| 1 |
//! Bindings to [libFuzzer](http://llvm.org/docs/LibFuzzer.html): a runtime for |
| 2 |
//! coverage-guided fuzzing. |
| 3 |
//! |
| 4 |
//! See [the `cargo-fuzz` |
| 5 |
//! guide](https://rust-fuzz.github.io/book/cargo-fuzz.html) for a usage |
| 6 |
//! tutorial. |
| 7 |
//! |
| 8 |
//! The main export of this crate is [the `fuzz_target!` |
| 9 |
//! macro](./macro.fuzz_target.html), which allows you to define targets for |
| 10 |
//! libFuzzer to exercise. |
| 11 |
|
| 12 |
#![deny(missing_docs, missing_debug_implementations)] |
| 13 |
|
| 14 |
pub use arbitrary; |
| 15 |
use std::sync::OnceLock; |
| 16 |
|
| 17 |
/// Indicates whether the input should be kept in the corpus or rejected. This |
| 18 |
/// should be returned by your fuzz target. If your fuzz target does not return |
| 19 |
/// a value (i.e., returns `()`), then the input will be kept in the corpus. |
| 20 |
#[derive(Debug)] |
| 21 |
pub enum Corpus { |
| 22 |
/// Keep the input in the corpus. |
| 23 |
Keep, |
| 24 |
|
| 25 |
/// Reject the input and do not keep it in the corpus. |
| 26 |
Reject, |
| 27 |
} |
| 28 |
|
| 29 |
impl From<()> for Corpus { |
| 30 |
fn from(_: ()) -> Self { |
| 31 |
Self::Keep |
| 32 |
} |
| 33 |
} |
| 34 |
|
| 35 |
impl Corpus { |
| 36 |
#[doc(hidden)] |
| 37 |
/// Convert this Corpus result into the [integer codes used by |
| 38 |
/// `libFuzzer`](https://llvm.org/docs/LibFuzzer.html#rejecting-unwanted-inputs). |
| 39 |
/// This is -1 for reject, 0 for keep. |
| 40 |
pub fn to_libfuzzer_code(self) -> i32 { |
| 41 |
match self { |
| 42 |
Corpus::Keep => 0, |
| 43 |
Corpus::Reject => -1, |
| 44 |
} |
| 45 |
} |
| 46 |
} |
| 47 |
|
| 48 |
extern "C" { |
| 49 |
// We do not actually cross the FFI bound here. |
| 50 |
#[allow(improper_ctypes)] |
| 51 |
fn rust_fuzzer_test_input(input: &[u8]) -> i32; |
| 52 |
|
| 53 |
fn LLVMFuzzerMutate(data: *mut u8, size: usize, max_size: usize) -> usize; |
| 54 |
} |
| 55 |
|
| 56 |
/// Do not use; only for LibFuzzer's consumption. |
| 57 |
#[doc(hidden)] |
| 58 |
#[export_name = "LLVMFuzzerTestOneInput"] |
| 59 |
pub unsafe fn test_input_wrap(data: *const u8, size: usize) -> i32 { |
| 60 |
let test_input = ::std::panic::catch_unwind(|| { |
| 61 |
let data_slice = ::std::slice::from_raw_parts(data, size); |
| 62 |
rust_fuzzer_test_input(data_slice) |
| 63 |
}); |
| 64 |
|
| 65 |
match test_input { |
| 66 |
Ok(i) => i, |
| 67 |
Err(_) => { |
| 68 |
// hopefully the custom panic hook will be called before and abort the |
| 69 |
// process before the stack frames are unwinded. |
| 70 |
::std::process::abort(); |
| 71 |
} |
| 72 |
} |
| 73 |
} |
| 74 |
|
| 75 |
#[doc(hidden)] |
| 76 |
pub fn rust_libfuzzer_debug_path() -> &'static Option<String> { |
| 77 |
static RUST_LIBFUZZER_DEBUG_PATH: OnceLock<Option<String>> = OnceLock::new(); |
| 78 |
RUST_LIBFUZZER_DEBUG_PATH.get_or_init(|| std::env::var("RUST_LIBFUZZER_DEBUG_PATH").ok()) |
| 79 |
} |
| 80 |
|
| 81 |
#[doc(hidden)] |
| 82 |
pub fn initialize(_argc: *const isize, _argv: *const *const *const u8) -> isize { |
| 83 |
// Registers a panic hook that aborts the process before unwinding. |
| 84 |
// It is useful to abort before unwinding so that the fuzzer will then be |
| 85 |
// able to analyse the process stack frames to tell different bugs appart. |
| 86 |
// |
| 87 |
// HACK / FIXME: it would be better to use `-C panic=abort` but it's currently |
| 88 |
// impossible to build code using compiler plugins with this flag. |
| 89 |
// We will be able to remove this code when |
| 90 |
// https://github.com/rust-lang/cargo/issues/5423 is fixed. |
| 91 |
let default_hook = std::panic::take_hook(); |
| 92 |
std::panic::set_hook(Box::new(move |panic_info| { |
| 93 |
default_hook(panic_info); |
| 94 |
std::process::abort(); |
| 95 |
})); |
| 96 |
0 |
| 97 |
} |
| 98 |
|
| 99 |
/// Define a fuzz target. |
| 100 |
/// |
| 101 |
/// ## Example |
| 102 |
/// |
| 103 |
/// This example takes a `&[u8]` slice and attempts to parse it. The parsing |
