On Wed, Aug 27, 2008 at 5:59 AM, Avi Kivity <avi at qumranet.com>
wrote:>
>> I was wondering if anyone had thought about creating a virtio file
system,
>
> There was some work using the plan 9 filesystem protocol over virtio. Not
> sure where that stands. You could base your work off that.
>
There is still work ongoing, but its mostly in short bursts. Its
operational under Linux, but I have not tested under the most recent
kernel/kvm combination. I presented some initial performance studies
at KVM Forum -- the slides are available:
http://kvm.qumranet.com/kvmwiki/KvmForum2008
The client code is already in kernel mainstream, there are patches for
an in-qemu-server, but they have not been merged yet -- I think you
can find them in the archives on the KVM mailing list. I'd be happy
to answer any questions on the code. There's plenty of room for
optimizations on both the client and server side, as well as more
complete support for the full Linux VFS interface.
>>
>> I have started to play with the idea a little myself, but my biggest
>> problem is that I have no clue how to write filesystem drives in my
Guest
>> OS, Window XP, or even if it is possible to get a Windows FileSystem
driver
>> to talk directly to a PCI device, or if some kind of dummy block device
>> would be needed.
>>
Windows is a tougher nut to crack here -- there is actually a 9P
filesystem driver for Windows (that works over TCP/IP) available from
Rangboom, but the license terms under which it was developed means
that it is closed source. I'm not sure if we could entice Rangboom to
add virtio support to their existing driver....
-eric