Today while trying to update some softwares on my NAS server I had issues with Debian Jessie's missing repositories. I though it was the right time to upgrade to OpenMediaVault 4, which was released last year and features a Debian Stretch.

The upgrade of packages went smoothly but as usual with this kind of upgrade, problems came after the reboot: the server became irresponsive.

Time to find a VGA cable and dig into it. The server was stuck in maintenance mode saying that the systemd unit responsible for mounting the data disk timed out, all subsequent dependencies failed too (other data mount points). I checked the status of the array, it was there and healthy but blkid was not returning it at all.

Don't forget to make backups before playing with your drives following a failure.

Trying to mount the array by hand gave me this pretty message:

mount: /dev/md0: more filesystems detected. This should not happen,
       use -t  to explicitly specify the filesystem type or
       use wipefs(8) to clean up the device.

Well… Executing wipefs /dev/md0 showed that there was a residual ZFS filesystem down there:

offset               type
0x2ba993ff000        zfs_member   [filesystem]

0x438                ext4   [filesystem]

I tried FreeNAS and a ZFS pool before installing OpenMediaVault and it seems that now the system is not really happy with that. Folks usually advise to make a backup of data, wipe all filesystem signatures from the disk and create the array again.

As I was quite unsatisfied with this method, I decided to give a try to just wiping out the ZFS signature with wipefs -b -o 0x2ba993ff000 /dev/md0.

-b here creates a backup of the deleted bits, can be useful if things go wrong.

Making a new wipefs /dev/md0 call showed the zfs_member filesystem signature again, but with another offset:

offset               type
0x2ba993fe000        zfs_member   [filesystem]

0x438                ext4   [filesystem]

I checked dirtily how many ZFS signatures were still on the drive:

hexdump -s 0x2ba99000000 -C /dev/md0 | grep "0c b1 ba"

This command printed out more than 40 signatures. I decided not to automate the wipe of these signatures and wiped out the zfs signatures one by one.

After 25 signatures wiped, wipefs stopped showing ZFS and the system automatically mounted the drive.

I made a last reboot to check that everything was ok and it was already time to take lunch.