webrtcprivate: add transport by sukunrt · Pull Request #2576 · libp2p/go-libp2p

sukunrt · 2023-09-20T15:33:59Z

Part of: #2009

Depends on: #2586

marten-seemann · 2023-10-13T07:52:21Z

#2586 was merged. @sukunrt Can you rebase please?

marten-seemann

Review of the non-transport parts of this PR.

marten-seemann · 2023-10-13T08:32:51Z

 	Constructor interface{}
 }

+type ICEServer = webrtc.ICEServer


Why do we need to redeclare this? Is it a requirement for Fx?

Nope. Just so that we redefine it a different way later if we want. But I think it's fine to remove this and just take the pion specific object. We can make a breaking change if this needs to be changed.

It’s a breaking change anyway if we change the underlying type. That should be fine.

Can you remove it?

marten-seemann · 2023-10-13T08:46:23Z

 	}
 }
+
+func EnableWebRTCPrivate(stunServers []config.ICEServer) Option {


Why is this a separate option? We already have libp2p.Transport to configure transports. I had hoped that we don't need to introduce a separate knob here.

marten-seemann · 2023-10-13T08:51:59Z

 	rf.metricsTracer.ScheduledWorkUpdated(&scheduledWorkTimes{})
 }
+
+var browserProtocols []int = []int{ma.P_WEBTRANSPORT, ma.P_WEBRTC_DIRECT, ma.P_WS, ma.P_WSS}


Are you sure about /ws? Is this supposed to catch the /ws/tls case?

Suggested change

var browserProtocols []int = []int{ma.P_WEBTRANSPORT, ma.P_WEBRTC_DIRECT, ma.P_WS, ma.P_WSS}

var browserProtocols = []int{ma.P_WEBTRANSPORT, ma.P_WEBRTC_DIRECT, ma.P_WS, ma.P_WSS}

I've removed ma.P_WS. That shouldn't be here.

marten-seemann · 2023-10-13T08:54:52Z

 		for _, addr := range addrs {
 			pub := addr.Encapsulate(circuit)
 			raddrs = append(raddrs, pub)
+			if rf.conf.webRTCSupport && isBrowserDialableAddr(addr) {


Why is this needed for WebRTC and not for WebTransport?

As a general design principle, it would be nice if AutoRelay didn't need to contain any transport-specific logic.

Yeah. This was a poor choice. I've moved this to host.Addrs. Even this is not great but I'm not sure what else to do without an address pipeline. The flow now is:

BasicHost.AllAddrs => RelayFinder.Addrs => BasicHost.AddrFactory => BasicHost.Addrs

marten-seemann · 2023-10-13T08:56:47Z

 	}
-	if isRelayAddr(a) {
+	if isProtocolAddr(a, ma.P_WEBRTC) {
+		return s.transports.m[ma.P_WEBRTC]


Why does this require extra logic here? Is it because of the certhash? Why doesn't it apply to WebTransport then?

The peer address is a circuitv2 address. So we need to check it before the circuitv2 check but the check is the same as the circuit v2 check.

The address is of the form
/ip4/1.2.3.4/udp/1/quic-v1/p2p/<relay-id>/p2p-circuit/webrtc
So we need to ensure that we don't use the quic-v1 or the p2p-circuit transport for dialing.

Didn't isRelayAddr already do that? Or phrased differently, the same problem would have occurred for a non-WebRTC address, why do we need special logic for WebRTC now?

Yes but we don't want the circuit-v2 transport to handle dialing to the webrtc address.
The previous code was

if isRelayAddr => Use circuitv2 transport.

But now we need

if isWebRTCAddr => Use webrtcprivate transport. else if isRelayAddr => Use circuitv2 transport.

This is because the webrtcprivate address has a p2p-circuit component but we want to use webrtcprivate transport. I could have written this as:

if isRelayAddr(addr): if isWebRTCAddr(addr): use webrtcprivate transport else use circuitv2 transport

The flow I've implemented is:
peerAddr: <relay-addr>/p2p-circuit/webrtc

webrtcprivate.Transport.Dial is called with peerAddr

webrtcprivate.Transport.Dial adds the relay address inferred from the webrtcprivate address inferred from the peerstore. Then makes a host.Connect call to establish a relay connection with the peer.

