]> git.karo-electronics.de Git - mv-sheeva.git/blobdiff - net/netfilter/nf_conntrack_core.c
[NETFILTER]: nfnetlink: support attribute policies
[mv-sheeva.git] / net / netfilter / nf_conntrack_core.c
index 992d0ef31fa39028f037dc598f7ad20a8dc54c51..9edaaf2d57e74a1adef211e19f352671954c930f 100644 (file)
 
 #define NF_CONNTRACK_VERSION   "0.5.0"
 
-#if 0
-#define DEBUGP printk
-#else
-#define DEBUGP(format, args...)
-#endif
-
 DEFINE_RWLOCK(nf_conntrack_lock);
 EXPORT_SYMBOL_GPL(nf_conntrack_lock);
 
@@ -85,7 +79,8 @@ static u_int32_t __hash_conntrack(const struct nf_conntrack_tuple *tuple,
        a = jhash2(tuple->src.u3.all, ARRAY_SIZE(tuple->src.u3.all),
                   (tuple->src.l3num << 16) | tuple->dst.protonum);
        b = jhash2(tuple->dst.u3.all, ARRAY_SIZE(tuple->dst.u3.all),
-                  (tuple->src.u.all << 16) | tuple->dst.u.all);
+                  ((__force __u16)tuple->src.u.all << 16) |
+                   (__force __u16)tuple->dst.u.all);
 
        return jhash_2words(a, b, rnd) % size;
 }
@@ -119,6 +114,36 @@ 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)
+{
+       struct nf_conntrack_l3proto *l3proto;
+       struct nf_conntrack_l4proto *l4proto;
+       unsigned int protoff;
+       u_int8_t protonum;
+       int ret;
+
+       rcu_read_lock();
+
+       l3proto = __nf_ct_l3proto_find(l3num);
+       ret = l3proto->get_l4proto(skb, nhoff, &protoff, &protonum);
+       if (ret != NF_ACCEPT) {
+               rcu_read_unlock();
+               return 0;
+       }
+
+       l4proto = __nf_ct_l4proto_find(l3num, protonum);
+
+       ret = nf_ct_get_tuple(skb, nhoff, protoff, l3num, protonum, tuple,
+                             l3proto, l4proto);
+
+       rcu_read_unlock();
+       return ret;
+}
+EXPORT_SYMBOL_GPL(nf_ct_get_tuplepr);
+
 int
 nf_ct_invert_tuple(struct nf_conntrack_tuple *inverse,
                   const struct nf_conntrack_tuple *orig,
@@ -141,7 +166,7 @@ EXPORT_SYMBOL_GPL(nf_ct_invert_tuple);
 static void
 clean_from_lists(struct nf_conn *ct)
 {
-       DEBUGP("clean_from_lists(%p)\n", ct);
+       pr_debug("clean_from_lists(%p)\n", ct);
        hlist_del(&ct->tuplehash[IP_CT_DIR_ORIGINAL].hnode);
        hlist_del(&ct->tuplehash[IP_CT_DIR_REPLY].hnode);
 
@@ -155,7 +180,7 @@ destroy_conntrack(struct nf_conntrack *nfct)
        struct nf_conn *ct = (struct nf_conn *)nfct;
        struct nf_conntrack_l4proto *l4proto;
 
-       DEBUGP("destroy_conntrack(%p)\n", ct);
+       pr_debug("destroy_conntrack(%p)\n", ct);
        NF_CT_ASSERT(atomic_read(&nfct->use) == 0);
        NF_CT_ASSERT(!timer_pending(&ct->timeout));
 
@@ -194,7 +219,7 @@ destroy_conntrack(struct nf_conntrack *nfct)
        if (ct->master)
                nf_ct_put(ct->master);
 
-       DEBUGP("destroy_conntrack: returning ct=%p to slab\n", ct);
+       pr_debug("destroy_conntrack: returning ct=%p to slab\n", ct);
        nf_conntrack_free(ct);
 }
 
