Skip to contentSkip to conversation

Wireguard address binding - how to fix?

DISCUSSION Nico Schottelius 7 participants 12 messages May 14, 2024 – Nov 18, 2024

lore
Share & tools
lore
#0 lore raw

Hello,

the problem of wireguard being unable to bind to an IP address are hunting us over and over again. Isolation via namespace is not always a solution and I would consider it a hack in the first place.

To the wireguard authors: what needs to be done to add IP address binding so that packets are always sent from a specific IP address?

Virtually every other network software out there has that option for a good reason, only wireguard does not support it and thus does not work properly on devices with multiple IP addresses.

What does it take to add IP address binding in wireguard?

Best regards,

Nico

Signature · 1 line
Sustainable and modern Infrastructures by ungleich.ch
#2 lore raw

Hello Jason,

do you mind applying the patch from Daniel? Or is there anything wrong with it?

Daniel: amazing work, I was not aware that you have already put in the hard work, thank you so very much!

The world (*) is suffering because of the lack of IP address binding in wireguard.

Best regards,

Nico

(*) With world I refer to every engineer that needs to run wireguard in non-trivial situations with multiple IP addresses on one host, which is extremely common for anything that routes.

Quoted 8 lines · Daniel Gröber
Quoted 2 lines · Nico Schottelius

To the wireguard authors: what needs to be done to add IP address binding so that packets are always sent from a specific IP address?

Implemented, but wating for Jason to respond:
https://lore.kernel.org/netdev/[email protected]/T/

--Daniel

Signature · 2 lines
-- 
Sustainable and modern Infrastructures by ungleich.ch
#3 lore raw

Den tis 21 maj 2024 kl 09:50 skrev Nico Schottelius <[email protected]>:

Quoted 11 lines

Hello Jason,
do you mind applying the patch from Daniel? Or is there anything wrong with it?

Daniel: amazing work, I was not aware that you have already put in the hard work, thank you so very much!

The world (*) is suffering because of the lack of IP address binding in wireguard.

(*) With world I refer to every engineer that needs to run wireguard in non-trivial situations with multiple IP addresses on one host, which is extremely common for anything that routes.

Well, the main reason for wg to NOT do anything special is because routing generally is done by looking at the destination ip and then using the routing rules which the kernel uses to tell the packet which interface is considered "best" in order to reach that ip, which is why icmp and udp acts like this. I am certain that many engineers hope/think it would be more logical (or fit their designs better) if it responded on the same interface as the packet came in or whatever but the reason for wg to act like it does is because this is how connection-less packets have been acting since ages. The point that many engineers design wg endpoints "the wrong way" is probably a fault of education when it comes to how TCP/IP stacks did and do manage IP output.

I have zero power to decide anything regarding how wg chooses to implement a workaround for IP being designed the way it is designed, but I can at least see why it hasn't immediately been accepted even if it would make some engineers suffer(*) less, at least in the short term.

There are a lot of workarounds for the times when you did design the udp service as if it would act like tcp does in multihomed situations, which includes firewall tricks, or VRF/namespace/routing-domains to make sure the inner and outer address-pairs for the wg tunnel see different views of the network to handle this without changing how wg sends packets.

Signature · 1 line
May the most significant bit of your life be positive.
#4 lore raw

Hello Janne,

Quoted 8 lines · Janne Johansson

Den tis 21 maj 2024 kl 09:50 skrev Nico Schottelius <[email protected]>:

Quoted 6 lines

Hello Jason,
do you mind applying the patch from Daniel? Or is there anything wrong with it?

Daniel: amazing work, I was not aware that you have already put in the hard work, thank you so very much!

Quoted 8 lines
Quoted 5 lines

The world (*) is suffering because of the lack of IP address binding in wireguard.

(*) With world I refer to every engineer that needs to run wireguard in non-trivial situations with multiple IP addresses on one host, which is extremely common for anything that routes.

Well, the main reason for wg to NOT do anything special is because routing generally is done by looking at the destination ip and then

No. Generally speaking that is incorrect.
It is not special to reply with the same IP address.

Generally speaking, when you have systems with multiple IP addresses you want to be able to steer the binding to an IP address. And even if you don't do that, you reply with the same IP address you have been contacted with. Wireguard does neither of it at the moment. I have written this already many times on this list, but the reason is very easy:

- A connection is initiated from device A, connecting to router B on IP adddress a.b.c.d
- The packet is correctly received by router B
- The router replies incorrectly with address f.d.g.h
- The reply packet is correctly blocked at the firewall of device A, because it comes from a random, unknown IP address

This is the basic 101 of networking is to reply with the same address you have been contacted with, there is no discussion necessary. The whole world does it, even A-patch-y httpd (*) supports it. Since 1980 or so.

