git.rot13.org Git - linux/blob - net/netfilter/nf_conncount.c

   1 /*
   2  * count the number of connections matching an arbitrary key.
   3  *
   4  * (C) 2017 Red Hat GmbH
   5  * Author: Florian Westphal <fw@strlen.de>
   6  *
   7  * split from xt_connlimit.c:
   8  *   (c) 2000 Gerd Knorr <kraxel@bytesex.org>
   9  *   Nov 2002: Martin Bene <martin.bene@icomedias.com>:
  10  *              only ignore TIME_WAIT or gone connections
  11  *   (C) CC Computer Consultants GmbH, 2007
  12  */
  13 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  14 #include <linux/in.h>
  15 #include <linux/in6.h>
  16 #include <linux/ip.h>
  17 #include <linux/ipv6.h>
  18 #include <linux/jhash.h>
  19 #include <linux/slab.h>
  20 #include <linux/list.h>
  21 #include <linux/rbtree.h>
  22 #include <linux/module.h>
  23 #include <linux/random.h>
  24 #include <linux/skbuff.h>
  25 #include <linux/spinlock.h>
  26 #include <linux/netfilter/nf_conntrack_tcp.h>
  27 #include <linux/netfilter/x_tables.h>
  28 #include <net/netfilter/nf_conntrack.h>
  29 #include <net/netfilter/nf_conntrack_count.h>
  30 #include <net/netfilter/nf_conntrack_core.h>
  31 #include <net/netfilter/nf_conntrack_tuple.h>
  32 #include <net/netfilter/nf_conntrack_zones.h>
  33
  34 #define CONNCOUNT_SLOTS         256U
  35
  36 #ifdef CONFIG_LOCKDEP
  37 #define CONNCOUNT_LOCK_SLOTS    8U
  38 #else
  39 #define CONNCOUNT_LOCK_SLOTS    256U
  40 #endif
  41
  42 #define CONNCOUNT_GC_MAX_NODES  8
  43 #define MAX_KEYLEN              5
  44
  45 /* we will save the tuples of all connections we care about */
  46 struct nf_conncount_tuple {
  47         struct list_head                node;
  48         struct nf_conntrack_tuple       tuple;
  49         struct nf_conntrack_zone        zone;
  50         int                             cpu;
  51         u32                             jiffies32;
  52         bool                            dead;
  53         struct rcu_head                 rcu_head;
  54 };
  55
  56 struct nf_conncount_rb {
  57         struct rb_node node;
  58         struct nf_conncount_list list;
  59         u32 key[MAX_KEYLEN];
  60         struct rcu_head rcu_head;
  61 };
  62
  63 static spinlock_t nf_conncount_locks[CONNCOUNT_LOCK_SLOTS] __cacheline_aligned_in_smp;
  64
  65 struct nf_conncount_data {
  66         unsigned int keylen;
  67         struct rb_root root[CONNCOUNT_SLOTS];
  68         struct net *net;
  69         struct work_struct gc_work;
  70         unsigned long pending_trees[BITS_TO_LONGS(CONNCOUNT_SLOTS)];
  71         unsigned int gc_tree;
  72 };
  73
  74 static u_int32_t conncount_rnd __read_mostly;
  75 static struct kmem_cache *conncount_rb_cachep __read_mostly;
  76 static struct kmem_cache *conncount_conn_cachep __read_mostly;
  77
  78 static inline bool already_closed(const struct nf_conn *conn)
  79 {
  80         if (nf_ct_protonum(conn) == IPPROTO_TCP)
  81                 return conn->proto.tcp.state == TCP_CONNTRACK_TIME_WAIT ||
  82                        conn->proto.tcp.state == TCP_CONNTRACK_CLOSE;
  83         else
  84                 return false;
  85 }
  86
  87 static int key_diff(const u32 *a, const u32 *b, unsigned int klen)
  88 {
  89         return memcmp(a, b, klen * sizeof(u32));
  90 }
  91
  92 enum nf_conncount_list_add
  93 nf_conncount_add(struct nf_conncount_list *list,
  94                  const struct nf_conntrack_tuple *tuple,
  95                  const struct nf_conntrack_zone *zone)
  96 {
  97         struct nf_conncount_tuple *conn;
  98
  99         if (WARN_ON_ONCE(list->count > INT_MAX))
 100                 return NF_CONNCOUNT_ERR;
 101
 102         conn = kmem_cache_alloc(conncount_conn_cachep, GFP_ATOMIC);
 103         if (conn == NULL)
 104                 return NF_CONNCOUNT_ERR;
 105
 106         conn->tuple = *tuple;
 107         conn->zone = *zone;
 108         conn->cpu = raw_smp_processor_id();
 109         conn->jiffies32 = (u32)jiffies;
 110         conn->dead = false;
 111         spin_lock_bh(&list->list_lock);
 112         if (list->dead == true) {
 113                 kmem_cache_free(conncount_conn_cachep, conn);
 114                 spin_unlock_bh(&list->list_lock);
 115                 return NF_CONNCOUNT_SKIP;
 116         }
 117         list_add_tail(&conn->node, &list->head);
 118         list->count++;
