Re: [PATCH] RFC: V4V Linux Driver

[Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>]

On Fri, Aug 10, 2012 at 09:37:15AM +0100, Jean Guyader wrote:
> On 6 August 2012 16:28, Konrad Rzeszutek Wilk <konrad.wilk@oracle.com> wrote:
> > On Fri, Aug 03, 2012 at 11:24:20PM +0100, Jean Guyader wrote:
> >> This is a Linux driver for the V4V inter VM communication system.
> >>
> >> I've posted the V4V Xen patches for comments, to find more info about
> >> V4V you can check out this link.
> >> http://osdir.com/ml/general/2012-08/msg05904.html
> >>
> >> This linux driver exposes two char devices one for TCP one for UDP.
> >> The interface exposed to userspace are made of IOCTLs, one per
> >> network operation (listen, bind, accept, send, recv, ...).
> >
> > I haven't had a chance to take a look at this and won't until next
> > week. But just a couple of quick questions:
> >
> >  - Is there a test application for this? If so where can I get it
> 
> I have a userspace library that talks to it, I'm in the process of
> cleaning it up.
> I'll send a patch series today that would add it in xen/tools.
> 
> >  - Is there any code in the Xen repository that uses it.
> 
> The Xen support is being upstream right now, but because it needs some
> userspace kernel to be useful it's kind a chicken and a egg problem, so I'm
> trying to upstream both at the same time.
> 
> You can find the last version of the Xen patches here:
> http://lists.xen.org/archives/html/xen-devel/2012-08/msg00385.html
> 
> >  - Who are the users?
> 
> Right now we use a close but not compatible version in XenClient.
> Potentially the users
> would be anyone that is looking to for a easy way to communicate
> between VMs with
> that has a feel of TCP/UDP.
> 
> Some background info about V4V could be found here:
> http://lists.xen.org/archives/html/xen-devel/2012-05/msg01866.html
> 
> >  - Why .. TCP and UDP ? Does that mean it masquarades as an Ethernet
> >    device? Why the choice of using a char device?
> >
> 
> Because of security concerns we didn't want to rely on the Linux
> networking code because it would
> have been hard for us to prove that a V4V packet could never end up on
> your network card.
> Although we understand that there is a need for a network like driver
> and we are working on a version
> of the V4V driver that will use SKBs and expose itself as a new socket type.
> 
> In fact we asked on the LKML if it would be acceptable to add a new
> type of socket in linux for
> inter-VM communication but we are still waiting for an answer.
> http://comments.gmane.org/gmane.linux.kernel/1337472

I saw that and wasn't sure what it meant. .. Why a new family?
You didn't really explain why it is neccessary and why you could
not create message sockets for example? Or just make your driver
be an network driver.
> 
> The really nice feature about V4V is it's ability leverage all the
> existing networking programs.
> We have a libc interposer library that wraps all the networking
> functions. Here is an example
> to access a ssh server running in another domain (domid=16)
> 
> LD_PRELOAD=/usr/lib/libv4v.so ssh 1.0.0.16

Wouldn't it be just easier to not have an interposer?

I mean, it all sounds like it is for networking, so.. it would
seem like doing the full networking (or even a partial simple
implemenation) would be the way to go?


> 
> Thanks,
> Jean
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel