All code has bugs. Xen has bugs. Qubes has bugs. And yes, OSes have bugs, although Linux local privilege escalation bugs are not an everyday occurrence, and OpenBSD bugs are very rare. You can't handwave a 0-day privilege escalation vulnerability into existence and claim that there are no 0-day privilege escalation vulnerabilities in Xen.

tar is pretty solid, actually, but, if you don't like it, make up your own trivial archive format (it's not hard), or don't use it and follow a one-disk-per-message protocol. And don't use a filesystem. dd if=email.eml of=/dev/cdrw (approximately), then dd if=/dev/cdrom of=email.eml

User error could happen with any system. You're really stretching here.

Where did I ever say any of these computers had hard discs? And of course there's an out-of-band way to store info. Just use a magical vulnerability in Xen I made up to write it to another VM's permanent storage.

I have no idea where you're getting that disc encryption has anything to do with anything here, and, no, you could definitely use the same three computers for all your emails. If you find burning discs to be too cumbersome, use floppy disc drivers or Zip drives or something, but it's really not that bad, especially if you don't bother to fixate.

This whole notion you have that "the operating system is insecure so let's put another layer on top of it" is just silly. If you make Xen your operating system, then Xen is your operating system. If you want a secure operating system with no privilege escalation, then that's what you need. Using a hypervisor doesn't magically make security vulnerabilities go away. There have and will be attacks against Xen, against Qubes, and against any other complex pieces of software you create. You want a secure OS, then you need a dead-simple OS written with security in mind. True security comes from simplicity, not complexity.

The one thing I think you're right about is that modern OSes are too complex to provide extreme levels of security. OpenBSD is the best, but even it is a quite complex piece of software. A braindead-simple POSIX-like OS kernel prizing security above all other values may have its uses. It could use generic hardware drivers for the hard disk and display adapter (VESA). Ethernet would be a problem but maybe there's a standard for USB Ethernet adapters? Wayland for the graphics API instead of X, braindead-simple written-from-scratch TCP stack, you get the idea. I'll have to make that when I have time. It shouldn't be too hard.

Burz,

I'm not saying it's impossible to customize a USB device. I'm saying rooting a machine to the point that it can customize an arbitrary USB key plugged in by the legitimate operator of the machine is impractical. You're also invoking speculative, unknown attacks against the USB host driver and firmware, which I will see you with my previously invoked unknown, speculative attacks against Xen. Also, you completely ignored my suggestion of using an optical disk if you are concerned about USB.

Safest way I can think of using airgapped machines right now for encrypted email:

1. Copy received email from networked machine C to write-once optical disk.
2. Decrypt received email on airgapped machine A.
3. Compose reply on different airgapped machine B, encrypt reply.
4. Copy encrypted reply to write-once optical disk.
5. Send encrypted reply from networked machine.

This involves three physical computers, but none has to be recent or expensive. Airgapped machine A has no ability to send information to C or B, and airgapped machine B is never touched by any devices from the outside world, and also never needs to know any secret keys, since all you need to encrypt an email message is the recipient's public key. Theoretically you could type such a key in by hand, but in any case it's a once-per-recipient transfer.

I would argue using USB sticks in this scenario gives only a very slight reduction in security, but write-many optical disks would be a practical approach if you're scared of USB. I will say that compiling out obscure and unused USB kernel drivers is a good move, as is disabling USB kernel module autoloading.

StackExchange says you're wrong about USB having DMA: http://security.stackexchange....

In any case, BadUSB would require reprogramming the actual device, so I still don't think it is a practical attack vector in this scenario. Moreover, if you're really paranoid, you can use write-once CD-Rs instead of USB devices.

QubesOS is an interesting idea, but it's more complicated and therefore more likely to have bugs than airgapping a machine. You're assuming there are no bugs in Xen, for instance.

As for filesystem bugs, this code has been around for 20 years or more. There are bugs everywhere, but I think especially popular Linux filesystem drivers are likely pretty solid. But go ahead and just dd the file to the optical disk directly and don't use a filesystem if it makes you feel better.

Bugs in the filesystem driver, yes, but those are probably pretty rare I'd think. BadUSB, not really. That attack works by emulating a keyboard/mouse HID controller. If you plug your USB drive in and all of a sudden your computer starts typing things and moving the mouse on its own, you would notice immediately. Also it typically requires special hardware; a rooted box couldn't just take a real USB drive and turn it into a HID controller.

I was thinking flash drive or possibly optical disk ... couldn't there theoretically be an exploitable buffer overflow in ZModem?

- It is technically possible to air-gap the machine you use to access your email, by copying the email over from an insecure computer to the air-gapped machine.
- TAILS is great, but they probably at least try to break it since it's popular. Will they succeed? Maybe. So use an OpenBSD live CD, it's more secure anyway. Or get creative: use Whonix. The FBI's pedestrian attempt at drive-by malware would have fallen flat on its face with an adversary using Whonix.
- Firejail. Google it. Won't protect you against local kernel privilege escalation attacks, though.

Yes, contingency planning is good. Yes, single points of failure are bad. But you can get very, very good communication security if you really try.

If you learned SunOS 2 or HP-UX 9, most of that knowledge is applicable to Solaris and modern Linux distributions today.

Uh, maybe PRIVATE schools can have content-based speech restrictions...

http://en.wikipedia.org/wiki/T...

Oh, yeah? Well, I'll make my own porn site. With blackjack! And hookers! Actually, forget the porn site. And the blackjack...

My undergrad curriculum had a class where we designed a pipelined processor, though we didn't actually build it. It was an interesting and educational experience for me, and I'm glad to have done it.

This. There is a good chance that Turing actually didn't commit suicide, but rather died of accidental cyanide inhalation. He had set up a chemical lab in his living space and wasn't exactly using OSHA-approved storage protocols for dangerous chemicals. His mother, at the time, said she didn't think he'd killed himself, and contemporary accounts were that he was doing pretty okay. The supposedly cyanide-poisoned apple was not tested for cyanide. None of this is conclusive.

IMO, any modern report on Turing should account for the possibility he didn't kill himself. The suicide angle makes a great story for gay rights activists, but it does a disservice to the memory of this great man to reduce him to a political talking point. The forced hormone treatment was abominable, whether or not it drove him to suicide. There's a chance it did, and a chance it did not.

Heh ... you're lucky. I seated a PCI card in wrong once and it shorted out. Fortunately, it was only $10 or so to replace.

But, you may have a point: it might be possible to electrically tap the PCI or PCI Express bus and do bad things with DMA, even if the bus wasn't built to support hot-swapping. You'd probably need custom hardware, a lot of time, and a lot of luck, though. Also, you'd need to keep power to the CPU on, meaning stuff like chassis intrusion detectors would be a sufficient countermeasure.

I intended no sarcasm. I think the Bill of Rights was a very good idea.

The story is actually very interesting. The Bill of Rights was enacted as a compromise to get the Constitution passed. The Constitution was not our first government -- that was the Articles of Confederation, but the Articles of Confederation basically wasn't working at all because it was a very poor design.

Some highlights: it gave the federal government so little power it couldn't do anything. It couldn't even pass taxes; the states were supposed to voluntarily pitch in. It also required unanimous consent in Congress to pass any law, and Congress was all there was; there was no executive or judicial branch.

So some of the leaders -- the Federalists -- drafted the Constitution to replace it. But there were Anti-Federalists, and they argued the central government would become so powerful it would eventually turn tyrannical. So, the Bill of Rights was added to placate them. We can see now that was a really, really Good Idea(TM).

I get writing a compiler, but why a text editor?