Make input loading fallible in SyncFromDiskStage (#3195)

* Make input loading fallible in SyncFromDiskStage

* fmt

* Add InvalidInput in Error enum and skip the Input in SyncFromDiskStage if it is encountered

* sync: remove file if error on loading in SyncFromDiskStage

* add reason to Error::InvalidInput

* sync make failure log a warning. clippy, fmt

* typo

* fmt

* fmt

---------

Co-authored-by: Dongjia "toka" Zhang <tokazerkje@outlook.com>
Co-authored-by: Dominik Maier <domenukk@gmail.com>

libafl/src/stages/sync.rs

@@ -58,6 +58,7 @@ impl SyncFromDiskMetadata {
 }
 
 /// A stage that loads testcases from disk to sync with other fuzzers such as AFL++
+/// When syncing, the stage will ignore `Error::InvalidInput` and will skip the file.
 #[derive(Debug)]
 pub struct SyncFromDiskStage<CB, E, EM, I, S, Z> {
     name: Cow<'static, str>,
@@ -125,15 +126,25 @@ where
         let to_sync = sync_from_disk_metadata.left_to_sync.clone();
         log::debug!("Number of files to sync: {:?}", to_sync.len());
         for path in to_sync {
-            let input = (self.load_callback)(fuzzer, state, &path)?;
             // Removing each path from the `left_to_sync` Vec before evaluating
             // prevents duplicate processing and ensures that each file is evaluated only once. This approach helps
-            // avoid potential infinite loops that may occur if a file is an objective.
+            // avoid potential infinite loops that may occur if a file is an objective or an invalid input.
             state
                 .metadata_mut::<SyncFromDiskMetadata>()
                 .unwrap()
                 .left_to_sync
                 .retain(|p| p != &path);
+            let input = match (self.load_callback)(fuzzer, state, &path) {
+                Ok(input) => input,
+                Err(Error::InvalidInput(reason, _)) => {
+                    log::warn!(
+                        "Invalid input found in {} when syncing; reason {reason}; skipping;",
+                        path.display()
+                    );
+                    continue;
+                }
+                Err(e) => return Err(e),
+            };
             log::debug!("Syncing and evaluating {}", path.display());
             fuzzer.evaluate_input(state, executor, manager, &input)?;
         }
@@ -161,6 +172,7 @@ where
 
 impl<CB, E, EM, I, S, Z> SyncFromDiskStage<CB, E, EM, I, S, Z> {
     /// Creates a new [`SyncFromDiskStage`]
+    /// To skip a file, you can return `Error::invalid_input` in `load_callback`
     #[must_use]
     pub fn new(sync_dirs: Vec<PathBuf>, load_callback: CB, interval: Duration, name: &str) -> Self {
         Self {

libafl_bolts/src/lib.rs

@@ -341,6 +341,8 @@ pub enum Error {
     InvalidCorpus(String, ErrorBacktrace),
     /// Error specific to a runtime like QEMU or Frida
     Runtime(String, ErrorBacktrace),
+    /// The `Input` was invalid.
+    InvalidInput(String, ErrorBacktrace),
 }
 
 impl Error {
@@ -369,6 +371,15 @@ impl Error {
         Error::EmptyOptional(arg.into(), ErrorBacktrace::new())
     }
 
+    /// The `Input` was invalid
+    #[must_use]
+    pub fn invalid_input<S>(reason: S) -> Self
+    where
+        S: Into<String>,
+    {
+        Error::InvalidInput(reason.into(), ErrorBacktrace::new())
+    }
+
     /// Key not in Map
     #[must_use]
     pub fn key_not_found<S>(arg: S) -> Self
@@ -580,6 +591,10 @@ impl Display for Error {
                 write!(f, "Runtime error: {0}", &s)?;
                 display_error_backtrace(f, b)
             }
+            Self::InvalidInput(s, b) => {
+                write!(f, "Encountered an invalid input: {0}", &s)?;
+                display_error_backtrace(f, b)
+            }
         }
     }
 }