PkmnLib_rs/src/script_implementations/wasm/script_resolver.rs

541 lines
21 KiB
Rust
Executable File

use std::fmt::{Debug, Formatter};
use std::mem::{align_of, forget, size_of};
use std::sync::{Arc, Weak};
use hashbrown::{HashMap, HashSet};
use parking_lot::lock_api::RwLockReadGuard;
use parking_lot::{RawRwLock, RwLock};
use unique_type_id::UniqueTypeId;
use wasmer::{
AsStoreMut, AsStoreRef, Cranelift, EngineBuilder, Extern, Features, Function, FunctionEnv, Imports, Instance,
Memory, Module, Store, StoreMut, StoreRef, TypedFunction, Value,
};
use crate::dynamic_data::{ItemScript, Script, ScriptOwnerData, ScriptResolver};
use crate::script_implementations::wasm::export_registry::register_webassembly_funcs;
use crate::script_implementations::wasm::extern_ref::ExternRef;
use crate::script_implementations::wasm::script::WebAssemblyScript;
use crate::script_implementations::wasm::script_function_cache::ScriptFunctionCache;
use crate::script_implementations::wasm::temp_wasm_allocator::{AllocatedObject, TempWasmAllocator};
use crate::script_implementations::wasm::WebAssemblyScriptCapabilities;
use crate::static_data::Item;
use crate::{PkmnResult, ScriptCategory, StringKey};
/// A WebAssembly script resolver implements the dynamic scripts functionality with WebAssembly.
pub struct WebAssemblyScriptResolver {
/// The global state storage of WASM.
_store: *mut Store,
/// The WASM modules we have loaded.
modules: Vec<Module>,
/// Our currently loaded WASM instances. Empty until finalize() is called, after which the loaded modules get turned
/// into actual instances.
instances: Vec<Instance>,
/// This is the WASM function to load a script.
load_script_fn: Option<TypedFunction<(u8, ExternRef<StringKey>), u32>>,
/// The data for use in the scripting function calls.
environment_data: Arc<WebAssemblyEnvironmentData>,
}
/// This struct allows us to index a hashmap with both a category and name of a script.
#[derive(Debug, Clone, Eq, PartialEq, Hash)]
struct ScriptCapabilitiesKey {
/// The category for the script we're looking for capabilities for.
category: ScriptCategory,
/// The name of the script we're looking for capabilities for.
script_key: StringKey,
}
impl WebAssemblyScriptResolver {
/// Instantiates a new WebAssemblyScriptResolver.
pub fn new() -> Box<WebAssemblyScriptResolver> {
let compiler = Cranelift::default();
let mut features = Features::new();
features.multi_value = true;
features.reference_types = true;
let engine = EngineBuilder::new(compiler).set_features(Some(features));
let mut store = Box::new(Store::new(engine));
let store_ptr: *mut Store = store.as_mut();
forget(store);
let environment = Arc::new(WebAssemblyEnvironmentData::new(store_ptr));
environment.self_arc.write().replace(Arc::downgrade(&environment));
let s = Self {
_store: store_ptr,
modules: Default::default(),
instances: Default::default(),
load_script_fn: None,
environment_data: environment,
};
Box::new(s)
}
/// Get an immutable reference to the current WASM Store.
fn store_ref(&self) -> StoreRef<'_> {
unsafe { self._store.as_ref().unwrap().as_store_ref() }
}
/// Get a mutable reference to the current WASM Store.
fn store_mut(&self) -> StoreMut<'_> {
unsafe { self._store.as_mut().unwrap().as_store_mut() }
}
/// Load a compiled WASM module.
pub fn load_wasm_from_bytes(&mut self, bytes: &[u8]) {
// FIXME: Error handling
let module = Module::new(&self.store_ref(), bytes).unwrap();
self.modules.push(module);
}
/// Initialise all the data we need.
