lnd.xprv/routing/router.go

package routing

import (
	"bytes"
	"fmt"
	"runtime"
	"sort"
	"sync"
	"sync/atomic"
	"time"

	"github.com/coreos/bbolt"
	"github.com/davecgh/go-spew/spew"
	"github.com/lightningnetwork/lnd/channeldb"
	"github.com/lightningnetwork/lnd/htlcswitch"
	"github.com/lightningnetwork/lnd/lnwallet"
	"github.com/lightningnetwork/lnd/lnwire"
	"github.com/lightningnetwork/lnd/multimutex"
	"github.com/lightningnetwork/lnd/routing/chainview"
	"github.com/roasbeef/btcd/btcec"
	"github.com/roasbeef/btcd/wire"
	"github.com/roasbeef/btcutil"

	"crypto/sha256"

	"github.com/go-errors/errors"
	"github.com/lightningnetwork/lightning-onion"
)

const (
	// DefaultFinalCLTVDelta is the default value to be used as the final
	// CLTV delta for a route if one is unspecified.
	DefaultFinalCLTVDelta = 9

	// defaultPayAttemptTimeout is a duration that we'll use to determine
	// if we should give up on a payment attempt. This will be used if a
	// value isn't specified in the LightningNode struct.
	defaultPayAttemptTimeout = time.Duration(time.Second * 60)
)

// ChannelGraphSource represents the source of information about the topology
// of the lightning network. It's responsible for the addition of nodes, edges,
// applying edge updates, and returning the current block height with which the
// topology is synchronized.
type ChannelGraphSource interface {
	// AddNode is used to add information about a node to the router
	// database. If the node with this pubkey is not present in an existing
	// channel, it will be ignored.
	AddNode(node *channeldb.LightningNode) error

	// AddEdge is used to add edge/channel to the topology of the router,
	// after all information about channel will be gathered this
	// edge/channel might be used in construction of payment path.
	AddEdge(edge *channeldb.ChannelEdgeInfo) error

	// AddProof updates the channel edge info with proof which is needed to
	// properly announce the edge to the rest of the network.
	AddProof(chanID lnwire.ShortChannelID, proof *channeldb.ChannelAuthProof) error

	// UpdateEdge is used to update edge information, without this message
	// edge considered as not fully constructed.
	UpdateEdge(policy *channeldb.ChannelEdgePolicy) error

	// IsStaleNode returns true if the graph source has a node announcement
	// for the target node with a more recent timestamp. This method will
	// also return true if we don't have an active channel announcement for
	// the target node.
	IsStaleNode(node Vertex, timestamp time.Time) bool

	// IsKnownEdge returns true if the graph source already knows of the
	// passed channel ID.
	IsKnownEdge(chanID lnwire.ShortChannelID) bool

	// IsStaleEdgePolicy returns true if the graph source has a channel
	// edge for the passed channel ID (and flags) that have a more recent
	// timestamp.
	IsStaleEdgePolicy(chanID lnwire.ShortChannelID, timestamp time.Time,
		flags lnwire.ChanUpdateFlag) bool

	// ForAllOutgoingChannels is used to iterate over all channels
	// emanating from the "source" node which is the center of the
	// star-graph.
	ForAllOutgoingChannels(cb func(c *channeldb.ChannelEdgeInfo,
		e *channeldb.ChannelEdgePolicy) error) error

	// CurrentBlockHeight returns the block height from POV of the router
	// subsystem.
	CurrentBlockHeight() (uint32, error)

	// GetChannelByID return the channel by the channel id.
	GetChannelByID(chanID lnwire.ShortChannelID) (*channeldb.ChannelEdgeInfo,
		*channeldb.ChannelEdgePolicy, *channeldb.ChannelEdgePolicy, error)

	// ForEachNode is used to iterate over every node in the known graph.
	ForEachNode(func(node *channeldb.LightningNode) error) error

	// ForEachChannel is used to iterate over every channel in the known
	// graph.
	ForEachChannel(func(chanInfo *channeldb.ChannelEdgeInfo,
		e1, e2 *channeldb.ChannelEdgePolicy) error) error
}

// FeeSchema is the set fee configuration for a Lightning Node on the network.
// Using the coefficients described within the schema, the required fee to
// forward outgoing payments can be derived.
type FeeSchema struct {
	// BaseFee is the base amount of milli-satoshis that will be chained
	// for ANY payment forwarded.
	BaseFee lnwire.MilliSatoshi

	// FeeRate is the rate that will be charged for forwarding payments.
	// This value should be interpreted as the numerator for a fraction
	// (fixed point arithmetic) whose denominator is 1 million. As a result
	// the effective fee rate charged per mSAT will be: (amount *
	// FeeRate/1,000,000).
	FeeRate uint32
}

// ChannelPolicy holds the parameters that determine the policy we enforce
// when forwarding payments on a channel. These parameters are communicated
// to the rest of the network in ChannelUpdate messages.
type ChannelPolicy struct {
	// FeeSchema holds the fee configuration for a channel.
	FeeSchema

	// TimeLockDelta is the required HTLC timelock delta to be used
	// when forwarding payments.
	TimeLockDelta uint32
}

// Config defines the configuration for the ChannelRouter. ALL elements within
// the configuration MUST be non-nil for the ChannelRouter to carry out its
// duties.
type Config struct {
	// Graph is the channel graph that the ChannelRouter will use to gather
	// metrics from and also to carry out path finding queries.
	// TODO(roasbeef): make into an interface
	Graph *channeldb.ChannelGraph

	// Chain is the router's source to the most up-to-date blockchain data.
	// All incoming advertised channels will be checked against the chain
	// to ensure that the channels advertised are still open.
	Chain lnwallet.BlockChainIO

	// ChainView is an instance of a FilteredChainView which is used to
	// watch the sub-set of the UTXO set (the set of active channels) that
	// we need in order to properly maintain the channel graph.
	ChainView chainview.FilteredChainView

	// SendToSwitch is a function that directs a link-layer switch to
	// forward a fully encoded payment to the first hop in the route
	// denoted by its public key. A non-nil error is to be returned if the
	// payment was unsuccessful.
	SendToSwitch func(firstHop [33]byte, htlcAdd *lnwire.UpdateAddHTLC,
		circuit *sphinx.Circuit) ([sha256.Size]byte, error)

	// ChannelPruneExpiry is the duration used to determine if a channel
	// should be pruned or not. If the delta between now and when the
	// channel was last updated is greater than ChannelPruneExpiry, then
	// the channel is marked as a zombie channel eligible for pruning.
	ChannelPruneExpiry time.Duration

	// GraphPruneInterval is used as an interval to determine how often we
	// should examine the channel graph to garbage collect zombie channels.
	GraphPruneInterval time.Duration

	// QueryBandwidth is a method that allows the router to query the lower
	// link layer to determine the up to date available bandwidth at a
	// prospective link to be traversed. If the  link isn't available, then
	// a value of zero should be returned. Otherwise, the current up to
	// date knowledge of the available bandwidth of the link should be
	// returned.
	QueryBandwidth func(edge *channeldb.ChannelEdgeInfo) lnwire.MilliSatoshi
}

// routeTuple is an entry within the ChannelRouter's route cache. We cache
// prospective routes based on first the destination, and then the target
// amount. We required the target amount as that will influence the available
// set of paths for a payment.
type routeTuple struct {
	amt  lnwire.MilliSatoshi
	dest [33]byte
}

// newRouteTuple creates a new route tuple from the target and amount.
func newRouteTuple(amt lnwire.MilliSatoshi, dest []byte) routeTuple {
	r := routeTuple{
		amt: amt,
	}
	copy(r.dest[:], dest)

	return r
}

// ChannelRouter is the layer 3 router within the Lightning stack. Below the
// ChannelRouter is the HtlcSwitch, and below that is the Bitcoin blockchain
// itself. The primary role of the ChannelRouter is to respond to queries for
// potential routes that can support a payment amount, and also general graph
// reachability questions. The router will prune the channel graph
// automatically as new blocks are discovered which spend certain known funding
// outpoints, thereby closing their respective channels.
type ChannelRouter struct {
	ntfnClientCounter uint64 // To be used atomically.

	started uint32 // To be used atomically.
	stopped uint32 // To be used atomically.

	bestHeight uint32 // To be used atomically.

