Hacker News Reader

Although the article is confused regarding dynamic IP addresses your solution of identifying the endpoint using a DNS name is also not a complete solution when using Wireguard. The Wireguard tool, wg, resolves the DNS name and then sets the ip address for the endpoint which the wireguard protocol implementation then uses. Since the wireguard protocol implementation doesn't do any DNS resolution if the endpoint address changes even if the DNS record is updated wireguard will never know. Instead another tool needs to check the DNS record of the endpoint periodically either at some fixed interval and/or based upon the TTL of the DNS record. If the record changes it then needs to update the configuration of the wireguard protocol implementation.

This is not always required. WireGuard peers can roam freely very similarly to Mosh (if one peer receives an authenticated message from one of its peers from a different IP address, it will update the gateway of that peer and will send all subsequent packets to that IP address).

But to be fair, this doesn't work perfectly (especially if you have certain firewall rules that block UDP packets if they're not part of the what the firewall thinks is the same "connection", if both peers change their IP at the same time, or if the peer whose IP changed was rebooted or reconfigured before the other peer received a packet from its new IP addreds) but it really does work surprisingly well. For those (important) corner-cases, I would imagine that higher-level tooling such as NetworkManager would do the DNS checks you're talking about.

Right, the free roaming tends to be on "the other side" though. This is a bit confusing to talk about because Wireguard does not have a concept of "client" and "server" and either side can nominally initiate the tunnel. However, in roaming configurations, typically only one side will have the information to initiate the connection (endpoint) and so there's a tendency to call this the "client."

The non-roaming end (end with a known endpoint) will update the IP of the roaming end every time a handshake occurs. The issue the OP is getting at though is when both ends are roaming. In this case the, er, "more roaming" peer needs some way to find the other peer. There are a lot of technically interesting ways that we could solve this problem, but in practice it's pretty much always solved by dynamically updating a DNS record. I would say Wireguard handles this scenario about as well as anything else (maybe a little worse, see sibling comment).

Another approach would be using a "meet-me" relay with a fixed IP, such as TURN. This also has the advantage of handling NAT on both sides (which would be a common issue with a roaming-to-roaming situation). I can't think of any reason it's impossible to run Wireguard through TURN but I can't find any examples of anyone doing so, likely in part because the TURN client is usually embedded in the application and is a bit complicated. Still, might be a hobby project to implement it for Wireguard.

This is an issue, but in typical dynamic IP situations the IP tends to only change when there's some sort of disruption that prevents renewal and not very frequently (e.g. I've held a Comcast consumer IP for over a year before), which tends to mitigate this effect since the IP changing "out from underneath" the tunnel won't occur. Most dynamic DNS schemes, with DNS caching added on top, also tend to result in pretty long (hour +) delays on IP updates to DNS which further make this a bit of a non-issue.

That said, I do believe both IPv6 and OpenVPN will have this exact same problem - IP change will not be reflected until something forces setup to repeat. The difference would be that these, at least I do believe, will re-resolve on retries. Wireguard doesn't really have a sense of "retrying the connection" which makes it sort of philosophically difficult for it to do so.

One very common scenario where WireGuard fails - you have a WG persistent VPN enabled and you actually move into the network that has the VPN endpoint. The IP of WG endpoint changes from a public one to a local one.

This keeps happening to me all the time, and it is no problem.

Once I move into the network that has the respective VPN endpoint, the subnets that are routable via wg become routable via ethernet/wifi which have higher priority, so the traffic will go directly.

I don't have this problem. If the endpoint is accessible through the public address even in the internal network then everything just works.

https://github.com/WireGuard/wireguard-tools/tree/master/con...

It's funny that the author mentions IPv6 as an argument when IPFire itself does not support it.

The article only directly compares WireGuard with IPsec and OpenVPN. IPFire was only mentioned twice in passing.