@@ -244,13 +269,12 @@ EXPORT_SYMBOL_GPL(__nf_conntrack_find);
 
 /* Find a connection corresponding to a tuple. */
 struct nf_conntrack_tuple_hash *
-nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple,
-                     const struct nf_conn *ignored_conntrack)
+nf_conntrack_find_get(const struct nf_conntrack_tuple *tuple)
 {
        struct nf_conntrack_tuple_hash *h;
 
        read_lock_bh(&nf_conntrack_lock);
-       h = __nf_conntrack_find(tuple, ignored_conntrack);
+       h = __nf_conntrack_find(tuple, NULL);
        if (h)
                atomic_inc(&nf_ct_tuplehash_to_ctrack(h)->ct_general.use);
        read_unlock_bh(&nf_conntrack_lock);
@@ -314,7 +338,7 @@ __nf_conntrack_confirm(struct sk_buff **pskb)
        /* No external references means noone else could have
           confirmed us. */
        NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
-       DEBUGP("Confirming conntrack %p\n", ct);
+       pr_debug("Confirming conntrack %p\n", ct);
 
        write_lock_bh(&nf_conntrack_lock);
 
@@ -378,21 +402,30 @@ nf_conntrack_tuple_taken(const struct nf_conntrack_tuple *tuple,
 }
 EXPORT_SYMBOL_GPL(nf_conntrack_tuple_taken);
 
+#define NF_CT_EVICTION_RANGE   8
+
 /* 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(struct hlist_head *chain)
+static int early_drop(unsigned int hash)
 {
        /* Use oldest entry, which is roughly LRU */
        struct nf_conntrack_tuple_hash *h;
        struct nf_conn *ct = NULL, *tmp;
        struct hlist_node *n;
+       unsigned int i, cnt = 0;
        int dropped = 0;
 
        read_lock_bh(&nf_conntrack_lock);
-       hlist_for_each_entry(h, n, chain, hnode) {
-               tmp = nf_ct_tuplehash_to_ctrack(h);
-               if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
-                       ct = tmp;
+       for (i = 0; i < nf_conntrack_htable_size; i++) {
+               hlist_for_each_entry(h, n, &nf_conntrack_hash[hash], hnode) {
+                       tmp = nf_ct_tuplehash_to_ctrack(h);
+                       if (!test_bit(IPS_ASSURED_BIT, &tmp->status))
+                               ct = tmp;
+                       cnt++;
+               }
+               if (ct || cnt >= NF_CT_EVICTION_RANGE)
+                       break;
+               hash = (hash + 1) % nf_conntrack_htable_size;
        }
        if (ct)
                atomic_inc(&ct->ct_general.use);
@@ -426,8 +459,7 @@ struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
        if (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])) {
+               if (!early_drop(hash)) {
                        atomic_dec(&nf_conntrack_count);
                        if (net_ratelimit())
                                printk(KERN_WARNING
@@ -439,7 +471,7 @@ struct nf_conn *nf_conntrack_alloc(const struct nf_conntrack_tuple *orig,
 
        conntrack = kmem_cache_zalloc(nf_conntrack_cachep, GFP_ATOMIC);
        if (conntrack == NULL) {
-               DEBUGP("nf_conntrack_alloc: Can't alloc conntrack.\n");
+               pr_debug("nf_conntrack_alloc: Can't alloc conntrack.\n");
                atomic_dec(&nf_conntrack_count);
                return ERR_PTR(-ENOMEM);
        }
@@ -478,37 +510,34 @@ init_conntrack(const struct nf_conntrack_tuple *tuple,
        struct nf_conntrack_expect *exp;
 
        if (!nf_ct_invert_tuple(&repl_tuple, tuple, l3proto, l4proto)) {
-               DEBUGP("Can't invert tuple.\n");
+               pr_debug("Can't invert tuple.\n");
                return NULL;
        }
 
