qcow2: fix flushing after dirty bitmap metadata writes

update_header_sync itself does not need to flush the caches to disk.
The only paths that allocate clusters are:

- bitmap_list_store with in_place=false, called by update_ext_header_and_dir

- store_bitmap_data, called by store_bitmap

- store_bitmap, called by qcow2_store_persistent_dirty_bitmaps and
  followed by update_ext_header_and_dir

So in the end the central place where we need to flush the caches
is update_ext_header_and_dir.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Kevin Wolf <kwolf@redhat.com>
diff --git a/block/qcow2-bitmap.c b/block/qcow2-bitmap.c
index 5127276..3010adb 100644
--- a/block/qcow2-bitmap.c
+++ b/block/qcow2-bitmap.c
@@ -110,7 +110,7 @@
         return ret;
     }
 
-    return bdrv_flush(bs);
+    return bdrv_flush(bs->file->bs);
 }
 
 static inline void bitmap_table_to_be(uint64_t *bitmap_table, size_t size)
@@ -882,7 +882,7 @@
             return ret;
         }
 
-        ret = bdrv_flush(bs->file->bs);
+        ret = qcow2_flush_caches(bs);
         if (ret < 0) {
             goto fail;
         }