2019-08-28 09:50:48 +00:00
|
|
|
// Copyright 2019 Intel Corporation. All Rights Reserved.
|
|
|
|
// SPDX-License-Identifier: Apache-2.0
|
|
|
|
|
2020-01-15 09:32:05 +00:00
|
|
|
use super::super::net_util::{build_net_config_space, CtrlVirtio, NetCtrlEpollHandler};
|
|
|
|
use super::super::Error as CtrlError;
|
2019-12-31 10:49:11 +00:00
|
|
|
use super::super::{ActivateError, ActivateResult, Queue, VirtioDevice, VirtioDeviceType};
|
|
|
|
use super::handler::*;
|
|
|
|
use super::vu_common_ctrl::*;
|
|
|
|
use super::Error as DeviceError;
|
|
|
|
use super::{Error, Result};
|
|
|
|
use crate::VirtioInterrupt;
|
|
|
|
use arc_swap::ArcSwap;
|
2019-08-28 09:50:48 +00:00
|
|
|
use libc;
|
|
|
|
use libc::EFD_NONBLOCK;
|
2020-01-09 11:56:21 +00:00
|
|
|
use net_util::MacAddr;
|
2019-08-28 09:50:48 +00:00
|
|
|
use std::cmp;
|
|
|
|
use std::io::Write;
|
2019-12-02 20:08:53 +00:00
|
|
|
use std::result;
|
|
|
|
use std::sync::atomic::{AtomicBool, Ordering};
|
2019-12-31 10:49:11 +00:00
|
|
|
use std::sync::Arc;
|
2019-08-28 09:50:48 +00:00
|
|
|
use std::thread;
|
|
|
|
use std::vec::Vec;
|
2019-09-23 23:39:41 +00:00
|
|
|
use vhost_rs::vhost_user::message::{VhostUserProtocolFeatures, VhostUserVirtioFeatures};
|
2019-08-30 18:47:24 +00:00
|
|
|
use vhost_rs::vhost_user::{Master, VhostUserMaster, VhostUserMasterReqHandler};
|
2019-08-28 09:50:48 +00:00
|
|
|
use vhost_rs::VhostBackend;
|
2019-09-19 13:42:29 +00:00
|
|
|
use virtio_bindings::bindings::virtio_net;
|
|
|
|
use virtio_bindings::bindings::virtio_ring;
|
2019-12-31 10:49:11 +00:00
|
|
|
use vm_device::{Migratable, MigratableError, Pausable, Snapshotable};
|
|
|
|
use vm_memory::GuestMemoryMmap;
|
|
|
|
use vmm_sys_util::eventfd::EventFd;
|
2019-08-28 09:50:48 +00:00
|
|
|
|
2020-01-09 17:29:00 +00:00
|
|
|
const DEFAULT_QUEUE_NUMBER: usize = 2;
|
|
|
|
|
2019-08-30 18:47:24 +00:00
|
|
|
struct SlaveReqHandler {}
|
|
|
|
impl VhostUserMasterReqHandler for SlaveReqHandler {}
|
|
|
|
|
2019-08-28 09:50:48 +00:00
|
|
|
pub struct Net {
|
|
|
|
vhost_user_net: Master,
|
2019-10-04 16:56:35 +00:00
|
|
|
kill_evt: Option<EventFd>,
|
2019-12-02 20:08:53 +00:00
|
|
|
pause_evt: Option<EventFd>,
|
2019-08-28 09:50:48 +00:00
|
|
|
avail_features: u64,
|
|
|
|
acked_features: u64,
|
2019-09-03 08:26:01 +00:00
|
|
|
backend_features: u64,
|
2019-08-28 09:50:48 +00:00
|
|
|
config_space: Vec<u8>,
|
|
|
|
queue_sizes: Vec<u16>,
|
