telegabber/calls/signaling/tgsig/adapter.go

// telegram side of signaling.Bridge.
// Adapter is per-session; Manager is keyed by (jid, callID).
// Caller = gateway issued CreateCall, Callee = gateway received Pending.
package tgsig

import (
	"sync"
	"time"

	"dev.narayana.im/narayana/telegabber/calls/audio"
	"dev.narayana.im/narayana/telegabber/calls/signaling"
	"gotgcalls/ntgcalls"

	log "github.com/sirupsen/logrus"
	"github.com/zelenin/go-tdlib/client"
)

// past this point ntgcalls is probably stuck and the user hears silence -
// audio sent before Connected is dropped
const ntgConnectStallThreshold = 7 * time.Second

// narrow upstream APIs to what we actually use, so tests can fake them

type TdlibClient interface {
	CreateCall(req *client.CreateCallRequest) (*client.CallId, error)
	AcceptCall(req *client.AcceptCallRequest) (*client.Ok, error)
	DiscardCall(req *client.DiscardCallRequest) (*client.Ok, error)
	SendCallSignalingData(req *client.SendCallSignalingDataRequest) (*client.Ok, error)
}

type NtgCallsClient interface {
	OnSignal(callback ntgcalls.SignalCallback)
	OnConnectionChange(callback ntgcalls.ConnectionChangeCallback)
	CreateP2PCall(chatId int64) error
	SkipExchange(chatId int64, encryptionKey []byte, isOutgoing bool) error
	ConnectP2P(chatId int64, rtcServers []ntgcalls.RTCServer, versions []string, p2pAllowed bool) error
	SendSignalingData(chatId int64, data []byte) error
	Stop(chatId int64) error
}

type Adapter struct {
	tdlib    TdlibClient
	ntg      NtgCallsClient
	audioNtg audio.NtgClient // nil disables audio (signaling-only tests)
	jid      string
	mgr      *Manager

	mu       sync.Mutex
	byUserID map[int64]*callBase

	// gates ntg/tdlib touches against Close()/Free() on the C++ object.
	// dispatchers take RLock via withCallContext; Close takes Lock.
	lifecycleLock sync.RWMutex
	closed        bool

	// per-chatID stall warning while ntgcalls is in Connecting
	ntgConnectTimersMu sync.Mutex
	ntgConnectTimers   map[int64]*time.Timer
}

// runs fn under the lifecycle RLock; returns false (and skips fn) if closed
func (a *Adapter) withCallContext(fn func()) bool {
	a.lifecycleLock.RLock()
	defer a.lifecycleLock.RUnlock()
	if a.closed {
		return false
	}
	fn()
	return true
}

func New(tdlib TdlibClient, ntg NtgCallsClient, audioNtg audio.NtgClient, jid string, mgr *Manager) *Adapter {
	a := &Adapter{
		tdlib:    tdlib,
		ntg:      ntg,
		audioNtg: audioNtg,
		jid:      jid,
		mgr:      mgr,
		byUserID: make(map[int64]*callBase),
	}
	a.attachNtgCallbacks()
	return a
}

func (a *Adapter) JID() string { return a.jid }

// outbound call; callID is unknown until CreateCall returns, so
// the entry registers with Manager from Caller.Start
func (a *Adapter) NewCaller(userID int64, isVideo bool) *Caller {
	return &Caller{callBase: callBase{
		adapter: a, userID: userID, isVideo: isVideo, side: signaling.CallerSide,
	}}
}

// inbound call; callID is known so the entry registers immediately
func (a *Adapter) NewCallee(callID int32, userID int64, isVideo bool) *Callee {
	c := &Callee{callBase: callBase{
		adapter: a, userID: userID, isVideo: isVideo, callID: callID, side: signaling.CalleeSide,
	}}
	a.registerSide(userID, &c.callBase)
	a.mgr.Register(a.jid, callID, &c.callBase)
	return c
}

func (a *Adapter) registerSide(userID int64, s *callBase) {
	a.mu.Lock()
	a.byUserID[userID] = s
	a.mu.Unlock()
}

func (a *Adapter) dropSide(userID int64) {
	a.mu.Lock()
	delete(a.byUserID, userID)
	a.mu.Unlock()
}

func (a *Adapter) lookupSideByUser(userID int64) (*callBase, bool) {
	a.mu.Lock()
	defer a.mu.Unlock()
	s, ok := a.byUserID[userID]
	return s, ok
}

