- Update to 2.6.25-rc3.
[linux-flexiantxendom0-3.2.10.git] / net / ipv4 / netfilter / nf_nat_core.c
index 86b465b..0d5fa3a 100644 (file)
 #include <net/netfilter/nf_conntrack_l3proto.h>
 #include <net/netfilter/nf_conntrack_l4proto.h>
 
-static DEFINE_RWLOCK(nf_nat_lock);
+static DEFINE_SPINLOCK(nf_nat_lock);
 
-static struct nf_conntrack_l3proto *l3proto = NULL;
+static struct nf_conntrack_l3proto *l3proto __read_mostly;
 
 /* Calculated at init based on memory size */
-static unsigned int nf_nat_htable_size;
+static unsigned int nf_nat_htable_size __read_mostly;
 static int nf_nat_vmalloced;
 
-static struct hlist_head *bysource;
+static struct hlist_head *bysource __read_mostly;
 
 #define MAX_IP_NAT_PROTO 256
-static struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO];
+static const struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO]
+                                               __read_mostly;
 
-static inline struct nf_nat_protocol *
+static inline const struct nf_nat_protocol *
 __nf_nat_proto_find(u_int8_t protonum)
 {
        return rcu_dereference(nf_nat_protos[protonum]);
 }
 
-struct nf_nat_protocol *
+const struct nf_nat_protocol *
 nf_nat_proto_find_get(u_int8_t protonum)
 {
-       struct nf_nat_protocol *p;
+       const struct nf_nat_protocol *p;
 
        rcu_read_lock();
        p = __nf_nat_proto_find(protonum);
@@ -66,7 +67,7 @@ nf_nat_proto_find_get(u_int8_t protonum)
 EXPORT_SYMBOL_GPL(nf_nat_proto_find_get);
 
 void
-nf_nat_proto_put(struct nf_nat_protocol *p)
+nf_nat_proto_put(const struct nf_nat_protocol *p)
 {
        module_put(p->me);
 }
@@ -76,10 +77,13 @@ EXPORT_SYMBOL_GPL(nf_nat_proto_put);
 static inline unsigned int
 hash_by_src(const struct nf_conntrack_tuple *tuple)
 {
+       unsigned int hash;
+
        /* Original src, to ensure we map it consistently if poss. */
-       return jhash_3words((__force u32)tuple->src.u3.ip,
+       hash = jhash_3words((__force u32)tuple->src.u3.ip,
                            (__force u32)tuple->src.u.all,
-                           tuple->dst.protonum, 0) % nf_nat_htable_size;
+                           tuple->dst.protonum, 0);
+       return ((u64)hash * nf_nat_htable_size) >> 32;
 }
 
 /* Is this tuple already taken? (not by us) */
@@ -105,7 +109,7 @@ static int
 in_range(const struct nf_conntrack_tuple *tuple,
         const struct nf_nat_range *range)
 {
-       struct nf_nat_protocol *proto;
+       const struct nf_nat_protocol *proto;
        int ret = 0;
 
        /* If we are supposed to map IPs, then we must be in the
@@ -150,8 +154,8 @@ find_appropriate_src(const struct nf_conntrack_tuple *tuple,
        struct nf_conn *ct;
        struct hlist_node *n;
 
-       read_lock_bh(&nf_nat_lock);
-       hlist_for_each_entry(nat, n, &bysource[h], bysource) {
+       rcu_read_lock();
+       hlist_for_each_entry_rcu(nat, n, &bysource[h], bysource) {
                ct = nat->ct;
                if (same_src(ct, tuple)) {
                        /* Copy source part from reply tuple. */
@@ -160,12 +164,12 @@ find_appropriate_src(const struct nf_conntrack_tuple *tuple,
                        result->dst = tuple->dst;
 
                        if (in_range(result, range)) {
-                               read_unlock_bh(&nf_nat_lock);
+                               rcu_read_unlock();
                                return 1;
                        }
                }
        }
-       read_unlock_bh(&nf_nat_lock);
+       rcu_read_unlock();
        return 0;
 }
 
@@ -210,12 +214,13 @@ find_best_ips_proto(struct nf_conntrack_tuple *tuple,
        maxip = ntohl(range->max_ip);
        j = jhash_2words((__force u32)tuple->src.u3.ip,
                         (__force u32)tuple->dst.u3.ip, 0);
-       *var_ipp = htonl(minip + j % (maxip - minip + 1));
+       j = ((u64)j * (maxip - minip + 1)) >> 32;
+       *var_ipp = htonl(minip + j);
 }
 