| 104 |
/// might fail and return an `Err`, but it shouldn't ever panic or segfault. |
| 105 |
/// |
| 106 |
/// ```no_run |
| 107 |
/// #![no_main] |
| 108 |
/// |
| 109 |
/// use libfuzzer_sys::fuzz_target; |
| 110 |
/// |
| 111 |
/// // Note: `|input|` is short for `|input: &[u8]|`. |
| 112 |
/// fuzz_target!(|input| { |
| 113 |
/// let _result: Result<_, _> = my_crate::parse(input); |
| 114 |
/// }); |
| 115 |
/// # mod my_crate { pub fn parse(_: &[u8]) -> Result<(), ()> { unimplemented!() } } |
| 116 |
/// ``` |
| 117 |
/// |
| 118 |
/// ## Rejecting Inputs |
| 119 |
/// |
| 120 |
/// It may be desirable to reject some inputs, i.e. to not add them to the |
| 121 |
/// corpus. |
| 122 |
/// |
| 123 |
/// For example, when fuzzing an API consisting of parsing and other logic, |
| 124 |
/// one may want to allow only those inputs into the corpus that parse |
| 125 |
/// successfully. To indicate whether an input should be kept in or rejected |
| 126 |
/// from the corpus, return either [Corpus::Keep] or [Corpus::Reject] from your |
| 127 |
/// fuzz target. The default behavior (e.g. if `()` is returned) is to keep the |
| 128 |
/// input in the corpus. |
| 129 |
/// |
| 130 |
/// For example: |
| 131 |
/// |
| 132 |
/// ```no_run |
| 133 |
/// #![no_main] |
| 134 |
/// |
| 135 |
/// use libfuzzer_sys::{Corpus, fuzz_target}; |
| 136 |
/// |
| 137 |
/// fuzz_target!(|input: String| -> Corpus { |
| 138 |
/// let parts: Vec<&str> = input.splitn(2, '=').collect(); |
| 139 |
/// if parts.len() != 2 { |
| 140 |
/// return Corpus::Reject; |
| 141 |
/// } |
| 142 |
/// |
| 143 |
/// let key = parts[0]; |
| 144 |
/// let value = parts[1]; |
| 145 |
/// let _result: Result<_, _> = my_crate::parse(key, value); |
| 146 |
/// Corpus::Keep |
| 147 |
/// }); |
| 148 |
/// # mod my_crate { pub fn parse(_key: &str, _value: &str) -> Result<(), ()> { unimplemented!() } } |
| 149 |
/// ``` |
| 150 |
/// |
| 151 |
/// ## Arbitrary Input Types |
| 152 |
/// |
| 153 |
/// The input is a `&[u8]` slice by default, but you can take arbitrary input |
| 154 |
/// types, as long as the type implements [the `arbitrary` crate's `Arbitrary` |
| 155 |
/// trait](https://docs.rs/arbitrary/*/arbitrary/trait.Arbitrary.html) (which is |
| 156 |
/// also re-exported as `libfuzzer_sys::arbitrary::Arbitrary` for convenience). |
| 157 |
/// |
| 158 |
/// For example, if you wanted to take an arbitrary RGB color, you could do the |
| 159 |
/// following: |
| 160 |
/// |
| 161 |
/// ```no_run |
| 162 |
/// #![no_main] |
| 163 |
/// # mod foo { |
| 164 |
/// |
| 165 |
/// use libfuzzer_sys::{arbitrary::{Arbitrary, Error, Unstructured}, fuzz_target}; |
| 166 |
/// |
| 167 |
/// #[derive(Debug)] |
| 168 |
/// pub struct Rgb { |
| 169 |
/// r: u8, |
| 170 |
/// g: u8, |
| 171 |
/// b: u8, |
| 172 |
/// } |
| 173 |
/// |
| 174 |
/// impl<'a> Arbitrary<'a> for Rgb { |
| 175 |
/// fn arbitrary(raw: &mut Unstructured<'a>) -> Result<Self, Error> { |
| 176 |
/// let mut buf = [0; 3]; |
| 177 |
/// raw.fill_buffer(&mut buf)?; |
| 178 |
/// let r = buf[0]; |
| 179 |
/// let g = buf[1]; |
| 180 |
/// let b = buf[2]; |
| 181 |
/// Ok(Rgb { r, g, b }) |
| 182 |
/// } |
| 183 |
/// } |
| 184 |
/// |
| 185 |
/// // Write a fuzz target that works with RGB colors instead of raw bytes. |
| 186 |
/// fuzz_target!(|color: Rgb| { |
| 187 |
/// my_crate::convert_color(color); |
| 188 |
/// }); |
| 189 |
/// # mod my_crate { |
| 190 |
/// # use super::Rgb; |
| 191 |
/// # pub fn convert_color(_: Rgb) {} |
| 192 |
/// # } |
| 193 |
/// # } |
| 194 |
/// ``` |
| 195 |
/// |
| 196 |
/// You can also enable the `arbitrary` crate's custom derive via this crate's |
| 197 |
/// `"arbitrary-derive"` cargo feature. |
| 198 |
/// |
| 199 |
/// ## Init Code |
| 200 |
/// |
| 201 |
/// Init code to the fuzz target by using the `init` keyword. This is called once before the fuzzer starts. |
| 202 |
/// Supports short |input| or |input: <type>| syntax. |
| 203 |
/// |
| 204 |
/// ```no_run |
| 205 |
/// #![no_main] |
| 206 |
/// |
| 207 |
/// use libfuzzer_sys::fuzz_target; |
| 208 |
/// use std::collections::HashSet; |
| 209 |
/// use std::sync::OnceLock; |
| 210 |
/// |
| 211 |
/// static DICTIONARY: OnceLock<HashSet<String>> = OnceLock::new(); |
| 212 |
/// |
| 213 |
/// fuzz_target!( |
| 214 |
/// init: { |
| 215 |
/// let read_dictionary = |_| unimplemented!(); |
| 216 |
/// let dictionary = read_dictionary("/usr/share/dict/words"); |
| 217 |
/// DICTIONARY.set(dictionary).unwrap(); |
| 218 |
/// }, |
| 219 |
/// |input| { |
| 220 |
/// // Use the initialized `DICTIONARY` here... |
| 221 |
/// } |
| 222 |
/// ); |
| 223 |
/// ``` |
| 224 |
/// |
| 225 |
#[macro_export] |
| 226 |
macro_rules! fuzz_target { |
| 227 |
(|$bytes:ident| $body:expr) => { |
| 228 |
$crate::fuzz_target!(init: (), |$bytes: &[u8]| -> () { $body }); |
| 229 |
}; |
| 230 |
|
| 231 |
(|$bytes:ident: &[u8]| $body:expr) => { |
| 232 |
$crate::fuzz_target!(init: (), |$bytes: &[u8]| -> () { $body }); |
| 233 |
}; |
| 234 |
|
| 235 |
(|$bytes:ident: &[u8]| -> $rty:ty $body:block) => { |
| 236 |
$crate::fuzz_target!(init: (), |$bytes: &[u8]| -> $rty { $body }); |
| 237 |
}; |
| 238 |
|
| 239 |
(init: $init:expr, |$bytes:ident| $body:expr) => { |
| 240 |
$crate::fuzz_target!(init: $init, |$bytes: &[u8]| -> () { $body }); |
| 241 |
}; |
| 242 |
|
| 243 |
(init: $init:expr, |$bytes:ident: &[u8]| $body:expr) => { |
| 244 |
$crate::fuzz_target!(init: $init, |$bytes: &[u8]| -> () { $body }); |
| 245 |
}; |
| 246 |
|
| 247 |
(init: $init:expr, |$bytes:ident: &[u8]| -> $rty:ty $body:block) => { |
| 248 |
const _: () = { |
| 249 |
/// Auto-generated functions |
| 250 |
/// LLVMFuzzerInitialize is called once before the fuzzer starts. |
| 251 |
#[no_mangle] |
| 252 |
pub extern "C" fn LLVMFuzzerInitialize(_argc: *const isize, _argv: *const *const *const u8) -> isize { |
| 253 |
$crate::initialize(_argc, _argv); |
| 254 |
|
| 255 |
// Supplied init code |
| 256 |
$init; |
| 257 |
0 |
| 258 |
} |
| 259 |
|
| 260 |
#[no_mangle] |
| 261 |
pub extern "C" fn rust_fuzzer_test_input(bytes: &[u8]) -> i32 { |
| 262 |
// When `RUST_LIBFUZZER_DEBUG_PATH` is set, write the debug |
| 263 |
// formatting of the input to that file. This is only intended for |
| 264 |
// `cargo fuzz`'s use! |
| 265 |
|
| 266 |
// `RUST_LIBFUZZER_DEBUG_PATH` is set in initialization. |
| 267 |
if let Some(path) = $crate::rust_libfuzzer_debug_path() { |
| 268 |
use std::io::Write; |
| 269 |
let mut file = std::fs::File::create(path) |
| 270 |
.expect("failed to create `RUST_LIBFUZZER_DEBUG_PATH` file"); |
| 271 |
writeln!(&mut file, "{:?}", bytes) |
| 272 |
.expect("failed to write to `RUST_LIBFUZZER_DEBUG_PATH` file"); |
| 273 |
return 0; |
| 274 |
} |
| 275 |
|
| 276 |
let result = ::libfuzzer_sys::Corpus::from(__libfuzzer_sys_run(bytes)); |
| 277 |
result.to_libfuzzer_code() |
| 278 |
} |
| 279 |
|
| 280 |
// Split out the actual fuzzer into a separate function which is |
| 281 |
// tagged as never being inlined. This ensures that if the fuzzer |
| 282 |
// panics there's at least one stack frame which is named uniquely |
| 283 |
// according to this specific fuzzer that this is embedded within. |
| 284 |
// |
| 285 |
// Systems like oss-fuzz try to deduplicate crashes and without this |
| 286 |
// panics in separate fuzzers can accidentally appear the same |
| 287 |
// because each fuzzer will have a function called |
| 288 |
// `rust_fuzzer_test_input`. By using a normal Rust function here |
| 289 |
// it's named something like `the_fuzzer_name::_::__libfuzzer_sys_run` which should |
| 290 |
// ideally help prevent oss-fuzz from deduplicate fuzz bugs across |
| 291 |
// distinct targets accidentally. |
| 292 |
#[inline(never)] |
| 293 |
fn __libfuzzer_sys_run($bytes: &[u8]) -> $rty { |
| 294 |
$body |
| 295 |
} |
| 296 |
}; |
| 297 |
}; |
| 298 |
|
| 299 |
(|$data:ident: $dty:ty| $body:expr) => { |
| 300 |
