ipv6 sit: RCU conversion phase I
Eric Dumazet [Fri, 23 Oct 2009 17:51:26 +0000 (17:51 +0000)]
SIT tunnels use one rwlock to protect their prl entries.

This first patch adds RCU locking for prl management,
with standard call_rcu() calls.

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

include/net/ipip.h
net/ipv6/sit.c

index 86f1c8b..b3db2fd 100644 (file)
@@ -45,6 +45,7 @@ struct ip_tunnel_prl_entry
        struct ip_tunnel_prl_entry      *next;
        __be32                          addr;
        u16                             flags;
+       struct rcu_head                 rcu_head;
 };
 
 #define IPTUNNEL_XMIT() do {                                           \
index 510d31f..8cdcc2a 100644 (file)
@@ -240,15 +240,22 @@ failed:
        return NULL;
 }
 
+static DEFINE_SPINLOCK(ipip6_prl_lock);
+
+#define for_each_prl_rcu(start)                        \
+       for (prl = rcu_dereference(start);      \
+            prl;                               \
+            prl = rcu_dereference(prl->next))
+
 static struct ip_tunnel_prl_entry *
 __ipip6_tunnel_locate_prl(struct ip_tunnel *t, __be32 addr)
 {
-       struct ip_tunnel_prl_entry *p = (struct ip_tunnel_prl_entry *)NULL;
+       struct ip_tunnel_prl_entry *prl;
 
-       for (p = t->prl; p; p = p->next)
-               if (p->addr == addr)
+       for_each_prl_rcu(t->prl)
+               if (prl->addr == addr)
                        break;
-       return p;
+       return prl;
 
 }
 
@@ -273,7 +280,7 @@ static int ipip6_tunnel_get_prl(struct ip_tunnel *t,
                kcalloc(cmax, sizeof(*kp), GFP_KERNEL) :
                NULL;
 
-       read_lock(&ipip6_lock);
+       rcu_read_lock();
 
        ca = t->prl_count < cmax ? t->prl_count : cmax;
 
@@ -291,7 +298,7 @@ static int ipip6_tunnel_get_prl(struct ip_tunnel *t,
        }
 
        c = 0;
-       for (prl = t->prl; prl; prl = prl->next) {
+       for_each_prl_rcu(t->prl) {
                if (c >= cmax)
                        break;
                if (kprl.addr != htonl(INADDR_ANY) && prl->addr != kprl.addr)
@@ -303,7 +310,7 @@ static int ipip6_tunnel_get_prl(struct ip_tunnel *t,
                        break;
        }
 out:
-       read_unlock(&ipip6_lock);
+       rcu_read_unlock();
 
        len = sizeof(*kp) * c;
        ret = 0;
@@ -324,12 +331,14 @@ ipip6_tunnel_add_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a, int chg)
        if (a->addr == htonl(INADDR_ANY))
                return -EINVAL;
 
-       write_lock(&ipip6_lock);
+       spin_lock(&ipip6_prl_lock);
 
        for (p = t->prl; p; p = p->next) {
                if (p->addr == a->addr) {
-                       if (chg)
-                               goto update;
+                       if (chg) {
+                               p->flags = a->flags;
+                               goto out;
+                       }
                        err = -EEXIST;
                        goto out;
                }
@@ -346,46 +355,63 @@ ipip6_tunnel_add_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a, int chg)
                goto out;
        }
 
+       INIT_RCU_HEAD(&p->rcu_head);
        p->next = t->prl;
-       t->prl = p;
-       t->prl_count++;
-update:
        p->addr = a->addr;
        p->flags = a->flags;
+       t->prl_count++;
+       rcu_assign_pointer(t->prl, p);
 out:
-       write_unlock(&ipip6_lock);
+       spin_unlock(&ipip6_prl_lock);
        return err;
 }
 
+static void prl_entry_destroy_rcu(struct rcu_head *head)
+{
+       kfree(container_of(head, struct ip_tunnel_prl_entry, rcu_head));
+}
+
+static void prl_list_destroy_rcu(struct rcu_head *head)
+{
+       struct ip_tunnel_prl_entry *p, *n;
+
+       p = container_of(head, struct ip_tunnel_prl_entry, rcu_head);
+       do {
+               n = p->next;
+               kfree(p);
+               p = n;
+       } while (p);
+}
+
 static int
 ipip6_tunnel_del_prl(struct ip_tunnel *t, struct ip_tunnel_prl *a)
 {
        struct ip_tunnel_prl_entry *x, **p;
        int err = 0;
 
-       write_lock(&ipip6_lock);
+       spin_lock(&ipip6_prl_lock);
 
        if (a && a->addr != htonl(INADDR_ANY)) {
                for (p = &t->prl; *p; p = &(*p)->next) {
                        if ((*p)->addr == a->addr) {
                                x = *p;
                                *p = x->next;
-                               kfree(x);
+                               call_rcu(&x->rcu_head, prl_entry_destroy_rcu);
                                t->prl_count--;
                                goto out;
                        }
                }
                err = -ENXIO;
        } else {
-               while (t->prl) {
+               if (t->prl) {
+                       t->prl_count = 0;
                        x = t->prl;
-                       t->prl = t->prl->next;
-                       kfree(x);
-                       t->prl_count--;
+                       call_rcu(&x->rcu_head, prl_list_destroy_rcu);
+                       t->prl = NULL;
                }
        }
 out:
-       write_unlock(&ipip6_lock);
+       spin_unlock(&ipip6_prl_lock);
        return err;
 }
 
@@ -395,7 +421,7 @@ isatap_chksrc(struct sk_buff *skb, struct iphdr *iph, struct ip_tunnel *t)
        struct ip_tunnel_prl_entry *p;
        int ok = 1;
 
-       read_lock(&ipip6_lock);
+       rcu_read_lock();
        p = __ipip6_tunnel_locate_prl(t, iph->saddr);
        if (p) {
                if (p->flags & PRL_DEFAULT)
@@ -411,7 +437,7 @@ isatap_chksrc(struct sk_buff *skb, struct iphdr *iph, struct ip_tunnel *t)
                else
                        ok = 0;
        }
-       read_unlock(&ipip6_lock);
+       rcu_read_unlock();
        return ok;
 }
 
@@ -1192,6 +1218,7 @@ static void __exit sit_cleanup(void)
        xfrm4_tunnel_deregister(&sit_handler, AF_INET6);
 
        unregister_pernet_gen_device(sit_net_id, &sit_net_ops);
+       rcu_barrier(); /* Wait for completion of call_rcu()'s */
 }
 
 static int __init sit_init(void)