2019-10-04 16:56:35 +00:00
|
|
|
queue_evts: Option<Vec<EventFd>>,
|
2020-01-13 17:52:19 +00:00
|
|
|
interrupt_cb: Option<Arc<dyn VirtioInterrupt>>,
|
2020-01-09 17:29:00 +00:00
|
|
|
epoll_thread: Option<Vec<thread::JoinHandle<result::Result<(), DeviceError>>>>,
|
2020-01-15 09:32:05 +00:00
|
|
|
ctrl_queue_epoll_thread: Option<thread::JoinHandle<result::Result<(), CtrlError>>>,
|
2019-12-02 20:08:53 +00:00
|
|
|
paused: Arc<AtomicBool>,
|
2019-08-28 09:50:48 +00:00
|
|
|
}
|
|
|
|
|
2019-09-23 17:42:52 +00:00
|
|
|
impl Net {
|
2020-01-09 17:29:00 +00:00
|
|
|
/// Create a new vhost-user-net device
|
2019-08-28 09:50:48 +00:00
|
|
|
/// Create a new vhost-user-net device
|
2019-09-23 17:42:52 +00:00
|
|
|
pub fn new(mac_addr: MacAddr, vu_cfg: VhostUserConfig) -> Result<Net> {
|
|
|
|
let mut vhost_user_net = Master::connect(&vu_cfg.sock, vu_cfg.num_queues as u64)
|
2019-08-28 09:50:48 +00:00
|
|
|
.map_err(Error::VhostUserCreateMaster)?;
|
|
|
|
|
|
|
|
// Filling device and vring features VMM supports.
|
|
|
|
let mut avail_features = 1 << virtio_net::VIRTIO_NET_F_GUEST_CSUM
|
|
|
|
| 1 << virtio_net::VIRTIO_NET_F_CSUM
|
|
|
|
| 1 << virtio_net::VIRTIO_NET_F_GUEST_TSO4
|
|
|
|
| 1 << virtio_net::VIRTIO_NET_F_GUEST_TSO6
|
|
|
|
| 1 << virtio_net::VIRTIO_NET_F_GUEST_ECN
|
|
|
|
| 1 << virtio_net::VIRTIO_NET_F_GUEST_UFO
|
|
|
|
| 1 << virtio_net::VIRTIO_NET_F_HOST_TSO4
|
|
|
|
| 1 << virtio_net::VIRTIO_NET_F_HOST_TSO6
|
|
|
|
| 1 << virtio_net::VIRTIO_NET_F_HOST_ECN
|
|
|
|
| 1 << virtio_net::VIRTIO_NET_F_HOST_UFO
|
|
|
|
| 1 << virtio_net::VIRTIO_NET_F_MRG_RXBUF
|
|
|
|
| 1 << virtio_net::VIRTIO_F_NOTIFY_ON_EMPTY
|
|
|
|
| 1 << virtio_net::VIRTIO_F_VERSION_1
|
|
|
|
| 1 << virtio_ring::VIRTIO_RING_F_EVENT_IDX
|
|
|
|
| VhostUserVirtioFeatures::PROTOCOL_FEATURES.bits();
|
|
|
|
|
|
|
|
vhost_user_net
|
|
|
|
.set_owner()
|
|
|
|
.map_err(Error::VhostUserSetOwner)?;
|
|
|
|
|
|
|
|
// Get features from backend, do negotiation to get a feature collection which
|
|
|
|
// both VMM and backend support.
|
2019-08-30 21:50:33 +00:00
|
|
|
let backend_features = vhost_user_net
|
|
|
|
.get_features()
|
|
|
|
.map_err(Error::VhostUserGetFeatures)?;
|
2019-08-28 09:50:48 +00:00
|
|
|
avail_features &= backend_features;
|
|
|
|
// Set features back is required by the vhost crate mechanism, since the
|
|
|
|
// later vhost call will check if features is filled in master before execution.
