
|
Hash: *txid,
|
|
Index: in.ChanPoint.OutputIndex,
|
|
}
|
|
|
|
// Next, we'll attempt to fetch a channel backup for this channel from
|
|
// the database. If this channel has been closed, or the outpoint is
|
|
// unknown, then we'll return an error
|
|
unpackedBackup, err := chanbackup.FetchBackupForChan(
|
|
chanPoint, r.server.chanDB,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
|
|
// At this point, we have an unpacked backup (plaintext) so we'll now
|
|
// attempt to serialize and encrypt it in order to create a packed
|
|
// backup.
|
|
packedBackups, err := chanbackup.PackStaticChanBackups(
|
|
[]chanbackup.Single{*unpackedBackup},
|
|
r.server.cc.keyRing,
|
|
)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("packing of back ups failed: %v", err)
|
|
}
|
|
|
|
// Before we proceed, we'll ensure that we received a backup for this
|
|
// channel, otherwise, we'll bail out.
|
|
packedBackup, ok := packedBackups[chanPoint]
|
|
if !ok {
|
|
return nil, fmt.Errorf("expected single backup for "+
|
|
"ChannelPoint(%v), got %v", chanPoint,
|
|
len(packedBackup))
|
|
}
|
|
|
|
return &lnrpc.ChannelBackup{
|
|
ChanPoint: in.ChanPoint,
|
|
ChanBackup: packedBackup,
|
|
}, nil
|
|
}
|
|
|
|
// VerifyChanBackup allows a caller to verify the integrity of a channel backup
|
|
// snapshot. This method will accept both either a packed Single or a packed
|
|
// Multi. Specifying both will result in an error.
|
|
func (r *rpcServer) VerifyChanBackup(ctx context.Context,
|
|
in *lnrpc.ChanBackupSnapshot) (*lnrpc.VerifyChanBackupResponse, error) {
|
|
|
|
switch {
|
|
// If neither a Single or Multi has been specified, then we have nothing
|
|
// to verify.
|
|
case in.GetSingleChanBackups() == nil && in.GetMultiChanBackup() == nil:
|
|
return nil, errors.New("either a Single or Multi channel " +
|
|
"backup must be specified")
|
|
|
|
// Either a Single or a Multi must be specified, but not both.
|
|
case in.GetSingleChanBackups() != nil && in.GetMultiChanBackup() != nil:
|
|
return nil, errors.New("either a Single or Multi channel " +
|
|
"backup must be specified, but not both")
|
|
|
|
// If a Single is specified then we'll only accept one of them to allow
|
|
// the caller to map the valid/invalid state for each individual Single.
|
|
case in.GetSingleChanBackups() != nil:
|
|
chanBackupsProtos := in.GetSingleChanBackups().ChanBackups
|
|
if len(chanBackupsProtos) != 1 {
|
|
return nil, errors.New("only one Single is accepted " +
|
|
"at a time")
|
|
}
|
|
|
|
// First, we'll convert the raw byte slice into a type we can
|
|
// work with a bit better.
|
|
chanBackup := chanbackup.PackedSingles(
|
|
[][]byte{chanBackupsProtos[0].ChanBackup},
|
|
)
|
|
|
|
// With our PackedSingles created, we'll attempt to unpack the
|
|
// backup. If this fails, then we know the backup is invalid for
|
|
// some reason.
|
|
_, err := chanBackup.Unpack(r.server.cc.keyRing)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("invalid single channel "+
|
|
"backup: %v", err)
|
|
}
|
|
|
|
case in.GetMultiChanBackup() != nil:
|
|
// We'll convert the raw byte slice into a PackedMulti that we
|
|
// can easily work with.
|
|
packedMultiBackup := in.GetMultiChanBackup().MultiChanBackup
|
|
packedMulti := chanbackup.PackedMulti(packedMultiBackup)
|
|
|
|
// We'll now attempt to unpack the Multi. If this fails, then we
|
|
// know it's invalid.
