ZFS woes

[Dick Hoogendijk]

  On 11-8-2010 0:52, Dale Scott wrote:
>>> wiped out the firt mb; i used sysinstall to create a fbsd slice; wiped
>>> it out again; booted knoppix to create an EFI / GPT; booted into
>>> opensolaris and created a zpool (v14), but nothing, nothing 
>>> did the trick.
> I was doing a vanilla fbsd install recently using a couple re-claimed 250GB IDE drives. The install completed without errors, but after reboot GEOM complained bitterly about the secondary GPT table on the boot drive being corrupted or invalid, and unrecoverable corrupted or invalid GPT tables on the 2nd drive. By trying something like above, I was able to get the system drive to rebuild the secondary GPT table, but nothing worked on the second drive. Google told me a targeted approach was technically possible (by calculating exactly where a specific drive stores its GPT metadata and zeroing just that bit), but also that the broader solution of zeroing out the entire drive would be faster for me than figuring out the calculation (about 18 hrs to zero the entire drive, at least it was mostly while sleeping): "dd if=/dev/zero of=/dev/ad3 bs=64K" (no idea if the block size is optimal or even relevant).

I did not want to overwrite two drives with /dev/zero, so I created a 
mirror with gmirror yesterday, folowing the steps from the freebsd manual.

After it was completed I just did:

# gmirror stop gm0
# gmirror clear /dev/ad12
# gmirror clear /dev/ad14

# dd if=/dev/zero of=/dev/ad12 bs=1m count=1
# dd if=/dev/zero of=/dev/ad14 bs=1m count=1
# kldload zfs
# zpool create store mirror ad12 ad14
## Wrote some data to /store ##
# zpool scrub store
# zpool history store
## No More Errors !!! ##

I guess creating the gmirror metadata / mirror and removing it cleared 
all data which caused me so much trouble.
I happely removed the geom_mirror_load="YES" with zfs_load="YES" and 
have what I wanted: FreeBSD/zfs