|
|
|
|
vhost_user_net
|
2019-08-30 18:55:28 +00:00
|
|
|
.set_features(avail_features)
|
2019-08-28 09:50:48 +00:00
|
|
|
.map_err(Error::VhostUserSetFeatures)?;
|
|
|
|
|
2020-01-09 17:29:00 +00:00
|
|
|
let protocol_features;
|
2019-08-28 09:50:48 +00:00
|
|
|
let mut acked_features = 0;
|
2019-09-03 08:26:01 +00:00
|
|
|
if avail_features & VhostUserVirtioFeatures::PROTOCOL_FEATURES.bits() != 0 {
|
2019-08-28 09:50:48 +00:00
|
|
|
acked_features |= VhostUserVirtioFeatures::PROTOCOL_FEATURES.bits();
|
2020-01-09 17:29:00 +00:00
|
|
|
protocol_features = vhost_user_net
|
2019-08-28 09:50:48 +00:00
|
|
|
.get_protocol_features()
|
|
|
|
.map_err(Error::VhostUserGetProtocolFeatures)?;
|
|
|
|
} else {
|
|
|
|
return Err(Error::VhostUserProtocolNotSupport);
|
|
|
|
}
|
|
|
|
|
2020-01-09 17:29:00 +00:00
|
|
|
let max_queue_number =
|
|
|
|
if protocol_features.bits() & VhostUserProtocolFeatures::MQ.bits() != 0 {
|
|
|
|
vhost_user_net
|
|
|
|
.set_protocol_features(protocol_features & VhostUserProtocolFeatures::MQ)
|
|
|
|
.map_err(Error::VhostUserSetProtocolFeatures)?;
|
|
|
|
match vhost_user_net.get_queue_num() {
|
|
|
|
Ok(qn) => qn,
|
|
|
|
Err(_) => DEFAULT_QUEUE_NUMBER as u64,
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
DEFAULT_QUEUE_NUMBER as u64
|
|
|
|
};
|
|
|
|
if vu_cfg.num_queues > max_queue_number as usize {
|
|
|
|
error!("vhost-user-net has queue number: {} larger than the max queue number: {} backend allowed\n",
|
|
|
|
vu_cfg.num_queues, max_queue_number);
|
|
|
|
return Err(Error::BadQueueNum);
|
|
|
|
}
|
|
|
|
|
2020-01-15 09:32:05 +00:00
|
|
|
avail_features |= 1 << virtio_net::VIRTIO_NET_F_CTRL_VQ;
|
|
|
|
let queue_num = vu_cfg.num_queues + 1;
|
|
|
|
|
2020-01-09 17:29:00 +00:00
|
|
|
let config_space = build_net_config_space(mac_addr, vu_cfg.num_queues, &mut avail_features);
|
2019-08-28 09:50:48 +00:00
|
|
|
|
|
|
|
// 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.
|
|
|
|
for i in 0..vu_cfg.num_queues {
|
|
|
|
vhost_user_net
|
|
|
|
.set_vring_base(i, 0)
|
|
|
|
.map_err(Error::VhostUserSetVringBase)?;
|
|
|
|
}
|
|
|
|
|
|
|
|
Ok(Net {
|
|
|
|
vhost_user_net,
|
2019-10-04 16:56:35 +00:00
|
|
|
kill_evt: None,
|
2019-12-02 20:08:53 +00:00
|
|
|
pause_evt: None,
|
2019-08-28 09:50:48 +00:00
|
|
|
avail_features,
|
|
|
|
acked_features,
|
2019-09-03 08:26:01 +00:00
|
|
|
backend_features,
|
2019-08-28 09:50:48 +00:00
|
|
|
config_space,
|
2020-01-15 09:32:05 +00:00
|
|
|
queue_sizes: vec![vu_cfg.queue_size; queue_num],
|
2019-10-04 16:56:35 +00:00
|
|
|
queue_evts: None,
|
|
|
|
interrupt_cb: None,
|
2019-12-02 20:08:53 +00:00
|
|
|
epoll_thread: None,
|
2020-01-15 09:32:05 +00:00
|
|
|
ctrl_queue_epoll_thread: None,
|
2019-12-02 20:08:53 +00:00
|
|
|
paused: Arc::new(AtomicBool::new(false)),
|
2019-08-28 09:50:48 +00:00
|
|
