On Thu, Mar 28, 2024 at 7:44 AM Andrew Melnychenko <andrew@xxxxxxxxxx> wrote: > > When the Qemu launched with vhost but without tap vnet_hdr, > vhost tries to copy vnet_hdr from socket iter with size 0 > to the page that may contain some trash. > That trash can be interpreted as unpredictable values for > vnet_hdr. > That leads to dropping some packets and in some cases to > stalling vhost routine when the vhost_net tries to process > packets and fails in a loop. > > Qemu options: > -netdev tap,vhost=on,vnet_hdr=off,... > > From security point of view, wrong values on field used later > tap's tap_get_user_xdp() and will affect skb gso and options. > Later the header(and data in headroom) should not be used by the stack. > Using custom socket as a backend to vhost_net can reveal some data > in the vnet_hdr, although it would require kernel access to implement. > > The issue happens because the value of sock_len in virtqueue is 0. > That value is set at vhost_net_set_features() with > VHOST_NET_F_VIRTIO_NET_HDR, also it's set to zero at device open() > and reset() routine. > So, currently, to trigger the issue, we need to set up qemu with > vhost=on,vnet_hdr=off, or do not configure vhost in the custom program. > > Signed-off-by: Andrew Melnychenko <andrew@xxxxxxxxxx> Acked-by: Jason Wang <jasowang@xxxxxxxxxx> It seems it has been merged by Michael. Thanks > --- > drivers/vhost/net.c | 3 +++ > 1 file changed, 3 insertions(+) > > diff --git a/drivers/vhost/net.c b/drivers/vhost/net.c > index f2ed7167c848..57411ac2d08b 100644 > --- a/drivers/vhost/net.c > +++ b/drivers/vhost/net.c > @@ -735,6 +735,9 @@ static int vhost_net_build_xdp(struct vhost_net_virtqueue *nvq, > hdr = buf; > gso = &hdr->gso; > > + if (!sock_hlen) > + memset(buf, 0, pad); > + > if ((gso->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) && > vhost16_to_cpu(vq, gso->csum_start) + > vhost16_to_cpu(vq, gso->csum_offset) + 2 > > -- > 2.43.0 >