Routing choices are independent of that, replying with the same IP address is a standard behaviour.

Nico

(*) As does ssh, nginx, ipsec protocols, openvpn, any rails application, any python application - I am not sure which software that binds to a socket does not support it, with the exception of wireguard.

Signature · 2 lines
-- 
Sustainable and modern Infrastructures by ungleich.ch
#5 lore raw

The reason wireguard does it like this I think is because when designing it there was no thought given to any client,server scenario.

Both sides are behaving like clients that can jump between IPs at any time. This is a flawed concept given that in 90% of scenarios there is at least one side acting as a server on a static ip. Unless the server side is a home user on dynamic ip and rebinding could be difficult.

I've also given a bit of thought to the security aspect of this for VPN providers. Since a remote party can override the configured "Endpoint" if there was a scenario where vpn provider privkeys are compromised. The attacker can then, by knowing the connecting clients ip, get him to shift over the tunnel to their server and perform a long term, most likely undetected, mitm attack.

Anyway. I've waited for this binding option for years. It's insane to me it gets ignored.

One product is for example Mikrotik hardware. They don't want to implement third party patches so they are waiting for this bind-patch to be included in the kernel. Until then we're forced to use OpenVPN in our setups.

Quoted 50 lines · Nico Schottelius

Hello Janne,

Quoted 8 lines · Janne Johansson

Den tis 21 maj 2024 kl 09:50 skrev Nico Schottelius <[email protected]>:

Quoted 6 lines

Hello Jason,
do you mind applying the patch from Daniel? Or is there anything wrong with it?

Daniel: amazing work, I was not aware that you have already put in the hard work, thank you so very much!

Quoted 7 lines
Quoted 5 lines

The world (*) is suffering because of the lack of IP address binding in wireguard.

(*) With world I refer to every engineer that needs to run wireguard in non-trivial situations with multiple IP addresses on one host, which is extremely common for anything that routes.

Well, the main reason for wg to NOT do anything special is because routing generally is done by looking at the destination ip and then

No. Generally speaking that is incorrect.
It is not special to reply with the same IP address.

Generally speaking, when you have systems with multiple IP addresses you want to be able to steer the binding to an IP address. And even if you don't do that, you reply with the same IP address you have been contacted with. Wireguard does neither of it at the moment. I have written this already many times on this list, but the reason is very easy:

- A connection is initiated from device A, connecting to router B on IP adddress a.b.c.d
- The packet is correctly received by router B
- The router replies incorrectly with address f.d.g.h
- The reply packet is correctly blocked at the firewall of device A, because it comes from a random, unknown IP address

This is the basic 101 of networking is to reply with the same address you have been contacted with, there is no discussion necessary. The whole world does it, even A-patch-y httpd (*) supports it. Since 1980 or so.

Routing choices are independent of that, replying with the same IP address is a standard behaviour.

Nico

(*) As does ssh, nginx, ipsec protocols, openvpn, any rails application, any python application - I am not sure which software that binds to a socket does not support it, with the exception of wireguard.

#6 lore raw

Hey Sebastian,

Quoted 3 lines · Sebastian Hyrvall

[...]
Anyway. I've waited for this binding option for years. It's insane to me it gets ignored.

I have to second this very strongly.

Quoted 4 lines

One product is for example Mikrotik hardware. They don't want to implement third party patches so they are waiting for this bind-patch to be included in the kernel. Until then we're forced to use OpenVPN in our setups.

As well as this one. As written in the previous mail, virtually every established networking software out there supports IP address binding, with the exception of wireguard.

Dear wireguard authors, can be go into the direction of accepting an already written patch (thanks again Daniel) or if that patch is not suitable for some reason at least discuss what needs to be changed so that IP address binding can make its way into wireguard?

BR,

Nico

Signature · 1 line
Sustainable and modern Infrastructures by ungleich.ch
#7 lore raw
Quoted 4 lines
Quoted 3 lines · Sebastian Hyrvall

[...]
Anyway. I've waited for this binding option for years. It's insane to me it gets ignored.