|
|
_, err := packedMulti.Unpack(r.server.cc.keyRing)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("invalid multi channel backup: "+
|
|
"%v", err)
|
|
}
|
|
}
|
|
|
|
return &lnrpc.VerifyChanBackupResponse{}, nil
|
|
}
|
|
|
|
// createBackupSnapshot converts the passed Single backup into a snapshot which
|
|
// contains individual packed single backups, as well as a single packed multi
|
|
// backup.
|
|
func (r *rpcServer) createBackupSnapshot(backups []chanbackup.Single) (
|
|
*lnrpc.ChanBackupSnapshot, error) {
|
|
|
|
// Once we have the set of back ups, we'll attempt to pack them all
|
|
// into a series of single channel backups.
|
|
singleChanPackedBackups, err := chanbackup.PackStaticChanBackups(
|
|
backups, r.server.cc.keyRing,
|
|
)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to pack set of chan "+
|
|
"backups: %v", err)
|
|
}
|
|
|
|
// Now that we have our set of single packed backups, we'll morph that
|
|
// into a form that the proto response requires.
|
|
numBackups := len(singleChanPackedBackups)
|
|
singleBackupResp := &lnrpc.ChannelBackups{
|
|
ChanBackups: make([]*lnrpc.ChannelBackup, 0, numBackups),
|
|
}
|
|
for chanPoint, singlePackedBackup := range singleChanPackedBackups {
|
|
txid := chanPoint.Hash
|
|
rpcChanPoint := &lnrpc.ChannelPoint{
|
|
FundingTxid: &lnrpc.ChannelPoint_FundingTxidBytes{
|
|
FundingTxidBytes: txid[:],
|
|
},
|
|
OutputIndex: chanPoint.Index,
|
|
}
|
|
|
|
singleBackupResp.ChanBackups = append(
|
|
singleBackupResp.ChanBackups,
|
|
&lnrpc.ChannelBackup{
|
|
ChanPoint: rpcChanPoint,
|
|
ChanBackup: singlePackedBackup,
|
|
},
|
|
)
|
|
}
|
|
|
|
// In addition, to the set of single chan backups, we'll also create a
|
|
// single multi-channel backup which can be serialized into a single
|
|
// file for safe storage.
|
|
var b bytes.Buffer
|
|
unpackedMultiBackup := chanbackup.Multi{
|
|
StaticBackups: backups,
|
|
}
|
|
err = unpackedMultiBackup.PackToWriter(&b, r.server.cc.keyRing)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to multi-pack backups: %v", err)
|
|
}
|
|
|
|
multiBackupResp := &lnrpc.MultiChanBackup{
|
|
MultiChanBackup: b.Bytes(),
|
|
}
|
|
for _, singleBackup := range singleBackupResp.ChanBackups {
|
|
multiBackupResp.ChanPoints = append(
|
|
multiBackupResp.ChanPoints, singleBackup.ChanPoint,
|
|
)
|
|
}
|
|
|
|
return &lnrpc.ChanBackupSnapshot{
|
|
SingleChanBackups: singleBackupResp,
|
|
MultiChanBackup: multiBackupResp,
|
|
}, nil
|
|
}
|
|
|
|
// ExportAllChannelBackups returns static channel backups for all existing
|
|
// channels known to lnd. A set of regular singular static channel backups for
|
|
// each channel are returned. Additionally, a multi-channel backup is returned
|
|
// as well, which contains a single encrypted blob containing the backups of
|
|
// each channel.
|
|
func (r *rpcServer) ExportAllChannelBackups(ctx context.Context,
|
|
in *lnrpc.ChanBackupExportRequest) (*lnrpc.ChanBackupSnapshot, error) {
|
|
|
|
// First, we'll attempt to read back ups for ALL currently opened
|
|
// channels from disk.
