Is it though? Is having your (and your co-workers) desktops/laptops/and possibly more directly able to access production servers at a packet level a good idea?

I would posit that it is not. A bastion host can be used to lock down more than firewalls, and reduces the attack surface living on your network considerably.

> VPN terminator in the VPC serves the same purpose as a jump box.

It certainly serves the same network separation purposes a jump box does, but is not able to dictate the access environment nearly as much. If only your jump box has access to servers (and production servers properly filtered), every bit of remote access must go through that single ingress/egress point. And that point can have security policy built-in as well for multi-person teams. For example: I can enforce certain key types being utilized on remote hosts, logging of ssh sessions, etc.

I also agree the "single packet authentication" is silly though. Most bastion hosts these days will be behind VPN already. Just another riff on port knocking really.

Most notable of these issues is the nature of either needing SSH agent forwarding enabled or storing a copy of the private keys on the bastion host (please don't do that). There are ofcourse ways of dealing with this but they aren't worth it.

It's fundamentally much better to just harden every system and implement proper federated authentication and authorization with LDAP. It also makes it much nicer to implement things like MFA and device level authentication.

Remember, it's very rarely that an attacker is going to be able to successfully walk in or break down the front door. Your security holes are very unlikely to be in your administrative channels like SSH/RDP but rather crappy applications or services you may be running on your systems.

Those crappy bits of software + insider threat are the 2 main things you generally need to deal with. Ensuring systems are sufficiently isolated from one another helps with both of these and bastion hosts just add another layer of shared stuff between workloads that should stay entirely separate.

I only said bastion hosts are bad and generally decrease overall security because it's almost impossible to craft one that doesn't create more security holes than it closes.

Much better? Yes. More practical? No.

Once you have a network of more than 3 hosts, host security can quickly become difficult to maintain. This means not just keeping up to date on patches but auditing all your configurations and carefully reviewing applications and services.

Every organization in the world has serious host security issues. For example, a company like Google has an excellent security team and presumably world class endpoint security, but there's almost no question that an attacker on some segment of their internal network would be able to pivot and take over hosts (the lateral movement would likely be noticed by their SOC, perhaps within an hour or less, but it would still be after-the-fact).

Proper network segmentation and architecture makes most of those issues much less risky though, as it is much harder to actually get to any of those hosts.

It's fundamentally much better to just harden every system

You must be one of those people that don't make mistakes and only use bug-free software across their entire deployment.

For the rest of us, DMZ's and bastion hosts are the only way to limit our attack surface to something that can be realistically managed.

I agree. I usually use simple port knocking for that purpose. Surely you can agree that "no ports listening" is a better circumstance than "tcp/22 open to the world, with key-based-auth only"?

A VPN is redundant (and requires special client), whereas port knocking to open a port for ssh (that only supports key-based auth) is a simple matter of a few "nc -w 1" lines in a script before you ssh...

EDIT: Saw your posts below. You presuppose that such OpenSSH-zero-day holding attackers are specifically targeting you, which is not invalid; however a spray-and-pray quickly after public release is much more probable in a lot of cases.

(tls-auth doesn't protect you from vulnerabilities in the OpenVPN data protocol, but that's a lot simpler.)

If you really believe that OpenSSH is too scary to expose to the Internet for fear of memory corruption flaws, don't stop at a half-measure like "port authentication". After all: once you "authenticate" the port, the attacker can simply hijack your TCP connection to trigger the flaw.

Instead, write a plugboard proxy, and have it apply a static-keyed AES-CTR transform to the data.

I think this is silly, but if you're trying to protect yourself from OpenSSH bugs, at least do something that actually protects you from OpenSSH bugs.

I also dispute that a plugboard proxy is "more complicated" than "cryptographic port knocking".

I'm not aware of any available plugboard proxy solution, so it would have the added complication of hacking it together. That's what I meant. And there are diminishing returns there: a non-criptographic port knocking scheme with a few packets could be good enough as a camouflage in many cases. But I still find it to be a time saver in the more general case (not APT). I think having cleaner logs is worth using this simple trick then.

is this because you expect this class of attackers to be able to thwart port knocking more or less trivially ?