This commit was manufactured by cvs2svn to create branch 'vserver'.
[linux-2.6.git] / net / ipv4 / netfilter / nf_nat_core.c
diff --git a/net/ipv4/netfilter/nf_nat_core.c b/net/ipv4/netfilter/nf_nat_core.c
new file mode 100644 (file)
index 0000000..8de7188
--- /dev/null
@@ -0,0 +1,646 @@
+/* NAT for netfilter; shared with compatibility layer. */
+
+/* (C) 1999-2001 Paul `Rusty' Russell
+ * (C) 2002-2006 Netfilter Core Team <coreteam@netfilter.org>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/module.h>
+#include <linux/types.h>
+#include <linux/timer.h>
+#include <linux/skbuff.h>
+#include <linux/vmalloc.h>
+#include <net/checksum.h>
+#include <net/icmp.h>
+#include <net/ip.h>
+#include <net/tcp.h>  /* For tcp_prot in getorigdst */
+#include <linux/icmp.h>
+#include <linux/udp.h>
+#include <linux/jhash.h>
+
+#include <linux/netfilter_ipv4.h>
+#include <net/netfilter/nf_conntrack.h>
+#include <net/netfilter/nf_conntrack_core.h>
+#include <net/netfilter/nf_nat.h>
+#include <net/netfilter/nf_nat_protocol.h>
+#include <net/netfilter/nf_nat_core.h>
+#include <net/netfilter/nf_nat_helper.h>
+#include <net/netfilter/nf_conntrack_helper.h>
+#include <net/netfilter/nf_conntrack_l3proto.h>
+#include <net/netfilter/nf_conntrack_l4proto.h>
+
+#if 0
+#define DEBUGP printk
+#else
+#define DEBUGP(format, args...)
+#endif
+
+static DEFINE_RWLOCK(nf_nat_lock);
+
+static struct nf_conntrack_l3proto *l3proto = NULL;
+
+/* Calculated at init based on memory size */
+static unsigned int nf_nat_htable_size;
+
+static struct list_head *bysource;
+
+#define MAX_IP_NAT_PROTO 256
+static struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO];
+
+static inline struct nf_nat_protocol *
+__nf_nat_proto_find(u_int8_t protonum)
+{
+       return nf_nat_protos[protonum];
+}
+
+struct nf_nat_protocol *
+nf_nat_proto_find_get(u_int8_t protonum)
+{
+       struct nf_nat_protocol *p;
+
+       /* we need to disable preemption to make sure 'p' doesn't get
+        * removed until we've grabbed the reference */
+       preempt_disable();
+       p = __nf_nat_proto_find(protonum);
+       if (!try_module_get(p->me))
+               p = &nf_nat_unknown_protocol;
+       preempt_enable();
+
+       return p;
+}
+EXPORT_SYMBOL_GPL(nf_nat_proto_find_get);
+
+void
+nf_nat_proto_put(struct nf_nat_protocol *p)
+{
+       module_put(p->me);
+}
+EXPORT_SYMBOL_GPL(nf_nat_proto_put);
+
+/* We keep an extra hash for each conntrack, for fast searching. */
+static inline unsigned int
+hash_by_src(const struct nf_conntrack_tuple *tuple)
+{
+       /* Original src, to ensure we map it consistently if poss. */
+       return jhash_3words((__force u32)tuple->src.u3.ip, tuple->src.u.all,
+                           tuple->dst.protonum, 0) % nf_nat_htable_size;
+}
+
+/* Noone using conntrack by the time this called. */
+static void nf_nat_cleanup_conntrack(struct nf_conn *conn)
+{
+       struct nf_conn_nat *nat;
+       if (!(conn->status & IPS_NAT_DONE_MASK))
+               return;
+
+       nat = nfct_nat(conn);
+       write_lock_bh(&nf_nat_lock);
+       list_del(&nat->info.bysource);
+       write_unlock_bh(&nf_nat_lock);
+}
+
+/* Is this tuple already taken? (not by us) */
+int
+nf_nat_used_tuple(const struct nf_conntrack_tuple *tuple,
+                 const struct nf_conn *ignored_conntrack)
+{
+       /* Conntrack tracking doesn't keep track of outgoing tuples; only
+          incoming ones.  NAT means they don't have a fixed mapping,
+          so we invert the tuple and look for the incoming reply.
