]> err.no Git - linux-2.6/blobdiff - net/netfilter/nf_conntrack_core.c
[NETFILTER]: nf_conntrack: split out protocol handling
[linux-2.6] / net / netfilter / nf_conntrack_core.c
index 8f2261965a68bb4e1bc039b8df37c890314f21fc..09c0e63110443ccc6358c2255dff3b9006830708 100644 (file)
@@ -55,9 +55,9 @@
 #include <net/netfilter/nf_conntrack.h>
 #include <net/netfilter/nf_conntrack_l3proto.h>
 #include <net/netfilter/nf_conntrack_protocol.h>
+#include <net/netfilter/nf_conntrack_expect.h>
 #include <net/netfilter/nf_conntrack_helper.h>
 #include <net/netfilter/nf_conntrack_core.h>
-#include <linux/netfilter_ipv4/listhelp.h>
 
 #define NF_CONNTRACK_VERSION   "0.5.0"
 
@@ -73,87 +73,15 @@ DEFINE_RWLOCK(nf_conntrack_lock);
 atomic_t nf_conntrack_count = ATOMIC_INIT(0);
 
 void (*nf_conntrack_destroyed)(struct nf_conn *conntrack) = NULL;
-LIST_HEAD(nf_conntrack_expect_list);
-struct nf_conntrack_protocol **nf_ct_protos[PF_MAX];
-struct nf_conntrack_l3proto *nf_ct_l3protos[PF_MAX];
-static LIST_HEAD(helpers);
-unsigned int nf_conntrack_htable_size = 0;
-int nf_conntrack_max;
-struct list_head *nf_conntrack_hash;
-static kmem_cache_t *nf_conntrack_expect_cachep;
+unsigned int nf_conntrack_htable_size __read_mostly = 0;
+int nf_conntrack_max __read_mostly;
+struct list_head *nf_conntrack_hash __read_mostly;
 struct nf_conn nf_conntrack_untracked;
-unsigned int nf_ct_log_invalid;
-static LIST_HEAD(unconfirmed);
-static int nf_conntrack_vmalloc;
+unsigned int nf_ct_log_invalid __read_mostly;
+LIST_HEAD(unconfirmed);
+static int nf_conntrack_vmalloc __read_mostly;
 
 static unsigned int nf_conntrack_next_id;
-static unsigned int nf_conntrack_expect_next_id;
-#ifdef CONFIG_NF_CONNTRACK_EVENTS
-ATOMIC_NOTIFIER_HEAD(nf_conntrack_chain);
-ATOMIC_NOTIFIER_HEAD(nf_conntrack_expect_chain);
-
-DEFINE_PER_CPU(struct nf_conntrack_ecache, nf_conntrack_ecache);
-
-/* deliver cached events and clear cache entry - must be called with locally
- * disabled softirqs */
-static inline void
-__nf_ct_deliver_cached_events(struct nf_conntrack_ecache *ecache)
-{
-       DEBUGP("ecache: delivering events for %p\n", ecache->ct);
-       if (nf_ct_is_confirmed(ecache->ct) && !nf_ct_is_dying(ecache->ct)
-           && ecache->events)
-               atomic_notifier_call_chain(&nf_conntrack_chain, ecache->events,
-                                   ecache->ct);
-
-       ecache->events = 0;
-       nf_ct_put(ecache->ct);
-       ecache->ct = NULL;
-}
-
-/* Deliver all cached events for a particular conntrack. This is called
- * by code prior to async packet handling for freeing the skb */
-void nf_ct_deliver_cached_events(const struct nf_conn *ct)
-{
-       struct nf_conntrack_ecache *ecache;
-
-       local_bh_disable();
-       ecache = &__get_cpu_var(nf_conntrack_ecache);
-       if (ecache->ct == ct)
-               __nf_ct_deliver_cached_events(ecache);
-       local_bh_enable();
-}
-
-/* Deliver cached events for old pending events, if current conntrack != old */
-void __nf_ct_event_cache_init(struct nf_conn *ct)
-{
-       struct nf_conntrack_ecache *ecache;
-       
-       /* take care of delivering potentially old events */
-       ecache = &__get_cpu_var(nf_conntrack_ecache);
-       BUG_ON(ecache->ct == ct);
-       if (ecache->ct)
-               __nf_ct_deliver_cached_events(ecache);
-       /* initialize for this conntrack/packet */
-       ecache->ct = ct;
-       nf_conntrack_get(&ct->ct_general);
-}
-
-/* flush the event cache - touches other CPU's data and must not be called
- * while packets are still passing through the code */
-static void nf_ct_event_cache_flush(void)
-{
-       struct nf_conntrack_ecache *ecache;
-       int cpu;
-
-       for_each_possible_cpu(cpu) {
-               ecache = &per_cpu(nf_conntrack_ecache, cpu);
-               if (ecache->ct)
-                       nf_ct_put(ecache->ct);
-       }
-}
-#else
-static inline void nf_ct_event_cache_flush(void) {}
-#endif /* CONFIG_NF_CONNTRACK_EVENTS */
 
