device_virtio_block/

asynchronous.rs

// Copyright 2021 The ChromiumOS Authors
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

use std::cell::RefCell;
use std::collections::BTreeMap;
use std::collections::BTreeSet;
use std::io;
use std::io::Write;
use std::mem::size_of;
use std::rc::Rc;
use std::result;
use std::sync::atomic::AtomicU64;
use std::sync::atomic::Ordering;
use std::sync::Arc;
use std::time::Duration;

use anyhow::Context;
use base::debug;
use base::error;
use base::info;
use base::warn;
use base::AsRawDescriptor;
use base::Error as SysError;
use base::Event;
use base::RawDescriptor;
use base::Result as SysResult;
use base::Timer;
use base::Tube;
use base::TubeError;
use base::WorkerThread;
use cros_async::sync::RwLock as AsyncRwLock;
use cros_async::AsyncError;
use cros_async::AsyncTube;
use cros_async::EventAsync;
use cros_async::Executor;
use cros_async::ExecutorKind;
use cros_async::IoOptions;
use cros_async::TimerAsync;
use data_model::Le16;
use data_model::Le32;
use data_model::Le64;
use devices::virtio::async_utils;
use devices::virtio::copy_config;
use devices::virtio::device_constants::block::virtio_blk_config;
use devices::virtio::device_constants::block::virtio_blk_discard_write_zeroes;
use devices::virtio::device_constants::block::virtio_blk_req_header;
use devices::virtio::device_constants::block::VIRTIO_BLK_DISCARD_WRITE_ZEROES_FLAG_UNMAP;
use devices::virtio::device_constants::block::VIRTIO_BLK_F_BLK_SIZE;
use devices::virtio::device_constants::block::VIRTIO_BLK_F_DISCARD;
use devices::virtio::device_constants::block::VIRTIO_BLK_F_FLUSH;
use devices::virtio::device_constants::block::VIRTIO_BLK_F_MQ;
use devices::virtio::device_constants::block::VIRTIO_BLK_F_RO;
use devices::virtio::device_constants::block::VIRTIO_BLK_F_SEG_MAX;
use devices::virtio::device_constants::block::VIRTIO_BLK_F_WRITE_ZEROES;
use devices::virtio::device_constants::block::VIRTIO_BLK_S_IOERR;
use devices::virtio::device_constants::block::VIRTIO_BLK_S_OK;
use devices::virtio::device_constants::block::VIRTIO_BLK_S_UNSUPP;
use devices::virtio::device_constants::block::VIRTIO_BLK_T_DISCARD;
use devices::virtio::device_constants::block::VIRTIO_BLK_T_FLUSH;
use devices::virtio::device_constants::block::VIRTIO_BLK_T_GET_ID;
use devices::virtio::device_constants::block::VIRTIO_BLK_T_IN;
use devices::virtio::device_constants::block::VIRTIO_BLK_T_OUT;
use devices::virtio::device_constants::block::VIRTIO_BLK_T_WRITE_ZEROES;
use devices::virtio::DescriptorChain;
use devices::virtio::DeviceType;
use devices::virtio::Interrupt;
use devices::virtio::Queue;
use devices::virtio::Reader;
use devices::virtio::VirtioDevice;
use devices::virtio::Writer;
use devices::PciAddress;
use disk::AsyncDisk;
use disk::DiskFile;
use futures::channel::mpsc;
use futures::channel::oneshot;
use futures::pin_mut;
use futures::stream::FuturesUnordered;
use futures::stream::StreamExt;
use futures::FutureExt;
use remain::sorted;
use snapshot::AnySnapshot;
use thiserror::Error as ThisError;
use virtio_sys::virtio_config::VIRTIO_F_RING_PACKED;
use vm_control::DiskControlCommand;
use vm_control::DiskControlResult;
use vm_memory::GuestMemory;
use zerocopy::IntoBytes;

use crate::interrupt_coalescer::InterruptCoalescer;
use crate::sys::*;
use crate::DiskOption;

const DEFAULT_QUEUE_SIZE: u16 = 256;
const DEFAULT_NUM_QUEUES: u16 = 16;

const SECTOR_SHIFT: u8 = 9;
const SECTOR_SIZE: u64 = 0x01 << SECTOR_SHIFT;

const MAX_DISCARD_SECTORS: u32 = u32::MAX;
const MAX_WRITE_ZEROES_SECTORS: u32 = u32::MAX;
// Arbitrary limits for number of discard/write zeroes segments.
const MAX_DISCARD_SEG: u32 = 32;
const MAX_WRITE_ZEROES_SEG: u32 = 32;
// Hard-coded to 64 KiB (in 512-byte sectors) for now,
// but this should probably be based on cluster size for qcow.
const DISCARD_SECTOR_ALIGNMENT: u32 = 128;

#[sorted]
#[derive(ThisError, Debug)]
enum ExecuteError {
    #[error("failed to copy ID string: {0}")]
    CopyId(io::Error),
    #[error("failed to perform discard or write zeroes; sector={sector} num_sectors={num_sectors} flags={flags}; {ioerr:?}")]
    DiscardWriteZeroes {
        ioerr: Option<disk::Error>,
        sector: u64,
        num_sectors: u32,
        flags: u32,
    },
    #[error("failed to flush: {0}")]
    Flush(disk::Error),
    #[error("not enough space in descriptor chain to write status")]
    MissingStatus,
    #[error("out of range")]
    OutOfRange,
    #[error("failed to read message: {0}")]
    Read(io::Error),
    #[error("io error reading {length} bytes from sector {sector}: {desc_error}")]
    ReadIo {
        length: usize,
        sector: u64,
        desc_error: disk::Error,
    },
    #[error("read only; request_type={request_type}")]
    ReadOnly { request_type: u32 },
    #[error("failed to recieve command message: {0}")]
    ReceivingCommand(TubeError),
    #[error("failed to send command response: {0}")]
    SendingResponse(TubeError),
    #[error("couldn't reset the timer: {0}")]
    TimerReset(base::Error),
    #[error("too many segments: {0} > {0}")]
    TooManySegments(usize, usize),
    #[error("unsupported ({0})")]
    Unsupported(u32),
    #[error("io error writing {length} bytes from sector {sector}: {desc_error}")]
    WriteIo {
        length: usize,
        sector: u64,
        desc_error: disk::Error,
    },
    #[error("failed to write request status: {0}")]
    WriteStatus(io::Error),
}

enum LogLevel {
    Debug,
    Error,
}

impl ExecuteError {
    fn status(&self) -> u8 {
        match self {
            ExecuteError::CopyId(_) => VIRTIO_BLK_S_IOERR,
            ExecuteError::DiscardWriteZeroes { .. } => VIRTIO_BLK_S_IOERR,
            ExecuteError::Flush(_) => VIRTIO_BLK_S_IOERR,
            ExecuteError::MissingStatus => VIRTIO_BLK_S_IOERR,
            ExecuteError::OutOfRange => VIRTIO_BLK_S_IOERR,
            ExecuteError::Read(_) => VIRTIO_BLK_S_IOERR,
            ExecuteError::ReadIo { .. } => VIRTIO_BLK_S_IOERR,
            ExecuteError::ReadOnly { .. } => VIRTIO_BLK_S_IOERR,
            ExecuteError::ReceivingCommand(_) => VIRTIO_BLK_S_IOERR,
            ExecuteError::SendingResponse(_) => VIRTIO_BLK_S_IOERR,
            ExecuteError::TimerReset(_) => VIRTIO_BLK_S_IOERR,
            ExecuteError::TooManySegments(_, _) => VIRTIO_BLK_S_IOERR,
            ExecuteError::WriteIo { .. } => VIRTIO_BLK_S_IOERR,
            ExecuteError::WriteStatus(_) => VIRTIO_BLK_S_IOERR,
            ExecuteError::Unsupported(_) => VIRTIO_BLK_S_UNSUPP,
        }
    }

    fn log_level(&self) -> LogLevel {
        match self {
            // Log disk I/O errors at debug level to avoid flooding the logs.
            ExecuteError::ReadIo { .. }
            | ExecuteError::WriteIo { .. }
            | ExecuteError::Flush { .. }
            | ExecuteError::DiscardWriteZeroes { .. } => LogLevel::Debug,
            // Log all other failures as errors.
            _ => LogLevel::Error,
        }
    }
}

/// Errors that happen in block outside of executing a request.
/// This includes errors during resize and flush operations.
#[sorted]
#[derive(ThisError, Debug)]
enum ControlError {
    #[error("failed to fdatasync the disk: {0}")]
    FdatasyncDisk(disk::Error),
    #[error("couldn't get a value from a timer for flushing: {0}")]
    FlushTimer(AsyncError),
}

/// Maximum length of the virtio-block ID string field.
const ID_LEN: usize = 20;

/// Virtio block device identifier.
/// This is an ASCII string terminated by a \0, unless all 20 bytes are used,
/// in which case the \0 terminator is omitted.
type BlockId = [u8; ID_LEN];

