wasi.rs

  1use std::{
  2    collections::HashMap, fs::File, future::Future, marker::PhantomData, path::Path, pin::Pin,
  3};
  4
  5use anyhow::{anyhow, Error};
  6use serde::{de::DeserializeOwned, Serialize};
  7
  8use wasi_common::{dir, file};
  9use wasmtime::{
 10    AsContext, AsContextMut, Caller, Config, Engine, Extern, Instance, Linker, Module, Store,
 11    StoreContext, StoreContextMut, Trap, TypedFunc,
 12};
 13use wasmtime::{IntoFunc, Memory};
 14use wasmtime_wasi::{Dir, WasiCtx, WasiCtxBuilder};
 15
 16pub struct WasiResource(u32);
 17
 18pub struct WasiFn<A: Serialize, R: DeserializeOwned> {
 19    function: TypedFunc<(u32, u32), u32>,
 20    _function_type: PhantomData<fn(A) -> R>,
 21}
 22
 23impl<A: Serialize, R: DeserializeOwned> Copy for WasiFn<A, R> {}
 24
 25impl<A: Serialize, R: DeserializeOwned> Clone for WasiFn<A, R> {
 26    fn clone(&self) -> Self {
 27        Self {
 28            function: self.function,
 29            _function_type: PhantomData,
 30        }
 31    }
 32}
 33
 34// impl<A: Serialize, R: DeserializeOwned> WasiFn<A, R> {
 35//     #[inline(always)]
 36//     pub async fn call(&self, runtime: &mut Wasi, arg: A) -> Result<R, Error> {
 37//         runtime.call(self, arg).await
 38//     }
 39// }
 40
 41// type signature derived from:
 42// https://docs.rs/wasmtime/latest/wasmtime/struct.Linker.html#method.func_wrap2_async
 43// macro_rules! dynHostFunction {
 44//     () => {
 45//         Box<
 46//             dyn for<'a> Fn(Caller<'a, WasiCtx>, u32, u32)
 47//                 -> Box<dyn Future<Output = u32> + Send + 'a>
 48//                     + Send
 49//                     + Sync
 50//                     + 'static
 51//         >
 52//     };
 53// }
 54
 55// macro_rules! implHostFunction {
 56//     () => {
 57//         impl for<'a> Fn(Caller<'a, WasiCtx>, u32, u32)
 58//             -> Box<dyn Future<Output = u32> + Send + 'a>
 59//                 + Send
 60//                 + Sync
 61//                 + 'static
 62//     };
 63// }
 64
 65pub struct WasiPluginBuilder {
 66    wasi_ctx: WasiCtx,
 67    engine: Engine,
 68    linker: Linker<WasiCtxAlloc>,
 69}
 70
 71impl WasiPluginBuilder {
 72    pub fn new(wasi_ctx: WasiCtx) -> Result<Self, Error> {
 73        let mut config = Config::default();
 74        config.async_support(true);
 75        let engine = Engine::new(&config)?;
 76        let mut linker = Linker::new(&engine);
 77
 78        Ok(WasiPluginBuilder {
 79            // host_functions: HashMap::new(),
 80            wasi_ctx,
 81            engine,
 82            linker,
 83        })
 84    }
 85
 86    pub fn new_with_default_ctx() -> Result<Self, Error> {
 87        let wasi_ctx = WasiCtxBuilder::new()
 88            .inherit_stdin()
 89            .inherit_stderr()
 90            .build();
 91        Self::new(wasi_ctx)
 92    }
 93
 94    pub fn host_function<A: DeserializeOwned + Send, R: Serialize + Send + Sync + Clone>(
 95        mut self,
 96        name: &str,
 97        function: impl Fn(A) -> R + Send + Sync + 'static,
 98    ) -> Result<Self, Error> {
 99        self.linker.func_wrap2_async(
100            "env",
101            name,
102            move |mut caller: Caller<'_, WasiCtxAlloc>, ptr: u32, len: u32| {
103                // TODO: use try block once avaliable
104                let result: Result<(Memory, Vec<u8>), Trap> = (|| {
105                    // grab a handle to the memory
106                    let mut plugin_memory = match caller.get_export("memory") {
107                        Some(Extern::Memory(mem)) => mem,
108                        _ => return Err(Trap::new("Could not grab slice of plugin memory"))?,
109                    };
110
111                    // get the args passed from Guest
112                    let args =
113                        Wasi::deserialize_from_buffer(&mut plugin_memory, &caller, ptr, len)?;
114
115                    // Call the Host-side function
116                    let result: R = function(args);
117
118                    // Serialize the result back to guest
119                    let result = Wasi::serialize(result).map_err(|_| {
120                        Trap::new("Could not serialize value returned from function")
121                    })?;
122                    Ok((plugin_memory, result))
123                })();
124
125                Box::new(async move {
126                    let (mut plugin_memory, result) = result?;
127
128                    // todo!();
129                    let (ptr, len) = Wasi::serialize_to_buffer(
130                        caller.data().alloc_buffer(),
131                        &mut plugin_memory,
132                        &mut caller,
133                        result,
134                    )
135                    .await?;
136
137                    Ok(7u32)
138                })
139            },
140        )?;
141        Ok(self)
142    }
143
144    pub async fn init<T: AsRef<[u8]>>(self, module: T) -> Result<Wasi, Error> {
145        Wasi::init(module.as_ref().to_vec(), self).await
146    }
147}
148
149// // TODO: remove
150// /// Represents a to-be-initialized plugin.