	// cfg is a copy of the configuration struct that the ChannelRouter was
	// initialized with.
	cfg *Config

	// selfNode is the center of the star-graph centered around the
	// ChannelRouter. The ChannelRouter uses this node as a starting point
	// when doing any path finding.
	selfNode *channeldb.LightningNode

	// routeCache is a map that caches the k-shortest paths from ourselves
	// to a given target destination for a particular payment amount. This
	// map is used as an optimization to speed up subsequent payments to a
	// particular destination. This map will be cleared each time a new
	// channel announcement is accepted, or a new block arrives that
	// results in channels being closed.
	//
	// TODO(roasbeef): make LRU
	routeCacheMtx sync.RWMutex
	routeCache    map[routeTuple][]*Route

	// newBlocks is a channel in which new blocks connected to the end of
	// the main chain are sent over, and blocks updated after a call to
	// UpdateFilter.
	newBlocks <-chan *chainview.FilteredBlock

	// staleBlocks is a channel in which blocks disconnected fromt the end
	// of our currently known best chain are sent over.
	staleBlocks <-chan *chainview.FilteredBlock

	// networkUpdates is a channel that carries new topology updates
	// messages from outside the ChannelRouter to be processed by the
	// networkHandler.
	networkUpdates chan *routingMsg

	// topologyClients maps a client's unique notification ID to a
	// topologyClient client that contains its notification dispatch
	// channel.
	topologyClients map[uint64]*topologyClient

	// ntfnClientUpdates is a channel that's used to send new updates to
	// topology notification clients to the ChannelRouter. Updates either
	// add a new notification client, or cancel notifications for an
	// existing client.
	ntfnClientUpdates chan *topologyClientUpdate

	// missionControl is a shared memory of sorts that executions of
	// payment path finding use in order to remember which vertexes/edges
	// were pruned from prior attempts. During SendPayment execution,
	// errors sent by nodes are mapped into a vertex or edge to be pruned.
	// Each run will then take into account this set of pruned
	// vertexes/edges to reduce route failure and pass on graph information
	// gained to the next execution.
	missionControl *missionControl

	// channelEdgeMtx is a mutex we use to make sure we process only one
	// ChannelEdgePolicy at a time for a given channelID, to ensure
	// consistency between the various database accesses.
	channelEdgeMtx *multimutex.Mutex

	rejectMtx   sync.RWMutex
	rejectCache map[uint64]struct{}

	sync.RWMutex

	quit chan struct{}
	wg   sync.WaitGroup
}

// A compile time check to ensure ChannelRouter implements the
// ChannelGraphSource interface.
var _ ChannelGraphSource = (*ChannelRouter)(nil)

// New creates a new instance of the ChannelRouter with the specified
// configuration parameters. As part of initialization, if the router detects
// that the channel graph isn't fully in sync with the latest UTXO (since the
// channel graph is a subset of the UTXO set) set, then the router will proceed
// to fully sync to the latest state of the UTXO set.
func New(cfg Config) (*ChannelRouter, error) {

	selfNode, err := cfg.Graph.SourceNode()
	if err != nil {
		return nil, err
	}

	r := &ChannelRouter{
		cfg:               &cfg,
		networkUpdates:    make(chan *routingMsg),
		topologyClients:   make(map[uint64]*topologyClient),
		ntfnClientUpdates: make(chan *topologyClientUpdate),
		channelEdgeMtx:    multimutex.NewMutex(),
		selfNode:          selfNode,
		routeCache:        make(map[routeTuple][]*Route),
		rejectCache:       make(map[uint64]struct{}),
		quit:              make(chan struct{}),
	}

	r.missionControl = newMissionControl(
		cfg.Graph, selfNode, cfg.QueryBandwidth,
	)

	return r, nil
}

// Start launches all the goroutines the ChannelRouter requires to carry out
// its duties. If the router has already been started, then this method is a
// noop.
func (r *ChannelRouter) Start() error {
	if !atomic.CompareAndSwapUint32(&r.started, 0, 1) {
		return nil
	}

	log.Tracef("Channel Router starting")

	// First, we'll start the chain view instance (if it isn't already
	// started).
	if err := r.cfg.ChainView.Start(); err != nil {
		return err
	}

	// Once the instance is active, we'll fetch the channel we'll receive
	// notifications over.
	r.newBlocks = r.cfg.ChainView.FilteredBlocks()
	r.staleBlocks = r.cfg.ChainView.DisconnectedBlocks()

	bestHash, bestHeight, err := r.cfg.Chain.GetBestBlock()
	if err != nil {
		return err
	}

	if _, _, err := r.cfg.Graph.PruneTip(); err != nil {
		switch {
		// If the graph has never been pruned, or hasn't fully been
		// created yet, then we don't treat this as an explicit error.
		case err == channeldb.ErrGraphNeverPruned:
			fallthrough
		case err == channeldb.ErrGraphNotFound:
			// If the graph has never been pruned, then we'll set
			// the prune height to the current best height of the
			// chain backend.
			_, err = r.cfg.Graph.PruneGraph(
				nil, bestHash, uint32(bestHeight),
			)
			if err != nil {
				return err
			}
		default:
			return err
		}
	}

	// Before we perform our manual block pruning, we'll construct and
	// apply a fresh chain filter to the active FilteredChainView instance.
	// We do this before, as otherwise we may miss on-chain events as the
	// filter hasn't properly been applied.
	channelView, err := r.cfg.Graph.ChannelView()
	if err != nil && err != channeldb.ErrGraphNoEdgesFound {
		return err
	}

	log.Infof("Filtering chain using %v channels active", len(channelView))
	if len(channelView) != 0 {
		err = r.cfg.ChainView.UpdateFilter(
			channelView, uint32(bestHeight),
		)
		if err != nil {
			return err
		}
	}

	// Before we begin normal operation of the router, we first need to
	// synchronize the channel graph to the latest state of the UTXO set.
	if err := r.syncGraphWithChain(); err != nil {
		return err
	}

	r.wg.Add(1)
	go r.networkHandler()

	return nil
}

// Stop signals the ChannelRouter to gracefully halt all routines. This method
// will *block* until all goroutines have excited. If the channel router has
// already stopped then this method will return immediately.
func (r *ChannelRouter) Stop() error {
	if !atomic.CompareAndSwapUint32(&r.stopped, 0, 1) {
		return nil
	}

	log.Infof("Channel Router shutting down")

	if err := r.cfg.ChainView.Stop(); err != nil {
		return err
	}

	close(r.quit)
	r.wg.Wait()

	return nil
}

// syncGraphWithChain attempts to synchronize the current channel graph with
// the latest UTXO set state. This process involves pruning from the channel
// graph any channels which have been closed by spending their funding output
// since we've been down.
func (r *ChannelRouter) syncGraphWithChain() error {
	// First, we'll need to check to see if we're already in sync with the
	// latest state of the UTXO set.
	bestHash, bestHeight, err := r.cfg.Chain.GetBestBlock()
	if err != nil {
		return err
	}
	r.bestHeight = uint32(bestHeight)

	pruneHash, pruneHeight, err := r.cfg.Graph.PruneTip()
	if err != nil {
		switch {
		// If the graph has never been pruned, or hasn't fully been
		// created yet, then we don't treat this as an explicit error.
		case err == channeldb.ErrGraphNeverPruned:
		case err == channeldb.ErrGraphNotFound:
		default:
			return err
		}
	}

	log.Infof("Prune tip for Channel Graph: height=%v, hash=%v", pruneHeight,
		pruneHash)

	switch {

	// If the graph has never been pruned, then we can exit early as this
	// entails it's being created for the first time and hasn't seen any
	// block or created channels.
	case pruneHeight == 0 || pruneHash == nil:
		return nil

	// If the block hashes and heights match exactly, then we don't need to
	// prune the channel graph as we're already fully in sync.
	case bestHash.IsEqual(pruneHash) && uint32(bestHeight) == pruneHeight:
		return nil
	}

	// If the main chain blockhash at prune height is different from the
	// prune hash, this might indicate the database is on a stale branch.
	mainBlockHash, err := r.cfg.Chain.GetBlockHash(int64(pruneHeight))
	if err != nil {
		return err
	}

	// While we are on a stale branch of the chain, walk backwards to find
	// first common block.
	for !pruneHash.IsEqual(mainBlockHash) {
		log.Infof("channel graph is stale. Disconnecting block %v "+
			"(hash=%v)", pruneHeight, pruneHash)
		// Prune the graph for every channel that was opened at height
		// >= pruneHeight.
		_, err := r.cfg.Graph.DisconnectBlockAtHeight(pruneHeight)
		if err != nil {
			return err
		}

		pruneHash, pruneHeight, err = r.cfg.Graph.PruneTip()
		if err != nil {
			switch {
			// If at this point the graph has never been pruned, we
			// can exit as this entails we are back to the point
			// where it hasn't seen any block or created channels,
			// alas there's nothing left to prune.
			case err == channeldb.ErrGraphNeverPruned:
				return nil
			case err == channeldb.ErrGraphNotFound:
				return nil
			default:
				return err
			}
		}
		mainBlockHash, err = r.cfg.Chain.GetBlockHash(int64(pruneHeight))
		if err != nil {
			return err
		}
	}

	log.Infof("Syncing channel graph from height=%v (hash=%v) to height=%v "+
		"(hash=%v)", pruneHeight, pruneHash, bestHeight, bestHash)

	// If we're not yet caught up, then we'll walk forward in the chain in
	// the chain pruning the channel graph with each new block in the chain
	// that hasn't yet been consumed by the channel graph.
	var numChansClosed uint32
	for nextHeight := pruneHeight + 1; nextHeight <= uint32(bestHeight); nextHeight++ {
		// Using the next height, request a manual block pruning from
		// the chainview for the particular block hash.
		nextHash, err := r.cfg.Chain.GetBlockHash(int64(nextHeight))
		if err != nil {
			return err
		}
		filterBlock, err := r.cfg.ChainView.FilterBlock(nextHash)
		if err != nil {
			return err
		}

		// We're only interested in all prior outputs that have been
		// spent in the block, so collate all the referenced previous
		// outpoints within each tx and input.
		var spentOutputs []*wire.OutPoint
		for _, tx := range filterBlock.Transactions {
			for _, txIn := range tx.TxIn {
				spentOutputs = append(spentOutputs,
					&txIn.PreviousOutPoint)
			}
		}