Once relay connection is established it opens a signaling stream and establishes a webrtc connection.

I've chosen this because it is similar to circuitv2 transport which first establishes a direct connection with the relay node and then a circuitv2 connection to the peer over the relayed connection. This also has the nice property that Transport.Dial can connect to the peer over the address no matter the state of the host when Transport.Dial is called. In this case, it means if no circuit-v2 connection exists, it'll make a circuit-v2 connection first.

marten-seemann · 2023-10-13T09:07:48Z

+			for _, addr := range s.host.Peerstore().Addrs(p) {
+				if _, err := addr.ValueForProtocol(ma.P_WEBRTC); err == nil {
+					ctx := network.WithForceDirectDial(s.ctx, "webrtc holepunch")
+					s.host.Connect(ctx, peer.AddrInfo{ID: p}) // address is already in peerstore


This will start multiple Connect calls if there are multiple WebRTC addresses.

It returns immediately after this line, so this should be fine, no?

marten-seemann · 2023-10-13T09:09:39Z

+				return
+			}
+			p := conn.RemotePeer()
+			// Peer supports DCUtR, let it trigger holepunch


What happens if the peer supports DCUtR, but there are only WebRTC addresses?

That case is handled here: https://github.com/libp2p/go-libp2p/blob/webrtcprivate/transport/p2p/protocol/holepunch/holepuncher.go#L142

In this case the peer(DCUtR initiator) is best suited to handle the call so I've kept it on that side.

interim commit

When running with docker, the listener might see a different remote port than the port the dialer observes locally and vice versa

Some failing tests are skipped. They're failing because they require a different testing strategy. A webrtcprivate dial also requires a relay dial, causing some of the expectations on mocks to be incorrectly setup

marten-seemann

Should we move the shared parts to a separate package that's imported by both webrtc and webrtcprivate?

marten-seemann · 2023-10-24T05:35:16Z

 	}
-	if isRelayAddr(a) {
+	if isProtocolAddr(a, ma.P_WEBRTC) {
+		return s.transports.m[ma.P_WEBRTC]


Didn't isRelayAddr already do that? Or phrased differently, the same problem would have occurred for a non-WebRTC address, why do we need special logic for WebRTC now?

marten-seemann · 2023-10-24T05:36:05Z

 type connection struct {
 	pc        *webrtc.PeerConnection
-	transport *WebRTCTransport
+	transport tpt.Transport


marten-seemann · 2023-10-24T05:48:37Z

+var _ net.Addr = NetAddr{}
+
+func (n NetAddr) Network() string {
+	return "libp2p-webrtc"


I don't think this is how it's intended in the net.Addr. The doc says:

name of the network (for example, "tcp", "udp")

marten-seemann · 2023-10-24T05:54:19Z

+	}
+
+	if l.transport.gater != nil && !l.transport.gater.InterceptSecured(network.DirInbound, s.Conn().RemotePeer(), conn) {
+		conn.Close()


Do we need to close the scope here?

marten-seemann · 2023-10-24T06:00:10Z

+		err = fmt.Errorf("failed to read offer: %w", err)
+		return nil, err
+	}
+	if msg.Type == nil || *msg.Type != pb.Message_SDP_OFFER {


Suggested change

if msg.Type == nil || *msg.Type != pb.Message_SDP_OFFER {

if msg.GetType() != pb.Message_SDP_OFFER {

marten-seemann · 2023-10-24T06:01:45Z

+				return
+			}
+
+			if msg.Type == nil || *msg.Type != pb.Message_ICE_CANDIDATE {


Suggested change

if msg.Type == nil || *msg.Type != pb.Message_ICE_CANDIDATE {

if msg.GetType() != pb.Message_ICE_CANDIDATE {

marten-seemann · 2023-10-24T06:02:12Z

+			// Ignore without Debuging on empty message.
+			// Pion has a case where OnCandidate callback may be called with a nil
+			// candidate
+			if msg.Data == nil || *msg.Data == "" {


