Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 19 Jan 2006 11:56:08 -0600
From:      Craig Boston <craig@feniz.gank.org>
To:        freebsd-stable@freebsd.org
Subject:   Re: Gmirror'd SWAP: no dump?
Message-ID:  <20060119175608.GA911@nowhere>
In-Reply-To: <20060119145342.GA86397@voi.aagh.net>
References:  <20060119121602.GC19305@voodoo.schug.net> <00c301c61cf2$8e9a7b70$68c8a8c0@lerctr.org> <20060119123428.GD19305@voodoo.schug.net> <20060119145342.GA86397@voi.aagh.net>

next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, Jan 19, 2006 at 02:53:42PM +0000, Thomas Hurst wrote:
> > IIRC I had the problem that 'gmirror insert' without '-h' not always
> > inserted the slice specified by the entire block device (e.g.
> > /dev/ad4s1 vs. /dev/ad4). Apparently there is some auto detection
> > code and/or gmirror cannot differ correctly, but that's just a
> > guess. Specifing '-h' fixed it in my case.
> 
> As I understand it, gmirror writes its metadata on the last sector of
> the provider; when tasting devices it will look at the last sector of
> ad4, find the metadata and use that as the provider for your mirror; you
> can either hardcode the provider name there to override it, or make the
> slice 1 sector smaller so gmirror tastes ad4, finds nothing, then goes
> on to taste ad4s1 correctly.

This was fixed for most cases by adding the size of the provider to the
metadata.  ad4 should be a different size than ad4s1 as the partition
table has to go somewhere...

Looks like it was fixed in HEAD in Feb 2005, and MFC'd to 5.x in March.
Shouldn't have ever been a problem for 6.x release.

http://www.freebsd.org/cgi/cvsweb.cgi/src/sys/geom/mirror/g_mirror.c#rev1.19.2.8

Craig



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20060119175608.GA911>