Re: File sharing over virtio-9p

On Tue, Mar 25, 2025 at 10:35 AM Greg A. Woods <[email protected]> wrote:
>
> At Thu, 24 Oct 2019 13:32:59 +0900, Ryota Ozaki <[email protected]> wrote:
> Subject: Re: File sharing over virtio-9p
> >
> > > A NetBSD guest can mount the exported directory with mount_9p.
> > >
> > >     mount_9p -cu /dev/vio9p0 /mnt/9p
>
> So I finally got a chance to try this the other day after uncommenting
> the config in GENERIC:
>
> diff --git a/sys/arch/amd64/conf/GENERIC b/sys/arch/amd64/conf/GENERIC
> index b9d864680b40..ea0770936914 100644
> --- a/sys/arch/amd64/conf/GENERIC
> +++ b/sys/arch/amd64/conf/GENERIC
> @@ -1133,7 +1133,7 @@ viocon* at virtio?                        # Virtio serial device
>  vioif* at virtio?                      # Virtio network device
>  viornd* at virtio?                     # Virtio entropy device
>  vioscsi* at virtio?                    # Virtio SCSI device
> -#vio9p*        at virtio?                      # Virtio 9P device
> +vio9p* at virtio?                      # Virtio 9P device
>
>  # Hyper-V devices
>  vmbus*         at acpi?                # Hyper-V VMBus
>
>
> It seemed to work wonderfully and seemed a bit faster than a QEMU usb
> umass block device (e.g. emulated CD/ROM or emulated USB stick) in
> scenario I tried it in, which was macOS with UTM (and QEMU doing the
> I/O), and with the shared directory in an APFS filesystem on fast SSD.
>
> [     1.053094] virtio4 at pci0 dev 8 function 0
> [     1.053094] virtio4: 9P transport device (id 9, rev. 0x00)
> [     1.053094] vio9p0 at virtio4: features: 0x10000001<INDIRECT_DESC,MOUNT_TAG>
> [     1.053094] virtio4: allocated 24576 byte for virtqueue 0 for vio9p, size 128
> [     1.053094] virtio4: using 16384 byte (1024 entries) indirect descriptors
> [     1.053094] vio9p0: tagged as share
>
>
> I used the mount to do an upgrade of the VM from the release just built
> on macOS.  (as an aside it didn't go as fast as I thought it should and
> 'systat vmstat' showed it wasn't keeping the devices fully occupied)

vio9p is not optimized for performance yet, for example, it can serve
each request
synchronously.

>
> However I did another reboot to get back to multi-user mode in the VM
> and I'm now greeted with the following error:
>
>         # mount_9p -cu /dev/vio9p0 /9pfs/
>         mount_9p: Rattach not received, got 107
>
> The kernel message in the new kernel look identical:
>
> [     1.016612] virtio4 at pci0 dev 8 function 0
> [     1.016612] virtio4: 9P transport device (id 9, rev. 0x00)
> [     1.016612] vio9p0 at virtio4: features: 0x10000001<INDIRECT_DESC,MOUNT_TAG>
> [     1.016612] virtio4: allocated 24576 byte for virtqueue 0 for vio9p, size 128
> [     1.016612] virtio4: using 16384 byte (1024 entries) indirect descriptors
> [     1.016612] vio9p0: tagged as share
>
>
> The only thing that it was a VM reboot, and maybe QEMU (which didn't
> restart) gets left in an unusable state?
>
> Trying a "cold" reboot now.....
>
> Ah ha, yes, that was the problem:
>
>         # mount_9p -cu /dev/vio9p0 /9pfs
>         # df
>         Filesystem      1K-blocks         Used        Avail %Cap Mounted on
>         /dev/dk1         27322780      6807586     19149056  26% /
>         /dev/dk3          5082862       601362      4227358  12% /var
>         /dev/dk5         20332078        62498     19252978   0% /home
>         /dev/dk4         60996394       612850     57333726   1% /usr/pkg
>         kernfs                  1            1            0 100% /kern
>         ptyfs                   1            1            0 100% /dev/pts
>         procfs                  4            4            0 100% /proc
>         tmpfs             2096140            4      2096136   0% /var/shm
>         /dev/vio9p0             0            0            0 100% /9pfs
>         # mount
>         /dev/dk1 on / type ffs (log, local)
>         /dev/dk3 on /var type ffs (log, local)
>         /dev/dk5 on /home type ffs (log, local)
>         /dev/dk4 on /usr/pkg type ffs (log, local)
>         kernfs on /kern type kernfs (local)
>         ptyfs on /dev/pts type ptyfs (local)
>         procfs on /proc type procfs (local)
>         tmpfs on /var/shm type tmpfs (local)
>         /dev/vio9p0 on /9pfs type puffs|9p
>
>
> I wonder is there anything NetBSD, i.e. vio9p(4), can do to cleanly shut
> down the 9p connection and allow it to be reused without having to
> restart the host QEMU process?  This would be nice as then a VM could
> reboot itself without having to use the host tools to restart.

umount /9pfs before rebooting might help you.

vio9p itself is just a bridge between a client and a server of 9p, so
it doesn't help for shutting down the 9p connection over it, I think.

  ozaki-r