/// Tracks the state of an anynchronous disk.
struct DiskState {
    disk_image: Box<dyn AsyncDisk>,
    read_only: bool,
    sparse: bool,
    id: BlockId,
    dontcache_read: bool,
    dontcache_write: bool,
    /// A DiskState is owned by each worker's executor and cannot be shared by workers, thus
    /// `worker_shared_state` holds the state shared by workers in Arc.
    worker_shared_state: Arc<AsyncRwLock<WorkerSharedState>>,
}

/// Disk state which can be modified by other worker threads
struct WorkerSharedState {
    disk_size: Arc<AtomicU64>,
}

async fn process_one_request(
    avail_desc: &mut DescriptorChain,
    disk_state: &AsyncRwLock<DiskState>,
    flush_timer: &RefCell<TimerAsync<Timer>>,
    flush_timer_armed: &RefCell<bool>,
) -> result::Result<usize, ExecuteError> {
    let reader = &mut avail_desc.reader;
    let writer = &mut avail_desc.writer;

    // The last byte of the buffer is virtio_blk_req::status.
    // Split it into a separate Writer so that status_writer is the final byte and
    // the original writer is left with just the actual block I/O data.
    let available_bytes = writer.available_bytes();
    let status_offset = available_bytes
        .checked_sub(1)
        .ok_or(ExecuteError::MissingStatus)?;
    let mut status_writer = writer.split_at(status_offset);

    let status = match BlockAsync::execute_request(
        reader,
        writer,
        disk_state,
        flush_timer,
        flush_timer_armed,
    )
    .await
    {
        Ok(()) => VIRTIO_BLK_S_OK,
        Err(e) => {
            match e.log_level() {
                LogLevel::Debug => debug!("failed executing disk request: {:#}", e),
                LogLevel::Error => error!("failed executing disk request: {:#}", e),
            }
            e.status()
        }
    };

    status_writer
        .write_all(&[status])
        .map_err(ExecuteError::WriteStatus)?;
    Ok(available_bytes)
}

/// Process one descriptor chain asynchronously.
async fn process_one_chain(
    queue: &RefCell<Queue>,
    mut avail_desc: DescriptorChain,
    disk_state: &AsyncRwLock<DiskState>,
    flush_timer: &RefCell<TimerAsync<Timer>>,
    flush_timer_armed: &RefCell<bool>,
    irq_coalescer: &InterruptCoalescer,
) {
    let len = match process_one_request(&mut avail_desc, disk_state, flush_timer, flush_timer_armed)
        .await
    {
        Ok(len) => len,
        Err(e) => {
            error!("block: failed to handle request: {:#}", e);
            0
        }
    };

    let mut queue = queue.borrow_mut();
    queue.add_used_with_bytes_written(avail_desc, len as u32);
    irq_coalescer.trigger_interrupt(|| {
        queue.trigger_interrupt();
    });
}

// There is one async task running `handle_queue` per virtio queue in use.
// Receives messages from the guest and queues a task to complete the operations with the async
// executor.
async fn handle_queue(
    disk_state: Rc<AsyncRwLock<DiskState>>,
    queue: Queue,
    evt: EventAsync,
    flush_timer: Rc<RefCell<TimerAsync<Timer>>>,
    flush_timer_armed: Rc<RefCell<bool>>,
    mut stop_rx: oneshot::Receiver<()>,
) -> Queue {
    let queue = RefCell::new(queue);
    let irq_coalescer = InterruptCoalescer::new();
    let mut background_tasks = FuturesUnordered::new();
    let evt_future = futures::future::Either::Left(std::future::ready(Ok(0))).fuse();
    pin_mut!(evt_future);
    loop {
        // Wait for the next signal from `evt` and process `background_tasks` in the meantime.
        // Requests that are ready are finished first and then their batch is ended. The stop
        // channel is checked before the queue event, so a guest that keeps kicking can't hold off a
        // stop.
        //
        // Interrupt the first and last completion. The first wakes the guest so it starts consuming
        // completions, and signalling only the last avoids the cost of a wake for each in the
        // batch.
        //
        // NOTE: We can't call `evt.next_val()` directly in the `select_biased!` expression. That
        // would create a new future each time, which, in the completion-based async backends like
        // io_uring, means we'd submit a new syscall each time (i.e. a race condition on the
        // eventfd).
        futures::select_biased! {
            _ = background_tasks.next() => continue,
            _ = irq_coalescer.active() => {
                irq_coalescer.complete(|| {
                    queue.borrow_mut().trigger_interrupt();
                });
                continue;
            }
            _ = stop_rx => {
                // Process all the descriptors we've already popped from the queue so that we leave
                // the queue in a consistent state.
                background_tasks.collect::<()>().await;
                irq_coalescer.complete(|| {
                    queue.borrow_mut().trigger_interrupt();
                });
                return queue.into_inner();
            }
            res = evt_future => {
                evt_future.set(futures::future::Either::Right(evt.next_val()).fuse());
                if let Err(e) = res {
                    error!("Failed to read the next queue event: {:#}", e);
                    continue;
                }
            }
        };
        while let Some(descriptor_chain) = queue.borrow_mut().pop() {
            background_tasks.push(process_one_chain(
                &queue,
                descriptor_chain,
                &disk_state,
                &flush_timer,
                &flush_timer_armed,
                &irq_coalescer,
            ));
        }
    }
}

async fn handle_command_tube(
    command_tube: &Option<AsyncTube>,
    interrupt: &RefCell<Option<Interrupt>>,
    disk_state: Rc<AsyncRwLock<DiskState>>,
) -> Result<(), ExecuteError> {
    let command_tube = match command_tube {
        Some(c) => c,
        None => {
            futures::future::pending::<()>().await;
            return Ok(());
        }
    };
    loop {
        match command_tube.next().await {
            Ok(command) => {
                let resp = match command {
                    DiskControlCommand::Resize { new_size } => resize(&disk_state, new_size).await,
                };

                let resp_clone = resp.clone();
                command_tube
                    .send(resp_clone)
                    .await
                    .map_err(ExecuteError::SendingResponse)?;
                if let DiskControlResult::Ok = resp {
                    if let Some(interrupt) = &*interrupt.borrow() {
                        interrupt.signal_config_changed();
                    }
                }
            }
            Err(e) => return Err(ExecuteError::ReceivingCommand(e)),
        }
    }
}

async fn resize(disk_state: &AsyncRwLock<DiskState>, new_size: u64) -> DiskControlResult {
    // Acquire exclusive, mutable access to the state so the virtqueue task won't be able to read
    // the state while resizing.
    let disk_state = disk_state.lock().await;
    // Prevent any other worker threads won't be able to do IO.
    let worker_shared_state = Arc::clone(&disk_state.worker_shared_state);
    let worker_shared_state = worker_shared_state.lock().await;

    if disk_state.read_only {
        error!("Attempted to resize read-only block device");
        return DiskControlResult::Err(SysError::new(libc::EROFS));
    }

    info!("Resizing block device to {} bytes", new_size);

    if let Err(e) = disk_state.disk_image.set_len(new_size) {
        error!("Resizing disk failed! {:#}", e);
        return DiskControlResult::Err(SysError::new(libc::EIO));
    }

    // Allocate new space if the disk image is not sparse.
    if !disk_state.sparse {
        if let Err(e) = disk_state.disk_image.allocate(0, new_size) {
            error!("Allocating disk space after resize failed! {:#}", e);
            return DiskControlResult::Err(SysError::new(libc::EIO));
        }
    }

    if let Ok(new_disk_size) = disk_state.disk_image.get_len() {
        worker_shared_state
            .disk_size
            .store(new_disk_size, Ordering::Release);
    }
    DiskControlResult::Ok
}

/// Periodically flushes the disk when the given timer fires.
async fn flush_disk(
    disk_state: Rc<AsyncRwLock<DiskState>>,
    timer: TimerAsync<Timer>,
    armed: Rc<RefCell<bool>>,
) -> Result<(), ControlError> {
    loop {
        timer.wait().await.map_err(ControlError::FlushTimer)?;
        if !*armed.borrow() {
            continue;
        }

        // Reset armed before calling fdatasync to guarantee that IO requests that started after we
        // call fdatasync will be committed eventually.
        *armed.borrow_mut() = false;

        disk_state
            .read_lock()
            .await
            .disk_image
            .fdatasync()
            .await
            .map_err(ControlError::FdatasyncDisk)?;
    }
}

enum WorkerCmd {
    StartQueue {
        index: usize,
        queue: Queue,
    },
    StopQueue {
        index: usize,
        // Once the queue is stopped, it will be sent back over `response_tx`.
        // `None` indicates that there was no queue at the given index.
        response_tx: oneshot::Sender<Option<Queue>>,
    },
    // Stop all queues without recovering the queues' state and without completing any queued up
    // work .
    AbortQueues {
        // Once the queues are stopped, a `()` value will be sent back over `response_tx`.
        response_tx: oneshot::Sender<()>,
    },
    // Flush any in-memory disk image state to file.
    FlushDisk {
        // Once the disk is flushed, a `()` value will be sent back over `response_tx`.
        response_tx: oneshot::Sender<()>,
    },
}

