Prevent hypercall handling from panicking

[jounathaen]

A panic in the hypercall causes uhyve to hang indefinitely, as only the specific CPU thread will panic, and all other threads are still waiting for the barrier in

uhyve/src/linux/mod.rs

Line 137 in 360b613

barrier.wait();

to pass, which never happens as the thread that could activate it has paniced. We should remove any unwrap() or expect() in the code that is called by run.

[charlottestinson]

Hello!

My partners and I are a team of three students currently studying Virtualization at the University of Texas at Austin. Part of our coursework is contributing to open-source repos like this. We are wondering if this issue is still open and available for contribution? If so, we would love to take a chance to work on this because it seems very interesting to us! If approved, we may be contributing to some other issues as well.

Thank you so much!

[jounathaen]

Sure, this would be much appreciated. If you need any help/advice, just post here. 🙂

[mkroening]

We should remove any unwrap() or expect() in the code that is called by run.

While that would be nice, I don't think we can and should remove all possible panics from the vCPUs comprehensively. I think it would be better to catch_unwind at a high level to then exit the application to resolve this issue. Better error handling is always good, of course, but that does not always mean less panicking, in my opinion.

[jounathaen]

True, but I think (without deeper looking into it) we actually have three cases here:

• Simple to fix unwraps that can be mitigated
• Errors in the hypervisor, which probably would be solved best by catch_unwind
• Errors resulting from erroneous hypercalls, that should be reported back to the kernel to be handled there. (this likely requires changes in uhyve-interface and the kernel)

So I think if any of these cases is at least partially tackled by @charlottestinson, it would be an improvement.