Cannot get ZFS pool created in OSX to import on Linux system

Question

I'm experimenting with different file systems for use between a dual-booting Linux/OSX laptop which will act as a testing platform. Despite being a BSD variant, I've had a lot of trouble finding a compatible file system but have settled on the implementation by OpenZFS.

The latest OpenZFS 0.6.3-1 is installed on both systems. I initially created a pool from within OSX using /dev/disk0s6. Everything works fine in OSX with the drive mounting and being writable:

$ zpool status
  pool: Data
 state: ONLINE
  scan: none requested
config:

    NAME        STATE     READ WRITE CKSUM
    Data        ONLINE       0     0     0
      disk0s6   ONLINE       0     0     0

errors: No known data errors

$ zpool import
   pool: Data
     id: 16636970642933363897
  state: ONLINE
 action: The pool can be imported using its name or numeric identifier.
 config:

    Data        ONLINE
      disk0s6   ONLINE

But when I export the zpool and reboot into Linux, I cannot import the pool, even with -f:

$ zpool import -f Data
cannot import 'Data': one or more devices are already in use

$ zpool import
   pool: Data
     id: 16636970642933363897
  state: UNAVAIL
 status: One or more devices contains corrupted data.
 action: The pool cannot be imported due to damaged devices or data.
   see: http://zfsonlinux.org/msg/ZFS-8000-5E
 config:

    Data                                         UNAVAIL  insufficient replicas
      ata-Corsair_Force_GT_135004090FF015790001  UNAVAIL

Rebooting into OSX shows that the pool is not corrupt and loads just fine.

What am I doing wrong?

score 2 · Accepted Answer · answered Sep 27 '14 at 03:01

After a lot of troubleshooting with some great people over at the OpenZFS GitHub, I can confirm that this is a bug.

The real problem is that I created the pool using the last partition of the disk which, on Linux, can be confused as corruption if the partition aligns closely enough with the end of the disk.

ZFS creates four labels on the target device for redundancy with two at the beginning and two at the end. When ZFS evaluates the disk after booting into Linux it will encounter /dev/sda first which will provide a partial match against the last two labels at the end of the disk (from the last partition). It then erroneously believes that the device is corrupt as there are no labels at the start of the disk.

The solution was to add a buffer of at least 10MB of free space at the end of the disk.

Full details can be found here: https://github.com/zfsonlinux/zfs/issues/2742

score 1 · Answer 2 · edited Apr 13 '17 at 12:14

Ok, so this is definitely due to my inexperience with ZFS. Unlike a traditional file system where you can reference a particular partition from any context, ZFS seems to retain the exact information used to reference the data location no matter where it is loaded from.

For example, I created the zpool and referenced the device by from OSX by "/dev/disk0s6" which is not present on Linux. I created a symlink from /dev/disk0s6 to /dev/sda6 and everything works OK now.

Referencing these questions:

ZFS RAID0 pool without redundancy

Why did rebooting cause one side of my ZFS mirror to become UNAVAIL?

It would appear as though I need to take portability into consideration when referencing the device. For example, using the partition's UUID which would remain unchanged between the two OSs.

I think this information is useful for others who are new to ZFS.

EDIT: This answer is WRONG! Please see the correct answer below.

Yea, I read one of the links in that question that pointed to: "Why did rebooting cause one side of my ZFS mirror to become UNAVAIL?" and it hit me like a ton of bricks. It's so obvious. :P — Zhro, Sep 26 '14 at 01:23

Cannot get ZFS pool created in OSX to import on Linux system

2 Answers2