Re: Device-mapper filesystem interface

[Greg KH <greg@kroah.com>]

On Thu, May 22, 2003 at 09:50:36AM +0100, Joe Thornber wrote:
> I thought I'd kick off a thread concerning the filesystem interface
> for device-mapper after it came up on last nights 'must-fix' meeting.
> 
> To recap:
> 
> Alasdair Kergon and I spent a lot of time thinking last autumn about
> how to best map the dm semantics onto an fs.  The end result was this
> very rough and ready patchset:
> 
> http://people.sistina.com/~thornber/patches/2.5-unstable/2.5.51/2.5.51-dmfs-1.tar.bz2
> 
> The reception was not favourable.  People didn't like the way creating
> a directory was analagous to creating a device, or the fact that these
> device directories were pre-populated with table, status and
> dependency files.  Gregkh was the only person who put forward
> alternatives ideas (sysfs), and I don't think even he had thought
> through how all of the dm functionality was going to be mapped.  eg,
> with dmfs as it stands the 'wait for event' ioctl has translated into
> a poll on the status file, ie wait until the status file changes - I
> think this is neat.

Yeah, I went down the sysfs path for a while, then got distracted by
other issues (driver core, etc.)  If you need this feature, then yes, a
dmfs does make sense to have, and not use sysfs.

I'm not opposed to your implementation, it was just a bit strange at
first glance.  Care to update your old patch for this?

thanks,

greg k-h