When using gretap, I am seeing that Path MTU discovery is not
working for packets going out over the tunnel interface. What happens
is that the driver correctly identifies the DF bit, and see IPv4
but since skb_dst(skb) is NULL, the icmp_send() ends up doing nothing.
The following fixes the problem but does not seem like the correct general
solution. IPv6 almost certainly has the same problem.
Perhaps we should just set the skb_dst() earlier (before the
MTU checks).
diff --git a/net/ipv4/ip_gre.c b/net/ipv4/ip_gre.c
index 1017460..da57bb0 100644
--- a/net/ipv4/ip_gre.c
+++ b/net/ipv4/ip_gre.c
@@ -838,8 +838,9 @@ static netdev_tx_t ipgre_tunnel_xmit(struct sk_buff *skb, struct net_device *dev
if ((old_iph->frag_off&htons(IP_DF)) &&
mtu < ntohs(old_iph->tot_len)) {
+ skb_dst_drop(skb);
+ skb_dst_set(skb, &rt->dst);
icmp_send(skb, ICMP_DEST_UNREACH, ICMP_FRAG_NEEDED, htonl(mtu));
- ip_rt_put(rt);
goto tx_error;
}
}