On Fri, 2016-05-06 at 18:19 -0700, Eric Dumazet wrote:
> From: Eric Dumazet <eduma...@google.com>
>
> When using ifb+netem on ingress on SIT/IPIP/GRE traffic,
> GRO packets are not properly processed.
>
> Segmentation should not be forced, since ifb is already adding
> quite a performance hit.
>
> Signed-off-by: Eric Dumazet <eduma...@google.com>
> ---
> drivers/net/ifb.c | 3 +++
> 1 file changed, 3 insertions(+)
>
> diff --git a/drivers/net/ifb.c b/drivers/net/ifb.c
> index cc56fac3c3f8..66c0eeafcb5d 100644
> --- a/drivers/net/ifb.c
> +++ b/drivers/net/ifb.c
> @@ -196,6 +196,7 @@ static const struct net_device_ops ifb_netdev_ops = {
>
> #define IFB_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | NETIF_F_FRAGLIST
> | \
> NETIF_F_TSO_ECN | NETIF_F_TSO | NETIF_F_TSO6 | \
> + NETIF_F_GSO_ENCAP_ALL | \
> NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX | \
> NETIF_F_HW_VLAN_STAG_TX)
>
> @@ -224,6 +225,8 @@ static void ifb_setup(struct net_device *dev)
> dev->tx_queue_len = TX_Q_LIMIT;
>
> dev->features |= IFB_FEATURES;
> + dev->hw_features |= dev->features;
> + dev->hw_enc_features |= dev->features;
> dev->vlan_features |= IFB_FEATURES & ~(NETIF_F_HW_VLAN_CTAG_TX |
> NETIF_F_HW_VLAN_STAG_TX);
>
>
BTW, encapsulated GRO traffic going through mirred+ifb is dropped
because segments get an incorrect skb->mac_len
(If TSO/GSO is disabled on ifb, as before the above patch)
SIT traffic for example : segments get mac_len set to 34 instead of 14
What do you think of :
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index e561f9f07d6d..bec5c32b2fe9 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -3176,7 +3176,7 @@ struct sk_buff *skb_segment(struct sk_buff *head_skb,
__copy_skb_header(nskb, head_skb);
skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
- skb_reset_mac_len(nskb);
+ nskb->mac_len = head_skb->mac_len;
skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
nskb->data - tnl_hlen,
