fpicker is a Frida-based fuzzing suite that offers a variety of fuzzing modes for in-process fuzzing, such as an AFL++ mode or a passive tracing mode. It should run on all platforms that are supported by Frida.

• Installation Instructions
• Building and Running
• Creating a Fuzzing Harness
• Modes and Configuration

Some background information and the thoughts and ideas behind fpicker can be found in a blogpost I wrote.

Fpicker is based on previous efforts on ToothPicker, which was developed during my master thesis. Most of fpicker was developed during working hours at my employer (ERNW).

Required for running fpicker:

• frida_compile to compile the harness script into one JS file
• The frida-core-devkit for the respective platform found at Frida releases on GitHub
  • depending on the platform you want to target store the library as frida-core-ios.a, frida-core-macos.a, or frida-core-linux.a. Also, linux and macOS/iOS apparently have different header files.

Required only when running in AFL++ mode:

• AFL++
  • on macOS:
    • Compile with CFLAGS="-DUSEMMAP=1".
  • on iOS:
    • Apply the aflpp-ios.patch. This changes the shared mem and out file mode to 666 instead of 600. Fpicker needs to be run as root on iOS. If the target is not running as root, it will not be able to read and write shared memory.
    • Compile with CFLAGS="-DUSEMMAP=1".

Fpicker can be built for macOS, iOS or Linux. The Makefile currently only supports building for iOS on macOS but it should be totally possible to build fpicker using an iOS toolchain on Linux.

Depending on the desired target run:

make fpicker-macos
make fpicker-ios
make fpicker-linux

to build fpicker.

Once fpicker is built, the fuzzing harness needs to be built next:

See the examples folder for different sample fuzzing cases. The general approach is as follows:

• Create a custom harness for the target (e.g. examples/test/test.js) (see here for more information on harnesses)
• Compile the custom harness using frida-compile frida-compile test.js -o harness.js

Now fpicker can start fuzzing. The exact command highly depends on the configuration and setup. In the following, a few example cases are given. These mostly correspond to the examples in the examples folder.

• Run fpicker as AFL++ proxy attaching to a target process fuzzing a specific function in process:

afl-fuzz -i examples/test-network/in -o ./examples/test-network/out -- \\
    ./fpicker --fuzzer-mode afl -e attach -p test-network -f ./examples/test-network/harness.js

• Run fpicker in standalone mode attaching to a server and running a client program to send the fuzzing input:

./fpicker --fuzzer-mode standalone -e attach -p server-process -f harness.js --input-mode cmd \\
    --command "./client-send @@" -i indir -o outdir

• Run fpicker in standalone mode attaching to a server, fuzzing in-process with a custom mutator cmd:

./fpicker --fuzzer-mode active --communication-mode shm -e attach -p server-process -f harness.js \\
    -i indir -o outdir --standalone-mutator cmd --mutator-command "radamsa"

• Run fpicker in passive mode attaching to a server collecting coverage and payloads:

./fpicker --fuzzer-mode passive --communication-mode send -e attach -p server-process -o outdir -f harness.js

• Run fpicker in standalone mode attaching to a running process on a remote device, fuzzing in-process with a custom mutator cmd:

./fpicker --fuzzer-mode active -e attach -p test -D remote -o examples/test/out/ -i examples/test/in/ \\
    -f fuzzer-agent.js --standalone-mutator cmd --mutator-command "radamsa"

Each target requires its own fuzzing harness. The most important part of this harness is defining the entry function of Frida's Stalker, which effectively determines at which point the instrumentation is inserted. In the in-process mode this is simple. The function would usually be the one that is called on each fuzzing iteration. However, it could also be a different one.

A minimalist harness implementation (in command mode) could be this:

// Import the fuzzer base class
const Fuzzer = require("harness/fuzzer.js");

// The custom fuzzer needs to subclass the Fuzzer class to work properly
class TestFuzzer extends Fuzzer.Fuzzer {
    constructor() {
        // The constructor needs to specify the address of the targeted function and a NativeFunction
        // object that can later be called by the fuzzer.

        const FUZZ_FUNCTION_ADDR = Module.getExportByName(null, "FUZZ_FUNCTION");
        const FUZZ_FUNCTION = new NativeFunction(
            FUZZ_FUNCTION_ADDR,
            "void", ["pointer", "int64"], {
        });

        super("test", FUZZ_FUNCTION_ADDR, FUZZ_FUNCTION);
    }
}

const f = new TestFuzzer();
exports.fuzzer = f;

This harness configures the instrumentation to follow the function FUZZ_FUNCTION. The instrumentation will start when this function is entered and stops when the function returns. This function should be chosen carefully as it is expensive and the more (potentially unimportant) parts of the process are instrumented, the slower the fuzzer gets. Of course, this is a consideration between speed and intended coverage. Additionally, the fuzzer currently only supports functions that are only entered once during one fuzzing iteration, i.e., the function should not be called more than once during one fuzz case, otherwise the coverage information might become unreliable.

When the in-process mode is used, another function is required in the fuzzer script. The fuzz method. It will get called on each iteration. It will be called with two parameters, a pointer to a buffer and the length of the buffer. Our exemplary target function takes two parameters, a pointer to a buffer and its length. Thus, we can just pass the parameters were getting in the fuzz method.

fuzz(payload, len) {
    this.target_function(payload, parseInt(len));
}

In passive mode, a callback needs to be specified that processes the required data. The fuzzer expects to receive a payload buffer and its length. Depending on the target function that is fuzzed, this data needs to be extracted. In the following example, we again have a function that has two parameters: a pointer to a buffer and its length. The args parameter contains all potential parameters the target function receives, so the length parameter (which is the second one in our case) can be accessed with args[1]. We then read the buffer as Uint8Array and send it back to the fuzzer using the sendPassiveCorpus method.

passiveCallback(args) {
    const len = args[1];
    const data = new Uint8Array(Memory.readByteArray(args[0], parseInt(len)));

    // this encodes the data and sends it back to the fuzzer
    this.sendPassiveCorpus(data, len);
}

In case the target needs some sort of preparation before the fuzzer can start, fpicker provides a prepare method that is called during the initialization of the fuzzer. Preparation could be the establishment of state, e.g., by instantiating an object. Such a preparation function could look like the following:

prepare() {
  // the object can be attached to the fuzzer instance so that it can be used within the
  // fuzz() method later on.
  this.required_object = call_native_function_that_creates_object();
}

pficker offers a large set of modes and configurations that are explained in the following. Most of these modes can be combined in different ways. At the end of this section is a table that shows which options can be combined and what their implementation status is.

Fpicker has three different fuzzing modes: AFL++ Mode, Standalone Active Mode and Standalone Passive Mode:

• AFL++ Mode: In AFL++ mode, fpicker acts as a proxy between AFL++ and the target process. Using Frida's instrumentation capabilities, AFL's coverage bitmap is populated while the target is fuzzed with input data generated by AFL++.

• Standalone Active Mode: In standalone active mode, the fuzzer uses Frida's Stalker call summaries to gather coverage in form of basic blocks that are executed during an iteration. This is nothing new and has been implemented in various forms before. However, in combination with some of the other fuzzer settings this can have various benefits. It is also a good alternative if AFL++ is not applicable or desired in a given environment or case.

• Standalone Passive Mode: Passive mode is less of a fuzzer and more of a tracer. Essentially, it does the same as standalone active mode. However, it does not send its own inputs. It just attaches to a certain function and collects coverage. Once new coverage is observed, both the coverage and the input is stored.

While fpicker is largely designed as an in-process fuzzer, it also supports fuzzing via an external command. For this fpicker offers two input modes.

• Input Mode In-Process: In in-process input mode, the harness directly calls a specified function in the target process. The fuzzer sends the payload to the harness and the harness prepares the payload in such a way that it can call the targeted function.

• Input Mode CMD: In command input mode, the payload is redirected to an external command. This is useful it is too complex to prepare the parameters other other state when directly calling the target function. The coverage collection still needs to be attached to a certain function. Maybe there is a client that can be supplied with a payload which then triggers the target function.

Communication mode determines how the injected harness communicates with the fuzzer. This largely depends on the target application. Frida offers an API to send and receive messages from the injected agent script. This type of communication is quite costly. One of the factors is that the transported message needs to be encoded in JSON. So sending binary data is straight-forward. Therefore, fpicker offers a second communicateion mode over shared memory. However, this only works if it is possible to establish shared memory between the fuzzer and the target application, which means that this mode cannot be used when the target is attached to the fuzzer host via USB. In CMD input mode, the communication mode only refers to how the coverage information is communicated back to the fuzzer, not how the payload is sent, as this is deferred to an external command.

• Communication Mode Send: In send communication mode the payload is sent by using Frida's RPC calling mechanism. This lets the fuzzer execute a JavaScript function within the injected harness script. This function inside the harness can then do all the necessary preparations to call the target function. Once the target function is returned from, coverage collection will stop and the harness can signal the fuzzer that the iteration is finished. This is done by sending the coverage information back to the fuzzer using Frida's send API.

• Communication Mode SHM: In SHM communication mode the fuzzer and the harness script communicate via shared memory and semaphores. A buffer in shared memory is used to send the payload and receive the coverage information. Instead of sending and receiving, the two components use waiting and posting to the semaphore. Depending on the system and the target, this introduces quite some perfomance gains. Especially, because the binary payload is written to memory once and does not have to be encoded and decoded or copied into other memory locations. Unfortunately, this mode sometimes leads to a low stability when running with AFL++. Not sure why, yet.

Exec mode can be either spawn or attach. This is pretty self-explanatory. fpicker can either attach to a runnning process or spawn a process. One thing that is a major difference between the two modes is that, should the attached target crash, fpicker will not try to respawn.

In standalone mode fpicker offers three different input mutation strategies. Nicely put, input mutation certainly has lots of room for improvement.

• Standalone Mutator NULL: This mutator does not mutate the payload and just returns a copy of the same payload. Mostly for testing purposes. Otherwise not really useful.

• Standalone Mutator Rand: A very bad random mutator. All it does is randomly replace values at random locations in the original payload. It does not change the payload length.

• Standalone Mutator Custom: This mutator can call an external command to mutate payloads. It writes the payload to stdin and receives the mutated payload from stdout. Due to its shallow implementation it has quite a performance impact.

With option -D remote it is possible to fuzz a process running on a network device. For this, the remote device must be running frida-server. As a sample configuration, use SSH with port forwarding to bind the frida-server default listening port 27042 on the remote device to a socket on the local client.

ssh -N [email protected] -L 127.0.0.1:27042:127.0.0.1:27042

Then use frida-ps to validate the configuration by listing processes on the remote device:

frida-ps -R