Refactor to new forkserver

[Evian-Zhang]

Description

Currently in the implementation of start_forkserver, the communication to AFL forkserver is coupled with communication with children. I'm improving unicornafl to better use LibAFL, and this makes it hard to custom communication between parent and children.

Moreover, I found the forkserver logic is written in C, which I didn't find some specific reason for that.

As a result, I use Rust to implement such logic (it's one-to-one rewrite, no logic change except for removing codes under USEMMAP macro that are never compiled), and add convenient trait for parent-children communication.

I'm poor at naming, so please correct me if I give name wrong, especially for forkserver-related names:

In my understanding, there are several entities in forkserver logic:

• The main process part

  This part conduct the mutation, scheduling, etc., and both AFL++ (afl-fuzz executable) and LibAFL (ForkserverExecutor in libafl::executors) have implementations.

• The parent process part

  This part communicate with main part with dedicated forkserver fd to indicate the process of each round. And it also in charge of spawning children and communicate with children. This part is where start_forkserver function should be invoked. ForkServerParent is used as a helper trait to communicate with children.

• The child process part

  This part is forked by the parent process, and does the real execution and observation.

Checklist

• I have run ./scripts/precommit.sh and addressed all comments

[Evian-Zhang]

Accidentallu deleted some global variables, will fix it few hours later. Fixed

The clippy CI says something wrong in executors/forkserver.c, but that code I did not touch. Strange...

[Evian-Zhang]

I will investigate into the broken pipe issue...

[wtdcode]

Does it surpass my #3087 ? If so I will take care of this. Just let me know when you are satisfied with the code.

[Evian-Zhang]

Maybe yes. This PR enables us to conduct the fallback in downstream. If this PR got merged, I will further modify unicornafl to fit in. :)

[Evian-Zhang]

Finally passed all tests... @wtdcode The code is ready for review. Thank you for your patience :)

[tokatoka]

can you do some benchmark comparing forkserver.rs and forkserver.c

[Evian-Zhang]

can you do some benchmark comparing forkserver.rs and forkserver.c

Sure, I'll do it asap.

[Evian-Zhang]

Tested fuzzers/forkserver_libafl_cc using just run with timeout 1h in Ubuntu 22.04 i9-13900K, the C version is 5.167K/s, the Rust version is 5.238K/s, which is acceptable for me :) (This PR is not primarily aimed at performance boosting; Instead, using Rust is to make it more generalized with help of trait system)

[tokatoka]

then it's ok

[wtdcode]

Very cool work. Thanks for your efforts =)

[wtdcode]

#3087 can be closed in favor of this I think.

[Evian-Zhang]

Thank you @wtdcode :) I already have a worked version of unicornafl for this patch that still needs some cleanup. Please wait me for about 4-5 days (some other stuff occupies me) and I will send a patch.

[wtdcode]

No hurry, I'm on academic travel.

[Evian-Zhang]

I have merged the latest commit in main branch, please notify me if there is anything I missed to get this PR accepted :)

[wtdcode]

Thanks =)