// The main worker thread. Initialized the asynchronous worker tasks and passes them to the executor
// to be processed.
//
// `disk_state` is wrapped by `AsyncRwLock`, which provides both shared and exclusive locks. It's
// because the state can be read from the virtqueue task while the control task is processing a
// resizing command.
async fn run_worker(
    ex: &Executor,
    disk_state: &Rc<AsyncRwLock<DiskState>>,
    control_tube: &Option<AsyncTube>,
    mut worker_rx: mpsc::UnboundedReceiver<WorkerCmd>,
    kill_evt: Event,
) -> anyhow::Result<()> {
    // One flush timer per disk.
    let timer = Timer::new().expect("Failed to create a timer");
    let flush_timer_armed = Rc::new(RefCell::new(false));

    // Handles control requests.
    let control_interrupt = RefCell::new(None);
    let control = handle_command_tube(control_tube, &control_interrupt, disk_state.clone()).fuse();
    pin_mut!(control);

    // Handle all the queues in one sub-select call.
    let flush_timer = Rc::new(RefCell::new(
        TimerAsync::new(
            // Call try_clone() to share the same underlying FD with the `flush_disk` task.
            timer.try_clone().expect("Failed to clone flush_timer"),
            ex,
        )
        .expect("Failed to create an async timer"),
    ));

    // Flushes the disk periodically.
    let flush_timer2 = TimerAsync::new(timer, ex).expect("Failed to create an async timer");
    let disk_flush = flush_disk(disk_state.clone(), flush_timer2, flush_timer_armed.clone()).fuse();
    pin_mut!(disk_flush);

    // Exit if the kill event is triggered.
    let kill = async_utils::await_and_exit(ex, kill_evt).fuse();
    pin_mut!(kill);

    // Running queue handlers.
    let mut queue_handlers = FuturesUnordered::new();
    // Async stop functions for queue handlers, by queue index.
    let mut queue_handler_stop_fns = std::collections::BTreeMap::new();

    loop {
        futures::select! {
            _ = queue_handlers.next() => continue,
            r = disk_flush => return r.context("failed to flush a disk"),
            r = control => return r.context("failed to handle a control request"),
            r = kill => return r.context("failed to wait on the kill event"),
            worker_cmd = worker_rx.next() => {
                match worker_cmd {
                    None => anyhow::bail!("worker control channel unexpectedly closed"),
                    Some(WorkerCmd::StartQueue{index, queue}) => {
                        if control_interrupt.borrow().is_none() {
                            *control_interrupt.borrow_mut() = Some(queue.interrupt().clone());
                        }

                        let (tx, rx) = oneshot::channel();
                        let kick_evt = queue.event().try_clone().expect("Failed to clone queue event");
                        let (handle_queue_future, remote_handle) = handle_queue(
                            Rc::clone(disk_state),
                            queue,
                            EventAsync::new(kick_evt, ex).expect("Failed to create async event for queue"),
                            Rc::clone(&flush_timer),
                            Rc::clone(&flush_timer_armed),
                            rx,
                        ).remote_handle();
                        let old_stop_fn = queue_handler_stop_fns.insert(index, move || {
                            // Ask the handler to stop.
                            tx.send(()).unwrap_or_else(|_| panic!("queue handler channel closed early"));
                            // Wait for its return value.
                            remote_handle
                        });

                        // If there was already a handler for this index, stop it before adding the
                        // new handler future.
                        if let Some(stop_fn) = old_stop_fn {
                            warn!("Starting new queue handler without stopping old handler");
                            // Unfortunately we can't just do `stop_fn().await` because the actual
                            // work we are waiting on is in `queue_handlers`. So, run both.
                            let mut fut = stop_fn().fuse();
                            loop {
                                futures::select! {
                                    _ = queue_handlers.next() => continue,
                                    _queue = fut => break,
                                }
                            }
                        }

                        queue_handlers.push(handle_queue_future);
                    }
                    Some(WorkerCmd::StopQueue{index, response_tx}) => {
                        match queue_handler_stop_fns.remove(&index) {
                            Some(stop_fn) => {
                                // NOTE: This await is blocking the select loop. If we want to
                                // support stopping queues concurrently, then it needs to be moved.
                                // For now, keep it simple.
                                //
                                // Unfortunately we can't just do `stop_fn().await` because the
                                // actual work we are waiting on is in `queue_handlers`. So, run
                                // both.
                                let mut fut = stop_fn().fuse();
                                let queue = loop {
                                    futures::select! {
                                        _ = queue_handlers.next() => continue,
                                        queue = fut => break queue,
                                    }
                                };

                                // If this is the last queue, drop references to the interrupt so
                                // that, when queues are started up again, we'll use the new
                                // interrupt passed with the first queue.
                                if queue_handlers.is_empty() {
                                    *control_interrupt.borrow_mut() = None;
                                }

                                let _ = response_tx.send(Some(queue));
                            }
                            None => { let _ = response_tx.send(None); },
                        }

                    }
                    Some(WorkerCmd::AbortQueues{response_tx}) => {
                        queue_handlers.clear();
                        queue_handler_stop_fns.clear();

                        *control_interrupt.borrow_mut() = None;

                        let _ = response_tx.send(());
                    }
                    Some(WorkerCmd::FlushDisk{response_tx}) => {
                        let disk = disk_state.read_lock().await;
                        if let Err(e) = disk.disk_image.flush().await {
                            error!("failed to flush disk image: {:#}", e);
                        }
                        let _ = response_tx.send(());
                    }
                }
            }
        };
    }
}

/// Virtio device for exposing block level read/write operations on a host file.
pub struct BlockAsync {
    // We need to make boot_index public bc the field is used by the main crate to determine boot
    // order
    boot_index: Option<usize>,
    // `None` iff `self.worker_per_queue == false` and the worker thread is running.
    disk_image: Option<Box<dyn DiskFile>>,
    disk_size: Arc<AtomicU64>,
    avail_features: u64,
    read_only: bool,
    sparse: bool,
    seg_max: u32,
    block_size: u32,
    id: BlockId,
    control_tube: Option<Tube>,
    queue_sizes: Vec<u16>,
    pub(super) executor_kind: ExecutorKind,
    // If `worker_per_queue == true`, `worker_threads` contains the worker for each running queue
    // by index. Otherwise, contains the monolithic worker for all queues at index 0.
    //
    // Once a thread is started, we never stop it, except when `BlockAsync` itself is dropped. That
    // is because we cannot easily convert the `AsyncDisk` back to a `DiskFile` when backed by
    // Overlapped I/O on Windows because the file becomes permanently associated with the IOCP
    // instance of the async executor.
    worker_threads: BTreeMap<usize, (WorkerThread<()>, mpsc::UnboundedSender<WorkerCmd>)>,
    shared_state: Arc<AsyncRwLock<WorkerSharedState>>,
    // Whether to run worker threads in parallel for each queue
    worker_per_queue: bool,
    // Indices of running queues.
    // TODO: The worker already tracks this. Only need it here to stop queues on sleep. Maybe add a
    // worker cmd to stop all at once, then we can delete this field.
    activated_queues: BTreeSet<usize>,
    #[cfg(windows)]
    pub(super) io_concurrency: u32,
    pci_address: Option<PciAddress>,
    dontcache_read: bool,
    dontcache_write: bool,
}

impl BlockAsync {
    /// Create a new virtio block device that operates on the given AsyncDisk.
    pub fn new(
        base_features: u64,
        disk_image: Box<dyn DiskFile>,
        disk_option: &DiskOption,
        control_tube: Option<Tube>,
        queue_size: Option<u16>,
        num_queues: Option<u16>,
    ) -> SysResult<BlockAsync> {
        let read_only = disk_option.read_only;
        let sparse = disk_option.sparse;
        let block_size = disk_option.block_size;
        let packed_queue = disk_option.packed_queue;
        let id = disk_option.id.unwrap_or_default();
        let mut worker_per_queue = disk_option.multiple_workers;
        // Automatically disable multiple workers if the disk image can't be cloned.
        if worker_per_queue && disk_image.try_clone().is_err() {
            base::warn!("multiple workers requested, but not supported by disk image type");
            worker_per_queue = false;
        }
        let executor_kind = disk_option.async_executor.unwrap_or_default();
        let boot_index = disk_option.bootindex;
        #[cfg(windows)]
        let io_concurrency = disk_option.io_concurrency.get();

