-----Original Message-----
From: Neil Brown [mailto:neilb@suse.de]
Sent: Friday, November 13, 2009 3:49 AM

On Thursday November 12, Maciej.Trela@intel.com wrote:

quoted

Hello Neil,
I'm sending the patch that is connected with intel metadata

compatibility.

quoted

The new code checks for existing partitions when creating an array.
If the metadata overwrites the last partition, the warning is shown to

the user. The patch supports only MBR and GPT partition table types.

quoted

Despite the patch was implemented as IMSM compatibility issue I've

decided to include the partition check for all metadata types as a generic
code.

quoted

I would appreciate any feedback.

Sorry for not responding the first time you posted this.

I agree with the principle.  However I would rather not add code for
reading different partition types to mdadm.  It would be much nicer if
that code came from a library.
Is there a good library that we could use?
libparted might be a possibility, but it seems rather over-weight.
/lib/libparted-1.9.so.12.0.0 is larger than /sbin/mdadm !!!

Are there any other libraries for reading different partition tables?
I guess we can open-code it if we have to but I'd be happy if another
solution could be found.

Hello Neil,
I made some research but unfortunately I couldn't find any other library except libparted.
Assuming there are no other possibilities, do you think we could stay with the current solution or should I switch the patch to libparted?

I have the impression that with libparted this simple patch would be much bigger (given the code size)...

I should probably change to using libblkid to check for the various
filesystem types...

Is this check really relevant for anything other RAID1 array.  For any
other level, the fact that the metadata lies inside or outside the
current partitions is quite irrelevant as the old data will not be
useful anyway.

Yes, that's right.
However, this check would be also relevant when creating a container with external metadata. Unfortunately in this case we do not know the raid level...
Thus, maybe the following condition would be appropriate:

if ((raid_level == CONTAINER) || ( !(external) && raid_level==1))
{
	Check partitions
}

And I don't think it is sensible to report "unknown partition table".
If there is no partition table, there is nothing to check.

NeilBrown

The idea behind this was that: maybe there is a partition table that we do not support so we should warn the user anyway, but probably this would be too paranoid... I'll change that.

Regards.
Maciek Trela.