[RFC] bhyve: modeling virtio-9p

[Daniel P. Berrangé]

On Tue, Oct 06, 2020 at 07:31:30PM +0400, Roman Bogorodskiy wrote:
>   Daniel P. Berrangé wrote:
> 
> > On Tue, Oct 06, 2020 at 06:51:52PM +0400, Roman Bogorodskiy wrote:
> > > Hi,
> > > 
> > > Recently bhyve got virtio-9p support. Modeling it appears to be pretty
> > > straight-forward, but probably I'm missing something, so decided to
> > > discuss first before proceeding with the implementation.
> > > 
> > > On the host side it looks like this:
> > > 
> > >   bhyve .... -s 25:0,virtio-9p,distfiles=/workspace/distfiles
> > > 
> > > Mounting it in a (Linux) guest looks this way:
> > > 
> > >   mount -t 9p distfiles /mnt/distfiles
> > > 
> > > lspci(8) shows it like this:
> > > 
> > > 00:1f.0 SCSI storage controller: Red Hat, Inc. Virtio filesystem
> > >         Subsystem: Red Hat, Inc. Virtio filesystem
> > >         Flags: bus master, fast devsel, latency 64, IRQ 20
> > >         I/O ports at 2200 [size=512]
> > >         Memory at c2004000 (32-bit, non-prefetchable) [size=8K]
> > >         Expansion ROM at c0007000 [virtual] [disabled] [size=2K]
> > >         Capabilities: [40] MSI-X: Enable+ Count=2 Masked-
> > >         Capabilities: [4c] MSI: Enable- Count=1/1 Maskable- 64bit+
> > >         Kernel driver in use: virtio-pci
> > > 
> > > I was thinking about presenting it like this:
> > > 
> > >   <filesystem type='mount'>
> > >     <driver type='virtiofs'/>
> > 
> > This driver type is for virtio-fuse, which is different from virtio-9p.
> > In QEMU we support type=path and type=handle as two different QEMU
> > bakends for 9p. Or simply omit  "type" entirely and it defaults to
> > 9p in QEMU. So I think you can just omit "type" for bhyve too.
> > 
> > >     <source dir='/workspace/distfiles'>
> > >     <target dir='distfiles'/>
> > >   </filesystem>
> > > 
> > > There's also an optional <readonly/> element for readonly mounts, which
> > > is also supported by bhyve.
> > 
> > Yep.
> > 
> > Also consider the "accessmode" attribute - you'll want to validate
> > whichever value(s) of that make sense given the bhyve impl of 9p.
> 
> It looks like the bhyve implementation doesn't do any uid/gid remapping,
> i.e. if I chown file on the host, I see same numeric ids (but they
> corresponding to different users/groups on my test Linux guest compared
> to the FreeBSD host). The same happens when I chown files in the guest
> and verify results in the host.
> 
> Sounds like 'passthrough' is the right choice for this behavior?

Yeah, that exposes the host uid/gid with no changes


Regards,
Daniel
-- 
|: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange :|
|: https://libvirt.org         -o-            https://fstop138.berrange.com :|
|: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange :|