On Thu, Mar 17, 2016 at 11:24:59AM +0100, Steffen Klassert wrote:

In IPv6 this check is missing, so this could be the
problem if this is IPv6.

indeed, this patch also fixes my problem:

--- a/net/ipv6/ip6_output.c
+++ b/net/ipv6/ip6_output.c

@@ -1353,6 +1353,7 @@ emsgsize:
 	     (skb && skb_is_gso(skb))) &&
 	    (sk->sk_protocol == IPPROTO_UDP) &&
 	    (rt->dst.dev->features & NETIF_F_UFO) &&
+	    !rt->dst.header_len &&
 	    (sk->sk_type == SOCK_DGRAM)) {
 		err = ip6_ufo_append_data(sk, queue, getfrag, from, length,
 					  hh_len, fragheaderlen,

I can't say which is better. Herbert originally seemed to like
the fix inside xfrm_output().

The IPv4 part is fixed by commit
c146066ab80267c3305de5dda6a4083f06df9265 (ipv4: Don't use ufo
handling on later transformed packets)

Thanks,

-- 
Jiri Bohac [off-list ref]
SUSE Labs, SUSE CZ