 DEFINE_PER_CPU(struct ip_conntrack_stat, nf_conntrack_stat);
 EXPORT_PER_CPU_SYMBOL(nf_conntrack_stat);
@@ -185,85 +113,6 @@ DEFINE_RWLOCK(nf_ct_cache_lock);
 /* This avoids calling kmem_cache_create() with same name simultaneously */
 static DEFINE_MUTEX(nf_ct_cache_mutex);
 
-extern struct nf_conntrack_protocol nf_conntrack_generic_protocol;
-struct nf_conntrack_protocol *
-__nf_ct_proto_find(u_int16_t l3proto, u_int8_t protocol)
-{
-       if (unlikely(l3proto >= AF_MAX || nf_ct_protos[l3proto] == NULL))
-               return &nf_conntrack_generic_protocol;
-
-       return nf_ct_protos[l3proto][protocol];
-}
-
-/* this is guaranteed to always return a valid protocol helper, since
- * it falls back to generic_protocol */
-struct nf_conntrack_protocol *
-nf_ct_proto_find_get(u_int16_t l3proto, u_int8_t protocol)
-{
-       struct nf_conntrack_protocol *p;
-
-       preempt_disable();
-       p = __nf_ct_proto_find(l3proto, protocol);
-       if (!try_module_get(p->me))
-               p = &nf_conntrack_generic_protocol;
-       preempt_enable();
-       
-       return p;
-}
-
-void nf_ct_proto_put(struct nf_conntrack_protocol *p)
-{
-       module_put(p->me);
-}
-
-struct nf_conntrack_l3proto *
-nf_ct_l3proto_find_get(u_int16_t l3proto)
-{
-       struct nf_conntrack_l3proto *p;
-
-       preempt_disable();
-       p = __nf_ct_l3proto_find(l3proto);
-       if (!try_module_get(p->me))
-               p = &nf_conntrack_generic_l3proto;
-       preempt_enable();
-
-       return p;
-}
-
-void nf_ct_l3proto_put(struct nf_conntrack_l3proto *p)
-{
-       module_put(p->me);
-}
-
-int
-nf_ct_l3proto_try_module_get(unsigned short l3proto)
-{
-       int ret;
-       struct nf_conntrack_l3proto *p;
-
-retry: p = nf_ct_l3proto_find_get(l3proto);
-       if (p == &nf_conntrack_generic_l3proto) {
-               ret = request_module("nf_conntrack-%d", l3proto);
-               if (!ret)
-                       goto retry;
-
-               return -EPROTOTYPE;
-       }
-
-       return 0;
-}
-
-void nf_ct_l3proto_module_put(unsigned short l3proto)
-{
-       struct nf_conntrack_l3proto *p;
-
-       preempt_disable();
-       p = __nf_ct_l3proto_find(l3proto);
-       preempt_enable();
-
-       module_put(p->me);
-}
-
 static int nf_conntrack_hash_rnd_initted;
 static unsigned int nf_conntrack_hash_rnd;
 
@@ -439,115 +288,13 @@ nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
        return protocol->invert_tuple(inverse, orig);
 }
 
