Add whole-archive feature to libafl_targets

[addisoncrump]

This addresses a very annoying issue with libafl_libfuzzer where linkage drops symbols from libafl_targets. It does this by forcing the rustc linkage step to add all object files from libafl_targets, regardless if they are directly referenced or not. Confirmed with GDB, this does cause __sanitizer_weak_*cmp to be linked. This should address the issues we see with performance in e.g. PROJ4: https://www.fuzzbench.com/reports/experimental/2023-09-21-libafl/index.html

This is feature-gated as it requires nightly, and the unstable feature cannot be added via rustversion.

[addisoncrump]

Fixes #1540

[tokatoka]

@andreafioraldi is this good to you?
I guess we might have a problem like multiple definition

[addisoncrump]

Fix is somehow incomplete 👻 We are still having some weak symbols not being overridden in fuzzbench. Testing locally, it works, but in Fuzzbench, it doesn't!

[addisoncrump]

^ This now adds the interceptors explicitly as well.

[domenukk]

error: code should be clang-formatted :P

[addisoncrump]

error: code should be clang-formatted :P

Dammit 😂

[addisoncrump]

I've confirmed that this resolves the linkage issues in https://github.com/AFLplusplus/LibAFL/tree/libfuzzer-best

[domenukk]

  error[E0554]: `#![feature]` may not be used on the stable release channel
   --> /home/runner/work/LibAFL/LibAFL/libafl_targets/src/lib.rs:2:40
    |
  2 | #![cfg_attr(feature = "whole_archive", feature(packed_bundled_libs))]
    |                                        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^

Doesn't seem to work on stable then?

[addisoncrump]

Yeah, sadly there's no way to do this with stable.

[domenukk]

Merge this then? And mark libfuzzer as nightly?

[addisoncrump]

Need to fix CI but I'm not sure what the problem is?

[tokatoka]

maybe

whole_archive = ["nightly"]

to disable it on stable?

[andreafioraldi]

For reference, years ago I solved this with a dynamic list in AFL++ https://github.com/AFLplusplus/AFLplusplus/blob/stable/dynamic_list.txt but ofc you need a compiler wrapper, not our case for rustc I guess

[domenukk]

So... @andreafioraldi merge or no merge?

[addisoncrump]

Rebased.

[domenukk]

So ready to merge?

[addisoncrump]

We shall see 🙂

[domenukk]

We don't want a dynamic list then?