Skip to contentSkip to conversation

include tests/debug.mk only if exists

PATCH In review Christian Hesse 4 participants 5 messages Dec 3, 2016 – Dec 4, 2016

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 include tests/debug.mk only if exists

From: Christian Hesse <[email protected]>

DKMS packages do not contain tests/debug.mk, so do not include if it does not exist.

---
 src/Makefile | 2 ++
 1 file changed, 2 insertions(+)
src/Makefile +2 -0
index 20947ac..370c2ea 100644
@@ -36,6 +36,8 @@ check:
 cloc: clean
 	cloc $(filter-out compat.h, $(wildcard *.c) $(wildcard *.h))
 
+ifeq ($(wildcard tests/debug.mk),tests/debug.mk)
 include tests/debug.mk
+endif
 
 .PHONY: all module module-debug tools install clean core-cloc check version.h
Signature · 1 line
2.10.2
#1 lore raw

DMVPM appreciation

When Wireguard was first announced, there were several comments like "Can you do DMVPM?"

So
What is DMVPN?
Do we care?
Can we do it better?

DMVPN
http://www.cisco.com/c/en/us/products/security/dynamic-multipoint-vpn-dmvpn/index.html
Is a Cisco product to lets spokes create spoke-to-spoke links in an otherwise spoke-hub VPN architecture.
If we have A<-> HUB <-> B then we can get A <-> B directly. This
unloads the hub.

So how do they do it?
They document 3 things
1) mGRE tunnels.
The tunnel is good old GRE on proto 47.
For every endpoint you create an interface. So hub with 1000 spokes has int gre0 -> int gre999.
Nope, not that dumb. They have a multipoint GRE, mGRE with one interface that can terminate all the tunnels.
--Just like Wireguard.

2)Ipsec. But GRE is encapsulation only, so they protect it with ipsec. Presumably ESP transport mode.
Wireguard doesn't need that.

3)NHRP. https://en.wikipedia.org/wiki/Next_Hop_Resolution_Protocol
This is a repurposed bit of the ATM suite. Sort of DB of remote internet addresses (endpoints) that spokes look up to set up tunnels. Wireguard. So much easier.

There is a 4th part. The actually doing it and the policy for controlling it.

When you look at it, a spoke-hub network that can overlay itself is a mesh network.
So, do we care about DMVPN? No, it's vendor specific.

Can we do it better? Sure, Jason has already done all the hard work.

The key thought going forward is that that security doesn't matter. It's outside the scope.
If A,B,C are /already/ in the VPN, they are /already/ authorised. A tunnel overlay does not impact /security/, it impacts /routing/.

I suggest looking up
Homenet https://www.irif.fr/~jch/software/homenet/howto.html

So lets consider a simplified case
A <-> B <-> C

A is sending a lot of data to C.

Policy triggers starting a direct A <-> C tunnel.

We need public key and endpoint to set up a tunnel.

First. A can talk to C just fine on the VPN. Thats all the authentication required.
A and C swap public keys
Endpoints? you could look up myip.com, but there is one perfect way. A asks B, because if two nodes are peered, they always know the endpoint of their peer.
In this case B knows how to contact A.
B passes that back to A
A then passes that to C along with A's public key

Ok, UDP Rendevous and we have a A <-> C tunnel, no NHRP required. The routing daemon engages and a new route becomes active.

--Dad

#2 lore raw

Re: DMVPM appreciation

I agree that wireguard is potentially a good substrate for what you describe. There are messy details.

My personal goal is to also get it to do good congestion control (adding fq_codel) next year while thinking hard about the problems that tor had in that department.

On Sat, Dec 3, 2016 at 10:07 AM, John Huttley <[email protected]> wro= te:

Quoted 11 lines

When Wireguard was first announced, there were several comments like "Can you do DMVPM?"

So
What is DMVPN?
Do we care?
Can we do it better?

DMVPN
http://www.cisco.com/c/en/us/products/security/dynamic-multipoint-vpn-dmv=

pn/index.html

Quoted 78 lines

Is a Cisco product to lets spokes create spoke-to-spoke links in an otherwise spoke-hub VPN architecture.
If we have A<-> HUB <-> B then we can get A <-> B directly. This
unloads the hub.

So how do they do it?
They document 3 things
1) mGRE tunnels.
The tunnel is good old GRE on proto 47.
For every endpoint you create an interface. So hub with 1000 spokes has int gre0 -> int gre999.
Nope, not that dumb. They have a multipoint GRE, mGRE with one interface that can terminate all the tunnels.
--Just like Wireguard.

2)Ipsec. But GRE is encapsulation only, so they protect it with ipsec. Presumably ESP transport mode.
Wireguard doesn't need that.

3)NHRP. https://en.wikipedia.org/wiki/Next_Hop_Resolution_Protocol
This is a repurposed bit of the ATM suite. Sort of DB of remote internet addresses (endpoints) that spokes look up to set up tunnels. Wireguard. So much easier.

There is a 4th part. The actually doing it and the policy for controlling it.

When you look at it, a spoke-hub network that can overlay itself is a mesh network.
So, do we care about DMVPN? No, it's vendor specific.

Can we do it better? Sure, Jason has already done all the hard work.

The key thought going forward is that that security doesn't matter. It's outside the scope.
If A,B,C are /already/ in the VPN, they are /already/ authorised. A tunnel overlay does not impact /security/, it impacts /routing/.

I suggest looking up
Homenet https://www.irif.fr/~jch/software/homenet/howto.html

So lets consider a simplified case
A <-> B <-> C

A is sending a lot of data to C.

Policy triggers starting a direct A <-> C tunnel.

We need public key and endpoint to set up a tunnel.