        conntrack = nf_conntrack_alloc(tuple, &repl_tuple);
        if (conntrack == NULL || IS_ERR(conntrack)) {
-               DEBUGP("Can't allocate conntrack.\n");
+               pr_debug("Can't allocate conntrack.\n");
                return (struct nf_conntrack_tuple_hash *)conntrack;
        }
 
        if (!l4proto->new(conntrack, skb, dataoff)) {
                nf_conntrack_free(conntrack);
-               DEBUGP("init conntrack: can't track with proto module\n");
+               pr_debug("init conntrack: can't track with proto module\n");
                return NULL;
        }
 
        write_lock_bh(&nf_conntrack_lock);
-       exp = find_expectation(tuple);
+       exp = nf_ct_find_expectation(tuple);
        if (exp) {
-               DEBUGP("conntrack: expectation arrives ct=%p exp=%p\n",
-                       conntrack, exp);
+               pr_debug("conntrack: expectation arrives ct=%p exp=%p\n",
+                        conntrack, exp);
                /* Welcome, Mr. Bond.  We've been expecting you... */
                __set_bit(IPS_EXPECTED_BIT, &conntrack->status);
                conntrack->master = exp->master;
                if (exp->helper) {
-                       help = nf_ct_ext_add(conntrack, NF_CT_EXT_HELPER,
-                                            GFP_ATOMIC);
+                       help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC);
                        if (help)
                                rcu_assign_pointer(help->helper, exp->helper);
-                       else
-                               DEBUGP("failed to add helper extension area");
                }
 
 #ifdef CONFIG_NF_CONNTRACK_MARK
@@ -524,14 +553,9 @@ init_conntrack(const struct nf_conntrack_tuple *tuple,
 
                helper = __nf_ct_helper_find(&repl_tuple);
                if (helper) {
-                       help = nf_ct_ext_add(conntrack, NF_CT_EXT_HELPER,
-                                            GFP_ATOMIC);
+                       help = nf_ct_helper_ext_add(conntrack, GFP_ATOMIC);
                        if (help)
-                               /* not in hash table yet, so not strictly
-                                  necessary */
                                rcu_assign_pointer(help->helper, helper);
-                       else
-                               DEBUGP("failed to add helper extension area");
                }
                NF_CT_STAT_INC(new);
        }
@@ -545,7 +569,7 @@ init_conntrack(const struct nf_conntrack_tuple *tuple,
        if (exp) {
                if (exp->expectfn)
                        exp->expectfn(conntrack, exp);
-               nf_conntrack_expect_put(exp);
+               nf_ct_expect_put(exp);
        }
 
        return &conntrack->tuplehash[IP_CT_DIR_ORIGINAL];
@@ -569,12 +593,12 @@ resolve_normal_ct(struct sk_buff *skb,
        if (!nf_ct_get_tuple(skb, skb_network_offset(skb),
                             dataoff, l3num, protonum, &tuple, l3proto,
                             l4proto)) {
-               DEBUGP("resolve_normal_ct: Can't get tuple\n");
+               pr_debug("resolve_normal_ct: Can't get tuple\n");
                return NULL;
        }
 