// terminates every active call and marks the adapter closed; after this
// returns no goroutine is mid-ntg/tdlib through here, safe to Free() the
// C++ ntgcalls object. Bridge.Terminate runs outside the lock and is
// synchronous so audio halves close (and ClearStreams runs) while C++
// is still alive
func (a *Adapter) Close() {
	a.lifecycleLock.Lock()
	if a.closed {
		a.lifecycleLock.Unlock()
		return
	}
	a.mu.Lock()
	sides := make([]*callBase, 0, len(a.byUserID))
	for _, s := range a.byUserID {
		sides = append(sides, s)
	}
	a.byUserID = make(map[int64]*callBase)
	a.mu.Unlock()
	a.closed = true
	a.lifecycleLock.Unlock()

	for _, s := range sides {
		if b := s.getBridge(); b != nil {
			b.Terminate(signaling.ReasonHangup)
		}
		if cid := s.getCallID(); cid != 0 {
			a.mgr.Drop(a.jid, cid)
		}
	}
	for _, s := range sides {
		if err := a.ntg.Stop(s.userID); err != nil {
			log.WithField("user_id", s.userID).WithError(err).Debug("ntgcalls Stop during teardown")
		}
	}
}

func (a *Adapter) attachNtgCallbacks() {
	a.ntg.OnSignal(func(chatID int64, data []byte) {
		// invoked from the ntgcalls C++ thread; guard against Close()/Free()
		a.withCallContext(func() {
			s, ok := a.lookupSideByUser(chatID)
			if !ok {
				log.WithFields(log.Fields{
					"jid":     a.jid,
					"user_id": chatID,
				}).Warn("ntgcalls OnSignal for unknown user")
				return
			}
			if _, err := a.tdlib.SendCallSignalingData(&client.SendCallSignalingDataRequest{
				CallId: s.getCallID(),
				Data:   data,
			}); err != nil {
				log.WithFields(log.Fields{
					"call_id": s.getCallID(),
					"user_id": chatID,
				}).WithError(err).Error("Failed to send signaling data to TDLib")
			}
		})
	})

	a.ntg.OnConnectionChange(func(chatID int64, info ntgcalls.NetworkInfo) {
		entry := log.WithFields(log.Fields{
			"jid":     a.jid,
			"user_id": chatID,
			"state":   info.State,
			"kind":    info.Kind,
		})
		entry.Info("ntgcalls connection state changed")

		// the TG peer can hear nothing while ntgcalls is still negotiating
		// with TG's TURN relays even though the XMPP side has RTP flowing
		switch info.State {
		case ntgcalls.Connecting:
			a.startNtgConnectStallTimer(chatID)
		case ntgcalls.Connected, ntgcalls.Failed, ntgcalls.Timeout, ntgcalls.Closed:
			a.stopNtgConnectStallTimer(chatID)
		}

		s, ok := a.lookupSideByUser(chatID)
		if !ok {
			return
		}
		b := s.getBridge()
		if b == nil {
			return
		}
		switch info.State {
		case ntgcalls.Connected:
			b.MediaConnected(s.side)
		case ntgcalls.Failed:
			b.MediaFailed(signaling.ReasonMediaFailed)
		case ntgcalls.Timeout:
			b.MediaFailed(signaling.ReasonConnectTimeout)
		}
	})
}

// (re)arms the stall timer; Connecting can fire more than once
func (a *Adapter) startNtgConnectStallTimer(chatID int64) {
	a.stopNtgConnectStallTimer(chatID)
	t := time.AfterFunc(ntgConnectStallThreshold, func() {
		log.WithFields(log.Fields{
			"jid":     a.jid,
			"user_id": chatID,
			"after":   ntgConnectStallThreshold,
		}).Warn("ntgcalls still Connecting - TG peer is likely hearing silence")
	})
	a.ntgConnectTimersMu.Lock()
	if a.ntgConnectTimers == nil {
		a.ntgConnectTimers = make(map[int64]*time.Timer)
	}
	a.ntgConnectTimers[chatID] = t
	a.ntgConnectTimersMu.Unlock()
}

func (a *Adapter) stopNtgConnectStallTimer(chatID int64) {
	a.ntgConnectTimersMu.Lock()
	t, ok := a.ntgConnectTimers[chatID]
	if ok {
		delete(a.ntgConnectTimers, chatID)
	}
	a.ntgConnectTimersMu.Unlock()
	if t != nil {
		t.Stop()
	}
}