]> bbs.cooldavid.org Git - net-next-2.6.git/blobdiff - net/netfilter/nf_conntrack_core.c
netfilter: nf_conntrack: add support for "conntrack zones"
[net-next-2.6.git] / net / netfilter / nf_conntrack_core.c
index 65351ed5d81515a9462137b54da757b989168d7d..0c9bbe93cc169c86ab2dbb241ae5a2654285e25f 100644 (file)
@@ -42,6 +42,7 @@
 #include <net/netfilter/nf_conntrack_extend.h>
 #include <net/netfilter/nf_conntrack_acct.h>
 #include <net/netfilter/nf_conntrack_ecache.h>
+#include <net/netfilter/nf_conntrack_zones.h>
 #include <net/netfilter/nf_nat.h>
 #include <net/netfilter/nf_nat_core.h>
 
@@ -68,7 +69,7 @@ static int nf_conntrack_hash_rnd_initted;
 static unsigned int nf_conntrack_hash_rnd;
 
 static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
-                                 unsigned int size, unsigned int rnd)
+                                 u16 zone, unsigned int size, unsigned int rnd)
 {
        unsigned int n;
        u_int32_t h;
@@ -79,16 +80,16 @@ static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
         */
        n = (sizeof(tuple->src) + sizeof(tuple->dst.u3)) / sizeof(u32);
        h = jhash2((u32 *)tuple, n,
-                  rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
-                         tuple->dst.protonum));
+                  zone ^ rnd ^ (((__force __u16)tuple->dst.u.all << 16) |
+                                tuple->dst.protonum));
 
        return ((u64)h * size) >> 32;
 }
 
-static inline u_int32_t hash_conntrack(const struct net *net,
+static inline u_int32_t hash_conntrack(const struct net *net, u16 zone,
                                       const struct nf_conntrack_tuple *tuple)
 {
-       return __hash_conntrack(tuple, net->ct.htable_size,
+       return __hash_conntrack(tuple, zone, net->ct.htable_size,
                                nf_conntrack_hash_rnd);
 }
 
@@ -292,11 +293,12 @@ static void death_by_timeout(unsigned long ul_conntrack)
  * - Caller must lock nf_conntrack_lock before calling this function
  */
 struct nf_conntrack_tuple_hash *
-__nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
+__nf_conntrack_find(struct net *net, u16 zone,
+                   const struct nf_conntrack_tuple *tuple)
 {
        struct nf_conntrack_tuple_hash *h;
        struct hlist_nulls_node *n;
-       unsigned int hash = hash_conntrack(net, tuple);
+       unsigned int hash = hash_conntrack(net, zone, tuple);
 
        /* Disable BHs the entire time since we normally need to disable them
         * at least once for the stats anyway.
@@ -304,7 +306,8 @@ __nf_conntrack_find(struct net *net, const struct nf_conntrack_tuple *tuple)
        local_bh_disable();
 begin:
        hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
-               if (nf_ct_tuple_equal(tuple, &h->tuple)) {
+               if (nf_ct_tuple_equal(tuple, &h->tuple) &&
+                   nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)) == zone) {
                        NF_CT_STAT_INC(net, found);
                        local_bh_enable();
                        return h;
@@ -326,21 +329,23 @@ EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 
 /* Find a connection corresponding to a tuple. */
 struct nf_conntrack_tuple_hash *
-nf_conntrack_find_get(struct net *net, const struct nf_conntrack_tuple *tuple)
+nf_conntrack_find_get(struct net *net, u16 zone,
+                     const struct nf_conntrack_tuple *tuple)
 {
        struct nf_conntrack_tuple_hash *h;
        struct nf_conn *ct;
 
        rcu_read_lock();
 begin:
-       h = __nf_conntrack_find(net, tuple);
+       h = __nf_conntrack_find(net, zone, tuple);
        if (h) {
                ct = nf_ct_tuplehash_to_ctrack(h);
                if (unlikely(nf_ct_is_dying(ct) ||
                             !atomic_inc_not_zero(&ct->ct_general.use)))
                        h = NULL;
                else {
-                       if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple))) {
+                       if (unlikely(!nf_ct_tuple_equal(tuple, &h->tuple) ||
+                                    nf_ct_zone(ct) != zone)) {
                                nf_ct_put(ct);
                                goto begin;
                        }
@@ -368,9 +373,11 @@ void nf_conntrack_hash_insert(struct nf_conn *ct)
 {
        struct net *net = nf_ct_net(ct);
        unsigned int hash, repl_hash;
+       u16 zone;
 