		// With the spent outputs gathered, attempt to prune the
		// channel graph, also passing in the hash+height of the block
		// being pruned so the prune tip can be updated.
		closedChans, err := r.cfg.Graph.PruneGraph(spentOutputs,
			nextHash,
			nextHeight)
		if err != nil {
			return err
		}

		numClosed := uint32(len(closedChans))
		log.Infof("Block %v (height=%v) closed %v channels",
			nextHash, nextHeight, numClosed)

		numChansClosed += numClosed
	}

	log.Infof("Graph pruning complete: %v channels were closed since "+
		"height %v", numChansClosed, pruneHeight)
	return nil
}

// pruneZombieChans is a method that will be called periodically to prune out
// any "zombie" channels. We consider channels zombies if *both* edges haven't
// been updated since our zombie horizon. We do this periodically to keep a
// health, lively routing table.
func (r *ChannelRouter) pruneZombieChans() error {
	var chansToPrune []wire.OutPoint
	chanExpiry := r.cfg.ChannelPruneExpiry

	log.Infof("Examining Channel Graph for zombie channels")

	// First, we'll collect all the channels which are eligible for garbage
	// collection due to being zombies.
	filterPruneChans := func(info *channeldb.ChannelEdgeInfo,
		e1, e2 *channeldb.ChannelEdgePolicy) error {

		// We'll ensure that we don't attempt to prune our *own*
		// channels from the graph, as in any case this should be
		// re-advertised by the sub-system above us.
		if info.NodeKey1Bytes == r.selfNode.PubKeyBytes ||
			info.NodeKey2Bytes == r.selfNode.PubKeyBytes {

			return nil
		}

		// If *both* edges haven't been updated for a period of
		// chanExpiry, then we'll mark the channel itself as eligible
		// for graph pruning.
		e1Zombie, e2Zombie := true, true
		if e1 != nil {
			e1Zombie = time.Since(e1.LastUpdate) >= chanExpiry
			if e1Zombie {
				log.Tracef("Edge #1 of ChannelPoint(%v) "+
					"last update: %v",
					info.ChannelPoint, e1.LastUpdate)
			}
		}
		if e2 != nil {
			e2Zombie = time.Since(e2.LastUpdate) >= chanExpiry
			if e2Zombie {
				log.Tracef("Edge #2 of ChannelPoint(%v) "+
					"last update: %v",
					info.ChannelPoint, e2.LastUpdate)
			}
		}
		if e1Zombie && e2Zombie {
			log.Debugf("ChannelPoint(%v) is a zombie, collecting "+
				"to prune", info.ChannelPoint)

			// TODO(roasbeef): add ability to delete single
			// directional edge
			chansToPrune = append(chansToPrune, info.ChannelPoint)

			// As we're detecting this as a zombie channel, we'll
			// add this to the set of recently rejected items so we
			// don't re-accept it shortly after.
			r.rejectCache[info.ChannelID] = struct{}{}
		}

		return nil
	}

	r.rejectMtx.Lock()
	defer r.rejectMtx.Unlock()

	err := r.cfg.Graph.ForEachChannel(filterPruneChans)
	if err != nil {
		return fmt.Errorf("Unable to filter local zombie "+
			"chans: %v", err)
	}

	log.Infof("Pruning %v Zombie Channels", len(chansToPrune))

	// With the set zombie-like channels obtained, we'll do another pass to
	// delete al zombie channels from the channel graph.
	for _, chanToPrune := range chansToPrune {
		log.Tracef("Pruning zombie chan ChannelPoint(%v)", chanToPrune)

		err := r.cfg.Graph.DeleteChannelEdge(&chanToPrune)
		if err != nil {
			return fmt.Errorf("Unable to prune zombie "+
				"chans: %v", err)
		}
	}

	return nil
}

// networkHandler is the primary goroutine for the ChannelRouter. The roles of
// this goroutine include answering queries related to the state of the
// network, pruning the graph on new block notification, applying network
// updates, and registering new topology clients.
//
// NOTE: This MUST be run as a goroutine.
func (r *ChannelRouter) networkHandler() {
	defer r.wg.Done()

	graphPruneTicker := time.NewTicker(r.cfg.GraphPruneInterval)
	defer graphPruneTicker.Stop()

	// We'll use this validation barrier to ensure that we process all jobs
	// in the proper order during parallel validation.
	validationBarrier := NewValidationBarrier(runtime.NumCPU()*4, r.quit)

	for {
		select {
		// A new fully validated network update has just arrived. As a
		// result we'll modify the channel graph accordingly depending
		// on the exact type of the message.
		case update := <-r.networkUpdates:
			// We'll set up any dependants, and wait until a free
			// slot for this job opens up, this allow us to not
			// have thousands of goroutines active.
			validationBarrier.InitJobDependencies(update.msg)

			r.wg.Add(1)
			go func() {
				defer r.wg.Done()
				defer validationBarrier.CompleteJob()

				// If this message has an existing dependency,
				// then we'll wait until that has been fully
				// validated before we proceed.
				err := validationBarrier.WaitForDependants(
					update.msg,
				)
				if err != nil {
					if err != ErrVBarrierShuttingDown {
						log.Warnf("unexpected error "+
							"during validation "+
							"barrier shutdown: %v",
							err)
					}
					return
				}

				// Process the routing update to determine if
				// this is either a new update from our PoV or
				// an update to a prior vertex/edge we
				// previously accepted.
				err = r.processUpdate(update.msg)
				update.err <- err

				// If this message had any dependencies, then
				// we can now signal them to continue.
				validationBarrier.SignalDependants(update.msg)
				if err != nil {
					return
				}

				// Send off a new notification for the newly
				// accepted update.
				topChange := &TopologyChange{}
				err = addToTopologyChange(
					r.cfg.Graph, topChange, update.msg,
				)
				if err != nil {
					log.Errorf("unable to update topology "+
						"change notification: %v", err)
					return
				}

				if !topChange.isEmpty() {
					r.notifyTopologyChange(topChange)
				}
			}()

			// TODO(roasbeef): remove all unconnected vertexes
			// after N blocks pass with no corresponding
			// announcements.

		case chainUpdate, ok := <-r.staleBlocks:
			// If the channel has been closed, then this indicates
			// the daemon is shutting down, so we exit ourselves.
			if !ok {
				return
			}

			// Since this block is stale, we update our best height
			// to the previous block.
			blockHeight := uint32(chainUpdate.Height)
			atomic.StoreUint32(&r.bestHeight, blockHeight-1)

			// Update the channel graph to reflect that this block
			// was disconnected.
			_, err := r.cfg.Graph.DisconnectBlockAtHeight(blockHeight)
			if err != nil {
				log.Errorf("unable to prune graph with stale "+
					"block: %v", err)
				continue
			}

			// Invalidate the route cache, as some channels might
			// not be confirmed anymore.
			r.routeCacheMtx.Lock()
			r.routeCache = make(map[routeTuple][]*Route)
			r.routeCacheMtx.Unlock()

			// TODO(halseth): notify client about the reorg?

		// A new block has arrived, so we can prune the channel graph
		// of any channels which were closed in the block.
		case chainUpdate, ok := <-r.newBlocks:
			// If the channel has been closed, then this indicates
			// the daemon is shutting down, so we exit ourselves.
			if !ok {
				return
			}

			// We'll ensure that any new blocks received attach
			// directly to the end of our main chain. If not, then
			// we've somehow missed some blocks. We don't process
			// this block as otherwise, we may miss on-chain
			// events.
			currentHeight := atomic.LoadUint32(&r.bestHeight)
			if chainUpdate.Height != currentHeight+1 {
				log.Errorf("out of order block: expecting "+
					"height=%v, got height=%v", currentHeight+1,
					chainUpdate.Height)
				continue
			}

			// Once a new block arrives, we update our running
			// track of the height of the chain tip.
			blockHeight := uint32(chainUpdate.Height)
			atomic.StoreUint32(&r.bestHeight, blockHeight)
			log.Infof("Pruning channel graph using block %v (height=%v)",
				chainUpdate.Hash, blockHeight)

			// We're only interested in all prior outputs that have
			// been spent in the block, so collate all the
			// referenced previous outpoints within each tx and
			// input.
			var spentOutputs []*wire.OutPoint
			for _, tx := range chainUpdate.Transactions {
				for _, txIn := range tx.TxIn {
					spentOutputs = append(spentOutputs,
						&txIn.PreviousOutPoint)
				}
			}

			// With the spent outputs gathered, attempt to prune
			// the channel graph, also passing in the hash+height
			// of the block being pruned so the prune tip can be
			// updated.
			chansClosed, err := r.cfg.Graph.PruneGraph(spentOutputs,
				&chainUpdate.Hash, chainUpdate.Height)
			if err != nil {
				log.Errorf("unable to prune routing table: %v", err)
				continue
			}

			log.Infof("Block %v (height=%v) closed %v channels",
				chainUpdate.Hash, blockHeight, len(chansClosed))

			// Invalidate the route cache as the block height has
			// changed which will invalidate the HTLC timeouts we
			// have crafted within each of the pre-computed routes.
			//
			// TODO(roasbeef): need to invalidate after each
			// chan ann update?
			//  * can have map of chanID to routes involved, avoids
			//    full invalidation
			r.routeCacheMtx.Lock()
			r.routeCache = make(map[routeTuple][]*Route)
			r.routeCacheMtx.Unlock()

			if len(chansClosed) == 0 {
				continue
			}

			// Notify all currently registered clients of the newly
			// closed channels.
			closeSummaries := createCloseSummaries(blockHeight, chansClosed...)
			r.notifyTopologyChange(&TopologyChange{
				ClosedChannels: closeSummaries,
			})

