Verifier integration

[Alan-Jowett]

This pull request integrates the eBPF verifier as a submodule and updates the libfuzzer harness to utilize the verifier. The changes include modifications to submodule configurations, build scripts, and the addition of new verification and debugging functionalities in the libfuzzer harness.

Integration of eBPF Verifier:

• .gitmodules: Added the ebpf-verifier submodule.
• CMakeLists.txt: Included the ebpf-verifier directory in the build when UBPF_ENABLE_LIBFUZZER is enabled.
• external/ebpf-verifier: Added the submodule commit reference for ebpf-verifier.

Build Script Updates:

• libfuzzer/CMakeLists.txt: Updated include directories and linked the ebpf-verifier library.

Enhancements to libfuzzer Harness:

• libfuzzer/libfuzz_harness.cc:
  • Included headers for the eBPF verifier and platform.
  • Added verify_bpf_byte_code function to verify BPF bytecode using the eBPF verifier.
  • Implemented context management and debug functions for the uBPF context.
  • Integrated verification and debug functions into the interpreter and JIT execution paths. [1] [2] [3]
  • Added a call to verify_bpf_byte_code in the main fuzzer test function.

API Changes:

• vm/inc/ubpf.h: Added register_mask parameter to the debug function signature.
• vm/ubpf_vm.c: Updated the call to the debug function to include register_mask.

[coveralls]

coverage: 79.503% (-0.1%) from 79.637%
when pulling 9ae1e5e on Alan-Jowett:verifier_integration
into 0642da6 on iovisor:main.

[hawkinsw]

I will review more as soon as I can, but I thought that I could send along these tiny little bits of feedback now! Thank you for doing this work -- I think that it is going to be tremendously helpful!

[hawkinsw]

I think that there is a problem with our CMake configuration when the developer wants to have both "regular" tests and "fuzzing" tests on Linux. In that configuration, I get a linker error because vm/test.c has a main and then the libfuzzer (which is linked) has a main, too. I will try to do some debugging and see whether my hunch is true.

[hawkinsw]

I think that there is a problem with our CMake configuration when the developer wants to have both "regular" tests and "fuzzing" tests on Linux. In that configuration, I get a linker error because vm/test.c has a main and then the libfuzzer (which is linked) has a main, too. I will try to do some debugging and see whether my hunch is true.

I have confirmed that this "cross up" is the problem -- when the user (on Linux) has conformance tests and fuzzing tests enabled at the same time, there is a linker error.

There also seems to be a need to do an additional sanity test in cmake/settings.cmake -- it has a test for whether the CC is set to clang (for fuzzing) but does not guarantee that CXX is also properly set. I accidentally set CC to clang but had CXX set to g++ and I got a linker error re:

c++: error: unrecognized argument to ‘-fsanitize=’ option: ‘fuzzer’

It was easily correctable, but I loved what you did with the detection for the C compiler and thought we could do the same with the C++ compiler.