On 12/9/12 5:51 PM, Dave Chinner wrote:
> On Sat, Dec 08, 2012 at 10:27:47AM -0600, Eric Sandeen wrote:
>> On 12/8/12 6:00 AM, Christoph Hellwig wrote:
>>> On Thu, Dec 06, 2012 at 04:02:16PM -0600, Eric Sandeen wrote:
>>>> Ensure that when mkfs.xfs is invoked with commandline geometry, it
>>>> triggers multidisk mode and creates more AGs.
>>>> Signed-off-by: Eric Sandeen <sandeen@xxxxxxxxxx>
>>> Looks good,
>>> Reviewed-by: Christoph Hellwig <hch@xxxxxx>
>>>> Maybe this should also test device-specified geometry, but not sure
>>>> how to do that easily within xfstests, any ideas?
>>> That would involved scsi_debug so that we could set them up. We had
>>> several ideas how to use scsi debug for these sort of tests but never
>>> really bothered to do it. If you have some spare cycles I'd love to see
>>> this sort of tests.
>> scsi_debug won't report geometry on its own today; we'd need to assemble
>> something out of scsi_debug luns I'm afraid, so it gets even more involved.
>> We do have a little bit of scsi_debug framework in xfstests today, though,
>> so if scsi_debug grew the capability it'd be easy to add.
> Does it really need scsi for this? To mkfs.xfs, all devices look the
> same, so if you want to make random geometries, wouldn't playing
> games with llopback devices and DM be sufficient?
Yeah that'd be fine too. If scsi_debug could just generate geometries
that'd be a bit easier. But it doesn't.