 119         spin_unlock_bh(&list->list_lock);
 120         return NF_CONNCOUNT_ADDED;
 121 }
 122 EXPORT_SYMBOL_GPL(nf_conncount_add);
 123
 124 static void __conn_free(struct rcu_head *h)
 125 {
 126         struct nf_conncount_tuple *conn;
 127
 128         conn = container_of(h, struct nf_conncount_tuple, rcu_head);
 129         kmem_cache_free(conncount_conn_cachep, conn);
 130 }
 131
 132 static bool conn_free(struct nf_conncount_list *list,
 133                       struct nf_conncount_tuple *conn)
 134 {
 135         bool free_entry = false;
 136
 137         spin_lock_bh(&list->list_lock);
 138
 139         if (conn->dead) {
 140                 spin_unlock_bh(&list->list_lock);
 141                 return free_entry;
 142         }
 143
 144         list->count--;
 145         conn->dead = true;
 146         list_del_rcu(&conn->node);
 147         if (list->count == 0) {
 148                 list->dead = true;
 149                 free_entry = true;
 150         }
 151
 152         spin_unlock_bh(&list->list_lock);
 153         call_rcu(&conn->rcu_head, __conn_free);
 154         return free_entry;
 155 }
 156
 157 static const struct nf_conntrack_tuple_hash *
 158 find_or_evict(struct net *net, struct nf_conncount_list *list,
 159               struct nf_conncount_tuple *conn, bool *free_entry)
 160 {
 161         const struct nf_conntrack_tuple_hash *found;
 162         unsigned long a, b;
 163         int cpu = raw_smp_processor_id();
 164         __s32 age;
 165
 166         found = nf_conntrack_find_get(net, &conn->zone, &conn->tuple);
 167         if (found)
 168                 return found;
 169         b = conn->jiffies32;
 170         a = (u32)jiffies;
 171
 172         /* conn might have been added just before by another cpu and
 173          * might still be unconfirmed.  In this case, nf_conntrack_find()
 174          * returns no result.  Thus only evict if this cpu added the
 175          * stale entry or if the entry is older than two jiffies.
 176          */
 177         age = a - b;
 178         if (conn->cpu == cpu || age >= 2) {
 179                 *free_entry = conn_free(list, conn);
 180                 return ERR_PTR(-ENOENT);
 181         }
 182
 183         return ERR_PTR(-EAGAIN);
 184 }
 185
 186 void nf_conncount_lookup(struct net *net,
 187                          struct nf_conncount_list *list,
 188                          const struct nf_conntrack_tuple *tuple,
 189                          const struct nf_conntrack_zone *zone,
 190                          bool *addit)
 191 {
 192         const struct nf_conntrack_tuple_hash *found;
 193         struct nf_conncount_tuple *conn, *conn_n;
 194         struct nf_conn *found_ct;
 195         unsigned int collect = 0;
 196         bool free_entry = false;
 197
 198         /* best effort only */
 199         *addit = tuple ? true : false;
 200
 201         /* check the saved connections */
 202         list_for_each_entry_safe(conn, conn_n, &list->head, node) {
 203                 if (collect > CONNCOUNT_GC_MAX_NODES)
 204                         break;
 205
 206                 found = find_or_evict(net, list, conn, &free_entry);
 207                 if (IS_ERR(found)) {
 208                         /* Not found, but might be about to be confirmed */
 209                         if (PTR_ERR(found) == -EAGAIN) {
 210                                 if (!tuple)
 211                                         continue;
 212
 213                                 if (nf_ct_tuple_equal(&conn->tuple, tuple) &&
 214                                     nf_ct_zone_id(&conn->zone, conn->zone.dir) ==
 215                                     nf_ct_zone_id(zone, zone->dir))
 216                                         *addit = false;
 217                         } else if (PTR_ERR(found) == -ENOENT)
 218                                 collect++;
 219                         continue;
 220                 }
 221
 222                 found_ct = nf_ct_tuplehash_to_ctrack(found);
 223
 224                 if (tuple && nf_ct_tuple_equal(&conn->tuple, tuple) &&
 225                     nf_ct_zone_equal(found_ct, zone, zone->dir)) {
 226                         /*
 227                          * We should not see tuples twice unless someone hooks
 228                          * this into a table without "-p tcp --syn".
