Re: [RFC] ipv6: dst_allfrag() not taken into account by TCP

[Eric Dumazet <eric.dumazet@gmail.com>]

Le mercredi 18 janvier 2012 à 13:42 +0100, Tore Anderson a écrit :
> * Eric Dumazet
> 
> > Could you please test following patch ?
> 
> Hi,
> 
> It seems to work fine, now all the outgoing packets that includes a
> Fragmentation header all are "atomic" or "non-fragmented" fragments,
> i.e., that they both have Offset=0 and More Fragments=0.
> 
> I'm uploaded a tcpdump showing the traffic with the patch here:
> 
> https://bugzilla.kernel.org/attachment.cgi?id=72106
> 
> Best regards,


Seems good, thanks !

Now I wonder if some RFC can allow us to send small frames without a
frag header, since this frag header forces us to perform tx checksum in
software. 

I guess 576 should be safe ?

diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
index d97e071..14cb938 100644
--- a/net/ipv6/ip6_output.c
+++ b/net/ipv6/ip6_output.c
@@ -153,7 +153,7 @@ static int ip6_finish_output2(struct sk_buff *skb)
 static int ip6_finish_output(struct sk_buff *skb)
 {
 	if ((skb->len > ip6_skb_dst_mtu(skb) && !skb_is_gso(skb)) ||
-	    dst_allfrag(skb_dst(skb)))
+	    (skb->len > 576 && dst_allfrag(skb_dst(skb))))
 		return ip6_fragment(skb, ip6_finish_output2);
 	else
 		return ip6_finish_output2(skb);