How to safely remove rest of GTP?

[Ian Lepore]

On Fri, 2016-12-30 at 20:48 +0300, Andrey V. Elsukov wrote:
> On 30.12.2016 20:37, Ian Lepore wrote:
> > 
> > > 
> > > In case when you have not valid primary header, `gpart destroy`
> > > will
> > > not
> > > touch first two sectors. In you case you can wipe only last
> > > sector,
> > > like
> > > Ian suggested, but use 'gpart destroy -F da6' instead of dd.
> > > 
> > If you gpart destroy -F da6 won't that also clear the valid MBR at
> > the
> > start of the disk?  I think Slawa needs to preserve that.
> As I said, it will not touch first two sectors, where PMBR and
> primary 
> GPT headers are.
> 
> > 
> > I think it would be really useful if gpart recover could help with
> > this
> > somehow.  There is a valid mbr, and a partially-valid gpt (only the
> > backup is valid).  So gpart recover should provide a way to recover
> > the
> > gpt by refreshing the primary gpt and the pbmr,
> `gpart recover` will restore GPT using information from backup
> header 
> and table and construct new PMBR. After that you need to write
> bootcode 
> if needed.
> 
> > 
> > or recover the mbr by
> > wiping the backup gpt.
> This is what the patch does. You need to use `gpart destroy -F` to 
> CORRUPTED GPT, this command will wipe last sector where GPT backup 
> header is located. Since GPT is in CORRUPT state, the primary header 
> will not be overwrited by this command.
> 
> When both primary and backup headers and tables are valid, `gpart 
> destroy` overwites PMBR, primary and backup headers.
> 
> > 
> > I'm not sure what the syntax would be to
> > specify that, though.
> > 
> > This didn't used to be a problem until the changes that made the
> > backup
> > gpt get used automatically without any sysadmin intervention.
> For some users the problem was that they could not find their GPT,
> that 
> declares its reliability.
> 

Do you seriously believe that people are going to just do a gpart
destroy -F on a device with a valid MBR as a method of preserving that
mbr?  If that's the kind of thinking behind the gpart interface, I
guess that helps explain why it's so nearly impossible to understand
and use already.

-- Ian