$crate::fuzz_target!(init: (), |$data: $dty| -> () { $body }); |
| 301 |
}; |
| 302 |
|
| 303 |
(|$data:ident: $dty:ty| -> $rty:ty $body:block) => { |
| 304 |
$crate::fuzz_target!(init: (), |$data: $dty| -> $rty { $body }); |
| 305 |
}; |
| 306 |
|
| 307 |
(init: $init:expr, |$data:ident: $dty:ty| $body:expr) => { |
| 308 |
$crate::fuzz_target!(init: $init, |$data: $dty| -> () { $body }); |
| 309 |
}; |
| 310 |
|
| 311 |
(init: $init:expr, |$data:ident: $dty:ty| -> $rty:ty $body:block) => { |
| 312 |
const _: () = { |
| 313 |
/// Auto-generated functions |
| 314 |
/// LLVMFuzzerInitialize is called once before the fuzzer starts. |
| 315 |
#[no_mangle] |
| 316 |
pub extern "C" fn LLVMFuzzerInitialize(_argc: *const isize, _argv: *const *const *const u8) -> isize { |
| 317 |
$crate::initialize(_argc, _argv); |
| 318 |
|
| 319 |
// Supplied init code |
| 320 |
$init; |
| 321 |
0 |
| 322 |
} |
| 323 |
|
| 324 |
#[no_mangle] |
| 325 |
pub extern "C" fn rust_fuzzer_test_input(bytes: &[u8]) -> i32 { |
| 326 |
use $crate::arbitrary::{Arbitrary, Unstructured}; |
| 327 |
|
| 328 |
// Early exit if we don't have enough bytes for the `Arbitrary` |
| 329 |
// implementation. This helps the fuzzer avoid exploring all the |
| 330 |
// different not-enough-input-bytes paths inside the `Arbitrary` |
| 331 |
// implementation. Additionally, it exits faster, letting the fuzzer |
| 332 |
// get to longer inputs that actually lead to interesting executions |
| 333 |
// quicker. |
| 334 |
if bytes.len() < <$dty as Arbitrary>::size_hint(0).0 { |
| 335 |
return -1; |
| 336 |
} |
| 337 |
|
| 338 |
let mut u = Unstructured::new(bytes); |
| 339 |
let data = <$dty as Arbitrary>::arbitrary_take_rest(u); |
| 340 |
|
| 341 |
// When `RUST_LIBFUZZER_DEBUG_PATH` is set, write the debug |
| 342 |
// formatting of the input to that file. This is only intended for |
| 343 |
// `cargo fuzz`'s use! |
| 344 |
|
| 345 |
// `RUST_LIBFUZZER_DEBUG_PATH` is set in initialization. |
| 346 |
if let Some(path) = $crate::rust_libfuzzer_debug_path() { |
| 347 |
use std::io::Write; |
| 348 |
let mut file = std::fs::File::create(path) |
| 349 |
.expect("failed to create `RUST_LIBFUZZER_DEBUG_PATH` file"); |
| 350 |
(match data { |
| 351 |
Ok(data) => writeln!(&mut file, "{:#?}", data), |
| 352 |
Err(err) => writeln!(&mut file, "Arbitrary Error: {}", err), |
| 353 |
}) |
| 354 |
.expect("failed to write to `RUST_LIBFUZZER_DEBUG_PATH` file"); |
| 355 |
return -1; |
| 356 |
} |
| 357 |
|
| 358 |
let data = match data { |
| 359 |
Ok(d) => d, |
| 360 |
Err(_) => return -1, |
| 361 |
}; |
| 362 |
|
| 363 |
let result = ::libfuzzer_sys::Corpus::from(__libfuzzer_sys_run(data)); |
| 364 |
result.to_libfuzzer_code() |
| 365 |
} |
| 366 |
// See above for why this is split to a separate function. |
| 367 |
#[inline(never)] |
| 368 |
fn __libfuzzer_sys_run($data: $dty) -> $rty { |
| 369 |
$body |
| 370 |
} |
| 371 |
}; |
| 372 |
}; |
| 373 |
} |
| 374 |
|
| 375 |
/// Define a custom mutator. |
| 376 |
/// |
| 377 |
/// This is optional, and libFuzzer will use its own, default mutation strategy |
| 378 |
/// if this is not provided. |
| 379 |
/// |
| 380 |
/// You might consider using a custom mutator when your fuzz target is very |
| 381 |
/// particular about the shape of its input: |
| 382 |
/// |
| 383 |
/// * You want to fuzz "deeper" than just the parser. |
| 384 |
/// * The input contains checksums that have to match the hash of some subset of |
| 385 |
/// the data or else the whole thing is invalid, and therefore mutating any of |
| 386 |
/// that subset means you need to recompute the checksums. |
| 387 |
/// * Small random changes to the input buffer make it invalid. |
| 388 |
/// |
| 389 |
/// That is, a custom mutator is useful in similar situations where [a `T: |
| 390 |
/// Arbitrary` input type](macro.fuzz_target.html#arbitrary-input-types) is |
| 391 |
/// useful. Note that the two approaches are not mutually exclusive; you can use |
| 392 |
/// whichever is easier for your problem domain or both! |
