mac80211: fix another race in aggregation start
[linux-2.6.git] / net / netfilter / xt_connlimit.c
index 6a9e2a3..c6d5a83 100644 (file)
@@ -5,18 +5,19 @@
  *   Nov 2002: Martin Bene <martin.bene@icomedias.com>:
  *             only ignore TIME_WAIT or gone connections
  *   (C) CC Computer Consultants GmbH, 2007
- *   Contact: <jengelh@computergmbh.de>
  *
  * based on ...
  *
  * Kernel module to match connection tracking information.
  * GPL (C) 1999  Rusty Russell (rusty@rustcorp.com.au).
  */
+#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
 #include <linux/in.h>
 #include <linux/in6.h>
 #include <linux/ip.h>
 #include <linux/ipv6.h>
 #include <linux/jhash.h>
+#include <linux/slab.h>
 #include <linux/list.h>
 #include <linux/module.h>
 #include <linux/random.h>
 #include <net/netfilter/nf_conntrack.h>
 #include <net/netfilter/nf_conntrack_core.h>
 #include <net/netfilter/nf_conntrack_tuple.h>
+#include <net/netfilter/nf_conntrack_zones.h>
 
 /* we will save the tuples of all connections we care about */
 struct xt_connlimit_conn {
-       struct list_head list;
-       struct nf_conntrack_tuple tuple;
+       struct hlist_node               node;
+       struct nf_conntrack_tuple       tuple;
+       union nf_inet_addr              addr;
 };
 
 struct xt_connlimit_data {
-       struct list_head iphash[256];
-       spinlock_t lock;
+       struct hlist_head       iphash[256];
+       spinlock_t              lock;
 };
 
-static u_int32_t connlimit_rnd;
-static bool connlimit_rnd_inited;
+static u_int32_t connlimit_rnd __read_mostly;
 
 static inline unsigned int connlimit_iphash(__be32 addr)
 {
-       if (unlikely(!connlimit_rnd_inited)) {
-               get_random_bytes(&connlimit_rnd, sizeof(connlimit_rnd));
-               connlimit_rnd_inited = true;
-       }
        return jhash_1word((__force __u32)addr, connlimit_rnd) & 0xFF;
 }
 
@@ -59,11 +57,6 @@ connlimit_iphash6(const union nf_inet_addr *addr,
        union nf_inet_addr res;
        unsigned int i;
 
-       if (unlikely(!connlimit_rnd_inited)) {
-               get_random_bytes(&connlimit_rnd, sizeof(connlimit_rnd));
-               connlimit_rnd_inited = true;
-       }
-
        for (i = 0; i < ARRAY_SIZE(addr->ip6); ++i)
                res.ip6[i] = addr->ip6[i] & mask->ip6[i];
 
@@ -72,10 +65,9 @@ connlimit_iphash6(const union nf_inet_addr *addr,
 
 static inline bool already_closed(const struct nf_conn *conn)
 {
-       u_int16_t proto = conn->tuplehash[0].tuple.dst.protonum;
-
-       if (proto == IPPROTO_TCP)
-               return conn->proto.tcp.state == TCP_CONNTRACK_TIME_WAIT;
+       if (nf_ct_protonum(conn) == IPPROTO_TCP)
+               return conn->proto.tcp.state == TCP_CONNTRACK_TIME_WAIT ||
+                      conn->proto.tcp.state == TCP_CONNTRACK_CLOSE;
        else
                return 0;
 }
@@ -83,9 +75,9 @@ static inline bool already_closed(const struct nf_conn *conn)
 static inline unsigned int
 same_source_net(const union nf_inet_addr *addr,
                const union nf_inet_addr *mask,
-               const union nf_inet_addr *u3, unsigned int family)
+               const union nf_inet_addr *u3, u_int8_t family)
 {
-       if (family == AF_INET) {
+       if (family == NFPROTO_IPV4) {
                return (addr->ip & mask->ip) == (u3->ip & mask->ip);
        } else {
                union nf_inet_addr lh, rh;
@@ -100,31 +92,32 @@ same_source_net(const union nf_inet_addr *addr,
        }
 }
 
-static int count_them(struct xt_connlimit_data *data,
+static int count_them(struct net *net,
+                     struct xt_connlimit_data *data,
                      const struct nf_conntrack_tuple *tuple,
                      const union nf_inet_addr *addr,
                      const union nf_inet_addr *mask,
-                     const struct xt_match *match)
+                     u_int8_t family)
 {
-       struct nf_conntrack_tuple_hash *found;
+       const struct nf_conntrack_tuple_hash *found;
        struct xt_connlimit_conn *conn;
-       struct xt_connlimit_conn *tmp;
+       struct hlist_node *pos, *n;
        struct nf_conn *found_ct;
-       struct list_head *hash;
+       struct hlist_head *hash;
        bool addit = true;
        int matches = 0;
 
-
-       if (match->family == AF_INET6)
+       if (family == NFPROTO_IPV6)
                hash = &data->iphash[connlimit_iphash6(addr, mask)];
        else
                hash = &data->iphash[connlimit_iphash(addr->ip & mask->ip)];
 
-       read_lock_bh(&nf_conntrack_lock);
+       rcu_read_lock();
 
        /* check the saved connections */
-       list_for_each_entry_safe(conn, tmp, hash, list) {
-               found    = __nf_conntrack_find(&conn->tuple, NULL);
+       hlist_for_each_entry_safe(conn, pos, n, hash, node) {
+               found    = nf_conntrack_find_get(net, NF_CT_DEFAULT_ZONE,
+                                                &conn->tuple);
                found_ct = NULL;
 
                if (found != NULL)
@@ -142,7 +135,7 @@ static int count_them(struct xt_connlimit_data *data,
 
                if (found == NULL) {
                        /* this one is gone */
-                       list_del(&conn->list);
+                       hlist_del(&conn->node);
                        kfree(conn);
                        continue;
                }
@@ -152,26 +145,28 @@ static int count_them(struct xt_connlimit_data *data,
                         * we do not care about connections which are
                         * closed already -> ditch it
                         */
-                       list_del(&conn->list);
+                       nf_ct_put(found_ct);
+                       hlist_del(&conn->node);
                        kfree(conn);
                        continue;
                }
 
-               if (same_source_net(addr, mask, &conn->tuple.src.u3,
-                   match->family))
+               if (same_source_net(addr, mask, &conn->addr, family))
                        /* same source network -> be counted! */
                        ++matches;
+               nf_ct_put(found_ct);
        }
 
-       read_unlock_bh(&nf_conntrack_lock);
+       rcu_read_unlock();
 
        if (addit) {
                /* save the new connection in our list */
-               conn = kzalloc(sizeof(*conn), GFP_ATOMIC);
+               conn = kmalloc(sizeof(*conn), GFP_ATOMIC);
                if (conn == NULL)
                        return -ENOMEM;
                conn->tuple = *tuple;
-               list_add(&conn->list, hash);
+               conn->addr = *addr;
+               hlist_add_head(&conn->node, hash);
                ++matches;
        }
 
@@ -179,12 +174,10 @@ static int count_them(struct xt_connlimit_data *data,
 }
 
 static bool
-connlimit_mt(const struct sk_buff *skb, const struct net_device *in,
-             const struct net_device *out, const struct xt_match *match,
-             const void *matchinfo, int offset, unsigned int protoff,
-             bool *hotdrop)
+connlimit_mt(const struct sk_buff *skb, struct xt_action_param *par)
 {
-       const struct xt_connlimit_info *info = matchinfo;
+       struct net *net = dev_net(par->in ? par->in : par->out);
+       const struct xt_connlimit_info *info = par->matchinfo;
        union nf_inet_addr addr;
        struct nf_conntrack_tuple tuple;
        const struct nf_conntrack_tuple *tuple_ptr = &tuple;
@@ -194,79 +187,86 @@ connlimit_mt(const struct sk_buff *skb, const struct net_device *in,
 
        ct = nf_ct_get(skb, &ctinfo);
        if (ct != NULL)
-               tuple_ptr = &ct->tuplehash[0].tuple;
+               tuple_ptr = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
        else if (!nf_ct_get_tuplepr(skb, skb_network_offset(skb),
-                                   match->family, &tuple))
+                                   par->family, &tuple))
                goto hotdrop;
 
-       if (match->family == AF_INET6) {
+       if (par->family == NFPROTO_IPV6) {
                const struct ipv6hdr *iph = ipv6_hdr(skb);
-               memcpy(&addr.ip6, &iph->saddr, sizeof(iph->saddr));
+               memcpy(&addr.ip6, (info->flags & XT_CONNLIMIT_DADDR) ?
+                      &iph->daddr : &iph->saddr, sizeof(addr.ip6));
        } else {
                const struct iphdr *iph = ip_hdr(skb);
-               addr.ip = iph->saddr;
+               addr.ip = (info->flags & XT_CONNLIMIT_DADDR) ?
+                         iph->daddr : iph->saddr;
        }
 
        spin_lock_bh(&info->data->lock);
-       connections = count_them(info->data, tuple_ptr, &addr,
-                                &info->mask, match);
+       connections = count_them(net, info->data, tuple_ptr, &addr,
+                                &info->mask, par->family);
        spin_unlock_bh(&info->data->lock);
 