-/* nf_conntrack_expect helper functions */
-void nf_ct_unlink_expect(struct nf_conntrack_expect *exp)
-{
-       struct nf_conn_help *master_help = nfct_help(exp->master);
-
-       NF_CT_ASSERT(master_help);
-       ASSERT_WRITE_LOCK(&nf_conntrack_lock);
-       NF_CT_ASSERT(!timer_pending(&exp->timeout));
-
-       list_del(&exp->list);
-       NF_CT_STAT_INC(expect_delete);
-       master_help->expecting--;
-       nf_conntrack_expect_put(exp);
-}
-
-static void expectation_timed_out(unsigned long ul_expect)
-{
-       struct nf_conntrack_expect *exp = (void *)ul_expect;
-
-       write_lock_bh(&nf_conntrack_lock);
-       nf_ct_unlink_expect(exp);
-       write_unlock_bh(&nf_conntrack_lock);
-       nf_conntrack_expect_put(exp);
-}
-
-struct nf_conntrack_expect *
-__nf_conntrack_expect_find(const struct nf_conntrack_tuple *tuple)
-{
-       struct nf_conntrack_expect *i;
-       
-       list_for_each_entry(i, &nf_conntrack_expect_list, list) {
-               if (nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask)) {
-                       atomic_inc(&i->use);
-                       return i;
-               }
-       }
-       return NULL;
-}
-
-/* Just find a expectation corresponding to a tuple. */
-struct nf_conntrack_expect *
-nf_conntrack_expect_find(const struct nf_conntrack_tuple *tuple)
-{
-       struct nf_conntrack_expect *i;
-       
-       read_lock_bh(&nf_conntrack_lock);
-       i = __nf_conntrack_expect_find(tuple);
-       read_unlock_bh(&nf_conntrack_lock);
-
-       return i;
-}
-
-/* If an expectation for this connection is found, it gets delete from
- * global list then returned. */
-static struct nf_conntrack_expect *
-find_expectation(const struct nf_conntrack_tuple *tuple)
-{
-       struct nf_conntrack_expect *i;
-
-       list_for_each_entry(i, &nf_conntrack_expect_list, list) {
-       /* If master is not in hash table yet (ie. packet hasn't left
-          this machine yet), how can other end know about expected?
-          Hence these are not the droids you are looking for (if
-          master ct never got confirmed, we'd hold a reference to it
-          and weird things would happen to future packets). */
-               if (nf_ct_tuple_mask_cmp(tuple, &i->tuple, &i->mask)
-                   && nf_ct_is_confirmed(i->master)) {
-                       if (i->flags & NF_CT_EXPECT_PERMANENT) {
-                               atomic_inc(&i->use);
-                               return i;
-                       } else if (del_timer(&i->timeout)) {
-                               nf_ct_unlink_expect(i);
-                               return i;
-                       }
-               }
-       }
-       return NULL;
-}
-
-/* delete all expectations for this conntrack */
-void nf_ct_remove_expectations(struct nf_conn *ct)
-{
-       struct nf_conntrack_expect *i, *tmp;
-       struct nf_conn_help *help = nfct_help(ct);
-
-       /* Optimization: most connection never expect any others. */
-       if (!help || help->expecting == 0)
-               return;
-
-       list_for_each_entry_safe(i, tmp, &nf_conntrack_expect_list, list) {
-               if (i->master == ct && del_timer(&i->timeout)) {
-                       nf_ct_unlink_expect(i);
-                       nf_conntrack_expect_put(i);
-               }
-       }
-}
-
 static void
 clean_from_lists(struct nf_conn *ct)
 {
-       unsigned int ho, hr;
-       
        DEBUGP("clean_from_lists(%p)\n", ct);
        ASSERT_WRITE_LOCK(&nf_conntrack_lock);
-
-       ho = hash_conntrack(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
-       hr = hash_conntrack(&ct->tuplehash[IP_CT_DIR_REPLY].tuple);
-       LIST_DELETE(&nf_conntrack_hash[ho], &ct->tuplehash[IP_CT_DIR_ORIGINAL]);
-       LIST_DELETE(&nf_conntrack_hash[hr], &ct->tuplehash[IP_CT_DIR_REPLY]);
+       list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
+       list_del(&ct->tuplehash[IP_CT_DIR_REPLY].list);
 
        /* Destroy all pending expectations */
        nf_ct_remove_expectations(ct);
@@ -617,16 +364,6 @@ static void death_by_timeout(unsigned long ul_conntrack)
        nf_ct_put(ct);
 }
 
-static inline int
-conntrack_tuple_cmp(const struct nf_conntrack_tuple_hash *i,
-                   const struct nf_conntrack_tuple *tuple,
-                   const struct nf_conn *ignored_conntrack)
-{
-       ASSERT_READ_LOCK(&nf_conntrack_lock);
-       return nf_ct_tuplehash_to_ctrack(i) != ignored_conntrack
-               && nf_ct_tuple_equal(tuple, &i->tuple);
-}
-
 struct nf_conntrack_tuple_hash *
 __nf_conntrack_find(const struct nf_conntrack_tuple *tuple,
                    const struct nf_conn *ignored_conntrack)
@@ -636,7 +373,8 @@ __nf_conntrack_find(const struct nf_conntrack_tuple *tuple,
 
