Files
cloud-hypervisor/block/src/raw_async_aio.rs
Wei Liu 2fe775fce2 block: use BLKDISCARD/BLKZEROOUT ioctls for block devices
Some block devices (ZFS volume) may require BLKDISCARD and BLKZEROOUT
ioctls for discard and write_zeroes operations respectively.

There is no good way to probe whether fallocate is supported on a block
device. Arguably, punch_hole and write_zeroes are rare. Instead of
having a complex scheme for the IO uring backend, we force it to always
use ioctls. The code can be changed if the synchronized ioctls become a
performance issue.

Changes:
- Detect block devices at construction time
- Use BLKDISCARD ioctl for punch_hole (discard) on block devices
- Use BLKZEROOUT ioctl for write_zeroes on block devices
- Add BLKDISCARD/BLKZEROOUT to VirtioBlock seccomp whitelist
- Keep fallocate() path for regular files (no behavior change)
- Consolidate some helper functions to the new sparse module

Signed-off-by: Wei Liu <liuwe@microsoft.com>
2026-05-14 22:35:02 +00:00

194 lines
6.1 KiB
Rust

// Copyright © 2023 Intel Corporation
//
// SPDX-License-Identifier: Apache-2.0 AND BSD-3-Clause
//
// Copyright © 2023 Crusoe Energy Systems LLC
//
use std::collections::VecDeque;
use std::os::unix::io::{AsRawFd, RawFd};
use vmm_sys_util::aio;
use vmm_sys_util::eventfd::EventFd;
use crate::async_io::{AsyncIo, AsyncIoError, AsyncIoResult};
use crate::error::{BlockError, BlockErrorKind, BlockResult};
use crate::sparse::{punch_hole, write_zeroes};
use crate::{SECTOR_SIZE, is_block_device};
pub struct RawFileAsyncAio {
fd: RawFd,
ctx: aio::IoContext,
eventfd: EventFd,
alignment: u64,
completion_list: VecDeque<(u64, i32)>,
is_block_device: bool,
}
impl RawFileAsyncAio {
pub fn new(fd: RawFd, queue_depth: u32) -> BlockResult<Self> {
let eventfd =
EventFd::new(libc::EFD_NONBLOCK).map_err(|e| BlockError::new(BlockErrorKind::Io, e))?;
let ctx =
aio::IoContext::new(queue_depth).map_err(|e| BlockError::new(BlockErrorKind::Io, e))?;
let is_block_device = is_block_device(fd);
Ok(RawFileAsyncAio {
fd,
ctx,
eventfd,
alignment: SECTOR_SIZE,
completion_list: VecDeque::new(),
is_block_device,
})
}
}
impl AsyncIo for RawFileAsyncAio {
fn notifier(&self) -> &EventFd {
&self.eventfd
}
fn alignment(&self) -> u64 {
self.alignment
}
fn read_vectored(
&mut self,
offset: libc::off_t,
iovecs: &[libc::iovec],
user_data: u64,
) -> AsyncIoResult<()> {
let iocbs = [&mut aio::IoControlBlock {
aio_fildes: self.fd.as_raw_fd() as u32,
aio_lio_opcode: aio::IOCB_CMD_PREADV as u16,
aio_buf: iovecs.as_ptr() as u64,
aio_nbytes: iovecs.len() as u64,
aio_offset: offset,
aio_data: user_data,
aio_flags: aio::IOCB_FLAG_RESFD,
aio_resfd: self.eventfd.as_raw_fd() as u32,
..Default::default()
}];
let _ = self
.ctx
.submit(&iocbs[..])
.map_err(AsyncIoError::ReadVectored)?;
Ok(())
}
fn write_vectored(
&mut self,
offset: libc::off_t,
iovecs: &[libc::iovec],
user_data: u64,
) -> AsyncIoResult<()> {
let iocbs = [&mut aio::IoControlBlock {
aio_fildes: self.fd.as_raw_fd() as u32,
aio_lio_opcode: aio::IOCB_CMD_PWRITEV as u16,
aio_buf: iovecs.as_ptr() as u64,
aio_nbytes: iovecs.len() as u64,
aio_offset: offset,
aio_data: user_data,
aio_flags: aio::IOCB_FLAG_RESFD,
aio_resfd: self.eventfd.as_raw_fd() as u32,
..Default::default()
}];
let _ = self
.ctx
.submit(&iocbs[..])
.map_err(AsyncIoError::WriteVectored)?;
Ok(())
}
fn fsync(&mut self, user_data: Option<u64>) -> AsyncIoResult<()> {
if let Some(user_data) = user_data {
let iocbs = [&mut aio::IoControlBlock {
aio_fildes: self.fd.as_raw_fd() as u32,
aio_lio_opcode: aio::IOCB_CMD_FSYNC as u16,
aio_data: user_data,
aio_flags: aio::IOCB_FLAG_RESFD,
aio_resfd: self.eventfd.as_raw_fd() as u32,
..Default::default()
}];
let _ = self.ctx.submit(&iocbs[..]).map_err(AsyncIoError::Fsync)?;
} else {
// SAFETY: FFI call with a valid fd
unsafe { libc::fsync(self.fd) };
}
Ok(())
}
fn next_completed_request(&mut self) -> Option<(u64, i32)> {
if self.completion_list.is_empty() {
// Drain pending AIO completions batched into the same queue.
let mut events = [aio::IoEvent::default(); 32];
let rc = self.ctx.get_events(0, &mut events, None).unwrap();
for event in &events[..rc] {
self.completion_list
.push_back((event.data, event.res as i32));
}
}
self.completion_list.pop_front()
}
fn punch_hole(&mut self, offset: u64, length: u64, user_data: u64) -> AsyncIoResult<()> {
// Linux AIO has no IOCB command for fallocate, so perform the
// operation synchronously and signal completion via the completion
// list, matching the pattern used by the sync backend (RawFileSync).
punch_hole(self.fd, self.is_block_device, offset, length)
.map_err(AsyncIoError::PunchHole)?;
self.completion_list.push_back((user_data, 0));
self.eventfd.write(1).unwrap();
Ok(())
}
fn write_zeroes(&mut self, offset: u64, length: u64, user_data: u64) -> AsyncIoResult<()> {
// Same as punch_hole().
write_zeroes(self.fd, self.is_block_device, offset, length)
.map_err(AsyncIoError::WriteZeroes)?;
self.completion_list.push_back((user_data, 0));
self.eventfd.write(1).unwrap();
Ok(())
}
}
#[cfg(test)]
mod unit_tests {
use std::os::unix::io::AsRawFd;
use vmm_sys_util::tempfile::TempFile;
use super::*;
use crate::raw_async_io_tests;
#[test]
fn test_punch_hole() {
let temp_file = TempFile::new().unwrap();
let mut file = temp_file.into_file();
let mut async_io = RawFileAsyncAio::new(file.as_raw_fd(), 128).unwrap();
raw_async_io_tests::test_punch_hole(&mut async_io, &mut file);
}
#[test]
fn test_write_zeroes() {
let temp_file = TempFile::new().unwrap();
let mut file = temp_file.into_file();
let mut async_io = RawFileAsyncAio::new(file.as_raw_fd(), 128).unwrap();
raw_async_io_tests::test_write_zeroes(&mut async_io, &mut file);
}
#[test]
fn test_punch_hole_multiple_operations() {
let temp_file = TempFile::new().unwrap();
let mut file = temp_file.into_file();
let mut async_io = RawFileAsyncAio::new(file.as_raw_fd(), 128).unwrap();
raw_async_io_tests::test_punch_hole_multiple_operations(&mut async_io, &mut file);
}
}