I remembered how exciting when I tested wireguard at 2017. until I asked muti-home question in the list.
wiregurad is beautiful,elegant,fast but not easy to get along with. openvpn is not so amazing but it can get the job done.

#8 lore raw
Quoted 4 lines · d tbsky

I remembered how exciting when I tested wireguard at 2017. until I asked muti-home question in the list.
wiregurad is beautiful,elegant,fast but not easy to get along with. openvpn is not so amazing but it can get the job done.

Nice summary, hits the nail quite well.

Jason, do you mind having a look at the submitted patches for IP address binding and comment on them? Or alternatively can you give green light for generally moving forward so that a direct inclusion in the Linux kernel would be accepted?

Best regards,

Nico

Signature · 2 lines
-- 
Sustainable and modern Infrastructures by ungleich.ch
#9 lore raw

Jason,

may I shortly ask what your opinion is on the patch and whether there is a way forward to make wireguard usable on systems with multiple IP addresses?

Best regards,

Nico

Quoted 16 lines · Nico Schottelius
Quoted 4 lines · d tbsky

I remembered how exciting when I tested wireguard at 2017. until I asked muti-home question in the list.
wiregurad is beautiful,elegant,fast but not easy to get along with. openvpn is not so amazing but it can get the job done.

Nice summary, hits the nail quite well.

Jason, do you mind having a look at the submitted patches for IP address binding and comment on them? Or alternatively can you give green light for generally moving forward so that a direct inclusion in the Linux kernel would be accepted?

Best regards,

Nico

Signature · 2 lines
-- 
Sustainable and modern Infrastructures by ungleich.ch
#11 lore raw
In reply to Janne Johansson

Hello Janne,

I know that most people dealing with wg do not really understand how a box works that has more than one IP and one interface.
A simple question: an interface has 100 IP addresses, how do you choose in wg on which IP your tunnel end is?
Can you please (all wg infected) understand that about every box that has a redundancy setup has several IPs per interface and you want wg connect to the _right_ one, because this is the one taken over in case of failover. The question is not about interface for routing, the question is about LISTEN_ADDR.

I always thought it cannot get worse, but then came wg... --
Regards,
Stephan

Quoted 44 lines · On Tue, 21 May 2024 13:11:00 +0200 Janne Johansson <[email protected]> wrote:

Den tis 21 maj 2024 kl 09:50 skrev Nico Schottelius <[email protected]>:

Quoted 7 lines

Hello Jason,
do you mind applying the patch from Daniel? Or is there anything wrong with it?

Daniel: amazing work, I was not aware that you have already put in the hard work, thank you so very much!

Quoted 6 lines

The world (*) is suffering because of the lack of IP address binding in wireguard.

(*) With world I refer to every engineer that needs to run wireguard in non-trivial situations with multiple IP addresses on one host, which is extremely common for anything that routes.

Well, the main reason for wg to NOT do anything special is because routing generally is done by looking at the destination ip and then using the routing rules which the kernel uses to tell the packet which interface is considered "best" in order to reach that ip, which is why icmp and udp acts like this. I am certain that many engineers hope/think it would be more logical (or fit their designs better) if it responded on the same interface as the packet came in or whatever but the reason for wg to act like it does is because this is how connection-less packets have been acting since ages. The point that many engineers design wg endpoints "the wrong way" is probably a fault of education when it comes to how TCP/IP stacks did and do manage IP output.

I have zero power to decide anything regarding how wg chooses to implement a workaround for IP being designed the way it is designed, but I can at least see why it hasn't immediately been accepted even if it would make some engineers suffer(*) less, at least in the short term.

There are a lot of workarounds for the times when you did design the udp service as if it would act like tcp does in multihomed situations, which includes firewall tricks, or VRF/namespace/routing-domains to make sure the inner and outer address-pairs for the wg tunnel see different views of the network to handle this without changing how wg sends packets.

Signature · 1 line
May the most significant bit of your life be positive.
#10 lore raw
In reply to Sebastian Hyrvall

Hello Sebastian,

great to hear that you came to the same conclusion regarding the security of wg. You have never read me on the list as I get continuously censored away from it.
Obviously because my first post 3 years ago was exactly this:

#########################

From: Stephan von Krawczynski <[email protected]>
To: [email protected]
Subject: How to set source ip of wireguard packets?
Date: Tue, 20 Oct 2020 13:27:46 +0200
Organization: ith Kommunikationstechnik GmbH
X-Mailer: Claws Mail 3.17.6 (GTK+ 2.24.32; x86_64-pc-linux-gnu)

Hello all,

configuring wireguard for the first time I ran into a question I could not find any answer in the docs, so:
Lets assume both client and server have several IPs on their outgoing interface. How do you setup wireguard so that a specific IP is used as source ip in the outgoing encrypted packets?
This is important in failover-setups where a VIP moves between some physical hosts and we don't want to listen on the physical IPs but only the VIP ...

