Re: [PATCH V2 2/2] ipvs: Extend MTU check to account for IPv6 NAT defrag changes

On Tue, 2012-08-28 at 07:49 -0700, Eric Dumazet wrote:

On Tue, 2012-08-28 at 16:23 +0200, Jesper Dangaard Brouer wrote:

quoted

This patch is necessary, to make IPVS work, after Patrick McHardys
IPv6 NAT defragmentation changes.

Signed-off-by: Jesper Dangaard Brouer <redacted>
---
In V2: the tunnel mode is no longer a special case.

 net/netfilter/ipvs/ip_vs_xmit.c |    9 ++++++++-
 1 files changed, 8 insertions(+), 1 deletions(-)

diff --git a/net/netfilter/ipvs/ip_vs_xmit.c b/net/netfilter/ipvs/ip_vs_xmit.c
index 67a3978..56f6d5d 100644
--- a/net/netfilter/ipvs/ip_vs_xmit.c
+++ b/net/netfilter/ipvs/ip_vs_xmit.c

@@ -88,7 +88,14 @@ __ip_vs_dst_check(struct ip_vs_dest *dest, u32 rtos)
 static inline bool
 __mtu_check_toobig_v6(const struct sk_buff *skb, u32 mtu)
 {
-	if (skb->len > mtu && !skb_is_gso(skb)) {
+	if (IP6CB(skb)->frag_max_size) {
+		/* frag_max_size tell us that, this packet have been
+		 * defragmented by netfilter IPv6 conntrack module.
+		 */
+		if (IP6CB(skb)->frag_max_size > mtu)
+			return true; /* largest fragment violate MTU */

Implicit:         else
     			return false

(if it makes it more clear, not sure)

quoted

+	}
+	else if (skb->len > mtu && !skb_is_gso(skb)) {
 		return true; /* Packet size violate MTU size */
 	}

Couldnt you use a single test ?

if (IP6CB(skb)->frag_max_size > mtu)
	return true;

if (skb->len > mtu && !skb_is_gso(skb))
	return true;

Nope, this will not work.

If (IP6CB(skb)->frag_max_size > 0) then we have a defragmented packet,
this means that skb->len cannot be used for MTU checking, because
skb->len is now the total length of all the fragments (which your
solution will fall-through to)

The unreadable version of the function is:

static inline bool
__mtu_check_toobig_v6_unreadable(const struct sk_buff *skb, u32 mtu)
{
       return !!((!IP6CB(skb)->frag_max_size &&
                  (skb->len > mtu && !skb_is_gso(skb)))
                 || IP6CB(skb)->frag_max_size > mtu);
}

Perhaps you like this version better (you seem to use constructions like
this), but I really dislike it, because I (personally) find it
harder/slower to read this kind of code, and I also believe its more
error prone when someone needs to extend this.

-- 
Best regards,
  Jesper Dangaard Brouer
  MSc.CS, Sr. Network Kernel Developer at Red Hat
  Author of http://www.iptv-analyzer.org
  LinkedIn: http://www.linkedin.com/in/brouer