Virtio File System.

Virtio File System.

[scott.kvm]

All

I am new to this mailing list, so please let me know if I am in the wrong 
place or asking questions in a way I shouldn't.

I was wondering if anyone had thought about creating a virtio file system, 
rather than using a block device and hence a single file on the host OS as 
the virtio block device already does, but map file system calls on the 
guest OS to file system calls on the host OS through the virtio interface.  
In this way the Guest OS could share files with the Host OS without needing 
to use Samba, which from my current experiences is rather slow.

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.

My host OS is Linux, Guest OS is Windows XP.  Maybe I should start with a 
Linux/Linux setup and get that working first ?

Any comments, suggests, problems that you can see with this idea would be 
great.

Thanks,

Scott.

Re: Virtio File System.

[Avi Kivity]

scott.kvm@scottrix.co.uk wrote:
>
> I am new to this mailing list, so please let me know if I am in the 
> wrong place or asking questions in a way I shouldn't.
>

Here is fine.  virtualization@lists.linux-foundation.org is also a good 
place since virtio is not kvm specific.

> I was wondering if anyone had thought about creating a virtio file 
> system, rather than using a block device and hence a single file on 
> the host OS as the virtio block device already does, but map file 
> system calls on the guest OS to file system calls on the host OS 
> through the virtio interface.  In this way the Guest OS could share 
> files with the Host OS without needing to use Samba, which from my 
> current experiences is rather slow.

There was some work using the plan 9 filesystem protocol over virtio.  
Not sure where that stands.  You could base your work off that.

>
> 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.
>

It ought to be possible; the Windows smb stack works without a block device.

> My host OS is Linux, Guest OS is Windows XP.  Maybe I should start 
> with a Linux/Linux setup and get that working first ?

That's definitely a good idea.

-- 
error compiling committee.c: too many arguments to function

Re: Virtio File System.

[Eric Van Hensbergen]

On Wed, Aug 27, 2008 at 5:59 AM, Avi Kivity <avi@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

Re: Virtio File System.

[Anthony Liguori]

Eric Van Hensbergen wrote:
> On Wed, Aug 27, 2008 at 5:59 AM, Avi Kivity <avi@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....
>   

If someone is enterprising, then 
http://uranus.it.swin.edu.au/~jn/linux/ext2ifs.htm is probably a good 
start as an example of how to build an IFS driver for Windows.

Would probably want to implement extensions to the 9p protocol to 
support this too.  And mapping file system semantics between Windows and 
Unix is hugely complicated.  In all honesty, CIFS over virtio-net is a 
better solution since Samba has already done the hard work of getting 
the details right.

Regards,

Anthony Liguori

>          -eric
> --
> To unsubscribe from this list: send the line "unsubscribe kvm" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>   

Re: Virtio File System.

[Anthony Liguori]

Eric Van Hensbergen wrote:
> On Wed, Aug 27, 2008 at 5:59 AM, Avi Kivity <avi@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....
>   

If someone is enterprising, then 
http://uranus.it.swin.edu.au/~jn/linux/ext2ifs.htm is probably a good 
start as an example of how to build an IFS driver for Windows.

Would probably want to implement extensions to the 9p protocol to 
support this too.  And mapping file system semantics between Windows and 
Unix is hugely complicated.  In all honesty, CIFS over virtio-net is a 
better solution since Samba has already done the hard work of getting 
the details right.

Regards,

Anthony Liguori

>          -eric
> --
> To unsubscribe from this list: send the line "unsubscribe kvm" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>   

Re: Virtio File System.

[Eric Van Hensbergen]

On Wed, Aug 27, 2008 at 5:59 AM, Avi Kivity <avi@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

Re: Virtio File System.

[Avi Kivity]

scott.kvm@scottrix.co.uk wrote:
>
> I am new to this mailing list, so please let me know if I am in the 
> wrong place or asking questions in a way I shouldn't.
>

Here is fine.  virtualization@lists.linux-foundation.org is also a good 
place since virtio is not kvm specific.

> I was wondering if anyone had thought about creating a virtio file 
> system, rather than using a block device and hence a single file on 
> the host OS as the virtio block device already does, but map file 
> system calls on the guest OS to file system calls on the host OS 
> through the virtio interface.  In this way the Guest OS could share 
> files with the Host OS without needing to use Samba, which from my 
> current experiences is rather slow.

There was some work using the plan 9 filesystem protocol over virtio.  
Not sure where that stands.  You could base your work off that.

>
> 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.
>

It ought to be possible; the Windows smb stack works without a block device.

> My host OS is Linux, Guest OS is Windows XP.  Maybe I should start 
> with a Linux/Linux setup and get that working first ?

That's definitely a good idea.

-- 
error compiling committee.c: too many arguments to function