[Virtio-fs] xfstest results for virtio-fs on aarch64

[Boeuf, Sebastien]

On Wed, 2019-10-16 at 15:14 +0100, Stefan Hajnoczi wrote:
> On Wed, Oct 16, 2019 at 03:21:52AM +0900, Chirantan Ekbote wrote:
> > On Wed, Oct 16, 2019 at 2:28 AM Boeuf, Sebastien
> > <sebastien.boeuf at intel.com> wrote:
> > > Thanks Dave for adding me to the loop here!
> > > 
> > > On Tue, 2019-10-15 at 17:26 +0100, Dr. David Alan Gilbert wrote:
> > > > * Chirantan Ekbote (chirantan at chromium.org) wrote:
> > > > > On Wed, Oct 16, 2019 at 12:58 AM Dr. David Alan Gilbert
> > > > > <dgilbert at redhat.com> wrote:
> > > > > > * Chirantan Ekbote (chirantan at chromium.org) wrote:
> > > > > > > I'm not using virtiofsd.  We have our own server for
> > > > > > > crosvm,
> > > > > > > which
> > > > > > > supports multiple queues.
> > > > > > 
> > > > > > Ah excellent; is that public anywhere?
> > > > > > 
> > > > > 
> > > > > It's currently under review here:
> > > > > https://chromium-review.googlesource.com/c/chromiumos/platform/crosvm/+/1758103
> > > > > 
> > > > > I expect to have it merged sometime this week.
> > > 
> > > Oh that's very nice to hear. We have some code from the cloud-
> > > hypervisor project where we have a vhost-user-net implementation
> > > relying on some sort of a wrapper code similar to what all vhost-
> > > user
> > > daemon implementations would reuse (libvhost-user kind of thing).
> > > 
> > > We're planning to see this wrapper code land on Rust-VMM at some
> > > point,
> > > which is why it would be interesting if you could rebase this
> > > virtio-fs
> > > daemon code on top of this small library. If you don't have the
> > > bandwidth, we could probably look at it if that's something you
> > > would
> > > be interested in.
> > > 
> > > Here is the code for the small wrapper crate:
> > > https://github.com/intel/cloud-hypervisor/blob/master/vhost_user_backend/src/lib.rs
> > > 
> > > And here is the vhost-user-net example using it:
> > > https://github.com/intel/cloud-hypervisor/blob/master/src/bin/vhost_user_net.rs
> > > 
> > 
> > Our virtio-fs server is not a vhost-user device (this was one of
> > the
> > reasons we didn't want to use virtiofsd).  It currently just runs
> > directly as a child process of crosvm (like all other crosvm
> > devices).
> > So unless I'm missing something I don't think there's much to be
> > gained by rebasing on vhost_user_backend.
> 
> I'm also interested in porting the server over to vhost-user so that
> other VMMs like QEMU and cloud-hypervisor can use it.

Yes I agree that'd be awesome!
Are you planning to start working on this soon? Just asking as I want
to avoid potential duplicated efforts :)

Thanks,
Sebastien
> 
> From a quick look at the code last night, here are things that were
> not
> carried over from virtiofsd:
>  * DAX is not supported
>  * The FUSE protocol version is a bit dated, e.g. copy_file_range()
> is
>    not supported.
>  * Requests are processed synchronously for each virtqueue and
> therefore
>    performance is limited.  The server won't saturate the queue depth
>    supported by the host kernel and hardware.
> 
> Have I missed anything?  Otherwise it looks quite close to
> passthrough_ll.c.
> 
> Stefan