Add support for async call hooks (#3876)

* Instead of simply panicking, return an error when we attempt to resume on a dying fiber.

This situation should never occur in the existing code base, but can be
triggered if support for running outside async code in a call hook.

* Shift `async_cx()` to return an `Option`, reflecting if the fiber is dying.

This should never happen in the existing code base, but is a nice
forward-looking guard. The current implementations simply lift the
trap that would eventually be produced by such an operation into
a `Trap` (or similar) at the invocation of `async_cx()`.

* Add support for using `async` call hooks.

This retains the ability to do non-async hooks. Hooks end up being
implemented as an async trait with a handler call, to get around some
issues passing around async closures. This change requires some of
the prior changes to handle picking up blocked tasks during fiber
shutdown, to avoid some panics during timeouts and other such events.

* More fully specify a doc link, to avoid a doc-building error.

* Revert the use of catchable traps on cancellation of a fiber; turn them into expect()/unwrap().

The justification for this revert is that (a) these events shouldn't
happen, and (b) they wouldn't be catchable by wasm anyways.

* Replace a duplicated check in `async` hook evaluation with a single check.

This also moves the checks inside of their respective Async variants,
meaning that if you're using an async-enabled version of wasmtime but
using the synchronous versions of the callbacks, you won't pay any
penalty for validating the async context.

* Use `match &mut ...` insead of `ref mut`.

* Add some documentation on why/when `async_cx` can return None.

* Add two simple test cases for async call hooks.

* Fix async_cx() to check both the box and the value for current_poll_cx.

In the prior version, we only checked that the box had not been cleared,
but had not ensured that there was an actual context for us to use. This
updates the check to validate both, returning None if the inner context
is missing. This allows us to skip a validation check inside `block_on`,
since all callers will have run through the `async_cx` check prior to
arrival.

* Tweak the timeout test to address PR suggestions.

* Add a test about dropping async hooks while suspended

Should help exercise that the check for `None` is properly handled in a
few more locations.

Co-authored-by: Alex Crichton <alex@alexcrichton.com>

crates/wasmtime/src/func.rs

@@ -272,8 +272,9 @@ macro_rules! generate_wrap_async_func {
         {
             assert!(store.as_context().async_support(), concat!("cannot use `wrap", $num, "_async` without enabling async support on the config"));
             Func::wrap(store, move |mut caller: Caller<'_, T>, $($args: $args),*| {
-                let async_cx = caller.store.as_context_mut().0.async_cx();
+                let async_cx = caller.store.as_context_mut().0.async_cx().expect("Attempt to start async function on dying fiber");
                 let mut future = Pin::from(func(caller, $($args),*));
+
                 match unsafe { async_cx.block_on(future.as_mut()) } {
                     Ok(ret) => ret.into_fallible(),
                     Err(e) => R::fallible_from_trap(e),
@@ -439,7 +440,12 @@ impl Func {
             "cannot use `new_async` without enabling async support in the config"
         );
         Func::new(store, ty, move |mut caller, params, results| {
-            let async_cx = caller.store.as_context_mut().0.async_cx();
+            let async_cx = caller
+                .store
+                .as_context_mut()
+                .0
+                .async_cx()
+                .expect("Attempt to spawn new action on dying fiber");
             let mut future = Pin::from(func(caller, params, results));
             match unsafe { async_cx.block_on(future.as_mut()) } {
                 Ok(Ok(())) => Ok(()),

crates/wasmtime/src/lib.rs

@@ -423,6 +423,8 @@ pub use crate::linker::*;
 pub use crate::memory::*;
 pub use crate::module::{FrameInfo, FrameSymbol, Module};
 pub use crate::r#ref::ExternRef;
+#[cfg(feature = "async")]
+pub use crate::store::CallHookHandler;
 pub use crate::store::{AsContext, AsContextMut, CallHook, Store, StoreContext, StoreContextMut};
 pub use crate::trap::*;
 pub use crate::types::*;

crates/wasmtime/src/linker.rs

@@ -146,7 +146,7 @@ macro_rules! generate_wrap_async_func {
                 ),
             );
             self.func_wrap(module, name, move |mut caller: Caller<'_, T>, $($args: $args),*| {
-                let async_cx = caller.store.as_context_mut().0.async_cx();
+                let async_cx = caller.store.as_context_mut().0.async_cx().expect("Attempt to start async function on dying fiber");
                 let mut future = Pin::from(func(caller, $($args),*));
                 match unsafe { async_cx.block_on(future.as_mut()) } {
                     Ok(ret) => ret.into_fallible(),
@@ -360,7 +360,12 @@ impl<T> Linker<T> {
             "cannot use `func_new_async` without enabling async support in the config"
         );
         self.func_new(module, name, ty, move |mut caller, params, results| {
-            let async_cx = caller.store.as_context_mut().0.async_cx();
+            let async_cx = caller
+                .store
+                .as_context_mut()
+                .0
+                .async_cx()
+                .expect("Attempt to spawn new function on dying fiber");
             let mut future = Pin::from(func(caller, params, results));
             match unsafe { async_cx.block_on(future.as_mut()) } {
                 Ok(Ok(())) => Ok(()),

crates/wasmtime/src/store.rs

@@ -200,7 +200,7 @@ pub struct StoreInner<T> {
     inner: StoreOpaque,
 
     limiter: Option<ResourceLimiterInner<T>>,
-    call_hook: Option<Box<dyn FnMut(&mut T, CallHook) -> Result<(), crate::Trap> + Send + Sync>>,
+    call_hook: Option<CallHookInner<T>>,
     // for comments about `ManuallyDrop`, see `Store::into_data`
     data: ManuallyDrop<T>,
 }
@@ -211,6 +211,21 @@ enum ResourceLimiterInner<T> {
     Async(Box<dyn FnMut(&mut T) -> &mut (dyn crate::ResourceLimiterAsync) + Send + Sync>),
 }
 
+/// An object that can take callbacks when the runtime enters or exits hostcalls.
+#[cfg(feature = "async")]
+#[async_trait::async_trait]
+pub trait CallHookHandler<T>: Send {
+    /// A callback to run when wasmtime is about to enter a host call, or when about to
+    /// exit the hostcall.
+    async fn handle_call_event(&self, t: &mut T, ch: CallHook) -> Result<(), crate::Trap>;
+}
+
+enum CallHookInner<T> {
+    Sync(Box<dyn FnMut(&mut T, CallHook) -> Result<(), crate::Trap> + Send + Sync>),
+    #[cfg(feature = "async")]
+    Async(Box<dyn CallHookHandler<T> + Send + Sync>),
+}
+
 // Forward methods on `StoreOpaque` to also being on `StoreInner<T>`
 impl<T> Deref for StoreInner<T> {
     type Target = StoreOpaque;
@@ -603,6 +618,27 @@ impl<T> Store<T> {
         inner.limiter = Some(ResourceLimiterInner::Async(Box::new(limiter)));
     }
 
+    #[cfg_attr(nightlydoc, doc(cfg(feature = "async")))]
+    /// Configures an async function that runs on calls and returns between
+    /// WebAssembly and host code. For the non-async equivalent of this method,
+    /// see [`Store::call_hook`].
+    ///
+    /// The function is passed a [`CallHook`] argument, which indicates which
+    /// state transition the VM is making.
+    ///
+    /// This function's future may return a [`Trap`]. If a trap is returned
+    /// when an import was called, it is immediately raised as-if the host
+    /// import had returned the trap. If a trap is returned after wasm returns
+    /// to the host then the wasm function's result is ignored and this trap is
+    /// returned instead.
+    ///
+    /// After this function returns a trap, it may be called for subsequent
+    /// returns to host or wasm code as the trap propagates to the root call.
+    #[cfg(feature = "async")]
+    pub fn call_hook_async(&mut self, hook: impl CallHookHandler<T> + Send + Sync + 'static) {
+        self.inner.call_hook = Some(CallHookInner::Async(Box::new(hook)));
+    }
+
     /// Configure a function that runs on calls and returns between WebAssembly
     /// and host code.
     ///
@@ -616,12 +652,12 @@ impl<T> Store<T> {
     /// instead.
     ///
     /// After this function returns a trap, it may be called for subsequent returns
-    /// to host or wasm code as the trap propogates to the root call.
+    /// to host or wasm code as the trap propagates to the root call.
     pub fn call_hook(
         &mut self,
         hook: impl FnMut(&mut T, CallHook) -> Result<(), Trap> + Send + Sync + 'static,
     ) {
-        self.inner.call_hook = Some(Box::new(hook));
+        self.inner.call_hook = Some(CallHookInner::Sync(Box::new(hook)));
     }
 
