`mkdir` test case fails at a high rate on LLVM-14 version of BinRec

[michaelbrownuc]

We need to explore this and determine to what degree it is different from other transient issues we have encountered with S2E.

[ameily]

mkdir is routinely failing with the "can't locate main" error (#160):

binrec.errors.BinRecLiftingError: failed to perform initial lifting of LLVM bitcode: s2e-out: [recover_functions] Failed to located main via entrypoint

I'm working now to see if this is really a side effect of the known issue where inconsistent trace info is being produced, trailofbits/binrec-tob#14, or if something else is going on.

[ameily]

With the latest update I cannot get mkdir to lift properly at all and I'm not sure if it was ever working to be honest. There was a typo in the batch file where the -f flag was being passed in, which doesn't exist in that version of mkdir. So, what was being run and lifted previous was an early exit with a usage error.

[ameily]

I've confirmed that this issues is related to trailofbits/binrec-tob#14. I finally got a working mkdir trace and then compared each failure against it. Each failure had a large trace info difference from the working trace info.

mkdir-diff.txt

For some reason mkdir reliably triggers this behavior in S2E and I don't see anything in the source code or lifted IR that points to why.

@michaelbrownuc For now I think we should disable the mkdir integration test since this appears to be part of a much larger issue, potentially within S2E.

[michaelbrownuc]

This is fine for now, we can look into this at the same time as trailofbits/binrec-tob#14.

[michaelbrownuc]

Also affecting server, now a high priority to repair.