Introduce LLVMFuzzerInitialize support

CHANGELOG.md

@@ -28,6 +28,20 @@ Released YYYY-MM-DD.
 
 --------------------------------------------------------------------------------
 
+## 0.4.9
+
+Released YYYY-MM-DD.
+
+### Added
+
+* The `example_init` demonstrates how to pass an initialization code block to the `fuzz_target!` macro.
+
+### Changed
+
+* The `fuzz_target!` macro now supports the generation of `LLVMFuzzerInitialize` to execute initialization code once before running the fuzzer. This change is not breaking and is completely backward compatible.
+
+--------------------------------------------------------------------------------
+
 ## 0.4.8
 
 Released 2024-11-07.

Cargo.toml

@@ -24,6 +24,7 @@ members = [
   "./example/fuzz",
   "./example_arbitrary/fuzz",
   "./example_crossover/fuzz",
+  "./example_init/fuzz",
   "./example_mutator/fuzz",
 ]
 

ci/script.sh

@@ -36,4 +36,10 @@ cargo fuzz build  --dev
 (! cargo fuzz run --release boom -- -runs=10000000)
 popd
 
+pushd ./example_init
+cargo fuzz build
+cargo fuzz build  --dev
+(! cargo fuzz run --release bigbang -- -runs=10000000)
+popd
+
 echo "All good!"

example_init/.gitignore

@@ -0,0 +1 @@
+crash-*

example_init/Cargo.toml

@@ -0,0 +1,6 @@
+[package]
+name = "example_init"
+version = "0.1.0"
+edition = "2021"
+
+[dependencies]

example_init/fuzz/Cargo.toml

@@ -0,0 +1,16 @@
+[package]
+name = "example_init-fuzz"
+version = "0.1.0"
+authors = ["Andrea Cappa"]
+edition = "2018"
+
+[package.metadata]
+cargo-fuzz = true
+
+[dependencies]
+libfuzzer-sys = { path = "../.." }
+example_init = { path = ".." }
+
+[[bin]]
+name = "bigbang"
+path = "fuzz_targets/bigbang.rs"

example_init/fuzz/fuzz_targets/bigbang.rs

@@ -0,0 +1,13 @@
+#![no_main]
+
+use libfuzzer_sys::fuzz_target;
+
+fuzz_target!(
+    init: {
+        // Custom initialization code here
+        println!("Initializing fuzzer...");
+        example_init::initialize();
+    },
+    |data: &[u8]| {
+    example_init::bigbang(data);
+});

example_init/src/lib.rs

@@ -0,0 +1,22 @@
+use std::sync::OnceLock;
+
+static EXTRA_DATA: OnceLock<&'static str> = OnceLock::new();
+
+pub fn bigbang(data: &[u8]) {
+    // The fuzzer needs to mutate input to be "bigbang!"
+    // Init needs to be called before bigbang() is called
+    // This actually proves that the fuzzer is calling init before bigbang
+    if data == &b"bigbang!"[..] && is_initialized() {
+        panic!("bigbang!");
+    }
+}
+
+pub fn initialize() {
+    EXTRA_DATA
+        .set("initialized")
+        .expect("should only initialize once");
+}
+
+pub fn is_initialized() -> bool {
+    EXTRA_DATA.get().is_some()
+}

src/lib.rs

@@ -79,7 +79,6 @@ pub fn rust_libfuzzer_debug_path() -> &'static Option<String> {
 }
 
 #[doc(hidden)]
-#[export_name = "LLVMFuzzerInitialize"]
 pub fn initialize(_argc: *const isize, _argv: *const *const *const u8) -> isize {
     // Registers a panic hook that aborts the process before unwinding.
     // It is useful to abort before unwinding so that the fuzzer will then be
@@ -89,10 +88,10 @@ pub fn initialize(_argc: *const isize, _argv: *const *const *const u8) -> isize
     // impossible to build code using compiler plugins with this flag.
     // We will be able to remove this code when
     // https://github.com/rust-lang/cargo/issues/5423 is fixed.
-    let default_hook = ::std::panic::take_hook();
-    ::std::panic::set_hook(Box::new(move |panic_info| {
+    let default_hook = std::panic::take_hook();
+    std::panic::set_hook(Box::new(move |panic_info| {
         default_hook(panic_info);
-        ::std::process::abort();
+        std::process::abort();
     }));
     0
 }
@@ -196,11 +195,37 @@ pub fn initialize(_argc: *const isize, _argv: *const *const *const u8) -> isize
 ///
 /// You can also enable the `arbitrary` crate's custom derive via this crate's
 /// `"arbitrary-derive"` cargo feature.
+/// 
+/// ## Init Code
+/// 
+/// Init code to the fuzz target by using the `init` keyword. This is called once before the fuzzer starts.
+/// Supports short |input| or |input: <type>| syntax.
+/// 
+/// ```no_run
+/// #![no_main]
+/// 
+/// use libfuzzer_sys::fuzz_target;
+/// 
+/// fuzz_target!(init: (), |input| {
+///     // ...
+/// });
+/// ```
+/// 
 #[macro_export]
 macro_rules! fuzz_target {
-    (|$bytes:ident| $body:expr) => {
+    (init: $init:expr, |$bytes:ident| $body:expr) => {
         const _: () = {
-            /// Auto-generated function
+            /// Auto-generated functions
+            /// LLVMFuzzerInitialize is called once before the fuzzer starts.
+            #[no_mangle]
+            pub extern "C" fn LLVMFuzzerInitialize(_argc: *const isize, _argv: *const *const *const u8) -> isize {
+                $crate::initialize(_argc, _argv);
+
+                // Supplied init code
+                $init;
+                0
+            }
+
             #[no_mangle]
             pub extern "C" fn rust_fuzzer_test_input(bytes: &[u8]) -> i32 {
                 // When `RUST_LIBFUZZER_DEBUG_PATH` is set, write the debug
@@ -240,17 +265,47 @@ macro_rules! fuzz_target {
         };
     };
 
+    (|$bytes:ident| $body:expr) => {
+        $crate::fuzz_target!(|$bytes: &[u8]| $body);
+    };
+
     (|$data:ident: &[u8]| $body:expr) => {
-        $crate::fuzz_target!(|$data| $body);
+        $crate::fuzz_target!(init: (), |$data| $body);
     };
 
     (|$data:ident: $dty:ty| $body:expr) => {
-        $crate::fuzz_target!(|$data: $dty| -> () { $body });
+        $crate::fuzz_target!(init: (), |$data: $dty| -> () { $body });
     };
 
     (|$data:ident: $dty:ty| -> $rty:ty $body:block) => {
+        $crate::fuzz_target!(init: (), |$data: $dty| -> $rty { $body });
+    };
+
+    (init: $init:expr, |$data:ident: &[u8]| $body:expr) => {
+        $crate::fuzz_target!(init: $init, |$data| $body);
+    };
+
+    (init: $init:expr, |$bytes:ident| $body:expr) => {
+        $crate::fuzz_target!(init: $init, |$bytes: &[u8]| $body);
+    };
+
+    (init: $init:expr, |$data:ident: $dty:ty| $body:expr) => {
+        $crate::fuzz_target!(init: $init, |$data: $dty| -> () { $body });
+    };
+
+    (init: $init:expr, |$data:ident: $dty:ty| -> $rty:ty $body:block) => {
         const _: () = {
-            /// Auto-generated function
+            /// Auto-generated functions
+            /// LLVMFuzzerInitialize is called once before the fuzzer starts.
+            #[no_mangle]
+            pub extern "C" fn LLVMFuzzerInitialize(_argc: *const isize, _argv: *const *const *const u8) -> isize {
+                $crate::initialize(_argc, _argv);
+
+                // Supplied init code
+                $init;
+                0
+            }
+
             #[no_mangle]
             pub extern "C" fn rust_fuzzer_test_input(bytes: &[u8]) -> i32 {
                 use $crate::arbitrary::{Arbitrary, Unstructured};
@@ -293,7 +348,6 @@ macro_rules! fuzz_target {
                 let result = ::libfuzzer_sys::Corpus::from(__libfuzzer_sys_run(data));
                 result.to_libfuzzer_code()
             }
-
             // See above for why this is split to a separate function.
             #[inline(never)]
             fn __libfuzzer_sys_run($data: $dty) -> $rty {