|
|
allUnpackedBackups, err := chanbackup.FetchStaticChanBackups(
|
|
r.server.chanDB,
|
|
)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to fetch all static chan "+
|
|
"backups: %v", err)
|
|
}
|
|
|
|
// With the backups assembled, we'll create a full snapshot.
|
|
return r.createBackupSnapshot(allUnpackedBackups)
|
|
}
|
|
|
|
// RestoreChannelBackups accepts a set of singular channel backups, or a single
|
|
// encrypted multi-chan backup and attempts to recover any funds remaining
|
|
// within the channel. If we're able to unpack the backup, then the new channel
|
|
// will be shown under listchannels, as well as pending channels.
|
|
func (r *rpcServer) RestoreChannelBackups(ctx context.Context,
|
|
in *lnrpc.RestoreChanBackupRequest) (*lnrpc.RestoreBackupResponse, error) {
|
|
|
|
// First, we'll make our implementation of the
|
|
// chanbackup.ChannelRestorer interface which we'll use to properly
|
|
// restore either a set of chanbackup.Single or chanbackup.Multi
|
|
// backups.
|
|
chanRestorer := &chanDBRestorer{
|
|
db: r.server.chanDB,
|
|
secretKeys: r.server.cc.keyRing,
|
|
chainArb: r.server.chainArb,
|
|
}
|
|
|
|
// We'll accept either a list of Single backups, or a single Multi
|
|
// backup which contains several single backups.
|
|
switch {
|
|
case in.GetChanBackups() != nil:
|
|
chanBackupsProtos := in.GetChanBackups()
|
|
|
|
// Now that we know what type of backup we're working with,
|
|
// we'll parse them all out into a more suitable format.
|
|
packedBackups := make([][]byte, 0, len(chanBackupsProtos.ChanBackups))
|
|
for _, chanBackup := range chanBackupsProtos.ChanBackups {
|
|
packedBackups = append(
|
|
packedBackups, chanBackup.ChanBackup,
|
|
)
|
|
}
|
|
|
|
// With our backups obtained, we'll now restore them which will
|
|
// write the new backups to disk, and then attempt to connect
|
|
// out to any peers that we know of which were our prior
|
|
// channel peers.
|
|
err := chanbackup.UnpackAndRecoverSingles(
|
|
chanbackup.PackedSingles(packedBackups),
|
|
r.server.cc.keyRing, chanRestorer, r.server,
|
|
)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to unpack single "+
|
|
"backups: %v", err)
|
|
}
|
|
|
|
case in.GetMultiChanBackup() != nil:
|
|
packedMultiBackup := in.GetMultiChanBackup()
|
|
|
|
// With our backups obtained, we'll now restore them which will
|
|
// write the new backups to disk, and then attempt to connect
|
|
// out to any peers that we know of which were our prior
|
|
// channel peers.
|
|
packedMulti := chanbackup.PackedMulti(packedMultiBackup)
|
|
err := chanbackup.UnpackAndRecoverMulti(
|
|
packedMulti, r.server.cc.keyRing, chanRestorer,
|
|
r.server,
|
|
)
|
|
if err != nil {
|
|
return nil, fmt.Errorf("unable to unpack chan "+
|
|
"backup: %v", err)
|
|
}
|
|
}
|
|
|
|
return &lnrpc.RestoreBackupResponse{}, nil
|
|
}
|
|
|
|
// SubscribeChannelBackups allows a client to sub-subscribe to the most up to
|
|
// date information concerning the state of all channel back ups. Each time a
|
|
// new channel is added, we return the new set of channels, along with a
|
|
// multi-chan backup containing the backup info for all channels. Each time a
|
|
// channel is closed, we send a new update, which contains new new chan back
|
|
// ups, but the updated set of encrypted multi-chan backups with the closed
|
|
// channel(s) removed.