        ASSERT_READ_LOCK(&nf_conntrack_lock);
        list_for_each_entry(h, &nf_conntrack_hash[hash], list) {
-               if (conntrack_tuple_cmp(h, tuple, ignored_conntrack)) {
+               if (nf_ct_tuplehash_to_ctrack(h) != ignored_conntrack &&
+                   nf_ct_tuple_equal(tuple, &h->tuple)) {
                        NF_CT_STAT_INC(found);
                        return h;
                }
@@ -667,10 +405,10 @@ static void __nf_conntrack_hash_insert(struct nf_conn *ct,
                                       unsigned int repl_hash) 
 {
        ct->id = ++nf_conntrack_next_id;
-       list_prepend(&nf_conntrack_hash[hash],
-                    &ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
-       list_prepend(&nf_conntrack_hash[repl_hash],
-                    &ct->tuplehash[IP_CT_DIR_REPLY].list);
+       list_add(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list,
+                &nf_conntrack_hash[hash]);
+       list_add(&ct->tuplehash[IP_CT_DIR_REPLY].list,
+                &nf_conntrack_hash[repl_hash]);
 }
 
 void nf_conntrack_hash_insert(struct nf_conn *ct)
@@ -690,7 +428,9 @@ int
 __nf_conntrack_confirm(struct sk_buff **pskb)
 {
        unsigned int hash, repl_hash;
+       struct nf_conntrack_tuple_hash *h;
        struct nf_conn *ct;
+       struct nf_conn_help *help;
        enum ip_conntrack_info ctinfo;
 
        ct = nf_ct_get(*pskb, &ctinfo);
@@ -720,41 +460,41 @@ __nf_conntrack_confirm(struct sk_buff **pskb)
        /* See if there's one in the list already, including reverse:
           NAT could have grabbed it without realizing, since we're
           not in the hash.  If there is, we lost race. */
-       if (!LIST_FIND(&nf_conntrack_hash[hash],
-                      conntrack_tuple_cmp,
-                      struct nf_conntrack_tuple_hash *,
-                      &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple, NULL)
-           && !LIST_FIND(&nf_conntrack_hash[repl_hash],
-                         conntrack_tuple_cmp,
-                         struct nf_conntrack_tuple_hash *,
-                         &ct->tuplehash[IP_CT_DIR_REPLY].tuple, NULL)) {
-               struct nf_conn_help *help;
-               /* Remove from unconfirmed list */
-               list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
+       list_for_each_entry(h, &nf_conntrack_hash[hash], list)
+               if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple,
+                                     &h->tuple))
+                       goto out;
+       list_for_each_entry(h, &nf_conntrack_hash[repl_hash], list)
+               if (nf_ct_tuple_equal(&ct->tuplehash[IP_CT_DIR_REPLY].tuple,
+                                     &h->tuple))
+                       goto out;
 
-               __nf_conntrack_hash_insert(ct, hash, repl_hash);
-               /* Timer relative to confirmation time, not original
-                  setting time, otherwise we'd get timer wrap in
-                  weird delay cases. */
-               ct->timeout.expires += jiffies;
-               add_timer(&ct->timeout);
-               atomic_inc(&ct->ct_general.use);
-               set_bit(IPS_CONFIRMED_BIT, &ct->status);
-               NF_CT_STAT_INC(insert);
-               write_unlock_bh(&nf_conntrack_lock);
-               help = nfct_help(ct);
-               if (help && help->helper)
-                       nf_conntrack_event_cache(IPCT_HELPER, *pskb);
+       /* Remove from unconfirmed list */
+       list_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].list);
+
+       __nf_conntrack_hash_insert(ct, hash, repl_hash);
+       /* Timer relative to confirmation time, not original
+          setting time, otherwise we'd get timer wrap in
+          weird delay cases. */
+       ct->timeout.expires += jiffies;
+       add_timer(&ct->timeout);
+       atomic_inc(&ct->ct_general.use);
+       set_bit(IPS_CONFIRMED_BIT, &ct->status);
+       NF_CT_STAT_INC(insert);
+       write_unlock_bh(&nf_conntrack_lock);
+       help = nfct_help(ct);
+       if (help && help->helper)
+               nf_conntrack_event_cache(IPCT_HELPER, *pskb);
 #ifdef CONFIG_NF_NAT_NEEDED
-               if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
-                   test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
-                       nf_conntrack_event_cache(IPCT_NATINFO, *pskb);
+       if (test_bit(IPS_SRC_NAT_DONE_BIT, &ct->status) ||
+           test_bit(IPS_DST_NAT_DONE_BIT, &ct->status))
+               nf_conntrack_event_cache(IPCT_NATINFO, *pskb);
 #endif
-               nf_conntrack_event_cache(master_ct(ct) ?
-                                        IPCT_RELATED : IPCT_NEW, *pskb);
-               return NF_ACCEPT;
-       }
+       nf_conntrack_event_cache(master_ct(ct) ?
+                                IPCT_RELATED : IPCT_NEW, *pskb);
+       return NF_ACCEPT;
 
