Not really, it’s been pretty effortless. Every couple months I have to make sure my renewed LetsEncrypt certs really got imported, but I don’t think I’ve had to intervene manually for anything in a long time.

I do, and I agree about their utility. My users and aliases are in OpenLDAP but it’s pretty easy to add new ones.

Separate accounts are preferable if you’re actually going to be responding to messages. I’ve had some embarrassing encounters where I’ve given an alias to a business that I didn’t realize was going to actually use it for real email conversations with a human. By default roundcube web mail lets you hit reply anyway and the reply goes out with your real address, which can lead to confusion.

I host my own for mspencer dot net, used this 15-ish step walkthrough from linuxbabe dot com. Only maybe three instances of spam in two years, gmail and outlook receive my messages just fine, etc. (Successful spammers were using legitimate services, and those services took action when notified. Greylist delays emails by a few minutes but it’s extremely effective against most spammers because they never come back to retry messages after a few minutes, while legitimate senders will.) I don’t know if I would accept blanket advice against self hosting.

Fundamentally if your mail server can see the addressee, it can see the content. SMTPS encrypts both in the same channel. So at the point where you accept messages and store them in a mailbox, the messages have to be readable.

Encrypting them at rest isn’t something I currently do, but if you’re going to later serve those messages to an email client that expects to receive clear text, your server needs both the keys and the messages. They can be stored in different places.

Most of your needs could be met with full disk encryption on the box hosting Dovecot. If you’re worried about being compelled to decrypt, there’s always the deck of cards trick: The pass phrase for full disk encryption consists of a memorized portion plus the letters and numbers of the top N cards in this deck of cards you keep by the server. If someone were to shuffle that deck of cards, and the server were powered down, the encrypted volume would be impossible to recover.

I’m eager to learn what other Dovecot tricks people can recommend to improve security.

Are you going to be hosting things for public use? Does it feel like you’re trying to figure out how to emulate what a big company does when hosting services? If so, I’ve been struggling with the same thing. I was recently pointed at NIST 800-207 describing a Zero Trust Architecture. It’s around 50 pages and from August 2020.

Stuff like that, your security architecture, helps describe how you set everything up and what practices you make yourself follow.

Mostly I’m scared I’ll write a firewall rule incorrectly and suddenly expose a bunch of internal infrastructure I thought wasn’t exposed.

In a general sense, you are discussing a way to control other people and organizations, and to make them stop talking about you. (Communicating and storing your information) This isn’t always possible or practical.

If you pay a merchant with your payment card, that merchant is allowed to know your payment card number. If you call a toll free number, the recipient of your call is allowed to know your phone number.

If they decide to share what they learn about you, and they do so legally, there’s not a whole lot you can do to stop them. I’m not saying this to antagonize or hurt you. I invite you to think differently about what you can control and what is worth worrying about.

Thank you for your reply, but to be clear, I’m not looking for individual details to be spelled out in comments. What you said is absolutely correct, thoughtful, and very helpful. But emotions are running a little high and I’m worried I’ll accidentally lash out at someone for helping. Apologies in advance.

But do you have any links? Beyond just the general subjects of security architecture, secure design, threat modeling, and attack surface identification, I’d love to see this hypothetical “generic VM and web application housing provider in a box” come with a reasonably secure default architecture. Not what you’re running, but how you’re running it.

Like, imagine decades in the future, internet historians uncover documentation and backups from a successful generic hosting company. They don’t necessarily care what their customers are hosting, their job is to make sure a breach in one customer’s stuff doesn’t impact any other customer. The documentation describes what policies and practices they used for networking, storage, compute, etc. They paid some expensive employees to come up with this and maintain it, it was their competitive advantage, so they guarded it jealously.

I’d want to see that, but (a) a public, community project and (b) now, while it’s still useful and relevant to emulate it in one’s own homelab.

If I can get some of that sweet, sweet dopamine from others liking the idea and wishing for my success, maybe I can build my own first version of it, publish my flawed version, and it can get feedback.

I’ve been struggling to wrap my head around a good security architecture for my mspencer.net replacement crap. Could I bug you for links?