		// A new notification client update has arrived. We're either
		// gaining a new client, or cancelling notifications for an
		// existing client.
		case ntfnUpdate := <-r.ntfnClientUpdates:
			clientID := ntfnUpdate.clientID

			if ntfnUpdate.cancel {
				r.RLock()
				client, ok := r.topologyClients[ntfnUpdate.clientID]
				r.RUnlock()
				if ok {
					r.Lock()
					delete(r.topologyClients, clientID)
					r.Unlock()

					close(client.exit)
					client.wg.Wait()

					close(client.ntfnChan)
				}

				continue
			}

			r.Lock()
			r.topologyClients[ntfnUpdate.clientID] = &topologyClient{
				ntfnChan: ntfnUpdate.ntfnChan,
				exit:     make(chan struct{}),
			}
			r.Unlock()

		// The graph prune ticker has ticked, so we'll examine the
		// state of the known graph to filter out any zombie channels
		// for pruning.
		case <-graphPruneTicker.C:
			if err := r.pruneZombieChans(); err != nil {
				log.Errorf("unable to prune zombies: %v", err)
			}

		// The router has been signalled to exit, to we exit our main
		// loop so the wait group can be decremented.
		case <-r.quit:
			return
		}
	}
}

// assertNodeAnnFreshness returns a non-nil error if we have an announcement in
// the database for the passed node with a timestamp newer than the passed
// timestamp. ErrIgnored will be returned if we already have the node, and
// ErrOutdated will be returned if we have a timestamp that's after the new
// timestamp.
func (r *ChannelRouter) assertNodeAnnFreshness(node Vertex,
	msgTimestamp time.Time) error {

	// If we are not already aware of this node, it means that we don't
	// know about any channel using this node. To avoid a DoS attack by
	// node announcements, we will ignore such nodes. If we do know about
	// this node, check that this update brings info newer than what we
	// already have.
	lastUpdate, exists, err := r.cfg.Graph.HasLightningNode(node)
	if err != nil {
		return errors.Errorf("unable to query for the "+
			"existence of node: %v", err)
	}
	if !exists {
		return newErrf(ErrIgnored, "Ignoring node announcement"+
			" for node not found in channel graph (%x)",
			node[:])
	}

	// If we've reached this point then we're aware of the vertex being
	// advertised. So we now check if the new message has a new time stamp,
	// if not then we won't accept the new data as it would override newer
	// data.
	if !lastUpdate.Before(msgTimestamp) {
		return newErrf(ErrOutdated, "Ignoring outdated "+
			"announcement for %x", node[:])
	}

	return nil
}

// processUpdate processes a new relate authenticated channel/edge, node or
// channel/edge update network update. If the update didn't affect the internal
// state of the draft due to either being out of date, invalid, or redundant,
// then error is returned.
func (r *ChannelRouter) processUpdate(msg interface{}) error {

	var invalidateCache bool

	switch msg := msg.(type) {
	case *channeldb.LightningNode:
		// Before we add the node to the database, we'll check to see
		// if the announcement is "fresh" or not. If it isn't, then
		// we'll return an error.
		err := r.assertNodeAnnFreshness(msg.PubKeyBytes, msg.LastUpdate)
		if err != nil {
			return err
		}

		if err := r.cfg.Graph.AddLightningNode(msg); err != nil {
			return errors.Errorf("unable to add node %v to the "+
				"graph: %v", msg.PubKeyBytes, err)
		}

		log.Infof("Updated vertex data for node=%x", msg.PubKeyBytes)

	case *channeldb.ChannelEdgeInfo:
		// If we recently rejected this channel edge, then we won't
		// attempt to re-process it.
		r.rejectMtx.RLock()
		if _, ok := r.rejectCache[msg.ChannelID]; ok {
			r.rejectMtx.RUnlock()
			return newErrf(ErrIgnored, "recently rejected "+
				"chan_id=%v", msg.ChannelID)
		}
		r.rejectMtx.RUnlock()

		// Prior to processing the announcement we first check if we
		// already know of this channel, if so, then we can exit early.
		_, _, exists, err := r.cfg.Graph.HasChannelEdge(msg.ChannelID)
		if err != nil && err != channeldb.ErrGraphNoEdgesFound {
			return errors.Errorf("unable to check for edge "+
				"existence: %v", err)
		} else if exists {
			return newErrf(ErrIgnored, "Ignoring msg for known "+
				"chan_id=%v", msg.ChannelID)
		}

		// Query the database for the existence of the two nodes in this
		// channel. If not found, add a partial node to the database,
		// containing only the node keys.
		_, exists, _ = r.cfg.Graph.HasLightningNode(msg.NodeKey1Bytes)
		if !exists {
			node1 := &channeldb.LightningNode{
				PubKeyBytes:          msg.NodeKey1Bytes,
				HaveNodeAnnouncement: false,
			}
			err := r.cfg.Graph.AddLightningNode(node1)
			if err != nil {
				return errors.Errorf("unable to add node %v to"+
					" the graph: %v", node1.PubKeyBytes, err)
			}
		}
		_, exists, _ = r.cfg.Graph.HasLightningNode(msg.NodeKey2Bytes)
		if !exists {
			node2 := &channeldb.LightningNode{
				PubKeyBytes:          msg.NodeKey2Bytes,
				HaveNodeAnnouncement: false,
			}
			err := r.cfg.Graph.AddLightningNode(node2)
			if err != nil {
				return errors.Errorf("unable to add node %v to"+
					" the graph: %v", node2.PubKeyBytes, err)
			}
		}

		// Before we can add the channel to the channel graph, we need
		// to obtain the full funding outpoint that's encoded within
		// the channel ID.
		channelID := lnwire.NewShortChanIDFromInt(msg.ChannelID)
		fundingPoint, err := r.fetchChanPoint(&channelID)
		if err != nil {
			r.rejectMtx.Lock()
			r.rejectCache[msg.ChannelID] = struct{}{}
			r.rejectMtx.Unlock()

			return errors.Errorf("unable to fetch chan point for "+
				"chan_id=%v: %v", msg.ChannelID, err)
		}

		// Now that we have the funding outpoint of the channel, ensure
		// that it hasn't yet been spent. If so, then this channel has
		// been closed so we'll ignore it.
		chanUtxo, err := r.cfg.Chain.GetUtxo(
			fundingPoint, channelID.BlockHeight,
		)
		if err != nil {
			r.rejectMtx.Lock()
			r.rejectCache[msg.ChannelID] = struct{}{}
			r.rejectMtx.Unlock()

			return errors.Errorf("unable to fetch utxo for "+
				"chan_id=%v, chan_point=%v: %v", msg.ChannelID,
				fundingPoint, err)
		}

		// Recreate witness output to be sure that declared in channel
		// edge bitcoin keys and channel value corresponds to the
		// reality.
		_, witnessOutput, err := lnwallet.GenFundingPkScript(
			msg.BitcoinKey1Bytes[:], msg.BitcoinKey2Bytes[:],
			chanUtxo.Value,
		)
		if err != nil {
			return errors.Errorf("unable to create funding pk "+
				"script: %v", err)
		}

		// By checking the equality of witness pkscripts we checks that
		// funding witness script is multisignature lock which contains
		// both local and remote public keys which was declared in
		// channel edge and also that the announced channel value is
		// right.
		if !bytes.Equal(witnessOutput.PkScript, chanUtxo.PkScript) {
			return errors.Errorf("pkScript mismatch: expected %x, "+
				"got %x", witnessOutput.PkScript, chanUtxo.PkScript)
		}

		// TODO(roasbeef): this is a hack, needs to be removed
		// after commitment fees are dynamic.
		msg.Capacity = btcutil.Amount(chanUtxo.Value)
		msg.ChannelPoint = *fundingPoint
		if err := r.cfg.Graph.AddChannelEdge(msg); err != nil {
			return errors.Errorf("unable to add edge: %v", err)
		}

		invalidateCache = true
		log.Infof("New channel discovered! Link "+
			"connects %x and %x with ChannelPoint(%v): "+
			"chan_id=%v, capacity=%v",
			msg.NodeKey1Bytes, msg.NodeKey2Bytes,
			fundingPoint, msg.ChannelID, msg.Capacity)

		// As a new edge has been added to the channel graph, we'll
		// update the current UTXO filter within our active
		// FilteredChainView so we are notified if/when this channel is
		// closed.
		filterUpdate := []wire.OutPoint{*fundingPoint}
		err = r.cfg.ChainView.UpdateFilter(
			filterUpdate, atomic.LoadUint32(&r.bestHeight),
		)
		if err != nil {
			return errors.Errorf("unable to update chain "+
				"view: %v", err)
		}

	case *channeldb.ChannelEdgePolicy:
		// If we recently rejected this channel edge, then we won't
		// attempt to re-process it.
		r.rejectMtx.RLock()
		if _, ok := r.rejectCache[msg.ChannelID]; ok {
			r.rejectMtx.RUnlock()
			return newErrf(ErrIgnored, "recently rejected "+
				"chan_id=%v", msg.ChannelID)
		}
		r.rejectMtx.RUnlock()

		channelID := lnwire.NewShortChanIDFromInt(msg.ChannelID)

