netfilter: nf_conntrack: add nf_ct_kill()
[safe/jmp/linux-2.6] / net / netfilter / nf_conntrack_core.c
index 4a2cce1..79b07c3 100644 (file)
@@ -94,7 +94,7 @@ static inline u_int32_t hash_conntrack(const struct nf_conntrack_tuple *tuple)
                                nf_conntrack_hash_rnd);
 }
 
-int
+bool
 nf_ct_get_tuple(const struct sk_buff *skb,
                unsigned int nhoff,
                unsigned int dataoff,
@@ -104,11 +104,11 @@ nf_ct_get_tuple(const struct sk_buff *skb,
                const struct nf_conntrack_l3proto *l3proto,
                const struct nf_conntrack_l4proto *l4proto)
 {
-       NF_CT_TUPLE_U_BLANK(tuple);
+       memset(tuple, 0, sizeof(*tuple));
 
        tuple->src.l3num = l3num;
        if (l3proto->pkt_to_tuple(skb, nhoff, tuple) == 0)
-               return 0;
+               return false;
 
        tuple->dst.protonum = protonum;
        tuple->dst.dir = IP_CT_DIR_ORIGINAL;
@@ -117,10 +117,8 @@ nf_ct_get_tuple(const struct sk_buff *skb,
 }
 EXPORT_SYMBOL_GPL(nf_ct_get_tuple);
 
-int nf_ct_get_tuplepr(const struct sk_buff *skb,
-                     unsigned int nhoff,
-                     u_int16_t l3num,
-                     struct nf_conntrack_tuple *tuple)
+bool nf_ct_get_tuplepr(const struct sk_buff *skb, unsigned int nhoff,
+                      u_int16_t l3num, struct nf_conntrack_tuple *tuple)
 {
        struct nf_conntrack_l3proto *l3proto;
        struct nf_conntrack_l4proto *l4proto;
@@ -134,7 +132,7 @@ int nf_ct_get_tuplepr(const struct sk_buff *skb,
        ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
        if (ret != NF_ACCEPT) {
                rcu_read_unlock();
-               return 0;
+               return false;
        }
 
        l4proto = __nf_ct_l4proto_find(l3num, protonum);
@@ -147,17 +145,17 @@ int nf_ct_get_tuplepr(const struct sk_buff *skb,
 }
 EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
 
-int
+bool
 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
                   const struct nf_conntrack_tuple *orig,
                   const struct nf_conntrack_l3proto *l3proto,
                   const struct nf_conntrack_l4proto *l4proto)
 {
-       NF_CT_TUPLE_U_BLANK(inverse);
+       memset(inverse, 0, sizeof(*inverse));
 
        inverse->src.l3num = orig->src.l3num;
        if (l3proto->invert_tuple(inverse, orig) == 0)
-               return 0;
+               return false;
 
        inverse->dst.dir = !orig->dst.dir;
 
@@ -194,8 +192,7 @@ destroy_conntrack(struct nf_conntrack *nfct)
         * destroy_conntrack() MUST NOT be called with a write lock
         * to nf_conntrack_lock!!! -HW */
        rcu_read_lock();
-       l4proto = __nf_ct_l4proto_find(ct->tuplehash[IP_CT_DIR_REPLY].tuple.src.l3num,
-                                      ct->tuplehash[IP_CT_DIR_REPLY].tuple.dst.protonum);
+       l4proto = __nf_ct_l4proto_find(nf_ct_l3num(ct), nf_ct_protonum(ct));
        if (l4proto && l4proto->destroy)
                l4proto->destroy(ct);
 
@@ -256,13 +253,19 @@ __nf_conntrack_find(const struct nf_conntrack_tuple *tuple)
        struct hlist_node *n;
        unsigned int hash = hash_conntrack(tuple);
 
+       /* Disable BHs the entire time since we normally need to disable them
+        * at least once for the stats anyway.
+        */
+       local_bh_disable();
        hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) {
                if (nf_ct_tuple_equal(tuple, &h->tuple)) {
                        NF_CT_STAT_INC(found);
+                       local_bh_enable();
                        return h;
                }
                NF_CT_STAT_INC(searched);
        }
+       local_bh_enable();
 
        return NULL;
 }
@@ -400,17 +403,20 @@ nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
        struct hlist_node *n;
        unsigned int hash = hash_conntrack(tuple);
 
-       rcu_read_lock();
+       /* Disable BHs the entire time since we need to disable them at
+        * least once for the stats anyway.
+        */
+       rcu_read_lock_bh();
        hlist_for_each_entry_rcu(h, n, &nf_conntrack_hash[hash], hnode) {
                if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
                    nf_ct_tuple_equal(tuple, &h->tuple)) {
                        NF_CT_STAT_INC(found);
-                       rcu_read_unlock();
+                       rcu_read_unlock_bh();
                        return 1;
                }
                NF_CT_STAT_INC(searched);
        }
-       rcu_read_unlock();
+       rcu_read_unlock_bh();
 
        return 0;
 }
@@ -420,7 +426,7 @@ EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 
 /* There's a small race here where we may free a just-assured
    connection.  Too bad: we're in trouble anyway. */
-static int early_drop(unsigned int hash)
+static noinline int early_drop(unsigned int hash)
 {
        /* Use oldest entry, which is roughly LRU */
        struct nf_conntrack_tuple_hash *h;
@@ -462,7 +468,7 @@ static int early_drop(unsigned int hash)
 struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
                                   const struct nf_conntrack_tuple *repl)
 {
-       struct nf_conn *conntrack = NULL;
+       struct nf_conn *ct = NULL;
 
        if (unlikely(!nf_conntrack_hash_rnd_initted)) {
                get_random_bytes(&nf_conntrack_hash_rnd, 4);
@@ -472,8 +478,8 @@ struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
        /* We don't want any race condition at early drop stage */
        atomic_inc(&nf_conntrack_count);
 
-       if (nf_conntrack_max
-           && atomic_read(&nf_conntrack_count) > nf_conntrack_max) {
+       if (nf_conntrack_max &&
+           unlikely(atomic_read(&nf_conntrack_count) > nf_conntrack_max)) {
                unsigned int hash = hash_conntrack(orig);
                if (!early_drop(hash)) {
                        atomic_dec(&nf_conntrack_count);
@@ -485,22 +491,21 @@ struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
                }
        }
 
-       conntrack = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC);
-       if (conntrack == NULL) {
+       ct = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC);
+       if (ct == NULL) {
                pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
                atomic_dec(&nf_conntrack_count);
                return ERR_PTR(-ENOMEM);
        }
 
-       atomic_set(&conntrack->ct_general.use, 1);
-       conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
-       conntrack->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
+       atomic_set(&ct->ct_general.use, 1);
+       ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
+       ct->tuplehash[IP_CT_DIR_REPLY].tuple = *repl;
        /* Don't set timer yet: wait for confirmation */
-       setup_timer(&conntrack->timeout, death_by_timeout,
-                   (unsigned long)conntrack);
-       INIT_RCU_HEAD(&conntrack->rcu);
+       setup_timer(&ct->timeout, death_by_timeout, (unsigned long)ct);
+       INIT_RCU_HEAD(&ct->rcu);
 
-       return conntrack;
+       return ct;
 }
 EXPORT_SYMBOL_GPL(nf_conntrack_alloc);
 
@@ -513,9 +518,9 @@ static void nf_conntrack_free_rcu(struct rcu_head *head)
        atomic_dec(&nf_conntrack_count);
 }
 
-void nf_conntrack_free(struct nf_conn *conntrack)
+void nf_conntrack_free(struct nf_conn *ct)
 {
-       call_rcu(&conntrack->rcu, nf_conntrack_free_rcu);
+       call_rcu(&ct->rcu, nf_conntrack_free_rcu);
 }
 EXPORT_SYMBOL_GPL(nf_conntrack_free);
 
@@ -528,7 +533,7 @@ init_conntrack(const struct nf_conntrack_tuple *tuple,
               struct sk_buff *skb,
               unsigned int dataoff)
 {
-       struct nf_conn *conntrack;
+       struct nf_conn *ct;
        struct nf_conn_help *help;
        struct nf_conntrack_tuple repl_tuple;
        struct nf_conntrack_expect *exp;
@@ -538,14 +543,14 @@ init_conntrack(const struct nf_conntrack_tuple *tuple,
                return NULL;
        }
 
-       conntrack = nf_conntrack_alloc(tuple, &repl_tuple);
-       if (conntrack == NULL || IS_ERR(conntrack)) {
+       ct = nf_conntrack_alloc(tuple, &repl_tuple);
+       if (ct == NULL || IS_ERR(ct)) {
                pr_debug("Can't allocate conntrack.\n");
-               return (struct nf_conntrack_tuple_hash *)conntrack;
+               return (struct nf_conntrack_tuple_hash *)ct;
        }
 
