Files
wasmtime/crates/cache/src/lib.rs
Alex Crichton c73be1f13a Use an mmap-friendly serialization format (#3257)
* Use an mmap-friendly serialization format

This commit reimplements the main serialization format for Wasmtime's
precompiled artifacts. Previously they were generally a binary blob of
`bincode`-encoded metadata prefixed with some versioning information.
The downside of this format, though, is that loading a precompiled
artifact required pushing all information through `bincode`. This is
inefficient when some data, such as trap/address tables, are rarely
accessed.

The new format added in this commit is one which is designed to be
`mmap`-friendly. This means that the relevant parts of the precompiled
artifact are already page-aligned for updating permissions of pieces
here and there. Additionally the artifact is optimized so that if data
is rarely read then we can delay reading it until necessary.

The new artifact format for serialized modules is an ELF file. This is
not a public API guarantee, so it cannot be relied upon. In the meantime
though this is quite useful for exploring precompiled modules with
standard tooling like `objdump`. The ELF file is already constructed as
part of module compilation, and this is the main contents of the
serialized artifact.

THere is some extra information, though, not encoded in each module's
individual ELF file such as type information. This information continues
to be `bincode`-encoded, but it's intended to be much smaller and much
faster to deserialize. This extra information is appended to the end of
the ELF file. This means that the original ELF file is still a valid ELF
file, we just get to have extra bits at the end. More information on the
new format can be found in the module docs of the serialization module
of Wasmtime.

Another refatoring implemented as part of this commit is to deserialize
and store object files directly in `mmap`-backed storage. This avoids
the need to copy bytes after the artifact is loaded into memory for each
compiled module, and in a future commit it opens up the door to avoiding
copying the text section into a `CodeMemory`. For now, though, the main
change is that copies are not necessary when loading from a precompiled
compilation artifact once the artifact is itself in mmap-based memory.

To assist with managing `mmap`-based memory a new `MmapVec` type was
added to `wasmtime_jit` which acts as a form of `Vec<T>` backed by a
`wasmtime_runtime::Mmap`. This type notably supports `drain(..N)` to
slice the buffer into disjoint regions that are all separately owned,
such as having a separately owned window into one artifact for all
object files contained within.

Finally this commit implements a small refactoring in `wasmtime-cache`
to use the standard artifact format for cache entries rather than a
bincode-encoded version. This required some more hooks for
serializing/deserializing but otherwise the crate still performs as
before.

