Add custom mutator macro

[epilys]

This adds a new macro that defines a custom mutator macro for libfuzzer.

Custom mutator

The mutator receives a mutable C array of bytes that is afterwards fed into the fuzzer. The user can use libfuzzer's standard mutator by calling the llvm_fuzzer_mutate wrapper function.

Handling of the C byte array

In Rust, the byte array can be exposed as a mutable slice of length max_len. In the C API, the user must return the actual length used by the array without any guarantee that len <= max_len. I have added two options in the mutator declaration macro, either provide a mutable Vec<u8> and then trim the vector if its length is more than max_len before copying it back into the C API's slice, or use a custom struct Buffer that works like a Vec but ignores any attempts at adding more bytes if the final length surpasses max_len. Buffer is not a very clean solution, but at least it prevents copying in each mutation step.

Next steps

If the Arbritary trait provides a method to turn objects back into bytes (like mentioned in rust-fuzz/arbitrary/issues/44) the macro can take a data: &mut T argument so that the user can work directly on an object rather than raw bytes.

[Manishearth]

r? @fitzgen

[fitzgen]

In Rust, the byte array can be exposed as a mutable slice of length max_len. In the C API, the user must return the actual length used by the array without any guarantee that len <= max_len. I have added two options in the mutator declaration macro, either provide a mutable Vec<u8> and then trim the vector if its length is more than max_len before copying it back into the C API's slice, or use a custom struct Buffer that works like a Vec but ignores any attempts at adding more bytes if the final length surpasses max_len. Buffer is not a very clean solution, but at least it prevents copying in each mutation step.

I would prefer to more closely match the C API.

Buffer is roughly a whole copy of std::vec::Vec, and I don't think maintaining that is the best use of our (this crate's maintainers) resources.

I would prefer to give custom mutators the &mut [u8] slice created from the data and size, the max length, and the seed, and then expect it to return a usize new length, and assert that the returned length is less than or equal the max length. Then, if users want to layer Buffer-style APIs on top of that, they have access to everything they need to do so.

[fitzgen]

Closing in favor of #79.