Compare commits

..

7 Commits

Author SHA1 Message Date
Steve Manuel
211d55337d feat: add rust-protobuf support to extism-convert (#653)
(code from @zshipko, thank you!)

---------

Co-authored-by: Zach Shipko <zach@dylibso.com>
2024-01-05 16:36:40 -07:00
zach
431bc4d8af cleanup: improve wat detection (again) (#651)
Merged the other one too quickly, this PR is updated with some
additional feedback from @chrisdickinson:

              (Ooh, this could also start with `(;`)
https://github.com/extism/extism/pull/650#discussion_r1443419249

It's also updated to accommodate for modules that start with an open
paren followed by some whitespace. I doubt this covers all of the
permitted syntax but it should be good enough.
2024-01-05 14:58:47 -08:00
zach
cd4fc39655 cleanup: improve wat detection (#650)
- Update to ensure `is_wat` is only true when the input is a valid
string
- Update to allow leading whitespace in a wat file
2024-01-05 14:00:41 -08:00
zach
03e761908c feat: add Plugin::call_get_error_code to get Extism error code along with the error (#649)
- Adds `Plugin::call_get_error_code` to get the Extism error code when a
call fails
2024-01-05 12:52:31 -08:00
zach
26542d5740 feat(kernel): add extism_length_unsafe (#648)
- Adds `length_unsafe` function to the extism kernel, a more performant
`length` for known-valid memory handles

After this is merged I will update go-sdk and js-sdk too.

Closes #643
2024-01-03 09:29:05 -08:00
CosmicHorror
950a0f449f Toggle off default clap feature for cbindgen (#644)
The only feature for `cbindgen` is the `clap` feature for using it as a
standalone binary and isn't needed when using it as a library
2023-12-26 19:40:32 -08:00
zach
c8868c37d8 chore: update wasmtime bounds to include 16.0.0 (#642) 2023-12-20 14:37:43 -08:00
17 changed files with 143 additions and 258 deletions

View File

@@ -14,6 +14,7 @@ anyhow = "1.0.75"
base64 = "~0.21"
bytemuck = {version = "1.14.0", optional = true }
prost = { version = "0.12.0", optional = true }
protobuf = { version = "3.2.0", optional = true }
rmp-serde = { version = "1.1.2", optional = true }
serde = "1.0.186"
serde_json = "1.0.105"
@@ -22,7 +23,6 @@ serde_json = "1.0.105"
serde = { version = "1.0.186", features = ["derive"] }
[features]
default = ["msgpack", "protobuf", "raw"]
default = ["msgpack", "prost", "raw"]
msgpack = ["rmp-serde"]
protobuf = ["prost"]
raw = ["bytemuck"]

View File

@@ -112,19 +112,19 @@ impl FromBytesOwned for Base64<String> {
/// Protobuf encoding
///
/// Allows for `prost` Protobuf messages to be used as arguments to Extism plugin calls
#[cfg(feature = "protobuf")]
#[cfg(feature = "prost")]
#[derive(Debug)]
pub struct Protobuf<T: prost::Message>(pub T);
pub struct Prost<T: prost::Message>(pub T);
#[cfg(feature = "protobuf")]
impl<T: prost::Message> From<T> for Protobuf<T> {
#[cfg(feature = "prost")]
impl<T: prost::Message> From<T> for Prost<T> {
fn from(data: T) -> Self {
Self(data)
}
}
#[cfg(feature = "protobuf")]
impl<'a, T: prost::Message> ToBytes<'a> for Protobuf<T> {
#[cfg(feature = "prost")]
impl<'a, T: prost::Message> ToBytes<'a> for Prost<T> {
type Bytes = Vec<u8>;
fn to_bytes(&self) -> Result<Self::Bytes, Error> {
@@ -132,10 +132,32 @@ impl<'a, T: prost::Message> ToBytes<'a> for Protobuf<T> {
}
}
#[cfg(feature = "protobuf")]
impl<T: Default + prost::Message> FromBytesOwned for Protobuf<T> {
#[cfg(feature = "prost")]
impl<T: Default + prost::Message> FromBytesOwned for Prost<T> {
fn from_bytes_owned(data: &[u8]) -> Result<Self, Error> {
Ok(Protobuf(T::decode(data)?))
Ok(Prost(T::decode(data)?))
}
}
/// Protobuf encoding
///
/// Allows for `rust-protobuf` Protobuf messages to be used as arguments to Extism plugin calls
#[cfg(feature = "protobuf")]
pub struct Protobuf<T: protobuf::Message>(pub T);
#[cfg(feature = "protobuf")]
impl<'a, T: protobuf::Message> ToBytes<'a> for Protobuf<T> {
type Bytes = Vec<u8>;
fn to_bytes(&self) -> Result<Self::Bytes, Error> {
Ok(self.0.write_to_bytes()?)
}
}
#[cfg(feature = "protobuf")]
impl<T: Default + protobuf::Message> FromBytesOwned for Protobuf<T> {
fn from_bytes_owned(data: &[u8]) -> Result<Self, Error> {
Ok(Protobuf(T::parse_from_bytes(data)?))
}
}

