diff --git a/block/src/formats/qcow/backing.rs b/block/src/formats/qcow/backing.rs
index ced5cd39b..a149b2145 100644
--- a/block/src/formats/qcow/backing.rs
+++ b/block/src/formats/qcow/backing.rs
@@ -139,12 +139,6 @@ impl Qcow2Backing {
     }
 }
 
-impl Drop for Qcow2Backing {
-    fn drop(&mut self) {
-        self.metadata.shutdown();
-    }
-}
-
 /// Construct a thread safe backing file reader.
 pub(super) fn shared_backing_from(bf: BackingFile) -> BlockResult<Arc<dyn BackingRead>> {
     let (kind, virtual_size) = bf.into_kind();
diff --git a/block/src/formats/qcow/engine_sync.rs b/block/src/formats/qcow/engine_sync.rs
index 488ebf7c4..9b3b7f766 100644
--- a/block/src/formats/qcow/engine_sync.rs
+++ b/block/src/formats/qcow/engine_sync.rs
@@ -2016,7 +2016,6 @@ mod unit_tests {
         assert!(result < 0, "host WriteZeroes failure must reach the guest");
 
         drop(aio);
-        metadata.shutdown();
         drop(metadata);
 
         let reopened = QcowDisk::new(
@@ -2314,7 +2313,6 @@ mod unit_tests {
         // L2, and only now may the old data cluster enter unref_clusters.
         aio.apply_dealloc_action(&actions[0]).unwrap();
         metadata.flush().unwrap();
-        metadata.shutdown();
         drop(aio);
         drop(metadata);
 
diff --git a/block/src/formats/qcow/metadata.rs b/block/src/formats/qcow/metadata.rs
index 1ce8145d7..2e7780e35 100644
--- a/block/src/formats/qcow/metadata.rs
+++ b/block/src/formats/qcow/metadata.rs
@@ -389,6 +389,12 @@ impl QcowMetadata {
     }
 }
 
+impl Drop for QcowMetadata {
+    fn drop(&mut self) {
+        self.shutdown();
+    }
+}
+
 impl QcowState {
     /// Fast path read mapping under read lock only. Returns None on cache
     /// miss.
diff --git a/block/src/formats/qcow/mod.rs b/block/src/formats/qcow/mod.rs
index a7ad66c20..3d09d27b3 100644
--- a/block/src/formats/qcow/mod.rs
+++ b/block/src/formats/qcow/mod.rs
@@ -240,12 +240,6 @@ impl QcowTempDisk {
     }
 }
 
-impl Drop for QcowDisk {
-    fn drop(&mut self) {
-        self.metadata.shutdown();
-    }
-}
-
 impl disk_file::DiskSize for QcowDisk {
     fn logical_size(&self) -> BlockResult<u64> {
         Ok(self.metadata.virtual_size())
@@ -349,6 +343,8 @@ impl disk_file::AsyncDiskFile for QcowDisk {
 
 #[cfg(test)]
 mod unit_tests {
+    use std::os::unix::fs::FileExt;
+
     use super::*;
     use crate::async_io::AsyncIo;
     use crate::disk_file::{AsyncDiskFile, DiskSize, PhysicalSize};
@@ -362,6 +358,13 @@ mod unit_tests {
             .into_file()
     }
 
+    fn dirty_bit_is_set(file: &File) -> bool {
+        let mut buf = [0u8; 8];
+        file.read_exact_at(&mut buf, header::V2_BARE_HEADER_SIZE as u64)
+            .unwrap();
+        u64::from_be_bytes(buf) & IncompatFeatures::DIRTY.bits() != 0
+    }
+
     #[test]
     fn new_sync_returns_correct_size() {
         let file = make_qcow_file();
@@ -401,6 +404,34 @@ mod unit_tests {
         assert_async_io_from_dyn(cloned.as_ref(), false);
     }
 
+    #[test]
+    fn dropping_clone_does_not_clear_dirty_bit() {
+        let file = make_qcow_file();
+        let disk = QcowDisk::new(file, false, false, true, false).unwrap();
+        let cloned = disk.try_clone().unwrap();
+
+        drop(cloned);
+
+        assert_ne!(
+            disk.metadata().header().incompatible_features & IncompatFeatures::DIRTY.bits(),
+            0
+        );
+    }
+
+    #[test]
+    fn async_io_clears_dirty_bit_when_last_metadata_owner_drops() {
+        let file = make_qcow_file();
+        let inspect = file.try_clone().unwrap();
+        let disk = QcowDisk::new(file, false, false, true, false).unwrap();
+        let async_io = disk.create_async_io(1).unwrap();
+
+        drop(disk);
+        assert!(dirty_bit_is_set(&inspect));
+
+        drop(async_io);
+        assert!(!dirty_bit_is_set(&inspect));
+    }
+
     #[cfg(feature = "io_uring")]
     #[test]
     fn try_clone_preserves_io_uring_dispatch() {