I would have expected this to be specified in "ListenPort" optionally like "ListenPort = IP:PORT", but that does not work currently.

--
Regards,
Stephan

###############################

Ever since I was censored away. I came to the same conclusion, which is that wg is in fact only an orchestrated way to get vpn over the world completely open to mitm attacks, just like your conclusion.
Even better the very same people always talk about the insecurity of PPTP, completely dumping the fact that every PPP since the very beginning has a (script) interface where you can check the connecting _IPs_ and users. We don't talk about the encryption security here, because I bet that we all cannot analyse this point in wg, away from the fact the special implementation should be deeply reviewed, too.
Therefore we judge the current wg as high security risk which should not be used in professional environment. A VPN where you cannot even verify the connecting IP? Gimme a break ...
Being one of the last "forkers" of the cipe project where all wg problems were solved decades ago we do really wonder why this project gets so much code support being so badly designed from the start. That is why we think it is orchestrated by interested parties.
While we're at it, the very same seems to be going on in the certificate business where it is prevented for decades by the browser people to use self-signed certificates which could be verified dead easy by a pointer in the corresponding domains' dns-record.
We are not amused.
--
Regards
Stephan

Quoted 78 lines · On Tue, 21 May 2024 21:11:31 +0700 Sebastian Hyrvall <[email protected]> wrote:

The reason wireguard does it like this I think is because when designing it there was no thought given to any client,server scenario.

Both sides are behaving like clients that can jump between IPs at any time. This is a flawed concept given that in 90% of scenarios there is at least one side acting as a server on a static ip. Unless the server side is a home user on dynamic ip and rebinding could be difficult.

I've also given a bit of thought to the security aspect of this for VPN providers. Since a remote party can override the configured "Endpoint" if there was a scenario where vpn provider privkeys are compromised. The attacker can then, by knowing the connecting clients ip, get him to shift over the tunnel to their server and perform a long term, most likely undetected, mitm attack.

Anyway. I've waited for this binding option for years. It's insane to me it gets ignored.

One product is for example Mikrotik hardware. They don't want to implement third party patches so they are waiting for this bind-patch to be included in the kernel. Until then we're forced to use OpenVPN in our setups.

Quoted 53 lines · Nico Schottelius

Hello Janne,

Quoted 9 lines · Janne Johansson

Den tis 21 maj 2024 kl 09:50 skrev Nico Schottelius <[email protected]>:

Quoted 7 lines

Hello Jason,
do you mind applying the patch from Daniel? Or is there anything wrong with it?

Daniel: amazing work, I was not aware that you have already put in the hard work, thank you so very much!

Quoted 8 lines
Quoted 6 lines

The world (*) is suffering because of the lack of IP address binding in wireguard.

(*) With world I refer to every engineer that needs to run wireguard in non-trivial situations with multiple IP addresses on one host, which is extremely common for anything that routes.

Well, the main reason for wg to NOT do anything special is because routing generally is done by looking at the destination ip and then

No. Generally speaking that is incorrect.
It is not special to reply with the same IP address.

Generally speaking, when you have systems with multiple IP addresses you want to be able to steer the binding to an IP address. And even if you don't do that, you reply with the same IP address you have been contacted with. Wireguard does neither of it at the moment. I have written this already many times on this list, but the reason is very easy:

- A connection is initiated from device A, connecting to router B on IP adddress a.b.c.d
- The packet is correctly received by router B
- The router replies incorrectly with address f.d.g.h
- The reply packet is correctly blocked at the firewall of device A, because it comes from a random, unknown IP address

This is the basic 101 of networking is to reply with the same address you have been contacted with, there is no discussion necessary. The whole world does it, even A-patch-y httpd (*) supports it. Since 1980 or so.

Routing choices are independent of that, replying with the same IP address is a standard behaviour.

Nico

(*) As does ssh, nginx, ipsec protocols, openvpn, any rails application, any python application - I am not sure which software that binds to a socket does not support it, with the exception of wireguard.