 229                          *
 230                          * Attempt to avoid a re-add in this case.
 231                          */
 232                         *addit = false;
 233                 } else if (already_closed(found_ct)) {
 234                         /*
 235                          * we do not care about connections which are
 236                          * closed already -> ditch it
 237                          */
 238                         nf_ct_put(found_ct);
 239                         conn_free(list, conn);
 240                         collect++;
 241                         continue;
 242                 }
 243
 244                 nf_ct_put(found_ct);
 245         }
 246 }
 247 EXPORT_SYMBOL_GPL(nf_conncount_lookup);
 248
 249 void nf_conncount_list_init(struct nf_conncount_list *list)
 250 {
 251         spin_lock_init(&list->list_lock);
 252         INIT_LIST_HEAD(&list->head);
 253         list->count = 0;
 254         list->dead = false;
 255 }
 256 EXPORT_SYMBOL_GPL(nf_conncount_list_init);
 257
 258 /* Return true if the list is empty */
 259 bool nf_conncount_gc_list(struct net *net,
 260                           struct nf_conncount_list *list)
 261 {
 262         const struct nf_conntrack_tuple_hash *found;
 263         struct nf_conncount_tuple *conn, *conn_n;
 264         struct nf_conn *found_ct;
 265         unsigned int collected = 0;
 266         bool free_entry = false;
 267         bool ret = false;
 268
 269         list_for_each_entry_safe(conn, conn_n, &list->head, node) {
 270                 found = find_or_evict(net, list, conn, &free_entry);
 271                 if (IS_ERR(found)) {
 272                         if (PTR_ERR(found) == -ENOENT)  {
 273                                 if (free_entry)
 274                                         return true;
 275                                 collected++;
 276                         }
 277                         continue;
 278                 }
 279
 280                 found_ct = nf_ct_tuplehash_to_ctrack(found);
 281                 if (already_closed(found_ct)) {
 282                         /*
 283                          * we do not care about connections which are
 284                          * closed already -> ditch it
 285                          */
 286                         nf_ct_put(found_ct);
 287                         if (conn_free(list, conn))
 288                                 return true;
 289                         collected++;
 290                         continue;
 291                 }
 292
 293                 nf_ct_put(found_ct);
 294                 if (collected > CONNCOUNT_GC_MAX_NODES)
 295                         return false;
 296         }
 297
 298         spin_lock_bh(&list->list_lock);
 299         if (!list->count) {
 300                 list->dead = true;
 301                 ret = true;
 302         }
 303         spin_unlock_bh(&list->list_lock);
 304
 305         return ret;
 306 }
 307 EXPORT_SYMBOL_GPL(nf_conncount_gc_list);
 308
 309 static void __tree_nodes_free(struct rcu_head *h)
 310 {
 311         struct nf_conncount_rb *rbconn;
 312
 313         rbconn = container_of(h, struct nf_conncount_rb, rcu_head);
 314         kmem_cache_free(conncount_rb_cachep, rbconn);
 315 }
 316
 317 static void tree_nodes_free(struct rb_root *root,
 318                             struct nf_conncount_rb *gc_nodes[],
 319                             unsigned int gc_count)
 320 {
 321         struct nf_conncount_rb *rbconn;
 322
 323         while (gc_count) {
 324                 rbconn = gc_nodes[--gc_count];
 325                 spin_lock(&rbconn->list.list_lock);
 326                 rb_erase(&rbconn->node, root);
 327                 call_rcu(&rbconn->rcu_head, __tree_nodes_free);
 328                 spin_unlock(&rbconn->list.list_lock);
 329         }
 330 }
 331
 332 static void schedule_gc_worker(struct nf_conncount_data *data, int tree)
 333 {
 334         set_bit(tree, data->pending_trees);
 335         schedule_work(&data->gc_work);
