1
0
Fork 0
mirror of https://github.com/denoland/deno.git synced 2024-11-27 16:10:57 -05:00
denoland-deno/core/shared_queue.rs
2020-04-23 03:35:52 +02:00

316 lines
9 KiB
Rust

// Copyright 2018-2020 the Deno authors. All rights reserved. MIT license.
/*
SharedQueue Binary Layout
+-------------------------------+-------------------------------+
| NUM_RECORDS (32) |
+---------------------------------------------------------------+
| NUM_SHIFTED_OFF (32) |
+---------------------------------------------------------------+
| HEAD (32) |
+---------------------------------------------------------------+
| OFFSETS (32) |
+---------------------------------------------------------------+
| RECORD_ENDS (*MAX_RECORDS) ...
+---------------------------------------------------------------+
| RECORDS (*MAX_RECORDS) ...
+---------------------------------------------------------------+
*/
use crate::bindings;
use crate::ops::OpId;
use rusty_v8 as v8;
const MAX_RECORDS: usize = 100;
/// Total number of records added.
const INDEX_NUM_RECORDS: usize = 0;
/// Number of records that have been shifted off.
const INDEX_NUM_SHIFTED_OFF: usize = 1;
/// The head is the number of initialized bytes in SharedQueue.
/// It grows monotonically.
const INDEX_HEAD: usize = 2;
const INDEX_OFFSETS: usize = 3;
const INDEX_RECORDS: usize = INDEX_OFFSETS + 2 * MAX_RECORDS;
/// Byte offset of where the records begin. Also where the head starts.
const HEAD_INIT: usize = 4 * INDEX_RECORDS;
/// A rough guess at how big we should make the shared buffer in bytes.
pub const RECOMMENDED_SIZE: usize = 128 * MAX_RECORDS;
pub struct SharedQueue {
buf: v8::SharedRef<v8::BackingStore>,
}
impl SharedQueue {
pub fn new(len: usize) -> Self {
let mut buf = Vec::new();
buf.resize(HEAD_INIT + len, 0);
let buf = buf.into_boxed_slice();
let buf = v8::SharedArrayBuffer::new_backing_store_from_boxed_slice(buf);
let mut q = Self {
buf: buf.make_shared(),
};
q.reset();
q
}
pub fn get_backing_store(&mut self) -> &mut v8::SharedRef<v8::BackingStore> {
&mut self.buf
}
pub fn bytes(&self) -> &[u8] {
unsafe {
bindings::get_backing_store_slice(&self.buf, 0, self.buf.byte_length())
}
}
pub fn bytes_mut(&mut self) -> &mut [u8] {
unsafe {
bindings::get_backing_store_slice_mut(
&self.buf,
0,
self.buf.byte_length(),
)
}
}
fn reset(&mut self) {
debug!("rust:shared_queue:reset");
let s: &mut [u32] = self.as_u32_slice_mut();
s[INDEX_NUM_RECORDS] = 0;
s[INDEX_NUM_SHIFTED_OFF] = 0;
s[INDEX_HEAD] = HEAD_INIT as u32;
}
fn as_u32_slice(&self) -> &[u32] {
let p = self.bytes().as_ptr();
// Assert pointer is 32 bit aligned before casting.
assert_eq!((p as usize) % std::mem::align_of::<u32>(), 0);
#[allow(clippy::cast_ptr_alignment)]
let p32 = p as *const u32;
unsafe { std::slice::from_raw_parts(p32, self.bytes().len() / 4) }
}
fn as_u32_slice_mut(&mut self) -> &mut [u32] {
let p = self.bytes_mut().as_mut_ptr();
// Assert pointer is 32 bit aligned before casting.
assert_eq!((p as usize) % std::mem::align_of::<u32>(), 0);
#[allow(clippy::cast_ptr_alignment)]
let p32 = p as *mut u32;
unsafe { std::slice::from_raw_parts_mut(p32, self.bytes().len() / 4) }
}
pub fn size(&self) -> usize {
let s = self.as_u32_slice();
(s[INDEX_NUM_RECORDS] - s[INDEX_NUM_SHIFTED_OFF]) as usize
}
fn num_records(&self) -> usize {
let s = self.as_u32_slice();
s[INDEX_NUM_RECORDS] as usize
}
fn head(&self) -> usize {
let s = self.as_u32_slice();
s[INDEX_HEAD] as usize
}
fn num_shifted_off(&self) -> usize {
let s = self.as_u32_slice();
s[INDEX_NUM_SHIFTED_OFF] as usize
}
fn set_meta(&mut self, index: usize, end: usize, op_id: OpId) {
let s = self.as_u32_slice_mut();
s[INDEX_OFFSETS + 2 * index] = end as u32;
s[INDEX_OFFSETS + 2 * index + 1] = op_id;
}
#[cfg(test)]
fn get_meta(&self, index: usize) -> Option<(OpId, usize)> {
if index < self.num_records() {
let s = self.as_u32_slice();
let end = s[INDEX_OFFSETS + 2 * index] as usize;
let op_id = s[INDEX_OFFSETS + 2 * index + 1];
Some((op_id, end))
} else {
None
}
}
#[cfg(test)]
fn get_offset(&self, index: usize) -> Option<usize> {
if index < self.num_records() {
Some(if index == 0 {
HEAD_INIT
} else {
let s = self.as_u32_slice();
let prev_end = s[INDEX_OFFSETS + 2 * (index - 1)] as usize;
(prev_end + 3) & !3
})
} else {
None
}
}
/// Returns none if empty.