+out:
        NF_CT_STAT_INC(insert_failed);
        write_unlock_bh(&nf_conntrack_lock);
        return NF_DROP;
@@ -777,24 +517,21 @@ nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 
 /* There's a small race here where we may free a just-assured
    connection.  Too bad: we're in trouble anyway. */
-static inline int unreplied(const struct nf_conntrack_tuple_hash *i)
-{
-       return !(test_bit(IPS_ASSURED_BIT,
-                         &nf_ct_tuplehash_to_ctrack(i)->status));
-}
-
 static int early_drop(struct list_head *chain)
 {
        /* Traverse backwards: gives us oldest, which is roughly LRU */
        struct nf_conntrack_tuple_hash *h;
-       struct nf_conn *ct = NULL;
+       struct nf_conn *ct = NULL, *tmp;
        int dropped = 0;
 
        read_lock_bh(&nf_conntrack_lock);
-       h = LIST_FIND_B(chain, unreplied, struct nf_conntrack_tuple_hash *);
-       if (h) {
-               ct = nf_ct_tuplehash_to_ctrack(h);
-               atomic_inc(&ct->ct_general.use);
+       list_for_each_entry_reverse(h, chain, list) {
+               tmp = nf_ct_tuplehash_to_ctrack(h);
+               if (!test_bit(IPS_ASSURED_BIT, &tmp->status)) {
+                       ct = tmp;
+                       atomic_inc(&ct->ct_general.use);
+                       break;
+               }
        }
        read_unlock_bh(&nf_conntrack_lock);
 
@@ -810,48 +547,6 @@ static int early_drop(struct list_head *chain)
        return dropped;
 }
 
-static inline int helper_cmp(const struct nf_conntrack_helper *i,
-                            const struct nf_conntrack_tuple *rtuple)
-{
-       return nf_ct_tuple_mask_cmp(rtuple, &i->tuple, &i->mask);
-}
-
-static struct nf_conntrack_helper *
-__nf_ct_helper_find(const struct nf_conntrack_tuple *tuple)
-{
-       return LIST_FIND(&helpers, helper_cmp,
-                        struct nf_conntrack_helper *,
-                        tuple);
-}
-
-struct nf_conntrack_helper *
-nf_ct_helper_find_get( const struct nf_conntrack_tuple *tuple)
-{
-       struct nf_conntrack_helper *helper;
-
-       /* need nf_conntrack_lock to assure that helper exists until
-        * try_module_get() is called */
-       read_lock_bh(&nf_conntrack_lock);
-
-       helper = __nf_ct_helper_find(tuple);
-       if (helper) {
-               /* need to increase module usage count to assure helper will
-                * not go away while the caller is e.g. busy putting a
-                * conntrack in the hash that uses the helper */
-               if (!try_module_get(helper->me))
-                       helper = NULL;
-       }
-
-       read_unlock_bh(&nf_conntrack_lock);
-
-       return helper;
-}
-
-void nf_ct_helper_put(struct nf_conntrack_helper *helper)
-{
-       module_put(helper->me);
-}
-
 static struct nf_conn *
 __nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
                     const struct nf_conntrack_tuple *repl,
@@ -866,11 +561,15 @@ __nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
                nf_conntrack_hash_rnd_initted = 1;
        }
 
