[wasmtime] Try fixing delivery of SIGILL/SIGSEGV to JIT

[alexcrichton]

This is an attempt to apply the suggestions from #3316 to the fuzzing
infrastructure for the wasmtime target. This will hopefully allow the
delivery of SIGSEGV and SIGILL signals to the wasmtime program itself.
These are expected signals when executing wasm code so we don't want the
fuzzer to treat all forms of the signal as a fatal error.

Unfortunately I'm not entirely sure how to test this locally, using the reproduction workflow this doesn't seem to have any effect, but it looks like this *.options file may not be picked up during the reproduce workflow? I figure a shot in the dark could hopefully help here and I'm trusting @Dor1s as well :)

[Dor1s]

Unfortunately I'm not entirely sure how to test this locally

yeah, sorry about that, this is pretty much an undocumented feature which exists but is used by <0.1% fuzz targets running on ClusterFuzz.

We would need to add support for .options file parsing in infra/base-images/base-runner/run_fuzzer.

There is a convenient way to parse it in Python https://github.com/google/clusterfuzz/blob/5e9b02204428cb1a81993d512844966d2bee0a80/src/python/bot/fuzzers/options.py#L96

but run_fuzzer is a bash script, so I guess we would need a small python script such as env_setup.py that would read a corresponding .options file (if exists) and set up the variables like:

  export ASAN_OPTIONS="$ASAN_OPTIONS:<whatever_parsed_in_the_options_file>"
  export MSAN_OPTIONS="$MSAN_OPTIONS:<whatever_parsed_in_the_options_file>"
  export UBSAN_OPTIONS="$UBSAN_OPTIONS:<whatever_parsed_in_the_options_file>"

but in Python :) That script would be called from run_fuzzer, and eventually run_fuzzer may be re-written into Python as well. If could upload a pull request with such script, that'd be appreciated :)

[Dor1s]

The change LGTM, waiting for Travis to finish

[alexcrichton]

Aha that makes sense! And no worries, just wanted to provide a warning that this was somewhat untested, but if anything goes awry can always have a follow-up PR.

Thanks so much for your help with this!

[TravisBuddy]

Hey @alexcrichton,
Your changes look good to me!

View build log

TravisBuddy Request Identifier: d061e430-483e-11ea-adaa-75531245892f