-/* Manipulate the tuple into the range given.  For NF_IP_POST_ROUTING,
- * we change the source to map into the range.  For NF_IP_PRE_ROUTING
- * and NF_IP_LOCAL_OUT, we change the destination to map into the
+/* Manipulate the tuple into the range given.  For NF_INET_POST_ROUTING,
+ * we change the source to map into the range.  For NF_INET_PRE_ROUTING
+ * and NF_INET_LOCAL_OUT, we change the destination to map into the
  * range.  It might not be possible to get a unique tuple, but we try.
  * At worst (or if we race), we will end up with a final duplicate in
  * __ip_conntrack_confirm and drop the packet. */
@@ -226,7 +231,7 @@ get_unique_tuple(struct nf_conntrack_tuple *tuple,
                 struct nf_conn *ct,
                 enum nf_nat_manip_type maniptype)
 {
-       struct nf_nat_protocol *proto;
+       const struct nf_nat_protocol *proto;
 
        /* 1) If this srcip/proto/src-proto-part is currently mapped,
           and that same mapping gives a unique tuple within the given
@@ -276,12 +281,11 @@ out:
 unsigned int
 nf_nat_setup_info(struct nf_conn *ct,
                  const struct nf_nat_range *range,
-                 unsigned int hooknum)
+                 enum nf_nat_manip_type maniptype)
 {
        struct nf_conntrack_tuple curr_tuple, new_tuple;
        struct nf_conn_nat *nat;
        int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK);
-       enum nf_nat_manip_type maniptype = HOOK2MANIP(hooknum);
 
        /* nat helper or nfctnetlink also setup binding */
        nat = nfct_nat(ct);
@@ -293,10 +297,8 @@ nf_nat_setup_info(struct nf_conn *ct,
                }
        }
 
-       NF_CT_ASSERT(hooknum == NF_IP_PRE_ROUTING ||
-                    hooknum == NF_IP_POST_ROUTING ||
-                    hooknum == NF_IP_LOCAL_IN ||
-                    hooknum == NF_IP_LOCAL_OUT);
+       NF_CT_ASSERT(maniptype == IP_NAT_MANIP_SRC ||
+                    maniptype == IP_NAT_MANIP_DST);
        BUG_ON(nf_nat_initialized(ct, maniptype));
 
        /* What we've got will look like inverse of reply. Normally
@@ -328,12 +330,12 @@ nf_nat_setup_info(struct nf_conn *ct,
                unsigned int srchash;
 
                srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
-               write_lock_bh(&nf_nat_lock);
+               spin_lock_bh(&nf_nat_lock);
                /* nf_conntrack_alter_reply might re-allocate exntension aera */
                nat = nfct_nat(ct);
                nat->ct = ct;
-               hlist_add_head(&nat->bysource, &bysource[srchash]);
-               write_unlock_bh(&nf_nat_lock);
+               hlist_add_head_rcu(&nat->bysource, &bysource[srchash]);
+               spin_unlock_bh(&nf_nat_lock);
        }
 
        /* It's done. */
@@ -355,7 +357,7 @@ manip_pkt(u_int16_t proto,
          enum nf_nat_manip_type maniptype)
 {
        struct iphdr *iph;
-       struct nf_nat_protocol *p;
+       const struct nf_nat_protocol *p;
 
        if (!skb_make_writable(skb, iphdroff + sizeof(*iph)))
                return 0;
@@ -372,10 +374,10 @@ manip_pkt(u_int16_t proto,
        iph = (void *)skb->data + iphdroff;
 
        if (maniptype == IP_NAT_MANIP_SRC) {
-               nf_csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
+               csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
                iph->saddr = target->src.u3.ip;
        } else {
-               nf_csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
+               csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
                iph->daddr = target->dst.u3.ip;
        }
        return 1;
@@ -515,29 +517,29 @@ int nf_nat_icmp_reply_translation(struct nf_conn *ct,
 EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
 
 /* Protocol registration. */
-int nf_nat_protocol_register(struct nf_nat_protocol *proto)
+int nf_nat_protocol_register(const struct nf_nat_protocol *proto)
 {
        int ret = 0;
 
-       write_lock_bh(&nf_nat_lock);
+       spin_lock_bh(&nf_nat_lock);
        if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) {
                ret = -EBUSY;
                goto out;
        }
        rcu_assign_pointer(nf_nat_protos[proto->protonum], proto);
  out:
-       write_unlock_bh(&nf_nat_lock);
+       spin_unlock_bh(&nf_nat_lock);
        return ret;
 }
 EXPORT_SYMBOL(nf_nat_protocol_register);
 
 /* Noone stores the protocol anywhere; simply delete it. */
-void nf_nat_protocol_unregister(struct nf_nat_protocol *proto)
+void nf_nat_protocol_unregister(const struct nf_nat_protocol *proto)
 {
-       write_lock_bh(&nf_nat_lock);
+       spin_lock_bh(&nf_nat_lock);
        rcu_assign_pointer(nf_nat_protos[proto->protonum],
                           &nf_nat_unknown_protocol);
-       write_unlock_bh(&nf_nat_lock);
+       spin_unlock_bh(&nf_nat_lock);
        synchronize_rcu();
 }
 EXPORT_SYMBOL(nf_nat_protocol_unregister);
@@ -547,10 +549,8 @@ int
 nf_nat_port_range_to_nlattr(struct sk_buff *skb,
                            const struct nf_nat_range *range)
 {
-       NLA_PUT(skb, CTA_PROTONAT_PORT_MIN, sizeof(__be16),
-               &range->min.tcp.port);
-       NLA_PUT(skb, CTA_PROTONAT_PORT_MAX, sizeof(__be16),
-               &range->max.tcp.port);
+       NLA_PUT_BE16(skb, CTA_PROTONAT_PORT_MIN, range->min.tcp.port);
+       NLA_PUT_BE16(skb, CTA_PROTONAT_PORT_MAX, range->max.tcp.port);
 
        return 0;
 
@@ -568,8 +568,7 @@ nf_nat_port_nlattr_to_range(struct nlattr *tb[], struct nf_nat_range *range)
 
        if (tb[CTA_PROTONAT_PORT_MIN]) {
                ret = 1;
-               range->min.tcp.port =
-                       *(__be16 *)nla_data(tb[CTA_PROTONAT_PORT_MIN]);
+               range->min.tcp.port = nla_get_be16(tb[CTA_PROTONAT_PORT_MIN]);
        }
 
        if (!tb[CTA_PROTONAT_PORT_MAX]) {
@@ -577,8 +576,7 @@ nf_nat_port_nlattr_to_range(struct nlattr *tb[], struct nf_nat_range *range)
                        range->max.tcp.port = range->min.tcp.port;
        } else {
                ret = 1;
-               range->max.tcp.port =
-                       *(__be16 *)nla_data(tb[CTA_PROTONAT_PORT_MAX]);
+               range->max.tcp.port = nla_get_be16(tb[CTA_PROTONAT_PORT_MAX]);
        }
 
        return ret;
@@ -596,25 +594,25 @@ static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
 
        NF_CT_ASSERT(nat->ct->status & IPS_NAT_DONE_MASK);
 
-       write_lock_bh(&nf_nat_lock);
-       hlist_del(&nat->bysource);
+       spin_lock_bh(&nf_nat_lock);
+       hlist_del_rcu(&nat->bysource);
        nat->ct = NULL;
-       write_unlock_bh(&nf_nat_lock);
+       spin_unlock_bh(&nf_nat_lock);
 }
 
-static void nf_nat_move_storage(struct nf_conn *conntrack, void *old)
+static void nf_nat_move_storage(void *new, void *old)
 {
-       struct nf_conn_nat *new_nat = nf_ct_ext_find(conntrack, NF_CT_EXT_NAT);
-       struct nf_conn_nat *old_nat = (struct nf_conn_nat *)old;
+       struct nf_conn_nat *new_nat = new;
+       struct nf_conn_nat *old_nat = old;
        struct nf_conn *ct = old_nat->ct;
 
        if (!ct || !(ct->status & IPS_NAT_DONE_MASK))
                return;
 
-       write_lock_bh(&nf_nat_lock);
+       spin_lock_bh(&nf_nat_lock);
        hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
        new_nat->ct = ct;
-       write_unlock_bh(&nf_nat_lock);
+       spin_unlock_bh(&nf_nat_lock);
 }
 
 static struct nf_ct_ext_type nat_extend __read_mostly = {
@@ -648,17 +646,13 @@ static int __init nf_nat_init(void)
        }
 
        /* Sew in builtin protocols. */
-       write_lock_bh(&nf_nat_lock);
+       spin_lock_bh(&nf_nat_lock);
        for (i = 0; i < MAX_IP_NAT_PROTO; i++)
                rcu_assign_pointer(nf_nat_protos[i], &nf_nat_unknown_protocol);
        rcu_assign_pointer(nf_nat_protos[IPPROTO_TCP], &nf_nat_protocol_tcp);
        rcu_assign_pointer(nf_nat_protos[IPPROTO_UDP], &nf_nat_protocol_udp);
        rcu_assign_pointer(nf_nat_protos[IPPROTO_ICMP], &nf_nat_protocol_icmp);
-       write_unlock_bh(&nf_nat_lock);
-
-       for (i = 0; i < nf_nat_htable_size; i++) {
-               INIT_HLIST_HEAD(&bysource[i]);
-       }
+       spin_unlock_bh(&nf_nat_lock);
 
        /* Initialize fake conntrack so that NAT will skip it */
        nf_conntrack_untracked.status |= IPS_NAT_DONE_MASK;