+       /* 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) {
+           && atomic_read(&nf_conntrack_count) > nf_conntrack_max) {
                unsigned int hash = hash_conntrack(orig);
                /* Try dropping from this hash chain. */
                if (!early_drop(&nf_conntrack_hash[hash])) {
+                       atomic_dec(&nf_conntrack_count);
                        if (net_ratelimit())
                                printk(KERN_WARNING
                                       "nf_conntrack: table full, dropping"
@@ -907,12 +606,6 @@ __nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
 
        memset(conntrack, 0, nf_ct_cache[features].size);
        conntrack->features = features;
-       if (helper) {
-               struct nf_conn_help *help = nfct_help(conntrack);
-               NF_CT_ASSERT(help);
-               help->helper = helper;
-       }
-
        atomic_set(&conntrack->ct_general.use, 1);
        conntrack->ct_general.destroy = destroy_conntrack;
        conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
@@ -921,10 +614,12 @@ __nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
        init_timer(&conntrack->timeout);
        conntrack->timeout.data = (unsigned long)conntrack;
        conntrack->timeout.function = death_by_timeout;
+       read_unlock_bh(&nf_ct_cache_lock);
 
-       atomic_inc(&nf_conntrack_count);
+       return conntrack;
 out:
        read_unlock_bh(&nf_ct_cache_lock);
+       atomic_dec(&nf_conntrack_count);
        return conntrack;
 }
 
@@ -994,8 +689,13 @@ init_conntrack(const struct nf_conntrack_tuple *tuple,
 #endif
                nf_conntrack_get(&conntrack->master->ct_general);
                NF_CT_STAT_INC(expect_new);
-       } else
+       } else {
+               struct nf_conn_help *help = nfct_help(conntrack);
+
+               if (help)
+                       help->helper = __nf_ct_helper_find(&repl_tuple);
                NF_CT_STAT_INC(new);
+       }
 
        /* Overload tuple linked list to put us in unconfirmed list. */
        list_add(&conntrack->tuplehash[IP_CT_DIR_ORIGINAL].list, &unconfirmed);
@@ -1146,244 +846,6 @@ int nf_ct_invert_tuplepr(struct nf_conntrack_tuple *inverse,
                                                     orig->dst.protonum));
 }
 
