The most important for me is to keep the P2P feature(auto NAT traverse) of wireguard while wrapping the wireguard UDP packets into udp2raw Fake TCP/Raw TCP packets. I think that keeping p2p feature of wireguard in wireguard+udp2raw is more complex or even impossible than Wireguard itself supports Fake TCP/Raw TCP packets: https://github.com/wangyu-/udp2raw-tunnel/issues/212

On Oct 2, 2018 11:11, "Breus Blaauwendraad" <b.blaauwendraad@gmail.com> wrote:
Hello StarBrilliant,

I think that what Kexianbin is trying to make clear is that it would be nice if  native TCP support would be added in WireGuard.
He (probably) currently uses udp2raw as an alternative, because an UDP-only VPN is sadly enough not working in every situation.
With him, there have been others pitching ideas on how to wrap the WireGuard UDP traffic in TCP, which in my opinion, is a poor idea.

So, why would UDP for a VPN service not be enough?

First of all, most VPN traffic obfuscation techniques require a TCP connection instead of UDP.
VPN usage is particularly useful in countries were Internet censorship is applied.
Since obfuscating VPN traffic comes with performance overhead, wrapping the UDP in TCP and obfuscating this would kill the performance advantage WireGuard has over other protocols.

However, what is arguably more important: various corporate firewalls (hotels/restaurants/offices) are known to block UDP (for random ports besides the standard DNS 53 port etc.). 
In these instances, it would be useful to have the possibility to build a fall back to TCP instead of UDP in a VPN service (or even the protocol itself).

Also, various researchers I have spoken to told me that sometimes they need a more reliable (TCP) connection, which is then more important than the speed UDP provides.

I'm looking into the possibilities to deploy the WireGuard VPN protocol for an existing VPN service currently using OpenVPN, but I keep stumbling on the problem that WIreGuard does not support TCP.
I (think I) have read all posts regarding TCP for WireGuard, but it seems that there do not exists a good way to tackle this problem.

Could someone tell whether or not TCP would be a future additional option for WireGuard, and why (not)?

Also, do you think there actually does exist a neat and fitting solution for the problems I described?

Kind regards,

Breus Blaauwendraad


On Tue, 25 Sep 2018 at 17:54, StarBrilliant <coder@poorlab.com> wrote:
On Tue, Sep 25, 2018 at 1:17 PM KeXianbin(http://diyism.com)
<kexianbin@diyism.com> wrote:
>
> Currently, I'm using udp2raw-tunnel to transform wireguard udp traffic into raw tcp (config files as follows),
> It's very stable on my home network than using wireguard alone,
> But if we can integrate RAW TCP feature into wireguard, it would significantly improve performance and stability for end users.
>
>
> from: https://gist.github.com/diyism/1b80903a83776675031c73ae499438d8#file-wireguard_config-txt-L145
>
> $wget https://github.com/wangyu-/udp2raw-tunnel/releases/download/20180830.2/udp2raw_binaries.tar.gz
> $tar xzvf udp2raw_binaries.tar.gz
> $sudo cp udp2raw_amd64 /usr/bin/
> $sudo udp2raw_amd64 -c -l127.0.0.2:24448 -r<server ip>:24447 -a
> $cat /etc/wireguard/wg0.conf
> [Interface]
> PrivateKey = <client privkey>
> Address = 10.0.0.3/32
> ListenPort = 24447
> MTU = 1300
> PostUp = ip route add 10.0.0.0/24 dev wg0 && wg set wg0 peer <server pubkey> allowed-ips 0.0.0.0/0
> PostDown = ip route del 10.0.0.0/24
>
> [Peer]
> #10.0.0.1
> PublicKey = <server pubkey>
> Endpoint = 127.0.0.2:24448
> #AllowedIPs = 0.0.0.0/0
>
> $sudo wg-quick down wg0 ; sudo wg-quick up wg0
> $ping 10.0.0.1
> 64 bytes from 10.0.0.1: icmp_seq=2113 ttl=64 time=183 ms
> $sudo ip route add 104.24.0.0/16 dev wg0
> $ping myip.ipip.net
> PING myip.ipip.net (104.24.20.50) 56(84) bytes of data.
> 64 bytes from 104.24.20.50 (104.24.20.50): icmp_seq=1 ttl=60 time=185 ms
> $curl http://myip.ipip.net
> IP:<server ip>
>
> #take care, "MTU = 1300" in wg0.conf is needed when wireguard over udp2raw, or else most https requests will be blocked because of mtu problem.


Hello Kexianbin,

This is an UNOFFICIAL response to your question. (But I think the
official developers may have similar answers.)

Wireguard probably will not accept an official integration to udp2raw.
The reasons are:

1) Wireguard wants to keep their kernel part code minimized, therefore
easy for security auditing, and less bugs.The UDP protocol is actually
very simple and straightforward. (By the way, if you intended to use
Wireguard in China, be informed that this is a protocol that is very
easy to block by the ISP.)

2) I have read the source code of udp2raw. To be frank, the code is of
very low quality. For this reason, I don't think udp2raw would be
integrated into Wireguard unless it's rewritten.

3) Udp2raw is not suitable for everyone or for every country. For
example udp2raw may have problems passing middleboxes, which is common
among satellite ISPs in Oceania. Middleboxes break and resemble TCP
segments thus make udp2raw literally unusable. Also it is not
congestion friendly (by design), so a massive deployment may affect
the global Internet ecology.

However the good news is, Wireguard provides an open control interface
(see https://www.wireguard.com/xplatform/ ). By utilizing this
interface, we can develop an alternate frontend application other than
the official command "wg", that automatically sets up the kernel
Wireguard kernel part and a userland udp2raw part, packaged as one
application.


My words for Wireguard developers:

1) In case you may not know the udp2raw protocol, here is a
description. Some ISPs in certain countries have strange QoS strategy
that deprioritize UDP packets during network congestion, resulting a
50% loss rate or more for UDP. The udp2raw protocol simulates a
three-way TCP handshake and add TCP header to UDP packets so they will
not be dropped. This protocol does not do congestion control or rate
control, neither does it understand any TCP semantics. It's a dirty
hack for dirty ISP, not suitable for everyone, but overwhelmingly
useful in certain countries.

2) Wireguard currently does not support binding to localhost. This is
required for any third-party plugins upon Wireguard to work. We might
need to consider binding to localhost an important feature to go in
the near future.


Best regards,
StarBrilliant
_______________________________________________
WireGuard mailing list
WireGuard@lists.zx2c4.com
https://lists.zx2c4.com/mailman/listinfo/wireguard

_______________________________________________
WireGuard mailing list
WireGuard@lists.zx2c4.com
https://lists.zx2c4.com/mailman/listinfo/wireguard