[NETFILTER]: xt_helper: use RCU
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / net / ipv4 / netfilter / nf_nat_core.c
index ac7e8abbbdeb0197d9b6374df5550eb5a22f90b7..f242ac61b3eb71cc741b1394edaf518df3fdcc3e 100644 (file)
@@ -12,7 +12,6 @@
 #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>
@@ -44,8 +43,9 @@ static struct nf_conntrack_l3proto *l3proto = NULL;
 
 /* Calculated at init based on memory size */
 static unsigned int nf_nat_htable_size;
+static int nf_nat_vmalloced;
 
-static struct list_head *bysource;
+static struct hlist_head *bysource;
 
 #define MAX_IP_NAT_PROTO 256
 static struct nf_nat_protocol *nf_nat_protos[MAX_IP_NAT_PROTO];
@@ -87,20 +87,6 @@ hash_by_src(const struct nf_conntrack_tuple *tuple)
                            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);
-       nat->info.ct = NULL;
-       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,
@@ -167,10 +153,11 @@ find_appropriate_src(const struct nf_conntrack_tuple *tuple,
        unsigned int h = hash_by_src(tuple);
        struct nf_conn_nat *nat;
        struct nf_conn *ct;
+       struct hlist_node *n;
 
        read_lock_bh(&nf_nat_lock);
-       list_for_each_entry(nat, &bysource[h], info.bysource) {
-               ct = nat->info.ct;
+       hlist_for_each_entry(nat, n, &bysource[h], bysource) {
+               ct = nat->ct;
                if (same_src(ct, tuple)) {
                        /* Copy source part from reply tuple. */
                        nf_ct_invert_tuplepr(result,
@@ -297,11 +284,20 @@ nf_nat_setup_info(struct nf_conn *ct,
                  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;
+       struct nf_conn_nat *nat;
        int have_to_hash = !(ct->status & IPS_NAT_DONE_MASK);
        enum nf_nat_manip_type maniptype = HOOK2MANIP(hooknum);
 
+       /* nat helper or nfctnetlink also setup binding */
+       nat = nfct_nat(ct);
+       if (!nat) {
+               nat = nf_ct_ext_add(ct, NF_CT_EXT_NAT, GFP_ATOMIC);
+               if (nat == NULL) {
+                       DEBUGP("failed to add NAT extension\n");
+                       return NF_ACCEPT;
+               }
+       }
+
        NF_CT_ASSERT(hooknum == NF_IP_PRE_ROUTING ||
                     hooknum == NF_IP_POST_ROUTING ||
                     hooknum == NF_IP_LOCAL_IN ||
@@ -338,8 +334,10 @@ nf_nat_setup_info(struct nf_conn *ct,
 
                srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
                write_lock_bh(&nf_nat_lock);
-               info->ct = ct;
-               list_add(&info->bysource, &bysource[srchash]);
+               /* nf_conntrack_alter_reply might re-allocate exntension aera */
+               nat = nfct_nat(ct);
+               nat->ct = ct;
+               hlist_add_head(&nat->bysource, &bysource[srchash]);
                write_unlock_bh(&nf_nat_lock);
        }
 
@@ -592,17 +590,69 @@ nf_nat_port_nfattr_to_range(struct nfattr *tb[], struct nf_nat_range *range)
 EXPORT_SYMBOL_GPL(nf_nat_port_range_to_nfattr);
 #endif
 
+/* Noone using conntrack by the time this called. */
+static void nf_nat_cleanup_conntrack(struct nf_conn *ct)
+{
+       struct nf_conn_nat *nat = nf_ct_ext_find(ct, NF_CT_EXT_NAT);
+
+       if (nat == NULL || nat->ct == NULL)
+               return;
+
+       NF_CT_ASSERT(nat->ct->status & IPS_NAT_DONE_MASK);
+
+       write_lock_bh(&nf_nat_lock);
+       hlist_del(&nat->bysource);
+       nat->ct = NULL;
+       write_unlock_bh(&nf_nat_lock);
+}
+
+static void nf_nat_move_storage(struct nf_conn *conntrack, void *old)
+{
+       struct nf_conn_nat *new_nat = nf_ct_ext_find(conntrack, NF_CT_EXT_NAT);
+       struct nf_conn_nat *old_nat = (struct nf_conn_nat *)old;
+       struct nf_conn *ct = old_nat->ct;
+       unsigned int srchash;
+
+       if (!(ct->status & IPS_NAT_DONE_MASK))
+               return;
+
+       srchash = hash_by_src(&ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple);
+
+       write_lock_bh(&nf_nat_lock);
+       hlist_replace_rcu(&old_nat->bysource, &new_nat->bysource);
+       new_nat->ct = ct;
+       write_unlock_bh(&nf_nat_lock);
+}
+
+static struct nf_ct_ext_type nat_extend __read_mostly = {
+       .len            = sizeof(struct nf_conn_nat),
+       .align          = __alignof__(struct nf_conn_nat),
+       .destroy        = nf_nat_cleanup_conntrack,
+       .move           = nf_nat_move_storage,
+       .id             = NF_CT_EXT_NAT,
+       .flags          = NF_CT_EXT_F_PREALLOC,
+};
+
 static int __init nf_nat_init(void)
 {
        size_t i;
+       int ret;
+
+       ret = nf_ct_extend_register(&nat_extend);
+       if (ret < 0) {
+               printk(KERN_ERR "nf_nat_core: Unable to register extension\n");
+               return ret;
+       }
 
        /* 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;
+       bysource = nf_ct_alloc_hashtable(&nf_nat_htable_size,
+                                        &nf_nat_vmalloced);
+       if (!bysource) {
+               ret = -ENOMEM;
+               goto cleanup_extend;
+       }
 
        /* Sew in builtin protocols. */
        write_lock_bh(&nf_nat_lock);
@@ -614,18 +664,18 @@ static int __init nf_nat_init(void)
        write_unlock_bh(&nf_nat_lock);
 
        for (i = 0; i < nf_nat_htable_size; i++) {
-               INIT_LIST_HEAD(&bysource[i]);
+               INIT_HLIST_HEAD(&bysource[i]);
        }
 
-       /* FIXME: Man, this is a hack.  <SIGH> */
-       NF_CT_ASSERT(rcu_dereference(nf_conntrack_destroyed) == NULL);
-       rcu_assign_pointer(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;
+
+ cleanup_extend:
+       nf_ct_extend_unregister(&nat_extend);
+       return ret;
 }
 
 /* Clear NAT section of all conntracks, in case we're loaded again. */
@@ -643,10 +693,10 @@ static int clean_nat(struct nf_conn *i, void *data)
 static void __exit nf_nat_cleanup(void)
 {
        nf_ct_iterate_cleanup(&clean_nat, NULL);
-       rcu_assign_pointer(nf_conntrack_destroyed, NULL);
        synchronize_rcu();
-       vfree(bysource);
+       nf_ct_free_hashtable(bysource, nf_nat_vmalloced, nf_nat_htable_size);
        nf_ct_l3proto_put(l3proto);
+       nf_ct_extend_unregister(&nat_extend);
 }
 
 MODULE_LICENSE("GPL");