-       if (connections < 0) {
+       if (connections < 0)
                /* kmalloc failed, drop it entirely */
-               *hotdrop = true;
-               return false;
-       }
+               goto hotdrop;
 
-       return (connections > info->limit) ^ info->inverse;
+       return (connections > info->limit) ^
+              !!(info->flags & XT_CONNLIMIT_INVERT);
 
  hotdrop:
-       *hotdrop = true;
+       par->hotdrop = true;
        return false;
 }
 
-static bool
-connlimit_mt_check(const char *tablename, const void *ip,
-                   const struct xt_match *match, void *matchinfo,
-                   unsigned int hook_mask)
+static int connlimit_mt_check(const struct xt_mtchk_param *par)
 {
-       struct xt_connlimit_info *info = matchinfo;
+       struct xt_connlimit_info *info = par->matchinfo;
        unsigned int i;
+       int ret;
 
-       if (nf_ct_l3proto_try_module_get(match->family) < 0) {
-               printk(KERN_WARNING "cannot load conntrack support for "
-                      "address family %u\n", match->family);
-               return false;
+       if (unlikely(!connlimit_rnd)) {
+               u_int32_t rand;
+
+               do {
+                       get_random_bytes(&rand, sizeof(rand));
+               } while (!rand);
+               cmpxchg(&connlimit_rnd, 0, rand);
+       }
+       ret = nf_ct_l3proto_try_module_get(par->family);
+       if (ret < 0) {
+               pr_info("cannot load conntrack support for "
+                       "address family %u\n", par->family);
+               return ret;
        }
 
        /* init private data */
        info->data = kmalloc(sizeof(struct xt_connlimit_data), GFP_KERNEL);
        if (info->data == NULL) {
-               nf_ct_l3proto_module_put(match->family);
-               return false;
+               nf_ct_l3proto_module_put(par->family);
+               return -ENOMEM;
        }
 
        spin_lock_init(&info->data->lock);
        for (i = 0; i < ARRAY_SIZE(info->data->iphash); ++i)
-               INIT_LIST_HEAD(&info->data->iphash[i]);
+               INIT_HLIST_HEAD(&info->data->iphash[i]);
 
-       return true;
+       return 0;
 }
 
-static void
-connlimit_mt_destroy(const struct xt_match *match, void *matchinfo)
+static void connlimit_mt_destroy(const struct xt_mtdtor_param *par)
 {
-       struct xt_connlimit_info *info = matchinfo;
+       const struct xt_connlimit_info *info = par->matchinfo;
        struct xt_connlimit_conn *conn;
-       struct xt_connlimit_conn *tmp;
-       struct list_head *hash = info->data->iphash;
+       struct hlist_node *pos, *n;
+       struct hlist_head *hash = info->data->iphash;
        unsigned int i;
 
-       nf_ct_l3proto_module_put(match->family);
+       nf_ct_l3proto_module_put(par->family);
 
        for (i = 0; i < ARRAY_SIZE(info->data->iphash); ++i) {
-               list_for_each_entry_safe(conn, tmp, &hash[i], list) {
-                       list_del(&conn->list);
+               hlist_for_each_entry_safe(conn, pos, n, &hash[i], node) {
+                       hlist_del(&conn->node);
                        kfree(conn);
                }
        }
@@ -277,7 +277,8 @@ connlimit_mt_destroy(const struct xt_match *match, void *matchinfo)
 static struct xt_match connlimit_mt_reg[] __read_mostly = {
        {
                .name       = "connlimit",
-               .family     = AF_INET,
+               .revision   = 0,
+               .family     = NFPROTO_UNSPEC,
                .checkentry = connlimit_mt_check,
                .match      = connlimit_mt,
                .matchsize  = sizeof(struct xt_connlimit_info),
@@ -286,7 +287,8 @@ static struct xt_match connlimit_mt_reg[] __read_mostly = {
        },
        {
                .name       = "connlimit",
-               .family     = AF_INET6,
+               .revision   = 1,
+               .family     = NFPROTO_UNSPEC,
                .checkentry = connlimit_mt_check,
                .match      = connlimit_mt,
                .matchsize  = sizeof(struct xt_connlimit_info),
@@ -308,8 +310,8 @@ static void __exit connlimit_mt_exit(void)
 
 module_init(connlimit_mt_init);
 module_exit(connlimit_mt_exit);
-MODULE_AUTHOR("Jan Engelhardt <jengelh@computergmbh.de>");
-MODULE_DESCRIPTION("netfilter xt_connlimit match module");
+MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
+MODULE_DESCRIPTION("Xtables: Number of connections matching");
 MODULE_LICENSE("GPL");
 MODULE_ALIAS("ipt_connlimit");
 MODULE_ALIAS("ip6t_connlimit");