Re: [Qemu-devel] [PATCH V2] virtio-net: unbreak any layout

["Michael S. Tsirkin" <mst@redhat.com>]

On Wed, Jul 15, 2015 at 03:56:07PM +0800, Jason Wang wrote:
> Commit 032a74a1c0fcdd5fd1c69e56126b4c857ee36611
> ("virtio-net: byteswap virtio-net header") breaks any layout by
> requiring out_sg[0].iov_len >= n->guest_hdr_len. Fixing this by
> copying header to temporary buffer if swap is needed, and then use
> this buffer as part of out_sg.
> 
> Fixes 032a74a1c0fcdd5fd1c69e56126b4c857ee36611
> ("virtio-net: byteswap virtio-net header")
> Cc: qemu-stable@nongnu.org
> Cc: clg@fr.ibm.com
> Signed-off-by: Jason Wang <jasowang@redhat.com>
> ---
> Changes from V1:
> - avoid header copying if there's no need to do header swap
> - don't write the header back
> ---
>  hw/net/virtio-net.c               | 17 ++++++++++++++---
>  include/hw/virtio/virtio-access.h |  9 +++++++++
>  2 files changed, 23 insertions(+), 3 deletions(-)
> 
> diff --git a/hw/net/virtio-net.c b/hw/net/virtio-net.c
> index e3c2db3..12322bd 100644
> --- a/hw/net/virtio-net.c
> +++ b/hw/net/virtio-net.c
> @@ -1142,7 +1142,8 @@ static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
>          ssize_t ret, len;
>          unsigned int out_num = elem.out_num;
>          struct iovec *out_sg = &elem.out_sg[0];
> -        struct iovec sg[VIRTQUEUE_MAX_SIZE];
> +        struct iovec sg[VIRTQUEUE_MAX_SIZE], sg2[VIRTQUEUE_MAX_SIZE];
> +        struct virtio_net_hdr hdr;
>  
>          if (out_num < 1) {
>              error_report("virtio-net header not in first element");
> @@ -1150,11 +1151,21 @@ static int32_t virtio_net_flush_tx(VirtIONetQueue *q)
>          }
>  
>          if (n->has_vnet_hdr) {
> -            if (out_sg[0].iov_len < n->guest_hdr_len) {
> +            if (iov_size(out_sg, out_num) < n->guest_hdr_len) {
>                  error_report("virtio-net header incorrect");
>                  exit(1);
>              }
> -            virtio_net_hdr_swap(vdev, (void *) out_sg[0].iov_base);
> +            if (virtio_needs_swap(vdev)) {
> +                iov_to_buf(out_sg, out_num, 0, &hdr, sizeof(hdr));
> +                virtio_net_hdr_swap(vdev, (void *) &hdr);
> +                sg2[0].iov_base = &hdr;
> +                sg2[0].iov_len = sizeof(hdr);
> +                out_num = iov_copy(&sg2[1], ARRAY_SIZE(sg2) - 1,
> +                                   out_sg, out_num,
> +                                   sizeof(hdr), -1);
> +                out_num += 1;
> +                out_sg = sg2;
> +            }
>          }
>  
>          /*

I guess this works but iov cpy is pretty expensive.

- When using the tun backend, we can just set VNETBE/VNETLE
  correctly and avoid swaps in userspace.
  It's easy to probe for - why don't we do this?

- OTOH when not using the tun backend, offloads are
  generally disabled, so the swap isn't needed
  since header is going to be all 0s.

- what's left is old kernels when using tun.
  how about probing for that explicitly?
  In other words, virtio_net_needs_swap as opposed to
  the generic virtio_needs_swap.


> diff --git a/include/hw/virtio/virtio-access.h b/include/hw/virtio/virtio-access.h
> index cee5dd7..1ec1dfd 100644
> --- a/include/hw/virtio/virtio-access.h
> +++ b/include/hw/virtio/virtio-access.h
> @@ -143,6 +143,15 @@ static inline uint64_t virtio_ldq_p(VirtIODevice *vdev, const void *ptr)
>      }
>  }
>  
> +static inline bool virtio_needs_swap(VirtIODevice *vdev)
> +{
> +#ifdef HOST_WORDS_BIGENDIAN
> +    return virtio_access_is_big_endian(vdev) ? false : true;
> +#else
> +    return virtio_access_is_big_endian(vdev) ? true : false;
> +#endif
> +}
> +
>  static inline uint16_t virtio_tswap16(VirtIODevice *vdev, uint16_t s)
>  {
>  #ifdef HOST_WORDS_BIGENDIAN
> -- 
> 2.1.4