 336 }
 337
 338 static unsigned int
 339 insert_tree(struct net *net,
 340             struct nf_conncount_data *data,
 341             struct rb_root *root,
 342             unsigned int hash,
 343             const u32 *key,
 344             u8 keylen,
 345             const struct nf_conntrack_tuple *tuple,
 346             const struct nf_conntrack_zone *zone)
 347 {
 348         enum nf_conncount_list_add ret;
 349         struct nf_conncount_rb *gc_nodes[CONNCOUNT_GC_MAX_NODES];
 350         struct rb_node **rbnode, *parent;
 351         struct nf_conncount_rb *rbconn;
 352         struct nf_conncount_tuple *conn;
 353         unsigned int count = 0, gc_count = 0;
 354         bool node_found = false;
 355
 356         spin_lock_bh(&nf_conncount_locks[hash % CONNCOUNT_LOCK_SLOTS]);
 357
 358         parent = NULL;
 359         rbnode = &(root->rb_node);
 360         while (*rbnode) {
 361                 int diff;
 362                 rbconn = rb_entry(*rbnode, struct nf_conncount_rb, node);
 363
 364                 parent = *rbnode;
 365                 diff = key_diff(key, rbconn->key, keylen);
 366                 if (diff < 0) {
 367                         rbnode = &((*rbnode)->rb_left);
 368                 } else if (diff > 0) {
 369                         rbnode = &((*rbnode)->rb_right);
 370                 } else {
 371                         /* unlikely: other cpu added node already */
 372                         node_found = true;
 373                         ret = nf_conncount_add(&rbconn->list, tuple, zone);
 374                         if (ret == NF_CONNCOUNT_ERR) {
 375                                 count = 0; /* hotdrop */
 376                         } else if (ret == NF_CONNCOUNT_ADDED) {
 377                                 count = rbconn->list.count;
 378                         } else {
 379                                 /* NF_CONNCOUNT_SKIP, rbconn is already
 380                                  * reclaimed by gc, insert a new tree node
 381                                  */
 382                                 node_found = false;
 383                         }
 384                         break;
 385                 }
 386
 387                 if (gc_count >= ARRAY_SIZE(gc_nodes))
 388                         continue;
 389
 390                 if (nf_conncount_gc_list(net, &rbconn->list))
 391                         gc_nodes[gc_count++] = rbconn;
 392         }
 393
 394         if (gc_count) {
 395                 tree_nodes_free(root, gc_nodes, gc_count);
 396                 /* tree_node_free before new allocation permits
 397                  * allocator to re-use newly free'd object.
 398                  *
 399                  * This is a rare event; in most cases we will find
 400                  * existing node to re-use. (or gc_count is 0).
 401                  */
 402
 403                 if (gc_count >= ARRAY_SIZE(gc_nodes))
 404                         schedule_gc_worker(data, hash);
 405         }
 406
 407         if (node_found)
 408                 goto out_unlock;
 409
 410         /* expected case: match, insert new node */
 411         rbconn = kmem_cache_alloc(conncount_rb_cachep, GFP_ATOMIC);
 412         if (rbconn == NULL)
 413                 goto out_unlock;
 414
 415         conn = kmem_cache_alloc(conncount_conn_cachep, GFP_ATOMIC);
 416         if (conn == NULL) {
 417                 kmem_cache_free(conncount_rb_cachep, rbconn);
 418                 goto out_unlock;
 419         }
 420
 421         conn->tuple = *tuple;
 422         conn->zone = *zone;
 423         memcpy(rbconn->key, key, sizeof(u32) * keylen);
 424
 425         nf_conncount_list_init(&rbconn->list);
 426         list_add(&conn->node, &rbconn->list.head);
 427         count = 1;
 428         rbconn->list.count = count;
 429
 430         rb_link_node_rcu(&rbconn->node, parent, rbnode);
 431         rb_insert_color(&rbconn->node, root);
 432 out_unlock:
 433         spin_unlock_bh(&nf_conncount_locks[hash % CONNCOUNT_LOCK_SLOTS]);