		// We make sure to hold the mutex for this channel ID,
		// such that no other goroutine is concurrently doing
		// database accesses for the same channel ID.
		r.channelEdgeMtx.Lock(msg.ChannelID)
		defer r.channelEdgeMtx.Unlock(msg.ChannelID)

		edge1Timestamp, edge2Timestamp, exists, err := r.cfg.Graph.HasChannelEdge(
			msg.ChannelID,
		)
		if err != nil && err != channeldb.ErrGraphNoEdgesFound {
			return errors.Errorf("unable to check for edge "+
				"existence: %v", err)

		}

		// As edges are directional edge node has a unique policy for
		// the direction of the edge they control. Therefore we first
		// check if we already have the most up to date information for
		// that edge. If so, then we can exit early.
		switch {

		// A flag set of 0 indicates this is an announcement for the
		// "first" node in the channel.
		case msg.Flags&lnwire.ChanUpdateDirection == 0:
			if edge1Timestamp.After(msg.LastUpdate) ||
				edge1Timestamp.Equal(msg.LastUpdate) {

				return newErrf(ErrIgnored, "Ignoring update "+
					"(flags=%v) for known chan_id=%v", msg.Flags,
					msg.ChannelID)

			}

		// Similarly, a flag set of 1 indicates this is an announcement
		// for the "second" node in the channel.
		case msg.Flags&lnwire.ChanUpdateDirection == 1:
			if edge2Timestamp.After(msg.LastUpdate) ||
				edge2Timestamp.Equal(msg.LastUpdate) {

				return newErrf(ErrIgnored, "Ignoring update "+
					"(flags=%v) for known chan_id=%v", msg.Flags,
					msg.ChannelID)
			}
		}

		if !exists {
			// Before we can update the channel information, we'll
			// ensure that the target channel is still open by
			// querying the utxo-set for its existence.
			chanPoint, err := r.fetchChanPoint(&channelID)
			if err != nil {
				r.rejectMtx.Lock()
				r.rejectCache[msg.ChannelID] = struct{}{}
				r.rejectMtx.Unlock()

				return errors.Errorf("unable to fetch chan "+
					"point for chan_id=%v: %v",
					msg.ChannelID, err)
			}
			_, err = r.cfg.Chain.GetUtxo(
				chanPoint, channelID.BlockHeight,
			)
			if err != nil {
				r.rejectMtx.Lock()
				r.rejectCache[msg.ChannelID] = struct{}{}
				r.rejectMtx.Unlock()

				return errors.Errorf("unable to fetch utxo for "+
					"chan_id=%v: %v", msg.ChannelID, err)
			}
		}

		// Now that we know this isn't a stale update, we'll apply the
		// new edge policy to the proper directional edge within the
		// channel graph.
		if err = r.cfg.Graph.UpdateEdgePolicy(msg); err != nil {
			err := errors.Errorf("unable to add channel: %v", err)
			log.Error(err)
			return err
		}

		invalidateCache = true
		log.Debugf("New channel update applied: %v", spew.Sdump(msg))

	default:
		return errors.Errorf("wrong routing update message type")
	}

	// If we've received a channel update, then invalidate the route cache
	// as channels within the graph have closed, which may affect our
	// choice of the KSP's for a particular routeTuple.
	if invalidateCache {
		r.routeCacheMtx.Lock()
		r.routeCache = make(map[routeTuple][]*Route)
		r.routeCacheMtx.Unlock()
	}

	return nil
}

// fetchChanPoint retrieves the original outpoint which is encoded within the
// channelID.
//
// TODO(roasbeef): replace with call to GetBlockTransaction? (would allow to
// later use getblocktxn)
func (r *ChannelRouter) fetchChanPoint(chanID *lnwire.ShortChannelID) (*wire.OutPoint, error) {
	// First fetch the block hash by the block number encoded, then use
	// that hash to fetch the block itself.
	blockNum := int64(chanID.BlockHeight)
	blockHash, err := r.cfg.Chain.GetBlockHash(blockNum)
	if err != nil {
		return nil, err
	}
	fundingBlock, err := r.cfg.Chain.GetBlock(blockHash)
	if err != nil {
		return nil, err
	}

	// As a sanity check, ensure that the advertised transaction index is
	// within the bounds of the total number of transactions within a
	// block.
	numTxns := uint32(len(fundingBlock.Transactions))
	if chanID.TxIndex > numTxns-1 {
		return nil, fmt.Errorf("tx_index=#%v is out of range "+
			"(max_index=%v), network_chan_id=%v\n", chanID.TxIndex,
			numTxns-1, spew.Sdump(chanID))
	}

	// Finally once we have the block itself, we seek to the targeted
	// transaction index to obtain the funding output and txid.
	fundingTx := fundingBlock.Transactions[chanID.TxIndex]
	return &wire.OutPoint{
		Hash:  fundingTx.TxHash(),
		Index: uint32(chanID.TxPosition),
	}, nil
}

// routingMsg couples a routing related routing topology update to the
// error channel.
type routingMsg struct {
	msg interface{}
	err chan error
}

// pruneNodeFromRoutes accepts set of routes, and returns a new set of routes
// with the target node filtered out.
func pruneNodeFromRoutes(routes []*Route, skipNode Vertex) []*Route {

	// TODO(roasbeef): pass in slice index?

	prunedRoutes := make([]*Route, 0, len(routes))
	for _, route := range routes {
		if route.containsNode(skipNode) {
			continue
		}

		prunedRoutes = append(prunedRoutes, route)
	}

	log.Tracef("Filtered out %v routes with node %x",
		len(routes)-len(prunedRoutes), skipNode[:])

	return prunedRoutes
}

// pruneChannelFromRoutes accepts a set of routes, and returns a new set of
// routes with the target channel filtered out.
func pruneChannelFromRoutes(routes []*Route, skipChan uint64) []*Route {
	prunedRoutes := make([]*Route, 0, len(routes))
	for _, route := range routes {
		if route.containsChannel(skipChan) {
			continue
		}

		prunedRoutes = append(prunedRoutes, route)
	}

	log.Tracef("Filtered out %v routes with channel %v",
		len(routes)-len(prunedRoutes), skipChan)

	return prunedRoutes
}

// pathsToFeeSortedRoutes takes a set of paths, and returns a corresponding set
// of routes. A route differs from a path in that it has full time-lock and
// fee information attached. The set of routes returned may be less than the
// initial set of paths as it's possible we drop a route if it can't handle the
// total payment flow after fees are calculated.
func pathsToFeeSortedRoutes(source Vertex, paths [][]*ChannelHop,
	finalCLTVDelta uint16, amt, feeLimit lnwire.MilliSatoshi,
	currentHeight uint32) ([]*Route, error) {

	validRoutes := make([]*Route, 0, len(paths))
	for _, path := range paths {
		// Attempt to make the path into a route. We snip off the first
		// hop in the path as it contains a "self-hop" that is inserted
		// by our KSP algorithm.
		route, err := newRoute(
			amt, feeLimit, source, path[1:], currentHeight,
			finalCLTVDelta,
		)
		if err != nil {
			// TODO(roasbeef): report straw breaking edge?
			continue
		}

		// If the path as enough total flow to support the computed
		// route, then we'll add it to our set of valid routes.
		validRoutes = append(validRoutes, route)
	}

	// If all our perspective routes were eliminating during the transition
	// from path to route, then we'll return an error to the caller
	if len(validRoutes) == 0 {
		return nil, newErr(ErrNoPathFound, "unable to find a path to "+
			"destination")
	}

	// Finally, we'll sort the set of validate routes to optimize for
	// lowest total fees, using the required time-lock within the route as
	// a tie-breaker.
	sort.Slice(validRoutes, func(i, j int) bool {
		// To make this decision we first check if the total fees
		// required for both routes are equal. If so, then we'll let
		// the total time lock be the tie breaker. Otherwise, we'll put
		// the route with the lowest total fees first.
		if validRoutes[i].TotalFees == validRoutes[j].TotalFees {
			timeLockI := validRoutes[i].TotalTimeLock
			timeLockJ := validRoutes[j].TotalTimeLock
			return timeLockI < timeLockJ
		}

		return validRoutes[i].TotalFees < validRoutes[j].TotalFees
	})

	return validRoutes, nil
}

// FindRoutes attempts to query the ChannelRouter for a bounded number
// available paths to a particular target destination which is able to send
// `amt` after factoring in channel capacities and cumulative fees along each
// route.  To `numPaths eligible paths, we use a modified version of
// Yen's algorithm which itself uses a modified version of Dijkstra's algorithm
// within its inner loop.  Once we have a set of candidate routes, we calculate
// the required fee and time lock values running backwards along the route. The
// route that will be ranked the highest is the one with the lowest cumulative
// fee along the route.
func (r *ChannelRouter) FindRoutes(target *btcec.PublicKey,
	amt, feeLimit lnwire.MilliSatoshi, numPaths uint32,
	finalExpiry ...uint16) ([]*Route, error) {

	var finalCLTVDelta uint16
	if len(finalExpiry) == 0 {
		finalCLTVDelta = DefaultFinalCLTVDelta
	} else {
		finalCLTVDelta = finalExpiry[0]
	}

	dest := target.SerializeCompressed()
	log.Debugf("Searching for path to %x, sending %v", dest, amt)

	// Before attempting to perform a series of graph traversals to find
	// the k-shortest paths to the destination, we'll first consult our
	// path cache
	rt := newRouteTuple(amt, dest)
	r.routeCacheMtx.RLock()
	routes, ok := r.routeCache[rt]
	r.routeCacheMtx.RUnlock()

	// If we already have a cached route, and it contains at least the
	// number of paths requested, then we'll return it directly as there's
	// no need to repeat the computation.
	if ok && uint32(len(routes)) >= numPaths {
		return routes, nil
	}

	// If we don't have a set of routes cached, we'll query the graph for a
	// set of potential routes to the destination node that can support our
	// payment amount. If no such routes can be found then an error will be
	// returned.