-/* Would two expected things clash? */
-static inline int expect_clash(const struct nf_conntrack_expect *a,
-                              const struct nf_conntrack_expect *b)
-{
-       /* Part covered by intersection of masks must be unequal,
-          otherwise they clash */
-       struct nf_conntrack_tuple intersect_mask;
-       int count;
-
-       intersect_mask.src.l3num = a->mask.src.l3num & b->mask.src.l3num;
-       intersect_mask.src.u.all = a->mask.src.u.all & b->mask.src.u.all;
-       intersect_mask.dst.u.all = a->mask.dst.u.all & b->mask.dst.u.all;
-       intersect_mask.dst.protonum = a->mask.dst.protonum
-                                       & b->mask.dst.protonum;
-
-       for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){
-               intersect_mask.src.u3.all[count] =
-                       a->mask.src.u3.all[count] & b->mask.src.u3.all[count];
-       }
-
-       for (count = 0; count < NF_CT_TUPLE_L3SIZE; count++){
-               intersect_mask.dst.u3.all[count] =
-                       a->mask.dst.u3.all[count] & b->mask.dst.u3.all[count];
-       }
-
-       return nf_ct_tuple_mask_cmp(&a->tuple, &b->tuple, &intersect_mask);
-}
-
-static inline int expect_matches(const struct nf_conntrack_expect *a,
-                                const struct nf_conntrack_expect *b)
-{
-       return a->master == b->master
-               && nf_ct_tuple_equal(&a->tuple, &b->tuple)
-               && nf_ct_tuple_equal(&a->mask, &b->mask);
-}
-
-/* Generally a bad idea to call this: could have matched already. */
-void nf_conntrack_unexpect_related(struct nf_conntrack_expect *exp)
-{
-       struct nf_conntrack_expect *i;
-
-       write_lock_bh(&nf_conntrack_lock);
-       /* choose the the oldest expectation to evict */
-       list_for_each_entry_reverse(i, &nf_conntrack_expect_list, list) {
-               if (expect_matches(i, exp) && del_timer(&i->timeout)) {
-                       nf_ct_unlink_expect(i);
-                       write_unlock_bh(&nf_conntrack_lock);
-                       nf_conntrack_expect_put(i);
-                       return;
-               }
-       }
-       write_unlock_bh(&nf_conntrack_lock);
-}
-
-/* We don't increase the master conntrack refcount for non-fulfilled
- * conntracks. During the conntrack destruction, the expectations are
- * always killed before the conntrack itself */
-struct nf_conntrack_expect *nf_conntrack_expect_alloc(struct nf_conn *me)
-{
-       struct nf_conntrack_expect *new;
-
-       new = kmem_cache_alloc(nf_conntrack_expect_cachep, GFP_ATOMIC);
-       if (!new) {
-               DEBUGP("expect_related: OOM allocating expect\n");
-               return NULL;
-       }
-       new->master = me;
-       atomic_set(&new->use, 1);
-       return new;
-}
-
-void nf_conntrack_expect_put(struct nf_conntrack_expect *exp)
-{
-       if (atomic_dec_and_test(&exp->use))
-               kmem_cache_free(nf_conntrack_expect_cachep, exp);
-}
-
-static void nf_conntrack_expect_insert(struct nf_conntrack_expect *exp)
-{
-       struct nf_conn_help *master_help = nfct_help(exp->master);
-
-       atomic_inc(&exp->use);
-       master_help->expecting++;
-       list_add(&exp->list, &nf_conntrack_expect_list);
-
-       init_timer(&exp->timeout);
-       exp->timeout.data = (unsigned long)exp;
-       exp->timeout.function = expectation_timed_out;
-       exp->timeout.expires = jiffies + master_help->helper->timeout * HZ;
-       add_timer(&exp->timeout);
-
-       exp->id = ++nf_conntrack_expect_next_id;
-       atomic_inc(&exp->use);
-       NF_CT_STAT_INC(expect_create);
-}
-
-/* Race with expectations being used means we could have none to find; OK. */
-static void evict_oldest_expect(struct nf_conn *master)
-{
-       struct nf_conntrack_expect *i;
-
-       list_for_each_entry_reverse(i, &nf_conntrack_expect_list, list) {
-               if (i->master == master) {
-                       if (del_timer(&i->timeout)) {
-                               nf_ct_unlink_expect(i);
-                               nf_conntrack_expect_put(i);
-                       }
-                       break;
-               }
-       }
-}
-
-static inline int refresh_timer(struct nf_conntrack_expect *i)
-{
-       struct nf_conn_help *master_help = nfct_help(i->master);
-
-       if (!del_timer(&i->timeout))
-               return 0;
-
-       i->timeout.expires = jiffies + master_help->helper->timeout*HZ;
-       add_timer(&i->timeout);
-       return 1;
-}
-
-int nf_conntrack_expect_related(struct nf_conntrack_expect *expect)
-{
-       struct nf_conntrack_expect *i;
-       struct nf_conn *master = expect->master;
-       struct nf_conn_help *master_help = nfct_help(master);
-       int ret;
-
-       NF_CT_ASSERT(master_help);
-
-       DEBUGP("nf_conntrack_expect_related %p\n", related_to);
-       DEBUGP("tuple: "); NF_CT_DUMP_TUPLE(&expect->tuple);
-       DEBUGP("mask:  "); NF_CT_DUMP_TUPLE(&expect->mask);
-
-       write_lock_bh(&nf_conntrack_lock);
-       list_for_each_entry(i, &nf_conntrack_expect_list, list) {
-               if (expect_matches(i, expect)) {
-                       /* Refresh timer: if it's dying, ignore.. */
-                       if (refresh_timer(i)) {
-                               ret = 0;
-                               goto out;
-                       }
-               } else if (expect_clash(i, expect)) {
-                       ret = -EBUSY;
-                       goto out;
-               }
-       }
-       /* Will be over limit? */
-       if (master_help->helper->max_expected &&
-           master_help->expecting >= master_help->helper->max_expected)
-               evict_oldest_expect(master);
-
-       nf_conntrack_expect_insert(expect);
-       nf_conntrack_expect_event(IPEXP_NEW, expect);
-       ret = 0;
-out:
-       write_unlock_bh(&nf_conntrack_lock);
-       return ret;
-}
-
-int nf_conntrack_helper_register(struct nf_conntrack_helper *me)
-{
-       int ret;
-       BUG_ON(me->timeout == 0);
-
-       ret = nf_conntrack_register_cache(NF_CT_F_HELP, "nf_conntrack:help",
-                                         sizeof(struct nf_conn)
-                                         + sizeof(struct nf_conn_help)
-                                         + __alignof__(struct nf_conn_help));
-       if (ret < 0) {
-               printk(KERN_ERR "nf_conntrack_helper_reigster: Unable to create slab cache for conntracks\n");
-               return ret;
-       }
-       write_lock_bh(&nf_conntrack_lock);
-       list_prepend(&helpers, me);
-       write_unlock_bh(&nf_conntrack_lock);
-
-       return 0;
-}
-
-struct nf_conntrack_helper *
-__nf_conntrack_helper_find_byname(const char *name)
-{
-       struct nf_conntrack_helper *h;
-
-       list_for_each_entry(h, &helpers, list) {
-               if (!strcmp(h->name, name))
-                       return h;
-       }
-
-       return NULL;
-}
-
-static inline int unhelp(struct nf_conntrack_tuple_hash *i,
-                        const struct nf_conntrack_helper *me)
-{
-       struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(i);
-       struct nf_conn_help *help = nfct_help(ct);
-
-       if (help && help->helper == me) {
-               nf_conntrack_event(IPCT_HELPER, ct);
-               help->helper = NULL;
-       }
-       return 0;
-}
-
-void nf_conntrack_helper_unregister(struct nf_conntrack_helper *me)
-{
-       unsigned int i;
-       struct nf_conntrack_expect *exp, *tmp;
-
-       /* Need write lock here, to delete helper. */
-       write_lock_bh(&nf_conntrack_lock);
-       LIST_DELETE(&helpers, me);
-
-       /* Get rid of expectations */
-       list_for_each_entry_safe(exp, tmp, &nf_conntrack_expect_list, list) {
-               struct nf_conn_help *help = nfct_help(exp->master);
-               if (help->helper == me && del_timer(&exp->timeout)) {
-                       nf_ct_unlink_expect(exp);
-                       nf_conntrack_expect_put(exp);
-               }
-       }
-
-       /* Get rid of expecteds, set helpers to NULL. */
-       LIST_FIND_W(&unconfirmed, unhelp, struct nf_conntrack_tuple_hash*, me);
-       for (i = 0; i < nf_conntrack_htable_size; i++)
-               LIST_FIND_W(&nf_conntrack_hash[i], unhelp,
-                           struct nf_conntrack_tuple_hash *, me);
-       write_unlock_bh(&nf_conntrack_lock);
-
-       /* Someone could be still looking at the helper in a bh. */
-       synchronize_net();
-}
-
 /* Refresh conntrack for this many jiffies and do accounting if do_acct is 1 */
 void __nf_ct_refresh_acct(struct nf_conn *ct,
                          enum ip_conntrack_info ctinfo,
@@ -1510,37 +972,41 @@ do_iter(const struct nf_conntrack_tuple_hash *i,
 }
 
