Re: [ROUTE] PMTU only works on half the time

[Julian Anastasov <ja@xxxxxx>]

        Hello,

On Mon, 1 Dec 2003, David S. Miller wrote:

> I disagree, and this is related to my most recent email
> in this thread.
>
> This packet we are reacting to for PMTU purposes could only
> have come from us if the TOS matches precisely.

        Here is what I have for today. I assume all ip_route_output
callers provide valid tos (not a wildcard). As result, only RTO_ONLINK and 
oif have wildcard value. I'm not sure if ip_rt_frag_needed needs an iif 
argument, may be yes?

        Also, it seems ip_rt_redirect needs the 'tos, tos | RTO_ONLINK'
array too as in ip_rt_frag_needed. Not included yet.

        Another problem: it seems __ip_route_output_key does not
hash with valid tos key bits, fix included below:

--- net/ipv4/route.c.orig       Tue Dec  2 03:25:59 2003
+++ net/ipv4/route.c    Tue Dec  2 03:37:27 2003
@@ -1239,19 +1239,25 @@
 
 unsigned short ip_rt_frag_needed(struct iphdr *iph, unsigned short new_mtu)
 {
-       int i;
+       int i, j, k;
        unsigned short old_mtu = ntohs(iph->tot_len);
        struct rtable *rth;
        u32  skeys[2] = { iph->saddr, 0, };
        u32  daddr = iph->daddr;
        u8   tos = iph->tos & IPTOS_RT_MASK;
        unsigned short est_mtu = 0;
+       u8 toskeys[2] = { tos, tos | RTO_ONLINK };
+       int iif = 0; // Can be argument
+       int  ikeys[2] = { iif, 0 };
 
        if (ipv4_config.no_pmtu_disc)
                return 0;
 
+       for (k = 0; k < (iif ? 2 : 1); k++)
+       for (j = 0; j < 2; j++)
        for (i = 0; i < 2; i++) {
-               unsigned hash = rt_hash_code(daddr, skeys[i], tos);
+               unsigned hash = rt_hash_code(daddr, skeys[i] ^ (ikeys[k] << 5),
+                                            toskeys[j]);
 
                rcu_read_lock();
                for (rth = rt_hash_table[hash].chain; rth;
@@ -1261,7 +1267,8 @@
                            rth->fl.fl4_src == skeys[i] &&
                            rth->rt_dst  == daddr &&
                            rth->rt_src  == iph->saddr &&
-                           rth->fl.fl4_tos == tos &&
+                           rth->fl.fl4_tos == toskeys[j] &&
+                           rth->fl.oif == ikeys[k] &&
                            rth->fl.iif == 0 &&
                            !(dst_metric_locked(&rth->u.dst, RTAX_MTU))) {
                                unsigned short mtu = new_mtu;
@@ -2214,7 +2221,8 @@
        unsigned hash;
        struct rtable *rth;
 
-       hash = rt_hash_code(flp->fl4_dst, flp->fl4_src ^ (flp->oif << 5), 
flp->fl4_tos);
+       hash = rt_hash_code(flp->fl4_dst, flp->fl4_src ^ (flp->oif << 5),
+               flp->fl4_tos & (IPTOS_RT_MASK | RTO_ONLINK));
 
        rcu_read_lock();
        for (rth = rt_hash_table[hash].chain; rth; rth = rth->u.rt_next) {

Regards

--
Julian Anastasov <ja@xxxxxx>