Hi Adam,
I've reviewed this set of patches, applied a few of them, and
made some patches of my own that should achieve the result that I think you
were aiming for. I haven't tested them so there might still be some
issues...
Two important points.
1/ I now require getinfo_super and container_content to present a
reshape_active array in a form that md can cope with it. So a
RAID0 array that is in the middle of a reshape must appear to be
a RAID4 or RAID5 array.
I have changed super-intel to do this, but I have probably missed
some cases - please check that all supported migrations that cannot
be handled directly are handled by the new code.
With this in place, I have changed sysfs_set_array to set up all
the geometry of the array, both old and new. I think this makes
a lot of things a lot cleaner.
2/ I've revised how to handle the restart of a container-wide
migration.
As there is only one array in such a container that is actually
migrating, the mdadm which assembles it can be the one that forks
and managed the whole container.
This now happens. The metadata informs metadata that a container-wide
reshape is needed by setting ->reshape_active to '2'.
I have added a 'freeze' call where I think it should go but I wouldn't be at
all surprised if I got it wrong.
Please review the patch - which are all in my devel-3.2 branch
http://neil.brown.name/git?p=mdadm;a=log;h=refs/heads/devel-3.2
and let me know what you think.
Thanks,
NeilBrown
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
