sysctl: remove "struct file *" argument of ->proc_handler
[safe/jmp/linux-2.6] / net / ipv4 / route.c
index f20060a..bb41992 100644 (file)
@@ -131,8 +131,8 @@ static int ip_rt_min_advmss __read_mostly   = 256;
 static int ip_rt_secret_interval __read_mostly = 10 * 60 * HZ;
 static int rt_chain_length_max __read_mostly   = 20;
 
-static void rt_worker_func(struct work_struct *work);
-static DECLARE_DELAYED_WORK(expires_work, rt_worker_func);
+static struct delayed_work expires_work;
+static unsigned long expires_ljiffies;
 
 /*
  *     Interface to generic destination cache.
@@ -787,9 +787,12 @@ static void rt_check_expire(void)
        struct rtable *rth, *aux, **rthp;
        unsigned long samples = 0;
        unsigned long sum = 0, sum2 = 0;
+       unsigned long delta;
        u64 mult;
 
-       mult = ((u64)ip_rt_gc_interval) << rt_hash_log;
+       delta = jiffies - expires_ljiffies;
+       expires_ljiffies = jiffies;
+       mult = ((u64)delta) << rt_hash_log;
        if (ip_rt_gc_timeout > 1)
                do_div(mult, ip_rt_gc_timeout);
        goal = (unsigned int)mult;
@@ -1082,8 +1085,35 @@ restart:
        now = jiffies;
 
        if (!rt_caching(dev_net(rt->u.dst.dev))) {
-               rt_drop(rt);
-               return 0;
+               /*
+                * If we're not caching, just tell the caller we
+                * were successful and don't touch the route.  The
+                * caller hold the sole reference to the cache entry, and
+                * it will be released when the caller is done with it.
+                * If we drop it here, the callers have no way to resolve routes
+                * when we're not caching.  Instead, just point *rp at rt, so
+                * the caller gets a single use out of the route
+                * Note that we do rt_free on this new route entry, so that
+                * once its refcount hits zero, we are still able to reap it
+                * (Thanks Alexey)
+                * Note also the rt_free uses call_rcu.  We don't actually
+                * need rcu protection here, this is just our path to get
+                * on the route gc list.
+                */
+
+               if (rt->rt_type == RTN_UNICAST || rt->fl.iif == 0) {
+                       int err = arp_bind_neighbour(&rt->u.dst);
+                       if (err) {
+                               if (net_ratelimit())
+                                       printk(KERN_WARNING
+                                           "Neighbour table failure & not caching routes.\n");
+                               rt_drop(rt);
+                               return err;
+                       }
+               }
+
+               rt_free(rt);
+               goto skip_hashing;
        }
 
        rthp = &rt_hash_table[hash].chain;
@@ -1118,7 +1148,7 @@ restart:
                        if (rp)
                                *rp = rth;
                        else
-                               skb->dst = &rth->u.dst;
+                               skb_dst_set(skb, &rth->u.dst);
                        return 0;
                }
 
@@ -1200,7 +1230,8 @@ restart:
 #if RT_CACHE_DEBUG >= 2
        if (rt->u.dst.rt_next) {
                struct rtable *trt;
-               printk(KERN_DEBUG "rt_cache @%02x: %pI4", hash, &rt->rt_dst);
+               printk(KERN_DEBUG "rt_cache @%02x: %pI4",
+                      hash, &rt->rt_dst);
                for (trt = rt->u.dst.rt_next; trt; trt = trt->u.dst.rt_next)
                        printk(" . %pI4", &trt->rt_dst);
                printk("\n");
@@ -1214,10 +1245,12 @@ restart:
        rcu_assign_pointer(rt_hash_table[hash].chain, rt);
 
        spin_unlock_bh(rt_hash_lock_addr(hash));
+
+skip_hashing:
        if (rp)
                *rp = rt;
        else
-               skb->dst = &rt->u.dst;
+               skb_dst_set(skb, &rt->u.dst);
        return 0;
 }
 
@@ -1481,13 +1514,17 @@ static struct dst_entry *ipv4_negative_advice(struct dst_entry *dst)
 void ip_rt_send_redirect(struct sk_buff *skb)
 {
        struct rtable *rt = skb_rtable(skb);
-       struct in_device *in_dev = in_dev_get(rt->u.dst.dev);
+       struct in_device *in_dev;
+       int log_martians;
 
-       if (!in_dev)
+       rcu_read_lock();
+       in_dev = __in_dev_get_rcu(rt->u.dst.dev);
+       if (!in_dev || !IN_DEV_TX_REDIRECTS(in_dev)) {
+               rcu_read_unlock();
                return;
-
-       if (!IN_DEV_TX_REDIRECTS(in_dev))
-               goto out;
+       }
+       log_martians = IN_DEV_LOG_MARTIANS(in_dev);
+       rcu_read_unlock();
 
        /* No redirected packets during ip_rt_redirect_silence;
         * reset the algorithm.
@@ -1500,7 +1537,7 @@ void ip_rt_send_redirect(struct sk_buff *skb)
         */
        if (rt->u.dst.rate_tokens >= ip_rt_redirect_number) {
                rt->u.dst.rate_last = jiffies;
-               goto out;
+               return;
        }
 