	// We can short circuit the routing by opportunistically checking to
	// see if the target vertex event exists in the current graph.
	targetVertex := NewVertex(target)
	if _, exists, err := r.cfg.Graph.HasLightningNode(targetVertex); err != nil {
		return nil, err
	} else if !exists {
		log.Debugf("Target %x is not in known graph", dest)
		return nil, newErrf(ErrTargetNotInNetwork, "target not found")
	}

	// We'll also fetch the current block height so we can properly
	// calculate the required HTLC time locks within the route.
	_, currentHeight, err := r.cfg.Chain.GetBestBlock()
	if err != nil {
		return nil, err
	}

	// Before we open the db transaction below, we'll attempt to obtain a
	// set of bandwidth hints that can help us eliminate certain routes
	// early on in the path finding process.
	bandwidthHints, err := generateBandwidthHints(
		r.selfNode, r.cfg.QueryBandwidth,
	)
	if err != nil {
		return nil, err
	}

	tx, err := r.cfg.Graph.Database().Begin(false)
	if err != nil {
		tx.Rollback()
		return nil, err
	}

	// Now that we know the destination is reachable within the graph,
	// we'll execute our KSP algorithm to find the k-shortest paths from
	// our source to the destination.
	shortestPaths, err := findPaths(
		tx, r.cfg.Graph, r.selfNode, target, amt, numPaths,
		bandwidthHints,
	)
	if err != nil {
		tx.Rollback()
		return nil, err
	}

	tx.Rollback()

	// Now that we have a set of paths, we'll need to turn them into
	// *routes* by computing the required time-lock and fee information for
	// each path. During this process, some paths may be discarded if they
	// aren't able to support the total satoshis flow once fees have been
	// factored in.
	sourceVertex := Vertex(r.selfNode.PubKeyBytes)
	validRoutes, err := pathsToFeeSortedRoutes(
		sourceVertex, shortestPaths, finalCLTVDelta, amt, feeLimit,
		uint32(currentHeight),
	)
	if err != nil {
		return nil, err
	}

	go log.Tracef("Obtained %v paths sending %v to %x: %v", len(validRoutes),
		amt, dest, newLogClosure(func() string {
			return spew.Sdump(validRoutes)
		}),
	)

	// Populate the cache with this set of fresh routes so we can reuse
	// them in the future.
	r.routeCacheMtx.Lock()
	r.routeCache[rt] = validRoutes
	r.routeCacheMtx.Unlock()

	return validRoutes, nil
}

// generateSphinxPacket generates then encodes a sphinx packet which encodes
// the onion route specified by the passed layer 3 route. The blob returned
// from this function can immediately be included within an HTLC add packet to
// be sent to the first hop within the route.
func generateSphinxPacket(route *Route, paymentHash []byte) ([]byte,
	*sphinx.Circuit, error) {
	// First obtain all the public keys along the route which are contained
	// in each hop.
	nodes := make([]*btcec.PublicKey, len(route.Hops))
	for i, hop := range route.Hops {
		// We create a new instance of the public key to avoid possibly
		// mutating the curve parameters, which are unset in a higher
		// level in order to avoid spamming the logs.
		nodePub, err := hop.Channel.Node.PubKey()
		if err != nil {
			return nil, nil, err
		}
		pub := btcec.PublicKey{
			Curve: btcec.S256(),
			X:     nodePub.X,
			Y:     nodePub.Y,
		}
		nodes[i] = &pub
	}

	// Next we generate the per-hop payload which gives each node within
	// the route the necessary information (fees, CLTV value, etc) to
	// properly forward the payment.
	hopPayloads := route.ToHopPayloads()

	log.Tracef("Constructed per-hop payloads for payment_hash=%x: %v",
		paymentHash[:], newLogClosure(func() string {
			return spew.Sdump(hopPayloads)
		}),
	)

	sessionKey, err := btcec.NewPrivateKey(btcec.S256())
	if err != nil {
		return nil, nil, err
	}

	// Next generate the onion routing packet which allows us to perform
	// privacy preserving source routing across the network.
	sphinxPacket, err := sphinx.NewOnionPacket(
		nodes, sessionKey, hopPayloads, paymentHash,
	)
	if err != nil {
		return nil, nil, err
	}

	// Finally, encode Sphinx packet using its wire representation to be
	// included within the HTLC add packet.
	var onionBlob bytes.Buffer
	if err := sphinxPacket.Encode(&onionBlob); err != nil {
		return nil, nil, err
	}

	log.Tracef("Generated sphinx packet: %v",
		newLogClosure(func() string {
			// We unset the internal curve here in order to keep
			// the logs from getting noisy.
			sphinxPacket.EphemeralKey.Curve = nil
			return spew.Sdump(sphinxPacket)
		}),
	)

	return onionBlob.Bytes(), &sphinx.Circuit{
		SessionKey:  sessionKey,
		PaymentPath: nodes,
	}, nil
}

// LightningPayment describes a payment to be sent through the network to the
// final destination.
type LightningPayment struct {
	// Target is the node in which the payment should be routed towards.
	Target *btcec.PublicKey

	// Amount is the value of the payment to send through the network in
	// milli-satoshis.
	Amount lnwire.MilliSatoshi

	// FeeLimit is the maximum fee in millisatoshis that the payment should
	// accept when sending it through the network. The payment will fail
	// if there isn't a route with lower fees than this limit.
	FeeLimit lnwire.MilliSatoshi

	// PaymentHash is the r-hash value to use within the HTLC extended to
	// the first hop.
	PaymentHash [32]byte

	// FinalCLTVDelta is the CTLV expiry delta to use for the _final_ hop
	// in the route. This means that the final hop will have a CLTV delta
	// of at least: currentHeight + FinalCLTVDelta. If this value is
	// unspecified, then a default value of DefaultFinalCLTVDelta will be
	// used.
	FinalCLTVDelta *uint16

	// PayAttemptTimeout is a timeout value that we'll use to determine
	// when we should should abandon the payment attempt after consecutive
	// payment failure. This prevents us from attempting to send a payment
	// indefinitely.
	PayAttemptTimeout time.Duration

	// RouteHints represents the different routing hints that can be used to
	// assist a payment in reaching its destination successfully. These
	// hints will act as intermediate hops along the route.
	//
	// NOTE: This is optional unless required by the payment. When providing
	// multiple routes, ensure the hop hints within each route are chained
	// together and sorted in forward order in order to reach the
	// destination successfully.
	RouteHints [][]HopHint

	// TODO(roasbeef): add e2e message?
}

// SendPayment attempts to send a payment as described within the passed
// LightningPayment. This function is blocking and will return either: when the
// payment is successful, or all candidates routes have been attempted and
// resulted in a failed payment. If the payment succeeds, then a non-nil Route
// will be returned which describes the path the successful payment traversed
// within the network to reach the destination. Additionally, the payment
// preimage will also be returned.
func (r *ChannelRouter) SendPayment(payment *LightningPayment) ([32]byte, *Route, error) {
	// Before starting the HTLC routing attempt, we'll create a fresh
	// payment session which will report our errors back to mission
	// control.
	paySession, err := r.missionControl.NewPaymentSession(
		payment.RouteHints, payment.Target,
	)
	if err != nil {
		return [32]byte{}, nil, err
	}

	return r.sendPayment(payment, paySession)
}

// SendToRoute attempts to send a payment as described within the passed
// LightningPayment through the provided routes. This function is blocking
// and will return either: when the payment is successful, or all routes
// have been attempted and resulted in a failed payment. If the payment
// succeeds, then a non-nil Route will be returned which describes the
// path the successful payment traversed within the network to reach the
// destination. Additionally, the payment preimage will also be returned.
func (r *ChannelRouter) SendToRoute(routes []*Route,
	payment *LightningPayment) ([32]byte, *Route, error) {

	paySession := r.missionControl.NewPaymentSessionFromRoutes(
		routes,
	)

	return r.sendPayment(payment, paySession)
}

// sendPayment attempts to send a payment as described within the passed
// LightningPayment. This function is blocking and will return either: when the
// payment is successful, or all candidates routes have been attempted and
// resulted in a failed payment. If the payment succeeds, then a non-nil Route
// will be returned which describes the path the successful payment traversed
// within the network to reach the destination. Additionally, the payment
// preimage will also be returned.
func (r *ChannelRouter) sendPayment(payment *LightningPayment,
	paySession *paymentSession) ([32]byte, *Route, error) {

	log.Tracef("Dispatching route for lightning payment: %v",
		newLogClosure(func() string {
			// Remove the public key curve parameters when logging
			// the route to prevent spamming the logs.
			if payment.Target != nil {
				payment.Target.Curve = nil
			}

			for _, routeHint := range payment.RouteHints {
				for _, hopHint := range routeHint {
					hopHint.NodeID.Curve = nil
				}
			}
			return spew.Sdump(payment)
		}),
	)

	var (
		preImage  [32]byte
		sendError error
	)

