Handbook mirroring section

[Warren Block]

On Sun, 10 Jun 2012, Hiroki Sato wrote:

> Warren Block <wblock at wonkity.com> wrote
>  in <alpine.BSF.2.00.1206060454200.13150 at wonkity.com>:
>
> wb> On Wed, 6 Jun 2012, Hiroki Sato wrote:
> wb> > BTW, do you (or anyone) know the common failure pattern when trying
> wb> > to use GPT + gmirror of the whole disk?  IIRC it was a metadata
> wb> > corruption but I do not remember when it happens.
> wb>
> wb> When gmirror is used to mirror two disks, metadata goes in the last
> wb> block of each.  If GPT partitioning is used on the mirror, the GPT
> wb> secondary partition table overwrites the mirror metadata because, by
> wb> specification, GPT puts that secondary table at the end of the
> wb> physical drive, rather than inside the gmirror logical device.
>
> I think the interpretation that "GPT specification says the backup
> header must be at the last LBA of the *physical* disk" is just making
> things too harder.  The problem is that BIOS and the loader does not
> support a gmirror logical volume, and the incompletely-configured
> volume is recognized for a short time during a boot due to it.

Agreed.  This doc should point out that BIOS boots from one of the 
individual drives, not the mirror.

> The location of the backup header is recorded in the Backup LBA field
> in the primary header, so everything will work fine as long as the
> pre-boot programs support it and the primary header is not corrupted.
> I do not think this situation is against standards conformance though
> it is suboptimal.

It would certainly make things a lot simpler.  gpart would need 
modification.

> If we can teach GEOM to BIOS and/or UEFI it would be the best, but it 
> is not likely.

> What I want to know is whether there are some practical difficulties
> when we apply the whole-disk mirroring procedure for MBR to a
> GPT-based system.  Overwrite of the last LBA should not happen if
> FreeBSD is the only one system on the disk.  One common failure
> scenario is when the primary header is corrupted.  If there are more,
> I want to know it.

I tried a whole-disk GPT mirror for a short time, and had no problems 
other than gptboot complaining about the corrupted secondary header.  It 
was not an exhaustive test, though.