Suggested change

if msg.Data == nil || *msg.Data == "" {

if len(msg.GetData()) == 0 {

marten-seemann · 2023-10-24T06:03:02Z

+			}
+
+			var init webrtc.ICECandidateInit
+			if err := json.Unmarshal([]byte(*msg.Data), &init); err != nil {


Suggested change

if err := json.Unmarshal([]byte(*msg.Data), &init); err != nil {

if err := json.Unmarshal([]byte(msg.GetData()), &init); err != nil {

marten-seemann · 2023-10-24T06:04:56Z

+
+var _ tpt.Transport = &transport{}
+
+func AddTransport(h host.Host, gater connmgr.ConnectionGater, stunServers []webrtc.ICEServer) (*transport, error) {


Exported function returning unexported type.

Why is this method needed? Why don't we have such a method for other transports?

We have this method for circuit-v2 transport. We need this method for the same reason for webrtc private transport.

The peer address is <relay-addr>/p2p-circuit/webrtc. This transport will first make a relayed connection with the peer, which is why it needs access to the host. After this point it establishes the connection over the signaling stream.

marten-seemann · 2023-10-24T06:15:18Z


-type connMultiaddrs struct {
-	local, remote ma.Multiaddr
+type ConnMultiaddrs struct {


Nit: Don't make this public. Redefine if necessary.

cpeliciari · 2025-03-30T22:43:38Z

@sukunrt What do we need for webrtcprivate to work?

sukunrt · 2025-03-31T14:48:58Z

We'll need to refresh/rewrite this PR, which is significant work. We can reuse parts of the code, but most likely it'll be better to start from scratch and use code from here as much as we can. This is because the webrtc-direct Transport has diverged significantly.

Do you have a usecase that's only solved by this transport?

cpeliciari · 2025-03-31T15:37:36Z

@sukunrt I would like to be able to create private meetings. Then I would share the video between the participants.

simsman2695 · 2025-06-25T00:38:01Z

@sukunrt currently webrtc-direct works only for public networks, my understanding now is that there is no means to peer browser's directly in private networks over the internet. Direct peering for web applications would reduce the overhead for relay requirements for private networks for pubsub. An example would be building a video meeting room with chat and file sharing.

MarcoPolo · 2025-09-19T21:42:22Z

Automatically closing old open PRs. If you're still interested in merging this please open the PR again and rebase from the latest master, otherwise do nothing.

sukunrt requested a review from marten-seemann September 20, 2023 15:34

sukunrt mentioned this pull request Sep 20, 2023

webrtc: private/browser-to-private/server #2009

Open

6 tasks

p-shahi mentioned this pull request Sep 20, 2023

feat: Implement webrtc private transport #2177

Closed

p-shahi linked an issue Sep 20, 2023 that may be closed by this pull request

webrtc: private/browser-to-private/server #2009

Open

6 tasks

sukunrt force-pushed the webrtcprivate/transport branch 9 times, most recently from 912a8ee to 7a47233 Compare September 26, 2023 06:11

sukunrt mentioned this pull request Sep 26, 2023

interop: run for /webrtc #2584

Closed

mxinden mentioned this pull request Sep 27, 2023

webrtc(private-to-private): leave the signalling stream open on successful connection libp2p/specs#580

Closed

sukunrt force-pushed the webrtcprivate/transport branch 2 times, most recently from 6325e4e to 5cad220 Compare October 2, 2023 15:51

sukunrt mentioned this pull request Oct 10, 2023

webrtcprivate: add transport integration tests #2601

Closed

sukunrt force-pushed the webrtcprivate/transport branch 2 times, most recently from 5adc210 to eb5a01d Compare October 12, 2023 09:47

sukunrt force-pushed the webrtcprivate/transport branch from eb5a01d to 962b26a Compare October 13, 2023 08:31

marten-seemann reviewed Oct 13, 2023

View reviewed changes

sukunrt force-pushed the webrtcprivate/transport branch 2 times, most recently from 6ac54f1 to db7c7b2 Compare October 16, 2023 16:40