 434         return count;
 435 }
 436
 437 static unsigned int
 438 count_tree(struct net *net,
 439            struct nf_conncount_data *data,
 440            const u32 *key,
 441            const struct nf_conntrack_tuple *tuple,
 442            const struct nf_conntrack_zone *zone)
 443 {
 444         enum nf_conncount_list_add ret;
 445         struct rb_root *root;
 446         struct rb_node *parent;
 447         struct nf_conncount_rb *rbconn;
 448         unsigned int hash;
 449         u8 keylen = data->keylen;
 450
 451         hash = jhash2(key, data->keylen, conncount_rnd) % CONNCOUNT_SLOTS;
 452         root = &data->root[hash];
 453
 454         parent = rcu_dereference_raw(root->rb_node);
 455         while (parent) {
 456                 int diff;
 457                 bool addit;
 458
 459                 rbconn = rb_entry(parent, struct nf_conncount_rb, node);
 460
 461                 diff = key_diff(key, rbconn->key, keylen);
 462                 if (diff < 0) {
 463                         parent = rcu_dereference_raw(parent->rb_left);
 464                 } else if (diff > 0) {
 465                         parent = rcu_dereference_raw(parent->rb_right);
 466                 } else {
 467                         /* same source network -> be counted! */
 468                         nf_conncount_lookup(net, &rbconn->list, tuple, zone,
 469                                             &addit);
 470
 471                         if (!addit)
 472                                 return rbconn->list.count;
 473
 474                         ret = nf_conncount_add(&rbconn->list, tuple, zone);
 475                         if (ret == NF_CONNCOUNT_ERR) {
 476                                 return 0; /* hotdrop */
 477                         } else if (ret == NF_CONNCOUNT_ADDED) {
 478                                 return rbconn->list.count;
 479                         } else {
 480                                 /* NF_CONNCOUNT_SKIP, rbconn is already
 481                                  * reclaimed by gc, insert a new tree node
 482                                  */
 483                                 break;
 484                         }
 485                 }
 486         }
 487
 488         if (!tuple)
 489                 return 0;
 490
 491         return insert_tree(net, data, root, hash, key, keylen, tuple, zone);
 492 }
 493
 494 static void tree_gc_worker(struct work_struct *work)
 495 {
 496         struct nf_conncount_data *data = container_of(work, struct nf_conncount_data, gc_work);
 497         struct nf_conncount_rb *gc_nodes[CONNCOUNT_GC_MAX_NODES], *rbconn;
 498         struct rb_root *root;
 499         struct rb_node *node;
 500         unsigned int tree, next_tree, gc_count = 0;
 501
 502         tree = data->gc_tree % CONNCOUNT_LOCK_SLOTS;
 503         root = &data->root[tree];
 504
 505         rcu_read_lock();
 506         for (node = rb_first(root); node != NULL; node = rb_next(node)) {
 507                 rbconn = rb_entry(node, struct nf_conncount_rb, node);
 508                 if (nf_conncount_gc_list(data->net, &rbconn->list))
 509                         gc_nodes[gc_count++] = rbconn;
 510         }
 511         rcu_read_unlock();
 512
 513         spin_lock_bh(&nf_conncount_locks[tree]);
 514
 515         if (gc_count) {
 516                 tree_nodes_free(root, gc_nodes, gc_count);
 517         }
 518
 519         clear_bit(tree, data->pending_trees);
 520
 521         next_tree = (tree + 1) % CONNCOUNT_SLOTS;
 522         next_tree = find_next_bit(data->pending_trees, next_tree, CONNCOUNT_SLOTS);
 523
 524         if (next_tree < CONNCOUNT_SLOTS) {
 525                 data->gc_tree = next_tree;
 526                 schedule_work(work);
 527         }
 528
 529         spin_unlock_bh(&nf_conncount_locks[tree]);
 530 }
 531
 532 /* Count and return number of conntrack entries in 'net' with particular 'key'.
 533  * If 'tuple' is not null, insert it into the accounting data structure.
 534  * Call with RCU read lock.
