fix(vsock): save state after sending a notification

This is a fix for a fix introduced in #4796
The issue was in vsock device hanging after snapshot
restoration due to the guest not being notified about
the termination packet. But there was bug in the fix, maily
we saved the vsock state before the notification was sent,
thus discarding all modifications made to sent the notification.

The reason original fix worked, is because we were only testing
with 1 iteration of snap/restore. This way even though we lost
synchronization with the guest in the event queue state, it worked
fine once. But doing more iterations causes vsock to hang
as before.

This commit fixes the issue by storing vsock state after the
notification is sent and modifies the vsock test to run
multiple iterations of snap/restore.

Signed-off-by: Egor Lazarchuk <[email protected]>

Author: ShadowCurse

src/vmm/src/device_manager/persist.rs

@@ -365,11 +365,6 @@ impl<'a> Persist<'a> for MMIODeviceManager {
                         .downcast_mut::<Vsock<VsockUnixBackend>>()
                         .unwrap();
 
-                    let vsock_state = VsockState {
-                        backend: vsock.backend().save(),
-                        frontend: vsock.save(),
-                    };
-
                     // Send Transport event to reset connections if device
                     // is activated.
                     if vsock.is_activated() {
@@ -378,6 +373,13 @@ impl<'a> Persist<'a> for MMIODeviceManager {
                         });
                     }
 
+                    // Save state after potential notification to the guest. This
+                    // way we save changes to the queue the notification can cause.
+                    let vsock_state = VsockState {
+                        backend: vsock.backend().save(),
+                        frontend: vsock.save(),
+                    };
+
                     states.vsock_device = Some(ConnectedVsockState {
                         device_id: devid.clone(),
                         device_state: vsock_state,

tests/integration_tests/functional/test_vsock.py

@@ -225,54 +225,61 @@ def test_vsock_transport_reset_g2h(uvm_nano, microvm_factory):
     test_vm.api.vsock.put(vsock_id="vsock0", guest_cid=3, uds_path=f"/{VSOCK_UDS_PATH}")
     test_vm.start()
 
-    host_socket_path = os.path.join(
-        test_vm.path, f"{VSOCK_UDS_PATH}_{ECHO_SERVER_PORT}"
-    )
-    host_socat_commmand = [
-        "socat",
-        "-dddd",
-        f"UNIX-LISTEN:{host_socket_path},fork",
-        "STDOUT",
-    ]
-    host_socat = subprocess.Popen(
-        host_socat_commmand, stdout=subprocess.PIPE, stderr=subprocess.PIPE
-    )
-
-    # Give some time for host socat to create socket
-    time.sleep(0.5)
-    assert Path(host_socket_path).exists()
-    test_vm.create_jailed_resource(host_socket_path)
-
-    # Create a socat process in the guest which will connect to the host socat
-    guest_socat_commmand = f"tmux new -d 'socat - vsock-connect:2:{ECHO_SERVER_PORT}'"
-    test_vm.ssh.run(guest_socat_commmand)
-
-    # socat should be running in the guest now
-    code, _, _ = test_vm.ssh.run("pidof socat")
-    assert code == 0
-
-    # Create snapshot.
+    # Create snapshot and terminate a VM.
     snapshot = test_vm.snapshot_full()
-    test_vm.resume()
-
-    # After `create_snapshot` + 'restore' calls, connection should be dropped
-    code, _, _ = test_vm.ssh.run("pidof socat")
-    assert code == 1
-
-    # Kill host socat as it is not useful anymore
-    host_socat.kill()
-    host_socat.communicate()
-
-    # Terminate VM.
     test_vm.kill()
 
-    # Load snapshot.
-    vm2 = microvm_factory.build()
-    vm2.spawn()
-    vm2.restore_from_snapshot(snapshot, resume=True)
-
-    # After snap restore all vsock connections should be
-    # dropped. This means guest socat should exit same way
-    # as it did after snapshot was taken.
-    code, _, _ = vm2.ssh.run("pidof socat")
-    assert code == 1
+    for _ in range(5):
+        # Load snapshot.
+        new_vm = microvm_factory.build()
+        new_vm.spawn()
+        new_vm.restore_from_snapshot(snapshot, resume=True)
+
+        # After snap restore all vsock connections should be
+        # dropped. This means guest socat should exit same way
+        # as it did after snapshot was taken.
+        code, _, _ = new_vm.ssh.run("pidof socat")
+        assert code == 1
+
+        host_socket_path = os.path.join(
+            new_vm.path, f"{VSOCK_UDS_PATH}_{ECHO_SERVER_PORT}"
+        )
+        host_socat_commmand = [
+            "socat",
+            "-dddd",
+            f"UNIX-LISTEN:{host_socket_path},fork",
+            "STDOUT",
+        ]
+        host_socat = subprocess.Popen(
+            host_socat_commmand, stdout=subprocess.PIPE, stderr=subprocess.PIPE
+        )
+
+        # Give some time for host socat to create socket
+        time.sleep(0.5)
+        assert Path(host_socket_path).exists()
+        new_vm.create_jailed_resource(host_socket_path)
+
+        # Create a socat process in the guest which will connect to the host socat
+        guest_socat_commmand = (
+            f"tmux new -d 'socat - vsock-connect:2:{ECHO_SERVER_PORT}'"
+        )
+        new_vm.ssh.run(guest_socat_commmand)
+
+        # socat should be running in the guest now
+        code, _, _ = new_vm.ssh.run("pidof socat")
+        assert code == 0
+
+        # Create snapshot.
+        snapshot = new_vm.snapshot_full()
+        new_vm.resume()
+
+        # After `create_snapshot` + 'restore' calls, connection should be dropped
+        code, _, _ = new_vm.ssh.run("pidof socat")
+        assert code == 1
+
+        # Kill host socat as it is not useful anymore
+        host_socat.kill()
+        host_socat.communicate()
+
+        # Terminate VM.
+        new_vm.kill()