External/cloud-hypervisor
1
0
Fork 0
mirror of https://github.com/cloud-hypervisor/cloud-hypervisor.git synced 2025-02-22 19:32:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

vm-virtio: Implement refactor for net devices and backend

Browse Source 
Since the common parts are put into net_util.rs under vm-virtio,
refactoring code for virtio-net device, vhost-user-net device
and backend to shrink the code size and improve readability
meanwhile.

Signed-off-by: Cathy Zhang <cathy.zhang@intel.com>
This commit is contained in:
Cathy Zhang 2020-01-09 19:56:21 +08:00 committed by Sebastien Boeuf
parent 6ae2597d19
commit 1ae7deb393
4 changed files with 146 additions and 470 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

212
src/bin/vhost_user_net.rs
View File

@ -18,44 +18,29 @@ use clap::{App, Arg};
use epoll;
use libc::{self, EAGAIN, EFD_NONBLOCK};
use log::*;
use std::cmp;
use net_util::Tap;
use std::fmt;
use std::io::Read;
use std::io::{self, Write};
use std::mem;
use std::io::{self};
use std::net::Ipv4Addr;
use std::os::unix::io::AsRawFd;
use std::process;
use std::sync::{Arc, RwLock};
use std::vec::Vec;
use vhost_rs::vhost_user::message::*;
use vhost_rs::vhost_user::Error as VhostUserError;
use vhost_user_backend::{VhostUserBackend, VhostUserDaemon, Vring, VringWorker};
use net_gen;
use net_util::{Tap, TapError};
use virtio_bindings::bindings::virtio_net::*;
use vm_memory::{Bytes, GuestAddress, GuestMemoryMmap};
use vm_memory::GuestMemoryMmap;
use vm_virtio::net_util::{
open_tap, RxVirtio, TxVirtio, KILL_EVENT, RX_QUEUE_EVENT, RX_TAP_EVENT, TX_QUEUE_EVENT,
};
use vm_virtio::Queue;
use vmm_sys_util::eventfd::EventFd;
/// The maximum buffer size when segmentation offload is enabled. This
/// includes the 12-byte virtio net header.
/// http://docs.oasis-open.org/virtio/virtio/v1.0/virtio-v1.0.html#x1-1740003
const MAX_BUFFER_SIZE: usize = 65562;
const QUEUE_SIZE: usize = 1024;
const NUM_QUEUES: usize = 2;
// The guest has made a buffer available to receive a frame into.
const RX_QUEUE_EVENT: u16 = 0;
// The transmit queue has a frame that is ready to send from the guest.
const TX_QUEUE_EVENT: u16 = 1;
// A frame is available for reading from the tap device to receive in the guest.
const RX_TAP_EVENT: u16 = 2;
// The device has been dropped.
const KILL_EVENT: u16 = 3;
pub type VhostUserResult<T> = std::result::Result<T, VhostUserError>;
pub type Result<T> = std::result::Result<T, Error>;
pub type VhostUserBackendResult<T> = std::result::Result<T, std::io::Error>;
@ -93,17 +78,7 @@ pub enum Error {
/// Failed to parse mask parameter.
ParseMaskParam,
/// Open tap device failed.
TapOpen(TapError),
/// Setting tap IP failed.
TapSetIp(TapError),
/// Setting tap netmask failed.
TapSetNetmask(TapError),
/// Setting tap interface offload flags failed.
TapSetOffload(TapError),
/// Setting vnet header size failed.
TapSetVnetHdrSize(TapError),
/// Enabling tap interface failed.
TapEnable(TapError),
OpenTap(vm_virtio::net_util::Error),
}
impl fmt::Display for Error {
@ -120,44 +95,6 @@ impl std::convert::From<Error> for std::io::Error {
}
}
#[derive(Clone)]
struct TxVirtio {
iovec: Vec<(GuestAddress, usize)>,
frame_buf: [u8; MAX_BUFFER_SIZE],
}
impl TxVirtio {
fn new() -> Self {
TxVirtio {
iovec: Vec::new(),
frame_buf: [0u8; MAX_BUFFER_SIZE],
}
}
}
#[derive(Clone)]
struct RxVirtio {
deferred_frame: bool,
deferred_irqs: bool,
bytes_read: usize,
frame_buf: [u8; MAX_BUFFER_SIZE],
}
impl RxVirtio {
fn new() -> Self {
RxVirtio {
deferred_frame: false,
deferred_irqs: false,
bytes_read: 0,
frame_buf: [0u8; MAX_BUFFER_SIZE],
}
}
}
fn vnet_hdr_len() -> usize {
mem::size_of::<virtio_net_hdr_v1>()
}
struct VhostUserNetBackend {
mem: Option<GuestMemoryMmap>,
vring_worker: Option<Arc<VringWorker>>,
@ -185,16 +122,6 @@ impl std::clone::Clone for VhostUserNetBackend {
impl VhostUserNetBackend {
/// Create a new virtio network device with the given TAP interface.
pub fn new_with_tap(tap: Tap) -> Result<Self> {
// Set offload flags to match the virtio features below.
tap.set_offload(
net_gen::TUN_F_CSUM | net_gen::TUN_F_UFO | net_gen::TUN_F_TSO4 | net_gen::TUN_F_TSO6,
)
.map_err(Error::TapSetOffload)?;
let vnet_hdr_size = vnet_hdr_len() as i32;
tap.set_vnet_hdr_size(vnet_hdr_size)
.map_err(Error::TapSetVnetHdrSize)?;
let rx = RxVirtio::new();
let tx = TxVirtio::new();
@ -212,10 +139,7 @@ impl VhostUserNetBackend {
/// Create a new virtio network device with the given IP address and
/// netmask.
pub fn new(ip_addr: Ipv4Addr, netmask: Ipv4Addr) -> Result<Self> {
let tap = Tap::new().map_err(Error::TapOpen)?;
tap.set_ip_addr(ip_addr).map_err(Error::TapSetIp)?;
tap.set_netmask(netmask).map_err(Error::TapSetNetmask)?;
tap.enable().map_err(Error::TapEnable)?;
let tap = open_tap(ip_addr, netmask).map_err(Error::OpenTap)?;
Self::new_with_tap(tap)
}
@ -223,10 +147,10 @@ impl VhostUserNetBackend {
// Copies a single frame from `self.rx.frame_buf` into the guest. Returns true
// if a buffer was used, and false if the frame must be deferred until a buffer
// is made available by the driver.
fn rx_single_frame(&mut self, vring: &mut Vring) -> Result<bool> {
fn rx_single_frame(&mut self, mut queue: &mut Queue) -> Result<bool> {
let mem = self.mem.as_ref().ok_or(Error::NoMemoryConfigured)?;
let mut next_desc = vring.mut_queue().iter(&mem).next();
let next_desc = queue.iter(&mem).next();
if next_desc.is_none() {
// Queue has no available descriptors
@ -244,52 +168,9 @@ impl VhostUserNetBackend {
}
return Ok(false);
}
let write_complete = self.rx.process_desc_chain(&mem, next_desc, &mut queue);
// We just checked that the head descriptor exists.
let head_index = next_desc.as_ref().unwrap().index;
let mut write_count = 0;
// Copy from frame into buffer, which may span multiple descriptors.
loop {
match next_desc {
Some(desc) => {
if !desc.is_write_only() {
break;
}
let limit = cmp::min(write_count + desc.len as usize, self.rx.bytes_read);
let source_slice = &self.rx.frame_buf[write_count..limit];
let write_result = mem.write_slice(source_slice, desc.addr);
match write_result {
Ok(_) => {
write_count = limit;
}
Err(e) => {
error!("Failed to write slice: {:?}", e);
break;
}
};
if write_count >= self.rx.bytes_read {
break;
}
next_desc = desc.next_descriptor();
}
None => {
warn!("Receiving buffer is too small to hold frame of current size");
break;
}
}
}
vring
.mut_queue()
.add_used(&mem, head_index, write_count as u32);
// Mark that we have at least one pending packet and we need to interrupt the guest.
self.rx.deferred_irqs = true;
Ok(write_count >= self.rx.bytes_read)
Ok(write_complete)
}
fn process_rx(&mut self, vring: &mut Vring) -> Result<()> {
@ -298,7 +179,7 @@ impl VhostUserNetBackend {
match self.read_tap() {
Ok(count) => {
self.rx.bytes_read = count;
if !self.rx_single_frame(vring)? {
if !self.rx_single_frame(&mut vring.mut_queue())? {
self.rx.deferred_frame = true;
break;
}
@ -328,7 +209,7 @@ impl VhostUserNetBackend {
fn resume_rx(&mut self, vring: &mut Vring) -> Result<()> {
if self.rx.deferred_frame {
if self.rx_single_frame(vring)? {
if self.rx_single_frame(&mut vring.mut_queue())? {
self.rx.deferred_frame = false;
// process_rx() was interrupted possibly before consuming all
// packets in the tap; try continuing now.
