[Qemu-devel] qemu-img convert : failed to convert an image which contains a backing file

[Akkarit Sangpetch]

Hi, I have a problem with qemu-img convert tool in the master branch.

The command 'qemu-img convert' failed to produce a valid image if the source 
image referenced a backing file.

To reproduce, suppose we have an image base.qcow2
1. qemu-img create -b base.qcow2 -f qcow2 temp.qcow2
2. do something with temp.qcow2 (run a vm, write a file, etc.)
3. qemu-img convert -O qcow2 temp.qcow2 rebase.qcow2

The content of rebase.qcow2 is exactly the same as temp.qcow2, but without any
backing file reference. This makes the file unusable.

In earlier version, qemu-img convert produced a rebased version (it also
removed reference to the backing files but the final image contains merged
contents from both base.qcow2 and temp.qcow2)

It seems that qemu-img convert assume '-B' option so it skips the unallocated 
part of the source file.

The attached patch seems to fix the problem. 
(it modifies the patch found here http://git.savannah.gnu.org/cgit/qemu.git/commit/?id=93c65b47a6fb9ba0e2b89269a751ba3433a33427)

--Akkarit

diff -uNrp qemu-kvm-devel-87.orig/qemu-img.c qemu-kvm-devel-87/qemu-img.c
--- qemu-kvm-devel-87.orig/qemu-img.c   2009-07-05 13:44:04.095124070 -0400
+++ qemu-kvm-devel-87/qemu-img.c        2009-07-05 20:40:07.282452250 -0400
@@ -747,14 +747,20 @@ static int img_convert(int argc, char **
                 n = bs_offset + bs_sectors - sector_num;
 
             if (strcmp(drv->format_name, "host_device")) {
-                if (!bdrv_is_allocated(bs[bs_i], sector_num - bs_offset,
-                                       n, &n1)) {
-                    sector_num += n1;
-                    continue;
-                }
-                /* The next 'n1' sectors are allocated in the input image. Copy
-                   only those as they may be followed by unallocated sectors. 
*/
-                n = n1;
+                /* If the output image is being created as a copy on write 
image,
+                   assume that sectors which are unallocated in the input image
+                   are present in both the output's and input's base images (no
+                   need to copy them). */
+               if (out_baseimg) {
+                  if (!bdrv_is_allocated(bs[bs_i], sector_num - bs_offset,
+                                         n, &n1)) {
+                      sector_num += n1;
+                      continue;
+                  }
+                  /* The next 'n1' sectors are allocated in the input image. 
Copy
+                     only those as they may be followed by unallocated 
sectors. */
+                  n = n1;
+               }
             } else {
                 n1 = n;
             }