     /// Returns the [`Engine`] that this store is associated with.
@@ -956,10 +992,19 @@ impl<T> StoreInner<T> {
     }
 
     pub fn call_hook(&mut self, s: CallHook) -> Result<(), Trap> {
-        if let Some(hook) = &mut self.call_hook {
-            hook(&mut self.data, s)
-        } else {
-            Ok(())
+        match &mut self.call_hook {
+            Some(CallHookInner::Sync(hook)) => hook(&mut self.data, s),
+
+            #[cfg(feature = "async")]
+            Some(CallHookInner::Async(handler)) => unsafe {
+                Ok(self
+                    .inner
+                    .async_cx()
+                    .ok_or(Trap::new("couldn't grab async_cx for call hook"))?
+                    .block_on(handler.handle_call_event(&mut self.data, s).as_mut())??)
+            },
+
+            None => Ok(()),
         }
     }
 }
@@ -1143,14 +1188,29 @@ impl StoreOpaque {
         panic!("trampoline missing")
     }
 
+    /// Yields the async context, assuming that we are executing on a fiber and
+    /// that fiber is not in the process of dying. This function will return
+    /// None in the latter case (the fiber is dying), and panic if
+    /// `async_support()` is false.
     #[cfg(feature = "async")]
     #[inline]
-    pub fn async_cx(&self) -> AsyncCx {
+    pub fn async_cx(&self) -> Option<AsyncCx> {
         debug_assert!(self.async_support());
-        AsyncCx {
-            current_suspend: self.async_state.current_suspend.get(),
-            current_poll_cx: self.async_state.current_poll_cx.get(),
+
+        let poll_cx_box_ptr = self.async_state.current_poll_cx.get();
+        if poll_cx_box_ptr.is_null() {
+            return None;
         }
+
+        let poll_cx_inner_ptr = unsafe { *poll_cx_box_ptr };
+        if poll_cx_inner_ptr.is_null() {
+            return None;
+        }
+
+        Some(AsyncCx {
+            current_suspend: self.async_state.current_suspend.get(),
+            current_poll_cx: poll_cx_box_ptr,
+        })
     }
 
     pub fn fuel_consumed(&self) -> Option<u64> {
@@ -1214,7 +1274,11 @@ impl StoreOpaque {
         // to clean up this fiber. Do so by raising a trap which will
         // abort all wasm and get caught on the other side to clean
         // things up.
-        unsafe { self.async_cx().block_on(Pin::new_unchecked(&mut future)) }
+        unsafe {
+            self.async_cx()
+                .expect("attempted to pull async context during shutdown")
+                .block_on(Pin::new_unchecked(&mut future))
+        }
     }
 
     fn add_fuel(&mut self, fuel: u64) -> Result<()> {
@@ -1649,22 +1713,15 @@ unsafe impl<T> wasmtime_runtime::Store for StoreInner<T> {
         desired: usize,
         maximum: Option<usize>,
     ) -> Result<bool, anyhow::Error> {
-        // Need to borrow async_cx before the mut borrow of the limiter.
-        // self.async_cx() panicks when used with a non-async store, so
-        // wrap this in an option.
-        #[cfg(feature = "async")]
-        let async_cx = if self.async_support() {
-            Some(self.async_cx())
-        } else {
-            None
-        };
         match self.limiter {
             Some(ResourceLimiterInner::Sync(ref mut limiter)) => {
                 Ok(limiter(&mut self.data).memory_growing(current, desired, maximum))
             }
             #[cfg(feature = "async")]
             Some(ResourceLimiterInner::Async(ref mut limiter)) => unsafe {
-                Ok(async_cx
+                Ok(self
+                    .inner
+                    .async_cx()
                     .expect("ResourceLimiterAsync requires async Store")
                     .block_on(
                         limiter(&mut self.data)
@@ -1700,7 +1757,7 @@ unsafe impl<T> wasmtime_runtime::Store for StoreInner<T> {
         // wrap this in an option.
         #[cfg(feature = "async")]
         let async_cx = if self.async_support() {
-            Some(self.async_cx())
+            Some(self.async_cx().unwrap())
         } else {
             None
         };