| 393 |
/// |
| 394 |
/// ## Implementation Contract |
| 395 |
/// |
| 396 |
/// The original, unmodified input is given in `data[..size]`. |
| 397 |
/// |
| 398 |
/// You must modify the data in place and return the new size. |
| 399 |
/// |
| 400 |
/// The new size should not be greater than `max_size`. If this is not the case, |
| 401 |
/// then the `data` will be truncated to fit within `max_size`. Note that |
| 402 |
/// `max_size < size` is possible when shrinking test cases. |
| 403 |
/// |
| 404 |
/// You must produce the same mutation given the same `seed`. Generally, when |
| 405 |
/// choosing what kind of mutation to make or where to mutate, you should start |
| 406 |
/// by creating a random number generator (RNG) that is seeded with the given |
| 407 |
/// `seed` and then consult the RNG whenever making a decision: |
| 408 |
/// |
| 409 |
/// ```no_run |
| 410 |
/// #![no_main] |
| 411 |
/// |
| 412 |
/// use rand::{rngs::StdRng, Rng, SeedableRng}; |
| 413 |
/// |
| 414 |
/// libfuzzer_sys::fuzz_mutator!(|data: &mut [u8], size: usize, max_size: usize, seed: u32| { |
| 415 |
/// let mut rng = StdRng::seed_from_u64(seed as u64); |
| 416 |
/// |
| 417 |
/// # let first_mutation = |_, _, _, _| todo!(); |
| 418 |
/// # let second_mutation = |_, _, _, _| todo!(); |
| 419 |
/// # let third_mutation = |_, _, _, _| todo!(); |
| 420 |
/// # let fourth_mutation = |_, _, _, _| todo!(); |
| 421 |
/// // Choose which of our four supported kinds of mutations we want to make. |
| 422 |
/// match rng.gen_range(0..4) { |
| 423 |
/// 0 => first_mutation(rng, data, size, max_size), |
| 424 |
/// 1 => second_mutation(rng, data, size, max_size), |
| 425 |
/// 2 => third_mutation(rng, data, size, max_size), |
| 426 |
/// 3 => fourth_mutation(rng, data, size, max_size), |
| 427 |
/// _ => unreachable!() |
| 428 |
/// } |
| 429 |
/// }); |
| 430 |
/// ``` |
| 431 |
/// |
| 432 |
/// ## Example: Compression |
| 433 |
/// |
| 434 |
/// Consider a simple fuzz target that takes compressed data as input, |
| 435 |
/// decompresses it, and then asserts that the decompressed data doesn't begin |
| 436 |
/// with "boom". It is difficult for `libFuzzer` (or any other fuzzer) to crash |
| 437 |
/// this fuzz target because nearly all mutations it makes will invalidate the |
| 438 |
/// compression format. Therefore, we use a custom mutator that decompresses the |
| 439 |
/// raw input, mutates the decompressed data, and then recompresses it. This |
| 440 |
/// allows `libFuzzer` to quickly discover crashing inputs. |
| 441 |
/// |
| 442 |
/// ```no_run |
| 443 |
/// #![no_main] |
| 444 |
/// |
| 445 |
/// use flate2::{read::GzDecoder, write::GzEncoder, Compression}; |
| 446 |
/// use libfuzzer_sys::{fuzz_mutator, fuzz_target}; |
| 447 |
/// use std::io::{Read, Write}; |
| 448 |
/// |
| 449 |
/// fuzz_target!(|data: &[u8]| { |
| 450 |
/// // Decompress the input data and crash if it starts with "boom". |
| 451 |
/// if let Some(data) = decompress(data) { |
| 452 |
/// if data.starts_with(b"boom") { |
| 453 |
/// panic!(); |
| 454 |
/// } |
| 455 |
/// } |
| 456 |
/// }); |
| 457 |
/// |
| 458 |
/// fuzz_mutator!( |
| 459 |
/// |data: &mut [u8], size: usize, max_size: usize, _seed: u32| { |
| 460 |
/// // Decompress the input data. If that fails, use a dummy value. |
| 461 |
/// let mut decompressed = decompress(&data[..size]).unwrap_or_else(|| b"hi".to_vec()); |
| 462 |
/// |
| 463 |
/// // Mutate the decompressed data with `libFuzzer`'s default mutator. Make |
| 464 |
/// // the `decompressed` vec's extra capacity available for insertion |
| 465 |
/// // mutations via `resize`. |
| 466 |
/// let len = decompressed.len(); |
| 467 |
/// let cap = decompressed.capacity(); |
| 468 |
/// decompressed.resize(cap, 0); |
| 469 |
/// let new_decompressed_size = libfuzzer_sys::fuzzer_mutate(&mut decompressed, len, cap); |
| 470 |
/// |
| 471 |
/// // Recompress the mutated data. |
| 472 |
