Autostarting arrays with V1 metadata

Autostarting arrays with V1 metadata

[Ben Winslow]

I've noticed that (as of 2.6.25, at least) the kernel doesn't autostart 
arrays with V1 metadata.  In autostart_arrays(), md_import_device() is 
called for 0.90 arrays only, and there's currently no sort of 
autodetection of the superblock position for V1 metadata.

Is this simply a matter of the code being unwritten, or are there other 
reasons that V1 arrays aren't autostarted (such as the SB format being 
unfinished)?  My only real motivation to use V1 metadata is the unhashed 
homehost field, which I'll give up if the SB format hasn't stabilized yet.

Thanks,
-- 
Ben Winslow <rain@bluecherry.net>

Re: Autostarting arrays with V1 metadata

[Neil Brown]

On Wednesday May 7, rain@bluecherry.net wrote:
> I've noticed that (as of 2.6.25, at least) the kernel doesn't autostart 
> arrays with V1 metadata.  In autostart_arrays(), md_import_device() is 
> called for 0.90 arrays only, and there's currently no sort of 
> autodetection of the superblock position for V1 metadata.
> 
> Is this simply a matter of the code being unwritten, or are there other 
> reasons that V1 arrays aren't autostarted (such as the SB format being 
> unfinished)?  My only real motivation to use V1 metadata is the unhashed 
> homehost field, which I'll give up if the SB format hasn't stabilized yet.

V1.x metadata (deliberately) doesn't contain sufficient information to
allow in-kernel autodetect.  In particular it doesn't store the 'minor
number' information so the kernel wouldn't know which md device to
use.

However it is quite possible to assemble V1.x arrays using mdadm in an
initrd.  This is the preferred method.

Ofcourse, if the array doesn't hold the root filesystem, you can just
assemble it as part of the regular boot scripts.

NeilBrown

Re: Autostarting arrays with V1 metadata

[David Greaves]

Ben Winslow wrote:
> I've noticed that (as of 2.6.25, at least) the kernel doesn't autostart
> arrays with V1 metadata.  In autostart_arrays(), md_import_device() is
> called for 0.90 arrays only, and there's currently no sort of
> autodetection of the superblock position for V1 metadata.
> 
> Is this simply a matter of the code being unwritten, or are there other
> reasons that V1 arrays aren't autostarted (such as the SB format being
> unfinished)?  My only real motivation to use V1 metadata is the unhashed
> homehost field, which I'll give up if the SB format hasn't stabilized yet.
> 
> Thanks,

Also, FYI:
  http://linux-raid.osdl.org/index.php/Autodetect

and
  http://linux-raid.osdl.org/index.php/RAID_Boot

Re: Autostarting arrays with V1 metadata

[Bryan Mesich]

On Thu, May 08, 2008 at 01:05:44PM +1000, Neil Brown wrote:

[snip...]

> However it is quite possible to assemble V1.x arrays using mdadm in an
> initrd.  This is the preferred method.


I wasn't aware there's a "preferred method" to assembling an
array at boot time.  Could you please elaborate as to why
assembling in user space would be better? (besides having the
ability to assemble non 0.9 metadata)

Just curious :-)

Bryan