SgtPooki mentioned this pull request Oct 16, 2023

[🏆 Golden path scenario] Browsers can reliably retrieve content from any modern Kubo node providing content ipfs/helia#255

Open

sukunrt added 4 commits October 23, 2023 17:40

webrtcprivate: add transport

3a54181

webrtcprivate: fix deadline, limit inflight connection requests

8f84284

interim commit

webrtcprivate: integrate connection gater

4f3c12e

webrtcprivate: factor establishConnection out in listener

5ab60e2

sukunrt added 18 commits October 23, 2023 17:40

webrtcprivate: setup addresses on connection properly

10624d1

webrtcprivate: set negotiated=true for the init data channel

0b20ffe

webrtcprivate: don't test for address equality

072b38a

When running with docker, the listener might see a different remote port than the port the dialer observes locally and vice versa

webrtcprivate: setup incoming data channel handlers early

8bf790f

webrtcprivate: fix comments

4622ab2

webrtcprivate: add rcmgr reservation on listen side

1448286

webrtcprivate: setup conn.ConnState correctly

6e464e5

webrtcprivate: reuse protobuf msg structs

8baee51

webrtcprivate: use context for closing listener

255fd3e

webrtcprivate: fix Multiple Dialers test to use listeners for dialing

f41c97c

swarm: integrate webrtc dialing

a2e3df3

libp2p: provide option for enabling webrtcprivate

212159b

libp2p: use webrtc.iceserver for configuration

7e077d3

holepunch: trigger holepunching for peers with webrtc addresses

13081dd

webrtcprivate: integrate transport integration tests

05f5d33

Some failing tests are skipped. They're failing because they require a different testing strategy. A webrtcprivate dial also requires a relay dial, causing some of the expectations on mocks to be incorrectly setup

webrtcprivate: fix bug with gater intercept secured

4a77bae

webrtcprivate: fix rcmgr bugs on listener

6249772

webrtcprivate: advertise /webrtc addresses

bc90626

sukunrt force-pushed the webrtcprivate/transport branch from db7c7b2 to bc90626 Compare October 23, 2023 12:12

marten-seemann reviewed Oct 24, 2023

View reviewed changes

MarcoPolo closed this Sep 19, 2025

	var browserProtocols []int = []int{ma.P_WEBTRANSPORT, ma.P_WEBRTC_DIRECT, ma.P_WS, ma.P_WSS}
	var browserProtocols = []int{ma.P_WEBTRANSPORT, ma.P_WEBRTC_DIRECT, ma.P_WS, ma.P_WSS}

	if msg.Type == nil \|\| *msg.Type != pb.Message_SDP_OFFER {
	if msg.GetType() != pb.Message_SDP_OFFER {

	if msg.Type == nil \|\| *msg.Type != pb.Message_ICE_CANDIDATE {
	if msg.GetType() != pb.Message_ICE_CANDIDATE {

	if msg.Data == nil \|\| *msg.Data == "" {
	if len(msg.GetData()) == 0 {

	if err := json.Unmarshal([]byte(*msg.Data), &init); err != nil {
	if err := json.Unmarshal([]byte(msg.GetData()), &init); err != nil {


		var _ tpt.Transport = &transport{}

		func AddTransport(h host.Host, gater connmgr.ConnectionGater, stunServers []webrtc.ICEServer) (*transport, error) {

Conversation

sukunrt commented Sep 20, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

marten-seemann commented Oct 13, 2023

Uh oh!

marten-seemann left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

sukunrt Oct 24, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

marten-seemann left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cpeliciari commented Mar 30, 2025

Uh oh!

sukunrt commented Mar 31, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

cpeliciari commented Mar 31, 2025

Uh oh!

simsman2695 commented Jun 25, 2025

Uh oh!

MarcoPolo commented Sep 19, 2025

sukunrt commented Sep 20, 2023 •

edited

Loading

sukunrt Oct 24, 2023 •

edited

Loading

sukunrt commented Mar 31, 2025 •

edited

Loading