+
+          We could keep a separate hash if this proves too slow. */
+       struct nf_conntrack_tuple reply;
+
+       nf_ct_invert_tuplepr(&reply, tuple);
+       return nf_conntrack_tuple_taken(&reply, ignored_conntrack);
+}
+EXPORT_SYMBOL(nf_nat_used_tuple);
+
+/* If we source map this tuple so reply looks like reply_tuple, will
+ * that meet the constraints of range. */
+static int
+in_range(const struct nf_conntrack_tuple *tuple,
+        const struct nf_nat_range *range)
+{
+       struct nf_nat_protocol *proto;
+
+       proto = __nf_nat_proto_find(tuple->dst.protonum);
+       /* If we are supposed to map IPs, then we must be in the
+          range specified, otherwise let this drag us onto a new src IP. */
+       if (range->flags & IP_NAT_RANGE_MAP_IPS) {
+               if (ntohl(tuple->src.u3.ip) < ntohl(range->min_ip) ||
+                   ntohl(tuple->src.u3.ip) > ntohl(range->max_ip))
+                       return 0;
+       }
+
+       if (!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
+           proto->in_range(tuple, IP_NAT_MANIP_SRC,
+                           &range->min, &range->max))
+               return 1;
+
+       return 0;
+}
+
+static inline int
+same_src(const struct nf_conn *ct,
+        const struct nf_conntrack_tuple *tuple)
+{
+       const struct nf_conntrack_tuple *t;
+
+       t = &ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple;
+       return (t->dst.protonum == tuple->dst.protonum &&
+               t->src.u3.ip == tuple->src.u3.ip &&
+               t->src.u.all == tuple->src.u.all);
+}
+
+/* Only called for SRC manip */
+static int
+find_appropriate_src(const struct nf_conntrack_tuple *tuple,
+                    struct nf_conntrack_tuple *result,
+                    const struct nf_nat_range *range)
+{
+       unsigned int h = hash_by_src(tuple);
+       struct nf_conn_nat *nat;
+       struct nf_conn *ct;
+
+       read_lock_bh(&nf_nat_lock);
+       list_for_each_entry(nat, &bysource[h], info.bysource) {
+               ct = (struct nf_conn *)((char *)nat - offsetof(struct nf_conn, data));
+               if (same_src(ct, tuple)) {
+                       /* Copy source part from reply tuple. */
+                       nf_ct_invert_tuplepr(result,
+                                      &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
+                       result->dst = tuple->dst;
+
+                       if (in_range(result, range)) {
+                               read_unlock_bh(&nf_nat_lock);
+                               return 1;
+                       }
+               }
+       }
+       read_unlock_bh(&nf_nat_lock);
+       return 0;
+}
+
+/* For [FUTURE] fragmentation handling, we want the least-used
+   src-ip/dst-ip/proto triple.  Fairness doesn't come into it.  Thus
+   if the range specifies 1.2.3.4 ports 10000-10005 and 1.2.3.5 ports
+   1-65535, we don't do pro-rata allocation based on ports; we choose
+   the ip with the lowest src-ip/dst-ip/proto usage.
+*/
+static void
+find_best_ips_proto(struct nf_conntrack_tuple *tuple,
+                   const struct nf_nat_range *range,
+                   const struct nf_conn *ct,
+                   enum nf_nat_manip_type maniptype)
+{
+       __be32 *var_ipp;
+       /* Host order */
+       u_int32_t minip, maxip, j;
+
+       /* No IP mapping?  Do nothing. */
+       if (!(range->flags & IP_NAT_RANGE_MAP_IPS))
+               return;
+
+       if (maniptype == IP_NAT_MANIP_SRC)
+               var_ipp = &tuple->src.u3.ip;
+       else
+               var_ipp = &tuple->dst.u3.ip;
+
+       /* Fast path: only one choice. */
+       if (range->min_ip == range->max_ip) {
+               *var_ipp = range->min_ip;
+               return;
+       }
+
+       /* Hashing source and destination IPs gives a fairly even
+        * spread in practice (if there are a small number of IPs
+        * involved, there usually aren't that many connections
+        * anyway).  The consistency means that servers see the same
+        * client coming from the same IP (some Internet Banking sites
+        * like this), even across reboots. */
+       minip = ntohl(range->min_ip);
+       maxip = ntohl(range->max_ip);
+       j = jhash_2words((__force u32)tuple->src.u3.ip,
+                        (__force u32)tuple->dst.u3.ip, 0);
+       *var_ipp = htonl(minip + j % (maxip - minip + 1));
+}
+
+/* Manipulate the tuple into the range given.  For NF_IP_POST_ROUTING,
+ * we change the source to map into the range.  For NF_IP_PRE_ROUTING
+ * and NF_IP_LOCAL_OUT, we change the destination to map into the
+ * range.  It might not be possible to get a unique tuple, but we try.
