Using a jump host is good practice for increased security and a simple solution when you don’t have a fixed IP-adress but need access to Safespring’s APIs.
Automation tools like Terraform require access to several of OpenStack's APIs which we're currently exposing only to whitelisted IPs for security reasons. This is a practice we share with several other small and independent cloud providers like ourselves. With many engineers working remote and not having a fixed IP, submitting a support ticket each time your public IP changes might feel tedious and unnecessary in 2022. However, using a jump host is not only a simple solution to this problem but also good practice for most automation scenarios where security is critical.
OpenStack is a software platform consisting of several components controlling a pool of different processing, storage, and networking resources along with helper components such as an identity service and a GUI dashboard. What might look like one unified product from a dashboard perspective are in reality different pieces of software developed by different community teams organized under a larger project - a common model in open source software. The components communicate over their respective REST APIs with the dashboard as a centerpiece for the user interaction. However, when using the CLI or automation software like Terraform you're not communicating through a single application interface like the dashboard, but with all the individual service's public API endpoints.
This doesn't imply that OpenStack is insecure - quite the contrary. OpenStack is one of the most transparent and secure cloud platforms there is, being developed in the open by a large number of people and organizations, and under constant scrutiny. However, all software is potentially vulnerable to security breaches and exploits, which is why there are ceaseless attempts to compromise practically anything exposed to the internet from every corner of the world. While we feel confident that the OpenStack components’ public APIs are reasonably secure and correctly configured on our side, they nevertheless constitute a relatively large attack vector. Given the resources we as a small company have available - and the potential damage a security breach may cause to our customers - it makes sense to limit this attack vector even if it causes a minor inconvenience when direct API access is needed.
Even if we hypothetically could guarantee that every public API endpoint was 100% bulletproof, there are good reasons to be cautious as a consumer. For example, there's no way to use multi-factor authentication between e.g. Terraform and OpenStack APIs, or with the OpenStack CLI for that matter. Credentials are often stored in clear text. The implications of a stolen laptop or credentials gone astray could be disastrous - a single command could wipe out an entire infrastructure.
Using a jump host is not a perfect solution to the above-mentioned challenges, but an excellent first step to better security - and there are other benefits as well.
Using a jump host
A jump host in a broader definition is a system on a network used to access and manage devices in a separate security zone. By using a jump host you can restrict access to internal resources to a single source with increased security hardening (including multi-factor authentication) and monitoring. Depending on how strictly you define a jump host, it can also be a work environment where e.g. deployments to internal servers are executed. In some cases, even full development environments are deployed on jump hosts using remote development tools like VSCode or a combination of terminal editors and multiplexers with the benefits of easy access to internal resources and collaborative coding.
The public instance network ranges in Safespring Compute are whitelisted for API access on our load-balancers which is why we always recommend our customers to start by creating a jump host through the web dashboard and utilize this for API-based provisioning (e.g. Terraform).
There are other benefits as well. For example, if you want to deploy to instances on the private (RFC1918) network, the only way to do that is through a public instance in the same site. We've written about that in detail in another blog post. You could of course create a specific jump for this purpose alone, however, doing all provisioning and deployments from a single host is far more convenient.
Combine all this and you have a very practical way of managing your Safespring infrastructure as well as your own security zone.
Working over SSH from a jump host doesn't fit every scenario. Another common option when working across security zones is a VPN. This has traditionally been a rather complicated task to configure involving complex networking, security and protocols, and is considered both archaic and a time-drain for many administrators.
Enter sshuttle, by its developer referred to as “a poor man's VPN”, but no less powerful for that. sshuttle is entirely based on SSH, is very simple to use, is reasonably fast, and requires no server-side configuration besides a working SSH server. Not only is sshuttle leveraging the security benefits of OpenSSH but also its in-built tunneling capabilities - without needing to set up individual forwarding for every host/port on the remote network.
The downside of sshuttle is that it currently only runs on macOS, Linux, and BSD. There's currently no support for Windows, including WSL (although there is a workaround).
Using sshuttle with a Safespring Compute instance hardly requires any set up. Just launch an instance, add your SSH-key and simply run:
sshuttle -r <username>@<instance ip> <range you want to forward>
The range could be as narrow or wide as you'd like, but for the OpenStack APIs it's sufficient to forward the IP the auth_url in your OpenStack config resolves to.
As sshuttle is an SSH client, it supports the same authentication methods and security hardening as a regular OpenSSH client, including MFA.
Other options include more full-fledged VPN solutions like Wireguard, ZeroTier or OpenVPN which are all a lot simpler to configure than traditional protocols like IPSec, GRE, PPTP, etc. If you're going to set up a VPN in Safespring Compute anyway the gateway can serve as a jump host has well. We've created some documentation on how to get started and we also provide an automated setup of a Wireguard gateway which can be used for this purpose and more, available here