gmirroring slices

[Miroslav Lachman]

Lorenzo Perone wrote:
>
> Hello,
>
> I was wondering if anyone could give me an advice on how viable and
> reliable it is, to use gmirror on a slice of an MBR-style partitioned
> disk, and use the second slice(s) within a zpool.
>
> I remember a discussion here on where metadata is kept (always at the
> end of the disk as opposed to the end of the given consumer?), so I
> wasn't sure about how much of a good idea this might be.

I think metadata is stored at the end of the provider (slice in this 
case), but I am not a GEOM expert.

> The reason I'd
> like to have it like this is, that I had mixed bad experiences in the
> effort of using ZFS as a boot and root volume, so I'd rather keep a
> traditional slice for booting/rooting, and a zpool for the production
> jails on that machine.
>
> The example would be
>
> provider: mirror/gm0
> consumers: ad6s1 and ad8s1
>
> zpool mirror made out of
> ad6s2 and ad8s2

I am running following setup for year without any configuration problems

# gmirror status
        Name    Status  Components
mirror/gms1  COMPLETE  ad4s1
                        ad6s1

# zpool status
   pool: tank
  state: ONLINE
  scrub: none requested
config:

         NAME        STATE     READ WRITE CKSUM
         tank        ONLINE       0     0     0
           mirror    ONLINE       0     0     0
             ad4s2   ONLINE       0     0     0
             ad6s2   ONLINE       0     0     0

The first slice is 20GB partitioned as usual:
# mount -t ufs
/dev/mirror/gms1a on / (ufs, local)
/dev/mirror/gms1e on /usr (ufs, local, soft-updates)
/dev/mirror/gms1d on /var (ufs, local, nosuid, soft-updates)
/dev/mirror/gms1f on /tmp (ufs, local, noexec, nosuid, soft-updates)

The rest (450GB) is used in ZFS mirrored zpool for jails (each jail has 
its own filesystem)

> while experimenting, I got into the problem that gmirror label -v -b
> round-robin gm0 ad6s1 got a permission denied (even with sysctl
> kern.geom.debugflags=16/17). Any hints on what can cause this (I might
> have screwed up something with fdisk/bsdlabel, but after doublechecking
> I wonder what it could be..)

I did it in non-standard way - converting already installed system on 
one disk to mirrored. So when I was in system running off ad6 I created 
two slices on ad4, setup gmirror gms1 from first slice of ad4, create 
partitions, newfs, mount it and transfer files from running system by 
dump & restore, edit fstab. Then I rebooted system from gms1, destroy 
content of ad6, create slices on ad6 and insert first slice in to gms1.
After this I had ad4s2 and ad6s2 ready for zpool.
All was done remotely through ssh.

Miroslav Lachman