Do not let tokio swallow forwarded panics

[s-arash]

We forward enclave panics (i.e., panic if the enclave panics) inside a tokio task, which results in tokio eating up the panic. This is the exact opposite of the intended behavior of forward_panics. This PR makes the runner panic outside tokio tasks, so the panic won't be caught by tokio.

Closes #763

[Goirad]

Nice!

[jethrogb]

Does this work? I'd think all aborts get converted to Secondary in syscall_loop.

[s-arash]

@jethrogb

Does this work? I'd think all aborts get converted to Secondary in syscall_loop.

Not all of them get converted to Secondary. If the mode is ExecutbaleMain (or Library), the abort remains an Exit:

rust-sgx/intel-sgx/enclave-runner/src/usercalls/mod.rs

Line 892 in 44823d2

(e, ReturnSource::Library) |

But you are correct that some aborts do get converted to Secondary (when mode is ExecutableNonMain or AsyncUsercall). To fix that, I can edit this line:

rust-sgx/intel-sgx/enclave-runner/src/usercalls/mod.rs

Line 894 in 44823d2

(e @ Err(EnclaveAbort::Exit { panic: None }), _)

to

(e @ Err(EnclaveAbort::Exit { panic: _ }), _) 

so that it also keeps panic-exits as is. I'm not sure why we made this distinction here in the first place.