ip6tnl: less dev_put() calls
[safe/jmp/linux-2.6] / net / ipv6 / ip6_tunnel.c
index 404d16a..1d61411 100644 (file)
@@ -88,8 +88,10 @@ struct ip6_tnl_net {
        struct ip6_tnl **tnls[2];
 };
 
-/* lock for the tunnel lists */
-static DEFINE_RWLOCK(ip6_tnl_lock);
+/*
+ * Locking : hash tables are protected by RCU and a spinlock
+ */
+static DEFINE_SPINLOCK(ip6_tnl_lock);
 
 static inline struct dst_entry *ip6_tnl_dst_check(struct ip6_tnl *t)
 {
@@ -130,6 +132,9 @@ static inline void ip6_tnl_dst_store(struct ip6_tnl *t, struct dst_entry *dst)
  *   else %NULL
  **/
 
+#define for_each_ip6_tunnel_rcu(start) \
+       for (t = rcu_dereference(start); t; t = rcu_dereference(t->next))
+
 static struct ip6_tnl *
 ip6_tnl_lookup(struct net *net, struct in6_addr *remote, struct in6_addr *local)
 {
@@ -138,13 +143,14 @@ ip6_tnl_lookup(struct net *net, struct in6_addr *remote, struct in6_addr *local)
        struct ip6_tnl *t;
        struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
 
-       for (t = ip6n->tnls_r_l[h0 ^ h1]; t; t = t->next) {
+       for_each_ip6_tunnel_rcu(ip6n->tnls_r_l[h0 ^ h1]) {
                if (ipv6_addr_equal(local, &t->parms.laddr) &&
                    ipv6_addr_equal(remote, &t->parms.raddr) &&
                    (t->dev->flags & IFF_UP))
                        return t;
        }
-       if ((t = ip6n->tnls_wc[0]) != NULL && (t->dev->flags & IFF_UP))
+       t = rcu_dereference(ip6n->tnls_wc[0]);
+       if (t && (t->dev->flags & IFF_UP))
                return t;
 
        return NULL;
@@ -186,10 +192,10 @@ ip6_tnl_link(struct ip6_tnl_net *ip6n, struct ip6_tnl *t)
 {
        struct ip6_tnl **tp = ip6_tnl_bucket(ip6n, &t->parms);
 
+       spin_lock_bh(&ip6_tnl_lock);
        t->next = *tp;
-       write_lock_bh(&ip6_tnl_lock);
-       *tp = t;
-       write_unlock_bh(&ip6_tnl_lock);
+       rcu_assign_pointer(*tp, t);
+       spin_unlock_bh(&ip6_tnl_lock);
 }
 
 /**
@@ -204,9 +210,9 @@ ip6_tnl_unlink(struct ip6_tnl_net *ip6n, struct ip6_tnl *t)
 
        for (tp = ip6_tnl_bucket(ip6n, &t->parms); *tp; tp = &(*tp)->next) {
                if (t == *tp) {
-                       write_lock_bh(&ip6_tnl_lock);
+                       spin_lock_bh(&ip6_tnl_lock);
                        *tp = t->next;
-                       write_unlock_bh(&ip6_tnl_lock);
+                       spin_unlock_bh(&ip6_tnl_lock);
                        break;
                }
        }
@@ -313,9 +319,9 @@ ip6_tnl_dev_uninit(struct net_device *dev)
        struct ip6_tnl_net *ip6n = net_generic(net, ip6_tnl_net_id);
 
        if (dev == ip6n->fb_tnl_dev) {
-               write_lock_bh(&ip6_tnl_lock);
+               spin_lock_bh(&ip6_tnl_lock);
                ip6n->tnls_wc[0] = NULL;
-               write_unlock_bh(&ip6_tnl_lock);
+               spin_unlock_bh(&ip6_tnl_lock);
        } else {
                ip6_tnl_unlink(ip6n, t);
        }
@@ -394,13 +400,13 @@ parse_tlv_tnl_enc_lim(struct sk_buff *skb, __u8 * raw)
 
