Failed TCP live migration can hang Cloud Hypervisor
Brief
Both sides now treat a dead socket as a stop instead of waiting forever.
Notes
6920bae274b1 (Philipp Schuster, 2026-08-26). vmm/src/migration/transport.rs. Fix hanging TCP migration workers on failure. If the worker queue is full, cleanup can run before workers drain it, Disconnect never enqueues, and join hangs the VMM. Co-authored-by Leander Kohler.
8b3868696fa6 (Schuster, 2026-08-26). Same file. Abort receive on unexpected EOF while receiving memory. A canceled sender made the receive loop treat a zero-length read as no progress and spin, hanging the destination API. Found by migration-cancellation tests that are not upstream yet.
Same-push cleanup, not scored: fbf2461df473, 175700b48442, dc6b45cd5a69, 9947e34e88ed. CH HEAD is now 8b386869 (v53.0). This is not the VHDX sequence-max abort.
Commit 6920bae274b1e1ef6e7b1050759470c4fee4b6cb