+ * At worst (or if we race), we will end up with a final duplicate in
+ * __ip_conntrack_confirm and drop the packet. */
+static void
+get_unique_tuple(struct nf_conntrack_tuple *tuple,
+                const struct nf_conntrack_tuple *orig_tuple,
+                const struct nf_nat_range *range,
+                struct nf_conn *ct,
+                enum nf_nat_manip_type maniptype)
+{
+       struct nf_nat_protocol *proto;
+
+       /* 1) If this srcip/proto/src-proto-part is currently mapped,
+          and that same mapping gives a unique tuple within the given
+          range, use that.
+
+          This is only required for source (ie. NAT/masq) mappings.
+          So far, we don't do local source mappings, so multiple
+          manips not an issue.  */
+       if (maniptype == IP_NAT_MANIP_SRC) {
+               if (find_appropriate_src(orig_tuple, tuple, range)) {
+                       DEBUGP("get_unique_tuple: Found current src map\n");
+                       if (!nf_nat_used_tuple(tuple, ct))
+                               return;
+               }
+       }
+
+       /* 2) Select the least-used IP/proto combination in the given
+          range. */
+       *tuple = *orig_tuple;
+       find_best_ips_proto(tuple, range, ct, maniptype);
+
+       /* 3) The per-protocol part of the manip is made to map into
+          the range to make a unique tuple. */
+
+       proto = nf_nat_proto_find_get(orig_tuple->dst.protonum);
+
+       /* Only bother mapping if it's not already in range and unique */
+       if ((!(range->flags & IP_NAT_RANGE_PROTO_SPECIFIED) ||
+            proto->in_range(tuple, maniptype, &range->min, &range->max)) &&
+           !nf_nat_used_tuple(tuple, ct)) {
+               nf_nat_proto_put(proto);
+               return;
+       }
+
+       /* Last change: get protocol to try to obtain unique tuple. */
+       proto->unique_tuple(tuple, range, maniptype, ct);
+
+       nf_nat_proto_put(proto);
+}
+
+unsigned int
+nf_nat_setup_info(struct nf_conn *ct,
+                 const struct nf_nat_range *range,
+                 unsigned int hooknum)
+{
+       struct nf_conntrack_tuple curr_tuple, new_tuple;
+       struct nf_conn_nat *nat = nfct_nat(ct);
+       struct nf_nat_info *info = &nat->info;
+       int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK);
+       enum nf_nat_manip_type maniptype = HOOK2MANIP(hooknum);
+
+       NF_CT_ASSERT(hooknum == NF_IP_PRE_ROUTING ||
+                    hooknum == NF_IP_POST_ROUTING ||
+                    hooknum == NF_IP_LOCAL_IN ||
+                    hooknum == NF_IP_LOCAL_OUT);
+       BUG_ON(nf_nat_initialized(ct, maniptype));
+
+       /* What we've got will look like inverse of reply. Normally
+          this is what is in the conntrack, except for prior
+          manipulations (future optimization: if num_manips == 0,
+          orig_tp =
+          conntrack->tuplehash[IP_CT_DIR_ORIGINAL].tuple) */
+       nf_ct_invert_tuplepr(&curr_tuple,
+                            &ct->tuplehash[IP_CT_DIR_REPLY].tuple);
+
+       get_unique_tuple(&new_tuple, &curr_tuple, range, ct, maniptype);
+
+       if (!nf_ct_tuple_equal(&new_tuple, &curr_tuple)) {
+               struct nf_conntrack_tuple reply;
+
+               /* Alter conntrack table so will recognize replies. */
+               nf_ct_invert_tuplepr(&reply, &new_tuple);
+               nf_conntrack_alter_reply(ct, &reply);
+
+               /* Non-atomic: we own this at the moment. */
+               if (maniptype == IP_NAT_MANIP_SRC)
+                       ct->status |= IPS_SRC_NAT;
+               else
+                       ct->status |= IPS_DST_NAT;
+       }
+