View File

@@ -18,6 +18,9 @@ pub use encoding::{Base64, Json};
#[cfg(feature = "msgpack")]
pub use encoding::Msgpack;
#[cfg(feature = "prost")]
pub use encoding::Prost;
#[cfg(feature = "protobuf")]
pub use encoding::Protobuf;

View File

@@ -382,11 +382,40 @@ pub unsafe fn free(p: Pointer) {
}
/// Get the length of an allocated memory block
///
/// Note: this should only be called on memory handles returned
/// by a call to `alloc` - it will return garbage on invalid offsets
#[no_mangle]
pub unsafe fn length_unsafe(p: Pointer) -> Length {
if p == 0 {
return 0;
}
if !MemoryRoot::pointer_in_bounds_fast(p) {
return 0;
}
let ptr = p - core::mem::size_of::<MemoryBlock>() as u64;
let block = &mut *(ptr as *mut MemoryBlock);
// Simplest sanity check to verify the pointer is a block
if block.status.load(Ordering::Acquire) != MemoryStatus::Active as u8 {
return 0;
}
block.used as Length
}
/// Get the length but returns 0 if the offset is not a valid handle.
///
/// Note: this function walks each node in the allocations list, which ensures correctness, but is also
/// slow
#[no_mangle]
pub unsafe fn length(p: Pointer) -> Length {
if p == 0 {
return 0;
}
if let Some(block) = MemoryRoot::new().find_block(p) {
block.used as Length
} else {

View File

@@ -9,8 +9,8 @@ repository.workspace = true
version.workspace = true
[dependencies]
wasmtime = ">= 14.0.0, < 16.0.0"
wasmtime-wasi = ">= 14.0.0, < 16.0.0"
wasmtime = ">= 14.0.0, < 17.0.0"
wasmtime-wasi = ">= 14.0.0, < 17.0.0"
anyhow = "1"
serde = {version = "1", features = ["derive"]}
serde_json = "1"
@@ -33,7 +33,7 @@ register-filesystem = [] # enables wasm to be loaded from disk
http = ["ureq"] # enables extism_http_request
[build-dependencies]
cbindgen = "0.26"
cbindgen = { version = "0.26", default-features = false }
[dev-dependencies]
criterion = "0.5.1"

View File

@@ -173,13 +173,6 @@ void extism_function_free(ExtismFunction *f);
*/
void extism_function_set_namespace(ExtismFunction *ptr, const char *namespace_);
/**
* Set the cost of an `ExtismFunction`, when set to 0 this has no effect, when set to `1` this will add
* the runtime of the function back to the plugin timer.
*/
void extism_function_set_cost(ExtismFunction *ptr,
double cost);
/**
* Create a new plugin with host functions, the functions passed to this function no longer need to be manually freed using
*

View File

@@ -246,6 +246,26 @@ impl CurrentPlugin {
Ok(len)
}
pub fn memory_length_unsafe(&mut self, offs: u64) -> Result<u64, Error> {
let (linker, mut store) = self.linker_and_store();
let output = &mut [Val::I64(0)];
if let Some(f) = linker.get(&mut store, EXTISM_ENV_MODULE, "length_unsafe") {
f.into_func()
.unwrap()
.call(&mut store, &[Val::I64(offs as i64)], output)?;
} else {
anyhow::bail!("unable to locate an extism kernel function: length_unsafe",)
}
let len = output[0].unwrap_i64() as u64;
trace!(
plugin = self.id.to_string(),
"memory_length_unsafe({}) = {}",
offs,
len
);
Ok(len)
}
/// Access a plugin's variables
pub fn vars(&self) -> &std::collections::BTreeMap<String, Vec<u8>> {
&self.vars

Binary file not shown.