pub fn finalize(&mut self) {
let mut imports = Imports::new();
//let mut exports = Exports::new();
let env = FunctionEnv::new(
&mut self.store_mut(),
WebAssemblyEnv::new(Arc::downgrade(&self.environment_data)),
);
register_webassembly_funcs(&mut imports, &mut self.store_mut(), &env);
//imports.register("env", exports);
for module in &self.modules {
for import in module.imports() {
if imports.get_export("env", import.name()).is_none() {
println!(
"\x1b[91mMissing import: \"{}\" with type: {:?} \x1b[0m",
import.name(),
import.ty()
);
}
}
let instance = Instance::new(&mut self.store_mut(), module, &imports).unwrap();
let exports = &instance.exports;
let init_fn = exports.get_extern("_init");
if let Some(Extern::Function(init_fn)) = init_fn {
init_fn.call(&mut self.store_mut(), &[]).unwrap();
}
let mut exported_functions = self.environment_data.exported_functions.write();
for export in exports.iter() {
match export.1 {
Extern::Function(f) => {
exported_functions.insert(export.0.as_str().into(), f.clone());
}
Extern::Memory(m) => {
let _ = self.environment_data.memory.write().insert(m.clone());
}
_ => {}
}
}
if let Some(m) = &self.environment_data.memory.read().as_ref() {
m.grow(&mut self.store_mut(), 32).unwrap();
}
if let Some(f) = exported_functions.get::<StringKey>(&"load_script".into()) {
self.load_script_fn = Some(f.typed(&self.store_ref()).unwrap())
}
if let Some(f) = exported_functions.get::<StringKey>(&"allocate_mem".into()) {
let _ = self
.environment_data
.allocate_mem_fn
.write()
.insert(f.typed(&self.store_ref()).unwrap());
let temp_memory_slab = self.environment_data.allocate_mem(128, 1);
let _ = self
.environment_data
.temp_allocator
.write()
.insert(TempWasmAllocator::new(temp_memory_slab.0, temp_memory_slab.1));
}
self.instances.push(instance);
}
}
/// Gets the data passed to every function as environment data.
pub fn environment_data(&self) -> &Arc<WebAssemblyEnvironmentData> {
&self.environment_data
}
}
impl ScriptResolver for WebAssemblyScriptResolver {
fn load_script(
&self,
owner: ScriptOwnerData,
category: ScriptCategory,
script_key: &StringKey,
) -> PkmnResult<Option<Arc<dyn Script>>> {
let script = self
.load_script_fn
.as_ref()
.unwrap()
.call(
&mut self.store_mut(),
category as u8,
ExternRef::new_with_resolver(self, script_key),
)
.unwrap();
self.environment_data.setup_script(script, category, script_key, owner)
}
fn load_item_script(&self, _key: &Item) -> PkmnResult<Option<Arc<dyn ItemScript>>> {
todo!()
}
}
impl Debug for WebAssemblyScriptResolver {
fn fmt(&self, f: &mut Formatter<'_>) -> std::fmt::Result {
f.write_str("WebAssemblyScriptResolver")
}
}
impl Drop for WebAssemblyScriptResolver {
fn drop(&mut self) {
unsafe {
drop(Box::from_raw(self._store));
}
}
}
/// This data is what is passed to every function that requires access to the global runtime context.
pub struct WebAssemblyEnvironmentData {
/// We currently have a hacky implementation of extern refs while we're waiting for ExternRef support to hit the
/// wasm32-unknown-unknown target of Rust. As we don't want to pass raw memory pointers to WASM for security reasons,
/// we instead keep track of all the data we've sent to WASM, and pass the ID of that data to WASM. This allows us
/// to only operate on data we know WASM owns. We currently store this data in this continuous Vec, and give the index
/// of the data as the ID.
extern_ref_pointers: RwLock<Vec<*const u8>>,
/// To make sure we send the same identifier to WASM when we send the same piece of data multiple times, we have a
/// backwards lookup on extern_ref_pointers. This allows us to get the index for a given piece of data.
extern_ref_pointers_lookup: RwLock<HashMap<ExternRefLookupKey, u32>>,
/// As an added security measure on our extern refs, we keep track of the types of the extern ref data we've sent.
