vhost-user-vsock: enable savevm/loadvm for VM snapshots

[omerzlabs]

Summary

This patch enables savevm/loadvm for the vhost-user-vsock device, allowing VM snapshots when vsock-based networking is in use (e.g., Penguin's VPN port forwarding).

Currently vhost-user-vsock is marked as non-migratable, blocking all snapshot operations even though the parent class (vhost-vsock-common) already provides correct save/restore hooks. The kernel-backed variant (vhost-vsock) uses these hooks and is fully migratable. This patch brings vhost-user-vsock to parity.

Changes

Single file: hw/virtio/vhost-user-vsock.c (19 lines added, 1 removed)

1. Replace unmigratable VMStateDescription with a proper one

The empty vuv_vmstate with .unmigratable = 1 is replaced with a vmstate that:

• Saves/restores VMSTATE_VIRTIO_DEVICE (virtio ring state, features, config)
• Uses vhost_vsock_common_pre_save() — asserts the vhost backend is stopped before save
• Uses vhost_vsock_common_post_load() — sends VIRTIO_VSOCK_EVENT_TRANSPORT_RESET after load so the guest kernel reconnects vsock sockets

This matches the kernel-backed vhost-vsock device exactly.

2. Remove the vhost-level migration blocker after device init

vhost_dev_init() registers a migration blocker when the backend lacks VHOST_F_LOG_ALL (dirty-page tracking for live migration). For savevm/loadvm, dirty tracking is unnecessary — the VM is paused and state is captured atomically. The blocker is removed after successful device initialization.

Note: migrate_del_blocker() removes the blocker for all migration modes. If live migration support is ever added for this device, the blocker handling should be revisited.

Why this is safe

• vm_stop() is called before savevm, which triggers virtio_vmstate_change() -> vuv_set_status(0) -> vhost_vsock_common_stop(). The backend is guaranteed stopped when pre_save runs.
• The chardev/vhost-user connection survives same-process savevm/loadvm — nothing tears it down.
• On loadvm, post_load schedules a VIRTIO_VSOCK_EVENT_TRANSPORT_RESET via timer, telling the guest kernel to reconnect. set_status then restarts the backend.
• Changes are scoped entirely to vhost-user-vsock — no other devices are affected.

Testing

• make check unit tests: 97 pass / 0 fail / 4 skip (unchanged from baseline)
• qtest-x86_64/migration-test: 40 subtests pass (unchanged from baseline)
• savevm with vhost-user-vsock-pci enabled: succeeds (previously blocked with "State blocked by non-migratable device")
• loadvm from snapshot: VM restores, guest receives transport reset
• End-to-end: validated with Penguin firmware emulations across multiple vendors (Netgear, D-Link, TP-Link) — services accessible after snapshot restore