-       if (!l4proto->new(conntrack, skb, dataoff)) {
-               nf_conntrack_free(conntrack);
+       if (!l4proto->new(ct, skb, dataoff)) {
+               nf_conntrack_free(ct);
                pr_debug("init conntrack: can't track with proto module\n");
                return NULL;
        }
@@ -554,30 +559,30 @@ init_conntrack(const struct nf_conntrack_tuple *tuple,
        exp = nf_ct_find_expectation(tuple);
        if (exp) {
                pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
-                        conntrack, exp);
+                        ct, exp);
                /* Welcome, Mr. Bond.  We've been expecting you... */
-               __set_bit(IPS_EXPECTED_BIT, &conntrack->status);
-               conntrack->master = exp->master;
+               __set_bit(IPS_EXPECTED_BIT, &ct->status);
+               ct->master = exp->master;
                if (exp->helper) {
-                       help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC);
+                       help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
                        if (help)
                                rcu_assign_pointer(help->helper, exp->helper);
                }
 
 #ifdef CONFIG_NF_CONNTRACK_MARK
-               conntrack->mark = exp->master->mark;
+               ct->mark = exp->master->mark;
 #endif
 #ifdef CONFIG_NF_CONNTRACK_SECMARK
-               conntrack->secmark = exp->master->secmark;
+               ct->secmark = exp->master->secmark;
 #endif
-               nf_conntrack_get(&conntrack->master->ct_general);
+               nf_conntrack_get(&ct->master->ct_general);
                NF_CT_STAT_INC(expect_new);
        } else {
                struct nf_conntrack_helper *helper;
 
                helper = __nf_ct_helper_find(&repl_tuple);
                if (helper) {
-                       help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC);
+                       help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
                        if (help)
                                rcu_assign_pointer(help->helper, helper);
                }
@@ -585,18 +590,17 @@ init_conntrack(const struct nf_conntrack_tuple *tuple,
        }
 
        /* Overload tuple linked list to put us in unconfirmed list. */
-       hlist_add_head(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].hnode,
-                      &unconfirmed);
+       hlist_add_head(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode, &unconfirmed);
 
        spin_unlock_bh(&nf_conntrack_lock);
 
        if (exp) {
                if (exp->expectfn)
-                       exp->expectfn(conntrack, exp);
+                       exp->expectfn(ct, exp);
                nf_ct_expect_put(exp);
        }
 
-       return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL];
+       return &ct->tuplehash[IP_CT_DIR_ORIGINAL];
 }
 
 /* On success, returns conntrack ptr, sets skb->nfct and ctinfo */
@@ -732,10 +736,10 @@ nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff *skb)
 }
 EXPORT_SYMBOL_GPL(nf_conntrack_in);
 
-int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
-                        const struct nf_conntrack_tuple *orig)
+bool nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
+                         const struct nf_conntrack_tuple *orig)
 {
-       int ret;
+       bool ret;
 
        rcu_read_lock();
        ret = nf_ct_invert_tuple(inverse, orig,
@@ -759,10 +763,10 @@ void nf_conntrack_alter_reply(struct nf_conn *ct,
        NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 
        pr_debug("Altering reply tuple of %p to ", ct);
-       NF_CT_DUMP_TUPLE(newreply);
+       nf_ct_dump_tuple(newreply);
 
        ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
-       if (ct->master || (help && help->expecting != 0))
+       if (ct->master || (help && !hlist_empty(&help->expectations)))
                return;
 
        rcu_read_lock();
@@ -844,6 +848,13 @@ acct:
 }
 EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
 
+void nf_ct_kill(struct nf_conn *ct)
+{
+       if (del_timer(&ct->timeout))
+               ct->timeout.function((unsigned long)ct);
+}
+EXPORT_SYMBOL_GPL(nf_ct_kill);
+
 #if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
 
 #include <linux/netfilter/nfnetlink.h>
@@ -904,14 +915,6 @@ static void nf_conntrack_attach(struct sk_buff *nskb, struct sk_buff *skb)
        nf_conntrack_get(nskb->nfct);
 }
 
-static inline int
-do_iter(const struct nf_conntrack_tuple_hash *i,
-       int (*iter)(struct nf_conn *i, void *data),
-       void *data)
-{
-       return iter(nf_ct_tuplehash_to_ctrack(i), data);
-}
-
 /* Bring out ya dead! */
 static struct nf_conn *
 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),