Selecting and Hardening Remote Access VPN Solutions [pdf]

With something like OpenVPN, they seem to have strong issues with some exploitable parts of the startup/bring-up process. There are two ways to look at this:

It's either too good for them or it's bad. :)

> I imagine these solutions are popular because they are more likely to function through corporate firewalls, where IPsec might be blocked.

I implement IPSec at work, because we need the real deal.

Though for my personal projects, my infra is using OpenVPN. The reason is simple: it's so much easier to configure and understand, and the desktop clients are usually more modern and stable.

With IPsec, there's always these plethora of extensions and subtleties that each client will support or not. Clients will support Cisco IPsec but will bug with strongswan, strongswan clients will not play well with windows, Android client will not support IKEv2, etc etc etc. From my experience, there's a lot of "finding the right client for the right setup for each platform" with IPsec.

For OpenVPN, usually the native windows/Linux(NM) clients will work out of the box, which is a much better experience than a third party client.

As for TLS VPNs being more prone to vulnerabilities, that does not scare me all that much, since there's always SSH behind the VPN anyway, so it's usually "good enough" for non corporate security.

> Unsurprisingly no mention of wireguard

The hard part of a VPN, the part that everybody makes money at, isn't the IP-level encapsulation. Yes, Wireguard is both conceptually and in implementation simpler and more elegant in this regard. But IPSec per se isn't actually a real pain point in real world corporate road warrior deployments, at least not any more than with Wireguard, which can have very similar issues with MTUs, NAT timeouts, etc when dealing with the diversity of remote user networks.

The hard part of a corporate VPN is authentication, address assignment, subnet routing, nameserver announcements, etc. Therein lies the hard, dirty, complex pieces. Many IPSec-based VPNs utilize IKE for that, although some don't and utilize bespoke, hacky solutions. Most of the horror stories of configuring IPSec relate to IKE, and much of the complexity there is irreducible. (Though I rejoiced when Android finally added native IKEv2 support--years after Windows, macOS, and iPhone--as IKEv2 substantially reduces some of the complexity relative to IKEv1.)

Wireguard provides none of that. Instead, you have to build that stuff separately, and in the realm of Wireguard it's still the wild west, with people slapping together management frameworks with all the dubious quality and opaqueness that have made the traditional VPN ecosystem a nightmare. (The Wireguard team is working on and publishes a separate project for that aspect, but AFAIU it's not widely used and in any event is far removed by its nature from the stark elegance of the transport layer Wireguard protocol.)

But the year is 2021. Every major platform today, including Debian- and RedHat-based Linux, FreeBSD, NetBSD, OpenBSD, macOS, Android, and iPhone, ships with a modern, native IPSec+IKEv2 stack that is highly interoperable and provides, at least in the simple case, certificate-based authentication, dynamic address assignment and subnet routing, and nameserver advertisement. And, importantly, they all support modern cipher suites so the old problem of finding compatible suites is (or could be, in new deployments) a distant memory--everybody supports ECDH P256, AES128-GCM, etc.

On my work-issued macOS laptop the third-party VPN software simply uses native IPSec and a custom IKE daemon. Though as far as I can tell from the configuration the custom daemon is just technical debt from building the product before macOS' IKE daemon (racoon fork) was more mature. And in fact, AFAICT there's absolutely nothing of value, beside IT familiarity, that macOS' IPSec+IKEv2 stack doesn't provide. An equivalent VPN setup using the native stack requires nothing more than a username, password, certificate, and a hostname for the endpoint. You don't need to install a third-party client to automate entering that information; heck, you don't really even need to automate it at all, not any more than you need to automate sending a person to gmail.com or outlook.com for their e-mail.

When you're comparing Apples + Apples, I personally don't think Wireguard is a serious choice for corporate, road warrior VPN access. On most key points, actual Wireguard-based systems are qualitatively worse. The story is different for static point-to-point, server-to-server communications, especially when you can do static address assignments on the tunnel interfaces. In such scenarios the simplicity of Wireguard shines through. But in those scenarios the popular solution is more often OpenVPN than IPSec.

I simply don't see Wireguard replacing IPSec for corporate VPNs. I hope I don't ever see that because good VPN solutions should be deeply integrated with the host system (e.g. to help prevent leakage, integrate DNS resolution, etc), and you're just not going to see that with Wireguard-based solutions. The situation isn't perfect on modern platforms, but it's much better than it was 10 or even 5 years ago, and its getting better and the pace of improvement has (relative to the past 20 years) picked up considerably.

I’ve found that ipsec/ike based VPNs are easily blocked and often unusable at airports, businesses, and with mobile data — pretty much every situation where you would want a VPN. However OpenVPN based protocols running over port 443 seem to magically work everywhere. I started testing Wireguard over ports 53 and 123 right before the pandemic so I can’t say as much regarding that one but I imagine it will have the same qualities as OpenVPN once http/3 becomes mainline and UDP traffic to port 443 becomes common.

I was thrilled to see that ProtonVPN recently added openvpn over tcp as an option — I really wanted to give them money but a vpn that only works on networks I control has limited utility. Express VPN has offered openvpn over tcp for a while and they have been my go-to because it worked everywhere.

That's because wireguard uses non FIPS 140 compliant algorithms. What would be really interesting is if the NSA told us their thoughts on the wireguard algos.

> Strange how this implies that the end-all-be-all of VPNs is IPsec

And no mention of poor configurations such as IKEv1, pre-shared keys and aggressive mode?

The NSA did like breaking IPSec VPN configurations according to documents exposed by Snowden

I think that the problem is temporary because I tried many other NSA URLs and they all 404s

Also,the reference [1] is also the same URL as references 6, 9, 10, 13, 16, 18, 19, 20, 21

You're expecting the NSA to recommend BeyondCorp?

Ah, "beyond corp" the Google marketing term that many industry misunderstood and then exposed their entire internal infra...

It would dilute the messages in the guidance, it was also be quite patronising for anyone who has just been through a lengthy decision making process and concluded they still need a VPN, to be told they shouldn't be using a VPN.

Any time I bother to use a VPN I also use a network slug[1].

I don't want to worry about implementation errors or weird bugs or DNS leaks or anything like that.

A slug, which is a "transparent layer 2 firewall running on a device with only two interfaces" generally breaks your entire network if your VPN is not working exactly as you hope it will.

[1] https://john.kozubik.com/pub/NetworkSlug/tip.html

Their RHEL hardening guide was pretty solid advice and well written.

Vendors have to balance usability and security. That's always been the trade-off between IPSec and TLS. I don't envy people who have to provide support for IPSec solutions.

SSL/TLS based ones will just be more interoperable and have fewer connectivity issues through intermediary network devices and access points.

But TLS has traditionally been a nightmare, crypto-wise, to say secure. POODLE, CRIME, BEAST et al.

At least WireGuard is either secure or off.