Code review feedback.

* Move `Module::compile` to `Engine::precompile_module`.
* Remove `Module::deserialize` method.
* Make `Module::serialize` the same format as `Engine::precompile_module`.
* Make `Engine::precompile_module` return a `Vec<u8>`.
* Move the remaining serialization-related code to `serialization.rs`.
This commit is contained in:
Peter Huene
2021-03-31 23:46:30 -07:00
parent abf3bf29f9
commit d1313b1291
10 changed files with 136 additions and 146 deletions

View File

@@ -185,13 +185,10 @@ pub extern "C" fn wasmtime_module_deserialize(
binary: &wasm_byte_vec_t,
ret: &mut *mut wasm_module_t,
) -> Option<Box<wasmtime_error_t>> {
handle_result(
Module::deserialize(&engine.engine, binary.as_slice()),
|module| {
let module = Box::new(wasm_module_t::new(module));
*ret = Box::into_raw(module);
},
)
handle_result(Module::new(&engine.engine, binary.as_slice()), |module| {
let module = Box::new(wasm_module_t::new(module));
*ret = Box::into_raw(module);
})
}
#[no_mangle]

View File

@@ -78,6 +78,34 @@ impl Engine {
pub fn same(a: &Engine, b: &Engine) -> bool {
Arc::ptr_eq(&a.inner, &b.inner)
}
/// Ahead-of-time (AOT) compiles a WebAssembly module.
///
/// The `bytes` provided must be in one of two formats:
///
/// * A [binary-encoded][binary] WebAssembly module. This is always supported.
/// * A [text-encoded][text] instance of the WebAssembly text format.
/// This is only supported when the `wat` feature of this crate is enabled.
/// If this is supplied then the text format will be parsed before validation.
/// Note that the `wat` feature is enabled by default.
///
/// [binary]: https://webassembly.github.io/spec/core/binary/index.html
/// [text]: https://webassembly.github.io/spec/core/text/index.html
pub fn precompile_module(&self, bytes: &[u8]) -> Result<Vec<u8>> {
const USE_PAGED_MEM_INIT: bool = cfg!(all(feature = "uffd", target_os = "linux"));
#[cfg(feature = "wat")]
let bytes = wat::parse_bytes(&bytes)?;
let (_, artifacts, types) = wasmtime_jit::CompilationArtifacts::build(
&self.inner.compiler,
&bytes,
USE_PAGED_MEM_INIT,
)?;
crate::module::SerializedModule::from_artifacts(&self.inner.compiler, &artifacts, &types)
.to_bytes()
}
}
impl Default for Engine {

