Booting from ZFS raidz

krad kraduk at
Wed Jan 6 20:37:18 UTC 2010

2010/1/6 Anselm Strauss <amsibamsi at>

> Hi,
> I'm experimenting with a ZFS only system and booting from it in VirtualBox.
> Thanks to various mails and forum posts from the net I have a working
> scenario with booting from a ZFS mirror. However, I can't get the thing to
> work with raidz with the exactly same setup, except that the pool is now
> raidz instead of mirror and there is one more disk. I feel sure I have all
> the stuff with partitioning, boot loader installation, etc. right. I tested
> this with version 8.0-RELEASE on 64bit.
> Now, before I go into detailed explaining, is raidz really supported? I
> always get the following error after it says "FreeBSD/i386 bootstrap loader,
> Revision 1.1":
> ZFS: i/o error - all block copies unavailable
> ZFS: can't read MOS object directory
> (repeats a lot)
> Can't find root filesystem - giving up
> can't load 'kernel'
> I think the "MOS" message comes from zfs_mount_root() in
> /usr/src/sys/boot/zfs/zfsimpl.c. I asume that is the point when /boot/loader
> has been loaded and now wants to load the kernel into memory. After that
> error I'm in the loader prompt. When I try to load any file I always get the
> same error as above.
> Anyone any ideas?
> Thanks,
> Anselm_______________________________________________
> freebsd-questions at mailing list
> To unsubscribe, send any mail to "
> freebsd-questions-unsubscribe at"

Opensolaris doesnt support booting off raidz yet so id be surprised if you
managed to, as i doubt all the relevent code is in the loader, and robust
enough yet. I have seen a few hacks mentioned in places that might get it to
work, bit these are unsupported and might flake out at any time.

Also why do you need a raidz for the os? It implys you might be mixing data
on the pool as well. This isnt best practice, and you are best off having a
separate pool for os and data. If you have 3+ drives, gpt it into 3 or 4
chunks dependent on whether you want swap on a zvol. Have and x way mirror
for the os and then the last and biggest gpt slice use for your raidz data.
Better still have the os and data on separate spindles

More information about the freebsd-questions mailing list