Re: [PATCH net v10 1/2] virtio-net: correct hdr_len handling for VIRTIO_NET_F_GUEST_HDRLEN

[Xuan Zhuo <xuanzhuo@linux.alibaba.com>]

On Tue, 17 Mar 2026 12:29:17 +0100, Paolo Abeni <pabeni@redhat.com> wrote:
> On 3/13/26 8:59 AM, Xuan Zhuo wrote:
> > The commit be50da3e9d4a ("net: virtio_net: implement exact header length
> > guest feature") introduces support for the VIRTIO_NET_F_GUEST_HDRLEN
> > feature in virtio-net.
> >
> > This feature requires virtio-net to set hdr_len to the actual header
> > length of the packet when transmitting, the number of
> > bytes from the start of the packet to the beginning of the
> > transport-layer payload.
> >
> > However, in practice, hdr_len was being set using skb_headlen(skb),
> > which is clearly incorrect. This commit fixes that issue.
> >
> > Fixes: be50da3e9d4a ("net: virtio_net: implement exact header length guest feature")
> > Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com>
> > ---
> >  drivers/net/tun_vnet.h     |  2 +-
> >  drivers/net/virtio_net.c   |  6 +++++-
> >  include/linux/virtio_net.h | 33 +++++++++++++++++++++++++++++----
> >  3 files changed, 35 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/net/tun_vnet.h b/drivers/net/tun_vnet.h
> > index a5f93b6c4482..fa5cab9d3e55 100644
> > --- a/drivers/net/tun_vnet.h
> > +++ b/drivers/net/tun_vnet.h
> > @@ -244,7 +244,7 @@ tun_vnet_hdr_tnl_from_skb(unsigned int flags,
> >
> >  	if (virtio_net_hdr_tnl_from_skb(skb, tnl_hdr, has_tnl_offload,
> >  					tun_vnet_is_little_endian(flags),
> > -					vlan_hlen, true)) {
> > +					vlan_hlen, true, false)) {
> >  		struct virtio_net_hdr_v1 *hdr = &tnl_hdr->hash_hdr.hdr;
> >  		struct skb_shared_info *sinfo = skb_shinfo(skb);
> >
> > diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> > index 72d6a9c6a5a2..7106333ef904 100644
> > --- a/drivers/net/virtio_net.c
> > +++ b/drivers/net/virtio_net.c
> > @@ -3267,8 +3267,12 @@ static int xmit_skb(struct send_queue *sq, struct sk_buff *skb, bool orphan)
> >  	struct virtio_net_hdr_v1_hash_tunnel *hdr;
> >  	int num_sg;
> >  	unsigned hdr_len = vi->hdr_len;
> > +	bool feature_hdrlen;
> >  	bool can_push;
> >
> > +	feature_hdrlen = virtio_has_feature(vi->vdev,
> > +					    VIRTIO_NET_F_GUEST_HDRLEN);
> > +
> >  	pr_debug("%s: xmit %p %pM\n", vi->dev->name, skb, dest);
> >
> >  	/* Make sure it's safe to cast between formats */
> > @@ -3288,7 +3292,7 @@ static int xmit_skb(struct send_queue *sq, struct sk_buff *skb, bool orphan)
> >
> >  	if (virtio_net_hdr_tnl_from_skb(skb, hdr, vi->tx_tnl,
> >  					virtio_is_little_endian(vi->vdev), 0,
> > -					false))
> > +					false, feature_hdrlen))
> >  		return -EPROTO;
> >
> >  	if (vi->mergeable_rx_bufs)
> > diff --git a/include/linux/virtio_net.h b/include/linux/virtio_net.h
> > index 75dabb763c65..48de4a16a96a 100644
> > --- a/include/linux/virtio_net.h
> > +++ b/include/linux/virtio_net.h
> > @@ -207,6 +207,22 @@ static inline int virtio_net_hdr_to_skb(struct sk_buff *skb,
> >  	return __virtio_net_hdr_to_skb(skb, hdr, little_endian, hdr->gso_type);
> >  }
> >
> > +static inline void virtio_net_set_hdrlen(const struct sk_buff *skb,
> > +					 struct virtio_net_hdr *hdr,
> > +					 bool little_endian)
> > +{
> > +	u16 hdr_len;
> > +
> > +	hdr_len = skb_transport_offset(skb);
> > +
> > +	if (hdr->gso_type == VIRTIO_NET_HDR_GSO_UDP_L4)
> > +		hdr_len += sizeof(struct udphdr);
> > +	else
> > +		hdr_len += tcp_hdrlen(skb);
> > +
> > +	hdr->hdr_len = __cpu_to_virtio16(little_endian, hdr_len);
> > +}
>
> The series LGTM, let's wait for Jason and or Michael.
>
> If another revision is needed, please consider factoring out a
> __virtio_net_set_hdrlen() helper taking an additional explicit
> transport_offset argument, to deduplicate a bit the code between patch
> 1/2 and 2.2. Not a blocked anyway for me.

Hi, currently, a new version seems necessary. However, I do not plan to
implement it this way. This is because we require two parameters: one for
transport and another for TCP offset (tcp_hdrlen or inner_tcp_hdrlen). I believe
it would be inappropriate to put the TCP offset into the parameters as well,
since we must first determine whether the protocol is TCP before calling
tcp_hdrlen or inner_tcp_hdrlen. Therefore, I think having two
separate functions is clearer.

Thanks.


>
> /P
>