-       hash = hash_conntrack(net, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
-       repl_hash = hash_conntrack(net, &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
+       zone = nf_ct_zone(ct);
+       hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
+       repl_hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 
        __nf_conntrack_hash_insert(ct, hash, repl_hash);
 }
@@ -387,6 +394,7 @@ __nf_conntrack_confirm(struct sk_buff *skb)
        struct hlist_nulls_node *n;
        enum ip_conntrack_info ctinfo;
        struct net *net;
+       u16 zone;
 
        ct = nf_ct_get(skb, &ctinfo);
        net = nf_ct_net(ct);
@@ -398,8 +406,9 @@ __nf_conntrack_confirm(struct sk_buff *skb)
        if (CTINFO2DIR(ctinfo) != IP_CT_DIR_ORIGINAL)
                return NF_ACCEPT;
 
-       hash = hash_conntrack(net, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
-       repl_hash = hash_conntrack(net, &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
+       zone = nf_ct_zone(ct);
+       hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
+       repl_hash = hash_conntrack(net, zone, &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
 
        /* We're not in hash table, and we refuse to set up related
           connections for unconfirmed conns.  But packet copies and
@@ -418,11 +427,13 @@ __nf_conntrack_confirm(struct sk_buff *skb)
           not in the hash.  If there is, we lost race. */
        hlist_nulls_for_each_entry(h, n, &net->ct.hash[hash], hnnode)
                if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
-                                     &h->tuple))
+                                     &h->tuple) &&
+                   zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
                        goto out;
        hlist_nulls_for_each_entry(h, n, &net->ct.hash[repl_hash], hnnode)
                if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
-                                     &h->tuple))
+                                     &h->tuple) &&
+                   zone == nf_ct_zone(nf_ct_tuplehash_to_ctrack(h)))
                        goto out;
 
        /* Remove from unconfirmed list */
@@ -469,15 +480,19 @@ nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
        struct net *net = nf_ct_net(ignored_conntrack);
        struct nf_conntrack_tuple_hash *h;
        struct hlist_nulls_node *n;
-       unsigned int hash = hash_conntrack(net, tuple);
+       struct nf_conn *ct;
+       u16 zone = nf_ct_zone(ignored_conntrack);
+       unsigned int hash = hash_conntrack(net, zone, tuple);
 
        /* Disable BHs the entire time since we need to disable them at
         * least once for the stats anyway.
         */
        rcu_read_lock_bh();
        hlist_nulls_for_each_entry_rcu(h, n, &net->ct.hash[hash], hnnode) {
-               if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
-                   nf_ct_tuple_equal(tuple, &h->tuple)) {
+               ct = nf_ct_tuplehash_to_ctrack(h);
+               if (ct != ignored_conntrack &&
+                   nf_ct_tuple_equal(tuple, &h->tuple) &&
+                   nf_ct_zone(ct) == zone) {
                        NF_CT_STAT_INC(net, found);
                        rcu_read_unlock_bh();
                        return 1;
@@ -540,7 +555,7 @@ static noinline int early_drop(struct net *net, unsigned int hash)
        return dropped;
 }
 
-struct nf_conn *nf_conntrack_alloc(struct net *net,
+struct nf_conn *nf_conntrack_alloc(struct net *net, u16 zone,
                                   const struct nf_conntrack_tuple *orig,
                                   const struct nf_conntrack_tuple *repl,
                                   gfp_t gfp)
@@ -558,7 +573,7 @@ struct nf_conn *nf_conntrack_alloc(struct net *net,
 
        if (nf_conntrack_max &&
            unlikely(atomic_read(&net->ct.count) > nf_conntrack_max)) {
-               unsigned int hash = hash_conntrack(net, orig);
+               unsigned int hash = hash_conntrack(net, zone, orig);
                if (!early_drop(net, hash)) {
                        atomic_dec(&net->ct.count);
                        if (net_ratelimit())
@@ -595,13 +610,28 @@ struct nf_conn *nf_conntrack_alloc(struct net *net,
 #ifdef CONFIG_NET_NS
        ct->ct_net = net;
 #endif
-
+#ifdef CONFIG_NF_CONNTRACK_ZONES
+       if (zone) {
+               struct nf_conntrack_zone *nf_ct_zone;
+
+               nf_ct_zone = nf_ct_ext_add(ct, NF_CT_EXT_ZONE, GFP_ATOMIC);
+               if (!nf_ct_zone)
+                       goto out_free;
+               nf_ct_zone->id = zone;
+       }
+#endif
        /*
         * changes to lookup keys must be done before setting refcnt to 1
         */
        smp_wmb();
        atomic_set(&ct->ct_general.use, 1);
        return ct;
+
+#ifdef CONFIG_NF_CONNTRACK_ZONES
+out_free:
+       kmem_cache_free(net->ct.nf_conntrack_cachep, ct);
+       return ERR_PTR(-ENOMEM);
+#endif
 }
 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 
@@ -631,13 +661,14 @@ init_conntrack(struct net *net, struct nf_conn *tmpl,
        struct nf_conntrack_tuple repl_tuple;
        struct nf_conntrack_ecache *ecache;
        struct nf_conntrack_expect *exp;
+       u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 
        if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
                pr_debug("Can't invert tuple.\n");
                return NULL;
        }
 