        if block_size % SECTOR_SIZE as u32 != 0 {
            error!(
                "Block size {} is not a multiple of {}.",
                block_size, SECTOR_SIZE,
            );
            return Err(SysError::new(libc::EINVAL));
        }
        let disk_size = disk_image.get_len()?;
        if disk_size % block_size as u64 != 0 {
            warn!(
                "Disk size {} is not a multiple of block size {}; \
                 the remainder will not be visible to the guest.",
                disk_size, block_size,
            );
        }
        let num_queues = num_queues.unwrap_or(DEFAULT_NUM_QUEUES);
        let multi_queue = match num_queues {
            0 => panic!("Number of queues cannot be zero for a block device"),
            1 => false,
            _ => true,
        };
        let q_size = queue_size.unwrap_or(DEFAULT_QUEUE_SIZE);
        if !q_size.is_power_of_two() {
            error!("queue size {} is not a power of 2.", q_size);
            return Err(SysError::new(libc::EINVAL));
        }
        let queue_sizes = vec![q_size; num_queues as usize];

        let avail_features =
            Self::build_avail_features(base_features, read_only, sparse, multi_queue, packed_queue);

        let seg_max = get_seg_max(q_size);

        let disk_size = Arc::new(AtomicU64::new(disk_size));
        let shared_state = Arc::new(AsyncRwLock::new(WorkerSharedState {
            disk_size: disk_size.clone(),
        }));

        let mut dontcache_read = disk_option.dontcache;
        let mut dontcache_write = disk_option.dontcache;
        if dontcache_write {
            let descriptors = disk_image.as_raw_descriptors();
            if descriptors.is_empty()
                || !descriptors
                    .iter()
                    .all(|&fd| check_dontcache_support(fd, true /* write */))
            {
                base::info!(
                    "dontcache requested, but not supported for writes by backing filesystem; falling
                    back to cached I/O"
                );
                dontcache_write = false;
            }
        }
        if dontcache_read {
            let descriptors = disk_image.as_raw_descriptors();
            if descriptors.is_empty()
                || !descriptors
                    .iter()
                    .all(|&fd| check_dontcache_support(fd, false /* write */))
            {
                base::info!(
                    "dontcache requested, but not supported for reads by backing filesystem; falling
                    back to cached I/O"
                );
                dontcache_read = false;
            }
        }

        Ok(BlockAsync {
            disk_image: Some(disk_image),
            disk_size,
            avail_features,
            read_only,
            sparse,
            seg_max,
            block_size,
            id,
            queue_sizes,
            worker_threads: BTreeMap::new(),
            shared_state,
            worker_per_queue,
            control_tube,
            executor_kind,
            activated_queues: BTreeSet::new(),
            boot_index,
            #[cfg(windows)]
            io_concurrency,
            pci_address: disk_option.pci_address,
            dontcache_read,
            dontcache_write,
        })
    }

    /// Returns the feature flags given the specified attributes.
    fn build_avail_features(
        base_features: u64,
        read_only: bool,
        sparse: bool,
        multi_queue: bool,
        packed_queue: bool,
    ) -> u64 {
        let mut avail_features = base_features;
        if read_only {
            avail_features |= 1 << VIRTIO_BLK_F_RO;
        } else {
            if sparse {
                avail_features |= 1 << VIRTIO_BLK_F_DISCARD;
            }
            avail_features |= 1 << VIRTIO_BLK_F_FLUSH;
            avail_features |= 1 << VIRTIO_BLK_F_WRITE_ZEROES;
        }
        avail_features |= 1 << VIRTIO_BLK_F_SEG_MAX;
        avail_features |= 1 << VIRTIO_BLK_F_BLK_SIZE;
        if multi_queue {
            avail_features |= 1 << VIRTIO_BLK_F_MQ;
        }
        if packed_queue {
            avail_features |= 1 << VIRTIO_F_RING_PACKED;
        }
        avail_features
    }

    // Execute a single block device request.
    // `writer` includes the data region only; the status byte is not included.
    // It is up to the caller to convert the result of this function into a status byte
    // and write it to the expected location in guest memory.
    async fn execute_request(
        reader: &mut Reader,
        writer: &mut Writer,
        disk_state: &AsyncRwLock<DiskState>,
        flush_timer: &RefCell<TimerAsync<Timer>>,
        flush_timer_armed: &RefCell<bool>,
    ) -> result::Result<(), ExecuteError> {
        // Acquire immutable access to prevent tasks from resizing disk.
        let disk_state = disk_state.read_lock().await;
        // Acquire immutable access to prevent other worker threads from resizing disk.
        let worker_shared_state = disk_state.worker_shared_state.read_lock().await;

        let req_header: virtio_blk_req_header = reader.read_obj().map_err(ExecuteError::Read)?;

        let req_type = req_header.req_type.to_native();
        let sector = req_header.sector.to_native();

        if disk_state.read_only && req_type != VIRTIO_BLK_T_IN && req_type != VIRTIO_BLK_T_GET_ID {
            return Err(ExecuteError::ReadOnly {
                request_type: req_type,
            });
        }

        /// Check that a request accesses only data within the disk's current size.
        /// All parameters are in units of bytes.
        fn check_range(
            io_start: u64,
            io_length: u64,
            disk_size: u64,
        ) -> result::Result<(), ExecuteError> {
            let io_end = io_start
                .checked_add(io_length)
                .ok_or(ExecuteError::OutOfRange)?;
            if io_end > disk_size {
                Err(ExecuteError::OutOfRange)
            } else {
                Ok(())
            }
        }

        let disk_size = worker_shared_state.disk_size.load(Ordering::Relaxed);
        match req_type {
            VIRTIO_BLK_T_IN => {
                let data_len = writer.available_bytes();
                if data_len == 0 {
                    return Ok(());
                }
                let offset = sector
                    .checked_shl(u32::from(SECTOR_SHIFT))
                    .ok_or(ExecuteError::OutOfRange)?;
                check_range(offset, data_len as u64, disk_size)?;
                let disk_image = &disk_state.disk_image;
                writer
                    .write_all_from_at_fut(
                        &**disk_image,
                        data_len,
                        offset,
                        IoOptions {
                            dontcache: disk_state.dontcache_read,
                        },
                    )
                    .await
                    .map_err(|desc_error| ExecuteError::ReadIo {
                        length: data_len,
                        sector,
                        desc_error,
                    })?;
            }
            VIRTIO_BLK_T_OUT => {
                let data_len = reader.available_bytes();
                if data_len == 0 {
                    return Ok(());
                }
                let offset = sector
                    .checked_shl(u32::from(SECTOR_SHIFT))
                    .ok_or(ExecuteError::OutOfRange)?;
                check_range(offset, data_len as u64, disk_size)?;
                let disk_image = &disk_state.disk_image;
                reader
                    .read_exact_to_at_fut(
                        &**disk_image,
                        data_len,
                        offset,
                        IoOptions {
                            dontcache: disk_state.dontcache_write,
                        },
                    )
                    .await
                    .map_err(|desc_error| ExecuteError::WriteIo {
                        length: data_len,
                        sector,
                        desc_error,
                    })?;

                if !*flush_timer_armed.borrow() {
                    *flush_timer_armed.borrow_mut() = true;

                    let flush_delay = Duration::from_secs(60);
                    flush_timer
                        .borrow_mut()
                        .reset_oneshot(flush_delay)
                        .map_err(ExecuteError::TimerReset)?;
                }
            }
            VIRTIO_BLK_T_DISCARD | VIRTIO_BLK_T_WRITE_ZEROES => {
                if req_type == VIRTIO_BLK_T_DISCARD && !disk_state.sparse {
                    // Discard is a hint; if this is a non-sparse disk, just ignore it.
                    return Ok(());
                }

                let seg_count =
                    reader.available_bytes() / size_of::<virtio_blk_discard_write_zeroes>();
                let seg_max = if req_type == VIRTIO_BLK_T_DISCARD {
                    MAX_DISCARD_SEG as usize
                } else {
                    MAX_WRITE_ZEROES_SEG as usize
                };
                if seg_count > seg_max {
                    return Err(ExecuteError::TooManySegments(seg_count, seg_max));
                }

                while reader.available_bytes() >= size_of::<virtio_blk_discard_write_zeroes>() {
                    let seg: virtio_blk_discard_write_zeroes =
                        reader.read_obj().map_err(ExecuteError::Read)?;

                    let sector = seg.sector.to_native();
                    let num_sectors = seg.num_sectors.to_native();
                    let flags = seg.flags.to_native();

                    let valid_flags = if req_type == VIRTIO_BLK_T_WRITE_ZEROES {
                        VIRTIO_BLK_DISCARD_WRITE_ZEROES_FLAG_UNMAP
                    } else {
                        0
                    };

                    if (flags & !valid_flags) != 0 {
                        return Err(ExecuteError::DiscardWriteZeroes {
                            ioerr: None,
                            sector,
                            num_sectors,
                            flags,
                        });
                    }

                    let offset = sector
                        .checked_shl(u32::from(SECTOR_SHIFT))
                        .ok_or(ExecuteError::OutOfRange)?;
                    let length = u64::from(num_sectors)
                        .checked_shl(u32::from(SECTOR_SHIFT))
                        .ok_or(ExecuteError::OutOfRange)?;
                    check_range(offset, length, disk_size)?;