@ -345,65 +226,10 @@ impl VhostUserNetBackend {
}
}
fn process_tx(&mut self, vring: &mut Vring) -> Result<()> {
fn process_tx(&mut self, mut queue: &mut Queue) -> Result<()> {
let mem = self.mem.as_ref().ok_or(Error::NoMemoryConfigured)?;
let mut used_desc_heads = [(0, 0); QUEUE_SIZE];
let mut used_count = 0;
while let Some(avail_desc) = vring.mut_queue().iter(&mem).next() {
let head_index = avail_desc.index;
let mut read_count = 0;
let mut next_desc = Some(avail_desc);
self.tx.iovec.clear();
while let Some(desc) = next_desc {
if desc.is_write_only() {
break;
}
self.tx.iovec.push((desc.addr, desc.len as usize));
read_count += desc.len as usize;
next_desc = desc.next_descriptor();
}
used_desc_heads[used_count] = (head_index, read_count);
used_count += 1;
read_count = 0;
// Copy buffer from across multiple descriptors.
// TODO(performance - Issue #420): change this to use `writev()` instead of `write()`
// and get rid of the intermediate buffer.
for (desc_addr, desc_len) in self.tx.iovec.drain(..) {
let limit = cmp::min((read_count + desc_len) as usize, self.tx.frame_buf.len());
let read_result = mem.read_slice(
&mut self.tx.frame_buf[read_count..limit as usize],
desc_addr,
);
match read_result {
Ok(_) => {
// Increment by number of bytes actually read
read_count += limit - read_count;
}
Err(e) => {
error!("Failed to read slice: {:?}", e);
break;
}
}
}
let write_result = self.tap.write(&self.tx.frame_buf[..read_count as usize]);
match write_result {
Ok(_) => {}
Err(e) => {
error!("net: tx: error failed to write to tap: {}", e);
}
};
}
if used_count > 0 {
for &(desc_index, _) in &used_desc_heads[..used_count] {
vring.mut_queue().add_used(&mem, desc_index, 0);
}
vring.signal_used_queue().unwrap();
}
self.tx.process_desc_chain(&mem, &mut self.tap, &mut queue);
Ok(())
}
@ -468,7 +294,7 @@ impl VhostUserBackend for VhostUserNetBackend {
}
TX_QUEUE_EVENT => {
let mut vring = vrings[1].write().unwrap();
self.process_tx(&mut vring)?;
self.process_tx(&mut vring.mut_queue())?;
}
RX_TAP_EVENT => {
let mut vring = vrings[0].write().unwrap();
@ -476,7 +302,7 @@ impl VhostUserBackend for VhostUserNetBackend {
// Process a deferred frame first if available. Don't read from tap again
// until we manage to receive this deferred frame.
{
if self.rx_single_frame(&mut vring)? {
if self.rx_single_frame(&mut vring.mut_queue())? {
self.rx.deferred_frame = false;
self.process_rx(&mut vring)?;
} else if self.rx.deferred_irqs {

388
vm-virtio/src/net.rs
View File

@ -5,22 +5,23 @@
// Use of this source code is governed by a BSD-style license that can be
// found in the THIRD-PARTY file.
use super::net_util::{
build_net_config_space, open_tap, RxVirtio, TxVirtio, KILL_EVENT, NET_EVENTS_COUNT,
PAUSE_EVENT, RX_QUEUE_EVENT, RX_TAP_EVENT, TX_QUEUE_EVENT,
};
use super::Error as DeviceError;
use super::{
ActivateError, ActivateResult, DeviceEventT, Queue, VirtioDevice, VirtioDeviceType,
VirtioInterruptType,
ActivateError, ActivateResult, Queue, VirtioDevice, VirtioDeviceType, VirtioInterruptType,
};
use crate::VirtioInterrupt;
use arc_swap::ArcSwap;
use epoll;
use libc::EAGAIN;
use libc::EFD_NONBLOCK;
use net_gen;
use net_util::{MacAddr, Tap, TapError, MAC_ADDR_LEN};
use net_util::{MacAddr, Tap};
use std::cmp;
use std::io::Read;
use std::io::{self, Write};
use std::mem;
use std::net::Ipv4Addr;
use std::os::unix::io::{AsRawFd, RawFd};
use std::result;
@ -30,93 +31,21 @@ use std::thread;
use std::vec::Vec;
use virtio_bindings::bindings::virtio_net::*;
use vm_device::{Migratable, MigratableError, Pausable, Snapshotable};
use vm_memory::{Bytes, GuestAddress, GuestMemoryMmap};
use vm_memory::GuestMemoryMmap;
use vmm_sys_util::eventfd::EventFd;
/// The maximum buffer size when segmentation offload is enabled. This
/// includes the 12-byte virtio net header.