 static int
 ip6_tnl_err(struct sk_buff *skb, __u8 ipproto, struct inet6_skb_parm *opt,
-           int *type, int *code, int *msg, __u32 *info, int offset)
+           u8 *type, u8 *code, int *msg, __u32 *info, int offset)
 {
        struct ipv6hdr *ipv6h = (struct ipv6hdr *) skb->data;
        struct ip6_tnl *t;
        int rel_msg = 0;
-       int rel_type = ICMPV6_DEST_UNREACH;
-       int rel_code = ICMPV6_ADDR_UNREACH;
+       u8 rel_type = ICMPV6_DEST_UNREACH;
+       u8 rel_code = ICMPV6_ADDR_UNREACH;
        __u32 rel_info = 0;
        __u16 len;
        int err = -ENOENT;
@@ -409,7 +415,7 @@ ip6_tnl_err(struct sk_buff *skb, __u8 ipproto, struct inet6_skb_parm *opt,
           in trouble since we might need the source address for further
           processing of the error. */
 
-       read_lock(&ip6_tnl_lock);
+       rcu_read_lock();
        if ((t = ip6_tnl_lookup(dev_net(skb->dev), &ipv6h->daddr,
                                        &ipv6h->saddr)) == NULL)
                goto out;
@@ -482,17 +488,17 @@ ip6_tnl_err(struct sk_buff *skb, __u8 ipproto, struct inet6_skb_parm *opt,
        *msg = rel_msg;
 
 out:
-       read_unlock(&ip6_tnl_lock);
+       rcu_read_unlock();
        return err;
 }
 
 static int
 ip4ip6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
-          int type, int code, int offset, __be32 info)
+          u8 type, u8 code, int offset, __be32 info)
 {
        int rel_msg = 0;
-       int rel_type = type;
-       int rel_code = code;
+       u8 rel_type = type;
+       u8 rel_code = code;
        __u32 rel_info = ntohl(info);
        int err;
        struct sk_buff *skb2;
@@ -586,11 +592,11 @@ out:
 
 static int
 ip6ip6_err(struct sk_buff *skb, struct inet6_skb_parm *opt,
-          int type, int code, int offset, __be32 info)
+          u8 type, u8 code, int offset, __be32 info)
 {
        int rel_msg = 0;
-       int rel_type = type;
-       int rel_code = code;
+       u8 rel_type = type;
+       u8 rel_code = code;
        __u32 rel_info = ntohl(info);
        int err;
 
@@ -652,6 +658,7 @@ static void ip6ip6_dscp_ecn_decapsulate(struct ip6_tnl *t,
                IP6_ECN_set_ce(ipv6_hdr(skb));
 }
 
+/* called with rcu_read_lock() */
 static inline int ip6_tnl_rcv_ctl(struct ip6_tnl *t)
 {
        struct ip6_tnl_parm *p = &t->parms;
@@ -662,15 +669,13 @@ static inline int ip6_tnl_rcv_ctl(struct ip6_tnl *t)
                struct net_device *ldev = NULL;
 
                if (p->link)
-                       ldev = dev_get_by_index(net, p->link);
+                       ldev = dev_get_by_index_rcu(net, p->link);
 
                if ((ipv6_addr_is_multicast(&p->laddr) ||
                     likely(ipv6_chk_addr(net, &p->laddr, ldev, 0))) &&
                    likely(!ipv6_chk_addr(net, &p->raddr, NULL, 0)))
                        ret = 1;
 
-               if (ldev)
-                       dev_put(ldev);
        }
        return ret;
 }
@@ -693,23 +698,23 @@ static int ip6_tnl_rcv(struct sk_buff *skb, __u16 protocol,
        struct ip6_tnl *t;
        struct ipv6hdr *ipv6h = ipv6_hdr(skb);
 