|
})
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
impl Drop for Net {
|
|
|
|
fn drop(&mut self) {
|
2019-10-04 16:56:35 +00:00
|
|
|
if let Some(kill_evt) = self.kill_evt.take() {
|
|
|
|
if let Err(e) = kill_evt.write(1) {
|
|
|
|
error!("failed to kill vhost-user-net: {:?}", e);
|
|
|
|
}
|
2019-08-28 09:50:48 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
impl VirtioDevice for Net {
|
|
|
|
fn device_type(&self) -> u32 {
|
|
|
|
VirtioDeviceType::TYPE_NET as u32
|
|
|
|
}
|
|
|
|
|
|
|
|
fn queue_max_sizes(&self) -> &[u16] {
|
|
|
|
&self.queue_sizes
|
|
|
|
}
|
|
|
|
|
|
|
|
fn features(&self, page: u32) -> u32 {
|
|
|
|
match page {
|
|
|
|
0 => self.avail_features as u32,
|
|
|
|
1 => (self.avail_features >> 32) as u32,
|
|
|
|
_ => {
|
|
|
|
warn!("Received request for unknown features page: {}", page);
|
|
|
|
0u32
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
fn ack_features(&mut self, page: u32, value: u32) {
|
|
|
|
let mut v = match page {
|
|
|
|
0 => u64::from(value),
|
|
|
|
1 => u64::from(value) << 32,
|
|
|
|
_ => {
|
|
|
|
warn!("Cannot acknowledge unknown features page: {}", page);
|
|
|
|
0u64
|
|
|
|
}
|
|
|
|
};
|
|
|
|
|
|
|
|
// Check if the guest is ACK'ing a feature that we didn't claim to have.
|
|
|
|
let unrequested_features = v & !self.avail_features;
|
|
|
|
if unrequested_features != 0 {
|
|
|
|
warn!("Received acknowledge request for unknown feature: {:x}", v);
|
|
|
|
// Don't count these features as acked.
|
|
|
|
v &= !unrequested_features;
|
|
|
|
}
|
|
|
|
self.acked_features |= v;
|
|
|
|
}
|
|
|
|
|
|
|
|
fn read_config(&self, offset: u64, mut data: &mut [u8]) {
|
|
|
|
let config_len = self.config_space.len() as u64;
|
|
|
|
if offset >= config_len {
|
|
|
|
error!("Failed to read config space");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
if let Some(end) = offset.checked_add(data.len() as u64) {
|
|
|
|
// This write can't fail, offset and end are checked against config_len.
|
|
|
|
data.write_all(&self.config_space[offset as usize..cmp::min(end, config_len) as usize])
|
|
|
|
.unwrap();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
fn write_config(&mut self, offset: u64, data: &[u8]) {
|
|
|
|
let data_len = data.len() as u64;
|
|
|
|
let config_len = self.config_space.len() as u64;
|
|
|
|
if offset + data_len > config_len {
|
|
|
|
error!("Failed to write config space");
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
let (_, right) = self.config_space.split_at_mut(offset as usize);
|
|
|
|
right.copy_from_slice(&data[..]);
|
|
|
|
}
|
|
|
|
|
|
|
|
fn activate(
|
|
|
|
&mut self,
|
2019-12-31 10:49:11 +00:00
|
|
|
mem: Arc<ArcSwap<GuestMemoryMmap>>,
|
2020-01-13 17:52:19 +00:00
|
|
|
interrupt_cb: Arc<dyn VirtioInterrupt>,
|
2020-01-15 09:32:05 +00:00
|
|
|
mut queues: Vec<Queue>,
