[dm-devel] [RFC] Message ioctl direct to device

Alasdair G Kergon agk at redhat.com
Fri Jul 9 17:51:23 UTC 2004

On Fri, Jul 09, 2004 at 11:52:00AM -0500, Kevin Corry wrote:
> On Thursday 08 July 2004 9:34 pm, Daniel Phillips wrote:

> >   - What is the advantage of passing the sector address instead
> >     a direct index into the target table?
I believe it's easier in general for applications to work with the
offset than it would be to work with the index (which will often 
be completely arbitrary).  The offset can be trusted to remain 
constant across a range of operations that change the index.  
[e.g. pvmove, remapping bad blocks etc.]

> >   - I don't see the purpose of the  __dev_status call in the
> >     target_message function

It's a standard for all the DM ioctls to return info and/or status
as appropriate.  This allows userspace to maintain its own cache
of the configuration cheaply (identical hook on return from every ioctl), 
detect problems sooner etc.  It costs very little to provide, avoids 
races, improves application efficiency etc.

> >   - Constructing a proper ioctl number is problematic since the
> >     length of the passed structure is dynamic.  
> >     The ioctl
> >     number construction kit stuff is stupid anyway, 

dm-ioctl does proper versioning itself, so it always ignores the length.

agk at redhat.com

More information about the dm-devel mailing list