                    if req_type == VIRTIO_BLK_T_DISCARD {
                        // Since Discard is just a hint and some filesystems may not implement
                        // FALLOC_FL_PUNCH_HOLE, ignore punch_hole errors.
                        let _ = disk_state.disk_image.punch_hole(offset, length).await;
                    } else {
                        disk_state
                            .disk_image
                            .write_zeroes_at(offset, length)
                            .await
                            .map_err(|e| ExecuteError::DiscardWriteZeroes {
                                ioerr: Some(e),
                                sector,
                                num_sectors,
                                flags,
                            })?;
                    }
                }
            }
            VIRTIO_BLK_T_FLUSH => {
                disk_state
                    .disk_image
                    .fdatasync()
                    .await
                    .map_err(ExecuteError::Flush)?;

                if *flush_timer_armed.borrow() {
                    flush_timer
                        .borrow_mut()
                        .clear()
                        .map_err(ExecuteError::TimerReset)?;
                    *flush_timer_armed.borrow_mut() = false;
                }
            }
            VIRTIO_BLK_T_GET_ID => {
                writer
                    .write_all(&disk_state.id)
                    .map_err(ExecuteError::CopyId)?;
            }
            t => return Err(ExecuteError::Unsupported(t)),
        };
        Ok(())
    }

    /// Builds and returns the config structure used to specify block features.
    fn build_config_space(
        disk_size: u64,
        seg_max: u32,
        block_size: u32,
        num_queues: u16,
    ) -> virtio_blk_config {
        virtio_blk_config {
            // If the image is not a multiple of the sector size, the tail bits are not exposed.
            capacity: Le64::from(disk_size >> SECTOR_SHIFT),
            seg_max: Le32::from(seg_max),
            blk_size: Le32::from(block_size),
            num_queues: Le16::from(num_queues),
            max_discard_sectors: Le32::from(MAX_DISCARD_SECTORS),
            discard_sector_alignment: Le32::from(DISCARD_SECTOR_ALIGNMENT),
            max_write_zeroes_sectors: Le32::from(MAX_WRITE_ZEROES_SECTORS),
            write_zeroes_may_unmap: 1,
            max_discard_seg: Le32::from(MAX_DISCARD_SEG),
            max_write_zeroes_seg: Le32::from(MAX_WRITE_ZEROES_SEG),
            ..Default::default()
        }
    }

    /// Get the worker for a queue, starting it if necessary.
    // NOTE: Can't use `BTreeMap::entry` because it requires an exclusive ref for the whole branch.
    #[allow(clippy::map_entry)]
    fn start_worker(
        &mut self,
        idx: usize,
    ) -> anyhow::Result<&(WorkerThread<()>, mpsc::UnboundedSender<WorkerCmd>)> {
        let key = if self.worker_per_queue { idx } else { 0 };
        if self.worker_threads.contains_key(&key) {
            return Ok(self.worker_threads.get(&key).unwrap());
        }

        let ex = self.create_executor();
        let control_tube = self.control_tube.take();
        let dontcache_read = self.dontcache_read;
        let dontcache_write = self.dontcache_write;
        let disk_image = if self.worker_per_queue {
            self.disk_image
                .as_ref()
                .context("Failed to ref a disk image")?
                .try_clone()
                .context("Failed to clone a disk image")?
        } else {
            self.disk_image
                .take()
                .context("Failed to take a disk image")?
        };
        let read_only = self.read_only;
        let sparse = self.sparse;
        let id = self.id;
        let worker_shared_state = self.shared_state.clone();

        let (worker_tx, worker_rx) = mpsc::unbounded();
        let worker_thread = WorkerThread::start("virtio_blk", move |kill_evt| {
            let async_control =
                control_tube.map(|c| AsyncTube::new(&ex, c).expect("failed to create async tube"));

            let async_image = match disk_image.to_async_disk(&ex) {
                Ok(d) => d,
                Err(e) => panic!("Failed to create async disk {e:#}"),
            };

            let disk_state = Rc::new(AsyncRwLock::new(DiskState {
                disk_image: async_image,
                read_only,
                sparse,
                id,
                dontcache_read,
                dontcache_write,
                worker_shared_state,
            }));

            if let Err(err_string) = ex
                .run_until(async {
                    let r = run_worker(&ex, &disk_state, &async_control, worker_rx, kill_evt).await;
                    // Flush any in-memory disk image state to file.
                    if let Err(e) = disk_state.lock().await.disk_image.flush().await {
                        error!("failed to flush disk image when stopping worker: {e:?}");
                    }
                    r
                })
                .expect("run_until failed")
            {
                error!("{:#}", err_string);
            }
        });
        match self.worker_threads.entry(key) {
            std::collections::btree_map::Entry::Occupied(_) => unreachable!(),
            std::collections::btree_map::Entry::Vacant(e) => {
                Ok(e.insert((worker_thread, worker_tx)))
            }
        }
    }

    pub fn start_queue(
        &mut self,
        idx: usize,
        queue: Queue,
        _mem: GuestMemory,
    ) -> anyhow::Result<()> {
        let (_, worker_tx) = self.start_worker(idx)?;
        worker_tx
            .unbounded_send(WorkerCmd::StartQueue { index: idx, queue })
            .expect("worker channel closed early");
        self.activated_queues.insert(idx);
        Ok(())
    }

    pub fn stop_queue(&mut self, idx: usize) -> anyhow::Result<Queue> {
        // TODO: Consider stopping the worker thread if this is the last queue managed by it. Then,
        // simplify `virtio_sleep` and/or `reset` methods.
        let (_, worker_tx) = self
            .worker_threads
            .get(if self.worker_per_queue { &idx } else { &0 })
            .context("worker not found")?;
        let (response_tx, response_rx) = oneshot::channel();
        worker_tx
            .unbounded_send(WorkerCmd::StopQueue {
                index: idx,
                response_tx,
            })
            .expect("worker channel closed early");
        let queue = cros_async::block_on(async {
            response_rx
                .await
                .expect("response_rx closed early")
                .context("queue not found")
        })?;
        self.activated_queues.remove(&idx);
        Ok(queue)
    }
}

impl VirtioDevice for BlockAsync {
    fn keep_rds(&self) -> Vec<RawDescriptor> {
        let mut keep_rds = Vec::new();

        if let Some(disk_image) = &self.disk_image {
            keep_rds.extend(disk_image.as_raw_descriptors());
        }

        if let Some(control_tube) = &self.control_tube {
            keep_rds.push(control_tube.as_raw_descriptor());
        }

        keep_rds
    }

    fn features(&self) -> u64 {
        self.avail_features
    }

    fn device_type(&self) -> DeviceType {
        DeviceType::Block
    }

    fn queue_max_sizes(&self) -> &[u16] {
        &self.queue_sizes
    }

    fn read_config(&self, offset: u64, data: &mut [u8]) {
        let config_space = {
            let disk_size = self.disk_size.load(Ordering::Acquire);
            Self::build_config_space(
                disk_size,
                self.seg_max,
                self.block_size,
                self.queue_sizes.len() as u16,
            )
        };
        copy_config(data, 0, config_space.as_bytes(), offset);
    }

    fn activate(
        &mut self,
        mem: GuestMemory,
        _interrupt: Interrupt,
        queues: BTreeMap<usize, Queue>,
    ) -> anyhow::Result<()> {
        for (i, q) in queues {
            self.start_queue(i, q, mem.clone())?;
        }
        Ok(())
    }

    fn reset(&mut self) -> anyhow::Result<()> {
        for (_, (_, worker_tx)) in self.worker_threads.iter_mut() {
            let (response_tx, response_rx) = oneshot::channel();
            worker_tx
                .unbounded_send(WorkerCmd::AbortQueues { response_tx })
                .expect("worker channel closed early");
            cros_async::block_on(async { response_rx.await.expect("response_rx closed early") });
        }
        self.activated_queues.clear();
        Ok(())
    }

    fn virtio_sleep(&mut self) -> anyhow::Result<Option<BTreeMap<usize, Queue>>> {
        // Reclaim the queues from workers.
        let mut queues = BTreeMap::new();
        for index in self.activated_queues.clone() {
            queues.insert(index, self.stop_queue(index)?);
        }
        if queues.is_empty() {
            return Ok(None); // Not activated.
        }
        if let Some((_, (_, worker_tx))) = self.worker_threads.iter().next() {
            let (response_tx, response_rx) = oneshot::channel();
            worker_tx
                .unbounded_send(WorkerCmd::FlushDisk { response_tx })
                .expect("worker channel closed early");
            cros_async::block_on(async { response_rx.await.expect("response_rx closed early") });
        }
        Ok(Some(queues))
    }

    fn virtio_wake(
        &mut self,
        queues_state: Option<(GuestMemory, Interrupt, BTreeMap<usize, Queue>)>,
    ) -> anyhow::Result<()> {
        if let Some((mem, _interrupt, queues)) = queues_state {
            for (i, q) in queues {
                self.start_queue(i, q, mem.clone())?
            }
        }
        Ok(())
    }

    fn virtio_snapshot(&mut self) -> anyhow::Result<AnySnapshot> {
        // `virtio_sleep` ensures there is no pending state, except for the `Queue`s, which are
        // handled at a higher layer.
        AnySnapshot::to_any(())
    }

    fn virtio_restore(&mut self, data: AnySnapshot) -> anyhow::Result<()> {
        let () = AnySnapshot::from_any(data)?;
        Ok(())
    }

    fn pci_address(&self) -> Option<PciAddress> {
        self.pci_address
    }

    fn bootorder_fw_cfg(&self, pci_slot: u8) -> Option<(Vec<u8>, usize)> {
        self.boot_index
            .map(|s| (format!("scsi@{pci_slot}/disk@0,0").as_bytes().to_vec(), s))
    }
}

#[cfg(test)]
mod tests {
    use std::fs::File;
    use std::mem::size_of_val;
    use std::sync::atomic::AtomicU64;

    use data_model::Le32;
    use data_model::Le64;
    #[cfg(any(target_os = "android", target_os = "linux"))]
    use devices::suspendable_virtio_tests;
    use devices::virtio::base_features;
    use devices::virtio::create_descriptor_chain;
    use devices::virtio::DescriptorType;
    use devices::virtio::QueueConfig;
    use disk::SingleFileDisk;
    use hypervisor::ProtectionType;
    use tempfile::tempfile;
    use tempfile::TempDir;
    use vm_memory::GuestAddress;

    use super::*;

    #[test]
    fn read_size() {
        let f = tempfile().unwrap();
        f.set_len(0x1000).unwrap();

        let features = base_features(ProtectionType::Unprotected);
        let disk_option = DiskOption::default();
        let b = BlockAsync::new(features, Box::new(f), &disk_option, None, None, None).unwrap();
        let mut num_sectors = [0u8; 4];
        b.read_config(0, &mut num_sectors);
        // size is 0x1000, so num_sectors is 8 (4096/512).
        assert_eq!([0x08, 0x00, 0x00, 0x00], num_sectors);
        let mut msw_sectors = [0u8; 4];
        b.read_config(4, &mut msw_sectors);
        // size is 0x1000, so msw_sectors is 0.
        assert_eq!([0x00, 0x00, 0x00, 0x00], msw_sectors);
    }

    #[test]
    fn read_block_size() {
        let f = tempfile().unwrap();
        f.set_len(0x1000).unwrap();

        let features = base_features(ProtectionType::Unprotected);
        let disk_option = DiskOption {
            block_size: 4096,
            sparse: false,
            ..Default::default()
        };
        let b = BlockAsync::new(features, Box::new(f), &disk_option, None, None, None).unwrap();
        let mut blk_size = [0u8; 4];
        b.read_config(20, &mut blk_size);
        // blk_size should be 4096 (0x1000).
        assert_eq!([0x00, 0x10, 0x00, 0x00], blk_size);
    }

    #[test]
    fn read_features() {
        let tempdir = TempDir::new().unwrap();
        let mut path = tempdir.path().to_owned();
        path.push("disk_image");

        // Feature bits 0-23 and 50-127 are specific for the device type, but
        // at the moment crosvm only supports 64 bits of feature bits.
        const DEVICE_FEATURE_BITS: u64 = 0xffffff;

        // read-write block device
        {
            let f = File::create(&path).unwrap();
            let features = base_features(ProtectionType::Unprotected);
            let disk_option = DiskOption::default();
            let b = BlockAsync::new(features, Box::new(f), &disk_option, None, None, None).unwrap();
            // writable device should set VIRTIO_BLK_F_FLUSH + VIRTIO_BLK_F_DISCARD
            // + VIRTIO_BLK_F_WRITE_ZEROES + VIRTIO_BLK_F_BLK_SIZE + VIRTIO_BLK_F_SEG_MAX
            // + VIRTIO_BLK_F_MQ
            assert_eq!(0x7244, b.features() & DEVICE_FEATURE_BITS);
        }

        // read-write block device, non-sparse
        {
            let f = File::create(&path).unwrap();
            let features = base_features(ProtectionType::Unprotected);
            let disk_option = DiskOption {
                sparse: false,
                ..Default::default()
            };
            let b = BlockAsync::new(features, Box::new(f), &disk_option, None, None, None).unwrap();
            // writable device should set VIRTIO_F_FLUSH + VIRTIO_BLK_F_RO
            // + VIRTIO_BLK_F_BLK_SIZE + VIRTIO_BLK_F_SEG_MAX + VIRTIO_BLK_F_MQ
            assert_eq!(0x5244, b.features() & DEVICE_FEATURE_BITS);
        }

        // read-only block device
        {
            let f = File::create(&path).unwrap();
            let features = base_features(ProtectionType::Unprotected);
            let disk_option = DiskOption {
                read_only: true,
                ..Default::default()
            };
            let b = BlockAsync::new(features, Box::new(f), &disk_option, None, None, None).unwrap();
            // read-only device should set VIRTIO_BLK_F_RO
            // + VIRTIO_BLK_F_BLK_SIZE + VIRTIO_BLK_F_SEG_MAX + VIRTIO_BLK_F_MQ
            assert_eq!(0x1064, b.features() & DEVICE_FEATURE_BITS);
        }
    }

    #[test]
    fn check_pci_adress_configurability() {
        let f = tempfile().unwrap();

        let features = base_features(ProtectionType::Unprotected);
        let disk_option = DiskOption {
            pci_address: Some(PciAddress {
                bus: 0,
                dev: 1,
                func: 1,
            }),
            ..Default::default()
        };
        let b = BlockAsync::new(features, Box::new(f), &disk_option, None, None, None).unwrap();

        assert_eq!(b.pci_address(), disk_option.pci_address);
    }

    #[test]
    fn check_runtime_blk_queue_configurability() {
        let tempdir = TempDir::new().unwrap();
        let mut path = tempdir.path().to_owned();
        path.push("disk_image");
        let features = base_features(ProtectionType::Unprotected);

        // Default case
        let f = File::create(&path).unwrap();
        let disk_option = DiskOption::default();
        let b = BlockAsync::new(features, Box::new(f), &disk_option, None, None, None).unwrap();
        assert_eq!(
            [DEFAULT_QUEUE_SIZE; DEFAULT_NUM_QUEUES as usize],
            b.queue_max_sizes()
        );

        // Single queue of size 128
        let f = File::create(&path).unwrap();
        let disk_option = DiskOption::default();
        let b = BlockAsync::new(
            features,
            Box::new(f),
            &disk_option,
            None,
            Some(128),
            Some(1),
        )
        .unwrap();
        assert_eq!([128; 1], b.queue_max_sizes());
        // Single queue device should not set VIRTIO_BLK_F_MQ
        assert_eq!(0, b.features() & (1 << VIRTIO_BLK_F_MQ) as u64);
    }

    #[test]
    fn read_last_sector() {
        let ex = Executor::new().expect("creating an executor failed");

        let f = tempfile().unwrap();
        let disk_size = 0x1000;
        f.set_len(disk_size).unwrap();
        let af = SingleFileDisk::new(f, &ex).expect("Failed to create SFD");

        let mem = Rc::new(
            GuestMemory::new(&[(GuestAddress(0u64), 4 * 1024 * 1024)])
                .expect("Creating guest memory failed."),
        );

        let req_hdr = virtio_blk_req_header {
            req_type: Le32::from(VIRTIO_BLK_T_IN),
            reserved: Le32::from(0),
            sector: Le64::from(7), // Disk is 8 sectors long, so this is the last valid sector.
        };
        mem.write_obj_at_addr(req_hdr, GuestAddress(0x1000))
            .expect("writing req failed");

        let mut avail_desc = create_descriptor_chain(
            &mem,
            GuestAddress(0x100),  // Place descriptor chain at 0x100.
            GuestAddress(0x1000), // Describe buffer at 0x1000.
            vec![
                // Request header
                (DescriptorType::Readable, size_of_val(&req_hdr) as u32),
                // I/O buffer (1 sector of data)
                (DescriptorType::Writable, 512),
                // Request status
                (DescriptorType::Writable, 1),
            ],
            0,
        )
        .expect("create_descriptor_chain failed");

        let timer = Timer::new().expect("Failed to create a timer");
        let flush_timer = Rc::new(RefCell::new(
            TimerAsync::new(timer, &ex).expect("Failed to create an async timer"),
        ));
        let flush_timer_armed = Rc::new(RefCell::new(false));

        let disk_state = Rc::new(AsyncRwLock::new(DiskState {
            disk_image: Box::new(af),
            read_only: false,
            sparse: true,
            id: Default::default(),
            dontcache_read: false,
            dontcache_write: false,
            worker_shared_state: Arc::new(AsyncRwLock::new(WorkerSharedState {
                disk_size: Arc::new(AtomicU64::new(disk_size)),
            })),
        }));

        let fut = process_one_request(
            &mut avail_desc,
            &disk_state,
            &flush_timer,
            &flush_timer_armed,
        );

        ex.run_until(fut)
            .expect("running executor failed")
            .expect("execute failed");

        let status_offset = GuestAddress((0x1000 + size_of_val(&req_hdr) + 512) as u64);
        let status = mem.read_obj_from_addr::<u8>(status_offset).unwrap();
        assert_eq!(status, VIRTIO_BLK_S_OK);
    }

