Struct NetworkFrameQueues 

pub struct NetworkFrameQueues {
    pub guest_to_host: ArrayQueue<Vec<u8>>,
    pub host_to_guest: ArrayQueue<Vec<u8>>,
    pub guest_wake: WakePipe,
    pub host_wake: WakePipe,
    pub relay_wake: WakePipe,
    /* private fields */
}

Shared queues and wake handles for the host-side virtio-net runtime.

One NetworkFrameQueues is shared across all helper threads for a single guest NIC.

A useful mental model is:

queues  = ownership transfer for frame bytes
wakes   = "go look at the queue now"
shutdown= sticky flag + wake all blocked waiters

Fields

guest_to_host: ArrayQueue<Vec<u8>>

Raw ethernet frames emitted by the guest and waiting for smoltcp.

host_to_guest: ArrayQueue<Vec<u8>>

Raw ethernet frames emitted by smoltcp and waiting for libkrun.

guest_wake: WakePipe

Wake the smoltcp poll loop when a guest frame arrives.

guest_wake and relay_wake deliberately share one underlying poller: the smoltcp loop blocks on that single poller and either wake unblocks it. The loop re-runs its whole pipeline on every wakeup, so it does not need to know which side fired.

host_wake: WakePipe

Wake the libkrun writer thread when a host frame is ready.

relay_wake: WakePipe

Wake the smoltcp poll loop when a TCP relay thread has new data.

Implementations

impl NetworkFrameQueues

pub fn shared(capacity: usize) -> Arc<Self>

Create a new shared queue set wrapped in Arc.

pub fn add_egress_bytes(&self, n: u64)

Add n guest-outbound bytes to the egress counter. Relaxed ordering is fine: the counter is a monotonic statistic, not a synchronization point.

pub fn egress_bytes(&self) -> u64

Cumulative guest-outbound bytes for this NIC since boot.

pub fn egress_counter(&self) -> Arc<AtomicU64>

A cheap, cloneable read handle to the egress counter, for a flush thread owned by the launcher (the runtime itself is not Clone).

pub fn begin_shutdown(&self)

Mark the runtime as shutting down and wake all waiters.

The wakes are part of shutdown correctness. Without them, a thread blocked waiting on socket readiness could sleep indefinitely even though the shutdown flag was already set.

pub fn is_shutting_down(&self) -> bool

Whether shutdown has been requested.