|
|
|
|
mut queue_evts: Vec<EventFd>,
|
2019-08-28 09:50:48 +00:00
|
|
|
) -> ActivateResult {
|
2020-01-15 09:32:05 +00:00
|
|
|
if queues.len() != self.queue_sizes.len() || queue_evts.len() != self.queue_sizes.len() {
|
|
|
|
error!(
|
|
|
|
"Cannot perform activate. Expected {} queue(s), got {}",
|
|
|
|
self.queue_sizes.len(),
|
|
|
|
queues.len()
|
|
|
|
);
|
|
|
|
return Err(ActivateError::BadActivate);
|
|
|
|
}
|
|
|
|
|
2019-12-02 20:08:53 +00:00
|
|
|
let (self_kill_evt, kill_evt) = EventFd::new(EFD_NONBLOCK)
|
|
|
|
.and_then(|e| Ok((e.try_clone()?, e)))
|
|
|
|
.map_err(|e| {
|
|
|
|
error!("failed creating kill EventFd pair: {}", e);
|
|
|
|
ActivateError::BadActivate
|
|
|
|
})?;
|
2019-10-04 16:56:35 +00:00
|
|
|
self.kill_evt = Some(self_kill_evt);
|
|
|
|
|
2019-12-02 20:08:53 +00:00
|
|
|
let (self_pause_evt, pause_evt) = EventFd::new(EFD_NONBLOCK)
|
|
|
|
.and_then(|e| Ok((e.try_clone()?, e)))
|
|
|
|
.map_err(|e| {
|
|
|
|
error!("failed creating pause EventFd pair: {}", e);
|
|
|
|
ActivateError::BadActivate
|
|
|
|
})?;
|
|
|
|
self.pause_evt = Some(self_pause_evt);
|
|
|
|
|
2019-10-04 16:56:35 +00:00
|
|
|
// Save the interrupt EventFD as we need to return it on reset
|
|
|
|
// but clone it to pass into the thread.
|
|
|
|
self.interrupt_cb = Some(interrupt_cb.clone());
|
|
|
|
|
|
|
|
let mut tmp_queue_evts: Vec<EventFd> = Vec::new();
|
|
|
|
for queue_evt in queue_evts.iter() {
|
|
|
|
// Save the queue EventFD as we need to return it on reset
|
|
|
|
// but clone it to pass into the thread.
|
|
|
|
tmp_queue_evts.push(queue_evt.try_clone().map_err(|e| {
|
|
|
|
error!("failed to clone queue EventFd: {}", e);
|
|
|
|
ActivateError::BadActivate
|
|
|
|
})?);
|
|
|
|
}
|
|
|
|
self.queue_evts = Some(tmp_queue_evts);
|
2019-08-28 09:50:48 +00:00
|
|
|
|
2020-01-15 09:32:05 +00:00
|
|
|
let queue_num = queue_evts.len();
|
|
|
|
|
|
|
|
if (self.acked_features & 1 << virtio_net::VIRTIO_NET_F_CTRL_VQ) != 0 && queue_num % 2 != 0
|
|
|
|
{
|
|
|
|
let cvq_queue = queues.remove(queue_num - 1);
|
|
|
|
let cvq_queue_evt = queue_evts.remove(queue_num - 1);
|
|
|
|
|
|
|
|
let mut ctrl_handler = NetCtrlEpollHandler {
|
|
|
|
mem: mem.clone(),
|
|
|
|
kill_evt: kill_evt.try_clone().unwrap(),
|
|
|
|
pause_evt: pause_evt.try_clone().unwrap(),
|
|
|
|
ctrl_q: CtrlVirtio::new(cvq_queue, cvq_queue_evt),
|
|
|
|
epoll_fd: 0,
|
|
|
|
};
|
|
|
|
|
|
|
|
let paused = self.paused.clone();
|
|
|
|
thread::Builder::new()
|
|
|
|
.name("virtio_net".to_string())
|
|
|
|
.spawn(move || ctrl_handler.run_ctrl(paused))
|
|
|
|
.map(|thread| self.ctrl_queue_epoll_thread = Some(thread))
|
|
|
|
.map_err(|e| {
|
|
|
|
error!("failed to clone queue EventFd: {}", e);
|
|
|
|
ActivateError::BadActivate
|
|
|
|
})?;
|
|
|
|
}
|
|
|
|
|
2020-01-09 17:29:00 +00:00
|
|
|