+       /* Place in source hash if this is the first time. */
+       if (have_to_hash) {
+               unsigned int srchash;
+
+               srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
+               write_lock_bh(&nf_nat_lock);
+               list_add(&info->bysource, &bysource[srchash]);
+               write_unlock_bh(&nf_nat_lock);
+       }
+
+       /* It's done. */
+       if (maniptype == IP_NAT_MANIP_DST)
+               set_bit(IPS_DST_NAT_DONE_BIT, &ct->status);
+       else
+               set_bit(IPS_SRC_NAT_DONE_BIT, &ct->status);
+
+       return NF_ACCEPT;
+}
+EXPORT_SYMBOL(nf_nat_setup_info);
+
+/* Returns true if succeeded. */
+static int
+manip_pkt(u_int16_t proto,
+         struct sk_buff **pskb,
+         unsigned int iphdroff,
+         const struct nf_conntrack_tuple *target,
+         enum nf_nat_manip_type maniptype)
+{
+       struct iphdr *iph;
+       struct nf_nat_protocol *p;
+
+       if (!skb_make_writable(pskb, iphdroff + sizeof(*iph)))
+               return 0;
+
+       iph = (void *)(*pskb)->data + iphdroff;
+
+       /* Manipulate protcol part. */
+       p = nf_nat_proto_find_get(proto);
+       if (!p->manip_pkt(pskb, iphdroff, target, maniptype)) {
+               nf_nat_proto_put(p);
+               return 0;
+       }
+       nf_nat_proto_put(p);
+
+       iph = (void *)(*pskb)->data + iphdroff;
+
+       if (maniptype == IP_NAT_MANIP_SRC) {
+               nf_csum_replace4(&iph->check, iph->saddr, target->src.u3.ip);
+               iph->saddr = target->src.u3.ip;
+       } else {
+               nf_csum_replace4(&iph->check, iph->daddr, target->dst.u3.ip);
+               iph->daddr = target->dst.u3.ip;
+       }
+       return 1;
+}
+
+/* Do packet manipulations according to nf_nat_setup_info. */
+unsigned int nf_nat_packet(struct nf_conn *ct,
+                          enum ip_conntrack_info ctinfo,
+                          unsigned int hooknum,
+                          struct sk_buff **pskb)
+{
+       enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
+       unsigned long statusbit;
+       enum nf_nat_manip_type mtype = HOOK2MANIP(hooknum);
+
+       if (mtype == IP_NAT_MANIP_SRC)
+               statusbit = IPS_SRC_NAT;
+       else
+               statusbit = IPS_DST_NAT;
+
+       /* Invert if this is reply dir. */
+       if (dir == IP_CT_DIR_REPLY)
+               statusbit ^= IPS_NAT_MASK;
+
+       /* Non-atomic: these bits don't change. */
+       if (ct->status & statusbit) {
+               struct nf_conntrack_tuple target;
+
+               /* We are aiming to look like inverse of other direction. */
+               nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
+
+               if (!manip_pkt(target.dst.protonum, pskb, 0, &target, mtype))
+                       return NF_DROP;
+       }
+       return NF_ACCEPT;
+}
+EXPORT_SYMBOL_GPL(nf_nat_packet);
+
+/* Dir is direction ICMP is coming from (opposite to packet it contains) */
+int nf_nat_icmp_reply_translation(struct nf_conn *ct,
+                                 enum ip_conntrack_info ctinfo,
+                                 unsigned int hooknum,
+                                 struct sk_buff **pskb)
+{
+       struct {
+               struct icmphdr icmp;
+               struct iphdr ip;
+       } *inside;
+       struct nf_conntrack_tuple inner, target;
+       int hdrlen = (*pskb)->nh.iph->ihl * 4;
+       enum ip_conntrack_dir dir = CTINFO2DIR(ctinfo);
+       unsigned long statusbit;
+       enum nf_nat_manip_type manip = HOOK2MANIP(hooknum);
+
+       if (!skb_make_writable(pskb, hdrlen + sizeof(*inside)))
+               return 0;
+
+       inside = (void *)(*pskb)->data + (*pskb)->nh.iph->ihl*4;
+
+       /* We're actually going to mangle it beyond trivial checksum
+          adjustment, so make sure the current checksum is correct. */
+       if (nf_ip_checksum(*pskb, hooknum, hdrlen, 0))
+               return 0;
+
+       /* Must be RELATED */
+       NF_CT_ASSERT((*pskb)->nfctinfo == IP_CT_RELATED ||
+                    (*pskb)->nfctinfo == IP_CT_RELATED+IP_CT_IS_REPLY);
+
+       /* Redirects on non-null nats must be dropped, else they'll
+           start talking to each other without our translation, and be
+           confused... --RR */
+       if (inside->icmp.type == ICMP_REDIRECT) {
+               /* If NAT isn't finished, assume it and drop. */
+               if ((ct->status & IPS_NAT_DONE_MASK) != IPS_NAT_DONE_MASK)
+                       return 0;
+
+               if (ct->status & IPS_NAT_MASK)
+                       return 0;
+       }
+
+       DEBUGP("icmp_reply_translation: translating error %p manp %u dir %s\n",
+              *pskb, manip, dir == IP_CT_DIR_ORIGINAL ? "ORIG" : "REPLY");
+
+       if (!nf_ct_get_tuple(*pskb,
+                            (*pskb)->nh.iph->ihl*4 + sizeof(struct icmphdr),
+                            (*pskb)->nh.iph->ihl*4 +
+                            sizeof(struct icmphdr) + inside->ip.ihl*4,
+                            (u_int16_t)AF_INET,
+                            inside->ip.protocol,
+                            &inner,
+                            l3proto,
+                            __nf_ct_l4proto_find((u_int16_t)PF_INET,
+                                                 inside->ip.protocol)))
+               return 0;
+
+       /* Change inner back to look like incoming packet.  We do the
+          opposite manip on this hook to normal, because it might not
+          pass all hooks (locally-generated ICMP).  Consider incoming
+          packet: PREROUTING (DST manip), routing produces ICMP, goes
+          through POSTROUTING (which must correct the DST manip). */
+       if (!manip_pkt(inside->ip.protocol, pskb,
+                      (*pskb)->nh.iph->ihl*4 + sizeof(inside->icmp),
+                      &ct->tuplehash[!dir].tuple,
+                      !manip))
+               return 0;
+
+       if ((*pskb)->ip_summed != CHECKSUM_PARTIAL) {
+               /* Reloading "inside" here since manip_pkt inner. */
+               inside = (void *)(*pskb)->data + (*pskb)->nh.iph->ihl*4;
+               inside->icmp.checksum = 0;
+               inside->icmp.checksum =
+                       csum_fold(skb_checksum(*pskb, hdrlen,
+                                              (*pskb)->len - hdrlen, 0));
+       }
+
+       /* Change outer to look the reply to an incoming packet
+        * (proto 0 means don't invert per-proto part). */
+       if (manip == IP_NAT_MANIP_SRC)
+               statusbit = IPS_SRC_NAT;
+       else
+               statusbit = IPS_DST_NAT;
+
+       /* Invert if this is reply dir. */
+       if (dir == IP_CT_DIR_REPLY)
+               statusbit ^= IPS_NAT_MASK;
+
+       if (ct->status & statusbit) {
+               nf_ct_invert_tuplepr(&target, &ct->tuplehash[!dir].tuple);
+               if (!manip_pkt(0, pskb, 0, &target, manip))
+                       return 0;
+       }
+
+       return 1;
+}
+EXPORT_SYMBOL_GPL(nf_nat_icmp_reply_translation);
+
+/* Protocol registration. */
+int nf_nat_protocol_register(struct nf_nat_protocol *proto)
+{
+       int ret = 0;
+
+       write_lock_bh(&nf_nat_lock);
+       if (nf_nat_protos[proto->protonum] != &nf_nat_unknown_protocol) {
+               ret = -EBUSY;
+               goto out;
+       }
+       nf_nat_protos[proto->protonum] = proto;