View File

@@ -155,7 +155,7 @@ unsafe impl<T> Sync for UserData<T> {}
unsafe impl Send for CPtr {}
unsafe impl Sync for CPtr {}
pub(crate) type FunctionInner = dyn Fn(wasmtime::Caller<CurrentPlugin>, &[wasmtime::Val], &mut [wasmtime::Val]) -> Result<(), Error>
type FunctionInner = dyn Fn(wasmtime::Caller<CurrentPlugin>, &[wasmtime::Val], &mut [wasmtime::Val]) -> Result<(), Error>
+ Sync
+ Send;
@@ -174,9 +174,6 @@ pub struct Function {
/// Function handle
pub(crate) f: Arc<FunctionInner>,
/// Function cost (in terms of time)
pub(crate) cost: f64,
/// UserData
pub(crate) _user_data: UserDataHandle,
}
@@ -207,12 +204,10 @@ impl Function {
ty,
f: Arc::new(
move |mut caller: Caller<_>, inp: &[Val], outp: &mut [Val]| {
let plugin = caller.data_mut();
let x = data.clone();
f(plugin, inp, outp, x)
f(caller.data_mut(), inp, outp, x)
},
),
cost: 0.0,
namespace: None,
_user_data: match &user_data {
UserData::C(ptr) => UserDataHandle::C(ptr.clone()),
@@ -244,23 +239,6 @@ impl Function {
self
}
/// Function cost
pub fn cost(&self) -> f64 {
self.cost
}
/// Set host function cost
pub fn set_cost(&mut self, cost: f64) {
trace!("Setting cost for {} to {cost}", self.name);
self.cost = cost;
}
/// Update host function cost
pub fn with_cost(mut self, cost: f64) -> Self {
self.set_cost(cost);
self
}
/// Get function type
pub fn ty(&self) -> &wasmtime::FuncType {
&self.ty

View File

@@ -14,7 +14,6 @@ pub(crate) mod manifest;
pub(crate) mod pdk;
mod plugin;
mod plugin_builder;
mod timeout_manager;
mod timer;
/// Extism C API
@@ -28,7 +27,6 @@ pub use plugin::{CancelHandle, Plugin, WasmInput, EXTISM_ENV_MODULE, EXTISM_USER
pub use plugin_builder::{DebugOptions, PluginBuilder};
pub(crate) use internal::{Internal, Wasi};
pub(crate) use timeout_manager::TimeoutManager;
pub(crate) use timer::{Timer, TimerAction};
pub(crate) use tracing::{debug, error, trace, warn};

View File

@@ -117,10 +117,17 @@ pub(crate) fn load(
match input {
WasmInput::Data(data) => {
let has_magic = data.len() >= 4 && data[0..4] == WASM_MAGIC;
let is_wat = data.starts_with(b"(module") || data.starts_with(b";;");
let s = std::str::from_utf8(&data);
let is_wat = s.is_ok_and(|s| {
let s = s.trim_start();
let starts_with_module = s.len() > 2
&& data[0] == b'(' // First character is `(`
&& s[1..].trim_start().starts_with("module"); // Then `module` (after any whitespace)
starts_with_module || s.starts_with(";;") || s.starts_with("(;")
});
if !has_magic && !is_wat {
trace!("Loading manifest");
if let Ok(s) = std::str::from_utf8(&data) {
if let Ok(s) = s {
let t = if let Ok(t) = toml::from_str::<extism_manifest::Manifest>(s) {
trace!("Manifest is TOML");
modules(engine, &t, &mut mods)?;