/// http://docs.oasis-open.org/virtio/virtio/v1.0/virtio-v1.0.html#x1-1740003
const MAX_BUFFER_SIZE: usize = 65562;
const QUEUE_SIZE: u16 = 256;
const NUM_QUEUES: usize = 2;
const QUEUE_SIZES: &[u16] = &[QUEUE_SIZE; NUM_QUEUES];
// A frame is available for reading from the tap device to receive in the guest.
const RX_TAP_EVENT: DeviceEventT = 0;
// The guest has made a buffer available to receive a frame into.
const RX_QUEUE_EVENT: DeviceEventT = 1;
// The transmit queue has a frame that is ready to send from the guest.
const TX_QUEUE_EVENT: DeviceEventT = 2;
// The device has been dropped.
pub const KILL_EVENT: DeviceEventT = 3;
// Number of DeviceEventT events supported by this implementation.
pub const NET_EVENTS_COUNT: usize = 4;
// The device should be paused.
const PAUSE_EVENT: DeviceEventT = 5;
#[derive(Debug)]
pub enum Error {
/// Open tap device failed.
TapOpen(TapError),
/// Setting tap IP failed.
TapSetIp(TapError),
/// Setting tap netmask failed.
TapSetNetmask(TapError),
/// Setting tap interface offload flags failed.
TapSetOffload(TapError),
/// Setting vnet header size failed.
TapSetVnetHdrSize(TapError),
/// Enabling tap interface failed.
TapEnable(TapError),
/// Failed to open taps.
OpenTap(super::net_util::Error),
}
pub type Result<T> = result::Result<T, Error>;
struct TxVirtio {
queue_evt: EventFd,
queue: Queue,
iovec: Vec<(GuestAddress, usize)>,
frame_buf: [u8; MAX_BUFFER_SIZE],
}
impl TxVirtio {
fn new(queue: Queue, queue_evt: EventFd) -> Self {
let tx_queue_max_size = queue.get_max_size() as usize;
TxVirtio {
queue_evt,
queue,
iovec: Vec::with_capacity(tx_queue_max_size),
frame_buf: [0u8; MAX_BUFFER_SIZE],
}
}
}
struct RxVirtio {
queue_evt: EventFd,
deferred_frame: bool,
deferred_irqs: bool,
queue: Queue,
bytes_read: usize,
frame_buf: [u8; MAX_BUFFER_SIZE],
}
impl RxVirtio {
fn new(queue: Queue, queue_evt: EventFd) -> Self {
RxVirtio {
queue_evt,
deferred_frame: false,
deferred_irqs: false,
queue,
bytes_read: 0,
frame_buf: [0u8; MAX_BUFFER_SIZE],
}
}
}
fn vnet_hdr_len() -> usize {
mem::size_of::<virtio_net_hdr_v1>()
}
struct NetEpollHandler {
mem: Arc<ArcSwap<GuestMemoryMmap>>,
tap: Tap,
@ -140,71 +69,35 @@ impl NetEpollHandler {
// Copies a single frame from `self.rx.frame_buf` into the guest. Returns true
// if a buffer was used, and false if the frame must be deferred until a buffer
// is made available by the driver.
fn rx_single_frame(&mut self) -> bool {
fn rx_single_frame(&mut self, mut queue: &mut Queue) -> bool {
let mem = self.mem.load();
let mut next_desc = self.rx.queue.iter(&mem).next();
let next_desc = queue.iter(&mem).next();
if next_desc.is_none() {
// Queue has no available descriptors
if self.rx_tap_listening {
self.unregister_tap_rx_listener().unwrap();
unregister_listener(
self.epoll_fd,
self.tap.as_raw_fd(),
epoll::Events::EPOLLIN,
u64::from(RX_TAP_EVENT),
)
.unwrap();
self.rx_tap_listening = false;
}
return false;
}
// We just checked that the head descriptor exists.