/// let compressed = compress(&decompressed[..new_decompressed_size]); |
| 473 |
/// |
| 474 |
/// // Copy the recompressed mutated data into `data` and return the new size. |
| 475 |
/// let new_size = std::cmp::min(max_size, compressed.len()); |
| 476 |
/// data[..new_size].copy_from_slice(&compressed[..new_size]); |
| 477 |
/// new_size |
| 478 |
/// } |
| 479 |
/// ); |
| 480 |
/// |
| 481 |
/// fn decompress(compressed_data: &[u8]) -> Option<Vec<u8>> { |
| 482 |
/// let mut decoder = GzDecoder::new(compressed_data); |
| 483 |
/// let mut decompressed = Vec::new(); |
| 484 |
/// if decoder.read_to_end(&mut decompressed).is_ok() { |
| 485 |
/// Some(decompressed) |
| 486 |
/// } else { |
| 487 |
/// None |
| 488 |
/// } |
| 489 |
/// } |
| 490 |
/// |
| 491 |
/// fn compress(data: &[u8]) -> Vec<u8> { |
| 492 |
/// let mut encoder = GzEncoder::new(Vec::new(), Compression::default()); |
| 493 |
/// encoder |
| 494 |
/// .write_all(data) |
| 495 |
/// .expect("writing into a vec is infallible"); |
| 496 |
/// encoder.finish().expect("writing into a vec is infallible") |
| 497 |
/// } |
| 498 |
/// ``` |
| 499 |
/// |
| 500 |
/// This example is inspired by [a similar example from the official `libFuzzer` |
| 501 |
/// docs](https://github.com/google/fuzzing/blob/master/docs/structure-aware-fuzzing.md#example-compression). |
| 502 |
/// |
| 503 |
/// ## More Example Ideas |
| 504 |
/// |
| 505 |
/// * A PNG custom mutator that decodes a PNG, mutates the image, and then |
| 506 |
/// re-encodes the mutated image as a new PNG. |
| 507 |
/// |
| 508 |
/// * A [`serde`](https://serde.rs/) custom mutator that deserializes your |
| 509 |
/// structure, mutates it, and then reserializes it. |
| 510 |
/// |
| 511 |
/// * A Wasm binary custom mutator that inserts, replaces, and removes a |
| 512 |
/// bytecode instruction in a function's body. |
| 513 |
/// |
| 514 |
/// * An HTTP request custom mutator that inserts, replaces, and removes a |
| 515 |
/// header from an HTTP request. |
| 516 |
#[macro_export] |
| 517 |
macro_rules! fuzz_mutator { |
| 518 |
( |
| 519 |
| |
| 520 |
$data:ident : &mut [u8] , |
| 521 |
$size:ident : usize , |
| 522 |
$max_size:ident : usize , |
| 523 |
$seed:ident : u32 $(,)* |
| 524 |
| |
| 525 |
$body:block |
| 526 |
) => { |
| 527 |
/// Auto-generated function. Do not use; only for LibFuzzer's |
| 528 |
/// consumption. |
| 529 |
#[export_name = "LLVMFuzzerCustomMutator"] |
| 530 |
#[doc(hidden)] |
| 531 |
pub unsafe fn rust_fuzzer_custom_mutator( |
| 532 |
$data: *mut u8, |
| 533 |
$size: usize, |
| 534 |
$max_size: usize, |
| 535 |
$seed: std::os::raw::c_uint, |
| 536 |
) -> usize { |
| 537 |
// Depending on if we are growing or shrinking the test case, `size` |
| 538 |
// might be larger or smaller than `max_size`. The `data`'s capacity |
| 539 |
// is the maximum of the two. |
| 540 |
let len = std::cmp::max($max_size, $size); |
| 541 |
let $data: &mut [u8] = std::slice::from_raw_parts_mut($data, len); |
| 542 |
|
| 543 |
// `unsigned int` is generally a `u32`, but not on all targets. Do |
| 544 |
// an infallible (and potentially lossy, but that's okay because it |
| 545 |
// preserves determinism) conversion. |
| 546 |
let $seed = $seed as u32; |
| 547 |
|
| 548 |
// Define and invoke a new, safe function so that the body doesn't |
| 549 |
// inherit `unsafe`. |
| 550 |
fn custom_mutator( |
| 551 |
$data: &mut [u8], |
| 552 |
$size: usize, |
| 553 |
$max_size: usize, |
| 554 |
$seed: u32, |
| 555 |
) -> usize { |
| 556 |
$body |
| 557 |
} |
| 558 |
let new_size = custom_mutator($data, $size, $max_size, $seed); |
| 559 |
|
| 560 |
// Truncate the new size if it is larger than the max. |
| 561 |
std::cmp::min(new_size, $max_size) |
| 562 |
} |
| 563 |
}; |
| 564 |
} |
| 565 |
|
| 566 |
/// The default `libFuzzer` mutator. |
| 567 |
/// |
| 568 |
/// You generally don't have to use this at all unless you're defining a |
| 569 |
/// custom mutator with [the `fuzz_mutator!` macro][crate::fuzz_mutator]. |
| 570 |
/// |
| 571 |
/// Mutates `data[..size]` in place such that the mutated data is no larger than |
| 572 |
/// `max_size` and returns the new size of the mutated data. |
| 573 |
/// |
| 574 |
/// To only allow shrinking mutations, make `max_size < size`. |
| 575 |
/// |
| 576 |
/// To additionally allow mutations that grow the size of the data, make |
| 577 |
/// `max_size > size`. |
| 578 |
/// |
| 579 |
/// Both `size` and `max_size` must be less than or equal to `data.len()`. |
| 580 |
/// |
| 581 |
/// # Example |
| 582 |
/// |
| 583 |
/// ```no_run |
| 584 |
/// // Create some data in a buffer. |
| 585 |
/// let mut data = vec![0; 128]; |
| 586 |
/// data[..b"hello".len()].copy_from_slice(b"hello"); |
| 587 |
/// |
| 588 |
/// // Ask `libFuzzer` to mutate the data. By setting `max_size` to our buffer's |
| 589 |
/// // full length, we are allowing `libFuzzer` to perform mutations that grow |
| 590 |
/// // the size of the data, such as insertions. |
| 591 |
/// let size = b"hello".len(); |
| 592 |
/// let max_size = data.len(); |
| 593 |
/// let new_size = libfuzzer_sys::fuzzer_mutate(&mut data, size, max_size); |
| 594 |
/// |
| 595 |
/// // Get the mutated data out of the buffer. |
| 596 |
/// let mutated_data = &data[..new_size]; |
| 597 |
/// ``` |
| 598 |
pub fn fuzzer_mutate(data: &mut [u8], size: usize, max_size: usize) -> usize { |
| 599 |
assert!(size <= data.len()); |
| 600 |
assert!(max_size <= data.len()); |
| 601 |
let new_size = unsafe { LLVMFuzzerMutate(data.as_mut_ptr(), size, max_size) }; |
| 602 |
assert!(new_size <= data.len()); |
| 603 |
new_size |
| 604 |
} |
| 605 |
|
| 606 |
/// Define a custom cross-over function to combine test cases. |
| 607 |
/// |
| 608 |
/// This is optional, and libFuzzer will use its own, default cross-over strategy |
| 609 |
/// if this is not provided. (As of the time of writing, this default strategy |
| 610 |
/// takes alternating byte sequences from the two test cases, to construct the |
| 611 |
/// new one) (see `FuzzerCrossOver.cpp`) |
| 612 |
/// |
| 613 |
/// This could potentially be useful if your input is, for instance, a |
| 614 |
/// sequence of fixed sized, multi-byte values and the crossover could then |
| 615 |
/// merge discrete values rather than joining parts of a value. |
| 616 |
/// |
| 617 |
/// ## Implementation Contract |
| 618 |
/// |
| 619 |
/// The original, read-only inputs are given in the full slices of `data1`, and |
| 620 |
/// `data2` (as opposed to the, potentially, partial slice of `data` in |
| 621 |
/// [the `fuzz_mutator!` macro][crate::fuzz_mutator]). |
| 622 |
/// |
| 623 |
/// You must place the new input merged from the two existing inputs' data |
| 624 |
/// into `out` and return the size of the relevant data written to that slice. |
| 625 |
/// |
| 626 |
/// The deterministic requirements from [the `fuzz_mutator!` macro][crate::fuzz_mutator] |
| 627 |
/// apply as well to the `seed` parameter |
| 628 |
/// |
| 629 |
/// ## Example: Floating-Point Sum NaN |
| 630 |
/// |
| 631 |
/// ```no_run |
| 632 |
/// #![no_main] |
| 633 |
/// |
| 634 |
/// use libfuzzer_sys::{fuzz_crossover, fuzz_mutator, fuzz_target, fuzzer_mutate}; |
| 635 |
/// use rand::{rngs::StdRng, Rng, SeedableRng}; |
| 636 |
/// use std::mem::size_of; |
| 637 |
/// |
| 638 |
/// fuzz_target!(|data: &[u8]| { |
| 639 |
/// let (_, floats, _) = unsafe { data.align_to::<f64>() }; |
| 640 |
/// |
| 641 |
/// let res = floats |
| 642 |
/// .iter() |
| 643 |
/// .fold(0.0, |a, b| if b.is_nan() { a } else { a + b }); |
| 644 |
/// |
| 645 |
/// assert!( |
| 646 |
/// !res.is_nan(), |
| 647 |
/// "The sum of the following floats resulted in a NaN: {floats:?}" |
| 648 |
/// ); |
| 649 |
/// }); |
| 650 |
/// |
| 651 |
/// // Inject some ...potentially problematic values to make the example close |
| 652 |
/// // more quickly. |
| 653 |
/// fuzz_mutator!(|data: &mut [u8], size: usize, max_size: usize, seed: u32| { |
| 654 |