151// /// Please use [`WasiPluginBuilder`], don't use this directly.
152// pub struct WasiPlugin {
153//     pub module: Vec<u8>,
154//     pub wasi_ctx: WasiCtx,
155//     pub host_functions:
156//         HashMap<String, Box<dyn Fn(&str, &mut Linker<WasiCtx>) -> Result<(), Error>>>,
157// }
158
159#[derive(Copy, Clone)]
160struct WasiAlloc {
161    alloc_buffer: TypedFunc<u32, u32>,
162    free_buffer: TypedFunc<u32, u32>,
163}
164
165struct WasiCtxAlloc {
166    wasi_ctx: WasiCtx,
167    alloc: Option<WasiAlloc>,
168}
169
170impl WasiCtxAlloc {
171    fn alloc_buffer(&self) -> TypedFunc<u32, u32> {
172        self.alloc
173            .expect("allocator has been not initialized, cannot allocate buffer!")
174            .alloc_buffer
175    }
176
177    fn free_buffer(&self) -> TypedFunc<u32, u32> {
178        self.alloc
179            .expect("allocator has been not initialized, cannot free buffer!")
180            .free_buffer
181    }
182
183    fn init_alloc(&mut self, alloc: WasiAlloc) {
184        self.alloc = Some(alloc)
185    }
186}
187
188pub struct Wasi {
189    engine: Engine,
190    module: Module,
191    store: Store<WasiCtxAlloc>,
192    instance: Instance,
193}
194
195impl Wasi {
196    pub fn dump_memory(data: &[u8]) {
197        for (i, byte) in data.iter().enumerate() {
198            if i % 32 == 0 {
199                println!();
200            }
201            if i % 4 == 0 {
202                print!("|");
203            }
204            if *byte == 0 {
205                print!("__")
206            } else {
207                print!("{:02x}", byte);
208            }
209        }
210        println!();
211    }
212}
213
214impl Wasi {
215    async fn init(module: Vec<u8>, plugin: WasiPluginBuilder) -> Result<Self, Error> {
216        // initialize the WebAssembly System Interface context
217        let engine = plugin.engine;
218        let mut linker = plugin.linker;
219        wasmtime_wasi::add_to_linker(&mut linker, |s| &mut s.wasi_ctx)?;
220
221        // create a store, note that we can't initialize the allocator,
222        // because we can't grab the functions until initialized.
223        let mut store: Store<WasiCtxAlloc> = Store::new(
224            &engine,
225            WasiCtxAlloc {
226                wasi_ctx: plugin.wasi_ctx,
227                alloc: None,
228            },
229        );
230        let module = Module::new(&engine, module)?;
231
232        // load the provided module into the asynchronous runtime
233        linker.module_async(&mut store, "", &module).await?;
234        let instance = linker.instantiate_async(&mut store, &module).await?;
235
236        // now that the module is initialized,
237        // we can initialize the store's allocator
238        let alloc_buffer = instance.get_typed_func(&mut store, "__alloc_buffer")?;
239        let free_buffer = instance.get_typed_func(&mut store, "__free_buffer")?;
240        store.data_mut().init_alloc(WasiAlloc {
241            alloc_buffer,
242            free_buffer,
243        });
244
245        Ok(Wasi {
246            engine,
247            module,
248            store,
249            instance,
250        })
251    }
252
253    /// Attaches a file or directory the the given system path to the runtime.