let head_index = next_desc.as_ref().unwrap().index;
let mut write_count = 0;
// Copy from frame into buffer, which may span multiple descriptors.
loop {
match next_desc {
Some(desc) => {
if !desc.is_write_only() {
break;
}
let limit = cmp::min(write_count + desc.len as usize, self.rx.bytes_read);
let source_slice = &self.rx.frame_buf[write_count..limit];
let write_result = mem.write_slice(source_slice, desc.addr);
match write_result {
Ok(_) => {
write_count = limit;
}
Err(e) => {
error!("Failed to write slice: {:?}", e);
break;
}
};
if write_count >= self.rx.bytes_read {
break;
}
next_desc = desc.next_descriptor();
}
None => {
warn!("Receiving buffer is too small to hold frame of current size");
break;
}
}
self.rx.process_desc_chain(&mem, next_desc, &mut queue)
}
self.rx.queue.add_used(&mem, head_index, write_count as u32);
// Mark that we have at least one pending packet and we need to interrupt the guest.
self.rx.deferred_irqs = true;
write_count >= self.rx.bytes_read
}
fn process_rx(&mut self) -> result::Result<(), DeviceError> {
fn process_rx(&mut self, queue: &mut Queue) -> result::Result<(), DeviceError> {
// Read as many frames as possible.
loop {
match self.read_tap() {
Ok(count) => {
self.rx.bytes_read = count;
if !self.rx_single_frame() {
if !self.rx_single_frame(queue) {
self.rx.deferred_frame = true;
break;
}
@ -225,22 +118,22 @@ impl NetEpollHandler {
}
if self.rx.deferred_irqs {
self.rx.deferred_irqs = false;
self.signal_used_queue(&self.rx.queue)
self.signal_used_queue(queue)
} else {
Ok(())
}
}
fn resume_rx(&mut self) -> result::Result<(), DeviceError> {
fn resume_rx(&mut self, queue: &mut Queue) -> result::Result<(), DeviceError> {
if self.rx.deferred_frame {
if self.rx_single_frame() {
if self.rx_single_frame(queue) {
self.rx.deferred_frame = false;
// process_rx() was interrupted possibly before consuming all
// packets in the tap; try continuing now.
self.process_rx()
self.process_rx(queue)
} else if self.rx.deferred_irqs {
self.rx.deferred_irqs = false;
self.signal_used_queue(&self.rx.queue)
self.signal_used_queue(queue)
} else {
Ok(())
}
@ -249,56 +142,10 @@ impl NetEpollHandler {
}
}
fn process_tx(&mut self) -> result::Result<(), DeviceError> {
fn process_tx(&mut self, mut queue: &mut Queue) -> result::Result<(), DeviceError> {
let mem = self.mem.load();
while let Some(avail_desc) = self.tx.queue.iter(&mem).next() {
let head_index = avail_desc.index;
let mut read_count = 0;
let mut next_desc = Some(avail_desc);
self.tx.iovec.clear();
while let Some(desc) = next_desc {
if desc.is_write_only() {
break;
}
self.tx.iovec.push((desc.addr, desc.len as usize));
read_count += desc.len as usize;
next_desc = desc.next_descriptor();
}
read_count = 0;
// Copy buffer from across multiple descriptors.
// TODO(performance - Issue #420): change this to use `writev()` instead of `write()`
// and get rid of the intermediate buffer.