        /* look for tuple match */
-       h = nf_conntrack_find_get(&tuple, NULL);
+       h = nf_conntrack_find_get(&tuple);
        if (!h) {
                h = init_conntrack(&tuple, l3proto, l4proto, skb, dataoff);
                if (!h)
@@ -592,13 +616,14 @@ resolve_normal_ct(struct sk_buff *skb,
        } else {
                /* Once we've had two way comms, always ESTABLISHED. */
                if (test_bit(IPS_SEEN_REPLY_BIT, &ct->status)) {
-                       DEBUGP("nf_conntrack_in: normal packet for %p\n", ct);
+                       pr_debug("nf_conntrack_in: normal packet for %p\n", ct);
                        *ctinfo = IP_CT_ESTABLISHED;
                } else if (test_bit(IPS_EXPECTED_BIT, &ct->status)) {
-                       DEBUGP("nf_conntrack_in: related packet for %p\n", ct);
+                       pr_debug("nf_conntrack_in: related packet for %p\n",
+                                ct);
                        *ctinfo = IP_CT_RELATED;
                } else {
-                       DEBUGP("nf_conntrack_in: new packet for %p\n", ct);
+                       pr_debug("nf_conntrack_in: new packet for %p\n", ct);
                        *ctinfo = IP_CT_NEW;
                }
                *set_reply = 0;
@@ -628,9 +653,12 @@ nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb)
 
        /* rcu_read_lock()ed by nf_hook_slow */
        l3proto = __nf_ct_l3proto_find((u_int16_t)pf);
-
-       if ((ret = l3proto->prepare(pskb, hooknum, &dataoff, &protonum)) <= 0) {
-               DEBUGP("not prepared to track yet or error occured\n");
+       ret = l3proto->get_l4proto(*pskb, skb_network_offset(*pskb),
+                                  &dataoff, &protonum);
+       if (ret <= 0) {
+               pr_debug("not prepared to track yet or error occured\n");
+               NF_CT_STAT_INC_ATOMIC(error);
+               NF_CT_STAT_INC_ATOMIC(invalid);
                return -ret;
        }
 
@@ -666,7 +694,7 @@ nf_conntrack_in(int pf, unsigned int hooknum, struct sk_buff **pskb)
        if (ret < 0) {
                /* Invalid: inverse of the return code tells
                 * the netfilter core what to do */
-               DEBUGP("nf_conntrack_in: Can't track with proto module\n");
+               pr_debug("nf_conntrack_in: Can't track with proto module\n");
                nf_conntrack_put((*pskb)->nfct);
                (*pskb)->nfct = NULL;
                NF_CT_STAT_INC_ATOMIC(invalid);
@@ -707,7 +735,7 @@ void nf_conntrack_alter_reply(struct nf_conn *ct,
        /* Should be unconfirmed, so not in hash table yet */
        NF_CT_ASSERT(!nf_ct_is_confirmed(ct));
 
-       DEBUGP("Altering reply tuple of %p to ", ct);
+       pr_debug("Altering reply tuple of %p to ", ct);
        NF_CT_DUMP_TUPLE(newreply);
 
        ct->tuplehash[IP_CT_DIR_REPLY].tuple = *newreply;
@@ -722,11 +750,9 @@ void nf_conntrack_alter_reply(struct nf_conn *ct,
        }
 
        if (help == NULL) {
-               help = nf_ct_ext_add(ct, NF_CT_EXT_HELPER, GFP_ATOMIC);
-               if (help == NULL) {
-                       DEBUGP("failed to add helper extension area");
+               help = nf_ct_helper_ext_add(ct, GFP_ATOMIC);
+               if (help == NULL)
                        goto out;
-               }
        } else {
                memset(&help->help, 0, sizeof(help->help));
        }
@@ -801,44 +827,43 @@ EXPORT_SYMBOL_GPL(__nf_ct_refresh_acct);
 #include <linux/netfilter/nfnetlink_conntrack.h>
 #include <linux/mutex.h>
 
-
 /* Generic function for tcp/udp/sctp/dccp and alike. This needs to be
  * in ip_conntrack_core, since we don't want the protocols to autoload
  * or depend on ctnetlink */
-int nf_ct_port_tuple_to_nfattr(struct sk_buff *skb,
+int nf_ct_port_tuple_to_nlattr(struct sk_buff *skb,
                               const struct nf_conntrack_tuple *tuple)
 {
-       NFA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t),
+       NLA_PUT(skb, CTA_PROTO_SRC_PORT, sizeof(u_int16_t),
                &tuple->src.u.tcp.port);
-       NFA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t),
+       NLA_PUT(skb, CTA_PROTO_DST_PORT, sizeof(u_int16_t),
                &tuple->dst.u.tcp.port);
        return 0;
 