254    /// Note that the resource must be freed by calling `remove_resource` afterwards.
255    pub fn attach_path<T: AsRef<Path>>(&mut self, path: T) -> Result<WasiResource, Error> {
256        // grab the WASI context
257        let ctx = self.store.data_mut();
258
259        // open the file we want, and convert it into the right type
260        // this is a footgun and a half
261        let file = File::open(&path).unwrap();
262        let dir = Dir::from_std_file(file);
263        let dir = Box::new(wasmtime_wasi::dir::Dir::from_cap_std(dir));
264
265        // grab an empty file descriptor, specify capabilities
266        let fd = ctx.wasi_ctx.table().push(Box::new(()))?;
267        let caps = dir::DirCaps::all();
268        let file_caps = file::FileCaps::all();
269
270        // insert the directory at the given fd,
271        // return a handle to the resource
272        ctx.wasi_ctx
273            .insert_dir(fd, dir, caps, file_caps, path.as_ref().to_path_buf());
274        Ok(WasiResource(fd))
275    }
276
277    /// Returns `true` if the resource existed and was removed.
278    pub fn remove_resource(&mut self, resource: WasiResource) -> Result<(), Error> {
279        self.store
280            .data_mut()
281            .wasi_ctx
282            .table()
283            .delete(resource.0)
284            .ok_or_else(|| anyhow!("Resource did not exist, but a valid handle was passed in"))?;
285        Ok(())
286    }
287
288    // pub fn with_resource<T>(
289    //     &mut self,
290    //     resource: WasiResource,
291    //     callback: fn(&mut Self) -> Result<T, Error>,
292    // ) -> Result<T, Error> {
293    //     let result = callback(self);
294    //     self.remove_resource(resource)?;
295    //     return result;
296    // }
297
298    // So this call function is kinda a dance, I figured it'd be a good idea to document it.
299    // the high level is we take a serde type, serialize it to a byte array,
300    // (we're doing this using bincode for now)
301    // then toss that byte array into webassembly.
302    // webassembly grabs that byte array, does some magic,
303    // and serializes the result into yet another byte array.
304    // we then grab *that* result byte array and deserialize it into a result.
305    //
306    // phew...
307    //
308    // now the problem is, webassambly doesn't support buffers.
309    // only really like i32s, that's it (yeah, it's sad. Not even unsigned!)
310    // (ok, I'm exaggerating a bit).
311    //
312    // the Wasm function that this calls must have a very specific signature:
313    //
314    // fn(pointer to byte array: i32, length of byte array: i32)
315    //     -> pointer to (
316    //            pointer to byte_array: i32,
317    //            length of byte array: i32,
318    //     ): i32
319    //
320    // This pair `(pointer to byte array, length of byte array)` is called a `Buffer`
321    // and can be found in the cargo_test plugin.
322    //
323    // so on the wasm side, we grab the two parameters to the function,
324    // stuff them into a `Buffer`,
325    // and then pray to the `unsafe` Rust gods above that a valid byte array pops out.
326    //
327    // On the flip side, when returning from a wasm function,
328    // we convert whatever serialized result we get into byte array,
329    // which we stuff into a Buffer and allocate on the heap,
330    // which pointer to we then return.
331    // Note the double indirection!
332    //
333    // So when returning from a function, we actually leak memory *twice*:
334    //
335    // 1) once when we leak the byte array
336    // 2) again when we leak the allocated `Buffer`
337    //
338    // This isn't a problem because Wasm stops executing after the function returns,
339    // so the heap is still valid for our inspection when we want to pull things out.
340
341    fn serialize<A: Serialize>(item: A) -> Result<Vec<u8>, Error> {
342        // serialize the argument using bincode
343        let item = bincode::serialize(&item)?;
344        Ok(item)
345    }
346
347    /// Takes an item, allocates a buffer, serializes the argument to that buffer,
348    /// and returns a (ptr, len) pair to that buffer.