View File

@@ -238,20 +238,6 @@ impl Plugin {
CurrentPlugin::new(manifest, with_wasi, available_pages, id)?,
);
store.set_epoch_deadline(1);
store.call_hook(|data, hook| {
if hook.entering_host() {
let tx = Timer::tx();
tx.send(TimerAction::EnterHost {
id: data.id.clone(),
})?;
} else if hook.exiting_host() {
let tx = Timer::tx();
tx.send(TimerAction::ExitHost {
id: data.id.clone(),
})?;
}
Ok(())
});
let mut linker = Linker::new(&engine);
linker.allow_shadowing(true);
@@ -298,13 +284,9 @@ impl Plugin {
for f in &mut imports {
let name = f.name().to_string();
let ns = f.namespace().unwrap_or(EXTISM_USER_MODULE);
let cost = f.cost();
let inner: &function::FunctionInner = unsafe { &*(f.f.as_ref() as *const _) };
linker.func_new(ns, &name, f.ty().clone(), move |caller, params, results| {
let _timeout =
TimeoutManager::new(caller.data()).map(|x| x.with_cost(cost.clone()));
inner(caller, params, results)
})?;
unsafe {
linker.func_new(ns, &name, f.ty().clone(), &*(f.f.as_ref() as *const _))?;
}
}
let instance_pre = linker.instantiate_pre(main)?;
@@ -842,6 +824,24 @@ impl Plugin {
.and_then(move |_| self.output())
}
/// Similar to `Plugin::call`, but returns the Extism error code along with the
/// `Error`. It is assumed if `Ok(_)` is returned that the error code was `0`.
///
/// All Extism plugin calls return an error code, `Plugin::call` consumes the error code,
/// while `Plugin::call_get_error_code` preserves it - this function should only be used
/// when you need to inspect the actual return value of a plugin function when it fails.
pub fn call_get_error_code<'a, 'b, T: ToBytes<'a>, U: FromBytes<'b>>(
&'b mut self,
name: impl AsRef<str>,
input: T,
) -> Result<U, (Error, i32)> {
let lock = self.instance.clone();
let mut lock = lock.lock().unwrap();
let data = input.to_bytes().map_err(|e| (e, -1))?;
self.raw_call(&mut lock, name, data)
.and_then(move |_| self.output().map_err(|e| (e, -1)))
}
/// Get a `CancelHandle`, which can be used from another thread to cancel a running plugin
pub fn cancel_handle(&self) -> CancelHandle {
self.cancel_handle.clone()

View File

@@ -258,18 +258,6 @@ pub unsafe extern "C" fn extism_function_set_namespace(
}
}
/// Set the cost of an `ExtismFunction`, when set to 0 this has no effect, when set to `1` this will add
/// the runtime of the function back to the plugin timer.
#[no_mangle]
pub unsafe extern "C" fn extism_function_set_cost(ptr: *mut ExtismFunction, cost: f64) {
let f = &mut *ptr;
if let Some(x) = f.0.get_mut() {
x.set_cost(cost);
} else {
debug!("Trying to set the cost of already registered function")
}
}
/// Create a new plugin with host functions, the functions passed to this function no longer need to be manually freed using
///
/// `wasm`: is a WASM module (wat or wasm) or a JSON encoded manifest

