block_util: Avoid intermediate completion queue allocation

Rather than aggregate the completion list into an intermediate vector
instead adjust the API to provide one completion item at a time.

With DHAT this shows the number of heap allocations has decreased.

Before:

    dhat: Total:     623,852 bytes in 8,157 blocks

After:

    dhat: Total:     380,444 bytes in 3,469 blocks

Signed-off-by: Rob Bradford <robert.bradford@intel.com>

block_util/src/async_io.rs

@@ -141,5 +141,5 @@ pub trait AsyncIo: Send {
         user_data: u64,
     ) -> AsyncIoResult<()>;
     fn fsync(&mut self, user_data: Option<u64>) -> AsyncIoResult<()>;
-    fn complete(&mut self) -> Vec<(u64, i32)>;
+    fn next_completed_request(&mut self) -> Option<(u64, i32)>;
 }

block_util/src/fixed_vhd_async.rs

@@ -104,7 +104,7 @@ impl AsyncIo for FixedVhdAsync {
         self.raw_file_async.fsync(user_data)
     }
 
-    fn complete(&mut self) -> Vec<(u64, i32)> {
+    fn next_completed_request(&mut self) -> Option<(u64, i32)> {
-        self.raw_file_async.complete()
+        self.raw_file_async.next_completed_request()
     }
 }

block_util/src/fixed_vhd_sync.rs

@@ -101,7 +101,7 @@ impl AsyncIo for FixedVhdSync {
         self.raw_file_sync.fsync(user_data)
     }
 
-    fn complete(&mut self) -> Vec<(u64, i32)> {
+    fn next_completed_request(&mut self) -> Option<(u64, i32)> {
-        self.raw_file_sync.complete()
+        self.raw_file_sync.next_completed_request()
     }
 }

block_util/src/qcow_sync.rs

@@ -97,7 +97,7 @@ impl AsyncIo for QcowSync {
             .fsync_sync(user_data, &self.eventfd, &mut self.completion_list)
     }
 
-    fn complete(&mut self) -> Vec<(u64, i32)> {
+    fn next_completed_request(&mut self) -> Option<(u64, i32)> {
-        self.completion_list.drain(..).collect()
+        self.completion_list.pop()
     }
 }

block_util/src/raw_async.rs

@@ -155,13 +155,10 @@ impl AsyncIo for RawFileAsync {
         Ok(())
     }
 
-    fn complete(&mut self) -> Vec<(u64, i32)> {
+    fn next_completed_request(&mut self) -> Option<(u64, i32)> {
-        let cq = self.io_uring.completion();
+        self.io_uring
-        let mut completion_list = Vec::with_capacity(cq.len());
+            .completion()
-        for cq_entry in cq {
+            .next()
-            completion_list.push((cq_entry.user_data(), cq_entry.result()));
+            .map(|entry| (entry.user_data(), entry.result()))
-        }
-
-        completion_list
     }
 }

block_util/src/raw_sync.rs

@@ -127,7 +127,7 @@ impl AsyncIo for RawFileSync {
         Ok(())
     }
 
-    fn complete(&mut self) -> Vec<(u64, i32)> {
+    fn next_completed_request(&mut self) -> Option<(u64, i32)> {
-        self.completion_list.drain(..).collect()
+        self.completion_list.pop()
     }
 }

block_util/src/vhdx_sync.rs

@@ -96,7 +96,7 @@ impl AsyncIo for VhdxSync {
             .fsync_sync(user_data, &self.eventfd, &mut self.completion_list)
     }
 
-    fn complete(&mut self) -> Vec<(u64, i32)> {
+    fn next_completed_request(&mut self) -> Option<(u64, i32)> {
-        self.completion_list.drain(..).collect()
+        self.completion_list.pop()
     }
 }

virtio-devices/src/block.rs

@@ -244,8 +244,7 @@ impl BlockEpollHandler {
         let mut read_ops = Wrapping(0);
         let mut write_ops = Wrapping(0);
 
-        let completion_list = self.disk_image.complete();
+        while let Some((user_data, result)) = self.disk_image.next_completed_request() {
-        for (user_data, result) in completion_list {
             let desc_index = user_data as u16;
 
             let mut request = self.find_inflight_request(desc_index)?;