fundingManager: use NotifyWhenOnline when sending fundingLocked.

The fundingManager will register with the server to get notified
when the targetted peer comes online, in case of a failed send
of the fundingLocked message. This is necessary because if the
peer is not connected yet (or was disconnected while we were
waiting for the funding tx to confirm), we cannot continue the
the opening process before the peer successfully has received
the fundingLocked.

fundingmanager.go

@@ -190,6 +190,12 @@ type fundingConfig struct {
 	// channel's funding transaction and initial commitment transaction.
 	SendToPeer func(target *btcec.PublicKey, msgs ...lnwire.Message) error
 
+	// NotifyWhenOnline allows the FundingManager to register with a
+	// subsystem that will notify it when the peer comes online.
+	// This is used when sending the fundingLocked message, since it MUST be
+	// delivered after the funding transaction is confirmed.
+	NotifyWhenOnline func(peer *btcec.PublicKey, connectedChan chan<- struct{})
+
 	// FindPeer searches the list of peers connected to the node so that
 	// the FundingManager can notify other daemon subsystems as necessary
 	// during the funding process.
@@ -1453,18 +1459,46 @@ func (f *fundingManager) sendFundingLockedAndAnnounceChannel(
 	}
 	fundingLockedMsg := lnwire.NewFundingLocked(chanID, nextRevocation)
 
-	err = f.cfg.SendToPeer(completeChan.IdentityPub, fundingLockedMsg)
-	if err != nil {
-		fndgLog.Errorf("unable to send fundingLocked to peer: %v", err)
-		return
+	// If the peer has disconnected before we reach this point, we will need
+	// to wait for him to come back online before sending the fundingLocked
+	// message. This is special for fundingLocked, since failing to send any
+	// of the previous messages in the funding flow just cancels the flow.
+	// But now the funding transaction is confirmed, the channel is open
+	// and we have to make sure the peer gets the fundingLocked message when
+	// it comes back online. This is also crucial during restart of lnd,
+	// where we might try to resend the fundingLocked message before the
+	// server has had the time to connect to the peer. We keep trying to
+	// send fundingLocked until we succeed, or the fundingManager is shut
+	// down.
+	for {
+		err = f.cfg.SendToPeer(completeChan.IdentityPub,
+			fundingLockedMsg)
+		if err == nil {
+			// Sending succeeded, we can break out and continue
+			// the funding flow.
+			break
+		}
+
+		fndgLog.Warnf("unable to send fundingLocked to peer %x: "+
+			"%v. Will retry when online",
+			completeChan.IdentityPub.SerializeCompressed(), err)
+
+		connected := make(chan struct{})
+		f.cfg.NotifyWhenOnline(completeChan.IdentityPub, connected)
+		select {
+		case <-connected:
+			// Retry sending.
+		case <-f.quit:
+			return
+		}
 	}
 
 	// As the fundingLocked message is now sent to the peer, the channel is
 	// moved to the next state of the state machine. It will be moved to the
 	// last state (actually deleted from the database) after the channel is
 	// finally announced.
-	err = f.saveChannelOpeningState(&completeChan.FundingOutpoint, fundingLockedSent,
-		shortChanID)
+	err = f.saveChannelOpeningState(&completeChan.FundingOutpoint,
+		fundingLockedSent, shortChanID)
 	if err != nil {
 		fndgLog.Errorf("error setting channel state to "+
 			"fundingLockedSent: %v", err)

lnd.go

@@ -186,10 +186,11 @@ func lndMain() error {
 				idPrivKey.PubKey())
 			return <-errChan
 		},
-		ArbiterChan:    server.breachArbiter.newContracts,
-		SendToPeer:     server.SendToPeer,
-		FindPeer:       server.FindPeer,
-		TempChanIDSeed: chanIDSeed,
+		ArbiterChan:      server.breachArbiter.newContracts,
+		SendToPeer:       server.SendToPeer,
+		NotifyWhenOnline: server.NotifyWhenOnline,
+		FindPeer:         server.FindPeer,
+		TempChanIDSeed:   chanIDSeed,
 		FindChannel: func(chanID lnwire.ChannelID) (*lnwallet.LightningChannel, error) {
 			dbChannels, err := chanDB.FetchAllChannels()
 			if err != nil {