View File

@@ -22,6 +22,20 @@ fn extism_length<T>(mut store: &mut wasmtime::Store<T>, instance: &mut Instance,
out[0].unwrap_i64() as u64
}
fn extism_length_unsafe<T>(
mut store: &mut wasmtime::Store<T>,
instance: &mut Instance,
p: u64,
) -> u64 {
let out = &mut [Val::I64(0)];
instance
.get_func(&mut store, "length_unsafe")
.unwrap()
.call(&mut store, &[Val::I64(p as i64)], out)
.unwrap();
out[0].unwrap_i64() as u64
}
fn extism_load_u8<T>(mut store: &mut wasmtime::Store<T>, instance: &mut Instance, p: u64) -> u8 {
let out = &mut [Val::I32(0)];
instance
@@ -173,6 +187,7 @@ fn test_kernel_allocations() {
let first_alloc = p;
assert!(p > 0);
assert_eq!(extism_length(&mut store, instance, p), 1);
assert_eq!(extism_length_unsafe(&mut store, instance, p), 1);
extism_free(&mut store, instance, p);
// 2 bytes
@@ -180,18 +195,21 @@ fn test_kernel_allocations() {
assert!(x > 0);
assert!(x != p);
assert_eq!(extism_length(&mut store, instance, x), 2);
assert_eq!(extism_length_unsafe(&mut store, instance, x), 2);
extism_free(&mut store, instance, x);
for i in 0..64 {
let p = extism_alloc(&mut store, instance, 64 - i);
assert!(p > 0);
assert_eq!(extism_length(&mut store, instance, p), 64 - i);
assert_eq!(extism_length_unsafe(&mut store, instance, p), 64 - i);
extism_free(&mut store, instance, p);
// should re-use the last allocation
let q = extism_alloc(&mut store, instance, 64 - i);
assert_eq!(p, q);
assert_eq!(extism_length(&mut store, instance, q), 64 - i);
assert_eq!(extism_length_unsafe(&mut store, instance, q), 64 - i);
extism_free(&mut store, instance, q);
}

View File

@@ -235,43 +235,6 @@ fn test_timeout() {
assert!(err == "timeout");
}
fn hello_world_timeout_manager(
plugin: &mut CurrentPlugin,
inputs: &[Val],
outputs: &mut [Val],
_user_data: UserData<()>,
) -> Result<(), Error> {
let mgr = plugin.timeout_manager().map(|x| x.add_on_drop());
std::thread::sleep(std::time::Duration::from_secs(3));
outputs[0] = inputs[0].clone();
drop(mgr);
Ok(())
}
#[test]
fn test_timeout_manager() {
let f = Function::new(
"hello_world",
[PTR],
[PTR],
UserData::default(),
hello_world_timeout_manager,
);
let manifest = Manifest::new([extism_manifest::Wasm::data(WASM)])
.with_timeout(std::time::Duration::from_secs(1));
let mut plugin = Plugin::new(manifest, [f], true).unwrap();
let start = std::time::Instant::now();
let output: Result<&[u8], Error> = plugin.call("count_vowels", "testing");
println!("Result {:?}", output);
assert!(output.is_ok());
let end = std::time::Instant::now();
let time = end - start;
println!("Plugin ran for {:?}", time);
assert!(time.as_secs() >= 3);
}
typed_plugin!(pub TestTypedPluginGenerics {
count_vowels<T: FromBytes<'a>>(&str) -> T
});

View File

@@ -1,59 +0,0 @@
use crate::*;
/// `TimeoutManager` is used to control `Plugin` timeouts from within host functions
///
/// It can be used to add or subtract time from a plug-in's timeout. If a plugin is not
/// configured to have a timeout then this will have no effect.
pub(crate) struct TimeoutManager {
start_time: std::time::Instant,
id: uuid::Uuid,
tx: std::sync::mpsc::Sender<TimerAction>,
cost: f64,
}
impl TimeoutManager {
/// Create a new `TimeoutManager` from the `CurrentPlugin`, this will return `None` if no timeout
/// is configured
pub fn new(plugin: &CurrentPlugin) -> Option<TimeoutManager> {
plugin.manifest.timeout_ms.map(|_| TimeoutManager {
start_time: std::time::Instant::now(),
id: plugin.id.clone(),
tx: Timer::tx(),
cost: 0.0,
})
}
/// Add the amount of time this value has existed to the configured timeout
pub fn add_elapsed(&mut self) -> Result<(), Error> {
let cost = self.cost.abs();
let d = self.start_time.elapsed().mul_f64(cost);
let mut d: timer::ExtendTimeout = d.into();
if self.cost.is_sign_negative() {
d = -d;
}
self.tx.send(TimerAction::Extend {
id: self.id.clone(),
duration: d,
})?;
self.start_time = std::time::Instant::now();
Ok(())
}
pub fn with_cost(mut self, cost: f64) -> Self {
self.cost = cost;
self
}
}
impl Drop for TimeoutManager {
fn drop(&mut self) {
let x = self.add_elapsed();
if let Err(e) = x {
error!(
plugin = self.id.to_string(),
"unable to extend timeout: {}",
e.to_string()
);
}
}
}

