CSI: fix potential state store corruptions

.changelog/16256.txt

@@ -0,0 +1,3 @@
+```release-note:bug
+csi: Fixed potential state store corruption when garbage collecting CSI volume claims or checking whether it's safe to force-deregister a volume
+```

nomad/core_sched.go

@@ -728,7 +728,7 @@ func (c *CoreScheduler) csiVolumeClaimGC(eval *structs.Evaluation) error {
 		// we only call the claim release RPC if the volume has claims
 		// that no longer have valid allocations. otherwise we'd send
 		// out a lot of do-nothing RPCs.
-		vol, err := c.snap.CSIVolumeDenormalize(ws, vol)
+		vol, err := c.snap.CSIVolumeDenormalize(ws, vol.Copy())
 		if err != nil {
 			return err
 		}

nomad/core_sched_test.go

@@ -2218,6 +2218,7 @@ func TestCoreScheduler_CSIPluginGC(t *testing.T) {
 	require.NoError(t, err)
 
 	// Empty the plugin
+	plug = plug.Copy()
 	plug.Controllers = map[string]*structs.CSIInfo{}
 	plug.Nodes = map[string]*structs.CSIInfo{}
 

nomad/state/state_store.go

@@ -1458,14 +1458,13 @@ func deleteNodeCSIPlugins(txn *txn, node *structs.Node, index uint64) error {
 
 // updateOrGCPlugin updates a plugin but will delete it if the plugin is empty
 func updateOrGCPlugin(index uint64, txn Txn, plug *structs.CSIPlugin) error {
-	plug.ModifyIndex = index
-
 	if plug.IsEmpty() {
 		err := txn.Delete("csi_plugins", plug)
 		if err != nil {
 			return fmt.Errorf("csi_plugins delete error: %v", err)
 		}
 	} else {
+		plug.ModifyIndex = index
 		err := txn.Insert("csi_plugins", plug)
 		if err != nil {
 			return fmt.Errorf("csi_plugins update error %s: %v", plug.ID, err)
@@ -2601,6 +2600,7 @@ func (s *StateStore) CSIVolumeDeregister(index uint64, namespace string, ids []s
 // volSafeToForce checks if the any of the remaining allocations
 // are in a non-terminal state.
 func (s *StateStore) volSafeToForce(txn Txn, v *structs.CSIVolume) bool {
+	v = v.Copy()
 	vol, err := s.csiVolumeDenormalizeTxn(txn, nil, v)
 	if err != nil {
 		return false

nomad/state/state_store_test.go

@@ -3490,8 +3490,10 @@ func TestStateStore_CSIVolume(t *testing.T) {
 	vs = slurp(iter)
 	require.False(t, vs[0].HasFreeWriteClaims())
 
-	claim0.Mode = u
-	err = state.CSIVolumeClaim(2, ns, vol0, claim0)
+	claim2 := new(structs.CSIVolumeClaim)
+	*claim2 = *claim0
+	claim2.Mode = u
+	err = state.CSIVolumeClaim(2, ns, vol0, claim2)
 	require.NoError(t, err)
 	ws = memdb.NewWatchSet()
 	iter, err = state.CSIVolumesByPluginID(ws, ns, "", "minnie")
@@ -3520,8 +3522,10 @@ func TestStateStore_CSIVolume(t *testing.T) {
 
 	// release claims to unblock deregister
 	index++
-	claim0.State = structs.CSIVolumeClaimStateReadyToFree
-	err = state.CSIVolumeClaim(index, ns, vol0, claim0)
+	claim3 := new(structs.CSIVolumeClaim)
+	*claim3 = *claim2
+	claim3.State = structs.CSIVolumeClaimStateReadyToFree
+	err = state.CSIVolumeClaim(index, ns, vol0, claim3)
 	require.NoError(t, err)
 	index++
 	claim1.Mode = u
@@ -3577,7 +3581,7 @@ func TestStateStore_CSIPlugin_Lifecycle(t *testing.T) {
 		require.Equal(t, counts.nodesHealthy, plug.NodesHealthy, "nodes healthy")
 		require.Equal(t, counts.controllersExpected, plug.ControllersExpected, "controllers expected")
 		require.Equal(t, counts.nodesExpected, plug.NodesExpected, "nodes expected")
-		return plug
+		return plug.Copy()
 	}
 
 	type allocUpdateKind int