performance-metrics: Add QCOW2 copy-on-write write micro benchmark

Add micro_bench_qcow_cow_write which writes clusters into a QCOW2
overlay backed by a raw file.  Each write triggers copy-on-write:
cluster allocation, L2 and refcount table updates, then the data
write.  This measures COW allocation overhead compared to writing
into a plain empty image.

Workloads: 128 and 256 clusters.
Signed-off-by: Anatol Belski <anbelski@linux.microsoft.com>

performance-metrics/src/main.rs

@@ -378,7 +378,7 @@ mod adjuster {
     }
 }
 
-const TEST_LIST: [PerformanceTest; 74] = [
+const TEST_LIST: [PerformanceTest; 76] = [
     PerformanceTest {
         name: "boot_time_ms",
         func_ptr: performance_boot_time,
@@ -1397,6 +1397,30 @@ const TEST_LIST: [PerformanceTest; 74] = [
         },
         unit_adjuster: adjuster::s_to_us,
     },
+    PerformanceTest {
+        name: "micro_block_qcow_cow_write_128_us",
+        func_ptr: micro_bench_block::micro_bench_qcow_cow_write,
+        control: PerformanceTestControl {
+            test_timeout: 10,
+            test_iterations: 20,
+            warmup_iterations: 5,
+            num_ops: Some(128),
+            ..PerformanceTestControl::default()
+        },
+        unit_adjuster: adjuster::s_to_us,
+    },
+    PerformanceTest {
+        name: "micro_block_qcow_cow_write_256_us",
+        func_ptr: micro_bench_block::micro_bench_qcow_cow_write,
+        control: PerformanceTestControl {
+            test_timeout: 10,
+            test_iterations: 20,
+            warmup_iterations: 5,
+            num_ops: Some(256),
+            ..PerformanceTestControl::default()
+        },
+        unit_adjuster: adjuster::s_to_us,
+    },
 ];
 
 fn run_test_with_timeout(

performance-metrics/src/micro_bench_block.rs

@@ -221,3 +221,29 @@ pub fn micro_bench_qcow_backing_read(control: &PerformanceTestControl) -> f64 {
 
     elapsed
 }
+
+/// Write num_ops clusters into a QCOW2 overlay backed by a raw file.
+///
+/// Each write triggers copy-on-write: the overlay must allocate a new
+/// cluster, update L2 and refcount tables, then write the data.  This
+/// measures the COW allocation overhead compared to writing into an
+/// empty image (no backing read needed since we overwrite the full
+/// cluster).
+///
+/// Returns the total write wall clock time in seconds.
+pub fn micro_bench_qcow_cow_write(control: &PerformanceTestControl) -> f64 {
+    let num_ops = control.num_ops.expect("num_ops required") as usize;
+    let (_backing, _overlay, disk) = util::qcow_overlay_tempfile(num_ops);
+    let mut async_io = disk.new_async_io(1).expect("new_async_io failed");
+
+    let buf = vec![0xBBu8; QCOW_CLUSTER_SIZE as usize];
+    let iovec = write_iovec(&buf);
+
+    let start = Instant::now();
+    submit_writes(async_io.as_mut(), num_ops, QCOW_CLUSTER_SIZE, &[iovec]);
+    let elapsed = start.elapsed().as_secs_f64();
+
+    drain_completions(async_io.as_mut(), num_ops);
+
+    elapsed
+}