-nfattr_failure:
+nla_put_failure:
        return -1;
 }
-EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nfattr);
+EXPORT_SYMBOL_GPL(nf_ct_port_tuple_to_nlattr);
 
-static const size_t cta_min_proto[CTA_PROTO_MAX] = {
-       [CTA_PROTO_SRC_PORT-1]  = sizeof(u_int16_t),
-       [CTA_PROTO_DST_PORT-1]  = sizeof(u_int16_t)
+static const size_t cta_min_proto[CTA_PROTO_MAX+1] = {
+       [CTA_PROTO_SRC_PORT]  = sizeof(u_int16_t),
+       [CTA_PROTO_DST_PORT]  = sizeof(u_int16_t)
 };
 
-int nf_ct_port_nfattr_to_tuple(struct nfattr *tb[],
+int nf_ct_port_nlattr_to_tuple(struct nlattr *tb[],
                               struct nf_conntrack_tuple *t)
 {
-       if (!tb[CTA_PROTO_SRC_PORT-1] || !tb[CTA_PROTO_DST_PORT-1])
+       if (!tb[CTA_PROTO_SRC_PORT] || !tb[CTA_PROTO_DST_PORT])
                return -EINVAL;
 
-       if (nfattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto))
+       if (nlattr_bad_size(tb, CTA_PROTO_MAX, cta_min_proto))
                return -EINVAL;
 
-       t->src.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_SRC_PORT-1]);
-       t->dst.u.tcp.port = *(__be16 *)NFA_DATA(tb[CTA_PROTO_DST_PORT-1]);
+       t->src.u.tcp.port = *(__be16 *)nla_data(tb[CTA_PROTO_SRC_PORT]);
+       t->dst.u.tcp.port = *(__be16 *)nla_data(tb[CTA_PROTO_DST_PORT]);
 
        return 0;
 }
-EXPORT_SYMBOL_GPL(nf_ct_port_nfattr_to_tuple);
+EXPORT_SYMBOL_GPL(nf_ct_port_nlattr_to_tuple);
 #endif
 
 /* Used by ipt_REJECT and ip6t_REJECT. */
@@ -921,8 +946,7 @@ static int kill_all(struct nf_conn *i, void *data)
        return 1;
 }
 
-static void free_conntrack_hash(struct hlist_head *hash, int vmalloced,
-                               int size)
+void nf_ct_free_hashtable(struct hlist_head *hash, int vmalloced, int size)
 {
        if (vmalloced)
                vfree(hash);
@@ -930,6 +954,7 @@ static void free_conntrack_hash(struct hlist_head *hash, int vmalloced,
                free_pages((unsigned long)hash,
                           get_order(sizeof(struct hlist_head) * size));
 }
+EXPORT_SYMBOL_GPL(nf_ct_free_hashtable);
 
 void nf_conntrack_flush(void)
 {
@@ -962,15 +987,15 @@ void nf_conntrack_cleanup(void)
        rcu_assign_pointer(nf_ct_destroy, NULL);
 
        kmem_cache_destroy(nf_conntrack_cachep);
-       kmem_cache_destroy(nf_conntrack_expect_cachep);
-       free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc,
-                           nf_conntrack_htable_size);
+       nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
+                            nf_conntrack_htable_size);
 
        nf_conntrack_proto_fini();
        nf_conntrack_helper_fini();
+       nf_conntrack_expect_fini();
 }
 
