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

[Julian Anastasov <ja@xxxxxx>]

        Hello,

        I'm appending new version that handles all missing cases for 
hashed values on icmp error:

- ip_rt_frag_needed now provides interface index
- ip_rt_redirect checks for tos|RTO_ONLINK too
- ip_rt_frag_needed checks also for tos|RTO_ONLINK and oif!=0
- __ip_route_output_key now ignores illegal bits (bit 1) from tos

        Please review and edit if needed.

diff -Nru a/include/net/route.h b/include/net/route.h
--- a/include/net/route.h       Tue Dec  2 23:37:17 2003
+++ b/include/net/route.h       Tue Dec  2 23:37:17 2003
@@ -122,7 +122,7 @@
 extern int             ip_route_output_key(struct rtable **, struct flowi 
*flp);
 extern int             ip_route_output_flow(struct rtable **rp, struct flowi 
*flp, struct sock *sk, int flags);
 extern int             ip_route_input(struct sk_buff*, u32 dst, u32 src, u8 
tos, struct net_device *devin);
-extern unsigned short  ip_rt_frag_needed(struct iphdr *iph, unsigned short 
new_mtu);
+extern unsigned short  ip_rt_frag_needed(int iif, struct iphdr *iph, unsigned 
short new_mtu);
 extern void            ip_rt_send_redirect(struct sk_buff *skb);
 
 extern unsigned                inet_addr_type(u32 addr);
diff -Nru a/net/ipv4/icmp.c b/net/ipv4/icmp.c
--- a/net/ipv4/icmp.c   Tue Dec  2 23:37:17 2003
+++ b/net/ipv4/icmp.c   Tue Dec  2 23:37:17 2003
@@ -626,7 +626,7 @@
                                                         "and DF set.\n",
                                               NIPQUAD(iph->daddr));
                        } else {
-                               info = ip_rt_frag_needed(iph,
+                               info = ip_rt_frag_needed(skb->dev->ifindex, iph,
                                                     ntohs(icmph->un.frag.mtu));
                                if (!info)
                                        goto out;
diff -Nru a/net/ipv4/route.c b/net/ipv4/route.c
--- a/net/ipv4/route.c  Tue Dec  2 23:37:17 2003
+++ b/net/ipv4/route.c  Tue Dec  2 23:37:17 2003
@@ -967,11 +967,12 @@
 void ip_rt_redirect(u32 old_gw, u32 daddr, u32 new_gw,
                    u32 saddr, u8 tos, struct net_device *dev)
 {
-       int i, k;
+       int i, j, k;
        struct in_device *in_dev = in_dev_get(dev);
        struct rtable *rth, **rthp;
        u32  skeys[2] = { saddr, 0 };
        int  ikeys[2] = { dev->ifindex, 0 };
+       u8 toskeys[2];
 
        tos &= IPTOS_RT_MASK;
 
@@ -992,11 +993,15 @@
                        goto reject_redirect;
        }
 
+       toskeys[0] = tos;
+       toskeys[1] = tos | RTO_ONLINK;
+       if (saddr && daddr)
+       for (j = 0; j < 2; j++)
        for (i = 0; i < 2; i++) {
                for (k = 0; k < 2; k++) {
                        unsigned hash = rt_hash_code(daddr,
                                                     skeys[i] ^ (ikeys[k] << 5),
-                                                    tos);
+                                                    toskeys[j]);
 
                        rthp=&rt_hash_table[hash].chain;
 
@@ -1007,7 +1012,7 @@
                                smp_read_barrier_depends();
                                if (rth->fl.fl4_dst != daddr ||
                                    rth->fl.fl4_src != skeys[i] ||
-                                   rth->fl.fl4_tos != tos ||
+                                   rth->fl.fl4_tos != toskeys[j] ||
                                    rth->fl.oif != ikeys[k] ||
                                    rth->fl.iif != 0) {
                                        rthp = &rth->u.rt_next;
@@ -1237,21 +1242,26 @@
        return 68;
 }
 
-unsigned short ip_rt_frag_needed(struct iphdr *iph, unsigned short new_mtu)
+unsigned short ip_rt_frag_needed(int iif, 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  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 +1271,8 @@
                            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;
@@ -2213,8 +2224,9 @@
 {
        unsigned hash;
        struct rtable *rth;
+       u8 tos = flp->fl4_tos & (IPTOS_RT_MASK | RTO_ONLINK);
 
-       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), tos);
 
        rcu_read_lock();
        for (rth = rt_hash_table[hash].chain; rth; rth = rth->u.rt_next) {
@@ -2226,8 +2238,7 @@
 #ifdef CONFIG_IP_ROUTE_FWMARK
                    rth->fl.fl4_fwmark == flp->fl4_fwmark &&
 #endif
-                   !((rth->fl.fl4_tos ^ flp->fl4_tos) &
-                           (IPTOS_RT_MASK | RTO_ONLINK))) {
+                   rth->fl.fl4_tos == tos) {
                        rth->u.dst.lastuse = jiffies;
                        dst_hold(&rth->u.dst);
                        rth->u.dst.__use++;

Regards

--
Julian Anastasov <ja@xxxxxx>