Linux Virtualization - Mar 2016 - [RFC v2 -next 1/2] virtio: Start feature MTU support

On Tue, 15 Mar 2016 17:04:12 -0400
Aaron Conole <aconole at redhat.com> wrote:
> --- a/include/uapi/linux/virtio_net.h
> +++ b/include/uapi/linux/virtio_net.h
> @@ -55,6 +55,7 @@
>  #define VIRTIO_NET_F_MQ	22	/* Device supports Receive Flow
>  					 * Steering */
>  #define VIRTIO_NET_F_CTRL_MAC_ADDR 23	/* Set MAC address */
> +#define VIRTIO_NET_F_MTU 25	/* Device supports Default MTU Negotiation */
>  
>  #ifndef VIRTIO_NET_NO_LEGACY
>  #define VIRTIO_NET_F_GSO	6	/* Host handles pkts w/ any GSO type */
> @@ -73,6 +74,8 @@ struct virtio_net_config {
>  	 * Legal values are between 1 and 0x8000
>  	 */
>  	__u16 max_virtqueue_pairs;
> +	/* Default maximum transmit unit advice */
> +	__u16 mtu;
>  } __attribute__((packed));
>  
>  /*
You can't change user visible headers without breaking ABI.
This structure might be used by other user code. Also how can this
work if host is using old size of structure.

On Wed, Mar 16, 2016 at 11:23:14AM -0700, Stephen Hemminger
wrote:
> On Tue, 15 Mar 2016 17:04:12 -0400
> Aaron Conole <aconole at redhat.com> wrote:
> 
> > --- a/include/uapi/linux/virtio_net.h
> > +++ b/include/uapi/linux/virtio_net.h
> > @@ -55,6 +55,7 @@
> >  #define VIRTIO_NET_F_MQ	22	/* Device supports Receive Flow
> >  					 * Steering */
> >  #define VIRTIO_NET_F_CTRL_MAC_ADDR 23	/* Set MAC address */
> > +#define VIRTIO_NET_F_MTU 25	/* Device supports Default MTU
Negotiation */
> >  
> >  #ifndef VIRTIO_NET_NO_LEGACY
> >  #define VIRTIO_NET_F_GSO	6	/* Host handles pkts w/ any GSO type */
> > @@ -73,6 +74,8 @@ struct virtio_net_config {
> >  	 * Legal values are between 1 and 0x8000
> >  	 */
> >  	__u16 max_virtqueue_pairs;
> > +	/* Default maximum transmit unit advice */
> > +	__u16 mtu;
> >  } __attribute__((packed));
> >  
> >  /*
> 
> You can't change user visible headers without breaking ABI.
> This structure might be used by other user code.
Then this userspace is broken.

If someone uses virtio code one has to follow virtio spec.

And Virtio spec makes it very clear that
1. config space size can change at any time
2. fields can only be accessed if the correct feature bit
   has been both advertized in host bitmap and acknowledged
   in guest bitmap
> Also how can this
> work if host is using old size of structure.
It works because access is guarded by feature bit check.

-- 
MST

Stephen Hemminger <stephen at networkplumber.org> writes:
> On Tue, 15 Mar 2016 17:04:12 -0400
> Aaron Conole <aconole at redhat.com> wrote:
>
>> --- a/include/uapi/linux/virtio_net.h
>> +++ b/include/uapi/linux/virtio_net.h
>> @@ -55,6 +55,7 @@
>>  #define VIRTIO_NET_F_MQ	22	/* Device supports Receive Flow
>>  					 * Steering */
>>  #define VIRTIO_NET_F_CTRL_MAC_ADDR 23	/* Set MAC address */
>> +#define VIRTIO_NET_F_MTU 25	/* Device supports Default MTU Negotiation
*/
>>  
>>  #ifndef VIRTIO_NET_NO_LEGACY
>>  #define VIRTIO_NET_F_GSO	6	/* Host handles pkts w/ any GSO type */
>> @@ -73,6 +74,8 @@ struct virtio_net_config {
>>  	 * Legal values are between 1 and 0x8000
>>  	 */
>>  	__u16 max_virtqueue_pairs;
>> +	/* Default maximum transmit unit advice */
>> +	__u16 mtu;
>>  } __attribute__((packed));
>>  
>>  /*
>
> You can't change user visible headers without breaking ABI.
> This structure might be used by other user code. Also how can this
> work if host is using old size of structure.
How else can this field be added and remain compliant with the spec? The
spec requires that mtu be passed in the virtio_net_config field.

As for old sizeof, I think the absence of the VIRTIO_NET_F_MTU bit being
asserted is confirmation that mtu is not valid (at least, it is implied
in the spec).

Thanks so much for the review, Stephen!

-Aaron

On Thu, 17 Mar 2016 17:10:55 -0400
Aaron Conole <aconole at redhat.com> wrote:
> Stephen Hemminger <stephen at networkplumber.org> writes:
> 
> > On Tue, 15 Mar 2016 17:04:12 -0400
> > Aaron Conole <aconole at redhat.com> wrote:
> >
> >> --- a/include/uapi/linux/virtio_net.h
> >> +++ b/include/uapi/linux/virtio_net.h
> >> @@ -55,6 +55,7 @@
> >>  #define VIRTIO_NET_F_MQ	22	/* Device supports Receive Flow
> >>  					 * Steering */
> >>  #define VIRTIO_NET_F_CTRL_MAC_ADDR 23	/* Set MAC address */
> >> +#define VIRTIO_NET_F_MTU 25	/* Device supports Default MTU
Negotiation */
> >>  
> >>  #ifndef VIRTIO_NET_NO_LEGACY
> >>  #define VIRTIO_NET_F_GSO	6	/* Host handles pkts w/ any GSO type
*/
> >> @@ -73,6 +74,8 @@ struct virtio_net_config {
> >>  	 * Legal values are between 1 and 0x8000
> >>  	 */
> >>  	__u16 max_virtqueue_pairs;
> >> +	/* Default maximum transmit unit advice */
> >> +	__u16 mtu;
> >>  } __attribute__((packed));
> >>  
> >>  /*
> >
> > You can't change user visible headers without breaking ABI.
> > This structure might be used by other user code. Also how can this
> > work if host is using old size of structure.
> 
> How else can this field be added and remain compliant with the spec? The
> spec requires that mtu be passed in the virtio_net_config field.
> 
> As for old sizeof, I think the absence of the VIRTIO_NET_F_MTU bit being
> asserted is confirmation that mtu is not valid (at least, it is implied
> in the spec).
Michael is right as long as the code checks for MTU flag before
referencing the mtu field, everything is fine.  Actually, the structure
is never used directly only by fetching fields with offsetof