8.2-RELEASE - gmirror and gpart issue. Metadata overlap?

[Andrey V. Elsukov]

On 28.02.2011 17:23, Joan Picanyol i Puig wrote:
> http://lists.freebsd.org/pipermail/freebsd-geom/2010-July/004278.html
> 
>     1. glabeled discs
>     2. two of this previous discs are gmirrored
>     3. this mirror is gparted 
> 
> We got distracted by side issues, but I believe this issue is still
> present, even though I have been unable to investigate further (it might
> well be that GPT specification is incompatible with gmirror'ed disks).
> For your convinience the interesting bits from that thread are:
> 
>     GEOM: da1: the secondary GPT table is corrupt or invalid.
>     GEOM: da1: using the primary only -- recovery suggested.
> 
>     So, what explains the messages? gpart probes the disk before gmirror
>     (or at least it prints later on dmesg), but since the offset is in
>     the header, it should not even know about the gmirror + glabel part.

When you create gmirror on the whole disk you have got:

                whole disk da0
+----------------------------------------+----+
|                 mirror/gm0             | MD |
+----------------------------------------+----+

MD - gmirror's metadata in the last sector of da0.
Now when you create GPT on the mirror/gm0:

               mirror/gm0
+-+-----+--------------------------+-----+
| | GPT |                          | GPT |
+-+-----+--------------------------+-----+

When mirror/gm0 provider is tasted all is good.
You got mirror with GPT without any warnings.
But when da0 provider is tasted you got this:

               da0
+-+-----+--------------------------+-----+----+
| | GPT |                          | GPT |    |
+-+-----+--------------------------+-----+----+

This provider has one sector in the end of the disk
with unknown data, but there should be secondary
GPT header. First of you got a message from gptboot
about corrupt GPT. The second message is from GEOM_PART
that it does not like that secondary GPT located not
in the end of disk.

This message
>     GEOM: da1: the secondary GPT table is corrupt or invalid.
>     GEOM: da1: using the primary only -- recovery suggested.

does mean that your secondary GPT header or table is lost,
or it is in disagree with primary one. AFAIR, it may mean different
things in 8.1 and 8.2.

-- 
WBR, Andrey V. Elsukov

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: OpenPGP digital signature
Url : http://lists.freebsd.org/pipermail/freebsd-stable/attachments/20110228/b0bbe60e/signature.pgp