/// This prevents illegal arbitrary memory operations, where we expect type X, but the actual type is Y, which would
/// allow for modifying memory we might not want to. If we get a type mismatch, we will panic, preventing this.
extern_ref_type_lookup: RwLock<HashSet<ExternRefLookupKey>>,
/// Additional security for data slices passed to WASM.
extern_vec_ref_lookup: RwLock<HashMap<u32, Vec<u32>>>,
/// The memory inside of the WASM container.
memory: RwLock<Option<Memory>>,
/// This is a map of all the functions that WASM gives us.
exported_functions: RwLock<HashMap<StringKey, Function>>,
/// A cache of all script functions for faster calls.
script_function_cache: ScriptFunctionCache,
/// This is the WASM function to allocate memory inside the WASM container.
allocate_mem_fn: RwLock<Option<TypedFunction<(u32, u32), u32>>>,
/// An allocator for quick short lifetime allocations within WASM.
temp_allocator: RwLock<Option<TempWasmAllocator>>,
/// The WASM store.
store: *mut Store,
/// Script capabilities tell us which functions are implemented on a given script. This allows us to skip unneeded
/// WASM calls.
script_capabilities: RwLock<HashMap<ScriptCapabilitiesKey, Arc<HashSet<WebAssemblyScriptCapabilities>>>>,
/// A weak reference to ourselves.
self_arc: RwLock<Option<Weak<Self>>>,
/// A lookup from WASM memory pointer to their actual script wrappers.
loaded_scripts: RwLock<HashMap<u32, Weak<WebAssemblyScript>>>,
}
/// A quick lookup so we can find the extern ref of the value.
#[derive(Clone, Eq, PartialEq, Hash)]
struct ExternRefLookupKey {
/// The raw pointer to the data
pub ptr: *const u8,
/// Whether or not the reference is a Vec
pub is_vec: bool,
/// The unique identifier of the type.
pub t: u64,
}
impl WebAssemblyEnvironmentData {
/// Instantiates new Environment data with a given store.
pub(crate) fn new(store: *mut Store) -> Self {
Self {
extern_ref_pointers: Default::default(),
extern_ref_pointers_lookup: Default::default(),
extern_ref_type_lookup: Default::default(),
extern_vec_ref_lookup: Default::default(),
memory: Default::default(),
exported_functions: Default::default(),
script_function_cache: Default::default(),
allocate_mem_fn: Default::default(),
temp_allocator: Default::default(),
store,
script_capabilities: Default::default(),
self_arc: Default::default(),
loaded_scripts: Default::default(),
}
}
/// This returns the memory of the WASM container.
pub fn memory(&self) -> *mut u8 {
self.memory.read().as_ref().unwrap().view(&self.store_ref()).data_ptr()
}
/// Return a pointer to something inside the WASM memory.
pub fn get_raw_pointer<T>(&self, offset: u32) -> *mut T {
unsafe {
self.memory
.read()
.as_ref()
.unwrap()
.view(&self.store_ref())
.data_ptr()
.offset(offset as isize) as *mut T
}
}
/// This returns the functions exported from WASM.
pub fn exported_functions(&self) -> RwLockReadGuard<'_, RawRwLock, HashMap<StringKey, Function>> {
self.exported_functions.read()
}
///
pub(super) fn script_function_cache(&self) -> &ScriptFunctionCache {
&self.script_function_cache
}
/// Allocates memory inside the WASM container with a given size and alignment. This memory is
/// owned by WASM, and is how we can pass memory references that the host allocated to WASM.
