fix(net): break tx loop on non-recoverable errors to prevent busy-loop

[areporeporepo]

Summary

Fixes #602.

When process_tx() returns a persistent error other than NothingWritten (e.g. Internal or ProcessNotRunning), the loop in process_tx_loop() continues spinning if the guest keeps supplying TX buffers, because tx_has_deferred_frame is false and has_new_entries is true:

process_tx() → Err(Internal) → tx_has_deferred_frame = false
enable_notification() → has_new_entries = true

Fix

Break immediately on non-recoverable errors. The worker will be re-entered on the next:

• TX queue kick (guest adds new descriptors)
• Backend socket writable event (edge-triggered epoll OUT)

This is safe because Internal and ProcessNotRunning errors are non-transient — retrying immediately won't resolve them, but the next epoll cycle gives the system a chance to recover or the guest driver to reset.

Change

 self.tx_has_deferred_frame = match self.process_tx() {
     Err(TxError::Backend(WriteError::NothingWritten)) => true,
     Err(e) => {
-        false
+        break;
     }
     _ => false,
 };

Testing

• cargo clippy --features net -- -D warnings passes
• cargo fmt -- --check passes
• CI unit tests (Linux-only, relying on PR checks)

[slp]

/gemini review

[slp]

@mtjhrc PTAL

[gemini-code-assist]

Code Review

This pull request modifies the virtio net worker to break the TX processing loop upon encountering non-recoverable errors. This change is intended to prevent potential busy-looping scenarios where a guest continues to provide TX buffers despite backend failures. I have no feedback to provide as there were no review comments to evaluate.