VIRTIO_VSOCK_OP_REQUEST with the same src_port from a guest

VIRTIO_VSOCK_OP_REQUEST with the same src_port from a guest

[Roman Kiryanov]

Hi virtio-dev,

we (Android Studio Emulator) have our own implementation of
virtio-vsock and we recently found that the virtio spec does not cover
the situation if a guest sends a VIRTIO_VSOCK_OP_REQUEST packet with
the same src_port that is already registered on the device side.

This sounds like an error to me and I don't think I ever saw this
happening, but it would be nice if the spec explicitly mentioned how
this situation should be handled on the device side.

The spec does mention "A VIRTIO_VSOCK_OP_RST reply is sent if ... the
destination has insufficient resources to establish the connection",
but I am not sure if this includes the duplicate src_port case. I
think this means the previously established connection will be broken.

Thank you.

Regards,
Roman.

Re: VIRTIO_VSOCK_OP_REQUEST with the same src_port from a guest

[Michael S. Tsirkin]

On Mon, Jul 01, 2024 at 10:35:57AM -0700, Roman Kiryanov wrote:
> Hi virtio-dev,
> 
> we (Android Studio Emulator) have our own implementation of
> virtio-vsock and we recently found that the virtio spec does not cover
> the situation if a guest sends a VIRTIO_VSOCK_OP_REQUEST packet with
> the same src_port that is already registered on the device side.
> 
> This sounds like an error to me and I don't think I ever saw this
> happening, but it would be nice if the spec explicitly mentioned how
> this situation should be handled on the device side.
> 
> The spec does mention "A VIRTIO_VSOCK_OP_RST reply is sent if ... the
> destination has insufficient resources to establish the connection",
> but I am not sure if this includes the duplicate src_port case. I
> think this means the previously established connection will be broken.
> 
> Thank you.
> 
> Regards,
> Roman.

Another way is just to forbid this from happening.

How does this happen on the driver side?
-- 
MST

Re: VIRTIO_VSOCK_OP_REQUEST with the same src_port from a guest

[Roman Kiryanov]

Hi Michael, thank you for looking.

as I mentioned, I have never seen this happening.

But this code path (on the device side) exists (e.g. insertion of
<dst_port, src_port> to a
hash table says that the key already exists) even if it is never taken
and I would be nice
if the virtio-spec mentioned explicitly what to do there.

Regards,
Roman.


On Mon, Jul 1, 2024 at 11:44 AM Michael S. Tsirkin <mst@redhat.com> wrote:
>
> On Mon, Jul 01, 2024 at 10:35:57AM -0700, Roman Kiryanov wrote:
> > Hi virtio-dev,
> >
> > we (Android Studio Emulator) have our own implementation of
> > virtio-vsock and we recently found that the virtio spec does not cover
> > the situation if a guest sends a VIRTIO_VSOCK_OP_REQUEST packet with
> > the same src_port that is already registered on the device side.
> >
> > This sounds like an error to me and I don't think I ever saw this
> > happening, but it would be nice if the spec explicitly mentioned how
> > this situation should be handled on the device side.
> >
> > The spec does mention "A VIRTIO_VSOCK_OP_RST reply is sent if ... the
> > destination has insufficient resources to establish the connection",
> > but I am not sure if this includes the duplicate src_port case. I
> > think this means the previously established connection will be broken.
> >
> > Thank you.
> >
> > Regards,
> > Roman.
>
> Another way is just to forbid this from happening.
>
> How does this happen on the driver side?
> --
> MST
>

Re: VIRTIO_VSOCK_OP_REQUEST with the same src_port from a guest

[Michael S. Tsirkin]

On Mon, Jul 01, 2024 at 12:16:15PM -0700, Roman Kiryanov wrote:
> Hi Michael, thank you for looking.
> 
> as I mentioned, I have never seen this happening.
> 
> But this code path (on the device side) exists (e.g. insertion of
> <dst_port, src_port> to a
> hash table says that the key already exists) even if it is never taken
> and I would be nice
> if the virtio-spec mentioned explicitly what to do there.
> 
> Regards,
> Roman.

Can Linux driver ever trigger this? Some other driver?
If this is something that drivers don't actually ever
trigger, why does it matter how exactly the device
behaves?

-- 
MST

Re: VIRTIO_VSOCK_OP_REQUEST with the same src_port from a guest

[Roman Kiryanov]

On Mon, Jul 1, 2024 at 12:40 PM Michael S. Tsirkin <mst@redhat.com> wrote:
> Can Linux driver ever trigger this? Some other driver?

I have never seen a driver with this behavior so far.

> If this is something that drivers don't actually ever
> trigger, why does it matter how exactly the device
> behaves?

I think there is a value in having the spec to cover all cases.

Re: VIRTIO_VSOCK_OP_REQUEST with the same src_port from a guest

[Michael S. Tsirkin]

On Mon, Jul 01, 2024 at 12:55:46PM -0700, Roman Kiryanov wrote:
> On Mon, Jul 1, 2024 at 12:40 PM Michael S. Tsirkin <mst@redhat.com> wrote:
> > Can Linux driver ever trigger this? Some other driver?
> 
> I have never seen a driver with this behavior so far.
> 
> > If this is something that drivers don't actually ever
> > trigger, why does it matter how exactly the device
> > behaves?
> 
> I think there is a value in having the spec to cover all cases.

Well we don't really do it for other devices.
We focus on how things fail when it's important e.g. for security.

Basically spec has been out there for a while and we can't
declare existing devices non-compliant.
We can add a recommendation.
Given no driver actually does this, the value seems low.

-- 
MST