/// The return is a tuple containing both the actual pointer to the memory (usable by the host),
/// and the WASM offset to the memory (usable by the client).
pub fn allocate_mem(&self, size: u32, align: u32) -> (*mut u8, u32) {
let wasm_ptr = self
.allocate_mem_fn
.read()
.as_ref()
.unwrap()
.call(&mut self.store_mut(), size, align)
.unwrap();
unsafe { (self.memory().offset(wasm_ptr as isize), wasm_ptr) }
}
/// Allocates memory inside the WASM container with a given size and alignment. This memory is
/// owned by WASM, and is how we can pass memory references that the host allocated to WASM.
/// The return is a tuple containing both the actual pointer to the memory (usable by the host),
/// and the WASM offset to the memory (usable by the client).
pub fn allocate_mem_typed<T>(&self) -> (*mut u8, u32) {
let wasm_ptr = self
.allocate_mem_fn
.read()
.as_ref()
.unwrap()
.call(&mut self.store_mut(), size_of::<T>() as u32, align_of::<T>() as u32)
.unwrap();
unsafe { (self.memory().offset(wasm_ptr as isize), wasm_ptr) }
}
/// Allocate a piece of memory inside WASM with a very short lifespan. This is mainly used for
/// rapid allocation of script function parameters, where WASM needs to write to a specific
/// pointer.
pub(super) fn allocate_temp<T>(&self, value: T) -> AllocatedObject<T> {
self.temp_allocator.read().as_ref().unwrap().alloc::<T>(value)
}
/// Get a numeric value from any given value. This is not a true Extern Ref from WASM, as this
/// is not supported by our current WASM platform (Rust). Instead, this is simply a way to not
/// have to send arbitrary pointer values back and forth with WASM. Only values WASM can actually
/// access can be touched through this, and we ensure the value is the correct type. In the future,
/// when extern refs get actually properly implemented at compile time we might want to get rid
/// of this code.
#[inline(always)]
pub fn get_extern_ref_index<T: UniqueTypeId<u64> + ?Sized>(&self, value: &T) -> u32 {
self.get_extern_ref_from_ptr(value as *const T as *const u8, T::id().0, false)
}
/// Get a numeric value from any given value. This is not a true Extern Ref from WASM, as this
/// is not supported by our current WASM platform (Rust). Instead, this is simply a way to not
/// have to send arbitrary pointer values back and forth with WASM. Only values WASM can actually
/// access can be touched through this, and we ensure the value is the correct type. In the future,
/// when extern refs get actually properly implemented at compile time we might want to get rid
/// of this code.
pub fn get_extern_vec_ref_index<T: UniqueTypeId<u64>>(&self, value: &[T]) -> u32 {
let mut vec = Vec::with_capacity(value.len());
for v in value {
vec.push(self.get_extern_ref_index(v));
}
let p = self.get_extern_ref_from_ptr(value as *const [T] as *const u8, T::id().0, true);
self.extern_vec_ref_lookup.write().insert(p, vec);
p
}
/// Get an extern ref belonging to a vector we have passed to WASM.
pub fn get_extern_vec_ref_extern_ref(&self, extern_vec_ref: u32, index: u32) -> u32 {
let r = self.extern_vec_ref_lookup.read();
let v = r.get(&extern_vec_ref).unwrap();
v[index as usize]
}
/// Gets the extern ref index belonging to a specific pointer. If none exists, this will create
/// a new one.
#[inline(always)]
fn get_extern_ref_from_ptr(&self, ptr: *const u8, type_id: u64, is_vec: bool) -> u32 {
if let Some(v) = self.extern_ref_pointers_lookup.read().get(&ExternRefLookupKey {
ptr,
is_vec,
t: type_id,
}) {
return *v as u32;
}
let index = {
let mut extern_ref_guard = self.extern_ref_pointers.write();
extern_ref_guard.push(ptr);
extern_ref_guard.len() as u32
};
self.extern_ref_pointers_lookup.write().insert(
ExternRefLookupKey {
ptr,
is_vec,
t: type_id,
},
index,
);
self.extern_ref_type_lookup.write().insert(ExternRefLookupKey {
ptr,
is_vec,
t: type_id,
});
index
}
/// Gets a value from the extern ref lookup. This turns an earlier registered index back into
/// its proper value, validates its type, and returns the value.