    #[test]
    fn read_beyond_last_sector() {
        let f = tempfile().unwrap();
        let disk_size = 0x1000;
        f.set_len(disk_size).unwrap();
        let mem = Rc::new(
            GuestMemory::new(&[(GuestAddress(0u64), 4 * 1024 * 1024)])
                .expect("Creating guest memory failed."),
        );

        let req_hdr = virtio_blk_req_header {
            req_type: Le32::from(VIRTIO_BLK_T_IN),
            reserved: Le32::from(0),
            sector: Le64::from(7), // Disk is 8 sectors long, so this is the last valid sector.
        };
        mem.write_obj_at_addr(req_hdr, GuestAddress(0x1000))
            .expect("writing req failed");

        let mut avail_desc = create_descriptor_chain(
            &mem,
            GuestAddress(0x100),  // Place descriptor chain at 0x100.
            GuestAddress(0x1000), // Describe buffer at 0x1000.
            vec![
                // Request header
                (DescriptorType::Readable, size_of_val(&req_hdr) as u32),
                // I/O buffer (2 sectors of data - overlap the end of the disk).
                (DescriptorType::Writable, 512 * 2),
                // Request status
                (DescriptorType::Writable, 1),
            ],
            0,
        )
        .expect("create_descriptor_chain failed");

        let ex = Executor::new().expect("creating an executor failed");

        let af = SingleFileDisk::new(f, &ex).expect("Failed to create SFD");
        let timer = Timer::new().expect("Failed to create a timer");
        let flush_timer = Rc::new(RefCell::new(
            TimerAsync::new(timer, &ex).expect("Failed to create an async timer"),
        ));
        let flush_timer_armed = Rc::new(RefCell::new(false));
        let disk_state = Rc::new(AsyncRwLock::new(DiskState {
            disk_image: Box::new(af),
            read_only: false,
            sparse: true,
            id: Default::default(),
            dontcache_read: false,
            dontcache_write: false,
            worker_shared_state: Arc::new(AsyncRwLock::new(WorkerSharedState {
                disk_size: Arc::new(AtomicU64::new(disk_size)),
            })),
        }));

        let fut = process_one_request(
            &mut avail_desc,
            &disk_state,
            &flush_timer,
            &flush_timer_armed,
        );

        ex.run_until(fut)
            .expect("running executor failed")
            .expect("execute failed");

        let status_offset = GuestAddress((0x1000 + size_of_val(&req_hdr) + 512 * 2) as u64);
        let status = mem.read_obj_from_addr::<u8>(status_offset).unwrap();
        assert_eq!(status, VIRTIO_BLK_S_IOERR);
    }

    #[test]
    fn get_id() {
        let ex = Executor::new().expect("creating an executor failed");

        let f = tempfile().unwrap();
        let disk_size = 0x1000;
        f.set_len(disk_size).unwrap();

        let mem = GuestMemory::new(&[(GuestAddress(0u64), 4 * 1024 * 1024)])
            .expect("Creating guest memory failed.");

        let req_hdr = virtio_blk_req_header {
            req_type: Le32::from(VIRTIO_BLK_T_GET_ID),
            reserved: Le32::from(0),
            sector: Le64::from(0),
        };
        mem.write_obj_at_addr(req_hdr, GuestAddress(0x1000))
            .expect("writing req failed");

        let mut avail_desc = create_descriptor_chain(
            &mem,
            GuestAddress(0x100),  // Place descriptor chain at 0x100.
            GuestAddress(0x1000), // Describe buffer at 0x1000.
            vec![
                // Request header
                (DescriptorType::Readable, size_of_val(&req_hdr) as u32),
                // I/O buffer (20 bytes for serial)
                (DescriptorType::Writable, 20),
                // Request status
                (DescriptorType::Writable, 1),
            ],
            0,
        )
        .expect("create_descriptor_chain failed");

        let af = SingleFileDisk::new(f, &ex).expect("Failed to create SFD");
        let timer = Timer::new().expect("Failed to create a timer");
        let flush_timer = Rc::new(RefCell::new(
            TimerAsync::new(timer, &ex).expect("Failed to create an async timer"),
        ));
        let flush_timer_armed = Rc::new(RefCell::new(false));

        let id = b"a20-byteserialnumber";

        let disk_state = Rc::new(AsyncRwLock::new(DiskState {
            disk_image: Box::new(af),
            read_only: false,
            sparse: true,
            id: *id,
            dontcache_read: false,
            dontcache_write: false,
            worker_shared_state: Arc::new(AsyncRwLock::new(WorkerSharedState {
                disk_size: Arc::new(AtomicU64::new(disk_size)),
            })),
        }));

        let fut = process_one_request(
            &mut avail_desc,
            &disk_state,
            &flush_timer,
            &flush_timer_armed,
        );

        ex.run_until(fut)
            .expect("running executor failed")
            .expect("execute failed");

        let status_offset = GuestAddress((0x1000 + size_of_val(&req_hdr) + 512) as u64);
        let status = mem.read_obj_from_addr::<u8>(status_offset).unwrap();
        assert_eq!(status, VIRTIO_BLK_S_OK);

        let id_offset = GuestAddress(0x1000 + size_of_val(&req_hdr) as u64);
        let returned_id = mem.read_obj_from_addr::<[u8; 20]>(id_offset).unwrap();
        assert_eq!(returned_id, *id);
    }

    #[test]
    fn reset_and_reactivate_single_worker() {
        reset_and_reactivate(false, None);
    }

    #[test]
    fn reset_and_reactivate_multiple_workers() {
        reset_and_reactivate(true, None);
    }

    #[test]
    #[cfg(windows)]
    fn reset_and_reactivate_overrlapped_io() {
        reset_and_reactivate(
            false,
            Some(
                cros_async::sys::windows::ExecutorKindSys::Overlapped { concurrency: None }.into(),
            ),
        );
    }

    fn reset_and_reactivate(
        enables_multiple_workers: bool,
        async_executor: Option<cros_async::ExecutorKind>,
    ) {
        // Create an empty disk image
        let f = tempfile::NamedTempFile::new().unwrap();
        f.as_file().set_len(0x1000).unwrap();
        // Close the file so that it is possible for the disk implementation to take exclusive
        // access when opening it.
        let path: tempfile::TempPath = f.into_temp_path();

        // Create an empty guest memory
        let mem = GuestMemory::new(&[(GuestAddress(0u64), 4 * 1024 * 1024)])
            .expect("Creating guest memory failed.");

        // Create a control tube.
        // NOTE: We don't want to drop the vmm half of the tube. That would cause the worker thread
        // will immediately fail, which isn't what we want to test in this case.
        let (_control_tube, control_tube_device) = Tube::pair().unwrap();

        // Create a BlockAsync to test
        let features = base_features(ProtectionType::Unprotected);
        let id = b"Block serial number\0";
        let disk_option = DiskOption {
            path: path.to_path_buf(),
            read_only: true,
            id: Some(*id),
            sparse: false,
            multiple_workers: enables_multiple_workers,
            async_executor,
            ..Default::default()
        };
        let disk_image = disk_option.open().unwrap();
        let mut b = BlockAsync::new(
            features,
            disk_image,
            &disk_option,
            Some(control_tube_device),
            None,
            None,
        )
        .unwrap();

        let interrupt = Interrupt::new_for_test();

        // activate with queues of an arbitrary size.
        let mut q0 = QueueConfig::new(DEFAULT_QUEUE_SIZE, 0);
        q0.set_ready(true);
        let q0 = q0
            .activate(&mem, Event::new().unwrap(), interrupt.clone())
            .expect("QueueConfig::activate");

        let mut q1 = QueueConfig::new(DEFAULT_QUEUE_SIZE, 0);
        q1.set_ready(true);
        let q1 = q1
            .activate(&mem, Event::new().unwrap(), interrupt.clone())
            .expect("QueueConfig::activate");

        b.activate(mem.clone(), interrupt, BTreeMap::from([(0, q0), (1, q1)]))
            .expect("activate should succeed");
        // assert resources are consumed
        if !enables_multiple_workers {
            assert!(
                b.disk_image.is_none(),
                "BlockAsync should not have a disk image"
            );
        }
        assert!(
            b.control_tube.is_none(),
            "BlockAsync should not have a control tube"
        );
        assert_eq!(
            b.worker_threads.len(),
            if enables_multiple_workers { 2 } else { 1 }
        );