I figured out a while ago to keep VM host management on a management VLAN, and I put each service VM on its own VLAN with heavy, service-specific firewalling and a private OS update repo mirror - but after hearing about ESXi jackpotting vulns and Broadcom shenanigans, I’ve gotten really disheartened. I’d love some safe defaults.

I think this needs to exist, but as a community supported system, not as a commercial product.

Pick a set of open technologies - but not the best, lightest weight, just pick something open.

Come up with a security architecture that’s reasonably safe and only adds a moderate amount of extra annoyance, and build out a really generic “self-hosted web hosting and VM company-like thingy” system people can rally around.

Biggest threat to this, I think, is that this isn’t the 90s and early 2000s any longer, and for a big project like this, most of the oxygen has been sucked out already by free commercial offerings like Facebook. The technical family friend offering to self-host email or forums or chat no longer gets gratitude and love, they get “why not Facebook?”

So… small group effort, resistant to bad actors joining the project to kill it, producing a good design with reasonably safe security architecture, that people can install step by step, and have fun using while they build and learn it.

Married, we both work from home, and we’re in an apartment.

First, all of my weird stuff is not between her work and living room pcs and the internet. Cable modem connects to normal consumer router (openwrt) with four lan ports. Two of those are directly connected to her machines (requiring a 150-ish foot cable for one), and two connect to my stuff. All of my stuff can be down and she still has internet.

Second, no rack mount servers with loud fans, mid tower cases only. Through command line tools I’ve found some of these are in fact capable of a lot of fan noise, but this never happens in normal operation so she’s fine with it.

Separately I’d say, have a plan for what she will need if something happens to you. Precious memories, backups, your utility and service accounts, etc. should remain accessible to her if you’re gone and everything is powered off - but not accessible to a burglar. Ideally label and structure things so a future internet installer can ignore your stuff and set her up with normal consumer internet after your business internet account is shut off.

Also keep in mind if you both switch over so every movie and show you watch only ever comes from Plex (which we both like), in an extended power outage situation all of your media will be inaccessible. It might be good to save a few emergency-entertainment shows to storage you can browse from your phone, usb or iXpand drive you can plug directly into your phone for example.

I’m probably thinking about this in a naive way. I’d love to see proprietary models, if trained using public information, be required to become public and free via legislation. AI companies can compete on selling GPU time, on ease of use.

And, if AI companies are required to figure out attribution in order to be able to use their work commercially, research will accelerate in that area because money. No I don’t know how that would work either.

Still probably a bad idea but I haven’t figured out why yet.

Thank you for your well written reply.

Are you sure? Maybe I’m using the wrong word. What is it called when, in an academic paper, the author states findings or conclusions the author got from some other source, in the author’s own words, but doesn’t cite their source?

Broadly this is preventing plagiarism. We don’t want someone to scrape all our knowledge, remove the human connection and reference back to experts and people, and serve the information itself, uncredited.

But if a human can read something, so can a bot. I think ultimately we need legislation.

Advice from most to least certain: If you want very long standby time (a reliably perfect first print after literally months of inactivity) and you have the space for an ugly cube of a printer, laser is the only option. Ink tank printers have unexpected wear parts, like internal ink sponges.

Black and white laser is stupid simple. Color laser “prints” four times in series onto an intermediate transfer belt (ITB) and then puts that onto the paper, still super reliable but bulkier, and your prints get watermarked with yellow dots because FBI or something. I’d go color.

Toner lock-in is becoming more common, not just for HP. If your page count is going to be low, just pay full price for name brand toner. If you don’t want to do that, like your use case could involve printing a single page or entire binders of paper between months of inactivity, read on.

Start your printer research by shopping for cheap off brand toner, get a sense for what they’re selling the most of and what that’s compatible with, and see what printers they support.

Some aftermarket toner just works, out of the box, because the printer isn’t crazy locked down. Those cartridges have normal sounding instructions. Some aftermarket toner requires you to transplant a chip from a first party cartridge, and their instructions include this. Avoid those printers.

And consider used printers. I have a used HP LaserJet Pro MFP M477fdw that I love, but I would never ever buy another HP printer, especially not one made later than this one. Be very careful before buying any HP printer, especially one made in the past 6-8 years. Even wear items (like the ITB) have modules with firmware and compatibility requirements, and I’m worried I could be one replacement component away from suddenly having a locked down printer.