Ceph » rbd

Since pacific, LIST_SNAPS_FLAG_WHOLE_OBJECT flag is treated the same as read_whole_object case. This causes bogus holes to be reported when the object is smaller than max object size, plus a few semantic inconsistencies in other places.

On an image without fast-diff feature:

$ rbd create -s 1G --image-shared foo
$ sudo rbd device map foo
/dev/rbd0
$ sudo xfs_io -d -c 'pwrite -b 16k 0 16k' /dev/rbd0
$ sudo rbd device unmap /dev/rbd0
$ rbd snap create foo@snap1
Creating snap: 100% complete...done.
$ rbd snap create foo@snap2
Creating snap: 100% complete...done.

Running a diff from snap1 to HEAD -- no diff is expected, but --whole-object deviates:

$ rbd diff --from-snap snap1 foo
$ rbd diff --from-snap snap1 --whole-object foo
Offset  Length   Type
16384   4177920  zero