Yet still it spreads lies. There's not only 4 pieces of information needed to establish IPsec (they even cannot count as they mention 5 pieces). In addition to named remote and local IP addresses, subnets and a pre-shared key you also need: * mode of operation (transport or tunnel) * exchange mode (base, main, aggressive or ikev2) * policy mode of operation (encrypt and/or protect - yes encryption is optional with IPsec) * Phase 1 parameters (hashing algorithm(s) + encryption algorithm(s)), key lifetime in bytes or seconds * Whether you/remote uses a PFS (if so, then also DH Group) * Phase 2 parameters (hashing algorithm(s) + encryption algorithm(s)), key lifetime in bytes or seconds * Whether or not to use NAT traversal (changes the protocols and ports needed to be opened on the firewall(s))

If you don't know some of these parameters chances are you won't be able to establish the tunnel no matter how hard you try.

Plus you need to know all the vendor quirks as establishing IPsec from Cisco to Fortigate or Sophos is not straightforward.

Then you may end up playing with NAT-exclusions, SNAT/DNAT if both sides of the tunnel have the same or overlapping IP ranges.

Setting up IPsec is definitely not an exchange of 4 parameters...

I think the point is that if you “WireGuard-ify” IPSec and just fix all those parameters (e.g. take the vendor defaults) then you’re left with 5 pieces of information.

If the value of WG is just that it forces all these parameters to be static then the best thing we could do is come up with the “WireGuard” IPSec profile call it a day.

Are you saying that fixed parameters chosen for no apparent reason by the WireGuard developers are better than modularity and interoperability of IPsec? What if in a few months ChaCha20 gets proven insecure due to collisions found or easy factorisation? What can WireGuard offer to mitigate that? Shouldn't then also browsers implement only TLS1.3 and ed25519 ciphers because they are currently the most secure?

> .. for no apparent reason by the WireGuard developers...

It's a very weird assertion that the Wireguard devs just threw the software together at random, choosing parameters for "no apparent reason".

> What if in a few months ChaCha20 gets proven insecure due to collisions found or easy factorisation? What can WireGuard offer to mitigate that? Shouldn't then also browsers implement only TLS1.3 and ed25519 ciphers because they are currently the most secure?

TLS is famously susceptible to downgrade attacks. (https://blog.ivanristic.com/2013/09/is-beast-still-a-threat.... etc.)

Ultimately, it's a value judgement. You can either have resistance to downgrade attacks (which have themselves proven to be quite problematic), or you have interoperability across multiple versions and configurations of endpoints. Increase the configurability of the protocol, and you massively increase the complexity. Increased complexity means increased testing burden and ultimately increased risk.

If Wireguard needs to be fixed in a backwards-incompatible way, then we'll find ourselves with a new version of Wireguard that doesn't work with the old version.

> new version of Wireguard that doesn't work with the old version.

Right, and maybe this is actually in improvement in security overall but it just externalizes the downgrade attack since once there are multiple versions of WG floating around with different vendors/clients only supporting a specific version you end up similarly vulnerable since you need to run multiple WG endpoints of different versions.

And since it’s a kernel module you’ve made the hassle of doing so very annoying compared to one line in a config file.

IPSec feels messy and complex specifically because the world is messy and complex. WG is fantastic and I love it dearly for “the 90% case” where I have total control over all peers.

>TLS is famously susceptible to downgrade attacks.

The article is from 8 years ago so I would suggest that the fame has mostly faded. TLS downgrade attacks are not a thing in practice. A system with non-upgradable and broken crypto is much worse than something that requires a MITM attack to get at the broken crypto. I am not sure why there are opinions to the contrary. In either case you still have to fix things. The non-upgradable case will just be much much harder.

Absolutely not! What I mean is that I don’t think that IPSec having a bunch of tunables like this should count as harshly as the parent implies when it comes to configuration complexity because WireGuard isn’t all that different — the tunables are just chosen earlier in a way that’s not user-visible.

Sounds like wireguard is just positioning into corporate clients.