 535  */
 536 unsigned int nf_conncount_count(struct net *net,
 537                                 struct nf_conncount_data *data,
 538                                 const u32 *key,
 539                                 const struct nf_conntrack_tuple *tuple,
 540                                 const struct nf_conntrack_zone *zone)
 541 {
 542         return count_tree(net, data, key, tuple, zone);
 543 }
 544 EXPORT_SYMBOL_GPL(nf_conncount_count);
 545
 546 struct nf_conncount_data *nf_conncount_init(struct net *net, unsigned int family,
 547                                             unsigned int keylen)
 548 {
 549         struct nf_conncount_data *data;
 550         int ret, i;
 551
 552         if (keylen % sizeof(u32) ||
 553             keylen / sizeof(u32) > MAX_KEYLEN ||
 554             keylen == 0)
 555                 return ERR_PTR(-EINVAL);
 556
 557         net_get_random_once(&conncount_rnd, sizeof(conncount_rnd));
 558
 559         data = kmalloc(sizeof(*data), GFP_KERNEL);
 560         if (!data)
 561                 return ERR_PTR(-ENOMEM);
 562
 563         ret = nf_ct_netns_get(net, family);
 564         if (ret < 0) {
 565                 kfree(data);
 566                 return ERR_PTR(ret);
 567         }
 568
 569         for (i = 0; i < ARRAY_SIZE(data->root); ++i)
 570                 data->root[i] = RB_ROOT;
 571
 572         data->keylen = keylen / sizeof(u32);
 573         data->net = net;
 574         INIT_WORK(&data->gc_work, tree_gc_worker);
 575
 576         return data;
 577 }
 578 EXPORT_SYMBOL_GPL(nf_conncount_init);
 579
 580 void nf_conncount_cache_free(struct nf_conncount_list *list)
 581 {
 582         struct nf_conncount_tuple *conn, *conn_n;
 583
 584         list_for_each_entry_safe(conn, conn_n, &list->head, node)
 585                 kmem_cache_free(conncount_conn_cachep, conn);
 586 }
 587 EXPORT_SYMBOL_GPL(nf_conncount_cache_free);
 588
 589 static void destroy_tree(struct rb_root *r)
 590 {
 591         struct nf_conncount_rb *rbconn;
 592         struct rb_node *node;
 593
 594         while ((node = rb_first(r)) != NULL) {
 595                 rbconn = rb_entry(node, struct nf_conncount_rb, node);
 596
 597                 rb_erase(node, r);
 598
 599                 nf_conncount_cache_free(&rbconn->list);
 600
 601                 kmem_cache_free(conncount_rb_cachep, rbconn);
 602         }
 603 }
 604
 605 void nf_conncount_destroy(struct net *net, unsigned int family,
 606                           struct nf_conncount_data *data)
 607 {
 608         unsigned int i;
 609
 610         cancel_work_sync(&data->gc_work);
 611         nf_ct_netns_put(net, family);
 612
 613         for (i = 0; i < ARRAY_SIZE(data->root); ++i)
 614                 destroy_tree(&data->root[i]);
 615
 616         kfree(data);
 617 }
 618 EXPORT_SYMBOL_GPL(nf_conncount_destroy);
 619
 620 static int __init nf_conncount_modinit(void)
 621 {
 622         int i;
 623
 624         BUILD_BUG_ON(CONNCOUNT_LOCK_SLOTS > CONNCOUNT_SLOTS);
 625         BUILD_BUG_ON((CONNCOUNT_SLOTS % CONNCOUNT_LOCK_SLOTS) != 0);
 626
 627         for (i = 0; i < CONNCOUNT_LOCK_SLOTS; ++i)
 628                 spin_lock_init(&nf_conncount_locks[i]);
 629
 630         conncount_conn_cachep = kmem_cache_create("nf_conncount_tuple",
 631                                            sizeof(struct nf_conncount_tuple),
 632                                            0, 0, NULL);
 633         if (!conncount_conn_cachep)
 634                 return -ENOMEM;
 635
 636         conncount_rb_cachep = kmem_cache_create("nf_conncount_rb",
 637                                            sizeof(struct nf_conncount_rb),
 638                                            0, 0, NULL);
 639         if (!conncount_rb_cachep) {
 640                 kmem_cache_destroy(conncount_conn_cachep);
 641                 return -ENOMEM;
 642         }
 643
 644         return 0;
 645 }
 646
 647 static void __exit nf_conncount_modexit(void)
 648 {
 649         kmem_cache_destroy(conncount_conn_cachep);
 650         kmem_cache_destroy(conncount_rb_cachep);
 651 }
 652
 653 module_init(nf_conncount_modinit);
 654 module_exit(nf_conncount_modexit);
 655 MODULE_AUTHOR("Jan Engelhardt <jengelh@medozas.de>");
 656 MODULE_AUTHOR("Florian Westphal <fw@strlen.de>");
 657 MODULE_DESCRIPTION("netfilter: count number of connections matching a key");
 658 MODULE_LICENSE("GPL");