Re: [Qemu-devel] [PATCH] NetKVM: fix for indirectc mode when vring is full

[Yan Vugenfirer]

Thanks!

The fix is correct, but I am not sure this patch is needed.
This is a frozen part of the code with XP and Windows 2003 support for NetKVM 
only.

The library that is used for newer OSes and other drivers has this fix: 
https://github.com/YanVugenfirer/kvm-guest-drivers-windows/commit/264f1b6c86f5eeca5e2c9fbd24e3de8dbd0bee1d

In the NetKVM driver code itself we are testing for those conditions before 
calling to add_buf (check ParaNdis_DoSubmitPacket function in 
https://github.com/YanVugenfirer/kvm-guest-drivers-windows/blob/stable/NetKVM/NDIS5/Common/ParaNdis-Common.c#L1528).

Best regads,
Yan.

> On Dec 10, 2014, at 9:28 AM, Ting Wang <address@hidden> wrote:
> 
> In function vring_add_indirect, there is no limiti
> about free entry in vring. If vring is full,
> vq->num_free will be less than zero, and
> the address of vq->vring.desc becomes illegal.
> 
> Signed-off-by: Ting Wang <address@hidden>
> ---
> NetKVM/NDIS5/VirtIO/VirtIORing.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/NetKVM/NDIS5/VirtIO/VirtIORing.c 
> b/NetKVM/NDIS5/VirtIO/VirtIORing.c
> index 90ace4c..0f3783c 100644
> --- a/NetKVM/NDIS5/VirtIO/VirtIORing.c
> +++ b/NetKVM/NDIS5/VirtIO/VirtIORing.c
> @@ -150,7 +150,7 @@ static int vring_add_buf(struct virtqueue *_vq,
>         return -1;
>     }
> 
> -    if (va_indirect)
> +    if (va_indirect && (out + in) > 1 && vq->num_free)
>     {
>         int ret = vring_add_indirect(_vq, sg, out, in, va_indirect, 
> phys_indirect);
>         if (ret >= 0)
> -- 
> 1.8.5
> 
>