dm-userspace

Hi List,

As you all know, I'm working on a userspace-controlled CoW
implementation using device-mapper.  The idea is to offer a
pseudo
device through device-mapper that has CoW behavior, but
where the
block allocation decisions are made from userspace.

My thoughts are that it might be best to abandon the concept
of the
"dm-cow" target and instead work on a
"dm-userspace" target.  The
userspace cow application I'm working on would remain
mostly the
same.  Similarly, dm-userspace would look almost identical
to my
dm-cow does right now, but more generic.  The target would
simply
present the details of the data passed to the map() function
to
userspace, which would respond with a target device and
sector of
where to send the request.

A generic dm-userspace target would allow for testing of new
algorithms (RAID, CoW, etc) from a userspace application, as
well as
some more interesting things involving distributed
applications.  Just
like FUSE allows for some neat (although not necessarily
high-performance) tricks, dm-userspace could allow the same
thing for
block devices.

Would the device-mapper maintainers be interested in
accepting
something like dm-userspace upstream?

-- 
Dan Smith
IBM Linux Technology Center
Open Hypervisor Team
email: danmsus.ibm.com
--
dm-devel mailing list
dm-develredhat.com
http
s://www.redhat.com/mailman/listinfo/dm-devel

On 4/19/06, Dan Smith <danmsus.ibm.com> wrote:
>
> A generic dm-userspace target would allow for testing
of new
> algorithms (RAID, CoW, etc) from a userspace
application, as well as
> some more interesting things involving distributed
applications.  Just
> like FUSE allows for some neat (although not
necessarily
> high-performance) tricks, dm-userspace could allow the
same thing for
> block devices.
>
> Would the device-mapper maintainers be interested in
accepting
> something like dm-userspace upstream?
>

It seems like this would be really useful for prototyping
and
debugging new device mapper modules (like the dm-cache ideas
I posted
about a few weeks back).  At the very least, I'd be
interested in
using it?

A couple of questions:

1) how would you handle permissions?  IIRC FUSE allows
normal users to
bind their own FUSE userspace file systems, would something
similar
happen for dm-userspace or would even binding a userspace
device-mapper module require root?

2) How close would the userspace API be to the kernel
device-mapper
API?  It'd be nice to have something close so that
userspace code
could easily be migrated into the kernel (for performance
reasons) as
appropriate.

3) When do you think you'll be able to post a patch for
RFC?

           -eric

--
dm-devel mailing list
dm-develredhat.com
http
s://www.redhat.com/mailman/listinfo/dm-devel