storage: add garbage collection in CasMgr

- Changed `delete_blobs` method in `CasDb` to take an immutable reference (`&self`) instead of a mutable reference (`&mut self`).
- Updated `dedup_chunk` method in `CasMgr` to correctly handle the deletion of non-existent blob files from both the file descriptor cache and the database.
- Implemented the `gc` (garbage collection) method in `CasMgr` to identify and remove blobs that no longer exist on the filesystem, ensuring the database and cache remain consistent.

Signed-off-by: Yadong Ding <ding_yadong@foxmail.com>

storage/src/cache/dedup/db.rs

@@ -129,7 +129,7 @@ impl CasDb {
         Ok(conn.last_insert_rowid() as u64)
     }
 
-    pub fn delete_blobs(&mut self, blobs: &[String]) -> Result<()> {
+    pub fn delete_blobs(&self, blobs: &[String]) -> Result<()> {
         let delete_blobs_sql = "DELETE FROM Blobs WHERE BlobId = (?1)";
         let delete_chunks_sql = "DELETE FROM Chunks WHERE BlobId = (?1)";
         let mut conn = self.get_connection()?;

storage/src/cache/dedup/mod.rs

@@ -124,6 +124,15 @@ impl CasMgr {
                         }
                     }
                 }
+            } else if d_file.as_ref().unwrap().metadata().is_err() {
+                // If the blob file no longer exists, delete if from fds and db.
+                let mut guard = self.fds.write().unwrap();
+                guard.remove(&path);
+                let blob_ids: &[String] = &[path];
+                if let Err(e) = self.db.delete_blobs(&blob_ids) {
+                    warn!("failed to delete blobs: {}", e);
+                }
+                return false;
             }
 
             if let Some(f) = d_file {
@@ -176,6 +185,33 @@ impl CasMgr {
             blob.digester().to_string() + ":" + &chunk.chunk_id().to_string()
         }
     }
+
+    // Check if blobs in the database still exist on the filesystem and perform garbage collection.
+    pub fn gc(&self) -> Result<()> {
+        let all_blobs = self.db.get_all_blobs()?;
+        let mut blobs_not_exist = Vec::new();
+        for (_, file_path) in all_blobs {
+            if !std::path::Path::new(&file_path).exists() {
+                blobs_not_exist.push(file_path);
+            }
+        }
+
+        // If there are any non-existent blobs, delete them from the database.
+        if !blobs_not_exist.is_empty() {
+            self.db.delete_blobs(&blobs_not_exist).map_err(|e| {
+                warn!("failed to delete blobs: {}", e);
+                e
+            })?;
+        }
+
+        let mut guard = self.fds.write().unwrap();
+        for path in blobs_not_exist {
+            // Remove the non-existent blob paths from the cache.
+            guard.remove(&path);
+        }
+
+        Ok(())
+    }
 }
 
 #[cfg(test)]