Files
cloud-hypervisor/block/src/raw_async_aio.rs
Saravanan D 00c05f4761 block: Use logical block size for alignment
O_DIRECT requires buffer addresses to be aligned to the backend
device's logical block size. The existing bounce buffer logic in
execute_async() hardcodes SECTOR_SIZE (512) for the alignment check
and bounce buffer allocation. This is insufficient for devices with
a 4096-byte logical block size, where misaligned buffers cause
-EINVAL from the host kernel.

Add an alignment() method to the AsyncIo trait that returns the
backend's logical block size, defaulting to SECTOR_SIZE. The three
raw I/O backends (io_uring, AIO, synchronous) probe the device
topology via DiskTopology::probe() at creation time and return the
actual logical block size. All image format backends would simply
use the default value of 512 bytes since their underlying are
not block devices.

execute_async() now queries disk_image.alignment() instead of using
the hardcoded SECTOR_SIZE

Fixes: #7720

Signed-off-by: Saravanan D <saravanand@crusoe.ai>
2026-02-26 15:47:17 +00:00

193 lines
5.4 KiB
Rust

// Copyright © 2023 Intel Corporation
//
// SPDX-License-Identifier: Apache-2.0 AND BSD-3-Clause
//
// Copyright © 2023 Crusoe Energy Systems LLC
//
use std::fs::File;
use std::io::{Seek, SeekFrom};
use std::os::unix::io::{AsRawFd, RawFd};
use log::warn;
use vmm_sys_util::aio;
use vmm_sys_util::eventfd::EventFd;
use crate::async_io::{
AsyncIo, AsyncIoError, AsyncIoResult, BorrowedDiskFd, DiskFile, DiskFileError, DiskFileResult,
};
use crate::{DiskTopology, SECTOR_SIZE, probe_sparse_support};
pub struct RawFileDiskAio {
file: File,
}
impl RawFileDiskAio {
pub fn new(file: File) -> Self {
RawFileDiskAio { file }
}
}
impl DiskFile for RawFileDiskAio {
fn logical_size(&mut self) -> DiskFileResult<u64> {
self.file
.seek(SeekFrom::End(0))
.map_err(DiskFileError::Size)
}
fn physical_size(&mut self) -> DiskFileResult<u64> {
self.file
.metadata()
.map(|m| m.len())
.map_err(DiskFileError::Size)
}
fn new_async_io(&self, ring_depth: u32) -> DiskFileResult<Box<dyn AsyncIo>> {
let mut raw = RawFileAsyncAio::new(self.file.as_raw_fd(), ring_depth)
.map_err(DiskFileError::NewAsyncIo)?;
raw.alignment = DiskTopology::probe(&self.file)
.map(|t| t.logical_block_size)
.unwrap_or(SECTOR_SIZE);
Ok(Box::new(raw) as Box<dyn AsyncIo>)
}
fn topology(&mut self) -> DiskTopology {
if let Ok(topology) = DiskTopology::probe(&self.file) {
topology
} else {
warn!("Unable to get device topology. Using default topology");
DiskTopology::default()
}
}
fn supports_sparse_operations(&self) -> bool {
probe_sparse_support(&self.file)
}
fn fd(&mut self) -> BorrowedDiskFd<'_> {
BorrowedDiskFd::new(self.file.as_raw_fd())
}
}
pub struct RawFileAsyncAio {
fd: RawFd,
ctx: aio::IoContext,
eventfd: EventFd,
alignment: u64,
}
impl RawFileAsyncAio {
pub fn new(fd: RawFd, queue_depth: u32) -> std::io::Result<Self> {
let eventfd = EventFd::new(libc::EFD_NONBLOCK)?;
let ctx = aio::IoContext::new(queue_depth)?;
Ok(RawFileAsyncAio {
fd,
ctx,
eventfd,
alignment: SECTOR_SIZE,
})
}
}
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)> {
let mut events: [aio::IoEvent; 1] = [aio::IoEvent::default()];
let rc = self.ctx.get_events(0, &mut events, None).unwrap();
if rc == 0 {
None
} else {
Some((events[0].data, events[0].res as i32))
}
}
fn punch_hole(&mut self, _offset: u64, _length: u64, _user_data: u64) -> AsyncIoResult<()> {
Err(AsyncIoError::PunchHole(std::io::Error::other(
"punch_hole not supported with AIO backend",
)))
}
fn write_zeroes(&mut self, _offset: u64, _length: u64, _user_data: u64) -> AsyncIoResult<()> {
Err(AsyncIoError::WriteZeroes(std::io::Error::other(
"write_zeroes not supported with AIO backend",
)))
}
}