for (desc_addr, desc_len) in self.tx.iovec.drain(..) {
let limit = cmp::min((read_count + desc_len) as usize, self.tx.frame_buf.len());
let read_result = mem.read_slice(
&mut self.tx.frame_buf[read_count..limit as usize],
desc_addr,
);
match read_result {
Ok(_) => {
// Increment by number of bytes actually read
read_count += limit - read_count;
}
Err(e) => {
error!("Failed to read slice: {:?}", e);
break;
}
}
}
let write_result = self.tap.write(&self.tx.frame_buf[..read_count as usize]);
match write_result {
Ok(_) => {}
Err(e) => {
warn!("net: tx: error failed to write to tap: {}", e);
}
};
self.tx.queue.add_used(&mem, head_index, 0);
}
self.tx.process_desc_chain(&mem, &mut self.tap, &mut queue);
Ok(())
}
@ -307,47 +154,73 @@ impl NetEpollHandler {
self.tap.read(&mut self.rx.frame_buf)
}
fn register_tap_rx_listener(&self) -> std::result::Result<(), std::io::Error> {
epoll::ctl(
self.epoll_fd,
epoll::ControlOptions::EPOLL_CTL_ADD,
self.tap.as_raw_fd(),
epoll::Event::new(epoll::Events::EPOLLIN, u64::from(RX_TAP_EVENT)),
)?;
Ok(())
fn handle_rx_event(&mut self, mut queue: &mut Queue, queue_evt: &EventFd) {
if let Err(e) = queue_evt.read() {
error!("Failed to get rx queue event: {:?}", e);
}
fn unregister_tap_rx_listener(&self) -> std::result::Result<(), std::io::Error> {
epoll::ctl(
self.resume_rx(&mut queue).unwrap();
if !self.rx_tap_listening {
register_listener(
self.epoll_fd,
epoll::ControlOptions::EPOLL_CTL_DEL,
self.tap.as_raw_fd(),
epoll::Event::new(epoll::Events::EPOLLIN, u64::from(RX_TAP_EVENT)),
)?;
Ok(())
epoll::Events::EPOLLIN,
u64::from(RX_TAP_EVENT),
)
.unwrap();
self.rx_tap_listening = true;
}
}
fn run(&mut self, paused: Arc<AtomicBool>) -> result::Result<(), DeviceError> {
fn handle_tx_event(&mut self, mut queue: &mut Queue, queue_evt: &EventFd) {
if let Err(e) = queue_evt.read() {
error!("Failed to get tx queue event: {:?}", e);
}
self.process_tx(&mut queue).unwrap();
}
fn handle_rx_tap_event(&mut self, mut queue: &mut Queue) {
if self.rx.deferred_frame
// Process a deferred frame first if available. Don't read from tap again
// until we manage to receive this deferred frame.
{
if self.rx_single_frame(&mut queue) {
self.rx.deferred_frame = false;
self.process_rx(&mut queue).unwrap();
} else if self.rx.deferred_irqs {
self.rx.deferred_irqs = false;
self.signal_used_queue(&queue).unwrap();
}
} else {
self.process_rx(&mut queue).unwrap();
}
}
fn run(
&mut self,
paused: Arc<AtomicBool>,
mut queues: Vec<Queue>,
queue_evts: Vec<EventFd>,
) -> result::Result<(), DeviceError> {
// Create the epoll file descriptor
self.epoll_fd = epoll::create(true).map_err(DeviceError::EpollCreateFd)?;
// Add events
// Add events
epoll::ctl(
self.epoll_fd,
epoll::ControlOptions::EPOLL_CTL_ADD,
self.rx.queue_evt.as_raw_fd(),
queue_evts[0].as_raw_fd(),
epoll::Event::new(epoll::Events::EPOLLIN, u64::from(RX_QUEUE_EVENT)),
)
.map_err(DeviceError::EpollCtl)?;
epoll::ctl(
self.epoll_fd,
epoll::ControlOptions::EPOLL_CTL_ADD,
self.tx.queue_evt.as_raw_fd(),
queue_evts[1].as_raw_fd(),
epoll::Event::new(epoll::Events::EPOLLIN, u64::from(TX_QUEUE_EVENT)),
)
.map_err(DeviceError::EpollCtl)?;
self.register_tap_rx_listener()
.map_err(DeviceError::EpollCtl)?;
self.rx_tap_listening = true;
epoll::ctl(
self.epoll_fd,
epoll::ControlOptions::EPOLL_CTL_ADD,
@ -363,8 +236,7 @@ impl NetEpollHandler {
)
.map_err(DeviceError::EpollCtl)?;
const EPOLL_EVENTS_LEN: usize = 100;
let mut events = vec![epoll::Event::new(epoll::Events::empty(), 0); EPOLL_EVENTS_LEN];
let mut events = vec![epoll::Event::new(epoll::Events::empty(), 0); NET_EVENTS_COUNT];
'epoll: loop {
let num_events = match epoll::wait(self.epoll_fd, -1, &mut events[..]) {
@ -389,43 +261,13 @@ impl NetEpollHandler {
match ev_type {
RX_QUEUE_EVENT => {
debug!("RX_QUEUE_EVENT received");
if let Err(e) = self.rx.queue_evt.read() {
error!("Failed to get rx queue event: {:?}", e);
break 'epoll;
}
self.resume_rx().unwrap();
if !self.rx_tap_listening {
self.register_tap_rx_listener().unwrap();
self.rx_tap_listening = true;
}
self.handle_rx_event(&mut queues[0], &queue_evts[0]);
}
TX_QUEUE_EVENT => {
debug!("TX_QUEUE_EVENT received");
if let Err(e) = self.tx.queue_evt.read() {
error!("Failed to get tx queue event: {:?}", e);
break 'epoll;
}
self.process_tx().unwrap();
self.handle_tx_event(&mut queues[1], &queue_evts[1]);
}
RX_TAP_EVENT => {
debug!("RX_TAP_EVENT received");
if self.rx.deferred_frame
// Process a deferred frame first if available. Don't read from tap again
// until we manage to receive this deferred frame.