-       read_lock(&ip6_tnl_lock);
+       rcu_read_lock();
 
        if ((t = ip6_tnl_lookup(dev_net(skb->dev), &ipv6h->saddr,
                                        &ipv6h->daddr)) != NULL) {
                if (t->parms.proto != ipproto && t->parms.proto != 0) {
-                       read_unlock(&ip6_tnl_lock);
+                       rcu_read_unlock();
                        goto discard;
                }
 
                if (!xfrm6_policy_check(NULL, XFRM_POLICY_IN, skb)) {
-                       read_unlock(&ip6_tnl_lock);
+                       rcu_read_unlock();
                        goto discard;
                }
 
                if (!ip6_tnl_rcv_ctl(t)) {
                        t->dev->stats.rx_dropped++;
-                       read_unlock(&ip6_tnl_lock);
+                       rcu_read_unlock();
                        goto discard;
                }
                secpath_reset(skb);
@@ -727,10 +732,10 @@ static int ip6_tnl_rcv(struct sk_buff *skb, __u16 protocol,
                t->dev->stats.rx_packets++;
                t->dev->stats.rx_bytes += skb->len;
                netif_rx(skb);
-               read_unlock(&ip6_tnl_lock);
+               rcu_read_unlock();
                return 0;
        }
-       read_unlock(&ip6_tnl_lock);
+       rcu_read_unlock();
        return 1;
 
 discard:
@@ -798,8 +803,9 @@ static inline int ip6_tnl_xmit_ctl(struct ip6_tnl *t)
        if (p->flags & IP6_TNL_F_CAP_XMIT) {
                struct net_device *ldev = NULL;
 
+               rcu_read_lock();
                if (p->link)
-                       ldev = dev_get_by_index(net, p->link);
+                       ldev = dev_get_by_index_rcu(net, p->link);
 
                if (unlikely(!ipv6_chk_addr(net, &p->laddr, ldev, 0)))
                        printk(KERN_WARNING
@@ -813,8 +819,7 @@ static inline int ip6_tnl_xmit_ctl(struct ip6_tnl *t)
                               p->name);
                else
                        ret = 1;
-               if (ldev)
-                       dev_put(ldev);
+               rcu_read_unlock();
        }
        return ret;
 }
@@ -1036,18 +1041,13 @@ ip6ip6_tnl_xmit(struct sk_buff *skb, struct net_device *dev)
        return 0;
 }
 
-static int
+static netdev_tx_t
 ip6_tnl_xmit(struct sk_buff *skb, struct net_device *dev)
 {
        struct ip6_tnl *t = netdev_priv(dev);
        struct net_device_stats *stats = &t->dev->stats;
        int ret;
 
-       if (t->recursion++) {
-               stats->collisions++;
-               goto tx_err;
-       }
-
        switch (skb->protocol) {
        case htons(ETH_P_IP):
                ret = ip4ip6_tnl_xmit(skb, dev);
@@ -1062,15 +1062,13 @@ ip6_tnl_xmit(struct sk_buff *skb, struct net_device *dev)
        if (ret < 0)
                goto tx_err;
 
-       t->recursion--;
-       return 0;
+       return NETDEV_TX_OK;
 
 tx_err:
        stats->tx_errors++;
        stats->tx_dropped++;
        kfree_skb(skb);
-       t->recursion--;
-       return 0;
+       return NETDEV_TX_OK;
 }
 
 static void ip6_tnl_set_cap(struct ip6_tnl *t)
@@ -1394,14 +1392,19 @@ static void ip6_tnl_destroy_tunnels(struct ip6_tnl_net *ip6n)
 {
        int h;
        struct ip6_tnl *t;
+       LIST_HEAD(list);
 
        for (h = 0; h < HASH_SIZE; h++) {
-               while ((t = ip6n->tnls_r_l[h]) != NULL)
-                       unregister_netdevice(t->dev);
+               t = ip6n->tnls_r_l[h];
+               while (t != NULL) {
+                       unregister_netdevice_queue(t->dev, &list);
+                       t = t->next;
+               }
        }
 
        t = ip6n->tnls_wc[0];
-       unregister_netdevice(t->dev);
+       unregister_netdevice_queue(t->dev, &list);
+       unregister_netdevice_many(&list);
 }
 
 static int ip6_tnl_init_net(struct net *net)