View File

@@ -1,51 +1,17 @@
use crate::*;
#[derive(Debug)]
pub(crate) struct ExtendTimeout {
duration: std::time::Duration,
negative: bool,
}
impl From<std::time::Duration> for ExtendTimeout {
fn from(value: std::time::Duration) -> Self {
ExtendTimeout {
duration: value,
negative: false,
}
}
}
impl std::ops::Neg for ExtendTimeout {
type Output = ExtendTimeout;
fn neg(mut self) -> Self::Output {
self.negative = !self.negative;
self
}
}
pub(crate) enum TimerAction {
Start {
id: uuid::Uuid,
engine: Engine,
duration: Option<std::time::Duration>,
},
Extend {
id: uuid::Uuid,
duration: ExtendTimeout,
},
Stop {
id: uuid::Uuid,
},
Cancel {
id: uuid::Uuid,
},
EnterHost {
id: uuid::Uuid,
},
ExitHost {
id: uuid::Uuid,
},
Shutdown,
}
@@ -65,8 +31,6 @@ extern "C" fn cleanup_timer() {
static mut TIMER: std::sync::Mutex<Option<Timer>> = std::sync::Mutex::new(None);
type TimerMap = std::collections::BTreeMap<uuid::Uuid, (Engine, Option<std::time::Instant>)>;
impl Timer {
pub(crate) fn tx() -> std::sync::mpsc::Sender<TimerAction> {
let mut timer = match unsafe { TIMER.lock() } {
@@ -85,8 +49,7 @@ impl Timer {
pub fn init(timer: &mut Option<Timer>) -> std::sync::mpsc::Sender<TimerAction> {
let (tx, rx) = std::sync::mpsc::channel();
let thread = std::thread::spawn(move || {
let mut plugins = TimerMap::new();
let mut in_host = TimerMap::new();
let mut plugins = std::collections::BTreeMap::new();
macro_rules! handle {
($x:expr) => {
@@ -107,16 +70,12 @@ impl Timer {
TimerAction::Stop { id } => {
trace!(plugin = id.to_string(), "handling stop event");
plugins.remove(&id);
in_host.remove(&id);
}
TimerAction::Cancel { id } => {
trace!(plugin = id.to_string(), "handling cancel event");
if let Some((engine, _)) = plugins.remove(&id) {
engine.increment_epoch();
}
if let Some((engine, _)) = in_host.remove(&id) {
engine.increment_epoch();
}
}
TimerAction::Shutdown => {
trace!("Shutting down timer");
@@ -124,42 +83,8 @@ impl Timer {
trace!(plugin = id.to_string(), "handling shutdown event");
engine.increment_epoch();
}
for (id, (engine, _)) in in_host.iter() {
trace!(plugin = id.to_string(), "handling shutdown event");
engine.increment_epoch();
}
return;
}
TimerAction::Extend { id, duration } => {
if let Some((_engine, Some(timeout))) = plugins.get_mut(&id) {
let x = if duration.negative {
timeout.checked_sub(duration.duration)
} else {
timeout.checked_add(duration.duration)
};
if let Some(t) = x {
*timeout = t;
} else {
error!(
plugin = id.to_string(),
"unable to extend timeout by {:?}", duration.duration
);
}
}
}
TimerAction::EnterHost { id } => {
trace!(plugin = id.to_string(), "enter host function");
if let Some(x) = plugins.remove(&id) {
in_host.insert(id, x);
}
}
TimerAction::ExitHost { id } => {
trace!(plugin = id.to_string(), "exit host function");
if let Some(x) = in_host.remove(&id) {
plugins.insert(id, x);
}
}
}
};
}
@@ -171,10 +96,6 @@ impl Timer {
}
}
for x in rx.try_iter() {
handle!(x)
}
plugins = plugins
.into_iter()
.filter(|(_k, (engine, end))| {
@@ -188,6 +109,10 @@ impl Timer {
true
})
.collect();
for x in rx.try_iter() {
handle!(x)
}
}
});
*timer = Some(Timer {