* Review comments
2021-08-30 09:19:20 -05:00

238 lines
8.4 KiB
Rust

use log::{debug, trace, warn};
use serde::{Deserialize, Serialize};
use sha2::{Digest, Sha256};
use std::fs;
use std::hash::Hash;
use std::hash::Hasher;
use std::io::Write;
use std::path::{Path, PathBuf};
#[macro_use] // for tests
mod config;
mod worker;
pub use config::{create_new_config, CacheConfig};
use worker::Worker;
/// Module level cache entry.
pub struct ModuleCacheEntry<'config>(Option<ModuleCacheEntryInner<'config>>);
struct ModuleCacheEntryInner<'config> {
root_path: PathBuf,
cache_config: &'config CacheConfig,
}
struct Sha256Hasher(Sha256);
impl<'config> ModuleCacheEntry<'config> {
/// Create the cache entry.
pub fn new<'data>(compiler_name: &str, cache_config: &'config CacheConfig) -> Self {
if cache_config.enabled() {
Self(Some(ModuleCacheEntryInner::new(
compiler_name,
cache_config,
)))
} else {
Self(None)
}
}
#[cfg(test)]
fn from_inner(inner: ModuleCacheEntryInner<'config>) -> Self {
Self(Some(inner))
}
/// Gets cached data if state matches, otherwise calls `compute`.
///
/// Data is automatically serialized/deserialized with `bincode`.
pub fn get_data<T, U, E>(&self, state: T, compute: fn(&T) -> Result<U, E>) -> Result<U, E>
where
T: Hash,
U: Serialize + for<'a> Deserialize<'a>,
{
self.get_data_raw(
&state,
compute,
|_state, data| bincode::serialize(data).ok(),
|_state, data| bincode::deserialize(&data).ok(),
)
}
/// Gets cached data if state matches, otherwise calls `compute`.
///
/// If the cache is disabled or no cached data is found then `compute` is
/// called to calculate the data. If the data was found in cache it is
/// passed to `deserialize`, which if successful will be the returned value.
/// When computed the `serialize` function is used to generate the bytes
/// from the returned value.
pub fn get_data_raw<T, U, E>(
&self,
state: &T,
// NOTE: These are function pointers instead of closures so that they
// don't accidentally close over something not accounted in the cache.
compute: fn(&T) -> Result<U, E>,
serialize: fn(&T, &U) -> Option<Vec<u8>>,
deserialize: fn(&T, Vec<u8>) -> Option<U>,
) -> Result<U, E>
where
T: Hash,
{
let inner = match &self.0 {
Some(inner) => inner,
None => return compute(state),
};
let mut hasher = Sha256Hasher(Sha256::new());
state.hash(&mut hasher);
let hash: [u8; 32] = hasher.0.finalize().into();
// standard encoding uses '/' which can't be used for filename
let hash = base64::encode_config(&hash, base64::URL_SAFE_NO_PAD);
if let Some(cached_val) = inner.get_data(&hash) {
if let Some(val) = deserialize(state, cached_val) {
let mod_cache_path = inner.root_path.join(&hash);
inner.cache_config.on_cache_get_async(&mod_cache_path); // call on success
return Ok(val);
}
}
let val_to_cache = compute(state)?;
if let Some(bytes) = serialize(state, &val_to_cache) {
if inner.update_data(&hash, &bytes).is_some() {
let mod_cache_path = inner.root_path.join(&hash);
inner.cache_config.on_cache_update_async(&mod_cache_path); // call on success
}
}
Ok(val_to_cache)
}
}
impl<'config> ModuleCacheEntryInner<'config> {
fn new<'data>(compiler_name: &str, cache_config: &'config CacheConfig) -> Self {
// If debug assertions are enabled then assume that we're some sort of
// local build. We don't want local builds to stomp over caches between
// builds, so just use a separate cache directory based on the mtime of
// our executable, which should roughly correlate with "you changed the
// source code so you get a different directory".
//
// Otherwise if this is a release build we use the `GIT_REV` env var
// which is either the git rev if installed from git or the crate
// version if installed from crates.io.
let compiler_dir = if cfg!(debug_assertions) {
fn self_mtime() -> Option<String> {
let path = std::env::current_exe().ok()?;
let metadata = path.metadata().ok()?;
let mtime = metadata.modified().ok()?;
Some(match mtime.duration_since(std::time::UNIX_EPOCH) {
Ok(dur) => format!("{}", dur.as_millis()),
Err(err) => format!("m{}", err.duration().as_millis()),
})
}
let self_mtime = self_mtime().unwrap_or("no-mtime".to_string());
format!(
"{comp_name}-{comp_ver}-{comp_mtime}",
comp_name = compiler_name,
comp_ver = env!("GIT_REV"),
comp_mtime = self_mtime,
)
} else {
format!(
"{comp_name}-{comp_ver}",
comp_name = compiler_name,
comp_ver = env!("GIT_REV"),
)
};
let root_path = cache_config.directory().join("modules").join(compiler_dir);
Self {
root_path,
cache_config,
}
}
fn get_data(&self, hash: &str) -> Option<Vec<u8>> {
let mod_cache_path = self.root_path.join(hash);
trace!("get_data() for path: {}", mod_cache_path.display());
let compressed_cache_bytes = fs::read(&mod_cache_path).ok()?;
let cache_bytes = zstd::decode_all(&compressed_cache_bytes[..])
.map_err(|err| warn!("Failed to decompress cached code: {}", err))
.ok()?;
Some(cache_bytes)
}
fn update_data(&self, hash: &str, serialized_data: &[u8]) -> Option<()> {
let mod_cache_path = self.root_path.join(hash);
trace!("update_data() for path: {}", mod_cache_path.display());
let compressed_data = zstd::encode_all(
&serialized_data[..],
self.cache_config.baseline_compression_level(),
)
.map_err(|err| warn!("Failed to compress cached code: {}", err))
.ok()?;
// Optimize syscalls: first, try writing to disk. It should succeed in most cases.
// Otherwise, try creating the cache directory and retry writing to the file.
if fs_write_atomic(&mod_cache_path, "mod", &compressed_data) {
return Some(());
}
debug!(
"Attempting to create the cache directory, because \
failed to write cached code to disk, path: {}",
mod_cache_path.display(),
);
let cache_dir = mod_cache_path.parent().unwrap();
fs::create_dir_all(cache_dir)
.map_err(|err| {
warn!(
"Failed to create cache directory, path: {}, message: {}",
cache_dir.display(),
err
)
})
.ok()?;
if fs_write_atomic(&mod_cache_path, "mod", &compressed_data) {
Some(())
} else {
None
}
}
}
impl Hasher for Sha256Hasher {
fn finish(&self) -> u64 {
panic!("Sha256Hasher doesn't support finish!");
}
fn write(&mut self, bytes: &[u8]) {
self.0.update(bytes);
}
}
// Assumption: path inside cache directory.
// Then, we don't have to use sound OS-specific exclusive file access.
// Note: there's no need to remove temporary file here - cleanup task will do it later.
fn fs_write_atomic(path: &Path, reason: &str, contents: &[u8]) -> bool {
let lock_path = path.with_extension(format!("wip-atomic-write-{}", reason));
fs::OpenOptions::new()
.create_new(true) // atomic file creation (assumption: no one will open it without this flag)
.write(true)
.open(&lock_path)
.and_then(|mut file| file.write_all(contents))
// file should go out of scope and be closed at this point
.and_then(|()| fs::rename(&lock_path, &path)) // atomic file rename
.map_err(|err| {
warn!(
"Failed to write file with rename, lock path: {}, target path: {}, err: {}",
lock_path.display(),
path.display(),
err
)
})
.is_ok()
}
#[cfg(test)]
mod tests;