Re: [PATCH] virtio: Work around frames incorrectly marked as gso

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

On Thu, Feb 13, 2020 at 07:44:06AM -0800, Eric Dumazet wrote:
> 
> 
> On 2/13/20 2:00 AM, Michael S. Tsirkin wrote:
> > On Wed, Feb 12, 2020 at 05:38:09PM +0000, Anton Ivanov wrote:
> >>
> >>
> >> On 11/02/2020 10:37, Michael S. Tsirkin wrote:
> >>> On Tue, Feb 11, 2020 at 07:42:37AM +0000, Anton Ivanov wrote:
> >>>> On 11/02/2020 02:51, Jason Wang wrote:
> >>>>>
> >>>>> On 2020/2/11 上午12:55, Anton Ivanov wrote:
> >>>>>>
> >>>>>>
> >>>>>> On 09/12/2019 10:48, anton.ivanov@cambridgegreys.com wrote:
> >>>>>>> From: Anton Ivanov <anton.ivanov@cambridgegreys.com>
> >>>>>>>
> >>>>>>> Some of the frames marked as GSO which arrive at
> >>>>>>> virtio_net_hdr_from_skb() have no GSO_TYPE, no
> >>>>>>> fragments (data_len = 0) and length significantly shorter
> >>>>>>> than the MTU (752 in my experiments).
> >>>>>>>
> >>>>>>> This is observed on raw sockets reading off vEth interfaces
> >>>>>>> in all 4.x and 5.x kernels I tested.
> >>>>>>>
> >>>>>>> These frames are reported as invalid while they are in fact
> >>>>>>> gso-less frames.
> >>>>>>>
> >>>>>>> This patch marks the vnet header as no-GSO for them instead
> >>>>>>> of reporting it as invalid.
> >>>>>>>
> >>>>>>> Signed-off-by: Anton Ivanov <anton.ivanov@cambridgegreys.com>
> >>>>>>> ---
> >>>>>>>    include/linux/virtio_net.h | 8 ++++++--
> >>>>>>>    1 file changed, 6 insertions(+), 2 deletions(-)
> >>>>>>>
> >>>>>>> diff --git a/include/linux/virtio_net.h b/include/linux/virtio_net.h
> >>>>>>> index 0d1fe9297ac6..d90d5cff1b9a 100644
> >>>>>>> --- a/include/linux/virtio_net.h
> >>>>>>> +++ b/include/linux/virtio_net.h
> >>>>>>> @@ -112,8 +112,12 @@ static inline int
> >>>>>>> virtio_net_hdr_from_skb(const struct sk_buff *skb,
> >>>>>>>                hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
> >>>>>>>            else if (sinfo->gso_type & SKB_GSO_TCPV6)
> >>>>>>>                hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
> >>>>>>> -        else
> >>>>>>> -            return -EINVAL;
> >>>>>>> +        else {
> >>>>>>> +            if (skb->data_len == 0)
> >>>>>>> +                hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
> >>>>>>> +            else
> >>>>>>> +                return -EINVAL;
> >>>>>>> +        }
> >>>>>>>            if (sinfo->gso_type & SKB_GSO_TCP_ECN)
> >>>>>>>                hdr->gso_type |= VIRTIO_NET_HDR_GSO_ECN;
> >>>>>>>        } else
> >>>>>>>
> >>>>>>
> >>>>>> ping.
> >>>>>>
> >>>>>
> >>>>> Do you mean gso_size is set but gso_type is not? Looks like a bug
> >>>>> elsewhere.
> >>>>>
> >>>>> Thanks
> >>>>>
> >>>>>
> >>>> Yes.
> >>>>
> >>>> I could not trace it where it is coming from.
> >>>>
> >>>> I see it when doing recvmmsg on raw sockets in the UML vector network
> >>>> drivers.
> >>>>
> >>>
> >>> I think we need to find the culprit and fix it there, lots of other things
> >>> can break otherwise.
> >>> Just printing out skb->dev->name should do the trick, no?
> >>
> >> The printk in virtio_net_hdr_from_skb says NULL.
> >>
> >> That is probably normal for a locally originated frame.
> >>
> >> I cannot reproduce this with network traffic by the way - it happens only if the traffic is locally originated on the host.
> >>
> >> A,
> > 
> > OK so is it code in __tcp_transmit_skb that sets gso_size to non-null
> > when gso_type is 0?
> >
> 
> Correct way to determine if a packet is a gso one is by looking at gso_size.
> Then only it is legal looking at gso_type
> 
> 
> static inline bool skb_is_gso(const struct sk_buff *skb)
> {
>     return skb_shinfo(skb)->gso_size;
> }
> 
> /* Note: Should be called only if skb_is_gso(skb) is true */
> static inline bool skb_is_gso_v6(const struct sk_buff *skb)
> ...
> 
> 
> There is absolutely no relation between GSO and skb->data_len, skb can be linearized
> for various orthogonal reasons.

The reported problem is that virtio gets a packet where gso_size
is !0 but gso_type is 0.

It currently drops these on the assumption that it's some type
of a gso packet it does not know how to handle.


So you are saying if skb_is_gso we can still have gso_type set to 0,
and that's an expected configuration?

So the patch should just be:


-        if (skb_is_gso(skb)) {
+        if (skb_is_gso(skb) && sinfo->gso_type) {



?