-       ct = nf_conntrack_alloc(net, tuple, &repl_tuple, GFP_ATOMIC);
+       ct = nf_conntrack_alloc(net, zone, tuple, &repl_tuple, GFP_ATOMIC);
        if (IS_ERR(ct)) {
                pr_debug("Can't allocate conntrack.\n");
                return (struct nf_conntrack_tuple_hash *)ct;
@@ -657,7 +688,7 @@ init_conntrack(struct net *net, struct nf_conn *tmpl,
                             GFP_ATOMIC);
 
        spin_lock_bh(&nf_conntrack_lock);
-       exp = nf_ct_find_expectation(net, tuple);
+       exp = nf_ct_find_expectation(net, zone, tuple);
        if (exp) {
                pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
                         ct, exp);
@@ -713,6 +744,7 @@ resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
        struct nf_conntrack_tuple tuple;
        struct nf_conntrack_tuple_hash *h;
        struct nf_conn *ct;
+       u16 zone = tmpl ? nf_ct_zone(tmpl) : NF_CT_DEFAULT_ZONE;
 
        if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
                             dataoff, l3num, protonum, &tuple, l3proto,
@@ -722,7 +754,7 @@ resolve_normal_ct(struct net *net, struct nf_conn *tmpl,
        }
 
        /* look for tuple match */
-       h = nf_conntrack_find_get(net, &tuple);
+       h = nf_conntrack_find_get(net, zone, &tuple);
        if (!h) {
                h = init_conntrack(net, tmpl, &tuple, l3proto, l4proto,
                                   skb, dataoff);
@@ -958,6 +990,14 @@ bool __nf_ct_kill_acct(struct nf_conn *ct,
 }
 EXPORT_SYMBOL_GPL(__nf_ct_kill_acct);
 
+#ifdef CONFIG_NF_CONNTRACK_ZONES
+static struct nf_ct_ext_type nf_ct_zone_extend __read_mostly = {
+       .len    = sizeof(struct nf_conntrack_zone),
+       .align  = __alignof__(struct nf_conntrack_zone),
+       .id     = NF_CT_EXT_ZONE,
+};
+#endif
+
 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 
 #include <linux/netfilter/nfnetlink.h>
@@ -1139,6 +1179,9 @@ static void nf_conntrack_cleanup_init_net(void)
 
        nf_conntrack_helper_fini();
        nf_conntrack_proto_fini();
+#ifdef CONFIG_NF_CONNTRACK_ZONES
+       nf_ct_extend_unregister(&nf_ct_zone_extend);
+#endif
 }
 
 static void nf_conntrack_cleanup_net(struct net *net)
@@ -1214,6 +1257,7 @@ int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
        unsigned int hashsize, old_size;
        struct hlist_nulls_head *hash, *old_hash;
        struct nf_conntrack_tuple_hash *h;
+       struct nf_conn *ct;
 
        if (current->nsproxy->net_ns != &init_net)
                return -EOPNOTSUPP;
@@ -1240,8 +1284,10 @@ int nf_conntrack_set_hashsize(const char *val, struct kernel_param *kp)
                while (!hlist_nulls_empty(&init_net.ct.hash[i])) {
                        h = hlist_nulls_entry(init_net.ct.hash[i].first,
                                        struct nf_conntrack_tuple_hash, hnnode);
+                       ct = nf_ct_tuplehash_to_ctrack(h);
                        hlist_nulls_del_rcu(&h->hnnode);
-                       bucket = __hash_conntrack(&h->tuple, hashsize,
+                       bucket = __hash_conntrack(&h->tuple, nf_ct_zone(ct),
+                                                 hashsize,
                                                  nf_conntrack_hash_rnd);
                        hlist_nulls_add_head_rcu(&h->hnnode, &hash[bucket]);
                }
@@ -1299,6 +1345,11 @@ static int nf_conntrack_init_init_net(void)
        if (ret < 0)
                goto err_helper;
 
+#ifdef CONFIG_NF_CONNTRACK_ZONES
+       ret = nf_ct_extend_register(&nf_ct_zone_extend);
+       if (ret < 0)
+               goto err_extend;
+#endif
        /* Set up fake conntrack: to never be deleted, not in any hashes */
 #ifdef CONFIG_NET_NS
        nf_conntrack_untracked.ct_net = &init_net;
@@ -1309,6 +1360,10 @@ static int nf_conntrack_init_init_net(void)
 
        return 0;
 
+#ifdef CONFIG_NF_CONNTRACK_ZONES
+err_extend:
+       nf_conntrack_helper_fini();
+#endif
 err_helper:
        nf_conntrack_proto_fini();
 err_proto: