#mutation #fuzzing #fuzzer #arbitrary #random-string

no-std mutatis

mutatis is a library for writing custom, structure-aware test-case mutators for fuzzers in Rust

3 releases (breaking)

0.3.0 Aug 19, 2024
0.2.0 Aug 15, 2024
0.1.0 Aug 15, 2024

#54 in Testing

Download history 215/week @ 2024-08-10 180/week @ 2024-08-17 12/week @ 2024-08-24 2/week @ 2024-08-31

409 downloads per month

MIT/Apache

145KB
1.5K SLoC

mutatis

Easily create custom, structure-aware mutators for fuzzing.

crates.io docs.rs supported rustc stable

Repository | Docs | Guide | Contributing

About

The most popular fuzzers — including libfuzzer and AFL — are coverage-guided and mutation-based.

Coverage-guided means that the fuzzer observes which code is dynamically executed while running an input through the system under test. When creating new inputs, it will try to make inputs that execute new code paths, maximizing the amount of code that's been explored. If a new input triggers new code paths to be executed, then it is added to the corpus. If a new input only exercises code paths that have already been discovered, then it is thrown away.

Mutation-based means that, when creating a new input, the fuzzer modifies an existing input from its corpus. The idea is that, if the existing input triggered interesting behavior in the system under test, then a modification of that input probably will as well, but might additionally trigger some new behavior as well. Consider the scenario where we are fuzzing a compiler: if some input made it all the way through the parser, type checker, and into code generation — rather than bouncing off early due to an invalid token — then a new input derived from this one is also likely to go deep into the compiler's pipeline. At least it is more likely to do so than a completely new, random string.

But what happens when we aren't fuzzing a text or binary interface? What happens when we have a custom input type that the fuzzer's built-in mutation strategies aren't very good at targeting? Many fuzzers will expose a hook for customizing the routine for mutating an existing input from its corpus to create a new candidate input, for example libfuzzer has the fuzz_mutator! hook.

mutatis exists to make writing these custom mutators easy and efficient.

Using Default Mutators

To randomly mutate a value with its default, off-the-shelf mutator:

  • Create a mutatis::Session.
  • Call session.mutate, passing in the value you wish to mutate.

Here's a simple example of using mutatis and its default mutators to randomly mutate a value:

# fn foo() -> mutatis::Result<()> {
let mut point = (42, 36);

let mut session = mutatis::Session::new();
for _ in 0..3 {
    session.mutate(&mut point)?;
    println!("mutated point is {point:?}");
}

// Example output:
//
//     mutated point is (-565504428, 36)
//     mutated point is (-565504428, 49968845)
//     mutated point is (-1854163941, 49968845)
# Ok(())
# }
# foo().unwrap()

Combining and Customizing Mutators

You can use the mutator combinators in the mutatis::mutators module to build more complex mutators from simpler ones or to customize mutation strategies to, for example, maintain a type's internal invariants or bound the resulting values into a particular range. The mutatis::mutators module is typically imported under the alias m.

To randomly mutate a value with a custom mutator:

  • Create the custom mutator with from mutatis::mutators combinators and Mutate trait adapter methods.
  • Create a mutatis::Session.
  • Call session.mutate_with, passing in the value you wish to mutate and the mutator you wish to use to perform the mutation.

Here's an example of using mutatis to define a custom mutator for a custom struct type that has multiple fields, and maintains a relationship between the fields' values:

# fn foo() -> mutatis::Result<()> {
use mutatis::{mutators as m, Mutate, Session};

/// A scary monster type.
#[derive(Debug)]
pub struct Monster {
    pos: [i32; 2],
    hp: u16,

    // Invariant: ghost's are already dead, so when `is_ghost = true` it must
    // always be the case that `hp = 0`.
    is_ghost: bool,
}

/// A mutator that mutates one of a monster's fields, while maintaining our
/// invariant that ghosts always have zero HP.
let mut mutator =
    // Mutate the `pos` field...
    m::array(m::i32()).proj(|x: &mut Monster| &mut x.pos)
        // ...or mutate the `hp` field...
        .or(
            m::u16()
                .proj(|x: &mut Monster| &mut x.hp)
                .map(|_ctx, monster| {
                    // If we mutated the `hp` such that it is non-zero, then the
                    // monster cannot be a ghost.
                    if monster.hp > 0 {
                        monster.is_ghost = false;
                    }
                    Ok(())
                }),
        )
        // ...or mutate the `is_ghost` field.
        .or(
            m::bool()
                .proj(|x: &mut Monster| &mut x.is_ghost)
                .map(|_ctx, monster| {
                    // If we turned this monster into a ghost, then its `hp`
                    // must be zero.
                    if monster.is_ghost {
                        monster.hp = 0;
                    }
                    Ok(())
                }),
        );

// Define a monster...
let mut monster = Monster {
    hp: 36,
    is_ghost: false,
    pos: [-8, 9000],
};

// ...and mutate it a bunch of times!
let mut session = Session::new();
for _ in 0..5 {
    session.mutate_with(&mut mutator, &mut monster)?;
    println!("mutated monster is {monster:?}");
}

// Example output:
//
//     mutated monster is Monster { pos: [-8, -1647191276], hp: 36, is_ghost: false }
//     mutated monster is Monster { pos: [-8, -1062708247], hp: 36, is_ghost: false }
//     mutated monster is Monster { pos: [-8, -1062708247], hp: 61401, is_ghost: false }
//     mutated monster is Monster { pos: [-8, -1062708247], hp: 0, is_ghost: true }
//     mutated monster is Monster { pos: [-8, 1487274938], hp: 0, is_ghost: true }
# Ok(())
# }
# foo().unwrap()

Automatically Deriving Mutators with #[derive(Mutate)]

First, enable this crate's derive feature, then slap #[derive(Mutate)] onto your type definitions:

# fn foo() -> mutatis::Result<()> {
#![cfg(feature = "derive")]
use mutatis::{Mutate, Session};

// An RGB color.
#[derive(Debug)]
#[derive(Mutate)] // Automatically derive a mutator for `Rgb`!
pub struct Rgb {
    r: u8,
    g: u8,
    b: u8,
}

// Create an RGB color: chartreuse.
let mut color = Rgb {
    r: 0x7f,
    g: 0xff,
    b: 0x00,
};

// ...and mutate it a bunch of times!
let mut session = Session::new();
for _ in 0..5 {
    session.mutate(&mut color)?;
    println!("mutated color is {color:?}");
}

// Example output:
//
//     mutated color is Rgb { r: 127, g: 45, b: 0 }
//     mutated color is Rgb { r: 127, g: 134, b: 0 }
//     mutated color is Rgb { r: 127, g: 10, b: 0 }
//     mutated color is Rgb { r: 127, g: 10, b: 29 }
//     mutated color is Rgb { r: 172, g: 10, b: 29 }
# Ok(())
# }
# #[cfg(feature = "derive")] foo().unwrap()

Writing Smoke Tests with mutatis::check

When you enable the check feature in Cargo.toml, the mutatis::check module provides a tiny property-based testing framework that is suitable for writing smoke tests that you use for local development and CI. It is not intended to replace a full-fledged, coverage-guided fuzzing engine that you'd use for in-depth, continuous fuzzing.

# #[cfg(feature = "check")]
#[cfg(test)]
mod tests {
    use mutatis::check::Check;

    #[test]
    fn test_that_addition_commutes() {
        Check::new()
            .iters(1000)
            .shrink_iters(1000)
            .run(|(a, b): &(i32, i32)| {
                if a + b == b + a {
                    Ok(())
                } else {
                    Err("addition is not commutative!")
                }
            })
            .unwrap();
    }
}

See the check module's documentation for more details.

Documentation

API Reference Documentation

The API reference documentation is available on docs.rs.

Guide

Check out the guide for tutorials, discussions, and recipes; everything else that doesn't fall into the API-reference category.

License

Licensed under dual MIT or Apache-2.0 at your choice.

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in this project by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.

Dependencies

~245–440KB