let mut vu_interrupt_list = setup_vhost_user(
|
2019-08-28 09:50:48 +00:00
|
|
|
&mut self.vhost_user_net,
|
2019-12-31 10:49:11 +00:00
|
|
|
mem.load().as_ref(),
|
2019-08-28 09:50:48 +00:00
|
|
|
queues,
|
|
|
|
queue_evts,
|
2020-01-14 07:18:35 +00:00
|
|
|
&interrupt_cb,
|
2019-09-03 08:26:01 +00:00
|
|
|
self.acked_features & self.backend_features,
|
2019-08-28 09:50:48 +00:00
|
|
|
)
|
|
|
|
.map_err(ActivateError::VhostUserNetSetup)?;
|
|
|
|
|
2020-01-09 17:29:00 +00:00
|
|
|
let mut epoll_thread = Vec::new();
|
|
|
|
for _ in 0..vu_interrupt_list.len() / 2 {
|
2020-01-14 07:18:35 +00:00
|
|
|
let mut interrupt_list_sub: Vec<(Option<EventFd>, Queue)> = Vec::with_capacity(2);
|
2020-01-09 17:29:00 +00:00
|
|
|
interrupt_list_sub.push(vu_interrupt_list.remove(0));
|
|
|
|
interrupt_list_sub.push(vu_interrupt_list.remove(0));
|
|
|
|
|
|
|
|
let mut handler = VhostUserEpollHandler::<SlaveReqHandler>::new(VhostUserEpollConfig {
|
|
|
|
interrupt_cb: interrupt_cb.clone(),
|
|
|
|
kill_evt: kill_evt.try_clone().unwrap(),
|
|
|
|
pause_evt: pause_evt.try_clone().unwrap(),
|
|
|
|
vu_interrupt_list: interrupt_list_sub,
|
|
|
|
slave_req_handler: None,
|
|
|
|
});
|
|
|
|
|
|
|
|
let paused = self.paused.clone();
|
|
|
|
thread::Builder::new()
|
|
|
|
.name("vhost_user_net".to_string())
|
|
|
|
.spawn(move || handler.run(paused))
|
|
|
|
.map(|thread| epoll_thread.push(thread))
|
|
|
|
.map_err(|e| {
|
|
|
|
error!("failed to clone queue EventFd: {}", e);
|
|
|
|
ActivateError::BadActivate
|
|
|
|
})?;
|
|
|
|
}
|
|
|
|
|
|
|
|
self.epoll_thread = Some(epoll_thread);
|
2019-12-02 20:08:53 +00:00
|
|
|
|
2019-08-28 09:50:48 +00:00
|
|
|
Ok(())
|
|
|
|
}
|
2019-10-04 16:56:35 +00:00
|
|
|
|
2020-01-13 17:52:19 +00:00
|
|
|
fn reset(&mut self) -> Option<(Arc<dyn VirtioInterrupt>, Vec<EventFd>)> {
|
2019-12-02 20:08:53 +00:00
|
|
|
// We first must resume the virtio thread if it was paused.
|
|
|
|
if self.pause_evt.take().is_some() {
|
|
|
|
self.resume().ok()?;
|
|
|
|
}
|
|
|
|
|
2019-10-04 16:56:35 +00:00
|
|
|
if let Err(e) = reset_vhost_user(&mut self.vhost_user_net, self.queue_sizes.len()) {
|
|
|
|
error!("Failed to reset vhost-user daemon: {:?}", e);
|
|
|
|
return None;
|
|
|
|
}
|
|
|
|
|
|
|
|
if let Some(kill_evt) = self.kill_evt.take() {
|
|
|
|
// Ignore the result because there is nothing we can do about it.
|
|
|
|
let _ = kill_evt.write(1);
|
|
|
|
}
|
|
|
|
|
|
|
|
// Return the interrupt and queue EventFDs
|
|
|
|
Some((
|
|
|
|
self.interrupt_cb.take().unwrap(),
|
|
|
|
self.queue_evts.take().unwrap(),
|
|
|
|
))
|
|
|
|
}
|
2019-08-28 09:50:48 +00:00
|
|
|
}
|
2019-12-02 20:08:53 +00:00
|
|
|
|
2020-01-09 17:29:00 +00:00
|
|
|
virtio_pausable!(Net, true, true);
|
2019-12-02 20:08:53 +00:00
|
|
|
impl Snapshotable for Net {}
|
|
|
|
impl Migratable for Net {}
|