        /* Check for load limit; set rate_last to the latest sent
@@ -1514,7 +1551,7 @@ void ip_rt_send_redirect(struct sk_buff *skb)
                rt->u.dst.rate_last = jiffies;
                ++rt->u.dst.rate_tokens;
 #ifdef CONFIG_IP_ROUTE_VERBOSE
-               if (IN_DEV_LOG_MARTIANS(in_dev) &&
+               if (log_martians &&
                    rt->u.dst.rate_tokens == ip_rt_redirect_number &&
                    net_ratelimit())
                        printk(KERN_WARNING "host %pI4/if%d ignores redirects for %pI4 to %pI4.\n",
@@ -1522,8 +1559,6 @@ void ip_rt_send_redirect(struct sk_buff *skb)
                                &rt->rt_dst, &rt->rt_gateway);
 #endif
        }
-out:
-       in_dev_put(in_dev);
 }
 
 static int ip_error(struct sk_buff *skb)
@@ -2251,7 +2286,7 @@ int ip_route_input(struct sk_buff *skb, __be32 daddr, __be32 saddr,
                        dst_use(&rth->u.dst, jiffies);
                        RT_CACHE_STAT_INC(in_hit);
                        rcu_read_unlock();
-                       skb->dst = &rth->u.dst;
+                       skb_dst_set(skb, &rth->u.dst);
                        return 0;
                }
                RT_CACHE_STAT_INC(in_hlist_search);
@@ -2934,7 +2969,7 @@ static int inet_rtm_getroute(struct sk_buff *in_skb, struct nlmsghdr* nlh, void
        if (err)
                goto errout_free;
 
-       skb->dst = &rt->u.dst;
+       skb_dst_set(skb, &rt->u.dst);
        if (rtm->rtm_flags & RTM_F_NOTIFY)
                rt->rt_flags |= RTCF_NOTIFY;
 
@@ -2975,15 +3010,15 @@ int ip_rt_dump(struct sk_buff *skb,  struct netlink_callback *cb)
                                continue;
                        if (rt_is_expired(rt))
                                continue;
-                       skb->dst = dst_clone(&rt->u.dst);
+                       skb_dst_set(skb, dst_clone(&rt->u.dst));
                        if (rt_fill_info(net, skb, NETLINK_CB(cb->skb).pid,
                                         cb->nlh->nlmsg_seq, RTM_NEWROUTE,
                                         1, NLM_F_MULTI) <= 0) {
-                               dst_release(xchg(&skb->dst, NULL));
+                               skb_dst_drop(skb);
                                rcu_read_unlock_bh();
                                goto done;
                        }
-                       dst_release(xchg(&skb->dst, NULL));
+                       skb_dst_drop(skb);
                }
                rcu_read_unlock_bh();
        }
@@ -3001,7 +3036,7 @@ void ip_rt_multicast_event(struct in_device *in_dev)
 
 #ifdef CONFIG_SYSCTL
 static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
-                                       struct file *filp, void __user *buffer,
+                                       void __user *buffer,
                                        size_t *lenp, loff_t *ppos)
 {
        if (write) {
@@ -3011,7 +3046,7 @@ static int ipv4_sysctl_rtcache_flush(ctl_table *__ctl, int write,
 
                memcpy(&ctl, __ctl, sizeof(ctl));
                ctl.data = &flush_delay;
-               proc_dointvec(&ctl, write, filp, buffer, lenp, ppos);
+               proc_dointvec(&ctl, write, buffer, lenp, ppos);
 
                net = (struct net *)__ctl->extra1;
                rt_cache_flush(net, flush_delay);
@@ -3071,12 +3106,11 @@ static void rt_secret_reschedule(int old)
 }
 
 static int ipv4_sysctl_rt_secret_interval(ctl_table *ctl, int write,
-                                         struct file *filp,
                                          void __user *buffer, size_t *lenp,
                                          loff_t *ppos)
 {
        int old = ip_rt_secret_interval;
-       int ret = proc_dointvec_jiffies(ctl, write, filp, buffer, lenp, ppos);
+       int ret = proc_dointvec_jiffies(ctl, write, buffer, lenp, ppos);
 
        rt_secret_reschedule(old);
 
@@ -3379,7 +3413,7 @@ int __init ip_rt_init(void)
                alloc_large_system_hash("IP route cache",
                                        sizeof(struct rt_hash_bucket),
                                        rhash_entries,
-                                       (num_physpages >= 128 * 1024) ?
+                                       (totalram_pages >= 128 * 1024) ?
                                        15 : 17,
                                        0,
                                        &rt_hash_log,
@@ -3397,6 +3431,8 @@ int __init ip_rt_init(void)
        /* All the timers, started at system startup tend
           to synchronize. Perturb it a bit.
         */
+       INIT_DELAYED_WORK_DEFERRABLE(&expires_work, rt_worker_func);
+       expires_ljiffies = jiffies;
        schedule_delayed_work(&expires_work,
                net_random() % ip_rt_gc_interval + ip_rt_gc_interval);
 
@@ -3407,7 +3443,7 @@ int __init ip_rt_init(void)
                printk(KERN_ERR "Unable to create route proc files\n");
 #ifdef CONFIG_XFRM
        xfrm_init();
-       xfrm4_init();
+       xfrm4_init(ip_rt_max_size);
 #endif
        rtnl_register(PF_INET, RTM_GETROUTE, inet_rtm_getroute, NULL);