3 ms·
It's the same foolish concept people have been reinventing since early secure UNIX's hosted on security kernels. Long story short, it was impossible without bre
by nickpsecurity 11y ago
It's the same foolish concept people have been reinventing since early secure UNIX's hosted on security kernels. Long story short, it was impossible without breaking applications and lots of staff back when UNIX had "dozens" of system calls. It's not going to work this time.
The solution high assurance security adopted was to virtually or physically isolate a UNIX/Linux stack that interacted with security-critical software that ran directly on microkernel and through mediated interfaces. This resulted in commercial work like INTEGRITY-178B and LynxSecure with Nizza, Perseus, and OKL4/OKLinux in OSS space. Look up "Mikro-SINA VPN" (yes with a k) for great case study. Middleware like Camkes was developed for integration. Similarly, kernels like INTEGRITY and PikeOS developed safe/secure runtimes that ran directly on kernel for Ada and Java.
What this project and all others like it need to do is this:
A. Drop all efforts at rebuilding UNIX in language X.
B. Get a user-mode version of Linux running on it with user-space drivers on microkernel for one, easily-acquired board.
C. Redirect safe language efforts on creating a secure runtime on the microkernel.
D. Create middleware tools along lines of Protocol Buffers, ZeroMQ, or LANGSEC stuff. These should allow seemless, efficient integration between untrusted (esp GUI) stuff in Linux VM and trusted stuff in safe lang uage runtime.
E. Begin on a POSIX-like runtime whose components can be stripped on a case-by-case basis. Focus on syscalls used by critical libraries and services first.
F. The current goal of full POSIX. 99% dont make it to E. If they get here, then great. If to E, then we get at least some apps. If to C and D, then we at least have trustworthy runtime.
That's my two cents worth and builds on what's historically happened in these efforts.
- wmf 11y agoYeah, Graphene on seL4 might be a more practical path for running legacy apps. http://graphene.cs.stonybrook.edu/ http://graphene.cs.stonybrook.edu/
- nickpsecurity 11y agoThat's interesting. Ill have to look at it on desktop. Seems backwards so far though: says it's implemented on top of Linux. My model has a stripped Linux running in user or guest mode on top of a microkernel. See OKLinux, L4Linux, Nizza Security Architecture, or LynxSecure's Linux for examples. Then you just need middleware which becomes a source file or library for your apps.
- pcwalton 11y ago> A. Drop all efforts at rebuilding UNIX in language X. ... OK, but then you say "full POSIX" later. So how does that differ? > B. Get a user-mode version of Linux running on it with user-space drivers on microkernel for one, easily-acquired board. How is this relevant to what happens later? The whole goal is to not rely on the Linux kernel. > C. Redirect safe language efforts on creating a secure runtime on the microkernel. There aren't "safe language efforts". There is a safe language that already exists, and the effort has already been expended to make that exist. The project is creating a secure runtime on the microkernel using that safe language. > D. Create middleware tools along lines of Protocol Buffers, ZeroMQ, or LANGSEC stuff You mean like serde? This is already done for the Rust ecosystem. > E. Begin on a POSIX-like runtime whose components can be stripped on a case-by-case basis. Focus on syscalls used by critical libraries and services first. How is this not what this project is doing? > F. The current goal of full POSIX. 99% dont make it to E. If they get here, then great. If to E, then we get at least some apps. If to C and D, then we at least have trustworthy runtime. ... in other words, exactly what the project is doing?
- cmrx64 11y agoYou've got it exactly right. AFAIK Linux can already be run in userspace on seL4, and there is also a VMM. This is a different project.
- nickpsecurity 11y agoNobody said it was same project and it would be orthogonal to others. My claim is redoing POSIX will likely fail while first doing a Rust runtime for seL4 kernel is much easier and straightforward to use. Essentially, like Nizza security architecture or MILS kernels but with Rust for apps that run directly on microkernel.
- tptacek 11y agoI didn't understand the parent comment either. For what it's worth: this is the first project I've seen that really made me want to start playing with Rust.