View File

@@ -1,9 +1,7 @@
use crate::types::{ExportType, ExternType, ImportType};
use crate::{Engine, ModuleType};
use anyhow::{bail, Context, Result};
use bincode::Options;
use std::fs;
use std::io::Write;
use std::path::Path;
use std::sync::Arc;
use wasmparser::Validator;
@@ -15,9 +13,7 @@ use wasmtime_jit::{CompilationArtifacts, CompiledModule, TypeTables};
mod serialization;
use serialization::SerializedModule;
const COMPILED_MODULE_HEADER: &[u8] = b"\0wasmtime-aot";
pub use serialization::SerializedModule;
/// A compiled WebAssembly module, ready to be instantiated.
///
@@ -111,14 +107,16 @@ struct ModuleInner {
impl Module {
/// Creates a new WebAssembly `Module` from the given in-memory `bytes`.
///
/// The `bytes` provided must be in one of three formats:
/// The `bytes` provided must be in one of the following formats:
///
/// * A [binary-encoded][binary] WebAssembly module. This is always supported.
/// * A [text-encoded][text] instance of the WebAssembly text format.
/// This is only supported when the `wat` feature of this crate is enabled.
/// If this is supplied then the text format will be parsed before validation.
/// Note that the `wat` feature is enabled by default.
/// * A module compiled with [`Module::compile`] or the `wasmtime compile` command.
/// * A module serialized with [`Module::serialize`].
/// * A module compiled with [`Engine::precompile_module`] or the
/// `wasmtime compile` command.
///
/// The data for the wasm module must be loaded in-memory if it's present
/// elsewhere, for example on disk. This requires that the entire binary is
@@ -177,8 +175,9 @@ impl Module {
/// ```
pub fn new(engine: &Engine, bytes: impl AsRef<[u8]>) -> Result<Module> {
let bytes = bytes.as_ref();
if bytes.starts_with(COMPILED_MODULE_HEADER) {
return Self::deserialize(engine, &bytes[COMPILED_MODULE_HEADER.len()..]);
if let Some(module) = SerializedModule::from_bytes(bytes)? {
return module.into_module(engine);
}
#[cfg(feature = "wat")]
@@ -267,8 +266,8 @@ impl Module {
/// # }
/// ```
pub fn from_binary(engine: &Engine, binary: &[u8]) -> Result<Module> {
if binary.starts_with(COMPILED_MODULE_HEADER) {
return Self::deserialize(engine, &binary[COMPILED_MODULE_HEADER.len()..]);
if let Some(module) = SerializedModule::from_bytes(binary)? {
return module.into_module(engine);
}
// Check to see that the config's target matches the host
@@ -344,41 +343,6 @@ impl Module {
Ok(())
}
/// Ahead-of-time (AOT) compiles a WebAssembly module.
///
/// The `bytes` provided must be in one of two formats:
///
/// * A [binary-encoded][binary] WebAssembly module. This is always supported.
/// * A [text-encoded][text] instance of the WebAssembly text format.
/// This is only supported when the `wat` feature of this crate is enabled.
/// If this is supplied then the text format will be parsed before validation.
/// Note that the `wat` feature is enabled by default.
///
/// See [`Module::new`] for errors that may be returned by this function.
///
/// [binary]: https://webassembly.github.io/spec/core/binary/index.html
/// [text]: https://webassembly.github.io/spec/core/text/index.html
pub fn compile(engine: &Engine, bytes: &[u8], mut output: impl Write) -> Result<()> {
const USE_PAGED_MEM_INIT: bool = cfg!(all(feature = "uffd", target_os = "linux"));
if bytes.starts_with(COMPILED_MODULE_HEADER) {
bail!("input is already a compiled module");
}
#[cfg(feature = "wat")]
let bytes = wat::parse_bytes(&bytes)?;
let (_, artifacts, types) =
CompilationArtifacts::build(engine.compiler(), &bytes, USE_PAGED_MEM_INIT)?;
// Write a header that marks this as a compiled module
output.write_all(COMPILED_MODULE_HEADER)?;
Self::serialize_module(
&SerializedModule::from_artifacts(engine.compiler(), &artifacts, &types),
output,
)
}
/// Returns the type signature of this module.
pub fn ty(&self) -> ModuleType {
let mut sig = ModuleType::new();
@@ -396,58 +360,12 @@ impl Module {
sig
}
/// Serialize compilation artifacts to the buffer. See also `deserialize`.
pub fn serialize(&self) -> Result<Vec<u8>> {
let mut buffer = Vec::new();
Self::serialize_module(&SerializedModule::new(self), &mut buffer)?;
Ok(buffer)
}
fn serialize_module(module: &SerializedModule, mut output: impl Write) -> Result<()> {
// Preface the data with a version so we can do a version check independent
// of the serialized data.
let version = env!("CARGO_PKG_VERSION");
assert!(
version.len() < 256,
"package version must be less than 256 bytes"
);
output.write(&[version.len() as u8])?;
output.write_all(version.as_bytes())?;
bincode_options().serialize_into(output, module)?;
Ok(())
}
/// Deserializes and creates a module from the compilation artifacts.
/// The `serialize` saves the compilation artifacts along with the host
/// fingerprint, which consists of target, compiler flags, and wasmtime
/// package version.
/// Serialize the module to a vector of bytes.
///
/// The method will fail if fingerprints of current host and serialized
/// one are different. The method does not verify the serialized artifacts
/// for modifications or corruptions. All responsibly of signing and its
/// verification falls on the embedder.
pub fn deserialize(engine: &Engine, serialized: &[u8]) -> Result<Module> {
if serialized.is_empty() {
bail!("serialized data data is empty");
}
let version_len = serialized[0] as usize;
if serialized.len() < version_len + 1 {
bail!("serialized data is malformed");
}
let version = std::str::from_utf8(&serialized[1..1 + version_len])?;
if version != env!("CARGO_PKG_VERSION") {
bail!(
"Module was compiled with incompatible Wasmtime version '{}'",
version
);
}
bincode_options()
.deserialize::<SerializedModule<'_>>(&serialized[1 + version_len..])
.context("Deserialize compilation artifacts")?
.into_module(engine)
/// Use `Module::new` or `Module::from_binary` to create the module
/// from the bytes.
pub fn serialize(&self) -> Result<Vec<u8>> {
SerializedModule::new(self).to_bytes()
}
/// Creates a submodule `Module` value from the specified parameters.
@@ -732,17 +650,6 @@ impl Module {
}
}
fn bincode_options() -> impl Options {
// Use a variable-length integer encoding instead of fixed length. The
// module shown on #2318 gets compressed from ~160MB to ~110MB simply using
// this, presumably because there's a lot of 8-byte integers which generally
// have small values. Local testing shows that the deserialization
// performance, while higher, is in the few-percent range. For huge size
// savings this seems worthwhile to lose a small percentage of
// deserialization performance.
bincode::DefaultOptions::new().with_varint_encoding()
}
fn _assert_send_sync() {
fn _assert<T: Send + Sync>() {}
_assert::<Module>();

View File

@@ -2,7 +2,8 @@
use super::ModuleInner;
use crate::{Engine, Module, OptLevel};
use anyhow::{anyhow, bail, Result};
use anyhow::{anyhow, bail, Context, Result};
use bincode::Options;
use serde::{Deserialize, Serialize};
use std::collections::HashMap;
use std::hash::{Hash, Hasher};
@@ -14,6 +15,19 @@ use wasmtime_jit::{
CompilationArtifacts, CompilationStrategy, CompiledModule, Compiler, TypeTables,
};
const HEADER: &[u8] = b"\0wasmtime-aot";
fn bincode_options() -> impl Options {
// Use a variable-length integer encoding instead of fixed length. The
// module shown on #2318 gets compressed from ~160MB to ~110MB simply using
// this, presumably because there's a lot of 8-byte integers which generally
// have small values. Local testing shows that the deserialization
// performance, while higher, is in the few-percent range. For huge size
// savings this seems worthwhile to lose a small percentage of
// deserialization performance.
bincode::DefaultOptions::new().with_varint_encoding()
}
// This exists because `wasmparser::WasmFeatures` isn't serializable
#[derive(Hash, Debug, Copy, Clone, Serialize, Deserialize)]
struct WasmFeatures {
@@ -273,6 +287,60 @@ impl<'a> SerializedModule<'a> {
}
}
pub fn to_bytes(&self) -> Result<Vec<u8>> {
use std::io::Write;
let mut bytes = Vec::new();
bytes.write_all(HEADER)?;
// Preface the data with a version so we can do a version check independent
// of the serialized data.
let version = env!("CARGO_PKG_VERSION");
assert!(
version.len() < 256,
"package version must be less than 256 bytes"
);
bytes.write(&[version.len() as u8])?;
bytes.write_all(version.as_bytes())?;
bincode_options().serialize_into(&mut bytes, self)?;
Ok(bytes)
}
pub fn from_bytes(bytes: &[u8]) -> Result<Option<Self>> {
if !bytes.starts_with(HEADER) {
return Ok(None);
}
let bytes = &bytes[HEADER.len()..];
if bytes.is_empty() {
bail!("serialized data data is empty");
}
let version_len = bytes[0] as usize;
if bytes.len() < version_len + 1 {
bail!("serialized data is malformed");
}
let version = std::str::from_utf8(&bytes[1..1 + version_len])?;
if version != env!("CARGO_PKG_VERSION") {
bail!(
"Module was compiled with incompatible Wasmtime version '{}'",
version
);
}
Ok(Some(
bincode_options()
.deserialize::<SerializedModule<'_>>(&bytes[1 + version_len..])
.context("deserialize compilation artifacts")?,
))
}
fn check_triple(&self, isa: &dyn TargetIsa) -> Result<()> {
let triple = target_lexicon::Triple::from_str(&self.target).map_err(|e| anyhow!(e))?;