        // reset and assert resources are still not back (should be in the worker thread)
        assert!(b.reset().is_ok(), "reset should succeed");
        if !enables_multiple_workers {
            assert!(
                b.disk_image.is_none(),
                "BlockAsync should not have a disk image"
            );
        }
        assert!(
            b.control_tube.is_none(),
            "BlockAsync should not have a control tube"
        );
        assert_eq!(
            b.worker_threads.len(),
            if enables_multiple_workers { 2 } else { 1 }
        );
        assert_eq!(b.id, *b"Block serial number\0");

        // re-activate should succeed
        let interrupt = Interrupt::new_for_test();
        let mut q0 = QueueConfig::new(DEFAULT_QUEUE_SIZE, 0);
        q0.set_ready(true);
        let q0 = q0
            .activate(&mem, Event::new().unwrap(), interrupt.clone())
            .expect("QueueConfig::activate");

        let mut q1 = QueueConfig::new(DEFAULT_QUEUE_SIZE, 0);
        q1.set_ready(true);
        let q1 = q1
            .activate(&mem, Event::new().unwrap(), interrupt.clone())
            .expect("QueueConfig::activate");

        b.activate(mem, interrupt, BTreeMap::from([(0, q0), (1, q1)]))
            .expect("re-activate should succeed");
    }

    #[test]
    fn resize_with_single_worker() {
        resize(false);
    }

    #[test]
    fn resize_with_multiple_workers() {
        // Test resize handled by one worker affect the whole state
        resize(true);
    }

    fn resize(enables_multiple_workers: bool) {
        // disk image size constants
        let original_size = 0x1000;
        let resized_size = 0x2000;

        // Create an empty disk image
        let f = tempfile().unwrap();
        f.set_len(original_size).unwrap();
        let disk_image: Box<dyn DiskFile> = Box::new(f);
        assert_eq!(disk_image.get_len().unwrap(), original_size);

        // Create an empty guest memory
        let mem = GuestMemory::new(&[(GuestAddress(0u64), 4 * 1024 * 1024)])
            .expect("Creating guest memory failed.");

        // Create a control tube
        let (control_tube, control_tube_device) = Tube::pair().unwrap();

        // Create a BlockAsync to test
        let features = base_features(ProtectionType::Unprotected);
        let disk_option = DiskOption {
            multiple_workers: enables_multiple_workers,
            ..Default::default()
        };
        let mut b = BlockAsync::new(
            features,
            disk_image.try_clone().unwrap(),
            &disk_option,
            Some(control_tube_device),
            None,
            None,
        )
        .unwrap();

        let interrupt = Interrupt::new_for_test();

        // activate with queues of an arbitrary size.
        let mut q0 = QueueConfig::new(DEFAULT_QUEUE_SIZE, 0);
        q0.set_ready(true);
        let q0 = q0
            .activate(&mem, Event::new().unwrap(), interrupt.clone())
            .expect("QueueConfig::activate");

        let mut q1 = QueueConfig::new(DEFAULT_QUEUE_SIZE, 0);
        q1.set_ready(true);
        let q1 = q1
            .activate(&mem, Event::new().unwrap(), interrupt.clone())
            .expect("QueueConfig::activate");

        b.activate(mem, interrupt.clone(), BTreeMap::from([(0, q0), (1, q1)]))
            .expect("activate should succeed");

        // assert the original size first
        assert_eq!(
            b.disk_size.load(Ordering::Acquire),
            original_size,
            "disk_size should be the original size first"
        );
        let mut capacity = [0u8; 8];
        b.read_config(0, &mut capacity);
        assert_eq!(
            capacity,
            // original_size (0x1000) >> SECTOR_SHIFT (9) = 0x8
            [0x8, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00],
            "read_config should read the original capacity first"
        );

        // assert resize works
        control_tube
            .send(&DiskControlCommand::Resize {
                new_size: resized_size,
            })
            .unwrap();
        assert_eq!(
            control_tube.recv::<DiskControlResult>().unwrap(),
            DiskControlResult::Ok,
            "resize command should succeed"
        );
        assert_eq!(
            b.disk_size.load(Ordering::Acquire),
            resized_size,
            "disk_size should be resized to the new size"
        );
        assert_eq!(
            disk_image.get_len().unwrap(),
            resized_size,
            "underlying disk image should be resized to the new size"
        );
        let mut capacity = [0u8; 8];
        b.read_config(0, &mut capacity);
        assert_eq!(
            capacity,
            // resized_size (0x2000) >> SECTOR_SHIFT (9) = 0x10
            [0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00],
            "read_config should read the resized capacity"
        );
        // Wait until the blk signals the interrupt
        interrupt
            .get_interrupt_evt()
            .wait()
            .expect("interrupt should be signaled");

        assert_eq!(
            interrupt.read_interrupt_status(),
            devices::virtio::INTERRUPT_STATUS_CONFIG_CHANGED as u8,
            "INTERRUPT_STATUS_CONFIG_CHANGED should be signaled"
        );
    }

    #[test]
    fn run_worker_threads() {
        // Create an empty duplicable disk image
        let f = tempfile().unwrap();
        f.set_len(0x1000).unwrap();
        let disk_image: Box<dyn DiskFile> = Box::new(f);

        // Create an empty guest memory
        let mem = GuestMemory::new(&[(GuestAddress(0u64), 4 * 1024 * 1024)])
            .expect("Creating guest memory failed.");

        // Create a BlockAsync to test with single worker thread
        let features = base_features(ProtectionType::Unprotected);
        let disk_option = DiskOption::default();
        let mut b = BlockAsync::new(
            features,
            disk_image.try_clone().unwrap(),
            &disk_option,
            None,
            None,
            None,
        )
        .unwrap();

        // activate with queues of an arbitrary size.
        let interrupt = Interrupt::new_for_test();
        let mut q0 = QueueConfig::new(DEFAULT_QUEUE_SIZE, 0);
        q0.set_ready(true);
        let q0 = q0
            .activate(&mem, Event::new().unwrap(), interrupt.clone())
            .expect("QueueConfig::activate");

        let mut q1 = QueueConfig::new(DEFAULT_QUEUE_SIZE, 0);
        q1.set_ready(true);
        let q1 = q1
            .activate(&mem, Event::new().unwrap(), interrupt.clone())
            .expect("QueueConfig::activate");

        b.activate(mem.clone(), interrupt, BTreeMap::from([(0, q0), (1, q1)]))
            .expect("activate should succeed");

        assert_eq!(b.worker_threads.len(), 1, "1 threads should be spawned.");
        drop(b);

        // Create a BlockAsync to test with multiple worker threads
        let features = base_features(ProtectionType::Unprotected);
        let disk_option = DiskOption {
            read_only: true,
            sparse: false,
            multiple_workers: true,
            ..DiskOption::default()
        };
        let mut b = BlockAsync::new(features, disk_image, &disk_option, None, None, None).unwrap();

        // activate should succeed
        let interrupt = Interrupt::new_for_test();
        let mut q0 = QueueConfig::new(DEFAULT_QUEUE_SIZE, 0);
        q0.set_ready(true);
        let q0 = q0
            .activate(&mem, Event::new().unwrap(), interrupt.clone())
            .expect("QueueConfig::activate");

        let mut q1 = QueueConfig::new(DEFAULT_QUEUE_SIZE, 0);
        q1.set_ready(true);
        let q1 = q1
            .activate(&mem, Event::new().unwrap(), interrupt.clone())
            .expect("QueueConfig::activate");

        b.activate(mem, interrupt, BTreeMap::from([(0, q0), (1, q1)]))
            .expect("activate should succeed");

        assert_eq!(b.worker_threads.len(), 2, "2 threads should be spawned.");
    }

    #[cfg(any(target_os = "android", target_os = "linux"))]
    struct BlockContext {}

    #[cfg(any(target_os = "android", target_os = "linux"))]
    fn modify_device(_block_context: &mut BlockContext, b: &mut BlockAsync) {
        b.avail_features = !b.avail_features;
    }

    #[cfg(any(target_os = "android", target_os = "linux"))]
    fn create_device() -> (BlockContext, BlockAsync) {
        // Create an empty disk image
        let f = tempfile().unwrap();
        f.set_len(0x1000).unwrap();
        let disk_image: Box<dyn DiskFile> = Box::new(f);

        // Create a BlockAsync to test
        let features = base_features(ProtectionType::Unprotected);
        let id = b"Block serial number\0";
        let disk_option = DiskOption {
            read_only: true,
            id: Some(*id),
            sparse: false,
            multiple_workers: true,
            ..Default::default()
        };
        (
            BlockContext {},
            BlockAsync::new(
                features,
                disk_image.try_clone().unwrap(),
                &disk_option,
                None,
                None,
                None,
            )
            .unwrap(),
        )
    }

    #[cfg(any(target_os = "android", target_os = "linux"))]
    suspendable_virtio_tests!(asyncblock, create_device, 2, modify_device);
}