Re: [PATCH] show an error instead of segfaulting on grub-probe -t partmap on a unsynced raid

[Felix Zielcke <fzielcke@z-51.de>]

Am Freitag, den 01.08.2008, 15:36 +0200 schrieb Robert Millan:

> The existing code is confusing.  What is total_devs for?  If we need to
> iterate up to 32, sounds like this variable is pointless?
> 
> If it's useless, it should be removed (but maybe it isn't!).
> 

total_devs is a fast check if the array is readable.

On fully synced raids the disks should be always in linear continues
order starting with 0 in the mdraid superblock.

If I added the sdb1 back to the md0 then mdadm did reassemble them to 0
and 1
Whereas when I removed it then disk 1 was the working one and disk 2 the
removed one.

So at least for raid1 it's not safe to just check disk 0 if the array
has only one device (which is totally legal on raid1)

For raid0 you always must have a fully synced raid.

On raid5 one disk can be missing, thanks to VMware and that I have enoug
time I can find it out how it works with raid5

By the way, this 32 seems to be oversafe
mdadm(8) says super 0.90 supports only 28 devices in one array

With implementing super 1.X this will be a problem because mdadm(8)
doestn't say any limit at all.