|
|
func (r *rpcServer) SubscribeChannelBackups(req *lnrpc.ChannelBackupSubscription,
|
|
updateStream lnrpc.Lightning_SubscribeChannelBackupsServer) error {
|
|
|
|
// First, we'll subscribe to the primary channel notifier so we can
|
|
// obtain events for new opened/closed channels.
|
|
chanSubscription, err := r.server.channelNotifier.SubscribeChannelEvents()
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
defer chanSubscription.Cancel()
|
|
for {
|
|
select {
|
|
// A new event has been sent by the channel notifier, we'll
|
|
// assemble, then sling out a new event to the client.
|
|
case e := <-chanSubscription.Updates():
|
|
// TODO(roasbeef): batch dispatch ntnfs
|
|
|
|
switch e.(type) {
|
|
|
|
// We only care about new/closed channels, so we'll
|
|
// skip any events for active/inactive channels.
|
|
case channelnotifier.ActiveChannelEvent:
|
|
continue
|
|
case channelnotifier.InactiveChannelEvent:
|
|
continue
|
|
}
|
|
|
|
// Now that we know the channel state has changed,
|
|
// we'll obtains the current set of single channel
|
|
// backups from disk.
|
|
chanBackups, err := chanbackup.FetchStaticChanBackups(
|
|
r.server.chanDB,
|
|
)
|
|
if err != nil {
|
|
return fmt.Errorf("unable to fetch all "+
|
|
"static chan backups: %v", err)
|
|
}
|
|
|
|
// With our backups obtained, we'll pack them into a
|
|
// snapshot and send them back to the client.
|
|
backupSnapshot, err := r.createBackupSnapshot(
|
|
chanBackups,
|
|
)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
err = updateStream.Send(backupSnapshot)
|
|
if err != nil {
|
|
return err
|
|
}
|
|
|
|
case <-r.quit:
|
|
return nil
|
|
}
|
|
}
|
|
}
|
|
|
|
// chanAcceptInfo is used in the ChannelAcceptor bidirectional stream and
|
|
// encapsulates the request information sent from the RPCAcceptor to the
|
|
// RPCServer.
|
|
type chanAcceptInfo struct {
|
|
chanReq *chanacceptor.ChannelAcceptRequest
|
|
responseChan chan bool
|
|
}
|
|
|
|
// ChannelAcceptor dispatches a bi-directional streaming RPC in which
|
|
// OpenChannel requests are sent to the client and the client responds with
|
|
// a boolean that tells LND whether or not to accept the channel. This allows
|
|
// node operators to specify their own criteria for accepting inbound channels
|
|
// through a single persistent connection.
|
|
func (r *rpcServer) ChannelAcceptor(stream lnrpc.Lightning_ChannelAcceptorServer) error {
|
|
chainedAcceptor := r.chanPredicate
|
|
|
|
// Create two channels to handle requests and responses respectively.
|
|
newRequests := make(chan *chanAcceptInfo)
|
|
responses := make(chan lnrpc.ChannelAcceptResponse)
|
|
|
|
// Define a quit channel that will be used to signal to the RPCAcceptor's
|
|
// closure whether the stream still exists.
|
|
quit := make(chan struct{})
|
|
defer close(quit)
|
|
|
|
// demultiplexReq is a closure that will be passed to the RPCAcceptor and
|
|
// acts as an intermediary between the RPCAcceptor and the RPCServer.
|
|
demultiplexReq := func(req *chanacceptor.ChannelAcceptRequest) bool {
|
|
respChan := make(chan bool, 1)
|
|
|
|
newRequest := &chanAcceptInfo{
|
|
chanReq: req,
|
|
responseChan: respChan,
|
|
}
|
|
|
|
// timeout is the time after which ChannelAcceptRequests expire.
|
|
timeout := time.After(defaultAcceptorTimeout)
|
|
|
|
// Send the request to the newRequests channel.
