guest_memory: fixes to Bytes<GuestAddress>::read_from

There are a bug and a small enhancement here.  The enhancement is that the
source need not provide the exact number of bytes; try_access supports
partial reads so we can use src.read() instead of src.read_exact().
The bug fix is in the else case, which uses src.read() but then throws
away the result and advances the iterator by the full amount that was
expected.

We could also fix the bug by using src.read_exact() in the else branch,
but read_from() has a return value for partial reads so it's unnecessary
to do that.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

Author: bonzini

coverage_config.json

@@ -1,5 +1,5 @@
 {
-  "coverage_score": 84.5,
+  "coverage_score": 84.3,
   "exclude_path": "mmap_windows.rs",
   "crate_features": "backend-mmap"
 }

src/guest_memory.rs

@@ -582,16 +582,15 @@ impl<T: GuestMemory> Bytes<GuestAddress> for T {
                 // This is safe cause `start` and `len` are within the `region`.
                 let start = caddr.raw_value() as usize;
                 let end = start + len;
-                src.read_exact(&mut dst[start..end])
-                    .map_err(Error::IOError)?;
-                Ok(len)
+                let bytes_read = src.read(&mut dst[start..end]).map_err(Error::IOError)?;
+                Ok(bytes_read)
             } else {
                 let len = std::cmp::min(len, MAX_ACCESS_CHUNK);
                 let mut buf = vec![0u8; len].into_boxed_slice();
                 let bytes_read = src.read(&mut buf[..]).map_err(Error::IOError)?;
                 let bytes_written = region.write(&buf[0..bytes_read], caddr)?;
                 assert_eq!(bytes_written, bytes_read);
-                Ok(len)
+                Ok(bytes_read)
             }
         })
     }