{
if self.rx_single_frame() {
self.rx.deferred_frame = false;
self.process_rx().unwrap();
} else if self.rx.deferred_irqs {
self.rx.deferred_irqs = false;
self.signal_used_queue(&self.rx.queue).unwrap();
}
} else {
self.process_rx().unwrap();
}
self.handle_rx_tap_event(&mut queues[0]);
}
KILL_EVENT => {
debug!("KILL_EVENT received, stopping epoll loop");
@ -446,11 +288,38 @@ impl NetEpollHandler {
}
}
}
Ok(())
}
}
pub fn register_listener(
epoll_fd: RawFd,
fd: RawFd,
ev_type: epoll::Events,
data: u64,
) -> result::Result<(), io::Error> {
epoll::ctl(
epoll_fd,
epoll::ControlOptions::EPOLL_CTL_ADD,
fd,
epoll::Event::new(ev_type, data),
)
}
pub fn unregister_listener(
epoll_fd: RawFd,
fd: RawFd,
ev_type: epoll::Events,
data: u64,
) -> result::Result<(), io::Error> {
epoll::ctl(
epoll_fd,
epoll::ControlOptions::EPOLL_CTL_DEL,
fd,
epoll::Event::new(ev_type, data),
)
}
pub struct Net {
kill_evt: Option<EventFd>,
pause_evt: Option<EventFd>,
@ -468,17 +337,7 @@ pub struct Net {
impl Net {
/// Create a new virtio network device with the given TAP interface.
pub fn new_with_tap(tap: Tap, guest_mac: Option<&MacAddr>, iommu: bool) -> Result<Self> {
// Set offload flags to match the virtio features below.
tap.set_offload(
net_gen::TUN_F_CSUM | net_gen::TUN_F_UFO | net_gen::TUN_F_TSO4 | net_gen::TUN_F_TSO6,
)
.map_err(Error::TapSetOffload)?;
let vnet_hdr_size = vnet_hdr_len() as i32;
tap.set_vnet_hdr_size(vnet_hdr_size)
.map_err(Error::TapSetVnetHdrSize)?;
pub fn new_with_tap(tap: Tap, guest_mac: Option<MacAddr>, iommu: bool) -> Result<Self> {
let mut avail_features = 1 << VIRTIO_NET_F_GUEST_CSUM
| 1 << VIRTIO_NET_F_CSUM
| 1 << VIRTIO_NET_F_GUEST_TSO4
@ -491,15 +350,9 @@ impl Net {
avail_features |= 1u64 << VIRTIO_F_IOMMU_PLATFORM;
}
let mut config_space;
let config_space;
if let Some(mac) = guest_mac {
config_space = Vec::with_capacity(MAC_ADDR_LEN);
// This is safe, because we know the capacity is large enough.
unsafe { config_space.set_len(MAC_ADDR_LEN) }
config_space[..].copy_from_slice(mac.get_bytes());
// When this feature isn't available, the driver generates a random MAC address.