|
|
select {
|
|
case newRequests <- newRequest:
|
|
case <-timeout:
|
|
rpcsLog.Errorf("RPCAcceptor returned false - reached timeout of %d",
|
|
defaultAcceptorTimeout)
|
|
return false
|
|
case <-quit:
|
|
return false
|
|
case <-r.quit:
|
|
return false
|
|
}
|
|
|
|
// Receive the response and return it. If no response has been received
|
|
// in defaultAcceptorTimeout, then return false.
|
|
select {
|
|
case resp := <-respChan:
|
|
return resp
|
|
case <-timeout:
|
|
rpcsLog.Errorf("RPCAcceptor returned false - reached timeout of %d",
|
|
defaultAcceptorTimeout)
|
|
return false
|
|
case <-quit:
|
|
return false
|
|
case <-r.quit:
|
|
return false
|
|
}
|
|
}
|
|
|
|
// Create a new RPCAcceptor via the NewRPCAcceptor method.
|
|
rpcAcceptor := chanacceptor.NewRPCAcceptor(demultiplexReq)
|
|
|
|
// Add the RPCAcceptor to the ChainedAcceptor and defer its removal.
|
|
id := chainedAcceptor.AddAcceptor(rpcAcceptor)
|
|
defer chainedAcceptor.RemoveAcceptor(id)
|
|
|
|
// errChan is used by the receive loop to signal any errors that occur
|
|
// during reading from the stream. This is primarily used to shutdown the
|
|
// send loop in the case of an RPC client disconnecting.
|
|
errChan := make(chan error, 1)
|
|
|
|
// We need to have the stream.Recv() in a goroutine since the call is
|
|
// blocking and would prevent us from sending more ChannelAcceptRequests to
|
|
// the RPC client.
|
|
go func() {
|
|
for {
|
|
resp, err := stream.Recv()
|
|
if err != nil {
|
|
errChan <- err
|
|
return
|
|
}
|
|
|
|
var pendingID [32]byte
|
|
copy(pendingID[:], resp.PendingChanId)
|
|
|
|
openChanResp := lnrpc.ChannelAcceptResponse{
|
|
Accept: resp.Accept,
|
|
PendingChanId: pendingID[:],
|
|
}
|
|
|
|
// Now that we have the response from the RPC client, send it to
|
|
// the responses chan.
|
|
select {
|
|
case responses <- openChanResp:
|
|
case <-quit:
|
|
return
|
|
case <-r.quit:
|
|
return
|
|
}
|
|
}
|
|
}()
|
|
|
|
acceptRequests := make(map[[32]byte]chan bool)
|
|
|
|
for {
|
|
select {
|
|
case newRequest := <-newRequests:
|
|
|
|
req := newRequest.chanReq
|
|
pendingChanID := req.OpenChanMsg.PendingChannelID
|
|
|
|
acceptRequests[pendingChanID] = newRequest.responseChan
|
|
|
|
// A ChannelAcceptRequest has been received, send it to the client.
|
|
chanAcceptReq := &lnrpc.ChannelAcceptRequest{
|
|
NodePubkey: req.Node.SerializeCompressed(),
|
|
ChainHash: req.OpenChanMsg.ChainHash[:],
|
|
PendingChanId: req.OpenChanMsg.PendingChannelID[:],
|
|
FundingAmt: uint64(req.OpenChanMsg.FundingAmount),
|
|
PushAmt: uint64(req.OpenChanMsg.PushAmount),
|
|
DustLimit: uint64(req.OpenChanMsg.DustLimit),
|
|
MaxValueInFlight: uint64(req.OpenChanMsg.MaxValueInFlight),
|
|
ChannelReserve: uint64(req.OpenChanMsg.ChannelReserve),
|
|
MinHtlc: uint64(req.OpenChanMsg.HtlcMinimum),
|
|
FeePerKw: uint64(req.OpenChanMsg.FeePerKiloWeight),
|
|
CsvDelay: uint32(req.OpenChanMsg.CsvDelay),
|
|
MaxAcceptedHtlcs: uint32(req.OpenChanMsg.MaxAcceptedHTLCs),
|
|
ChannelFlags: uint32(req.OpenChanMsg.ChannelFlags),
|
|
}
|
|
|
|
if err := stream.Send(chanAcceptReq); err != nil {
|
|
return err
|
|
}
|
|
case resp := <-responses:
|
|
// Look up the appropriate channel to send on given the pending ID.