+ out:
+       write_unlock_bh(&nf_nat_lock);
+       return ret;
+}
+EXPORT_SYMBOL(nf_nat_protocol_register);
+
+/* Noone stores the protocol anywhere; simply delete it. */
+void nf_nat_protocol_unregister(struct nf_nat_protocol *proto)
+{
+       write_lock_bh(&nf_nat_lock);
+       nf_nat_protos[proto->protonum] = &nf_nat_unknown_protocol;
+       write_unlock_bh(&nf_nat_lock);
+
+       /* Someone could be still looking at the proto in a bh. */
+       synchronize_net();
+}
+EXPORT_SYMBOL(nf_nat_protocol_unregister);
+
+#if defined(CONFIG_NF_CT_NETLINK) || defined(CONFIG_NF_CT_NETLINK_MODULE)
+int
+nf_nat_port_range_to_nfattr(struct sk_buff *skb,
+                           const struct nf_nat_range *range)
+{
+       NFA_PUT(skb, CTA_PROTONAT_PORT_MIN, sizeof(__be16),
+               &range->min.tcp.port);
+       NFA_PUT(skb, CTA_PROTONAT_PORT_MAX, sizeof(__be16),
+               &range->max.tcp.port);
+
+       return 0;
+
+nfattr_failure:
+       return -1;
+}
+EXPORT_SYMBOL_GPL(nf_nat_port_nfattr_to_range);
+
+int
+nf_nat_port_nfattr_to_range(struct nfattr *tb[], struct nf_nat_range *range)
+{
+       int ret = 0;
+
+       /* we have to return whether we actually parsed something or not */
+
+       if (tb[CTA_PROTONAT_PORT_MIN-1]) {
+               ret = 1;
+               range->min.tcp.port =
+                       *(__be16 *)NFA_DATA(tb[CTA_PROTONAT_PORT_MIN-1]);
+       }
+
+       if (!tb[CTA_PROTONAT_PORT_MAX-1]) {
+               if (ret)
+                       range->max.tcp.port = range->min.tcp.port;
+       } else {
+               ret = 1;
+               range->max.tcp.port =
+                       *(__be16 *)NFA_DATA(tb[CTA_PROTONAT_PORT_MAX-1]);
+       }
+
+       return ret;
+}
+EXPORT_SYMBOL_GPL(nf_nat_port_range_to_nfattr);
+#endif
+
+static int __init nf_nat_init(void)
+{
+       size_t i;
+
+       /* Leave them the same for the moment. */
+       nf_nat_htable_size = nf_conntrack_htable_size;
+
+       /* One vmalloc for both hash tables */
+       bysource = vmalloc(sizeof(struct list_head) * nf_nat_htable_size);
+       if (!bysource)
+               return -ENOMEM;
+
+       /* Sew in builtin protocols. */
+       write_lock_bh(&nf_nat_lock);
+       for (i = 0; i < MAX_IP_NAT_PROTO; i++)
+               nf_nat_protos[i] = &nf_nat_unknown_protocol;
+       nf_nat_protos[IPPROTO_TCP] = &nf_nat_protocol_tcp;
+       nf_nat_protos[IPPROTO_UDP] = &nf_nat_protocol_udp;
+       nf_nat_protos[IPPROTO_ICMP] = &nf_nat_protocol_icmp;
+       write_unlock_bh(&nf_nat_lock);
+
+       for (i = 0; i < nf_nat_htable_size; i++) {
+               INIT_LIST_HEAD(&bysource[i]);
+       }
+
+       /* FIXME: Man, this is a hack.  <SIGH> */
+       NF_CT_ASSERT(nf_conntrack_destroyed == NULL);
+       nf_conntrack_destroyed = &nf_nat_cleanup_conntrack;
+
+       /* Initialize fake conntrack so that NAT will skip it */
+       nf_conntrack_untracked.status |= IPS_NAT_DONE_MASK;
+
+       l3proto = nf_ct_l3proto_find_get((u_int16_t)AF_INET);
+       return 0;
+}
+
+/* Clear NAT section of all conntracks, in case we're loaded again. */
+static int clean_nat(struct nf_conn *i, void *data)
+{
+       struct nf_conn_nat *nat = nfct_nat(i);
+
+       if (!nat)
+               return 0;
+       memset(nat, 0, sizeof(nat));
+       i->status &= ~(IPS_NAT_MASK | IPS_NAT_DONE_MASK | IPS_SEQ_ADJUST);
+       return 0;
+}
+
+static void __exit nf_nat_cleanup(void)
+{
+       nf_ct_iterate_cleanup(&clean_nat, NULL);
+       nf_conntrack_destroyed = NULL;
+       vfree(bysource);
+       nf_ct_l3proto_put(l3proto);
+}
+
+MODULE_LICENSE("GPL");
+
+module_init(nf_nat_init);
+module_exit(nf_nat_cleanup);