Catch panics in threads as oops

kernel/src/lib.rs

@@ -66,6 +66,7 @@ pub mod events;
 pub mod fs;
 pub mod ipc;
 pub mod net;
+pub mod oops;
 pub mod prelude;
 mod process;
 mod sched;

kernel/src/oops.rs

@@ -0,0 +1,104 @@
+// SPDX-License-Identifier: MPL-2.0
+
+//! Kernel "oops" handling.
+//!
+//! In Asterinas, a Rust panic leads to a kernel "oops". A kernel oops behaves
+//! as an exceptional control flow event. If kernel oopses happened too many
+//! times, the kernel panics and the system gets halted.
+//!
+//! Though we can recover from the Rust panics. It is generally not recommended
+//! to make Rust panics as a general exception handling mechanism. Handling
+//! exceptions with [`Result`] is more idiomatic.
+
+use alloc::{
+    boxed::Box,
+    string::{String, ToString},
+    sync::Arc,
+};
+use core::{
+    result::Result,
+    sync::atomic::{AtomicBool, AtomicUsize, Ordering},
+};
+
+use ostd::panic;
+
+use crate::{current_thread, Thread};
+
+// TODO: Control the kernel commandline parsing from the kernel crate.
+// In Linux it can be dynamically changed by writing to
+// `/proc/sys/kernel/panic`.
+static PANIC_ON_OOPS: AtomicBool = AtomicBool::new(true);
+
+/// The kernel "oops" information.
+pub struct OopsInfo {
+    /// The "oops" message.
+    pub message: String,
+    /// The thread where the "oops" happened.
+    pub thread: Arc<Thread>,
+}
+
+/// Executes the given function and catches any panics that occur.
+///
+/// All the panics in the given function will be regarded as oops. If a oops
+/// happens, this function returns `None`. Otherwise, it returns the return
+/// value of the given function.
+///
+/// If the kernel is configured to panic on oops, this function will not return
+/// when a oops happens.
+pub fn catch_panics_as_oops<F, R>(f: F) -> Result<R, OopsInfo>
+where
+    F: FnOnce() -> R,
+{
+    if PANIC_ON_OOPS.load(Ordering::Relaxed) {
+        return Ok(f());
+    }
+
+    let result = panic::catch_unwind(f);
+
+    match result {
+        Ok(result) => Ok(result),
+        Err(err) => {
+            let info = err.downcast::<OopsInfo>().unwrap();
+
+            log::error!("Oops! {}", info.message);
+
+            let count = OOPS_COUNT.fetch_add(1, Ordering::Relaxed);
+            if count >= MAX_OOPS_COUNT {
+                // Too many oops. Panic the kernel.
+                //
+                // Note that for nested `catch_panics_as_oops` it still works as
+                // expected. The outer `catch_panics_as_oops` will catch the panic
+                // and found that the oops count is too high, then panic the kernel.
+                panic!("Too many oops. The kernel panics.");
+            }
+
+            Err(*info)
+        }
+    }
+}
+
+/// The maximum number of oops allowed before the kernel panics.
+///
+/// It is the same as Linux's default value.
+const MAX_OOPS_COUNT: usize = 10_000;
+
+static OOPS_COUNT: AtomicUsize = AtomicUsize::new(0);
+
+#[ostd::panic_handler]
+fn panic_handler(info: &core::panic::PanicInfo) -> ! {
+    let message = info.message().to_string();
+    let thread = current_thread!();
+
+    // Raise the panic and expect it to be caught.
+    // TODO: eliminate the need for heap allocation.
+    panic::begin_panic(Box::new(OopsInfo {
+        message: message.clone(),
+        thread,
+    }));
+
+    // Halt the system if the panic is not caught.
+    log::error!("Uncaught panic! {:#?}", message);
+
+    panic::print_stack_trace();
+    panic::abort();
+}

kernel/src/thread/kernel_thread.rs

@@ -45,8 +45,8 @@ impl KernelThreadExt for Thread {
 pub fn create_new_kernel_task(mut thread_options: ThreadOptions) -> Arc<Task> {
     let task_fn = thread_options.take_func();
     let thread_fn = move || {
-        task_fn();
+        let _ = crate::oops::catch_panics_as_oops(task_fn);
-        // Ensures the thread is exit
+        // Ensure that the thread exits.
         current_thread!().exit();
     };
 

kernel/src/thread/task.rs

@@ -89,7 +89,11 @@ pub fn create_new_user_task(user_space: Arc<UserSpace>, thread_ref: Arc<Thread>)
         debug!("exit user loop");
     }
 
-    TaskOptions::new(user_task_entry)
+    TaskOptions::new(|| {
+        // TODO: If a kernel "oops" is caught, we should kill the entire
+        // process rather than just ending the thread.
+        let _ = crate::oops::catch_panics_as_oops(user_task_entry);
+    })
     .data(thread_ref)
     .user_space(Some(user_space))
     .build()