|
|
// If a channel is found, send the response over it.
|
|
var pendingID [32]byte
|
|
copy(pendingID[:], resp.PendingChanId)
|
|
respChan, ok := acceptRequests[pendingID]
|
|
if !ok {
|
|
continue
|
|
}
|
|
|
|
// Send the response boolean over the buffered response channel.
|
|
respChan <- resp.Accept
|
|
|
|
// Delete the channel from the acceptRequests map.
|
|
delete(acceptRequests, pendingID)
|
|
case err := <-errChan:
|
|
rpcsLog.Errorf("Received an error: %v, shutting down", err)
|
|
return err
|
|
case <-r.quit:
|
|
return fmt.Errorf("RPC server is shutting down")
|
|
}
|
|
}
|
|
}
|
|
|
|
// BakeMacaroon allows the creation of a new macaroon with custom read and write
|
|
// permissions. No first-party caveats are added since this can be done offline.
|
|
func (r *rpcServer) BakeMacaroon(ctx context.Context,
|
|
req *lnrpc.BakeMacaroonRequest) (*lnrpc.BakeMacaroonResponse, error) {
|
|
|
|
rpcsLog.Debugf("[bakemacaroon]")
|
|
|
|
// If the --no-macaroons flag is used to start lnd, the macaroon service
|
|
// is not initialized. Therefore we can't bake new macaroons.
|
|
if r.macService == nil {
|
|
return nil, fmt.Errorf("macaroon authentication disabled, " +
|
|
"remove --no-macaroons flag to enable")
|
|
}
|
|
|
|
helpMsg := fmt.Sprintf("supported actions are %v, supported entities "+
|
|
"are %v", validActions, validEntities)
|
|
|
|
// Don't allow empty permission list as it doesn't make sense to have
|
|
// a macaroon that is not allowed to access any RPC.
|
|
if len(req.Permissions) == 0 {
|
|
return nil, fmt.Errorf("permission list cannot be empty. "+
|
|
"specify at least one action/entity pair. %s", helpMsg)
|
|
}
|
|
|
|
// Validate and map permission struct used by gRPC to the one used by
|
|
// the bakery.
|
|
requestedPermissions := make([]bakery.Op, len(req.Permissions))
|
|
for idx, op := range req.Permissions {
|
|
if !stringInSlice(op.Action, validActions) {
|
|
return nil, fmt.Errorf("invalid permission action. %s",
|
|
helpMsg)
|
|
}
|
|
if !stringInSlice(op.Entity, validEntities) {
|
|
return nil, fmt.Errorf("invalid permission entity. %s",
|
|
helpMsg)
|
|
}
|
|
|
|
requestedPermissions[idx] = bakery.Op{
|
|
Entity: op.Entity,
|
|
Action: op.Action,
|
|
}
|
|
}
|
|
|
|
// Bake new macaroon with the given permissions and send it binary
|
|
// serialized and hex encoded to the client.
|
|
newMac, err := r.macService.Oven.NewMacaroon(
|
|
ctx, bakery.LatestVersion, nil, requestedPermissions...,
|
|
)
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
newMacBytes, err := newMac.M().MarshalBinary()
|
|
if err != nil {
|
|
return nil, err
|
|
}
|
|
resp := &lnrpc.BakeMacaroonResponse{}
|
|
resp.Macaroon = hex.EncodeToString(newMacBytes)
|
|
|
|
return resp, nil
|
|
}
|