	// errFailedFeeChans is a map of the short channel ID's that were the
	// source of fee related routing failures during this payment attempt.
	// We'll use this map to prune out channels when the first error may
	// not require pruning, but any subsequent ones do.
	errFailedFeeChans := make(map[lnwire.ShortChannelID]struct{})

	// We'll also fetch the current block height so we can properly
	// calculate the required HTLC time locks within the route.
	_, currentHeight, err := r.cfg.Chain.GetBestBlock()
	if err != nil {
		return [32]byte{}, nil, err
	}

	var finalCLTVDelta uint16
	if payment.FinalCLTVDelta == nil {
		finalCLTVDelta = DefaultFinalCLTVDelta
	} else {
		finalCLTVDelta = *payment.FinalCLTVDelta
	}

	var payAttemptTimeout time.Duration
	if payment.PayAttemptTimeout == time.Duration(0) {
		payAttemptTimeout = defaultPayAttemptTimeout
	} else {
		payAttemptTimeout = payment.PayAttemptTimeout
	}

	timeoutChan := time.After(payAttemptTimeout)

	// We'll continue until either our payment succeeds, or we encounter a
	// critical error during path finding.
	for {
		// Before we attempt this next payment, we'll check to see if
		// either we've gone past the payment attempt timeout, or the
		// router is exiting. In either case, we'll stop this payment
		// attempt short.
		select {
		case <-timeoutChan:
			errStr := fmt.Sprintf("payment attempt not completed "+
				"before timeout of %v", payAttemptTimeout)

			return preImage, nil, newErr(
				ErrPaymentAttemptTimeout, errStr,
			)

		case <-r.quit:
			return preImage, nil, fmt.Errorf("router shutting down")

		default:
			// Fall through if we haven't hit our time limit, or
			// are expiring.
		}

		route, err := paySession.RequestRoute(
			payment, uint32(currentHeight), finalCLTVDelta,
		)
		if err != nil {
			// If we're unable to successfully make a payment using
			// any of the routes we've found, then return an error.
			if sendError != nil {
				return [32]byte{}, nil, fmt.Errorf("unable to "+
					"route payment to destination: %v",
					sendError)
			}

			return preImage, nil, err
		}

		log.Tracef("Attempting to send payment %x, using route: %v",
			payment.PaymentHash, newLogClosure(func() string {
				return spew.Sdump(route)
			}),
		)

		// Generate the raw encoded sphinx packet to be included along
		// with the htlcAdd message that we send directly to the
		// switch.
		onionBlob, circuit, err := generateSphinxPacket(
			route, payment.PaymentHash[:],
		)
		if err != nil {
			return preImage, nil, err
		}

		// Craft an HTLC packet to send to the layer 2 switch. The
		// metadata within this packet will be used to route the
		// payment through the network, starting with the first-hop.
		htlcAdd := &lnwire.UpdateAddHTLC{
			Amount:      route.TotalAmount,
			Expiry:      route.TotalTimeLock,
			PaymentHash: payment.PaymentHash,
		}
		copy(htlcAdd.OnionBlob[:], onionBlob)

		// Attempt to send this payment through the network to complete
		// the payment. If this attempt fails, then we'll continue on
		// to the next available route.
		firstHop := route.Hops[0].Channel.Node.PubKeyBytes
		preImage, sendError = r.cfg.SendToSwitch(
			firstHop, htlcAdd, circuit,
		)
		if sendError != nil {
			// An error occurred when attempting to send the
			// payment, depending on the error type, we'll either
			// continue to send using alternative routes, or simply
			// terminate this attempt.
			log.Errorf("Attempt to send payment %x failed: %v",
				payment.PaymentHash, sendError)

			fErr, ok := sendError.(*htlcswitch.ForwardingError)
			if !ok {
				return preImage, nil, sendError
			}

			errSource := fErr.ErrorSource

			log.Tracef("node=%x reported failure when sending "+
				"htlc=%x", errSource.SerializeCompressed(),
				payment.PaymentHash[:])

			switch onionErr := fErr.FailureMessage.(type) {
			// If the end destination didn't know they payment
			// hash, then we'll terminate immediately.
			case *lnwire.FailUnknownPaymentHash:
				return preImage, nil, sendError

			// If we sent the wrong amount to the destination, then
			// we'll exit early.
			case *lnwire.FailIncorrectPaymentAmount:
				return preImage, nil, sendError

			// If the time-lock that was extended to the final node
			// was incorrect, then we can't proceed.
			case *lnwire.FailFinalIncorrectCltvExpiry:
				return preImage, nil, sendError

			// If we crafted an invalid onion payload for the final
			// node, then we'll exit early.
			case *lnwire.FailFinalIncorrectHtlcAmount:
				return preImage, nil, sendError

			// Similarly, if the HTLC expiry that we extended to
			// the final hop expires too soon, then will fail the
			// payment.
			//
			// TODO(roasbeef): can happen to to race condition, try
			// again with recent block height
			case *lnwire.FailFinalExpiryTooSoon:
				return preImage, nil, sendError

			// If we erroneously attempted to cross a chain border,
			// then we'll cancel the payment.
			case *lnwire.FailInvalidRealm:
				return preImage, nil, sendError

			// If we get a notice that the expiry was too soon for
			// an intermediate node, then we'll prune out the node
			// that sent us this error, as it doesn't now what the
			// correct block height is.
			case *lnwire.FailExpiryTooSoon:
				update := onionErr.Update
				if err := r.applyChannelUpdate(&update); err != nil {
					log.Errorf("unable to apply channel "+
						"update for onion error: %v", err)
				}

				pruneVertexFailure(
					paySession, route, errSource, false,
				)
				continue

			// If we hit an instance of onion payload corruption or
			// an invalid version, then we'll exit early as this
			// shouldn't happen in the typical case.
			case *lnwire.FailInvalidOnionVersion:
				return preImage, nil, sendError
			case *lnwire.FailInvalidOnionHmac:
				return preImage, nil, sendError
			case *lnwire.FailInvalidOnionKey:
				return preImage, nil, sendError

			// If the onion error includes a channel update, and
			// isn't necessarily fatal, then we'll apply the update
			// and continue with the rest of the routes.
			case *lnwire.FailAmountBelowMinimum:
				update := onionErr.Update
				if err := r.applyChannelUpdate(&update); err != nil {
					log.Errorf("unable to apply channel "+
						"update for onion error: %v", err)
				}

				return preImage, nil, sendError

			// If we get a failure due to a fee, so we'll apply the
			// new fee update, and retry our attempt using the
			// newly updated fees.
			case *lnwire.FailFeeInsufficient:
				update := onionErr.Update
				if err := r.applyChannelUpdate(&update); err != nil {
					log.Errorf("unable to apply channel "+
						"update for onion error: %v", err)

					pruneEdgeFailure(
						paySession, route, errSource,
					)
				}

				// We'll now check to see if we've already
				// reported a fee related failure for this
				// node. If so, then we'll actually prune out
				// the vertex for now.
				chanID := update.ShortChannelID
				_, ok := errFailedFeeChans[chanID]
				if ok {
					pruneVertexFailure(
						paySession, route, errSource, false,
					)
					continue
				}

				// Finally, we'll record a fee failure from
				// this node and move on.
				errFailedFeeChans[chanID] = struct{}{}
				continue

			// If we get the failure for an intermediate node that
			// disagrees with our time lock values, then we'll
			// prune it out for now, and continue with path
			// finding.
			case *lnwire.FailIncorrectCltvExpiry:
				update := onionErr.Update
				if err := r.applyChannelUpdate(&update); err != nil {
					log.Errorf("unable to apply channel "+
						"update for onion error: %v", err)
				}

				pruneVertexFailure(
					paySession, route, errSource, false,
				)
				continue

			// The outgoing channel that this node was meant to
			// forward one is currently disabled, so we'll apply
			// the update and continue.
			case *lnwire.FailChannelDisabled:
				update := onionErr.Update
				if err := r.applyChannelUpdate(&update); err != nil {
					log.Errorf("unable to apply channel "+
						"update for onion error: %v", err)
				}

				pruneEdgeFailure(paySession, route, errSource)
				continue

			// It's likely that the outgoing channel didn't have
			// sufficient capacity, so we'll prune this edge for
			// now, and continue onwards with our path finding.
			case *lnwire.FailTemporaryChannelFailure:
				update := onionErr.Update
				if err := r.applyChannelUpdate(update); err != nil {
					log.Errorf("unable to apply channel "+
						"update for onion error: %v", err)
				}

				pruneEdgeFailure(paySession, route, errSource)
				continue

			// If the send fail due to a node not having the
			// required features, then we'll note this error and
			// continue.
			case *lnwire.FailRequiredNodeFeatureMissing:
				pruneVertexFailure(
					paySession, route, errSource, false,
				)
				continue

			// If the send fail due to a node not having the
			// required features, then we'll note this error and
			// continue.
			case *lnwire.FailRequiredChannelFeatureMissing:
				pruneVertexFailure(
					paySession, route, errSource, false,
				)
				continue