First. A can talk to C just fine on the VPN. Thats all the authentication required.
A and C swap public keys
Endpoints? you could look up myip.com, but there is one perfect way. A asks B, because if two nodes are peered, they always know the endpoint of their peer.
In this case B knows how to contact A.
B passes that back to A
A then passes that to C along with A's public key

Ok, UDP Rendevous and we have a A <-> C tunnel, no NHRP required. The routing daemon engages and a new route becomes active.

--Dad

Quoted 4 lines

_______________________________________________
WireGuard mailing list
[email protected]
https://lists.zx2c4.com/mailman/listinfo/wireguard

--=20
Dave T=C3=A4ht
Let's go make home routers and wifi faster! With better software!
http://blog.cerowrt.org

#3 lore raw

Re: DMVPM appreciation

In reply to John Huttley

Hmm...
Really good high level theory ...

Quoted 9 lines · John Huttley

So lets consider a simplified case
A <-> B <-> C

A is sending a lot of data to C.

Policy triggers starting a direct A <-> C tunnel.

We need public key and endpoint to set up a tunnel.

Don't forget we need two more things:
* A --> C (over UDP)
* C --> A (over UDP)

Throw a few weird NAT/PAT and other ACLs in between and try again.

Quoted 3 lines

First. A can talk to C just fine on the VPN. Thats all the authentication required.

Yes, but in the background A only talks to B over UDP... and B to C over UDP.

Quoted 9 lines

A and C swap public keys
Endpoints? you could look up myip.com, but there is one perfect way. A asks B, because if two nodes are peered, they always know the endpoint of their peer.
In this case B knows how to contact A.
B passes that back to A
A then passes that to C along with A's public key

Quoted 3 lines

Ok, UDP Rendevous and we have a A <-> C tunnel, no NHRP required. The routing daemon engages and a new route becomes active.

I think there may be cases where such situation is possible and highly desirable.
The logic being, try to talk directly (using the protocol described above), if it doesn't work, continue talking via B.
Reminds me of STUN and friends.

And an alternative will be centralized database/service advertising public keys and endpoints (think PGP keyserver).
And then comes the WoT and then... Jason will shoot us for bringing the complexity he is so fond of avoiding (and I am supporting him) :-D

Since this functionality can be implemented outside of Wireguard, as a simple script (I guess 5 lines of Bash, but will leave the challenge open), it is all a matter of convenience.
Nothing can be gained by "building it in", may be except tiny convenience. If you try to script it, then provide clean documentation/test cases, I am sure it can be included in
https://git.zx2c4.com/WireGuard/plain/contrib/examples/

Also consider the trust model you are changing, because security DOES matter :-) As an example, think buying on Amazon from 3rd party, vs. talking to them directly:
(1st -> 2nd case may be possible, 2nd ->1st is highly improbable)

* 1st case is more secure, but more expensive
* 1st provides a way to find your party, plus it provides some mediatior/trust relationship (at cost)
* How often did you buy something via Amazon an then went directly to the seller to buy more?

It is all about balancing security/cost and convenience of setup a transaction, within some (implicit) trust/insurance model.

Cheers,
Kalin.

#4 lore raw

Re: DMVPM appreciation

In reply to John Huttley

Hmm...
Really good high level theory ...

Quoted 5 lines

Don't forget we need two more things:
* A --> C (over UDP)
* C --> A (over UDP)

Throw a few weird NAT/PAT and other ACLs in between and try again.

In one direction there is no need, because that's how we established the tunnel. See below, we could make exposed endpoints a requirement.

Quoted 5 lines
Quoted 3 lines

First. A can talk to C just fine on the VPN. Thats all the authentication required.

Yes, but in the background A only talks to B over UDP... and B to C over UDP.

Sorry I don't get what you mean. Anyone /could/ talk to anyone, firewalls permitting.

Quoted 5 lines

I think there may be cases where such situation is possible and highly desirable.
The logic being, try to talk directly (using the protocol described above), if it doesn't work, continue talking via B.
Reminds me of STUN and friends.

Actually we can simplify it further. Just make it a requirement of overlaying that you have a valid listener. Then no UDP magic. Overly works or it doesn't.

Quoted 4 lines

And an alternative will be centralized database/service advertising public keys and endpoints (think PGP keyserver).
And then comes the WoT and then... Jason will shoot us for bringing the complexity he is so fond of avoiding (and I am supporting him) :-D

Agreed. I don't think Jason will be more than mildly interested and it shouldn't be a part of Wireguard itself.

Quoted 7 lines

Since this functionality can be implemented outside of Wireguard, as a simple script (I guess 5 lines of Bash, but will leave the challenge open), it is all a matter of convenience.
Nothing can be gained by "building it in", may be except tiny >convenience. If you try to script it, then provide clean documentation/test cases, I am sure it can be included in
https://git.zx2c4.com/WireGuard/plain/contrib/examples/

Agreed.

Quoted 5 lines

Also consider the trust model you are changing, because security DOES matter :-)
As an example, think buying on Amazon from 3rd party, vs. talking to them directly:
(1st -> 2nd case may be possible, 2nd ->1st is highly improbable)

If you are on the VPN, you are by definition, trusted as much as anyone is.

"Improbable" is a policy decision. Overlaying might not even be of benefit. If C is on bad, slow connection, an overlay to C isn't going to help. Policy needs to figure that out.

Quoted 5 lines

* 1st case is more secure, but more expensive
* 1st provides a way to find your party, plus it provides some mediatior/trust relationship (at cost)
* How often did you buy something via Amazon an then went directly to the seller to buy more?

This isn't the use case I'm considering. I'm not quite sure what you mean.

Quoted 2 lines

It is all about balancing security/cost and convenience of setup a transaction, within some (implicit) trust/insurance model.

Definitely don't know what you mean.

Regards
-Dad