Linux

Depends on your threat model and actual realistic concerns.

Ultimately, if it comes down to it, there's very little you can do that's failsafe and 100% guaranteed: the provider has access to your disk, all data in your instances RAM (including encryption keys), and can watch your processes execute in real time and see even the specific instructions your vCPU is executing.

Don't put illegal shit on hardware you do not physically own and have physical control over, and encrypt everything else but like, if the value of your shit is high enough, you're fucked if you're using someone else's computer.

You don't really. Treat it as totally untrusted

Encrypt them before they’re ever put there. One example I can think of is in resilio sync, which has the option for sharing a folder to an encrypted peer. Other peers encrypt it before sending anything, that peer doesn’t have the decryption keys at all.

It depends what you want to do with it. If it's just for storing files/backups then encrypt them before uploading and make sure the key never goes anywhere near the VPS. If it's for serving up something like a simple website, you probably care more about data integrity than exfiltration, so make sure you have the security, including selinux or equivalent, locked down, and regularly run integrity checks. If it's for running something interactive, or where data will be generated or downloaded to the machine, you're out of luck, there's no even theoretical way of securing that against an adversary with that much access.

The only way you can do this, is if the only service you use the provider for is storage. Encrypt the data before you send it to the provider and then they don't know what they're storing.

If they have to do any processing on it at all, then conceptually they need a plain text copy of it to feed into the CPU. And if they have that, there is nothing you can do to stop them from stealing it or using it.

There has been some research in this field, the concept is called homomorphic encryption. That is where you encrypt something in a way that allows a third party to manipulate the data without possessing a key. It is still very limited, and likely always will be due to the extreme difficulty of the question.

Ultimately, you can’t. Even if everything you’re doing is encrypted, they have access to the RAM that’s holding your encryption keys.

Syncthing has a concept of untrusted node, which only gets to store files, not see them

If you're only talking about Storage (data at rest) or Network (data in transit) then encrypt/decrypt offsite and never let symmetric keys (or asymmetric private keys) near the VPS, or for in-transit you could similarly setup encrypted tunnels (symmetric/private keys offsite only) where neither end of the tunnel terminates at the VPS. If you're talking about Compute then whatever does the processing inherently needs access to decrypted data (in RAM, cache, etc) to do anything meaningful. Although there are lots of methods for delegating, compartmentalising, obfuscating, etc (like enclaves, TPM/vTPM...) the unavoidable truth is that you must trust whomever owns the base-infra ultimately processing your data. The one vaguely useful way to use "other people's computers" trustlessly is with SMPC (secure multi-party computation) spread sufficiently widely across multiple independent (preferably competing - or even adversarial!) virtual-computation providers, with an "N-of-M keys" policy that avoids any single provider being able to attain a meaningful level of access to your data independently, or being able to view tangible portions of your data while providing functionality during SMPC. That stuff gets super-niche though.

I've done a lot of thinking about this over the years.

Ultimately the answer is you cannot, at least with certainty. If you don't own the host, you cannot trust anything that runs on the machine.

A few people have said similar, and that for me is the right answer here. I'll expand on how I used to run my servers, but eventually decided it wasn't worth the effort.

Having said that, there are some things you can do to protect yourself, although it depends on how much you care about your data Vs how much effort you want to put in.

For example, you can disguise your data on disk, by creating an encrypted file on Linux that you mount as a filesystem. Everything you care about runs from there. The ideal solution is you have an encryption key that you store somewhere trusted, that you use to decrypt the volume.

But then of course you have to insert that key each time your machine reboots, such as a kernel update.

You also have to manage and protect that key yourself, otherwise 💥 your data is gone.

Another thing to consider is, is your key in memory or on disk at any time. You need to decrypt the disk without the key ending up on the machine. I passed it over SSH and I assume the LUKS folks know what they're doing about disguising the key in memory, but I don't know for certain. I never looked.

My expectation was that I was doing something outside the norms of how these tools were designed to function, so expect unexpected results.

This isn't to say you cannot trust any provider, it really depends how much you want to trust them.

Only real option is to crypt them before putting them on the VPS, but at this point a VPS is pretty useless.

Clevis and Tang but even that can only really do so much.

Just encrypt storage on-site

not a technical but you can't just do full disk encryption and put the password manually at every single boot?

It seems very unlikely that a reputable hosting company would snoop even in that case

If we're talking about 3 letter agencies, for the dedicated servers they'll directly seize the disks...

Thanks for the comments. I agree on the general consensus, that once an encryption key enters the VPS, the encryption is compromised.

However, I'm thinking more in practical terms, eg. the service provider doing just casual scanning across all disks of VPS instances. Some examples could be: cloud authentication keys, torrc files, specific installed software, SSH private keys, TLS certificates.

LUKS

VPN

Encrypt sensitive files