pub fn get_extern_ref_value<T: UniqueTypeId<u64>>(&self, index: u32) -> &T {
let read_guard = self.extern_ref_pointers.read();
let ptr = read_guard.get((index - 1) as usize).unwrap();
if self
.extern_ref_type_lookup
.read()
.get(&ExternRefLookupKey {
ptr: *ptr,
is_vec: false,
t: T::id().0,
})
.is_none()
{
panic!(
"Extern ref was accessed with wrong type. Requested type {}, but this was not the type the extern ref was stored with.",
std::any::type_name::<T>()
);
}
unsafe { (*ptr as *const T).as_ref().unwrap() }
}
/// The WASM store.
pub fn store_ref(&self) -> StoreRef<'_> {
unsafe { self.store.as_ref().unwrap().as_store_ref() }
}
/// The mutable WASM store.
pub fn store_mut(&self) -> StoreMut<'_> {
unsafe { self.store.as_mut().unwrap().as_store_mut() }
}
/// Find a loaded script based on the pointer in WASM memory.
pub(crate) fn get_loaded_script(&self, wasm_ptr: u32) -> Option<Arc<WebAssemblyScript>> {
if let Some(script) = self.loaded_scripts.read().get(&wasm_ptr) {
script.upgrade()
} else {
None
}
}
/// Wrap a script pointer in WASM memory into a host managed script.
pub fn setup_script(
&self,
script_ptr: u32,
category: ScriptCategory,
script_key: &StringKey,
owner: ScriptOwnerData,
) -> PkmnResult<Option<Arc<dyn Script>>> {
if script_ptr == 0 {
return Ok(None);
}
let key = ScriptCapabilitiesKey {
category,
script_key: script_key.clone(),
};
if !self.script_capabilities.read().contains_key(&key) {
let mut capabilities = HashSet::new();
unsafe {
if let Some(get_cap) = self
.exported_functions
.read()
.get::<StringKey>(&"get_script_capabilities".into())
{
let res = get_cap
.call(&mut self.store_mut(), &[Value::I32(script_ptr as i32)])
.unwrap();
let ptr =
(self.memory() as *const WebAssemblyScriptCapabilities).offset(res[0].i32().unwrap() as isize);
let length = res[1].i32().unwrap() as usize;
for i in 0..length {
capabilities.insert(*ptr.add(i));
}
}
}
self.script_capabilities
.write()
.insert(key.clone(), Arc::new(capabilities));
}
let read_guard = self.script_capabilities.read();
let capabilities = read_guard.get(&key).unwrap();
let script = Arc::new(WebAssemblyScript::new(
owner,
script_ptr,
capabilities.clone(),
self.self_arc.read().as_ref().unwrap().upgrade().unwrap(),
script_key.clone(),
));
self.loaded_scripts.write().insert(script_ptr, Arc::downgrade(&script));
Ok(Some(script))
}
}
/// The runtime environment for script execution. This is passed to most of the host functions being called.
#[derive(Clone)]
pub(crate) struct WebAssemblyEnv {
/// A pointer to the WebAssemblyScriptResolver belonging to the current script environment.
data: Weak<WebAssemblyEnvironmentData>,
}
impl WebAssemblyEnv {
/// Instantiates a new Environment with the requested data.
pub fn new(data: Weak<WebAssemblyEnvironmentData>) -> Self {
Self { data }
}
/// Get the actual data belonging to the current context.
pub fn data(&self) -> Arc<WebAssemblyEnvironmentData> {
self.data.upgrade().unwrap()
}
}
unsafe impl Sync for WebAssemblyEnv {}
unsafe impl Send for WebAssemblyEnv {}
unsafe impl Sync for WebAssemblyEnvironmentData {}
unsafe impl Send for WebAssemblyEnvironmentData {}