disk?s1 was already unmounted (was: ZFSDatasetProxy virtual mounts are broken)

[lundman]

Somewhere along the line, maybe Mojave, our pseudo disks for mountpoints appear to no longer work.

# ./cmd.sh zpool create -f zones disk3
# ./cmd.sh zpool export -a
disk5s1 was already unmounted
Fallback umount called

Which can be tested with

# ./cmd.sh zpool create -f zones disk3
# diskutil list
/dev/disk5 (internal, virtual):
   #:                       TYPE NAME                    SIZE       IDENTIFIER
   0:      GUID_partition_scheme                        +15.6 GB    disk5
   1:                ZFS Dataset                         15.6 GB    disk5s1
# diskutil unmount disk5s1
disk5s1 was already unmounted

Creating the pool with -O com.apple.devdisk=off makes diskutil happy. So it appears there is something with our pseudo disk that diskarb objects to.

[lundman]

Have you come across this yet @evansus ?

[JMoVS]

I’m not evansus but I come across this “calling fallback unmount” and then “already unmounted“ shenanigans quite often on export.

[lundman]

Ok turns out it is working. I had not installed Filesystems/zfs.fs and zfs_bundle. That it runs can be confirmed by checking you get a dataset name in diskutil list:

# diskutil list
/dev/disk5 (internal, virtual):
   #:                       TYPE NAME                    SIZE       IDENTIFIER
   0:      GUID_partition_scheme                        +20.9 GB    disk5
   1:                ZFS Dataset zones                   20.9 GB    disk5s1
==================================^^^  poolname means zfs.util was called ok.

then diskutil unmount works as usual.

[JMoVS]

what is the supposed benefit of giving it a disk id? I cannot point an Installer to such a dataset and Time Machine already works with com.apple.mimic_hfs for these...

[lundman]

Can those that experience the occasional unmount error please try the df6e22e fix.

[JMoVS]

I haven't seen a single "already unmounted" error since, looks good to me