On 10/05/2017 09:58 PM, Duyck, Alexander H wrote: > On Thu, 2017-10-05 at 20:06 +0300, Alexey Kodanev wrote: >> When gso_size reset to zero for the tail segment in skb_segment(), later >> in ipv6_gso_segment(), we will get incorrect payload_len for that segment. >> inet_gso_segment() already has a check for gso_size before calculating >> payload so fixing only IPv6 part. >> >> The issue was found with LTP vxlan & gre tests over ixgbe NIC. >> >> Fixes: 07b26c9454a2 ("gso: Support partial splitting at the frag_list >> pointer") >> Signed-off-by: Alexey Kodanev <alexey.koda...@oracle.com> >> --- >> net/ipv6/ip6_offload.c | 2 +- >> 1 file changed, 1 insertion(+), 1 deletion(-) >> >> diff --git a/net/ipv6/ip6_offload.c b/net/ipv6/ip6_offload.c >> index cdb3728..4a87f94 100644 >> --- a/net/ipv6/ip6_offload.c >> +++ b/net/ipv6/ip6_offload.c >> @@ -105,7 +105,7 @@ static struct sk_buff *ipv6_gso_segment(struct sk_buff >> *skb, >> >> for (skb = segs; skb; skb = skb->next) { >> ipv6h = (struct ipv6hdr *)(skb_mac_header(skb) + nhoff); >> - if (gso_partial) >> + if (gso_partial && skb_is_gso(skb)) >> payload_len = skb_shinfo(skb)->gso_size + >> SKB_GSO_CB(skb)->data_offset + >> skb->head - (unsigned char *)(ipv6h + 1); > So looking over this change it looks good to me. I'm just wondering if > you have looked at the code in __skb_udp_tunnel_segment or > gre_gso_segment? It seems like if you needed this change here you > should need to make similar changes to those functions as well. I'm > wondering if we just aren't seeing issues due to the segments already > being MSS sized before being handed to us for segmentation.
Right, it can happen in __skb_udp_tunnel_segment as well. I wasn't able to reproduce with gre, it looks like it doesn't go to that part of code, skipping it, possibly, on gso_type & SKB_GSO_GRE_CSUM check, though the NIC has tx-gre-csum-segmentation enabled... I'll send new version after testing completed. Thanks, Alexey