/// let mut gen = StdRng::seed_from_u64(seed.into()); |
| 655 |
/// |
| 656 |
/// let (_, floats, _) = unsafe { data[..size].align_to_mut::<f64>() }; |
| 657 |
/// |
| 658 |
/// let x = gen.gen_range(0..=1000); |
| 659 |
/// if x == 0 && !floats.is_empty() { |
| 660 |
/// floats[0] = f64::INFINITY; |
| 661 |
/// } else if x == 1000 && floats.len() > 1 { |
| 662 |
/// floats[1] = f64::NEG_INFINITY; |
| 663 |
/// } else { |
| 664 |
/// return fuzzer_mutate(data, size, max_size); |
| 665 |
/// } |
| 666 |
/// |
| 667 |
/// size |
| 668 |
/// }); |
| 669 |
/// |
| 670 |
/// fuzz_crossover!(|data1: &[u8], data2: &[u8], out: &mut [u8], _seed: u32| { |
| 671 |
/// // Decode each source to see how many floats we can pull with proper |
| 672 |
/// // alignment, and destination as to how many will fit with proper alignment |
| 673 |
/// // |
| 674 |
/// // Keep track of the unaligned prefix to `out`, as we will need to remember |
| 675 |
/// // that those bytes will remain prepended to the actual floats that we |
| 676 |
/// // write into the out buffer. |
| 677 |
/// let (out_pref, out_floats, _) = unsafe { out.align_to_mut::<f64>() }; |
| 678 |
/// let (_, d1_floats, _) = unsafe { data1.align_to::<f64>() }; |
| 679 |
/// let (_, d2_floats, _) = unsafe { data2.align_to::<f64>() }; |
| 680 |
/// |
| 681 |
/// // Put into the destination, floats first from data1 then from data2, ...if |
| 682 |
/// // possible given the size of `out` |
| 683 |
/// let mut i: usize = 0; |
| 684 |
/// for float in d1_floats.iter().chain(d2_floats).take(out_floats.len()) { |
| 685 |
/// out_floats[i] = *float; |
| 686 |
/// i += 1; |
| 687 |
/// } |
| 688 |
/// |
| 689 |
/// // Now that we have written the true floats, report back to the fuzzing |
| 690 |
/// // engine that we left the unaligned `out` prefix bytes at the beginning of |
| 691 |
/// // `out` and also then the floats that we wrote into the aligned float |
| 692 |
/// // section. |
| 693 |
/// out_pref.len() * size_of::<u8>() + i * size_of::<f64>() |
| 694 |
/// }); |
| 695 |
/// ``` |
| 696 |
/// |
| 697 |
/// This example is a minimized version of [Erik Rigtorp's floating point |
| 698 |
/// summation fuzzing example][1]. A more detailed version of this experiment |
| 699 |
/// can be found in the `example_crossover` directory. |
| 700 |
/// |
| 701 |
/// [1]: https://rigtorp.se/fuzzing-floating-point-code/ |
| 702 |
#[macro_export] |
| 703 |
macro_rules! fuzz_crossover { |
| 704 |
( |
| 705 |
| |
| 706 |
$data1:ident : &[u8] , |
| 707 |
$data2:ident : &[u8] , |
| 708 |
$out:ident : &mut [u8] , |
| 709 |
$seed:ident : u32 $(,)* |
| 710 |
| |
| 711 |
$body:block |
| 712 |
) => { |
| 713 |
/// Auto-generated function. Do not use; only for LibFuzzer's |
| 714 |
/// consumption. |
| 715 |
#[export_name = "LLVMFuzzerCustomCrossOver"] |
| 716 |
#[doc(hidden)] |
| 717 |
pub unsafe fn rust_fuzzer_custom_crossover( |
| 718 |
$data1: *const u8, |
| 719 |
size1: usize, |
| 720 |
$data2: *const u8, |
| 721 |
size2: usize, |
| 722 |
$out: *mut u8, |
| 723 |
max_out_size: usize, |
| 724 |
$seed: std::os::raw::c_uint, |
| 725 |
) -> usize { |
| 726 |
let $data1: &[u8] = std::slice::from_raw_parts($data1, size1); |
| 727 |
let $data2: &[u8] = std::slice::from_raw_parts($data2, size2); |
| 728 |
let $out: &mut [u8] = std::slice::from_raw_parts_mut($out, max_out_size); |
| 729 |
|
| 730 |
// `unsigned int` is generally a `u32`, but not on all targets. Do |
| 731 |
// an infallible (and potentially lossy, but that's okay because it |
| 732 |
// preserves determinism) conversion. |
| 733 |
let $seed = $seed as u32; |
| 734 |
|
| 735 |
// Define and invoke a new, safe function so that the body doesn't |
| 736 |
// inherit `unsafe`. |
| 737 |
fn custom_crossover( |
| 738 |
$data1: &[u8], |
| 739 |
$data2: &[u8], |
| 740 |
$out: &mut [u8], |
| 741 |
$seed: u32, |
| 742 |
) -> usize { |
| 743 |
$body |
| 744 |
} |
| 745 |
|
| 746 |
custom_crossover($data1, $data2, $out, $seed) |
| 747 |
} |
| 748 |
}; |
| 749 |
} |