			// If the next hop in the route wasn't known or
			// offline, we'll only the channel which we attempted
			// to route over. This is conservative, and it can
			// handle faulty channels between nodes properly.
			// Additionally, this guards against routing nodes
			// returning errors in order to attempt to black list
			// another node.
			case *lnwire.FailUnknownNextPeer:
				pruneEdgeFailure(paySession, route, errSource)
				continue

			// If the node wasn't able to forward for which ever
			// reason, then we'll note this and continue with the
			// routes.
			case *lnwire.FailTemporaryNodeFailure:
				pruneVertexFailure(
					paySession, route, errSource, false,
				)
				continue

			case *lnwire.FailPermanentNodeFailure:
				pruneVertexFailure(
					paySession, route, errSource, false,
				)
				continue

			// If we get a permanent channel or node failure, then
			// we'll note this (exclude the vertex/edge), and
			// continue with the rest of the routes.
			case *lnwire.FailPermanentChannelFailure:
				pruneEdgeFailure(paySession, route, errSource)
				continue

			default:
				return preImage, nil, sendError
			}
		}

		return preImage, route, nil
	}
}

// pruneVertexFailure will attempt to prune a vertex from the current available
// vertexes of the target payment session in response to an encountered routing
// error.
func pruneVertexFailure(paySession *paymentSession, route *Route,
	errSource *btcec.PublicKey, nextNode bool) {

	// By default, we'll try to prune the node that actually sent us the
	// error.
	errNode := NewVertex(errSource)

	// If this failure indicates that the node _after_ the source of the
	// error was not found. As a result, we'll locate the vertex for that
	// node itself.
	if nextNode {
		nodeToPrune, ok := route.nextHopVertex(errSource)

		if ok {
			errNode = nodeToPrune
		}
	}

	// Once we've located the vertex, we'll report this failure to
	// missionControl and restart path finding.
	paySession.ReportVertexFailure(errNode)
}

// pruneEdgeFailure will attempts to prune an edge from the current available
// edges of the target payment session in response to an encountered routing
// error.
func pruneEdgeFailure(paySession *paymentSession, route *Route,
	errSource *btcec.PublicKey) {

	// As this error indicates that the target channel was unable to carry
	// this HTLC (for w/e reason), we'll query the index to find the
	// _outgoing_ channel the source of the error was meant to pass the
	// HTLC along to.
	badChan, ok := route.nextHopChannel(errSource)
	if !ok {
		// If we weren't able to find the hop *after* this node, then
		// we'll attempt to disable the previous channel.
		prevChan, ok := route.prevHopChannel(
			errSource,
		)

		if !ok {
			return
		}

		badChan = prevChan
	}

	// If the channel was found, then we'll inform mission control of this
	// failure so future attempts avoid this link temporarily.
	paySession.ReportChannelFailure(badChan.ChannelID)
}

// applyChannelUpdate applies a channel update directly to the database,
// skipping preliminary validation.
func (r *ChannelRouter) applyChannelUpdate(msg *lnwire.ChannelUpdate) error {
	// If we get passed a nil channel update (as it's optional with some
	// onion errors), then we'll exit early with a nil error.
	if msg == nil {
		return nil
	}

	err := r.UpdateEdge(&channeldb.ChannelEdgePolicy{
		SigBytes:                  msg.Signature.ToSignatureBytes(),
		ChannelID:                 msg.ShortChannelID.ToUint64(),
		LastUpdate:                time.Unix(int64(msg.Timestamp), 0),
		Flags:                     msg.Flags,
		TimeLockDelta:             msg.TimeLockDelta,
		MinHTLC:                   msg.HtlcMinimumMsat,
		FeeBaseMSat:               lnwire.MilliSatoshi(msg.BaseFee),
		FeeProportionalMillionths: lnwire.MilliSatoshi(msg.FeeRate),
	})
	if err != nil && !IsError(err, ErrIgnored) {
		return fmt.Errorf("Unable to apply channel update: %v", err)
	}

	return nil
}

// AddNode is used to add information about a node to the router database. If
// the node with this pubkey is not present in an existing channel, it will
// be ignored.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) AddNode(node *channeldb.LightningNode) error {
	rMsg := &routingMsg{
		msg: node,
		err: make(chan error, 1),
	}

	select {
	case r.networkUpdates <- rMsg:
		select {
		case err := <-rMsg.err:
			return err
		case <-r.quit:
			return errors.New("router has been shut down")
		}
	case <-r.quit:
		return errors.New("router has been shut down")
	}
}

// AddEdge is used to add edge/channel to the topology of the router, after all
// information about channel will be gathered this edge/channel might be used
// in construction of payment path.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) AddEdge(edge *channeldb.ChannelEdgeInfo) error {
	rMsg := &routingMsg{
		msg: edge,
		err: make(chan error, 1),
	}

	select {
	case r.networkUpdates <- rMsg:
		select {
		case err := <-rMsg.err:
			return err
		case <-r.quit:
			return errors.New("router has been shut down")
		}
	case <-r.quit:
		return errors.New("router has been shut down")
	}
}

// UpdateEdge is used to update edge information, without this message edge
// considered as not fully constructed.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) UpdateEdge(update *channeldb.ChannelEdgePolicy) error {
	rMsg := &routingMsg{
		msg: update,
		err: make(chan error, 1),
	}

	select {
	case r.networkUpdates <- rMsg:
		select {
		case err := <-rMsg.err:
			return err
		case <-r.quit:
			return errors.New("router has been shut down")
		}
	case <-r.quit:
		return errors.New("router has been shut down")
	}
}

// CurrentBlockHeight returns the block height from POV of the router subsystem.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) CurrentBlockHeight() (uint32, error) {
	_, height, err := r.cfg.Chain.GetBestBlock()
	return uint32(height), err
}

// GetChannelByID return the channel by the channel id.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) GetChannelByID(chanID lnwire.ShortChannelID) (
	*channeldb.ChannelEdgeInfo,
	*channeldb.ChannelEdgePolicy,
	*channeldb.ChannelEdgePolicy, error) {

	return r.cfg.Graph.FetchChannelEdgesByID(chanID.ToUint64())
}

// ForEachNode is used to iterate over every node in router topology.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) ForEachNode(cb func(*channeldb.LightningNode) error) error {
	return r.cfg.Graph.ForEachNode(nil, func(_ *bolt.Tx, n *channeldb.LightningNode) error {
		return cb(n)
	})
}

// ForAllOutgoingChannels is used to iterate over all outgoing channels owned by
// the router.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) ForAllOutgoingChannels(cb func(*channeldb.ChannelEdgeInfo,
	*channeldb.ChannelEdgePolicy) error) error {

	return r.selfNode.ForEachChannel(nil, func(_ *bolt.Tx, c *channeldb.ChannelEdgeInfo,
		e, _ *channeldb.ChannelEdgePolicy) error {

		return cb(c, e)
	})
}

// ForEachChannel is used to iterate over every known edge (channel) within our
// view of the channel graph.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) ForEachChannel(cb func(chanInfo *channeldb.ChannelEdgeInfo,
	e1, e2 *channeldb.ChannelEdgePolicy) error) error {

	return r.cfg.Graph.ForEachChannel(cb)
}

// AddProof updates the channel edge info with proof which is needed to
// properly announce the edge to the rest of the network.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) AddProof(chanID lnwire.ShortChannelID,
	proof *channeldb.ChannelAuthProof) error {

	info, _, _, err := r.cfg.Graph.FetchChannelEdgesByID(chanID.ToUint64())
	if err != nil {
		return err
	}

	info.AuthProof = proof
	return r.cfg.Graph.UpdateChannelEdge(info)
}

// IsStaleNode returns true if the graph source has a node announcement for the
// target node with a more recent timestamp.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) IsStaleNode(node Vertex, timestamp time.Time) bool {
	// If our attempt to assert that the node announcement is fresh fails,
	// then we know that this is actually a stale announcement.
	return r.assertNodeAnnFreshness(node, timestamp) != nil
}

// IsKnownEdge returns true if the graph source already knows of the passed
// channel ID.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) IsKnownEdge(chanID lnwire.ShortChannelID) bool {
	_, _, exists, _ := r.cfg.Graph.HasChannelEdge(chanID.ToUint64())
	return exists
}

// IsStaleEdgePolicy returns true if the graph soruce has a channel edge for
// the passed channel ID (and flags) that have a more recent timestamp.
//
// NOTE: This method is part of the ChannelGraphSource interface.
func (r *ChannelRouter) IsStaleEdgePolicy(chanID lnwire.ShortChannelID,
	timestamp time.Time, flags lnwire.ChanUpdateFlag) bool {

	edge1Timestamp, edge2Timestamp, exists, err := r.cfg.Graph.HasChannelEdge(
		chanID.ToUint64(),
	)
	if err != nil {
		return false

	}

	// If we don't know of the edge, then it means it's fresh (thus not
	// stale).
	if !exists {
		return false
	}

	// As edges are directional edge node has a unique policy for the
	// direction of the edge they control. Therefore we first check if we
	// already have the most up to date information for that edge. If so,
	// then we can exit early.
	switch {

	// A flag set of 0 indicates this is an announcement for the "first"
	// node in the channel.
	case flags&lnwire.ChanUpdateDirection == 0:
		return !edge1Timestamp.Before(timestamp)

	// Similarly, a flag set of 1 indicates this is an announcement for the
	// "second" node in the channel.
	case flags&lnwire.ChanUpdateDirection == 1:
		return !edge2Timestamp.Before(timestamp)
	}

	return false
}