349    async fn serialize_to_buffer(
350        alloc_buffer: TypedFunc<u32, u32>,
351        plugin_memory: &mut Memory,
352        mut store: impl AsContextMut<Data = WasiCtxAlloc>,
353        item: Vec<u8>,
354    ) -> Result<(u32, u32), Error> {
355        // allocate a buffer and write the argument to that buffer
356        let buffer_len = item.len() as u32;
357        let buffer_ptr = alloc_buffer.call_async(&mut store, buffer_len).await?;
358        plugin_memory.write(&mut store, buffer_ptr as usize, &item)?;
359        Ok((buffer_ptr, buffer_len))
360    }
361
362    /// Takes `ptr to a `(ptr, len)` pair, and returns `(ptr, len)`.
363    fn deref_buffer(
364        plugin_memory: &mut Memory,
365        store: impl AsContext<Data = WasiCtxAlloc>,
366        buffer: u32,
367    ) -> Result<(u32, u32), Error> {
368        // create a buffer to read the (ptr, length) pair into
369        // this is a total of 4 + 4 = 8 bytes.
370        let raw_buffer = &mut [0; 8];
371        plugin_memory.read(store, buffer as usize, raw_buffer)?;
372
373        // use these bytes (wasm stores things little-endian)
374        // to get a pointer to the buffer and its length
375        let b = raw_buffer;
376        let buffer_ptr = u32::from_le_bytes([b[0], b[1], b[2], b[3]]);
377        let buffer_len = u32::from_le_bytes([b[4], b[5], b[6], b[7]]);
378
379        return Ok((buffer_ptr, buffer_len));
380    }
381
382    /// Takes a `(ptr, len)` pair and returns the corresponding deserialized buffer.
383    fn deserialize_from_buffer<R: DeserializeOwned>(
384        plugin_memory: &mut Memory,
385        store: impl AsContext<Data = WasiCtxAlloc>,
386        buffer_ptr: u32,
387        buffer_len: u32,
388    ) -> Result<R, Error> {
389        let buffer_ptr = buffer_ptr as usize;
390        let buffer_len = buffer_len as usize;
391        let buffer_end = buffer_ptr + buffer_len;
392
393        // read the buffer at this point into a byte array
394        // deserialize the byte array into the provided serde type
395        let result = &plugin_memory.data(store.as_context())[buffer_ptr..buffer_end];
396        let result = bincode::deserialize(result)?;
397
398        // TODO: this is handled wasm-side
399        // // deallocate the argument buffer
400        // self.free_buffer.call(&mut self.store, arg_buffer);
401
402        Ok(result)
403    }
404
405    pub fn function<A: Serialize, R: DeserializeOwned, T: AsRef<str>>(
406        &mut self,
407        name: T,
408    ) -> Result<WasiFn<A, R>, Error> {
409        let fun_name = format!("__{}", name.as_ref());
410        let fun = self
411            .instance
412            .get_typed_func::<(u32, u32), u32, _>(&mut self.store, &fun_name)?;
413        Ok(WasiFn {
414            function: fun,
415            _function_type: PhantomData,
416        })
417    }
418
419    // TODO: dont' use as for conversions
420    pub async fn call<A: Serialize, R: DeserializeOwned>(
421        &mut self,
422        handle: &WasiFn<A, R>,
423        arg: A,
424    ) -> Result<R, Error> {
425        // dbg!(&handle.name);
426        // dbg!(serde_json::to_string(&arg)).unwrap();
427
428        let mut plugin_memory = self
429            .instance
430            .get_memory(&mut self.store, "memory")
431            .ok_or_else(|| anyhow!("Could not grab slice of plugin memory"))?;
432
433        // write the argument to linear memory
434        // this returns a (ptr, lentgh) pair
435        let arg_buffer = Self::serialize_to_buffer(
436            self.store.data().alloc_buffer(),
437            &mut plugin_memory,
438            &mut self.store,
439            Self::serialize(arg)?,
440        )
441        .await?;
442
443        // call the function, passing in the buffer and its length
444        // this returns a ptr to a (ptr, lentgh) pair
445        let result_buffer = handle
446            .function
447            .call_async(&mut self.store, arg_buffer)
448            .await?;
449        let (result_buffer_ptr, result_buffer_len) =
450            Self::deref_buffer(&mut plugin_memory, &mut self.store, result_buffer)?;
451
452        Self::deserialize_from_buffer(
453            &mut plugin_memory,
454            &mut self.store,
455            result_buffer_ptr,
456            result_buffer_len,
457        )
458    }
459}