-static struct hlist_head *alloc_hashtable(int *sizep, int *vmalloced)
+struct hlist_head *nf_ct_alloc_hashtable(int *sizep, int *vmalloced)
 {
        struct hlist_head *hash;
        unsigned int size, i;
@@ -993,6 +1018,7 @@ static struct hlist_head *alloc_hashtable(int *sizep, int *vmalloced)
 
        return hash;
 }
+EXPORT_SYMBOL_GPL(nf_ct_alloc_hashtable);
 
 int set_hashsize(const char *val, struct kernel_param *kp)
 {
@@ -1010,7 +1036,7 @@ int set_hashsize(const char *val, struct kernel_param *kp)
        if (!hashsize)
                return -EINVAL;
 
-       hash = alloc_hashtable(&hashsize, &vmalloced);
+       hash = nf_ct_alloc_hashtable(&hashsize, &vmalloced);
        if (!hash)
                return -ENOMEM;
 
@@ -1038,7 +1064,7 @@ int set_hashsize(const char *val, struct kernel_param *kp)
        nf_conntrack_hash_rnd = rnd;
        write_unlock_bh(&nf_conntrack_lock);
 
-       free_conntrack_hash(old_hash, old_vmalloced, old_size);
+       nf_ct_free_hashtable(old_hash, old_vmalloced, old_size);
        return 0;
 }
 
@@ -1067,8 +1093,8 @@ int __init nf_conntrack_init(void)
                 * entries. */
                max_factor = 4;
        }
-       nf_conntrack_hash = alloc_hashtable(&nf_conntrack_htable_size,
-                                           &nf_conntrack_vmalloc);
+       nf_conntrack_hash = nf_ct_alloc_hashtable(&nf_conntrack_htable_size,
+                                                 &nf_conntrack_vmalloc);
        if (!nf_conntrack_hash) {
                printk(KERN_ERR "Unable to create nf_conntrack_hash\n");
                goto err_out;
@@ -1082,27 +1108,23 @@ int __init nf_conntrack_init(void)
 
        nf_conntrack_cachep = kmem_cache_create("nf_conntrack",
                                                sizeof(struct nf_conn),
-                                               0, 0, NULL, NULL);
+                                               0, 0, NULL);
        if (!nf_conntrack_cachep) {
                printk(KERN_ERR "Unable to create nf_conn slab cache\n");
                goto err_free_hash;
        }
 
-       nf_conntrack_expect_cachep = kmem_cache_create("nf_conntrack_expect",
-                                       sizeof(struct nf_conntrack_expect),
-                                       0, 0, NULL, NULL);
-       if (!nf_conntrack_expect_cachep) {
-               printk(KERN_ERR "Unable to create nf_expect slab cache\n");
+       ret = nf_conntrack_proto_init();
+       if (ret < 0)
                goto err_free_conntrack_slab;
-       }
 
-       ret = nf_conntrack_proto_init();
+       ret = nf_conntrack_expect_init();
        if (ret < 0)
-               goto out_free_expect_slab;
+               goto out_fini_proto;
 
        ret = nf_conntrack_helper_init();
        if (ret < 0)
-               goto out_fini_proto;
+               goto out_fini_expect;
 
        /* For use by REJECT target */
        rcu_assign_pointer(ip_ct_attach, __nf_conntrack_attach);
@@ -1116,15 +1138,15 @@ int __init nf_conntrack_init(void)
 
        return ret;
 
+out_fini_expect:
+       nf_conntrack_expect_fini();
 out_fini_proto:
        nf_conntrack_proto_fini();
-out_free_expect_slab:
-       kmem_cache_destroy(nf_conntrack_expect_cachep);
 err_free_conntrack_slab:
        kmem_cache_destroy(nf_conntrack_cachep);
 err_free_hash:
-       free_conntrack_hash(nf_conntrack_hash, nf_conntrack_vmalloc,
-                           nf_conntrack_htable_size);
+       nf_ct_free_hashtable(nf_conntrack_hash, nf_conntrack_vmalloc,
+                            nf_conntrack_htable_size);
 err_out:
        return -ENOMEM;
 }