Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 28 Feb 2011 19:31:22 +0200
From:      Marin Atanasov Nikolov <dnaeon@gmail.com>
To:        "Andrey V. Elsukov" <bu7cher@yandex.ru>
Cc:        Johan Hendriks <Johan@double-l.nl>, freebsd-stable@freebsd.org
Subject:   Re: 8.2-RELEASE - gmirror and gpart issue. Metadata overlap?
Message-ID:  <AANLkTi=uSrqs57ihn%2B9DLwYmtSa_V9OGB_ok1Hh0zBiG@mail.gmail.com>
In-Reply-To: <4D6BD634.5070602@yandex.ru>
References:  <AANLkTimsRDV=Z4Ansd2seWUGN7C7f-5SZB=6w13kAK%2BY@mail.gmail.com> <4D694336.3090203@yandex.ru> <57200BF94E69E54880C9BB1AF714BBCBDD318F@w2003s01.double-l.local> <4D6B96E4.1050709@yandex.ru> <AANLkTinEr71fC-SjCcSBLEi=A1DXCot1yWaNd-QORGN6@mail.gmail.com> <4D6BD634.5070602@yandex.ru>

next in thread | previous in thread | raw e-mail | index | archive | help
@Johan Hendriks

> As far as i know there is no warning anywhere that you can not use gpart
> and gmirror the whole disk.

Well, there is actually :)

>From gpart(8):

     NOTE: The GEOM class PART can detect the same partition table on diffe=
r-
     ent GEOM providers and some of them will marked as corrupt. Be careful
     when choising a provider for recovering. If you did incorrect choise y=
ou
     can destroy metadata of another GEOM class, e.g. GEOM MIRROR or GEOM
     LABEL.

@Cliffton Royston

> I think this only ever worked accidentally at best.  It would work
> fine with the older fdisk-style disk partition because that doesn't
> touch the end of the disk, but any time you tell two different programs
> that they both have absolute control over the last sector on the disk
> and can write critical data there - which is what this is doing -
> that's begging for trouble.

I think I've tried this with bsdlabel(8) and gmirror(8) some time ago,
and didn't have any issues, but that
of course should be because bsdlabel does not touch the last sectors,
so gmirror would work.

@Freddie Cash

> The above process is operator error, as both your gpart and gmirror
> commands are working on the same GEOM (ad0).  You need to stack /
> layer your GEOMs (ie, do one operation on the disk, the other
> operations on the sub-parts).

You are right. Typo mistake I made in my previous mail :)

What you describe is what I actually did. gmirror'ed the disk, and
then gpart'ed it. See my very first mail, for exact steps I made to do
this.

@Andrey V. Elsukov

> What do you mean when you said 'gpart the first disk'?
> gpart(8) is the default tool to make any type of partitions and
> partition tables. The list of exact commands and at least a photo of
> screen with the panic message will be good.

Please check my first mail for all the details.

I'm currently running the system with mirrored partitions, instead of disks=
.

I'll setup a test system as well in the following days and give you
all the details of the steps, commands and output of them.. screenshot
will be made as well :)

Regards,
Marin

2011/2/28 Andrey V. Elsukov <bu7cher@yandex.ru>:
> On 28.02.2011 19:23, Marin Atanasov Nikolov wrote:
>> I can confirm as well that I get kernel panic =A0if I gpart and then
>> gmirror a disk on 8.2-RELEASE.
>>
>> To reproduce it, I just did the following:
>>
>> 1) Boot a system with a Fixit image
>> 2) Remove all gpart partitions
>> 3) gpart the first disk (ad0)
>> 4) Restored my data to the partitions from backups
>> 5) Reboot
>> 6) gmirror the ad0 disk
>>
>> And that's where I got kernel panic.
>>
>> gpart'ing the disk and the mirroring the partitions works just as
>> fine, but not when you mirror the whole disk.
>
> What do you mean when you said 'gpart the first disk'?
> gpart(8) is the default tool to make any type of partitions and
> partition tables. The list of exact commands and at least a photo of
> screen with the panic message will be good.
>
> --
> WBR, Andrey V. Elsukov
>
>



--=20
Marin Atanasov Nikolov

dnaeon AT gmail DOT com
daemon AT unix-heaven DOT org
http://www.unix-heaven.org/



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?AANLkTi=uSrqs57ihn%2B9DLwYmtSa_V9OGB_ok1Hh0zBiG>