Frida API Fuzzer - This Experimetal Fuzzer Is Meant To Be Used For API In-Memory Fuzzing

This experimental fuzzer is meant to be used for API in-memory fuzzing.
The design is highly inspired and based on AFL/AFL++.
ATM the mutator is quite simple, just the AFL's havoc and splice stages.
I tested only the examples under tests/, this is a WIP project but is known to works at least on GNU/Linux x86_64 and Android x86_64.
You need Frida >= 12.8.1 to run this (pip3 install -U frida) and frida-tools to compile the harness.

Usage
The fuzz library has to be imported into a custom harness and then compiled with frida-compile to generate the agent that frida-fuzzer will inject into the target app.
The majority of the logic of the fuzzer is in the agent.
A harness has the following format:

var fuzz = require("./fuzz");

var TARGET_MODULE = "test_linux64";
var TARGET_FUNCTION = DebugSymbol.fromName("target_func").address;;
var RET_TYPE = "void";
var ARGS_TYPES = ['pointer', 'int'];

var func_handle = new NativeFunction(TARGET_FUNCTION, RET_TYPE, ARGS_TYPES, { traps: 'all' });

fuzz.target_module = TARGET_MODULE;

var payload_mem = Memory.alloc(fuzz.config.MAX_FILE);

fuzz.fuzzer_test_one_input = function (/* Uint8Array */ payload) {

  Memory.writeByteArray(payload_mem, payload, payload.length);

  func_handle(payload_mem, payload.length);

}

fuzz.fuzzer_test_one_input is mandatory. If you don't specify fuzz.target_module, all the code executed will be instrumented.
You can also set fuzz.manual_loop_start = true to tell the fuzzer that you will call fuzz.fuzzing_loop() in a callback and so it must not call it for you (e.g. to start fuzzing when a button is clicked in the Android app).
The callback fuzz.init_callback can be set to execute code when the fuzzer is ready to begin. See tests/test_java.js for an example.
fuzz.dictionary is a classic fuzzer dictionary, an array in which you can add items (accepted types are Array, ArrayBuffer, Uint8Array, String) that are used as additional values in the mutator. See tests/test_libxml2.js for an example.
frida-fuzzer accepts the following arguments:

-i FOLDER	Folder with initial seeds
-o FOLDER	Output folder with intermediate seeds and crashes
-U	Connect to USB
-spawn	Spawn and attach instead of simply attach
-script SCRIPT	Script filename (default is fuzzer-agent.js)

If you don't specify the output folder, a temp folder is created under /tmp. If you don't specify the folder with the initial seed, an uninformed seed 0000 is used as starting seed.
If you are fuzzing a local application, you may want to execute system-config before frida-fuzzer to tune the parameters of your system and speed-up the things.
Running ./frida-fuzzer -spawn ./tests/test_linux64 you will see something like the following status screen on your terminal:

You can also easily add a custom stage in fuzz/fuzzer.js and add it to the stages list in fuzz/index.js.
To customize the fuzzer, edit fuzz/config.js. The variables that you may want to change are MAP_SIZE (If the code that you are fuzzing is small you can reduce it and gain a bit of speed), MAX_FILE (the maximum size of generated input) and QUEUE_CACHE_MAX_SIZE (increase the queue cache size for more speed, especially on Android).

Example
Let's fuzz the native shared library in the example Android app in tests.
Make sure you have root on your virtual device:

host$ adb root

Download the Android x86_64 frida-server from the repo release page and copy it on the device under /data/local/tmp (use adb push).
Start a shell and run the frida-server:

device# cd /data/local/tmp
device# ./frida-server

Now install the test app tests/app-debug.apk using the drag & drop into the emulator window.
Then, open the app.
Compile the agent script wiht frida-compile:

host$ frida-compile -x tests/test_ndk_x64.js -o fuzzer-agent.js

Open the app in the emulator.
Fuzz the test_func function of the libnative-lib.so library shipped with the test app with the command:

host$ ./frida-fuzzer -U -o output_folder/ com.example.ndktest1

Interesting testcases and crashes are both saved into output_folder.
Enjoy.

TODO
Hey OSS community, there are a lot of TODOs if someone wants to contribute.

~~Java code fuzzing (waiting for additional exposed methods in frida-java-bridge, should be easy, almost done)~~
~~splice stage (merge two testcase in queue and apply havoc on it)~~
~~support dictionaries (and so modify also havoc)~~
~~seed selection~~
inlined instrumentation for arm64
performance scoring (explore schedule of AFL)
structural mutator (mutate bytes based on a grammar written in JSON)
CompareCoverage (sub-instruction profiling to bypass fuzzing roadblocks)
rewrite frida-fuzzer in C with frida-core to be able to run all stuff on the mobile device