 /* Bring out ya dead! */
-static struct nf_conntrack_tuple_hash *
+static struct nf_conn *
 get_next_corpse(int (*iter)(struct nf_conn *i, void *data),
                void *data, unsigned int *bucket)
 {
-       struct nf_conntrack_tuple_hash *h = NULL;
+       struct nf_conntrack_tuple_hash *h;
+       struct nf_conn *ct;
 
        write_lock_bh(&nf_conntrack_lock);
        for (; *bucket < nf_conntrack_htable_size; (*bucket)++) {
-               h = LIST_FIND_W(&nf_conntrack_hash[*bucket], do_iter,
-                               struct nf_conntrack_tuple_hash *, iter, data);
-               if (h)
-                       break;
+               list_for_each_entry(h, &nf_conntrack_hash[*bucket], list) {
+                       ct = nf_ct_tuplehash_to_ctrack(h);
+                       if (iter(ct, data))
+                               goto found;
+               }
        }
-       if (!h)
-               h = LIST_FIND_W(&unconfirmed, do_iter,
-                               struct nf_conntrack_tuple_hash *, iter, data);
-       if (h)
-               atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use);
+       list_for_each_entry(h, &unconfirmed, list) {
+               ct = nf_ct_tuplehash_to_ctrack(h);
+               if (iter(ct, data))
+                       goto found;
+       }
        write_unlock_bh(&nf_conntrack_lock);
-
-       return h;
+       return NULL;
+found:
+       atomic_inc(&ct->ct_general.use);
+       write_unlock_bh(&nf_conntrack_lock);
+       return ct;
 }
 
 void
 nf_ct_iterate_cleanup(int (*iter)(struct nf_conn *i, void *data), void *data)
 {
-       struct nf_conntrack_tuple_hash *h;
+       struct nf_conn *ct;
        unsigned int bucket = 0;
 
-       while ((h = get_next_corpse(iter, data, &bucket)) != NULL) {
-               struct nf_conn *ct = nf_ct_tuplehash_to_ctrack(h);
+       while ((ct = get_next_corpse(iter, data, &bucket)) != NULL) {
                /* Time to push up daises... */
                if (del_timer(&ct->timeout))
                        death_by_timeout((unsigned long)ct);