#[cfg(test)]
pub fn shift(&mut self) -> Option<(OpId, &[u8])> {
let u32_slice = self.as_u32_slice();
let i = u32_slice[INDEX_NUM_SHIFTED_OFF] as usize;
if self.size() == 0 {
assert_eq!(i, 0);
return None;
}
let off = self.get_offset(i).unwrap();
let (op_id, end) = self.get_meta(i).unwrap();
if self.size() > 1 {
let u32_slice = self.as_u32_slice_mut();
u32_slice[INDEX_NUM_SHIFTED_OFF] += 1;
} else {
self.reset();
}
println!(
"rust:shared_queue:shift: num_records={}, num_shifted_off={}, head={}",
self.num_records(),
self.num_shifted_off(),
self.head()
);
Some((op_id, &self.bytes()[off..end]))
}
/// Because JS-side may cast popped message to Int32Array it is required
/// that every message is aligned to 4-bytes.
pub fn push(&mut self, op_id: OpId, record: &[u8]) -> bool {
let off = self.head();
assert_eq!(off % 4, 0);
let end = off + record.len();
let aligned_end = (end + 3) & !3;
debug!(
"rust:shared_queue:pre-push: op={}, off={}, end={}, len={}, aligned_end={}",
op_id,
off,
end,
record.len(),
aligned_end,
);
let index = self.num_records();
if aligned_end > self.bytes().len() || index >= MAX_RECORDS {
debug!("WARNING the sharedQueue overflowed");
return false;
}
assert_eq!(aligned_end % 4, 0);
self.set_meta(index, end, op_id);
assert_eq!(end - off, record.len());
self.bytes_mut()[off..end].copy_from_slice(record);
let u32_slice = self.as_u32_slice_mut();
u32_slice[INDEX_NUM_RECORDS] += 1;
u32_slice[INDEX_HEAD] = aligned_end as u32;
debug!(
"rust:shared_queue:push: num_records={}, num_shifted_off={}, head={}",
self.num_records(),
self.num_shifted_off(),
self.head()
);
true
}
}
#[cfg(test)]
mod tests {
use super::*;
use crate::ops::Buf;
#[test]
fn basic() {
let mut q = SharedQueue::new(RECOMMENDED_SIZE);
let h = q.head();
assert!(h > 0);
let r = vec![1u8, 2, 3, 4].into_boxed_slice();
let len = r.len() + h;
assert!(q.push(0, &r));
assert_eq!(q.head(), len);
let r = vec![5, 6, 7, 8].into_boxed_slice();
assert!(q.push(0, &r));
let r = vec![9, 10, 11, 12].into_boxed_slice();
assert!(q.push(0, &r));
assert_eq!(q.num_records(), 3);
assert_eq!(q.size(), 3);
let (_op_id, r) = q.shift().unwrap();
assert_eq!(r, vec![1, 2, 3, 4].as_slice());
assert_eq!(q.num_records(), 3);
assert_eq!(q.size(), 2);
let (_op_id, r) = q.shift().unwrap();
assert_eq!(r, vec![5, 6, 7, 8].as_slice());
assert_eq!(q.num_records(), 3);
assert_eq!(q.size(), 1);
let (_op_id, r) = q.shift().unwrap();
assert_eq!(r, vec![9, 10, 11, 12].as_slice());
assert_eq!(q.num_records(), 0);
assert_eq!(q.size(), 0);
assert!(q.shift().is_none());
assert!(q.shift().is_none());
assert_eq!(q.num_records(), 0);
assert_eq!(q.size(), 0);
}
fn alloc_buf(byte_length: usize) -> Buf {
let mut v = Vec::new();
v.resize(byte_length, 0);
v.into_boxed_slice()
}
#[test]
fn overflow() {
let mut q = SharedQueue::new(RECOMMENDED_SIZE);
assert!(q.push(0, &alloc_buf(RECOMMENDED_SIZE - 5)));
assert_eq!(q.size(), 1);
assert!(!q.push(0, &alloc_buf(6)));
assert_eq!(q.size(), 1);
assert!(q.push(0, &alloc_buf(1)));
assert_eq!(q.size(), 2);
let (_op_id, buf) = q.shift().unwrap();
assert_eq!(buf.len(), RECOMMENDED_SIZE - 5);
assert_eq!(q.size(), 1);
assert!(!q.push(0, &alloc_buf(1)));
let (_op_id, buf) = q.shift().unwrap();
assert_eq!(buf.len(), 1);
assert_eq!(q.size(), 0);
}
#[test]
fn full_records() {
let mut q = SharedQueue::new(RECOMMENDED_SIZE);
for _ in 0..MAX_RECORDS {
assert!(q.push(0, &alloc_buf(1)))
}
assert_eq!(q.push(0, &alloc_buf(1)), false);
// Even if we shift one off, we still cannot push a new record.
let _ignored = q.shift().unwrap();
assert_eq!(q.push(0, &alloc_buf(1)), false);
}
#[test]
fn allow_any_buf_length() {
let mut q = SharedQueue::new(RECOMMENDED_SIZE);
// Check that `record` that has length not a multiple of 4 will
// not cause panic. Still make sure that records are always
// aligned to 4 bytes.
for i in 1..9 {
q.push(0, &alloc_buf(i));
assert_eq!(q.num_records(), i);
assert_eq!(q.head() % 4, 0);
}
}
}