Ceph » rbd

rbd_diff_iterate/rbd_diff_iterate2() API documentation says:

If the source snapshot name is NULL, we interpret that as the beginning of time and return all allocated regions of the image.

Based on this sentence, the user can rightfully assume that the callback would only be invoked for allocated areas (i.e. with exists=true). However, if there is a snapshot which contains a discarded object, diff-iterate reports holes (exists=false) even when diffing against the beginning of time:

$ rbd create --size 16M foo
$ sudo rbd map foo
/dev/rbd0
$ sudo xfs_io -d -c 'pwrite -b 4M 0 16M' /dev/rbd0
wrote 16777216/16777216 bytes at offset 0
16 MiB, 4 ops; 0.4607 sec (34.726 MiB/sec and 8.6814 ops/sec)
$ rbd snap create foo@snap
Creating snap: 100% complete...done.
$ rbd diff foo
Offset    Length   Type
0         4194304  data
4194304   4194304  data
8388608   4194304  data
12582912  4194304  data

$ blkdiscard -o 4M -l 8M /dev/rbd0
$ rbd diff foo
Offset    Length   Type
0         4194304  data
4194304   4194304  zero
8388608   4194304  zero
12582912  4194304  data

Removing such a snapshot makes these "holes" disappear from the report:

$ rbd snap rm foo@snap
Removing snap: 100% complete...done.
$ rbd diff foo
Offset    Length   Type
0         4194304  data
12582912  4194304  data

If nothing depends on this quirk, it should be fixed. Otherwise, the documentation needs to be updated to mention this.