ipv6 sit: RCU conversion phase II
authorEric Dumazet <eric.dumazet@gmail.com>
Fri, 23 Oct 2009 17:52:14 +0000 (17:52 +0000)
committerDavid S. Miller <davem@davemloft.net>
Sat, 24 Oct 2009 13:07:56 +0000 (06:07 -0700)
SIT tunnels use one rwlock to protect their hash tables.

This locking scheme can be converted to RCU for free, since netdevice
already must wait for a RCU grace period at dismantle time.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/ipv6/sit.c

index 8cdcc2a..b6b1626 100644 (file)
@@ -77,8 +77,17 @@ struct sit_net {
        struct net_device *fb_tunnel_dev;
 };
 
-static DEFINE_RWLOCK(ipip6_lock);
+/*
+ * Locking : hash tables are protected by RCU and a spinlock
+ */
+static DEFINE_SPINLOCK(ipip6_lock);
+
+#define for_each_ip_tunnel_rcu(start) \
+       for (t = rcu_dereference(start); t; t = rcu_dereference(t->next))
 
+/*
+ * Must be invoked with rcu_read_lock
+ */
 static struct ip_tunnel * ipip6_tunnel_lookup(struct net *net,
                struct net_device *dev, __be32 remote, __be32 local)
 {
@@ -87,26 +96,26 @@ static struct ip_tunnel * ipip6_tunnel_lookup(struct net *net,
        struct ip_tunnel *t;
        struct sit_net *sitn = net_generic(net, sit_net_id);
 
-       for (t = sitn->tunnels_r_l[h0^h1]; t; t = t->next) {
+       for_each_ip_tunnel_rcu(sitn->tunnels_r_l[h0 ^ h1]) {
                if (local == t->parms.iph.saddr &&
                    remote == t->parms.iph.daddr &&
                    (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
                    (t->dev->flags & IFF_UP))
                        return t;
        }
-       for (t = sitn->tunnels_r[h0]; t; t = t->next) {
+       for_each_ip_tunnel_rcu(sitn->tunnels_r[h0]) {
                if (remote == t->parms.iph.daddr &&
                    (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
                    (t->dev->flags & IFF_UP))
                        return t;
        }
-       for (t = sitn->tunnels_l[h1]; t; t = t->next) {
+       for_each_ip_tunnel_rcu(sitn->tunnels_l[h1]) {
                if (local == t->parms.iph.saddr &&
                    (!dev || !t->parms.link || dev->iflink == t->parms.link) &&
                    (t->dev->flags & IFF_UP))
                        return t;
        }
-       t = sitn->tunnels_wc[0];
+       t = rcu_dereference(sitn->tunnels_wc[0]);
        if ((t != NULL) && (t->dev->flags & IFF_UP))
                return t;
        return NULL;
@@ -143,9 +152,9 @@ static void ipip6_tunnel_unlink(struct sit_net *sitn, struct ip_tunnel *t)
 
        for (tp = ipip6_bucket(sitn, t); *tp; tp = &(*tp)->next) {
                if (t == *tp) {
-                       write_lock_bh(&ipip6_lock);
+                       spin_lock_bh(&ipip6_lock);
                        *tp = t->next;
-                       write_unlock_bh(&ipip6_lock);
+                       spin_unlock_bh(&ipip6_lock);
                        break;
                }
        }
@@ -155,10 +164,10 @@ static void ipip6_tunnel_link(struct sit_net *sitn, struct ip_tunnel *t)
 {
        struct ip_tunnel **tp = ipip6_bucket(sitn, t);
 
+       spin_lock_bh(&ipip6_lock);
        t->next = *tp;
-       write_lock_bh(&ipip6_lock);
-       *tp = t;
-       write_unlock_bh(&ipip6_lock);
+       rcu_assign_pointer(*tp, t);
+       spin_unlock_bh(&ipip6_lock);
 }
 
 static void ipip6_tunnel_clone_6rd(struct net_device *dev, struct sit_net *sitn)
@@ -447,9 +456,9 @@ static void ipip6_tunnel_uninit(struct net_device *dev)
        struct sit_net *sitn = net_generic(net, sit_net_id);
 
        if (dev == sitn->fb_tunnel_dev) {
-               write_lock_bh(&ipip6_lock);
+               spin_lock_bh(&ipip6_lock);
                sitn->tunnels_wc[0] = NULL;
-               write_unlock_bh(&ipip6_lock);
+               spin_unlock_bh(&ipip6_lock);
                dev_put(dev);
        } else {
                ipip6_tunnel_unlink(sitn, netdev_priv(dev));
@@ -502,7 +511,7 @@ static int ipip6_err(struct sk_buff *skb, u32 info)
 
        err = -ENOENT;
 
-       read_lock(&ipip6_lock);
+       rcu_read_lock();
        t = ipip6_tunnel_lookup(dev_net(skb->dev),
                                skb->dev,
                                iph->daddr,
@@ -520,7 +529,7 @@ static int ipip6_err(struct sk_buff *skb, u32 info)
                t->err_count = 1;
        t->err_time = jiffies;
 out:
-       read_unlock(&ipip6_lock);
+       rcu_read_unlock();
        return err;
 }
 
@@ -540,7 +549,7 @@ static int ipip6_rcv(struct sk_buff *skb)
 
        iph = ip_hdr(skb);
 
-       read_lock(&ipip6_lock);
+       rcu_read_lock();
        tunnel = ipip6_tunnel_lookup(dev_net(skb->dev), skb->dev,
                                     iph->saddr, iph->daddr);
        if (tunnel != NULL) {
@@ -554,7 +563,7 @@ static int ipip6_rcv(struct sk_buff *skb)
                if ((tunnel->dev->priv_flags & IFF_ISATAP) &&
                    !isatap_chksrc(skb, iph, tunnel)) {
                        tunnel->dev->stats.rx_errors++;
-                       read_unlock(&ipip6_lock);
+                       rcu_read_unlock();
                        kfree_skb(skb);
                        return 0;
                }
@@ -565,12 +574,12 @@ static int ipip6_rcv(struct sk_buff *skb)
                nf_reset(skb);
                ipip6_ecn_decapsulate(iph, skb);
                netif_rx(skb);
-               read_unlock(&ipip6_lock);
+               rcu_read_unlock();
                return 0;
        }
 
        icmp_send(skb, ICMP_DEST_UNREACH, ICMP_PORT_UNREACH, 0);
-       read_unlock(&ipip6_lock);
+       rcu_read_unlock();
 out:
        kfree_skb(skb);
        return 0;