// Otherwise, it should attempt to read the device MAC address from the config space.
avail_features |= 1 << VIRTIO_NET_F_MAC;
config_space = build_net_config_space(mac, &mut avail_features);
} else {
config_space = Vec::new();
}
@ -523,13 +376,10 @@ impl Net {
pub fn new(
ip_addr: Ipv4Addr,
netmask: Ipv4Addr,
guest_mac: Option<&MacAddr>,
guest_mac: Option<MacAddr>,
iommu: bool,
) -> Result<Self> {
let tap = Tap::new().map_err(Error::TapOpen)?;
tap.set_ip_addr(ip_addr).map_err(Error::TapSetIp)?;
tap.set_netmask(netmask).map_err(Error::TapSetNetmask)?;
tap.enable().map_err(Error::TapEnable)?;
let tap = open_tap(ip_addr, netmask).map_err(Error::OpenTap)?;
Self::new_with_tap(tap, guest_mac, iommu)
}
@ -658,15 +508,17 @@ impl VirtioDevice for Net {
}
self.queue_evts = Some(tmp_queue_evts);
let rx_queue = queues.remove(0);
let tx_queue = queues.remove(0);
let rx_queue_evt = queue_evts.remove(0);
let tx_queue_evt = queue_evts.remove(0);
let mut queues_v = Vec::new();
let mut queue_evts_v = Vec::new();
queues_v.push(queues.remove(0));
queues_v.push(queues.remove(0));
queue_evts_v.push(queue_evts.remove(0));
queue_evts_v.push(queue_evts.remove(0));
let mut handler = NetEpollHandler {
mem,
tap,
rx: RxVirtio::new(rx_queue, rx_queue_evt),
tx: TxVirtio::new(tx_queue, tx_queue_evt),
rx: RxVirtio::new(),
tx: TxVirtio::new(),
interrupt_cb,
kill_evt,
pause_evt,
@ -677,7 +529,7 @@ impl VirtioDevice for Net {
let paused = self.paused.clone();
thread::Builder::new()
.name("virtio_net".to_string())
.spawn(move || handler.run(paused))
.spawn(move || handler.run(paused, queues_v, queue_evts_v))
.map(|thread| self.epoll_thread = Some(thread))
.map_err(|e| {
error!("failed to clone the virtio-net epoll thread: {}", e);

8
vm-virtio/src/vhost_user/net.rs
View File

@ -1,6 +1,7 @@
// Copyright 2019 Intel Corporation. All Rights Reserved.
// SPDX-License-Identifier: Apache-2.0
use super::super::net_util::build_net_config_space;
use super::super::{ActivateError, ActivateResult, Queue, VirtioDevice, VirtioDeviceType};
use super::handler::*;
use super::vu_common_ctrl::*;
@ -10,7 +11,7 @@ use crate::VirtioInterrupt;
use arc_swap::ArcSwap;
use libc;
use libc::EFD_NONBLOCK;
use net_util::{MacAddr, MAC_ADDR_LEN};
use net_util::MacAddr;
use std::cmp;
use std::io::Write;
use std::result;
@ -98,10 +99,7 @@ impl Net {
return Err(Error::VhostUserProtocolNotSupport);
}
let mut config_space = Vec::with_capacity(MAC_ADDR_LEN);
unsafe { config_space.set_len(MAC_ADDR_LEN) }
config_space[..].copy_from_slice(mac_addr.get_bytes());
avail_features |= 1 << virtio_net::VIRTIO_NET_F_MAC;
let config_space = build_net_config_space(mac_addr, &mut avail_features);
// Send set_vring_base here, since it could tell backends, like OVS + DPDK,
// how many virt queues to be handled, which backend required to know at early stage.

4
vmm/src/device_manager.rs
View File

@ -993,7 +993,7 @@ impl DeviceManager {
let virtio_net_device = if let Some(ref tap_if_name) = net_cfg.tap {
let tap = Tap::open_named(tap_if_name).map_err(DeviceManagerError::OpenTap)?;
Arc::new(Mutex::new(
vm_virtio::Net::new_with_tap(tap, Some(&net_cfg.mac), net_cfg.iommu)
vm_virtio::Net::new_with_tap(tap, Some(net_cfg.mac), net_cfg.iommu)
.map_err(DeviceManagerError::CreateVirtioNet)?,
))
} else {
@ -1001,7 +1001,7 @@ impl DeviceManager {
vm_virtio::Net::new(
net_cfg.ip,
net_cfg.mask,
Some(&net_cfg.mac),
Some(net_cfg.mac),
net_cfg.iommu,
)
.map_err(DeviceManagerError::CreateVirtioNet)?,