Safe way to repair corrupted GPT partition table?

[Warren Block]

On Sat, 19 Jan 2013, Bob Willcox wrote:

> On Sat, Jan 19, 2013 at 07:25:09AM +0700, Erich Dollansky wrote:
>> Hi,
>>
>> On Fri, 18 Jan 2013 14:08:25 -0600
>> Bob Willcox <bob at immure.com> wrote:
>>
>>> Is there a way to repair a GPT partition table that has gotten
>>> corrupted (following a system hang during heavy I/O to a ZFS
>>> filesystem)?
>>>
>> I would use a hex editor. Of course, try it out on another disk before
>> working on that disk. You can even copy the data with dd from the other
>> disk after you are sure it will work. Of course, the size must match or
>> must be made matching.
>>
>> Ok, it is not a safe way but it is a working way.
>
> Have to say I was hoping that there was some programatic way to do this.
> Certainly if I go down this path I'll have to practice on a disk that doesn't
> contain data that I care about. Getting the size right as this is the only
> disk of this size I have. (Actually, it's an Areca RAID 5 Volume Set.)

If the primary table at the start of the disk is okay, 'gpart recover' 
can copy it to the backup table at the end of the disk.  I thought it 
would do that the other way around also.  Neither table should be 
affected by a power failure, as they are almost never written.

How it got into a state where it could be recognized as GPT but not 
recoverable, don't know.  Could be the disk device (ada0) was given to 
ZFS rather than the partition (ada0p1).  ZFS is supposed to leave some 
space at the end of the disk to allow for slightly differing nominal 
disk sizes, which could have left the backup GPT table intact.

ZFS has its own metadata, so it's not necessary to partition a drive 
with GPT unless you want to put more than one partition on it, or maybe 
control the size of space used.