Archive for January, 2008

ZFS on Mac OS X 10.5

Friday, January 4th, 2008

Have you ever seen a Drobo? Cool stuff, eh? Yes, if it weren’t be USB only (let’s be honest, it sucks), therefore had a limit of 2 TeraByte per Volume and used some kind of proprietary “SAFE” algorithm (”it’s like RAID.5, but better”). No thanks…

Hey, I’m a geek, I can do this on my own, can’t I? - So let’s start again…

Have you ever seen ZFS? Cool stuff, eh?Yes, if it didn’t run primarily on Solaris and FreeBSD (sorry, too much stuff to learn at the same time). Linux, well, yeah, userspace… you wouldn’t want that, would you?

Fortunately, Apple (well, yeah, most likely Steve “the dictator” himself) decided to like ZFS and to port it to OS X. It wasn’t ready in an Apple-kind of way for 10.5, but it’s available for developers, which I am not, but Apple doesn’t know that.


It sucks. There is a pkg installation file in the developer connection, but it only runs on 10.5.0. So, after some screwing around with pax (seriously, who wrote that man page, and what is a member of an archive file?!), I was able to extract zfs.kext (if google is your friend, you might be able to find Shawn Ferry’s documentation on how to do that). Victory is mine!

Playing around:

I wouldn’t want to bore you with the umptiest description on how to create a pool - the PDF “ZFS on Mac OS X” from Apple says it all and is a very good start (thanks Chris). Let’s just say I had two hours of fun creating pools, making snapshots and clones and swapping disks.Some additional reading:


Well, my first plan was to rebuild a Drobo. Unfortunately, ZFS does not yet allow you to add an arbitrary amount of disks to a RAID.Z. The possibility to substitute every disk in a RAID.Z with a bigger one and afterwards just using the additional space is nice, but definitely not the same. A friend of mine suggested to use a stripeset of disks as one or more parts of the RAID.Z, which would give me at least the possibility to use all the 120, 160, 200 and 250GB disks lying around, but:

“Virtual devices cannot be nested, so a mirror or raidz virtual device can only contain files or disks. Mirrors of mirrors (or other combinations) are not allowed.”

Oh come on, where’s the fun in ZFS if you can’t do something stupid?

The OS X implementation is… well, yes, Beta. A source at Apple told me the data is stored “rock solid”, but I might have some kernel panics to get it there and from there. It’s not that bad, at least it’s predictable what you shouldn’t do to avoid kernel panics (primarily: do not unplug (even unmounted) ZFS devices), but you can see that it is not yet integrated in the rest of the system:

  • The mounted volume has drwxr-xr-x root:sys
  • The Finder can’t use the Trashcan on ZFS volumes. Everything gets deleted at once.
  • zfs umount doesn’t work most of the time. Surprisingly, dragging the volume in the Trashcan does?!
  • The afp share names for pool and filesystems are just the poolname plus an increment:




    Funnily enough, unmounting and remounting filesystems results in higher increments:




  • SMB, on the other hand, doesn’t show the filesystems at all. But, if the filesystems are mounted inside the pool, it is shown as a directory
  • Does compression work? It seems so, but a 30MB file consisting of zeroes takes either:

     30MB (ls -la),

     1MB (zfs list) or

     1KB (31.817.728 bytes) (Get Info on SMB-mount).

    Unfortunately, zfs get all still says: compressratio 1.00x
  • to be continued