tunneling: Always use skb->len for PMTUD.

Currently IPv4 uses the length of the packet in the IP header
and IPv6 uses actual length of the packet for path MTU discovery.
This changes both to use the length of the packet to be consistent
with each other and the the Linux IP stack and also to match what
actually causes fragmentation.

Signed-off-by: Jesse Gross <jesse@nicira.com>
Acked-by: Ben Pfaff <blp@nicira.com>

diff --git a/datapath/tunnel.c b/datapath/tunnel.c
index 7bf46e07e9052d557045bebfda023a2c32d60898..40577fbd516c26d65b2825544a712b8c27bf5dfc 100644 (file)
--- a/datapath/tunnel.c
+++ b/datapath/tunnel.c
@@ -718,6 +718,10 @@ static bool check_mtu(struct sk_buff *skb,
        bool pmtud = mutable->flags & TNL_F_PMTUD;
        __be16 frag_off = 0;
        int mtu = 0;
+       unsigned int packet_length = skb->len - ETH_HLEN;
+
+       if (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q))
+               packet_length -= VLAN_HLEN;
 
        if (pmtud) {
                frag_off = htons(IP_DF);
@@ -737,7 +741,7 @@ static bool check_mtu(struct sk_buff *skb,
                if (pmtud && iph->frag_off & htons(IP_DF)) {
                        mtu = max(mtu, IP_MIN_MTU);
 
-                       if (ntohs(iph->tot_len) > mtu &&
+                       if (packet_length > mtu &&
                            tnl_frag_needed(vport, mutable, skb, mtu,
                                            OVS_CB(skb)->tun_id))
                                return false;
@@ -745,10 +749,6 @@ static bool check_mtu(struct sk_buff *skb,
        }
 #if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
        else if (skb->protocol == htons(ETH_P_IPV6)) {
-               unsigned int packet_length = skb->len - ETH_HLEN
-                       - (eth_hdr(skb)->h_proto == htons(ETH_P_8021Q) ?
-                               VLAN_HLEN : 0);
-
                /* IPv6 requires PMTUD if the packet is above the minimum MTU. */
                if (packet_length > IPV6_MIN_MTU)
                        frag_off = htons(IP_DF);