'zfs destroy -r' of parent dataset destroys clones of subdatasets, while destroy -r of subdataset throws error/prompts to use -R

[maanloper]

System information

Type	Version/Name
OpenZFS Version	zfs-2.2.2-0ubuntu9.1, zfs-kmod-2.2.2-0ubuntu9.1

Describe the problem you're observing

zfs destroy -r pool/dataset will happily destroy pool/dataset/subdataset when it has clones, while zfs destroy -r pool/dataset/subdataset will not destroy the dataset, with the warning that it has clones and -R has to be used to destroy the dataset.

Describe how to reproduce the problem

1. Create a pool, with pool/dataset and pool/dataset/subdataset
2. Take a recursive snapshot of the pool: zfs snapshot -r pool@snapshot
3. Create a clone: zfs clone pool/dataset/subdataset@snapshot pool/dataset/cloneofsubdataset
4. Try to destroy pool/dataset/subdataset: zfs destroy -r pool/dataset/subdataset. This will throw an error and prompting to use -R.
5. Try to destroy pool/dataset: zfs destroy -r pool/dataset and observe how it happily destroys both pool/dataset and pool/dataset/subdataset

Shouldn't the behaviour be consistent whether it concerns the dataset or any of its parents?

[scineram]

5. What is problem? Why wouldn't it destroy the whole hierarchy?

[maanloper]

Because there is a difference in behaviour for the same command depending if it is executed directly on a dataset or on its parent?

[scineram]

RTM

[maanloper]

RTM

I'm unfamiliar with that abbreviation, what does it mean?

[pcd1193182]

-R  Recursively  destroy  all dependents, including cloned file systems outside the
             target hierarchy.
...

-r  Recursively destroy all children.

4. cloneofsubdataset is not a child of subdataset, it's a clone, which in this case is a dependent outside of the heirarchy of subdataset. Thus, -r fails and directs you to -R.
5. cloneofsubdataset and subdataset are both children of dataset, so -r destroys them all. None of the dependents are outside of the heirarchy, so it doesn't complain.